Introduction

Although cardiomyocytes account for ≈25% to 35% of all cells in the heart,^1,2 there is a lack of consensus on the composition of the cardiac nonmyocyte cell population. In mammals, cardiomyocytes reside in close proximity to capillaries, and the calculated ratio of endothelial cells (ECs) to cardiomyocytes is 3:1.^3,4. This estimated number of ECs contradicts studies that have previously characterized heart cellular composition, where findings suggested that fibroblasts constitute the principal nonmyocyte cell type.^1,2,5,6 Often these analyses relied on the expression of singular proteins for identifying specific cell types; this approach may have resulted in exclusion of or loss of distinct cell populations. Because of the incongruence between these data sets and recent advances in our knowledge of heart resident immune cells,^7,8 we revisited the issue of cardiac cellular composition. Using newly available genetic tracers and enhanced flow cytometry techniques to analyze the relative abundance of different cell types in the human and mouse heart, we document that fibroblasts comprise a relatively minor cell population and that ECs are the most abundant cell type.

In This Issue, see p 367

Editorial, see p 368

Methods

Cardiac Single-Cell Preparation

Mouse hearts were isolated for single-cell preparation as previously described⁹ with atria and valves removed. Isolated mouse hearts were digested using 1 of 3 protocols (designated protocol 1, 2, and 3). For protocol 1, each mouse heart was divided into ≈20 pieces, placed in 10 mL of protocol 1 digestion buffer (2-mg/mL collagenase type II in 1× HBSS [Worthington Biochemical Corporation]) in gentleMacs C-tubes (Miltenyi Biotec), dissociated using Heart 1 program of a gentleMacs Dissociator, and incubated at 37°C for 40 minutes with gentle agitation. After incubation, tissue was further dissociated using Heart 2 program before placing on ice. For protocol 2, isolated hearts were finely minced using forceps to ≈2-mm pieces and placed in 3 mL of protocol 2 digestion buffer (2-mg/mL collagenase type IV [Worthington Biochemical Corporation] and 1.2 U/mL dispase II [Sigma-Aldrich or Thermofisher Scientific] in Dulbecco’s phosphate-buffered saline supplemented with 0.9 mmol/L CaCl₂). Tissue was incubated at 37°C for 15 minutes with gentle rocking. After incubation, tissue digestion buffer with tissue clusters was triturated by pipetting 12 times using a 10-mL serological pipette. Dishes were again incubated at 37°C and triturated twice more (45 minutes of total digestion time). The final trituration was conducted by pipetting 30× with a p1000 pipette. For protocol 3, isolated mouse hearts were finely minced using forceps, placed in 10-mL protocol 3 digestion buffer (125-U/mL collagenase type XI [Sigma-Aldrich], 60-U/mL hyaluronidase type I-s [Sigma-Aldrich], and 60-U/mL DNase 1 [Sigma-Aldrich] in Dulbecco’s phosphate-buffered saline supplemented with 0.9-mmol/L CaCl₂ and 20-mmol/L HEPES) incubated at 37°C for 1 hour with gentle agitation, triturated 20× using a 10-mL serological pipette, and placed on ice. All cell suspensions were filtered using a 40-μm cell strainer. Filtered suspensions were placed into 50-mL tubes with 40 mL of Dulbecco’s phosphate-buffered saline and centrifuged at 200g for 20 minutes with centrifuge brakes deactivated to remove small tissue debris. Cell pellets were resuspended in 250-μL 2% FBS/HBSS solution before staining with various antibodies and reagents for flow cytometry or fluorescence activated cell sorting.

