To compare Endo- and SV-derived ECs during development and in adult hearts, we combined scRNAseq with lineage-specific fluorescent labeling (Figure 1A and B). A tamoxifen-inducible Bmx^CreER (Ehling et al., 2013) mouse was crossed with the Rosa^tdTomato Cre reporter, which specifically labels a high percentage of the Endo (94.44% of Endo cells labeled at e12.5), but does not mark the SV (3.61% of SV cells labeled at e12.5) (D’Amato et al., 2021). Labeling was induced before e11.5, when coronary development begins, so Endo-derived ECs expressed tdTomato while SV-derived ECs did not (Materials and methods). Cells from e12, e17.5, and adult hearts were sorted using fluorescence-activated cell sorting (FACS) and processed using the 10× Genomics platform (Figure 1A–B and Figure 1—figure supplement 1A-E). These strategies captured the expected EC subtypes at each stage (including coronary, valve, Endo, and SV) (Figure 1—figure supplement 2A-D) and contained a large number of cells that passed standard quality controls (Materials and methods). ScRNAseq analyses are most accurate when the specific cell populations of interest are extracted and re-analyzed without the influence of other cell types in the dataset, including cycling cells (Luecken and Theis, 2019). Thus, we isolated non-cycling coronary ECs (for e12—Pecam1+, Cldn5+, Npr3-, Top2a-, Mki67, Tbx20-, Cldn11-, Bmp4-, Vwf-; for e17.5—Pecam1+, Npr3-, Tbx20-, Pdgfra-, Top2a-, Bmp4-, Mki67-) (Figure 1—figure supplement 2E and F) and performed direct comparisons of cell states between Endo- and SV-enriched coronary ECs. The remaining cells in the dataset will be reported by D’Amato et al., which addresses experimental questions outside the scope of this study.

We first used unbiased clustering to identify coronary EC subtypes within the e12, e17.5, and adult datasets. Clustering resolution was determined individually for each dataset as the highest resolution at which every cluster expressed at least one unique marker gene. E12 coronary ECs separated into three clusters—capillary plexus 1 (Cap1), capillary plexus 2 (Cap2), and pre-artery (Figure 1C). Markers used to identify these populations matched previous reports and are shown in Figure 1—figure supplement 3A. The absence of venous ECs and the higher numbers of SV-enriched cells present at this stage are also consistent with previous studies (Red-Horse et al., 2010; Su et al., 2018). Separating plots by sample revealed that Cap2 was exclusively from the SV-enriched sample (Figure 1D and E). Consistent with this was its increased expression of Aplnr (Apj) (Figure 1—figure supplement 3A), which we previously demonstrated to be enriched in SV-derived vessels (Sharma et al., 2017). All but one of the Endo-enriched capillary cells were in Cap1, but Cap1 also contained cells from the SV-enriched sample (Figure 1D and E). Although recombination rates in the Endo were very high (Materials and methods), we cannot exclude the possibility that a small number of Endo-derived ECs are tdTomato-negative due to some un-recombined Endo ECs. These data show that shortly after coronary development is initiated, lineage is correlated with transcriptionally distinct capillary populations within the immature capillary plexus.

To test whether this phenomenon persists into late development, we similarly analyzed coronary ECs at e17.5. A larger number of coronary ECs were captured due to the increase in cardiac vasculature by this stage. EC clusters in this sample included two artery (Art1 and Art2), one vein, and two capillary (Cap1 and Cap2) clusters (Figure 1F and Figure 1—figure supplement 3B). If Cap1 and Cap2 continued to reflect different lineages, we would expect at least one cluster to contain only Endo- or SV-enriched ECs. There was skewed contribution with a higher percentage of Endo-enriched cells in Cap1 and a higher percentage of SV-enriched cells in Cap2 (Figure 1G and H), but no cluster was lineage exclusive, suggesting that additional factors were driving transcriptional heterogeneity. Veins were much more represented in the SV-enriched sample (Figure 1H), which is expected since most veins reside on the surface of the heart and SV angiogenesis progresses outside-in while Endo angiogenesis is in the opposite direction. We also considered the e17.5 dataset with cycling cells included, but with cell cycle effects regressed out, in order to evaluate whether there are differences in proliferation between Endo- and SV-enriched cells, and to rule this out as a cause for the difference in cluster distribution between the two lineages (Figure 1—figure supplement 4A and B). This analysis showed that cycling capillary cells also segregate into the Cap1 and Cap2 clusters (Figure 1—figure supplement 4C), and that their distribution into these clusters is biased by lineage, similar to non-cycling cells (Figure 1—figure supplement 4D). Additionally, there is no difference in proliferation between Cap1 and Cap2 (Figure 1—figure supplement 4E).

