A missense in HSF2BP causing Primary Ovarian Insufficiency affects meiotic recombination by its novel interactor C19ORF57/MIDAP

Primary Ovarian Insufficiency (POI) is a major cause of infertility, but its etiology remains poorly understood. Using whole-exome sequencing in a family with 3 cases of POI, we identified the candidate missense variant S167L in HSF2BP, an essential meiotic gene. Functional analysis of the HSF2BP-S167L variant in mouse, compared to a new HSF2BP knock-out mouse showed that it behaves as a hypomorphic allele. HSF2BP-S167L females show reduced fertility with small litter sizes. To obtain mechanistic insights, we identified C19ORF57/MIDAP as a strong interactor and stabilizer of HSF2BP by forming a higher-order macromolecular structure involving BRCA2, RAD51, RPA and PALB2. Meiocytes bearing the HSF2BP-S167L mutation showed a strongly decreased expression of both MIDAP and HSF2BP at the recombination nodules. Although HSF2BP-S167L does not affect heterodimerization between HSF2BP and MIDAP, it promotes a lower expression of both proteins and a less proficient activity in replacing RPA by the recombinases RAD51/DMC1, thus leading to a lower frequency of cross-overs. Our results provide insights into the molecular mechanism of two novel actors of meiosis underlying non-syndromic ovarian insufficiency. Summary Felipe-Medina et al. describe a missense variant in the meiotic gene HSF2BP in a consanguineous family with Premature Ovarian Insufficiency, and characterize it as an hypormorphic allele, that in vivo impairs its dimerization with a novel meiotic actor, MIDAP/ C19ORF57, and affect recombination at double-strand DNA breaks.


Introduction
The process of gametogenesis is one of the most complex and highly regulated differentiation programs. It involves a unique reductional cell division, known as meiosis, to generate highly specialized gametes. The outcome of meiosis is the production of oocytes and spermatozoa, which are the most distinctive cells of an adult organism and are essential for the faithful transmission of the genome across generations.
The meiotic division is an orderly process that results in the pairing and synapsis of homologous chromosomes and cross-over (CO) formation, which ultimately enable homologous chromosomes segregation (Hunter, 2015;Loidl, 2016;Zickler and Kleckner, 2015). In mammals, pairing of homologs is dependent on the repair of self-induced doublestrand breaks (DSBs) during prophase I by homologous recombination (Handel and Schimenti, 2010) and it precedes the intimate alignment of homologous chromosomes (synapsis) through the zipper-like structure that is the synaptonemal complex (SC) (Cahoon and Hawley, 2016). The SC proteinaceous tripartite assembly also provides the structural framework for DSBs repair , as epitomized by the tight association of the recombination nodules (RN, multicomponent recombinogenic factories) and the axial elements of the SC (Zickler and Kleckner, 2015).
Meiotic DSBs repair is an evolutionarily conserved pathway that is highly regulated to promote the formation of at least one crossover per bivalent. Chromosome junction together with cohesion ensure tension between sister kinetochores during the first reductional division. As other DNA repair processes, proper meiotic recombination is essential for genome stability and alterations can result in infertility, miscarriage and birth defects (Geisinger and Benavente, 2016;Handel and Schimenti, 2010;Webster and Schuh, 2017).
Infertility refers to failure of a couple to conceive and affects 10-15% of couples (Isaksson and Tiitinen, 2004). Infertility can be due to female factors, male factors, a combination of both or to unknown causes, each category representing approximately 25% of cases (Isaksson and Tiitinen, 2004;Matzuk and Lamb, 2008). There are several underlying causes and physiological, genetic and even environmental and social factors can play a role.
Forward and reverse genetic analyses in model organisms have identified multiple molecular pathways that regulate fertility and have allowed inferring reasonable estimates of the number of protein-coding genes essential for fertility (de Rooij and de Boer, 2003;Schimenti and Handel, 2018).

Primary ovarian insufficiency (POI) is a major cause of female infertility and affects
about 1-3% of women under 40 years of age. It is characterized by cessation of menstruation before the age of 40 years. POI results from a depletion of the ovarian follicle pool and can be isolated or syndromic. Genetic causes of POI account for approximately 20% of cases (Rossetti et al., 2017). Whereas individual infertility-causing pathogenic variants are inherently unlikely to spread in a population, they can be observed within families, especially in the event of consanguinity. They provide crucial insights into the function of the genes and molecular mechanisms that they disrupt. Over the last decade, several causative genes have been found using whole exome sequencing (WES) in POI pedigrees. In particular, pathogenic variants in genes involved in DNA replication, recombination or repair, such as STAG3, SYCE1, HFM1, MSH5 and MEIOB have been formally implicated in this condition by ourselves and others (Caburet et al., 2014;Caburet et al., 2019b;de Vries et al., 2014;Guo et al., 2017;Wang et al., 2014).
In this study, we identified in a consanguineous family with POI a candidate missense variant in HSF2BP, an essential yet poorly-studied meiotic gene. HSF2BP gene encodes a binding protein of the heat shock response transcription factor HSF2 (Yoshima et al., 1998).
During the course of this work and in agreement with our results, two independent groups showed that HSF2BP is essential for meiotic recombination through its ability to interact with BRCA2 (Brandsma et al., 2019;Zhang et al., 2019). Here, we report that the introduction of the missense variant HSF2BP-S167L in the mouse leads to subfertility and DNA repair defects during prophase I. In addition, we identified a protein complex composed of BRCA2, HSF2BP, and the as yet unexplored C19ORF57/MIDAP (meiotic double-stranded break BRCA2/HSF2BP complex associated protein) as a key component of the meiotic recombination machinery. Our studies show that a single substitution (S167L) in HSF2BP leads to a reduction in the loading of both MIDAP and HSF2BP at the recombination nodules. Furthermore, our results suggest that meiotic progression requires a critical threshold level of MIDAP for the ulterior loading of recombinases to the RN.

