Abstract
Polycomb group proteins are transcriptional repressors that are important regulators of cell fate during embryonic development. Among them, Ezh2 is responsible for catalyzing the epigenetic repressive mark H3K27me3 and is essential for animal development. The ability of zebrafish embryos lacking both maternal and zygotic ezh2 to form a normal body plan provides a unique model to comprehensively study Ezh2 function during early development in vertebrates. By using a multi-omics approach, we found that Ezh2 is required for the recruitment of H3K27me3 and Polycomb group protein Rnf2. However, in absence of Ezh2, only minor changes in global H3K4me3 levels and gene and protein expression occurred. These changes were mainly due to local deregulation of transcription factors outside their normal expression boundaries. Altogether, our results in zebrafish show that Polycomb-mediated gene repression is important only after the body plan is formed to maintain spatially restricted transcriptional profiles of Polycomb-targeted transcription factors.
Introduction
Development of multi-cellular organisms involves highly dynamic and controlled processes during which one single totipotent cell will multiply and differentiate into all the cells composing the adult individual. Specification of cell identity is controlled through the establishment of spatially and temporally restricted transcriptional profiles, which are subsequently maintained by epigenetic mechanisms(1). Epigenetic maintenance of gene expression can act through modifications of the chromatin, the complex of DNA wrapped around an octamer of histones H2A, H2B, H3, and H4 and its associated proteins and non-coding RNAs, creating an epigenetic landscape, often referred to as the epigenome(2). These modifications can be propagated from mother to daughter cells and thereby maintain gene expression profiles by controlling the accessibility of the DNA to the transcriptional machinery(3).
Important regulators of the epigenome during development are the Polycomb Group (PcG) proteins. First identified in Drosophila melanogaster, PcG proteins were found to maintain the pre-established pattern of hox gene expression(4). Subsequent studies showed that PcG proteins are important for proper patterning during early embryonic development, tissue-specific development, and maintenance of the balance between pluripotency and differentiation of stem cells in multiple species(5). Two main PcG complexes have been described(6). The Polycomb Repressive Complex 2 (PRC2) is composed of the core subunits EZH1/2 (Enhancer of Zeste Homologue 1/2), SUZ12 (Supressor of Zeste 12), and EED (Embryonic Ectoderm Development). EZH2 has a catalytically active SET domain that trimethylates lysine 27 of histone H3 (H3K27me3), an epigenetic mark associated with gene repression and found along gene coding sequences(7). The catalytic subunits of PRC2 are mutually exclusive and EZH1 is postulated to complement the function of EZH2 in non-proliferative adult organs(8, 9). H3K27me3 can be recognized by the Polycomb Repressive Complex 1 (PRC1). A diversity of PRC1 compositions has been described and canonical PRC1 is composed of the core subunits RING1/RNF2 (Ring Finger Protein 2 a/b), PCGF1-6 (Polycomb Group RING fingers 1-6), PHC (Polyhomeotic), and CBX (Chromobox homolog)(10, 11). PRC1 catalyzes the ubiquitination of lysine 119 of histone H2A (H2AK1119ub) and strengthens gene repression. In contrast to this canonical view, recent studies implicate that PRC1 is also active in the absence of PRC2(12-14).
Trithorax Group (TrxG) proteins antagonize PcG protein function through the deposition of a trimethyl group on lysine 4 of histone H3 (H3K4me3) on promoters and enhancers from virtually all transcribed genes(15-17). Acetylation of H3K27 (H3K27ac) is a different epigenetic mark associated with promoters and active enhancers and constitutes a dynamic signature marking developmentally regulated genes(18).
In mice, loss of PRC2 genes Ezh2, EED, or Suz12 or PRC1 gene Rnf2 leads to post-implantation embryonic lethality during early gastrulation(19-22), making it difficult to study transcriptional regulation by PcG complexes during early development. Lately, the zebrafish embryo has emerged as a model of choice to study developmental epigenetics in vertebrates, although epigenomics studies on mutant lines are still rare(23-26). Others and we previously showed that ezh2 is essential for zebrafish development(27-29). More particularly, zebrafish embryos mutant for both maternal and zygotic ezh2, referred as MZezh2 mutant embryos, develop seemingly normal until 1 dpf, forming a proper body plan. These mutants ultimately die at 2 dpf, exhibiting a 100% penetrant pleiotropic phenotype associated with a loss of tissue maintenance(28). This makes zebrafish MZezh2 mutant embryos a unique model to study the function of Ezh2 during early development, from fertilization to tissue specification, in the unique context of a vertebrate embryo in which trimethylation of H3K27 has never occurred, unlike cell culture, conditional, or zygotic mutant models.
We conducted a multi-omics approach in these MZezh2 mutant embryos to study how PcG-mediated gene regulation controls axis formation and tissue specification. We focused our study on 24 hours post fertilization (hpf) embryos, when the first phenotypes become visible, and the anterior-posterior patterning of the embryos is properly established. Surprisingly, our data show that, despite the complete absence of PcG associated epigenetic mark and proteins, the transcriptional and proteomic profile in MZezh2 mutant embryos remains largely unchanged compared to wildtype embryos. The changes mainly affect transcription factors essential for developmental processes. Closer analysis of spatial expression of the genes deregulated in MZezh2 mutants revealed that aberrant gene expression is primarily local. Our results show that zebrafish embryo development is initially independent of PcG repression until the stage of tissue maintenance, where PcG proteins maintain precise spatial repression of transcription factor expression.
Results
The repressive epigenetic mark H3K27me3 is absent in MZezh2 embryos
To study the function of Ezh2 during development, we used the ezh2 nonsense mutant allele ezh2(hu5670) containing a premature stop codon within the catalytic SET domain and resulting in the absence of Ezh2 protein(28). Elimination of both maternal and zygotic contribution of Ezh2, by using the germ cell transplantation technique described previously(28, 30), allowed us to study the function of Ezh2 during early development. As previously shown, MZezh2 mutant embryos display normal body plan formation and a mild phenotype at 24 hpf. They die at 48 hpf, at which point pleiotropic phenotypes are observed, such as smaller eyes, smaller brain, blood coagulation, and absence of pectoral fins (Figure. 1A). Western Blot analysis at 3.3 hpf and 24 hpf confirmed the absence of both maternal and zygotic Ezh2 in these mutants, respectively (Figure 1B).
