Loss of the Polycomb group protein Rnf2 results in derepression of tbx-transcription factors and defects in embryonic and cardiac development

The Polycomb group (PcG) protein family is a well-known group of epigenetic modifiers. We used zebrafish to investigate the role of Rnf2, the enzymatic subunit of PRC1. We found a positive correlation between loss of Rnf2 and upregulation of genes, especially of those whose promoter is normally bound by Rnf2. The heart of rnf2 mutants shows a tubular shaped morphology and to further understand the underlying mechanism, we studied gene expression of single wildtype and rnf2 mutant hearts. We detected the most pronounced differences at 3 dpf, including upregulation of heart transcription factors, such as tbx2a, tbx2b, and tbx3a. These tbx genes were decorated by broad PcG domains in wildtype whole embryo lysates. Chamber specific genes such as vmhc, myh6, and nppa showed downregulation in rnf2 mutant hearts. The marker of the working myocard, nppa, is negatively regulated by Tbx2 and Tbx3. Based on our findings and literature we postulate that loss of Rnf2-mediated repression results in upregulation and ectopic expression of tbx2/3, whose expression is normally restricted to the cardiac conductive system. This could lead to repression of chamber specific gene expression, a misbalance in cardiac cell types, and thereby to cardiac defects observed in rnf2 mutants.

mono-ubiquitinylates H2AK119 (H2AK119ub), via its RING-domain and this mark stabilizes H3K27me3 6 . Both H3K27me3 and H2AK119ub are epigenetic marks associated with transcriptional repression [7][8][9][10] . H3K27me3 represses gene expression by changing the chromatin structure and by antagonizing the H3K27ac mark, which is a mark known to be present at active enhancers 10,11 . Because H3K27me3 and H3K27ac reside at the same amino acid of the same histone tail, they are mutually exclusive. Furthermore, recent studies showed that the recruitment of PRC1 to the chromatin can also be H3K27me3-independent 12,13 . PRC1 variants that contain subunits which have a DNA-binding domain are described to be involved in this process 14 . H3K27me3-independent recruitment of PRC1 can repress gene expression by different mechanism: by condensation of the chromatin structure, by preventing RNA polymerase II elongation, and by recruiting PRC2 13,[15][16][17] . However, much still remains unknown about the interplay between PRC1 and H3K27me3.
Studies using systems in which PRC1 is disrupted indicate a crucial role in cellular differentiation across species. Mice have two homologous of the RING-domain containing proteins that both can assemble in PRC1: Rnf2 (Ring1b) and Ring1 (Ring1a) 18 . Loss of Rnf2 in mice results in developmental arrest during gastrulation 19 . Murine Ring1 homozygous mutants are viable 20 , and similar to PcG mutants in Drosophila, Ring1 heterozygous mice display homeotic transformations and skeletal defects 21 . In mice, the loss of Ring1/Rnf2 postnatally results in dental defects, but no lethality, when the mice are studied up to 17 days 22 . Additionally, studies in mouse embryonic stem cells showed that Rnf2 and Ring1 are essential for maintaining cells in a pre-mature state, by repressing genes involved in differentiation pathways 23,24 . In zebrafish, only one Ring1 orthologue is identified, which shows most homology with Rnf2 3 . Therefore, ablation of Rnf2 in zebrafish results in loss of functional PRC1 and the H2AK119ub mark 8 . Zinc-finger nuclease induced rnf2 null-mutant zebrafish embryos and rnf2 morphant embryos gastrulate normally, which makes it possible to study development in the absence of Rnf2 8,25 . Rnf2 morphants have an overall normal morphology and, although their primitive erythropoiesis was largely unaffected, the number of hematopoietic stem and thrombocytes was shown to be smaller at 36 hpf 25 . An rnf2 mutant allele has been generated, and the rnf2 mutation results in pre-mature stop codon. These rnf2 null-mutant zebrafish embryos show lethality around 4-5 dpf and display defects in terminal differentiation of the pectoral fins, likely due to interference with Fgf-signaling 8 . In addition, it was found that Cranial Neural Crest (CNC) cells do not properly differentiate into chondrocytes in rnf2 mutants, resulting in cartilage malformation in the head 26 . These defects in pectoral fin and chondrocyte development upon loss of Rnf2 both arise during terminal tissue differentiation.
