1. Genetics and Genomics
Download icon

ZCWPW1 is recruited to recombination hotspots by PRDM9 and is essential for meiotic double strand break repair

  1. Daniel Wells  Is a corresponding author
  2. Emmanuelle Bitoun  Is a corresponding author
  3. Daniela Moralli
  4. Gang Zhang
  5. Anjali Hinch
  6. Julia Jankowska
  7. Peter Donnelly
  8. Catherine Green
  9. Simon R Myers  Is a corresponding author
  1. The Wellcome Centre for Human Genetics, Roosevelt Drive, University of Oxford, United Kingdom
  2. Department of Statistics, University of Oxford, United Kingdom
Research Article
  • Cited 1
  • Views 538
  • Annotations
Cite this article as: eLife 2020;9:e53392 doi: 10.7554/eLife.53392

Abstract

During meiosis, homologous chromosomes pair and recombine, enabling balanced segregation and generating genetic diversity. In many vertebrates, double-strand breaks (DSBs) initiate recombination within hotspots where PRDM9 binds, and deposits H3K4me3 and H3K36me3. However, no protein(s) recognising this unique combination of histone marks have been identified. We identified Zcwpw1, containing H3K4me3 and H3K36me3 recognition domains, as having highly correlated expression with Prdm9. Here, we show that ZCWPW1 has co-evolved with PRDM9 and, in human cells, is strongly and specifically recruited to PRDM9 binding sites, with higher affinity than sites possessing H3K4me3 alone. Surprisingly, ZCWPW1 also recognises CpG dinucleotides. Male Zcwpw1 knockout mice show completely normal DSB positioning, but persistent DMC1 foci, severe DSB repair and synapsis defects, and downstream sterility. Our findings suggest ZCWPW1 recognition of PRDM9-bound sites at DSB hotspots is critical for synapsis, and hence fertility.

eLife digest

Sexual reproduction – that is, the combination of sex cells from two different individuals to produce an embryo – is one of the many mechanisms that have evolved to maintain genetic diversity. Most human cells contain 23 pairs of chromosomes, with each chromosome in a pair carrying either a paternal or maternal copy of the same gene. To form an embryo with the right number of chromosomes, each sex cell (the egg or sperm cell) must only contain one chromosome from each pair.

Sex cells are produced from parent cells containing two sets of paternal and maternal chromosomes: these cells then divide twice to form four sex cells which contain only one chromosome from each pair. Before the parent cell divides, a process known as ‘recombination’ takes place, which allows chromosomes in a pair to exchange bits of genetic information. This reshuffling ensures that each chromosome in a sex cell is unique. A protein called PRDM9 helps control which sections of genetic information are recombined by modifying proteins attached to the chromosomes, marking them as locations for exchange. The DNA at each of these sites is then broken and repaired using the genetic sequence of the chromosome it is paired with as a template, thus causing the two chromosomes to swap genes.

In 2019, a group of researchers found a set of genes in the testis of mice that are expressed at the same time as the gene for PRDM9. This suggested that another protein called ZCWPW1 is likely involved in recombination, but the precise role of this protein was unclear. To answer this question, Wells, Bitoun et al. – including many of the researchers involved in the 2019 study – examined human cells grown in the laboratory to determine where ZCWPW1 binds to in the chromosome.

This revealed that ZCWPW1 can be found at the same sites as PRDM9, which is responsible for bringing it there. Furthermore, cells from male mice lacking the gene for ZCWPW1 cannot complete the exchange of genetic information between chromosomes, meaning that the mice are infertile. As such, ZCWPW1 seems to connect location selection by PRDM9 to the DNA repair mechanisms needed for gene exchange between chromosomes.

Infertility is a significant issue for humans affecting as many as one in every six couples. Fertility is complex and many of the biological mechanisms involved are not fully understood. This work suggests that both PRDM9 and ZCWPW1 are key to the production of sex cells and may be worth investigating as factors that affect fertility in humans.

Introduction

Meiosis is a specialised cell division, producing haploid gametes essential for reproduction. Uniquely, during this process homologous maternal and paternal chromosomes pair and exchange DNA (recombine) before undergoing balanced independent segregation. Alongside de novo mutations, this generates all genetic diversity – providing a substrate on which natural selection can act.

In humans, mice, and likely many other vertebrates (Baker et al., 2017), the locations at which recombination occur are determined by the binding locations of PRDM9, which recognises a specific sequence motif encoded by its zinc finger array (Baudat et al., 2010; Myers et al., 2010; Parvanov et al., 2010). At these sites, SPO11 catalyses recombination-initiating double strand breaks (DSBs), a subset of which are repaired by recombination with the homologous chromosome. The sister chromatid can also provide a repair substrate, for example for DSBs on the X chromosome in males, and there is evidence that this also occurs at some autosomal hotspots (Allen and Latt, 1976Li et al., 2019b; Lu and Yu, 2015).

In Prdm9-null mice, DSBs form at promoter regions rich in H3K4me3 (Brick et al., 2012), and fail to repair resulting in severe asynapsis, meiotic arrest at pachytene, and sterility in both males and females on the B6 background (Hayashi et al., 2005). We previously showed (Davies et al., 2016) that across DSB sites, increased binding of PRDM9 to the homologous chromosome aids synapsis and fertility in hybrid mice, implying an impact of binding downstream of DSB positioning. Mice for which most DSBs occur at sites where PRDM9 does not bind the homologue show widespread asynapsis (Davies et al., 2016; Gregorova et al., 2018).

Following their formation, DSB sites are processed by resection, resulting in single-stranded DNA (ssDNA) that becomes decorated with DMC1 (Hong et al., 2001; Neale and Keeney, 2006; Sehorn et al., 2004). In wild-type (WT) mice, DMC1 foci start to appear in early zygotene cells. From mid-zygotene to early pachytene, as part of the recombinational repair process, DMC1 dissociates from the ssDNA and counts decrease until all breaks (except those on the XY chromosomes) are repaired at late pachytene (Moens et al., 2002). At DSB sites where the homologous chromosome is not bound by PRDM9, the DMC1 signal is strongly elevated (Davies et al., 2016), suggesting persistent DMC1 foci and consistent with delayed DSB repair at these sites. In addition, fewer homologous recombination events occur at these sites (Hinch et al., 2019; Li et al., 2019b) implying that (eventual) DSB repair may sometimes use a sister chromosome pathway. However, the underlying mechanism(s) by which PRDM9 effectively contributes to DSB repair are not yet known.

PRDM9 deposits both H3K4me3 and H3K36me3 histone methylation marks at the sites it binds, and this methyltransferase activity is essential for its role in DSB positioning (Diagouraga et al., 2018; Powers et al., 2016). What factors ‘read’ this unique combination of marks at recombination sites, however, is currently unknown. Notably, outside of hotspots and the pseudoautosomal region (PAR) on sex chromosomes, H3K4me3 and H3K36me3 occur at largely non-overlapping locations (Powers et al., 2016), suggesting potentially highly specialised reader(s). Indeed, in somatic cells H3K4me3 is deposited mainly at promoters, in particular by the SET1 complex targeted by CXXC1/CFP1 and Wdr82 which binds Ser5 phosphorylated polymerase II (Barski et al., 2007; Lee and Skalnik, 2008; Lee and Skalnik, 2005). H3K36me3 is deposited by different methyltransferases, including SETD2 bound to Ser2 phosphorylated (elongating) polymerase II, and is enriched at exon-bound nucleosomes, particularly for 3’ exons (reviewed in McDaniel and Strahl, 2017; Wagner and Carpenter, 2012). H3K36me3 has multiple important roles, including in directing DNA methylation by recruiting DNMT3B, somatic DSB repair by homologous recombination (Aymard et al., 2014; Carvalho et al., 2014; Pfister et al., 2014), mismatch repair by recruiting MSH6 (Huang et al., 2018; Li et al., 2013), and V(D)J recombination during lymphopoiesis (Ji et al., 2019).

Using single-cell RNA-sequencing (RNA-seq) of mouse testis, we identified a set of genes co-expressed in (pre)leptotene cells which are highly enriched for genes involved in meiotic recombination (Jung et al., 2019). Zcwpw1, which ranks 3rd in this set after Prdm9 (2nd), is of unknown function but contains two recognised protein domains: CW and PWWP, shown to individually bind H3K4me3 and H3K36me3 respectively (He et al., 2010a; Rona et al., 2016). This raises the attractive possibility that ZCWPW1 might recognise and physically associate with the same marks deposited by PRDM9 (Jung et al., 2019).

In humans, ZCWPW1 is expressed in testis (Carithers et al., 2015; Uhlén et al., 2015Figure 1—figure supplement 1). It is also one of 104 genes specifically expressed in meiotic prophase within murine fetal ovaries (Soh et al., 2015), further suggesting a conserved meiotic function. This was confirmed by a recent study showing that ZCWPW1 is required for male fertility in mice, with ZCWPW1 hypothesised to recruit the DSB machinery to hotspot sites (Li et al., 2019a). Here, we show that ZCWPW1 co-evolves with PRDM9 and is recruited to recombination hotspots by the combination of histone marks deposited by PRDM9. However ZCWPW1 is not in fact required for the positioning of DSBs at PRDM9-bound sites, which occurs normally in Zcwpw1-null mice. Instead, ZCWPW1 is required for proper inter-homologue interactions: synapsis and the repair of DSBs. In Zcwpw1-null mice, DMC1 dynamics are strongly perturbed, with signals at autosomal hotspots resembling those on the X-chromosome, which does not have a homologue. Thus, ZCWPW1 represents the first protein directly positioned by PRDM9’s dual histone marks, but impacting homologous DSB repair, not positioning.

Results

ZCWPW1 co-evolves with PRDM9

Based on publicly available databases, we identified likely ZCWPW1 orthologues in 167 species (Materials and methods, Figure 1—source data 1), aligned each to the human reference ZCWPW1 protein sequence, and compared against a previous analysis of PRDM9 (Baker et al., 2017). The regions containing the CW and PWWP domains are well-conserved (Figure 1A,B), while other parts of the protein appear absent in some species. Additionally, there is a region of moderate conservation downstream of the PWWP domain, not overlapping any known domain. Further, an SYCP1 (SCP1) domain is annotated in the mouse protein only which, although only suggestive, is notable given that SYCP1 physically connects homologous chromosomes in meiosis. The C-terminal end of ZCWPW1 may include a methyl-CpG-binding domain (Materials and methods, Lobley et al., 2009), at least in some species.

Figure 1 with 1 supplement see all
Domain organisation (A) and evolutionary conservation (B) of ZCWPW1, and co-evolution with PRDM9 (C,D).

(A) Protein domains in the human and mouse proteins (source: UniProt). Amino acid start and end positions of each domain are shown above and below the rectangles, respectively. Prediction of SCP-1 (SYCP1) domain from Marchler-Bauer and Bryant, 2004 and of MBDs (methyl-CpG binding domain) from Lobley et al., 2009 (Materials and methods). (B) Conservation of human amino acids, normalised Jensen-Shanon divergence normalised to mean of 0 and standard deviation of 1 is shown on the y-axis (a measure of sequence conservation, see Capra and Singh, 2007 and Johansson and Toh, 2010) computed from using multiple alignment of 167 orthologues (Materials and methods). (C) All species we identified as possessing ZCWPW1 copies were phylogenetically grouped into clades as previously (Baker et al., 2017) (x-axis) and each clade divided (stacked bars) according to whether ZCWPW1-possessing species within it also possess PRDM9 (‘Species’, red) or instead their closest PRDM9-possessing relative is respectively in the same genus/family, order, clade or order/phylum, with colours as given in the ‘Closest PRDM9’ legend. (D) As (C), but now showing the closest relative possessing ZCWPW1 (‘Closest ZCWPW1’ legend) for species possessing complete, partial or no identified PRDM9 copies. As in (C), the x-axis groups species into clades, now further divided based on data from Baker et al., 2017 into subclades according to the domains of PRDM9 lost or mutated across that subclade in all observed copies, reflecting multiple partial losses of particular PRDM9 domains, or complete loss of all PRDM9 copies (Main text). The x-axis labels are ordered and coloured according to the PRDM9 domains present (‘PRDM9 domains’ legend, where ‘SET’ refers to PRDM9’s PR/SET domain and the KRAB and SSXRD domains are grouped as ‘non-SET’, and ‘partial’ losses are seen in some but not all PRDM9 copies in that species). Further details are presented in Materials and methods, and the raw data in Figure 1—source datas 1 and 2.

Incomplete currently available DNA and protein sequences mean we are almost certain to miss some species where either ZCWPW1 and/or PRDM9 (Baker et al., 2017) are present. Despite this, we see extremely high overlap between PRDM9 and ZCWPW1 occurrence (Figure 1C,D). ZCWPW1 appears to mainly occur in species possessing PRDM9 (and even more clearly in groups of species possessing PRDM9). Furthermore, with the exception of a small number of amphibians, ZCWPW1 orthologues are only identified in clades previously identified to possess PRDM9 (Baker et al., 2017), and the majority of species (78% of 167) possessing ZCWPW1 also possess a close relative (species or family) where PRDM9 is identified (Figure 1C).

We leveraged a previous analysis of PRDM9 evolution (Baker et al., 2017) to extend this comparison (Figure 1D, Figure 1—source datas 1 and 2). Among seven clades previously shown to possess PRDM9 copies with all N-terminal domains (KRAB, SSXRD and PR/SET domains) intact, ZCWPW1 is found in each case, typically in the same species (Figure 1D). In some species with several PRDM9 copies, including many teleost fish and two other confirmed independent events, PRDM9 has lost N-terminal domain sequences in some but not all copies, while in at least three further species clades, particular N-terminal domains have been lost in all PRDM9 copies, but the PR/SET domain remains intact. ZCWPW1 is also found across these cases (Figure 1D), so it is widely conserved across species possessing one or more copies of PRDM9 with an intact PR/SET domain.

Previously (Baker et al., 2017), three independent groups of species have been identified possessing mutations in the PR/SET domain of PRDM9, at key catalytic residues for the deposition of the H3K4me3/H3K36me3 mark(s). Complete loss of PRDM9 is also identified in three groups: bird/crocodiles, canids, and (tentatively) amphibians. Given that ZCWPW1 is predicted to recognise the H3K4me3/H3K36me3 marks catalysed by the PR/SET domain of PRDM9, we would expect each of these six events to disrupt any potential interaction between the two proteins. Strikingly, we failed to identify ZCWPW1 orthologues in any of the previously examined species impacted by these six events (Figure 1D), except in canids (and one amphibian species). Thus, at least four of the six may be completely mirrored by ZCWPW1 loss. Even if PRDM9 is retained, species with PR/SET domain mutations are much more likely not to possess an identified ZCWPW1 orthologue within their class/phylum compared to species without such mutations (odds ratio = 285; indicative p<10−15 by FET, although we note that these observations are not all independent).

In conclusion, among vertebrates at least, it appears the set of species with ZCWPW1 is extremely similar to those species possessing PRDM9 with an intact PR/SET domain, with considerable evidence of co-evolution of gain/loss events for each protein. In some species with apparently intact PR/SET domains (for example Tasmanian Devil, Mexican Tetra, Platypus), it is hypothesised that PRDM9 does not position DSBs due to loss of other N-terminal domains and/or the PRDM9 zinc-finger array (Baker et al., 2017). These example species still possess identified ZCWPW1 orthologues (Figure 1—source data 1), so it does not seem that PRDM9’s ability to position DSBs is essential for retention of ZCWPW1. Instead, our results are consistent with ZCWPW1’s main function involving recognition of the histone modifications catalysed by PRDM9-SET during early meiosis, in agreement with the functional evidence from ZCWPW1’s CW and PWWP domains.

Localisation of ZCWPW1 in meiosis and analysis of asynapsis in infertile Zcwpw1−/− male mice

We previously detected Zcwpw1 transcript expression in pre-leptotene to leptotene spermatocytes of adult WT mice (Jung et al., 2019). To investigate the role of ZCWPW1 during meiosis in vivo, we produced and validated the specificity of an antibody against the full-length recombinant mouse protein (Figure 2—figure supplement 1), and studied the phenotype of a newly generated knockout (KO) mouse line for Zcwpw1, with a particular focus on fertility and meiotic recombination.

In testes from WT mice, we observe a dynamic localisation of ZCWPW1 protein (Figure 2), similar but non-identical to that reported in a recent study (Li et al., 2019a). ZCWPW1 shows a strong, punctate nuclear staining excluding the pericentromeric regions (clustered into chromocenters brightly stained with DAPI) in zygotene and early pachytene cells. In mid-pachytene cells, ZCWPW1 levels drop, with the protein now mainly localised in the XY body and as bright foci at the ends of the synaptonemal complex labeled by SYCP3, not previously observed using an antibody raised against a 174 amino acid C-terminal region of the protein (Li et al., 2019a). By diplotene, little expression is visible. Using fluorescent in-situ hybridisation (FISH) to label centromeric and telomeric (distal and proximal to centromeres) regions of chromosomes, we established that these discrete foci of ZCWPW1 are consistently positioned at the ends of the synaptonemal complex, with the majority at both ends simultaneously, lying close to telomeres and centromeres themselves (Figure 2—figure supplement 2).

Figure 2 with 2 supplements see all
Expression of ZCWPW1 across meiosis prophase I in mouse testis.

Nuclear spreads from 9 to 10 weeks old WT mice were immunostained with antibodies against ZCWPW1 (red) and the synaptonemal complex protein SYCP3 (green) which labels the chromosome axis, and counterstained with DAPI (blue) to visualise nuclei. Developmental stages are indicated above and below. Yellow arrows point to ZCWPW1 foci clearly visible at both ends of the synaptonemal complex in mid-late pachytene. Additional evidence is provided in Figure 2—figure supplement 2. The dashed circle shows staining in the XY body. Images for the individual channels are provided in Figure 3—figure supplement 1. These images are representative of the results obtained in three mice. Scale bar: 10 μm.

We next studied mice from a constitutive Zcwpw1 KO line (Materials and methods), carrying a ~ 1.5 kb frameshift deletion encompassing exons 5 to 7 upstream of the CW and PWWP domains. This creates a premature stop codon that would produce a predicted heavily truncated 492bp (vs 1893bp for WT) transcript lacking either of these domains (Figure 3A), and a candidate for nonsense-mediated mRNA decay (Kurosaki et al., 2019). Indeed, we observed a complete lack of detectable ZCWPW1 protein expression in testis protein extracts (Figure 3—figure supplement 1A) and chromosome spreads from Zcwpw1−/− mice, in zygotene cells (Figure 3B) and all other meiotic stages of prophase I where expression is detected in WT mice (Figure 3—figure supplement 1B). Confirming recent findings in a different Zcwpw1−/− mouse (Li et al., 2019a), we observed no overt fertility phenotype in either sex in the heterozygous Zcwpw1+/− mice (data not shown). However, Zcwpw1−/− male mice were sterile with complete azoospermia and reduced testis size and weight (Figure 3C,D, Figure 3—source data 1), while female mice retained fertility until around 7–8 months of age (Figure 3—source data 2), and otherwise both sexes develop normally. As in Li et al., 2019a, in male mice we observe no meiotic progression beyond (pseudo)pachytene with widespread asynapsed chromosomes marked by γ-H2AX and HORMAD2, failure to form the sex body (Figure 3B, Figure 3—figure supplement 2A), a complete absence of MLH1 foci marking recombination crossover sites (Figure 3—figure supplement 2B), and persistent DMC1 foci marking unrepaired DSBs (Figure 4).

Figure 3 with 2 supplements see all
Zcwpw1−/ male mice show reduced testis size and asynapsis, similar to the Prdm9−/− mutant.

(A) Schematic of the Zcwpw1 knockout (KO) mouse line. E: Exon. gRNA: guideRNA. Sanger sequencing DNA chromatograms of wild-type (WT) and KO mice encompassing the deletion are shown. The intron-exon organisation is not to scale. (B) Immunofluorescence staining of testis nuclear spreads from 9- to 10-week-old Zcwpw1+/+ and Zcwpw1−/− mice for ZCWPW1 or HORMAD2 (red) which marks asynapsed chromosomes, and the synaptonemal complex protein SYCP3 (green) which labels the chromosome axis. Cells were counterstained with DAPI (blue) to visualise nuclei (top images). These images are representative of the data obtained for three mice per genotype. Scale bar: 10 μm. (C) Representative testes from 9- to 10-week-old WT (+/+), Het (+/−) and Hom (−/−) Zcwpw1 KO mice are shown. (D) Paired testes weight was normalised to lean body weight. Each datapoint represents one mouse. The p-value is from Welch’s two sided, two sample t-test. Raw data in Figure 3—source data 1. (E) Synapsis quantification in testis chromosome spreads immunostained with HORMAD2, as in (B). The percentage of mid-Pachytene (WT) or pseudo-Pachytene (Zcwpw1−/− and Prdm9−/−) cells with all autosomes fully synapsed is plotted by genotype; each datapoint represents one mouse, each with n≥ 49 cells analysed. Vertical lines are 95% Wilson binomial confidence intervals. Raw data in Figure 3—figure supplement 2—source data 1.

Figure 4 with 3 supplements see all
Similar DMC1 count elevation in Zcwpw1−/− and Prdm9−/− mice, compared to wild-type.

(A) Testis chromosome spreads from 9- to 10-week-old Zcwpw1+/+ and Zcwpw1−/− mice were immunostained for DMC1 and SYCP3. Late (pseudo)-Pachytene cells are shown. These images are representative of the data obtained for three mice per genotype. Scale bar: 10 μm. (B) The number of DMC1 foci in cells from the indicated stages of prophase I were counted; see Figure 4—source data 1 for number of cells per stage per mouse. p-values are from Welch’s two sided, two sample t-test. L: Leptotene, Z: Zygotene, P: Pachytene. n = 3 mice for Zcwpw1−/− and wild-type, n = 2 for Prdm9−/−. Raw data in Figure 4—source data 1.

