Cross-species comparison reveals that Hmga1 reduces H3K27me3 levels to promote cardiomyocyte proliferation and cardiac regeneration

Main

Myocardial infarction (MI) causes a massive loss of cardiomyocytes (CMs) that cannot be regenerated, an event that culminates in the formation of non-regenerative fibrotic scar tissue and a decline in heart function. The inability of the adult mammalian heart to regenerate is attributed to the very low CM turnover and cell-intrinsic properties, such as DNA content, and environmental conditions, such as oxygen levels, which collectively restrict the proliferative capacity of mammalian CMs^{1,2,3,4,5,6,7}. Although the adult mammalian heart is unable to regenerate lost myocardium after injury, embryonic and neonatal hearts do possess regenerative capacity⁸. This observation suggests the presence of a latent cardiac regenerative program in mammals that is silenced shortly after birth.

Several approaches to induce CM proliferation in mammals, including inhibiting the Hippo pathway^9,10, activating erb-b2 receptor tyrosine kinase (ErbB2) signaling¹¹ and overexpressing microRNAs¹², have demonstrated promise in regenerating the injured heart. However, the widespread and injury-independent CM proliferation induced by these methods can also lead to detrimental outcomes, including cardiomegaly^11,13,14,15, reduced cardiac contractility¹⁶ and arrhythmias¹⁷. This underscores the pressing need for spatiotemporal control over pro-regenerative stimuli to avert these adverse effects, thus enabling their therapeutic application¹⁸.

In contrast to mammals, zebrafish regenerate an injured adult heart efficiently without adverse effects¹⁹. In the injured zebrafish heart, CMs within the injury vicinity, also known as the border zone (BZ), undergo a transformative process characterized by changes in chromatin organization, de-differentiation and proliferation, all of which facilitate the replacement of lost CMs^{5,20,21,22,23,24,25,26}. Although a BZ also forms in the adult mouse heart after MI^27,28,29,30, the observed transcriptional changes fail to reactivate the cell cycle in CMs. The enigma of species-related differences in injury response remains a largely uncharted territory, with a paucity of knowledge about the molecular mechanisms underlying these variations.

Interspecies and intraspecies comparisons of transcriptional programs have proven very effective in uncovering mechanisms that drive tissue regeneration^31,32. For instance, an interspecies transcriptome analysis between medaka fish, which lack efficient heart regeneration, and zebrafish revealed differences in the innate immune response after cardiac injury, highlighting the importance of the immune response during regeneration versus scarring³³. Moreover, an intraspecies comparison in Astyanax mexicanus fish revealed that efficient heart regeneration in a surface population and its absence in a cave-dwelling population correlates well with the expression of lrrc10 (leucine-rich repeat containing 10), a gene pivotal for cardiomyocyte redifferentiation and maturation^26,34. In our study, we set out to unravel the shared and species-specific responses to injury within the BZ between mouse and zebrafish, aiming to reveal the mechanisms that drive natural heart regeneration. Our hypothesis posits that harnessing these natural responses in the mammalian heart will induce heart regeneration without detrimental effects, such as cardiomegaly or lethal arrhythmias.

Our study unveils substantial transcriptional shifts within the BZ of injured zebrafish and mouse hearts, marked by activation of a stress program, extracellular matrix remodeling and a simultaneous reduction in metabolic gene expression linked to mitochondrial oxidative phosphorylation. Notably, we identified high-mobility group AT-hook protein 1 (Hmga1), a conserved architectural chromatin protein, whose expression correlates with regenerative potential in zebrafish, neonatal mouse and neonatal human hearts. In zebrafish, Hmga1 is a prerequisite for the re-expression of embryonic genes, CM proliferation, scar resolution and the induction of a regenerative program. Hmga1a achieves this by clearing repressive histone 3 lysine 27 trimethylation (H3K27me3) marks from gene bodies of embryonic and metabolic genes, typically silenced in adult CMs but reactivated during regeneration. Notably, inducing Hmga1 expression in BZ CMs of injured mouse hearts triggers a regenerative program, fostering CM proliferation and functional recovery. Moreover, in mouse BZ CMs overexpressing Hmga1, we found a similar decrease in H3K27me3 repressive marks consistent with findings in zebrafish, indicating that Hmga1 has a conserved function. This insight into Hmga1-driven mechanisms that decrease these repressive histone marks introduces transformative potential for regenerative therapies in repairing injured hearts.

Results

Hmga1a is an essential regulator of zebrafish heart regeneration

The BZ in the injured zebrafish heart can be regarded as a local environment composed of various cell types (for example, immune cells, endothelial cells, cardiomyocytes and fibroblasts) that is permissive for CM proliferation. Although these cell types are also present in the BZ of injured mouse and human hearts, their presence and interactions do not lead to cell cycle re-entry of CMs^30,35. To reveal the molecular differences among BZ microenvironments, we performed spatial transcriptomic analysis on injured zebrafish and mouse hearts. For the spatial transcriptomics, we used a method that we developed previously, called TOMO-seq^24,36, on injured zebrafish and mouse hearts at 3, 7 and 14 days post injury (dpi). Although, for zebrafish, the entire ventricle was processed, we used BAC-Nppb-Katushka mice to visualize the BZ after MI³⁷ and isolated a region of the heart including the injury area (IA), the BZ and the remote zone (RZ) for TOMO-seq (Fig. 1a). Pearson’s correlation analysis across all genes for each pairwise combination of sections revealed clusters of genes with expression in the different areas (IA, BZ and RZ) (Fig. 1b). Based on these gene clusters as well as marker gene expression, we identified the locations of the IA (cluster 1, Rhoc (ras homolog family member C), Fstl1 (follistatin-like 1) and Tmsb4x (thymosin beta 4)); the BZ (cluster 2, Nppa (natriuretic peptide type A), Des (desmin) and Ankrd1 (ankyrin repeat domain 1)); and the RZ (cluster 3, Tnnt2 (troponin T2), Tnni3 (troponin I) and Ech1 (enoyl coenzyme A hydratase 1)) within the TOMO-seq datasets, which were validated through in situ hybridization (ISH) (Fig. 1c).

**Fig. 1: TOMO-seq reveals transcriptionally distinct regions in the injured mouse heart.**

To allow for an interspecies comparison, we identified all genes that were differentially expressed in the BZ compared to the IA and RZ for both the zebrafish and mouse datasets (Supplementary Tables 1 and 2) and pooled the different timepoints to mitigate any temporal differences between both species. We identified all differentially expressed BZ genes with an annotated homolog in the mouse and zebrafish genomes and plotted, for each of these gene pairs, the log fold change (logFC) (BZ versus the rest of the tissue) (Fig. 2a,b). This comparison of BZ transcriptomes combined with Gene Ontology (GO) analysis revealed that genes with a function in oxygen-dependent metabolism (OXPHOS) were downregulated in the BZ of both species, reflecting the metabolic reprogramming of BZ CMs⁵^,21,24,30,38 (Extended Data Fig. 1a–d and Supplementary Tables 3–6). In addition, genes with a role in extracellular matrix remodeling and calcium binding were upregulated in the BZ of both species. To validate the overlap between the transcriptomic profile of the zebrafish and mammalian BZ, we compared our TOMO-seq data with previously published transcriptomics datasets obtained from human hearts after MI³⁹. This comparison identified a significant overlap between the published human BZ genes and the mouse and zebrafish BZ genes revealed by our TOMO-seq analysis. In total, we found 102 genes overlapping among the human, mouse and zebrafish BZ lists, including NPPB, ANKRD1 and DES (Extended Data Fig. 1e), which were validated through ISH (Extended Data Fig. 1f).

