PDGFRα demarcates the cardiogenic clonogenic Sca1+ stem/progenitor cell in adult murine myocardium

Cardiac progenitor/stem cells in adult hearts represent an attractive therapeutic target for heart regeneration, though (inter)-relationships among reported cells remain obscure. Using single-cell qRT–PCR and clonal analyses, here we define four subpopulations of cardiac progenitor/stem cells in adult mouse myocardium all sharing stem cell antigen-1 (Sca1), based on side population (SP) phenotype, PECAM-1 (CD31) and platelet-derived growth factor receptor-α (PDGFRα) expression. SP status predicts clonogenicity and cardiogenic gene expression (Gata4/6, Hand2 and Tbx5/20), properties segregating more specifically to PDGFRα+ cells. Clonal progeny of single Sca1+ SP cells show cardiomyocyte, endothelial and smooth muscle lineage potential after cardiac grafting, augmenting cardiac function although durable engraftment is rare. PDGFRα− cells are characterized by Kdr/Flk1, Cdh5, CD31 and lack of clonogenicity. PDGFRα+/CD31− cells derive from cells formerly expressing Mesp1, Nkx2-5, Isl1, Gata5 and Wt1, distinct from PDGFRα−/CD31+ cells (Gata5 low; Flk1 and Tie2 high). Thus, PDGFRα demarcates the clonogenic cardiogenic Sca1+ stem/progenitor cell.

F ate-mapping studies provide evidence that adult mammalian cardiac regeneration exists, though at a level insufficient to rescue damaged hearts, occurring at least partly through a lineage decision by progenitor/stem cells [1][2][3][4] , not proliferation of pre-formed myocytes as in zebrafish or newborn mice 5,6 . This view is supported by evidence using transgenic fluorescent anillin, that cardiomyocytes in damaged adult hearts increase in ploidy but do not divide 7 . Characterizing the dormant adult cardiac progenitors is arguably still in its infancy, despite identifiers including the orphan receptor stem cell antigen-1 (Sca1; refs 2,3,8,9), c-kit 4,10 , side population (SP) dye-efflux phenotype [11][12][13] , Isl1 (ref. 14), cardiosphere-15 and colonyforming assays 16 , aldehyde dehydrogenase 17 , or re-expression of the embryonic epicardial marker Wt1 (ref. 18). Notwithstanding these uncertainties, cardiac progenitor/stem cells have begun to be used in human trials 19 . Unlike cells from bone marrow, intrinsic progenitor/stem cells residing in the heart are predisposed to convert to the cardiac muscle lineage after grafting 5 and are, uniquely, a possible target for activation in situ by developmental catalysts 5,18 .
Existing work on endogenous cardiac progenitor cells has chiefly relied on purified but potentially mixed populations. Where clonal growth was reported, this was often achieved at a prevalence r0.1% for fresh cells, or contingent on prior adaptation to culture 10,[20][21][22][23][24] . In one study, only 0.03% of adult cardiac Sca1 þ cells proliferated beyond 14 days 20 . Sheets of clonally expanded Sca1 þ cells improve cardiac function after infarction 21 . Sca1 þ cells have cardiogenic and vascular differentiation potential 2,8,9,12 , though whether their single-cell progeny have multilineage potential is uncertain.
Tracking cell progeny with Cre recombinase suggests that Sca1-fated cells generate cardiac muscle in situ during normal ageing 3 and that Sca1 þ cells are a major source of new myocytes after ischaemic injury 2 . Fate mapping with R26R-Confetti, which labels clonal derivatives randomly with one of four fluorescent proteins 25 , suggests that most single Sca1 cells have limited expansion potential and give rise to a single cell type 3 . Given this rare clonogenic potential, infrequent differentiation and variable expression of other markers 8,9,12,26 , Sca1 þ cells, if not further refined, are presumptively heterogeneous. More remains to be learned about the native cells' capacity for clonal expansion and differentiation, including what prospective markers best predict the dual features of clonal growth plus enrichment for cardiogenic genes.
One candidate is the SP dye-efflux phenotype, which distinguishes hematopoietic stem cells with long-term selfrenewal potential 27 and is mediated by ATP-binding cassette (ABC) transport proteins associated with 'stemness.' SP cells are found in developing and adult hearts 8,11 and have greater colonyforming activity in methylcellulose than non-SP cells 12 , yet little information exists for cardiac SP single-cell derivatives. A predictor of the cardiac colony-forming units (cCFU-F) amid adult Sca1 þ cells is platelet-derived growth factor receptor-a (PDGFRa); these multipotent cells resemble other mesenchymal stem cells (MSCs) and originate from the epicardium 16 . Relative to MSCs from other tissues, cCFU-Fs express greater Mef2c, but lack most cardiogenic transcription factors 26 .
In short, there is insufficient detailed information about the molecular characteristics of progenitor/stem cells residing in the adult heart, whether cloned cardiac cells faithfully resemble their in situ precursors and whether they resemble the multipotent cardiovascular progenitors in embryos and differentiating embryonic stem cells (ESCs). Despite the need to define more clearly the putative reservoirs of adult cardiac cells with differentiation potential, too little is known about how the various reported progenitors relate to one another. In particular, can one identify a more homogenous population at the single-cell level?
Here we have dissected the cardiac Sca1 þ cells-based on their SP phenotype, PECAM-1 (CD31) and PDGFRa-using single-cell expression profiles and rigorous clonal analysis. SP status predicted clonogenicity plus the cardiogenic signature. However, both properties map even more selectively to PDGFRa þ cells.