Each of these properties resembles observations in the Prdm9−/− mutant (Hayashi et al., 2005), and so we compared to this mutant. In our Zcwpw1−/− male mice, >98% of pachytene cells failed to properly synapse at least one pair of chromosomes (Figure 3B,E), similar to Prdm9−/− males. However, the nature of the synaptic defects observed differed (Figure 3—figure supplement 2—source data 1). In the Prdm9−/− mutant, 63.2% of the pseudo-pachytene cells showed mispairing of non-homologous chromosomes in a typical branched structure (referred to as ‘tangled’, Figure 3—figure supplement 2—source data 1). In contrast, we only observed 24.8% of Zcwpw1−/− pseudo-pachytene cells with this type of error, while the majority (74.5%) of cells contained multiple bundles of HORMAD2-positive asynapsed chromosomes, which resemble the XY body and may merge with the sex chromosomes (thus referred to as ‘multibodies’, Figure 3—figure supplement 2). These results imply that the Prdm9−/− mutant often mispairs chromosomes, while Zcwpw1−/− spermatocytes mainly fail to pair a subset of chromosomes at all. The expression levels and staining pattern of ZCWPW1 were not visibly altered in testis nuclear spreads (data not shown) and protein extracts (Figure 3—figure supplement 1A) from Prdm9−/− mice. However, since the Zcwpw1 KO and Prdm9 KO lines are on a different genetic background (C57/BL6N and B6×129, respectively), we cannot rule out the possibility that this explains partly or fully the differences in meiotic defects observed between the two mutants.

Comparing levels of DMC1 foci as a proxy for DSB repair, or a potential cause for the asynapsis in Zcwpw1−/− males, in both Zcwpw1−/− and Prdm9−/− mice foci count was significantly elevated from zygotene onwards, indicating delayed repair of DSBs (Figure 4). However, we observed a wider spread for Zcwpw1−/− males. A similar increase was observed at pachytene stage in the levels of RAD51 (Figure 4—figure supplement 1), a strand exchange protein which functions in concert with DMC1, with the large majority of RAD51 and DMC1 foci co-localising (Brown et al., 2015; Tarsounas et al., 1999). Like DMC1, RPA2 levels were also significantly elevated in the Zcwpw1−/− mouse from zygotene onwards (Figure 4—figure supplement 2). In Zcwpw1−/− males, like in the Prdm9−/− mutant, DSBs form and recruit RPA2, RAD51 and DMC1 in similar numbers (Huang et al., 2020; Li et al., 2019a; Mahgoub et al., 2020), but fail to repair efficiently, accompanied by asynapsis and meiotic arrest at pseudo-pachytene. Indeed, we observe late unrepaired DMC1 foci mainly on HORMAD2-positive asynapsed chromosomes (Figure 4—figure supplement 3).

ZCWPW1 is recruited to PRDM9-binding sites in an allele-specific manner

We previously studied the binding properties of human PRDM9 and established a genome-wide map in transfected human mitotic (HEK293T) cells by ChIP-seq (Altemose et al., 2017), observing binding to the majority of human meiotic recombination hotspots. Based on the presence of H3K4me3 and H3K36me3 recognition domains in ZCWPW1, we hypothesised that it would be recruited to PRDM9-bound genomic sites, where these marks are deposited upon binding in HEK293T cells (Altemose et al., 2017).

To test this, we co-transfected HEK293T cells with full-length human HA-tagged ZCWPW1 and either no other protein, or full-length PRDM9 alleles carrying the human or chimpanzee zinc finger array, as studied previously (Altemose et al., 2017), and then performed ChIP-seq against the ZCWPW1 tag (the endogenous levels of these proteins in HEK293T cells, and of the closely related family member ZCWPW2, are extremely low so unlikely to confound the analysis - see Figure 5—source data 1). Confirming the recruitment hypothesis, in the presence of human PRDM9 and compared to cells lacking human PRDM9, ZCWPW1 shows a strong enrichment at human PRDM9-binding sites with higher enrichment at sites with higher PRDM9 enrichment (Figure 5A,B, Pearson’s correlation = 0.43 - Figure 5—figure supplement 1). Notably, even without PRDM9 we observed many ZCWPW1 binding peaks across the genome, some coinciding with PRDM9-binding sites (Figure 5C and Figure 5—figure supplement 2). However, upon co-transfection with PRDM9, 92% of the strongest 10,000 ZCWPW1 peaks are at PRDM9-bound sites (98% of the strongest 3,000 vs 7% expected overlap with randomised peaks, and only 47% overlap for the strongest 3,000 ZCWPW1 peaks in cells without PRDM9 transfection). Hence, PRDM9 is able to strongly reprogram ZCWPW1 binding (Figure 5C, see also Figure 5—figure supplement 3), suggesting that loci with peaks in cells without PRDM9 transfection are bound more weakly. Because our transfection has <100% efficiency (Figure 5—figure supplement 4), some cells containing ZCWPW1 will not possess PRDM9, and so it is possible that an even greater fraction of ZCWPW1 is redirected to PRDM9 binding sites in cells where both proteins are present.

Figure 5 with 6 supplements see all
Enrichment and binding profiles of ZCWPW1 and other factors.

(A) Enrichment of ZCWPW1 (with vs without PRDM9) at PRDM9-binding sites when co-transfected with PRDM9 with either Human or Chimp Zinc Finger (Materials and methods section ‘Enrichment Profiles’). Q = quartile. Human PRDM9 sites are centered and stranded by the motif. Y-axis is log10 scale (y-axis labels remaining in linear space). (B) Profiles and heatmaps of reads from cells co-transfected with human (h) or chimp (c) PRDM9 around the top 25% of individual human PRDM9-binding sites (rows). Heatmaps: log-fold change of target (indicated in column titles, Materials and methods) vs input, for various labelled target proteins, ordered by human PRDM9. ZCWPW1, H3K4me3 and H3K36me3 each become enriched at human PRDM9 sites, following (co-)transfection with human PRDM9. Profiles: Sum of all target coverage divided by sum of all input coverage for all regions shown in the heatmap, shown on a linear scale. w\o, without. (C) ChIP-seq data and annotation in a genome plot illustrate the behaviour of ZCWPW1 and other factors. ChIP-seq tracks show fragment coverage. Tracks where PRDM9 is present are labeled ‘w/PRDM9’, and below, corresponding tracks without PRDM9. ZCWPW1 binds to Alus, CpG islands and other CpG-rich sequences even in the absence of PRDM9. On addition of PRDM9, ZCWPW1 becomes strongly enriched at PRDM9 binding locations (center left peak within DIO1). mCpG, methylated CpG.

Importantly, when co-transfecting with a modified version of PRDM9 in which the zinc finger array is replaced with that from chimp (which binds different locations in the genome [Altemose et al., 2017]), we find that the enrichment at human binding sites disappears, but instead ZCWPW1 is enriched at chimp PRDM9-binding sites (Figure 5A and Figure 5—figure supplement 5). This perturbation experiment provides strong evidence that PRDM9 causes recruitment of ZCWPW1 (as opposed to for example independent recruitment of both proteins). Notably, the strength of ZCWPW1 binding in HEK293T cells at human PRDM9-binding sites provides a better predictor of human meiotic DMC1 binding status (a proxy for DSB formation) than does human PRDM9 binding strength itself (Figure 5—figure supplement 6). We show (see below) that ZCWPW1 is not directly involved in DSB positioning, this result might therefore instead suggest that similar features to those ZCWPW1 recognises are involved in recruiting DSB formation machinery.

We tested whether the recruitment of ZCWPW1 to sites bound by PRDM9 might be mediated by the dual histone modifications H3K4me3 and H3K36me3. Consistent with this idea, ZCWPW1 binding is positively associated with levels of both H3K4me3 and H3K36me3 marks (Figure 6—figure supplements 1 and 2, tested separately, not necessarily ‘dual’ marks coincident within an individual cell, although locations of these marks in meiotic cells are rarely coincident except at hotspots [Powers et al., 2016]). While we cannot rule out a direct interaction between ZCWPW1 and PRDM9, others have shown that dual modified peptides have a greater affinity to ZCWPW1 than peptides carrying the single modifications, supporting that the marks themselves are responsible for ZCWPW1’s recruitment (Mahgoub et al., 2020). We examined transcription start sites, which possess H3K4me3 at high levels, but lack H3K36me3, observing some ZCWPW1 signal at these sites, but with a uniformly lower mean ZCWPW1 enrichment compared to those with evidence of PRDM9 binding (and hence both histone marks) (Figure 6). Although it is not possible to measure H3K4me3 presence/absence in individual cells in our system, even the most weakly bound PRDM9 sites – which must therefore possess H3K4me3 in only a fraction of cells – show stronger ZCWPW1 enrichment than the strongest promoters, which are likely to have near 100% H3K4me3 marking, and possess >2 fold more H3K4me3 than even the strongest PRDM9-binding sites. We conclude that H3K4me3 alone endows only relatively weak binding, while PRDM9 is therefore able to recruit ZCWPW1 with a much greater efficiency than sites marked by H3K4me3 alone, suggesting that both histone modifications might aid efficient binding. While the increase in ZCWPW1 enrichment at sites with pre-existing H3K4me3 is consistent with both marks contributing to recruitment, we have no evidence that these weaker binding sites serve a functional role, and indeed concurrent investigations did not find strong evidence of ZCWPW1-driven transcriptional changes (Huang et al., 2020).

Figure 6 with 2 supplements see all
PRDM9-bound regions (H3K4me3 and H3K36me3) are a stronger recruiter of ZCWPW1 than promoters (H3K4me3 only).

For any given level of H3K4me3 (x-axis), ZCWPW1 enrichment (y-axis) is higher at PRDM9-bound regions (red) than regions with pre-existing H3K4me3 (promoters, blue). H3K4me3 and ZCWPW1 were force called in 100bp windows across all autosomes. These windows were split into two sets defined as indicated in the legend (where ‘p’ is the p-value from peak-calling required for a window to be included in the subset) with the additional constraint of requiring input fragment coverage >5 for ZCWPW1 and >15 for H3K4me3. p: p-value for non-zero level of input corrected coverage in that bin. ‘pre-existing H3K4me3’ refers to H3K4me3 that is present without transfection (of either PRDM9 or ZCWPW1), which is mainly found at promoter regions. For each subset, H3K4me3 was split into 25 bins with equal number of data points. Horizontal bars: two standard errors of the mean. Vertical dotted bars: upper and lower quartiles. Grey ribbons show two standard errors for a Generalized additive model on log(mean H3K4me3 enrichment + 0.1). Dashed black horizontal line highlights that the mean enrichment of the highest bin for promoters is similar to that of the lowest bin for PRDM9-bound sites.

DSBs occur at their normal locations in Zcwpw1−/− mice but show increased DMC1 levels

Previous work has shown that Prdm9−/− mice use a new set of DSB hotspots, localising at CpG islands and/or promoter regions (Brick et al., 2012). Given that PRDM9 recruits ZCWPW1, one possible function of ZCWPW1 may be that, in turn, it recruits the DSB machinery and hence forms part of the causal chain in normal positioning of DSBs at PRDM9-specified hotspots. Alternatively, the Zcwpw1−/− mutant phenotypes we observe might reflect a more downstream role. To distinguish these hypotheses, we gathered data on DSB positioning and repair dynamics genome-wide, by carrying out single ssDNA sequencing (SSDS) by ChIP-seq against DMC1 (Khil et al., 2012).

In the Zcwpw1−/− mutant males, we observed normal localisation of DMC1 at the expected B6 WT hotspots for this background, with no signal at Prdm9−/− hotspots (Figure 7A). We only fail to see evidence of DMC1 signal in a subset of the weakest hotspots, where we are likely to lack statistical power (Figure 7—figure supplement 1). Thus, DSBs occur in unchanged hotspot regions in Zcwpw1−/− males, relative to WT. To check if they occur at the same locations within hotspot regions, we leveraged data for SPO11-mapped DSB sites (Lange et al., 2016). Specifically, we identified three sets of hotspots whose respective mapped breaks occur mainly upstream, central, or downstream of the PRDM9-binding site (Materials and methods), and compared their DMC1 ChIP-seq signal profiles, as well as their SPO11 signals (Figure 7—figure supplement 2). This revealed that in the Zcwpw1−/− males, as in the WT, DMC1 signals mirror this break positioning. Therefore, ZCWPW1 is not required to specify hotspot locations, and neither does it strongly influence where DSBs occur within hotspots.

Figure 7 with 4 supplements see all
DMC1 levels in the Zcwpw1−/− mouse compared to DMC1 and SPO11 levels in WT.

(A) DSBs occur at normal hotspot locations in the Zcwpw1−/− male mouse. Average coverage of reads from DMC1 SSDS ChIP-seq in a 10-week-old mouse at previously mapped regions (Materials and methods) in B6 WT (left) and Prdm9−/− (right) mice is shown, centered at the PRDM9 motif (left). DMC1 profiles from a WT mouse are shown in red, data from Brick et al., 2012. (B) Normalised DMC1 profile (both strands combined) is plotted for WT and Zcwpw1−/−, stratified by H3K4me3 (a proxy for PRDM9 binding). Low: <50th percentile cumulative enrichment, High: >75th percentile cumulative enrichment, with Medium being the remaining data. Greyed out lines show the alternative genotype for comparison. (C) Relationship between WT SPO11-oligos (measuring the number of DSBs) vs DMC1 (a measure of the number and persistence of DSBs) at each B6 hotspot for WT and Zcwpw1−/−. Unlike WT mice, DMC1 signals in Zcwpw1−/− mice are approximately linearly associated with WT SPO11. The DMC1 enrichment was force called at the positions of B6 WT hotspots. Black dashed line is y = x for reference. SPO11 and DMC1 enrichment have been scaled by dividing by the mean autosomal enrichment. Large dark blue and dark red points show mean DMC1 signal, binned into groups containing equal numbers of hotspots by WT SPO11 signal (vertical lines: corresponding 95% CIs), for X (10 bins) and autosomal data (100 bins) respectively (smaller lighter dots represent individual hotspots).

One perhaps important difference between the Zcwpw1−/− and WT DMC1 profiles is that the latter signal is slightly wider (up to ~200bp) in hotspots with increasing PRDM9-induced H3K4me3 (Figure 7B), while this effect is absent in the Zcwpw1−/− case. This might be explained by small differences in chromatin accessibility subtly impacting SPO11 locations, DNA end resection which generates the 3’ ssDNA tail to which DMC1 binds, or downstream processing differences altering DMC1 span in the mutant mice.

Despite very similar DSB locations, we saw much greater, and systematic, differences in the strength of the DMC1 signal at individual hotspots. As previously (Davies et al., 2016; Khil et al., 2012), we note that observed average DMC1 signal strength at a hotspot reflects the product of the frequency at which DSBs occur there, and the average length of time DMC1 remains bound to the ssDNA repair intermediate, with the latter reflecting DSB processing/repair time. In contrast, available SPO11 oligo-seq data (Lange et al., 2016) reflect mainly the frequency of DSBs. We therefore compared DMC1 and SPO11 signal strength at each autosomal and X-chromosome hotspot, in WT and Zcwpw1−/− male mice. In WT mice, the non-PAR X-chromosome shows a very strong elevation of DMC1 signal strength (as seen previously in other mice [Davies et al., 2016] including, albeit somewhat more weakly, even sterile hybrids). This elevation reflects the persistent DMC1 foci on this chromosome also visible using microscopy, at DSB sites that eventually repair using the sister chromatid. Moreover, in WT mice there is a sub-linear (curved) relationship between SPO11 and DMC1 at hotspots (Figure 7C). This is thought to reflect a wider phenomenon of faster DSB repair occurring within those hotspots whose homologue is more strongly bound by PRDM9, that is those hotspots with a stronger H3K4me3 signal in the WT mouse (Davies et al., 2016; Hinch et al., 2019Li et al., 2019b). The X-chromosome DMC1 elevation is in a sense an extreme case of slower repair of DSBs whose homologue is not PRDM9-bound, because no homologue exists in this case.

However, we see a striking departure in the Zcwpw1−/− mouse, where the X-chromosome behaves more similarly to the autosomes with respect to DMC1 vs WT SPO11 signal strength (Figure 7C). Moreover, we see a simple linear relationship between DMC1 and WT SPO11 binding in this mouse (Figure 7C). This implies that the effect of PRDM9 binding on DMC1 removal is eliminated in this mouse (assuming SPO11 levels are not greatly affected by Zcwpw1 KO); so PRDM9 appears unable to ‘assist’ in homologue pairing and synapsis in this mouse (whether because ZCWPW1 directly aids the process, or indirectly). Indeed, the data imply a widespread perturbation of DSB repair in this mouse, with autosomal DMC1 foci persisting as long as those on the X-chromosome. In the previously studied Hop2−/− mouse (Khil et al., 2012; Petukhova et al., 2003; Smagulova et al., 2011), we identified a very similar pattern (Figure 7—figure supplement 3): in this mouse, DMC1 is loaded onto ssDNA, but DSBs are not repaired at all, indicating that DSB repair pathways in Zcwpw1−/− mice are profoundly altered.

These results are consistent with, and extend, our microscopy observations (Figure 4) that many DSBs persist in the Zcwpw1−/− mouse, and some may never repair. We are unable to say whether DSB repair involves the homologue or not in this mouse, but the partial synapsis we observe suggests some repair does likely occur. To further understand DMC1 persistence at individual hotspots, we estimated the relative DMC1 heat of autosomal hotspots in the Zcwpw1−/− mouse, compared to the WT mouse. The best individual predictor of this ratio among H3K4me3, DMC1, and SPO11 in WT was the level of H3K4me3 (r = 0.60 Generalised Additive Model; Figure 7—figure supplement 4, while r = 0.63 when using all three predictors together). The average KO:WT DMC1 ratio increases around ninefold from the most weakly to the most strongly H3K4me3-marked hotspots. Therefore, if DMC1 signal changes are indeed explained by slower DSB processing in the Zcwpw1−/− mouse at hotspots bound strongly by PRDM9, then this implies a very strong effect of PRDM9’s ‘assistance’ in aiding repair. Whatever the cause, the overall strong correlation implies perturbation of DMC1 behaviour is widespread, rather than impacting any small subset of hotspots. Moreover, it appears to be mainly controlled by local levels of PRDM9 histone modification, in keeping with the evidence that ZCWPW1 recognises this mark at PRDM9-bound sites, on either the broken or (identical) homologous chromosome.

ZCWPW1 binds CpG dinucleotides

In addition to the strong PRDM9-dependent ZCWPW1 peaks described earlier, there are many locations in HEK293T cells at which ZCWPW1 binds, typically more weakly, and independently of PRDM9. Indeed we identified over 800,000 ZCWPW1 peaks. Surprisingly, a large proportion of these binding sites overlap Alu repeats (Figure 8A, Figure 5C, and Figure 5—figure supplement 2) (of which there are 1.1 million in the human genome [Deininger, 2011]). The weakest ZCWPW1 peaks overlap Alus most frequently, whilst the strongest peaks are depleted of Alus relative to chance overlap (Figure 8A).

Figure 8 with 1 supplement see all
ZCWPW1 binds CpG-rich sequences such as Alu repeats.

(A) Fraction of overlap of ZCWPW1 binding peaks, , with Alusin HEK293T cells transfected with ZCWPW1 alone, ordered by enrichment in ZCWPW1 binding. ZCWPW1 peaks are binned into 25 bins with equal number of data points, and means of both enrichment and overlap are plotted. Solid ribbons: prediction from GAM logistic regression. Dotted lines: overall means. Red points show actual observed peaks, blue points the same number of peaks placed at random genomic positions. (B) Rate of overlap of Alu repeats with ZCWPW1 peaks, for Alus with different numbers of CpG dinucleotides. Other details as A. (C) The probability of a 300bp window on an autosome overlapping a ZCWPW1 peak increases with increasing CpG count in that window. Windows overlapping (by 10bp or more) Alus, other repeats, or CpG islands have been excluded. Methylated CpG regions (full colour) are those with a methylated to unmethylated reads ratio of >0.75, and unmethylated <0.25 (semi-transparent, Materials and methods). (D) Relative proportion of peaks with given numbers of CpGs (stacked bars) +/− 150bp from peak center, within peaks binned by ZCWPW1 enrichment (x-axis). ZCWPW1 peaks are enriched in CpGs compared to random peak locations (leftmost bar). ‘Meth’, methylated; ‘UnMeth’, unmethylated.

Because Alus are rich in CpGs (containing >23% of human CpG dinucleotides, with deamination depleting CpGs in older Alus [Luo et al., 2014]), we tested if binding correlates with CpG presence/absence. Indeed, the binding of ZCWPW1 to Alus depends on the presence of CpGs, which are mainly methylated (Xie et al., 2011). Alus with no CpGs were bound at a rate lower than by chance, but almost all Alus containing 10–20 CpGs are bound (Figure 8B). This dependence is not Alu-specific: even outside genomic repeats, 300bp regions with zero CpGs have negligible probability of overlapping a ZCWPW1 binding site, while the overlap probability rises to >50% for regions containing 10 or more CpGs, suggesting CpGs are essential for PRDM9-independent ZCWPW1 binding (Figure 8C,D and Figure 8—figure supplement 1). ZCWPW1 appears to have greater affinity for methylated CpG pairs, but retains some affinity even for non-methylated regions. This binding mode was unexpected a priori, and does not appear to be easily explained by patterns of H3K4me3/H3K36me3 in the genome, which do not concentrate as strongly in these regions.

Discussion

Here, we exploited previous work on single-cell transcript profiling of mouse testis, which has established spatio-temporal (and likely functional) components of co-expressed genes (Jung et al., 2019), to identify co-factors relevant to PRDM9 function in specification of recombination hotspots and DSB repair during meiosis. We show that the histone methylation reader ZCWPW1, which is both highly co-expressed and co-evolving with PRDM9, is recruited by PRDM9 to its binding sites, likely mediated through the recognition of the dual H3K4me3-H3K36me3 histone mark that PRDM9 deposits.

Like Mahgoub and colleagues in a parallel analysis (Mahgoub et al., 2020), we find co-evolution of ZCWPW1 with PRDM9, consistent with proteins functioning in the same pathway. We find that this association appears to depend on particular domains. ZCWPW1 co-occurs with PRDM9 in species where PRDM9 possesses a SET domain predicted to retain catalytic activity for H3K4me3/H3K36me3, which the domains of ZCWPW1 are predicted to recognise, strongly suggesting a partnership totally dependent on this recognition. This co-occurrence can persist even in species where PRDM9 is not believed to position DSBs (Baker et al., 2017), suggesting the co-operation between these proteins does not completely depend on this aspect of PRDM9’s function. Interestingly, a predicted functional SET domain of PRDM9 mainly occurs when PRDM9 possesses a SSXRD domain, which in other proteins (Banito et al., 2018) recruits CXXC2 which binds to CpG islands – whether this connects to our observation of a potential CpG-binding domain in ZCWPW1, and CpG-binding behaviour in a human cell line, remains to be explored.

