Dedifferentiation and Proliferation of Artery Endothelial Cells Drive Coronary Collateral Development in Mice

Raw sequence files and Cell Ranger processed data are available on GEO (accession ID: GSE210307).

Scripts used for the analysis are available at https://github.com/Snehasrivatsa/scRNA.

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Neonatal and adult mouse hearts were dissected, and atria were removed from each heart. Tissue digestion was performed only with cardiac ventricles. Adult mouse hearts were additionally perfused before harvesting hearts. Approximately 15 to 20 neonatal (for sham and MI each) and ≈10 to 12 adult (for sham and MI each) Cx40CreER; Rosa26^TdTomato mice were used for tissue digestion. Post explant, ventricles of the hearts were minced using razor blade and fine forceps and transferred to a tissue digestion buffer. The tissue digestion buffer was composed of 500 U/mL Collagenase IV (catalog number LS004186; Worthington), 1.2 U/mL Dispase (catalog number LS02100; Worthington), 32 U/mL DNase I (catalog number LS002007; Worthington) in commercial grade sterile 1× DPBS with Ca²⁺ and Mg²⁺. Each heart was transferred to ≈300 µL of digestion buffer. Hearts bathing in digestion buffer were incubated in 37 °C shaker for 40 to 45 minutes, with periodic resuspension using micropipettes to break down tissue lumps. Digested tissue was then equilibrated with 4 volumes of 5% FBS prepared in 1× DPBS, gently mixed with a micropipette, and filtered using 40 µm of sterile cell strainer. The filtered tissue digest was spun down for 5 minutes at 400g at 4 °C, and the cell pellet was gently dissolved in 5 mL of 3% FBS in 1× DPBS. The resuspended cell pellet was spun down for 5 minutes at 400g at 4 °C. Supernatant was discarded and the cell pellet was suspended in ≈500 to 600 µL of 3% FBS in 1×DPBS. These suspended single cells were stained with Ter119-Alexafluor-647 (catalog number 116218, 1:100 dilution; BioLegend) for 30 minutes at 4 °C. Poststaining, cells were washed with and resuspended in fresh 3% FBS/1× DPBS, followed by single-cell sorting with fluorescence-activated cell sorting. DAPI (4′,6-diamidino-2-phenylindole) was added before starting the cell sort. Live single ECs (DAPI⁻ Ter119⁻ TdTomato⁺) were collected, in that order, using BD Fluorescence-Activated Cell Sorting Aria II cell sorter. Single cells were counted, and submitted to Stanford Genome Sequencing Service Center for 10× single-cell V3 library preparation. Sequencing was performed using HiSeq 4000.

Data were log normalized with scale factor set to 10 000, followed by finding variable features with n set to 2000 and finally, scaled with scaledata (a command from Surat package to scale and center the gene features in the data set). We focused on artery ECs, by subsetting the data using different markers. Cadherin5 or Pecam1 (pan endothelial markers) expression >1 was analyzed. A filter for >1 was used for Cadherin5 or Pecam1 to exclude non-ECs expressing these genes at a minimal level. As the cells from both filters echoed the same clustering pattern, we used Cadherin5 cells for further analyses. The pattern acquired by clustering sham and MI data separately or together (merged) echoed the same patterns. Hence, the sham and MI data were integrated using the IntegrateData feature (a command from the Seurat package). Normalization, finding variable features, and scaling data were repeated on the integrated data, using the same settings as above.

We performed principal component analysis and top 30 PCs were computed and stored for the downstream analyses. Clusters for the integrated data were obtained with Louvain algorithm with resolution set to 0.5 (https://satijalab.org/seurat/articles/pbmc3k_tutorial.html). The cells were projected and visualized in 2-dimension using Uniform Manifold Approximation and Projection (UMAP) for Dimension Reduction; parameters were n.neighbors set to 10, dims set to 1:20, spread set to 2, and min.dist set to 0.3. Attributes group.by and split.by were used to observe sham and MI cells. This was done for optimal spread of cells and best visualization.

TdTomato sequence was aligned with and checked in our data sets. Thus, cells with TdTomato expression >0, were analyzed. Since the MI cell number was 10-fold lesser than that of the sham, data were analyzed by subsetting the merged data based on its orig.ident and reclustering the respective cell populations. The steps used with Cadherin>1 cells were used for all clustering steps with TdTomato >0.

For expression of genes, data were subsetted for gene >0, the cell numbers per cluster were obtained using table ([email protected]).

The communities detected were identified with Violin plots (StackedVlnPlot from CellChat v1.1.0 package (https://github.com/sqjin/CellChat), DotPlots (a feature of Seurat package to visualize gene expression), and feature plots (Seurat argument). Clusters were identified with established markers for a variety of cardiac cell–specific genes. Cluster identifications were confirmed using AddModuleScore feature from Seurat package. A list of cell type–specific, established markers were used for this analysis on Cadherin5>1, Pecam1>1, and TdTomato>0 data sets in both neonate and adult data sets. Average expression of the listed genes for each cell type was visualized using a feature plot. Neonatal trans-EC cluster did not show unique expression of any known cardiac cell types.

Heatmaps were used to check the variation in expression of dedifferentiation genes across sham and MI data sets. DoHeatmap feature from Seurat package was used to plot the genes of interest. Violin plots generated from Seurat package were analyzed by using the split attribute.

Updated G2M and S-phase–specific gene lists (from 2019) on Seurat were used. As the genes were in human gene format, useMart function from biomaRt v3.12 (R package) was used to obtain hsapiens_gene_ensembl and mmusculus_gene_ensembl from ensembl database. Homology mapping and gene symbol conversion of genes from 2 gene sets were performed using getLDS function. Genes were converted to mouse gene format using convertHumanGeneList. Cell cycle analysis was performed using standard cell cycle regression (CellCycleScoring) on Seurat.

Monocle3²² was used for trajectory analysis to understand possible pseudo time relationship between the cell types. Seurat object split into sham and MI from the integrated data, subsetted for ECs alone, was used for the analysis. as.cell_data_set feature from SeuratWrappers was used to convert Seurat object to Monocle3 object. Furthermore, the Monocle3 object was subjected to pseudo time analysis using learn_graph and order_cells functions (https://cole-trapnell-lab.github.io/monocle3/docs/getting_started/). Results were visualized on UMAPs with the aid of plot_cells function.

Spliced and unspliced counts were generated from raw Fastq files using kallisto bustools wrapper kb (kb-python v0:27:0, https://www.kallistobus.tools/).²³ Mouse genomic (DNA) FASTA and GTF annotations were downloaded from Ensembl and a transcriptome index was generated by kallisto using kb ref command. Kb count command was used to generate the count matrix in the form of h5ad file. The h5ad file was converted into a Seurat object using SeuratDisk (v0.0.0.9019). The preprocessing, normalization, scaling, and clustering were done as described earlier. The RNA velocity maps were generated using Velocyto package (velocyto.R v6).²⁴

R package Cellular (Cyto) Trajectory Reconstruction Analysis using Gene counts and Expression (CytoTRACE) v0.3.3²⁵ (https://cytotrace.stanford.edu/) was used for cell differentiation analysis. Cell counts and UMAP embeddings were obtained as data frames from sham and MI clusters separately, for the analysis. A CytoTRACE score was assigned to each cell with the CytoTRACE function, which was visualized on the UMAP embeddings with plotCytoTRACE feature. We compared the mean CytoTRACE scores between all EC clusters. To do so, CytoTRACE score for each cell and its cluster number was obtained. The mean CytoTRACE score was then calculated for each cluster. These mean values along with standard deviations were plotted as boxplots using ggboxplot feature from ggpubr v0.4.0 package.