**Fig. 2: Interspecies comparison identifies Hmga1a, which spatially and temporally correlates with cardiac regenerative capacity.**

To identify genes that may promote regeneration, we focussed on genes that are upregulated only in the zebrafish BZ, with no known upregulated mouse homolog (Extended Data Fig. 1c). The zebrafish-specific BZ profile was enriched in genes related to actin binding, myofibrils and heart development as well as genes related to regulation of proliferation and spindle formation, which is consistent with CM dedifferentiation and redifferentiation^22,26,40 and CM proliferation^5,41, respectively, in the BZ of the zebrafish heart. To identify potential drivers of cardiac regeneration from the list of 371 zebrafish-specific BZ genes, we focused on genes with a potential regulatory function, specifically transcription factors, chromatin-associated factors and genes with a known regulatory role in major signaling pathways. As a result, we selected 20 genes for further experimental investigation (Extended Data Fig. 2a–c and Supplementary Table 7). To enrich for genes with a potential cell-autonomous function in BZ CMs, we explored their spatial expression pattern in zebrafish using ISH. Of the 20 candidates, nine showed expression predominantly in BZ CMs, whereas 11 candidates showed expression mostly in non-CMs. Additionally, quantitative polymerase chain reaction (qPCR) analysis confirmed, for six of nine candidates, that their expression was not upregulated in the mouse BZ. Finally, three candidates were selected for functional follow-up experiments, which had not been implicated in cardiac regeneration. For the remaining three candidate genes, knockout zebrafish lines were produced: khdrbs1a (KH domain-containing, RNA-binding, signal transduction-associated 1a), znfx1 (zinc finger NFX1-type containing 1) and hmga1a (high-mobility group AT-hook 1a). While khdrbs1a mutant fish displayed a lethal phenotype preventing the assessments of its role during heart regeneration, znfx1 and hmga1a mutants were viable as adults. To investigate the potential functional role of Znfx1 and Hmga1a on zebrafish heart regeneration, we performed cryoinjuries on znfx1 and hmga1a mutants and investigated CM proliferation in the BZ and scar regression. Although znfx1 mutants showed no difference in CM proliferation and scar regression (Extended Data Fig. 2d,e), hmga1a mutants showed reduced BZ CM proliferation as well as impaired scar regression (Fig. 2c–f).

As the pipeline narrowed our focus on hmga1a, we sought to analyze expression in injured zebrafish, mouse and human hearts in more detail. We observed that hmga1a expression in both uninjured and 1-dpi zebrafish hearts was undetectable, whereas, at 3 dpi and 7 dpi, hmga1a expression was consistent in the BZ with robust hmga1a expression in BZ CMs at 7 dpi (Fig. 2g). As the zebrafish genome contains two hmga1 genes (hmga1a and hmga1b), expression of both genes was analyzed, revealing that hmga1a but not hmga1b expression is induced in BZ CMs upon injury (Extended Data Fig. 2f–h). In addition, hmga1b expression was not detectable in the hmga1a mutant hearts, suggesting that hmga1b does not compensate for the loss of hmga1a (Extended Data Fig. 2i). Notably, Hmga1/HMGA1 expression was undetectable by ISH in injured adult mouse and human hearts (Fig. 3a,b). Reanalyzing previously published spatial transcriptomics data of human heart after MI³⁹ confirmed this absence of HMGA1 expression (Supplementary Fig. 1). Contrary to adult hearts, ISH on mouse and human neonatal hearts showed abundant expression of Hmga1/HMGA1 (Fig. 3a,b). Additionally, qPCR analysis demonstrated that Hmga1 is expressed in neonatal mouse hearts at 1 day after birth (P1) and that its expression declines rapidly in the first week after birth, coinciding with the loss of regenerative capacity⁸ (Fig. 3c). HMGA1 protein levels were confirmed to follow this pattern, with high protein abundance in a P3 mouse heart and significantly lower abundance in 14 dpi as well as sham hearts (Fig. 3d).

**Fig. 3: *Hmga1* correlates with regenerative capacity of the mammalian heart.**

From these results, we conclude that hmga1a/Hmga1/HMGA1 expression spatially and temporally correlates with the regenerative capacity of the zebrafish and mammalian heart and that hmga1a is required to stimulate efficient CM proliferation and heart regeneration in zebrafish.

BZ gene expression program is regulated by Hmga1a

Hmga proteins (Hmga1/Hmga2) are architectural chromatin proteins that contain three AT-hooks, enabling them to bind to the minor groove of AT-rich DNA⁴². By doing so, Hmga proteins can compete with histone H1, which is a linker histone that compacts chromatin, leading to more open and accessible chromatin and increased expression of undifferentiated and stem-cell-related genes^{43,44,45,46,47,48}. To gain insight into the function of Hmga1a in the BZ, we performed single-cell RNA sequencing (scRNA-seq) on sorted BZ CMs using Tg(nppa:mCitrine) of both wild-type and hmga1a mutants at 7 dpi (Fig. 4a). We used uniform manifold approximation and projection (UMAP) to visualize expression of the pan-cardiomyocyte marker myl7 (myosin light chain 7), which was uniformly high across all cells (Extended Data Fig. 3a). Additionally, the expression of BZ markers nppa and desma was widespread (Extended Data Fig. 3b,c), confirming that these cells are indeed BZ CMs. Unsupervised Seurat analysis on the scRNA-seq data identified six transcriptionally distinct cell clusters, indicating that BZ CMs adopt different cell states. Notably, hmga1a mutant and wild-type BZ CMs were unequally distributed across these clusters (Fig. 4b,c and Extended Data Fig. 3d,e), indicating that Hmga1a influences the transition into these different cell states. To further investigate this, we applied RNA velocity, which estimates the future transcriptional state of cells by modeling transcriptional dynamics⁴⁹, to our data. The model predicts whether the abundance of spliced mRNA will increase or decrease in the future, providing direction and magnitude for transcriptional changes. The resulting velocity vectors indicate the direction of predicted transcriptional shifts (Fig. 4d). We also performed pseudo-temporal ordering of the cells using Monocle, which corroborated the RNA velocity findings (Fig. 4e). Both RNA velocity and pseudo-temporal ordering revealed a directional trend from the bottom of the UMAP plot toward the top, aligning with the distribution of hmga1a mutant cells (enriched at the bottom) and wild-type cells (enriched at the top). Based on these results, we conclude that (1) BZ CMs undergo transcriptional changes and (2) these changes are dependent on the presence of Hmga1a.

**Fig. 4: Hmga1a regulates progression of the regeneration program in BZ CMs.**

Next, we used the pseudo-temporal ordering to identify processes that occur in BZ CMs during regeneration, in which order they occur and how they depend on the presence of Hmga1a. Differential expression analysis of genes over pseudo-time and unsupervised clustering led to the identification of eight gene modules that consist of co-expressed genes (Fig. 4f and Supplementary Table 8). Modules 1–4 consisted of genes mainly involved in cell–cell interactions, immune response, stress and hypoxia (Fig. 4g). Pseudo-time modules 5–8 were enriched for genes with a role in cell cycle regulation, chromatin organization, heart development, energy metabolism and protein translation (Fig. 4g,h). Increased rates of translation⁵⁰ and oxidative phosphorylation⁵¹ are correlated to cell cycle progression toward mitosis, and the reduced CM proliferation observed in hmga1a mutant BZ CMs may explain their observed downregulation. hmga1a was found in module 5, indicating that genes in modules 1–4 could potentially act upstream of Hmga1a. Corroborating such an upstream function, we observed that module 1 contained genes related to neuregulin (Nrg1)/ErbB2 signaling, which induces CM proliferation when activated⁵² (Fig. 4g,h). Indeed, pharmacological inhibition of the Nrg1/ErbB2 pathway with the ErbB2 inhibitor AG1478 resulted in a significant reduction in hmga1a expression in BZ CMs (Extended Data Fig. 4a,b). In addition, intraperitoneal injection of recombinant NRG1 protein into uninjured zebrafish resulted in ectopic hmga1a expression throughout the entire heart (Extended Data Fig. 4c,d) and enhanced CM proliferation (Extended Data Fig. 4e). Finally, we also observed that NRG1-induced CM proliferation is largely dependent on Hmga1a activity, as NRG1 injection in hmga1a mutant hearts had no effect on CM proliferation compared to NRG1 injection in wild-type fish (Extended Data Fig. 4e). Of interest in modules 5–8 was the presence of tbx5a (T-box transcription factor 5a), tbx20 (T-box transcription factor 20) and nkx2.5 (NK2 homeobox 5), encoding cardiac transcription factors that are induced by injury⁴⁰, as well as hk1 (hexokinase 1), encoding a rate-limiting enzyme for glycolysis, which is essential for BZ CM proliferation^5,41. We confirmed reduced hk1 and tbx20 expression and reduced levels of phosphorylated ribosomal protein S6 (pS6, Ser253/263), a marker for translation rate⁵³, in hmga1a mutant BZ CMs, corroborating that genes in modules 5–8 are regulated by Hmga1a (Fig. 4i–l). Interestingly, mouse BZ CM gene expression determined by Calgagno et al.⁵⁴ showed a strong overlap with modules 1–4 (22%) but much less so with modules 5–8 (11%) (Supplementary Fig. 2), which is in good accordance with a lack of Hmga1 expression and the low proliferative capacity of mouse BZ CMs.