We next performed the same analyses on adult coronary ECs. Similar to e17.5, clustering revealed one vein and two artery clusters. It additionally revealed three capillary clusters, Cap1, Cap2 and Cap3 (Figure 1I), which were distinguishable by expression of unique gene markers (Figure 1—figure supplement 3C and Figure 1—figure supplement 5A). While Cap1 contained the majority of capillary cells, Cap2 was distinguished by expression of pro-angiogenic genes including Apln and Adm, and Cap3 was distinguished by expression of interferon-induced genes such as Ifit3 (Figure 1—figure supplement 5A). Both of these clusters have been reported previously in adult hearts (Kalucka et al., 2020). However, unlike the clusters at earlier stages, there was a similar distribution of cells into each of these clusters in both the Endo- and SV-enriched samples (Figure 1j–k), and there were no appreciable gene expression differences between the samples. This observation is consistent with another lineage-specific adult scRNAseq dataset we produced with Smart-seq2 (unpublished results). Thus, we concluded that there is no lineage-based heterogeneity in adult coronary ECs.

We next investigated the genes driving coronary ECs into two capillary cell states at the different stages of development. One hypothesis was that cells retained gene expression patterns from their progenitors. To test this, a list of genes defining the progenitor states (the Endo and SV) was compiled by directly comparing gene expression in the Endo and the SV at e12 (Figure 1—figure supplement 2A) and using differentially expressed genes (DEGs) passing significance thresholds described in the Materials and methods (Table 1). The expression of these genes was then assessed in capillary clusters. At e12, Cap1 cells expressed higher levels of Endo-specific genes, while Cap2 cells expressed higher levels of SV-specific genes (Figure 2A and Figure 2—figure supplement 1A-C). Indeed, 40% of Cap1 genes and 3% of Cap2 genes overlapped with the Endo, while 47% of Cap2 genes and 1% of Cap1 genes overlapped with the SV (Figure 2B). We concluded that the transcriptional identities of Cap1 and Cap2 cells derive at least in part by gene expression patterns retained from the SV or Endo.

This pattern was not observed at e17.5. There was no clear pattern between e17.5 Cap1 and Cap2 in the expression of SV- and Endo-specific genes (Figure 2A and Figure 2—figure supplement 1D), and there was little to no overlap between Cap1 and Cap2 differential genes and SV or Endo genes (Figure 2C). Although it may appear on the heatmap in Figure 2A that there is a lineage-based distinction in Cap1 for a small subset of the genes at e17.5, this is not the case. As we will demonstrate later in Figure 3—figure supplement 2, the data indicated that these results were primarily due to Cap1 containing septum ECs, which are mostly derived from the Endo, and the septum imparting a location-specific effect on transcription. Furthermore, the minimal overlap between the e12 and e17.5 Cap1- and Cap2-defining genes is consistent with e17.5 coronary ECs not retaining the progenitor-type genes enriched in e12 clusters (Figure 2D).

Calculating Pearson correlations using Endo and SV genes revealed that e12 coronary ECs were similar to their progenitor sources while e17.5 coronary ECs were much less so (Figure 2E). We also calculated Pearson correlations as a function of lineage at e12. As expected, tdTomato-positive cells were highly similar to the Endo while tdTomato-negative cells in Cap2 were more similar to the SV than to the Endo (Figure 2F). Interestingly, tdTomato-negative cells in Cap1 were more similar to the Endo than to the SV (Figure 2F). This could result from a number of reasons that we cannot currently distinguish including: (1) SV-derived cells migrating close to the Endo take on Endo-type gene expression or (2) there is a rare Endo population that does not express Bmx. Point 2 is supported over point 1 due to the observation that Cap2-like cells are present in a previous scRNAseq dataset of SV-derived ECs (Su et al., 2018), while Cap1 cells are not (Figure 2—figure supplement 2A-E). In total, these data indicate that e17.5 capillary heterogeneity is not a remnant of the heterogeneity present at e12, and lineage-related differences are not apparent in adults.