List from FindAllMarkers was used to recognize processes that appear on publicly available online tool Gene Ontology Enrichment Analysis and Visualization Tool (GOrilla)²⁶ (http://cbl-gorilla.cs.technion.ac.il/). Gene Ontology analyses were performed on MI gene list, with sham genes serving as background. Default settings with a filter of P<10⁻³ were used. Processes with significant P values (<0.0005), and relevant to our study, were plotted on a bar graph.

Pathway analysis was performed using an online tool g:Profiler.²⁷ The gene list was obtained for neonatal cluster aEC1 (from merged Cadherin5>1 data set), using FindMarkers feature, with test.use set to “MAST.” The organism was set to Mus musculus and the pathways were obtained from databases like KEGG, reactome, WikiPathways, etc.

Wilcoxon test was performed using RunPresto feature from SeuratWrappers package. The statistical test was performed on cycling aECs (cyc-aEC) and aEC1 cells from neonatal Cadherin5⁺ MI group. Additionally, the test was performed on Cadherin5⁺ cyc-aECs from neonatal and adult MI data sets. For this, the data were merged using the merge feature from Seurat and retained as such without any further modifications. Test was done to check statistical significance for expression of Cx40, Cxcr4, Aplnr, Car4, Apln, Foxm1, Vegfa, Vegfc, and Kdr across different treatments (sham and MI in Figure 4) or clusters (cyc-aEC and aEC1 in neonates, in Figure S7A).

Following mouse lines were used in this study. Rosa26^TdTomato Cre reporter line (The Jackson Laboratory, B6.Cg-Gt[ROSA]26Sortm9[CAG-TdTomato]Hze/J, stock number 007909), Cx40CreER,²⁸ VegfR2 flox (The Jackson Laboratory, Kdr^tm2Sato/J, stock number 018977), Cx40^eGFP/+.²⁹

Mice with mixed strain background were used. All mice were housed and bred either at Stanford (in accordance with Institutional Animal Care and Use Committee) or National Centre for Biological Sciences animal facility (in accordance with Institutional Animal Ethics Committee). Mice were fed gamma irradiated vegetal rodent maintenance diet SAFEA40. All tamoxifen injections (6 mg dissolved in corn oil) were administered intraperitoneally to nursing mothers for Cre activation in new born pups. Efficiency of tamoxifen and Cre activation has been shown earlier to be ≈84% using Cx40CreER mice and ≈86% using ApjCreER mice.¹⁷ Data shown in this study is compiled from age-matched males and females from several litters. Data from both males and females were combined.

Sample size for the current experiment was determined from our previous study conducted with the same animal model with similar treatments.¹⁷ In any given litter, animals were categorized as control or knockout based on their genotype, and irrespective of the sex. Severity of MI was determined by first checking the location of left coronary artery (LCA) ligation under fluorescence stereo-microscope (to check knot around TdTomato⁺ arteries). Hearts with severe MI (where either the main LCA or the 2 primary branches of LCA were captured) were included in this study. Hearts with mild MI (where secondary or tertiary branches were ligated) were excluded from the study. Biological replicates from multiple litters were included in the study and are mentioned below for each in vivo experiment (Table).

Adult mice were anesthetized in a 4% isoflurane chamber and then intubated with a 20G angiocatheter. General anesthesia was maintained with 1.5% to 4% isoflurane during mechanical ventilation at a respiratory rate of 180 breaths per minute. The mice were placed in right lateral decubitus position, and the left chest was shaved and sterilely prepped. 1 mg/kg bupivacaine and 0.5 mg/kg Buprenorphine were injected subcutaneously for perioperative analgesia. A 1 cm left thoracotomy incision was made, and the pericardium was opened. The LCA was identified and permanently ligated using a 6-0 polypropylene suture. The chest was then closed in layers using 5-0 polypropylene sutures. Isoflurane was then weaned off and the mice were extubated after regaining physiological respirations. 5 mg/kg Carprofen was injected subcutaneously daily for the first 3 days after surgery for postoperative analgesia. For adult MI experiments, 2 to 3 months old mice were used (both males and females). To activate Cre, a single dose of 6 mg tamoxifen intraperitoneal injections were administered to each mouse, 2 days before MI. Echocardiograms were obtained from P32/P33 adult mice (sham or MI surgeries at P2) ≈30 days post-MI using VisualSonics Vevo 3100 imaging system (FUJIFILM VisualSonics, Inc) with an MX400 20–46 MHz transducer. Data were analyzed using Vevo LAB Version 3.1.0 (Build 13029).

Technical details of the procedure are referred and modified from Mahmoud et al.^17,30 To induce hypothermic arrest, neonatal mice (P2 or P7) were incubated on ice. Neonates were wrapped with a 3-layer, precooled, gauze piece for ≈3 minutes. Anesthesia was confirmed using toe pinch. Surgery was performed in sterile conditions from start to end, using sterile surgical tools at all times. Animals were placed in a supine position, on a sterile surgical drape, wrapped around cold ice pack. Animals were then prepped with betadine and 70% ethanol. Left anterolateral thoracotomy was performed by making an incision on fourth intercostal space to access heart, under a dissecting microscope. The LCA was identified and ligated by passing a needle attached to 8-0 nonabsorbable prolene suture (catalog number NW3322; Ethicon Ethilon). Post-ligation, the 8-0 prolene suture was used to close the ribs, intercostal muscles, and skin separately. The neonate was then transferred to warm plate (37 °C) to gain consciousness and recover completely. When conscious, neonates were returned to their mother’s cage and closely observed over the next few days.

The procedure has been described earlier.¹⁷ Neonatal whole hearts were dissected and immediately fixed in 4% paraformaldehyde at 4 °C for 1 hour on a rocker. Hearts were then subjected to 3- and 15-minute washes, with 1× PBS at 4 °C. Primary antibody staining was done by incubating hearts in respective dilutions of primary antibodies, prepared in at least 5 volumes of 1× PBS containing 0.5% Triton-X 100 (0.5% PBT). Incubations were done overnight at 4 °C, on a shaker. Hearts were washed in 20 volumes of 0.5% PBT for 12 hours, with a change in wash buffer (0.5% PBT), every 2 hours at 4 °C. The secondary antibody staining was performed by incubating hearts in 1:250 dilution prepared in 0.5% PBT. Incubation was done on a shaker. Next, hearts were washed in 20 volumes of 0.5% PBT, for 4 days with a change in wash buffer, every 2 hours at 4 °C. All washes were performed with continuous but gentle shaking. Hearts were cleared in at least 2 volumes of anti-fade mounting media (https://www.jacksonimmu number com/technical/products/protocols/anti-fade) for 2 hours, in dark, at room temperature. Whole hearts were then stored at −20 °C till imaged. Poststaining washes were done in 15- and 50-mL tubes, whereas incubation in antibodies or mounting media were performed in 1.5 mL microtubes.

Neonates were injected with 30 µL of 10 mM 5-ethynyl-2′-deoxyuridine (EdU; prepared in 1× commercial grade PBS, catalog number 10010023) for 1 hour, before harvesting their hearts. EdU label detection was combined with postsecondary immunostaining steps as described above and detected using Click-iT staining kit (catalog number C10340; Invitrogen).