Together, these results indicate that Hmga1a acts downstream of Nrg1/ErbB2 signaling to regulate the expression of cardiac transcription factors, cell cycle regulators and genes that regulate the metabolic reprogramming of BZ CMs during heart regeneration.

Hmga1a promotes cell cycle re-entry of CMs

As we observed that Hmga1a is required for cryoinjury-induced expression of embryonic cardiac genes and CM proliferation, we wanted to address whether hmga1a expression is sufficient to induce a regenerative program. Therefore, we generated a zebrafish line with tamoxifen-inducible and CM-specific hmga1a overexpression (OE), Tg(ubi:Loxp-BFP-stop-Loxp-hmga1a-eGFP, myl7:CreERT2), which we hereafter refer to as hmga1a OE. CMs from this line showed robust nuclear protein localization 14 days post tamoxifen (dpT) (Fig. 5a,b). To investigate transcriptional changes, we sorted CMs from hmga1a OE and control Tg(myl7:CreERT2) fish at 14 dpT and performed bulk mRNA sequencing (mRNA-seq). Analysis of the mRNA-seq data identified 1,280 upregulated genes (P < 0.05, logFC > 1) and 1,203 downregulated genes (P < 0.05, logFC < −1) in hmga1a OE CMs (Extended Data Fig. 5a and Supplementary Table 9). There was a striking correlation between the upregulated genes and the scRNA-seq modules 5–8, containing BZ genes downstream of Hmga1a, and between the downregulated genes with scRNA-seq modules 1–4, which contain genes upstream of Hmga1a (Extended Data Fig. 5b,c). In addition, a comparison with previously identified BZ genes⁵ revealed that 28% of these BZ genes were upregulated in hmga1a OE CMs (Extended Data Fig. 5d,e). Together, this suggests that hmga1a OE stimulates a more embryonic-like BZ transcriptome.

**Fig. 5: *hmga1a* OE stimulates CM proliferation resulting in myocardial expansion without pathological remodeling.**

Next, we quantified the fraction of proliferating cell nuclear antigen (PCNA)⁺ CMs at 14 dpT of hmga1a OE to investigate the consequences of the Hmga1a-induced transcriptional changes on CM proliferation. We observed that Hmga1a OE resulted in a significant induction of PCNA⁺ CMs, without affecting sarcomere organization or heart morphology (Fig. 5b–d). Long-term Hmga1a OE for 1 year affected cardiac growth, as we observed a significant increase in myocardial surface area (Fig. 5e,f), mainly due to an expansion of the trabecular region at the expense of the cardiac lumen (Fig. 5g,h). This expansion was due to the modest but significant increase in CM proliferation (Fig. 5i), not due to an increase in CM size (Fig. 5j). Notably, although hearts from long-term hmga1a OE fish displayed enhanced growth, they did not show any pathological remodeling (Fig. 5k), which is a striking difference when compared to 5 months of Nrg1 OE in Tg(β-actin2:loxPmTagBFP-STOP-loxP-Nrg1) zebrafish (Fig. 5l). Together, these results demonstrate that hmga1a OE in CMs is sufficient to induce a partial BZ-like gene expression program, including the induction of CM proliferation without any pathological consequences.

Epigenetic remodeling by Hmga1a

Given the role of Hmga1 in chromatin organization^44,55,56, we investigated whether the transcriptional changes observed in hmga1a OE are associated with changes in epigenetic modifications in CMs. Ex vivo time-lapse imaging on cardiac slices revealed that the nuclear Hmga1a–eGFP remained bound to chromatin during CM division (Extended Data Fig. 6a), indicating that it is a structural chromatin component. To further explore Hmga1a-induced epigenetic changes, we employed sort-assisted single-cell chromatin immunocleavage (sortChIC), a technique combining cell sorting, which allowed us to sort for CMs, with chromatin immunocleavage⁵⁷, to map changes in three histone marks: histone 3 lysine 4 trimethylation (H3K4me3) (marking active promotors), histone 3 lysine 9 trimethylation (H3K9me3) (marking constitutive heterochromatin) and H3K27me3 (marking facultative heterochromatin) (Fig. 6a). As expected, we found that H3K4me3 was enriched at promoter regions, H3K9me3 at distal intergenic regions and H3K27me3 in intergenic regions, gene bodies and promoter regions (Fig. 6b and Extended Data Fig. 7a). Additionally, gene expression levels correlated with H3K4me3 and H3K27me3 marks: highly expressed genes showed high H3K4me3 levels and low H3K27me3 levels (Fig. 6b). In contrast, no such correlation was observed for H3K9me3, likely due to its predominant localization in intergenic regions (Extended Data Fig. 7a,b). Comparing hmga1a OE and control CMs, we observed a significant increase in H3K4me3 marks on promoter regions and a significant decrease in H3K27me3 marks on gene bodies in hmga1a OE CMs (Fig. 6c). To further assess whether hmga1a OE affects H3K27me3 levels, we performed immunohistochemistry on tissue sections of zebrafish hearts. We observed a significant reduction of H3K27me3 levels in zebrafish hearts with hmga1a OE compared to controls (Fig. 6d,e), demonstrating that hmga1a OE reduces repressive H3K27me3 marks, potentially leading to increased gene expression. Indeed, analysis of H3K27me3 levels on gene bodies in a subset of genes upregulated by hmga1a OE showed a strong and significant reduction in H3K27me3, whereas genes downregulated by hmga1a OE did not exhibit such a reduction (Fig. 6f).

**Fig. 6: Reduction of repressive H3K27me3 marks by *hmga1a* OE in zebrafish CMs.**

Among the genes both downstream of Hmga1a and upregulated by hmga1a OE, as revealed by our RNA-seq analyses, are embryonic cardiac genes (for example, nppa), cardiac transcription factors (for example, nkx2.5 and tbx20) and genes with a role in energy metabolism (for example, aldoab (aldolase a, fructose-bisphosphate, b) and mb (myoglobin)). Genome track of these genes revealed a clear correlation among a reduced H3K27me3 signal over the gene body and promoter region, an increased H3K4me3 signal at the promoter region and increased mRNA reads upon hmga1a OE (Fig. 6g). Conversely, genome tracks of genes such as hox genes show no reduction in H3K27me3 levels upon hmga1a OE (Extended Data Fig. 7c), highlighting the specificity of hmga1a OE. Together, these results support a model in which hmga1a OE leads to a reduction in H3K27me3 and an increase in H3K4me3 on genes involved in cardiac development and energy metabolism, resulting in upregulation of these genes.

Hmga1 stimulates mammalian heart regeneration

Given the observed effects of hmga1a OE in zebrafish, we tested the potential of Hmga1 to stimulate mammalian heart regeneration. Because the neonatal mammalian heart can regenerate, and neonatal CMs are susceptible to proliferative stimuli^8,58,59, we addressed whether ectopic Hmga1 expression in primary isolated neonatal rat ventricular cardiomyocytes (NRVMs) stimulates proliferation (Extended Data Fig. 8a). Interestingly, a significant increase in 5-ethynyl-2′-deoxyuridine (EdU) incorporation and Ki67 labeling was observed in the Hmga1–eGFP transduced NRVMs compared to the eGFP-only transduced CMs (Extended Data Fig. 8b,c), indicating that delivery of Hmga1 in neonatal mammalian CMs can increase their proliferative capacity.