To study the role of PRC1 and PRC2 during embryogenesis is challenging due to lethality of mutants in many species before gastrulation 19,27 . Therefore, in this study, rnf2 mutant zebrafish embryos are used to investigate the effects of loss of Rnf2 on development by studying the transcriptome and correlate this to the Rnf2 binding pattern in wildtype embryos at 3 dpf. We find an important regulatory role for Rnf2 at the chromatin level. The loss of Rnf2 results in upregulation of the genes normally occupied by Rnf2; these include genes associated with transcriptional regulation. In order to gain insight in a tissue specific role of Rnf2 we studied the heart in more detail. Transcriptome analysis of single hearts of wildtype and rnf2 mutant embryos at 1, 2, and 3 dpf indicates that at 1 and 2 dpf the transcriptional difference between wildtype and rnf2 mutant hearts are minor and at 3 dpf these differences are more prominent. At 3 dpf upregulation of transcription factors like tbx2a, tbx2b, and tbx3a was detected and, in addition, a downregulation of cardiac chamber genes, such as nppa was observed. We suggest that the upregulation of the tbx transcription factors is a direct consequence of the loss of Rnf2-mediated repression and we hypothesize that these transcription factors are responsible for the downregulation of chamber genes, resulting in malformation of the rnf2 mutant hearts. This finding sheds new light on the molecular mechanisms underlying heart development and the role of Rnf2 during vertebrate embryogenesis.

Results
Phenotypical differences between rnf2 mutant and wildtype zebrafish embryos. To gain additional insight in the role of PRC1 in development we used previously identified rnf2 ibl31/ibl31 (referred to as rnf2) mutant zebrafish, which harbor a mutation in the enzymatic subunit of PRC1. A 14 base pair deletion in the rnf2 gene, results in a premature stop codon, and the mutant embryos were shown to lack rnf2 gene expression at 3 dpf 8 . We observed the same pleiotropic phenotype in rnf2 mutants, as reported before 8 . This includes motility defects, defects in craniofacial development, the lack of pectoral fins, and a pronounced heart edema (Fig. 1a).
Immunohistochemistry for Rnf2 in wildtype siblings shows that expression of Rnf2 protein at 2 dpf is mainly detected anteriorly and in the notochord ( Supplementary Fig. S1, left panel). The rnf2 mutants lack Rnf2 protein at 2 dpf ( Supplementary Fig. S1, right panel).
To obtain more insight in the pleiotropic phenotype we assessed the expression of four organ markers by whole mount in situ hybridization (WISH) in wildtype and rnf2 mutant embryos at 3 dpf. Results from WISH for fatty acid-binding protein type 2 (fabp2), an intestinal marker, suggests a smaller intestine in the rnf2 mutants compared to the wildtypes (Fig. 1b). Expression of the liver specific marker fatty acid-binding protein type 10 (fabp10) is present in wildtypes, whilst it cannot be detected in rnf2 mutants. This suggests that liver terminal differentiation is abrogated in rnf2 mutants. The terminal differentiation marker of the exocrine pancreas, trypsin (try), is present in both wildtype and rnf2 mutant embryos. However, the shape of the exocrine pancreas is different in rnf2 mutants: the pancreatic lobe is not detected. Lastly, the expression of the cardiomyocyte marker myosin light chain 7 (myl7) was assessed. Wildtype embryos show pronounced myl7 expression in the atrium and the ventricle of the heart. Expression of myl7 is detected in rnf2 mutant embryos at 3 dpf; however, the expression pattern of myl7 in rnf2 mutants indicates malformation of the heart. The rnf2 mutant heart shows a stringy morphology and appears smaller based on the myl7 expression pattern (Fig. 1b).
Rnf2 binds the same targets as H3K27me3 and H3K27me3 deposition is present in rnf2 mutants.
To allow quantitative normalization and to demonstrate the efficiency of the method, we added Drosophila melanogaster spike-in chromatin during the ChIP-seq procedure 35 . After ChIP-seq, k-means clustering revealed five different classes of binding of Rnf2 and H3K27me3 at promoter regions (Fig. 2a). The first cluster represents Rnf2 and H3K72me3 positive promoters. The second, fourth, and fifth cluster contain promoter regions that are positive for H3K27me3 and show close to background levels for Rnf2. The third cluster contains broad PcG domains, in which both Rnf2 and H3K27me3 are present. The intensity of the peaks for Rnf2 and H3K27me3 was analyzed and visualized with bandplots (Fig. 2b). Rnf2 is present at the chromatin in wildtype embryos and is at around background levels in rnf2 mutants. The levels of H3K27me3 are similar in rnf2 mutants compared to wildtypes. H3K27me3 presence has been retained upon loss of Rnf2, which could suggest that its deposition does not rely on Rnf2 (Fig. 2b).

Loss of Rnf2 is associated with upregulation of genes decorated by broad PcG domains.