Antibody, Nuclear, and Metabolically Active Cell Staining for Flow Cytometry

Antibody staining for specific antigens (Online Table I) was conducted in 100 μL of single-cell suspension (in 2% FBS/HBSS) using protocols 1, 2, or 3 after FC receptor blocking with CD16/CD32 antibody. After 1-hour antibody incubation at 4°C, calcein (calcein-AM or calcein-violet; Life Technologies) and Vybrant DyeCycle Orange (VDO; Life Technologies) were added to antibody/cell suspensions at final concentrations of 5 and 2.5 μmol/L, respectively. Samples with dyes added were incubated in a 37°C water bath for 10 minutes, before placing samples on ice. Samples were washed with 2% FBS/HBSS and resuspended in 2% FBS/HBSS with or without the viability dyes 4',6-diamidino-2-phenylindole (DAPI) or 7-aminoactinomycin D (7-AAD). All flow cytometry was conducted on LSR II Fortessa Flow Cytometers (BD Biosciences) or LSR II Flow Cytometers (BD Biosciences). For compensation of fluorescence spectral overlap, UltraComp eBeads (eBioscience, Inc.) were used following the manufacturer’s protocols. FCS 3.0 files generated by flow cytometry were initially processed using FlowJo Software (Tree Star, Ashland) for automated compensation. Dye-positive or dye-negative cell populations were gated and exported as new FCS 3.0 files and uploaded to Cytobank Premium for subsequent spanning-tree progression analysis of density-normalized event (SPADE) analyses. For analysis of total cardiac cells (Figure 2), metabolically active (calcein⁺), nucleated (VDO⁺), and viable (DAPI⁻) events were gated as shown (Figure 2A). Events were gated on viable (7-AAD⁻) single cells, before gating on CD31⁺CD45⁻ or CD45⁺ events for analysis of ECs and leukocytes, respectively (Figure 3A–3F). Events were gated on calcein⁺, viable (7-AAD⁻), CD31⁻CD45⁻ events for analysis of resident mesenchymal cells (RMC; Figure 3G–3J; Online Figures V and VI). All SPADE analyses were conducted with the target number of nodes set at 200 or 100 and down-sampled events target set at 100%.

Statistical Analyses

All statistical analyses were conducted using Prism 6 for Windows software (GraphPad Software, Inc., La Jolla, CA). Statistical variability expressed as mean±SD. Additional information is available in the Online Data Supplement.

Results

Immunohistochemical Analysis of Cardiac Cellular Composition

To determine the proportions of the major types of nonmyocyte cells in the adult murine heart, we conducted histological analysis of cardiac tissue from adult Cx₃cr1^GFP/+ mice,¹⁰ which express green fluorescent protein (GFP) in monocytes, macrophages, and dendritic cells including subsets of cardiac tissue macrophages.^11,12 In addition to GFP, we stained tissue sections with a hematopoietic lineage antibody cocktail (Lin1) targeting leukocyte antigens: CD45, Mrc1, B220, and major histocompatibility complex class II, to distinguish a broad cross section of cardiac leukocytes (Online Figure IA). Cardiomyocytes and ECs were identified by wheat germ agglutinin (WGA) and isolectin B4 (IB4), respectively. Using this combination of reagents, we were able to examine regional differences in cardiac cellular composition (cardiomyocytes, ECs, and leukocytes) in both ventricles and the interventricular septum by immunohistochemistry (Figure 1A and 1B; Online Figures IA and IIA–IIC). We found that 31.0±4.2% of nuclei were cardiomyocytes, 43.6±4.1% ECs, and 4.7±1.5% leukocytes. These proportions remained similar in all regions of the heart (Online Figure IIB). Unmarked cells accounted for 20.7±4.5% of nuclei. When considering only nonmyocytes (Online Figure IIC), our analysis indicates that 63.3±5.4% are ECs, 6.8±2.1% are leukocytes, and 29.9±5.9% were unmarked.

• Download figure
• Download PowerPoint

As many studies have suggested that fibroblasts constitute the majority of nonmyocytes in the heart, similar analyses were performed in cardiac tissues from mice containing either a nuclear localized H2B-eGFP expressed from the platelet-derived growth factor receptor alpha (PDGFRα) locus,¹³PDGFRα^GFP, (Figure 1C; Online Figure IB) or GFP driven by a Col1a1 promoter (3.2 kb) and HS 4,5 enhancer (1.7 kb), Col1a1-GFP.¹⁴ These 2 mouse lines drive expression of GFP in both epicardial and endocardial lineages of fibroblasts.^15–17 Using these 2 strains and focusing on the left ventricle, we independently stained for cardiomyocytes (ACTN2), ECs (isolectin B4 or dachshund homolog 1 [DACH1]), hematopoietic lineage cocktail (Lin2), which contains anti-CD5, anti-CD11b, anti-B220, anti-7-4, anti-Gr-1 (Ly6G/C), and anti–Ter-119 antibodies, and vascular smooth muscle cells (VSMCs)/pericytes¹⁸ (anti– platelet-derived growth factor receptor beta [PDGFRβ]; Figure 1C; Online Figure IB, and data not shown). 32.6±4.8% of nuclei corresponded to cardiomyocytes, 55.0±6.0% to isolectin B4⁺ ECs, 5.7±1.1% to VSMCs/pericytes, 8.5±1.5% to leukocytes, and 12.6±2.5%, and 12.7±3.6% to fibroblasts (PDGFRα^GFP/+ and Col1a1-GFP mice, respectively; Figure 1D). Because the endothelial values were considerably higher than previously reported,⁵ the expression of DACH1 was used as an independent marker for ECs (Figure 1C and 1D). DACH1 is expressed in the nuclei of coronary vessel ECs (Online Figure IB) but not in endocardial cells^19,20 and, therefore, was a reliable marker for quantification of coronary ECs. Figure 1C illustrates the frequency of ECs and fibroblasts relative to other nuclei. DACH1⁺ ECs constituted 51.8±3.9% of the nuclei. The sum of these data may be >100% because of inaccuracies in attributing nuclei to cells stained for plasma membrane components. Nonetheless, these independent data sets demonstrate that ECs are the major cell type in the adult murine ventricle. Surprisingly, these results also show that fibroblasts, marked by 2 unrelated mouse GFP reporter lines, contributed to <20% of total nonmyocyte nuclei.

