A cardiac transcriptional enhancer is repurposed during regeneration to activate an anti-proliferative program

Zebrafish have a profound ability to regenerate damaged heart muscle. Heart regeneration proceeds via the proliferation of pre-existing cardiomyocytes (CMs) that replace myocardium lost to injury (Jopling et al., 2010; Kikuchi et al., 2010). Thousands of genes change expression levels in a coordinated, injury-responsive program that leads to CM proliferation and other key cell behaviors. Several groups have identified tissue regeneration enhancer elements (TREEs) in zebrafish, regulatory sequences that activate gene expression in different cell types within the regenerating heart and in other organs in response to injury and during regeneration (Cao et al., 2022; Goldman et al., 2017; Kang et al., 2016; Lee et al., 2020; Sun et al., 2022; Thompson et al., 2020). In each case, enhancer activity was demonstrated using ectopic reporter genes. However, understanding enhancer biology in its endogenous environment is more challenging because regeneration phenotypes rarely result after enhancer deletions (Sun et al., 2022; Wang et al., 2020). Phenotypes are likely difficult to recover because genes are rarely dependent on a single enhancer and multiple enhancers increase the possibility of functional redundancy (Bolt and Duboule, 2020; Kvon et al., 2021). Some examples of enhancer-related regeneration phenotypes do exist and, in each case, these loss-of-function experiments have uncovered new and interesting biology (Sun et al., 2022; Zlatanova et al., 2023). For example, in killifish, deletion of an enhancer upstream from inhibin beta resulted in impaired heart regeneration and suggested a model for how evolutionary changes in enhancer elements develop together with regeneration capacity (Wang et al., 2020). However, more enhancer deletions will be necessary to reveal mechanistic activities at the core of how endogenous enhancers function during regeneration.

We produced an atlas of nucleosome turnover in zebrafish CMs to identify regeneration-specific genes and regulatory elements (Goldman et al., 2017). Using a transgenic histone H3.3, we profiled changes in chromatin accessibility in CMs. Dozens of loci containing enrichment of H3.3 were validated in transgenic reporter assays as enhancers, some of which were exclusive to CMs and others that also stimulated reporter expression in other cell types. One such reporter contained a 1265 bp region found 103 kb upstream from the runx1 locus with H3.3 enrichment that increased during heart regeneration. This runx1-linked enhancer (or REN) activated GFP in proliferating cells within the wound of a regenerating heart. REN was not just injury responsive but was also activated in proliferating CMs in a transgenic model of CM hyperplasia, in which the Nrg1 protein stimulates proliferation rather than injury (Goldman et al., 2017). Thus, H3.3 profiling in regenerating hearts identified REN as a genetic marker for proliferating CMs in the adult. Interestingly, the REN enhancer is perfectly functional in mammalian hearts, indicating that its regulatory machinery is likely conserved (Yan et al., 2022). However, we still have not identified what gene REN regulates endogenously and the importance that interaction has to regeneration.

One crucial question that is incompletely understood is where regeneration enhancers come from. In part, the regeneration genetic program involves reactivation of embryonic genetic elements, for example, promoter sequences for the cardiac transcription factor gata4 (Kikuchi et al., 2010). After all, regeneration is largely a recapitulation of development (Goldman and Poss, 2020; Viragova et al., 2024). However, in the case of the heart, it is unclear how much of the developmental program is necessary. The transcription factor Klf1 is dispensable for embryogenesis but required for regeneration (Ogawa et al., 2021). Moreover, Klf1 expression alone is sufficient to cause massive cardiomyocyte hyperplasia by binding mostly to enhancers that are already present in the adult (Ogawa et al., 2021). However, the enhancer(s) most crucial to Klf1-driven CM hyperplasia have yet to be identified. Previously, we have shown that during regeneration there is a doubling of enhancers in CMs, with tens of thousands of previously inaccessible enhancers emerging (Goldman et al., 2017). The origin of these novel CM enhancers and their importance to heart regeneration remains unknown.

Here, we demonstrate that the role of the REN enhancer unexpectedly changes between uninjured hearts and those that are regenerating. REN stimulates gene expression in a subdomain of cardiac tissue that surrounds valves leading to the outflow tract of the heart. During regeneration, activity around the valve is inversely correlated with REN stimulation in regenerating tissue. Surprisingly, zebrafish mutants with REN deleted have improved proliferation of CMs after injury and we show that, during regeneration, REN controls the nearby runx1 gene, mutants of which have similar phenotypes. However, in uninjured hearts, REN deletion mutants have increased Collagen within cardiac valves. This phenotype is complemented by runx1 mutants, demonstrating that valve phenotypes are independent of runx1. Instead in uninjured hearts and around valves, REN controls adamts1, a metalloprotease that can degrade Collagen. Taken together, this suggests that REN is an enhancer that is repurposed from one cardiac domain to stimulate expression from a different gene in regenerating tissue.

Previously, we have shown that a transgenic reporter containing the REN enhancer cloned upstream of a minimal promoter was able to activate GFP in proliferating cells during heart regeneration (Goldman et al., 2017). REN was identified using a CM-specific profiling method that is consistent with REN:GFP expression, co-staining with an antibody targeting the myosin heavy chain in heart muscle 7 days post-amputation (dpa) (Fig. 1A). However, we also observed a significant amount of GFP that is not in the muscle from other cardiac cell types (Fig. 1B). The epicardium, a cell layer enveloping the heart, and endocardium, a second single-cell layer covering the heart lumen, are important tissues required for signaling to the myocardium during regeneration (Kikuchi and Poss, 2012; Kikuchi et al., 2011a; Lowe et al., 2019, 2021; Wang et al., 2015). To determine whether REN:GFP expression also occurs in the epicardium, we injured fish containing both REN:GFP and tcf21:dsRed reporter transgenes and found extensive colocalization in every injured heart (Fig. 1C,D) (Kikuchi et al., 2011b). Similarly, we tested REN:GFP and kdrl:mCherry double reporter fish for REN-directed expression in endocardium (Wang et al., 2010). REN:GFP activated in a few isolated endocardial cells and not in every heart (Fig. 1E). Only a few isolated REN:GFP and kdrl:mCherry colocalized cells are detected within the injury area, representing less than 1% of the total kdrl-positive cells (Fig. 1F; mean 3 dpa=2.0, mean 7 dpa=2.3; N=9,10). We conclude that REN is a regulatory element with predominantly myocardial and epicardial enhancer activity during heart regeneration.