Presence of Rnf2 and H3K27me3 on the chromatin has a repressive effect on the underlying genes [7][8][9][10] . Therefore, we compared the transcriptome of wildtype embryos and rnf2 mutant embryos at 3 dpf. Our data shows both upand downregulated genes upon loss of Rnf2 (Fig. 3a). In total, 492 genes were found to be differentially expressed (LFC ≥ 1; padj ≤ 0.1). Of these, 292 were identified to be upregulated and 200 genes to be downregulated in rnf2 mutant embryos (Supplementary Table S1). Hierarchical clustering confirms good homology of the replicates (Euclidian distance; Supplementary Fig. S2a). The organ markers tested by WISH, were also studied in the whole embryo RNA-sequencing dataset. A downregulation of marker expression in rnf2 mutants was observed for fabp2, fabp10, and myl7. Expression of try was not found to be affected by the loss of Rnf2 ( Supplementary  Fig. S2b).
The promoters of the five different clusters identified after ChIP-seq ( Fig. 2a) have been linked to the genes they regulate. The expression of these genes was analyzed by Gene Set Enrichment Analysis (GSEA). We tested the differences in expression between rnf2 mutants and wildtypes in the five clusters (genes within the five clusters are enlisted in Supplementary Table S2). Genes belonging to the promoters of cluster 1, 2, 4, and 5 do not show significant enrichment for differential regulation upon the rnf2 mutation (p-value = 0.176, 0.679, 0.605, and www.nature.com/scientificreports www.nature.com/scientificreports/ 0.249, respectively). Cluster 3 contains 69 promoter regions, regulating 112 genes ('cluster 3 genes'). This cluster is significantly enriched with genes that are upregulated in the rnf2 mutant embryos at 3 dpf (p-value < 0.001) (Fig. 3b). Gene ontology analysis shows that these 'cluster 3 genes' are associated with regulation of transcription and embryonic organ development (Fig. 3c). Examples of tracks of two 'cluster 3 genes' are shown in Fig. 3d. These genes are decorated by Rnf2 and H3K27me3 and the RNA-seq tracks indicate a significant upregulation of gene expression in rnf2 mutants at 3 dpf. In total, 24 genes were identified in cluster 3 that were significantly upregulated in rnf2 mutant embryos which are decorated by Rnf2 in the promoter region of wildtype embryos at 3 dpf (LFC ≥ 1; padj ≤ 0.1) (Supplementary Table S3). www.nature.com/scientificreports www.nature.com/scientificreports/ The cardiac phenotype in rnf2 mutant zebrafish embryos shows looping defects. To gain insight into a tissue specific role of Rnf2 during development, we performed more extensive studies on the heart, which was one of the organs severely affected upon the rnf2 mutation. The heart is a well-studied organ in zebrafish, due to its similarity in cardiac development to other species and its regenerative capacities. Since we observe a severe cardiac phenotype upon loss of Rnf2, we studied a potential role of Rnf2-mediated regulation of heart development.
We analyzed the development of the heart of wildtype and rnf2 mutant embryos at 3 dpf in a Tg(myl7::GFP) background. This transgene specifically marks cardiomyocytes 36 . During normal cardiac development, cardiac www.nature.com/scientificreports www.nature.com/scientificreports/ looping ensures the proper positioning of the atrium and ventricle and it marks the transition from a linear heart tube to a two-chambered heart separated by the atrioventricular canal (AVC), which occurs between 28 and 50 hours post fertilization 37 . In contrast to the wildtype situation, the hearts of rnf2 mutant embryos display defective looping morphogenesis, resulting in a stringy heart phenotype with no clear defined cardiac chambers or AVC at 3 dpf (Fig. 4a). To investigate the underlying developmental dynamics, we used high speed selective plane illumination microscopy (SPIM) to time-lapse image wildtype and rnf2 mutant embryos from 1 to 2 dpf (Fig. 4b). Indeed, whereas wildtype embryos showed cardiac looping around 36 hpf the mutants fail to loop properly (Fig. 4b).
Single HeartsRNA-seq highlights differences between wildtype and rnf2 mutant hearts over time. We next aimed at getting a better understanding of the molecular mechanisms underlying the heart defects detected in rnf2 mutants. To start with, we show that disturbing the epigenetic repressor Rnf2 shows a global effect on gene expression, especially of genes decorated by Rnf2 in wildtype embryos (Fig. 3a). In order to address this finding and in connection with the observed cardiac phenotype, we took genes positive for Rnf2 in their promoter, as identified by ChIP-seq (n = 206), and all genes detected by RNA-seq (n = 32,266) and searched for the presence of heart transcription factors (n = 18) within these two groups 38 , finding a significant enrichment of this category (chi-squared test; p-value < 0.001).
