Rare DMRT1 regulatory variants in severe spermatogenic failure

Abstract

The DMRT1 (doublesex and mab-3 related transcription factor 1) gene has long been linked to sex-determining pathways across vertebrates and is known to play an essential role in gonadal development and maintenance of spermatogenesis in mice. In humans, the genomic region harboring the DMRT gene cluster has been implicated in disorders of sex development and recently DMRT1 deletions were shown to be associated with non-obstructive azoospermia (NOA). In this work we have employed different methods to screen a cohort of Portuguese NOA patients for DMRT1 exonic insertions and deletions (by MLPA; n=68) and point mutations (by Sanger sequencing; n=155). We have found three novel patient-specific non-coding variants in heterozygosity that were absent from 357 geographically matched controls. One of these is a complex variant with a putative regulatory role (c.-223_-219CGAAA>T), located in the promoter region within a conserved sequence involved in Dmrt1 repression. Moreover, while DMRT1 domains are highly conserved across vertebrates and show reduced levels of diversity in human populations, 2 rare synonymous substitutions (rs376518776 and rs34946058) and 2 rare non-coding variants that potentially affect DMRT1 expression and splicing (rs144122237 and rs200423545) were overrepresented in patients when compared to 376 Portuguese controls (301 fertile and 75 normozoospermic). Overall our previous and present results suggest a role of changes in DMRT1 dosage in NOA potentially also through a process of gene misregulation, even though DMRT1 deleterious variants seem to be rare.

Introduction

Male infertility is a complex phenotype often arising from the inability of men to produce viable spermatozoa capable of fertilization. Non-obstructive azoospermia (NOA) is a severe form of male infertility characterized by the lack of sperm in the ejaculate, which results from conditions other than hypothalamic-pituitary disease or obstruction of the male genital tract (Jungwirth et al. 2013). It is estimated that NOA accounts for 10-15% of the cases of male infertility, affecting 1% of male individuals worldwide (Practice Committee of American Society for Reproductive Medicine in collaboration with Society for Male Reproduction and Urology 2008). Importantly, over the past years several studies have been showing that a significant portion of the genes involved in severe spermatogenic failure (SSF) are spread throughout the genome rather than restricted to the sex chromosomes [Reviewed in (Lima & Lopes 2014)]. One particular genomic region (9p), harboring the doublesex and mab-3 related transcription factor (DMRT) gene cluster, has been linked with disorders of sex development (Õunap et al. 2004, Raymond et al. 1999, Tannour-Louet et al. 2010, Veitia et al. 1997). It appears that isolated genetic defects in DMRT1, one of the genes in the cluster, may be the cause of alterations in gonadal development in some cases (Ledig et al. 2010, 2012) but not all, as no inactivating mutations or deletions were found in this gene in a study of patients with gonadal dysgenesis (Machado et al. 2012). However, we have recently identified recurrent 9p24.3 deletions spanning DMRT1 in patients with idiopathic azoospermia, suggesting that loss of function of this gene is a risk factor and potential genetic cause of human SSF (Lopes et al. 2013).

DMRT proteins are transcription factors that bind to the promoters of target genes through their zinc-finger DM (doublesex and mab-3)-domain. This DNA-binding motif is highly conserved and requires dimerization for recognition of a target consensus palindromic sequence (Erdman et al. 1996). Also, in association with the DM-domain, the DMRT protein family shares a less conserved DMRT1-like domain of yet uncharacterized function. Particularly, DMRT1 has been extensively studied in vertebrates and is known to be crucial in the pathway of sex determination and gonadal development in most, if not all, species of this group [Reviewed in (Zarkower 2013)]. Indeed DMRT1 has acquired specialized functions in different clades. It is essential for male gonad differentiation but not sex determination in mice, showing testis specific expression restricted to Sertoli and premeiotic germ cells. A similar pattern is observed for human DMRT1 with the exception that it is also part of the network of sex determination [Reviewed in (Bratuś 2012)], being co-expressed with SRY (sex determining region) in the genital ridge of the male embryo (Moniot et al. 2000). Interestingly, a total of three cDNA isoforms have been detected in the human testis - DMRT1a, DMRT1b, DMRT1c - and result from the inclusion of intronic regions and Alu elements (Cheng et al. 2006), suggesting that the transcriptional diversity of this gene might be involved in the regulation of spermatogenesis.

