Cell types, subtypes, and cell states in the adult heart
When discussing cardiac cell types and their functions, we first need to define cellular identities. However, this is far from trivial. Traditionally, cell types (e.g. cardiomyocytes) and specialized subtypes (e.g. ventricular cardiomyocytes, atrial cardiomyocytes, pacemaker cardiomyocytes) were annotated according to their morphological phenotype, the presence of certain surface antigens, their function, and/or later genetic lineage tracing. The rise of single-cell RNA-seq (scRNA-seq) technologies and the accompanying possibility of making ever more subtle distinctions, but also the user-dependence of RNA-seq data analysis, presentation and interpretation, have refuelled the ongoing debate on how to define a ‘cell type’ [
123,
195,
218]. For scRNA-seq, RNA of each cell is converted into complementary DNA (cDNA) by reverse transcription, individually barcoded by the insertion of a short nucleotide sequence, amplified, and processed for sequencing [
185]. Bioinformatics analyses of the resulting raw sequencing data are then used to determine gene expression profiles of individual cells. Similarity of cells is defined according to the similarity in their gene expression, and this information is then used to define their positions in a multidimensional virtual space. To reduce dimensionality, genes that show high correlation in expression are grouped into modules [
195]. Depending on how far the dimensionality of sequencing data is reduced, each cell can be assigned to a larger or smaller cluster of cells with similar gene expression profiles (Fig.
1B). This approach has two important implications: first, the number of cell types detected by this kind of analysis depends on the extent of dimensionality reduction prescribed in the analysis. Second, newly defined cell types, inferred from transcriptome analysis, require validation by a different technique.
Since changes in gene expression can be highly dynamic within one and the same cell population, a different transcriptomic signature does not necessarily define a distinct cell type. Differences that are determined by a cell’s local environment, or that occur in response to stimuli—but which do not alter cell identity—are often referred to as ‘cell states’ [
123,
195]. Thus, one cell type can have multiple states—which, in the absence of agreed standards on defining cell types—has given rise to large variability in the way RNA-seq based information is presented visually and conceptually. Their identification generally requires a priori knowledge about the relevant biology, which is not always available when exploring hitherto unknown biological processes. Furthermore, gene expression data can be integrated with information from in situ hybridization or targeted barcoding to localize cells in the tissue, providing spatial—albeit thus far largely two-dimensional—information [
83,
100]. Finally, epigenetic information such as chromatin accessibility can be included in the analysis to discriminate cell types and states [
156,
187,
220]. Both scRNA-seq and single nucleus RNA-seq (snRNA-seq) techniques are used by researchers world-wide, including large international consortia, aiming to identify and spatially map all cell types in the human body [
165].
In 2016, first reports on scRNA-seq of the developing mouse heart were published [
40,
98] and since then, the technology has spread quickly in the cardiovascular field [
28,
53,
60,
73,
99,
141,
166,
181,
188], right through to spatio-temporally resolved RNA-seq [
10,
95,
111]. For technical reasons that are discussed in more detail below, nuclei instead of cells have been used in several studies for snRNA-seq of the heart [
60,
73,
188]. Transcriptomics studies have described a large variety of cell types in adult human and rodent heart tissue, each of them with different subtypes and states. These include well-known players such as cardiomyocytes, endothelial cells, smooth muscle cells, fibroblasts, and myeloid or lymphoid immune cells. Other cardiac cell types, such as pericytes, adipocytes, neurons or Schwann cells, and their roles in physiology or pathophysiology of the heart, are also gaining increasing interest. An overview of commonly used cell type marker genes or surface antigens is provided in Table
1.