Further analyses provided additional evidence that the differences between e17.5 Cap1 and Cap2 are not primarily due to lineage. With the prediction that significant lineage-related heterogeneity would be accompanied by substantial differences in gene expression, we compared the DEGs between e17.5 Cap1 and Cap2 to the DEGs between Endo- and SV-enriched capillary cells. There were 202 DEGs between e17.5 Cap1 and Cap2, but only 24 DEGs between all Endo-enriched and SV-enriched capillaries (Figure 2H). Inspecting DEG identities provided further support that lineage is not retained. Eighteen of the DEGs between all Endo- and SV-enriched cells were also DEGs between Cap1 and Cap2 (Table 2). If the differential patterns of these genes were due to a lineage effect, we would expect a greater log-fold change in the all Endo- versus SV-enriched comparison than in the Cap1 versus Cap2 comparison. However, 16 of the 18 genes have a greater log-fold change in the Cap1 versus Cap2 comparison (Table 2). These data indicate that differential expression between the Endo- and SV-enriched capillaries mostly stems from the differential contribution of the Endo and SV lineages to Cap1 and Cap2.

Since there was not strong evidence that lineage was a significant factor, we next considered whether differential localization in the heart might underlie e17.5 heterogeneity. During development, different regions of the heart show varying levels of hypoxia and signaling factors, for example, at some stages, the septum is more hypoxic and expresses higher levels of Vegfa (Miquerol et al., 2000; Sharma et al., 2017). Localization-driven heterogeneity would also explain the bias in cluster distribution between the Endo and SV lineages (Figure 1H) because they contribute ECs to complementary regions of the heart. This is most dramatic in the septum where almost all ECs are derived from the Endo (Wu et al., 2012; Chen et al., 2014; Zhang et al., 2016). The top DEGs between e17.5 Cap1 and Cap2 included hypoxia-induced genes (Mif, Adm, Igfbp3, Kcne3) (Tazuke et al., 1998; Lee et al., 1999; Keleg et al., 2007; Simons et al., 2011; Heng et al., 2019) and tip-cell markers (Apln, Plaur, Lamb1, Dll4) (Hellström et al., 2007; del Toro et al., 2010) in Cap1, and flow-induced genes (Klf2, Klf4, Thbd, Lims) (Dekker et al., 2002; Hamik et al., 2007; Kumar et al., 2014; Wang and Zhang, 2020) in Cap2 (Figure 3A and B, Figure 3—figure supplement 1A), suggesting pathways that could be consistent with localization-driven heterogeneity.

We next used Car4 to localize Cap2 within the tissue because it was specifically expressed in this cluster and there was an antibody available for immunostaining (Figure 3A and C). Erg-positive ECs expressing Car4 were located mainly in the right and left ventricle-free walls and dorsal side while ECs in the septum and ventral wall were mostly Car4-negative (Figure 3D–F and Figure 3—figure supplement 1B). This indicated that Cap1 localizes primarily to the septum and ventral wall and Cap2 to the remaining walls of the ventricle. The Kcne3 pattern provided further support for this localization because it was specific to Cap1, and in situ hybridization revealed specific septal and ventral expression (Figure 3G). Car4 analysis revealed a further segregation of Cap1 into septum and non-septum regions on the UMAP. Specifically, one side of the Cap1 cluster was almost completely devoid of Car4 and comprised almost all Endo-enriched cells (Figure 3C), which are two features specific to the septum as shown in Figure 3D–F and Figure 3—figure supplement 1B. Additionally, these cells have lower expression of Aplnr compared to the non-septal cells of Cap1 (Figure 1—figure supplement 3B), which is also characteristic of the septum (unpublished observations). Although unbiased clustering did not distinguish this group of cells as a separate cluster, this could be because we used a combination of gene expression, protein staining, and lineage information (Figure 3B–G), which the clustering algorithms do not take into account. Interestingly, this proposed septum region is where gene expression indicates both decreased blood flow and local hypoxia (Figure 3B), the former of which would cause the latter. Thus, the coronary vasculature in the septum (and where the rest of Cap1 localizes) may not receive full blood flow at this late stage in development.