Clinical cases
The parents are first-degree cousins of Moslem-Arab origin. Of the five daughters, three are affected with POI and presented with early secondary amenorrhea. They had menarche at normal age (at 13-14) but with irregular menstruation that stopped around 25.
Only one of the POI patients could conceive with the help of a fertility treatment (see pedigree in Figure S1). In order to identify the genetic basis of this familial POI case, we performed WES on genomic DNA from two POI patients, III-2 and III-3, and their fertile sister III-10 (Table S1). Variants were filtered on the basis of i) their homozygosity in the patients, ii) their heterozygosity or absence in the fertile sister, iii) their absence in unrelated fertile in-house controls and iv) a minor allele frequency (MAF) below 0.01 in all available databases (Table S2). The identified variant was a missense substitution located in the HSF2BP gene, rs200655253 (21:43630396 G>A, GRCh38). The variant lies within the sixth exon of the reference transcript ENST00000291560.7 (NM_007031.2:c.500C>T) and changes a TCG codon into a TTG (NP_008962.1:p.Ser167Leu). It is very rare (VAF 0.0001845 in the GnomAD database and 0.0005 in the GME Variome dedicated to Middle-East populations) and absent at the homozygous state in all databases. The variant was verified by Sanger sequencing and was found to segregate in a Mendelian fashion within the family: the affected twin III-1 was homozygous for the variant and both parents and fertile siblings were heterozygous carriers ( Figure S2). Serine 167 is a highly conserved position and the S167L variant is predicted to be pathogenic or deleterious by 11 out of the 18 pathogenicity predictors available in dbNSFP 3.5. (Table S3, Figures S3 and S4).
HSF2BP was first identified as a testis-specific interactor of HSF2, able to sequester this transcription factor in the cytoplasm (Yoshima et al., 1998). Another partner, basonuclin 1 (BCN1), has been recently implicated in POI (Tsuiko et al., 2016;Wu et al., 2013;Zhang et al., 2018). We therefore tested a possible impact of the S167L variant in HSF2BP on its interaction with HSF2 and BCN1 by a two-hybrid system in cell culture, but did not observe any difference with the wild-type (WT) protein.

Mice with the HSF2BP S167L variant show a partial reduction of fertility
In order to confirm the causality of the S167L variant in this POI family, we generated a knock-in mouse Hsf2bp S167L/S167L by genome editing ( Figure S5a). We also generated a loss of function model (Hsf2bp -/-) for direct comparison ( Figure S5b-e).
Hsf2bp S167L/S167L male and female mice were able to reproduce but females showed a significant reduction in the number and size of litters ( Figure 1a) whilst males only showed a slight non-significant reduction in fertility (Figure 1a). This is in agreement with the fertility defects observed in Hsf2bp -/mice (Figure 1c, e-f and (Brandsma et al., 2019;Zhang et al., 2019)), and suggests that the S167L variant might impact more specifically female fertility.
Histological analysis of Hsf2bp S167L/S167L ovaries revealed no apparent difference in the number of follicles in comparison with WT animals, in contrast with the drastic reduction of the follicle pool in Hsf2bp -/ovaries (Figure 1b-c). Testes from Hsf2bp S167L/S167L mice displayed a reduced size (only 70% of WT mice: S167L 67,1mg ±10,4 vs 95,6 mg ±10,1 for WT controls; n=6, Figure 1d) and this reduction was higher in Hsf2bp -/testes (25% of the wild-type, Figure 1e). Accordingly, histological analysis of adult Hsf2bp S167L/S167L testes revealed seminiferous tubules with a partial arrest with apoptotic spermatocytes (meiotic divisions) and their epididymis exhibited scarcer spermatozoa ( Figure 1f). Hsf2bp -/males show a total meiotic arrest at epithelial stage IV with massive apoptosis (Figure 1f). The presence of spermatogonia, spermatocytes, Sertoli and Leydig cells was not altered in any of the mutants (Figure 1f). These results suggest that mice bearing the POI-causing variant only partially phenocopy the human disease and fits well with the exacerbated sexual phenotypic dimorphism observed in the Hsf2bp -/mice which might be less pronounced or absent in humans. However, the existence of a mild subfertility in females and of an overt spermatogenic phenotype prompted us to analyse their meiotic progression in more detail.
To characterize the meiotic defect in detail, Hsf2bp S167L/S167L meiocytes were first analyzed for the assembly/disassembly of the SC by monitoring the distribution of SYCP1 and SYCP3. We did not observe any difference in synapsis and desynapsis from leptotene to diplotene in both oocytes and spermatocytes ( Figure S6a-b). However, we observed an elevated number of apoptotic meiotic divisions in Hsf2bp S167L/S167L males (30% in Hsf2bp S167L/S167L vs 5% in WT) as well as some metaphases I with free univalents (Figure 1gh and S7). These results are consistent with the partial arrest observed in the histological analysis ( Figure 1f). As expected, this phenotype was exacerbated in Hsf2bp -/spermatocytes that were arrested at a zygotene-like stage ( Figure S6c) and in the Hsf2bp-deficient oocytes that showed a delay in prophase I progression with the majority of cells at zygotene stage in 17,5 dpp females (whilst the WT oocytes are mainly at pachytene stage). Additionally, we observed increased numbers of oocytes showing synapsis defects in the Hsf2bp -/oocytes (45,5%±1,5 vs 7,5%±1,5 in WT, Figure S6d). These results strongly suggest that the POI variant S167L is a hypomorphic allele.
We next analyzed if the POI-inducing variant affects the loading/stability of HSF2BP by immunolabelling meiocytes from Hsf2bp S167L/S167L mice. We observed a striking reduction of HSF2BP staining at the axes during prophase I in both oocytes and spermatocytes ( Figure   2a-b). This reduction and the specificity of the antibodies used were validated by Western blot of whole testis extracts from wild-type, Hsf2bp S167L/S167L and Hsf2bp -/animals ( Figure   2c) and by immunofluorescence (IF) of Hsf2bp -/spermatocytes ( Figure S5e).
Given that HSF2BP is essential for DNA repair in spermatocytes and that its deficiency provokes the accumulation of DNA repair proteins such as H2AX ('tagging' the DSBs), the single stranded-DNA binding protein RPA, a strong reduction of the recombinases DMC1 and RAD51 at the RNs, and the lack of cross-overs (COs) by labelling MLH1 (Baker et al., 1996) (Figures S8, 3, 4 and 5), we carried out a comparative analysis of these proteins in meiocytes from Hsf2bp S167L/S167L , Hsf2bp -/and WT animals. Our results reveal that Hsf2bp S167L/S167L spermatocytes also showed an increased labelling of H2AX at pachytene (Figure 3a), an accumulation of RPA at the chromosome axis (Figure 3b), a reduction of the recombinases DMC1 and RAD51 staining (Figure 4a-b), and a decreased number of COs (measured as MLH1 or interstitial CDK2 foci, Figure 5a) in comparison with the WT. Consequently with the reduction of COs, we observed the presence of free XY univalents at pachytene in Hsf2bp S167L/S167L males (Figure 5a), which explains the univalents observed at metaphase I and the elevated number of apoptotic cells observed (Figure 1g-h and S7). However, our female analysis showed no accumulation in RPA labelling in Hsf2bp S167L/S167L oocytes (Figure 5b), whereas RPA accumulates in the synapsis-defective cells from KO oocytes ( Figure S8b) and DMC1 shows lower staining in both Hsf2bp S167L/S167L and Hsf2bp -/- (Figures 5c and S8c). MLH1 foci show a strong reduction and no differences (though with a clear trend towards a reduction) in the number of COs in oocytes from Hsf2bp -/and Hsf2bp S167L/S167L , respectively ( Figure 5d). Overall, male and female Hsf2bp S167L/S167L mice show similar molecular alterations although with different severity. In humans, and given the absence of men with the homozygous variant in the POI family ( Figure S1), a similar difference between the sexes cannot be ruled out.
We next sought to understand how the HSF2BP pathogenic variant is mediating the observed meiotic alteration. HSF2BP has been shown to bind BRCA2, an essential protein for homologous meiotic recombination (Martinez et al., 2016;Sharan et al., 2004), in a direct manner tnatahrough its tryptophane at position 200 and the C-term (Gly2270-Thr2337) (Brandsma PhD thesis 2016;(Brandsma et al., 2019)). Given the impossibility to detect endogenous BRCA2 by IF in mouse spermatocytes, we carried out co-localization/interaction assays in a heterologous system by transfecting BRCA2-C (C-term) and HSF2BP in U2OS/HEK293T. The results showed that BRCA2-C co-immunoprecipitates with both HSF2BP-WT and HSF2BP-S167L in similar ways ( Figure S9a). When transfected alone, HSF2BP localization was cytoplasmic and BRCA2-C was nuclear. This pattern changed drastically to a nuclear dotted pattern when co-transfected ( Figure S9b). This re-localization is independent of the HSF2BP variant, strongly suggesting that the HSF2BP variant effects are not directly mediated by BRCA2 delocalization.

