Single epicardial cell transcriptome sequencing identifies Caveolin 1 as an essential factor in zebrafish heart regeneration

Adult zebrafish regenerate lost cardiomyocytes (CMs) with high efficiency (Poss et al., 2002). Pre-existing CMs are the primary cellular source of new muscle (Jopling et al., 2010; Kikuchi et al., 2010), while the epicardium, a thin mesothelial cell layer covering the chambers, is activated to express embryonic markers, proliferate, and colonize the injury site (Lepilina et al., 2006). Studies over the past decade have identified the epicardium as a crucial player in heart regeneration and cardiac repair. During heart regeneration, epicardial cells provide paracrine signals, including retinoic acid (RA), Neuregulin 1 (Nrg1) and extracellular matrix (ECM) components such as fibronectin, while also functioning in vasculogenesis (Lepilina et al., 2006; Kikuchi et al., 2011a,,b; Wang et al., 2013; Gemberling et al., 2015). During mammalian cardiac repair, epicardial cells also support cardiac cell survival and vascularization and are a major supply of cardiac fibroblasts, which have the capacity to be directly reprogrammed into CM-like cells (Smart et al., 2007,, 2011; Ieda et al., 2010; Zhou et al., 2011; Huang et al., 2012; Qian et al., 2012; Song et al., 2012).

A recent report demonstrated that the epicardial cell population of adult zebrafish vigorously regenerates after targeted genetic ablation, explaining in part its dynamism upon myocardial injury (Wang et al., 2015). One of several key remaining questions in adult epicardial biology is to what extent the epicardium is heterogeneous at the cellular and molecular levels, and whether potential subpopulations of epicardial cells play distinct roles in regeneration. The definition of specific epicardial subsets would lead to the definition of new molecular markers, the development of genetic tools involving subset-specific regulatory sequences, and the identification of candidate genes that underlie epicardial cell responses to injuries. In zebrafish, the transcription factor tcf21 is specifically expressed in epicardial cells and epicardial-derived cells (EPDCs) (as well as in other mesothelial tissues) throughout development and regeneration, whereas two other epicardial markers widely used in murine research, tbx18 and wt1b, lack such epicardial specificity (Kikuchi et al., 2011a). In adult ventricles, tcf21 expression domains include the outermost epicardial cell layer, as well as inner EPDCs, many of which associate with coronary vessels (Kikuchi et al., 2011a). There is a clear need for a more sophisticated understanding of the diversity and molecular signature of the epicardium beyond this information.

Many recent studies have exploited technologies to acquire transcriptome data from single purified cells (Navin et al., 2011; Shalek et al., 2013; Yan et al., 2013; Knouse et al., 2014). In this study, we generated single-cell transcriptomes from the tcf21⁺ population isolated from adult zebrafish ventricles, and from these data we obtained evidence of at least three cell subsets defined by expression signatures. Several new markers that were present in all epicardial cells, or specific to apparent subsets, were visualized by in situ hybridization. We pursued one gene with strong pan-epicardial expression, caveolin 1 (cav1), by creating a transgenic reporter line and targeted genetic mutants. Our findings reveal a requirement for cav1 in heart regeneration and indicate a strategy for functional exploration of the adult epicardium.

We dissociated epicardial cells from uninjured adult hearts of tcf21:nucEGFP fish and isolated EGFP-positive live cells by FACS (Fig. 1A-C). Thirty-nine single cells were captured via the Fluidigm C1 chip. Transcriptome analysis was performed with 3.6-8.8 million 50 nt read pairs obtained for each cell library. A pooled sample of ∼5000 cells was also analyzed, with sequencing yielding 10 million read pairs. To identify outliers, the Fluidigm SINGuLAR package first identified a set of core genes detected in at least half of the samples. Eight samples had median expression values below the fifteenth percentile for these core genes, suggesting technical failure. These samples were removed prior to further analysis (Fig. 1D) (Pollen et al., 2014). We detected 12,903 genes (expected count ≥1) in pooled samples, with 2826-6001 of these genes (average 3879) in each of the remaining single cells. Additionally, 15,147 unique genes (expected count ≥1 in at least one cell) were detected by single-cell sequencing. Comparison of pooled and single-cell data demonstrated a strong correlation in detected genes. Principal components analysis (PCA) and hierarchical clustering were performed using the Fluidigm SINGuLAR package for the remaining 31 single-cell sequencing libraries (Pollen et al., 2014). Both PCA and clustering indicated three main subpopulations (Fig. 1E,F). One subpopulation (Fig. 1E,F, green circle or frame) was further divided into two groups with apparent differential gene expression (Fig. 1F, dark green and light green frames). These results suggest heterogeneity within tcf21⁺ cells in the uninjured adult zebrafish ventricle, and further suggest at least three subtypes of tcf21⁺ cells.