With the knowledge that half of Cap1 was almost completely represented by Endo-derived cells from the septum, we revisited the issue of lineage-based heterogeneity by manually searching for genes enriched in the proposed septum and analyzing whether they were correlated with lineage. A few of the Endo-specific genes from the heatmap in Figure 2A were expressed at a higher level in the septum while a few SV-specific genes were expressed at a lower level (Figure 3—figure supplement 2A-D). If this is due to retention of lineage information, we would expect to see the Endo-specific genes expressed in higher percentages of Endo-enriched cells compared to SV-enriched cells both inside and outside the septum. However, there was generally a decrease in the expression of these genes in both Endo- and SV-enriched cells outside of the septum, supporting the location, but not the lineage, hypothesis. Only two genes somewhat followed a pattern indicating lineage retention—Gucy1b3 and Hand2— but these differences did not pass our pre-set significance thresholds (Materials and methods) and were expressed in a low percentage of cells (<20% outside the septum) (Figure 3—figure supplement 2B).

We performed additional analysis comparing the number of DEGs between Endo- and SV-enriched cells within different capillary subgroups defined by transcriptional states (i.e. clustering) and in different locations, specifically, the proposed septal and non-septal cells of Cap1 and Cap2 (Figure 2H). If lineage was a major contributor to e17.5 heterogeneity, we would expect to see a substantial number of DEGs between Endo- and SV-enriched cells within a specific location. Instead, once the effect of location was removed by only comparing Endo- and SV-enriched cells either inside or outside of the septum, there were only six DEGs between the lineages (Figure 2H). Further supporting the impact of location in transcription, the second largest number of DEGs was between the septum and non-septum cells of Cap1 (Figure 2H). Thus, although we cannot exclude that a few genes retain lineage information in a small number of cells, the overwhelming evidence of this analysis supports that location has a greater effect on cell state at e17.5, with little to no contribution by lineage.

To probe whether the septal portion of Cap1 (see Figure 3C, dotted line) segregates based on cell-autonomous, lineage-specific features of Endo-derived ECs or regional environments, we took advantage of the fact that some Endo-lineage-labeled cells migrate into SV-biased territories during development (Chen et al., 2014). Cell-autonomous lineage differences would be supported if these cells remained Car4-negative in the ventricular walls. However, the opposite was true. Endo-lineage cells outside the septum were more likely to express Car4 than those in the septum (on average, 23% outside septum versus 4% inside septum) (Figure 3F, Figure 3—figure supplement 1B). Endo-derived ECs in SV-biased territories such as the dorsal side of the heart start to express Car4 (Figure 3D, arrows), and they can exist in the Cap2 state (Figure 3C, arrow). In addition, the rates of Car4 positivity in both Endo- and SV-enriched ECs correlate similarly with location in the heart (Figure 3E). Thus, the combination of lineage labeling, scRNAseq, and histology allowed us to ascertain that Endo-lineage ECs are biased toward a separate cell state based on their preferential location in hypoxic regions. However, due to the strong association between EC lineage and location, we cannot rule out that some minor degree of lineage-based heterogeneity exists at e17.5.

Since regional hypoxia would be incompatible with adult heart function, we predicted that the absence of strong heterogeneity in adult capillary ECs could be explained by a resolution of regional environmental differences after development. To test this, we examined Car4 in the adult dataset, and found that it is broadly expressed in all adult capillary populations (Figure 3H). Additionally, immunostaining confirmed that septum ECs had become positive for Car4 in adults (Figure 3I, Table 3). Altogether, our data support the following model—that over the course of development, transcriptional heterogeneity in coronary ECs is first influenced by lineage, then by location, until both lineage- and location-based heterogeneity disappear in the static adult heart (Figure 3J).