To further confirm the absence of Ezh2 in MZezh2 mutants and assess its effect on H3K27me3 deposition, we performed ChIP-sequencing (ChIP-seq) for Ezh2 and H3K27me3 at 24 hpf in both wildtype and MZezh2 mutant embryos. ChIP-seq analyses for Ezh2 and H3K27me3 revealed 816 and 3,353 peaks in wildtype embryos, respectively (Figure 1C,D, Supplementary Table 1). Although the number of peaks differed between the two proteins, their binding profiles greatly overlap (Figure 1C). Quantification showed that 84.6% of Ezh2 peaks also contain H3K27me3 (Supplementary Figure 1A). Known PcG target genes such as the hoxab gene cluster, tbx genes, and gsc presented similar binding profiles for Ezh2 as for H3K27me3 (Figure 1F,G, Supplementary Figure 1B), whereas the ubiquitously expressed genes eif1ad and tbp showed absence of Ezh2 and H3K27me3 (Supplementary Figure 1B).
In MZezh2 mutant embryos, the binding of Ezh2 and H3K27me3, as detected by ChIP-seq, was virtually absent, with 3 and 22 peaks detected for Ezh2 and H3K27me3, respectively (Figure 1C). Manual inspection of these remaining peaks revealed that they are present in gene deserts and low complexity regions and are probably artefacts (Supplementary Figure 1C). Ezh2 and H3K27me3 coverage was reduced to background levels in MZezh2 mutants compared to wildtype (Figure 1C). Finally, the hoxab gene cluster, tbx3a, tbx5a, gsc, and isl1 loci, targeted by PcG repression in wildtypes, also showed a complete absence of Ezh2 and H3K27me3 binding in MZezh2 mutants (Figure 1F,G, Supplementary Figure 1B).
Altogether, these results demonstrate that the absence of maternal and zygotic Ezh2 results in a complete absence of Ezh2 and H3K27me3 from chromatin.
Loss of PRC2-mediated repression results in loss of PRC1 recruitment during early development
It is postulated that PRC1 is recruited to chromatin by PRC2-deposited H3K27me3 but can also have a function independent of PRC2(12-14). As both Ezh2 and H3K27me3 are absent from MZezh2 mutant embryos, we investigated whether PRC1 is still recruited to chromatin in these mutants. In zebrafish, Rnf2 is the only catalytic subunit of PRC1(31). ChIP-seq for Rnf2 in wildtype embryos at 24 hpf reveals 4,095 peaks (Figure 1C, Supplementary Table 1) which are present at Ezh2 and H3K27me3 positive regions (Figure 1E). We found that 84.9% of Ezh2 peaks were also positive for Rnf2 in wildtype embryos (Supplementary Figure 1A).
In MZezh2 mutant embryos, only 11 binding sites could be detected for Rnf2 (Figure 1C) and Rnf2 average binding (RPKM) was reduced to background level, as observed for Ezh2 and H3K27me3 binding (Figure 1C,E). This loss of Rnf2 was observed at both gene clusters such as hoxab (Figure 1F) and individual transcription factors such as tbx3a, tbx5a, and gsc (Figure 1G, Supplementary Figure 1B). As for Ezh2 and H3K27me3, Rnf2 remaining peaks in MZezh2 mutant embryos were detected in intergenic regions with repeat sequences and are probably artefacts (Supplementary Figure 1C).
Thus, the loss of PRC2 binding and H3K27me3 deposition in MZezh2 mutant embryos is associated with a complete loss of Rnf2 on the chromatin.
Loss of H3K27me3 in MZezh2 mutant embryos induces gene specific gain of H3K4me3
As PcG and TrxG complexes are known to have an antagonistic effect on gene expression(32), we investigated whether the loss of H3K27me3 in MZezh2 mutant embryos changed the deposition H3K4me3, a mark associated with gene activation.
To this aim, we performed ChIP-seq for H3K4me3 in both wildtype and MZezh2 mutant embryos at 24 hpf. In total, 11,979 H3K4me3 peaks were detected in wildtype embryos, and 13,401 in MZezh2 mutants (Figure 2A, Supplementary Table 1). A majority of 11,710 peaks were shared between wildtype and MZezh2 mutant embryos, whereas 1,691 were specific for MZezh2 mutants and 269 peaks for wildtype embryos (Figure 2A,B). Comparison of the RPKM-normalized H3K4me3 binding levels between MZezh2 mutant and wildtype embryos showed an overall increase in H3K4me3 enrichment upon loss of Ezh2 on the peak locations shared among them (Figure 2B,C). The wildtype specific peak locations presented no difference in RPKM coverage between wildtype and MZezh2 mutant embryos (Figure 2B). Detection of these wildtype-specific peaks could therefore be explained by the limitation of the peak detection algorithm in accurately identifying low enriched regions, rather than being true wildtype-specific peaks. In contrast, in MZezh2 mutant embryos, mutant-specific peak locations showed a significant increase in H3K4me3 enrichment level compared to wildtype embryos (Figure 2B). Finally, the binding intensity on the mutant specific H3K4me3 peak locations remained relatively low in MZezh2 mutant embryos and never reached the enrichment level detected on the H3K4me3 peak locations shared by wildtype and MZezh2 mutants (Figure 2B,C).