Bearing in mind the expression pattern of DMRT1, its crucial role in the progression of spermatogenesis and previous findings of an association of DMRT1 defects with SFF, we sought to evaluate the contribution of sequence variants in the promoter and coding regions of the DMRT1 gene to NOA phenotypes. Here, we describe novel non-coding variants and rare non-coding and synonymous variants overrepresented in NOA patients when compared to Portuguese fertile and normozoospermic controls, which are predicted to affect the regulation of the canonical DMRT1 transcript (DMRT1a) and alternative splicing.

Materials and Methods

Patient and Control Populations

DNA samples of 155 male individuals with idiopathic non-obstructive azoospermia (NOA) were collected at the Genetics Department from INSA-IP and at the Genetics Department from FMUP. Patients with known causes of infertility, including chromosome anomalies and Yq microdeletions were excluded from this study. The majority of these samples (n=110) had been analyzed for CNVs by SNP microarrays (Lopes et al., 2013) but none presented CNVs within DMRT1.

As controls, we obtained DNA of 376 Portuguese men: 75 normozoospermic (normal sperm parameters) and 301 males who fathered at least one child.

Ancestry of one NOA patient harboring two potential regulatory variants was determined by genotyping of 46 ancestry informative markers (insertions and deletions). Biogeographical ancestry was assigned using Snipper 2.0 (http://mathgene.usc.es/snipper/; See supplementary data and supplementary Fig.1)

The study was included in the project ‘Copy number variation in infertile men genomic regions: screening in the Portuguese population’ (PTDC/SAU-GMG/101229/2008), approved by the INSA Ethics Committee (Lisbon, Portugal on 6 November 2007). Molecular studies were performed after informed consent and all DNA samples were anonymized.

Analysis of DMRT1 Sequence

Multiplex Ligation Probe Assay (MLPA)

To detect deletions/duplications in the coding region, MLPA was performed in 68 randomly picked NOA patients. SALSA MLPA P334 Gonadal probemix (MRC-Holland, Amsterdam, the Netherlands), which contains two probes for each exon of the gene, was employed according to the manufacturer's instructions. Typically 150-200 ng of genomic DNA were used for amplification.

Sanger Sequencing

Primers for amplification and sequencing (Supplementary Table 1) of the DMRT1 genomic reference sequence (NM_021951.2) were designed in Primer3 v.0.4.0 (Koressaar & Remm 2007, Untergasser et al. 2012). DNA fragments were amplified by standard PCR, sequenced with the BigDye Terminator v.3.1 Cycle Sequencing Kit (Applied Biosystems®, Life Technologies, Carlsbad, California, USA) and run on an ABI PRISM 3130xl Genetic Analyzer (Applied Biosystems®, Life Technologies, Carlsbad, California, USA). Due to low DNA quality, in 23 patients only a partial gene sequence was obtained, which was taken into account in the variant frequency calculations, performed independently for each exon (See supplementary data). All individuals showing novel/rare variants had the full DMRT1 sequence analyzed. Sequences were assembled and analyzed using Geneious version 5.5.8 (Biomatters, Auckland, New Zealand) and all putative polymorphisms were manually inspected and individually confirmed.

Genotyping of Portuguese fertile and normozoospermic controls

Rare variants (<1%) in European populations which were overrepresented in patients and novel mutations identified exclusively in NOA cases were genotyped in our Portuguese controls (Tables 1 and 2). The presence of two indels (c.354+38_insG with a juxtaposed poly-T and rs59834456) was detected by fragment size separation in polyacrylamide and agarose gels respectively; variants in the promoter, 5′UTR and CDS of Exon 1 were genotyped by Sanger sequencing; 6 other variants were analyzed in a single SNaPshot Multiplex reaction. All variants detected by SNaPshot (SupplementaryFig.2) were confirmed by Sanger sequencing. See extended materials and methods section and Tables 1-3 from Supplementary material for more details on experiments and primers used.