Whole neonatal (immunostained or genetically labeled/traced with fluorescent reporter) hearts were mounted on a double concave microscope slide (catalog number 7104; Sail Brand) using a microscope coverslip (0.13–0.17 mm, catalog number 48367-059; Thermo Scientific) so that the ventral watershed regions spanning left and right coronary arteries can be imaged. Imaging was performed using Olympus FV3000 upright confocal microscope. Whole hearts were imaged using ×4, ×10, ×20, or ×60 (oil) objectives. Multiple z stacks were acquired using FV31S-SW software. Multiple images of the whole heart were processed on Fiji-ImageJ software, and stitched with Adobe Photoshop.

Primary antibodies: Anti-CX40 (1:500, catalog number CX40-A, RRID: AB_1616128; Alpha Diagnostics Int, Inc), Anti-VEGFR2 (1:125, catalog number AF644, RRID: AB_355500; R&D Systems). Secondary antibodies: Alexa fluor-conjugated antibodies (488, 555, and 633) from Invitrogen were used at 1:250 dilutions.

Proliferating TdTomato⁺ Cx40⁺ cardiac aECs (4 days post-MI) were identified with EdU⁺ nuclear labeling. Cytoplasmic intensity of Cx40 in these (TdTomato⁺ Cx40⁺ EdU⁺) ECs was quantified using images acquired with ×60 (oil) objective. Mean fluorescent intensity of cytoplasmic Cx40 signal was measured using a selection tool available with Fiji-ImageJ software. Quantification was performed with 68 single TdTomato⁺ Cx40⁺ EdU⁺ aECs, found in the watershed of 4 hearts, collected 4 days post-MI. Comparison between proliferative (EdU⁺) and nonproliferative (EdU⁻) aECs were made within the same field of view. A paired t test was done to check statistical significance. Similar analysis was done on 67 aECs within artery tips or integrated into mature arteries, as identified by Cx40 immunostaining. Data were compiled from 5 MI hearts from different 4 litters.

EdU⁺ TdTomato⁺ single aECs in the watershed regions were quantified using images acquired with ×20 objective. On an average 20 to 30 TdTomato⁺ single aECs were observed per watershed (each field of view), which was consistent across genotypes. Quantification was performed on 6 control (36 field of view) and 5 VegfR2 knockout (24 field of view) hearts. A total of 784 control TdTomato⁺ aECs and 512 VegfR2 knockout TdTomato⁺ aECs were counted for this quantification. Data were compiled from 3 to 5 litters from each genotype. Six control and 5 knockout hearts were used for quantification. An unpaired t test was performed to check statistical significance.

EdU⁺ TdTomato⁺ proliferating aECs were counted in artery tips and single aECs (saECs) from 3 Cx40CreER; Rosa26^TdTomato hearts. ≈80 to 120 µm of distal ends of arteries were identified as artery tips. Twenty-five fields of view from P11 watershed regions, from images taken with ×60 (oil) objective, were used to count a total of 113 TdTomato⁺ residual single cells.

Mean fluorescence intensity was measured using Fiji for Cx40CreER lineage–traced TdTomato⁺ single aECs found in watershed area 4 days post-MI. Images of watershed regions (each field of view) lacking any large or collateral arteries were acquired with ×20 objective. Watershed images only with single artery cells were considered for this quantification. 3 hearts from each genotype were used for this quantification. An unpaired t test was performed to check statistical significance.

LCA width was measured using a line selection tool on Fiji. The analysis was done for 3 control and 2 VegfR2 knockout hearts. Width was measured for the main branch of LCA in all hearts and genotypes. An unpaired t test was performed to check statistical significance.

Molecular profiling of cardiac ECs that participate in collateral artery formation can improve our understanding of ways in which complete cardiac reperfusion and regeneration can be attained. We, thus, performed single-cell RNA sequencing of neonatal post-MI heart cells. Specifically, Cx40CreER; Rosa26^TdTomato mice were used, where a single dose of tamoxifen to the lactating mothers induced expression of TdTomato fluorescence in Cx40 expressing aECs in neonates. Neonates were then subjected to either sham or MI surgeries at P2, and hearts were harvested at P4 or P5 (Figure 1A). TdTomato⁺ cardiac cells were sorted using fluorescence-activated cell sorting method, followed by sequencing using ×10 genomics platform. Sequences were processed using Cell Ranger, and the subsequent output files were analyzed using Seurat on RStudio.²¹

Sorting TdTomato⁺ cells during fluorescence-activated cell sorting, enriched our samples for Cx40CreER⁺ lineage-labeled/traced aECs. When analyzed, these TdTomato+ cells were additionally identified to be cardiomyocytes, SMCs, and cells undergoing endo-MT (data not shown). We also obtained some TdTomato⁻ nonartery cells. To specifically study EC biology postinjury, we thus analyzed Cadherin5⁺ ECs, which included, both, TdTomato⁺ aECs, genetically labeled between P0 and P2 (Figure 1A), and TdTomato⁻ aECs (if any), identified by expression of known arterial EC genes. Preprocessing and quality control steps (Figure S1A) resulted in 5432 sham and 2200 MI ECs (Figure 1B through 1D), out of which 4423 sham, and 262 MI ECs, expressed at least 1 copy of TdTomato (Figure 1E and 1F). As previously observed,^31,32 compared with sham, we obtained fewer cells upon MI. This is likely due to extensive cell death following coronary ligation and subsequent tissue ischemia.

Unbiased grouping of Cadherin5⁺ ECs from sham and MI resulted in 10 distinct clusters (Figure 1G and 1H). Merged data from sham and MI were projected onto 2-dimensional space using UMAP. The identified cell types were—mature aECs (aEC1–4), transitioning ECs, cyc-aEC, cells undergoing endo-MT, SMC1,2, and endo-EC (Figure 1G and 1H).

We identified the cell types with known vascular and cardiac cell markers and their relative expression levels (Figure 1I; Figures S1 through S3). aEC1–4 expressed markers for mature aECs (Cx40, Cxcr4) and SMC1,2 expressed markers for both pericytes and SMCs (Pdgfrb, Notch3). Cells undergoing endo-MT were identified by upregulated expression of mesenchymal markers such as Col1a1 and Cdh2 and endo-MT–associated transcription factors like Serpine1, Twist1, and Snai1 (Figure 1I; Figure S1B). ECs have been reported to express perivascular markers, such as αSMA (α-smooth muscle actin) and Tagln.^33–38 Thus, it was not surprising to find that some of these markers were coexpressed by SMC clusters (Figure S1B). An EC cluster expressed several genes specific to G2/M (Top2a, Mki67, Birc5, Cenpa, Kif2c, Bub1, Nuf2, and Hmmr; Figure 1I; Figure S1C through S1F) and G1/S (Tyms, Mcm4, Mcm6, Mcm7, Ung, Chaf1b, Fen1, and Gmnn; Figure 1I; Figure S1G through S1J) transition. Sham cells in this particular cluster mostly expressed genes like Cx40, Gja4, Jag1, and Cxcr4 (Figure S2) known to be significantly expressed and being specific to mouse³⁹ and human⁴⁰ coronary arteries. The top 20 genes in this cluster included artery-specific genes⁴¹ like Fgfr3, Slc45a4, Gja4, and Ssu2 (data not shown). We hence, named the cluster cyc-aEC. Endo-EC expressed Npr3,⁴² Cdh11, Tgfbi,⁴³ and Adgrg6⁴⁴ (Figure 1I) and some venous markers (Figure S1K). Nonendothelial cardiac cells were mostly absent in our data set (Figure S1K).

