Uncovering the molecular identity of cardiosphere-derived cells (CDCs) by single-cell RNA sequencing

Human primary cells (CDCs, AFs, CFs, SMCs, ECs) were generated from biopsies derived from adipose, atrial or vessel tissue from patients undergoing heart surgery (age: 5 days to 76 years) (Suppl Table 1 and 2). Human right atrial appendage tissue for single nuclei (sn) and single cell (sc) RNA sequencing (RNAseq) was also obtained from patients undergoing heart surgery (Suppl Table 1, Suppl Table 9). All patients had signed an informed consent. In the case of infants, their parents or legal guardians signed the informed consents. The local ethics committee of the Technical University of Munich Medical School supervised and approved the study (project number 570/16S). Tissue sampling within the framework of the cardiovascular biobank at the German Heart Center Munich was also approved by the local ethics committee of the Medical School of the Technical University of Munich (project number 5943/13). All experimental procedures were performed in accordance with the principles outlined in the Declaration of Helsinki.

Murine cardiac fibroblasts (CFs) were generated from adult hearts of transgenic Nkx2.5 cardiac enhancer eGFP mice [68]. The mice were anesthetized with isoflurane and then euthanized by cervical dislocation to extract their hearts. Neonatal rat cardiomyocytes (NRCMs) were isolated from 0–1 day old Sprague Dawley rats after decapitation.

Mice and rats were housed in accredited facilities in compliance with the European Community Directive related to laboratory animal protection (2010/63/EU). All animals sacrificed for harvesting organs were approved by the relevant authority “Regierung von Oberbayern” [Regional Government of Upper Bavaria], German TierSchG (Animal protection law). All animal experiments (organ extractions) were performed in accordance with the European guidelines and regulations for animal care and handling (Directive 2010/63/EU).

Cardiosphere-derived cells (CDCs) were generated from adult right atrial appendage tissue from patients undergoing cardiac surgery (Suppl Table 1, mean age: 63.92 years ± 8.23 years). CDCs were established by the production of “3D-cardiospheres” including stimulation with growth factors (Fig. 1A) [30, 43]. To assess the CDCs’ distinct molecular profile, they were compared to cardiac fibroblasts (CFs), smooth muscle cells (SMCs), endothelial cells (ECs), and adipose tissue-derived fibroblasts (AFs) as a non-cardiac cell type. CFs and AFs were isolated from cardiac or subcutaneous adipose tissue, respectively (Fig. 1A; Suppl Table 1). ECs and SMCs were generated from thoracic vessels of patients undergoing coronary artery bypass graft (CABG) surgery (Fig. 1A; Suppl Table 1). Cardiac progenitor cells derived from differentiated human induced pluripotent stem cells (hiPSCs [57]) on day six (DIFF D6) and immature cardiomyocytes on day eight (DIFF D8) were further used as controls for the expression of cardiac transcription factors [7] (Suppl Fig. S1A).

First, gene expression levels of several markers were assessed by qRT-PCR (Fig. 1B–F; Suppl Fig. S1B–E). Cardiac transcription factors such as TBX5, NKX2-5 and GATA4 have been described to be highly expressed in CDCs compared to fibroblasts [30]. Here, GATA4, TBX5 and NKX2-5 [22, 24, 27] were significantly upregulated in DIFF D6 and DIFF D8 (only TBX5 and NKX2-5) compared to CDCs. No significant differences were detected between CDCs and CFs (Fig. 1B, Suppl Fig. S1B). GATA4 was albeit significantly lower expressed in AFs, SMCs and ECs compared to CDCs. CF markers DDR2 and PDGFRA [19] or cardiac fibrosis associated microRNA (miR)-21 [49, 62] were not significantly higher expressed in CFs compared to CDCs, whereas CF markers ALDH1A2 [18] and THY1 (CD90) [19] were significantly increased in CFs compared to CDCs (Suppl Fig. S1C, Fig. 1C). Well-known SMC markers TAGLN [3] and PDGFRB [42] were also expressed in fibroblasts and CDCs (Fig. 1D; Suppl Fig. S1D). ECs were clearly distinguishable from all other cell types by high expression of typical EC markers PECAM1 (CD31) and CDH5 (Suppl Fig. S1E) [18]. Interestingly, CDCs exhibited significantly higher CDH5 expression than AFs, CFs and SMCs. MiR-146a, described as a marker for CDC-derived exosomes and as partially responsible for their beneficial effects [28, 64], was threefold higher expressed in CDCs compared to CFs but did not reach significance (Fig. 1E). MiR-132, enriched in EVs derived from cardiac outgrowth cells [4], was likewise not upregulated in CDCs versus CFs (Fig. 1E). The mesenchymal marker ENG (CD105), a quality marker for CDCs [41], was equally expressed in all non-myocyte cells and CDCs but was downregulated in DIFF D6 and DIFF D8 (Fig. 1F).