To identify markers representing these potential subpopulations, we filtered our dataset for transcription factors (or nuclear proteins), secreted factors (or membrane proteins) and other genes differentially expressed across subsets, and also included published epicardial-related genes. Upon gross inspection, genes expressed in group IV (Fig. 1F, blue frame) were also strongly expressed in group II (Fig. 1F, Table S1), making it difficult to differentiate groups II and IV without combinatorial expression patterns. From these filtered genes and known expression domains, we designated the remaining three tcf21⁺ subpopulations as representing the outermost (epithelial) epicardial cells (I, dark green frame), internal cells (middle of the ventricular wall) likely to contain perivascular components (II, red frame), and an innermost layer (inner part of the wall) of EPDCs (III, light green frame) (Fig. 1F, Fig. 2A). Based on these designations, we could further divide the expressed genes into pan-epicardial and subset-specific markers (Table S1, Fig. 2).

Among these selected genes, several have been described previously as epicardial markers. tcf21 was strongly expressed in all cells that we assessed, whereas tbx18 and wt1b were detected in a portion of these cells (Table S1) (Kikuchi et al., 2011a). The two fibronectin paralogs fn1a and fn1b, which are induced by injury in epicardial cells, were both detected in our profile. fn1b is highly expressed in all cells, whereas fn1a was detected at lower levels in a portion of cells; this is consistent with the finding that fn1a is injury induced (Table S1) (Wang et al., 2013). The epicardial (and endocardial) marker raldh2 (aldh1a2 – Zebrafish Information Network) is detected only in the outermost epithelial layer (Kikuchi et al., 2011a), and high-level detection of raldh2 was a signature of group I (Fig. 2B, Table S1). nrg1, which is reported to be expressed in tcf21⁺ perivascular cells, was detected in one cell of group II (Table S1) (Gemberling et al., 2015). cxcl12 encodes a CXC-motif chemokine ligand that is expressed in both epicardium and epicardial-derived smooth muscle surrounding coronary blood vessels during mouse heart development (Cavallero et al., 2015). A recent report in zebrafish described upregulation of cxcl12 in epicardial-derived perivascular cells, where it aids formation of the coronary vasculature (Harrison et al., 2015). Group III has high cxcl12a expression, whereas groups I and II have no detectable and low expression, respectively (Table S1). Further visual confirmation is required to definitively assign groups II and III. Other genes reported to have epicardial expression are igf2b, with expression in epicardial and endocardial cells in the injury site by 7 days post amputation (dpa) (Choi et al., 2013), and zTB2, with expression around the wound and surrounding compact myocardium at 3 and 7 dpa (Lien et al., 2006). These genes are predicted to be pan-epicardial expression markers by our transcriptome data.

To examine our dataset using a different technique, we made probes representing 40 genes and performed in situ hybridization to detect cardiac expression, both in uninjured ventricles and in ventricles 7 days after induced genetic ablation of ∼50% of CMs (Wang et al., 2011). We detected expression signals representing 24 genes in or around the epicardium, often representing different apparent domains. Interestingly, many genes were only detected in injured ventricles, suggesting possible roles in heart regeneration (Fig. 2, Fig. S1). Fifteen of these 24 expression domains were indicative of pan-epicardial expression. Among these genes, mmp2 exhibited a multilayer expression pattern in the walls of uninjured and injured ventricles, resembling the tcf21 reporter expression pattern (Fig. 2A). tmsb4x (zTB4), assessed for roles in epicardial cell activation in mammals (Smart et al., 2007,, 2011), also had a clear two-layer expression domain in the uninjured ventricle wall, and it was strongly induced after injury in multiple cardiac cell types (Fig. 2A). zTB2 displayed weak signals in the ventricular wall after injury (Fig. S1). Other apparent pan-epicardial expression genes revealed by our dataset included cav1, p4hb, sepp1a, vmo1, mmp14a, timp2a, cyr61, hspa5, pcolce2b, lxn, mdkb, twist1 and id3. The genes cav1, p4hb, vmo1, cyr61, mdkb and hspa5 were expressed in additional cardiac tissues, such as endocardium (Fig. S1).