To determine whether human tissue mirrors these findings, we analyzed healthy, adult human heart samples, using antibodies that recognize cardiomyocytes, ECs, and leukocytes (Figure 1E; Online Figures IC and IID). Unfortunately, none of the antibodies surveyed consistently delineated human RMC. Examination of human tissues demonstrated that 31.2±5.6% of nuclei correspond to cardiomyocytes (ACTN2), 53.8±6.4% to ECs (CD31), and 2.8±1.2% to leukocytes (CD45). DACH1 also identified human ECs, and 51.1±2.9% of the nuclei in the human heart were DACH1⁺ (Online Figure IID). Thus, estimations of endothelial abundance in the human heart reflect those observed in the murine heart.

Flow Cytometry Analysis of Nucleated Metabolically Active Cells

We next performed flow cytometry analysis of single-cell preparations as an alternative approach for evaluating cardiac cellular composition. When using conventional methods, many technical issues limit the accurate assessment of relative cell proportions by flow cytometry. These issues include the ability to clearly discriminate viable, nucleated cells from tissue debris and cell clusters. Normally, this issue is mitigated by using light-scattering properties of cells such as forward- and side-scatter. However, such approaches can skew data on the relative proportions of cell types. Therefore, for analysis of relative proportions of broad cardiac nonmyocyte cell types (ECs, leukocytes, and RMC), we developed a reliable approach for resolving nucleated cells from tissue debris using Vybrant DyeCycle Orange (VDO), calcein-AM (metabolically active cells) and DAPI (cells with compromised membranes; Online Figure IIIA). To determine the correct gating position for nucleated VDO⁺ cells within the DAPI⁻, calcein⁺ gated population (Online Figure IIIB), we isolated and analyzed cells from 4 regions (R1–4) with differential VDO labeling (Online Figure IIIB–IIID) by fluorescence activated cell sorting. Microscopy showed that R1 and R4 contained predominantly small cell fragments or clusters, whereas R2 and R3 contained mainly nucleated single cells. Analysis of forward scatter-height and forward scatter-area parameters confirmed that R1 and R4 comprise small and large elements, respectively; R2 and R3 were cells with a narrower size distribution (Online Figure IIIC). Further analysis of R1 events showed that almost all of these entities are CD31⁺ and CD105⁺ (data not shown), indicating that a potentially significant proportion of these entities may be non-nucleated endothelial microparticles or apoptotic bodies.²¹ These results point to reliable criteria for distinguishing dead cells and debris from viable, single cells. However, it should be noted that the VDO component could be replaced by conventional singlet gating with some contamination of R1 and R4 elements. In consideration of the enhanced accuracy conferred by the incorporation of the nuclear-staining dye, we have used it for flow cytometric survey of all nucleated nonmyocytes in the myocardium.

We used this approach to identify a tissue dissociation protocol that yields the greatest number of viable, nucleated cells. Three cardiac single-cell suspension protocols with distinct dissociation enzyme cocktails were compared: protocol 1,¹¹ protocol 2,²² and protocol 3.²³ The total number of nucleated cells isolated per milligram of tissue was 4241.8±517.4, 11223.3±1194.3, and 6728.9±491.7 for protocols 1, 2, and 3, respectively (Online Figure IVA). All 3 protocols demonstrated >95% calcein labeling in nucleated/viable cells (Online Figure IVB). Given the high yield of nucleated/viable cells, we used protocol 2 for subsequent analyses. It should be noted that during the process of protocol screening, we assessed Langendorff perfusion for cell isolation using several commonly referenced digestion cocktails. We found that these methods yielded fewer viable noncardiomyocytes and contained contamination by valvular interstitial cells (data not shown).