Table 1
Main cell populations and their functions in the healthy adult heart
Cardiomyocytes | | PLN, PCM-1 | |
Ventricular | Heart contraction | αMHC |
Atrial | Heart rhythm, heart contraction | MYL7 |
Endothelial cells | Regulation of vascular contraction Trafficking of cells and metabolites | CDH5, PECAM-1 | |
Smooth muscle cells | Vascular contraction (arteries, arterioles) | MYH11, ACTA2, TAGLN | |
Pericytes | Vascular contraction (capillaries) | NG2, PDGFRβ | |
Fibroblasts | Extracellular matrix formation | PDGFRα, TCF21, DDR2 | |
Myofibroblasts | ACTA2 |
Leukocytes | | CD45 | |
Myeloid | Phagocytosis of cellular debris, immune surveillance | F4/80, CD11b | |
Lymphoid | Adaptive immune response | IL7R | |
Adipocytes | Energy supply, mechanical protection | LEP | |
Neurons | Sympathetic and parasympathetic regulation of heart rate and contraction | NRXN1, NRXN2 | |
Schwann cells | Insulating nerve fibres | PLP-1 | |
Mesothelial cells | Quiescent | WT-1 | |
The relative cell numbers of each population in the myocardium are still a matter of debate. Single-cell preparations are affected by differences in cell viability, caused by the techniques used for tissue disruption, cell isolation, and sorting. Early flow cytometry-based analyses of single-cell preparations of adult murine myocardium reported 56% cardiomyocytes (identified by α myosin heavy chain expression), 27% fibroblasts (discoidin-domain receptor 2; DDR2), 10% smooth muscle cells (α smooth muscle actin; αSMA), and 7% endothelial cells (platelet endothelial cell adhesion molecule-1, generally referred to as CD31) [
14]. Note that these percentages relate to ‘cells identified and sorted’, not ‘cells in situ’, as is evident from the fact that they add up to 100% for just 4 of the many cell types known to be present in cardiac tissue. Slightly different results were obtained in snRNA-seq experiments on adult human ventricular myocardium, identifying 49% cardiomyocytes, 21% smooth muscle cells, 16% fibroblasts, 8% endothelial cells, and 5% immune cells [
102]. In contrast, other studies using nucleus-labelling techniques found only about 30% cardiomyocytes in adult mouse and human hearts [
15,
58], as had been indicated in earlier histomorphology-based studies on rat myocardium [
125]. The current understanding suggests that an overwhelming majority of cardiac non-myocytes, identified using a combination of immunohistochemistry and flow cytometry, consists of endothelial cells (60% of non-myocytes), followed by mesenchymal cells including fibroblasts, smooth muscle cells, and pericytes (27% of non-myocytes) and leukocytes (9% of non-myocytes [
148]. Numbers of rare cell types (neurons, Schwann cells, adipocytes) have not yet been reliably quantified. It is important to note that in addition to experimental aspects, other factors such as age or species may influence cardiac cell type composition. For instance, a modest decline of endothelial and mesenchymal cell numbers in the aging human heart has been reported [
15]. In mice, the embryo-derived population of cardiac macrophages is gradually replaced by monocyte-dependent macrophages [
120]. Both observations illustrate the dynamic turnover of cells in the adult heart.
Canonical and non-canonical functions of cardiac cells
Cardiomyocytes are the main actors in cardiac pump function. So-called ‘working cardiomyocytes’ are comparatively large cells (> 104 µm3) that are typically brick-shaped. They are connected end-to-end via intercalated discs, where desmosomes and fascia adherens junctions provide mechanical, and connexins electrical coupling. In addition, lateral mechanical connections link the contractile machinery at the z-disks to the extracellular matrix (ECM) as the main force-bearing structural element of the myocardium (a ‘deformable skeleton’), while lateral electrical connections occur mainly within the 4–6 cells thick layers of myocardium, referred to as ‘sheets’ or, more fittingly, ‘sheetles’.
