In eukaryotic cells, most DNA is wrapped around an octamer of histone proteins to form nucleosomes, the basic unit of chromatin. Chromatin functions as a physical barrier to DNA, and regulating access to DNA is an essential aspect to normal cellular function. Numerous proteins have evolved to carry out this function; these include pioneer factors, chromatin remodelers, and epigenetic modifiers (Iwafuchi-Doi and Zaret 2014;Clapier et al. 2017). Pioneer factors initiate formation of accessible chromatin by engaging partial recognition motifs present within nucleosomal DNA (Soufi et al. 2012;Zaret and Mango 2016). Binding of the pioneer factor leads to increased accessibility accompanied by deposition of active histone modifications, ultimately enabling other DNA-binding proteins to gain access to regulatory elements (Iwafuchi-Doi and Zaret 2014). Recruitment of chromatin remodeling enzymes by pioneer factors is a critical step in creating accessible DNA (King and Klose 2017). The Snf2 family of chromatin remodelers use energy from ATP hydrolysis to exchange, evict, and reposition nucleosomes within chromatin (Clapier and Cairns 2009). To date, most studies of pioneer factors have focused on their role as master regulators of cellular differentiation and lineage reprogramming (Soufi et al. 2012;Zaret and Mango 2016), leaving open their necessity in other cellular processes requiring access to DNA such as meiotic recombination.

Meiosis, which is essential for all sexual reproduction, involves major reorganization of chromatin, along with chromosome condensation, pairing, and, ultimately, genetic recombination. The outcome of recombination is the exchange of genetic material between parental chromosomes, resulting in the shuffling of alleles between generations. In mammals, fungi, and plants, recombination is concentrated at specialized sites termed hot spots (Baudat et al. 2013;Tock and Henderson 2018). The physical exchange of DNA is initiated when the topoisomerase SPO11 is recruited to hot spots, where it induces the double-strand breaks (DSBs) required for DNA exchange between homologous chromosomes (Keeney 2001). Significant chromatin remodeling is likely a prerequisite to overcome chromatin barriers at hot spots and allow the proper targeting of SPO11.

In most mammals, hot spot locations are determined by the DNA-binding protein PRDM9 (Baudat et al. 2010;Myers et al. 2010;Parvanov et al. 2010). PRDM9 contains a SET domain, which catalyzes both histone H3 lysine 4 trimethylation (H3K4me3) and H3K36me3 (Buard et al. 2009;Eram et al. 2014), and a series of tandem C-terminal zinc finger domains that bind DNA. Prdm9 is highly polymorphic with the majority of differences in regions encoding the zinc finger domains (Berg et al. 2011;Kono et al. 2014). Allelic variants bind different DNA sequences, ultimately dictating the location of hot spots. PRDM9 binding occurs within nucleosome-depleted regions (NDRs) that are subsequently targeted for DSBs by SPO11 (Baker et al. 2014;Lange et al. 2016). Loss of Prdm9 results in the persistence of unrepaired DSBs, incomplete synapsis of homologous chromosomes, and pachytene arrest in meiotic prophase I (Hayashi et al. 2005;Sun et al. 2015). Interestingly, meiosis-specific loss of two Snf2 family chromatin remodeling factors, Hells (Lsh) and Ino80, each results in phenotypes similar to loss of Prdm9 (Zeng et al. 2011;Serber et al. 2016), indicating that these chromatin remodelers might play direct roles in nucleosome remodeling at recombination hot spots.

Given the broad similarity in requirements to create accessible DNA for both pioneer factor-mediated control of gene regulation and meiotic recombination, we postulated that PRDM9 functions as a pioneer factor to overcome chromatin barriers at hot spots by recruitment of chromatin remodeling enzymes.

To functionally annotate the chromatin landscape of male germ cells, we determined co-occupancy of histone modifications using ChromHMM (Ernst and Kellis 2012).

ChromHMM uses a hidden Markov model to classify genomic regions into "states" by identifying combinatorial patterns of chromatin features, in this case histone modifications. Using ChromHMM, we solved an 11-state model using new and our previously published nucleosome-resolution ChIP-seq data of 10 histone modifications or histone variants (Fig. 1A, Supplemental Table 1). ChIP-seq was performed using enriched germ cells from 12-to 14-d postpartum (dpp) C57BL/6J (B6) mice. After birth, the first wave of meiosis occurs semisynchronously (Bellve et al. 1977), allowing enrichment of meiotic stages when PRDM9 is active (Ball et al. 2016). This germ cell model recapitulated many of the previously observed chromatin states found in other cell types, including promoters and insulators (states 1), enhancers (states 2-5), transcriptionally active regions (state 6), Polycomb repressive complex-repressed (state 8), and heterochromatin (states 9 and 10) (Fig. 1A,B).

Of these 11 chromatin states, only state 7 was enriched for PRDM9-binding sites; it is distinguished by a unique combination of histone modifications. In addition to H3K4me3 and H3K36me3, state 7 is characterized by H3K4me1 and H3K9ac (Fig. 1A,B). In contrast to other acetylation marks often found at enhancers (i.e., H3K27ac and H2BK120ac), only H3K9ac is enriched at hot spots. The majority (78%, 14,773/18,900) of all state 7 locations overlap with previously reported locations of B6 meiosis-specific DSBs (Smagulova et al. 2016) and PRDM9-dependent H3K4me3 modification (Fig. 1C; Baker et al. 2014), highlighting that state 7 represents recombination hot spots. Interestingly, compared with other phyla (Choi et al. 2013;Yamada et al. 2018), we did not detect H2A.Z at hot spots in mice. Together, these data identify a unique combination of histone features that biochemically distinguish recombination hot spots from other regulatory elements and may be involved in creating chromatin accessibility for DSB formation.