Two identification methods were used to determine whether mural cells (such as VSMCs and pericytes) could be recovered from protocol 2 isolates. To mark VSMCs, we used a lineage tagging system in a transgenic mouse that expresses Cre controlled by an αSMA promoter (SMACreERT2)²⁴ and a Cre reporter (ROSA26R^tdTomato).²⁵ An anti-NG2 antibody was used to identify pericytes.²⁶ We found that VSMCs and pericytes could be observed after isolation with protocol 2 (Online Figure V), suggesting that this population of cells will be included when determining relative cardiac cell proportions.

Cardiac Nonmyocyte Cellular Composition

To determine cardiac nonmyocyte cell composition, we conducted flow cytometry with antibodies for ECs (CD31 and CD102), leukocytes (CD45 and CD11b), and RMC (CD90 and Sca-1) in addition to nuclear and metabolic activity stains described above. Flow cytometric analyses with manual gating of nucleated, calcein⁺ cells indicated that CD31⁺CD45⁻ cells (ECs) constituted 62.1±3.9%, CD45⁺ cells (leukocytes) 9.6±1.3%, and CD31⁻CD45⁻ cells (RMC) 27.3±5.3% of all nonmyocytes (Figure 2A).

• Download figure
• Download PowerPoint

To better define cell populations represented by this antibody panel (Figure 2B), we conducted high-dimensional analysis of our flow cytometry data using the SPADE algorithm.^27,28 SPADE uses agglomerative clustering to identify groups of cells (referred to as nodes) that are phenotypically similar based on multiple surface or genetic markers (described in detail previously²⁸). Schematically, nodes are represented as colored circles that are interconnected within a dendrogram. Direct connection of 1 node to another signifies phenotypic similarity between the 2 nodes, and their size represents the number of cells within the node. Node color may indicate a range of statistical parameters. For example, all nodes within the dendrogram may be colored relative to the expression level of specific markers that define cell populations. This enables systematic annotation and categorization of individual nodes or groups of nodes as specific cell types. After annotation, nodes may be selected to derive statistical data, such as relative cell proportion.

After generation of SPADE dendrograms for cardiac cells using the aforementioned markers (Figure 2B), dendrogram branches corresponding to leukocytes, ECs, and RMC were manually annotated based on marker expression levels (Figure 2C and 2D). Branches with high CD45 expression, encompassing nodes and branches with high CD11b (myeloid cells) and CD90 (T cells), were classified as leukocytes (Figure 2D). Similarly, branches expressing CD31 and CD102 but not CD45 were identified as ECs. Remaining nodes, including nodes with high expression of commonly used fibroblast/mesenchymal markers, Sca-1 and CD90, were classified as RMC. Statistical analysis of SPADE branches identified 63.9±3.4% of cardiac cells as ECs, 9.4±1.6% as leukocytes, and 26.7±4.0% as RMC (Figure 2E). These values were consistent with histological analyses described above. Moreover, the SPADE data are consistent with the flow cytometry analyses conducted on nucleated cardiac cells manually gated for ECs (CD31⁺CD45⁻ cells), leukocytes (CD45⁺), and RMC (CD31⁻CD45^−;Figure 2A). Together, these findings confirm that ECs are the most prevalent cell type in the adult murine heart.