To gain more detailed insight into the role of Rnf2 in cardiac development, single embryonic hearts, both from rnf2 mutant and wildtype, serve as a useful model. The transcriptome of single hearts dissected from wildtype and rnf2 mutants was assessed, using a low-input RNA-seq method based on CEL-seq [39][40][41][42] . This method is very suitable for the number of cells present in a single embryonic heart, which is between 150 and 350 cells at 1 to 3 dpf 41,43,44 . We performed manual dissection of single embryonic hearts at 1, 2, and 3 dpf and prepared the individual rnf2 mutant and wildtypes dissected hearts for RNA-sequencing (Single HeartsRNA-seq, Fig. 5a). In total 63 single hearts were sequenced ( Supplementary Fig. S3a). Samples were filtered based on the number of mRNAs they express, and genes were filtered based on the number of samples that express them (see Materials and Methods and Supplementary Fig. S3b). In total 5 samples were excluded and at least 8 replicates remained per genotype per developmental time point (Supplementary Fig. S3c). After this, gene counts were normalized to avoid, among others, differences derived from unequal amounts of cardiac tissue developed by the rnf2 mutant and wildtype embryos. Hierarchical clustering based on Euclidian Distances at the 3 different developmental time points indicates that differences between wildtype and rnf2 mutant hearts are minor at 1 and 2 dpf and more pronounced at 3 dpf ( Supplementary Fig. S3d-f). This is also reflected in the number of genes that are differentially expressed at these different time points (Fig. 5b). At 1 dpf no genes were found to be upregulated and 7 genes were significantly downregulated (|Log2FC > 0|; padj < 0.01). This number increased to 82 upregulated and 30 downregulated genes at 2 dpf. At 3 dpf 284 genes were detected to be significantly upregulated and 269 genes to be significantly downregulated in rnf2 mutant heart. Overall, we observed that the number of differentially expressed genes between wildtype and rnf2 mutant hearts increased over time (Fig. 5b, Supplementary Table S4).
Cardiac chamber identity is disrupted in rnf2 mutants at 3 dpf. Recently, Hill et al. generated a hand-curated list of cardiac markers 38 . This list contains genes that are expressed in the developing heart and includes 26 annotations about the function and the location of these genes. These 26 annotations were studied for the gene expression changes of the genes falling into these categories. We used GSEA using the Single HeartsRNA-sequencing results of 3 dpf hearts (Fig. 6a). This analysis revealed that rnf2 mutant hearts are enriched for genes expressed in the atrioventricular canal (FDR q-value = 0.089; NES = 1.46) and for heart transcription factors (FDR q-value = 0.001; NES = 1.96; Fig. 6a). These two annotations show overlap in the genes they contain. Three transcription factors that repress myocardial genes (tbx2a, tbx2b, and tbx3a) are present in the top 4 of genes that are upregulated in rnf2 mutant hearts. Single HeartsRNA-seq results at 1, 2, and 3 dpf indicate that the www.nature.com/scientificreports www.nature.com/scientificreports/ expression of tbx3a is significantly upregulated at 2 dpf and that expression of tbx2a, tbx2b, and tbx3a is significantly upregulated at 3 dpf in rnf2 mutants (|Log2FC > 0|; padj < 0.01, Supplementary Fig. S4).
The GSEA using the 3 dpf Single HeartsRNA-sequencing results additionally indicated an enrichment for downregulation of structural genes (FDR q-value = 0.054; NES = −1.67; Fig. 6b) and myocardial genes (FDR q-value = 0.043; NES −1.62). The group of genes downregulated upon the rnf2 mutation is enriched in structural genes such as: vmhc, ttn.1, myh7ba, myh6, nppa, ttn.2, mylk3, and camk2a (Fig. 6b). Amongst the genes that are leading there is the marker for the working myocard nppa, the atrium marker myh6, and the ventricle marker vmhc. We analyzed the expression of nppa, myl7, myh6, and vmhc by Single HeartsRNA-seq and additionally tested the spatio-temporal expression of these genes by whole mount in situ hybridization (WISH) (Fig. 6c,  Supplementary Fig. S5). At 1 and 2 dpf the expression differences between wildtype and rnf2 mutant hearts for these four genes are relatively small. We found nppa and vmhc to be significantly downregulated in rnf2 mutant hearts at 1 dpf and 2 dpf, respectively. Single HeartsRNA-seq results at 3 dpf indicate that the expression of nppa, myh6, and vmhc is significantly decreased in rnf2 mutants (|Log2FC > 0|; padj < 0.01, Fig. 6c).

Discussion
Rnf2 mutants show a pleiotropic phenotype. In this study zygotic rnf2 mutant embryos (rnf2 ibl31/ibl31 ) are used as a model for loss of PRC1 and H2AK119ub 8,26 . These mutant embryos display defects in the maintenance of cellular identity and organ integrity 8,26 . Rnf2 is the only catalytic subunit of PRC1 in zebrafish and therefore disrupting Rnf2 is informative of PRC1's putative role in zebrafish development.
