Genome-wide transcriptional change during the mitosis-to-meiosis transition
To assess genome-wide gene expression changes during the mitosis-to-meiosis transition in spermatogenesis, we have recently performed RNA-seq analysis at three representative time points before, during, and after meiosis (Fig. 1a) [23]. Because purified spermatogonia consist of a heterogeneous cell population, it is difficult to obtain a large number of homogenous cells for ChIP-seq analysis. Because of this, we used cultured GS cells [24] as a representative stage of the mitotic phase of spermatogenesis. Our cultured GS cells exhibited a gene expression profile similar to THY1+ undifferentiated spermatogonia cells purified from mouse testes (Fig. 1b), confirming that the GS cells recapitulate undifferentiated spermatogonia in vivo. For meiotic and postmeiotic stages, we used purified PS and RS, respectively (Fig. 1a). To identify the unique features of germline transcriptomes during spermatogenesis, we compared RNA-seq data from these cell types to the published RNA-seq data obtained from THY1+ undifferentiated spermatogonia, ES cells, somatic cells, and tissues (see Methods section). A heatmap analysis of 17,213 genes expressed (reads per kilobase per million (RPKM) >3) in at least one condition revealed that a significant transcriptional change occurs during the mitosis-to-meiosis transition, and that the transcriptomes of PS and RS are largely different from that of the mitotic phase of spermatogenesis as well as other somatic cells and tissues (Fig. 1b). Two distinct features that are common in PS and RS transcriptomes are activation of late spermatogenesis genes, as previously described [17, 25, 26], and suppression of a large group of genes that are commonly expressed in the somatic phase and spermatogenesis progenitor cells. Herein we will refer to the latter group of genes as somatic/progenitor genes. This analysis suggests that there is a massive transcriptional change at the mitosis-to-meiosis transition during differentiation of mitotic spermatogonia into meiotic spermatocytes, and that transcriptomes during the late stages of spermatogenesis are significantly different from that of somatic lineages [23].
Unique features of transcriptomes during meiosis and postmeiosis
To define lists of somatic/progenitor genes and late spermatogenesis genes, the transcriptomes were separated into gene groups according to the following criteria (Fig. 1c,d,e): (1) four-fold change in pairwise comparisons between GS and PS, GS and RS, or PS and RS; (2) adjusted P value ≤0.05 for significance of differential expression among the cell types; and (3) RPKM in at least one cell type ≥5. Because sex chromosomes are subject to unique epigenetic programming during meiosis and postmeiosis, we analyzed expression profiles of genes located on autosomes and sex chromosomes separately.
On autosomes, we found that 2,826 genes were commonly active in PS and RS (abbreviated as PS/RS active genes hereafter), and further defined the list of PS- or RS-specific active and inactive genes (abbreviated as PS active, PS inactive, RS active, RS inactive, respectively; Fig. 1c). Gene ontology (GO) enrichment analysis revealed that male reproduction-associated genes are significantly enriched in PS/RS active, PS active genes, and RS active genes (Fig. 1d). On the other hand, 2,636 autosomal genes were commonly inactive in PS and RS (abbreviated as PS/RS inactive genes hereafter; Fig. 1c). GO enrichment analysis reveals that the PS/RS inactive gene set is enriched with genes involved in somatic functions such as blood vessel development and tissue morphogenesis, suggesting that they are presumably dispensable during the late stages of spermatogenesis (Fig. 1d). We also identified 1,044 constitutively active genes and 8,910 constitutively inactive genes on autosomes in all three cell types (see Methods).
Because of the paucity of Y-linked genes, we focused on the X chromosome for detailed analysis. On the X chromosomes, 225 genes were significantly repressed in both PS and RS, consistent with sex chromosome inactivation (Fig. 1c), and the GO analysis demonstrated that this group of genes was highly associated with chromatin modification (Fig. 1d). On the other hand, 102 X-linked genes escaped sex chromosome inactivation and were predominantly expressed in RS. Interestingly, this group of genes is specifically expressed in the germline (Additional file 1: Figure S1), and is disproportionately enriched on the X chromosome (102/994, 10.3 %, P <2.2e-16, chi-square test) when compared with those on the autosomes (Fig. 1f). Additionally, we identified 500 X-linked constitutively inactive genes in all three cell types (see Methods). Taken together, our RNA-seq data are in accord with previous gene expression studies [12, 17, 21, 25–27], and these results confirm distinct regulation between autosomes and the X chromosome in spermatogenesis.