Endothelial and Leukocyte Diversity

To determine the diversity of cardiac ECs and leukocytes, we also conducted SPADE analysis after gating on viable, nucleated CD31⁺CD45− ECs or CD45⁺ leukocytes. ECs were clustered based on surface expression of CD102, CD105, podoplanin, and CD90. We found that 94.3±0.7% corresponded to vascular ECs (CD102⁺CD105⁺ nodes) and 5.2±0.7% corresponded to lymphatic ECs (podoplanin⁺ nodes; Figure 3A–3C). To characterize leukocyte subsets, we clustered CD45⁺ cells based on surface expression of CD11b, CD64, B220, IgM, CD3ε, and CD90 (Figure 3D and 3E). Myeloid cells (CD11b⁺ nodes), B cells (CD11b⁻B220⁺ nodes), T cells (CD11b⁻CD3ε⁺ nodes), and nonmyeloid/lymphoid (CD11b⁻B220⁻CD3ε⁻ nodes) cells were clearly identified within dendrograms. Consistent with cell identity, CD64, IgM, and CD90 were highly expressed in nodes corresponding to myeloid cell, B-cell, and T-cell clusters, respectively. Statistical analysis showed that 81.4±1.4%, 8.9±0.6%, 3.1±0.4%, and 6.6±0.6% of leukocytes corresponded to myeloid cells (CD11b⁺), B cells (B220⁺), T cells (CD3ε⁺), and nonmyeloid/lymphoid cells (CD11b⁻B220⁻CD3ε⁻), respectively (Figure 3F).

• Download figure
• Download PowerPoint

RMC Composition

Although ECs and leukocytes can be identified using a variety of antibodies that recognize surface proteins, reagents for classifying fibroblasts and mural cells (VSMC and pericytes) are limited. Consequently, we attempted to identify uniform and specific antibodies suitable for flow cytometric identification of these 2 populations. In the murine context, an anti-PDGFRβ antibody would be a candidate for mural cells. Flow cytometry demonstrated that the PDGFRβ monoclonal antibody, APB5, did not label primary cells to a satisfactory level (data not shown for 6 independent sources) although the same monoclonal antibody can be used for immunohistochemistry.

Unlike the mural cell population, we were able to clearly identify the proportion of RMC that were cardiac fibroblasts by using genetically tagged Col1a1-GFP, PDGFRα^GFP/+, and Col1a1-GFP;PDGFRα^GFP/+ transgenic mice that express GFP in resident fibroblasts as described above. By gating on the CD45⁻CD31⁻ population, we observed that more than half of the RMC were GFP⁺ (Figure 3G and 3H). To determine whether GFP cells from Col1a1-GFP and PDGFRα^GFP/+ mice represent similar cardiac cell populations, we conducted SPADE analysis by clustering on CD31, CD102, CD45, CD11b, CD90, and Sca-1. A high degree of overlap occurred when comparing dendrograms for Col1a1-GFP and PDGFRα^GFP labeled cells (Online Figure VI).

Although cells from the 2 GFP fibroblast mouse lines were convenient for categorizing and surveying the fibroblast population, the use of these transgenic mice as a universal tool to analyze fibroblasts is not always feasible. Because antibodies validated for flow cytometry were readily available for 3 commonly used fibroblast surface antigens, Sca-1, CD90, and PDGFRα, we surveyed these by flow cytometry. PDGFRα antibodies were assessed on NIH3T3 fibroblasts, mouse embryonic fibroblasts (MEFs), including PDGFRα null MEFs,²⁹ and primary cardiac single-cell isolates from wild-type and PDGFRα^GFP/+ mice. None of the PDGFRα antibodies examined clearly and uniquely distinguished a positive cell population (Online Table I and data not shown) although PDGFRα expression was observed in cardiac fibroblasts by immunohistochemistry (Online Figure VII). Interestingly, after unsupervised SPADE clustering of RMC from Col1a1-GFP mice, we found that most GFP⁺ nodes were also MEFSK4⁺ (Figure 3I). Notably, similar correlation of gene expression within these nodes was not observed for CD90 or Sca-1. The MEFSK4 antibody has been used to identify and remove MEFs during culture of induced pluripotent stem cells and embryonic stem cells. This antibody identifies MEFs from C57BL/6, CF1, CF6, and DR4 mouse strains and an epitope on NIH3T3 cells.³⁰ The close correlation of GFP expression and MEFSK4 detection was further underscored when comparing GFP expression with MEFSK4, CD90, or Sca-1 profiles (Figure 3J). Indeed, almost all GFP⁺ RMC from Col1a1-GFP mice, PDGFRα^GFP/+ mice, or Col1a1-GFP;PDGFRα^GFP/+ double transgenic mice were MEFSK4⁺, and most MEFSK4⁺ RMC were GFP⁺ (Online Figure VIII). MEFSK4 antigen expression was independent of activation status, as surface staining was not decreased after transforming growth factor-β1 stimulation of MEFs or after myocardial infarction (data not shown). It should be noted, however, that the antibody also stains a subpopulation of CD11b⁺ leukocytes (data not shown). Therefore, this population must be excluded from analysis before considering the fibroblast lineage by flow cytometry. We have not been successful in tissue staining with the MEFSK4 antibody. To further investigate the MEFSK4⁺ RMC population by flow cytometry, we costained for the pericyte marker, NG2. NG2⁺ cells express the MEFSK4 epitope at lower levels than GFP⁺ cells, and NG2⁺ cells account for a majority of the MEFSK4⁺ GFP⁻ RMC population (Online Figure IX). To further define MEFSK4⁺ cells, we isolated CD31⁻,CD45⁻, and MEFSK4⁺ cells by fluorescence activated cell sorting and found that fibroblast gene expression is enriched in this cell population when compared to whole ventricle (Online Figure X). These findings indicate that Sca-1 and CD90 are suboptimal markers of resident cardiac fibroblasts and that MEFSK4 may be a suitable surrogate marker for cardiac fibroblasts.