The rnf2 mutation results in a pleiotropic phenotype in zebrafish embryos at 3 dpf 8 . A multitude of genes and processes were described to be affected upon the loss of Rnf2 8,26 , however no genome-wide molecular expression datasets are available for these mutants. Since Rnf2 is an epigenetic modifier, ChIP-sequencing can give insight in its mode of action and RNA-sequencing can elucidate its downstream effects. With bright field microscopy and WISH analyses we studied the rnf2 mutant phenotype on a global level. Next to the described phenotypes of rnf2 mutants 8,26 , we observed a heart edema accompanied by a tubular heart. WISH analyses suggest developmental organ defects of the intestine, pancreas, and liver at 3 dpf. If the observed differences are due to organ defects, they could result from defects in differentiation, cell proliferation, or tissue maintenance, as well as a combination of these three, as PcG proteins are described to play a role in these processes 45 . In this study we focused on heart development.
H3K27me3 is retained upon rnf2 mutation. H2AK119ub, the mark deposited by PRC1, and more specifically, by its enzymatic subunit Rnf2, is described to stabilize H3K27me3 9 . In addition, also a role for PRC1 in the recruitment of PRC2 is proposed 12,13 . Therefore, loss of H3K27me3 was considered to occur in the absence of Rnf2. However, many studies also reported that the loss of PRC1 does not affect H3K27me3 deposition 23,[30][31][32][33][34] . We used our PRC1 null model (rnf2 mutants) to study this process in vivo and show that H3K27me3 deposition at 3 dpf is retained upon loss of Rnf2/PRC1, which is thus in line with the majority of these previous reports 23,30-34 .

Figure 5.
Single HeartsRNA-seq is used to assess transcriptional differences between wildtype and rnf2 mutant hearts over time. (a) Workflow of Single HeartsRNA-seq. Zebrafish hearts were manually dissected at 1, 2, and 3 dpf as described previously 85 . The remaining tissue was used for genotyping and the rnf2 mutant and wildtype hearts were sequenced. (b) Differentially expressed genes |Log2FC > 0|; padj < 0.01 between rnf2 mutant and wildtype hearts are visualized in the bar graphs. Yellow: upregulated (up) in rnf2 mutants compared to wildtypes. Blue: downregulated (dn) in rnf2 mutants compared to wildtypes.
www.nature.com/scientificreports www.nature.com/scientificreports/ Based on our data one could even argue that the rnf2 mutation results in a slight increase in H3K27me3 levels. We study whole embryo lysates and therefore this slight increase can be due to more H3K27me3 deposition within the cell, however it can also be due to changes in the abundance of cells that repress these genes via an H3K27me3-mediated manner. This could also be the reason why we do not detect a loss of H3K27me3 in the absence of Rnf2, as changes in individual cells will be overshadowed by the signal detected in the bulk of all cells analyzed.
the rnf2 mutation results in derepression of genes decorated by Rnf2. A subset of genes was found to lose Rnf2-mediated repression and this subset was detected to be significantly upregulated in rnf2 mutants; however, they retain H3K27me3-mediated repression at the whole embryo lysate level. A potential explanation for gene upregulation, whilst they are decorated by H3K27me3, could be derived from the type of sample that has been used. The samples are lysates from whole embryos at 3 dpf, in which many cell types www.nature.com/scientificreports www.nature.com/scientificreports/ are present and the sequencing results give an average of the signal coming from across all those different cell types. Therefore, overall gene upregulation could be caused by cells in which these genes are not repressed by H3K27me3.
The role of epigenetics in heart development. The role of epigenetics in cardiac development has gained more attention over the years, as reviewed by Vallaster et al.,and Shirai et al. 46,47 . Zebrafish hearts are of high interest due to their regenerative capacity and epigenetics is also implied to be important in this process, as reviewed Quaife-Ryan et al. 48 . Therefore, it is of essence to better understand the role of epigenetics in heart tissue specification, maintenance, and regeneration. To unravel important regulators of cardiac development, studies have been performed that aimed to make a roadmap of the transcriptome and epigenome during myocardial differentiation 49,50 . Other approaches focus on chromatin remodelers and the identification of enhancers to gain insight in the role of epigenetics in heart development 51,52 . Many histone modifiers are described to play a role during cardiac development, such as histone deacetylases, HDACs 53,54 , H3K4me3 methyltransferase 55 , and also the PRC2-component Ezh2 [56][57][58] .