We next assessed if the H3K4me3 peaks gained upon loss of Ezh2 correlated with the presence of H3K27me3 in wildtype embryos. In wildtype embryos, 6.1% of H3K4me3 peaks were also covered by H3K27me3, which is more than expected by random chance (P-value < 0.001). In MZezh2 mutant embryos, a significant increase to 8.3% of the H3K4me3 peaks was found on regions positive for H3K27me3 in wildtypes embryos (P-value < 0.001). This enrichment was even higher when taking into account only the MZezh2 mutant specific H3K4me3 peaks (25.1% with P-value < 0.001) (Figure 2D,E). This result shows that the regions positive for H3K27me3 in wildtype are more susceptible to gain H3K4me3 upon loss of H3K27me3.
Next, we determined if the epigenetic landscape at the transcription start sites (TSS) in wildtype embryos had an influence on the ability to gain of H3K4me3 upon loss of Ezh2. We classified all the TSS from the zebrafish in four categories according to our ChIP-seq data in wildtype embryos: active TSS positive for H3K4me3, repressed TSS positive for H3K27me3, TSS positive for both marks, and TSS negative for both marks. Comparison of the level of H3K4me3 binding in MZezh2 mutant and wildtype embryos on these TSS revealed that 3.9% (P-value < 0.001) of the H3K27me3 positive and 4.4% (P-value < 0.001) of the H3K4me3/H3K27me3 double positive TSS were enriched (Fold Change: FC ≥ 2) for H3K4me3 in MZezh2 mutant embryos compared with random TSS (0.9%). TSS positive for only H3K4me3 or negative for both H3K27me3 and H3K4me3 were less likely to show an increase in H3K4me3 binding in MZezh2 mutant embryos compared with random TSS (0.6%, P-value < 0.01 and 0.4%, P-value < 0.001 respectively) (Figure 2F,G).
To find all the potential direct targets of PcG-mediated repression gaining H3K4me3 in absence of Ezh2, we searched for the closest genes of the H3K4me3 mutant-specific peaks, which had an H3K27me3 peak in wildtype embryos, and identified 463 genes. Gene ontology analysis revealed that these genes were mainly involved in transcriptional regulation and organismal development (Figure 2H). Among these 463 identified genes, 143 encode for transcription factors, among which were members of the hox, tbx, sox, and pax gene families, known targets of PcG complexes.
During development, gene transcription is also controlled by enhancer activation(33). H3K27me3 and H3K27ac are known to have an opposite effect on enhancer activation, the former being associated to poised enhancers and the latter to active enhancers(34). We studied the binding of H3K27ac in 24 hpf embryos lacking H3K27me3. Unlike H3K4me3, the number of peaks enriched for H3K27ac decreased by more than a half in MZezh2 mutants compared with wildtype (4,155 and 8,952 peaks detected, respectively) (Supplementary Figure 2A, Supplementary Table 1). This loss of H3K27ac was associated with a decrease in coverage intensity (Supplementary Figure 2B).
These results suggest that loss of PcG-mediated repression has a specific effect rather than a general impact on deposition of activating epigenetic marks.
Epigenetic changes in MZezh2 mutant embryos induce upregulation of transcription factors
As MZezh2 mutant embryos show a complete lack of the H3K27me3 repressive mark and a selective increase in H3K4me3 activating mark on genes coding for transcription factors, we investigated the effect of loss of Ezh2 on the transcriptome by RNA-sequencing (RNA-seq) of wildtypes and MZezh2 mutants at 24 hpf. Differential gene expression analysis between the two conditions revealed a limited effect on the transcriptome upon the loss of Ezh2. Only 60 genes were detected to be significantly upregulated (log2FC ≥ 1 and P-adj < 0.05) and 28 genes downregulated (log2FC ≤ −1 and P-adj < 0.05) in MZezh2 mutants (Figure 3A). When inspecting the upregulated genes, we found 60% of the upregulated genes (compared to 9.3% in all genes, P-value < 0.001) to be targeted by H3K27me3 in wildtype embryos (Figure 3B,C). On the other hand, the upregulated genes were less likely to be targeted by H3K4me3 in wildtype embryos compared to all zebrafish genes (21.7% compared to 47.8%, P-value < 0.001), yet gained significant binding of H3K4me3 in MZezh2 mutant embryos (23.3% compared to 5.3%, P-value < 0.001, Figure 3B,C). No gain of H3K4me3 was detected on genes downregulated in MZezh2 mutants (Figure 3B,C).
As a complement, we studied the relation between changes in transcriptome and epigenome by performing Gene Set Enrichment Analysis (GSEA)(35, 36). This analysis revealed that the genes upregulated in MZezh2 mutants showed enrichment for the set of genes occupied by H3K27me3 in wildtype embryos (n=2336, Fig 3d, left panel). The set of genes positive for H3K4me3 in wildtype embryos (n=11979) showed no enrichment for neither upregulated nor downregulated genes in MZezh2 mutants (Figure 3D, middle panel). The genes with higher expression in MZezh2 mutants showed a clear association with the presence of MZezh2 mutant specific H3K4me3 peaks (n=1317, Figure 3D, right panel). Thus, through the GSEA, we can confirm the association between the loss of H3K27me3 repressive chromatin and the gain of new H3K4me3 positive chromatin on specific loci with the gain in gene expression in MZezh2 mutant embryos at 24 hpf.
GO analysis of the deregulated genes in MZezh2 mutants revealed enrichment in biological processes of regulation of transcription and development (Figure 3E), a majority of these genes encoding for transcription factors. Anatomical terms associated with the deregulated genes indicated that these genes were expressed in organs showing strong phenotypes in MZezh2 mutant embryos, such as fin bud, retina, and heart tube (Supplementary Figure 3A).