Immunocytochemical stainings or flow cytometry evaluated protein levels of selected markers. Whereas DDR2 was ubiquitously abundant in all analyzed cell types (Suppl Fig. S1F), flow cytometry revealed that only 40–60% of the CDCs expressed CD90 (Fig. 1G–I). Smooth-muscle actin alpha (α-SMA, ACTA2), a SMC marker [69], was expressed to a comparable level in fibroblasts and CDCs (Fig. 1J). High abundance of CD31 (PECAM1) in ECs was confirmed by immunocytochemistry and flow cytometry (Suppl Fig. S1G–I). CD105 was found to be expressed ubiquitously in all analyzed cell types (Fig. 1K, L). Flow cytometry (Suppl Fig. S1J) confirmed the absence of the hematopoietic marker CD45 in CDCs [41] and for all non-myocyte cell types.

To conclude, CDCs showed similar molecular characteristics like non-myocyte cell types. However, CDCs clearly differed from veritable cardiac progenitor cells (DIFF D6) and immature cardiomyocytes (DIFF D8). Well-described CDC markers, such as CD105, were equally expressed in AFs, CFs, SMCs, and ECs.

We next assessed the differences between CDCs and the three main cardiac non-myocyte cell types CFs, SMCs and ECs in more detail using sc-RNAseq. All cells were generated from discarded tissue during CABG surgery as mentioned above (Suppl Table 5, age: 61–66 years; Fig. 1A). As a quality control, we carefully checked cell morphology (Suppl Fig. S2A) and some of the above-mentioned specific markers (Suppl Fig. S2B, C). Single-cell transcriptional profiling was performed using the 10 × chromium platform, followed by bioinformatical analyses with the Seurat software suite. To assess transcriptional differences between cell types single-cell data were integrated and canonical correlation analysis (CCA) was performed [8] to remove potential batch effects. To remove low-quality cells and doublets, cells with very high mitochondrial gene percentage were filtered out [38] (Suppl Fig. S2D–F, Suppl Table 6). Finally, 2815 cells were analyzed with a median of 3428 genes per cell and 17,078 UMI counts per cell (Suppl Fig. S2G, H, Suppl Table 7).

CDCs exhibited a higher median percentage of mitochondrial genes (6%) compared to CFs (1.3%), SMCs (2.5%) and ECs (3.0%) (Suppl Fig. S2F, Suppl Table 7) pointing to a mitochondria-rich cell type with high energy needs.

Subsequently, we performed unsupervised clustering using Seurat [8, 39, 52]. Figure 2A depicts the cell types according to their cell identity. For CDCs, CFs, SMCs and ECs, the top ten of upregulated genes were identified (Fig. 2B). Among those cell types, CDCs exclusively expressed genes encoding for chemokine ligands (such as CXCL1/8/6) or cytokines (IL1B and CSF3). CXCL6/8 as well as IL1B expression was specific for CDCs compared to the other non-myocyte cell types (Suppl Fig. S2I, S3A). Only CXCL1 was also expressed in a few ECs (Suppl Fig. S2I). FBLN2, S100A4, ACTA2 and TAGLN were not specific for CFs, but also expressed in SMCs and partly in CDCs (not FBLN2) (Fig. 2B, Suppl Fig. S3B, C). Many genes highly expressed in SMCs were also expressed in CFs and CDCs (Suppl Fig. S3D). The most upregulated genes in ECs IFI27, CLDN5, and PECAM1 (CD31) were specific for ECs (Fig. 2B; Suppl Fig. S3E, F). Looking at these 40 markers in Fig. 2B CDCs express 17 markers in common with CFs (42.5%), 18 markers in common with SMCs (45%) but only 10 markers in common with ECs (25%).