Table 1. Novel DMRT1 non-coding variants restricted to patients.

Name	Genomic location	Genomic context	Variant Type	Minor allele	MAF NOA	Global MAF	MAF PT Controls
c.-223_-219CGAAA>T	chr9:841616-841620	Promoter Region	Upstream gene variant	T	1/278	0	0/714
c.355-6T>C	chr9:846954	Intron 1	Intronic (Splice region)	C	1/280	0	0/700
c.823-64_823-62delATT	chr9:916699-916701	Intron 3	Intronic	-ATT	1/290	0	0/660

MAF: Minor allele frequency

Table 2. Annotated variants detected in the cohort of Portuguese NOA patients.

Reference name	Genomic location	Genomic context	Variant Type	Minor allele detected	Patients MAF a	Global MAF b	EUR MAF c	PT Controls MAF d	p-value*
rs144122237	chr9:841785	5′ UTR	5′ UTR variant	T	0.0036	0.004	0.0000	0.0000	0.2875
rs3739584	chr9:841825	5′ UTR	5′ UTR variant	T	0.1475	0.141	0.1378	0.1362	.
rs3739583	chr9:841971	Exon 1	Missense	A	0.1475	0.132	0.1227	0.1344	.
rs376518776	chr9:842051	Exon 1	Synonymous	A	0.0036	0.000	0.0006	0.0000	0.3440
rs55905583	chr9:846952	Intron 1	Intronic (Splice Region)	C	0.0071	0.011	0.0006	0.0057	1.0000
rs2273929	chr9:847208	Intron 2	Intronic	G	0.4607	0.417	0.5409	.	.
rs16925431	chr9:894036	Exon 3	Synonymous	C	0.0109	0.032	0.0222	.	.
rs146975077	chr9:894147	Exon 3	Synonymous	C	0.0036	0.009	0.0002	0.0014	0.4866
rs34946058	chr9:894156	Exon 3	Synonymous	G	0.0109	0.042	0.0007	0.0043	0.3596
rs59168737	chr9:894224	Intron 3	Intronic	A	0.0109	0.039	0.0220	.	.
rs1033836	chr9:894297	Intron 3	Intronic	A	0.0471	0.101	0.0409	.	.
rs112866575	chr9:967939-967940	Intron 4	Intronic	+CTCCCTTT	0.0229	0.149	0.0985	.	.
rs200423545	chr9:967959	Intron 4	Intronic	A	0.0099	0.002	0.0024	0.0057	0.4291
rs376062302	chr9:967953-967954	Intron 4	Intronic	+TTCTCTCT	0.0915	0.1236 e	n.a	.	.
rs279895	chr9:967981	Intron 4	Intronic (Splice Region)	G	0.1765	0.111	0.1658	.	.
rs79358387	chr9:968150-968151	3′ UTR	3′ UTR variant	+T	0.1144	0.223	0.1135	.	.
rs279894	chr9:968334	3′ UTR	3′ UTR variant	G	0.1800	0.077	0.1541	.	.

MAF: Minor allele frequency detected in:

^athe NOA Portuguese cohort

^ball populations screened in large genome sequencing projects

^cEuropean populations of the 1000 Genomes (EUR and CEU), ESP (European-American) and CLINSEQ (CSAgilent)

^dthe Portuguese (PT) control population

^econtrol population used by Machado et al. 2012

^*Fisher's exact test using only the PT population as control

Note: rs59834456 was detected in two Bantu individuals but no population data is available in the Ensembl release of March 2015

In silico and Statistical Analysis

Sequence data were retrieved from the 1000 genomes Project (Genomes Project Consortium et al. 2012), NHLBI GO Exome Sequencing Project (ESP) and CLINSEQ projects to check for all the reported variants in the regions screened and used to set a group of unphenotyped controls of European ancestry (Supplementary material and Supplementary Table 4).