Deregulations in epigenome and transcriptome are linked to changes in protein expression in MZezh2 mutant embryos
We next performed proteomic analysis of the MZezh2 mutant embryos and compared the result with wildtype embryos at 24 hpf. Differential analysis revealed 111 upregulated (log2FC ≥ 1.5 and Padj < 0.05) and 110 downregulated (log2FC ≤ −1.5 and P-adj < 0.05) proteins (Figure 4A, Supplementary Table 1). After ranking the proteins based on their difference in expression between the two conditions, we explored those differences in the context of different epigenetic marks and changes in the transcriptome by GSEA. Results showed that upregulated proteins were enriched in H3K27me3 targets in wildtypes and presented mutant-specific H3K4me3 peaks in MZezh2 mutant embryos (Figure 4B). Upregulation and downregulation of protein expression also correlated with deregulation of gene expression in transcriptome (Figure 4C). For example, hoxa9b was one of the top overexpressed targets in both the transcriptome and proteome analyses (Figure 3A, Figure 4A, Supplementary Table 1).
Furthermore, in addition to Ezh2, Suz12b was found to be downregulated in MZezh2 mutant embryos (Figure 4A). Other PRC2 subunits were either not detected or (not significantly) downregulated. Subunits of the canonical PRC1 complex were mostly overexpressed, although not significant (Supplementary Figure 3B).
The low number of deregulated genes detected by both transcriptomic and proteomic analyses suggests that changes in expression could be either global and low in intensity or limited to specific cell types or tissues. To test these hypotheses, we carried out a spatial expression analysis on selected target genes.
Loss of ezh2 results in expression of hox genes outside their normal expression domains
To start with, we focused on expression of different genes from the hox gene family. These genes are known targets of Polycomb-mediated repression(37) and some of them have been previously shown to be deregulated in MZezh2 mutant embryos(28). Every hox gene has an expression pattern that is restricted along the anterior-posterior axis(38). To obtain spatially resolved data along the anterior-posterior axis, we performed RT-qPCR on the anterior half and the posterior half of 24 hpf wildtype and MZezh2 mutant embryos. We then compared the normalized relative expression levels between the different halves of the MZezh2 mutant and wildtype embryos. The tested hox genes were selected based on their domain of expression along the anterior-posterior axis (Figure 5A-D). The hoxa9a gene, whose expression extends to anterior, until slightly outside the posterior half of the embryos, showed, as expected, a higher expression in the posterior part than in the anterior part in wildtype embryos (Figure 5A). In MZezh2 mutants, hoxa9a was overexpressed in the anterior compartment compared with wildtype embryos, reaching levels similar to those observed in the posterior part of wildtype embryos. No significant differences were detected in the level of expression when comparing the posterior compartment of MZezh2 mutant and wildtype embryos (Figure 5A). Similar results were obtained for hoxa9b, where overexpression was detected in the anterior compartment of MZezh2 mutant embryos compared to the anterior compartment of wildtype embryos (Figure 5B). The hoxa11b and hoxa13b genes, which are expressed primarily posterior, showed, as expected, higher expression in the posterior half of the wildtype embryos compared to the anterior half (Figure 5C,D). In the MZezh2 mutant embryos, both hox genes were upregulated in the anterior half of the MZezh2 mutant embryos compared to wildtypes (Figure 5C,D) but their expression level remained lower than in the posterior half of the wildtype embryos (Figure 5C,D).
These comparative analyses of anterior and posterior parts of the embryo suggest that, upon loss of Ezh2, hox genes show an ectopic anterior expression while keeping wildtype expression levels in their normal expression domains.
Different transcription factors show various profiles of deregulation in the absence of Ezh2
To further pursue our investigation on the changes in gene expression patterns in absence of Ezh2, we performed in situ hybridization (ISH) on members from the tbx gene family of transcription factors. The tbx2a, tbx2b, tbx3a, and tbx5a genes have partial overlapping expression patterns in wildtype embryos, but also display gene specific expression domains (Figure 6A). At 24 hpf, these tbx gene family members are expressed in the dorsal region of the retina, in the heart, and the pectoral fins(40, 41). In addition, the genes tbx2a, tbx2b, and tbx3a are expressed in the otic vesicle. The genes tbx2b and tbx3a are expressed in different ganglions and neurons in anterior and posterior regions of wildtype embryos(40). Finally, expression of tbx2b can also be detected in part of pharyngeal arches 3-7 and the distal region of the pronephros and tbx3a expression can be detected in the brachial arches(39). This spatial prevalence of tbx gene expression in the anterior half of the embryo was also detected by RT-qPCR at 24 hpf, where tbx2a, tbx2b and tbx5a expression was significantly higher in the anterior than in the posterior part of wildtype embryos (Figure 6B).
ISH for these tbx genes on MZezh2 mutant embryos at 24 hpf suggests ectopic expression of these transcription factors around their normal expression pattern in the eye, the otic vesicle, and the heart, except for tbx2b (Figure 6A). This scattering in gene expression was reflected in a trend towards a higher expression in the anterior half of MZezh2 mutant embryos as detected by RT-qPCR, although only tbx2a and tbx5a results were significant (Figure 6B). In addition, ISH for tbx5a, and to a lesser extent tbx3a, showed ubiquitous expression throughout the entire body of MZezh2 mutants which was not visible in wildtypes (Figure 6A). RT-qPCR results confirmed increased expression of tbx5a in both the anterior and posterior half of the MZezh2 mutant embryos (Figure 6B).
Beside the observed ectopic expression, all tested tbx genes showed absence of expression in specific structures upon Ezh2 loss. For example, in MZezh2 mutant embryos, all four tbx genes showed no expression in the fin bud (Figure 6A). In MZezh2 mutant embryos, the gene tbx2b showed no expression in the pharyngeal arches 3-7 and the lateral line ganglions, and tbx3a was not observed in the branchial arches (Fig 6a). This absence of expression was not detected by RT-qPCR (Figure 6B) but a trend toward downregulation for tbx2b was observed in RNA-seq results on whole MZezh2 mutant embryo lysates (Figure 6C).