Some other genes, including bmp4 and tfa, were detected only in the outermost layer, both before (very weak) and after injury, resembling the epicardial expression of raldh2 (Fig. 2B). A third gene, clu, also showed weak expression in the outer epithelial layer before injury, with enhanced expression in this layer upon injury. col18a1 and ltbp1, which labeled a cluster of cells within the ventricular wall, are likely to mark the group II cells as identified by transcriptome data. Three other genes, loxl2a, loxl2b and timp2b, were induced within the ventricular wall after injury, with no obvious expression on the outer edge, are in domains consistent with group II and III markers. Importantly, several of these markers are induced by injury, and several appear to be expressed in cell types in addition to epicardium. This latter point is not surprising, as single-cell profiling approaches do not select for restricted expression. In several cases, in situ hybridization signals were weak and could not definitively specify the expressing epicardial cell group.

During heart regeneration, adult epicardial cells can reactivate expression of genes induced in epicardial cells of developing embryos (Lepilina et al., 2006). Also, a recent publication indicated that adult murine epicardial cells restore elements of embryonic signatures (Bollini et al., 2014). To test if the markers we describe are expressed in zebrafish embryos, we performed whole-mount in situ hybridization on embryos at 3 and 4 days post fertilization (dpf). We observed signals indicative of cardiac expression for several genes, including tmsb4x, col18a1 and mdkb (Fig. S2).

In total, we have identified several new epicardial markers that appear to have different expression domains. The groupings indicated by cluster analysis are largely supported by in situ hybridization. These new markers will need further assessment to probe the identities of subsets, especially groups II and III. We speculate that these apparent subsets have distinct functions in cardiac homeostasis and/or regeneration, an idea that can be explored in future functional studies.

Among the listed genes, cav1 was identified as highly expressed in each epicardial group (Fig. 2A, Table S1). cav1 encodes a scaffolding protein and main component of caveolae, which are specialized membrane domains that function in vascular transport, signal transduction and membrane trafficking events (Parton and del Pozo, 2013; Shvets et al., 2014). To test cav1 expression in heart, we generated a BAC transgenic construct with EGFP fused in frame to the C-terminus of Cav1, driven by the endogenous cav1 regulatory sequences (Fig. 3A). We established a stable transgenic line, TgBAC(cav1:cav1-EGFP)^pd1096 (referred to hereafter as cav1:cav1-EGFP) and examined its expression in embryos and adult heart. After crossing with tcf21:DsRed2 or flk1:DsRed2 reporter strains to label epicardial or vascular tissue, we identified conspicuous Cav1-EGFP signals in larval heart, vasculature, notochord, branchial arches, skin and neuromasts (Fig. 3B-E), largely recapitulating what has been described for endogenous cav1 expression (Fang et al., 2006; Nixon et al., 2007). Whole-mount views of adult heart revealed that the entire cardiac surface and coronary vessels showed Cav1-EGFP fluorescence (Fig. 4A). In tissue sections, EGFP signals were strong in all three suggested groups of tcf21⁺ cells, overlapping with tcf21:DsRed2 reporter fluorescence, which is consistent with our single-cell analysis (Fig. 2A, Fig. 4B,D). Cav1-EGFP was also detected in smooth muscle cells of the outflow tract (Fig. 4B,C), in coronary vascular endothelium, and faintly in the endocardium (Fig. 4E). Upon resection of the ventricular apex, cav1-driven epicardial EGFP expression was strong in epicardial cells colonizing the injury site (Fig. 4F). We did not detect Cav1-EGFP in cardiac muscle. Thus, although cav1 expression is also present in other selected cardiac cell types, it is a prominent pan-epicardial marker.