Sc-RNAseq expression profiles of markers already analyzed by qRT-PCR (Fig. 1; Suppl Fig. S1) corresponded to previous results (Suppl Fig. S4A, B). The reliability of sc-RNAseq data was confirmed by validating selected gene expression by qRT-PCR in independent samples (Suppl Fig. S4C–E).

Unsupervised global subclustering of CFs, SMCs, ECs, and CDCs according to their specific gene expression subdivided them into eleven distinct clusters (Fig. 2C). Visually, all cell types split up into two main clusters sometimes with further sub-clusters, especially for CDCs, but also ECs (CDCs: Cl 3, 4, 5, 7, 8; CFs: Cl 2, 9; ECs: Cl 1, 6, 10; SMCs: Cl 0, 4, 5) (Fig. 2C).

To analyze the biological features of the clustered cells, we performed gene set enrichment analysis (GSEA) [56] of the selectively upregulated differentially expressed genes (uDEGs) in each cluster (Suppl Table S8). The main clusters for SMCs (Cl 0, 90% of SMC), ECs (Cl 1, 61% of EC) and CFs (Cl 2, 87% of CF) revealed biological properties such as angiogenesis (Cl0, Cl1), extracellular matrix organization (Cl0, Cl2), locomotion (Cl0,Cl1), or adhesion (Cl0, Cl2) and by this confirmed the identity of selected cell samples for sc-RNAseq (Suppl Fig. S4F, Suppl Fig. S5A, B, Suppl Table S8).

To better understand CDCs, we went into GSEA of all five CDC clusters. All clusters exhibited main immunomodulatory properties such as “response to cytokine” (Fig. 2D, Suppl Fig. S5C). The major difference between the two main clusters Cl4/7/3 and Cl5/8 was that the CDCs in Cl5/8 seemed to be involved in more energetic processes (“mitochondrion”, “cellular respiration”, Fig. 2D, Suppl Table S8), corresponding to the fact that CDCs exhibited a higher median percentage of mitochondrial genes than CFs, ECs and SMCs as noticed above (Fig. S2F). Energy-consuming processes for Cl5-CDCs (20% of CDCs) might correspond to “cell activation” while the energy in Cl8-CDCs (14% of CDCs) seemed to be necessary for cytoskeleton organization (Fig. 2D, Suppl Table S8). Cl3, Cl5 and Cl7 were enriched for processes associated with the immune system and “secretion” (Fig. 2D, Suppl Table S8). To elucidate which molecules might be secreted by CDCs we analyzed upregulated genes in Cl3, Cl5 and Cl7 in comparison to the other clusters in Fig. 2C. Several genes in this list encode for proteins known to be secreted by CDCs, for example VEGFA, VEGFB and FGF2 [37]. Interestingly, we observed many genes crucial for angiogenesis, such as TAGLN2, ICAM1, FGF2, VEGFA, and VEGFB. Besides, several chemokines (CXCL6, CXCL1, CXCL8, CXCL3, CXCL5, CXCL2), interleukins (IL1B, IL1A, IL24, IL6, IL32, IL6ST, IL11, IL1R1) and TGFB1/TGFB2 were found, suggesting that CDCs secrete immunomodulatory molecules.

Cl4-CDCs (24% of CDCs) showed “negative regulation of the cell-cycle” (Fig. 2D, Suppl Table S8) and thus might be quiescent cells. Cl7-CDCs (15% of CDCs) included cells that were still proliferative but also started catabolic processes (Fig. 2D, Suppl Table S8). Cl3-CDCs (26% of CDCs) were activated cells, able to migrate (“cell locomotion”), and involved in cell signaling as well as in developmental processes (Fig. 2D, Suppl Table S8).

To better understand whether significant similarities between CDCs and CFs, ECs or SMCs persist, cell similarity was calculated [65]. Three peaks appeared in the CDC sample (Fig. 2E): Whereas the highest peak stood out, the second-highest peak had approximately the same similarity score as CFs and SMCs, and the smallest peak paralleled the small peak of the SMC curve. Similarly upregulated genes in CDCs, CFs and SMCs were e.g. associated with GO terms such as “extracellular matrix” and “biological adhesion” (Suppl Fig. S6A). Associated exemplary genes COL6A2 and COL3A1 were expressed in CDCs, CFs and SMCs, but barely in ECs (Suppl Fig. S6B, C). qRT-PCR confirmed sc-RNAseq expression patterns in independent samples (Suppl Fig. S6D, E).