A recent study reported phenotyping of an independent Zcwpw1−/− mouse (Li et al., 2019a), and we observe similar complete loss of fertility in males, but not in young females. Why females deficient for ZCWPW1 are initially fertile (but lose fertility by ~8 months; Li et al., 2019a) is unknown, but resembles sexual dimorphism in loss-of-function mutants for other meiotic genes involved in homologous recombination, where females exhibit a milder fertility phenotype (Cahoon and Libuda, 2019; Morelli and Cohen, 2005; Zhang et al., 2019). However, we observe a distinct intracellular localisation of ZCWPW1 from that reported (Li et al., 2019a): while we also detect ZCWPW1 as a diffuse nuclear signal across the nucleus (although excluding the chromocenters) and localisation to the XY body in pachytene, we see an additional signal in mid-pachytene to diplotene cells at the ends of the synaptonemal complex. This might reflect the specificity of our polyclonal antibody, which is raised against the full-length mouse protein vs a monoclonal antibody raised against a C-terminal region, hence recognising a wider spectrum of ZCWPW1. Given that the Zcwpw1−/− mouse arrests in early pachytene, it is not possible to test later stages, so a caveat is that we cannot completely exclude off-target antibody recognition. If accurate, the late-stage localisation of ZCWPW1 at subtelomeric regions resembles other proteins (Speedy A, CDK2, TFR1, SUN1, TERB1/2, MAJIN, and KASH), which form a complex tethering the telomeres to the nuclear envelope to facilitate chromosome movement and pairing — essential for synapsis and fertility (Ashley et al., 2001; Ding et al., 2007; Horn et al., 2013; Shibuya et al., 2014; Tu et al., 2017). However, no evidence of impaired telomeric attachment was reported in the other Zcwpw1−/− mutant, based on normal localisation of TRF1 (Li et al., 2019a).

In parallel with this work, two other studies of ZCWPW1 Huang et al., 2020; Mahgoub et al., 2020 have performed complementary analyses, and together with our own results allow some clear conclusions to be drawn. We mapped positions of ZCWPW1 recruitment (of the human protein) in mitotic HEK293T cells, in the presence of human or chimpanzee PRDM9 and compared to assayed peaks for PRDM9, H3K4me3 and H3K36me3 (Altemose et al., 2017). We find that in co-transfected HEK293T cells ZCWPW1 is strongly recruited to PRDM9-bound sites, in an allele-specific manner. Mahgoub et al., 2020 also report PRDM9 allele-specific recruitment of ZCWPW1, but in hybrid mice carrying Prdm9Dom2 and Prdm9Cast alleles which encode proteins with distinct zinc finger array DNA binding specificities. Together with ZCWPW1 binding measurements from Huang et al., 2020, these results imply that across PRDM9 alleles and species - humans, mice, and likely chimpanzees - PRDM9 binding alone is sufficient to strongly recruit ZCWPW1 without requiring DSB formation (which does not occur in our system) or meiosis-specific co-factors (which would not be expected to be present in our cellular system).

A likely explanation of this recruitment is recognition of H3K4me3 and H3K36me3 that PRDM9 deposits by the CW and PWWP domains of ZCWPW1 respectively. Previously, Li and colleagues (Li et al., 2019a) showed co-immunoprecipitation of H3K4me3 with ZCWPW1. In HEK293T cells we observed that ZCWPW1 recruitment correlates with the levels of H3K4me3 and H3K36me3 that PRDM9 deposits, while Huang et al., 2020 reported that in mice with a non-functional CW domain no ZCWPW1 binding peaks were observed, suggesting that H3K4me3 recognition at least might be necessary for the recruitment of ZCWPW1 at hotspots in vivo. However, H3K4me3 cannot be the only factor because we observed much stronger recruitment of ZCWPW1 to PRDM9 binding sites than to other sites possessing similar H3K4me3 levels, which might be explained by the additional presence of H3K36me3 at PRDM9 binding sites increasing ZCWPW1’s recognition and/or its stable association at these sites. Indeed in cells with PRDM9, the strongest ZCWPW1-binding sites are almost all PRDM9-bound sites. Similarly in mice, >85% of ZCWPW1 peaks overlap both marks concomitantly (Huang et al., 2020; Mahgoub et al., 2020). Mahgoub et al., 2020 additionally reported that in vitro binding affinity of recombinant mouse ZCWPW1 is at least five fold greater for dual-modified H3K4me3/K36me3 peptides compared to peptides carrying the single modifications. Together, this paints a clear picture that ZCWPW1 likely directly recognises sites possessing the dual histone marks deposited by the PRDM9-HELLS complex (Spruce et al., 2020), the first such protein identified. Our evolutionary analyses suggest that this function might occur across almost all species where PRDM9 deposits these marks, even species where PRDM9 does not specify DSB locations (Baker et al., 2017).

Although it was suggested that ZCWPW1 might recruit the DSB machinery to PRDM9 binding sites (Li et al., 2019a; Spruce et al., 2020), we and Mahgoub et al., 2020 (using distinct assays) find that DSB positioning is completely unaffected by loss of ZCWPW1. However, the additional role PRDM9 plays besides DSB positioning, in enabling homologous chromosome pairing (Davies et al., 2016; Hinch et al., 2019Li et al., 2019b), apparently malfunctions in Zcwpw1−/− mice. Our DMC1 data, together with microscopy data, are most easily explained if, unlike in WT, DSBs at sites bound more strongly by PRDM9 do not repair more quickly. Instead, in Zcwpw1−/− males, all DSBs behave similarly to those on the X-chromosome — which repairs DSBs very slowly, even in WT animals, and does not synapse. This implies an unexpected role for ZCWPW1: in aiding PRDM9 in its second function, synapsis. ZCWPW1 therefore offers a (thus far) unique protein, directly recruited by PRDM9 binding and so linking PRDM9 binding with its downstream functions in aiding rapid DSB processing/repair at hotspots whose homologue is bound by PRDM9 (Davies et al., 2016; Hinch et al., 2019Li et al., 2019b).

PRDM9 is thought to aid homology search/chromosome pairing by such binding to the homologous chromosome, and hybrids where hotspots are highly ‘asymmetric’ (i.e. the homologues are not bound at the sites where DSBs occur, due to evolutionary hotspot erosion) show multiple features identical to Zcwpw1−/− mice. These include asynapsis, persistent DMC1 foci (at the asymmetric hotspots), and complete sterility in males only (Davies et al., 2016). In Zcwpw1−/− mice, unlike in WT mice, there are many persistent DMC1 foci and we observe that even strong PRDM9 binding to the homologue causes no reduction in DMC1 persistence. These mice therefore behave, in many ways, as if all hotspots are completely asymmetric. Thus, a parsimonious explanation of our findings is that ZCWPW1 functions by attaching to PRDM9-bound sites on the matching homologous chromosome, thus marking possible repair partner locations for DSB sites, to help pair homologues. This might for example lead to recruitment to the chromosomal axis (where DMC1 foci are found), for potential use as the homologous repair template of DSBs, perhaps mediated by the SCP-1-like domain identified in some (although not all) orthologous copies. Given we observe ZCWPW1 throughout the nucleus in early meiosis, such recruitment would likely be transient in nature. Alternatively, ZCWPW1’s part in mediating PRDM9’s downstream roles might also depend (or even, depend only) on it binding the chromosome on which DSBs occur.

Given that we still observe some synapsis, similar to Prdm9−/− mice, we suggest that a ‘back-up’ mechanism, which is independent of PRDM9 binding level, is acting to attempt homology search/chromosome pairing in Zcwpw1−/− mice. In fact, some such mechanism is likely necessary, to repair DSBs happening within a subset of hotspots that are asymmetric due to naturally occurring variation (as seen in for example humans and heterozygous mice). If our observed >9 fold changes in relative DMC1 levels for strongly vs weakly bound hotspots purely reflect increased DMC1 persistence, it might be that this mechanism is greatly slower than the PRDM9-dependent pathway for WT mice. In turn, this could explain many – or all – of the downstream phenotypes in these mice, as consequences of delayed homology search.

The simplest explanation of our observation that ZCWPW1 binds weakly to many sites in a CpG-dependent manner in HEK293T cells is that these sites might be directly bound by a methyl-CpG binding domain (MBD) putatively observed in the human, mouse, and coelacanth Zcwpw1, among others. Alternatively, recruitment might be indirect, with ZCWPW1 forming a complex with an MBD containing protein, for example SETDB1 which is expressed in HEK293T cells, co-expressed with ZCWPW1 and PRDM9 (Jung et al., 2019; Schultz et al., 2002), and interacts with several other identified PRDM9 partner proteins (Mulligan et al., 2008; Parvanov et al., 2017). Given the precisely regulated developmental changes in methylation status, ZCWPW1/PRDM9 protein abundance, and chromatin during meiosis (Gaysinskaya et al., 2018; Seisenberger et al., 2013) it is unclear whether ZCWPW1 binding will also be affected by CpGs in vivo. Although more modest, Huang et al., 2020 did also report enrichment of ZCWPW1 at B1 elements in mice, the closest equivalent of human Alus. Unlike Mahgoub et al., 2020, Huang et al., 2020 also observed overlap of ZCWPW1 peaks with CpG islands (CGI) more frequently than expected. Given (i) they and Mahgoub et al., 2020 detect a much lower number of peaks (14,688 and 4,487 respectively) than we do in our HEK293T system (800,000), and (ii) we only find Alu enrichment in weaker ZCWPW1 peaks (of our top 10,000 peaks, >90% would be PRDM9-bound sites), these results are not conflicting, and instead might depend on detection/behaviour of weaker peaks, and/or presence of PRDM9 in ZCWPW1-containing cells. If CpG dinucleotides do (although weakly, relative to PRDM9) affect in vivo ZCWPW1 binding affinity and given that CpG methylation is associated with transposons, it is interesting to speculate whether ZCWPW1 might play a role in germline transposon recognition, silencing, or facilitate repair of DSBs occurring within transposable elements.

Zcwpw1 possesses a paralogue, Zcwpw2, which is also co-expressed in testis with Prdm9 (although at a lower level), contains both CW and PWWP domains, and is able to recognise H3K4me3 at least (Liu et al., 2016). Further studies are required to investigate what function if any Zcwpw2 has in meiosis. One interesting possibility is that ZCWPW2, rather than ZCWPW1, might act to help position DSBs – if so, its low abundance relative to ZCWPW1 would be analogous to the low number of DSBs per meiosis (~300) relative to the number of PRDM9 binding sites.

In this study, we have shown that ZCWPW1 co-evolves with PRDM9, binds to sites marked by PRDM9, and is required for the proper processing, removal of DMC1, and repair of DSBs. How exactly it mediates repair, if for example by recruitment of other proteins, remains to be determined. This study also further demonstrates that genes co-expressed with Prdm9 represent a rich source of undiscovered meiotic genes, with important functional implications for recombination, meiosis progression and ultimately fertility. ZCWPW1 impacts synapsis, downstream of DSB formation. Therefore the protein(s) responsible for recognising PRDM9-marked sites for DSB formation, and for aiding synapsis, are at least partly different, and we anticipate that future studies will likely uncover additional PRDM9-recruited proteins.

Materials and methods

Key resources table
Reagent type
(species) or resource
DesignationSource or referenceIdentifiersAdditional information
Gene (167 species)ZCWPW1This paper using BlastP and tBLASTn (www.blast.ncbi.nlm.nih.gov), NCBI (www.ncbi.nlm.nih.gov) and Ensembl (www.ens.embl.org)Details in Materials and methodsAlso see Figure 1—source data 1
Gene (225 species)PRDM9Baker et al., 2017 (doi: 10.7554/eLife.24133)Also see Figure 1—source data 2
Genetic reagent (Mus musculus)Zcwpw1-/-Toronto Centre for Phenogeno-mics (Canada)RRID:IMSR_CMMR:ADVN;
Strain name C57BL/6N-Zcwpw1em1(IMPC)Tcp
Constitutive knock out for Zcwpw1 carrying a 1485bp CRISPR/Cas9-induced deletion (chr5:137799545–13780101029)
Genetic reagent (Mus musculus)Prdm9-/-RIKEN BioResource Research Center (Japan)RRID:MGI:3624989; Strain name B6.129P2-Prdm9 < tm1Ymat>, strain number RBRC05145Originating article Hayashi et al., 2005.
Cell line (Homo sapiens)Embryonic Epithelial KidneyATCCCat. CRL-3216
Transfected construct (Homo sapiens)hPRDM9-V5-YFPAltemose et al., 2017 (doi: 10.7554/eLife.28383)Human PRDM9 B allele cloned into pLENTI CMV/TO Puro DEST vector (Addgene plasmid #17293; Campeau et al., 2009) in frame with a Twin-strep tag, a V5 tag, and a self-cleaving YFP tag due to the presence of an upstream P2A sequence
Transfected construct (Homo sapiens/Pan troglodyte hybrid)cPRDM9-V5-YFPAltemose et al., 2017 (doi: 10.7554/eLife.28383)hPRDM9-V5-YFP construct where Exon 10 encoding the human zinc finger array was replaced with the equivalent sequence from the chimpPRDM9 w11a allele
Transfected construct (Homo sapiens)hZCWPW1-HAGenScriptClone ID OHu16813ZCWPW1, transcript variant 1, mRNA (NM_017984.5) cloned into pCDNA3.1+/C-HA
Transfected construct (Homo sapiens)hZCWPW2-HAGenScriptClone ID OHu31001CZCWPW2, transcript variant 1, mRNA (NM_001040132.3) cloned into pCDNA3.1+/C-HA
Transfected construct (M. musculus)mZCWPW1-FLAGOriGeneClone ID MR209594Zcwpw1, Transcript variant 2 mRNA (NM_001005426) cloned with a C-terminal Myc-DDK(FLAG) tag
Transfected construct (M. musculus)mZCWPW2-FLAGThis paperpCMV6-Entry (OriGene, Cat. PS100001)Generated by cloning custom-synthesised mZCWPW2 into pCMV6-Entry
Transfected construct (M. musculus)mZCWPW1-HisThis paperpET22b(+) Novagen (Sigma-Aldrich, Cat. 69744)Generated by sub-cloning mZCWPW1 from clone ID MR209594 (OriGene) into pET22b(+) in frame with a C-terminal 6-histidines tag
Recombinant DNA reagentmZCWPW2OrigeneMouse Zcwpw2-206; Transcript ID ENSMUST00000238919.1Custom synthesis of full-length cDNA sequence
Strain, strain background (Escherichia coli)BL21(DE3)Thermo Fisher ScientificCat. C600003Chemically competent cells
Peptide, recombinant proteinmZCWPW1-HisThis paperUsed to produce a rabbit polyclonal antibody against mouse ZCWPW1 by immunisation (Eurogentec)
AntibodyAnti-mouse ZCWPW1 antiserum, and pre-immune serum (rabbit polyclonal)This paperCustom generation (Eurogentec)IF (1:100), WB (1:1000), IP (5 μl on transfected cells, 10 μl on mouse testis)
AntibodyAnti-Human ZCWPW1
(mouse monoclonal)
Sigma-AldrichCat. SAB1409478WB (1:2000)
AntibodyAnti-SYCP3 (mouse monoclonal)Santa Cruz BiotechnologyCat. sc-74569,
RRID:AB_2197353
IF (1:100)
AntibodyAnti-SYCP3 (biotinylated, rabbit polyclonal)NovusCat. NB300-232, RRID:AB_2087193IF (1:100)
AntibodyAnti-DMC1 (rabbit polyclonal)Santa Cruz BiotechnologyCat. sc-22768, RRID:AB_2277191,
Discontinued
IF (1:100)
AntibodyAnti-DMC1 2H12/4 (mouse monoclonal)NovusCat. NB100-2617, RRID:AB_2245859ChIP (5 μg)
AntibodyAnti-HORMAD2 (rabbit polyclonal)Santa Cruz BiotechnologyCat. sc-282192,
RRID:AB_2121124
IF (1:300)
AntibodyAnti-RAD51 (mouse monoclonal)AbcamCat. ab88572,
RRID:AB_2042762
IF (1:50)
AntibodyAnti-RPA2 (rabbit polyclonal)AbcamCat. ab10359,
RRID:AB_297095
IF (1:1000)
AntibodyAnti-phospho-H2AX
(mouse monoclonal)
Sigma-AldrichCat 05–636, RRID:AB_309864IF (1:250)
AntibodyAnti-phospho γ-H2AX
(chicken polyclonal)
BiorbytCat. orb195374
Discontinued
IF (1:1000)
AntibodyAnti-rabbit IgG Alexa Fluor 488 secondary
(goat polyclonal)
Thermo Fisher ScientificCat. A-11008,
RRID:AB_143165
IF (1:250)
AntibodyAnti-mouse IgG Alexa Fluor 488 secondary
(goat polyclonal)
Thermo Fisher ScientificCat. A-11001,
RRID:AB_2534069
IF (1:250)
AntibodyAnti-rabbit IgG Alexa Fluor 594 secondary
(goat polyclonal)
Thermo Fisher ScientificCat. A-11012,
RRID:AB_141359
IF (1:250)
AntibodyAnti-mouse IgG Alexa Fluor 594 secondary
(goat polyclonal)
Thermo Fisher ScientificCat. A-11005,
RRID:AB_141372
IF (1:250)
AntibodyAnti-mouse IgG Alexa Fluor 647 secondary
(goat polyclonal)
Thermo Fisher ScientificCat. A-21235,
RRID:AB_2535804
IF (1:250)
AntibodyAnti-chicken IgY Alexa Fluor 647 secondary
(goat polyclonal)
Thermo Fisher ScientificCat. A-21449,
RRID:AB_2535866
IF (1:250)
AntibodyStreptavidin, Alexa Fluor 647Thermo Fisher ScientificCat. S32357IF (1:50)
AntibodyAnti-poly-His (mouse monoclonal)Sigma-AldrichCat. H1029,
RRID:AB_260015
WB (1:2000)
AntibodyAnti-HA (rabbit polyclonal)AbcamCat. ab9110, RRID:AB_307019IF (1:100), WB (1:1000), IP (2 μg), ChIP (5 μg)
AntibodyAnti-HA (mouse monoclonal)Sigma-AldrichCat. H3663, RRID:AB_262051IF (1:500)
AntibodyAnti-V5 (rabbit polyclonal)AbcamCat. ab9116, RRID:AB_307024IF (1:500)
AntibodyAnti-FLAG M2 (mouse monoclonal)Sigma-AldrichCat. F3165,
RRID:AB_259529
IF (1:500), WB (1:2000), IP (3 μg)
AntibodyAnti-β-Actin (mouse monoclonal)Sigma-AldrichCat. A1978,
RRID:AB_476692
WB (1:2000)
AntibodyECL Rabbit IgG, HRP-linked whole Ab (donkey polyclonal)GE HealthcareCat. NA934,
RRID:AB_772206
WB (1:10000)
AntibodyECL Mouse IgG, HRP-linked whole Ab (sheep polyclonal)GE HealthcareCat. NA931,
RRID:AB_772210
WB (1:10000)
Sequence-based reagentpIRESMinorChan et al., 2017biotin labelled minor satellite probe
Sequence-based reagentGAPDH_F
(Human)
OriGenePCR primers, transcript detection, NM_002046GCTCCTCTGACTTCAACAGCGGCT
Sequence-based reagentGAPDH_R
(Human)
OriGenePCR primers, transcript detection, NM_002046ACCACCCTGTTGCTGTAGCCAA
Sequence-based reagentPRDM9_F
(Human)
OriGenePCR primers, transcript detection, NM_020227ACGAAGAGGCAGCCAACAATGG
Sequence-based reagentPRDM9_R
(Human)
OriGenePCR primers, transcript detection, NM_020227GCCACCAGGTTCTGCTCTTCAT
Sequence-based reagentZCWPW1_F
(Human)
OriGenePCR primers, transcript detection,
NM_017984
GATGGCTCAAGAGGCAGAACAG
Sequence-based reagentZCWPW1_R
(Human)
OriGenePCR primers, transcript detection,
NM_017984
TGGGCTGTTCAAACCAGAGAGC
Sequence-based reagentZCWPW2_F
(Human)
OriGenePCR primers, transcript detection,
NM_001040432
AAGAGCTGGAGCAAATGCTGCAG
Sequence-based reagentZCWPW2_R
(Human)
OriGenePCR primers, transcript detection,
NM_001040432
CAGGAGCTTCTGGGCTGCATTT
Commercial assay or kitTelomere PNA FISH Kit/Cy3AgilentCat. K5326
Commercial assay or kitPierce BCA protein assay kitThermo Fisher ScientificCat. 23227
Commercial assay or kitECL Prime Western Blotting Detection ReagentGE HealthcareCat. 10308449
Commercial assay or kitMinelute Reaction Cleanup KitQIAGENCat. 28204
Commercial assay or kitQubit dsDNA HS Assay kitThermo Fisher ScientificCat. Q32851
Chemical compound, drugIPTGSigma-AldrichCat. I55020.5 mM final
OtherFast SYBR Green Master MixApplied BiosystemsCat. 4385610RNA extraction and RT-qPCR
OtherDynabeads M-280 Sheep anti-Rabbit IgGThermo Fisher ScientificCat. 11203D,
RRID:AB_2783009
IP and ChIP experiments;
IP (25–75 ul), ChIP (65 ul)
OtherDynabeads M-280 Sheep anti-Mouse IgGThermo Fisher ScientificCat. 11202D,
RRID:AB_2783640
IP and ChIP experiments;
IP (25 ul), ChIP (65 ul)
OtherTALON Metal Affinity ResinTakaraCat. 635502Expression and purification of ZCWPW1 recombinant protein; 2 ml per L of IPTG-induced bacterial culture
OtherTRI ReagentSigma-AldrichCat. T9424RNA extraction and RT-qPCR
OtherProtease Inhibitor CocktailSigma-AldrichCat. P8340IP and WB detection; 1:100 dilution
OtherComplete Mini Protease Inhibitor CocktailSigma-AldrichCat. 11697498001ChIP; 1 tablet in 10 ml volume
OtherNovex WedgeWell 4%to 20%, Tris-Glycine, Protein GelThermo Fisher ScientificCat. XP04200BOXIP and WB detection
OtherNovex WedgeWell 8%, Tris-Glycine, Protein GelThermo Fisher ScientificCat. XP00080BOXIP and WB detection
Software, AlgorithmMAPeakCallerAltemose et al., 2017 (doi: 10.7554/eLife.28383)https://github.com/MyersGroup/PeakCaller/ (archived at https://doi.org/10.5281/zenodo.3783600)
Software, AlgorithmBWA MEMLi, 2013 (arXiv:1303.3997)bwa mem (version 0.7.17-r1188)
Software, AlgorithmbwtoolPohl and Beato, 2014 (doi:10.1093/bioinformatics/btu056)RRID:SCR_003035; v1.0https://github.com/CRG-Barcelona/bwtool
Software, AlgorithmPicard‘Picard Toolkit.’ 2019. Broad Institute, GitHub Repository. http://broadinstitute.github.io/picard/; Broad InstituteRRID:SCR_006525; version 2.20.4-SNAPSHOT
Software, AlgorithmSAMtoolsPMID:19505943RRID:SCR_002105; v1.9https://www.htslib.org/download/
Software, AlgorithmBEDtoolsQuinlan and Hall, 2010 (doi:10.1093/bioinformatics/btq033)RRID:SCR_006646; v2.28.0bedtools.readthedocs.io
Software, AlgorithmSEQkitShen et al., 2016 (doi:10.1371/journal.pone.0163962)
Software, AlgorithmIGVThorvaldsdóttir et al., 2013 (doi: 10.1093/bib/bbs017)

Orthologue alignment

Request a detailed protocol

We identified ZCWPW1 orthologues across species using four data sources: first, we used BlastP (blast.ncbi.nlm.nih.gov), against the full-length human reference ZCWPW1 sequence (identifier NP_060454.3, against nr_v5 database) storing the top 1000 hits using the default parameters (set 1). Secondly, we downloaded the two sets of Ensembl identified ZCWPW1 orthologues (against ENSMUSG00000037108.13; ensembl.org, 105 orthologues), and identified NCBI ZCWPW1 protein orthologues (146 species, www.ncbi.nlm.nih.gov). Initial examination of identified orthologues revealed conservation among orthologues mainly of the CW and PWWP domains; therefore, to find additional orthologous sequences we performed tBLASTn (against the nr/nt nucleotide collection 17th July 2019; top 1000 hits) to identify orthologues in the NCBI nucleotide database, to the partial sequence (amino acids 256–339 of the reference sequence NP_060454.3), corresponding to the CW and PWWP domains of ZCWPW1. Protein sequences were then aligned against full-length human ZCWPW1 (NP_060454.3) using BLASTP2.9 (Altschul et al., 2005; Altschul et al., 1997). We obtained taxonomy information from the NIH classification https://ftp.ncbi.nlm.nih.gov/pub/taxonomy/new_taxdump/, for comparisons among species and between ZCWPW1 and PRDM9.