We also identified a cluster of cells (named trans-EC in Figure 1G and 1H) that did not express markers for any single EC subtype. Reclustering this group resulted in 4 distinct EC subclusters, those highly expressed markers for arteries, capillaries, venous/capillaries, and myofibrillar genes⁴⁵ (Figure S3A and S3B). This cluster especially was checked for doublets. No doublets were present in MI trans-EC cluster (data not shown). Trans-EC cluster in the sham data set had 3 doublets (data not shown). 42.8% of artery-like ECs, in this cluster, were TdTomato⁺ (Figure S3C) indicating their origin from preexisting aECs, those labeled between P0 and P2 (Figure 1A). Additionally, this artery-like EC subcluster had fewer cells expressing Cx40 or Cxcr4 (Figure S3B). The indistinct molecular nature of this entire cluster, and its proximity to both, a proliferative (cyc-aEC) and mature (aEC1/3) aEC clusters (Figure 1G and 1H), led us to hypothesize that cells in this cluster are in an intermediate state, that is, between a proliferative and mature state. We, hence, called this cluster, trans-EC. Bioinformatics analyses of Pecam1⁺ cells, instead of Cadherin5⁺ cells, resulted in identification of the same cell types, clustered in a similar manner (data not shown).

Artery cells rarely proliferate, and exiting cell cycle is required for artery cell specification.^17,46–48 However, we identified an aEC cluster (cyc-aEC) in our data set, expressing several genes that mark G1/S and G2/M transitions (Figure 1I; Figure S1C through S1F). Cell cycle analyses of Cadherin5⁺ cells showed a 12.4% increase in MI cells undergoing G2 to M transition as compared to the sham group (Figure 1J). The number of MI cells in S-phase increased by 3.3% and G0/G1 phase reduced by 15.6% (Figure 1J), suggesting a higher number of MI cells were likely mitotic.

We next performed cluster-wise quantification of mitotic cells (expressing G2/M markers), in sham and MI groups. In sham group, 2.5% of all mitotic cells were present in artery clusters (aEC1–4 and cyc-aEC; Figure 1K). In MI group, we observed 11% of all mitotic cells in artery clusters (Figure 1K), indicating a 4.4-fold increase in mitotic aECs, upon MI. 15.9% of all Cadherin5⁺ cells were mitotic in the MI group, and majority of these cells (≈69.2%) were aECs (Figure 1K), a stably differentiated cell type in neonatal hearts.

Post-MI aEC proliferation is associated with artery reassembly and complete cardiac regeneration.¹⁷ Use of aEC-specific Cx40CreER mouse line, allowed us to mark the lineage of preexisting aECs with TdTomato fluorescent reporter (Figure 1A). We performed cell cycle analysis and quantified the number of Cadherin5⁺TdTomato⁺ ECs that expressed genes for G2/M transition and were likely to be mitotic. 13.4% of all Cadherin5⁺TdTomato⁺ MI cells were mitotic, ≈63% of which were present in aEC clusters (aEC1–4 and cyc-aEC; Figure 1L). Analyzing mitotic cell distribution within the MI group showed 69.3% of all mitotic cells to be present in artery clusters but very few (6.3%) of them were TdTomato⁺ aECs (Figure 1M). This suggests that MI induces expansion of an aEC population, not lineage labeled by, or traced from Cx40CreER⁺ cells. This is consistent with our previous observation that injury causes organ-wide artery growth (especially, artery tips), independent of artery reassembly, and through capillary differentiation.¹⁷

We next performed in vivo experiments to confirm if an aEC population, originating from nonartery sources (as seen in Figure 1M) is indeed proliferating, upon MI. EdU labeled P6 hearts from Cx40CreER; Rosa26^TdTomato neonates (tamoxifen at P0, MI at P2) were harvested and immunostained for Cx40. Cx40 immunolabelled all mature aECs, including Cx40CreER lineage–traced TdTomato⁺ preexisting aECs. We indeed, observed numerous Cx40⁺TdTomato⁻ arteries (Figure S4A through S4C), especially in the ischemic regions. Furthermore, we identified these aECs to be proliferating (EdU⁺) and integrated into Cx40⁺ vessels (Figure 1N and 1O; Figure S4A through S4C). As expected, we also observed Cx40⁺TdTomato⁺EdU⁺ aECs (Figure 1N and 1P; Figure S4D and S4E). Thus, ischemia can induce formation of new coronary arteries via multiple mechanisms (Figure S4F). First, coronary collateral artery segments are generated via artery reassembly, a process driven by preexisting aECs. Second, an overall organ-wide artery growth such as extension of artery tips occurs by 2 processes: (1) incorporation of capillary ECs as observed earlier¹⁷ (also suggested in Figure S4A through S4C) and by (2) proliferation of preexisting (TdTomato⁺) aECs (as shown in Figure S4D and S4E).

New collateral arteries in adult mice, form in response to coronary occlusion within a week of injury.⁴⁹ However, adult coronary collateral arteries are fewer in number, have smaller diameters, and are predicted to perfuse poorly than those in neonate.⁵⁰ The reason for this age-dependent difference in which aECs respond to ischemia is unclear. One possibility could be that adult collateral arteries do not form naturally through artery reassembly. Keeping this in mind, we investigated the molecular nature of adult coronary aECs, and their derivative cell types, in sham and MI mice.

Adult hearts execute artery reassembly only in response to exogeneous Cxcl12. As a result, aEC lineage–traced coronary collateral arteries are observed only after 14 days post-MI.¹⁷ This is also the time point when the number of adult coronary collateral arteries peak.³⁹ To capture single aEC response that precedes new collateral formation, we harvested adult hearts 9 and 11 days post-MI. This stage could reveal adult molecular events, different from that of neonates’. We performed single-cell analyses of cardiac cells sorted from adult Cx40CreER; Rosa26^TdTomato ventricles (Figure 2A; Figure S5). Similar to neonatal data, we obtained both, TdTomato⁺ and some unlabeled cells. After performing quality control steps (Figure S5A), we analyzed 6256 Cadherin5⁺ sham and 1690 Cadherin5⁺ MI cells (Figure 2B and 2C). Like neonatal data, our adult data set was enriched for aECs (Figure 2D and 2E) and we obtained fewer Cadherin5⁺ (Figure 2C) and TdTomato⁺ (Figure 2F and 2G) cells in MI group than sham.

Based on cell-specific gene expression patterns, we obtained 12 adult cell populations (Figure 2D); 7 of which, constituted mature aEC1 to 7, 1 cyc-aEC, 1 capillary EC (cap-EC), 1 endo-MT, and 2 SMC clusters (Figure 2D and 2E). Unlike neonates, trans-EC cluster was absent in adults. Instead, we identified a prominent adult capillary EC population, which when further clustered generated 3 distinct microvascular EC populations—artery-cap-EC, cap-EC, and venous-cap-EC (Figure S5B through S5D). A significant number of art-cap-EC cluster (34%) expressed TdTomato fluorescence (Figure S5D), indicating an origin from preexisting mature arteries. Identity of cycling aECs was confirmed with comparable expression of artery-specific genes (Gja4, Cx40, Cxcr4, Jag1) in cyc-aEC and aEC1 to 7 (Figure S5E). A similar cell type distribution was also observed when bioinformatics analyses of Pecam1⁺ cells, instead of Cadherin5⁺ cells, were performed (data not shown).