So far, sc-RNAseq results showed that CDCs exhibited the highest amount of mitochondrial genes and were distinguished from CFs, SMCs and ECs by secretory and immunomodulatory characteristics. However, CDCs also showed certain similarities to non-myocyte cell types.

To compare CDCs with veritable cardiac progenitor cells (CPCs) and early cardiomyocytes (CMs) in more detail we included single-cell transcriptome data of differentiating human ESCs (hESCs) generated for a trajectory mapping of lineage decisions during hESC differentiation [44]. A similar “cardiac” differentiation protocol was used for the hESCs [44] like for hiPSCs described in Fig. 1 and Suppl Fig. S1. Sc-RNAseq data of CDCs and hESC DIFF D6/D7/D8/D9/D15 were integrated and UMAP plots were generated (Fig. 2F, G). Eight clusters emerged by unsupervised clustering (Fig. 2G). As we assumed before (Fig. 1, Suppl Fig. S1) CDCs did not overlap with CPCs or early CMs clearly marked by TBX5, GATA4 and NKX2-5 (cluster 3 associated with heart development) (Fig. 2F–H, Suppl Fig. S6F). In addition, there is no overlap of CDCs with epithelial (cluster 5) or angiogeneic progenitor cells (cluster 7) (Fig. 2F, G, Suppl Fig. S6F–G). Interestingly, these progenitor cells expressed c-KIT (Suppl Fig. S6G), which was completely absent in CDCs. However, CDCs overlap with differentiating hESCs in clusters, that show CF- (cluster 2) or proliferating cell (cluster 0, 6) characteristics (Fig. 2F, G, Suppl Fig. S6F, H). Cluster 1 and cluster 4 mainly consist of CDCs and show enrichment for GO terms such as “cell migration”, “anchoring junction”, “translation elongation” and “ribosome” (Fig. 2F, G, Suppl Fig. S6F). Interestingly, earlier defined markers for CDCs such as CXCL1, CXCL6 and IL1B (Fig. 2B) were highly specific for CDCs in this setting (Fig. 2I).

In conclusion, no overlap was found with cardiac progenitor cells or early cardiomyocytes by comparing sc-RNAseq data of CDCs to differentiating hESCs.

Next, we sought to elucidate the originating cell population of CDCs in the adult human heart. Therefore, we used right atrial human biopsies from four different individuals and performed either single nuclei (RA-1, RA-2) or single cell (RA-3, RA-4) RNAseq on them (Suppl Table 9). Single nuclei (sn) RNAseq data were generated before [34] and were included to analyze cardiomyocytes (CMs) which cannot be investigated with our protocol for sc-RNAseq due to methodological reasons (size limitation of 10X single cell controller ~ 40 µm). Sn- and sc-RNAseq biopsy data were integrated with adult CDC sc-RNAseq data and UMAP plots were generated. UMAP places related cell types near one another [9]. First of all, single cells and single nuclei clustered nicely together, so that we were able to identify all major cardiac cell types in the four biopsy samples: CMs, CFs, ECs, SMCs/pericytes (PCs), monocytes/macrophages (MPs) and B-/T-cells (Fig. 3A–C, Suppl Fig. S7A). In addition, several small cell populations were identified (Fig. 3B, Suppl Fig. S7B). The new CDC markers CXCL1/6/8 and IL1B (Fig. 2B) were also expressed in macrophages, dendritic cells and mast cells (except CXCL6) (Fig. 3C). Interestingly, CDCs were located in immediate proximity to CFs (Fig. 3A, B, Suppl Fig. S7A). 77% of CDCs even share the CF-1 cluster. 23% of CDCs build their own cluster (Cl7: CDC-2). CMs were situated far away from CDCs and ECs and pericytes/SMCs were located at bigger distances to CDCs than the CF populations.