Allele counts of rare variants that were overrepresented in patients were compared to the Portuguese control population applying Fisher's exact test, using RStudio v.0.98.1080 (R Development Core Team 2012).

Also, we predicted in silico the effect of novel non-coding variants in mRNA splicing [Human Splicing Finder v.2.4.1 (Desmet et al. 2009); BGDP: Splice Site Prediction by Neural Network] and the transcription factor binding sites (TFBSs) within the promoter [MatInspector, included in the Genomatix software suite (Genomatix Software GmbH, Munich, Germany); TFSEARCH 1.3, available at http://www.cbrc.jp/research/db/TFSEARCH.html.

We have retrieved the protein sequences of 12 vertebrates and C. elegans from Ensembl database of DMRT1 orthologs. Manually curated protein sequences were then aligned using ClustalW.

Results

In order to characterize the mutational spectrum of potentially deleterious DMRT1 variants in men with idiopathic SSF, we performed a survey of intragenic deletions/duplications by MLPA and point mutations by Sanger sequencing. By MLPA we aimed to find intragenic DMRT1 rearrangements that might be more frequent than the large CNVs previously described in American and Chinese NOA patients (Lopes et al. 2013). No alterations were detected by MLPA after screening approximately half of our patient cohort (n=68) and therefore we conclude that, if present, intragenic DMRT1 rearrangements are also rare in our population.

A New DMRT1 Promoter Variant Restricted to NOA

Given the potential deleterious effect of base substitutions and smaller deletions/duplications, we have screened the DMRT1 sequence of 155 NOA patients by Sanger sequencing. In total we detected 4 novel non-coding variants in heterozygosity in patients (Supplementary Table 2), three of which were absent from Portuguese controls (Table 1). All novel variants were further analyzed in silico for a regulatory role (see Methods and Supplementary material) and those with a predicted effect are described.

Interestingly, one of the variants detected only in a single NOA patient presents a complex genotypic configuration and is located within a region of the DMRT1 known to be involved in the transcriptional repression of the gene in mice (Lei & Heckert 2002). We identified, by allele-specific amplification followed by Sanger sequencing, a single variant (c.-223_-219CGAAA>T; Table 1) in one haplotype (i.e. inherited from the same parent, Fig. 1A) that most likely resulted from a complex mutation event. Importantly, this variant was not present in a sample of 357 geographically matched control individuals (301 fertile and 56 normozoospermic men). To assess its potential impact, we performed a more refined in silico analysis of the human DMRT1 promoter (HG_KWN:62427), located within a CpG-island (chr9:840690-842192; NCBI37/hg19), a feature often associated with TATA-less promoters. Typically the core promoter activity and basal gene transcription of TATA-less genes is regulated by an initiator element (Inr), with a consensus sequence Py Py A+ 1N T/A Py Py (Smale & Baltimore 1989), which we have found in the promoter region of DMRT1 (Fig. 1A) within the canonical transcription start site.

Figure 1. DMRT1 promoter sequence in the proximal region of the transcription start site (TSS).

The DMRT1 promoter region in humans (A) harbors an Inr element (purple filled box), which is characteristic of TATA-less genes, and also one region associated with DMRT1 repression from -88 to -63 bp upstream the TSS (green filled boxes). We have found by allele specific amplification a complex variant in the promoter of one azoospermic patient (A; red filled boxes). Analysis of the proximal region of TSS has predicted two binding sites for HSF1 within the conserved motif (A; blue boxes), one of which is disrupted by this novel patient-specific variant. Aligning the human sequence of this motif with the sequences from several species representative of mammalian clades (B) shows a high degree of conservation for this region (green filled boxes). Sequence annotations were performed using Geneious v.5.5.8 software and alignments were obtained applying the ClustalW algorithm available in the Geneious v.5.5.8 software.