Although Endo- and SV-derived capillary ECs were transcriptionally very similar in normal adult hearts, they could behave differently when challenged with hypoxia or other injury. To test this, we performed an IR injury by temporarily occluding the left anterior descending (LAD) coronary artery in adult Bmx^CreER;Rosa^tdTomato mouse hearts in which the Endo was labeled prior to coronary development (Figure 4a and Figure 1—figure supplement 1b). Proliferation in Endo- and SV-enriched ECs was quantified using EdU incorporation assays on day 5 post-injury. This time point was chosen based on prior data showing regrowth of the majority of coronary vessels 5 days after IR injury (Merz et al., 2019). Observing overall EdU incorporation revealed the site of injury, which was prominent in the mid-myocardial region between the inner and outer wall (Figure 4B). Proliferation was assessed in areas just below the ligation and apex of the heart, as this is the expected distribution of ischemia following LAD ligation (Merz et al., 2019). Regions of interest (ROIs) were chosen in myocardial areas containing a mix of Endo- and SV-derived ECs to ensure adequate representation from both lineages (Figure 4C and D). This analysis found no difference between the lineages (Figure 4E). We next performed EdU quantification in the inner and outer walls of the myocardium where most ECs derive from either the Endo or SV, respectively (Wu et al., 2012; Tian et al., 2014; Zhang et al., 2016; Sharma et al., 2017; Figure 4B and Figure 1—figure supplement 1B). In contrast to the lineage comparison, the outer wall showed a consistent, though non-significant, increase in EC proliferation over the inner wall, both within and adjacent to the injury site and regardless of lineage, indicating that this injury and proliferation assay can reveal differences (Figure 4F). These data support the notion that the local environment, rather than lineage, regulates capillary responses to injury in adult hearts, at least with respect to EC proliferation in the days after injury.

ECs exit the cell cycle during their differentiation into mature coronary arteries (Fang et al., 2017; Su et al., 2018). As a result, the proliferation of artery ECs is rare in the normal adult heart vasculature. Because IR injury induced the proliferation of capillary ECs, we investigated whether IR injury also induced the proliferation of artery ECs (Figure 4G). Approximately 69% of large arteries (n = 32) identified in the injured regions that we analyzed (as indicated in Figure 4B) contained at least one EdU-positive EC, and there was a significantly higher rate of EdU-positive ECs in the arteries of injured compared to uninjured hearts (Figure 4H). This observation could be due either to proliferating capillary cells which transitioned into artery cells but retained EdU or to artery ECs which began proliferating in response to the injury. Further studies will be necessary to determine whether one or both of these processes are occurring.

We next sought to investigate whether comparing mouse and human scRNAseq datasets could provide insights into human development. ScRNAseq was performed using Smart-seq2 on PECAM1-positive ECs sorted from human fetal hearts at 11, 14, and 22 weeks of gestation. In addition, a capillary-specific marker, CD36, allowed for enrichment of PECAM1+ CD36- arterial ECs (Figure 5A and Figure 5—figure supplement 1A and B; Cui et al., 2019). After initial filtering, 2339 high-quality, high-coverage single EC transcriptomes were obtained, of which 713 were arterial. The data included 12 clusters of the expected cell types—artery, capillary, vein, cycling, Endo, valve EC—identified by known markers (Figure 5B and Figure 5—figure supplement 1C). As with the mouse data, the analysis was restricted to non-cycling arteries, capillaries, and veins in order to specifically compare cell states and trajectories in coronary vessels. Similar to mouse, the data contained one vein and two capillary clusters, but in contrast to mouse, there was an additional arterial cluster for a total of three (Figure 5C). There was approximately equal representation of the clusters in the PECAM1-positive fraction at each gestational age, consistent with the data being from later stages (mouse equivalent of e15.5–18.5)(Krishnan et al., 2014) when the developing heart is growing in size rather than producing new coronary cell subtypes (Figure 5D and Figure 5—figure supplement 2A). Consequently, we pooled data from the three time points for all further analyses. To begin comparing human and mouse, the Seurat Label Transfer workflow (Stuart et al., 2019) was utilized to reference map each human cell to its closest mouse cluster and vice versa (Figure 5E). This showed close concordance between the two species (Figure 5F and G). With respect to the two capillary clusters that were extensively studied above, the majority of human Cap1 cells mapped to mouse Cap1, while the majority of human Cap2 cells mapped to mouse Cap2 (Figure 5F). When assigning mouse cells to human clusters, mouse Cap2 almost completely mapped to human Cap2, while mouse Cap1 mapped substantially to human Cap1, Cap2, and Art3 (Figure 5G). Analyzing specific gene expression revealed several enriched genes shared between corresponding mouse and human capillary clusters (including KIT, ODC1, CD300LG, RAMP3), and showed that human Cap1, like its mouse correlate, displayed patterns indicative of experiencing low blood flow conditions and potentially increased hypoxia, that is, lower LIMS2, THBD, KLF2, KLF4 and higher ADM, IGFBP3, KCNE3, LAMB1 (Figure 5H and Figure 5—figure supplement 1D). Notably, CA4 (the human homolog of mouse Car4) was not differentially expressed between human Cap1 and Cap2.