MIDAP is a novel interactor of HSF2BP
In order to further understand the mechanism underlying the pathogenicity of the HSF2BP-S167L variant, we searched for proteins that interact with the murine HSF2BP through a yeast two hybrid (Y2H) screening. The analysis of the clones with putative interactors revealed that 19 contained an uncharacterized protein named C19ORF57, herein renamed MIDAP for meiotic double-stranded break BRCA2/HSF2BP complex associated protein. This interactor consists of 600 amino acids with a high content of acidic residues, has no recognizable functional domains and is intrinsically disordered. The interaction was validated by transiently transfecting HSF2BP-and MIDAP-encoding expression plasmids.
Both HSF2BP-S167L and WT interacted with MIDAP ( Figure 6a). To identify the regions required for this interaction, we splitted the MIDAP protein into three fragments (N-terminal, central region and C-terminal). We mapped the HSF2BP/MIDAP-interacting domain to the C-term fragment of MIDAP (spanning residues 475-600 of the murine protein, Figure 6b-c).
Next, we asked whether this interacting domain was sufficient for HSF2BP function in vivo. We generated the corresponding mutant mice (Midap Δ/Δ ), which expressed the We next sought to analyse its function in meiosis through IF by using specific antibodies against MIDAP. MIDAP localized on the chromosome axes of meiocytes from zygotene to pachytene with a pattern of discrete foci that mimics RNs ( Figure S11a-b). In agreement with the Y2H and co-IP results, MIDAP perfectly co-localized with HSF2BP on the chromosome axes ( Figure 6d). This co-localization was verified by super-resolution microscopy ( Figure 6e). To further delineate if HSF2BP and/or MIDAP had DNA-binding activity (targeting to DSBs), we carried out an in vitro binding assay using HSF2BP and MIDAP proteins expressed in a transcription and translation coupled reticulocyte system (TNT) in which there are no nuclear protein and chromatin (Melton et al., 1984) and used RPA as positive control. The results show that both proteins lack DNA-binding abilities in contrast to the strong activity of RPA ( Figure S11c-d).
To determine the role of MIDAP in recombination and DNA repair, we analyzed its cytological pattern of distribution in different synaptic/recombination mutants (REC8, SIX6OS1, RNF212, HEI10 and PSMA8 (from more severe to less severe phenotypes) and in DSBs-deficient spermatocytes (SPO11 mutant; (Spo11 -/-, Rnf212 -/and Hei10 -/described in this work, see methods and Figures S12 and S13)). HSF2BP staining was also carried out for a direct comparison. The results showed that none of the recombination-deficient mutants abrogate MIDAP labelling at zygotene, in contrast to its absence of loading in SPO11deficient mice ( Figure S14, left). These results are very similar to those obtained for HSF2BP in these mutants ( Figure S14, right) and indicate that SPO11-dependent DSBs are essential for targeting HSF2BP/MIDAP to the RNs and both can be genetically positioned at the early events very soon after DSBs generation.
To functionally analyse the role of MIDAP in HSF2BP signalling and fertility, we Immunostaining of Midap -/spermatocytes for H2AX, RPA and the recombinases RAD51 and DMC1 revealed an accumulation of H2AX and RPA on zygonema-like spermatocytes and a drastic reduction of RAD51/DMC1 foci in early and late zygonema (Figure S16a-b and Figure 8c-d). Accordingly with the arrest, MLH1 staining revealed absence of COs ( Figure S17a). These results are similar to the phenotypes described for HSF2BP mutants (Figures 3, 4, 5, S8 and (Brandsma et al., 2019)). We also analyzed the localization of SPATA22, another ssDNA binding protein complexed to RPA during resection, and found a strong accumulation in both HSF2BP and MIDAP null mutants and milder staining in the Hsf2bp S167L/S167L spermatocytes supporting early meiotic defects before strand invasion ( Figure S18). Similar to the HSF2BP, oocytes from Midap -/females show a reduced staining of DMC1 and consequently a reduced number of MLH1 foci (Figure 8e and S17b). Thus, both HSF2BP and MIDAP mouse mutant show a meiotic phenotype with a strong sexual dimorphism in the mouse.
In order to study HSF2BP signalling through MIDAP, we analyzed its localization in Hsf2bp -/and Hsf2bp S167L/S167L spermatocytes and oocytes (Figure 9a-b). Our results showed a total absence of MIDAP loading in Hsf2bp -/meiocytes and a drastic reduction as a consequence of the POI variant (Figure 9a-b), strongly suggesting a causative relationship.