Cardiomyocytes are densely packed with arrays of structural proteins that form the sarcomeres, membrane systems that enable Ca
2+-mediated excitation–contraction coupling, and mitochondria. Cardiomyocyte contraction is a highly orchestrated process, involving multiple extra- and intracardiac feed-forward and feedback mechanisms [
161]. Following cardiomyocyte depolarization, extracellular Ca
2+ enters the cytosol via voltage-dependent Ca
2+ channels and triggers further Ca
2+ release from the sarcoplasmic reticulum. For relaxation, Ca
2+ has to be removed from the cytosol. In steady-state, the amount released from intracellular stores is returned by the sarcoplasmic/endoplasmic reticulum calcium ATPase, while the initial ‘trigger-Ca
2+’ is extruded back to the extracellular space via the sodium Na
+/Ca
2+ exchanger [
17,
49]. The rapid and efficient activation of intracellular contractile units of a cardiomyocyte depends on transverse (T-) tubules, a network of membrane invaginations containing ion channels and transporters. The narrow and tortuous structure of the T-tubular system may impede diffusion and cause regional microdomains of different ion concentrations [
171]. Similar to the heart as a whole that serves as a pressure-suction-pump in the circulation, T-tubules are squeezed by contracting cardiomyocytes, pushing and pulling extracellular fluid between T-tubular lumen and bulk extracellular space. This recently described mechanism accelerates T-tubular diffusion dynamics, which may be important in particular at high heart rates [
94,
171].
Sarcomeres are highly organized structures composed of actin, myosin, and titin filaments, and a number of accessory proteins [
202]. During cardiomyocyte contraction, myosin filament heads interact with actin filaments, forming so-called ‘crossbridges’ that govern force development and sarcomere shortening. This crossbridge cycling requires the hydrolysis of one ATP per power-stroke, leading to an immense flux of ATP that is reflected by a high mitochondrial density in cardiomyocytes [
42,
121]. Cardiomyocyte mitochondria communicate with one-another [
75], and they participate in Ca
2+ handling [
177,
179].
While cardiomyocytes from left or right ventricle share high similarity [
102], atrial cardiomyocytes have a different phenotype, with distinct Ca
2+-handling, contractile and electrophysiological properties [
21,
128,
204]. Atrial cardiomyocyte sarcomeres contain different contractile protein isoforms, and while they develop less mechanical force, compared to ventricular myocytes [
128], they contract faster [
20,
127].
Working cardiomyocytes in the atria and ventricles are activated by a well-coordinated action potential wave, which originates from sino-atrial node pacemaker cells, and spreads through the atria, the atrio-ventricular node and His–Purkinje system, to the ventricular myocardium. The cells of sinus node and conducting system are specialized cardiomyocyte subtypes that show a distinct repertoire of ion channel expression and activity, which conveys upon them the spontaneous rhythmic depolarization that underlies pacemaking [
204]. These cells are also cardiomyocytes, showing cross striations caused by sarcomeric arrangement of their contractile filaments, even if that had been questioned or overlooked in some of the early studies.
During pre-natal heart growth, cardiomyocytes show a high proliferation rate [
215]. After birth, cardiomyocyte proliferation rate declines rapidly and remains at no more than about ~ 1% in adult human [
51,
201]. When the cell cycle abrogates before cytokinesis, cardiomyocytes become multinucleated, polyploid, or both. The extent of bi- or multinucleated (human 26%, mouse 78%) and polyploid cardiomyocytes (human 58%, mouse 10%) in the adult heart varies between species [
6,
15]. Postnatal growth of the adult heart is largely a function of cardiomyocyte hypertrophy, often associated with increased ploidy [
69,
215]. Cardiomyocyte growth can be triggered by a variety of physiological or pathological stimuli, including mechanical and biochemical factors. For example, physiological stimuli caused by exercise or pregnancy can promote cardiomyocyte growth while maintaining or improving cardiac function, whereas pathological stimuli such as sustained volume or pressure overload (in valve disease of hypertension) can lead to hypertrophy with impaired function, altered metabolism, and dysregulated intracellular signalling [
126]. Cardiomyocytes not only receive growth signals, but they may stimulate local blood vessel sprouting and innervation via secretion of growth factors such as vascular endothelial cell growth factor (VEGF) [
134] or nerve growth factor [
44].