Given that recombination hot spots are marked by histone modifications associated with active chromatin, we measured DNA accessibility in germ cells using the assay for transposase-accessible chromatin (ATAC-seq) (Buenrostro et al. 2013). Biological replicates proved highly reproducible (r = 0.96), showed enrichment of open chromatin at promoters, and revealed typical nucleosome banding patterns (Supplemental Fig. 1A-D), all indicators of high-quality libraries. At PRDM9-dependent H3K4me3 sites, we detected increased DNA accessibility overlapping with PRDM9 motif locations (Fig. 1D-F). Overall, ATAC-seq identified fewer hot spots compared with H3K4me3 ChIP-seq (2902 ATAC peaks overlapping 13,498 PRDM9-dependent H3K4me3). Hot spots with greater PRDM9 binding and histone modification generally have more open chromatin (Fig. 1D). On average, the location of accessible DNA showed an inverse relationship to nucleosome positions at hot spots (Fig. 1E,F). Filtering ATAC-seq reads for those that fall within NDRs (i.e., <120 bp) showed that open chromatin was highest where PRDM9 binds. Together, these data build on an earlier observation that detected increased DNase hypersensitivity at a single hot spot (Shenkar et al. 1991), and show that generally hot spots are sites of open chromatin.

To determine whether the unique epigenomic state at hot spots is PRDM9-dependent, we used a strategy to maintain the same genetic background while changing Prdm9 alleles. To do so, we used a "knock-in" (KI) mouse strain that exchanges the zinc finger domain of the endogenous B6 Prdm9 Dom2 allele with the zinc finger domain from CAST/EiJ mice (Prdm9 Cst ) (Supplemental Fig. 1A; Baker et al. 2014). We measured chromatin accessibility and H3K9ac in enriched KI germ cells (Supplemental Fig. 1E,F), combined with H3K4me3 (Baker et al. 2014), and compared these with the B6 results (Fig. 2A,B). Quantifying differences between B6 and KI identified 5775 ATAC peaks and 14,249 H3K9ac peaks increased in KI (FDR < 0.01). Of these, 92% of the ATAC peaks (n = 5305) and 94% of the H3K9ac peaks (n = 13,408) overlap with reported PRDM9 Cst -dependent H3K4me3 peaks (Baker et al. 2014). Reciprocally, 1756 ATAC peaks and 7049 H3K9ac peaks were increased in B6 compared with KI, with 94% of the ATAC peaks (n = 1643) and 91% of the H3K9ac peaks (n = 6431) overlapping known PRDM9 Dom2 -dependent H3K4me3 peaks. Similar to the case with B6, open chromatin at hot spots in the KI strain is highest at PRDM9-binding sites (Supplemental Fig. 1H). These data show that hot spot chromatin that is accessible in B6 is not accessible in the KI. Instead, in the KI there are unique sites of accessible chromatin that overlap with known Prdm9 Cst hot spots, sites that are not open in B6.

Next, if PRDM9 binding facilitated open chromatin, we reasoned that variants that disrupt PRDM9 binding should change DNA accessibility. B6 and DBA/2J (D2) strains share the Prdm9 Dom2 allele, thus providing a natural experiment to determine the effect of genetic variation on chromatin accessibility. Recently we mapped quantitative trait loci (QTL) that regulate H3K4me3 levels in a genetic reference population whose genomes are homozygous mosaics of B6 and D2 (Baker et al. 2019). That study identified 1331 hot spots where the H3K4me3 level is regulated by local variation (local-QTL). We performed ATAC-seq on D2 mice and compared H3K4me3 levels and open chromatin at local-QTL to that from B6 (Fig. 2C). Hot spots with biased H3K4me3 levels also had biased open chromatin corresponding to increased frequency of single-nucleotide variants near PRDM9-binding sites. At each hot spot, we separately calculated scores for the PRDM9 motif for both B6 and D2 sequences. In general, the strain with the highest motif scores had the highest chromatin accessibility (Fig. 2C,right).

The programmed DSBs in meiosis are repaired primarily using the homologous chromatid as a template. This enabled us to determine whether the unique epigenomic landscape at hot spots occurs on both chromatids or only on the initiating chromatid. We performed ATAC-seq and H3K9ac ChIP-seq on spermatocytes from (B6xCAST)F1 hybrids and compared them with our published PRDM9 and H3K4me3 ChIP-seq data (Baker et al. 2015a). At asymmetric hot spots-i.e., those sites where PRDM9 preferentially binds one parental chromatid (Baker et al. 2015a;Smagulova et al. 2016) -both acetylation and chromatin accessibility are largely restricted to the parental chromatid to which PRDM9 binds (Fig. 2D,E). These data show that open chromatin is concordant with PRDM9 binding. Interestingly, however, they also suggest that hot spot chromatin is not open on the homologous chromosome, implying another mechanism is required to facilitate DNA accessibility for repair.

In somatic cells, DSBs lead to nucleosome remodeling and increased chromatin accessibility (Price and D'Andrea 2013). To determine whether this is also true of meiotic DSBs or whether chromatin remodeling precedes DSB formation, we tested whether the epigenomic state at hot spots is dependent on DSB formation. We performed ATAC-seq and ChIP-seq for H3K9ac and H3K4me3 on spermatocytes collected from 12-to 14-dpp mice lacking SPO11 (Baudat et al. 2000). In the absence of SPO11, we found wild-type levels of open chromatin and H3K9ac that show no quantitative differences compared with heterozygous littermate controls (Fig. 2F-H). These data agree with previous observations showing that H3K4me3 and H3K36me3 modifications at hot spots are independent of and therefore precede SPO11 binding and subsequent DSB formation (Grey et al. 2017).