Adult cardiomyocytes have limited proliferation capacity.¹⁶ To assess whether the same holds true for cardiac ECs, we performed cell cycle analyses on sham and MI adult Cadherin5⁺ ECs. Unlike neonates, we observed only a minimal increase of 2.5% in the number of ECs undergoing G2 to M transition, upon MI (Figure 2H). Further cluster-wise analyses revealed only 1.8% increase in the mitotic (G2/M) aECs (4.5%–6.3%; Figure 2I), compared with 8.5% increase observed in neonates (Figure 1K), upon MI. 4.6% of adult sham cells were mitotic (Figure 2I), suggesting that the preinjury mitotic baseline for adult Cadherin5⁺ aECs was higher than that observed in neonates (2.5%; Figure 1K). Thus, in comparison to neonates, fewer adult aECs undergo G2/M transition in response to MI.

P7 or older hearts are nonregenerative^15,16 and do not illicit artery reassembly.¹⁷ Cell cycle analyses of Cadherin5⁺ TdTomato⁺ lineage–traced adult ECs revealed minimal changes in G2/M ECs including in aEC clusters (aEC1–7, cyc-aEC; Figure 2J). Furthermore, we checked which cell types originate from preexisting aECs, upon MI. For this, we analyzed all TdTomato⁺ cells (labeled and traced from Cx40CreER⁺ resident aECs) in adult MI data set and compared it with neonatal MI data set (Figure 2K). Although the neonatal MI group showed a distinct cyc-aEC cluster, this population was absent in the adult MI group (Figure 2K; Figure S5F and S5G). Thus, adult hearts lack a proliferating aEC subpopulation, known to be crucial for artery reassembly, and complete cardiac regeneration.¹⁷

We next performed in vivo experiments to check if lack of proliferating artery cell population in injured adult hearts (Figure 2J and 2K), is also observed in nonregenerative P7 hearts. Cx40CreER; Rosa26^TdTomato neonates were administered with tamoxifen to lineage label aECs at P5, through P7, followed by MI at P7 (Figure 2L). Proliferation status of Cx40CreER⁺ lineage–traced, TdTomato⁺, aECs were analyzed with EdU label, before harvesting hearts at P11, 4 days post-MI. As shown earlier,¹⁷ these hearts did not demonstrate artery reassembly (Figure 2L and 2M), which was evident from lack of lineage-traced (TdTomato⁺) collateral arteries and single aECs in the injured watershed regions (Figure 2M). Very few single aECs were observed adjacent to artery tips (arrowheads, Figure 2M); we called them residual single aECs. EdU⁺ cells in TdTomato⁺ aEC population—in both, artery tips (Figure 2N) and residual single aECs in the watershed (Figure 2O)—were rarely observed. Together, data from both bioinformatics analyses and in vivo experiments indicate that aECs in nonregenerative P7 or adult hearts have limited self-renewing capacities.

Stably differentiated cells may attain a less differentiated state and reenter cell cycle in response to pathophysiological cues. The acquired proliferative abilities allow them to transform into less specialized cell type. They can then, either re-differentiate into cells within the same lineage or into cells of a different lineage.⁵¹ This phenomenon can often be detected through changes in gene expression patterns. Presence of an actively cycling differentiated cell cluster in neonates (cyc-aEC) led us to hypothesize that cells in this cluster dedifferentiate and contribute to new arteries, in response to MI.

Dedifferentiation is indicated by loss of differentiation markers, cell cycle reentry, and expression of early genes, suggesting lack of cell fate specification. We checked whether MI cells when compared with sham cells, manifested any of these dedifferentiation indicators. In sham, Cx40, a mature aEC marker, was expressed by cyc-aECs, at levels comparable to mature artery clusters—aEC1 to 4 (Figure 3A and 3B). In MI group, cells in cyc-aEC cluster expressed Cx40, at levels much lower than what was observed in sham cyc-aEC (Figure 3B) or the rest of the aEC1 to 4 (Figure 3B and 3C). The reduction of average expression levels of Cx40 from mature aECs to cycling aECs was of higher magnitude in the MI group (21.6-fold reduced) than in sham (1.3-fold reduced). The percentage of cells expressing Cx40, upon MI, reduced in all clusters (Figure 3A and 3C), but this decrease was most prominent in trans-EC (19% in sham, 0% in MI) and cyc-aEC (57% in sham, 3% in MI) clusters (Figure S6A). A similar reduction in number of Cxcr4⁺ MI cells was observed specifically in cyc-aEC and trans-EC clusters (Figure S6A). The average expression level of Cxcr4, in MI group, reduced by 3.83-fold in trans-EC and 6.7-fold in cyc-aEC, when compared with mature aECs (aEC1–4). This change was negligible in sham group (≈1.1-fold decrease). Thus, there is an overall reduction in number of cells expressing mature artery markers, and their expression levels, in cyc-aEC and trans-EC cluster, upon MI.

During development, an immature coronary vessel plexus undergoes morphogenesis to build coronary arteries.⁵² We hypothesized that cyc-aECs, if dedifferentiating, would reexpress some early genes expressed by this vessel plexus that mostly consists of venous/capillary ECs and from which coronaries originate during development. We, hence, analyzed sham and MI cells for cardiac microvascular genes, such as Aplnr⁵³ and Car4.^40,54 The number of cells expressing the gene Aplnr was increased upon MI (Figure S6A), but the average expression of AplnR across these cells was considerably low in mature aECs (aEC1–4) as compared to cyc-aECs or trans-ECs (Figure 3B). Car4, also followed the same pattern (Figure 3A and 3C; Figure S6A), indicating both cyc-aEC and trans-EC express venous-capillary marker genes, upon injury.

Angiogenic properties of ECs facilitate vascular remodeling and subsequent cardiac regeneration. We next investigated if cyc-aEC (or trans-EC) clusters expressed any proangiogenic genes, yet another cellular feature of dedifferentiation. Upon MI, 61% of the cyc-aECs and 47% of trans-ECs expressed proangiogenic Apln gene (Figure 3A and 3C; Figure S6A), known to drive vascular development^55,56 and regeneration^57–59 in vivo. Expression of other angiogenic genes like Foxm1,^60,61 Nrp2,⁴¹ Col15a1,⁴¹ and Cd34 were enriched in these 2 clusters (Figure S6B through S6E). Additionally, ECs in cyc-aEC and trans-EC clusters expressed various capillary-like/venous specific genes, such as Gpihbp1, Fabp3, and Pcdh17 (Figure S6F through S6H). All changes in expression of Cxcr4, Cx40, Aplnr, Car4, Apln in cyc-aEC between sham and MI groups were found to be statistically significant using Wilcoxon sum-rank test (Table S1). Thus, cyc-aEC and trans-EC clusters are likely to possess angiogenic properties, which are absent in stably differentiated aECs.