Activation of epicardium (1-3 dpa) precedes the peak of CM proliferation (7 dpa) during zebrafish heart regeneration (Kikuchi and Poss, 2012). To determine the temporal regulation of REN-driven expression, we performed a time course on REN:GFP reporter fish. The peak of total REN:GFP expression occurred at 3 dpa, coinciding with activated epicardium throughout the heart (Fig. S1A), and was slightly decreased at 7 dpa, plateauing by 14-30 dpa (Fig. S1B). Increased expression of GFP at 3 dpa likely reflects a broader distribution of REN:GFP positive cells. To measure the potency of transcriptional stimulation, we calculated mean fluorescence intensity of GFP in colocalized cells. REN:GFP intensity did not change in tcf21:dsRed cells that colocalized with REN:GFP between 3 and 7 dpa, despite the fact that there were fewer cells at 7 dpa [Fig. 1G; magenta dots: 3 dpa, tcf21+GFP mean=20,395 arbitrary density units (ADU)/pixel²; 7 dpa, tcf21+GFP mean=24,269 ADU/pixel²; Welch's t-test, P=0.4481]. Nor was there any change in mean fluorescence intensity in CMs that colocalized with REN:GFP (Fig. 1G; blue dots: 3 dpa, MHC+GFP mean=2770 ADU/pixel²; 7 dpa: MHC+GFP mean=2515 ADU/pixel²; Mann–Whitney test, P>0.9999). At 3 dpa there was 7.36-fold more GFP intensity in epicardial cells versus CMs and 9.65-fold more GFP intensity at 7 dpa, suggesting that REN has more potent activity in the epicardium (Fig. 1G; 3 dpa, Mann–Whitney test, P<0.0001, N=15,14; 7 dpa: Mann–Whitney test, P<0.0001, N=10,10). However, we cannot exclude the possibility that the observed intensity differences were impacted by cell-specific qualities such as size or shape that may influence how GFP is localized. We conclude that REN:GFP peaks throughout the heart at 3 dpa, is focused at the site of injury by 7 dpa, and is brighter in the epicardium than the myocardium.

To identify the minimal sequence components that promote REN activity we produced new transgenic reporter lines with sub-fragments of REN cloned upstream of a minimal promoter driving GFP. Previously, we found DNA sequence motifs enriched in regeneration-specific H3.3 peaks and ranked them by their specificity to regeneration (Goldman et al., 2017). Using the FIMO analysis tool, we searched the REN enhancer for enrichment of these cardiac regeneration motifs (CRMs) and found four clusters of CRMs that included some of the most regeneration-specific sequences (Cuellar-Partida et al., 2012). Using the four clusters of CRMs as a guide, we divided REN into four blocks and made transgenic reporters containing these blocks in different combinations (Fig. 2A).

To define a minimal fragment responsible for REN epicardial expression we looked for fragments of REN that retained expression in epicardium. A fragment containing an extended region encompassing the first two blocks, called REN-b12+, is sufficient for activating GFP in tcf21-positive cells (Fig. 2B). A second fragment containing just the blocks 1 and 2, called REN-b12, is not able to activate in the epicardium (Fig. 2C), nor is the smaller REN-b1x fragment (Fig. 2D). Therefore, the 375 bp extended region of REN-b12+ beyond blocks 1 and 2 is necessary for epicardial activity. However, a transgenic reporter containing block 2 and the 375 bp extended region (b2+) does not activate GFP at all during regeneration. This suggests that the 375 bp region is necessary but not sufficient for epicardial GFP activation.

Amputation of the ventricle apex is a standard injury model that is sufficient for full-length REN activity. However, it is possible that more severe injury models would uncover GFP activity from the sub-fragments of REN that do not activate after amputation. Therefore, we crossed our REN fragments into the zebrafish cardiac ablation transgenic system (ZCAT), in which CM-specific Cre releases the diphtheria toxin stochastically throughout the heart resulting in ablation of up to 60% of CMs (Wang et al., 2011). Most of the REN sub-fragments that are silent by amputation remain silent during genetic ablation (REN-b34, REN-b2, REN-b2+; Fig. S2A-C). The one fragment that does activate GFP, REN-b3/b4, does so in non-CM cells within the heart wall (Fig. S2D). The ZCAT transgenic ablation system includes a red transgenic marker, so we could not confirm whether these cells were epicardial using our tcf21 reporter. However, the expression pattern was highly reminiscent of epicardium, raising the possibility that the region responsible for REN expression in epicardium is a 140 bp fragment that overlaps with REN-b3/b4 and REN-b12+. Further investigation will be required to confirm whether this 140 bp region directs epicardial expression from REN.

The minimal fragment required for CM expression from REN was more straightforward. The three transgenic fragments containing block 1 can drive GFP in CMs at 7 dpa (Fig. 2E-H). Any fragment of REN that does not have block 1 also does not activate GFP in CMs (Fig. S2A-C). The most regeneration-specific motif within block 1, CRM17, has homology to binding sites of the AP1 transcription factor complex (JASPAR, P=0.0175) previously reported to be required for CM proliferation (Beisaw et al., 2020). Therefore, we isolated a 267 bp minimal fragment of that is both necessary and sufficient for REN myocardial activity during regeneration and that contains binding sites for known regulators of CM proliferation.

Multiple lines of evidence suggest that REN promotes a pro-regenerative gene expression program. First, REN directs expression in proliferating CMs independent from injury (Goldman et al., 2017). Second, REN activity peaks just before and during the peak of CM proliferation during regeneration (Fig. 1). Finally, a minimal fragment of REN that retains activity in CMs also harbors binding sites for transcription factors already shown to be required for regeneration (Fig. 2). To address whether REN is part of a pro-regeneration gene regulatory network, we used CRISPR to stably delete a 3162 bp region of the genome encompassing REN, that we call ΔREN (Fig. S3A). We note that the 3162 deletion encompasses the entire 1265 bp of the reporter region. There is no reported evidence of other cis-regulatory elements being located within the 3162 bp in adult zebrafish cardiac tissues (Cao et al., 2022; Cordero et al., 2024; Goldman et al., 2017). First, we tested ΔREN mutants for their ability to complete regeneration. Based on the evidence from the reporter, we hypothesized that homozygous mutants lacking the REN enhancer would have impaired regeneration. We amputated the apex hearts from ΔREN mutants and their wild-type siblings and observed whether regeneration was completed by 30 days. Surprisingly, the ΔREN mutants had complete regrowth of cardiac muscle and absence of scar that is indistinguishable from their wild-type siblings (Fig. S3B,C). We conclude that REN is not necessary to complete regeneration.