Pioneer factors recognize partial DNA-binding motifs within nucleosomes (Soufi et al. 2012). Based on DNA sequence, PRDM9 Dom2 -binding sites (Smagulova et al. 2011) and PRDM9 Cst -binding sites (Supplemental Fig. S2A) are predicted to be occupied by nucleosomes. To measure nu-cleosome occupancy in the absence of PRDM9 binding in vivo, we combined data from a total of eight MNaseseq libraries from four Mus musculus domesticus mouse strains (B6, D2, [B6xD2]F1, and WSB/EiJ) that do not carry the Prdm9 Cst allele, resulting in >328 million reads. Hot spots were sorted based on MNase signal strength across the central 100-bp overlapping PRDM9 Cst -binding sites (Supplemental Fig. S2B). We found that ∼50% of PRDM9 Cst -binding sites have nucleosome signals at the PRDM9-binding site and that these nucleosomes are remodeled in spermatocytes that express PRDM9 Cst to become nucleosome-depleted regions. These data are in agreement with a recent study (Hinch et al. 2019) that found higher levels of nucleosome occupancy at PRDM9-binding sites in the absence of PRDM9. In addition to exploring PRDM9 in vivo, expression of PRDM9 ex vivo recapitulates aspects of hot spot activation, including allele-specific H3K4me3 deposition (Baker et al. 2015b). To further investigate whether PRDM9 binds nucleosomal DNA in an orthologous system, we aggregated MNase-seq signal from HEK293 cells expressing different PRDM9 alleles (Supplemental Fig. S2C; Baker et al. 2015b). HEK293 cells that either do not express PRDM9 or are transfected with the PRDM9 A allele have no obvious nucleosome pattern at PRDM9 C hot spots. In contrast, these same genomic regions have depleted nucleosomes at PRDM9 C -binding sites upon expression of PRDM9 C (Supplemental Fig. S2C). These observations support the germ cell data and suggest that nearly half of PRDM9-binding sites must be buried within nucleosomes.

Residence time of DNA-binding proteins can be estimated in vivo using nuclease protection assays such as ATAC-seq and DNAse hypersensitivity (Sung et al. 2014). Proteins with long residence times will protect DNA from nucleases, resulting in a characteristic footprint observed by decreased cutting frequency at binding sites. In vitro, PRDM9 has a long residency time when bound to synthetic DNA oligonucleotides (Striedner et al. 2017). Footprint analysis of ATAC-seq data from germ cells found that PRDM9 Cst motif locations showed little evidence for nuclease protection (Supplemental Fig. 2D,E), relative to long-resident proteins like CTCF (Supplemental Fig. 2F,G), similar to other transiently binding pioneer factors (Swinstead et al. 2016). Together, these data suggest that PRDM9 Cst can access target motifs even within nucleosomal DNA, and the lack of a clear footprint is consistent with PRDM9 having a short residency time in vivo.

Our observations that hot spots are marked by histone modifications associated with active chromatin and show PRDM9-dependent formation of accessible chromatin all point to an unknown mechanism controlling chromatin remodeling. To identify candidate remodeling complexes, we reasoned that if chromatin remodeling is required for hot spot activation, loss of the remodeling enzyme would result in a phenotype similar to loss of Prdm9. Loss of either Ino80 (Serber et al. 2016) or Hells (Zeng et al. 2011) fulfills these criteria. Given Ino80 plays a major role in histone variant exchange, specifically H2A. Z and H2A (Papamichos-Chronakis et al. 2011), and we found that recombination hot spots in mice lack H2A.Z (Fig. 1A), we focused our attention on Hells.

To examine HELLS localization during spermatogenesis, we used immunohistochemistry to detect protein abundance in seminiferous tubules (Fig. 3A). HELLS showed high expression in both spermatogonia and meiotic cells, similar to a previous result (Zeng et al. 2011). HELLS expression overlaps with phosphorylated H2AX (γH2AX), a marker of DSBs, with broad temporal colocalization in leptotene and zygotene stages. During pachytene, both HELLS and γH2AX are sequestered to the sex body (Fig. 3A, arrow). In addition, HELLS is coexpressed with PRDM9 in spermatocytes (Fig. 3A). The timing and colocalization of PRDM9 and HELLS are supported by recent single-cell transcript analysis (Jung et al. 2019) showing individual leptotene cells coexpress Hells and Prdm9 (Supplemental Fig. 3). Together, these data show that HELLS and PRDM9 are both present when recombination hot spots are activated.

Next, we used a conditional strategy to study the meiosis-specific function of HELLS. We used Stra8-iCre (Sadate-Ngatchou et al. 2008) to create male mice with a homozygous loss of Hells at the onset of meiosis (Hells CKO). Adult CKO mice had reduced testis weight (Fig. 3B) with extensive loss of postmeiotic germ cells (Fig. 3C), while testes from Hells heterozygous littermates (control) appear normal. This block in germ cell development recapitulates an earlier observation using allographs of testis tissue between Hells -/-mice and wild-type donors (Zeng et al. 2011). Cross sections of adult Hells CKO seminiferous tubules show loss of HELLS staining (Supplemental Fig. 4A). The loss of later stage germ cells correlates with a large increase in germ cell apoptosis in Hells CKO compared with heterozygous littermate controls (Fig. 3C,D). Finally, Hells CKO male mice did not sire any offspring when mated to normal B6 females. In summary, we confirm that HELLS is required for sperm production and fertility of male mice.

Using surface-spread meiotic chromatin in combination with immunohistochemistry, we found that HELLS is diffusely localized in leptotene and zygotene spermatocytes (Fig. 3F), similar to localization of PRDM9 in the same stages (Parvanov et al. 2017). Meiotic spreads showed that later in prophase I, after repair of the majority of autosomal DSBs, HELLS is restricted to the sex body. Staining for HELLS is absent from Hells CKO mice (Fig. 3F). To determine at which stage meiotic progression is blocked in Hells CKO mice, we staged spermatocytes using spreads stained for both γH2AX as a marker of DSBs and SYCP3 to characterize chromosome synapsis in adult male mice (Supplemental Fig. 4B). We found that Hells CKO mice have an increased proportion of spermatocytes in leptotene and zygotene stages and a nearly complete loss of normal pachytene and diplotene stages (Fig. 3F). Hells CKO cells scored as pachytene consistently show persistent γH2AX phosphorylation and incomplete synapsis absent from littermate controls (Supplemental Fig. 4B). These pachytene-like cells are negative for the testis-specific histone variant H1t (Cobb et al. 1999), a marker for late pachytene (Supplemental Fig. 4C). Overall, our data confirm that loss of Hells leads to meiotic arrest at the late zygotene to early pachytene stage (Zeng et al. 2011), with incomplete synapsis and persistent DSBs leading to spermatocyte apoptosis.