The PRC1-variant containing Mel18 was described to be essential for specification of mesodermal cell fate, by preventing alternative lineage commitment 29 . In line with that, the PRC1-component Bmi1 was shown to act as a barrier during cardiac reprogramming in mouse cells 59 . Interestingly, a role in cardiac development in vivo for both Mel18 and Bmi1 has not been established 60,61 . That Mel18-PRC1 and Bmi1-PRC1 do not play a role in cardiac development does not exclude the possibility that any of the other PRC1 variants do. A role for Rnf2 in cardiac development has not been described in vivo, so far. In our current study we show that depletion of Rnf2 affects heart morphology and gene expression during zebrafish development.
A study on H3K4me3 methyltransferases in zebrafish heart development has shown that a decrease in H3K4me3 results in a linear shaped heart, similar to the rnf2 mutants. However, in that study the heart markers myl7, vmhc, and myh6 were reported not to be affected at 2 dpf upon decrease of H3K4me3 55 . This is in contrast to the rnf2 mutants, as we observe at 2 dpf significant downregulation of vmhc expression. A previous study from our lab using maternal zygotic ezh2 mutants also shows a linear shaped heart upon the mutation of the catalytic subunit of PRC2. Altogether, these results indicate that a similar phenotype can have different causes and different epigenetic signatures, as both a loss of H3K4me3, H3K27me3, and H2AK119ub result in a linear shaped heart phenotype 55,58 .

Repression of tbx-genes by Rnf2 is important for cardiac development.
At early stages (1 and 2 dpf) we found the expression of cardiac genes in rnf2 mutant hearts to be more or less similar to wildtype hearts, therefore we hypothesize that maternal rnf2 RNA and Rnf2 protein are sufficient for correct cell specification in the rnf2 mutants. On top of that, as most PcG proteins, Rnf2 is expected to be mostly involved in tissues maintenance rather than tissue specification 8,19,23,26,58 .
Since we observe an upregulation of tbx-genes in a system in which we mutate a transcriptional repressor, we analyzed our Rnf2 ChIP-seq results on whole embryo lysates and found that the tbx3a gene is bound by Rnf2 and decorated by H3K27me3 in the wildtype situation. This observation strongly hints towards Polycomb-mediated regulation of Tbx3. Literature describes that this is not zebrafish-specific, since Pcl2 (PRC2 subunit) knock-out murine ESC show an upregulation of Tbx3 62 . Research has revealed that when Tbx2/3 forms a complex with Gata and Nkx, it locally represses chamber myocardial gene expression and thereby enhances the formation of the conduction system 63 . Tbx2/3 are described to directly repress, amongst others, the myocardial gene nppa to allow for the formation of the conductive system in the heart 64,65 . Zebrafish rnf2 mutants show a malformed heart and our 3 dpf Single HeartsRNA-seq dataset shows that nppa is significantly downregulated, which we validated by WISH experiments. Tbx2/3 overexpression in mouse embryos results in a heart looping defect and the lack of cardiac chambers 66,67 . In these murine embryos the chamber-myocardial gene program is not correctly set up. Interestingly, the rnf2 mutant zebrafish embryos also show a heart in which the atrium and ventricle are not well developed.
Furthermore, Tbx5 competes with Tbx2/3 to form a complex with Nkx and Gata, and this complex has an activating effect on gene expression 66 . Early stages of rnf2 mutants show correct tbx5 levels, most likely due to the maternal load of rnf2 8 . However, at 3 dpf we detect tbx5a to be upregulated in the heart upon the rnf2 mutation and whole embryo lysate ChIP-seq indicates that the tbx5a gene is decorated by Rnf2 and H3K27me3. Overexpression of tbx5 in vitro has been reported to represses proliferation and cell growth 68 , which is in line with our observations. In contrast, rnf2 mutant zebrafish were reported to have greatly reduced expression of tbx5 at the pectoral fin mesenchyme, which is correlated to the absence of pectoral fins in these mutants 8 and interestingly, the loss of Tbx5 results in a stringy heart in zebrafish embryos 69 .
These studies and our observations indicate that tight regulation of tbx2a, tbx2b, tbx3a, and tbx5a is required for proper heart development. Since three Tbx2/3 variants are overexpressed in the rnf2 mutant heart to a larger extent than tbx5a, we suggest that overexpression of tbx2/3 is the main driver of the observed myocardial phenotype. We therefore postulate that the overall downregulation of myocardial genes is the result of inadequate activation or maintenance of the chamber-myocardial gene expression program, which results in defects in maintenance of cell identity. Studies by others indicate that Rnf2 is an important player in the maintenance of tissue integrity in a wide variety of systems, and the zebrafish data on single hearts from rnf2 mutants adds to this list 8,19,23,26 . We hypothesize that the molecular pathway that allows for the formation of the conductive system is partially regulated by Rnf2 and that this ensures the correct balance in chamber and conductive cell identity within the heart. Disruption in this balance results in defects in cardiac development and functioning. (

Methods
Zebrafish genetics and strains. Zebrafish (Danio rerio), were housed at 27.5 °C in a 14/10 h light/dark cycle. The evening before spawning, one male and one female were placed into a tank with a divider and the following morning, at the moment the light switched on, the fish were placed together for breeding. Embryos were collected and staged according to Kimmel et al. 70 . The rnf2 ibl31/ibl31 zebrafish were out-crossed with wildtype (TLF) or with Tg(myl7::GFP) 8,36 . All methods were carried out in accordance with relevant guidelines and regulations of national animal welfare laws.