MIDAP and HSF2BP form a multimeric complex with PALB2 and BRCA2
To further delineate the interactome of MIDAP, we immuno-precipitated MIDAP from mouse testis extracts coupled to mass spectrometry (IP-MS). Among the list of putative interactors, we identified HSF2BP as the main hit but also BRCA2, PALB2, RAD51 and RPA, strongly suggesting that they form a large multimeric complex (Table S4 and S5 for extended data). To confirm this result, we transfected the corresponding expression plasmids in HEK293T cells for co-immunoprecipitation analysis. MIDAP co-immunoprecipitated with BRCA2 and HSF2BP when they were all co-transfected but not when co-transfected only with BRCA2 (Figure 10a and S19a). The reciprocal co-IP of BRCA2 with HSF2BP and MIDAP was also positive. However, co-immunoprecipitation of MIDAP with BRCA2 was stronger in the presence of HSF2BP-S167L than in the presence of HSF2BP-WT ( Figure   10a). In addition, we observed modest but positive co-immunoprecipitations of HSF2BP with RPA, PALB2 and RAD51; and of MIDAP with RAD51 and RPA but not with PALB2 ( Figure S19b).
To test more precisely their direct interactions, MIDAP, HSF2BP, PALB2, RPA and RAD51 were produced in a TNT and co-immunoprecipitated. The results showed an absence of direct interaction between any of them with the exception of MIDAP and HSF2BP (as expected from the Y2H analysis) ( Figure S19c). Overall, these results suggest that despite the absence of direct interactions, these proteins are complexed in vivo likely through BRCA2 as observed in immunoprecipitations of testis extracts and provides clues on how the HSF2BP variant could be mechanistically operating by modulating MIDAP interaction with partners of major BRCA2-containing recombination complexes.
Given the close interaction of MIDAP and HSF2BP, we analyzed their interdependence in the heterologous system U2OS. We overexpressed MIDAP, HSF2BP (WT and S167L) and analyzed their cellular localization. Transfected HSF2BP localized to the cytoplasm and nucleus in a diffuse manner ( Figure 10b). However, when HSF2BP was co-overexpressed with MIDAP, its pattern changed drastically to an intense nucleoplasm staining that also revealed the appearance of nuclear invaginations that resemble nucleoplasmic reticulum ( Figure 10b) (Malhas et al., 2011). Interestingly, such invaginations were reduced when MIDAP was co-transfected with HSF2BP-S167L instead of WT ( Figure   10b). In addition, the intensity of the fluorescence signal of the transfected HSF2BP-S167L was lower than for HSF2BP-WT. The fluorescence intensity of HSF2BP (both S167L and WT) was increased when co-transfected with MIDAP ( Figure 10b). To further explore this observation, we analyzed by Western blot the expression levels of HSF2BP-WT and HSF2BP-S167L alone or in combination with MIDAP ( Figure 10c). The results clearly showed that HSF2BP-S167L was expressed at lower amounts than the WT, indicating a reduced protein stability of the S167L variant and are in accordance also with the reduced expression of HSF2BP-S167L in the knock-in mouse in vivo ( Figure 2). Interestingly, the expression of HSF2BP in transfected HEK293T cells is increased when co-transfected with MIDAP (greater effect in the S167L variant) and was partially dependent on proteasomedependent degradation (MG132 treatment, Figure 10c), indicating a direct role of MIDAP in stabilizing HSF2BP. Taken altogether and given the low expression level of MIDAP in the Hsf2bp S167L/S167L , these results suggest that a strong functional interdependence between MIDAP and HSF2BP promotes their lower stability / expression in meiocytes, which might lead to the RPA accumulation, reduced RAD51 and DMC1 loading and consequently reduced COs.