Although the adult mammalian heart does not regenerate, and BZ CMs display very limited cell cycle activation upon injury, BZ CMs do undergo drastic changes in terms of their morphology^27,28,29 as well as their transcriptome and chromosomal organization³⁰, which might potentiate their susceptibility to mitogenic stimuli^{10,11,35,60,61}. To address whether introducing Hmga1 expression in injured adult mouse hearts induces CM cell cycle re-entry in vivo, we performed permanent left anterior descending artery (LAD) ligation to induce an MI and injected an adeno-associated virus 9 (AAV9) carrying a control CMV:GFP (referred to as GFP virus) or a CMV:HA-Hmga1 cassette (referred to as Hmga1 virus) in two opposing regions bordering the area at risk of ischemic injury. Co-staining with anti-pericentriolar material 1 (anti-PCM1), which marks the nuclear membrane of CMs, revealed that 95–100% of transduced cells were CMs (Extended Data Fig. 8d), and the intracardiac injection resulted in transduction of 20–30% of all CMs in the BZ (Extended Data Fig. 8e). To assess CM cell cycle activity, mice were injected with EdU bi-daily for 2 weeks after MI (Fig. 7a). At 14 days after MI, EdU incorporation in CMs located at the BZ was increased more than 10-fold in CMs expressing HA-HMGA1 (Fig. 7b,c). In addition, HA-HMGA1 expression in BZ CMs resulted in a more than nine-fold increase Ki67⁺ CMs and a 10-fold increase in Aurora B⁺ CMs, further validating that Hmga1 virus injection induces CM cell cycle re-entry and progression through mitosis and cytokinesis in BZ CMs (Fig. 7d,e and Extended Data Fig. 9a,b). Notably, the percentage of EdU⁺, Ki67⁺ or Aurora B⁺ CMs was not affected in hearts injected with GFP virus (Extended Data Fig. 9c–e). Contrary to the observations in the BZ, we did not observe a stimulatory effect of the Hmga1 virus for these cell cycle markers in the CMs located in the RZ (Fig. 7c–e) nor did we observe an induction of CM proliferation in sham-operated hearts (Extended Data Fig. 9f,g), suggesting that, in the mouse heart, the BZ creates a microenvironment that allows HMGA1-induced cell cycle reactivation of CMs. Immune cells, including macrophages, accumulate in the BZ, and their presence is required for CM proliferation during zebrafish and neonatal heart regeneration^33,62. As Hmga1 OE can exacerbate inflammation⁶³, we investigated whether Hmga1 OE had a similar effect on MI-induced inflammation, but we did not find any evidence for this (Extended Data Fig. 8f–h).

**Fig. 7: HMGA1 promotes CM proliferation and cardiac regeneration in injured adult mice.**

Next, we addressed whether injection of Hmga1 virus also leads to functional improvement after MI (Fig. 7f). Histological analysis to assess scar size at 42 dpi revealed a modest decrease in scar size in Hmga1 virus-injected hearts compared to control, albeit not significant (Fig. 7g,h). Sham and MI mice injected with virus were subjected to echocardiography at baseline, 14 dpi and 42 dpi. At 14 dpi, there was no significant difference in ejection fraction (EF) between GFP and Hmga1 virus-treated animals, but, at 42 dpi, the EF of Hmga1 virus-treated animals was significantly improved, whereas the EF of GFP virus-treated animals continued to decline (Fig. 7i,j). Interestingly, cardiac output and stroke volume improved to sham levels in MI mice injected with Hmga1 virus (Extended Data Fig. 9h–j). No difference in heart weight or CM size was detected after Hmga1 OE (Extended Data Fig. 9k–m). Together, these results indicate that a single local injection of Hmga1 virus in the adult mouse heart after MI promotes mammalian heart regeneration by inducing cell cycle re-entry of BZ CMs and improving cardiac function.

Epigenetic remodeling by Hmga1 in the mouse BZ

Because HMGA1 binds to chromatin in mouse cells⁶⁴, a finding that we confirmed (Extended Data Fig. 6b), we investigated whether Hmga1 OE in mouse hearts leads to similar epigenetic changes as observed in zebrafish hearts with hmga1a OE. We first performed immunohistochemistry for H3K27me3 on tissue sections of MI mouse hearts injected with either GFP or Hmga1 virus. Notably, we observed a specific and significant reduction of H3K27me3 levels in BZ CMs of injured mouse hearts expressing HA-HMGA1 (Fig. 8a,b). To further explore whether this reduction in H3K27me3 levels occurs at specific genomic loci, we performed sortChIC on GFP⁺ BZ CM nuclei isolated from 14-dpi hearts of mice injected with control (AAV9Myo4a-hTNNT2p-mCherry-p2A-H2BGFP) or Hmga1 OE (AAV9Myo4a-hTNNT2p-HA-mHmga1-p2A-H2BGFP) virus, which allow sorting on H2B GFP (Fig. 8c). Similar as in zebrafish, we found that H3K27me3 was enriched in intergenic regions, gene bodies and promoter regions (Extended Data Fig. 10a). Comparing H3K27me3 levels between control and Hmga1 OE BZ CMs, we observed a significant decrease in H3K27me3 marks on gene bodies in Hmga1 OE CMs (Fig. 8d). Interestingly, analyzing a subset of genes orthologous to genes upregulated or downregulated in zebrafish hmga1a OE CMs revealed a significant reduction of H3K27me3 levels upon Hmga1 OE for genes upregulated but not for those downregulated upon hmga1a OE (Extended Data Fig. 10b). Genome tracks for Tbx20, Mb, Nppa, Aldoa and Nkx2.5 showed this reduction in H3K27me3 levels (Fig. 8e). Similar to our observations in the zebrafish, Hox genes did not exhibit this reduction in H3K27me3 marks, indicating a level of specificity in the effect of Hmga1 OE (Extended Data Fig. 10c). These results indicate that Hmga1 OE reduces repressive H3K27me3 marks from specific genomic loci in mammalian CMs.

**Fig. 8: Reduction of repressive H3K27me3 marks by HMGA1 in mouse BZ CMs.**

Taken together, our data bring forward a model where Hmga1 reduces repressive H3K27me3 marks on developmental genes, thereby allowing for transcription initiation, which ultimately leads to increased CM proliferation and heart regeneration (Extended Data Fig. 10d).

Discussion

CM proliferation has long been a focal point of research owing to its potential in regenerating lost CMs after heart injury⁵⁸. Our study underscores the value of comparing regenerative and non-regenerative species to unveil critical mechanisms and regulators of CM proliferation. Our findings highlight the pivotal role of Hmga1a in zebrafish as both necessary and sufficient for promoting CM proliferation and regeneration. Furthermore, analysis of epigenetic modifications suggests that Hmga1a acts on the chromatin by effectively removing repressive H3K27me3 marks on genes typically constrained to developmental stages, thereby inducing a pro-regenerative gene expression program. In contrast, Hmga1 expression in mammals is restricted to neonatal hearts during the regenerative window, with its absence in the adult injured heart. Our research reveals that introducing Hmga1 expression in the injured adult mouse heart activates a regenerative response, leading to CM proliferation specifically within the BZ. Moreover, a reduction in repressive H3K27me3 marks could be observed in BZ CMs upon Hmga1 OE, reminiscent of that found in zebrafish.

Diverse roles for Hmga1-driven chromatin organization and gene regulation have been postulated, showing both activating and repressive effects in different contexts^{43,44,45,46,65,66,67,68}. One proposed mechanism by which Hmga1 may influence chromatin state and gene expression is by displacing the linker histone H1 from chromatin^44,55,56. Histone H1 promotes PRC2-mediated H3K27me3 enrichment and chromatin condensation^69,70. The displacement of histone H1 results in chromatin opening and a reduction in repressive H3K27me3 marks⁴⁷. Furthermore, after embryonic development, overall H3K27me3 levels rise, which is linked to silencing of developmental genes and enabling cell differentiation⁷¹. In addition, PRC1/2-mediated repression is known to target transcription factors with crucial roles in development, such as those from Gata and Tbx transcription factor families⁷². Together, this aligns with our in vivo observation where Hmga1 reduces H3K27me3 marks from chromatin in both zebrafish and mouse CMs, leading to the re-expression of embryonic genes, the reactivation of the cell cycle and ultimately heart regeneration. Indeed, the injection of a virus expressing EZH1, a catalytic subunit of the PRC2 complex, in mouse hearts after an MI results in a genome-wide reduction of H3K27me3 and an increase in H3K27me1 chromatin marks in CMs, promoting heart regeneration⁷³. However, expression of a mutant histone H3.3^k27m in adult zebrafish heart impairs heart regeneration independently from CM cell cycle re-entry⁷⁴, likely due to a reduction in both H3K27me1 and H3K27me3. It is worth noting that prolonged depletion of H3K27me3 from developmentally regulated genes may affect not only proliferation but also CM redifferentiation and maturation, crucial for successful regeneration^16,26. The precise mechanisms conferring the specificity of Hmga1 in gene regulation remain to be fully elucidated, but they may involve specific post-translational modifications, such as phosphorylation and acetylation^42,67,73,74, and interactions with chromatin transcriptional regulators, such as chromatin remodeling complexes and transcription factors^42,67,75.