Genotyping. DNA was purified from embryos or from caudal fin tissue, taken from anesthetized adult zebrafish. Genotype analysis was performed by PCR using the primer set forward: 5′-TCTAAGCGCTCT CTTCGTCCAGA-3′ and reverse: 5′-ACAAGAGGATTTGTAACAAAGCCG-3′, followed by digestion of the PCR product with restriction enzyme Taq1 to identify the rnf2 ibl31/ibl31 allele 8 . The agarose gel was imaged using the Gel Doc XR+ Imaging System (Bio-Rad), in combination with Image Lab Software (Bio-Rad). After acquisition, the image color was inverted and the levels were adjusted, using Photoshop, to visualize all bands for correct genotyping.
Whole mount in situ hybridization. Dechorionated embryos were fixed overnight at 4 °C in 4% PFA (Aurion, 151710) in PBST (PBS with 0.1% Tween-20), after which they were gradually transferred to and stored in 100% methanol. To prevent probe trapping, the heart edema of 3 dpf rnf2 mutant embryos was pierced with watchmaker forceps (INOX5). Embryos were treated with proteinase K. Whole mount in situ hybridization was performed as described previously 71 . Trypsin and fabp10 probes were generated by PCR from cDNA from 1 dpf wildtype embryos using the following primers: forward trypsin CAGG CCCTTTAGTGAGGGTTAATT TGTCTGCTGCTCACTGGTAC; reverse trypsin CAGG TAATACGACTCACTATAGGG GTCCTTGCCTCCCT CCATAA. Forward fabp10 CAGG CCCTTTAGTGAGGGTTAATT GTTGAGCTTCTCCAGAAAGCATG, reverse fabp10 CAGG TAATACGACTCACTATAGGG GATCATGGTGGTTCCTCCGA. T7 polymerase was used to generate the anti-sense probes. After WISH the embryos were mounted in 4% methylcellulose and imaged by light microscopy on a Leica MZFLIII, equipped with a DFC450 camera. The embryos were genotyped after imaging.
Immunostainings. Dechorionated embryos were fixed overnight in 4% PFA in PBST at 4 °C. After fixation, embryos were gradually transferred to and stored in 100% methanol. Before immunostaining embryos were transferred stepwise to PBST. Rabbit anti-Ring1b antibody from Cell Signaling Technology was used (RING1B Cell Signaling D22F2 1:4000). Antibody incubation was followed by a secondary antibody and subsequent DAB staining (EnVision+ System-HRP (DAB) k4010). The embryos were mounted in 4% methylcellulose and imaged by light microscopy on a Leica MZFLIII, equipped with a DFC450 camera. The embryos were genotyped after imaging.
ChIp-sequencing. Embryos from a rnf2 heterozygous incross were sorted for the rnf2 pleiotropic phenotype at 3 dpf. These phenotypical mutants are used for the rnf2 mutant sample. The rnf2 mutation has a 100% penetrance and no false positives have been detected by genotyping after phenotypic screening. As controls we used a wildtype strain from the same genetic background. Pools of 80 to 100 embryos of 3 dpf were fixed, deyolked, and homogenized using pestles and sonicated to release and isolate the chromatin. Chromatin was stored at −80 °C. For spike-in experiments 30 µg zebrafish chromatin was mixed with Drosophila spike-in chromatin and incubated overnight with H2Aγ and RING1B antibodies (H2Aγ Active Motif 104597; RING1B cell signaling D22F2) or with H2Aγ and H3K27me3 antibodies (H2Aγ Active Motif 104597; Millipore 07-449). After overnight antibody incubation, the chromatin was extensively washed. The resulting ChIP-DNA was used as input for KAPA-HYPERprep library preparation. Libraries were paired-end sequenced (43 bp read-length) on an Illumina NextSeq500 platform. For wildtype and rnf2 mutant samples, two and three replicates were used, respectively.