Previous work has identified highly conserved ‘KWR’ and ‘PWWP’ patterns among CW and PWWP domains, respectively (He et al., 2010b; Qin and Min, 2014). We thus identified an initial set of ‘clear’ ZCWPW1 orthologues, and then used these to identify further less precise matches. ‘Clear’ orthologues are defined as proteins within set one containing perfect matches to both these sequences, and such that in the NCBI alignment, at least 39% of each sequence aligns to the human protein, and conversely. This second step is required to avoid spurious matches to, for example, ZCWPW2, which overlaps 19% of ZCWPW1, by requiring >2 fold this match length; although several ZCWPW2 copies are in our initial list, no gene annotated as most similar to ZCWPW2 attains the 39% overlap. For proteins within the ‘clear’ set, we identified the longest alignment for each given species, resulting in 136 species with a likely ZCWPW1 orthologue. This initial screen identified three groups of fish, placental mammals, marsupials, monotremes and reptiles as likely possessing ZCWPW1, similar to our final conclusions. Moreover, it defined amino acids 247–428 in the human reference sequence as being conserved in the alignment (at most one sequence not aligning), including the annotated CW and PWWP domains.

Using this initial set, we identified additional orthologous sequences and refined our results, by identifying conserved bases within ZCWPW1. Specifically, we divided the 136 species into major clades, as in Baker et al., 2017, and gave each sequence a weighting so that the overall weight for each clade was the same (so, for example, each Placental mammal sequence was downweighted as this clade was over-represented). Within the consistently aligned region 247–428, we calculated the overall weighted probability of each of the 20 amino acids, or a gap (adding 10−5 to exclude zero weights). This identified 31 completely conserved amino acids, and 78 amino acids whose entropy was below 1 (equivalent to two amino acids having equal probability, so implying one amino acid present in >50% of species). Finally, we defined ZCWPW1 orthologues as those sequences matching at least 90% of those 31 perfectly conserved bases which were aligned, and aligning to at least 50% of these bases (this last condition allows for inclusion of incomplete sequencing or protein assembly). We note that while this 90% condition is arbitrary, by definition all annotated orthologues exceed this threshold, while thresholds below ~70–80% are exceeded by orthologues of ZCWPW2 among other genes, offering some justification. Nonetheless, it is worth pointing out that our analysis does not rule out the existence of more poorly conserved ZCWPW1 copies in more distantly related species. This approach identified a final set of 167 genes, which we annotated as likely ZCWPW1 orthologues and used for the majority of results. For each ZCWPW1 orthologue, we also identified the taxonomic relationship of the closest species possessing such a PRDM9 copy (Figure 1—source data 1). We also reciprocally annotated each PRDM9 copy previously identified (Baker et al., 2017) according to the closest species also possessing ZCWPW1 (Figure 1—source data 1).

Our analyses revealed a relationship between ZCWPW1 predicted functional domains and PRDM9 histone modifications. Therefore, we identified additional potential conserved sites by identifying perfectly conserved bases among those species possessing both ZCWPW1 and PRDM9 orthologues, and where three key SET domain catalytic amino acids within PRDM9 are intact, meaning PRDM9 is predicted to have normal histone modification activity (Baker et al., 2017). This identified a slightly larger number of conserved amino acids (37). Only eight of these varied in any of the 167 species with a potential ZCWPW1 orthologue: 260, 404 and 411 (in two Xenopus frogs, with 411 also in white-headed capuchin), 19 and 22 (in three canines), 25 (in Ocelot gecko), 257 (in Anolis lizard and in Wombat), 325 (in elephantfish). While these changes might alter ZCWPW1 function, the true behaviour of ZCWPW1 is uncertain in these cases, although it is interesting that canids and frogs represent clades that all appear to have lost PRDM9, and possess multiple, clustered, amino acid changes.

Domain search

Request a detailed protocol

pDomThreader (Lobley et al., 2009) was used via the PSIPRED server (http://bioinf.cs.ucl.ac.uk/psipred/) on the following uniProt amino acid sequences (Q9H0M4|ZCPW1_HUMAN, Q6IR42|ZCPW1_MOUSE, E2RFJ2|E2RFJ2_CANLF, M3XJ39_LATCH, A0A3S5ZP38_BOVIN, M3WDY6_FELCA, and G3ULT5|G3ULT5_LOXAF) all of which identified a match to 1ub1A00 in the C terminal section after the PWWP domain, except in LOXAF in which this match was slightly below the p-value threshold of 0.001.

SCP1 prediction used NCBI conserved domain search server (https://www.ncbi.nlm.nih.gov/Structure/cdd/wrpsb.cgi?) with the uniProt amino acid sequences Q9H0M4|ZCPW1_HUMAN and Q6IR42|ZCPW1_MOUSE.

Mice and genotyping

Request a detailed protocol

KO mice for Zcwpw1 (C57BL/6N-Zcwpw1em1(IMPC)Tcp) were generated by the Toronto Centre for Phenogenomics (Canada). A 1485bp deletion on chromosome 5 spanning exons 5 to 7 of Zcwpw1 (chr5:137799545–13780101029) was engineered by CRISPR/Cas9 using guide RNAs 5’-GACTGCACTCACGGCCATCT-3’ and 5’-GCCCGGTTCTTCATCCAATT-3’. The frameshift deletion introduces a stop codon in Exon 8, leading to a predicted unstable short truncated 492bp transcript. Mice were genotyped at the Zcwpw1 locus using the following primers, and standard cycling conditions: KO allele-Forward, 5’-CACAGGCTCATGTATGTTTGTCTC-3’; KO allele-Reverse, 5’-CTGCTTCGTCCTCTTTCCTTATCTC-3’; WT allele-Forward, 5’-TGCCACCACACTTCATTTGT-3’; WT allele-Reverse CCTGTTTCCTTCCCAACTCA-3’. The deletion was verified by direct Sanger sequencing of the KO genotyping PCR product (Source Bioscience, UK), following purification with the QIAquick PCR Purification Kit (Qiagen). Sequence analysis was carried out using Chromas LITE (version 2.1.1).

KO mice for Prdm9 were described previously (Hayashi et al., 2005; Mihola et al., 2019) and obtained from the RIKEN BioResource Research Center in Japan (strain name B6.129P2-Prdm9 < tm1Ymat>, strain number RBRC05145). Mice were genotyped at the Prdm9 locus using the following primers, and standard cycling conditions: WT allele-Forward 5’-AGGAATCTTCCTTCCTTGCTGTCG-3’; WT allele-Reverse 5’-ATTTCCCTGTATCTTCTTCAGGACT-3’; KO allele-Reverse 5’-CGCCATTCAGGCTGCGCAACTGTT-3’.

All animal experiments received local ethical review approval from the University of Oxford Animal Welfare and Ethical Review Body (Clinical Medicine board) and were carried out in accordance with the UK Home Office Animals (Scientific Procedures) Act 1986.

Fertility measurements

Request a detailed protocol

Fertility was assessed in mice ranging from 9 to 12 weeks of age, either by mating with WT littermates and recording the average litter size and frequency, or by measuring paired testes weight (normalized to lean body weight), sperm count (per paired epididymides) and chromosome synapsis rate (by immunostaining of pachytene spermatocytes) in males. Lean body weight was measured using the EchoMRI-100 Small Animal Body Composition Analyzer.

Immunostaining of spermatocytes

Request a detailed protocol

Mouse testis chromosome spreads were prepared using surface spreading (Barchi et al., 2008; Peters et al., 1997) and immunostained as previously described (Davies et al., 2016). The following primary antibodies were used: custom ZCWPW1 rabbit antiserum (1:100 dilution), mouse anti-SYCP3 (Santa Cruz Biotechnology sc-74569, RRID:AB_2197353) or biotinylated rabbit anti-SYCP3 (Novus NB300-232, RRID:AB_2087193), rabbit anti-DMC1 (Santa Cruz Biotechnology sc-22768, RRID:AB_2277191, discontinued; 1:100 dilution), rabbit anti-HORMAD2 (Santa Cruz Biotechnology sc-82192, RRID:AB_2121124; 1:300 dilution), mouse anti-RAD51 (Abcam ab88572, RRID:AB_2042762; 1:50 dilution), rabbit anti-RPA2 (Abcam ab10359, RRID:AB_297095; 1:1000 dilution), mouse (Sigma-Aldrich 05–636, RRID:AB_309864; 1:250 dilution) or chicken (Biorbyt orb195374, discontinued; 1:1000 dilution) anti-phospho γ-H2AX and Alexa Fluor 488-, 647- or 594-conjugated secondary antibodies against rabbit, mouse or chicken IgG/Y (A-11008, RRID:AB_143165; A-11001, RRID:AB_2534069; A11012, RRID:AB_141359; A-1105, RRID:AB_141372; A-21235, RRID:AB_2535804; A-21449, RRID:AB_2535866; Thermo Fisher Scientific; 1:250 dilution), as well as streptavidin Alexa Fluor 647 (Thermo Fisher Scientific) were used to detect the primary antibodies. Images were acquired using either a BX-51 upright wide-field microscope equipped with a JAI CVM4 B and W fluorescence CCD camera and operated by the Leica Cytovision Genus software, or a Leica DM6B microscope for epifluorescence, equipped with a DFC 9000Gt B and W fluorescence CCD camera, and operated via the Leica LASX software. Image analysis was carried out using Fiji (ImageJ-win64). The specificity of the ZCWPW1 signal was verified using the pre-immune serum from the same rabbit: no staining was visible at any stage of prophase I (Figure 2—figure supplement 1B shows representative images of mid-zygotene and early pachytene cells). Staging and sub- (early, mid and late) staging of spermatocytes during prophase I was carried out according to defined staining patterns against SYCP1, SYCP3, and DAPI-stained DNA appearance, and morphological criteria previously published (Gaysinskaya et al., 2014). Briefly, cells showing large chromocentres (brightly stained by DAPI and corresponding to pericentromeric regions) with short tracts of SYCP3 were assigned as (mid-)leptotene. Cells with larger pooled chromocentres, associated with longer tracts of SYCP3 were identified as early zygotene. Cells where some synapsis was initiated (with more than two paired SYCP3 tracts, resulting in a thicker synaptonemal complex region), were assigned as (mid-)zygotene. Cells where the majority of chromosomes were synapsed, with three or less completely or partially asynapsed, were identified as late zygotene.

Fluorescent in-situ hybridisation (FISH)

Request a detailed protocol

Following immunostaining of spermatocytes, centromeres and telomeres were labelled using a biotin labelled minor satellite probe (pIRESMinor, Chan et al., 2017) together with the Telomere PNA FISH Kit/Cy3 (Agilent), following the manufacturer’s instructions, but without protease treatment. The hybridisation and signal detection were carried out using standard techniques.

Plasmids

Constructs encoding full-length human PRDM9 (B allele) in frame with a C-terminal V5 tag and a self-cleaving YFP tag (hPRDM9-V5-YFP), or the chimp construct (cPRDM9-V5-YFP) where the sequence that encodes the human zinc finger array in hPRDM9-V5-YFP was replaced with the equivalent sequence from the chimp PRDM9 w11a allele, for expression in mammalian cells (pLENTI CMV/TO Puro DEST backbone vector) were described previously (Altemose et al., 2017). Constructs encoding full-length human (h) and mouse (m) ZCWPW1 or ZCWPW2 for expression in mammalian cells were purchased from GenScript (hZCWPW1 and hZCWPW2 cDNA with a C-terminal HA tag in pCDNA3.1 clones ID OHu16813 and OHu31001C, respectively) and OriGene (mZCWPW1 cDNA with a FLAG-Myc dual tag in pCMV6-Entry; clone ID MR209594), respectively. mZCWPW2 cDNA was synthesised by GenScript and subcloned into pCMV6-Entry in-frame with a C-terminal FLAG tag. For expression in E. coli, full-length mZCWPW1 cDNA was subcloned into the pET22b(+) vector (Novagen) with a C-terminal poly-His tag.

ZCWPW1 antibody production and validation

Request a detailed protocol

pET22b-mZCWPW1-His construct was transformed into BL21 (DE3) E. coli cells (Thermo Fisher Scientific). Bacterial cultures were grown to a density with O.D600 ~0.7, and expression of recombinant His-tagged ZCWPW1 protein was induced overnight at 20°C by addition of IPTG to 0.5 mM. The protein was purified using TALON metal affinity resin, according to the manufacturer’s instructions (Takara 635502). Further purification was carried out by size exclusion chromatography (Superdex HiLoad 200 16/60, GE Life Sciences). WB validation was carried out as described previously (Altemose et al., 2017), using mouse anti-human ZCWPW1 (Sigma-Aldrich SAB1409478; 1:2000 dilution) and mouse anti-polyhistidine (Sigma-Aldrich H1029, RRID:AB_260015; 1:2000 dilution) antibodies. The purified protein was used to immunise two rabbits (Eurogentec, Belgium), and the resulting immune antisera (and pre-immmune sera from the same rabbits) were tested against the recombinant antigen by ELISA (Eurogentec), the overexpressed FLAG-tagged mouse ZCWPW1 protein (and the closely related ZCWPW2 protein to address specificity) in transfected HEK293T cells, and the endogenous protein in mouse testes (Figure 2—figure supplement 1). Both antisera showed high titers and similar reactivity by IP, WB and IF staining. The antiserum with the lowest level of background (corresponding to the signal detected by the pre-immune serum) was chosen to carry out the experiments.

Cell line, transfection and immunofluorescence staining

Request a detailed protocol

Human embryonic kidney (HEK) 293 T cells were purchased from the ATCC (ATCC CRL-3216), with a certificate of analysis confirming cell line identity by Short Tandem Repeat profiling and lack of mycoplasma contamination. All experiments were carried out on cells cultured for less than five passages from the purchased stock reference strain. Cells were cultured and transfected using Fugene HD as previously described (Altemose et al., 2017). High and comparable expression of the target proteins across all samples was verified by IF staining of duplicate transfected cultures before proceeding to ChIP (Figure 5—figure supplement 4). PRDM9 expression was visualised either directly from live cells through YFP fluorescence, or by IF staining of fixed cells using rabbit anti-V5 (Abcam ab9116, RRID:AB_307024; 1:500 dilution) as described previously (Altemose et al., 2017). ZCWPW1-HA was detected by immunostaining as above, using mouse anti-HA (Sigma-Aldrich H3663, RRID:AB_262051, 1:500 dilution). The fraction of cells co-expressing ZCWPW1-HA and PRDM9-V5 was determined by co-staining with both antibodies, as above (Figure 5—figure supplement 4). To assess the specificity of the ZCWPW1 antiserum, cells transfected with FLAG-tagged mZCWPW1 or mZCWPW2 were co-immunostained with the antiserum (1:100 dilution) and a mouse anti-FLAG (Sigma-Aldrich F3165, RRID:AB_259529; 1:500 dilution). Images were acquired using a ZOE fluorescent cell imager (Bio-Rad).

RNA extraction and reverse transcription-quantitative PCR (RT-qPCR)

Request a detailed protocol

RT-qPCR analysis of human PRDM9, ZCWPW1 and ZCWPW2 transcript expression in HEK293T cells was performed as previously described (Altemose et al., 2017), with minor modifications listed here. RNA was extracted from one biological replicate per sample using TRI Reagent (Sigma-Aldrich T9424), following the manufacturer’s guidelines. For qPCR, each sample was analysed in triplicate (technical replicates) using primer sets from Origene: GAPDH (NM_002046) forward 5’-GCTCCTCTGACTTCAACAGCG-3’ and reverse 5’-ACCACCCTGTTGCTGTAGCCAA-3’; PRDM9 (NM_020227) forward 5’-ACGAAGAGGCAGCCAACAATGG-3’ and reverse 5’-GCCACCAGGTTCTGCTCTTCAT-3’; ZCWPW1 (NM_017984) forward 5’-GATGGCTCAAGAGGCAGAACAG-3’ and reverse 5’-TGGGCTGTTCAAACCAGAGAGC-3’; ZCWPW2 (NM_001040432) forward 5’- AAGAGCTGGAGCAAATGCTGCAG-3’ and reverse 5’-CAGGAGCTTCTGGGCTGCATTT-3’. PCR reactions were carried out in triplicate (three technical replicates per sample), using Fast SYBR Green Master Mix according to the manufacturer’s instructions (Applied Biosystems 4385610). Relative gene expression was calculated using the ΔΔCt method after averaging the technical replicates for each sample and normalising to the GAPDH gene. Raw Ct values and detailed calculations are given in Figure 5—source data 1.

Immunoprecipitation and western blot detection

Request a detailed protocol

Cell and testes protein extracts were prepared in lysis buffer containing 50 mM Tris-HCl pH8.0, 150 mM NaCl, 1% Triton X-100 and a cocktail of protease inhibitors (Sigma-Aldrich P8340), followed by gentle rotation at 4°C for 30 to 45 min, respectively. Cell debris were pelleted by centrifugation at 4°C for 20 min at 20,000 g. Protein extracts were quantified using the Pierce BCA protein assay kit (Thermo Fisher Scientific 23227) and equal quantities were incubated overnight at 4°C with 5 or 10 µl (on transfected cells and mouse testis, respectively) of rabbit ZCWPW1 immune antiserum or pre-immune serum, or 3 µg of mouse anti-FLAG antibody (Sigma-Aldrich F3165, RRID:AB_259529), setting aside 5–100 µg of protein extract for direct western blot detection. Immunocomplexes were pulled down with 25–75 µl of Dynabeads M-280 sheep anti-rabbit (11203D, RRID:AB_2783009) or anti-mouse (11202D, RRID:AB_2783009) IgG (Thermo Fisher Scientific) for 2 hr at 4°C with gentle rotation. After five washes in lysis buffer, they were eluted from the beads by boiling in Laemmli sample buffer for 5 min. Immunocomplexes and total protein extracts were resolved on either 8% or 4–20% Novex Tris-Glycine precast gels (Thermo Fisher Scientific XP00080BOX, XP04200BOX). Proteins were transferred onto PVDF membranes and the proteins of interest were detected by western blotting following standard procedures. Blots were blocked for 1 hr at room temperature (RT) in PBS containing 0.2% Tween-20 (PBS-T) and 5% milk, and incubated for 1 hr at RT with rabbit ZCWPW1 immune antiserum or pre-immune serum (1:1000 dilution), mouse anti-β-actin (Sigma-Aldrich A1978, RRID:AB_476692, 1:2000 dilution), or mouse anti-FLAG antibody (Sigma-Aldrich F3165, RRID:AB_259529; 1,2000 dilution), washed three times in PBS-Tween buffer and incubated for 1 hr at RT with HRP-conjugated donkey anti-rabbit (NA934, RRID:AB_772206) or sheep anti-mouse (NA931, RRID:AB_772210) IgG antibody (1:10,000 dilution; GE Healthcare). Protein signals were revealed using the ECL Prime western blotting detection reagent according to the manufacturer’s recommendations (GE Healthcare 10308449). The specificity of the ZCWPW1 signal observed in B6 testis extract was verified by immunoprecipitation using the pre-immune serum from the same rabbit: no signal was detected (Figure 2—figure supplement 1C). No cross-reactivity was observed with ZCWPW2 in either B6 testis extract or transfected HEK293T cells overexpressing FLAG-tagged ZCWPW2 (Figure 2—figure supplement 1C,D).