In Hells-null mice, meiotic DSBs are targeted to other functional elements Because loss of Hells leads to incomplete synapsis and persistent γH2AX, we next determined the levels and locations of meiotic DSB in mutant spermatocytes. To characterize DSBs cytologically, we used spread meiotic chromatin labeled with SYCP3 and DMC1. DMC1 is a meiosis-specific recombinase that binds to single-stranded DNA at the sites of DSBs (Neale and Keeney 2006). While zygotene spermatocytes from Hells CKO overall had similar levels of DMC1 foci as littermate controls (Fig. 4A), these foci persist in Hells CKO pachytene-like cells on synapsed chromosomes (Fig. 4B), indicating incomplete DSB repair.

While the number of DSBs were similar between Hells CKO and control, their genomic locations were dramatically different. To determine locations of meiotic DSBs, we identified genomic regions bound by DMC1 (Khil et al. 2012) in testes from adult CKO and littermate controls. In total, we identified 13,643 DMC1 peaks in Hells CKO and 15,718 in heterozygous littermates. Overall, Hells CKO germ cells lost DSBs at canonical hot spots and gained DSBs at other sites (Fig. 4C). For example, loss of Hells resulted in a 12-fold increase in DSBs at promoters and a concomitant decrease in DSBs at distal sites (Fig. 4D). Nearly all DMC1 peaks identified in Hells CKO germ cells overlap with the location of DSBs in Prdm9 -/-(93.7%, 13,206/13,643), while only 15.7% (2475/15,718) of DMC1 sites identified in control germ cells overlap Prdm9 -/-DMC1 peaks (Fig. 4E).

To determine the epigenomic landscape targeted for DSBs in Hells CKO, we calculated enrichment of DMC1 sites and open chromatin within the ChromHMM states (Fig. 4F). While hot spots (state 7) were enriched for DSBs in the Hells control, in both Hells and Prdm9 mutants, DSBs were enriched in states 1 and 2, which are annotated as promoters, insulators, and enhancers. Interestingly, although H3K4me1, H3K4me3, and H3K9ac are present in states 1 and 2 (Fig. 1A), state 3 had a combination of histone modifications similar to classically defined enhancers that lack H3K4me3, and was less enriched for DSBs. States 1 and 2 were enriched for PRDM9-independent accessible chromatin; whereas state 7 only had PRDM9-dependent accessible chromatin in B6 compared with KI (Fig. 4F). Notably, unlike hot spots, states 1 and 2 were not enriched for H3K36me3. Furthermore, in Hells CKO, in which DSBs largely occur at state 1 and 2 locations, spermatocytes undergo apoptosis similar to loss of Prdm9. Together, these data suggest that in the absence of Hells, DSBs are retargeted to regions of open chromatin that lack the proper epigenome to become competent hot spots.

Because DSBs were retargeted in Hells CKO germ cells, we next examined the epigenome of the mutant spermatocytes. We first confirmed loss of HELLS in 12-dpp whole testes (Fig. 5A), when PRDM9 is active but prior to the Hells CKO meiotic arrest. ChIP-seq for H3K4me3 and ATAC-seq on enriched germ cell populations from 12-dpp male mice identified a significant reduction of epigenomic modification in Hells CKO mice at recombination hot spots (Fig. 5B,C; Supplemental Fig. 5). While 11,812 H3K4me3-modified sites are significantly reduced in Hells CKO compared with control (log 2 fold change < -1, FDR < 0.01), only 73 sites showed higher modification in Hells CKO. Annotating peaks using state 7, we found that both H3K4me3 (n = 11,125) (Fig. 5D) and chromatin accessibility (n = 1851) (Fig. 5E) sites at hot spots were re-duced in Hells CKO. In contrast, among peaks overlapping HMM state 1 (n = 31,161 H3K4me3 peaks), there were virtually no sites that changed. Histological sections of 12-dpp seminiferous tubules from both Hells CKO and heterozygous control littermates showed similar stages in development (Supplemental Fig. S6A). Although measuring PRDM9 in whole testes from CKO mice by Western blot detected a modest reduction in protein abundance (Supplemental Fig. 6B,C), PRDM9 was readily detected in individual leptotene and zygotene spermatocytes (Supplemental Fig. 6D). These data show that HELLS is required for establishment of the epigenomic state and chromatin accessibility at hot spots.

To develop a model for temporal molecular characterization of hot spot activation, we chose a human cell line that expresses HELLS and created a stably integrated Flag-PRDM9 C allele under inducible control of the tetracycline promoter (HEK293-P9 C ) (Fig. 6A). Addition of doxycycline led to robust expression of Flag-PRDM9 C and increased H3K4me3 at PRDM9 C hot spots (Fig. 6A,B). PRDM9 C expression also increased MNase sensitivity at PRDM9 Cbinding sites compared with adjacent regions (Fig. 6C), indicating nucleosome remodeling. Reciprocal coimmunoprecipitation (co-IP) using antibodies directed to either HELLS or PRDM9 (Flag) identified a doxycycline-dependent interaction between the two proteins (Fig. 6A right). We confirmed PRDM9-HELLS interaction in vivo by performing co-IP on protein lysates prepared from 12-dpp mouse testes. We detected a HELLS-PRDM9 interaction in testes from B6 mice and littermate controls but not in Hells CKO (Fig. 6D) or Prdm9 -/-(Fig. 6E). To test whether PRDM9-HELLS interaction requires assembly on chromatin, co-IP was repeated using a mouse strain expressing a PRDM9 variant that lacks the DNA-binding domain (PRDM9 ΔZF ) (Parvanov et al. 2017). We found that HELLS did interact with PRDM9 ΔZF (Fig. 6E), suggesting that this interaction is independent of PRDM9-directed DNA binding.

We next showed that HELLS is required for robust PRDM9 binding at hot spots (Fig. 6F-H). HELLS expression in HEK293-P9 C cells was reduced using small interfering RNA (siRNA), followed by induced expression of PRDM9 C (Fig. 6G). ChIP for Flag-PRDM9 C in HEK293-P9 C cells successfully detected robust binding at C-dependent hot spots and no binding at a PRDM9 A hot spot or at the GAPDH promoter (Fig. 6H). Critically, upon loss of HELLS, PRDM9 C binding at hot spots was reduced to background levels (Fig. 6H). Together, these data show that PRDM9 and HELLS form a complex in vivo and, by extension, suggests that active chromatin remodeling is required for robust PRDM9 binding at hot spots.