To examine whether ΔREN mutants have delayed regeneration, we measured CM proliferation levels at the peak of regeneration in ΔREN mutants and their wild-type clutch mates. We co-stained hearts recovering from amputation of the ventricle apex with Mef2, a marker for CM nuclei, and EdU, a marker of cell cycling (Fig. 3A). Unexpectedly, we found that ΔREN mutants had a significant increase in CM cycling compared to wild-type clutch mates (Fig. 3B). The fraction of Mef2-positive CMs that were also positive with the proliferation marker EdU was increased by 28.5% at 7 dpa (Fig. 3B; mean: wild type=9.11% and mutant=11.71%, P-value=0.0094, N=24 versus 19). There was no difference in CM cycling levels between ΔREN mutants and wild-type siblings in uninjured hearts (Fig. S3D). Also, there was no increase in overall CM numbers in adult ΔREN mutant hearts compared to their wild-type siblings (Fig. S3E). Thus, we conclude that ΔREN mutants have increased levels of CM proliferation that are specific to heart regeneration. This is in stark contrast to our expectations and suggests that REN is inducing expression of an anti-proliferative program.

As REN is a transcriptional enhancer, we expected that its deletion would cause gene expression changes during regeneration. Bulk RNA-sequencing (RNA-seq) of ΔREN mutant hearts and their wild-type siblings identified transcripts that were relatively increasing or decreasing in ΔREN mutants during regeneration (Fig. S3E). We performed linear regression analysis on the 2036 transcripts changing during regeneration in wild-type hearts and the 2159 transcripts changing during regeneration in the mutant and found 305 transcripts for which the fold-change during regeneration was significantly different (Fig. 3C; Table S1). There were 141 transcripts where the fold-change was relatively decreased in ΔREN mutants (Fig. 3C, blue dots) and 164 transcripts where the fold-change was relatively increased in the ΔREN mutants (Fig. 3C, red dots). We conclude that deletion of REN results in dysregulation of gene expression during heart regeneration.

To identify cell-type-specific expression of the differential genes, we looked for enrichment of CM-specific H3.3 at their promoters (Goldman et al., 2017). Of the 305 transcripts with differential fold-change, 187 had the CM-specific H3.3 in their promoters during regeneration, suggesting that the genes were expressed in CMs (Fig. 3D). The promoters of another 23 genes expressing transcripts with a change in levels in ΔREN mutants had active chromatin marks (H3K27ac) but not H3.3-CM (Goldman et al., 2017), suggesting expression in other cell types besides CMs. Taken together, 61% of transcripts disrupted in ΔREN mutant hearts were likely expressed in CMs, with only 7.5% specific to other cell-types. We cannot exclude that H3.3-positive genes are being expressed in other cells in addition to the muscle, but do conclude that, upon deletion of REN, dysregulated genes largely occur in CMs.

Several of the 50 transcripts that increase during regeneration in ΔREN mutants are known regulators of cardiac regeneration. For example, in mice, epicardial expression of Fstl1 increases CM proliferation and, in ΔREN mutants, fstl1a mRNA is 4× more abundant during regeneration (increased 22.14-fold in ΔREN mutants, increased 5.48-fold in wild type) (Wei et al., 2015). Also, fibronectin 1a (fn1a) is required for zebrafish heart regeneration, and fn1a transcripts increase 11.57-fold in ΔREN mutants, but only 6.12-fold in wild-type siblings (Wang et al., 2013). Expression of ankrd1a (also known as CARP) increases 52.82-fold in ΔREN mutants but only 12.96-fold in wild-type regeneration. An enhancer for CARP called 2ankrd1aEN was found together with REN using H3.3 profiling, and 2ankrd1aEN can drive gene expression at the site of injury in both mouse and porcine hearts (Yan et al., 2022). Finally, there is anln, which encodes a protein required for cytokinesis (increased 6.19-fold in ΔREN mutants, increased 2.41-fold in wild-type siblings, residual=1.72) (Oegema et al., 2000; Takayama et al., 2003). The increased induction of pro-regeneration and pro-proliferation genes supports our observation of improved proliferation of CMs (Fig. 3D) in ΔREN mutants.

Based on the expression of the transgenic reporter line, genes that are a direct target(s) of REN would be expected to decrease in abundance when REN is deleted. There were 103 transcripts that decreased relatively in mutant hearts during regeneration (Fig. 3C, red dots; Table S1). Only three of these transcripts are encoded in genes on chromosome 1 with REN; however, it is unlikely that they are direct targets. For example, Meis1 is the transcription factor involved in maturation of mouse CMs (Mahmoud et al., 2013), that increases 5.67-fold in wild-type regeneration but does not increase in ΔREN mutants. However, the meis1 gene is 49.97Mbp away from REN on the other end of the chromosome. Enhancers have been reported to interact with promoters at incredible distances including between different chromosomes (Markenscoff-Papadimitriou et al., 2014), although by chromosome capture, promoters and enhancers interact within 1 Mb ∼80% of the time (Rao et al., 2014). From experimentally validated enhancer-gene associations, the largest reported distance between an enhancer and its cis-regulated promoter is 1.7 Mb away for the Myc gene in mouse (Bahr et al., 2018; Gasperini et al., 2019). We find it unlikely that the three genes downregulated on chromosome 1 in REN mutants are direct targets of REN from 25-50 Mbp away and conclude that RNA-seq alone is insufficient to identify the direct target(s) of REN.

Published experiments using Hi-C have detailed topologically associating domains (TADs) of self-associating chromatin from adult zebrafish brain and skeletal muscle (Yang et al., 2020). Cis-chromatin interactions within TADs helps delineate enhancer-gene pairing from chromosome looping, which is remarkably similar between cell and tissue-types (Rao et al., 2014), developmental stages (Yang et al., 2020) and even between species (Harmston et al., 2017; Kikuta et al., 2007). In both zebrafish brain and skeletal muscle, the first TAD at the end of chromosome 1 encompasses runx1 and REN in a 1.28 Mbp domain (Yang et al., 2020) suggesting they may interact (Fig. S3F; Table S2). Activity by contact modeling of single-cell ATAC-sequencing (ATAC-seq) from adult zebrafish brain (Yang et al., 2020) shows that REN and the runx1 promoter are accessible within the same cells and therefore predicts REN to be an enhancer of runx1 (Table S2).