In the alignment of the DMRT1 promoter sequences from several species representative of different mammalian clades (Fig. 1B), a high degree of conservation is apparent, in accordance with previous findings (Lei & Heckert 2002). Moreover, no variants have been described in this region in human populations (chr9:841602-841626; GRCh37/hg19 assembly). In addition, we found that the G nucleotide (chr9:841617; hg19 assembly), deleted in the patient presenting the c.-223_-219CGAAA>T variant, is only found in the human sequence, whereas the ancestral allele A is present in all the other species. A screening for TFBSs within this conserved region of the DMRT1 promoter was performed using two bioinformatics tools: MatInspector and TFSEARCH 1.3 (see Methods and Supplementary Table 5). In the human reference sequence, binding of Heat-shock factors 1 (HSF1; + strand) and 2 (HSF2; - strand) was predicted by both programs. When the human-specific G is replaced by the ancestral A at this position or by the sequence variant found in the patient, a disruption of the HSF1 binding site (Fig. 1A) is predicted, while the score for HSF2 binding remains unchanged. Moreover, in the proximal region of the DMRT1 transcription start site, the human reference sequence contains a second binding site for HSF1 (HSF and TFSEARCH 1.3 predictions) that is conserved between human (chr9:841,690-969,090; GRCh37/hg19 assembly) and mouse (chr19:25,505,706-25,604,328; GRCm38/mm10 assembly).

Another patient was heterozygous for the insertion of a G nucleotide (c.354+38_insG) and for repeat length in a juxtaposed poly-T (>10 T) in intron 1. While this poly-T region seems to have small length variability in both Portuguese patients and controls, it is ≥ 10 bp longer than the reference sequence on both chromosomes of this patient. Length variability in this region alters the sequence of a DMRT1 alternative transcript (DMRT1c) that has been previously detected in human testis (Cheng et al. 2006) and may lead to a frameshift in the corresponding protein isoform depending on the number of repeats.

Rare DMRT1 Variants May Play a Role in Splicing Regulation

A subset of variants identified in the NOA cohort (n=18) had been previously found in unphenotyped individuals from large genome sequencing projects (Table 2). We obtained the frequency of all DMRT1 variants detected and set a control group of populations of European ancestry (See Supplementary material). Since any genetic variant leading to NOA should be rare in the population, we have selected those at ≤1% in European populations and overrepresented in patients for genotyping in the Portuguese controls (Table 2; rs3739584 and rs3739583 were genotyped in the same assay as rs144122237 and rs376518776). While only the rs59834456 deletion was present in Portuguese normozoospermic controls, two variants (one synonymous substitution and a 5′UTR variant) were also absent from fertile individuals with unknown sperm counts (Supplementary Table 2). The synonymous substitution, rs376518776 (c.213G>A), juxtaposed to the DNA binding DM domain, was found in heterozygosity in a single NOA patient. It showed a 6-fold increase in frequency in the NOA Portuguese cohort when compared to unphenotyped database controls of European ancestry and was absent from fertile and normozoospermic Portuguese controls. In silico predictions using the Human Splicing Finder software [HSF; (Desmet et al. 2009)] suggest that this substitution creates a new branch point site, a cis-element required for spliceosome assembly, with a score (89.44) above the average of naturally occurring branch point sequences [83.4±8.6; (Desmet et al. 2009)]. This new branch point site may interfere with the interplay of splicing elements in its vicinity and potentiate the occurrence of new alternative transcripts.

The single rare 5′-UTR variant (rs144122237; c. -54C>T) resides within a putative binding site for the transcription factor RFX3 (Regulatory factor X, 3; Supplementary Table 6) and was found in one NOA patient (0.34%) but was absent from Portuguese fertile and normozoospermic controls and from European populations. Given the presence of the same allele in Africans at low frequency (1.8%) we genotyped a panel of 46 ancestry informative markers (Pereira et al. 2012), and assigned an African ancestry for this patient (LR African versus European= 7.712×10²²; see supplementary data). This was the same individual who harbored two longer poly-T alleles (>10bp) within intron 1, which may reflect his African genetic background.