In addition, we tested transcription factors from other families which were targeted by H3K27me3 in wildtype. The transcription factor isl1, expressed in all primary neurons(42), showed a similar absence of expression in the fin bud and the cranial motor neurons in the midbrain (trigeminal, facial and vagal motor neurons), as observed for tbx2a. Its expression was also absent in the ventral region of the eye, the facial ganglia, and in the pronephros from MZezh2 mutant embryos, where it is normally expressed in wildtype embryos(43, 44) (Figure 6A). This loss in expression in MZezh2 mutant embryos was not detected by RT-qPCR but a clear tendency toward downregulation was detected by RNA-seq (Figure 6B,C). Even more surprising was the expression pattern of gsc in the MZezh2 mutant embryos. Whereas all the wildtype embryos show highly specific expression in the telencephalon and diencephalon nuclei, the branchial arches, and the otic vesicle(39), gsc expression was lost and replaced by a weak but ubiquitous expression in MZezh2 mutant embryos (Figure 6A). This observation was confirmed by RT-qPCR and RNA-seq where upregulation of gsc was clearly detected in MZezh2 mutant embryos (Figure 6B,C).
Taken together, these spatial expression analyses showed that the tested transcription factors are expressed outside their normal wildtype expression boundaries in MZezh2 mutant embryos at 24 hpf. Furthermore, expression of some of these genes is lost in specific tissues in the MZezh2 mutant embryos.
Discussion
Here, we showed for the first time the genome-wide binding patterns of Ezh2 and Rnf2, the catalytic subunits of PRC2 and PRC1, respectively, in 24 hpf zebrafish embryos. The overall overlap between the binding patterns of Ezh2, Rnf2, and the PcG related epigenetic mark H3K27me3 suggests that the PcG-mediated gene repression mechanisms(6) are evolutionary conserved in zebrafish development. The complete loss of H3K27me3 in MZezh2 mutant embryos reveals that Ezh2 is the only methyltransferase involved in trimethylation of H3K27 during early zebrafish development. This result was expected as Ezh1, the only other H3K27me3 methyltransferase, is not maternally loaded nor expressed in the zebrafish embryo until at least after 1 dpf(28, 45-47). In addition, proteomics results showed decreased protein expression of most PRC2 subunits. This could indicate a destabilization of PRC2 in absence of the catalytic subunit in MZezh2 mutant embryos. We could therefore confirm that zebrafish embryos can form a normal body plan in absence of PRC2-mediated gene repression.
The total loss of Rnf2 binding in the MZezh2 mutants suggests that only the canonical pathway, in which PRC2 is required for PRC1 recruitment, is active during this stage of development. This absence of PRC1 recruitment to the chromatin is not caused by an absence of the complex in the MZezh2 mutants, since most of the PRC1 subunits were detectable and even upregulated as shown by proteomic analysis. This is in contrast with studies in cultured mouse embryonic stem cells where non-canonical PRC1 complexes were shown to be recruited to developmental regulated genes independently of PRC2(12, 14). This difference could be explained by the complete absence of H3K27me3 since fertilization in MZezh2 mutant embryos, whereas other studies used conditional knockdown. Therefore, our model suggests that the PRC2-independent recruitment of PRC1 during early development can occur if PRC1 recruitment was first primed by a PRC2-dependent mechanism happening earlier during development.
As repressive and activating marks are known to antagonize each other(16), one could expect an increase in the H3K4me3 level deposited by TrxG proteins in absence of H3K27me3 associated with an increase in gene activation. However, the effects on H3K4me3 deposition, gene expression, and protein expression are limited in MZezh2 mutant embryos at 24 hpf. This observation is in agreement with the near complete absence of phenotype at this developmental time point. Thus, it appears that transcriptional regulation during zebrafish development is largely PRC2-independent until later stages of development, when maintenance of cellular identity is required. These results were unexpected, as PRC2 is described to be essential during mammalian development already during gastrulation(19-21). Exploring distribution of other repressive epigenetic marks in MZezh2 mutant embryos could reveal compensation mechanisms safeguarding gene repression during early development, as it was previously shown that absence of both DNA methylation and H3K9me3 could influence H3K27me3, and reciprocally(48-50). Possibly, the external development of the zebrafish could also explain this difference in phenotype.
Although limited, genes that show a gain in H3K4me3 deposition or in expression upon loss of ehz2 are mainly transcription factors targeted by H3K27me3 in wildtype embryos. That only a minor fraction of all H3K27me3 target genes gained expression (4.1%) suggests different mechanisms of regulation of PcG target genes. Our hypothesis is that control of gene expression by signaling pathways and transcription factor networks(51) is a robust mechanism and can be maintained until 1 dpf in absence of repression by PcG complexes. At 1 dpf, in absence of PcG-mediated repression, the first derepressed genes will be the genes subjected to the most fine-tuned transcriptional control, such as genes controlled by precise morphogen gradients. For example, it was shown that PRC2 attenuates expression of genes controlled by retinoic acid signaling(52, 53). In vertebrates, and most particularly zebrafish, retinoic acid signaling is responsible for induction of formation of, among others, the forelimb field(54, 55), dorsoventral patterning of eyes(56, 57), hindbrain patterning(58), hox gene expression(59), and the development of other organs(60). All these processes are affected in MZezh2 mutant embryos at 24 hpf and onwards and, therefore, could be explained by a defect in the response to retinoic acid signaling.
Spatial analysis of gene expression revealed different effects on gene expression patterns caused by loss of Ezh2, which could not be detected by RNA-seq on whole embryo lysates. Anterior-posterior specific RT-qPCR showed that hox genes become abnormally expressed in the anterior half of the MZezh2 mutant embryos; whereas expression levels in the posterior half remained unchanged. These results are in agreement with previous studies showing ectopic expression of hox genes in PRC1 and PRC2 zebrafish mutants(28, 61). Other transcription factors, such as the tbx gene family members, showed more diverse patterns of deregulation compared to hox genes. ISH and RT-qPCR showed that, among the tbx genes examined, some were overexpressed outside their normal expression domains (tbx2a, tbx3a, and tbx5a), whereas others were also ubiquitously upregulated (tbx3a and tbx5a). The case of eye patterning is a good example of the defect in control of gene expression pattern in MZezh2 mutant embryos. In wildtype embryos, at 24 hpf, tbx genes are expressed in the dorsal part of the eye whereas isl1 is expressed in the ventral part. Upon loss of Ezh2, our ISH results showed that the expression of the tbx genes expands to the whole eye whereas isl1 disappears from the ventral region. We concluded that Polycomb-mediated repression is therefore responsible for maintenance of expression domains rather than control of expression level.