To determine if REN regulates runx1 in regenerating hearts, we measured changes in runx1 mRNA in ΔREN mutants using the more sensitive droplet digital PCR (ddPCR) assay. For ddPCR, tens of thousands of individual PCR reactions are performed in parallel within separate lipid vesicles. The fraction of droplets that fluoresce from a successful PCR reaction are used to then extrapolate the original number of transcripts using the Poisson distribution. Using ddPCR, we observed a 4.3-fold increase in the number of runx1 transcripts during regeneration of wild-type hearts (Fig. 3E). Mutant ΔREN hearts, however, showed only a 2.4-fold increase, or 57%, of the wild-type during regeneration (Fig. 3E). The remaining increase in runx1 abundance in ΔREN mutant hearts may come from endocardium, where the runx1-BAC reporter is induced (Koth et al., 2020) but REN regulates little to no expression (Fig. 1F). Therefore, we suggest that REN is an enhancer that regulates runx1 expression in most cell types but not likely in the endocardium. Interestingly, mRNA for cbfb, a binding partner and co-factor required for Runx1 transcriptional activity, increases during regeneration in our RNA-seq of ΔREN mutant hearts (Fig. 3C, 5.2-fold, P-value=2.84×10⁻⁴; Table S1). Likely, a feedback loop is activated without the presence of runx1, but only during regeneration and not in uninjured hearts.

To determine the cell-type-specific distribution of runx1 expression we used RNAscope and a runx1-specific probe on wild-type and ΔREN mutant hearts (Fig. 3F). Co-staining of the runx1 probe with an antibody towards the myosin heavy chain demonstrated that runx1 mRNA decreased 65% in CMs of ΔREN mutant hearts (Fig. 3G; mean: wild type=183.3 and mutant=82.13, P-value<0.0001, N=10 versus 8). We note that the remaining signal in the muscle may largely be background (see Materials and Methods). We could not find a co-staining strategy that worked for epicardium or endocardium with RNAscope. However, non-CM signal from runx1 mRNA also decreased by 56%, suggesting that REN controls expression in either epicardium or endocardium (Fig. 3G; mean: wild type=272.6 and mutant=147, P-value=0.0035, N=10 versus 8). Due to the reporter expression pattern (see Fig. 1) we predict that it is in epicardial cells. The disappearing non-CM runx1 mRNA occurred in cells invading into the wound in ways similar to what is reported in the literature for epicardial cells (Fig. 3F, boxes 2 and 3 wild type versus boxes 5 and 6 ΔREN) (Lepilina et al., 2006). In contrast, some of the remaining runx1 mRNA is localized adjacent to cardiac muscle in regions reminiscent of endocardial cells (Fig. 3F, white arrowheads in boxes 4, 5 and 6). We conclude that, during regeneration, REN is a transcriptional enhancer of runx1 expression in CMs and another cell type(s) that is likely epicardium.

The expression of runx1 is minimal around cardiac valves in uninjured hearts (Fig. 3H). If detected, it is mainly confined to endocardial interstitial cells that compromise cardiac valves but is nearly absent from REN-positive regions. This confirms that the runx1-BAC reporter is consistent with in vivo expression patterns of runx1 (Koth et al., 2020). The runx1-BAC reporter contains the REN locus but does not express around valves, suggesting that this activity of REN is unrelated to runx1.

Previously, it was reported that runx1 mutants, like ΔREN mutants, have increased CM proliferation during heart regeneration, supporting a role for REN enhancing runx1 expression (Koth et al., 2020). A major conclusion of Koth et al. was that runx1 regulates CM proliferation through expression in endocardial cells that regulate the composition of temporary scar left within the clot after injury (Koth et al., 2020). Some of the remaining runx1 mRNAs in ΔREN mutant hearts likely came from runx1 expression in endocardial cells (Fig. 3). We did not observe similar changes in the abundance of collagen and fibrin in our ΔREN mutants. Acid Fuchsin Orange G (AFOG) staining of ΔREN mutant hearts was indistinguishable from their wild-type siblings at either 3 dpa (Fig. S3G) or 7 dpa (Fig. S3H). We calculated fibrin and collagen within the scars using the described methodology and, although we can detect expected differences between scarring at 3 dpa and 7 dpa, there were no differences between ΔREN mutants and their wild-type siblings at any time points (Fig. S3I, 3 dpa, chi-square P-value=0.378; Fig. S3J, 7 dpa, chi-square P-value=0.825). These data suggest that the observed differences in scarring in runx1 mutants are not the sole reason for observed increases in CM proliferation (please see Discussion for more details). REN-controlled regulation by runx1 in the epicardium and/or myocardium also affects CM proliferation. We note that the abundance of col12a1b, which was previously reported to be pro-regenerative (Hu et al., 2022), increased only 11× during wild-type regeneration and 31× during regeneration of ΔREN mutant hearts (Table S1). Thus, it is possible that runx1 may regulate CM proliferation in epicardium or muscle in part by changing the composition of the scar.

Expression of GFP in REN:GFP reporters is not unique to regeneration. In uninjured hearts, we observed GFP expression in CMs surrounding the valves leading from the ventricle to the outflow tract (Fig. 4A). Similar expression was observed in each of the three independent REN reporter lines that we originally generated (Goldman et al., 2017). As seen during regeneration, REN:GFP also expressed in non-CMs, including in epicardial cells that emanate above the valve around the outflow tract (Fig. 4B). Calculation of relative GFP intensities showed that REN:GFP had 10× more GFP around uninjured valves than the rest of the uninjured ventricle (Fig. S4A). Interestingly, REN:GFP expression disappeared in the myocardium surrounding the valve during regeneration – with its nadir at 7 dpa, it began to return by 14 dpa (Fig. 4C-F). At 30 dpa, when most of the muscle had finished regrowing (Fig. S3A), REN:GFP expression reached an equilibrium where it was on slightly less than half of its maximal expression in CM both around valves and in the freshly regrown site of injury (Fig. 4G). The expression of REN recovered to an uninjured distribution by 60 dpa (Fig. 4G, yellow line), but at lower overall levels (Fig. 4G, red and blue lines). In summary, the peak of REN-regulated gene expression is inversely correlated between CMs adjacent to valves and CMs that are undergoing regeneration. This suggests that REN activity is mechanistically connected between the expression domains.