Our experiments indicate that HELLS and PRDM9 interact and open chromatin at hot spots. This suggests a model in which PRDM9 recruits HELLS to hot spots to facilitate nucleosome remodeling for DSBs. To confirm this, we first tested whether HELLS binding at hot spots requires the presence of PRDM9 using our HEK293-P9 C cells. In the absence of PRDM9 C , ChIP for HELLS did not identify binding at any of the five genomic locations tested by qPCR. In contrast, upon expression of PRDM9 C , we observed significantly increased binding of HELLS only at C-hot spots and no change at negative controls (Fig. 7A), suggesting HELLS binding at hot spots is PRDM9 dependent.

To test for HELLS binding of hot spots in vivo, we performed ChIP using enriched germ cells from 12-dpp KI mice. Biological replicates were highly correlated (Pearson's r = 0.91) (Fig. 7B) and identified 2675 HELLS-binding sites. Visually, we detected clear HELLS ChIP signal at recombination hot spots (Fig. 7C) and promoters (Supplemental Fig. 7A,B). HELLS peaks were annotated based on overlap with PRDM9-dependent H3K4me3 peaks from KI mice (Baker et al. 2014) and TSSs representing promoters (Fig. 7D). The largest class of HELLS-binding sites (n = 1234, 46.1%) overlapped known PRDM9 Cst hot spots. HELLS binding was stronger at hot spots compared with promoters (p = 5.6 × 10 -25 ) or other unannotated regions (p = 6.6 × 10 -24 ) (Fig. 7E). Of all HELLS binding regions, 97% (n = 2591) fell within regions of open chromatin identified in KI germ cells. Moreover, hot spots with a HELLS ChIP-seq peak (n = 1230) had significantly higher levels of open chromatin (p = 1.3 × 10 -244 ) compared with those without (n = 4529) (Fig. 7F). In addition, genes with a HELLS peak at their promoters are generally expressed at significantly higher levels than those without (Supplemental Fig. 7C). Given that HELLS is a chromatin remodeling factor, HELLS might bind at hotpots via nucleosomes; however, the maximum average HELLS signal is found at the PRDM9-binding site (Fig. 7G,H). Together with the protein-protein interaction experiments, these results show that PRDM9 recruits HELLS to hot spots.

Pioneer function is defined by three features: (1) the ability to recognize partial motifs within nucleosomal DNA, (2) acquisition of active histone modifications, and (3) increased chromatin accessibility. Here we show that together HELLS and PRDM9 fulfill these criteria at hot spots. PRDM9 brings two of the salient features of canonical pioneer factor-mediated chromatin reorganization into one molecule: (1) a sequence-specific targeting domain that provides locational specificity and can interact with nucleosomal DNA, and (2) the ability to catalyze epigenetic marks associated with active chromatin. Both of these features are required for meiotic recombination (Parvanov et al. 2017;Diagouraga et al. 2018). While the number of recombination hot spots are generally depleted from repressive regions of chromatin marked by H3K9me2/me3 (Walker et al. 2015;Patel et al. 2019), hot spots that are found in these domains show nearly similar levels of PRDM9-dependent histone modification as those found outside of H3K9me2/3 domains, supporting the idea that even within heterochromatin, PRDM9 and HELLS have the capacity to create open chromatin. Additionally, a previous experiment characterizing the chromatin landscape using purified populations of male germ cells at stages earlier or later than leptotene/zygotene did not detect open chromatin at hot spots (Maezawa et al. 2018). This observation supports the role of a PRDM9-HELLS complex in creating de novo open chromatin at hot spots during a developmentally narrow window. Further evidence for a pioneer factor role for PRDM9 comes from our ex vivo analyses. Ectopic expression of PRDM9 in cells expressing HELLS recapitulates allelespecific histone modification and nucleosome remodeling. This suggests that the ability for PRDM9 to function as a pioneer factor is independent of chromatin structure that might be unique to meiosis.

The ability to target DSBs to open chromatin is critical for meiotic recombination. In mice, meiotic DSBs (Baker et al. 2014;Lange et al. 2016) and crossovers identified through single-sperm sequencing (Hinch et al. 2019) are concentrated within hot spot NDRs. In the absence of open chromatin at hot spots, by loss of either HELLS or PRDM9 (Brick et al. 2012), SPO11 preferentially targets other regions of accessible chromatin to create DSBs. Hot spots are marked by a unique combination of histone modifications not found at other regulatory elements, specifically H3K4me1, H3K4me3, H3K36me3, and H3K9ac. These observations support a precedent set by an early study that found enrichment of H3K9ac at an individual hot spot (Buard et al. 2009), and a recent survey of histone modifications in purified populations of male meiotic cells found that out of six histone acetylation antibodies tested, only H3K9ac showed high enrichment at hot spots (Lam et al. 2019). It is possible that this unique combination of histone modifications provides an epigenomic addressing system for directing placement or repair of DSBs through recruitment of an epigenetic reader. A plausible candidate for such a reader is ZCWPW1 (Jung et al. 2019;Li et al. 2019), which is coexpressed with PRDM9 during the leptotene/zygotene phase and carries two protein domains that recognize H3K4me3 and H3K36me3. While the redirection of DSBs to other regulatory elements usually results in a meiotic block and infertility, notably, the severity of the infertility phenotype caused by loss of PRDM9 depends on the genetic background (Mihola et al. 2019), suggesting other components in action. Finally, through creation of aberrant sites of open chromatin, the pioneer function of the HELLS-PRDM9 complex could provide a molecular mechanism by which particular alleles are associated with genomic instability in certain cancers (Houle et al. 2018).