Expression analysis by ISH for tbx and hox genes as well as for isl1 also showed loss of expression in specific structures in MZezh2 mutant embryos. We reasoned that the absence of expression of hox and tbx genes in the fin bud could be a secondary effect due to the absence of this structure in MZezh2 mutants(28). The same phenomenon could explain the lack of detection of tbx2b and isl1 in pharyngeal arches, pronephros, and lateral line ganglions. The case of gsc expression is more striking, as its normal expression pattern is totally abolished and replaced by a ubiquitous expression pattern. The gsc gene is known to be expressed in the Spemann organizer during gastrulation and therefore all cells will transiently express gsc when undergoing gastrulation(62, 63). In absence of Ezh2, gsc expression could remain active in all cells after leaving the Spemann organizer, leading to a ubiquitous expression pattern and impaired tissue specific expression in 24 hpf MZezh2 mutant embryos.
To conclude, our results show that early embryonic development, including germ layer formation and cell fate specification, is independent of PcG-mediated gene repression until axis are formed and organs specified. PcG-mediated gene repression is then required to control precise spatial restricted expression of specific transcription factors. We hypothesize that subtle changes in expression of these master gene regulators subsequently will lead to progressive and accumulating changes in gene network regulation and result in loss of tissue identity maintenance. Our results constitute a major advance in the understanding of the mechanisms of PcG-mediated epigenetic gene regulation during vertebrate development.
Materials and methods
Zebrafish genetics and strains
Zebrafish (Danio rerio), were housed according to standard conditions(64) and staged according to Kimmel et al.(65). The ezh2 nonsense mutant (hu5670)(28), Tg(H2A::GFP)(66), and Tg(vas::eGFP)(67) zebrafish lines have been described before. Genotyping of the ezh2 allele was performed as previously described(28) with following adaptations: different primer pairs were used for PCR and nested PCR (Supplementary Table S2), of which the restriction profile is shown on Supplementary Figure 1F. All experiments were carried out in accordance with animal welfare laws, guidelines, and policies and were approved by the Radboud University Animal Experiments Committee.
Germ cell transplantation
Germ cell transplantation was performed as described previously(28). For all experiments below, ezh2 germline mutant females were crossed with ezh2 germline mutant males to obtain 100% MZezh2 mutant progeny. The germline wild-type sibling males and females obtained during transplantation were used to obtain 100% wildtype progeny with similar genetic background and are referred to as wildtype. The embryos used were all from the first generation after germline transplantation.
Western blotting
At 3.3 hpf, 50 embryos were collected, resuspended in in 500 µl ½ Ringer solution (55 mM NaCl, 1.8 mM KCl, 1.25 mM NaHCO3) and forced through a 21G needle and a cell strainer in order to remove the chorion and disrupt the yolk. At 24 hpf, 20 embryos were collected and resuspended by thorough pipetting in 500µl ½ Ringer solution in order to disrupt the yolk. The samples of 3.3 and 24 hpf were centrifuged for 5 minutes at 3,500 g at 4°C and washed two additional times with 500 µl ½ Ringer solution. The embryo pellet was frozen in liquid nitrogen and stored at −80°C. Whole protein extraction was performed by adding 40 µl of RIPA buffer (100 mM Tris-HCl pH 8, 300 mM NaCl, 2% NP-40, 1% Sodium Deoxycholate, 0.2% SDS, 20% glycerol, 1x cOmplete EDTA-free protease inhibitor cocktails from Sigma) and sonication for 2 cycles of 15s ON and 15s OFF on medium power at 4°C on a PicoBioruptor (Diagenode). After 10 minutes incubation at 4°C, embryo lysates were centrifuged for 12 minutes at 16,000 g at 4°C and supernatant was transferred in a new tube. 20 µg protein was mixed with SDS containing sample loading buffer, denaturated at 95°C for 5 minutes and analyzed by Western blot analysis. Antibodies used for immunoblotting are described in Supplementary Table S3 HRP-conjugated anti-rabbit secondary antibody was used (Supplementary Table S3) and protein detection was performed with ECL Select Western Blotting Detection Reagent (GE Healthcare, RPN2235) on an ImageQuant LAS 4000 (GE Healthcare).
ChIP-sequencing
For chromatin preparation, embryos from a germline mutant or germline wildtype incross were collected at 24 hpf and processed per batches of 300 embryos. Embryos were first dechorionated by pronase (0.6 µg/µl) treatment and then extensively washed with E3 medium. Subsequently, embryos were fixed in 1% PFA (EMS, 15710) for 15 minutes at room temperature and fixation was terminated by adding 0.125M glycine and washed 3 times in cold PBS. Yolk from fixed embryos was disrupted by pipetting the fixed embryos 10 times with a 1 ml tip in 600 µl of ½ Ringer solution (55 mM NaCl, 1.8 mM KCl, 1.25 mM NaHCO3) and incubated for 5 minutes at 4°C on a rotating wheel. Embryos were pelleted by centrifuging 30 seconds at 300 g and the supernatant was removed. De-yolked embryos were resuspended in 600 µl sonication buffer (20 mM Tris-HCl pH 7.5, 70 mM KCl, 1 mM EDTA, 10% glycerol, 0.125% NP40, 1x cOmplete EDTA-free protease inhibitor cocktails from Sigma) and homogenized with a Dounce homogenizer (6 strokes with pestle A, followed by 6 strokes with pestle B). Homogenates were sonicated for 6 cycles of 30 seconds ON/30 seconds OFF on a PicoBioruptor (Diagenode), centrifuged for 10 minutes at 16,000 g at 4°C, and the supernatant containing the chromatin was stored at −80°C. 20 µl of the supernatant was subjected to phenol-chloroform extraction and ran on an agarose gel to verify that a proper chromatin size of 200-400 bp was obtained.