Knowing that Cap1 and Cap2 segregate spatially in the mouse heart (Figure 3D–F), we examined whether human Cap1 and Cap2 also represent cells in different locations. We performed in situ hybridization for TINAGL1, a gene which is enriched in human Cap2 (Figure 5I). This revealed a statistically significant difference in the amount of TINAGL1 RNA detection between the septum and the heart walls in both an 18-week and a 20-week gestational heart, with the septum having dramatically lower expression (Figure 5J–K). This difference was especially pronounced between the septum and the right ventricular free wall, similar to the Car4 pattern in mouse (Figure 3E and Figure 3—figure supplement 1B). In contrast to mouse Car4, there was no difference in TINAGL1 detection between the ventral and dorsal walls (Figure 5K). Thus, in situ hybridization with TINAGL1 supports a bias in the localization of human Cap1 to the septum and human Cap2 to the ventricular wall. Since lineage data and in situ immunofluorescence confirmed a subset of mouse Cap1 as containing the Endo-derived cells present in the septum (Figure 3C), the human expression data in Figure 5H–K suggested that human Cap1 may also represent an enrichment of septal cells with less blood flow and could also be biased toward the Endo lineage, although the latter cannot be confirmed with gene expression alone. When using the Label Transfer workflow to specifically map the putative mouse septal ECs to the human data, a higher percentage of human Cap1 than Cap2 cells matched to this septal EC group (35% of Cap1 cells versus 14% of Cap2 cells) (Figure 5—figure supplement 1E). We concluded that similar developmental environments and EC states exist between mouse and human capillaries, including those unique to the septum.

Because unbiased clustering produced three artery clusters in human but only two in mouse, we next investigated whether human hearts contained an artery cell state not present in mouse. This could occur because mouse and human arteries have some structural differences, for example, large conducting arteries in humans are on the surface rather than within the myocardium as in mouse (Wessels and Sedmera, 2003; Kumar et al., 2005; Fernández et al., 2008; Sorop et al., 2020). If these differences translated into an artery transcriptional state unique to human, we would expect one of the human artery clusters to not be represented in the mouse mapping. Instead, mouse cells mapped to all three human artery clusters (Figure 6a). There was also evidence indicating that human Art3 cells were in a less mature arterial state compared to human Art1 and Art2. This is because: (1) Art3 was unique among the human artery clusters in that a high proportion of Art3 cells matched mouse capillaries (Figure 6A). (2) Trajectory analysis with RNA velocity (La Manno et al., 2018), Slingshot (Street et al., 2018), and partition-based graph abstraction (PAGA) (Wolf et al., 2019) indicated a transition from human Art3 -> Art2 -> Art1 (Figure 6B and Figure 5—figure supplement 2B). (3) Comparing the fetal dataset with a publicly available adult dataset (Litviňuková et al., 2020) showed that almost no adult human artery ECs mapped to Art3, which would be predicted if Art3 were an immature developmental state (Figure 6C and Figure 5—figure supplement 1F). The direction of the arterial trajectory was determined by the arrows from the RNA velocity analysis, which provides this directional information as a consequence of comparing spliced (mature) to unspliced (immature) transcripts (La Manno et al., 2018). Additionally, this trajectory is supported by previous lineage analyses in mouse of a trajectory from capillaries to Gja5- arteries to Gja5+ arteries (Su et al., 2018). Together, these data suggest that human hearts do not contain a dramatically unique artery transcriptional state when compared to mouse, but they do have an immature state (Art3) that matures into Art1 and Art2 in adults.

Reference mapping from the fetal to adult human datasets also revealed a notable reduction in Cap1 cells (28% of fetal capillary cells are Cap1 versus 6% of adult capillary cells) (Figure 6C). In mice, e17.5 Cap1 and Cap2 converged into a relatively homogenous capillary population in adult (Figure 1I). The small percentage of adult human coronary ECs mapping to the fetal Cap1 cluster, as well as the spatial overlap between cells mapping to Cap1 and Cap2 in the adult UMAP reduction, indicates that these developmental states related to flow and oxygenation also converge in humans.