Other coding variants overrepresented in patients are rs146975077 (c.774G>C) and rs34946058 (c.783C>G), the latter possibly interfering with splicing. In fact, using the HSF software two novel splicing enhancer motifs were predicted for the binding of SC35 (score 82.18) and SRp40 (score 79.28), both with scores above the respective thresholds (75.05 and 78.08). These proteins are members of the SR protein family known to be involved in the regulation of alternative splicing (Liu et al. 1998). Nonetheless, this substitution was present in 2 Portuguese fertile controls, one of them homozygous for the minor allele.

Three rare (≤1%) intronic variants found in patients were also detected in unphenotyped individuals of large genome sequencing projects and are also predicted to affect splicing: rs55905583 (c.355-8G>C) increases the likelihood of expression of the canonical isoform, rs59834456 (c.822+141_822+144delATAT) creates a new enhancer site for SRp40 (score 91.86) and rs200423545 (c.968-26T>A) creates a new branch point in the vicinity of exon 5 (HSF score of 95.11). While rs200423545 is present at nearly twice the frequency in patients compared to Portuguese controls, rs55905583 and rs59834456 are found at similar or higher frequencies in the Portuguese controls (Table 2).

DMRT1 Domains are Highly Conserved in Mammals and Show Lower Genetic Variation in Controls

Even though DMRT1 is known to be involved in the sex determination pathway from C. elegans to humans, there is much uncertainty regarding the role of each functional domain. In order to grasp the potential functional impact of DMRT1 variants, we analyzed the spectrum of coding variants found in unphenotyped control populations (Supplementary Table 7). In Figure 2 we show the alignment of the available vertebrate DMRT1 sequences, and the invertebrate C. elegans, with variants found in human populations marked along the sequence. The zinc finger-like DM domain is involved in binding to the promoter of its target genes and therefore missense variants in this region are expected to have a functional impact. Indeed, only one missense variant (rs201947617) is described in the DM domain, in a less conserved region of the sequence. The DMRT1-like domain, typically found in association with the DM domain but of unknown function, is also conserved in mammals, even though some sequence variability can be found in the central region. Missense mutations within the conserved region are predicted to be deleterious, with an average frequency in the populations surveyed (0.0307%) similar to those without predicted functional impact on the protein (missense tolerated, 0.0266%; and synonymous variants, 0.0225%). Outside the domains, in regions not predicted to affect protein function, missense as well as synonymous substitutions are rare but more frequent (1.75% and 1.23% respectively) than potentially damaging missense variants (0.026%), which present a similar frequency within the domains (0.028%).

Figure 2. Alignment of DMRT1 protein sequences from several species representative of vertebrates clades, including human.

A high degree of identity is notably shared by the DM domain within vertebrates. Actually, a stretch of 83 amino acids containing the DM and part of the DMRT1-like domains seems to be fairly maintained amongst these clades. Interestingly, the DMRT1-like domain is greatly conserved in its extremities and more variability can be found at its center. This alignment was obtaining applying the ClustalW algorithm available with the Geneious v5.5.8 software. Amino acids are colored when they match at least 75% of the sequences. Purple filled boxes mark the domains; yellow and red filled boxes represent missense variants annotated in the Ensembl database, respectively predicted by at least one software (SIFT or PolyPhen) as not affecting the protein or as deleterious.

Compared to the rest of the DMRT1 coding sequence (67/729 = 0.0919) domains also show less variants per base pair (DM: 3/141 = 0.0212; DMRT1-like: 17/252 = 0.0675), likely reflecting functional constraints. Also, the high conservation of a stretch of 83 amino acids comprising the DM and part of the DMRT1-like domains within vertebrates supports their importance for protein function (Fig. 2A). Another region (19 amino acids) in the vicinity of DMRT1-like domain stands out in the alignment as highly conserved in mammals, also exhibiting shared homology with avian DMRT1 (Supplementary Fig. 3).