To determine the minimal component of REN required for expression around valves, we looked at REN fragment reporter expression in uninjured hearts. REN fragments b12+ (Fig. S4A) and b1x (Fig. S4B) both had activity in myocardium surrounding valves in uninjured hearts and the b2+ fragment did not (Fig. S4C). Thus, the 267 bp minimal region of REN that is necessary and sufficient during regeneration is also necessary and sufficient for myocardial expression in uninjured hearts. We conclude that the same region of the genome is repurposed to activate genes in cardiac tissue, between uninjured valves and during regeneration.

On occasion, reporters for transcriptional enhancers have subdomains of ectopic expression that do not reflect endogenous expression patterns (Bolt and Duboule, 2020). To confirm that the endogenous REN locus regulates genes in cardiac tissue adjacent to valves, we analyzed our RNA-seq comparing ΔREN mutants and their wild-type siblings, only looking at the uninjured replicates. There were 158 transcripts whose abundance significantly decreased by 75% and another 137 transcripts whose abundance significantly increased at least 3-fold in uninjured hearts of ΔREN mutants (Fig. S5A; Table S3). The differential abundance of mRNA transcripts suggests that REN does indeed regulate gene expression in vivo in uninjured hearts. Likely, many of the transcripts changing in abundance are not direct targets and are secondary effects of REN disruption. Interestingly, gene ontology analysis of transcripts with increased abundance in ΔREN mutants included three members of the AP1 transcription factor complex (fosab, fosb and atf3) each of which increased 26.3-fold, 11.62-fold and 9.08-fold, respectively, in uninjured mutant hearts (Fig. S5A, green dots). Therefore, disruption of REN resulted in increased levels of mRNAs encoding a transcription factor complex with motifs present in the deleted enhancer (Fig. S2A). This suggests that a feedback loop may be in place regulating REN activity in uninjured hearts.

The REN:GFP reporter showed that REN activates genes, suggesting that a direct target of REN in vivo would be among the transcripts that are less abundant in uninjured ΔREN mutant hearts. Of the 158 less abundant transcripts, six are encoded by genes found on chromosome 1 and none is found within the reported TAD with REN, where most enhancer-promoter interactions lie (Sun et al., 2019; Symmons et al., 2014) (Fig. S3F; Table S2). However, another gene called adamts1, 672 Mb away from REN, did have decreased abundance (Table S1). The adamts1 gene encodes an extracellular matrix protein that regulates cardiac valves by degrading collagen (Hong-Brown et al., 2015; Hulin et al., 2012, 2019; Tian et al., 2015). We used RNAscope to determine changes in the spatial distribution of adamts1 mRNA in wild-type and ΔREN mutant hearts. As expected, expression of adamts1 mRNA was strong within endocardial interstitial cells that compromise cardiac valves (Fig. 5A, asterisks). It was also abundant in neighboring CMs (Fig. 5A, cyan arrow) and likely epicardial cells (Fig. 5A, yellow arrow). In ΔREN mutant hearts, the abundance of adamts1 transcripts decreased 62% in CM surrounding valves (Fig. 5B; mean: wild type=81.1 and mutant=31.1, Mann–Whitney P-value=0.0001, N=8 versus 9). Expression of adamts1 also decreased to 66% in non-CM cells (Fig. 5C; mean: wild type=955.4 and mutant=321.8, Mann–Whitney P-value=0.0015, N=9 versus 10). Therefore, we conclude that REN functions as a transcriptional enhancer for the adamts1 gene in CMs and possibly in the epicardium surrounding cardiac valves. The expression of adamts1 also increased at the site of injury during regeneration (Table S1). Interestingly, there was no significant difference between adamts1 abundance in CMs at the site of injury between wild-type and ΔREN mutant hearts (Fig. 5C; mean: wild type=28.0 and mutant=23.9, Welch's P-value=0.634). The average level of adamts1 mRNA decreased 40% in epicardial cells, although the P-value was not significant (Fig. 5C; mean: wild type=48.6 and mutant=29.5, Welch's P-value=0.088). We conclude that REN is not a crucial enhancer for inducing adamts1 in regenerating tissue, but it is possible that REN plays a more minor role as a shadow enhancer that stabilizes expression of adamts1 in non-CMs.

To see if ΔREN mutants had observable phenotypes associated with Adamts1, we stained ΔREN mutant hearts for collagen using AFOG and compared them to their wild-type siblings (Fig. 5D; Fig. S5C). Uninjured valves surrounding the outflow tract were visibly more blue in ΔREN mutant hearts, indicating that they have more Collagen. Several pieces of data suggest that REN expression in uninjured valves is distinct from runx1 (Koth et al., 2020). First, from the runx1-BAC reporter, there is no detectable fluorescence in uninjured hearts (Koth et al., 2020). Second, there was no detectable difference in runx1 expression between uninjured wild-type and uninjured ΔREN mutant hearts (Fig. 3F). To determine if REN and runx1 are in different pathways in uninjured hearts, we performed a genetic complementation experiment (Fig. S5D). We crossed ΔREN heterozygotes to a runx1 mutant line and compared compound heterozygotes containing one of each of the REN and runx1 mutant alleles to their wild-type siblings. Uninjured hearts were stained for collagen by AFOG, and compound heterozygotes displayed similar levels of collagen as their wild-type siblings, demonstrating that uninjured phenotypes in ΔREN mutant hearts did not depend on runx1 (Fig. 5E). To quantitatively assess Collagen differences in ΔREN mutant valves, we used immunofluorescence with an antibody towards Collagen I (Fig. 5F). In ΔREN mutant hearts, Collagen I increased 56% in valves [Fig. 5G; mean ADU: wild type (blue)=5,691,576,603 and mutant (light blue)=8,904,512,333, Mann–Whitney P-value=0.0091, N=21 versus 20]. This difference was rescued by crossing ΔREN to a Δrunx1 mutant, demonstrating that Collagen changes were independent of runx1 [Fig. 5G; mean ADU: wild type=6,545,145,907 and double het (purple)=6,671,972,330, Welch's P-value=0.799, N=13 versus 16]. Taken together, deletion of REN results in both decreased abundance of a cardiac valve associated gene, adamts1, and in phenotypes associated with reduced Adamts1 function. We conclude that REN regulates adamts1 expression around uninjured cardiac valves.