ChIP-seq

ChIP

Request a detailed protocol

ChIP against ZCWPW1-HA was carried out from transfected HEK293T cells as follows. Cells were crosslinked for 10 min in 1% formaldehyde, the reaction was quenched for 5 min by the addition of glycine to a final concentration of 125 mM, and the cells were washed twice in cold PBS. The cell pellet was resuspended in cold sonication buffer (50 mM Tris-HCl pH8, 10 mM EDTA, 1% SDS) supplemented with Complete Mini protease inhibitor cocktail (Sigma-Aldrich 11697498001), and chromatin was sheared to an average size of 200–500bp by sonication for 35 cycles (30 s ON/30 s OFF) using a Bioruptor Twin (Diagenode). After centrifugation for 10 min at 20,000 g, 4°C, the sonicate was diluted 10 fold in ChIP buffer (16.7 mM Tris pH8 1.2 mM EDTA, 167 mM NaCl, 1.1% Triton X-100) supplemented with protease inhibitors, pre-cleared for 2 hr at 4°C with 65 μl of Dynabeads M-280 sheep anti-rabbit IgG (Thermo Fisher Scientific 11203D, RRID:AB_2783009) and a 1% input chromatin sample was set aside. The rest of the sample was incubated overnight at 4°C with 5 μg of rabbit anti-HA antibody (Abcam ab9110, RRID:AB_307019). Immunocomplexes were washed once with each of low salt buffer (20 mM Tris pH8, 150 mM NaCl, 1% Triton X- 100, 0.1% SDS, 2 mM EDTA), high salt buffer (20 mM Tris pH8, 500 mM NaCl, 1% Triton X-100, 0.1% SDS, 2 mM EDTA), LiCl buffer (10 mM Tris pH8, 0.25 M LiCl, 1% NP-40, 1% sodium deoxycholate, 1 mM EDTA) and TE buffer (10 mM Tris pH8, 1 mM EDTA) and eluted from the beads in 100 mM NaHCO3, 1% SDS for 30 min at 65°C with shaking. Both input and ChIP samples were reverse crosslinked overnight at 65°C in the presence of 200 mM NaCl, and proteins were digested for 90 min at 45°C by addition of proteinase K (0.3 mg/ml final concentration). DNA was purified using the MinElute Reaction Cleanup Kit (QIAGEN, 28204), and quantified using the Qubit dsDNA HS Assay Kit (Q32851) and a Qubit 2.0 Fluorometer (Thermo Fisher Scientific).

ChIP against DMC1 was performed from Zcwpw1−/− testes using the published method by Khil et al., 2012 with some modifications listed here. Chromatin shearing was carried out in 20 mM Tris-HCl pH8, 2 mM EDTA, 0.1% SDS using a Bioruptor Pico sonicator (Diagenode) for 4 cycles of 15 s ON/45 s OFF. ChIP was performed in 10 mM Tris-HCl pH8, 1 mM EDTA, 0.1% Sodium Deoxycholate, 1% Triton X-100, 500 mM NaCl using 5 μg of mouse anti-DMC1 2H12/4 (Novus NB100-2617, RRID:AB_2245859) pre-bound to 50 μl of Dynabeads M-280 sheep anti-mouse IgG (Thermo Fisher Scientific 11202D, RRID:AB_772210).

Sequencing

Request a detailed protocol

ZCWPW1 ChIP and input libraries from transfected cells were prepared by the Oxford Genomics Centre at the Wellcome Centre for Human Genetics (Oxford, UK) using the Apollo Prep System (Wafergen, PrepX ILMN 32i, 96 sample kit) and standard Illumina multiplexing adapters following the manufacturer’s protocol up to pre-PCR amplification, and sequenced on a HiSeq 4000 platform (75bp paired end reads, 48 million reads/sample). DMC1 ChIP libraries from Zcwpw1−/− testes were prepared and sequenced as described previously (Davies et al., 2016) on an Illumina HiSeq2500 platform (Rapid Run, 51bp paired end reads, 110 million reads/sample).

Read mapping

Request a detailed protocol

For the HEK293T experiments, reads were mapped to either hg38 (NCBI’s GCA_000001405.15_GRCh38_no_alt_plus_hs38d1_analysis_set.fna.gz) using bwa mem (version 0.7.17-r1188) (Li, 2013). Duplicates were removed using picard’s markDuplicates (version 2.20.4-SNAPSHOT). Unmapped, mate unmapped, non primary alignment, failing platform, and low MAPQ reads were removed using samtools with parameters ‘-q 30 F 3852 f 2 u’ (version 1.9 (using htslib 1.9)). Other unmapped reads and secondary alignments were removed using samtools fixmate. Fragment position bed files were created using bedtools bamtobed (v2.28.0).

For the DMC1 mouse experimental data, we processed the data following the algorithm provided by Khil et al., 2012 to map the reads to the mouse mm10 reference genome (Lunter and Goodson, 2011), and obtain type I reads.

Peak calling

Request a detailed protocol

We called DMC1 peaks, as described previously (Davies et al., 2016). For the HEK293T experiments, peaks for ZCWPW1, PRDM9, H3K4me3, and H3K36me3 were called using a peak caller previously described (Davies et al., 2016) and available at https://github.com/MyersGroup/PeakCaller (archived at https://doi.org/10.5281/zenodo.3783600). Single base peaks were called with parameters pthresh 10−6 and peakminsep 250 on 22 autosomes. For each experiment, we used available IP replicates, and sequenced input DNA to estimate background, in calling peaks.

Most of our analyses refer to these peaks, except in Figure 5A,B which plot the change in ZCWPW1 occupancy following PRDM9 transfection, for either the chimp or human PRDM9 alleles. To estimate ZCWPW1 levels with vs without PRDM9, we called ZCWPW1 peaks in HEK293T cells also transfected with PRDM9 as before, but now replacing the ‘input’ lane with the IP data for ZCWPW1 in cells not transfected with PRDM9. Figure 5—figure supplement 6 also uses this measure, of ZCWPW1 recruitment attributable to PRDM9 binding, to predict DMC1 levels, while Figure 5—figure supplement 2 (right panel) checks that changes in ZCWPW1 occupancy occur largely at PRDM9-bound sites, and Figure 5—figure supplement 1 shows the correlation of PRDM9 binding strength with the change in ZCWPW1 enrichment.

The peak calling algorithm allows for calculation of enrichment and p-values at arbitrary pre-specified locations or windows in the genome, we refer to these as ‘force called’.

Note that the enrichment values from peak calling represent coverage due to signal, after subtracting coverage due to background, relative to background coverage. Hence a value less than one does not imply depletion — any value greater than 0 is enriched. For more details see Altemose et al., 2017.

Enrichment profiles for HEK293T experiments

Request a detailed protocol

Peaks were filtered if: center within 2.5kb of PRDM9-independent H3K4me3 (promoters), input coverage ≤5, in top five by likelihood, greater than 99.9 percentile input coverage. For human PRDM9 peaks the PRDM9 motif position (Altemose et al., 2017) was inferred using the getmotifs function in MotifFinder within the 300bp region around the peak (with parameters alpha = 0.2, maxits = 10, seed = 42, stranded_prior = T), and peaks were recentered and stranded at these locations. Peaks within 4kb of one another are removed to avoid double counting. Mean coverage was calculated with bwtool 1.0 using the aggregate command with parameters ‘-fill = 0 -firstbase’ at a width of +/− 2kb. Each profile was normalised by the total coverage of all of the fragments and the normalised sample profile was divided by the relevant normalised comparison/control dataset to calculate enrichment. Random profiles were created using bedtools random with 10,000 locations, with seed 72346, and were also filtered such that none were within 4kb of one another.

DSB profiles

Request a detailed protocol

Mapped type one reads were filtered by removing non-canonical chromosomes (with underscores in names), sex chromosomes, and mitochondrial reads. Bedgraphs were converted to bigWig format using UCSC bedGraphToBigWig v4. Profiles were created using bwtool 1.0 using the aggregate command with parameters 5000:2000 and -fill = 0. B6 WT hotspot locations were filtered to remove non-autosomal chromosomes, retain only B6 allele hotspots, and remove hotspots with PRDM9-independent H3K4me3. Prdm9−/− hotspots were filtered to remove the X chromosome. Motif centered coordinates were used for WT hotspots where the motif positions were inferred using the MotifFinder software and PRDM9 position weight matrices from Altemose et al., 2017. To normalise background signal, the mean signal between −5000 and −3000 was subtracted from each strand-mouse combination. Mean coverage was normalised by the sum of coverage over each strand-mouse combination across both WT and KO locations.

Mapping of alu CpGs

Request a detailed protocol

Alu locations were downloaded from UCSC tables and filtered for Alu repeats with a width between 250 and 350bp. DNA sequence was extracted from the genome (Ensembl 95 h38 primary assembly) at these locations using the bedtools getfasta command and CpG dinucleotides were counted using ‘stringr::str_locate_all’ command in R.

CpG methylation

Request a detailed protocol

Data is from Libertini et al., 2015 (GSE51867), non CpG methylated sites were removed, and the two strands were summed. High copy number (>1.5) CpGs were removed.

DMC1 prediction using ZCWPW1 and PRDM9 binding strength

Request a detailed protocol

PRDM9-dependent ZCWPW1 enrichment was force-called at the human PRDM9 peaks. Human testis DMC1 sites are from Pratto et al., 2014 (GSE59836) and were subsetted to ‘A or B’ autosomal peaks with a width less than 3kb. These peaks were then trimmed to a maximum of 800bp before overlapping with PRDM9 peaks to create a binary target variable. Regions with input coverage ≤5 in either predictor were removed in addition to outliers with input >200 and/or ZCWPW1 enrichment >10. A binomial generalised linear model with a logit link (logistic regression) was fitted using the ‘glm’ function from the ‘stats’ package in R (3.6.0). Chromosomes 1, 3, and 5 were used as the test data with the remaining autosomes being the training data.

Heatmaps

Request a detailed protocol

Regions were extracted from bigWig using bwtool matrix-fill = 0 -decimals = 1 -tiled-averages=5, and width parameters 2000. Coverage was normalised by total coverage (scaled by 1010). A pseudocount of 1 was added to both input and sample, and the sample was then normalised by the input for each region. Values outside the quantile range 0.01–0.99 were thresholded. The profile plots are created by taking the ratio of the mean coverage of the sample and input separately. Ordering of the regions was determined by the mean coverage of a 200bp window centered on the peak center.

Software

Computational analysis was performed using R (R Development Core Team, 2018) and snakemake (Köster and Rahmann, 2012). BigWig aggregations by bwtool (Pohl and Beato, 2014), BED manipulation with BEDtools (Quinlan and Hall, 2010), and FASTA processing with seqkit (Shen et al., 2016). We also used BigWig tools (Kent et al., 2010), IGV (Thorvaldsdóttir et al., 2013), and GNU parallel (Tange, 2018). Plots were created using the ggplot2 package (Wickham, 2016) and extensions ggforce (Pedersen, 2016), RColorBrewer (Neuwirth, 2014), viridis (Garnier, 2018), drawProteins (Brennan, 2018), and cowplot (Wilke, 2018). In addition, the following R packages were used: data.table (Dowle and Srinivasan, 2019), mgcv (Wood, 2011), PRROC (Grau et al., 2015), sangerseqR (Hill et al., 2014), and ComplexHeatmap (Gu et al., 2016).

External datasets summary

Request a detailed protocol

DMC1 SSDS ChIP-seq in human testis is from Pratto et al., 2014 (GSE59836). DMC1 SSDS ChIP-seq (Brick et al., 2012) (GSE35498), H3K4me3 ChIP-seq (Davies et al., 2016) (GSE73833), and SPO11 oligo-seq (Lange et al., 2016) (GSE84689) are in WT B6 mouse testis. PRDM9 (h/c-V5 tagged), H3K4me3, and H3K36me3 ChIP-seq in HEK293T cells is from Altemose et al., 2017 (GSE99407). Whole-genome bisulfite sequencing (BS-seq) of HEK293 cells is from Libertini et al., 2015 (GSE51867).

References

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
    Localization of two mammalian cyclin dependent kinases during mammalian meiosis
    1. T Ashley
    2. D Walpita
    3. DG de Rooij
    (2001)
    Journal of Cell Science 114:685–693.
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
    data.table: Extension of `data.frame`
    1. M Dowle
    2. A Srinivasan
    (2019)
    data.table: Extension of `data.frame`.
  26. 26
    viridis: Default Color Maps from “matplotlib
    1. S Garnier
    (2018)
    viridis: Default Color Maps from “matplotlib.
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
  35. 35
  36. 36
  37. 37
  38. 38
  39. 39
  40. 40
  41. 41
  42. 42
  43. 43
  44. 44
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49
  50. 50
  51. 51
  52. 52
  53. 53
  54. 54
  55. 55
  56. 56
  57. 57
  58. 58
  59. 59
  60. 60
  61. 61
  62. 62
  63. 63
  64. 64
  65. 65
    The time course and chromosomal localization of recombination-related proteins at meiosis in the mouse are compatible with models that can resolve the early DNA-DNA interactions without reciprocal recombination
    1. PB Moens
    2. NK Kolas
    3. M Tarsounas
    4. E Marcon
    5. PE Cohen
    6. B Spyropoulos
    (2002)
    Journal of Cell Science 115:1611–1622.
  66. 66
  67. 67
  68. 68
  69. 69
  70. 70
    RColorBrewer: ColorBrewer Palettes
    1. E Neuwirth
    (2014)
    RColorBrewer: ColorBrewer Palettes.
  71. 71
  72. 72
  73. 73
    ggforce: Accelerating “ggplot2
    1. TL Pedersen
    (2016)
    ggforce: Accelerating “ggplot2.
  74. 74
    A drying-down technique for the spreading of mammalian meiocytes from the male and female germline
    1. AH Peters
    2. AW Plug
    3. MJ van Vugt
    4. P de Boer
    (1997)
    Chromosome Research : An International Journal on the Molecular, Supramolecular and Evolutionary Aspects of Chromosome Biology 5:66–68.
    https://doi.org/10.1023/a:1018445520117
  75. 75
  76. 76
  77. 77
  78. 78
  79. 79
  80. 80
  81. 81
  82. 82
    R: A Language and Environment for Statistical Computing
    1. R Development Core Team
    (2018)
    R Foundation for Statistical Computing, Vienna, Austria.
  83. 83
  84. 84
  85. 85
  86. 86
  87. 87
  88. 88
  89. 89
  90. 90
  91. 91
  92. 92
    GNU Parallel
    1. O Tange
    (2018)
    GNU Parallel.
  93. 93
  94. 94
  95. 95
  96. 96
  97. 97
  98. 98
  99. 99
    cowplot: Streamlined Plot Theme and Plot Annotations for “ggplot2.
    1. CO Wilke
    (2018)
    cowplot: Streamlined Plot Theme and Plot Annotations for “ggplot2..
  100. 100
  101. 101
  102. 102

Decision letter

  1. Bernard de Massy
    Reviewing Editor; CNRS UM, France
  2. Jessica K Tyler
    Senior Editor; Weill Cornell Medicine, United States
  3. Bernard de Massy
    Reviewer; CNRS UM, France

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

The identification of Zcwpw1 is an important step in understanding the regulation of meiotic DSB repair, and the analysis performed both in mice and in cells provide novel information and complementary to those of the two other papers published on this same topic.

Decision letter after peer review:

Thank you for submitting your article "ZCWPW1 is recruited to recombination hotspots by PRDM9, and is essential for meiotic double strand break repair" for consideration by eLife. Your article has been reviewed by three peer reviewers, including Bernard de Massy as the Reviewing Editor and Reviewer #1, and the evaluation has been overseen by Jessica Tyler as the Senior Editor.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

The authors characterise a new protein, ZCWPW1, that is co-expressed with PRDM9 in mouse, a protein that has been shown to specify meiotic SPO11-DSB hotspot locations via the zinc-finger domain, and histone-trimethylation activity. The authors demonstrate the parallel phylogeny between ZCWPW1 and PRDM9. The authors analyse ZCWPW1 KO mice and show DSB repair and synapsis defects, with a proper DSB activity and localization. By assaying ZCWPW1 chromatin binding in HEK cells, the authors also show that ZCWPW1 has binding affinity for CpG rich DNA.

This study provides novel and interesting information about meiotic DSB control in mice. Most of the conclusions are convincing, however several interpretations required additional experimental support and the presentation of the data requires revision and clarification.

Importantly, as the authors have submitted their manuscript at the same time as the ones from Mahgoub et al. and Huang et al. , the results from these two papers have to be integrated in the Discussion to provide a more coherent view or to point out discrepancies if any — i.e. what is the meaning of the ZCWPW1 enrichment detected in HEK as compared to the signal detected in spermatocytes? What are the potential implications on the role of ZCWPW1 on gene and TE expression? What is the dynamic of RPA2 in ZCWPW1 KO?

Essential revisions:

1) Show the specificity of the antibody for ZCWPW1 vs ZCWPW2.

2) Revise cytology of ZCWPW1 KO: solve the discrepancy about DMC1 vs RPA2, and refer to other manuscripts for complementary data.

3) Revise interpretation about DSB repair phenotype: Conclusion should be toned down. ZCWPW1 may also affect DSB formation.

4) Clarify several points in data presentation (in particular HEK cells data, genomic data and figures). It is unfortunate that many figures are poorly prepared, difficult to understand and thus requiring a lot of time to read and evaluate.

5) Consider presenting DSB analysis by ChIP-seq DMC1 after the cytology results.

Other points:

1) General comment about mouse data presentation: for all data, provide age, number of mice and number of nuclei counted (complete the transparent reporting form accordingly).

2) General comment about NGS data: for all figures, make sure to label properly x and y axes with explanation of units and scales, explain color labels and how curves were calculated, explained which regions are being analysed.

3) Abstract: "persistent DMC1 foci, particularly those more strongly bound by PRDM9": This sentence is a very incorrect shortcut: persistence of DMC1 at specific hotspots has not been evaluated by cytology but by ChIP, and no direct correlation with strength of PRDM9 binding is presented.

4) Introduction final paragraph. There are many proteins that are positioned in response to DSBs generated at PRDM9-binding sites (e.g. MRE11, RAD51, DMC1 etc). Please revise.

5) Results first paragraph. It would help to have a clearer description of these data and how they relate to the graph in Figure 1B. What do the values on the Y-axis indicate?

6) Results paragraph two-four are hard to follow without small summary tables and/or simple bar plots or possibly pie-charts. Pointing at the large Table S1 was not helpful.

It is not fully clear if the six independent losses are birds + crocodiles, three groups of fishes, canids, amphibians; please clarify. Then, what is defined as the first four losses? Birds + crocodiles and fishes? What are these mutations? Are they predicted to be loss of function? The succession of comparisons of groups of species is confusing. What is the group discussed and mentioned as: "they mainly retain ZCWPW1"?. What is the need to add this statement at this point of the presentation of the results? In addition, it is not correct that this pattern is expected. The a priori expectation depends on whether ZCWPW1 is the only reader of those marks, for instance, it could also depend on whether this reading activity is essential.

7) Figure 2: The pachytene nucleus should be defined: if it is neither early nor late, it should be mid-?

Add antigen names on figure. Provide separate channels in supplement.

8) Paragraph two subsection “Localisation of ZCWPW1 in meiosis and details of asynapsis in infertile male Zcwpw1-/- mice” (with Jung reference). This is confusingly placed. I suggest opening this paragraph with this published information, then describe the new observations within the manuscript.

9) Figure 3—figure supplement 1B is referred to before Figure 3—figure supplement 1A.

10) The text mentions lack of MLH1 foci (Figure 3—figure supplement 2), yet no mention of these observations within this figure or elsewhere can be found in the manuscript. Please amend/clarify.

11) Please add relevant references to the PRDM9 phenotype/observations.

12) It is not the best definition of RAD51 to qualify it as a ssDNA binding protein. It is above all a strand exchange protein that binds to both ss and dsDNA (and three-strand DNA as well).

13) “ In contrast, there was no difference in the levels of RPA2 (Figure 4—figure supplement 2).”. This statement is inappropriate without clear evidence (and in contradiction with Li et al.). Please quantify RPA2 foci as graphs in a similar way to the DMC1 and RAD51 data. If unclear, the authors could also test MEIOB foci.

14) Figure 3E. I assume it is implicit that each value is for one mouse, and data are in triplicate or duplicate for Prdm9? Please clarify.

15) Figure 4. How were the categories Early Z, Z (MidZ?), Late Z defined?

16) Please aid reader by clarifying what asynapsed chromosomes are (HORMAD2-positive). Question: Sometimes unsynapsed is used, sometimes asynapsed is used. If there is a distinction, please clarify for the reader what is meant, or instead, please consolidate and choose a single term for clarity.

17) Please define FISH.

18) Figure S9. Why do most (but not all) groups overlap? (e.g. Chr 2, 2-3, 3-4, 4-5, 6-7, 7-8, 8.9, 16-17, 17-18, 18-19). Are observations sometimes placed in two bins, and sometimes not? This seems problematic and unnecessary to make the point.

19) Comparisons of synapsis between ZCWPW1 KO and Prdm9 KO. An alternative hypothesis is for the distinct phenotype is a difference in the genetic background. ZCWPW1 is B6N and Prdm9 is B6x129 hybrid probably with a certain number (?) of backcrosses. Subtle differences in checkpoint, cell cycle progression could affect the phenotype. Were mice compared at identical ages? Maybe complementary information could be available from other Prdm9 deficient alleles. What about sterile hybrids?

20) Supplementary table 1 should have a legend, explain columns titles (for instance Y276, 341, 357, Max sum…)

21) Figure 2—figure supplement 2. Revise: This spread must be from WT not from ZCWPW1-/-.

22) ZCWPW1 staining is only at one of the two ends and it looks as if it is the centromeric end; please provide data to solve this ambiguity by labeling centromeres for instance. Also, as mentioned above, prove that this signal is not ZCWPW2.