Discussion

The role of DMRT1 orthologs in sex determination as well as in gonadal development and maintenance throughout evolution is well established in the literature. In fact, DMRT1 is crucial in the maintenance of the mammalian gonad and required for normal progression of spermatogenesis. Expression of Dmrt1 is necessary in Sertoli cells to prevent female reprogramming of the adult testis (Matson et al. 2011) and in spermatogonia to maintain their proliferative state by avoiding meiotic entry (Zarkower 2013). Recently, Murphy and co-workers have shown that DMRT1 is a bifunctional transcriptional regulator, activating genes required for proper development of the male phenotype while repressing others that are specific of the female gonad (Murphy et al. 2010). These observations, together with the existence of human phenotypes associated with DMRT1 alterations make this gene a strong candidate for a causal role in some of the cases of idiopathic NOA.

Following our initial discovery of recurrent DMRT1 whole gene deletions in azoospermic patients from Utah (USA) and China (Lopes et al. 2013), we now describe novel and rare potentially regulatory variants in Portuguese patients, reinforcing that rare variants affecting DMRT1 function may underlie some cases of severe spermatogenic impairment. In one NOA patient we uncovered a non-coding variant in the promoter (c.-223_-219CGAAA>T) that resides within a region shown to be involved in the repression of Dmrt1 expression in rat and mouse Sertoli cells (Lei & Heckert 2002). Our in silico analysis not only supports the high conservation of this region across mammals (Fig. 1B) but also predicts that the human sequence is recognized by two heat-shock transcription factors (Fig. 1A and Supplementary Table 5), HSF1 and HSF2. Heat-shock transcription factors are known to activate heat-shock response genes, under stress conditions. Interestingly, these two transcription factors (HSF1 and HSF2) are highly expressed in testis, showing germ cell type specificity and are required for the progression of spermatogenesis. In adult mice testis, HSF2 expression is limited to spermatocytes and spermatogonia (Kallio et al. 2002) while HSF1 is highly and specifically expressed in the nuclei of spermatocytes and round spermatids (Akerfelt et al. 2010). Apparently, Hsf1 is specifically expressed in those stages of spermatogenesis where Dmrt1 is not, suggesting that HSF1 might block the transcription of Dmrt1, therefore allowing spermatogenesis to progress.

Our analysis indicates that the base substitution within the repression motif of the human sequence has resulted in the acquisition of a second HSF1 binding site (Fig. 1 and Supplementary Table 5) absent in the ancestral sequence (Fig. 1). Remarkably, this human-specific HSF1 binding site is disrupted by the c.-223_-219CGAAA>T variant, suggesting that it may result in DMRT1 misregulation and the development of NOA. Indeed, the histology of this patient revealed maturation arrest, resembling the phenotype of Hsf1-/- male mice which display regions of the seminiferous tubules containing only spermatogonia (Akerfelt et al. 2010). It is noteworthy that a mutation in HSF2, also predicted to bind the DMRT1 promoter, has been associated with azoospermia (Mou et al. 2013). It would be interesting to further explore the role for HSF1 in DMRT1 regulation during spermatogenesis in order to understand how its dynamics can contribute to the phenotype.

Three alternatively spliced DMRT1 transcripts have been described in the human testis: DMRT1a, DMRT1b and DMRT1c (Cheng et al. 2006). DMRT1a is the canonical DMRT1 isoform, showing the highest expression, and translates into a 375 amino acid protein. Both DMRT1b and DMRT1c result from the incorporation of intronic sequences in the mature mRNA (intron 3 and intron 1 respectively) and are less expressed. Canonical splice signals are known to be insufficient for proper splicing, requiring the regulation provided by cis-elements such as enhancers or repressors, especially important when alternative splicing is involved (Cartegni et al. 2003). Indeed, the creation of a new consensus exonic splicing enhancer (ESE) motif by a deep intronic mutation has already been shown to result in the production of an aberrant mRNA involved in afibrinogenemia (Davis et al. 2009).

We have identified a novel indel that is predicted to alter splicing either by interfering with the reading frame of DMRT1c (c.354+38_insG and linked poly-T in intron 1, Supplementary Table 2) However, the length variation in the poly T in intron 1, was also observed in our controls makes them less likely candidates for a strong deleterious effect on DMRT1 function. Also, an African ancestry was determined for this NOA patient suggesting that this locus harbors longer alleles in African populations when compared to Europeans.