Results
A cardiogenic signature in SP cells by single-cell profiling. To address the innate heterogeneity of the cardiac Sca1 þ population, single-cell qRT-PCR (PCR with quantitative reverse transcription) was performed on fresh cells, obviating potential bias from in vitro expansion. Given that adult cardiac Sca1 þ cells are enriched for SP cells with cardiogenic potential in vivo 8,11-13 , we tested whether SP cells possess a distinguishable profile. Adult cardiac Lin À /Sca1 þ cells were isolated by immunomagnetic purification, further purified by flow cytometry (Fig. 1a, left), and stained with Hoechst 33342 (Fig. 1a, centre, right). SP cells comprised roughly 1% of the Lin À /Sca1 þ population. The SP phenotype was suppressed by reserpine and verapamil, broadspectrum inhibitors of ABC transport proteins, and by fumitremorgin C, more selective for ABCG2 27 . Gates for preparative sorting were widely spaced and non-contiguous, to obviate crossover by the 'tails' of SP and non-SP cells.
By principal component analysis (PCA; Fig. 1d and Supplementary Fig. 2), SP cells, non-SP cells and cardiomyocytes were resolved as discrete groups, with the mixed Sca1 þ  population straddling its SP and non-SP fractions (Fig. 1d, upper panel). This separation of SP cells, non-SP cells and cardiomyocytes is concordant with their distinct phenotypes, and preferential clustering of Sca1 þ cells with non-SP cells consistent with the predominance of non-SP cells in the Sca1 þ population. Separation visualized by principal component (PC)2 and PC3 was attributable to four subsets of genes, which collectively define the main differences (Myh6, Myl2; Gata4, Hand2, Tbx20, Tbx5; Cdh5, Kdr; Pdgfra, Tcf21; Fig. 1d, below). For most genes, expression or its absence was homogeneous within each respective population (Fig. 1b,c). However, cardiogenic genes were commonly bimodal in SP cells. Considering the functionally significant tetrad of Gata4, Mef2c, Tbx5 and Hand2 (ref. 30), only 8 of 43 cardiac SP cells expressed all four-a 'mosaic' transcription factor phenotype in 480% of the cells. Nkx2-5, Isl1 and Hand1 were not detected. Of the cardiogenic genes identified, only Mef2a and Mef2c were expressed equivalently in SP and non-SP cells, each in a bimodal pattern (Fig. 1c). Thus, unlike cardiomyocytes, fresh single SP cells show highly mosaic expression of key cardiogenic genes, a potential block to their differentiation. Conversely, the existence of any cells with all four factors yet not target gene activation suggests that these factors do not suffice at their native levels to drive differentiation, or that other barriers exist.
Several stem cell-associated markers showed bimodal expression in SP and non-SP cells, equivalently, including ABC transporters (Abcg2 and Abcb1) and two key genes for pluripotency (Nanog and Pou5f1); a third, Klf4, was present in nearly all Sca1 þ cells regardless of SP status ( Fig. 1c; Supplementary Fig. 1). Thus, differences in Abcg2 and Abcb1 mRNA do not establish the dye-effux phenotype-implicating potential differences in pump protein, activity or alternative transporters. Pluripotency genes showed no relation to the cardiogenic signature.
In summary, SP cells largely-but imprecisely-represent the adult cardiac Sca1 þ cells having the molecular signature suggesting a cardiogenic phenotype. The dichotomous expression of Pdgfra and Tcf21 versus Cdh5 and Kdr defines more exactly the cells enriched for cardiogenic transcripts. Cardiac SP cells (more accurately, Pdgfra þ /Sca1 þ cells) express Gata4/6, Mef2a/c, Tbx5/20 and Hand2, but rarely Hand1, Isl1 and Nkx2-5, features of the first, second and both heart fields, respectively. Hence, these adult cells resemble a persistent but incomplete form of the developing cardiac mesoderm.
Clonogenic self-renewing phenotype of cardiac Sca1 þ SP cells. To pinpoint where the rare clonogenic capacity of Lin À /Sca1 þ cells resides, and allow stringent production of single-cell derivatives, preparative flow sorting was performed. The direct cloning efficiency for single cardiac SP cells was B1% (21 clones/ 1,995 cells plated), in the absence of a feeder layer or prior adaptation to culture, ten-fold higher than for non-SP cells, analogous to the greater formation of colonies by SP cells in methylcellulose 12 (Fig. 2a).
Cloned cardiac SP cells were propagated for 410 months and 300 doublings, without crisis or replicative senescence (Fig. 2b). At passage 15-16, o5% showed senescence-associated b-galactosidase activity; four clones were retested later, with 0.3% of cells positive at passage 25, indicating selection against this subpopulation (Fig. 2c). Clones even at 29 passages were highly enriched for Sca1 and the SP phenotype (Fig. 2d) and, at 14 passages, generated secondary clones with an efficiency 440% (6 primary clones tested, 1,425 cells plated; Fig. 2e,f). Thus, cloned cardiac SP cells are self-renewing and maintain their phenotype after long-term propagation.
Given the presence of Gata4, Mef2c, Tbx5, Hand2 and Nkx2-5 in the ancestral network for cardiogenesis 31 , their importance to cardiogenic differentiation by ectopic transcription factors 30,32 , and their heterogeneous expression in fresh cardiac SP cells, these genes were plotted by two-dimensional clustering to visualize more clearly their state of co-expression in clonal progeny (Fig. 3d). Expression was highly mosaic, with permutations of 2-3 factors, but never all five simultaneously, and Nkx2-5 being rarest. Hence, the clones recapitulate the single cells' phenotype of mosaic co-expression. Cardiac transcription factors were well maintained over time, apart from loss of Tcf21 ( Supplementary  Fig. 3b,c). Given the existence of rare fresh cells co-expressing Gata4, Mef2c, Tbx5 and Hand2, but never clones, such cells might be least clonable.
Thus, by single-cell qRT-PCR and clonal analysis, cardiac SP cells resemble an incomplete but stable form of heart-forming mesoderm, a transient phenotype in embryos and embryoid bodies. What maintains their arrested development is unknown. Possibilities include failure to express the transcription factors as nuclear proteins. Hence, western blotting was performed on nuclear and cytoplasmic lysates of four clones (indicated with * in Fig. 3d). TBX20 and MEF2A were detected in all, as expected from their mRNA expression (Fig. 3d), and were nuclearlocalized ( Fig. 3e; Supplementary Fig. 4). GATA4, TBX5 and NKX2-5 were nuclear in all clones expressing these and absent from the remainder (Fig. 3d,e). Weak or absent expression of Mef2c was paralleled by lack of MEF2C protein. For expressed genes, protein expression was highly uniform (485% positive; Supplementary Fig. 5). Thus, cloned cardiac SP cells resemble the molecular signature of freshly isolated SP cells, and clonal variations recapitulate the fresh cells' microheterogeneity.