Cardiac vascular cells (ECs and SMCs) and cardiomyocytes have common precursor cells. In addition to classic angiogenic EC markers, cyc-aEC, and trans-EC cells also expressed cardiac myofibrillar genes (Tnnt2, Myl3, and Myh6; Figure S5I through S5K). Thus, cardiac ECs maintain low, but active transcription of these cardiomyocyte-specific markers as observed earlier.⁴⁵

The differentiation status of cells and the projected pseudotime, together, can determine the root of differentiation and hence can predict the direction of path taken by cell populations.²⁵ A package in RStudio − CytoTRACE,²⁵ uses the number of genes and their differential expression, to determine relative differentiation status of each cell with respect to others. CytoTRACE score for cycling aEC, in MI data set, was highest among all clusters, making it the least differentiated cluster (Figure 3D). We also performed pseudotime analysis of sham and MI cells using monocle3 package in RStudio. In sham, the predicted trajectory connected trans-EC to aEC1 to 4. Sham cyc-aECs were excluded from the projected path (Figure 3E). In contrast to sham, in MI, cyc-aEC, trans-EC, and aEC1 clusters were associated with the predicted path (Figure 3F). If multiple clusters are connected through a single trajectory, it is likely that, a less specialized cluster differentiates into a terminal cell type. This prediction is supported by our observations from genetic lineage tracing experiments in vivo, where upon MI, Cx40CreER lineage-labeled, proliferating aECs give rise to mature coronary collateral arteries¹⁷ and contribute to organ-wide artery growth (Figure S4F).

We next assessed the relative differentiation status of the 3 clusters participating in the trajectory obtained for MI cells (Figure 3F)—cyc-aEC, tans-EC, and aEC1. We compared the cellular and molecular features indicating dedifferentiation, across these 3 clusters (Figure 3G). As expected, cyc-aEC and aEC1 were the least and most differentiated clusters, respectively (Figure 3G). Trans-EC cluster showed dedifferentiation features intermediate between cyc-aEC and aEC1 (Figure 3G). Thus, it is likely that proliferating (less specialized) aECs in cyc-aEC cluster are differentiating into mature aECs in aEC1 (Figure 3F and 3G). These results were supported by RNA velocity analyses, which, additionally predict directionality of projected trajectories based on the ratio of spliced and unspliced transcripts (Figure S7A and S7B). Together, we hypothesized that upon neonatal MI, at least a subpopulation of mature aECs originates from dedifferentiation and proliferation of preexisting/resident artery cells.

We checked if aECs, which enter cell cycle upon MI (as seen in our single-cell analyses in Figure 1M), downregulate Cx40 protein expression, in an in vivo setting (Figure 4A). We compared Cx40 protein levels in EdU⁺ and EdU⁻ aECs labeled with Cx40 immunostaining in Cx40CreER; Rosa26^TdTomato neonates, 4 days post-MI (Figure 4B through 4D). Both categories of artery cells, single and associated with lumenized artery vessels, were analyzed. Single aECs observed in ischemic watershed are derived from resident arteries and contribute to artery reassembly. We observed that Cx40 staining was significantly reduced in these EdU⁺ TdTomato⁺ proliferative single aECs (Figure 4B and 4C) when compared with nonproliferative single aECs or artery tips (Figure 4B). In contrast, EdU⁺ proliferating aECs, associated with well-formed arteries, did not show any significant change in their Cx40 expression (Figure 4D). Together, this suggests that Cx40 is downregulated in proliferating single aECs, upon MI.

We further checked if MI-induced single aECs express early markers such as Apj, which labels capillary/venous ECs from which arteries originate during coronary development via arterialization. We used Cx40^eGFP/+; ApjCreER; Rosa26^TdTomato mice to label arteries with GFP and capillary/venous ECs with TdTomato upon tamoxifen administration. Transgenic neonates were subjected to tamoxifen and MI at P2, to lineage label and trace ApjCreER⁺ cells (Figure 4E). Hearts were harvested at P6 when both single aECs and coronary collateral arteries are present (Figure 4E) and analyzed for Apj expression in preexisting aECs, upon MI. We identified single artery cells based on their location, that is, proximity to distal branches of arteries spanning the ischemic watershed area and their morphology, that is, single (not incorporated into any lumenized artery vessels) and spindle-shaped as observed earlier with Cx40CreER-lineage trace (Figure 4B). Indeed, many GFP⁺ single aECs expressed TdTomato, indicative of Apj expression at 4 days post-MI (Figure 4F and 4G). These GFP⁺ ECs were not only restricted to artery tips but also observed in watershed regions (data not shown). Thus, several aECs upon MI upregulate Apj expression indicating change in cell fate, in response to MI.

Together, our in vivo data suggest that single aECs exiting preexisting artery tips which are the drivers of artery reassembly, show reduced Cx40 expression, increased Apj expression, and proliferate during the course of neonatal artery reassembly. All 3 features, that is, lack of mature cell marker, increase in early angiogenic marker, and cell cycle reentry are indicative of dedifferentiation. Thus, we propose that preexisting aECs dedifferentiate and proliferate to build coronary collateral arteries during artery reassembly (Figure 4H).

We next checked if adult MI Cadherin5⁺ cycling aECs (Figure 2I) demonstrate any characteristics indicative of dedifferentiation. Although adult MI cycling aEC cluster was the least differentiated aEC cluster (Figure 5A and 5B), its CytoTRACE score was comparable to a mature artery cluster—aEC5 in MI (Figure 5B). Expression of proangiogenic marker Apln, and its receptor Aplnr were seemingly upregulated in this cluster, but there was no change in expression of Car4, Cx40, or Cxcr4 (Figure 5C). We analyzed the cellular and molecular features of neonatal cyc-aECs, which were indicative of dedifferentiation, and checked those indicators in adult cyc-aECs (Figure 5D). Some of the key dedifferentiation features absent in adult cyc-aECs were—(1) their adequate number, which is representative of their ability to self-renew/ reenter cell cycle, (2) loss of mature cell markers, and (3) gain of microvascular/ proangiogenic markers (shown in red, Figure 5D). For instance, in contrast to neonatal MI cyc-aECs, adult MI cyc-aECs, showed almost no change in the number of Cx40-expressing cells (Figure 5D). Adult cells expressing Cxcr4 in this group, increased by 1.76-fold (Figure 5D). The number of cells expressing Apln, Aplnr, and Car4 in the adult cyc-aECs was increased upon MI. However, these changes in cell numbers or gene expression levels, were statistically insignificant when compared between adult sham and MI groups, using Wilcoxon sum-rank test (Figure 5D; Table S1). Furthermore, cycling aEC cluster did not contribute to the predicted path associated with mature aECs (aEC1–7), even in MI group (Figure 5E and 5F). We confirmed the same with RNA velocity analyses (Figure S5H). Thus, unlike neonate, adult cycling aECs were neither connected to aECs in trajectory analyses (Figure 5E and 5F) nor were lineage traced from preexisting arteries (Figure 2J and 2K). Together, adult cycling aEC population fails to drive artery regeneration (Figure 5G) due to its limited number and inability to dedifferentiate (Figure 5D).