Here we identified that the chromatin remodeling enzyme HELLS is a second critical component of recombination initiation. Chromatin accessibility and histone modification at hot spots require HELLS, and furthermore, HELLS localizes at hot spots through interaction with PRDM9. HELLS is a member of the SNF2 family of ATP-dependent chromatin remodeling enzymes that use the energy of ATP to reposition, remodel, and remove histones from DNA substrate (Clapier and Cairns 2009). HELLS is misregulated in human tumors (von Eyss et al. 2012) and mutated in patients with immunodeficiency-centromeric instability-facial anomalies syndrome (Thijssen et al. 2015). Mice with systemic loss of HELLS exhibit prenatal lethality (Geiman et al. 2001); therefore, most functional studies of HELLS have required ex vivo cell culture. Notable exceptions are two studies showing that loss of HELLS in both male and female germ cells results in errors in synapsis and DSB repair during meiosis (De La Fuente et al. 2006;Zeng et al. 2011). Here, using conditional loss of function, we confirmed the requirement of HELLS in meiotic progression and further defined the mechanistic role of HELLS in chromatin remodeling at hot spots.

HELLS has been implicated in the establishment of genome-wide methylation patterns and chromatin repression (Yu et al. 2014;Termanis et al. 2016). HELLS ATPase domain is required for these functions (Ren et al. 2015;Termanis et al. 2016), and ATP is necessary for nucleosome remodeling in vitro (Jenness et al. 2018). If HELLS functioned to close chromatin in meiotic cells, we would have expected to see an increase in open chro-matin in CKO mice; instead, we detected a near universal closing of chromatin at hot spots upon loss of HELLS. Given that our CKO strategy ablated HELLS primarily in meiotic cells, these epigenomic experiments are not optimal for determining the consequences of loss of HELLS in other cell types but support the idea that the primary role of HELLS in meiosis is creating open chromatin at hot spots. In contrast, ChIP for HELLS was performed on a population of male germ cells, including spermatogonia. In addition to hot spots, we found that HELLS is enriched at promoters associated with higher gene expression, in agreement with previous results in fibroblasts (von Eyss et al. 2012). However, like many cell types, promoters are constitutively open during spermatogenesis (Maezawa et al. 2018), supporting the idea that HELLS role in opening chromatin in prophase I is specific to hot spots. Together these data show that, in addition to its wellcharacterized role in generating repressive chromatin, HELLS is associated with increased chromatin accessibility. These disparate functions likely depend on HELLS protein binding partners.

Our evidence suggest that HELLS is recruited to hot spots through PRDM9, rather than being recruited as a consequence of PRDM9-dependent epigenomic modification. The HELLS-PRDM9 interaction persists in cells with a mutated PRDM9 lacking zinc fingers, supporting the idea that their interaction is independent of DNA binding by PRDM9. In addition, on average the maximum HELLS enrichment is found in the center of NDR at the PRDM9-binding site, rather than at the flanking nucleosomes. These observations support a model in which HELLS is recruited by direct interaction with PRDM9, and agree with the previous observation that HELLS-mediated chromatin remodeling in vitro requires interaction with the zinc finger protein CDCA7 (Jenness et al. 2018).

Overall, data reported here give rise to a model (Fig. 7I) in which HELLS and PRDM9 form a meiosis-specific pioneer complex to create open chromatin at recombination hot spots. This model not only expands our knowledge about diversity in function of pioneer complexes in general but, more specifically, also poses interesting challenges with respect to meiotic mechanisms. Clearly further work is required to determine how the HELLS-PRDM9 complex is temporally regulated, how the epigenomic environment and open chromatin at hot spots recruits programmed DSBs, and how chromatin barriers are overcome on the template chromatid used to repair these breaks.

C57BL/6J (B6, stock no. 000664), DBA/2J (D2, 000671), CAST/EiJ (CAST, 000928), B6.129X1-Spo11 tm1Mjn /J (Spo11 -/-, 019117), Tg (Stra8-iCre)1Reb/J (Stra8-iCre, 008208), B6.Cg-Tg(ACTFLPe) 9205Dym/J (ATCFLPe, 005703), and B6;129P2-Prdm9 tm1Ymat /J (Prdm9 KO , 010719) are all available through The Jackson Laboratory. Mice carrying various Prdm9 alleles B6.Cg-Prdm9 tm1.1Kpgn / Kpgn (Baker et al. 2014) and B6.Cg-Prdm9 tm3.1Kpgn /Kpgn (Prdm9 ΔZnf ) (Parvanov et al. 2017) were described previously. Frozen embryos of the EUCOMM allele C57BL/6NTac-Hells tm1a(EUCOMM)Wtsi /Ieg were obtained through the European Mouse Mutant Archive (EMMA 04583) and implanted into pseudopregnant B6 females at The Jackson Laboratory. Hells tm1a(EUCOMM)Wtsi/+ pups were crossed to ATCFLPe to remove the FRT flanked neomycin selection cassette, and resulting mice were inbred to make the conditional ready Hells tm1c(EUCOMM)Wtsi allele. These Hells conditionally ready mice are available through The Jackson Laboratory (B6.Cg-Hells tm1c(EUCOMM)Wtsi /Bakr, stock no. 30483). Mice were made homozygous for a germ cell-specific loss of function in two steps by crossing Hells tm1c homozygotes to Stra8-Cre to create male mice with germ cells heterozygous for Hells +/tm1d and Stra8-Cre. These male mice were backcrossed to female Hells tm1c homozygotes to produce Hells tm1c/tm1d ; Stra8-Cre offspring, of which testes from male mice would lack functional Hells (Hells tm1d homozygotes) and were used throughout this study as CKO. Fertility testing was performed on three Hells CKO male mice. Starting at 8 wk of age, the male CKO mice were each mated to two or more confirmed B6 breeder females for at least 4 mo, and no pregnancy or litters were ever recorded. All animal experiments were approved by the animal care and use committee of The Jackson Laboratory (Animal Use Summary 16043).