Discussion
Using exome sequencing, we identified the S167L missense variant in HSF2BP in a consanguineous family with three cases of POI with secondary amenorrhea. All affected family members are homozygous for the variant, and the healthy relatives are heterozygous carriers. The causality of the HSF2BP-S167L variant is supported by the meiotic phenotype and the subfertility observed in Hsf2bp S167L/S167L female mice. Furthermore, the DNA repair defects in murine KI spermatocytes, displayed by the accumulation of H2AX, the accumulation of RPA and the loss of loading of the RAD51/DMC1 recombinases on DSBs, provide direct evidence that this missense variant alters recombination. This conclusion was further supported by the comparative analysis of the POI allele with the Hsf2bp null allele that revealed that the missense variant can be considered as an hypomorphic allele of HSF2BP. This is in agreement with the secondary POI observed in the patients, and the residual (medically-assisted) fertility in one of the affected sisters. Our identification of We have shown by IP-MS that HSF2BP together with its novel interactor MIDAP forms a supramolecular complex with PALB2, RAD51, RPA and BRCA2. These interactions possibly are possibly mediated by the multidomain hub protein BRCA2 (Siaud et al., 2011) as we only detected direct interactions of HSF2BP with BRCA2 and MIDAP. In this regard, BRCA2 directly interacts with the DSBs recruiter PALB2, with the recombinases RAD51 and DMC1 (though through different specific domains), with DNA, with the DNAmimicking protein and proteasome subunit DSS1 (which binds RPA complexes), and with HSF2BP (which binds MIDAP) (Siaud et al., 2011). This supramolecular complex participates in the orderly recruitment of essential players to the DSBs such as the initial binding of RPA to the resected DNA, exchange of RPA by RAD51 in a DSS1-dependent manner, and loading of the complex MEIOB-SPATA22 to the RPA complexes (Martinez et al., 2016;Zhao et al., 2015). Interestingly, genes with recently-identified variants in POI patients are implicated in the repair of induced DSBs at the early stages of meiosis and encode BRCA2-interacting factors, such as MEIOB, DMC1 and HSF2BP, or BRCA2 itself (Caburet et al., 2019a;He et al., 2018). This highlights the crucial importance and the high sensitivity of this particular meiotic step, and the hub role of BRCA2 as a tightly-regulated platform for correct meiotic recombination.
We have shown that genetic depletion of HSF2BP and MIDAP leads to meiotic arrest at zygotene-like stage. The arrested spermatocytes are not able to load the recombinases RAD51/DMC1, impairing the repair of DSBs and the generation of Cos, as shown by the absence of MLH1 foci). As a consequence, zygonema-like spermatocytes accumulate the SSB RPA. RPA, as part of a trimeric replication protein complex, binds and stabilizes ssDNA intermediates that forms during DNA repair. In meiosis, RPA is forming a heterodimer with the essential meiotic players MEIOB (homologue of RPA) and SPATA22. However, the loading of the complex formed by MEIOB/SPATA22 to DSBs is RPA-independent (Shi et al., 2019). It is postulated that RPA functions in meiosis at two different stages; i) during the early recombination stages when the DSBs ends are resected by the MRN complex and ii) during the strand invasion into the homologue duplex that is carried out by RAD51 and DMC1, ssDNA is generated at the displacement loops (Shi et al., 2019). The accumulation of RPA observed in the Hsf2bp -/and Midap -/zygonema-like and in the Hfs2bp S167L/S167L spermatocytes is likely to occur at the early stages of recombination, because SPATA22 loading to the DSBs is also increased in the HSF2BP and MIDAP KOs and also to a lesser extent in the Hsf2bp S167L/S167L . We propose that this accumulation is not directly mediated by either HSF2BP or MIDAP, given the absence of direct interaction of HSF2BP/MIDAP to RPA, RAD51 and PALB2. In addition, the observed lack of DNA-binding ability of HSF2BP/MIDAP also points towards a model in which the absence of the heterodimer HSF2BP/MIDAP through a direct interaction with BRCA2 impairs the replacement of RPA by RAD51/DMC1 in the foci that form on the DSBs. Similarly, the POI variant, which does not affect the heterodimerization of HSF2BP/MIDAP, promotes a lower expression of both proteins of the heterodimer. This reduced amount of complex is thus less proficient in replacing RPA by the recombinases RAD51/DMC1 leading to a lower frequency of COs.
The observed sexual dimorphism of HSF2BP and MIDAP mutants has also been described for several other meiotic genes in the mouse (Cahoon and Libuda, 2019). These differences can have a structural basis, given that the organization of the axial elements is known to be different between sexes. This is supported by the difference in length of the axes and by the essential role that the meiotic cohesin subunit RAD21L plays in males but not females (Herran et al., 2011). This distinct axial organization might generate different recombination landscapes in spermatocytes and oocytes. Accordingly, and similar to MIDAP/HSF2BP, the early meiotic recombination protein TEX15 is critical for loading recombinases in males but not in females, strongly suggesting a parallelism in the early recombination pathway where HSF2BP/MIDAP work (Yang et al., 2008). In general, meiotic recombination mutants seem to proceed further in female than in male meiosis perhaps because of the presence of less stringent checkpoints (Hunt and Hassold, 2002). In addition, human female meiosis seems to be especially error-prone (Bennabi et al., 2016) in comparison with other organisms (including the mouse), explaining the increased incidence of aneuploidy in the oocytes of aging women. It has recently been shown that this high level of metaphase I mis-segregation in women is due to a "female-specific maturation inefficiency" . This idea is based on the paradox that despite the total number of COs is higher in females than in males the frequency of bivalents without CO is also higher in women, and could also explain the relative milder phenotype of the mouse Hsf2bp S167L/S167L oocytes compared with that of POI patients.
We have shown that the phenotype of female mice with S167L variants does not show a severe meiotic phenotype but only a slight reduction of fertility and a trend towards a reduction in the number of COs that can lead to univalents at metaphase I. Conversely, male mice bearing the mutation S167L did not show a statistically significant reduction in fertility but did show a reduction in the number of spermatozoa in the epididymis, the presence of apoptotic spermatocytes and a clear meiotic phenotype that resembles the KO phenotype but in a milder manner (reduction of COs that causes the presence of univalents at metaphase I).
Interestingly, it is known that a reduction of the spermatozoa count (up to 60%) does not affect male mouse fertility which can be used to understand the molecular mechanism underlying a mutation (Schurmann et al., 2002). This can explain the normal fertility of male mice bearing the HSF2BP-S167L genetic variant despite a strong meiotic alteration.
Very recently, a high resolution genome-wide recombination map revealed novel loci involved in the control of meiotic recombination and highlighted genes involved in the formation of the synaptonemal complex (SYCE2, RAD21L, SYCP3, SIX6OS1) and the meiotic machinery itself as determinants of COs (Halldorsson et al., 2019). Within the second category, variants of the SUMO ligase RNF212 and the ubiquitin ligase HEI10 have been largely documented as genetic determinants of the recombination rate in humans, and importantly also of HSF2BP. Consequently, gene dosage of RNF212 and HEI10 affect CO frequency through their activity in CO designation and maturation (Lake and Hawley, 2013;Reynolds et al., 2013). We found that both MIDAP and HSF2BP localization are unaffected in the loss-of-function mutants of Rnf212 and Hei10. This significant observation together with the proper co-localization of MIDAP/HSF2BP with RPA allows us to map these proteins upstream in the recombination pathway. Interestingly, a genetic variant of HSF2BP (Gly224Ter) affects recombination rate in males (Halldorsson et al., 2019). Two siblings homozygous for this HSF2BP variant in the analysed Icelandic population were healthy but none of them had descendants, suggesting they were infertile (Halldorsson et al., 2019).
It is interesting to note that some of the genes affecting the recombination rate have also been described as 'fertility genes', such as SYCP3, HFM1 and HSF2BP ((Geisinger and Benavente, 2016;Wang et al., 2014) and this work). Altogether, we propose that different variants of the same meiotic gene (alleles responsible for mild or strong phenotypes) can give rise to either an altered genome-wide recombination rate with no detrimental effect, or cause infertility when the decreased recombination rate falls below the lower limit of one COs per bivalent. In the present POI family, the S167L variant in HSF2BP seems to be under that limit. To our knowledge, HSF2BP is one of the very few human genes with variants known to affect both the genome-wide recombination rate in the human population and meiotic chromosome missegregation (fertility) through a reduction of the recombination rate. Along similar lines, it is conceivable that variants with additive effects (Schimenti and Handel, 2018) can lead to a genome-wide reduction of the recombination rate and thus to aneuploidy and infertility. Specifically, variants in genes involved in meiotic recombination and SC constituents could be responsible for a large fraction of genetic idiopathic infertilities. These variants should be under purifying selection and would be removed or substantially reduced from the population. However, this is not the case for genes with sexual phenotypic dimorphism (Gershoni and Pietrokovski, 2014) as is apparent for a wide number of meiotic genes, including HSF2BP and MIDAP (Cahoon and Libuda, 2019), where individuals of one of the sexes are fertile carriers.
In summary, we describe for the first time a human family where POI co-segregates with a genetic variant in HSF2BP (S167L) in a Mendelian fashion and reveal that this variant promotes a lower expression of the heterodimer formed by HSF2BP and its novel stabilizer and interactor MIDAP. This hypomorphic expression of HSF2BP/MIDAP phenocopy in a mild manner the meiotic arrest observed in mice lacking either of HSF2B or MIDAP.