Tri-lineage cardiovascular markers develop after grafting. Intact myocardium provides the best-agreed environment for differentiation of cardiac progenitors into cardiomyocytes and vascular cells 8,10,13,15,16,21 . To test whether cloned cardiac SP cells maintain their differentiation capacity and ascertain with singlecell progeny their multilineage potential, representative clones expressing mOrange were injected into the infarct border zone (clones 3, 5, 15 and 16). Cell retention 1 day later varied from 1 to 8%, declined to 0.1-0.5% at 2 weeks, and was no greater in ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms7930 control uninjured hearts. Retention r10% with later further loss is consistent with other studies 33 .
One day after grafting, no precocious expression was seen for seven markers of differentiation ( Supplementary Fig. 6a). At 2 weeks, 10% of donor cells expressed cTnI and sarcomeric a-actin, of which half were double stained; similar results were obtained for myosin light chain 2v plus sarcomeric myosin heavy chains ( Fig. 4b; Supplementary Fig. 6a, b). The differentiating mOrange þ cells were mononucleated and lacked organized sarcomeres, indicating immaturity.
By contrast, at 12 weeks, cTnI and sarcomeric a-actin were expressed in up to 50% of donor-derived cells, with organized sarcomeres and binucleation also detected ( Fig. 4a,b). The mOrange þ cells detected histologically at this later time point were scarce, however, and only 5-10 mOrange þ rod-shaped cells were recovered from each heart following enzymatic dissociation at 12-14 weeks (Fig. 4c,d).
Endothelial markers CD31 and vWF were expressed in 4-8% of cells beginning within 2 weeks ( Although not easily identifiable within vascular structures after short-term grafting, donor-derived differentiating cells were found at 12 weeks in the endothelial and SM layers (Fig. 4e,g).
Overall, no difference was seen between injured and uninjured hearts, or between clones, except increased SM formation after infarction with one. Little or no staining was seen after injection of tail tip fibroblasts ( Supplementary Fig. 6g).
To assess whether grafting-cloned SP cells had a functional impact following infarction, serial magnetic resonance imaging (MRI) measurements were performed (Fig. 5). At one day, no differences were seen between the infarcted groups with and without SP cells, both having similar infarct area and reduction of ejection fraction. Notably, at 12 weeks, cell grafting preserved ejection fraction (22 ± 2%), compared with the vehicle-treated control (12±1%), reduced infarct size (38±1% versus 45±2%) and reduced the prevalence of severe left ventricle (LV) These data using single-cell clones support the tri-lineage potential of cardiac SP cells 24 , and suggest that their 'developmental arrest' can be overridden despite incomplete expression of the cardiogenic programme, within the range of factors' represented here. Infarct size, LV function and the prevalence of severe remodelling all were improved by cell grafting. However, biomechanical benefits under these conditions of delivery are more plausibly due to paracrine effects (as in related publications 34,35 ) than to direct actions of the infrequently persistent, newly formed mOrange-positive cardiomyocytes.
Cardiac SP cells derive from Isl1-and Nkx2-5-expressing cells. Potential origins of cardiac SP cells include the specialization of primitive mesoderm denoted by Mesp1, a determinant of multipotent cardiovascular progenitor cells 36 . Like fresh adult cardiac Sca1 þ cells 8 , fresh single SP cells and clones typically lack Nkx2-5 and Isl1 (Figs 1 and 3; Supplementary Figs. 1,3) yet could be the progeny of cells expressing these formerly, that is, after specification by Mesp1. To evaluate these and other possibilities, we performed Cre/lox fate mapping, converting antecedent expression of Cre recombinase into persistent activation of a sensitive reporter, the Ai14 form of Rosa26-tdTomato (tdTom) 37 . The fidelity of each bigenic line (Supplementary Table 3) was confirmed by induction of tdTom (Fig. 6a) and contributions to cardiac Lin À /Sca1 þ /SP cells were quantitated by flow cytometry (Fig. 6b,c). Cre driven by the viral EIIa promoter evoked ubiquitous tdTom. Without Cre, no recombination occurred. With Mesp1-Cre, most cardiac cells expressed tdTom, including myocytes, vessels and interstitial cells. Virtually all Lin À /Sca1 þ cells derived from Mesp1 þ precursors, whether SP or not, suggesting an origin from pre-cardiac mesoderm. Neither SP nor non-SP cells were derived from neural crest, hematopoietic cells or pre-existing cardiomyocytes, defined using Wnt1-, Vav-and Myh6-Cre.    . Expression within each sample is scaled to Z-score, 0 indicating mean expression of the four genes. Red indicates higher expression than the mean, while blue indicates lower expression. The two-dimensional hierarchical clustering algorithm orders the clones and genes based on co-expression profiles. Asterisks denote four clones taken forward to more detailed studies. (e) Western blots for cardiac transcription factors in the clonal lines. Cytoplasmic (C) and nuclear (N) fractions were analysed using glyceraldehyde 3-phosphate dehydrogenase and histone H1 to authenticate the fractions and b-actin as a loading control.
The bands for TBX20 correspond to isoforms with or without the C-terminal 145 a.a. extension.