For ChIP, 100 µl of chromatin preparation (corresponding to 50 embryos) was mixed with 100 µl IP-buffer (50 mM Tris-HCL pH 7.5, 100 mM NaCl, 2 mM EDTA, 1% NP-40, 1x cOmplete EDTA-free protease inhibitor cocktails from Sigma) and antibody (for details on antibodies used see Supplementary Table S3) and incubated overnight at 4°C on a rotating wheel. For immunoprecipitation, 20 µl of protein G magnetic beads (Invitrogen, 1003D) were washed in IP buffer and then incubated with the chromatin mix for 2 hours at 4°C on a rotating wheel. Samples were washed in 500 µl washing buffer 1 (IP-buffer + 0.1% Sodium Deoxycholate), followed by washing in washing buffer 2 (washing buffer 1 + 400mM NaCl), washing buffer 3 (washing buffer 1 + 250mM LiCl), washing buffer 1 and a final wash in 250 µl of TE buffer. All washes were 5 minutes at 4°C on a rotating wheel. Chromatin was eluted from the beads by incubation in 100 µl of elution buffer (50 mM NaHCO3 pH 8.8, 1% SDS) for 15 minutes at 65°C at 900 rpm in a thermomixer. The supernatant was transferred in a clean 1.5 ml tube. Elution was repeated a second time and both supernatants were pooled. The eluate was treated with 0.33 µg/µl RNaseA for 2 hours at 37°C. Samples were then decrosslinked by adding 10 µl of 4M NaCL and 1 µl of 10mg/ml proteinase K and incubated overnight at 65°C. DNA was then purified using MinElute Reaction Clean-Up kit (Qiagen, 28204).
1-5 ng of DNA was used to prepare libraries with the KAPA Hyper Prep Kit (KAPABiosystems, KK8504) and NEXTflex ChIP-Seq Barcodes for Illumina (Bioo Scientific, 514122) followed by paired-end 43bp sequencing on an Illumina NextSeq500 platform. All ChIP-seq were performed in two biological replicates, except for Rnf2 in wildtype embryos which was performed once and H3K27ac which was performed in triplicate in both wildtype and mutant embryos.
RNA-sequencing
Ten to twenty manually dechorionated 24 hpf embryos of a germline mutant incross and a germline wildtype incross were homogenized in TRIzol (Ambion, 15596018). Subsequently, the Quick RNA microprep kit (Zymo Research, R1051) was used to isolate RNA and treat the samples with DNAseI. Most samples were depleted from rRNA using the Ribo-Zero rRNA Removal Kit (Illumina, MRZH11124), followed by fragmentation, cDNA synthesis, and libraries were generated using the KAPA Hyper Prep Kit (KAPABiosystems, KK8504). Sequencing libraries were paired-end sequenced (43 bp read-length) on an Illumina NextSeq500 platform. However, two samples per genotype were generated with the TruSeq Stranded Total RNA Library Prep Kit with Ribo-Zero (Illumina, RS-122-2201) and single-end sequenced (50 bp read-length) on an Illumina HiSeq 2500. For wildtype and MZezh2 mutant embryos, 6 and 7 biological replicates were used, respectively.
Mass spectrometry
At 24 hpf, 50 embryos were collected, dechorionated, and resuspended by gently pipetting in 500 µl deyolking buffer (1/2 Ginzburg Fish Ringer without Calcium: 55 mM NaCl, 1.8 mM KCl, 1.25 mM NaHCO3, 1x cOmplete EDTA-free protease inhibitor cocktail from Sigma) and incubated for 5 minutes in a Thermomixer at RT at 1,100 rpm to disrupt the yolk. The samples were then centrifuged for 30 seconds at 400 g and the pellet was washed two times in 0.5 ml wash buffer (110 mM NaCl, 3.5 mM KCl, 2.7 mM CaCl2, 10mM Tris/Cl pH8.5, 1x cOmplete EDTA-free protease inhibitor cocktail from Sigma) for 2 minutes in a Thermomixer at RT and 1,100 rpm, followed by 30 seconds centrifugation at 400 g. Washed pellets were lysed in 100 µl RIPA buffer (50 mM Tris pH8.0, 150 mM NaCl, 0.1% SDS, 1% NP-40, 0.5% DOC, 20% glycerol, 1 mM Sodium Orthovanadate, 1x cOmplete EDTA-free protease inhibitor cocktails from Sigma) and sonicated for 2 cycles of 15s ON and 15s OFF on full power at 4°C on a Bioruptor (Diagenode). Samples were incubated for 1 hour on a rotating wheel at 4°C and centrifuged 10 minutes at 12,000 g and 4°C. Supernatant was flash frozen and stored at −80°C. After Bradford analysis, 100 µg protein lysate was used for FASP-SAX as previously described(68). The peptide fractions were separated on an Easy nLC 1000 (Thermo Scientific) connected to a Thermo scientific Orbitrap Fusion Tribrid mass spectrometer. MS and MS/MS spectra were recorded in a top speed modus with a run cycle of 3s using Higher-energy Collision Dissociation (HCD) fragmentation. The raw mass spectrometry data were analyzed using the MAXQuant software version 1.6.0.1 (http://www.ncbi.nlm.nih.gov/pubmed/19029910) with default settings. Data was searched against the Danio rerio data base (UniProt June 2017). The experiment was performed with biological triplicates for each condition.