23) Figure 3—figure supplement 1A. Where is the wild type? Wild type nuclei that have been treated in parallel and image acquisition with identical parameters must be shown.

24) Data not shown: please show data referred to in lanes 855 (western blots), 891 (IP with pre-immune serum).

25) Supplementary Figure 9: This analysis requires more documentation: representative images of the nuclei with immuno and FISH signals, and table of raw data (values for axis lengths and synapsis). Why are 18 and 19, or 18 and 17 not distinguished?

26) Figure 5—figure supplement 3. Clarify the legend title. It is unclear where the conclusion comes from. One assumes blue hexagons (actually misleading: the shape is invisible, just look like spheres) are Prdm9 bound. I also understand the X axis is data from Zcpwp1 transfected alone. So, the binding at these non-Prdm9 site does not change when Prdm9 is co-transfected. Please clarify.

27) Are ZCWPW1, ZCWPW2 and Prdm9 expressed in HEK cells? If yes, at which level and can it have consequence on interpretation of the data? Is the HA-ZCWPW1 fully functional?

28) Figure 5A. How is "Enrichment" calculated? It is mentioned with vs without PRDM9. Is it a ratio? Is this a log scale? For ZCWPW1 around human PRDM9 sites, it appears that the enrichment spans much more than just the peak region where PRDM9 binds. It would be extremely helpful to be able to compare these analyses to a similar aggregation of PRDM9 and H3K4me3 around the same sites. Y axis: homogenize: 2 or 2.0.

Why is the enrichment not down to 1 with Prdm9 transfection?

Provide western blots to assay whether ZCWPW1 levels are identical in different analysis (with normalization for transfection efficiency).

Analyze quantitatively the correlation between ZCWPW1 and Prdm9 binding.

29) Please clarify what is meant by "sequence-specific binding", and please describe how the data presented in Figure 5C and Figure 5—figure supplement 3 support this interpretation.

30) “ However, upon transfection, 91% (98%) of the strongest 10,132 (3,016) peaks are PRDM9 binding sites (cf. 7% expected overlap with randomised peaks), so PRDM9 is able to strongly reprogram ZCWPW1 binding, suggesting that non-PRDM9 peaks are bound more weakly.”. What does the value in parentheses (3,016) mean? This sentence as written is hard to follow. Please simplify.

31) Figure 5—figure supplement 5. Add labels to A and B panels.

Panel A: It seems that ZCWPW1 correlates poorly with cPRDM9. Why is this? What about transfecting chimp ZCWPW1? Do the same correlation plot with human proteins. Could there be some species-specific interactions, despite the nearly identical sequences of human and chimp proteins? Please provide the plot (heatmap/average plot) of H3K4me3 at cPRDM9 peaks.

Panel B: What is the difference with Figure 5B?

32) Paragraph three of subsection “ZCWPW1 is recruited to PRDM9 binding sites in an allele-specific manner”. Please expand this section so that the average reader can appreciate and understand the analysis that was performed. Figure 5—figure supplement 6 is hard to follow, please clarify. Where is the logistic regression methods?

33) You mean DMC1 sites in meiosis? Clarify to avoid confusion between transfected cells and meiotic cells.

34) “…although we show (see below) that ZCWPW1 is not directly involved in DSB positioning, this might suggest involvement of similar features to those it recognizes, in recruiting DSBs”. Is hard to follow. It will help if the last "it" was clarified to read "ZCWPW1".

35) Figure 6—figure supplement 1. Please label and explain the dashed horizontal lines.

36) The casual reader may be misled into thinking that the authors have tested for recruitment of ZCWPW1 to sites that have "dual" modifications. Unless I am mistaken, these are independent correlations to each mark, with no evidence that the histone marks exist coincidentally at the same locus in the same cell. Please clarify this section accordingly. Also, a direct interaction of PRDM9 with ZCWPW1 would be enough to explain specificity.

37) Figure 6—figure supplement 2. Please clarify the X-axis. Is the "+0.01" an offset added to each value in order to present on log scale? Please describe what analysis was performed and why.

38) Figure 6. Please clarify the analysis presented here. Do the 100bp windows cover the whole genome without other filtering? The legend is unintelligible, so it is very hard to interpret what is being presented. Untransfected means what: no PRDM9 and no ZCWPW1? "No untransfected H3K4me3" does not make sense. Both plots (red point and blue points) show ~2.5 fold increase from left to right (from 2 to 5 vs from 0.8 to 2). As such, is there really a difference if "predictive" power of the H3K4me3 at the red vs blue sites? Plotting the data on a Y-log scale (since enrichment is a ratio after all isn't it?) might help demonstrate this lack of difference more clearly. i.e. the blue curve only looks different to the red because it has been flattened by the linear Y scale.

Why is the distribution of H3K4me3 shifted to lower values in the red subset?

What does input fragment coverage mean? Do we agree that Input is chromatin before IP?

39) Figure S16. Please label panels. It was unclear what these data were attempting to test/demonstrate. Perhaps some manual annotations (and clarified legend and axes labels) would help? Explaining what the subsets are (and how they were obtained) is critical.

40) Figure 7C is referred to before 7B.

41) Figure 7A. The shaded labelling of 5' and 3' strands is unclear. Please replace with a clearer presentation style (different colours or dashed perhaps?). 7B presumably merges both strands, please clarify. Please homogenize X-axis labelling in A and B to be clearer (i.e. PRDM9 motif centre). Please use same units in both graphs (bp or kb).

42) Please show some raw DMC1 SSDS data from some example loci in order to demonstrate the lack of change ±ZCWPW1.

43) Paragraph two of subsection “DSBs occur at their normal locations in Zcwpw1-/- mice but show altered DMC1 persistence”. Broader? Based on what? Why is the DMC1 signal so irregular in the ZCWPW1 KO? What are the number of reads (Type I and Type II) in WT and KO?

44) Figure 7—figure supplement 1. This figure is very hard to understand. What does Heat>0 mean in legend? What does "Fraction of hotter hotspots" on Y-axis mean?

45) Figure 7C. Like many other figures, it is unclear what is being presented here. The legend has no explanation for the different colours (blue and black curves for example). The red/orange colour scale itself means what? What are the units? Why is the H3K4me3 relevant here?

In addition, it should be made to the reader that interpretation form this figure requires one additional assumption: i.e. that SPO11 oligos levels are not affected in ZCWPW1 KO. The authors should make clear that only WT data of SPO11 oligos is available.

46) Figure 7—figure supplement 2. Please label panels A, B, C. Please revise the text in the legend. It is very hard to follow. Adding manual annotations to the plots would help greatly – something as simple as: Hotspots relative to PRDM9 binding motif: upstream (red), downstream (black), central (green).

47) The Lange data would be more accurately described as SPO11-oligo seq.

48) Please refer to Figure 7C.

49) Figure 7—figure supplement 3. Please add panel labels. Please add units for axes. Are they log-scale? Please add wild type presented in exactly the same way as the two mutants for comparison.

It seems that the DMC1 data is interpreted as indication that ZCWPW1 promotes repair at hotter hotspots preferentially (the Abstract says: persistent DMC1 foci, particularly those more strongly bound by PRDM9, see also comment about this sentence). This conclusion is not convincing, ZCWPW1 KO may have the same effect as Hop2 KO on DMC1 persistence (although the conclusion that the pattern is similar is an oversimplification, Y axis value are very different in the two strains): as a result of DSB repair defect, DMC1 level is proportional to DSB activity.

In addition, the authors cannot exclude that ZCWPW1 has a role on DSB activity and that DSB activity is increased in ZCWPW1 KO: it could be a direct effect: ZCWPW1 could compete with other readers, and/or indirect: synapsis defect allowing for a prolonged window for DSB activity with effects that may not be uniform across hotspots.

Accordingly, the statement, that DSB formation and positioning are completely unaffected is too strong.

50) Figure 7—figure supplement 4. Please present the other correlation plots that are mentioned in the text (DMC1 and SPO11) for comparison. Y-axis is a ratio so should be in log scale (or Log2 values)?

51) Please explain where the 9-fold effect is shown in Figure 7—figure supplement 4. Please clarify how the authors concluded this implies a "very strong effect" and a very strong effect of what? Of slower repair? Of H3K4me3? Of DMC1?

52) It is worth noting that the correspondence of ZCWPW1 occupancy with Alu repeats is also readily apparent in Figure 5C. And easier to visually interpret than the data presented in Figure 8.

53) Figure 5—figure supplement 2. Like many others, this figure needs revision. The labelling and presentation are poor and confusing: What utH3K4me3 means? In the middle, cells are cotransfected with ZCWPW1 and PRDM9 but peaks in red are ZCWPW1 alone? How is this possible?

54) Subsection “ZCWPW1 binds CpG dinucleotides” paragraph one. Please explain in the main text how the Figures 8A and Figure 5—figure supplement 2 support this statement.

55) Is it possible that the apparent Alu and CpG enrichment of ZCWPW1 by ChIP is due to lack of antibody specificity?

56) Use of "in vitro" is misleading. Instead ex vivo could be used, or in somatic cells.

57) Figure 8. Is this CpG binding still detected in the presence of Prdm9?

58) Extract the curve of ZCWPW1 enrichment at non-Prdm9 sites in the presence of Prdm9, and test whether it follows the same distribution as in Figure 8A.

59) Discussion paragraph three: present differences with the cited paper and explain.

60) Discussion paragraph four. To quote HEK293 cells as a model system is very misleading. It is in no way a model system for meiosis. The authors have to indicate that this cell nuclei have a completely different environment as compared to meiotic cells. Therefore, extrapolation of any data to meiosis is only hypothetical, even it is coherent and follows some expectations.

61) ZCWPW1 does not directly recognize Prdm9 binding sites: A Prdm9 binding site is a DNA sequence.

62) Since the authors cite Spruce et al. paper they cannot write that ZCWPW1 is the first protein that recognizes regions bound by PRDM9.

63) Graphical abstract: how could the authors write that the functional importance of the marks has remained unknown? They have been shown to be essential for meiotic DSB formation.

https://doi.org/10.7554/eLife.53392.sa1

Author response

Summary:

[…]

Importantly, as the authors have submitted their manuscript at the same time as the ones from Mahgoub et al. and Huang et al. , the results from these two papers have to be integrated in the Discussion to provide a more coherent view or to point out discrepancies if any — i.e.: what is the meaning of the ZCWPW1 enrichment detected in HEK as compared to the signal detected in spermatocytes? What are the potential implications on the role of ZCWPW1 on gene and TE expression? What is the dynamic of RPA2 in ZCWPW1 KO?

We thank the reviewers for their kind words, and we believe we have thoroughly addressed these substantive points and the specific points in our revised manuscript. All changes are tracked in the revised manuscript file.

As requested, we have now added a number of references to the related work on ZCWPW1 by Maghoub et al. and Huang et al. in the Results and Discussion sections, highlighting relevant evidence to integrate these studies into a coherent view:

i) “While we cannot rule out a direct interaction between ZCWPW1 and PRDM9, others have shown that dual modified peptides have a greater affinity to ZCWPW1 than peptides carrying the single modifications, supporting that the marks themselves are responsible for ZCWPW1’s recruitment (Mahgoub et al., 2020).”

ii) “Like Mahgoub and colleagues in a similar analysis (Mahgoub et al., 2020), we find co-evolution of ZCWPW1 with PRDM9, consistent with proteins functioning in the same pathway.”

iii) “In parallel with this work, two other studies of ZCWPW1 (Mahgoub et al., 2020; Huang et al., 2020) have reached complementary conclusions, and together with our own results allow some clear conclusions to be drawn.”

iv) “We find that in co-transfected HEK293T cells ZCWPW1 is strongly recruited to PRDM9-bound sites, in an allele-specific manner. Mahgoub et al., 2020, also report PRDM9 allele-specific recruitment of ZCWPW1, but in hybrid mice carrying Prdm9Dom2 and Prdm9Cast alleles which encode proteins with distinct zinc finger array DNA binding specificities. Together with ZCWPW1 binding measurements from Huang et al., 2020, these results imply that across PRDM9 alleles and species – humans, mice, and likely chimpanzees – PRDM9 binding alone is sufficient to strongly recruit ZCWPW1 without requiring DSB formation (which does not occur in our system) or meiosis-specific co-factors (which would not be expected to be present in our cellular system).”

v) “In HEK293T cells we observed ZCWPW1 recruitment depends on the levels of H3K4me3 and H3K36me3 that PRDM9 deposits, while Huang et al., 2020, reported that in mice with a non-functional CW domain no ZCWPW1 binding peaks were observed, suggesting that H3K4me3 recognition at least might be necessary for the recruitment of ZCWPW1 at hotspots in vivo.”

vi) “Indeed in cells with PRDM9, the strongest ZCWPW1 binding sites are almost all PRDM9-bound sites. Similarly in mice, >85% of ZCWPW1 peaks overlap both marks concomitantly (Huang et al., 2020; Mahgoub et al., 2020). Mahgoub et al., 2020, additionally reported that in vitro binding affinity of recombinant mouse ZCWPW1 is at least 5 fold greater for dual-modified H3K4me3/K36me3 peptides compared to peptides carrying the single modifications.”

vii) “Although it was suggested that ZCWPW1 might recruit the DSB machinery to PRDM9 binding sites (M. Li et al., 2019; Spruce et al., 2020), we and Mahgoub et al., 2020, (using distinct assays) find that DSB positioning is completely unaffected by loss of ZCWPW1.”

viii) “Although more modest, Huang et al., 2020, did also report enrichment of ZCWPW1 at B1 elements in mice, the closest equivalent of human Alus. Unlike Mahgoub et al., 2020, Huang et al., 2020, also observed overlap of ZCWPW1 peaks with CpG islands (CGI) more frequently than expected. Given (i) they and Mahgoub et al., 2020, detect (Mahgoub et al., 2020) a much lower number of peaks (14,688 and 4,487 respectively) than we do in our HEK293T system (800,000), and (ii) we only find Alu enrichment in weaker ZCWPW1 peaks (of our top 10,000 peaks, >90% would instead be PRDM9-bound sites), these results are not conflicting, and instead might depend on detection/behaviour of weaker peaks, and/or presence of PRDM9 in ZCWPW1-containing cells.”

In their published manuscript Huang et al., 2020, conclude from their RNA-seq analysis of Zcwpw1-/- mice “These results strongly suggest that ZCWPW1 may not affect the transcription level of many genes sharing promoter overlap with ZCWPW1”, and so we have not commented on the implications of ZCWPW1 as a transcription co-factor as previously speculated by Huang et al. in their preprint.

See point 13 regarding the analysis of RPA2 dynamics in the Zcwpw1 KO mouse.

Essential revisions:

1) Show the specificity of the antibody for ZCWPW1 vs ZCWPW2.

We agree it is essential to exclude any potential cross-reactivity with ZCWPW2 of our polyclonal antibody. In Figure 2—figure supplement 1, we now include evidence (previously mentioned only as “data not shown”) that: (i) the pre-immune serum does not detect any signal in mid-zygotene or early pachytene cells strongly labelled by the ZCWPW1 antiserum by immunofluorescence (IF) in B6 mouse testis nuclear spreads (panel B); (ii) there is no detection of ZCWPW2 by IP and WB at the 38KDa molecular weight range predicted for mouse ZCWPW2, only a single band is observed at the expected molecular weight for ZCWPW1 (panel C). We also (iii) use HEK293T cells transfected with either FLAG-tagged mouse ZCWPW1 or ZCWPW2 to show detection of ZCWPW1 by IP and WB (panel D) and IF (panel E), but not ZCWPW2, by the antiserum (and no detection by the pre-immune serum), even at the high expression levels likely in this system.

Because we think it is likely ZCWPW2 (whose expression timing is similar to ZCWPW1) is still expressed in the Zcwpw1 KO mouse, the complete loss of signal detected in IF staining by the ZCWPW1 antiserum in this mouse also suggests strong specificity to ZCWPW1.

The validation data are now referred to, in the relevant section, as: “To investigate the role of ZCWPW1 during meiosis in vivo, we produced and validated the specificity of an antibody against the full-length recombinant mouse protein (Figure 2—figure supplement 1), and … meiotic recombination.”

2) Revise cytology of ZCWPW1 KO: solve the discrepancy about DMC1 vs RPA2, and refer to other manuscripts for complementary data.

3) Revise interpretation about DSB repair phenotype: Conclusion should be toned down. ZCWPW1 may also affect DSB formation

4) Clarify several points in data presentation (in particular HEK cells data, genomic data and figures). It is unfortunate that many figures are poorly prepared, difficult to understand and thus requiring a lot of time to read and evaluate.

5) Consider presenting DSB analysis by ChIP-seq DMC1 after the cytology results.

We address these points where they arise below.

Other points:

1) General comment about mouse data presentation: for all data, provide age, number of mice and number of nuclei counted (complete the transparent reporting form accordingly).

We have revised the legend of all relevant figures (Figures 2-4, 7 and Supplements) to include this information which is also provided for each individual mouse used in the study in Figure 3—source data 1 and 2. For instance, Figure 2 legend now reads “Nuclear spreads from 9-10 weeks old B6 mice were immunostained with antibodies against ZCWPW1… These images are representative of the results obtained in 3 mice.” We have completed the transparent reporting form accordingly.

2) General comment about NGS data: for all figures, make sure to label properly x and y axes with explanation of units and scales, explain color labels and how curves were calculated, explained which regions are being analysed.

Thanks. The legends for Figures 1, 5 (and supplement 3 and 6), 6 (and supplements 1 and 2), 7 (and supplements 1, 2, 3 and 4), and 8 have been revised to include this information on the graphs and/or in the relevant legend. For example, we revised the legend of Figure 7C to add the sentences “Black dashed line is y=x for reference. SPO11 and DMC1 enrichment have been scaled by dividing by the mean autosomal enrichment. Large blue and dark red points show mean DMC1 signal binned into groups containing equal numbers of hotspots by WT SPO11 signal (vertical lines: corresponding 95% CIs), for X and autosomal data respectively.” in addition to revising the axis titles to specify “Scaled” and adding a legend for the (now simplified) colour scheme of the points.

3) Abstract: "persistent DMC1 foci, particularly those more strongly bound by PRDM9": This sentence is a very incorrect shortcut: persistence of DMC1 at specific hotspots has not been evaluated by cytology but by ChIP, and no direct correlation with strength of PRDM9 binding is presented.

We agree there are some caveats needed for this sentence, so due to lack of space in the Abstract we have simply deleted the phrase “particularly those more strongly bound by PRDM9”. We do think there is some evidence for this statement in the sense that we observe persistent foci by cytology but fairly normal break formation (both in the cytology and normal DMC1 ChIP-seq hotspot locations), implying an overall increase in DMC1 persistence. Then we observe a similar behaviour in DMC1 data for the X-chromosome and autosomes in the KO mouse, with a relative increase in those autosomal hotspots which are more active in DSB formation – which we would argue generally reflects stronger binding, and noting this phenomenon happens widely across all hotspots fairly similarly. It is then difficult not to conclude both (i) persistence increases overall for all hotspots, and (ii) persistent breaks occur particularly at the hotter hotspots (which will also contribute most of the breaks overall, i.e. cytologically visible signals of recombination). Indeed, many weaker hotspots become less detectable in our DMC1 ChIP-seq data in the KO mouse, so common, persistent breaks cannot be occurring at these hotspots even if the explanation is, they don’t break so often.

4) Introduction final paragraph. There are many proteins that are positioned in response to DSBs generated at PRDM9-binding sites (e.g. MRE11, RAD51, DMC1 etc). Please revise.

As suggested, we have edited this to read “directly positioned by PRDM9’s dual histone marks”.

5) Results first paragraph. It would help to have a clearer description of these data and how they relate to the graph in Figure 1B. What do the values on the Y-axis indicate?

We have added a note to the legend to properly define the Y-axis: “Jensen-Shanon divergence normalised to mean of 0 and standard deviation of 1 is shown on the y-axis (a measure of sequence conservation, see (Capra and Singh, 2007; Johansson and Toh, 2010))”.

6) Results paragraph two-four are hard to follow without small summary tables and/or simple bar plots or possibly pie-charts. Pointing at the large Table S1 was not helpful.

It is not fully clear if the six independent losses are birds + crocodiles, three groups of fishes, canids, amphibians; please clarify. Then, what is defined as the first four losses? Birds + crocodiles and fishes? What are these mutations? Are they predicted to be loss of function? The succession of comparisons of groups of species is confusing. What is the group discussed and mentioned as: "they mainly retain ZCWPW1 "?. What is the need to add this statement at this point of the presentation of the results? In addition, it is not correct that this pattern is expected. The a priori expectation depends on whether ZCWPW1 is the only reader of those marks, for instance, it could also depend on whether this reading activity is essential.

This section has now been reworded throughout, to clarify the points raised. We have also added two new panels to Figure 1, showing bar plots summarising overlap between different groups of species and (i) whether PRDM9 is observed, when ZCWPW1 is present (panel C), and (ii) whether ZCWPW1 is observed, when PRDM9 is present and shows various independent evolutionary events (panel D). For the last point above, we wanted to relate the co-evolution data to the various domains of the two proteins and their predicted functions, but agree that the a priori expectation was clumsily worded, and it is now stated differently.

We now state that the likely “complete” losses (of PRDM9’s ability to catalyse H3K4me3/K36me3 deposition) are in the three groups of fishes (3 distinct losses), bird/crocodiles, amphibians, canids. We also now more clearly plot comparable loss events of other PRDM9 domains. We also show the loss events on the bar plot (the evolutionary history of PRDM9 is complex, involves many duplications and complete or partial losses of the protein, and there is considerable uncertainty for individual species, but we have attempted to summarize accurately the known events and overall patterns).

7) Figure 2: The pachytene nucleus should be defined: if it is neither early nor late, it should be mid-?

Add antigen names on figure. Provide separate channels in supplement.

Figure 2 was revised as suggested, now indicating “mid-Leptotene”, “mid-Zygotene”, “mid-Pachytene” and mid-Diplotene stages, and the target proteins have been added directly to the panels. Separate channels are now provided for this staining in revised Figure 3—figure supplement 1B, and this is referred to in Figure 2 legend as “Images for the individual channels are provided in Figure 3—figure supplement 1”.