Here, we describe a transcriptional enhancer called REN that is active in cardiac tissue around the valve in uninjured hearts but is repurposed to activate at the site of injury during heart regeneration (Figs 1 and 4). Targeted removal of this enhancer using CRISPR resulted in diminished expression of adamts1 only in uninjured hearts and diminished expression of runx1 only during regeneration (Figs 3E-H and 5A-C). We were able to localize these changes to the same regions where the enhancer dictates expression in a reporter. This data conclusively shows that REN regulates adamts1 and runx1 alternatively in the two different conditions. Changes in underlying gene expression together with genetic complementation experiments show that phenotypes regulated by REN in uninjured hearts do not require runx1, demonstrating that uninjured and regeneration pathways for REN are independent of each other (Fig. 5D-G). The activity of transcriptional enhancers during regeneration has been well characterized (Sun et al., 2022; Wang et al., 2020; Yan et al., 2022; Zlatanova et al., 2023). Yet the source of regenerative enhancers, how they are recruited and then subsequently targeted to specific genes, is less well understood. Previously, it has been shown that regenerative enhancers reemerge from developmental enhancers or by invigoration of pre-existing adult enhancers (Kikuchi et al., 2010; Ogawa et al., 2021). Here, we show that, during regeneration, a particular enhancer is repurposed from functioning around cardiac valves and recruited to activate a different gene that is important for CM proliferation in regenerating tissue (Fig. 6).

Several lines of evidence demonstrate that REN very likely directly regulates runx1 during regeneration. First, deletion of either gene or enhancer results in a similar phenotype; CM proliferation improves during regeneration (Koth et al., 2020). Second, single cell ATAC-seq from zebrafish brains uncovered an ‘enhancer hub’ where the promoters for runx1 and genes for calcium channel homologs atp1a1a.4, atp1a1a.3 and atp1a1a.5 all associate with REN (Fig. S3F) (James et al., 1999; Uyehara and Apostolou, 2023; Yang et al., 2020). This suggests that, at least in the brain, REN forms a ‘hub’ to co-regulate these four genes across ∼350 Mbp. Third, when REN is deleted, runx1 expression is decreased during regeneration (Fig. 3E,F), confirming that, also in the heart, REN regulates runx1 expression. There must be additional enhancer(s) that govern runx1 expression in endocardial cells during regeneration (Fig. 1F) and runx1 expression during hematopoietic development, as ΔREN and runx1^ΔTSS double heterozygous mutants develop at Mendelian ratios unlike a runx1 mutant homozygote (Koth et al., 2020; Lam et al., 2009; Sood et al., 2010). REN regulates runx1 expression in CMs during regeneration (Fig. 3G). However, from the RNAscope alone, we cannot conclusively say whether additional runx1 mRNA changes are occurring in epicardial or in endocardial cells (Fig. 3G). The localization of the changing runx1 mRNA suggests that deletion of REN affects the epicardium (Fig. 3F) which is consistent with both REN:GFP reporter expression (Fig. 1F) and the reported data from Koth et al., (2020). We conclude that REN is a CM- and epicardial-specific enhancer of runx1 specifically during regeneration, and not in uninjured hearts (Fig. S6A).

Around cardiac valves in uninjured hearts, REN has additional roles independent of runx1. First, unlike REN, runx1 is not expressed around uninjured cardiac valves (Koth et al., 2020). Second, runx1 mRNA levels do not change when REN is deleted from uninjured hearts (Fig. 3F). Third, the runx1 co-regulator cbfb is only increased during regeneration in mutants and not in uninjured hearts, indicating that a feedback loop that invigorates missing runx1 function is specific to regeneration. Finally, and most importantly, the runx1 mutant allele compensates for ΔREN mutant phenotypes around valves, demonstrating that valvular phenotypes are independent of runx1 (Fig. 5E,G). We note that the 3162 bp deletion in the ΔREN mutant may remove a previously undetected cis-regulatory element that regulates valvular genes. However, the same minimal fragment of REN necessary and sufficient to activate during regeneration is also necessary and sufficient to activate in uninjured valves (Fig. 4). Thus, the ΔREN deletion mutant and reporter experiments are consistent with one another, suggesting that the same enhancer element is driving expression in both contexts.

What is the gene that REN regulates in uninjured hearts if it is not runx1? It is possible that changes in valvular Collagen are resulting from other gene(s) within in the REN enhancer ‘hub’ (Fig. S3F), for example, misexpression of sodium potassium ion channels like atp1a1a (James et al., 1999). Yet, we do not detect expression of mRNA from the cluster of atp1a1a homologs and there are no active chromatin marks at the three atp1a1a promoters in zebrafish hearts (Goldman et al., 2017). We postulate that REN regulates the adamts1 gene either directly or indirectly for several reasons. First, in uninjured ΔREN mutant hearts, valves have more collagen, a phenotype consistent with decreased Adamts1 function (Santamaria and de Groot, 2020) (Fig. 5B,C). Second, while adamts1 does not sit in the same TAD as REN in brain and muscle, there are interactions across the REN-containing TAD boundary between both the adamts1 and runx1 promoters (Yang et al., 2020) (Fig. S6B; Table S3). The runx1 promoter interacts with an enhancer upstream of adamts1 and the adamts1 promoter interacts with an enhancer within the REN-containing TAD (Yang et al., 2020). Third, the abundance of adamts1 mRNA decreases in uninjured ΔREN mutant hearts around cardiac valves in the same region the REN:GFP reporter is expressed (Fig. 5A). This raises the possibility that REN regulates adamts1 across 672 Mbp in cardiac tissue around valves in uninjured hearts.

Regeneration phenotypes for enhancers mutants are difficult to recover. There is a double requirement to have both a gene that when mutated would result in a phenotype and whose expression is dominated by a single enhancer. Here, we show that differences between enhancer mutants and gene mutants can reveal new mechanistic insights of gene function. In mutants for runx1, Koth et al. reported changes in clot composition in the wound during regeneration that is not observed in our ΔREN mutants (Fig. S3G-J). Since, unlike runx1, REN does not robustly activate in endocardial cells (Fig. 1F), we surmise that observed changes in scar deposition result from perturbing endocardial runx1 expression (Koth et al., 2020). However, the shared increases in CM proliferation in the runx1 and ΔREN mutants demonstrate that runx1 impacts CM proliferation through either epicardium or muscle. We cannot exclude a possible role for endocardial runx1 in regulating CM proliferation as well. However, the ΔREN mutant clearly establishes epicardial and/or muscle expression of runx1 as being essential for negatively regulating CM proliferation during regeneration. More work is needed to dissect the individual cell-type-specific contributions and to identify direct targets of Runx1 that modify the ability to regenerate hearts.