Cardiac Fibroblast Diversity

To further examine the spectrum of fibroblasts labeled by PDGFRα^GFP, Col1a1-GFP, and MEFSK4, we performed lineage tracing by intercrossing Tcf21^iCre/+; ROSA26R^tdTomato mice with the fibroblast GFP-expressing mouse lines. In the Tcf21^iCre mouse model, resident fibroblasts and their descendants are indelibly tagged with tdTomato fluorescence after Cre induction.^15,31 We induced Cre-mediated recombination at either embryonic day 16.5 (Tcf21^iCre/+;ROSA26R^tdTomato;Col1a1-GFP) or in the adult (Tcf21^iCre/+; ROSA26R^tdTomato;PDGFRα^GFP/+) and conducted flow cytometry 2 to 6 months after induction to determine the proportion of tdTomato⁺ cells (cardiac fibroblasts) that were GFP⁺. Gating on the tdTomato⁺ cells revealed that ≈95% were positive for GFP and MEFSK4 (Figure 4, data not shown), demonstrating that cardiac GFP⁺ cells from PDGFRα^GFP/+ and Col1a1-GFP mice comprise Tcf21 lineage fibroblasts. Taken together, these data suggest that the population of cells identified as fibroblasts is relatively uniform as determined by 4 independent markers (PDGFRα^GFP, Col1a1-GFP, MEFSK4, and Tcf21 lineage). Furthermore, our analysis indicates that this cell population constitutes 15% of the nonmyocyte cell pool, equating to ≈11% of the total cells of the heart when assuming ≈30% of the cells are cardiomyocytes.

• Download figure
• Download PowerPoint

Discussion

A comprehensive understanding of cardiac cellular composition will guide development of therapeutics that promote heart repair and regeneration. To date, attempts to survey the cellular composition of the heart, even in model organisms such as the mouse and rat, have been hindered by difficulties in cell identification and isolation. Here, using recently developed techniques and genetic tools, we demonstrate that ECs outnumber other cell types in the adult mouse heart ventricles (Figure 5) and that the cardiac fibroblast population is much smaller than previously reported. Nonetheless, classic fibroblast markers such as CD90 and Sca-1 under-represent the resident fibroblast population.

• Download figure
• Download PowerPoint

Although the accuracy and reliability of immunohistochemical data depend on the specificity and sensitivity of the antibodies used, it can be a powerful approach for evaluating cell populations within a tissue. Here, 2 histological approaches in separate colonies of mice generated highly analogous results that were further corroborated by the analysis of human tissue sections. Using histological analyses, however, we were unable to evaluate the proportion of myocytes relative to nonmyocytes with a high level of precision. This is because of the variability and dynamism of the number of nuclei per cardiomyocyte in addition to the differences between rodents and humans—most murine cardiomyocytes are binucleated,^32,33 whereas most human cardiomyocytes are mononucleated.^2,34 Therefore, estimations of myocyte to nonmyocyte ratios will differ depending on the species and age.