Nkx2-5 þ precursors generate most cardiomyocytes in the heart, plus endothelium and SM; Isl1 þ progenitors give rise chiefly to the right ventricle, outflow tract and atria (from the second heart field), contributing less to the left ventricle (from the first heart field) 38 . In all chambers, as expected, most cells were Nkx2-5-fated (Fig. 6a). Left ventricular labelling was patchy ARTICLE with Isl1-Cre, whereas second heart field derivatives labelled uniformly. By flow cytometry, more than half the Lin À /Sca1 þ cells were Nkx2-5-derived-fivefold more than for cCFU-Fs 16with an equal contribution from Isl1-fated cells (Fig. 6b,c). Less than 20% of SP and non-SP cells were fated by anterior heart field-restricted Mef2c-AHF-Cre. Whereas Nkx2-5 and Isl1 were expressed rarely in fresh single cells and cloned cardiac SP cells (Figs 1 and 3), their large contribution to the fate map suggests an origin from cardiac mesoderm, involving both heart fields or just the second. Alternatively, as Nkx2-5 and Isl1 contribute to the pro-epicardial organ 39 , a pro-epicardial origin is possible. Indeed, 50% of cardiac SP cells were labelled by proepicardium-expressed cGATA5-Cre (Fig. 6b,c), albeit less than in cCFU-Fs 16 . By contrast, non-SP cells showed greater contributions from Flk1-and Tie2-Cre, consistent with their endothelial lineage markers Cdh5 and Kdr. Given that differences between cardiogenic and endothelial lineage genes in SP and non-SP cells mapped to the Pdgfra þ and Cdh5 þ /Kdr þ phenotypes, respectively, we refined the Cre/lox analysis, adding antibodies against PDGFRa and the endothelial lineage marker CD31. SP and non-SP cells were each divided into two subpopulations, PDGFRa þ /CD31 À and PDGFRa À / CD31 þ (Fig. 7). These two features can be mutually exclusive 16,40 or partially overlapping ( Fig. 7a; cf. Fig. 8a), hence their combined use for prospective sorting is more stringent than PDGFRa alone. Regardless of the SP phenotype, B90% of PDGFRa À /CD31 þ cells were labelled by Flk1-Cre, unsurprising given this fraction's expression of Flk1/Kdr. By     Conversely, PDGFRa þ /CD31 À cells were highly labelled by cGata5-Cre and by activating Wt1-CreERT2 with tamoxifen at E10.5 (70-80% and 40-60%, respectively). Tamoxifen at 8 weeks labelled none of the four Sca1 þ populations. PDGFRa À /CD31 þ cells were infrequently labelled by cGata5-Cre (10%), with no contribution from embryonic activation of Wt1-CreERT2; this may reflect lesser sensitivity of CreERT2, especially at permissible doses of tamoxifen. Small differences were seen using Nkx2.5and Isl1-Cre: 60-70% of PDGFRa þ /CD31 À cells versus 40-50% of PDGFRa À /CD31 þ ones.
PDGFRa tracks specifically with the cardiogenic signature. Given that fresh Lin À /Sca1 þ /SP cells and their clonal derivatives are enriched for Pdgfra, and that Sca1 and PDGFRa are attributes of MSC-like cardiac progenitors identified by the CFU-F assay 16 , PDGFRa and MSC markers were analysed by fluorescenceactivated cell sorting (FACS; Supplementary Fig. 7). CD105 and CD90 were expressed in Z70% of fresh cardiac SP cells and equally in non-SP cells. CD73, CD44 and PDGFRa were detected in just 20-40% of fresh cardiac SP cells, but two-to three-fold more often than in non-SP cells. Greater labelling for PDGFRa concurs with SP cells' greater expression of Pdgfra.

Prospective sorting of cardiogenic cells using PDGFRa.
Because the SP phenotype was dispensable for clonogenicity and requires a mutagenic dye that is unsuited to translation, we tested the alternative that separating Lin À /Sca1 þ cells based on PDGFRa þ /CD31 À (omitting Hoechst 33342) would prospectively purify cells with a consistent cardiogenic signature (Fig. 8d-f). The single-cell profiles of all six populations-the PDGFRa þ /CD31 À versus PDGFRa À /CD31 þ subsets of Sca1 þ , SP and non-SP cells respectively-were compared with PCA (Fig. 8e). PDGFRa þ and PDGFRa À populations were separated clearly, attributable to antithetical expression of Pdgfra/ Tcf21 versus Cdh5/Kdr, the former genes co-segregating with Gata4/6, Hand2, and Tbx5/20. Density histograms showed equivalent impact of PDGFRa on gene expression regardless of SP status (Fig. 8f). The 'hinge' population between the rigorous SP and non-SP gates also was enriched for PDGFRa þ /CD31 À cells, and cardiogenic gene expression here likewise tracked with PDGFRa ( Supplementary Fig. 8). Conversely, gene enrichment in PDGFRa À /CD31 þ cells included Tal1 and Gata2, hemangioblast transcription factors that, respectively, repress the cardiac fate and assist the binding of Tal1 51 .
Given the cardiogenic gene profile of PDGFRa þ /CD31 À cells across all the various Sca1 þ populations, we sought to test whether PDGFRa and CD31 might suffice for prospective sorting of cells capable of cardiac differentiation in vivo, as reported for CD31 À SP cells if co-cultured with cardiomyocytes 12 . PDGFRa þ /CD31 À Sca1 þ cells were purified as in Fig. 7d, expanded for r10 passages as was essential to produce a sufficient number for injection, and were then delivered to recipient hearts, with and without infarction. In both settings, engraftment and co-expression of a-sarcomeric actin were confirmed at 2 weeks (Supplementary Fig. 9). The reciprocal PDGFRa À /CD31 þ population could not be tested under comparable conditions, however, as these do not undergo equivalent expansion.

Discussion
For a decade, investigations of adult cardiac progenitor/stem cells have proposed diverse criteria to define and purify these cells and, recently, progress them into human trials of cardiac repair 19 . The relationships among reported cardiac progenitor cells remain frustratingly elusive, even among various Sca1 þ cells. By microarray profiling of cardiac Sca1 þ , SP and c-kit þ cells, Sca1 þ cells were closest to cardiomyocytes in their transcriptome-wide molecular phenotype and c-kit cells most remote 52 . Gaps in the present understanding of cardiac stem/ progenitor cells may largely arise from molecular heterogeneities that can be unmasked using single-cell expression profiles as a guide. Such evidence would be essential to develop a dendrogram relating the potential cardiogenic reservoirs in adult myocardium.