Endothelial cells form the inner layer of the vasculature. They have distinct functions according to their localization and their association with different vascular beds. In the heart, capillary endothelial cells represent the largest cell population [
102]. Other subtypes include arterial, venous, lymphatic, and endocardial endothelial cells [
102]. All endothelial cells share high expression of cadherin 5 [
102,
140] and platelet endothelial cell adhesion molecule-1 (which shows weak expression also in the hematopoietic lineage; [
140]. Subspecification of endothelial cells is tightly regulated by the activity of transcription factors, in particular SOX (SRY-related HMG-box) and FOX (forkhead box) family members [
112,
154].
Endothelial cells sense and respond to mechanical and biochemical stimuli [
34,
41,
167]. Factors such as blood flow, mechanical stretch, and the interaction of cell membrane proteins with the extracellular environment influence transcription factor activity and gene expression in endothelial cells [
143]. It has been proposed that the presence of pulsatile or laminar flow, and of high or low shear stress determine arterial versus venous endothelial cell differentiation [
154]. Moreover, it has been shown recently that mechanical stress can induce endothelial-to-mesenchymal transition by activating the force-sensitive transforming growth factor β type I receptor kinase [
115]. This represents an example of how cells can change their phenotype in response to external stimuli, taking different functions in homeostasis or patho/physiological remodelling [
217].
Endothelial cells are the main source of nitric oxide in the heart, which via cyclic guanosine monophosphate (cGMP)-dependent signalling induces smooth muscle cell relaxation and vascular dilation [
203]. Endothelin-1 (ET-1) from endothelial cells is a potent vasoconstrictor, primarily by acting on ET
A receptors on smooth muscle cells. On the other hand, autocrine ET-1 signalling, acting via ET
B receptors on endothelial cells, increases nitric oxide synthesis and inhibits ET-1 production in a negative feedback loop [
203].
Cardiac capillary endothelial cells form a dense layer, with cells connected via tight junctions, which controls vascular permeability and trafficking of cells between blood and the surrounding cardiac tissue [
47]. The high energy demand of cardiomyocytes demands high flux rates for oxygen and carbon dioxide, which benefits from a tight capillary network [
134] with a capillaries-to-myocyte ratio of between 1.3:1 and 1.5:1 [
90,
107]. During angiogenesis, a complex network of angiocrine factors such as VEGF guides endothelial cells to alter their state, proliferate, migrate, and form new capillaries [
48].
Smooth muscle cells are a major constituent of coronary arterial and arteriolar walls. Their physiological function involves active changes in vessel cross-sectional area by contraction of circularly oriented cells, thereby controlling vascular resistance and—hence—regional blood flow (of note, the resistance R to flow is a fourth power inverse function of vessel radius r: R ~ r
−4) [
46,
104]. Efficient autoregulation of myocardial perfusion is of pivotal importance as it is positively correlated with contractile function [
71]. In the healthy heart, arterioles are the primary site of flow regulation. Several autoregulatory mechanisms (endothelial, neural, metabolic, myogenic) adjust myocardial blood supply through effects on vascular smooth muscle cells [
82]. In addition, smooth muscle cells respond to a number of paracrine and endocrine factors such as nitric oxide, endothelin-1, angiotensin II, aldosterone, or norepinephrine, secreted by cardiac cells or delivered through the circulation [
46].
Traditionally, a contractile and a synthetic smooth muscle cell phenotype have been distinguished, though these are more likely to represent two ends of a continuum. Contractile smooth muscle cells are packed with (non-sarcomeric) myofilaments, which can be identified by expression of myosin heavy chain 11 or smooth muscle cell actin α 2; [
25,
184]). Similar to cardiomyocytes, smooth muscle cell contraction depends on Ca
2+ fluxes, but it is more than an order of magnitude slower [
46].