The introduction of Hmga1 in mouse hearts via intramyocardial AAV9-mediated delivery resulted in CM cell cycle re-entry in the BZ but not in transduced CMs in the RZ or uninjured hearts. In contrast, Hippo inactivation⁹ or ErbB2 activation¹¹ induces CM cell cycle re-entry also in remote myocardium and even in uninjured mouse hearts. The BZ-specific effect of Hmga1-induced CM proliferation might be aided by the general susceptibility of mouse BZ CMs to mitogenic stimuli^10,35,60,61. In addition, due to the use of the CMV promoter Hmga1, transduction of non-CMs and subsequent cell–cell signaling might contribute to the induced CM proliferation, although low non-CM transduction rate makes this unlikely. Furthermore, differences in immune response and fibroblast activation in the BZ versus the remote myocardium might provide a permissive microenvironment for the pro-proliferative effect of Hmga1 (refs. ^76,77). This distinction holds promise in clinical applications, as Hmga1-induced proliferation would naturally cease once the permissive BZ microenvironment has resolved³⁰, preventing uncontrolled CM proliferation, cardiomegaly and adverse remodeling, issues that have plagued previous strategies^{11,13,14,15,18}. Remarkably, Hmga1 OE is sufficient to improve cardiac function of the left ventricle (LV), underscoring its pro-regenerative effect in the mammalian heart.

Finally, we acknowledge several limitations of our study. Although sortChIC provided valuable insight into epigenetic marks in the context of ectopic hmga1a/Hmga1 expression in zebrafish as well as mouse, it did not allow us to directly compare histone marks on BZ CMs in zebrafish and mouse due to technical constraints. Additionally, to fully elucidate the role of Hmga1 in epigenetic regulation, we need to optimize techniques to map Hmga1 binding sites in zebrafish as well as mouse CMs, which is a priority for future experiments. Finally, although we demonstrated that Hmga1 OE stimulates CM proliferation only in the context of an injury, this suggests that an injury-induced signal cooperates with Hmga1 to reduce H3K27me3 marks and induce CM proliferation. Identifying the nature of this injury-induced signal will be key to understanding the full mechanism by which Hmga1 drives cardiac regeneration.

In conclusion, this study underscores the pivotal role of Hmga1 in driving CM proliferation and regeneration in both zebrafish and mice, offering promising prospects for regenerative therapies.

Methods

Animal experiments

All procedures involving animals were approved by the Animal Welfare Body of the Royal Dutch Academy of Sciences and Arts and were performed in compliance with animal welfare laws, guidelines and policies, according to national and European law.

Zebrafish and mouse lines

The following zebrafish lines were used: TL, TgBAC(nppa:mCitrine)⁵, Tg(myl7:CreER)^pd10 (ref. ²³), Tg(myl7:DsRed2-NLS)⁷⁹, Tg(myl7:LATdTomato) and Tg(β-actin2:loxPmTagBFP-STOP-loxP-Nrg1)⁵². Both males and females were used for zebrafish experiments. The khdrbs1a^−/− and znfx1^−/− were produced using CRISPR–Cas9-based strategies to introduce, respectively, a 14-base pair (bp) deletion in exon 3 of khdrbs1a and, for znfx1, a 4-bp deletion in exon 1, causing a frameshift and pre-mature stop codon. The hmga1a^−/− was produced using a TALEN-based strategy, targeting the region adjacent to the transcription start site. The resulting 8-bp deletion directly after the start codon caused a frameshift and introduction of a pre-mature stop codon, resulting in a truncated Hmga1a protein (7 amino acids (aa) instead of 101 aa) (Extended Data Fig. 2e). The Tg(ubi:Loxp-stop-Loxp-hmga1a-eGFP) was produced using gBlocks and Gibson Assembly, using the pDESTp3A destination vector⁸⁰ and the p5E ubi promotor⁸¹.

The following mouse lines were used: C57BL/6J males (Charles River Laboratories) for all experiments with AAV9-mediated delivery of GFP or Hmga1-HA, Tg(Nppb:katushka)³⁷ for TOMO-seq experiments and C57BL/6N males for experiments with AAV9 Myo4a delivery.

Cryoinjuries in zebrafish

To induce cardiac injury in zebrafish, cryoinjuries were performed on fish of approximately 4–18 months of age. The cryoinjuries were performed as described in ref. ⁸², with the exception of the use of a copper filament (0.3 mm) cooled in liquid nitrogen instead of dry ice. Animals were excluded from the study in case of signs of aberrant behavior/sickness/infection, according to humane endpoints in animal guidelines.

MI in mice

Experiments with intracardiac injections of AAV9:

Cardiac ischemic injuries were accomplished by permanent occlusion of the LAD, previously described in ref. ³⁷, using adult male mice between 7 weeks and 12 weeks of age. After the mice were anesthetized with a mix of fentanyl (0.05 mg kg⁻¹), midazolam (5 mg kg⁻¹) and dexmedetomidine (0.125 mg kg⁻¹) via intraperitoneal injections, a tracheal tube was placed, and the mouse was connected to a ventilator (Uno Microventilator, UMV-03). Hair was removed from the thorax and neck with VEET (‘silk and fresh gevoelige huid’) hair removal. The surgical site was cleaned with iodine and 70% ethanol. Using aseptic techniques with sterile instruments, the skin was incised left of midline to allow access to the third intercostal space (Tough Cut Scissors and Delicate Moria (MC31) forceps). Pectoral muscles were retracted, and the intercostals muscles were cut caudal to the third rib (2× SuperGrip Forceps, Angled). Wound hooks were placed to allow access to the heart. The pericardium was incised longitudinally, and the LAD was identified. A 7–0 silk suture was placed beneath the LAD for MI, followed by intracardiac injection of 2 × 15 µl AAV9. Sham animals only received the intracardiac AAV9 injections. The rib cage was closed with a 5–0 silk suture, and the skin was closed with a wound clip. The animal was disconnected from the ventilator; the tracheal tube was removed; and the animal was placed unrestrained on a nose cone with 100% oxygen in a warm recovery cage until fully ambulatory, at which point the oxygen was turned off. During the whole surgery and recovery, mice were placed on a heating pad of 38–39 °C. To alleviate pain or discomfort, mice were injected subcutaneously with 0.05–0.1 mg kg⁻¹ buprenorphine 30 min before surgery as well as 8–12 h after the surgery and 24 h after the surgery.

Experiments with intravenous injections of AAV9Myo4a:

Seven- to nine-week-old male C57BL/6N mice were used for experiments. Mice were injected subcutaneously with buprenorphine (0.075 mg kg⁻¹) and carprofen (0.05 mg kg⁻¹; Rimadyl Cattle) for analgesia at least 30 min before surgery. Anaesthesia was induced with 4% isoflurane in 1 l min⁻¹ O₂. Mice were shaved, intubated and placed on a heating pad to maintain body temperature. Intubation was connected to a ventilator (Harvard Apparatus, MiniVent Model 845), and hair removal cream was applied on the surgery area. Subsequently, ropivacaine (3 mg kg⁻¹) was applied subcutaneously at the site of the incision as analgetic. Anaesthesia was maintained using ventilation with 2% isoflurane in 1 l min⁻¹ O₂. Left thoracotomy was performed to expose the heart at the third intercostal space. The LAD was identified, and a 8–0 polyamide 6 suture was used to make a permanent ligation for MI. The thoracotomy and skin were closed with a 6–0 reverse cutting needle with a polyamide 6 wire (Ethilon). Post-surgery analgesia consisted of 4 days of ad libitum carprofen (0.06 mg ml⁻¹) in drinking water and high caloric wet food.

Transthoracic echocardiography

Two-dimensional transthoracic echocardiography was performed on sedated (1–2% isoflurane) mice to address heart function, using a VisualSonics ultrasound system with a 30-MHz transducer (VisualSonics). The heart was imaged in a parasternal long-axis as well as short-axis view at the level of the papillary muscles, to record B-mode as well as M-mode measurements and to determine heart rate, wall thickness and end-diastolic and end-systolic dimensions. Fractional shortening (defined as the end-diastolic dimension minus the end-systolic dimension normalized for the end-diastolic dimension) as well as EF (defined as the stroke volume normalized for the end-diastolic volume), cardiac output (defined as amount of blood pumped per minute) and stroke volume (defined as end-diastolic volume minus end systolic volume) were used as an index of cardiac contractile function.

Virus in neonatal rat CMs

Ventricular myocytes of 1-day-old neonatal rat hearts were isolated by enzymatic dissociation with trypsin (Thermo Fisher Scientific, 15400054) and cultured as described in ref. ⁸³. After 2 days of culturing, OE of GFP/Hmga1 was accomplished through lentivirus-mediated delivery of, respectively, pHAGE2- EF1a:GFP / pHAGE2- EF1a:Hmga1-T2A-GFP constructs. Cells were fixed and analyzed after two more days.