Whole Exome Sequencing
Written informed consent was received from participants prior to inclusion in the study and the institutions involved. Genomic DNA was extracted from blood samples by standards protocols.
For individuals III-3 and III-10, library preparation, exome capture, sequencing and initial data processing were performed by Beckman Coulter Genomics (Danvers, USA). Exon capture was performed using the hsV5UTR kit target enrichment kit. Libraries were sequenced on an Illumina HiSEQ instrument as paired-end 100bp reads. For individual III-2, library preparation, exome capture, sequencing and data processing were performed by IntegraGen SA (Evry, France) according to their in-house procedures. Target capture, enrichment and elution were performed according to manufacturer's instructions and protocols (SureSelect Human All Exon Kits Version CRE, Agilent). The library was sequenced on an Illumina HiSEQ 2500 as paired-end 75bp reads. Image analysis and base calling was performed using Illumina Real Time Analysis (RTA 1.18.64) with default parameters.

Bioinformatic analysis
For the 3 individuals, sequence reads were mapped onto the human genome build (hg38 / GRCh38) using the Burrows-Wheeler Aligner (BWA) tool. Duplicated reads were removed using sambamba tools. WES metrics are provided in Table S1. The number of variants fulfilling those criteria is provided in Table S2. Visual inspection of the variant was performed using the IGV viewer ( Figure S1).

Sanger Sequencing Analysis
To confirm the presence and segregation of the variant, direct genomic Sanger DNA sequencing of HSF2BP was performed in the patients, the parents and non-affected siblings using specific primers: HSF2BP-EX6F: 5'-CTAGAATCTTCTGTATCCTGCA-3' and HHSF2BP-EX6R2: 5'-GGTCTGGAAGCAAACAGGCAA-3'. The resulting chromatograms are shown in Figure S2.