Distinct epigenetic landscapes between autosomes and the X chromosome during spermatogenesis
To elucidate the epigenetic principles of mouse spermatogenesis, we performed ChIP-seq chromatin profiling in GS, PS, and RS cells. In particular, we examined the distribution of RNA polymerase II (RNAPII) and representative active epigenetic modifications such as H3K4me2, H3K4me3, H4K8ac, H4K16ac, and histone lysine crotonylation (Kcr). In addition to active modification, we examined representative silent modifications H3K27me3 and H3K9me2.
To account for the distinct regulation of gene expression from autosomes and the X chromosome during spermatogenesis, we first compared the average tag density (ATD) profiles of these modifications around transcription start sites (TSSs) between all autosomal genes and all X chromosome-linked genes during spermatogenesis (Fig. 2a,b,c). Based on the limited availability of annotated sequences on the Y chromosome, Y chromosome data were excluded from our analysis hereafter. Consistent with the phenomena of almost complete silencing in MSCI and postmeiotic RS-specific escape gene activation, RNAPII was largely depleted from the X chromosome in PS and slightly increased in RS (Fig. 2b,c). Although distribution of H3K4me2 and H3K4me3 was comparable between autosomes and X chromosome in GS (Fig. 2a), H3K4me2 was slightly enriched on the X chromosome in PS consistent with the cytological localization that H3K4me2 starts to accumulate on the sex chromosomes during the late pachytene stage [21, 28], whereas H3K4me3 was enriched on autosomes at this stage (Fig. 2b). H4K8ac, Kcr, and H4K16ac were also enriched on autosomes in PS, and, curiously, H4K16ac was continuously low on the X chromosomes compared to autosomes in all three cell types (Additional file 1: Figure S2). In contrast to these active modifications, H3K9me2 was slightly enriched on the X chromosome compared to autosomes in GS cells (Fig. 2a). In PS and RS, consistent with cytological localization [17], H3K27me3 was largely depleted from the X chromosome, whereas H3K9me2 was enriched there (Fig. 2b,c). In contrast, H3K27me3 accumulated on autosomes in PS and was highly enriched on autosomal TSSs in RS. Therefore, these results suggest that autosomes and the X chromosomes are subject to distinct modes of epigenetic regulation during spermatogenesis. Because of the distinct regulation, we proceeded to analyze the epigenomes of autosomes and sex chromosomes separately.
Autosomal late spermatogenesis genes are silent, but poised in GS cells for later activation in PS
We first focused on the events on the autosomes and investigated whether the genes specifically regulated during spermatogenesis undergo epigenetic changes during differentiation. Heatmap analyses revealed that, in GS cells, H3K4me2 and H3K4me3 were highly accumulated on the active genes (both constitutively active genes and PS/RS inactive genes) (Fig. 3a). ATD analysis revealed that RNAPII and H3K4me3 were highly accumulated on TSSs of these genes, and that H3K4me2 localization is broader than localization of RNAPII and H3K4me3, and accumulated on the region surrounding TSSs of these genes (Fig. 3b). H4K8ac and Kcr were also accumulated around TSSs of constitutively active genes, but were less intense on the PS/RS inactive genes that are highly expressed in GS cells (Additional file 1: Figure S3), suggesting that, in GS cells, gene activation is distinctly regulated between PS/RS inactive genes and constitutively active genes. Consistent with this notion, H4K16ac accumulated on TSSs of constitutively active genes, but not on the TSSs of PS/RS inactive genes (Additional file 1: Figure S3).
Notably, RNAPII, H3K4me2, and H3K4me3 were largely present on PS/RS active genes even though these genes were silent in GS cells (Fig. 3a,b). Notably, ATD of these modifications on PS/RS active genes overlapped with that of PS active genes in GS cells, but RS active genes did not exhibit enrichment of active modifications in GS cells (Fig. 3c). Further, the H3K27me3 level of PS/RS active genes was lower than that of constitutively inactive genes, whereas H3K9me2 did not show this reduction (Fig. 3d). These results suggest that the autosomal genes activated in PS are already epigenetically poised by deposition of active modifications and RNAPII, as well as by reduction of H3K27me3, for future activation. Similar epigenetic gene poising was observed in T cells for genes that are inducible during T cell activation and in other systems [29, 30]. Taken together, we conclude that activation of autosomal late spermatogenesis genes in PS is preprogrammed in GS cells.