AAV9 production and injections in mice

Recombinant AAV9 vectors used in this study, carrying a CMV:GFP or a CMV:HA-Hmga1 cassette, prepared by the AAV Vector Unit at the International Centre for Genetic Engineering and Biotechnology Trieste (http://www.icgeb.org/avu-core-facility.html), as described previously⁸⁴ with a few modifications. In brief, infectious AAV vector particles were generated in HEK293T cells (American Type Culture Collection) cultured in roller bottles by a three-plasmid transfection cross-packaging approach, whereby the vector genome was packaged into AAV capsid serotype-9 (ref. ⁸⁵). Purification of viral particles was obtained by PEG precipitation and two subsequent CsCl₂ gradient centrifugations⁸⁶. The physical titer of recombinant AAVs was determined by quantifying vector genomes (vg) packaged into viral particles, by real-time PCR against a standard curve of a plasmid containing the vector genome⁸⁷; values obtained were in the range of 1 × 10¹³ to 1 × 10¹⁴ vg per milliliter. Directly after LAD ligation during MI surgery, hearts were injected twice with 15 μl of AAV9(CMV:GFP) or AAV9(CMV:HA-Hmga1) (1 × 10¹² virus particles per mouse) in opposing regions bordering the area at risk of ischemic injury.

For the ChIC experiment using AAV9 Myo4a, AAV9Myo4a-hTNNT2p-mCherry-p2A-H2BGFP and AAV9Myo4a-hTNNT2p-HA-mHmga1-p2A-H2BGFP were produced in the Academic Medical Center Amsterdam. One day before LAD, mice were injected into the retro-orbital sinus with 1 × 10¹¹ vg of AAV9Myo4a per animal. For this, animals were anesthetized in an induction chamber with of 4% isoflurane in 1 l min⁻¹ O₂. After being fully sedated, the animal was taken out of the chamber, and the thumb and index finger were used to pull back the skin around the eye socket until the eye slightly protruded. A 0.3-ml (30-gauge) × 8-mm U-100 insulin needle (BD Micro-Fine) was inserted at an angle of 45° starting around the medial canthus toward the retro-orbital sinus. The construct was slowly injected in one smooth motion in a maximum volume of 100 µl, after which the animal was placed on a heating pad until it fully regained consciousness.

EdU injections in mice

To assess cell cycle re-entry at 14 days after MI, adult mice received bi-daily intraperitoneal injections of EdU in PBS starting at day 2 (resulting in six EdU injections in total, per mouse). EdU concentrations were determined based on the individual weight of each mouse (50 μg g⁻¹).

Tamoxifen-induced hmga1a OE in zebrafish

To induce OE of Hmga1a in CMs specifically, the Tg(ubi:Loxp-BFP-stop-Loxp-hmga1a-eGFP) line was crossed to the Tg(myl7:CreER)^pd10 line. Adult zebrafish (4–12 months) received two overnight pulses of tamoxifen by swimming them in a 4 μM tamoxifen solution in E3 medium.

NRG1 injections in zebrafish

Intraperitoneal injections of human recombinant NRG1 (PeptroTech: recombinant human heregulin-b1, catalog number 100-03) were performed as described in ref. ⁸⁸. Fish were sedated using tricaine (0.0168%, w/v). Injections were performed using a Hamilton syringe (30-gauge), cleaned before use by washing in 70% ethanol, followed by two washes in PBS. Injection volumes were adjusted on the weight of the fish (30µl g⁻¹), and a single injection contained 60 µg kg⁻¹ of human recombinant NRG1 (diluted in PBS/BSA 0.1%).

Live imaging cardiac slices

Live imaging was performed as described previously⁸⁹. Hmga1a OE hearts were extracted 14 dpT, and imaging was performed using a Leica SP8 confocal microscope in a temperature-controlled chamber at 28 °C. z-stacks with a z-step size of 1 μm were acquired every 10 min.

Heart collection for histological analysis

In mice, mouse hearts were isolated and washed in PBS, after which they were fixed in 4% paraformaldehyde (PFA) (room temperature, overnight, on shaker), dehydrated through an ethanol series, embedded in paraffin and sectioned at 6 µm to use for histology.

In zebrafish, adult zebrafish ventricles were isolated and fixed in 4% PFA (4 °C, overnight, on shaker). The next day, the hearts were washed 3 × 10 min in 4% sucrose phosphate buffer, 5 h in 30% sucrose at room temperature and then embedded in cryo-medium (OCT). Cryo-sectioning of the hearts was performed at 10-µm thickness. For zebrafish paraffin sections, hearts were fixed in 4% PFA (4 °C, overnight, on shaker), followed by dehydration in ethanol series and subsequent paraffin embedding and sectioning at 10-μm thickness.

Human

Paraffin-embedded infarcted human heart tissue from three individuals who had died of MI were retrieved from the pathology archive of the University Medical Center Utrecht. Material was handled in a coded manner that met the criteria of the Code of Conduct used in The Netherlands for the responsible use of human tissue in medical research. Collection of the archive material was approved by the local biobank review committee (protocol 15–252).

ISH

In zebrafish paraffin sections, ISH was performed on paraffin sections as previously described⁹⁰ except that the hybridization buffer used did not contain heparin and yeast total RNA. Zebrafish cryosections: ISH was performed as for paraffin; however, sections were pre-fixed for 10 min in 4% PFA + 0.25% glutaraldehyde (Sigma-Aldrich) before Proteinase K treatment. Moreover, slides were fixed for 1 h in 4% PFA directly after staining. Slides were mounted in Entellan (Merck) mounting medium and imaged on either a DM4000 (Leica) or a VS200 Slide Scanner (Olympus). Zebrafish Dig probes were made via PCR for: hmga1a using 5′-TACTGTGTCTCGGGGCAAAA-3′ and 5′-GAGtaatacgactcactatagggACCCTTTGAGTGCGAGACAT-3′, hmga1b using 5′-CCCATCCAAGAGAAAATCATCGA-3′ and 5′-GAGtaatacgactcactatagggAAGCACCTCAGTCCAATTTAGA-3′, hk1 using 5′- TGGGTGGCTCTAATTTCCGT-3′ and 5′- GAGattaaccctcactaaagggaAGAGGCATACACTTTGGGCT-3, znfx1 using 5′-AAATGCTGTCCACCGTCCTA-3′ and 5′-GAGattaaccctcactaaagggaCATGGAGACGGAATGCACAG-3′, khdrbs1a using 5′-ATTTCCACCATCGCTCTCCA-3′ and 5′-GAGtaatacgactcactatagggACAGTCAGGAATGGGAGCAA-3′ and tbx20 using linearized plasmid. Mouse Dig probes were made via PCR for: Hmga1 using 5′-GGGAAGCAAGAATAAGGGCG-3′ and 5′-GAGtaatacgactcactatagggAAAACAAAGCGCCCAGAGAG-3′, Rhoc using 5′-CCGAAAGAAGCTGGTGATCG-3′ and 5′-GAGtaatacgactcactatagggGTGGCCATCTCAAACACCTC-3′, Fstl1 using 5′-GCCGAGGAAGAGCTAAGGAG-3′ and 5′-GAGtaatacgactcactatagggGAGCTCATCACGGTTGGACT-3′, Tmsb4x using 5′-CCGCCAATATGCACTGTACA-3′ and 5′-GAGtaatacgactcactatagggTGGCACTCTGATTAAACTGCA-3′, Nppa using 5′-GCATTCCAGCTCCTAGGTCA-3′ and 5′-GAGtaatacgactcactatagggTCAGTACCGGAAGCTGTTACA-3′, Des using 5′-GAGCTGCTGGACTTCTCACT-3′ and 5′-GAGtaatacgactcactatagggTCATACTGAGCCCGGATGTC-3′, Ankrd1 using 5′-GGGGAGCAACAGTGGAAAAG-3′ and 5′-GAGtaatacgactcactatagggTCCTTCTCTGTCTTTGGCGT-3′, Tnnt2 using 5′-TGTTGAAGAGCAGGAGGAGG-3′ and 5′-GAGtaatacgactcactatagggCTCCTTCTCCCGCTCATTCC-3′ and Ech1 using 5′-GGGATAGTGGCTTCTCGCAG-3′ and 5′-GAGattaaccctcactaaagggaGATAGCCGCAGACTCACCTC-3′.

qPCR

Total RNA was isolated from heart ventricles using TRIzol reagent (Life Technologies) according to the manufacturer’s instructions. Total RNA (1 μg) was reverse transcribed using an iScript cDNA Synthesis Kit (Bio-Rad). Real-time PCR was performed using an iQ SYBR Green kit and a CFX96 real-time PCR detection system (Bio-Rad). Data were normalized using reference genes Gapdh (Fig. 3c) or Hprt and Eefe1e (Extended Data Fig. 2c).