Predictions of pathogenicity and sequence conservation
The S167L variant was predicted to be pathogenic or deleterious and highly conserved by 11 out of the 18 pathogenicity predictors available in dbNSFP 3.5 (Table S3). Upon verification, it appears that the conflicting interpretation of this variant might stem from the single occurrence of a Leu at this position in zebrafish. As the change in zebrafish is the variant that we have in the human family, we checked all the available sequences (Ensembl Release 99, January 2020, removing the one-to-many relationships). Ser167 is very highly conserved in mammals, birds and reptiles and fish and is present in 208 of 212 orthologous sequences ( Figures S3 and S4).
For the Hsf2bp S167L we introduced a mutation in the mouse counterpart residue (p.Ser171Leu) of the POI mutation found in the clinical case (p.Ser167Leu). However, on this manuscript we will refer to the mutant allele by the acronym of the human mutation (S167L) to simplify.
The ssODN contains the mutation on the corresponding position of the mouse sequence (c.512C>T, p.Ser171Leu, see character in red in Table S7) and the PAM mutations avoiding amino acid changes (see characters in bold in the Table S7). For the Spo11 -/mice generation, the ssODN contains the mutations in the active site (TACTAC>TTCTTC p.YY137-138FF, see Table S7) and the PAM mutations (bold characters in Table S7). In all cases the crRNA and tracrRNA were annealed to obtain the mature sgRNA. A mixture containing the sgRNAs, recombinant Cas9 protein (IDT) and the ssODN (30 ng/μl Cas9, 20 ng/μl of each annealed sgRNA and 10 ng/μl ssODN) were microinjected into B6/CBA F2 zygotes (hybrids between strains C57BL/6J and CBA/J) (Singh et al., 2015) at the Transgenic Facility of the University of Salamanca. Edited founders were identified by PCR amplification (Taq polymerase, NZYtech) with primers flanking the edited region (see Table S8 for primer sequences). The PCRs products were direct sequenced or subcloned into pBlueScript (Stratagene) followed by Sanger sequencing, selecting the founders carrying the desired alelles. The selected founders were crossed with wild-type C57BL/6J to eliminate possible unwanted off-targets.
Heterozygous mice were re-sequenced and crossed to give rise to edited homozygous.
Genotyping was performed by analysis of the PCR products produced from genomic DNA extracted from tail biopsies. The primers and the expected amplicon sizes are listed in the welfare. Blinded experiments were applied when possible but for the Hsf2bp -/and Midap -/mutants it was not possible since the phenotype was obvious between wild-type and these mutant mice for all of the experimental procedures used. No randomization methods were applied since the animals were not divided in groups/treatments. The minimum size used for each analysis was two animals/genotype. The mice analyzed were between 2 and 4 months of age, except in those experiments where is indicated.
Histology. For histological analysis, after the necropsy of the mice their testes or ovaries were removed and fixed in Bouin´s fixative or formol 10%, respectively. They were processed into serial paraffin sections and stained with haematoxylin-eosin (ovaries) or Periodic acid-Schiff (PAS) and hematoxylin (testes). The samples were analysed using a microscope OLYMPUS BX51 and images were taken with a digital camera OLYMPUS DP70.
Fertility assessment. Hsf2bp +/+ and Hsf2bp S167L/S167L males and females (8 weeks old) were mated with wild type females and males, respectively, over the course of 4-12 months. 6 mice per genotype (7 mice for Hsf2bp S167L/S167L females) were crossed. The presence of copulatory plug was examined daily and the number of pups per litter was recorded.
Immunocytology and antibodies. Testes were detunicated and processed for spreading DNA pull down assay. ssDNA/dsDNA pull down assays were performed using the protocol previously described by (Souquet et al., 2013 Then the beads were washed three times ( antibodies were used at 1:5000 dilution. Antibodies were detected by using Immobilon Western Chemiluminescent HRP Substrate from Millipore.

Testis immunoprecipitation coupled to MS/MS analysis. 200 µg of antibodies R1 and R2
(against HSF2BP and MIDAP) were crosslinked to 100 ul of sepharose beads slurry (GE Healthcare). Testis extracts were prepared in 50mM Tris-HCl (pH8) Potential contaminants, reverse decoy sequences and proteins identified by site were removed. Proteins with less than two unique peptides in the Ab1 and Ab2 groups were not considered for ulterior analysis. Proteins with less than two unique peptides in the control group and more than two in both groups Ab1 and Ab2 were selected as high-confidence candidates (group Ab1 and Ab2 only). An additional group of putative candidates was selected for those proteins with two or more unique peptides in one of the Ab1 or Ab2 groups and no unique peptides in the control sample (groups Ab1 only and Ab2 only, respectively).

Statistics.
In order to compare counts between genotypes, we used the Welch´s t-test (unequal variances t-test), which was appropriate as the count data were not highly skewed (i.e., were reasonably approximated by a normal distribution) and in most cases showed unequal variance. We applied a two-sided test in all the cases. Asterisks denote statistical significance: *p-value <0.05, **p-value <0.01, ***p-value<0.001 and ****p-value<0.0001. Table S1 shows the WES and mapping metrics for the 3 genomic samples. Table S2 shows the numbers of variants from the WES analysis and passing the various filters. Table S3 shows the predictions of pathogenicity and conservations by 18 computational predictors. Tables S4 and S5 show the candidate MIDAP interactors identified by IP-MS. Table S6 and S7 show respectively the crRNAs and the ssODN employed in the generation of the various mouse models. Table S8 shows the primers and expected product sizes for genotyping mouse models. Figure S1 shows the pedigree for the consanguineous family with 3 POI cases. Figure S2 shows the chromatograms obtained by Sanger sequencing of the HSF2BP-S167L variant in the family. Figure S3 and S4 show the conservation of the S167 position in mammals, birds and reptiles, and fish. Figure S5 shows the generation and genetic characterization of Hsf2bp Ser167Leu and Hsf2bp-deficient mouse models. Figure S6 shows the absence of synaptic defects in Hsf2bp S167L/S167L meiocytes. Figure S7 shows apoptotic metaphases I in Hsf2bp S167L/S167L males. Figure S8 shows the defective DNA repair in

Online Supplemental material
Hsf2bp -/mice. Figure S9 shows Comparative interaction of HSF2BP-S167L and HSF2BP-WT with BRCA2. Figure S10 shows the absence of MIDAP and HSF2BP loading defects in the Midap Δ142-472 mutant. Figure S11 shows the localization of C19orf57/MIDAP at meiotic recombination nodules and the absence of DNA binding abilities in both HSF2BP and MIDAP. Figure S12 and S13 show the generation and genetic characterization of Spo11 -/-, Rnf212 -/and Hei10 -/mice. Figure S14 shows the loading of MIDAP and HSF2BP on in different synapsis/recombination-defective mutant mice. Figure S15 shows the generation and genetic characterization of Midap -/mice. Figure S16 shows the accumulation of H2AX and RPA in Midap -/males. Figure S17 shows the defect on MLH1 localization in Midap -/mice. Figure 18 shows the accumulation of SPATA22 in Midap -/-, Hsf2bp S167L/S167L and Hsf2bp -/mice. Figure S19 shows the co-immunoprecipitations between HSF2BP, MIDAP, RPA, RAD51 and PALB2. (a) Fertility assessment of Hsf2bp S167L/S167L mice were performed by crossing Hsf2bp +/+ and Hsf2bp S167L/S167L males and females (8 weeks old) with wild type females and males, respectively. The number of pups per litter was recorded. Both mutant males (right plots) and