Here we combined single-cell qRT-PCR profiling with preparative flow sorting, single-cell deposition, clonogenic assays and systematic investigation of the resultant clones to pinpoint Sca1 þ cells that were enriched for the attributes of a cardiac stem cell. We specifically link Pdgfra with the cardiogenic signature of Gata, Hand and Tbx genes, features mutually exclusive with the endothelial markers Cdh5 and Kdr in Pdgfra À cells. This relation in turn led us to the PDGFRa þ /SP cell as remarkably enriched for the cardiogenic gene signature and clonogenicity, the latter being not only just an attribute of 'stemness' but also instrumental to prove multilineage potential after grafting.
Our results provide clarity into the intricate relationships among at least those cardiac progenitor cells with Sca1 in common, unmasking subpopulations and microheterogeneities previously unknown: (i) fresh single SP cells express the pluripotency markers Nanog, Oct4 and Klf4, in concert with cardiac transcription factors; (ii) their cardiogenic signature is incomplete, resembling a forme fruste of cardiogenic mesoderm during embryogenesis; (iii) cardiac SP cells are highly enriched for Pdgfra mRNA and protein; (iv) Pdgfra and PDGFRa demarcate the cells expressing cardiac transcription factors with greater sensitivity and precision than does the SP phenotype; (v) the inverse correlation between cardiac and endothelial lineage genes suggests a mechanistic basis for differences in SP cells expressing or lacking CD31 12 ; (vi) clonal progeny faithfully recapitulate the cardiogenic signature of the starting SP cells, including mosaic expression of Gata, Hand and Tbx factors; and (vii) PDGFRa is sufficient to define clonogenic cells prospectively, with ten-fold synergy for clonal growth in cells also possessing the SP phenotype. In physiological O 2 , the direct cloning efficiency of fresh single PDGFRa þ /SP cells was425%.
The consensus signature of cardiac PDGFRa þ /Sca1 þ cells is the presence of Gata4/6, Mef2a/c, Tbx5/20 and Hand2, genes essential for normal cardiogenesis 31,53 and encompassing four (Gata4, Mef2c, Tbx5 and Hand2) whose supraphysiological expression drives fibroblasts to a cardiomyocyte-like phenotype 30 . Fresh single SP cells, like single clones, chiefly express just 2-3 of these, unsuspected heterogeneity we unmasked by single-cell profiling. Conceivably, such mosaicism might function to restrain precocious differentiation and exhaustion of the stem cell pool. While the existence of even a few cells coexpressing all four factors without their targets might seem paradoxical, it would be misleading to assume that the four suffice for cardiogenesis at native levels. Our results do not exclude heterogeneity in genes not measured, or epigenetic and posttranslational impediments.
The cardiac SP cell produced by preparative cell sorting partially resembles the cCFU-F produced by colony formation (sharing clonogenicity, multilineage potential, co-expression of PDGFRa with Sca1), but also differs significantly, as the cCFU-F lacks enrichment for the cardiogenic genes emphasized here 16,26 . This may reflect differences in cell selection, if the colony-forming assay discerns or elicits a more primitive cell. Most aspects of the fate map were similar, both derived from mesodermal cells expressing Mesp1, and at least in part from the pro-epicardial organ based on cGata5 and Wt1. Innately, Cre-lox fate-mapping can define only the minimal derivation from precursors expressing a given gene, contingent on the duration and level of Cre expression and on reporter gene functionality in a given context 39 . The few fate-mapping differences between freshly isolated cardiac PDGFRa þ /Sca1 þ cells and the cultured cCFU-F-chiefly, greater derivation from Nkx2-5 þ cells-are quantitative, not qualitative. Biological distinctions are possible, consistent with the gene profiles, but technical differences in recombination are an alternative and perhaps more parsimonious interpretation. As fate mapping does not exclude dual origins from the cardiac crescent and pro-epicardial organ, it would be intriguing to compare further the Nkx2-5-and Wt1-fated cells.
Notably, the SP phenotype was largely but incompletely specific as a predictor of cardiogenic gene expression in single cells, whereas Pdgfra co-segregated with cardiac transcription factors with absolute precision. Tcf21, a feature of cardiac SP cells 11 , was strictly co-expressed in the Pdgfra þ /PDGFRa þ cells enriched for heart-forming factors. Tcf21-fated cells contribute to the epicardium, coronary vessels and interstitium of the adult heart 54 . It is unknown whether Pdgfra and Tcf21 function in adult cardiac Sca1 þ cells' clonal growth and cardiovascular differentiation. Assuming a human counterpart exists for Sca1 þ /PDGFRa þ cells, we speculate that PDGFR may be useful towards obtaining cardiogenic cells from adult human hearts without need for the mutagen Hoecsht 33342 or selection in culture. PDGFRa þ progenitor cells exist in human hearts, whose capacity for cardiac differentiation is not settled 17,40 , and supplemental markers may be required. At the least, PDGFRa provides a rational means, in mice, to purify fresh adult cardiac cells that are uniformly enriched for cardiogenic genes, in greater yield and homogeneity than achievable with SP cells, as an enhanced platform for 'omic' studies and the search for developmental catalysts.
Cell isolation and flow sorting. Adult male 8-to 13-week-old C57BL/6 mice (Charles River) were used for cell purification, analysis and cloning. Lines used for flow cytometry in conjunction with fate mapping are detailed in Supplementary  Table 3. For activation of Wt1-CreERT2 in utero, tamoxifen was given to gestating females at E10.5 as a single dose of 0.04 mg g À 1 (1 mg per mouse), reduced from 0.12 mg g À 1 (ref. 55) and with the addition of progesterone (1 mg per mouse) 16 to improve embryo survival. For recombination in adult mice, tamoxifen was given at 8 weeks for 5 days consecutively (2 mg per day per mouse). Tamoxifen and progesterone were diluted in corn oil and administered by gavage.