Bioinformatics analyses
For ChIP-sequencing analysis, fastq files were aligned to GRCz10 zebrafish genome version using BWA-MEM (version 0.7.10-r789) for paired-end reads(69). Duplicated and multimapping reads were removed using samtools(70) version 1.2 and Picard tools (http://broadinstitute.github.io/picard) version 2.14.1. MACS2(71) version 2.1.1 was used to call peaks from each aligned bam files using an Input track from 24 hpf wild-type embryos as control sequence. Peaks separated by less than 1kb distance were merged, peaks that were called using Input alone were removed from all data sets using bedtools suit version 2.20.1, and the intersection between the two replicates for each antibody in each condition was used to define the definitive peak sets. For visualization, fastq files from duplicate ChIP-sequencing were merged, aligned as described above, and transformed into bigwig alignment files using bam2bw version 1.25. Peak lists were analyzed using bedtools and heatmaps were produced using deepTools plotHeatmap(72) version 2.5.3.
For RNA-sequencing analysis, read counts per gene were retrieve using GeneCounts quantification method from STAR(73) version 2.4.0 and the GRCz10 zebrafish genome version with Ensembl annotation version 87 as reference. Differential expression analysis was calculated with DESeq2(74) version 1.14.1.
For proteomics analysis, differential expression of protein between conditions was assessed with DEP(75) version 1.2.0.
Gene Ontology analyses on selected genes were performed using DAVID bioinformatics resources(76) version 6.8 and anatomical term enrichment was done using ZEOGS(77). GSEA were performed with the GSEA software from the Broad Institute(35, 36) version 3.0. To analyze the proteomics data, a pre-ranked list was generated by z-scoring the proteins based on their log2FC calculated with DEP. Only proteins with an Ensembl accession number were consider for further analyses.
Whole mount in situ hybridization
Embryos at 24 hpf were dechorionated and fixed overnight at 4°C in 4% PFA in PBST (0.1% Tween), after which they were gradually transferred to 100% methanol. Prior to ISH, embryos were gradually transferred back to PBST and, subsequently, ISH was performed as described previously(78). The embryos were imaged by light microscopy on a Leica MZFLIII, equipped with a DFC450 camera.
RT-qPCR analyses
Total RNA was isolated using Trizol from 20 flash-frozen dechorionated 24 hpf wildtype and MZezh2 mutant embryos cut in two with tweezers. Reverse transcription was achieved using Superscript III (Invitrogen, 18080093) and poly-dT primers. Standard qPCR using SYBR Green (iQ SYBR Green Supermix, BioRad, 1708880) was performed using the primers shown in Supplementary Table S2. Relative expression was calculated based on expression of housekeeping genes β-actin. Calculations were based on at least 3 independent replicates for both conditions.
Data availability
The sequencing data have been submitted to the NCBI Gene Expression Omnibus (GEO; http://www.ncbi.nlm.nih.gov/geo/) under accession number GSE119070. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE(79) partner repository with the dataset identifier PXD010922. Reviewers can obtain access to the datasets via login information provided to the editor.
Authors’ contributions
JR contributed to conception and design of the study, performed and analyzed experiments, and wrote and edited the manuscript. NDC contributed to conception and design of the study, performed RNA extraction and ISH imaging, and edited the manuscript. MA performed library preparation and RT-qPCR. KA performed Western blot experiments, acquired funding, and edited the manuscript. DME assisted with bioinformatics analyses and edited the manuscript. PJM and BRC performed RNA-seq experiments and analysis, acquired funding, and edited the manuscript. PWTCJ and MV performed proteomics experiments and analysis and edited the manuscript. LMK contributed to conception and design of the study, performed experiments, acquired funding, and wrote and edited the manuscript.
Competing interests
The authors declare no competing interests.
Acknowledgements
We thank J. Bakkers, from the Hubrecht Institute, for providing the tbx2a, tbx2b, tbx3a, tbx5a, and isl1 plasmids and J. den Hertog from the Hubrecht Institute for providing the gsc plasmid for ISH probe generation. We thank T. Spanings and A. van der Horst from the Radboud University for excellent zebrafish husbandry and E. Janssen-Megens from the Radboud University for excellent technical support. We thank R. Lindeboom, from the Radboud University, for computational advice. We thank Dr. G.J.C. Veenstra, from the Radboud University, and his team for fruitful discussions. We thank Dr. R. Knight, from the King’s College London, for his help with ISH analysis. The work was funded by the Innovative Research scheme of the Netherlands Organisation for Scientific research (www.nwo.nl, NWO-Vidi 864.12.009, NWO-Meervoud 836.13.003 L.M.K.), the Radboud University Nijmegen Medical Centre tenure track fellowship (www.radboudumc.nl, L.M.K.), the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie Grant (Agreement No. 705939, K.A.), the Howard Hughes Medical Institute and the Huntsman Cancer Institute core facilities (CA24014, B.R.C.), and the Eunice Kennedy Shriver National Institute of Child Health and Human Development of the NIH (T32HD007491, P.J.M.).
References
- 1.↵
- 2.↵
- 3.↵
- 4.↵
- 5.↵
- 6.↵
- 7.↵
- 8.↵
- 9.↵
- 10.↵
- 11.↵
- 12.↵
- 13.
- 14.↵
- 15.↵
- 16.↵
- 17.↵
- 18.↵
- 19.↵
- 20.
- 21.↵
- 22.↵
- 23.↵
- 24.
- 25.
- 26.↵
- 27.↵
- 28.↵
- 29.↵
- 30.↵
- 31.↵
- 32.↵
- 33.↵
- 34.↵
- 35.↵
- 36.↵
- 37.↵
- 38.↵
- 39.↵
- 40.↵
- 41.↵
- 42.↵
- 43.↵
- 44.↵
- 45.↵
- 46.
- 47.↵
- 48.↵
- 49.
- 50.↵
- 51.↵
- 52.↵
- 53.↵
- 54.↵
- 55.↵
- 56.↵
- 57.↵
- 58.↵
- 59.↵
- 60.↵
- 61.↵
- 62.↵
- 63.↵
- 64.↵
- 65.↵
- 66.↵
- 67.↵
- 68.↵
- 69.↵
- 70.↵
- 71.↵
- 72.↵
- 73.↵
- 74.↵
- 75.↵
- 76.↵
- 77.↵
- 78.↵
- 79.↵