Loss-of-function mutations that result in improved heart regeneration are not commonly described (Koth et al., 2020; Missinato et al., 2018). Why an anti-proliferative gene would be activated during regeneration is not obvious and poses a bit of a paradox. For example, at the peak of regeneration the REN enhancer is both a genetic marker of CM proliferation and activates an anti-proliferative program (Fig. 3) (Goldman et al., 2017). A priori, one might expect that a program to slow regeneration would be activated towards the conclusion of regeneration rather than at the peak. We saw no phenotypes in our ΔREN mutant hearts that would suggest regeneration is halted abnormally; there was no additional heart growth beyond 30 days (Fig. S7A) or increased CM proliferation at the end of regeneration (Fig. S7B). However, we cannot exclude that there are redundancies regulating the end of regeneration. In mice, Runx1 is induced in certain models of dilated cardiomyopathy and knockout of Runx1 in heart muscle improves recovery after heart injury (Kubin et al., 2011; McCarroll et al., 2018). However, changes in CM proliferation levels have not been described. Recently, Runx1 has been reported to be linked to the abundance of more proliferative mono-nucleus diploid CMs (MNDCMs), which raises the possibility that expression of Runx1 in early stages of mouse postnatal development increase numbers of MNDCMs by preventing binucleation through inhibiting cytokinesis (González-Rosa et al., 2018; Patterson et al., 2017; Swift et al., 2023).

For reporter fish strains, the regulatory element being studied was cloned upstream of the mouse fos minimal promoter as previously described (Goldman et al., 2017). Stable transgenic zebrafish lines were produced using the I-Sce method of random genomic integration. The tcf21:Red reporter and the flk:Red reporters were previously published. All zebrafish (Danio rerio) used in this study derive from the Ekwill strain. Adults less than 1 year old were used for all the experiments. Males and females were mixed in similar proportions in each of the conditions. All experiments were performed under university supervision according to the Institutional Animal Care and Use Committee protocol #2018R00000090-R2 of Ohio State University. For enhancer reporter lines, we selected at least two and as many as six independent insertions to exclude potential insertional effects.

Zebrafish were anesthetized using tricaine and placed ventral side up on a sponge to carry out resection of the ventricular apex. Iridectomy scissors were used to make an incision through the skin and pericardial sac. Gentle abdominal pressure exposed the heart and ∼20% of the apex was removed with scissors, penetrating the chamber lumen (Poss et al., 2002). Hearts were harvested 1, 3, 7, 14, 21 or 30 days after injury depending on the experiment. To genetically ablate CMs, cmlc2:CreER^pd10; bactin2:loxp-mCherry-STOP-loxp-DTA^pd36 (ZCAT) fish were incubated in 0.5 μM tamoxifen for 17 h (Wang et al., 2011).

Primary antibodies used in this study were: rabbit anti-Mef2 (1:100, Abcam, ab197070), rabbit anti-GFP (1:200, Life Technologies, A11122), mouse anti-MF20 (1:100, Developmental Studies Hybridoma Bank, MF20) and mouse anti-Collagen I (1:10, Developmental Studies Hybridoma Bank, SP1.D8-s). For anti-Mef2, hearts were embedded fresh frozen. For collagen, slides were boiled for 5 min in citrate buffer for epitope unmasking using a steamer. Secondary antibodies were: goat anti-mouse Alexa Fluor 546 (1:200, Thermo Fisher Scientific, A-11030) and goat anti-rabbit Alexa Fluor 488 (1:200, Thermo Fisher Scientific, A-11034).

Briefly, using ImageJ, the red (or blue) channel was thresholded and used to select a region of interest. We despeckled and then analyzed particles >50. Average intensity was calculated using the equation RawIntDen/SumArea=AvgInt/micron squared.

Injured fish were injected into the abdominal cavity once every 24 h for 3 days (4-6 dpa) with 10 μl of a 10 mM solution of EdU diluted in PBS. Hearts were removed on day 7, embedded and cryosectioned. Slides were stained with Alexa Fluor 594 Azide using click chemistry (Breinbauer and Köhn, 2003) and then immunostained for Mef2c. Briefly, sections were blocked with 1% bovine serum albumin (Fraction V; BSA) and 5% goat serum and washed in PBS with 0.2% Triton X-100. Three sections representing the largest wound area were selected from each heart and imaged using a 20× objective. The number of Mef2⁺ and Mef2⁺EdU⁺ cells were counted using MIPAR image analysis software and the CM proliferation index was calculated as the number of Mef2⁺EdU⁺ cells/total Mef2⁺ cells (Sosa et al., 2014). The CM proliferation index was averaged across two to four appropriate sections from each heart.

To derive the ΔREN allele (os⁷⁶), guide RNAs (sgRNAs) were designed against two regions flanking the REN element on chromosome 1 (sgUpstream: gTAgTgTTgAggATAgACAg; sgDownstream: gAAAACAgCTACAgCTCCCT). DNA templates of the respective sgRNA fused to a tracRNA were produced by PCR. T7-transcribed sgRNA were then injected with Cas9 protein into newly fertilized embryos from EK parents. Mutant strains were genotyped using PCR oligos (Fwd: gCCACTgCCTCgCCCCTgCg; Rev: AATCgATgATTCTTgAggTCAAAgATgTgTACT) for the mutant allele. A third oligo (ACAgCTACAgCTCCTCggAC) was added to the PCR to identify the wild-type allele. REN deletion was near perfect with the addition of 10 nucleotides (CACACCCgCT) between the fused cut sites. The 3162 bp deletion encompasses the entire 1265 bp fragment from the reporter plus some adjacent sequence that consists of mostly transposons. Finding reliable sgRNA and genotyping primer pairs within highly repetitive regions was non-trivial.

Demultiplexed and quality-filtered reads were aligned to the Danio rerio reference genome GRCz10 using Hierarchical Indexing for Spliced Alignment of Transcripts 2 (HISAT2) (Kim et al., 2015). Read counts for each gene were quantified using featureCounts software (Liao et al., 2014). Differential gene expression analysis was performed using R package edgeR (McCarthy et al., 2012). The read counts were normalized using the TMM method (Robinson and Oshlack, 2010). Differentially expressed genes were selected based on adjusted P-value and log2 fold change.