One rare (≤1%) non-coding variants more frequent in patients than in Portuguese fertile controls were also predicted to affect splicing (rs200423545, Table 2).

Furthermore, the single rare variant detected in the 5′-UTR (rs144122237) of one patient potentially disrupts a binding site for the transcription factor RFX3, which is highly expressed in the testis. This variant has not been detected in the European control populations surveyed (our own and those included in databases) and was only present in the patient with African ancestry. The relatively low frequency of this variant in Africans (1.8%; 6 female and 3 male carriers) does not preclude an effect on fertility and overall our results highlight the need to consider the combinatorial effect of regulatory variants, given the interplay between factors with different roles (expression regulation, splicing, etc), as well as the genetic background of each individual.

Increasing data has been showing that synonymous variants might have an effect on transcription, splicing, mRNA transport or translation [Reviewed in (Goymer 2007)]. In accordance, we have found two synonymous variants overrepresented in patients that were predicted to potentially alter splicing by the creation of either new branch point sequences (rs376518776; c.213G>A) or new motifs for ESEs (rs34946058; c.783C>G). At the rs34946058 triallelic locus the T allele, and not the G allele found in our patients, has been recently reported in two men of German ancestry with cryptozoospermia (sperm concentration <0.1 million/mL), but also in one normozoospermic control (Tewes et al. 2014). Interestingly, the Portuguese population seems to harbor only two alleles at this locus, the reference C and the minor allele G that reaches a frequency of approximately 6% in African populations. Three patients carried the G allele, all of which had been previously analyzed for genome-wide SNPs (Lopes et al. 2013) and show no evidence for African ancestry/admixture. One fertile control also carried the G allele and another fertile control was homozygous for the same allele. Thus, even though the sperm parameters for the fertile G homozygous individual are unknown, the relatively high frequency of this allele in Africans does not support a strong impact on fertility.

Overall, our in silico functional analysis suggests that some of the of DMRT1 variants found in patients, which are not predicted to affect the protein (synonymous and non-coding) may however be involved in alternative splicing and other regulatory mechanisms. Per se, these variants are likely to have a mild impact on fertility, probably affecting the regulatory network controlling spermatogenesis but may also be in LD with intronic cis-acting SNPs and therefore a combinatorial effect of regulatory variants could be implicated in NOA. Moreover, the lack of deleterious coding mutations and intragenic rearrangements (screened by MLPA) in our Portuguese cohort supports the view that DMRT1 genetic defects are rare (Lopes et al. 2013, Tewes et al. 2014) and strong association for individual variants may only be detected in multicenter studies with larger cohorts.

Concordantly, our analysis of the DMRT1 protein amongst vertebrates (Fig. 2) reinforces the previous observations. The highly conserved DM domain as well as some regions of DMRT1-like domain seem to be under stronger functional constraints displaying less variation. Together with our previous results for CNVs (Lopes et al. 2013), and a recent screening for DMRT1 variants in German patients (Tewes et al. 2014), the finding of variants potentially affecting DMRT1 expression supports a model where changes in dosage in this gene may underlie some cases of NOA. The association of DMRT1 hemizygosity with human disorders of sex development (Barbaro et al. 2009, Õunap et al. 2004, Raymond et al. 1999, Tannour-Louet et al. 2010, Veitia et al. 1997) but also NOA (Lopes et al. 2013, Tewes et al. 2014), further points to other cis- and trans-acting factors contributing to these more complex phenotypes. It seems therefore that both DMRT1 copies must be functional for proper gonadal development and spermatogenesis maintenance and that variation affecting one or both alleles would result in phenotypes of increasing severity.

In summary, we provide evidence suggesting that rare regulatory variants in coding and non-coding regions of the DMRT1 gene and its promoter might be involved in idiopathic non-obstructive azoospermia (NOA). Altogether our results draw attention to the potential contribution of regulatory variants to male infertility, namely NOA, reinforcing the complexity and significance of regulatory networks involved in the control of spermatogenesis.