Thus, we supposed that CDCs might be a cell-state transition of CFs. To test this, we sought to use single-cell trajectory analysis. We used monocle, an algorithm that applies reversed graph embedding to describe multiple fate decisions in a fully unsupervised manner [48]. Cells will be arranged along reconstructed “trajectories” of any type of inferred biological transition [48]. These trajectories describe gene expression changes that occur while a cell proceeds through the biological process under study [48]. In the two clusters that contain CDCs (Cl7: CDC and Cl2: CF-1/CDC) we detected three trajectory branches (numbered 1, 2 and 3) (Fig. 3D). CDCs and CFs build three major trajectories where CDCs are at the one end and CFs at the other end of the pseudospace. To understand the trajectory branches we investigated the associated genes. For this, we performed additional subcluster analyses in Seurat and analyzed the Top 25 specific genes in cluster 0 (mapping to trajectory 1), 7 (mapping to trajectory 2) and 12 (mapping trajectory 3) (Suppl Table S10, Suppl Fig. S7D, E, Fig. 3E). GSEA revealed that trajectory 1 was mainly associated with “External encapsulating structure” and ECM organization. Trajectory 2 was related to “locomotion” and “development”. Trajectory 3, the “CDC-trajectory”, was linked to “response to cytokine”, immune system processes and secretion (Fig. 3D, Suppl Fig. S7E).

To further reveal whether CDCs constitute a transcriptional transition state of CFs we finally performed RNA velocity analysis, a derivative corresponding to the gene expression state, to predict the potential directionality of cell state transitions [25, 33]. RNA velocity describes the rate of gene expression change for an individual gene at a given time point based on the ratio of its spliced (mature) and unspliced (native) mRNA [5, 33]. We used Velocyto that is based on transcriptional dynamics and accounts for the direction of motion [5, 33]. RNA velocity analysis is not feasible for sn-RNAseq in comparison with scRNAseq data since nuclei still contain unspliced pre-mRNA molecules including introns. Thus, we had to exclude sn-RNAseq-samples (RA-1, RA-2) and compared the CDCs to the sc-RNAseq data of two adult right atrial biopsies (RA-3, RA-4). Usually, RNA velocity analysis is applied to model developmental processes, such as neurogenesis [33]. However, to prove the assumption that such an algorithm can also be used to analyze the origin of cultivated cells in their originating biopsy we analyzed the cultivated CFs, SMCs and ECs (analyzed in detail in Fig. 2) together with the data of the atrial biopsies (RA-1—RA-4). The clustering and the identification of the major cell types for this analysis are displayed in Suppl Fig. S8A–D. Next, we analyzed single-cell trajectories and found, similar to the CDCs, that cultivated and atrial CFs (CF-3/Cl7: mainly cultivated CFs), as well as cultivated and atrial ECs (EC-2/Cl5: mainly cultivated ECs) were arranged at opposing ends of trajectorial pseudospaces (Suppl Fig. S9A). We further analyzed the trajectory branches according to their associated top-specific DEGs. For this, we performed an additional subcluster analysis in Seurat (Suppl Fig. S9B). Subcluster 12 mapped to trajectory 1, subclsuter 0 to trajectory 2, etc. We found reasonable gene expression and GO terms for CF trajectories (Tr 1, 2, 3) as well as EC trajectories (Tr 8, 9) (Suppl Fig. S9C, D). CF trajectories 1 and 2 are highly similar to the trajectory branches 1 and 2 in Fig. 3D. Tr 3 and Tr 9 are the trajectory end points of the cultivated CFs and ECs, respectively, exhibiting terms such as “collagen containing extracellular matrix” (CF Tr 3) or “cell cycle” (EC Tr 9) (Suppl Fig. S9D). Finally, we analyzed RNA velocity by Velocyto in this setting. We found that CFs and ECs were each linked to their atrial equivalent (atrial CFs (CF-1/2), atrial ECs (EC-1)) by velocity streamlines (arrows in Suppl Fig. 9E). The directions of the arrows pointed from the cultivated cells to the atrial cells (Suppl Fig. S9E). Then, by analyzing CDCs and atrial samples, we found that CDCs were directly linked to the CF-1 population by velocity streamlines (Fig. 3F, arrows) indicating that CDCs might originate from CFs. The bulk part of CDCs (77% of CDCs in Cluster CDC-1/CF-1) seemed to act like immature CFs since the direction of the arrows points from CDCs to CFs. The smaller part of CDCs (23% of CDCs in the cluster CDC-2) seemed to develop more into the direction of ECs (arrows from CDC-2 point into the direction of EC), even if they are still far away from those.