Western blot

Protein samples (20 µg per sample) were separated on a 15% SDS-PAGE gel (separation gel: 40% acrylamide, 2% Bis solution, 3 M Tris pH 8.8, MQ, 10% SDS, 10% APS, 1% TEMED; stacking gel: 40% acrylamide, 2% Bis solution, 3 M Tris pH 6.8, MQ, 10% SDS, 10% APS, 1% TEMED). Protein was transferred to methanol-activated PVDF membrane using semi-dry electroblotting with an Amersham Imager 600 (65 mA, 1.5 h). PVDF membrane was incubated in blocking solution (5% BSA in 0.1% TBS-T) for 30 min, after which the membrane was incubated in blocking solution with primary antibody (Abcam Rb mAb HMGA1, AB129153, 1:1,000, and Calbiochem anti-α-tubulin mouse mAb, CP06-100UG, 1:1,000) at 4 °C overnight. PVDF membrane was incubated with secondary antibody (BD Pharmingen HRP goat anti-rabbit, 554021, 1:10,000, and BD Pharmingen HRP goat anti-mouse, 554002, 1:10,000) for 1 h at room temperature. PVDF membrane was imaged using SuperSignal West Pico Chemiluminescence Signal (Thermo Fisher Scientific) on an Amersham ImageQuant 800 western blot imaging system.

Immunohistochemistry

On zebrafish cryosections, antigen retrieval was performed by heating slides containing heart sections at 85 °C in 10 mM sodium citrate buffer (pH 6) for 15 min. Primary antibodies used included anti-PCNA (Dako, M0879, 1:800), anti-GFP (Aves Labs, GFP-1010, 1:1,000), anti-Mef2c (Santa Cruz Biotechnology, sc313, and Biorbyt, orb256682, both 1:1000), anti-phosphor-S6 ribosomal protein (ser235/236) (Cell Signaling Technology, 4858, 1:500), anti-tri-methyl-histone H3 (Lys27) (C36B11) (Cell Signaling Technology, 9733T, 1:300), anti-α-actinin (Sigma-Aldrich, A7811, 1:500) and anti-tropomyosin (Sigma-Aldrich, T9283, 1:500). On mouse paraffin sections, antigen retrieval was performed by heating slides containing heart sections under pressure at 120 °C in 10 mM sodium citrate buffer (pH 6) for 1 h. Primary antibodies used included anti-HA (Abcam, ab9111, or BioLegend, 901501, both 1:200), anti-PCM (Atlas Antibodies, HPA023370, 1:400), anti-tri-methyl-histone H3 (Lys27) (C36B11) (Cell Signaling Technology, 9733T, 1:300), anti-Ki-67 (SolA15) (eBioscience Invitrogen, 15227437, 1:1,000), anti-Aurora B (AIM1) (BD Biosciences, 611082, 1:300), anti-mouse CD45 (BD Biosciences, 53076, 1:200) and anti-rabbit CD68 (Abcam, ab125212, 1:200). EdU was visualized with a Click-iT EdU Cell Proliferation Imaging Kit, Alexa Fluor 647 (Thermo Fisher Scientific, C10340), according to the instructions. For both mouse and zebrafish tissue, secondary antibodies included anti-chicken Alexa Fluor 488 (Thermo Fisher Scientific, A11039, 1:500), anti-mouse Alexa Fluor 488 (Thermo Fisher Scientific, A21133, 1:500), anti-rabbit Alexa Fluor 555 (Thermo Fisher Scientific, A21428, 1:500), anti-mouse Alexa Fluor 555 (Thermo Fisher Scientific, A21127, 1:500), anti-mouse Alexa Fluor 633 (Thermo Fisher Scientific, A21050, 1:500) and anti-mouse Cy5 (Jackson ImmunoResearch, 118090, 1:500). Phalloidin (Merck, P1951, 1:200) was used as an actin marker. Wheat gluten albumin (WGA) was used as a membrane marker (Merck, 100 µg ml⁻¹, 30 min). Nuclei were shown by DAPI (Invitrogen, D1306, 1:1,000). Images of immunofluorescence staining are single optical planes acquired with a VS200 Slide Scanner (Olympus), an Sp8 microscope (Leica) or an LSM900 AiryScan (Zeiss).

Acid fuchsin orange G staining

Zebrafish paraffin sections were rehydrated in dH₂O and post-fixed with Bouin’s solution (Sigma-Aldrich) at 60 °C for 2 h. Slides were cooled and rinsed under running water, followed by incubation with 1% phosphomolybdic acid for 2 min, washed in dH₂O and stained in acid fuchsin orange G (AFOG) solution (1:1:1 ratio of analine blue, orange G and acid fuchsin, pH 1.09) for 2 min, rinsed in dH₂O, dehydrated in an ethanol series and mounted in Pertex (Sigma-Aldrich). Slides were imaged on the VS200 Slide Scanner (Olympus).

TOMO-seq

Under a fluorescence stereoscope, injured mouse hearts were isolated, and tissue was selected based on the Katushka signal. Tissue was isolated from the injured hearts (n = 3) containing part of the injury, Katushka signal and part of the remote myocardium, respectively. TOMO-seq was conducted as previously described³⁶. Mapping was performed against the zebrafish reference assembly version 9 (Zv9) and the mouse reference assembly version 9 (mm9). Analysis was done based on the log₂-transformed fold change (zlfc) of the z-score (number of standard deviations above the mean) of all genes. Bioinformatic analyses were largely performed with R software using custom-written code. Hierarchical cluster analysis on the entire dataset (after z-score transformation) was performed on all genes with a peak in more than four consecutive sections (z-score > 1). Based on hierarchical clustering analysis, together with maker gene expression (BZ markers Nppa, Nppb and Des), we defined the locations of the IA, the BZ and the RZ within our datasets. Injured zebrafish hearts were isolated and processed as described in ref. ²⁴ to obtain IA, BZ and RZ. To transcriptionally compare the zebrafish and mouse BZ, we first pooled all IA, BZ and RZ regions from the different timepoints into one species-specific dataset per species, resulting in 14 (IA), 43 (BZ) and 43 (RZ) sections in the zebrafish dataset and 14 (IA), 65 (BZ) and 54 (RZ) sections in the mouse dataset. GO analysis was performed on these combined lists using the R package edgeR, which uses generalized linear models (GLMs) and empirical Bayes methods to identify differentially expressed genes⁹¹. These gene lists were subjected to GO analysis using the online tool DAVID⁷⁸. The transcriptional comparison between the zebrafish and the mouse BZ was performed by plotting the logFC (BZ versus the rest of the tissue) of annotated homologs (11,779 gene pairs) annotated in Ensembl (version 89) in a scatterplot (Fig. 1a,b). Genes with no annotated homolog were excluded from analysis. Genes with multiple annotated homologs were plotted as separate gene pairs. Next, gene pairs were selected using the following thresholds. Upregulated in both the mouse and zebrafish BZ (331 gene pairs): zebrafish logFC > 0.5, P < 0.05; mouse: logFC > 0.5, P < 0.05. Downregulated in both the mouse and zebrafish BZ (326 gene pairs): zebrafish: logFC < −0.5, P < 0.05; mouse: logFC < −0.5, P < 0.05. Upregulated in the zebrafish BZ but not the mouse BZ (371 gene pairs): zebrafish: logFC > 0.5, P < 0.05; mouse: logFC < 0. Upregulated in the mouse BZ but not the zebrafish BZ (366 gene pairs): zebrafish: logFC < 0; mouse: logFC > 0.5, P < 0.05. After determining zebrafish-specific and mouse-specific gene pairs, gene pairs were removed, of which at least one paralogous gene showed expression outside of the selection thresholds, accounting for redundant functions between paralogous genes. These gene pairs were subjected to GO analysis using their mouse name in DAVID. All datasets can be accessed via the TOMO-seq website (http://mouse.genomes.nl/tomoseq/2018).