To assess requirements for cav1 in heart regeneration, we generated a mutant using TALENs. We identified a founder containing a 10 nt deletion from nucleotides 77-86 of the open reading frame (ORF) of isoform a, producing a truncated protein lacking the conserved Caveolin domains that are required for caveolae formation (Fig. 5A,B,D) (Fang et al., 2006; Parton et al., 2006). Isoform b, with an alternative translational start site generating a protein truncated by 33 amino acids, should not be affected, as the 10 nt deletion is in the 5′ UTR. The two isoforms are reported to perform nonredundant roles in zebrafish early development. Morpholino depletion of either isoform disrupts embryonic development, and each is incapable of substituting for the other (Fang et al., 2006; Nixon et al., 2007). Isoform a drives caveolae formation more efficiently than isoform b (Fujimoto et al., 2000). Also, these two isoforms have different interaction partners and effects on signaling pathways (Nohe et al., 2005; Gabor et al., 2013). Through quantitative RT-PCR using primers pairs targeting the same region of both isoforms or isoform a-specific sequences, we found that the mRNA level of cav1a, not total cav1, is slightly reduced in homozygous mutants. mRNA levels are more variable in heterozygotes (Fig. 5C).

cav1 homozygous mutants survived normally through embryogenesis and adulthood and were fertile. No differences were apparent in adult cardiac size and morphology, epicardial cell density, or vascularization among wild types, heterozygotes and homozygous mutants (Fig. 5E-H). To test for function during heart regeneration, we resected ventricles of 4- to 6-month-old animals and allowed 30 days for repair. Strikingly, injured ventricles of both homozygous and heterozygous animals commonly displayed severely reduced muscularization, with gaps in the regenerated wall (22 of 29 for heterozygous mutants, and 27 of 32 for homozygotes); by contrast, control animals consistently regenerated a contiguous wall of heart muscle by this period (22 of 30) (Fig. 6A,B, Fig. S3). Instead of visible regeneration, both heterozygous and homozygous mutant animals retained pronounced fibrin and/or collagen deposits at the injured apex (Fig. 6A, Fig. S3). These defects were confirmed semi-quantitatively using a blinded scoring system (Fig. 6B).

Proliferation of existing CMs is the primary, if not exclusive, cellular source mechanism for heart regeneration in zebrafish (Jopling et al., 2010; Kikuchi et al., 2010). To assess CM proliferation in cav1 mutants, we co-stained sections of 7 dpa ventricles with antibodies for the nuclear marker Mef2 and the proliferation marker PCNA (Fig. 6C) (Wills et al., 2008; Kikuchi et al., 2011b; Wang et al., 2011). CM proliferation indices were reduced by 54% in heterozygous fish (5.8% versus 12.6%) and by 39% in homozygous animals (7.7% versus 12.6%) (Fig. 6C,D). There was no significant difference between heterozygous and homozygous mutants.

To further test requirements of cav1, we made a second allele, cav1^pd1104, which disrupts both isoforms and is homozygous adult viable (Fig. S4A; data not shown). Heterozygous cav1^pd1104 mutants also displayed defective heart regeneration (Fig. S4B,C). The fact that zebrafish heterozygous for strong cav1 alleles have a phenotype suggests that the expression level of cav1 is important. We speculate that there may be activation in cav1 homozygous mutants of compensatory networks, as was reported recently in zebrafish (Rossi et al., 2015), that assist embryo survival and might affect injury-induced phenotypes. By this reasoning, heterozygous mutations would activate these compensatory responses to a lesser extent. These results demonstrate that cav1 is essential for injury-induced CM proliferation and natural heart muscle regeneration.