Preexisting aECs drive collateral artery formation in neonatal mice, only upon MI.^17,62 We isolated all genes from Cx40CreER lineage–traced TdTomato⁺ injured aECs (Figure 1F) and performed gene ontology enrichment analysis and visualization (GOrilla). This allowed us to identify biological processes triggered upon MI. Key vascular responses, such as angiogenesis, migration, and proliferation, were significantly upregulated upon MI (Figure 6A). Artery reassembly is a systematic step-wise cellular event where aEC migration and proliferation are crucial. Thus, our neonatal single-cell data set contains the aEC population, including those that drive artery reassembly, and subsequent neonatal cardiac regeneration.

aEC1, in neonatal MI data set, is the only aEC cluster associated with the predicted trajectory and connected with cycling aECs (Figure 3F). We performed pathway analyses on Cadherin5⁺ aEC1 cells using g:Profiler and filtered pathways from various databases, based on their statistical significance (P>0.05). Many pathways relevant to vascular development and injury response were enriched in aEC1 such as apelin, TNF, and Interleukin signaling⁶³ (Figure 6B). Vegfa-VegfR2 signaling (P=0.0011) was of particular interest (Figure 6B), as it has established molecular functions in mouse coronary artery development^48,64 and neovascularization upon injury in different tissues.^65,66

We checked the expression levels of ligands (Vegfa, Vegfc) and a major receptor (VegfR2 or Kdr) associated with coronary morphogenesis, in our data set. While Vegfa and Vegfc were expressed in aECs (aEC1–4), VegfR2 was specifically enriched in cycling aEC and trans-EC population (Figure 6C). The VegfR2 expression, was significantly different in cyc-aECs, when compared between neonatal sham and MI groups, using Wilcoxon sum-rank test (Table S1). When immunostained for VegfR2, the protein was located in coronary aECs of small arteries, artery tips, and capillary ECs (Figure S8A). The levels of VegfR2 protein in arteries were lower as compared to capillaries, and the protein was almost always absent in large arteries (Figure S8A). This could mean an overall reduced expression of arterial VegfR2 (as compared to capillaries) or expression of VegfR2 by a smaller subset of arterial ECs. Both EC types (arteries and capillaries) facilitate artery reassembly, either by directly participating in the process (aECs) or providing the Cxcl12 ligand (cap-ECs).¹⁷ Given VegfR2 expression in both, capillaries and small arteries, we hypothesized that Vegf pathway has a role in artery reassembly, post-MI.

We analyzed the role of arterial VegfR2 in artery reassembly, in vivo. Control (Cx40CreER; Rosa26^TdTomato) and aEC-specific, conditional VegfR2 knockout (VegfR2^L/L; Cx40CreER; Rosa26^TdTomato) neonates received tamoxifen at P0/P1 which allowed labeling of preexisting aECs. These hearts were subjected to MI at P2/P3 and hearts were harvested 4 days post-MI, at P6/P7 (Figure 6D). Whole heart imaging allowed us to visualize the entire arterial network of the heart (Figure 6E and 6F) as well as focus into regions of interest such as the watershed region where new collaterals are expected to appear after injury (Figure 6G through 6I). Control and VegfR2 artery-specific knockout hearts did not illustrate any gross differences in their baseline artery architecture, without MI (Figure S8B through S8D). Watershed regions from sham (data not shown) or no injury (Figure 6G) hearts, did not show any collateral arteries lineage labeled with TdTomato, suggesting absence of artery reassembly in nonischemic hearts. Upon MI, control hearts consistently showed several (average of 8) Cx40CreER lineage–traced (TdTomato⁺) collateral arteries as observed earlier¹⁷ (Figure 6E and 6H). In contrast, we rarely observed any such collateral arteries in VegfR2 artery-specific knockout hearts (Figure 6F, 6I, and 6J). This suggests an artery-specific role of VegfR2 in executing artery reassembly, upon MI.

We next investigated the cause for lack of TdTomato⁺ collateral arteries in VegfR2 knockout watershed regions. Artery reassembly is a 3-step cellular event—migration, proliferation, and coalescence, of aECs. We first assessed the ability of TdTomato⁺ cells to migrate into the watershed. Qualitative and quantitative observations suggested that the number of Cx40CreER lineage–traced single aECs, observed in the watershed regions, were comparable in control and knockout MI hearts (Figure 6H and 6I; Figure S8E through S8I). This suggests that in absence of arterial VegfR2, the migration of aECs from their resident arteries into the watershed region is less likely affected. We next investigated the proliferation step and injected neonates with EdU, 1 hour before harvesting the hearts. We then quantified the number of proliferating single aECs in the MI watershed regions. Interestingly, VegfR2 knockout watershed regions showed a ≈2.4-fold reduction in the number of proliferating single aECs as compared to control hearts (Figure 6K through 6M; Figure S9). Such reduction in number of proliferating single aECs was also observed in nonregenerative P7 injured hearts (Figure 2O). Thus, arterial VegfR2 regulates neonatal aEC proliferation, post-MI. A defect in VegfR2-mediated artery cell proliferation could also affect subsequent coalescence of TdTomato⁺ cells and successful execution of artery reassembly, as observed in VegfR2 knockout hearts (Figure 6J).

Functional coronary collateral arteries built by artery reassembly promote cardiac regeneration. Since arterial deletion of VegfR2 reduced number of collateral arteries, we tested if this neonatal VegfR2 deletion affects cardiac function in adults. We performed MIs at neonatal stage (P2) and checked several parameters for cardiac function 30 days post-MI. Although parameters like ejection fraction and fractional shortening did not change significantly (Figure 6N and 6O), cardiac output and stroke volume were significantly reduced (Figure 6P and 6Q) in VegfR2 arterial EC-specific knockouts. Thus, arterial VegfR2 facilitates neonatal artery reassembly that regulates cardiac function in adult mice.

We next investigated if Vegf ligands and receptors are expressed in adult ECs. The expression of Vegf ligands, Vegfa and Vegfc, were mostly observed in mature aEC clusters (Figure S10A) like in neonatal data (Figure 6C). The receptor VegfR2 (Kdr), was upregulated in capillary ECs, as expected (Figure S10A). Interestingly, at least at the mRNA level, VegfR2 was downregulated in ≈75% of actively cycling aECs (Figure S10A). This is in contrast to the upregulated expression of VegfR2 in neonatal cyc-aECs (Figure 6C). When compared with sham group, VegfR2 was significantly downregulated in adult MI cyc-aEC cluster (Table S1). Given the functional significance of arterial VegfR2 in neonatal aEC proliferation (Figure 6K through 6M), the lack of VegfR2 expression in residual adult cycling aECs, could explain depleted potential for aEC proliferation in injured adults. In summary, our data indicate that arterial VegfR2 drives, specifically, the proliferation step of artery reassembly in injured neonatal mice (Figure S10B).

New cardiomyocytes originate from preexisting cardiomyocytes upon injury^15,16 by dedifferentiating and reentering cell cycle.^67,68 Cardiomyocyte dedifferentiation has been elucidated upon adult zebrafish cardiac injury via disassembly of sarcomeric structures, detachment of cardiomyocytes from one another, and expression of early embryonic genes (αSMA, embryonic myosin heavy chain, and Gata4).^67–69 Post-MI adult cardiomyocytes also demonstrate somewhat limited dedifferentiation potential indicated by downregulation of mature cardiomyocyte functions such as hypertrophy, contractility, electrical conduction, and upregulation of several early genes like αSMA, Gata4, Runx1, and Dab2.⁷⁰ Unlike cardiomyocytes, EC dedifferentiation has not been reported in any cardiac injury model. We show here, for the first time, that neonatal mouse coronary aECs undergo dedifferentiation upon MI; a process limited in adults.