ChIp-sequencing analyses. In order to avoid biases due to differences in the efficiency of the sequencing runs, spike-in ChIP-seq reads were mapped to the Drosophila melanogaster genome v6 using bwa mem version 0.7.15 72 with default settings. Multimapping reads were excluded using samtools version 1.3.1 73 and duplicated reads were removed with Picard (http://broadinstitute.github.io/picard/). Once we obtained the number of reads mapped per sample, we normalized them based on the sample of each experiment with lowest number of mapped reads, by removing random reads accordingly. After this, remaining ChIP-seq reads were mapped to the GRCz10/ danRer10 using bwa mem version 0.7.15 72 with default settings. Multimapping reads were excluded using samtools version 1.3.1 73 and duplicated reads were removed with Picard (http://broadinstitute.github.io/picard/). Peaks were called using MACS 2.1.1.20160309 74 relative to the input track using the options -f BAMPE -g 1.3e9 -q 1e-2-broad -broad-cutoff 1e-1. Peaks 1 kb or closer from each other were merged, and H3K27me3 peaks narrower than 100 nt were discarded. Intersecting peaks were considered for replicates using GenomicRanges 75 . Peaks overlapping peaks called in the input track were excluded. Clustering of peaks was done using the union of the remaining peaks found in wildtype Rnf2 and H3K27me3 ChIPs and rnf2 mutant H3K27me3 ChIP, considering only those overlapping promoter regions (400 nt upstream-100 nt downstream of the transcription start site). Clustering and visualization of the peaks (i.e. heatmaps, bandplots and profiles) was done using fluff version 2.1.3 76 . GO term analysis for biological process on genes of cluster 3 was performed with Cytoscape 3.3.0 using the ClueGO 2.2.4 plugin with default settings. To reduce redundancy in biological process GO terms, GO term fusion and GO term grouping was applied and the groups were plotted with group p-value corrected with Bonferroni step down. Fluorescent imaging. Embryos from a Tg(myl7::GFP);rnf2 heterozygous incross were anaesthetized in MS-222 and embedded in 1.5% low-melting-point agarose (Sigma). The embryos were imaged by light microscopy on a Leica MZFLIII, equipped with a DFC450 camera. The embryos were genotyped after imaging. spIM-imaging. Embryos from a Tg(myl7::GFP)_rnf2 heterozygous incross were injected with α-Bungarotoxin at the one-cell stage 82 . As controls we used embryos from a Tg(myl7::GFP) incross, which were also injected with α-Bungarotoxin at the one-cell stage 82 . At 1 dpf the embryos were embedded for SPIM imaging in 1.5% low-melting-point agarose (Sigma) in FEP tubes (Bola, S1815-04). We used the custom build multidirectional selective plane illumination microscopy (mSPIM) as described before 83 . Photos were taken with a 20-minute interval and the images were synchronized. The stages in which both the ventricle and atrium are dilated were used for data visualization by Imaris software (bitmap). The embryos were genotyped after imaging.
single HeartsRNA-sequencing. Hearts were manually dissected from 1, 2, or 3 dpf Tg(myl7::GFP) positive embryos from an rnf2 heterozygous incross using watchmaker forceps (INOX5) and placed into Eppendorf LoBind tubes with TRIzol (Ambion), rapidly frozen in liquid nitrogen, and stored at −80 °C prior to further processing. The remainder of the embryos was individually collected in methanol and used for genotyping. RNA was extracted from the wildtype and rnf2 mutant hearts using TRIzol reagent (Ambion) according to the manufacturer's manual. After RNA extraction, pellets were resuspended with barcoded primers. Primers consisted of a 24 bp polyT stretch, a 4 bp random barcode, a unique 8 bp sample-specific barcode, the 50 Illumina adaptor (as used in the TruSeq small RNA kit), and a T7 promoter for in vitro transcription 84 . The RNA samples were subsequently reverse transcribed, pooled, and in vitro transcribed for linear amplification with the MessageAmpII kit (Ambion) according to the CEL-seq protocol 42 . Illumina sequencing libraries were prepared with the TruSeq small RNA sample prep kit (Illumina) and sequenced single-end at 75 bp read length on an Illumina NextSeq. 500 instrument.
single HeartsRNA-sequencing analyses. Raw reads were processed to obtain expression levels following the pipeline designed for the CEL-Seq method 42 using the scripts available at https://github.com/yanailab/ CEL-Seq-pipeline. Data normalization and differential gene expression analyses were done with Monocle version 2.4.0 79 . Samples were filtered according to +/−2 SD cut-off in the amounts of mRNA that is detected in the sample after log transformation. Genes expressed in at least 25% of the samples at a given developmental stage were used for differential gene expression analysis and GSEA.

Data Availability
The ChIP-seq data and RNA-seq data from whole embryo lysates and the single hearts discussed in this manuscript have been deposited in NCBI's Gene Expression Omnibus and are accessible through GEO Series accession number GSE114038.