Hearts were harvested, minced and enzymatically dissociated using 100 mg.ml À 1 Liberase TH Research grade and 50 mg ml À 1 DNAse I (Roche Applied Science), with four to five cycles of digestion for a total maximum of 45 min at 37°C. The resulting cardiomyocyte-depleted cell preparation was filtered through 70 mm nylon mesh (BD Falcon). Hematopoietic lineage (Lin) depletion and Sca1 enrichment were performed by immunomagnetic separation (AutoMACS Pro Separator, Miltenyi Biotec). For Lin depletion, a cocktail of biotinylated antibodies was used with super-paramagnetic microbeads (Miltenyi Biotec). For Sca1 enrichment, cells were subsequently labelled with anti-Sca1-FITC antibody and anti-FITC microbeads (Miltenyi Biotec), then purified by four cycles of positive selection. In fate-mapping analyses, to minimize cell loss before flow sorting, no Sca1 enrichment was performed. To resolve SP and non-SP cells, cells were stained with 5 mg ml À 1 Hoechst 33342 (Sigma-Aldrich) and filtered as above. Control samples for the dye-efflux assay included Hoechst 33342 with one or more of the following inhibitors: 10 mM fumitremorgin C (Merck), 50 mm verapamil (Sigma-Aldrich) or 5 mm reserpine (Sigma-Aldrich). The final step comprised staining with antigen-presenting cell (APC)-conjugated streptavidin (eBioscience), to resolve any residual contaminating Lin þ cells. Propidium iodide was added as the routine dead cell marker. For fate-mapping analyses, TO-PRO3 was used instead, for compatibility with the tdTom reporter; for cGATA5-Cre and Mef2c-AHF-Cre, TO-PRO5 was used for additional compatibility with APC. Where indicated, other minor modifications were made for specific experiments.
For isolation of neonatal mouse cardiomyocytes (1-3 days old), the hearts were placed in Ca 2 þ -and Mg 2 þ -free phosphate-buffered saline containing 10 g l À 1 D-glucose (PBS-Dg), washed 3 Â in PBS-Dg and minced to obtain fragments of o1 mm. This was followed by enzymatic dissociation in PBS-Dg containing 0.2% collagenase type 2 (B50 U mg À 1 ), 0.05% DNAse (B3,200 U mg À 1 ) and 0.05% Trypsin (B250 U mg À 1 ; #4179, #2009 and #3707, respectively; Worthington). Digestion was performed for five rounds Â 10 min each with agitation at 37°C. At the end of each round, dissociation was facilitated by titruation and the dissociated cell supernatant was transferred to D-MEM/F-12 media containing GlutaMAX, 50 mg ml À 1 gentamicin (both, Invitrogen) and 10% fetal bovine serum (Hyclone). The enzymatic solution was replaced with fresh solution for each round 56 . At the end of the mechanical and enzymatic dissociation, myocytes were collected by centrifugation, followed by Percoll purification using 1.05, 1.06 and 1.82 g ml À 1 densities at 2,000g for 30 min. The myocyte-enriched fraction (interface between the 1.06 and 1.082 g ml À 1 layers) was harvested, washed and further enriched by differential plating on standard tissue culture plates, upon which the cardiac fibroblasts adhere rapidly. After 45 min, the myocyte-enriched suspension was collected and redeposited onto Primaria tissue culture plates in D-MEM/F-12 containing 5% horse serum (Hyclone). Tail tip fibroblasts were obtained by mechanical and enzymatic dissociation, using 0.25% Trypsin-EDTA at 37°C for 30 min with continuous rocking. PDGFRa þ Sca1 þ CD45 À TER119 À bone marrow MSCs were kindly provided by S. Rankin 57 .
The FACSAriaII flow sorter and LSRII flow cytometer (Becton Dickinson), optimized for use in the Hoechst 33342 assay, were equipped identically with 355 nm ultraviolet, 405 nm violet, 488 nm blue, 561 nm yellow-green and 638 nm red lasers. FlowJo vX was used for data analysis (versions 9.3.1 and 10.0.06; Tree Star).
Following initial pilot studies done manually, clone generation and propagation were performed using a customized Industrial Robot Integrated System (Beckman Coulter), comprising two Cytomat24C cell culture incubators with capability for hypoxic conditions (Thermo Fisher), compact HP 3JC Motoman robotic arm (Yaskawa), Cytomat Microplate Hotel (Thermo Fisher), Biomek FXP liquid handling system (Beckman Coulter), CloneSelect Imager (Molecular Devices) to evaluate confluency, Vi-CELL XR cell viability analyser (Beckman Coulter) for trypan blue dye exclusion assays, and Bigneat enclosure. Secondary clones were plated as above. Senescence-associated beta-galactosidase activity was determined, using the Abcam Senescence Detection Kit.
Quantitative RT-PCR. mRNA was extracted manually using TRI reagent or robotically using the Agencourt RNAdvance Tissue system and a Biomek FXP liquid handler (Beckman). Samples were treated with TURBODNase (Applied Biosystems) and RNA quantitated using a Nanodrop 8000 Spectrophotometer (Thermo Fisher Scientific). Reverse transcription was performed with 0.25-1 mg of RNA using the High Capacity cDNA kit (Applied Biosystems). qRT-PCR was performed using TaqMan primer/probe sets (Supplementary Table 2), customized TaqMan low-density array cards and an ABI PRISM 7900HT Sequence Detection System (Applied Biosystems). Cycle number at threshold was calculated using RQ Manager 1.2.1 (Applied Biosystems).
Data for TaqMan low-density array cards and single-cell qRT-PCR were processed by the DCt method 58 , in the R/Bioconductor platform (version 2.15.2) using custom R code. The heatmaps and density plots were visualized using the gplots (CRAN repository) and beanplot packages, respectively.