Preclinical studies with human CDCs derived from neonatal or adult atrial appendages revealed that neonatal CDCs had stronger repair abilities than adult CDCs after transplantation in immunodeficient infarcted rat hearts. Neonatal CDCs maintained myocardial function, prevented adverse remodeling and promoted angiogenesis [53]. We sought to elucidate whether this might be attributable to certain molecular characteristics of “infant” CDCs (age: 5 days—5 years, Suppl Table 2) compared to “adult” CDCs (age: 55–76 years, Suppl Table 1).

Interestingly, gene expression analysis showed lower expression of the cardiac transcription factors GATA4 and TBX5 in infant CDCs compared to adult CDCs, and infant CFs, respectively (Fig. 4A, left and middle panel). In contrast, NKX2-5 was significantly higher expressed in the infant group (CDCs and CFs) (Fig. 4A, right panel). CF marker ALDH1A2 was upregulated in adult cells compared to infant cells, and CFs compared to CDCs in both age groups (Fig. 4B, left panel). S100A4 was also upregulated in CFs versus CDCs in both age groups (Fig. 4B, middle panel). Comparing corresponding cell types, THY1 (CD90) expression was interestingly higher in the infant group (Fig. 4B, right panel). CXCL6 was specific for CDCs in both age groups while miR-146a proved to be specific in CDCs compared to CFs in the infant group only (Fig. 4C).

To analyze molecular differences of infant and adult CDCs in more detail, a neonatal CDC sample (patient age: 7 days) was compared to the adult single-cell CDC sample (patient age: 61 years, see also Fig. 2) by sc-RNAseq. Quality control of infant and adult CDCs was performed by morphological evaluation and ALDH1A2 (specific CF-marker) expression compared to CFs generated from the same patient (Suppl Fig. S10A, B). After low-quality cells were filtered out (Suppl Fig. S10C–E, Suppl Table 6), 2069 cells were analyzed with a median of 5649 genes and 33,731 UMI counts per cell (Suppl Fig. S10F, G). As already noticed for adult CDCs, infant CDCs also exhibited a high amount of mitochondrial genes (Suppl Fig. S10E, Suppl Table S7).

Unsupervised global subclustering of the two CDC-samples (infant and adult, Fig. 4D) according to their specific gene expression subdivided the CDCs into six clusters (Fig. 4E, Suppl Table 12). Mainly two clusters (0, 3) overlapped between the infant and adult sample (Fig. 4F). The cells in these clusters were enriched for GO-terms such as adhesion, cellular respiration, cell cycle and catabolic processes (Fig. 4F). The small clusters 4 and 5 mainly consisted of adult CDCs and corresponded to biological processes such as “response to cytokine”, secretion but also adhesion, cellular respiration, cell cycle and catabolic processes equally to cluster 0 and 3 (Fig. 4F). The high similarity between infant and adult CDC samples was confirmed by calculating cell similarity [65] (Fig. 4G) and CXCL6 was again found to be expressed in all CDCs (Suppl Fig. 10H). However, cluster 1 (56% of infant CDCs) was specific for infant CDCs while cluster 2 (33% of adult CDCs) was only observed in adult CDCs (Fig. 4D, E, Suppl Table 12). Interestingly, GSEA revealed several similar GO terms for both clusters (Suppl Table 13, Fig. 4H). Infant as well as adult CDCs from cluster 1 and 2 were involved into angiogenesis, extracellular matrix organization, adhesion, cell cycle processes, cell activation, cell motility and developmental processes. The main differences between cluster 1 (infant CDCs) and cluster 2 (adult CDCs) were immunomodulatory properties, “response to cytokine” and secretion for adult CDCs while only infant CDCs exhibited GO terms related to cardiac development (Fig. 4H). Exemplary violin and gene expression plots (WNT5A, CSF3), as well as heat maps confirmed those results (Suppl Fig. S10I–L).

In summary, sc-RNAseq analysis revealed cardiac developmental processes in neonatal CDCs while adult CDCs were more involved in immunomodulation and secretion.

To analyze paracrine effects of CDCs such as pro-angiogenic and anti-fibrotic effects or inhibition of cardiomyocyte apoptosis [28, 35, 64], we evaluated the impact of CDC-EVs on different cardiac cell types. Effects of CDC-EVs were directly compared to CF-EVs. The comparison of infant and adult CDC- and CF-EVs further allowed us to assess age-dependent effects.