Since there are no objective indicators of aEC pluripotency, one can argue that changes in gene expression or phenotypes could be a reflection of their plasticity and not necessarily reversion of cell fate into a dedifferentiated state. It is also possible that the number of dedifferentiated cells observed in cyc-aEC and trans-EC clusters of uninjured hearts simply increases upon MI. To address this, we combine single-cell analyses and genetic lineage tracing. Lineage-labeled noncycling aECs in MI hearts can be sequentially traced into cycling single aECs in the watershed, and consequently into noncycling aECs on the coronary collaterals. We analyzed the transiently cycling aECs and show that several indicators for dedifferentiation were enhanced in neonates. One such indicator was reduced expression of Cx40 in cyc-aECs upon MI. This expression changed minimally in sham cyc-aEC group when compared with sham mature aECs. This could be an indication of their low cycling potential due to lack of appropriate ischemia-induced molecular signals which drive dedifferentiation and artery reassembly. A second indicator was presence of Apj lineage in Cx40⁺ aECs. It is possible that Apj lineage–traced capillaries differentiate into aECs, a phenomenon called arterialization already shown to contribute (though minimally) to collateral formation by artery reassembly.¹⁷ Our in vivo data cannot distinguish between differentiation of capillary ECs into aECs and upregulation of capillary marker (Apj) in arteries. However, the unique location and morphology of cells allowed us to confidently identify Cx40⁺ single aECs and conclude that at least a subpopulation of aECs, upon MI, express Apj—an indicative of vascular dedifferentiation.

The dedifferentiation indicators were absent in adults. This finding positively correlated with the ability of aECs to reenter cell cycle. Neonates had much higher number of cycling aECs than P7 or adults. This finding was verified via 2 independent methods—single-cell gene expression analyses of cyc-aECs and presence of Cx40CreER⁺ EdU⁺ cells in an in vivo cardiac MI model. Thus, lack of dedifferentiation limits the number of cycling aECs in nonregenerative hearts and consequently affects their ability to form collateral arteries by artery reassembly.

Several molecular pathways relevant to vascular development regulate EC response to injury.⁷¹ For instance, Notch,⁷² Apelin,⁵⁵ Cxcl12^73–76 signaling, in mice, regulate different aspects of vascular development and remodeling during cardiac morphogenesis; ranging from EC fate determination to regulation of different EC responses such as migration, proliferation, and tube formation. Interestingly, some of these pathways are also associated with EC regeneration upon injury,^17,58,77 making them potential therapeutic candidates for treating vascular conditions in human patients.⁷⁸ Further studies are required, and will elucidate if (and how) the cellular and molecular mechanisms associated with these signaling pathways, during development and injury responses, are similar.

Vegf signaling drives coronary development in zebrafish and mouse hearts. In mice, Vegf signaling drives migration of coronary endothelial^53,79 and endo-ECs⁶⁴ in developing hearts, and proliferation of coronary ECs in heart explant cultures.⁸⁰ In zebrafish, vegfa mutants show reduced EC proliferation possibly via PI3K signaling.⁸¹ Upon cardiac injury, endocardial Vegfa signaling⁷⁷ and Vegfc-mediated (Cxcl8a/Cxcr1–driven) coronary EC proliferation⁸² execute revascularization and cardiac regeneration of ischemic hearts. In this study, we highlight the role of VegfR2-mediated signaling in aEC response triggered by neonatal cardiac injury. We show that VegfR2 in neonates is expressed in coronary aECs, deletion of which, significantly reduced the number of collateral arteries formed via artery reassembly.

Our data show that artery tips do not proliferate. Proliferation is rather demonstrated by single aECs which originate from these artery tips. This suggests that migration of preexisting aECs could precede aEC proliferation in the watershed. Appearance of VegfR2 depleted single aECs in ischemic watershed regions suggested that aEC migration was not affected. In fact, the TdTomato+ cells were increased in the VegfR2-depleted hearts as compared to wild-type hearts. This could be because TdTomato+ single aECs accumulated in the watershed due to inability to coalesce and form collateral arteries. VegfR2-depleted single aECs showed reduced proliferation potential, which could account for lack of lineage-traced collateral arteries in the knockout hearts. Proliferation of aECs is followed by their coalescence into collateral arteries. Thus, we cannot exclude the possible role of Vegf in coalescence of artery cells, or the vascular steps that follow, like lumenization or maturation of collateral arteries. It is also possible that because of reduced rate of arterial coalescence, VegfR2 null saECs reduce their proliferation rate as a feedback mechanism. This is also evident from higher number of VegfR2 null saECs present in post-MI watershed.

The precise source of Vegf ligand in injured mouse hearts could be the endocardium,⁸³ myocardium (Vegfa),^64,84 epicardium (Vegfc⁵³ or Vegfa⁸⁵), or cardiac ECs,⁸² thus suggesting an autocrine or paracrine signaling as observed in injured zebrafish hearts.⁸² In our neonatal mouse single-cell data set, we found both Vegfa and Vegfc being expressed by mature aECs (aEC1–4), but at lower levels by cyc-aECs (or by trans-ECs, SMCs, endo-MT ECs, endo ECs), indicating a possible cell autonomous mechanism. However, an alternate source of Vegf ligand, such as cardiomyocytes, fibroblasts, or immune cells, which are excluded from our data set, cannot be neglected.

Single nucleotide polymorphisms in both introns and exons of KDR (murine VegfR2) are strongly associated with coronary heart disease (CHD).⁸⁶ These single nucleotide polymorphisms can either modulate the affinity to ligand or alter E2F binding to KDR promoter, subsequently, affecting KDR activity and downstream signaling.⁸⁶ Elevated levels of circulating VEGF ligands are positively and significantly associated with CHD,⁸⁷ coronary artery disease,^88,89 and coronary artery lesions.^90,91 Several single nucleotide polymorphisms affect VEGF levels^87,89 and may account for the variation observed in revascularization abilities in individuals across different genetic backgrounds. At least 3 such single nucleotide polymorphisms in VEGFA are strongly associated with increased risk of CHD/MI⁹² and one with coronary collateral vessel perfusion.⁹³ How the association of VEGF/KDR signaling axis with CHD may serve for better prognosis or indicate the need for vascularization, remains unclear.

Preclinical studies in animals with sparse⁹⁴ to no⁴⁹ preexisting collateral arteries have consistently elucidated role of Vegf in ischemia-induced collateral formation. On the contrary, numerous past and ongoing clinical trials with VEGF alone, or combined with proangiogenic/regenerative factors, have been concluded with little to no success (data available at clinicaltrials.gov).^94,95 The delivery methods appear to be safe, but the angiogenic effects, particularly the efficacy of VEGF gene therapies in stimulating reperfusion via collateral arteries are minimal.⁹⁴ Mechanistic details of Vegf’s mode of action, in orchestrating vascular cell responses that lead to collateral formation, could aid designing clinical trials with higher efficacies.

We show that mouse coronary ECs in MI hearts dedifferentiate and then expand in VegfR2-dependent manner. The mere presence of a significant number of single aECs in a limited watershed area could be sufficient to create new arteries.³⁹ Thus, it will be interesting to see if this threshold density of aECs is accomplished by Vegf-mediated proliferation. Similarly, identifying the molecular drivers of EC dedifferentiation is also crucial, and when combined with Vegf, could promote collaterals in individuals with CHD and coronary artery disease.

Figure S1-10 and legends

Table S1

Major Resources Table

Graphical abstract