For the analyses of cardiac SP clones in comparison to a selection of reference tissues (Fig. 3, Supplementary Fig. 2), the expression of 40 test genes was measured together with 3 ubiquitous normalization controls (18 S, Hmbs and Ubc) for each sample. A total of 20 independent clonal cardiac SP lines was investigated, at varying passage number o30 (40 groups) and including replicated samples (for a total number of 87). The expression of each sample was normalized to the combined expression (geometric mean) of the two most stable reference genes (Hmbs and Ubc) to obtain the normalized value of DCt 59 .
The correlation between the cardiac SP clones and reference tissues was evaluated by calculating the Pearson correlation (r) for all sample pairs from the DCt values for all 40 test genes for each sample. Results were then arranged by hierarchical clustering and visualized as a heatmap (Fig. 3a). The Pearson correlation (r) can extend from À 1 to 1. A value close to 1 indicates a high positive correlation (all genes expressed similarly between the two samples), close to 0 (blue) a weak correlation and close to À 1 an anti-correlated gene expression profile.
The molecular signature of the cardiac SP clones is shown as bar plots on a linear scale (2 À DCt ) in Fig. 3b, which shows the mean expression of all 40 measured genes in the 20 clones. The genes were grouped and colour coded on the basis of functional association and tissue specificity. The nparcomp R package was used to perform nonparametric multiple comparisons based on relative contrasts effects. Dunnett's contrasts were specified, where each reference tissue is compared with the cloned SP samples (Supplementary Fig. 2). The statistic was then computed using multivariate transformation (Probit transformation function).
To elucidate the expression profile of key cardiac transcription factors across the single cardiac SP clones, inverted DCt values for each gene were visualized in a density plot (Fig. 3c). To estimate the population density of expression for each gene, we used a Gaussian kernel density estimation procedure applied to each of the 20 SP clones (short vertical lines). Again, the genes were ordered using hierarchical clustering. During this process, the expression of cardiogenic transcription factors within each sample was first scaled by their mean and s.d. This results in a Z-score, with zero mean and unit variance, which normalizes the sample-specific variability. Subsequently, the clustering algorithm was applied to separate genes with heterogeneous (multimodal) expression (Gata4, Mef2c, Tbx5 and Hand2) from genes with homogeneous low/negative expression (Isl1, Hand1and Nkx2-5) and homogeneous high expression (Tbx20, Tbx2, Mef2a and Gata6).
Single-cell qRT-PCR. Single cells were sorted directly (FACSAria II) into 96-well plates containing 10 ml of the reaction mixture for pre-amplification, using CellDirect One-Step qRT-PCR Kits (Invitrogen). Pre-amplification was performed in a Veriti Thermal Cycler (Applied Biosystems) for 22 cycles. As negative controls, at least 3-5 non-template samples were included in each run at the pre-amplification stage. Quantitative amplification was performed using Dynamic Array chips for 48 assays Â 48 samples, the BioMark HD system (Fluidigm) and TaqMan primer/probe sets as above. The stability of endogenous controls was estimated 59 and each sample was consequently normalized to DCt using the expression of Ubc. Samples were centred around their individual means by subtracting the mean expression of all the genes within the sample. This results in a standardized gene expression with zero mean in each sample, thus correcting the technical error associated with batch acquisition.
To visualize the expression of key genes in the populations tested, data were plotted in colour-coded heatmaps of inverted DCt values (Figs 1 and 8): blue (gene expression low/none) to red (high). Samples were ordered on the basis of cell type, and the genes grouped by a hierarchical clustering algorithm according to the underlying co-expression pattern (Fig. 1) or variance between the sample classes (Fig. 8). Individual gene expression levels between selected samples were compared using density plots. To identify differential expression, the genes were classified as present or absent using a Gaussian mixture model classification procedure. The frequency distribution of each gene was then calculated from the combination of two sample classes: SP versus non-SP ( Fig. 1) or PDGFRa þ /CD31 À versus PDGFRa À /CD31 þ (within SP, non-SP and Sca1 þ groups; Fig. 8). The level of significance was obtained with Fisher's Exact test and the resulting P-values adjusted using the Bonferroni correction.
Differences between samples were investigated using PCA 60 . PCA applies multiple linear transformations (singular value decomposition) to the expression profiles (standardized DCt values) of individual samples and identifies a series of PCs that elucidate the most distinguishing features between the samples. The linear projections (PC scores) attempt to maximize the variation between the samples, whereas the coefficients of those projections (PC loadings) measure the importance of genes in defining the underlying variability associated with each component.
Western blotting. For western blotting, cells and tissues were lysed for 45 min at 4°C in radioimmunoprecipitation assay (RIPA) buffer (Sigma-Aldrich), with the addition of cOmplete, EDTA-free, Protease Inhibitor Cocktail and PhosSTOP Phosphatase Inhibitor Cocktail tablets (Roche). Nuclear and cytoplasmic lysates was isolated using Nuclear/Cytosol Fractionation kits (Biovision). Proteins were quantitated by the Pierce BCA assay kit (Thermo Fisher). Lysates were sizefractionated by SDS-polyacrylamide gel electrophoresis (mg per lane), transferred to Hybond ECL nitrocellulose membranes (GE Healthcare), probed with the indicated antibodies and analysed by enhanced chemiluminescence (GE Healthcare). Glyceraldehyde 3-phosphate dehydrogenase and histone H1 were used, respectively, to authenticate the purity of cytoplasmic and nuclear lysates. For a full view of the blots including the location of molecular weight markers, see Supplementary Fig. 3.
Immunocytochemistry. Cultured cells were fixed in 4% paraformaldehyde in PBS for 15 min then treated for 1 h with blocking buffer containing 4% fetal bovine serum albumin (Sigma-Aldrich) and 0.2% Triton X-100 (VWR) in PBS. Cells were incubated with primary antibodies overnight at 4°C in blocking buffer, washed (3 Â 5 min), and incubated with species-specific secondary antibodies for 1 h at room temperature. Species-specific isotype controls were used for all primary antibodies. Cells were counterstained with 4 mg ml À 1 Hoechst 33342 or 4 mg ml À 1 4 0 , 6-diamidino-2-phenylindole (DAPI) for 2 min to visualize nuclei. Image acquisition was performed on an Axio Observer Z1 inverted epifluoresence microscope at 20 Â magnification using AxioVision v 4.8 (Zeiss).