Cav1 has been reported to negatively regulate Tgfβ signaling through interaction with Tgfβ receptors, modulation of receptor expression and receptor complex turnover, and prevention of downstream Smad phosphorylation (Razani et al., 2001; Di Guglielmo et al., 2003; Wang et al., 2006; Lee et al., 2007; Tourkina et al., 2008; Chen, 2009; Miyasato et al., 2011). Cav1 was also reported to be involved in ECM remodeling and turnover (Wang et al., 2006; Tourkina et al., 2008; Miyasato et al., 2011), cell proliferation (Razani et al., 2001; Cho et al., 2004), cell migration and adhesion (Echarri et al., 2007; Grande-Garcia and del Pozo, 2008). Both Tgfβ signaling and ECM remodeling have been suggested to be indispensable for zebrafish heart regeneration (Chablais and Jazwinska, 2012; Wang et al., 2013). To test possible mechanisms of regenerative defects, we assessed fibronectin (Fn) and phospho-Smad3 (pSmad3; the antibody recognizes both Smad2 and Smad3 when phosphorylated at the equivalent sites) presence in cav1 mutants. Injured 7 dpa cav1 hearts displayed similar Fn deposition as hearts from wild-type siblings (Fig. 7A). Similarly, 14 dpa cav1 mutant and wild-type sibling hearts displayed pSmad3 signals in both CMs and non-CM cells, mainly adjacent to the wound area, indicating activation of signaling. As cav1 is expressed in coronary vessels, and coronary neovascularization is important for myocardial regeneration (Lepilina et al., 2006), we also assessed the presence of the endothelial cell marker fli1a:EGFP in all three cav1 genotypes at 14 dpa. Interestingly, wound areas of all three genotypes displayed similar vascular densities (Fig. 8A,B).

As cav1 is predominantly expressed in epicardial cells, we were curious whether the regenerative capacity of epicardial cells themselves is affected in cav1 mutants. To test this, we crossed tcf21:NTR and tcf21:nucEGFP transgenes into the cav1 homozygous mutant background, of which tcf21:NTR enables inducible genetic ablation of ∼90% of the adult tcf21⁺ cardiac lineage (Wang et al., 2015). We then examined epicardial regeneration in cardiac explants by treating freshly extracted and cultured hearts with 1 mM metronidazole (Mtz) for 24 h, before daily imaging (Fig. 9A). We observed similar extents and dynamics of epicardial cell repopulation between all three cav1 genotypes, suggesting that epicardial regenerative capacity is not affected (Fig. 9B).

It is likely that the current binning of cardiac cells into designations such as myocardium, endocardium and epicardium greatly underrepresents the actual cellular diversity of the heart. Here, we have applied single-cell transcriptome profiling toward a better understanding of this diversity in adult epicardial cells, from which our results suggest at least three epicardial subsets. We identified several new epicardial markers with apparently different expression domains, many of which were upregulated after cardiac injury, suggesting possible roles in regeneration. It will also be interesting to examine the expression or function of mammalian orthologs in murine models of cardiac repair. Our single-cell transcriptomes represented 31 epicardial cells from uninjured hearts, but the analysis of hundreds or thousands of individual cells under different injury contexts would unquestionably provide higher-resolution information regarding epicardial heterogeneity. These types of analyses can be applied to other cardiac cell types, such as CMs and endocardium, and ultimately promise to radially expand the genetic toolbox and our molecular understanding of the heart and its regenerative capacity.

Cav1 functions in many contexts by the spatiotemporal regulation of signaling molecules, either by direct interaction with the signaling molecule, by compartmentalization of signaling platforms in caveolae, or by caveolar endocytosis (Gvaramia et al., 2013). The precise mechanistic basis for the CM proliferation and myocardial regeneration defects in cav1 mutants requires further investigation. These functions might overlap with the reported roles for murine Cav1 in mounting responses to myocardial injury that limit heart failure (Zhao et al., 2002; Cohen et al., 2003; Miyasato et al., 2011; Shivshankar et al., 2014).

Adult zebrafish of the Ekkwill strain were maintained as described, and resection injuries and CM ablation were performed as described (Poss et al., 2002; Wang et al., 2011). Animals between 4 and 12 months of age of both sexes were used. All surgeries, histological analyses and proliferation analyses of cav1 mutants were performed in a blinded fashion, and experiments were repeated using different clutches of animals. Transgenic lines used in this study were Tg(tcf21:mCherry-NTR)^pd108 (Wang et al., 2015), Tg(tcf21:nucEGFP)^pd41 (Wang et al., 2011), Tg(tcf21:DsRed2)^pd37 (Kikuchi et al., 2011a), Tg(cmlc2:CreER)^pd10 (Kikuchi et al., 2010), Tg(bactin2:loxP-mCherry-STOP-loxP-DTA176)^pd36 (Wang et al., 2011), Tg(flk1:DsRed2)^pd27 (Kikuchi et al., 2011b) and Tg(fli1a:EGFP)^y1 (Lawson and Weinstein, 2002). Newly constructed strains are described below. Genetic CM ablation injuries were performed as described (Wang et al., 2011). All transgenic strains were analyzed as hemizygotes. Animal procedures were performed in accordance with Duke University guidelines.