Substantial plasticity of smooth muscle cells has been observed, including conversion to myofibroblast- or macrophage-like phenotypes [
184]. Although smooth muscle cells are not thought to give rise to fibroblasts in the healthy heart [
84], they modulate ECM remodelling by paracrine signalling [
13,
150].
Pericytes are mesenchymal cells that share phenotypical similarities with smooth muscle cells and it has been suggested both cell types have a shared lineage, deriving from endocardial progenitor cells [
30]. In pericytes, but not exclusively there, high expression of neuron–glial antigen 2 and platelet-derived growth factor receptor β can be detected [
11]. Due to the lack of specific markers, pericytes and smooth muscle cells are usually distinguished by anatomical localization and sometimes summarized as mural cells [
133,
148]. Pericytes are largely restricted to the microvasculature [
9,
12,
27]. They discontinuously surround capillaries, forming circumferential and longitudinal processes that can contribute to vessel contraction, including contributions to post-ischaemic non-reflow phenomena [
38,
70,
133].
Together with smooth muscle cells and pericytes,
fibroblasts belong to the group of mesenchymal cells. They are characterized by high expression of platelet-derived growth factor receptor α, DDR2, and transcription factor 21 [
84,
141,
192]. Cardiac fibroblasts may be distinguished by their localization (e.g. ventricular or atrial interstitial fibroblasts versus valvular fibroblasts), activation status (e.g. fibroblast, myofibroblast), or origin. While the vast majority of fibroblasts in the healthy heart is of epicardial origin, there is an ongoing debate whether other sources contribute to the cardiac fibroblast population after injury, including endothelial cells, bone marrow-derived cells, and other mesenchymal cells such as smooth muscle cells or pericytes [
84,
141,
182,
191].
Fibroblasts secrete collagens and proteoglycans [
55] and therefore they are traditionally regarded as the ‘cells that create and maintain... extracellular matrix’ [
152]. However, over the last 30 years our view has substantially evolved, as it is evident that other cell types contribute to the ECM, while fibroblasts are now seen as much more versatile cells. Upon mechanical or biochemical activation, fibroblasts express contractile proteins such as αSMA and are then considered to be myofibroblasts [
152]. In the context of cardiac lesion repair, up to four states have been suggested: resident fibroblast, activated fibroblast, myofibroblast, and matrifibrocyte, all with distinct functional properties [
56]. Fibroblasts interact with one-another, and with other cell types, via multiple biochemical (cytokines, growth factors) [
196] and biophysical cues (mechanical and electrical contacts) [
24] to steer cardiomyocyte and non-myocyte functions as reviewed before and addressed in more detail below.
Cardiac immune cells comprise all major leukocyte classes of the innate and adaptive immune system, including myeloid cells such as macrophages, monocytes, dendritic cells, neutrophils, or mast cells, and lymphoid cells such as B cells and T cells. Each of these major cell populations contains several subtypes and states [
102,
186,
189]. The cardiac macrophage population stems from different origins and can be distinguished from other leukocyte populations by combined surface expression of respective markers (CD45
+, F4/80
+, CD11b
+) [
189]. In steady-state, the population mainly consists of tissue-resident macrophages derived from yolk-sac progenitors, maintained by local proliferation, and a smaller, C–C chemokine receptor 2 (CCR2)
+ macrophage population derived from circulating monocytes [
50,
189]. After myocardial injury, CCR2
+ monocytes make a major contribution to replenishment of the cardiac macrophage pool [
67]. In the healthy heart, tissue-resident macrophages have essential roles in maintaining tissue homeostasis, as they are involved in ECM turnover and in the removal of cellular debris [
189], which is of particular importance for organs that contain cells that are maintained for the life-time of an individual (see information on post-natal cardiomyocyte proliferation, above). Accordingly, experimental depletion of resident macrophages impairs the elimination of dysfunctional mitochondrial fragments from cardiomyocytes, leading to disturbed metabolism and ventricular dysfunction [
131].