EVs were isolated from serum-free medium conditioned by CDCs or CFs for seven days (Suppl Fig. S11A). Fewer than 5% of dead cells were detected after one week in serum-free medium for CDCs and CFs (Suppl Fig. S11B). Although no significant difference of cell numbers of CDCs and CFs per flask was observed (Suppl Fig. S11C), nanoparticle tracking revealed that CDCs secreted a significant higher particle number compared to CFs (Suppl Fig. S11D) corresponding to previous results of GSEA (GO terms associated with secretion, Suppl Table 8). Most EVs of CDCs and CFs from infants and adults ranged in size from 40 to 200 nm (mean: 140 nm), indicating that mainly exosomes and not apoptotic bodies were isolated [26, 50] (Suppl Fig. S11E, F). Presence of typical exosomal surface markers CD63 and CD81 [61] was confirmed by flow cytometry of selected CDC-EV preparations (Suppl Fig. S11G). MiR-146a was upregulated in CDC-EVs compared to CF-EVs both in infant and adult samples (Suppl Fig. S11H), whereas neither miR-132 nor miR-21 showed significant differences between cell types or age groups (Suppl Fig. S11I, J).

Functional effects of CDC- and CF-EVs on various cardiac cell types were assessed using well established in vitro assays. For all assays, cells were either seeded in their normal growth medium (positive control, PosCtr), serum-free medium (negative control, NegCtr) or serum-free medium supplemented with infant or adult CDC- or CF-EVs.

To study angiogenesis in vitro, tube formation assays on matrigel and migration assays with primary human ECs were utilized (Fig. 5A–I).

The tube formation assay is based on the ability of ECs to differentiate into tube-like structures on matrigel [47], an important prerequisite for vessel formation [15]. Primary ECs were seeded on matrigel and an ImageJ angiogenesis analyzer tool [1, 10] assessed the extent of tube formation on the next day (Fig. 5A). Performing the assay with infant and adult CDC- and CF-EVs showed that EC tube formation was significantly promoted by infant CDC-EVs, but not CF-EVs (Fig. 5B–E). Adult CDC-EVs tended to perform in a similar manner than infant CDC-EVs but did not reach significance.

The ability of ECs to migrate constitutes another important angiogenic factor (scratch assay). After EC proliferation was prevented by mitomycin C, a scratch was conducted in the EC layer. The closure of the scratch was observed over 48 h (Fig. 5G). Comparing the effects of infant and adult CDC- and CF-EVs, only infant CDC-EVs significantly promoted EC migration (Fig. 5H–J, Suppl Videos V1–6). However, adult CDC-EVs acted similar than infant CDC-EVs but did not reach significance.

Notably, in both EC assays (tube formation and scratch assay), no significant difference between CDC-EVs and CF-EVs was observed.

Several studies reported the ability of CDC-EVs to decrease cardiomyocyte apoptosis [35, 64]. Gene expression of the death surface receptor Fas [66] and the pro-apoptotic Bax assessed the anti-apoptotic effect of CDC- and CF-EVs on neonatal rat cardiomyocytes (NRCMs). Quality of NRCMs was demonstrated by the expression of cardiomyocyte markers (Suppl Fig. S12A, B). Only beating NRCMs were used for apoptosis assays (Suppl Video V7). Cobalt chloride treatment induced NRCM apoptosis and then either NRCM growth medium (PosCtr), serum-free medium (NegCtr) or serum-free medium supplemented with infant/adult CDC/CF-EVs was added for four days (Suppl Fig. S12C). Pro-apoptotic Bax was significantly reduced in NRCMs by all EVs (CDC and CF, adult and child) compared to the negative control (Suppl Fig. S12D) while the death surface receptor Fas was only significantly reduced by adult CF-EVs (Suppl Fig. S12E).

Since anti-fibrotic properties of CDC-EVs have been previously described [23, 35, 64], we assessed this effect in a so-called wound healing assay (scratch/migration assay with CFs). Migration of CFs is a relevant process during scar building after injury in the heart. CF scratch assays were performed analogically to the above-described EC scratch assay (Suppl Fig. S12E). No significant effect was observed, neither for infant nor adult CF- or CDC-EV-treated samples (Suppl Fig. S12F, G, Suppl Videos V8–13). However, adult as well as infant CDC-EVs stimulated CF migration to a certain extent (Suppl Fig. S12F, G) indicating a potential for wound healing/scarring, even if the effect was not significant.