H3K4me2 remained on somatic/progenitor genes after gene inactivation in PS
Next, we sought to examine how the meiosis-specific transcriptome is regulated for autosomes during the PS stage. At the PS active gene synaptonemal complex protein 3 (Sycp3), active modifications such as H3K4me3, H4K8ac, H4K16ac, and Kcr were highly accumulated at the TSS, and H3K4me2 exhibited a broader peak of enrichment near the TSS (Fig. 4a). These profiles of active modifications were common among PS/RS active genes, PS active genes, and constitutively active genes, suggesting that PS-specific gene activation is regulated by a similar epigenetic mechanism with that of constitutively active genes in PS (Fig. 4b, Additional file 1: Figure S4). On the other hand, genes inactivated in PS such as vimentin (Vim) exhibited a distinct feature compared to the constitutively inactive genes: H3K4me2 largely remained in PS at PS/RS inactive genes and PS inactive genes although RNAPII and H3K4me3 were largely depleted (Fig. 4a,c). In addition, the silent modification H3K27me3, but not H3K9me2, was highly enriched at PS/RS inactive genes (Fig. 4a,e). These results suggest that bivalent chromatin signatures such as H3K27me3 with H3K4me2 are associated with PS/RS inactive genes in PS.
We further investigated whether there is any epigenetic signature that predicts RS-specific gene activation in PS. Both H3K4me2 and Kcr were broadly enriched at RS active genes in PS, but H3K4me3 did not show enrichment compared to that of constitutively inactive genes (Fig. 4d, Additional file 1: Figure S4). Thus, on RS active genes, H3K4me2 and Kcr, but not H3K4me3, are already established in PS for future activation in RS.
Unique epigenetic landscape of late spermatogenesis genes on autosomes in RS
We next investigated the epigenetic signature of autosomes in postmeiotic RS specifically focusing on the RS active genes. Interestingly, active epigenetic modifications were present not just at the promoters of the RS active genes, such as spermatogenesis-associated 20 (Spata20), but were spread out broadly from TSSs to transcription end sites (TESs) (Fig. 5a). In the heatmap and ATP analysis, H3K4me2 was highly enriched in the gene bodies but less enriched at the TSSs of RS active genes compared to constitutively active genes (Fig. 5b,c). Contrastingly, H3K4me2, H4K8ac, and Kcr were highly enriched on the TSSs of RS active genes even compared to constitutively active genes (Fig. 5c), while H4K16ac did not show such a difference (Additional file 1: Figure S5). These results reveal the unique epigenetic landscape of late spermatogenesis genes on autosomes in RS.
Autosomal somatic/progenitor genes are silent in RS via the deposition of bivalent epigenetic marks
We next examined the epigenetic signature of somatic/progenitor genes inactivated in RS. H3K4me2 remained around the TSSs of PS/RS inactive genes such as Vim (Fig. 5a). RNAPII and H3K4me2/3 were present on both RS inactive genes and PS/RS inactive genes (Fig. 5b,d). Importantly, H3K27me3 was deposited at the TSS of PS/RS inactive genes in RS compared to PS/RS inactive genes in PS, while H3K9me2 did not exhibit this feature (Figs. 4e and 5a,e). This suggests that the deposition of bivalent marks at the TSS of somatic/progenitor genes occurred in the transition between PS and RS without expression changes. To determine whether somatic/progenitor genes are poised for activation after fertilization, we compared their gene expression during spermatogenesis and in ES cells, which represent the post-fertilization inner cell mass of blastocysts. Consistent with our global expression analysis (Fig. 1b), somatic/progenitor genes (RS and PS/RS inactive genes) are expressed in ES cells, whereas late spermatogenesis genes (PS/RS and RS active genes) are not expressed in ES cells (Fig. 5f). Therefore, these results suggest that the somatic/progenitor program is suppressed in late spermatogenesis, but poised for activation after fertilization. Importantly, in contrast to the class of bivalent domains on developmental promoters that are consistently silent throughout the male germline [11, 13], our analysis reveals a new class of bivalent genes that are expressed in ES and GS cells but are temporarily suppressed in PS and RS. This suggests a novel function of bivalent domains: suppression of the somatic/progenitor program during late spermatogenesis.
X-linked genes subject to MSCI and postmeiotic silencing are poised for activation after fertilization without the formation of bivalent domains
Because sex chromosomes undergo MSCI and are regulated separately from autosomes in spermatogenesis (Fig. 2), we investigated the epigenetic landscape of the X chromosomes separately from that of autosomes. In Fig. 1, we classified X-linked genes into two major categories: the major group consists of 225 genes that are active in GS but are subject to both MSCI and postmeiotic silencing (PS/RS inactive); the other group consists of 102 genes that are not expressed in GS, but are highly activated in RS (RS active). This latter group is also referred to as escape genes because they escape chromosome-wide postmeiotic silencing in RS and are activated [20, 21].