The heart is covered by the epicardium, a thin layer of
mesothelial cells, and in particular in the perivascular areas by
adipocytes [
102]. The epicardium has been recognized as a multipotent cardiac progenitor tissue [
26]. During cardiac development, Wilms’ tumour 1 expressing epicardial cells undergo epithelial-to-mesenchymal transition and give rise to multiple cell types, including fibroblasts, smooth muscle cells, and pericytes [
26,
159]. In addition, epicardial cells steer coronary vessel formation via paracrine signalling [
135]. In the adult heart, the epicardium is largely quiescent. However, it becomes re-activated upon injury and contributes to cardiac repair via cytokine signalling [
159]. Whether epicardial progenitors are able, after tissue injury, to form cardiomyocytes or fibroblasts at significant numbers remains controversial [
26,
84,
159].
Epicardial adipose tissue has several roles in cardiac physiology, such as energy supply and mechanical protection of the heart [
7,
77]. In addition, there is increasing evidence for more complex interactions between adipocytes and myocardial cells. Via secreted factors such as adiponectin or leptin, adipocytes can influence cardiomyocyte Ca
2+ cycling, metabolism, and redox state, and thereby alter contractility or contribute to arrhythmogenesis [
7,
136]. This refers to circulating, visceral tissue-derived adipocytokines, but also to the paracrine action of epicardial adipocytes. In obesity, accumulation and inflammation of epicardial fat contribute to adverse cardiac remodelling, fibrosis, and arrhythmia (as recently reviewed elsewhere [
77,
137]). Epicardial adipose tissue is most abundant in the atrio-ventricular and the interventricular sulcus, surrounding the coronary arteries [
77]. More recently, a population of cardiac adipocytes has been identified that derives from cardiac mesenchymal cells and is located in subendocardial myocardium [
81]. Whether these adipocytes have distinct functions remains to be explored. In diseases such as atrial fibrillation, fibro-fatty infiltration may contribute to electrical isolation of myocardial tissue regions, affecting action potential conduction and, potentially, facilitating the onset of arrhythmias [
39,
52,
110].
Heart function is tightly controlled by the cardiac autonomous nervous system. Sympathetic and parasympathetic
neurons located in ganglia of the sympathetic chain or in epicardial plexus form postganglionic nerve fibres reaching the myocardial tissue and modulating heart rate and contractility [
5,
216]. In addition to this extrinsic part, the heart contains a large number of neurons forming an intrinsic cardiac nervous system. High-resolution imaging with subsequent 3D modelling revealed the majority of neurons to be localized in a compact region posterior of both atria with a dense network of neuronal processes reaching distant cardiac tissue [
1]. Others suggested additional neuronal bodies within the ventricular myocardium [
102]. Cardiac nerves also include
Schwann cells and perimysial fibroblasts [
74]. Their roles are assumed to be similar to those in non-cardiac nerves, although development, regulation, and remodelling of intracardiac nerves during aging and/or disease [
174] are underinvestigated, beyond anatomical characterization.
Each cardiomyocyte is considered to be targeted by several neuronal processes via multiple neurotransmitter release sites [
216]. More recently it has been reported that the cholinergic transdifferentiation of sympathetic neurons that occurs after myocardial injury may harmonize action potential durations and thereby reduce the risk for ventricular arrhythmia [
209]. In addition, sympathetic neurons may, via direct cell–cell interactions (neuro-cardiac junctions), promote cardiomyocyte growth [
147], while cardiomyocytes, via neuronal growth factor release, may help to sustain local innervation [
44]. Thus, it is becoming increasingly clear that the intracardiac nervous system is subject of highly dynamic bidirectional neuro-muscular cross-talk.