Cell culture and HEK293-P9 C cells HEK293-P9 C cells were created using the T-REx system (Thermo Fisher K102001) and the T-REx-293 (HEK293) cell line (Thermo Fisher R71007). Human Flag-tagged PRDM9 C (Baker et al. 2015b) was subcloned into the pcDNA4/TO vector. T-REx-293 cells were transfected using Lipofectamine 3000 (Invitrogen), allowed to recover for 48 h, and subsequently split and plated on 100-cm plates maintained with 5 µg/mL blasticidin (Thermo Fisher R210-01) and 300 µg/µL zeocin (Thermo Fisher R250-01) to allow for clonal expansion. Individual colonies were picked, expanded, and tested by Western blot for expression of Flag-PRDM9 C . A single cell line was subsequently used for all studies. Flag-PRDM9 C expression was induced by replacing culture media with fresh media containing 1 µg/mL doxycycline (Sigma D9891) in DMSO (Sigma D2650-5 5 mL) and grown for an additional 24 h. Human Hells siRNA (Origene SR302088) knockdown was performed in HEK293-P9 C cells with either 100 nM siRNA or universal scrambled control (Origene SR30004) using Lipofectamine RNAiMAX (Thermo Fisher 13778030).

Protein extracts for Western blots from cultured HEK293-P9 C cells were prepared as described previously (Baker et al. 2015b). Protein extracts for co-IP from HEK293 cells was prepared using HEPES-based buffer as described previously (Baker et al. 2015b). For co-IP, protein lysates were diluted to 500 µL using HEPES buffer and treated with 4 µL antibody and 20 µL of Protein-A/G magnetic beads (Fisher). Immunocomplexes were incubated overnight at 4°C with rotation, washed three times with HEPES wash buffer (50 mM Hepes-KOH at pH 7.4, 150 mM NaCl, 0.4% NP-40), and eluted with 20 µL of 2× sample loading buffer (50 mM Tris-HCL at pH 6.8, 2% SDS, 10% glycerol, 100 mM DTT, 0.05% bromophenol blue) heated for 10 min at 95°C. Primary antibodies for Western blots and immunoprecipitation included anti-Flag M2 (Millipore/Sigma F3165), anti-PRDM9 (custom) (Parvanov et al. 2017), anti-HELLS (Millipore/Sigma ABD41, lot 3069868), and anti-β-Tubulin (Sigma T4026, lot 125M4884V). Secondary antibodies for Western blots included goat antimouse HRP (Bio-Rad 170-6516), goat antirabbit HRP (Bio-Rad 172-1019), and donkey antiguinea pig (Millipore AP193P).

Testes were fixed for 12-16 h with Bouin's solution and embedded in paraffin wax. Five-micrometer sections were prepared and stained with periodic acid Schiff diastase (PAS). Samples prepared for immunofluorescence were fixed in 4% PFA in PBS for 12-16 h as described previously (Parvanov et al. 2017). Prior to immunofluorescent staining, sections were deparaffinized with xylene washes and rehydrated through a gradient of ethanol washes (95%, 70%, and 50%, PBS) and then boiled in 10 mM sodium citrate buffer (pH 6.0) for 10 min. After being cooled and washed in PBS, the slides were blocked with 10% normal donkey serum (Jackson ImmunoResearch Laboratories 017-000-121), 3% BSA, 0.2× protease inhibitor cocktail, and 0.05% Triton-X 100 for 1 h at room temperature before adding primary antibodies and incubating overnight at 4°C. Slides were washed three times for 5 min in PBS and incubated in secondary antibody and then washed three times for 10 min in PBS. Slides were treated with ProLong Gold antifade reagent with DAPI (Invitrogen P36935).

Chromosome spreads were prepared using the drying-down technique (Peters et al. 1997). For immunofluorescence, slides were blocked with 3% BSA, 1% normal donkey serum, and 0.005% Triton-X 100, and primary antibodies were incubated overnight. Slides were washed and treated with secondary antibodies as described above and treated with ProLong Gold antifade reagent with DAPI (Invitrogen P36935).

Primary antibodies used for immunofluorescence on cross sections and chromosome spreads included anti-PRDM9 (1:100) (Parvanov et al. 2017), anti-HELLS (1:250; Millipore/Sigma ABD41), anti-γH2AX (1:2000; Millipore 05636I, lot 2888552), anti-SYCP3 (1:500; Santa Cruz Biotechnology SC74569, lot J1314), anti-H1t (1:300; a gift from Mary Ann Handel), and anti-DMC1 (1:200; Santa Cruz Biotechnology sc-8973). Secondary antibodies included goat antirabbit immunoglobulin G (IgG), Alexa fluor 594 conjugate (1:300; A-11037), goat antirabbit IgG, Alexa fluor 488 conjugate (1:500; ab150077), goat antimouse IgG, Alexa fluor 488 conjugate (ab150113), goat antimouse IgG, Alexa fluor 647 conjugate (1:300; A-21236), goat anti-guinea pig IgG, and Alexa fluor 488 conjugate (1:300; A-11073).

TUNEL assay was performed on testis sections from 6-wk-old Hells heterozygous control or CKO mice using the in situ cell death detection kit (Roche 11684795910) according to the manufacturer's protocol.

Isolation of enriched population of germ cells for ChIP for histone modifications and ATAC-seq was performed as described previously (Baker et al. 2014) on 12-to 14-dpp mice. Antibodies used for histone ChIP experiments were anti-H3K4me1 (Millipore/ Sigma 07-436, lot 2289129), anti-H3K4me3 (Millipore/Sigma 07-473, lot 3018770), anti-H3K9ac (Active Motif 39137, lot 09811002), anti-H3K27ac (Abcam ab4729, lot GR211893-1), anti-H3K27me3 (Millipore/Sigma 07-449, lot 2475696), anti-H2BK120ac (Active Motif 39120, lot 01008001), and anti-H2A.Z (Millipore/Sigma 07-594, lot 2455725). For ChIP of H3K9ac, H3K27ac, and H2BK120ac used in the ChromHMM analysis, germ cell isolation and ChIP buffers were supplemented with 20 mM sodium butyrate.

ChIP for Flag-PRDM9 was described previously (Baker et al. 2015a) and modified for HELLS to include a dual cross-linking step using disuccinimidyl glutarate (DSG) to first capture proteinto-protein interactions. Seminiferous tubules were placed in 5 mL of PBS (pH 7.4) supplemented with 1 mM MgCl 2 . A fresh 0.25 M stock of DSG was prepared in DMSO, added at a final concentration of 2 mM, and incubated for 30 min at room temperature with rotation. After incubation, fresh paraformaldehyde was added to a final concentration of 1%, and tubules were incubated for 5 min with rotation and quenched with 125 mM glycine. The remaining procedure for ChIP was followed as described (Baker et al. 2015a) using anti-HELLS (Millipore/Sigma ABD41, lot 3069868) or normal mouse IgG (Millipore/Sigma 12-371, lot 2880788).