Figure 5. Recombination is clearly affected in
Hsf2bp S167L/S167L males and partially affected in females. (a) Double immunofluorescence of MLH1 (green) and SYCP3 (red) in spermatocyte spreads from Hsf2bp S167L/S167L and Hsf2bp +/+ . MLH1 foci are significantly reduced in the Hsf2bp S167L/S167L spermatocytes and showed XY univalents at pachytene stage (white rectangle, see also figure 1g). The plot on the right shows the quantification of MLH1 foci in both genotypes. (b) Double immunolabeling of RPA (green) and SYCP3 (red) in oocyte spreads from Hsf2bp +/+ and Hsf2bp S167L/S167L females showing normal RPA localization and number of foci. Plot on the right of the panel represents the quantification of RPA foci on each genotype and stage. (c) Double immunofluorescence of DMC1 (green) and SYCP3 (red) in Hsf2bp +/+ and Hsf2bp S167L/S167L oocytes showing a reduction in the number of DMC1 foci. Plot on the right of the panel represents the quantification of DMC1 foci on each genotype and stage. (d) Double immunolabeling of MLH1 (green) and SYCP3 (red) in oocyte spreads from Hsf2bp +/+ and Hsf2bp S167L/S167L mice (19.5 dpp) showing no significant differences between both genotypes in the number of cross-overs (see plot under the panel for quantification). Bar in panels, 10 μm. Welch´s t-test analysis: ** p<0.01, **** p<0.0001.    oocytes from Hsf2bp +/+ , Hsf2bp S167L/S167L and Hsf2bp -/showing a strong reduction of MIDAP staining in the S167L mutant and a total absence in the Hsf2bp knock-out. Plots under the panel represent the quantification of the number of MIDAP foci or intensity on each genotype and stage. Bar in panels, 10μm. Welch´s t-test analysis: * p<0.05,**** p<0.0001.  (WT and S167L) alone or with GFP-MIDAP. Additionally, cells transfected with Flag-HSF2BP were treated with the proteasome inhibitor (MG132, 10μM) during 4 hours and analysed by immunoblotting with a mouse anti-Flag antibody. Cherry was used as transfection efficiency control. HSF2BP-WT was expressed at higher levels than HSF2BP-S167L suggesting a reduced stability of the S167L variant. The level of HSF2BP expression (both the WT and the S167L variant) was increased when co-transfected with MIDAP. This increase of expression was greater in the HSF2BP-S167L variant in comparison with the WT (4,9 times in WT vs 8,6 times in S167L, see plots under the blot). After incubation with the proteasome inhibitor MG132, the expression levels of transfected HSF2BP were also increased mimicking the effect of MIDAP. This effect is also greater in the S167L variant than in the WT (2,9 times in WT and 4,7 in the S167L, see plot on the right). Welch´s t-test analysis: * p<0.05, ** p<0.01, *** p<0.001, **** p<0.0001. Figure S1. Pedigree of the consanguineous family with the variant HSF2BP-S167L.

Legends to Supplementary figures
III-1 and III-2 are monozygotic twins, who appear phenotypically dizygotic. Clinical investigation confirmed POI, with normal 46, XX karyotype (500 bands and SKY spectral karyotyping).
Year of birth and age of menarche are indicated when known.
Hsf2bp -/males show a spermatogenic arrest at zygotene-like displaying synapsis between non-homologous chromosomes (partner switch phenotype). (d) Immunofluorescence of oocyte spreads (17,5 dpp) of wild-type and Hsf2bp -/with SYCP3 (red) and SYCP1 (green). Hsf2bp -/females show a delay in prophase I progression with an increase of cells with synapsis defects. Bar in panels, 10 µm. Figure S7. Metaphase I defects in Hsf2bp S167L/S167L males. Co-labeling of SYCP3 (green) and ACA (red) in squashed preparations of tubules from WT and S167L mice.
In the mutant mice, a large number of metaphases I were apoptotic with partial or total loss of immunoreactivity for the labelled proteins. See quantification of apoptotic cells in Figure 1h. Bar in panel, 10 µm.  encoding Flag-HSF2BP (WT or S167L) and EGFP-BRCA2-C alone or together and immuno-detected with antibodies against Flag (Flag-HSF2BP, red) and against EGFP (EGFP-BRCA2-C, green). Transfected HSF2BP alone (both WT and S167L version) is delocalized and labels the whole cell (S167L less intense) whereas BRCA2-C shows a nuclear localization. When we co-transfected BRCA2-C with HSF2BP or HSF2BP-S167L they both gave rise to a similar nuclear punctate pattern (see quantification in the graph on the right of the panel). Bar in panel, 20µm. Welch´s t-test analysis: * p<0.05; *** p<0.001.        Midap -/and Hsf2bp S167L/S167L was done in parallel. For this reason the data for the WT are the same on this figure and the figure 5d. Welch´s t-test analysis: **** p<0.0001. Figure S18. The ssDNA binding protein SPATA22 is accumulated in Hsf2bp -/-, Hsf2bp S167L/S167L and Midap -/mutants. Double labelling of SPATA22 (green) and SYCP3 (red) in spermatocyte spreads from WT, Midap-and Hsf2bp-deficient mice and in the Hsf2bp S167L/S167L mutant. SPATA22 is highly accumulated in both knock-out spermatocytes and shows a milder accumulation in the Hsf2bp S167L/S167L spermatocytes.    Table S6. crRNAs employed for the generation of the different mouse models.