Ventricles were collected from adult tcf21:nucEGFP fish at 6 months of age. Ventricular nucEGFP⁺ epicardial cells were isolated as described previously (Zhou and Pu, 2012; Wang et al., 2015). Briefly, ventricles were collected on ice and washed several times to remove blood cells. Ventricles were digested in an Eppendorf tube with 0.5 ml HBSS plus 0.13 U/ml Liberase DH (Roche) and 1% sheep serum at 37°C, while stirring gently with a Spinbar magnetic stirring bar (Bel-Art Products). Supernatants were collected every 5 min and neutralized with sheep serum. Dissociated cells were spun down (300 g) and resuspended in DMEM plus 10% fetal bovine serum (FBS) medium with 1.5 µg/ml propidium iodide (PI), and sorted using a BD FACSVantage SE sorter for EGFP-positive and PI-negative cells. Some sorted cells were plated into a culture dish for 24 h to show cell survival, morphology and EGFP signals. Single epicardial cells were captured in a Fluidigm C1 platform. A pool of ∼5000 cells was also captured as a control. Libraries were prepared according to Fluidigm protocols, and sequencing was performed using an Illumina HiSeq 2000 with 50 nt paired-end reads. Bioinformatics analyses were performed with sequences aligned to the zebrafish genome (Zv9) using Bowtie 2 and transcript expressions were quantified using RSEM 1.2.12 (Li and Dewey, 2011; Langmead and Salzberg, 2012). Outlier identification, PCA and hierarchical clustering of genes were performed in the Fluidigm SINGuLAR package, using default settings. The tool package chose 500 genes with the most differential expression patterns across single cells for PCA and hierarchical clustering (Pollen et al., 2014). RNA sequencing data have been deposited at GEO under accession number GSE75583.

An EGFP-SV40 poly(A) cassette with a 20 amino acid spacer was recombined into the translational stop codon of cav1 in the BAC clone CH211-226K23 using SW105 cells, to generate a C-terminal fusion protein as previously reported (Navis et al., 2013). For transgenesis, inverted Tol2 sites were inserted into the BAC backbone (Suster et al., 2009). The primers for recombination were (5′-3′): forward, ccatggggaaatgctttagtaacgtccgggtcactgctactaaggtggtgGATCTCCCCGCCGAACAGAAA; reverse, cattcccccatcccacccgatccctctggctcttctctttcactctgtccATTGGAGCTCCACCGCGGTG (lowercase, homologous to the BAC; uppercase, homologous to the EGFP cassette construct). The full name of this transgenic line is TgBAC(cav1:cav1-EGFP)^pd1096.

cav1 mutants were generated using TALENs targeting exon 2 of isoform a (exon 1 of isoform b; Fig. 5A). The TALEN was designed using TALEN Targeter (Doyle et al., 2012); the sequences are: left, NG NI NG HD NI NN NN NI NI NI HD NI NN NN NN; right, HD HD NI NG NG NG HD NG NG NG NI NG NG NN NG. The TALENs were assembled using Golden Gate assembly (Cermak et al., 2011) into destination vectors expressing heterodimeric FokI nuclease, pCS2TAL3DD and pCS2TAL3RR, optimized for production of in vitro RNA (Dahlem et al., 2012). Zebrafish embryos were injected at the one-cell stage with 100 pg total TALEN RNA. Genotyping was performed using primers: forward, AGTCTCCCACTCCACGCTCAT; reverse, TGTCCTACCTTCACCACATCGT. The wild-type product is 230 bp, whereas the mutant is 220 bp (Fig. 5B). This mutant allele designation is cav1^pd1094. The second allele was generated using the following TALEN pairs to target to the common sequence of both isoforms: left, NN NN NI NG NI NI HD NN NI HD NI NN HD NI NG; right, HD NG HD HD NG NG NN NN NG NN NG NN NN NI HD NI NG. A mutant with a 23 nt deletion, cav1^pd1104, was identified. Genotyping was performed using primers: forward, CTTTGGAACCGCAGGAATAC; reverse, CCATGCAGTTAAACAAATGTCC. The wild-type product is 282 bp, whereas the mutant is 259 bp (Fig. S4A).