There are many explanations as to why our findings on the number and proportion of ECs in the heart differ from those of previous studies. First, access to new and enhanced reagents enabled more accurate identification of cell populations than was previously possible. Second, the use of an isolation protocol that yielded the greatest number of viable cells at the expense of cardiomyocytes may explain why flow cytometry results strongly correlated with results from immunohistochemical analyses but produced different outcomes when compared with previous studies.^2,5,35 Many published protocols use digestion conditions that favor the isolation of cardiomyocytes with noncardiomyocytes being the byproducts of the isolation. Third, establishment of the criteria of cell nucleation and vitality to determine cell frequency enabled us to perform a more accurate analysis. This approach eliminates the dependency on using light-scattering (forward- and side-scatter) properties of cells for identifying viable cells, a practice that could lead to over- or under-representation of cell types and inclusion of tissue debris as cellular events. Fourth, in classifying cell populations, we used an unsupervised clustering algorithm (SPADE) after objective cell gating. SPADE and similar clustering algorithms enable superimposition of expression data for multiple cellular markers to visualize and quantify nonmyocyte cardiac cell populations. The identity of ECs was confirmed using multiple endothelial markers including CD31, CD102, and CD105, in addition to exclusion of leukocyte markers, CD45 and CD11b.

Although the recovery of cell populations and flow cytometry data presented here correlates with the cellular distribution documented by immunohistochemistry, the possibility remains that not all cells have been included in the current assessment. These may include cells that are more sensitive to the digestion protocol or are not recovered because of insufficient digestion. Notably, we have not tested the efficacy of the nuclear and metabolic activity (VDO and calcein) staining approach to distinguish between activated, proliferating, and quiescent cell populations.

The discrepancy in leukocyte proportions determined by flow cytometry and immunohistochemistry may be attributed to inefficiencies in antigen detection in fixed tissues. Indeed, this is likely for leukocyte staining using anti-CD45 antibodies, which do not reveal stellate-shaped myeloid cells in fixed human or mouse tissue. The use of antibody cocktails for staining mouse cardiac tissue may overcome this issue; nevertheless, immunohistochemical analyses of leukocytes may still fail to detect all cells. By flow cytometry analysis—focusing on nucleated metabolically active cells and SPADE—we have estimated that leukocytes comprise ≈7% to 10% of all nonmyocytes. The vast majority of these cells were identified as myeloid cells and, in particular, macrophages. Several lymphocytes were also detected although at a much lower prevalence.

Histological and flow cytometric analysis also suggests that the relative number of fibroblasts in the heart may have been overestimated in the past. The lack of a clear definition for fibroblasts has been at the root of many difficulties in quantifying and tracking these cells in vivo. In some cases, cardiac fibroblasts have been quantified by excluding cells that lack structural elements corresponding to ECs, VSMCs, or cardiomyocytes.¹ According to this definition, pericytes and leukocytes would also be classified as fibroblasts. Another confounding factor is that markers such as DDR2, CD90, Sca-1, and vimentin are not unique to fibroblasts, and not all fibroblasts express these proteins. Although resident cardiac fibroblasts derive from 2 embryonic sources, the consensus is that PDGFRα is expressed in both populations. The use of CD90 for identification of fibroblasts, on the contrary, remains a point of debate.^17,35 Previously, we have shown that cardiac Tcf21 lineage cells, and GFP⁺ cells from PDGFRα^GFP and Col1a1-GFP mouse hearts have similar levels of fibroblast gene expression including Col3a1, Col6a1, Dcn, and MMP2 with lack of expression of smooth muscle, endothelial, and cardiomyocyte genes.¹⁵ As flow cytometric results demonstrate that the MEFSK4 antibody detects Tcf21 lineage cells and GFP⁺ fibroblasts in mouse hearts, when excluding leukocytes, this antibody can be used to identify cardiac fibroblasts by flow cytometry when use of the genetic systems is not feasible. Our results also suggest that CD90 and Sca-1 only capture a fraction of the resident fibroblasts. This analysis represents the most comprehensive estimation of cardiac fibroblasts to date.

Recent reports have proposed that pericytes contribute significantly to the cardiovascular remodeling process.^36,37 In the uninjured heart, we show that these cells account for ≈5% of noncardiomyocytes and that they also express MEFSK4, albeit at lower levels than fibroblasts. Further investigation will be required to determine MEFSK4 expression in the activated pericyte population. The establishment of these cellular proportions at baseline will pave the way for future work examining how these populations vary in the postnatal, aged, and injured heart.

Although the role of the EC as a dynamic regulator of tissue responses is increasingly recognized, their abundance and roles in the heart are commonly underappreciated. The proximity of ECs to the coronary circulation and cardiomyocytes provides an access point for therapeutic manipulation. Therefore, a more comprehensive understanding of these potential cellular interactions will be required. Taken together, our findings redefine the cellular composition of the adult murine and human heart and point to the cardiac EC as a potentially important protagonist in cardiac homeostasis, disease, and aging.

Footnotes