As we hypothesized that CDCs develop their properties mainly by the exposure to growth factors and 3D-sphere formation, we assessed whether a similar cell type could be generated when treating CFs alike (Suppl Fig. S13A). Therefore, CFs were replated to poly-d-lysine-coated plates with the same growth factor mix than CDCs. Interestingly, this resulted in 86% of experiments in sphere formation which had a similar appearance and size than CDC cardiospheres (Suppl Fig. S13B). Cells grown out of CF-derived spheres were termed CF sphere-derived cells (CFSPhs) (Suppl Fig. S13A, B). CFSPhs were generated from neonatal patients only and compared to CDCs and CFs of the same age group (≤ 21 days). Gene expression of cardiac transcription factors, CF- and CDC-markers showed that CFSphs became more similar to CDCs than to their originating CF population (Suppl Fig. S13C–F). They lost their fibroblast identity and adopted a more CDC-like phenotype (Suppl Fig. S13E, F). Merely, NKX2.5 was expressed in an even lower level in CFSPhs than in CFs (Suppl Fig. S13C, right panel). Flow cytometry showed ubiquitous abundance of CD90 and CD105 in CDCs, CFSPhs and CFs (Suppl Fig. S13G, H). Immunocytochemical staining revealed no considerable difference between CFSPhs and CDCs or CFs regarding the abundance of DDR2 and CD90 (Suppl Fig. S13I, J).

To assess whether CFSPh-EVs had comparable effects to CDC-EVs, we isolated EVs according to the same protocol as for CDC- and CF-EVs (Suppl Fig. S11A). Percentage of dead cells and cell number per culture flask were comparable between CFSPhs, CDCs and CFs (Suppl Fig. S13K, L). However, CFSPh-EV yield approached the yield of CDC-EVs (Suppl Fig. S13M). CFSPh-EV size ranged between 40 and 200 nm, as seen for CDC- and CF-EVs (Suppl Fig. S13N, Suppl Fig. S11E). The expression level of miR-146a of CFSPh-EVs was between CDC-EVs and CF-EVs (Suppl Fig. S13O).

Finally, we compared CDC-, CF- and CFSPh-EVs by assessing their function as described above (tube formation assay, scratch assay with ECs and CFs). Results revealed that CFSPh-EVs were not significantly more beneficial than the NegCtr concerning their potential to augment angiogenesis assessed by tube formation assays (Suppl Fig. S14A–G, Suppl Video V14). However, they tended to stimulate EC migration to a similar level than CDC-EVs, without reaching significance (Suppl Fig. S14E–G). Further, CFSph-EVs did not promote migration of CFs, similar than CDC-EVs (Suppl Fig. S14H, I, Suppl Video V15).

Thus, incubation of CFs with growth factors and their cultivation in three-dimensional culture conditions (spheres), altered their gene expression, making them more similar to CDCs, but did not augment their EV-mediated angiogenesis potential to the same level of CDC-EVs.

Finally, we have to mention some limitations of our study.

In the present study, we generated CDCs from human right atrial tissue according to the protocol used by Messina et al. [43]. The same protocol was utilized for the generation of CDCs from children with single ventricle physiology in the TICAP and PERSEUS trial, where significant functional improvement of the right ventricular ejection fraction was detected [29, 30, 51]. However, recent clinical trials in adult patients, such as HOPE (Duchenne muscular dystrophy patients), DYNAMIC (dilated cardiomyopathy patients), ALLSTAR (ischemic left ventricular dysfunction patients) or the trial with Covid-19 patients [11, 40, 54, 60] all used an allogenic CDC cell product (CAP-1002) that was generated from ventricular tissue (see also [55]). These trials showed variable types of disease-modifying bioactivity of CAP-1002. However, we could not find a publication about a direct comparison of “atrial” versus “ventricular” CDCs and which cell type showed more beneficial effects. Thus, the results of our study cannot be compared directly to the CAP-1002 CDCs. Nevertheless, our results might elucidate the cellular origin of CDCs in the heart and their molecular identity compared to non-myocyte, non-hematopoietic cell types.