scRNA-seq

TgBAC(nppa:mCitrine)^hu8889Tg-positive cells showing high mCitrine expression were isolated from cryo-injured zebrafish hearts (7 days after injury). From 12 hmga1a^−/− hearts, 768 cells were isolated. From 12 hmga1a^+/+ wild-type hearts, 768 cells were isolated. Single-cell sequencing libraries were prepared by Single Cell Discoveries using the SORT-seq protocol⁹². The CEL-Seq2 protocol was used for library preparation⁹³. Illumina sequencing libraries were prepared with TruSeq small RNA primers (Illumina) and paired-end sequenced at 75-bp read length on an Illumina NextSeq platform. In total, seven 384-well plates were sequenced containing one cell per well, of which three were obtained from wild-type cells and four from hmga1a^−/− cells. Mapping was performed against the zebrafish reference genome assembly version 9 (Zv9). Based on the distribution of the log₁₀ total reads plotted against the frequency, we introduced a cutoff at minimally 600 reads per cell to be included for further analysis, which left us with a total of 1,310 cells (653 wild-type and 657 hmga1a^−/− cells). Next, single-cell data were analyzed using Seurat⁹⁴. The following parameters were used: variable features = all genes, dimensions = 8 and resolution = 0.7. This resulted in the identification of six clusters that were plotted in a two-dimensional UMAP. The Seurat object was subsequently subjected to RNA velocity⁴⁹. For this, reads were re-mapped using STAR mapping⁹⁵ to generate a dataset where intronic and exonic reads were separated. RNA velocity subsequently used the ratio of these reads to generate a vector that predicts the future state of a cell. These velocity vectors were plotted on the Seurat UMAP. To confirm findings from RNA velocity, we next subjected our cells to pseudo-temporal ordering using Monocle 2 (ref. ⁹⁶). Monocle 2 was used to identify genes that were differentially expressed over pseudo-time and organized in a self-organizing map with eight modules. Pseudo-temporal values assigned to single cells were integrated in our Seurat object, and the pseudo-temporal ordering was plotted on the UMAP. DAVID was used to perform GO analysis for the eight modules.

Bulk RNA-seq and ChIC-seq

Heart isolation was performed on whole zebrafish hearts 14 days after tamoxifen treatment, after which hearts were dissociated into single cells. For RNA-seq, fish either contained three transgenes allowing CM-specific OE of hmga1a-eGFP Tg(ubi:Loxp-stop-Loxp-hmga1a-eGFP), Tg(myl7:DsRed2-NLS) and Tg(myl7:CreER)^pd10 or formed the control fish containing only two transgenes (Tg(myl7:DsRed2-NLS) and Tg(myl7:CreER)^pd10). For ChIC experiments, fish contained Tg(myl7:LiveAct-TdTomato) instead of Tg(myl7:DsRed2-NLS) because of fluorophore compatibility with the sortChIC pipeline. Use of these lines allows fluorescence-activated cell sorting (FACS) of CMs after antibody incubation.

For bulk RNA-seq, 4 × 1,000 DsRed⁺ cells were sorted from pooled Cre-only control (n = 14) hearts and 4 × 1,000 DsRed⁺ cells from pooled hmga1a–eGFP (n = 14) hearts. Bulk RNA-seq was performed by Single Cell Discoveries. Cells were lysed in TRIzol; RNA was extracted; and libraries were prepared and sequenced on the Illumina platform. FASTQ files were mapped with the STARandGO pipeline (https://github.com/anna-alemany/VASAseq/blob/main/mapping/map_star.sh) against the danRer11 Ensembl genome with the zebrafish Lawson V4.3.2 annotation. Normalization and downstream analysis were performed in R. Owing to low read count, 1/4 control and 1/4 OE technical replicates were excluded from analysis. Using the R package edgeR, differentially expressed genes were obtained (FC < −1 or FC > 1 and P < 0.05). Gene lists were subjected to GO analysis using the online tool DAVID.

Bulk sortChIC-seq was performed by Single-Cell Core (Oncode Institute). In short, for zebrafish experiments using bulk sortChIC⁵⁷, single-cell suspensions from Cre-only control (n = 10) and hmga1a–eGFP (n = 10) hearts were subjected to the sortChIC protocol and incubated with antibodies against H3K4me3 (Invitrogen, MA-5-11199, 1:400), H3K9me3 (Invitrogen, MA5-33395, 1:200) or H3K27me3 (Cell Signaling Technology, C36B11, 1:200). The next day, cells were FACS sorted based on LiveAct-TdTomato⁺ post-antibody incubation in tubes of 100 cells. ChIC-seq libraries were subsequently prepared on sorted cells. Data pre-processing was performed using the SingleCellMultiOmics package (https://github.com/BuysDB/SingleCellMultiOmics), and the sequences were mapped to the zebrafish danRer11 Ensembl genome using Burrows–Wheeler Aligner (BWA). Data were uploaded to the Galaxy web platform, and the public server at https://usegalaxy.org/ was used to analyze the data. Reads were normalized to counts per million (CPM). Minimum mapping quality threshold was put at 50, and duplicates and quality control fails were filtered out (flag 1536) during BAM to BigWig conversion. To integrate ChIC-seq data with RNA-seq data, locations of all zebrafish genes were obtained from Ensembl BioMart, danRer11. For the mouse ChIC-seq experiment, BZ regions were manually excised from 14-dpi hearts of control mice (n = 4) and Hmga1 OE mice (n = 3), and CM nuclei isolation was performed as described previously³⁰. Single-nuclei suspensions were subjected to the sortChIC protocol and incubated with H3K27me3 antibody (Cell Signaling Technlology, 36B11, 1:200). The next day, CM nuclei were FACS sorted on GFP to select transduced CMs (1,000 cells per sample) and used for ChIC-seq library preparation. Data pre-processing was performed using the SingleCellMultiOmics package (https://github.com/BuysDB/SingleCellMultiOmics), and the sequences were mapped to the Mus musculus GRC38m38.p6 Ensembl genome using BWA. Data were uploaded to the Galaxy web platform, and the public server at https://usegalaxy.org/ was used to analyze the data. Reads were normalized to CPM. Minimum mapping quality threshold was put at 50, and duplicates and quality control fails were filtered out (flag 1536) during BAM to BigWig conversion.

Statistics and reproducibility

All experiments were performed with at least three biological replicates per group and were performed twice with similar results, with some exceptions. Mouse TOMO-seq, zebrafish scRNA-seq, zebrafish/mouse ChIC-seq, mouse qPCR for Hmga1 in neonatal samples and long-term hmga1a OE experiments were performed once.

All data were quantified in a double-blinded fashion. Statistical testing methods are indicated in the description of each figure. Statistical analyses were performed using GraphPad Prism (GraphPad Software). Histological quantifications of CM proliferation in zebrafish were performed using Imaris 64 (version 3.2.1) software (Oxford Instruments) and were performed either in the BZ, which was defined as 200 μm from the wound border, or throughout the ventricle for uninjured contexts, in both cases on three sections per heart, of at least three hearts. Quantifications of scar size on AFOG-stained zebrafish hearts were performed by staining all sections of each heart and measuring the remaining scar area in ImageJ software (National Institutes of Health). pS6 signal in zebrafish hearts was quantified using both Imaris 64 (version 3.2.1) software tools to mask tropomyosin and obtain CM-specific pS6 signal and ImageJ software to measure the intensity and area of the pS6⁺/tropomyosin⁺ area in the BZ (300 μm). H3K27me3 staining was quantified for both zebrafish and mouse sections using ImageJ software to measure signal intensity in individual nuclei. Quantification of cell size using WGA staining was performed using ImageJ to measure the surface area of transversely cut CMs. Quantification of CM proliferation in NRVMs was performed using Imaris 64 (version 3.2.1) software, counting every CM across three slides per condition, representing biological replicates. CM proliferation in mouse heart sections was performed using Imaris 64 (version 3.2.1) software, where the BZ was defined as 500 μm from the wound border, and at least three squares of 500 μm × 500 μm were selected for quantification in the BZ and one for quantification in the RZ. Analysis of echocardiography in the mouse was performed using Vevo software to measure the end-diastolic volume and the end-systolic volume, using Simpsonʼs method. Quantification of scar size on mouse heart sections at 42 dpi was performed using ImageJ software to obtain a midline length measurement readout (% MI midline of total LV midline⁹⁷).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.