In situ hybridization (ISH) was performed on 10 µm cryosections of paraformaldehyde-fixed hearts using digoxigenin-labeled cRNA probes as described (Poss et al., 2002) with the aid of an InSituPro robot (Intavis). Whole-mount in situ hybridization of 3 and 4 dpf embryos was also performed with the aid of the robot. The raldh2 probe was described previously (Lepilina et al., 2006). twist1 ORF was obtained from Open Biosystems and cloned into pBlueScript II vector (Agilent Technologies), and the probe was generated using T3 RNA polymerase. All other probes were cloned from 2 dpf zebrafish cDNA using the primers listed in Table S2.

Primary and secondary antibody staining were performed as previously described (Lepilina et al., 2006; Kikuchi et al., 2011b). Primary antibodies were mouse anti-Myosin heavy chain (MHC; F59, Developmental Studies Hybridoma Bank; 1:100), mouse anti-troponin T (Neomarkers, MS-295-PABX; 1:100), rabbit anti-Mef2 (Santa Cruz, sc-313; 1:100), mouse anti-PCNA (Sigma, P8825; 1:100), rabbit anti-fibronectin (Sigma, F3648; 1:200) and rabbit anti-phospho-Smad3 (Abcam, ab52903; 1:200). Secondary antibodies (Invitrogen; 1:200) were Alexa Fluor 488 or Alexa 594 goat anti-rabbit and goat anti-mouse. Acid Fuchsin-Orange G staining was performed as described (Poss et al., 2002).

Whole-mount and sectioned ventricular tissues were imaged using a Zeiss LSM 700 confocal microscope, a Zeiss Axio Zoom V16 stereomicroscope, or a Leica DM6000 compound microscope. Quantification of CM proliferation was performed as described previously, by counting a defined region typically including almost all Mef2⁺ PCNA⁺ CMs near the injury (314 µm in the y-dimension of the image). The percentages of Mef2⁺ PCNA⁺ cells from three sections showing the largest wounds were averaged to determine a proliferation index for each heart (Kikuchi et al., 2011b). The vascular density was quantified as described previously (Wang et al., 2015).

Quantitative RT-PCR was carried out in biological triplicate using a Roche LightCycler 480 II system with LightCycler 480 Probes Master Mix and the Universal Probe Library (UPL) (Roche). ef1a (eef1a1l1) served as the control. Primers (f, forward; r, reverse) and the UPL# were: ef1a-f CCTCTTTCTGTTACCTGGCAAA and ef1a-r CTTTTCCTTTCCCATGATTGA, UPL #73; cav1-f ATGACGATGTGGTGAAGGTG and cav1-r TGGTTACTGTGAAGGTGGTGA, UPL #97; cav1a-f TTTCCTCACACTCATCTGATCG and cav1a-r GGCGAGTGAGCGTATTCCT, UPL #124.

Heart explant culture and ex vivo epicardial ablation were performed as described previously (Wang et al., 2015). Briefly, adult hearts were cultured in DMEM medium plus 10% fetal bovine serum, 1% non-essential amino acids, 100 U/ml penicillin, 100 µg/ml streptomycin, and 50 µM 2-mercaptoethanol, while rotating at 150 rpm. For epicardial ablation, dissected hearts were bathed for 24 h in 1 mM Mtz before washout. Explants were imaged using a Leica MZ05FA stereofluorescence microscope.

We thank J. Burris, N. Lee, S. Davies, T. Hume Thoren and N. Benkaci for fish care; D. Corcoran for bioinformatic analysis; and J. Wang and J. A. Goldman for comments on the manuscript.