Immunohistochemistry. For tissue staining, 10 mm sections were blocked with normal serum (Vector) for 30 min before incubation with antibodies. For the 12-week-time-point hearts, which were paraffin embedded, sections were deparaffinized in xylene, rehydrated through ethanol and antigens were retrieved with citrate buffer (pH 6.0, 1.92 mg ml À 1 citric acid) in a microwave. Primary antibodies were incubated for 1 h at room temperature or overnight at 4°C and secondary antibodies for 1 h at room temperature. Sections were mounted in Prolong Gold with DAPI (Invitrogen). Routine imaging and quantification were performed on a Zeiss Axio Observer Z1 widefield fluorescent microscope and confocal imaging with a Leica SP5 or Zeiss LSM-780 microscope. For each combination of donor cell and antibody tested, 200 mOrange-positive donor cells were counted, encompassing at least five fields at 400 Â magnification, and the expression of cardiomyocyte, endothelial and SM markers was scored manually by two independent observers blinded to the conditions. Two-way analysis of variance with Bonferonni post-hoc tests was used for statistical analysis, with Po0.05 as the threshold for significant results.
For details and concentrations of the antibodies used, see Supplementary  Table 1.
Cardiac injury and cell grafting. Cells were transduced in the presence of hexadimethrine bromide (Polybrene, Sigma-Aldrich) with a modified version of the lentiviral vector pLL3.7 (ref. 61), replacing the CMV promoter with the phosphoglycerate kinase promoter and the green fluorescent protein reporter with mOrange. Successfully transduced cells were identified and fractionated by preparative flow sorting. For all four clones investigated, the resulting purity of mOrange þ cells was 497%.
For coronary artery ligation 62,63 , C57BL/six females aged 10-14 weeks were anesthetized with 2% isofluorane and given buprenorphine (0.05 mg kg À 1 ) for analgesia. The trachea was intubated for mechanical ventilation (120 strokes per min, 200 ml stroke volume, Harvard Apparatus). The thorax was opened at the left fourth intercostal space and the proximal left coronary artery was ligated with 8-0 Ethilon suture (Ethicon). Sham-operated animals were treated similarly but did not have the ligature tied. Immediately after infarction, mice received two intramural injections into the infarct border zone using a 30 gauge Hamilton syringe, each with 250,000 mOrange-labelled cells in 10 ml saline (Hamelyn Pharmaceuticals). Control mice were injected with saline alone. The chest was closed with 5-0 Mersilk suture (Ethicon).
For histological analysis, mice were euthanized by cervical dislocation 1, 14 days and after surgery and perfusion-fixed with PBS followed by 4% paraformaldehyde in PBS. The hearts were then removed, bisected longitudinally, fixed in 4% paraformaldehyde for 6 h at 4°C, then placed in 10% sucrose in distilled water overnight. Hearts were embedded in optimal cutting temperature compound (Sakura) and frozen in liquid nitrogen-cooled isopentane. For immunostaining, 10 mm frozen sections were prepared using a Leica cryostat (Leica Microsystems). For the 12-week time point, hearts were paraffin embedded to improve tissue preservation.
A modified protocol was used to analyse a-sarcomeric actin expression in donor-derived cardiomyocytes by immunofluorescence and better preserve their sarcomeric structure (Fig. 4d). Retrograde perfusion was performed in Ca 2 þ -free buffer, a recombinant enzyme mix was used (Liberase; Roche), Ca 2 þ re-introduction was performed at room temperature and cells were plated on dishes coated with 2 mg ml À 1 laminin (Invitrogen).
Fate mapping. A Cre-lox system was used for fate mapping (see Supplementary  Table 3 for the lines used). Typically, F1 double transgenic mice were generated by crossing Cre deleter males with R26R-tdTom females, to prevent ectopic reporter expression caused by nonspecific deletion of floxed alleles 65 . Due to the reported 'parent-of-origin' effect 66 , female EIIa-Cre mice were crossed with R26R-tdTom males. Because inconsistent recombination even between littermates was reported for several of the Cre strains used (Tek-Cre, Vav-Cre 67 ), multiple bigenic specimens from at least two independent litters were analysed for all crosses, to avoid possible misinterpretation caused by sample variation, and samples with anomalous reporter activity were excluded from further analysis.
Magnetic resonance imaging. Mice were imaged using a 9.4 T MRI system (Agilent, Palo Alto, CA, USA) and a 38 mm quadrature-driven birdcage RF coil (Rapid Biomedical, Rimpar, Germany). Multi-slice cardiac and respiratory gated cine-MRI was performed in the true short-axis orientation and covered the whole LV 68 . Late gadolinium-enhanced MRI was performed 20 min after intraperitoneal (i.p.) injection of 0.5 mmol kg À 1 Gd-DTPA-BMA (Omniscan, GE Healthcare, Hatfield, UK), using a multi-slice inversion recovery sequence 69 . Data were analysed in a blinded fashion using ImageJ (NIH, Bethesda, MD, US). Standard measures of left and right ventricular morphology and function were made from cine stacks, and infarct area determined by thresholding to three s.d. above the mean remote myocardial signal intensity 70 . Infarct size was quantified as epicardial circumferential length of the enhanced tissue, expressed as a percentage of total epicardial circumference of the left ventricle.
Statistics. The statistical approaches to RNA profiling are given in the sections on qRT-PCR above. For other experimental procedures, unless otherwise stated, Student's two-tailed t-test was used for pairwise comparisons and one-way (Fig. 6) or two-way analysis of variance with Bonferroni's correction for multiple comparisons.