8) Paragraph two subsection “Localisation of ZCWPW1 in meiosis and details of asynapsis in infertile male Zcwpw1-/- mice” (with Jung reference). This is confusingly placed. I suggest opening this paragraph with this published information, then describe the new observations within the manuscript.

The paragraph was revised according to the suggested structure.

9) Figure 3—figure supplement 1B is referred to before Figure 3—figure supplement 1A.

Thanks. We have revised this Figure (now Figure 3—figure supplement 1), swapping panels A and B to respect the chronological order of the data described in the text.

10) The text mentions lack of MLH1 foci (Figure 3—figure supplement 2), yet no mention of these observations within this figure or elsewhere can be found in the manuscript. Please amend/clarify.

We amended Figure 3—figure supplement 2, adding the missing data in panel B which shows lack of MLH1 foci in pseudo-pachytene cells in the Zcwpw1 KO mouse.

11) Please add relevant references to the PRDM9 phenotype/observations.

We have added a reference to Hayashi et al., 2005.

12) It is not the best definition of RAD51 to qualify it as a ssDNA binding protein. It is above all a strand exchange protein that binds to both ss and dsDNA (and three-strand DNA as well).

We agree and have corrected the description to “a strand exchange protein”.

13) “In contrast, there was no difference in the levels of RPA2 (Figure 4—figure supplement 2).”. This statement is inappropriate without clear evidence (and in contradiction with Li et al.). Please quantify RPA2 foci as graphs in a similar way to the DMC1 and RAD51 data. If unclear, the authors could also test MEIOB foci.

We have now quantified RPA2 foci in the Zcwpw1 KO mouse across the meiotic stages of prophase 1. The counts are presented as a graph in Figure 4—figure supplement 2B with the raw data in Figure 4—source data 1. In the light of the quantification results, the previous statement has been revised to: “Like DMC1, RPA2 levels were also significantly elevated in the Zcwpw1−/− mouse from zygotene onwards (Figure 4—figure supplement 2).”

14) Figure 3E. I assume it is implicit that each value is for one mouse, and data are in triplicate or duplicate for Prdm9? Please clarify.

Correct. We have added the clarification “each datapoint represents one mouse, each with n≥49 cells analysed.” to the legend. The corresponding raw data are presented in Figure 3—source data 3.

15) Figure 4. How were the categories Early Z, Z (MidZ?), Late Z defined?

Staging and sub- (early, mid and late) staging of spermatocytes during prophase I was carried out according to defined staining patterns against SYCP1 and SYCP3, and morphological criteria previously published (Gaysinskaya et al., 2014). We added more information about cell staging in section “Immunostaining of spermatocytes” of the Materials and methods.

16) Please aid reader by clarifying what asynapsed chromosomes are (HORMAD2-positive). Question: Sometimes unsynapsed is used, sometimes asynapsed is used. If there is a distinction, please clarify for the reader what is meant, or instead, please consolidate and choose a single term for clarity.

For consistency, the word “asynapsed” is now used across the manuscript. As suggested, we have also revised this sentence to read: “Indeed, we observe late unrepaired DMC1 foci mainly on HORMAD2-positive asynapsed chromosomes (Figure 4—figure supplement 3).”

17) Please define FISH.

We have expanded the FISH acronym (Fluorescent In-Situ Hybridisation).

18) Figure S9. Why do most (but not all) groups overlap? (e.g. Chr 2, 2-3, 3-4, 4-5, 6-7, 7-8, 8.9, 16-17, 17-18, 18-19). Are observations sometimes placed in two bins, and sometimes not? This seems problematic and unnecessary to make the point.

A single observation (one chromosome, in one cell) was only placed in one bin. However, because of the difficulty in measuring chromosome length with accuracy via microscopy, these bins overlapped several possible chromosomes. For simplicity, we have now decided to remove this analysis altogether.

19) Comparisons of synapsis between ZCWPW1 KO and Prdm9 KO. An alternative hypothesis is for the distinct phenotype is a difference in the genetic background. ZCWPW1 is B6N and Prdm9 is B6x129 hybrid probably with a certain number (?) of backcrosses. Subtle differences in checkpoint, cell cycle progression could affect the phenotype. Were mice compared at identical ages? Maybe complementary information could be available from other Prdm9 deficient alleles. What about sterile hybrids?

We agree differences in genetic background are a potential source of phenotypic variation, and indeed could explain partly or fully the apparent differences in meiotic defects observed between the two mutants. We have revised the paragraph to include this alternative. The mice from the 2 mutant lines were reasonably age-matched (within 2-3 weeks, see detail in Figure 3—source data 1). For the last point we agree, though time has not permitted us so far to perform a paired comparison of the meiotic phenotype of the Zcwpw1 KO mouse with that of other Prdm9 deficient strains (KRAB or SSXRD deletion mutants for instance), or mice with different combinations of alleles such as (PWDxB6) F1 hybrid infertile males where high levels of asymmetric PRDM9 binding are present. We hope to do so in future.

20) Supplementary table 1 should have a legend, explain columns titles (for instance Y276, 341, 357, Max sum…)

We did have some explanation of this in “Notes” within this table:

“Data are reproduced from Baker et al., 2017; for each species, we calculated the maximum domain architecture and sum of matching amino acids in the SET domain among all identified PRDM9 orthologues in that species (columns D and E)”. However, we have now reworded for increased clarity, and also added additional references to this note for each column separately, alongside retitling “Notes” to “Notes on table columns” in now Figure 1— source data 1 and 2.

21) Figure 2—figure supplement 2. Revise: This spread must be from WT not from ZCWPW1-/-.

It is indeed a wild-type sample. Note that this panel has been removed from the revised figure.

22) ZCWPW1 staining is only at one of the two ends and it looks as if it is the centromeric end; please provide data to solve this ambiguity by labeling centromeres for instance. Also, as mentioned above, prove that this signal is not ZCWPW2.

We do observe ZCWPW1 foci at both ends of the synaptonemal complex in many chromosomes, but not all. To quantify whether foci are at one, both ends or neither, in the revisions we quantified the number of chromosomes showing ZCWPW1 foci at either, or both, ends by FISH using telomeric and centromeric probes (revised Figure 2—figure supplement 2; raw data in Figure 2—source data 1). The results show that in some stages (late pachytene-diplotene) most chromosomes exhibit ZCWPW1 foci at both ends, while if foci (at least those observable via microscopy) are at one end only, this is mainly the centromeric end. We have revised the text now reading “Using Fluorescent In-Situ Hybridisation (FISH) to label centromeric and telomeric (distal and proximal to centromeres) regions of chromosomes, we established that these discrete foci of ZCWPW1 are consistently positioned at the ends of the synaptonemal complex, with the majority at both ends simultaneously, lying close to telomeres and centromeres themselves (Figure 2—figure supplement 2).” The specificity of the antibody for ZCWPW1 versus ZCWPW2 is now demonstrated in Figure 2—figure supplement 1.

23) SupFig4A. Where is the wild type? Wild type nuclei that have been treated in parallel and image acquisition with identical parameters must be shown.

Agreed; this dataset is now included in Figure 3—figure supplement 1B.

24) Data not shown: please show data referred to in lanes 855 (western blots), 891 (IP with pre-immune serum).

These data are now provided in Figure 2—figure supplement 1.

25) Supplementary figure 9: This analysis requires more documentation: representative images of the nuclei with immuno and FISH signals, and table of raw data (values for axis lengths and synapsis). Why are 18 and 19, or 18 and 17 not distinguished?

Chromosome ID was assigned based on the best match by predicted (Mb) length with a +/- 10% margin (calculated based on an inch to Mb ratio for chromosome 18 as a reference in the same cell), excluding chromosomes outside these boundaries. Where bins overlapped two possible chromosomes, these were grouped. For consistency across the analysis and the data presentation, we had retained (in the original Supplementary Figure 9) the grouping of chr17 or chr19 with chr18 (their closest chromosome in size) despite, in these instances, a potential ID as chr18 being excluded based on its identification by FISH. Due to the inaccuracy in length measurements, and likely high error rate in chromosome assignment, we have now removed this figure – see response to comment 18.

26) Figure 5—figure supplement 3. Clarify the legend title. It is unclear where the conclusion comes from. One assumes blue hexagons (actually misleading: the shape is invisible, just look like spheres) are Prdm9 bound.

We have added the clarification to the caption of what is now Figure 5—figure supplement 3: “(there are no peaks with high co-transfected enrichment [y-axis] when the untransfected enrichment [x-axis] is close to 0)”. Sorry about the figure resolution. The hexagons are clearer in the publication-quality image.

I also understand the X axis is data from Zcpwp1 transfected alone. So, the binding at these non-Prdm9 site does not change when Prdm9 is co-transfected. Please clarify.

The reviewer’s assumptions are correct. The red shows locations with no evidence of PRDM9 binding, and indeed these sites do not change much if at all when PRDM9 is co-transfected. The blue (locations with evidence of PRDM9 binding) shows many peaks above the diagonal hence showing that these sites do change when PRDM9 is co-transfected. i.e. some of the peaks from the ZCWPW1 transfected alone experiment do not change when PRDM9 is co-transfected (the ones which do not occur in locations with evidence of PRDM9 binding); but some of the peaks from the ZCWPW1 transfected alone experiment do change (those which have evidence of PRDM9 binding).

It is probably also worth pointing out that because ZCWPW1 binds at some level at many genomic sites, the sites of ZCWPW1 binding when transfected alone include a mixture of sites that are not bound by PRDM9, and sites that are bound by PRDM9, and both types are shown on this plot, not just sites not bound by PRDM9.

27) Are ZCWPW1, ZCWPW2 and Prdm9 expressed in HEK cells? If yes, at which level and can it have consequence on interpretation of the data?

All three genes have a highly restricted expression to reproductive glands in humans with comparatively low to no expression in kidney where the HEK293T cell line originates from (data from gtexportal PRDM9 https://www.gtexportal.org/home/gene/PRDM9; ZCWPW1 see Figure 1—figure supplement 1; ZCWPW2 https://www.gtexportal.org/home/gene/ZCWPW2). To ensure this is the case in our cell line stock, we have carried out a quantitative analysis of transcript expression by RT-PCR. The results presented in Figure 5—source data 1 show extremely low (PRDM9 and ZCWPW1) to no (ZCWPW2) detectable gene expression in untransfected cells. Such levels are highly unlikely to confound the interpretation of the data in cells transfected with ZCWPW1 and/or PRDM9 where the expression is on average 800-1000 or 75,000-84,000 fold higher, respectively. Similarly, in a dataset from 10X genomics (https://support.10xgenomics.com/single-cell-gene-expression/datasets/1.1.0/293t) these three genes have very low, almost undetectable, expression in HEK293T cells (mean transcript count per cell of 0.036, 0.013, and 0.001, where median total transcript count per cell was 14,338). We added a statement in the relevant section of the results : “To test this, we co-transfected HEK293T cells … and then performed ChIP-seq against the ZCWPW1 tag (the endogenous levels of these proteins in HEK293T cells, and of the closely related family member ZCWPW2, are extremely low so unlikely to confound the analysis – see Figure 5—source data 1).”

Author response image 1

Is the HA-ZCWPW1 fully functional?

We are uncertain of all the functional properties of this protein but note that we validated the predicted function of full length (HA-tagged) human ZCWPW1 as a dual histone methyl reader by demonstrating its preferential recruitment to PRDM9-bounds sites enriched in both histone marks. These results are in line with the expected behaviour. The C-terminal 9 amino acid small HA tag is located away from the internal CW and PWWP functional domains, and so is we hope unlikely to strongly confound the normal function of the protein. We note that others have specifically demonstrated binding of recombinant (His-tagged) mouse ZCWPW1 (residues 1-440 encompassing the CW and PWWP domains) to dual-modified H3K4me3/K36me3 peptides with a greater (over 5 fold) affinity compared to peptides carrying the single modifications in histone peptide assays in vitro (Mahgoub et al., 2020). We have added a reference to this work in the revised Discussion.

28) Figure 5A. How is "Enrichment" calculated? It is mentioned with vs without PRDM9. Is it a ratio? Is this a log scale?

We have added a clearer definition in the Materials and methods section which now reads “Each profile was normalised by the total coverage of all of the fragments and the normalised sample profile was divided by the relevant normalised comparison / control dataset to calculate enrichment.” It is a ratio and is presented on a log scale, which we have added a notification of in the legend as we agree with the point, that the axis breaks don’t make this clear: “Y-axis is log10 scale (y-axis labels remaining in linear space).”

For ZCWPW1 around human PRDM9 sites, it appears that the enrichment spans much more than just the peak region where PRDM9 binds. It would be extremely helpful to be able to compare these analyses to a similar aggregation of PRDM9 and H3K4me3 around the same sites. Y axis: homogenize: 2 or 2.0.

The profile plots (and heatmaps) of PRDM9 and H3K4me3 are shown in panel B of this figure. Indeed, they show a distance scale broadly consistent with binding of ZCWPW1 to H3K4me3 (and H3K36me3), but wider than binding by PRDM9 itself. (Given that the association between ZCWPW1 and chromatin likely reflects this indirect binding/recruitment, it is not clear in advance exactly what enrichment distance scale we expect.) Scale on Y axis is now consistent between panels.

Why is the enrichment not down to 1 with Prdm9 transfection?

The profile in Author response image 2, in orange, shows random sites and hits 1, so this is not simply an issue in normalisation. Instead, the higher background enrichment reflects higher regional (but not fully local) background effects, for example driven by some broad regions of the genome being more generally accessible, and other low level peaks also falling in the region surrounding one peak, on average. To demonstrate this, if we instead take locations randomly between 15kb and 10kb downstream of each peak (blue line) and calculate the profile there and overlay, we find that the locally higher background is consistent with the value the peak profile reaches.

Author response image 2
Black: Profile of peaks.

Blue: profile of locations 15 to 10kb downstream. Red: profile of globally (whole genome) random locations.

Provide western blots to assay whether ZCWPW1 levels are identical in different analysis (with normalization for transfection efficiency).

Unfortunately, we no longer have any material from these experiments to carry out the requested western blot quantification. At the time of the experiment, we verified similar transfection efficiency/levels of protein expression across the various experimental samples by immunofluorescence staining (on duplicate transfections). We have revised Figure 5—figure supplement 4 to include representative images of the transfected samples used in the study.

Analyze quantitatively the correlation between ZCWPW1 and Prdm9 binding.

We have calculated the correlation between PRDM9 enrichment and ZCWPW1 enrichment at sites of PRDM9 binding, and refer to this analysis as “ZCWPW1 shows a strong enrichment at human PRDM9 binding sites with higher enrichment at sites with higher PRDM9 enrichment (Figure 5A, B, Pearson's correlation = 0.43 – Figure 5—figure supplement 1)” where Figure 5—figure supplement 1 is now a scatter plot of PRDM9 enrichment against ZCWPW1 enrichment showing a linear effect.

29) Please clarify what is meant by "sequence-specific binding", and please describe how the data presented in Figure 5C and Figure 5—figure supplement 3 support this interpretation.

We have removed “sequence-specific” as we do not intend to imply a sequence motif per se which we suspect is what the reviewer is concerned about. We have also replaced the Figure 5—figure supplement 3 citation with Figure 5—figure supplement 2 to support the following new statement “Notably, even without PRDM9 we observed many ZCWPW1 binding peaks across the genome, some coinciding with PRDM9 binding sites (Figure 5C and Figure 5—figure supplement 2)”, and have moved the S10 (now Figure 5 —figure supplement 3) citation further down now reading “Hence, PRDM9 is able to strongly reprogram ZCWPW1 binding (Figure 5C, see also Figure 5—figure supplement 3), suggesting that peaks in cells without PRDM9 transfection are bound more weakly.”

30) “However, upon transfection, 91% (98%) of the strongest 10,132 (3,016) peaks are PRDM9 binding sites (cf. 7% expected overlap with randomised peaks), so PRDM9 is able to strongly reprogram ZCWPW1 binding, suggesting that non-PRDM9 peaks are bound more weakly.”. What does the value in parentheses (3,016) mean? This sentence as written is hard to follow. Please simplify.

Apologies. The parenthetical value corresponds to the 98% figure also in parentheses earlier in the sentence. We have re-written this sentence to clarify: “However, upon co-transfection with PRDM9, 92% of the strongest 10,000 ZCWPW1 peaks are at PRDM9-bound sites (98% of the strongest 3,000, vs 7% expected overlap with randomised peaks, and only 47% overlap for the strongest 3,000 ZCWPW1 peaks in cells without PRDM9 transfection).”

31) Figure 5—figure supplement 5. Add labels to A and B panels.

Added, thanks.

Panel A: It seems that ZCWPW1 correlates poorly with cPRDM9. Why is this?

Overall, ZCWPW1 in cells co-transfected with cPRDM9 still exhibits a large and clear enrichment at cPRDM9 sites, so we certainly do see allele-specific co-localisation. However, there are quantitative differences, with ZCWPW1 showing weaker enrichment for cPRDM9 than hPRDM9, in co-transfected cells. Although we do not know with certainty the reasons for this effect (which might operate differently in true meiotic cells with a different background chromatin environment), we have noted that cPRDM9 binds at sites with different broad-scale properties to hPRDM9, including more heterochromatic sites. Also, for example, cPRDM9 binding is not enriched at promoter regions. These properties might impact PRDM9’s deposition of histone modifications, and through this or another mechanism slightly weaken ZCWPW1’s ability to recognize bound sites.

What about transfecting chimp ZCWPW1?

We have not been able to explore chimp ZCWPW1 in this publication for reasons of time, but we agree it would indeed be an interesting avenue of research for the future.

Do the same correlation plot with human proteins.

We believe the reviewer means the heatmaps for human PRDM9 locations, which are provided as Figure 5B (see also comment below about panel B in reply to 31f).

Could there be some species-specific interactions, despite the nearly identical sequences of human and chimp proteins?

Although this is a good question, we don’t yet have data to answer it. Our working model is that the mechanism of action of ZCWPW1 is through the histone methylation marks deposited by PRDM9 rather than by directly binding PRDM9 itself – for example allowing flexibility in populations where PRDM9 is polymorphic. In this case the interaction might be expected to be quite similar for both human and chimp PRDM9. It is though possible that PRDM9 may directly bind ZCWPW1, or ZCWPW1 might have additional specificities (e.g. the CpG recognition) and/or recruit different proteins depending on human and chimp versions, but we have not explored this hypothesis.

Please provide the plot (heatmap/average plot) of H3K4me3 at cPRDM9 peaks.

H3K4me3 (without any transfection) at cPRDM9 peaks is shown in column 7 “Untransfected H3K4me3”. We do not have data available for H3K4me3 when co-transfected with cPRDM9, so have not been able to present this.

Panel B: What is the difference with Figure 5B?

Panel B shows the top 25% of all ZCWPW1 peaks (when co-transfected with human PRDM9). Figure 5B is focussed on PRDM9, so shows the top 25% of human PRDM9 peaks. There will be overlaps between these locations, but they are different in general.

32) Paragraph three of subsection “ZCWPW1 is recruited to PRDM9 binding sites in an allele-specific manner”. Please expand this section so that the average reader can appreciate and understand the analysis that was performed. Figure 5—figure supplement 6 is hard to follow, please clarify.

We have modified the section to clarify, it now reads: “Notably, the strength of ZCWPW1 binding in HEK293T cells at human PRDM9 binding sites provides a better predictor of human meiotic DMC1 binding status (a proxy for DSB formation) than does human PRDM9 binding strength itself (Figure 5—figure supplement 6). We show (see below) that ZCWPW1 is not directly involved in DSB positioning, this result might therefore instead suggest that similar features to those ZCWPW1 recognises are involved in recruiting DSB formation machinery.”

We also introduced additional details in the legend to interpret Figure 5—figure supplement 6 which now reads: “At identified human PRDM9 binding sites, we identified those at which male recombination hotspots occur, defined by the presence/absence of an overlapping human DMC1 peak, and fit a linear model to predict this hotspot status based on PRDM9 binding strength (PRDM9 Only), ZCWPW1 enrichment (with human PRDM9 vs without, referring to enrichment of ZCWPW1 cotransfected with PRDM9 relative to ZCWPW1 transfected alone), or both (see Materials and methods “DMC1 prediction”). […] Estimated PRDM9-dependent ZCWPW1 enrichment (green) provides a better predictor than does PRDM9 binding strength (blue).”

Where is the logistic regression methods?

The corresponding methods were under the title of “DMC1 prediction”. It was hard to find these, so we have added the term “logistic regression” in the Materials and methods section to make this section easier to find by searching as well as quoting the method section title in the caption of Figure 5—figure supplement 6 “(see Materials and methods “DMC1 prediction”)”.

33) You mean DMC1 sites in meiosis? Clarify to avoid confusion between transfected cells and meiotic cells.

Correct, we have added the clarification “human meiotic” here.

34) “…although we show (see below) that ZCWPW1 is not directly involved in DSB positioning, this might suggest involvement of similar features to those it recognizes, in recruiting DSBs”. Is hard to follow. It will help if the last "it" was clarified to read "ZCWPW1".

Thanks. Changed as suggested.

35) Figure 6—figure supplement 1. Please label and explain the dashed horizontal lines.

We have added the phrase “Dotted lines show overall means for each colour.” to the legend.

36) The casual reader may be misled into thinking that the authors have tested for recruitment of ZCWPW1 to sites that have "dual" modifications. Unless I am mistaken, these are independent correlations to each mark, with no evidence that the histone marks exist coincidentally at the same locus in the same cell. Please clarify this section accordingly. Also, a direct interaction of PRDM9 with ZCWPW1 would be enough to explain specificity.

Agreed (in principle). We have added the caveat “tested separately, not necessarily “dual” marks coincident within an individual cell, although locations of these marks are rarely coincident except at hotspots (Powers et al., 2016)”.

37) Figure 6—figure supplement 2. Please clarify the X-axis. Is the "+0.01" an offset added to each value in order to present on log scale? Please describe what analysis was performed and why.

We have added clarification to the plot legend of this figure: “This is in some sense opposite (but complementary) to Figure 6—figure supplement 1 in which the subject of the axis is reversed.” and “0.01 has been added to the x-axis values in order to display enrichment estimates of zero on the log scale.”