We next performed trajectory analysis to investigate whether arteries develop through the differentiation of capillary ECs, which is the developmental pathway in mice (Red-Horse et al., 2010; Su et al., 2018). Two common methods for estimating trajectories, PAGA and Slingshot, identified connections between human Cap1 and Art3 and Cap2 and Art2 (Figure 6B and Figure 5—figure supplement 1B). RNA velocity suggested directionality going from the capillaries into arteries (Figure 6B). These two predicted capillary to artery transitions suggested that Cap1/Art3 and Cap2/Art2 may be differentiation trajectories occurring in two different locations in the heart, that is, septum versus ventricle walls. This is supported by the observation that several genes are specifically co-expressed in Cap1 and Art3 (including CXCR7, MCAM, and TNFAIP8L1, PGF) or in Cap2 and Art2 (including TINAGL1, SLC9A3R2, SGK1, THBD, LIMS2, CALCRL), some of which were shared with mouse (Figures 5H and 6D, and Figure 5—figure supplement 1G). From these data, we propose a model where, in both mouse and human, two distinct subtypes of capillary cells at different locations in the developing heart initially produce two subtypes of artery cells, one of which eventually matures into the other (Figure 6E).

Since coronary artery disease is a leading cause of death and developmental information could suggest regenerative pathways, we next focused on the gene pathways present in developing human coronary arteries. As described above, unbiased clustering resulted in three artery states, each expressing known artery markers such as GJA4 and HEY1, but also containing unique genes (Figure 7A). The SCENIC package (Aibar et al., 2017), which uses gene expression information to identify transcription factor ‘regulons’ present in cells, implicated SOX17 as being strongly enriched in developing artery ECs (Figure 7B), which is consistent with previous reports on artery development (Corada et al., 2013; González-Hernández et al., 2020). Transcription factors of potential importance that have not been previously implicated in artery development were PRDM16 and GATA2 (Figure 7B). Interestingly, the IRF6 regulon was specific to the most mature population suggesting a potential role in artery maturation (Figure 7C). All of these regulons were similarly enriched in mouse artery cells (Figure 7B and C). We also identified several genes with strong expression patterns in human artery ECs that were not found in mouse (Figure 7D and Table 4). Interestingly, these included a GABA receptor, GABBR2, which was enriched in Art2, and a Glutamate receptor, GRIA2, which was is enriched in Art1. The human cells expressing GABBR2 also co-expressed SLC6A6, a transporter that imports the GABBR2 ligand into cells (Tomi et al., 2008; Figure 7D). Finally, we localized different types of arteries in sections of human fetal hearts. In order to identify the artery subtypes, we used in situ hybridization for GJA4 and GJA5. We found that GJA5-positive ECs, marking Art1, are in a small number of large arteries always covered with smooth muscle, while GJA5-negative/GJA4-positive ECs, marking Art2 and Art3, are numerous and in some cases not covered with smooth muscle (Figure 7E and F). This observation is consistent with the trajectory showed in Figure 6B, with the interpretation that the more mature GJA5+ arteries in human are larger and more proximal than GJA5- arteries, as they are in mice. This supports the conclusion that Art1 represents artery ECs in larger, more proximal branches, and Art2 and Art3 are smaller arterioles (Figure 7G).

Cluster hArt1 (GJA5+ GJA4+) localizes to the largest arteries that are covered by mature SMMHC-positive smooth muscle (white arrows). Clusters hArt2 and hArt3 (GJA4- GJA4+) are smaller and can be either covered (yellow arrowheads) or not (red arrowheads) by smooth muscle. Scale bar = 100 μm. Scale bar from (E) also applies to (D). (G) Schematic of human coronary artery hierarchy.

Under IRB approved protocols, human fetal hearts were collected for developmental analysis from elective terminations. Gestational age was determined by standard dating criteria by last menstrual period and ultrasound (ACOG, 2009). Tissue was processed within 1 hr following procedure. Tissue was extensively rinsed with cold, sterile PBS, and placed on ice in cold, sterile PBS before further processing as described below. Pregnancies complicated by multiple gestations and known fetal or chromosomal anomalies were excluded.