A master list of transcripts was created using the following criteria: (1) the transcript of interest had a P-value<0.05 for both the wild-type regeneration and Mut regeneration; (2) the transcript of interest had a log₂FC>1 for either wild-type regeneration or Mut regeneration. Each transcript was assigned its wild-type regeneration log₂FC as the x-value, and its Mut regeneration log₂FC as the y-value. A line of best fit was calculated using the master list, and residuals were calculated for each transcript using the line of best fit. If the residual >1, then the transcript was colored red. If the residual <−1, then the transcript was colored blue.

AFOG staining was performed as previously described (Rao et al., 2023). For calculation of relative fibrin and collagen levels, we adopted the methodology described in Koth et al. (2020). Briefly, ImageJ was used to analyze the color of the wound area of sections stained with AFOG. Channels were split and images were color thresholded for red and blue using the same settings for all hearts. Red and blue particles were analyzed and %Area was recorded for each heart. To determine the orange area, %Red and %Blue values were subtracted from 100%. Percentages were averaged from at least five hearts from each condition.

For each analysis, we used the Welch's parametric t-test to calculate significance. If the variance between the comparison groups was also significant (F test for unequal variances), we switched to using a non-parametric test (Mann–Whitney). For the CM counting and CM proliferation assays, one researcher embedded hearts and sectioned slides and a separate researcher who was unaware of the sample identity carried out imaging and quantification.

mRNA was isolated from wild-type and ΔREN uninjured and injured hearts using Trizol (Goldman et al., 2017). cDNA was made using 1 µg RNA in Superscript II reverse transcriptase reactions incubated at 44°C for 50 min. ddPCR assays were developed to detect runx1, adamts1 and mob4 cDNA. cDNA for Mob4 served as loading control as it does not change in RNA-seq datasets in response to injury or between cell types (Rao et al., 2023). The primer and probe sequences for ddPCR are: runx1 forward primer 5′-CgAgAgCCACgACgCCAC-3′; runx1 reverse primer 5′-CgACTgCTCATACgACCAggATgg-3′; runx1 probe 5′-/56-FAM/CATgCggTg/ZEN/CAgCCCACACCACg/3IABkFQ/-3′; mob4 forward primer 5′-AgTATTTTCCCAgCCgCgTCAgC-3′; mob4 reverse primer 5′-TCACgAAACgggTgAAgCgATgAC-3′; mob4 probe 5′-/HEX/TCCCATgCg/ZEN/TACTTTCACCATCgCCAg/3IABkFQ/-3′; adamts1 forward primer 5′-gAgACCTgCCCTgATAGCAATgg-3′; adamts1 reverse primer 5′-ggTgTCCCATCAgCCACCT-3′; adamts1 probe 5′-/56-FAM/CTGCAAgTT/ZEN/ggTgTGCCgAgCgAAGG/3IABkFQ/-3′. We used 1 µl of reverse transcriptase reactions for detection of runx1 and adamts1 cDNA, and 1 µl of a 1:10 dilution of the cDNA was used to quantify mob4. Standard Bio-Rad reaction conditions were used: 10 µl 2× ddPCR Supermix for Probes (No dUTP) (#1863023), 0.9 µM forward primer, 0.9 µM reverse primer, 0.25 µM probe, desired template amount and remaining volume of water to achieve 20 µl reaction volume. Reactions were partitioned into droplets using Bio-Rad QX200 Droplet Generator (#1864002), mixing 20 µl of reaction with 70 µl Droplet Generation Oil for Probes (#1863005). Droplets were transferred to Eppendorf 96-well twin.tec semi-skirted 96-well PCR Plates (#951020389) and sealed with Bio-Rad Pierceable Foil Heat Seal (#1814040) using a Vitl Life Science Solutions Variable Temperature Sealer (#V902001). PCR thermocycling conditions were performed as follows: 50°C for 2 min, 95°C for 2 min, 55 cycles of 95°C 30 s, 60°C for 1 min and 72°C for 30 s, followed by 72°C for 30 s and then 12°C, using Bio-Rad T100 Thermal Cycler (#1861096). Reactions were read using Bio-Rad QX200 Droplet Reader (#1864003) and thresholds were drawn between the positive and negative droplet populations using the Bio-Rad QX Software Version 2.1. Data were exported to Apple Numbers for further processing. Analysis of resulting data had to conform to >10,000 accepted reaction droplets, >1000 negative reaction droplets, and the number of positive droplets had to be greater than samples that did not receive reverse transcriptase. The calculated copies/µl for runx1 and adamts1 were normalized to mob4. These ratios were directly imported into GraphPad Prism to generate the figures.

Slides with hearts fixed in paraformaldehyde (4% overnight at 4°C) were sectioned, placed at −20°C for 2 h and then moved to −80°C overnight. From there we strictly followed the ACD (Biotechne) Technical Note Sample preparation for ‘fixed from tissue’ using RNAscope 2.5 Chromogenic assay. For the RNAscope protocol itself, we used the RNAscope Multiplex Fluorescent Reagent Kit v2 with the following modifications. We used protease IV, diluted fluorophores 1:3000, washed 3× each for 5 min with gentle mixing after. This was to get rid of as much background signal as possible in the muscle. We boiled slides for 5 min for the adamts1 probe (made for this project) or 10 min for the runx1 probe (Koth et al., 2020). After the last developing and wash for the RNAscope, we washed slides 3× in PBS with 0.05% Tween, left them to block for 1 h at room temperature in 1% goat serum, 1% BSA, and then in primary with the MHC (MF20) antibody (1:100 overnight at 4°C). Slides were washed and coverslipped the next day as normal. Hearts were then imaged on a Zeiss LSM900 Airyscan2 confocal.

The ΔREN mutant strain and some REN fragment reporter lines were originally derived in the Ken Poss lab, and we are grateful for his support and comments on the manuscript. Also, thanks to Arthur Burghes for access to the Droplet Digital PCR machine, loaning us Anton Blatnik, and for manuscript comments. Thanks to Maria Mihaylova and Kubra Akkaya for training with the RNAscope. Thanks to Dr Joshua Waxman for advice on the Collagen antibody. Thank you to the Neuroscience Imaging Core at the Ohio State University (#P30-NS104177) for use of the confocal microscope.

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.204458.reviewer-comments.pdf