38) Figure 6. Please clarify the analysis presented here. Do the 100bp windows cover the whole genome without other filtering?

Correct, the only filtering applied is that which is already stated in the caption “defined as indicated in the legend… with the additional constraint of requiring input fragment coverage >5 for ZCWPW1 and >15 for H3K4me3.” To eliminate and check for e.g. X-chromosome artifacts we have additionally re-run this analysis excluding the sex chromosomes, resulting in a near-identical plot.

The legend is unintelligible, so it is very hard to interpret what is being presented. Untransfected means what: no PRDM9 and no ZCWPW1?

Correct: untransfected means no transfection with either PRDM9 or ZCWPW1. We have changed this in the legend to “pre-existing”, see next response.

We have also added the following sentence to the caption “(where “p” is the p-value from peak-calling required for a window to be included in the subset)”.

"No untransfected H3K4me3" does not make sense.

We have changed this to “No pre-existing H3K4me3” and hope that this is clearer (it means no H3K4me3 is seen in untransfected cells). An alternative would be “No PRDM9-independent or ZCWPW1-independent H3K4me3”, we settled on the former and added an explanation to the legend: “Pre-existing H3K4me3 refers to H3K4me3 that is present without transfection (of either PRDM9 or ZCWPW1), which is mainly found at promoter regions.”

Both plots (red point and blue points) show ~2.5 fold increase from left to right (from 2 to 5 vs from 0.8 to 2). As such, is there really a difference if "predictive" power of the H3K4me3 at the red vs blue sites? Plotting the data on a Y-log scale (since enrichment is a ratio after all isn't it?) might help demonstrate this lack of difference more clearly. i.e. the blue curve only looks different to the red because it has been flattened by the linear Y scale.

It may be that there is a misunderstanding here: we are not claiming that there is a difference in the slope, i.e. the fold-impact of increasing H3K4me3 on relative ZCWPW1 presence. Instead, we are noting the absolute level of ZCWPW1 is much higher, for any given level of H3K4me3, at PRDM9 binding sites i.e. that the “intercept” is higher for PRDM9 bound regions. The parallel behaviour emphasizes that this property remains for any level of H3K4me3. The relevant variable is the presence of PRDM9 (perhaps mediated by H3K36me3 or another mark), and not H3K4me3 which is similarly present in both subsets. We have as suggested replaced the plot with a y-log version although as explained above our interpretation of the data is unchanged: the red and blue subsets are still almost completely separated by the black dotted line. We have added the following sentence to the legend which we hope will guide the reader on the interpretation of this plot “For any given level of H3K4me3 (x-axis), ZCWPW1 enrichment (y-axis) is higher at PRDM9 bound regions (red) than regions with pre-existing H3K4me3 (promoters, blue).”

Why is the distribution of H3K4me3 shifted to lower values in the red subset?

The blue subset is, in effect, promoter regions – which have very high H3K4me3 levels (plausibly close to ~100% of some histones being marked at the highest levels). The red subset has H3K4me3 only due to PRDM9 binding nearby, which is limited by both PRDM9 binding strength and also our achieved transfection efficiency, so in practice appears unable to achieve the same level.

What does input fragment coverage mean?

Fragment refers to the DNA fragment that was sequenced as inferred by the two ends of paired end sequencing, and coverage is the number of reads covering a given position.

Do we agree that Input is chromatin before IP?

Yes, Input refers to the starting material (crosslinked, sonicated chromatin from lysed cells) before IP.

39) Figure S16. Please label panels. It was unclear what these data were attempting to test/demonstrate. Perhaps some manual annotations (and clarified legend and axes labels) would help? Explaining what the subsets are (and how they were obtained) is critical.

For simplicity, we have just removed this faceted figure – as it made a similar point to Figure 6.

40) Figure 7C is referred to before 7B.

We have deleted this additional forward reference, as it is not essential.

41) Figure 7A. The shaded labelling of 5' and 3' strands is unclear. Please replace with a clearer presentation style (different colours or dashed perhaps?).

We have changed to use dotted lines as suggested.

7B presumably merges both strands, please clarify.

Yes – we have added “(Both strands combined)” to the legend of Figure 7B.

Please homogenize X-axis labelling in A and B to be clearer (i.e. PRDM9 motif centre).

Thanks – we have homogenized the axis labels to say “Position relative to PRDM9 motif (kb)”.

Please use same units in both graphs (bp or kb).

Done.

42) Please show some raw DMC1 SSDS data from some example loci in order to demonstrate the lack of change ±ZCWPW1.

In Author response image 3 are four representative regions among those with high coverage, showing identical general patterns. We believe it’s not clear what this figure adds in addition to Figure 7, because this figure shows the lack of change without the additional noise when looking at individual hotspots. However, we could add this as a supplement if the reviewer thinks it would be helpful having seen Author response image 3:

Author response image 3

43) Paragraph two of subsection “DSBs occur at their normal locations in Zcwpw1-/- mice but show altered DMC1 persistence”. Broader? Based on what?

We have removed the broader comment here, as this is discussed with more precision in the following paragraph “slightly wider […] (Figure 7B).”

Why is the DMC1 signal so irregular in the ZCWPW1 KO? What are the number of reads (Type I and Type II) in WT and KO?

As guessed by the reviewer the KO DMC1 signal has more variance, due to the lower number of both read types obtained (< ½), from a single infertile mouse:

44) Figure 7—figure supplement 1. This figure is very hard to understand. What does Heat>0 mean in legend?

As per the legend, heat refers to DMC1 signal, so Heat>0 therefore refers to hotspots with a DMC1 signal above 0. We have relabeled to “DMC1>0” for ease of understanding of this figure.

What does "Fraction of hotter hotspots" on Y-axis mean?

We have changed the axis titles to be clearer as follows. Y: “Fraction of WT hotspots falling in each KO DMC1 bucket for all hotspots with WT DMC1 ≥ a given x-axis value” and X: “WT hotspot DMC1 heat relative to strongest”. We also added the following explanation to the caption to guide the reader in how to read the plot: “Y-axis values at x = 0 show the fraction of all hotspots falling into the buckets shown in the inset colour legend. […] Therefore, almost all WT hotspots with activity >20% of the hottest hotspot are observed, and non-observed hotspots show only weak activity in WT, and so our power to detect them is expected to be reduced.”

What this figure shows is that although we do not detect all hotspots present in the WT mouse in the KO, we do detect many, and almost all the high-heat hotspots, suggesting those not observed are explained by being weaker and fewer reads meaning we lack power in the KO.

45) Figure 7C. Like many other figures, it is unclear what is being presented here. The legend has no explanation for the different colours (blue and black curves for example). The red/orange colour scale itself means what? What are the units? Why is the H3K4me3 relevant here?

Apologies: we have simplified this figure by removing H3K4me3 colouring, so the colours of the points representing the binned means are more easily pairable with the data points. We have also added a legend directly to the figure to label the autosomal vs X chromosome points. We have also added the following to the caption:

“Unlike WT mice, DMC1 signals in Zcwpw1-/- mice are approximately linearly associated with WT SPO11. […] Large blue and dark red points show mean DMC1 signal binned into groups containing equal numbers of hotspots by WT SPO11 signal (vertical lines: corresponding 95% CIs), for X and autosomal data respectively.”

In addition, it should be made to the reader that interpretation form this figure requires one additional assumption: i.e. that SPO11 oligos levels are not affected in ZCWPW1 KO. The authors should make clear that only WT data of SPO11 oligos is available.

We agree and have added the following clarification to the main text “assuming SPO11 levels are not greatly affected by ZCWPW1 KO”.

46) Figure 7—figure supplement 2. Please label panels A, B, C. Please revise the text in the legend. It is very hard to follow. Adding manual annotations to the plots would help greatly – something as simple as: Hotspots relative to PRDM9 binding motif: upstream (red), downstream (black), central (green).

We have added panel labels and reworded the legend including adding the suggested phrase to Figure 7—figure supplement 2.

47) The Lange data would be more accurately described as SPO11-oligo seq.

We have corrected this.

48) Please refer to Figure 7C.

We have added the reference.

49) Figure 7—figure supplement 3. Please add panel labels. Please add units for axes. Are they log-scale? Please add wild type presented in exactly the same way as the two mutants for comparison.

We have added panel labels, explained the scaling in the legend ("SPO11 and DMC1 enrichment have been scaled by dividing by the mean autosomal enrichment.”) and added a WT version as panel C to Figure 7—figure supplement 3.

It seems that the DMC1 data is interpreted as indication that ZCWPW1 promotes repair at hotter hotspots preferentially (the Abstract says: persistent DMC1 foci, particularly those more strongly bound by PRDM9, see also comment about this sentence). This conclusion is not convincing, ZCWPW1 KO may have the same effect as Hop2 KO on DMC1 persistence (although the conclusion that the pattern is similar is an oversimplification, Y axis value are very different in the two strains): as a result of DSB repair defect, DMC1 level is proportional to DSB activity.

We think the alternative hypothesis suggested by the reviewer “ZCWPW1 KO may have the same effect as Hop2 KO […] as a result of DSB repair defect, DMC1 level is proportional to DSB activity.” is exactly what we are proposing. It’s because in the absence of a DBS repair defect hotter hotspots are repaired faster (in a non-linear manner, Figure 7C) (i.e. normally there is a non-linear effect), so if interpreted as a delay or failure in repair increasing DMC1 signal (as observed via microscopy also), the difference/fold-effect of the KO is greater at the hottest hotspots. The y-axis scaling is arbitrary (based on number of ChIP-seq reads, for the previously published Hop2 KO data), and so does not impact this interpretation.

In addition, the authors cannot exclude that ZCWPW1 has a role on DSB activity and that DSB activity is increased in ZCWPW1 KO: it could be a direct effect: ZCWPW1 could compete with other readers, and/or indirect: synapsis defect allowing for a prolonged window for DSB activity with effects that may not be uniform across hotspots.

Accordingly, the statement, that DSB formation and positioning are completely unaffected is too strong.

We have removed “formation” so it currently reads “DSB positioning is completely unaffected”.

50) Figure 7—figure supplement 4. Please present the other correlation plots that are mentioned in the text (DMC1 and SPO11) for comparison. Y-axis is a ratio so should be in log scale (or Log2 values)?

We have added these new plots as additional panels to Figure 7—figure supplement 4. We have also used log scales for these plots (although the relationship appears no longer linear on this scale).

51) Please explain where the 9-fold effect is shown in Figure 7—figure supplement 4.

We have scaled each axis by dividing by the median, to make the 9 fold effect more obvious. Now the mean ratio starts at 1 on the left (y-axis) and increases to around 9 on the right.

Please clarify how the authors concluded this implies a "very strong effect" and a very strong effect of what? Of slower repair? Of H3K4me3? Of DMC1?

A “very strong effect” is based on the 9 fold change (almost an order of magnitude) which seems reasonably described as very strong.

We have added the clarification “of PRDM9’s “assistance” in aiding repair”, regarding the effect referred to.

52) It is worth noting that the correspondence of ZCWPW1 occupancy with Alu repeats is also readily apparent in Figure 5C. And easier to visually interpret than the data presented in Figure 8.

We thank the reviewer(s) for pointing this out, and we have additionally referenced 5C to aid the reader.

53) Figure 5—figure supplement 2. Like many others, this figure needs revision. The labelling and presentation are poor and confusing: What utH3K4me3 means?

We have modified Figure 5—figure supplement 2 legend by expanding all abbreviations rather than explaining them in the caption – using "Pre-existing” H3K4me3 instead of “untransfected”, and visually structuring the colour categories. We have also updated the caption to provide a more explicit description of what the panels and colours within show, now reading “For example dark green peaks are those which overlap with ZCWPW1 peaks when transfected alone, but not overlapping Human PRDM9 peaks, and not overlapping pre-existing H3K4me3 peaks but do overlap with Alu repeats. Pre-existing H3K4me3 refers to peaks found without PRDM9 or ZCWPW1 transfection.” and “In cells expressing ZCWPW1 in the presence (middle plot) but not in the absence (left plot) of PRDM9, the strongest peaks are dominated by PRDM9-bound sites marked by H3K4me3 (pink), while ZCWPW1 occupancy increases occur nearly exclusively at these sites, following co-transfection with PRDM9 (right plot).”

In the middle, cells are cotransfected with ZCWPW1 and PRDM9 but peaks in red are ZCWPW1 alone? How is this possible?

The graph shows “Proportion of ZCWPW1 peaks, ordered by enrichment of ZCWPW1 binding over input, overlapping various other marks”; only the light blue category represents peaks “Not [in] ZCWPW1 Alone” i.e. the co-transfected peaks are almost completely a subset of the “ZCWPW1 Alone” peaks (to rephrase the reviewers interpretation, the peaks in red are also present in “ZCWPW1 Alone”, see also Figure 5—figure supplement 3) resolving the impossibility.

54) Subsection “ZCWPW1 binds CpG dinucleotides” paragraph one. Please explain in the main text how the Figures 8A and Figure 5—figure supplement 2 support this statement.

We looked at this – the Y-axis of Figure 8A directly shows “Fraction of ZCWPW1 peaks that overlap Alu repeats” with values of greater than 75% for weaker ZCWPW1 peaks e.g. enrichment <1 (much greater than expected by chance, as stated in the following sentence). Similarly, the y-axis of Figure 5—figure supplement 2 shows the fraction of ZCWPW1 peaks overlapping various marks, including Alus (in dark green). So, we think this is clear, but would be happy to include a specific edit if preferred.

55) Is it possible that the apparent Alu and CpG enrichment of ZCWPW1 by ChIP is due to lack of antibody specificity?

The anti-HA antibody from Abcam (Ab9110) used for ChIP against (HA-tagged) ZCWPW1 is a highly specific and widely used ChIP-grade antibody. The negative control ChIP anti-HA in untransfected HEK293T cells for this experiment did not yield DNA in quantifiable amounts (below the range of detection of the Qubit dsDNA high sensitivity assay), therefore we believe ZCWPW1 does indeed (often weakly) bind Alus and CpGs.

56) Use of "in vitro" is misleading. Instead ex vivo could be used, or in somatic cells.

We have replaced this with “in HEK293T cells”. (We did not opt for “ex vivo” as this is preferred by some to only be used for experiments performed on primary tissue outside of the organism.)

57) Figure 8. Is this CpG binding still detected in the presence of Prdm9?

In these experiments, yes. However, we note that as they represent an average of many cells, with variable transfection efficiency, some cells have PRDM9 and some don’t. It may then be that in individual cells with PRDM9, ZCWPW1 is very strongly sequestered to PRDM9 sites away from non-hotspot CpG sites.

58) Extract the curve of ZCWPW1 enrichment at non-Prdm9 sites in the presence of Prdm9, and test whether it follows the same distribution as in Figure 8A.

In Author response image 4 we have calculated the curve both in the presence of PRDM9 (left column) and in absence (right column), as well as at all sites (top row) and at sites not bound by PRDM9 (bottom row) [as even in the absence of PRDM9 many ZCWPW1 binding sites overlap PRDM9 binding sites]. The curves all show effectively the same thing which is expected given the large overlap of site positions.

Author response image 4

59) Discussion paragraph three: present differences with the cited paper and explain.

In the revised manuscript, these differences are now discussed.

60) Discussion paragraph four. To quote HEK293 cells as a model system is very misleading. It is in no way a model system for meiosis. The authors have to indicate that this cell nuclei have a completely different environment as compared to meiotic cells. Therefore, extrapolation of any data to meiosis is only hypothetical, even it is coherent and follows some expectations.

We have revised this paragraph to give a clearer and more accurate description of this cellular system and its intended use. Given the mitotic nature of HEK293T cells, of course we never meant to claim these could be considered a model system of meiosis (as a process). However, they provide a suitable cellular system in which to study the direct DNA and histone binding properties of ectopically-expressed meiotic proteins, which are our focus in those experiments. The absence of a large number of meiotic proteins within these cells is consistent with the idea that recruitment of ZCWPW1 might require only PRDM9 and certainly rules out their necessity, and so in that sense is advantageous. In practice we do observe a strong overlap between human meiotic recombination hotspots and binding sites of both PRDM9, and (more strongly) sites where ZCWPW1 is present, attesting that though very different, mitotic cells can nevertheless give useful insights into molecular properties of these proteins.

61) ZCWPW1 does not directly recognize Prdm9 binding sites: A Prdm9 binding site is a DNA sequence.

We have changed this to “PRDM9-marked sites”, making reference to the histone marks more explicit.

62) Since the authors cite Spruce et al. paper they cannot write that ZCWPW1 is the first protein that recognizes regions bound by PRDM9.

We are making a distinct claim here. HELLS is required for stable PRDM9 binding and recognises PRDM9 itself, independently of whether PRDM9 is bound to DNA. Once PRDM9 has bound however, ZCWPW1 is the first protein which recognises these PRDM9 marked sites. To clarify this we have reworded to “ZCWPW1 likely directly recognises sites marked by the dual histone marks deposited by the PRDM9-HELLS complex (Spruce et al., 2020), the first such protein identified.”

63) Graphical abstract: how could the authors write that the functional importance of the marks has remained unknown? They have been shown to be essential for meiotic DSB formation.

Apologies for this slip – we have changed this to state “but the mechanistic role of these marks”, because the protein(s) recognizing the marks have remained unclear.

https://doi.org/10.7554/eLife.53392.sa2

Article and author information

Author details

  1. Daniel Wells

    1. The Wellcome Centre for Human Genetics, Roosevelt Drive, University of Oxford, Oxford, United Kingdom
    2. Department of Statistics, University of Oxford, Oxford, United Kingdom
    Contribution
    Conceptualization, Resources, Data curation, Software, Formal analysis, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing - original draft, Writing - review and editing
    Contributed equally with
    Emmanuelle Bitoun
    For correspondence
    wells@stats.ox.ac.uk
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-2007-8978
  2. Emmanuelle Bitoun

    1. The Wellcome Centre for Human Genetics, Roosevelt Drive, University of Oxford, Oxford, United Kingdom
    2. Department of Statistics, University of Oxford, Oxford, United Kingdom
    Contribution
    Conceptualization, Resources, Data curation, Formal analysis, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing - original draft, Project administration, Writing - review and editing
    Contributed equally with
    Daniel Wells
    For correspondence
    ebitoun@well.ox.ac.uk
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-3439-2113
  3. Daniela Moralli

    The Wellcome Centre for Human Genetics, Roosevelt Drive, University of Oxford, Oxford, United Kingdom
    Contribution
    Resources, Data curation, Formal analysis, Validation, Investigation, Visualization, Methodology
    Competing interests
    No competing interests declared
  4. Gang Zhang

    The Wellcome Centre for Human Genetics, Roosevelt Drive, University of Oxford, Oxford, United Kingdom
    Contribution
    Resources, Data curation, Investigation, Methodology
    Competing interests
    No competing interests declared
  5. Anjali Hinch

    The Wellcome Centre for Human Genetics, Roosevelt Drive, University of Oxford, Oxford, United Kingdom
    Contribution
    Resources, Data curation, Software, Formal analysis, Validation, Investigation, Visualization, Methodology
    Competing interests
    No competing interests declared
  6. Julia Jankowska

    The Wellcome Centre for Human Genetics, Roosevelt Drive, University of Oxford, Oxford, United Kingdom
    Contribution
    Resources, Data curation, Formal analysis, Validation, Investigation
    Competing interests
    No competing interests declared
  7. Peter Donnelly

    1. The Wellcome Centre for Human Genetics, Roosevelt Drive, University of Oxford, Oxford, United Kingdom
    2. Department of Statistics, University of Oxford, Oxford, United Kingdom
    Contribution
    Resources, Supervision
    Competing interests
    No competing interests declared
  8. Catherine Green

    The Wellcome Centre for Human Genetics, Roosevelt Drive, University of Oxford, Oxford, United Kingdom
    Contribution
    Resources, Supervision
    Competing interests
    No competing interests declared
  9. Simon R Myers

    1. The Wellcome Centre for Human Genetics, Roosevelt Drive, University of Oxford, Oxford, United Kingdom
    2. Department of Statistics, University of Oxford, Oxford, United Kingdom
    Contribution
    Conceptualization, Resources, Data curation, Software, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing - original draft, Project administration, Writing - review and editing
    For correspondence
    myers@stats.ox.ac.uk
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-2585-9626

Funding

Wellcome (098387/Z/12/Z)

  • Simon R Myers

Wellcome (212284/Z/18/Z)

  • Simon R Myers

Wellcome (109109/Z/15/Z)

  • Daniel Wells

Wellcome (095552/Z/11/Z)

  • Peter Donnelly

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We thank the High-Throughput Genomics Group at the Wellcome Centre for Human Genetics (funded by Wellcome Trust grant 203141/Z/16/Z) for the generation of sequencing data. We also thank Dr. Benjamin Bishop and Prof. Christian Siebold in the Division of Structural Biology (Wellcome Centre for Human Genetics, Oxford) for their help with the purification of recombinant mouse ZCWPW1 (size exclusion) towards antibody production. Funding: This work was supported by the Wellcome Trust through Investigator Awards to S.R.M. (098387/Z/12/Z and 212284/Z/18/Z), a PhD Studentship to D.W. (109109/Z/15/Z), a Senior Investigator Award to P.D (095552/Z/11/Z), and core support grants to the Wellcome Centre for Human Genetics (20314/Z/16/Z and 090532/Z/09/Z).

Ethics

Animal experimentation: All animal experiments received local ethical review approval from the University of Oxford Animal Welfare and Ethical Review Body (Clinical Medicine board) and were carried out in accordance with the UK Home Office Animals (Scientific Procedures) Act 1986. The specific protocols used were authorised by the UK Home Office under Project Licence PPL 3003437.

Senior Editor

  1. Jessica K Tyler, Weill Cornell Medicine, United States

Reviewing Editor

  1. Bernard de Massy, CNRS UM, France

Reviewer

  1. Bernard de Massy, CNRS UM, France

Publication history

  1. Received: November 7, 2019
  2. Accepted: July 31, 2020
  3. Accepted Manuscript published: August 3, 2020 (version 1)
  4. Version of Record published: September 16, 2020 (version 2)

Copyright

© 2020, Wells et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 538
    Page views
  • 143
    Downloads
  • 1
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)