Single-strand DMC1 ChIP-seq was performed as described (Khil et al. 2012) using DMC1 antibody (Santa Cruz Biotechnology sc-8973, lot K2614) on testes from 6-wk-old male mice collected from two independent biological replicates of Hells CKO and heterozygous littermate controls.

ChIP libraries were prepared for high-throughput sequencing either using Bioo Scientific's NEXTflex ChIP-seq kit without size selection (for ChromHMM data) or the KAPA hyper kit (Kapa Biosystems) for all other libraries. Library quality and size distribution were visualized using a Bioanalyzer (Agilent). All samples were sequenced using either the Illumina HiSeq2500 or NextSeq platforms.

Quantitative PCR was performed using PowerUp SYBR Green 2× master mix (ThermoFisher) on a ViiA 7 real-time PCR system (Thermo Fisher). PCR was carried out for 40 cycles followed by melting curve analysis. All samples were run with triplicate technical reactions, and cycle number was determined using an automated threshold analysis. MNase sensitivity assay in HEK293-P9 C cells before and after doxycycline addition was performed as described for spermatocytes (Baker et al. 2014). Oligonucleotides used for qPCR in this study are listed in Supplemental Table 2.

For B6, KI, D2, and (BxC)F1 hybrid genotypes, ATAC-seq was performed on male germ cells using the original protocol (Buenrostro et al. 2013). ATAC-seq for Spo11 and Hells genotypes was performed using the modified FAST-ATAC protocol (Corces et al. 2016). For Hells genotypes, ATAC-seq libraries were prepared from two independent homozygous Hells wild-type (Hells tm1c/+ ) and three independent Hells CKO (Hells tm1c/Δ ; Stra8-iCre). Tn5 transposition was performed using 1 × 10 5 enriched male germ cell populations prepared from pooling testes from three to four individual mice similar to ChIP. ATAC-seq libraries were amplified using a total of eight to 10 amplification cycles and DNA purified using Agencourt AMPure XP beads (Beckman Coulter).

All reference sequence coordinates in this study correspond to UCSC mm10 (release 3/18/2013).

Chromatin segmentation was performed using ChromHMM as described (Ernst and Kellis 2012) solving for an 11-state solution on histone ChIP-seq libraries (Supplemental Table 1). Duplicate ChIP libraries were merged prior to ChromHMM processing. Fold enrichments were calculated using whole-testis CTCF occupancy and predicted cis-regulatory elements (Shen et al. 2012), lamina-associated domains from mouse embryonic stem cells (Meuleman et al. 2013), and in vitro PRDM9-binding sites (Walker et al. 2015). Gene expression levels from germ cells of agematched mice were downloaded from GEO GSE72833 (Ball et al. 2016). Expressed genes were defined by threshold transcripts per million (TPM) of more than three (n = 13,739). All remaining genes were annotated as "silent." ATAC-seq fastq files were trimmed using trimmomatic (Bolger et al. 2014) using TRAILING:3 SLIDINGWINDOW:4:15 MIN-LEN:36 and removing Nextera adapters. For paired-end ATACseq samples, trimmed sequences were aligned using bowtie (version 0.12.9) (Langmead et al. 2009) with the following settings: -chunkmbs 2000 -S -X2000 -m1. Mapped reads were adjusted for Tn5 insertion site as described (Buenrostro et al. 2013). For single-end ATAC-seq, trimmed sequences were aligned using bwa mem (version 0.7.15) (Li and Durbin 2009) with default settings. Picard Tools was used to remove duplicates and calculate insert size metrics. ATAC-seq alignments were filtered to remove reads with MAPQ < 10 and mitochondrial reads.

ChIP libraries were aligned using bwa mem with default settings (version 0.7.15) (Li and Durbin 2009). D2 ChIP-seq and ATAC-seq data were aligned to an in silico pseudogenome incorporating known variants (R78-REL1505) (Wu et al. 2010) and converted to mm10 reference coordinates using G2Gtools (accessible online at https://github.com/churchill-lab/g2gtools). Allele-specific ChIP analysis was performed using variant-aware alignment strategy EMASE (Raghupathy et al. 2018). To identify hot spots and ensure robust haplotype calls, peaks were selected that overlap with (BxC)F1 PRDM9 ChIP-seq summits and filtered for > 30 reads that map uniquely to either B6 or CAST.

Both ChIP-seq and ATAC-seq peaks were called using MACS version 1.4.2 (Zhang et al. 2008) using a cutoff of 10 -5 . Duplicate reads were retained for ChIP against histone modifications and removing for HELLS ChIP. To build reference peakomes for each experiment, peaks were called on merged bam files from replicate experiments. Peak files across different strains or Prdm9 alleles were concatenated, sorted, and merged using bedtools (Quinlan and Hall 2010). Peaks overlapping ENCODE blacklist regions were removed for subsequent analysis. Read counts for each reference peakome were extracted from sample bam files using Bed-Tools multicoverage and are available as processed data at NCBI GEO (GSE135896). Genomic profiles were visualized using the UCSC browser.

Statistical analysis was performed using R (release 3.4.1). For differential analysis, read counts were normalized using the trimmed mean of M-values method and differential analysis performed using edgeR (Robinson et al. 2010). Genome-wide false discovery rates (FDRs) were calculated by adjusting P-values following the Benjamini-Hochberg method. Annotation of peaks using genomic locus overlap was performed with bedtools intersect or HOMER mergePeaks functions (Heinz et al. 2013) and visualized using the UpSet package (Lex et al. 2014). Read matrices for heat maps were generated using the CoverageView R package after RPM normalization with 10-bp bins and visualized using Java TreeView (Saldanha 2004). ATAC-seq footprints were aggregated and normalized using ATACseqQC (Ou et al. 2018). Calculation of motif scores for predicted PRDM9-binding sites was performed by generating a position weighted matrix as outlined (Wasserman and Sandelin 2004).