First, we examined the developmental changes in the epigenetic landscape of representative genes from each group during spermatogenesis. The PS/RS inactive gene Timp1 was marked with H3K4me2 and H3K4me3 in GS, where it is transcribed (Fig. 6a). Upon entry into meiosis, Timp1 was silenced by MSCI, RNAPII disappeared from the TSS, and the H3K9me2 level increased. However, H3K4me2 and other active modifications remained on the Timp1 locus, albeit at a lower level in PS. In agreement with the Timp1 locus, H3K4me2 remained near the TSS of PS/RS inactive X-linked genes in PS despite the disappearance of RNAPII (Fig. 6c,d). These results suggest that MSCI and postmeiotic silencing are established without complete removal of active modifications.
In RS, H3K4me2 and H4K16ac are still present on the Timp1 locus (Fig. 6a). Paradoxically, the H3K4me2 level at PS/RS inactive X-linked genes was higher than that on RS active X-linked genes at the TSS, and PS/RS inactive X-linked genes had levels of RNAPII and H3K4me3 comparable to that of RS active X-linked genes (Fig. 6e). Importantly, this group of genes was highly expressed in ES cells (Fig. 6f). These results suggest that PS/RS inactive X-linked genes are poised for activation after fertilization, as is the case with PS/RS inactive autosomal genes. However, H3K27me3 was largely depleted from the X chromosome in PS and RS (Fig. 2, Additional file 1: Figure S6). Therefore, PS/RS inactive X-linked genes are poised in RS without the formation of typical bivalent domains.
Distinct epigenetic features underlie RS-specific gene activation on the X chromosome
As described above, we found that PS/RS active autosomal genes are poised in GS cells for activation during meiosis (Fig. 3). Unlike autosomal genes, X-linked genes that are expressed at later stages possess only a low level of active modifications in GS cells (Fig. 6c, Additional file 1: Figure S6). This result suggests that the X-linked genes are not poised in GS cells for activation during meiosis, which is in accordance with the existence of MSCI and which supports the notion that autosomes and the X chromosome are distinctly regulated in GS cells prior to entry into meiosis.
Next, we examined changes in the epigenetic landscape of RS active X-linked genes during spermatogenesis. On the RS active X-linked gene Akap4, which regulates sperm motility, active epigenetic modifications were largely absent in GS cells (Fig. 6b). Upon entry into meiosis, active modifications started to accumulate broadly on the Akap4 locus with the induction of modest transcription. In RS, H3K4me3, Kcr, and RNAPII were highly accumulated around the TSS, and Akap4 was robustly expressed. Consistent with this, in ATD analysis, Kcr started to accumulate on the TSSs of RS active X-linked genes in PS (Fig. 6d), and reached a higher level in RS (Fig. 6e). H3K4me2 became enriched on the gene bodies of RS active X-linked genes (Fig. 6e), and RS active X-linked genes were not expressed in ES cells (Fig. 6f). Unlike RS active autosomal genes (Fig. 5c), RS active X-linked genes did not gain a high level of H4K8ac accumulation (Fig. 6e). Therefore, H4K8ac is specifically associated with RS active autosomal genes.
A unique feature of the RS active X-linked genes is that this group of genes escapes postmeiotic silencing of the sex chromosomes. To determine how this group can escape the chromosome-wide silencing of the sex chromosomes, we investigated the profiles of H3K9me2 on the X chromosome. H3K9me2 was consistently high in both groups of X-linked genes compared to autosomal genes and did not exhibit a difference between active and inactive genes in RS, whereas H3K27me3 levels were low (Additional file 1: Figure S6). This result suggests that RS active X-linked gene escape is activated from silent X chromosomes without removing H3K9me2 and instead depends on unique profiles of active modifications.
Previously, we have shown that RNF8 is required for the activation of a subset of escape genes from postmeiotic silencing [21]. To elucidate the regulatory mechanism underlying expression of RS active X-linked genes, we examined how unique profiles of active modifications are established on the X chromosome using the testes of Rnf8 knockout (KO) mice. Both H3K4me2 and Kcr accumulate on gene bodies and TSSs of RNF8-dependent escape genes (identified in [21]) in an RNF8-dependent manner in PS and RS (Fig. 6g,h). These results further support the conclusion that the unique localization of H3K4me2 and Kcr is important for RS-specific gene activation from the X chromosome.