Background

Life-long production of sperm through the process of spermatogenesis is supported by spermatogonia, the most undifferentiated adult germ cell type. The pool of spermatogonia ensures male fertility, as these cells have the potential to self-renew as well as to give rise to differentiating daughter cells. Spermatogonia originate from primordial germ cells (PGCs), which are specified very early during embryonic development. These PGCs undergo erasure of DNA methylation, which allows the establishment of sperm-specific DNA methylation profiles during later stages of gametogenesis [1].

Erasure of DNA methylation takes place in two sequential stages. During the initial stage, a global decrease in methylated cytosines occurs, whereas in the second stage methylation is removed from imprinting control regions and meiotic genes [2,3,4]. In human male PGCs, the epigenetic ground state of global methylation levels has been found in foetuses between 7 and 11 weeks of age [5,6,7]. Currently, it is not known when the de novo DNA methylation rise commences in human fetal germ cells, [8] but importantly, this process of de novo global methylation continues until well after birth in primates. [9]

A number of publications have reported that male infertility is associated with aberrant sperm DNA methylation profiles, particularly in imprinted genes [10,11,12,13,14,15]. However, we found no recurrent epimutations by deep bisulfite sequencing analysis of sperm from patients with severely impaired spermatogenesis (n = 93) and controls (n = 40), combined with whole genome bisulfite sequencing (WGBS) of selected samples [16]. This study, which is one of the largest in this field, rather revealed that the presence of residual somatic DNA in swim-up purified sperm samples and genetic variation are major confounders of methylation studies in sperm. In line with the recommendations made by Åsenius et al. [17], these two confounders need to be considered in prospective studies to clarify if there is indeed an increased prevalence of aberrant methylation in infertile men.

Only few studies have used whole genome bisulfite sequencing (WGBS) to investigate DNA methylation of human spermatozoa at base pair resolution [16, 18,19,20]. The main findings of the methylation studies were that there is no significant methylation at non-CpG sites, that there are large regions of low methylation in a manner independent of genomic features such as CpG islands (CGIs) and promoters, and that CGI shores in sperm are more shallow compared to CGI shores in embryonal stem cells (ESCs). WGBS has also been performed on PGCs and advanced germ cells (AGCs) of human embryos highlighting, among other features, the importance of DNA methylation at transposons [5]. With regard to the adult germline, Hammoud et al. [21] have studied murine spermatogonia but genome-wide analysis of DNA methylation in testicular germ cells (TGCs) of adult men and, in particular, its comparison to TGCs from patients with impaired spermatogenesis constitutes a research gap.

Publications of high-quality single-cell RNA sequencing (scRNA-seq) transcriptomes obtained from human testicular tissues with intact spermatogenesis [22,23,24,25] have greatly advanced our knowledge with regard to the transcriptional changes associated with human germ cell differentiation. Moreover, comparative analyses of scRNA-seq results from men with intact and severely impaired spermatogenesis have provided insight into the molecular mechanisms associated with failure of germ cell differentiation [25, 26]. In order to assess the presence of aberrant epigenetic patterns in TGCs and the potential association with aberrant transcriptional profiles, we combined the two powerful approaches of WGBS data with scRNA-seq in samples with intact and impaired spermatogenesis.

Results

Isolation and characterization of human testicular germ cells

To analyse DNA methylation differences between human testicular germ cells (TGCs) in normal and impaired spermatogenesis, we isolated germ cells from patients with obstructive azoospermia (CTR, n = 24) and cryptozoospermia (CZ, n = 10) (Fig. 1a, Additional file 1: Table S1). By deep bisulfite sequencing (DBS) of H19, MEST, DDX4 and XIST, we identified 21 CTR (87.5%) and five CZ (50%) samples with pure germ cell fractions, of which we selected four from each group for whole genome bisulfite sequencing (WGBS) (Table 1, Fig. 1a, Additional file 1: Tables S2 and S3, Additional file 2: Fig. S1). These samples were found to have normal methylation values in the four regions, consistent with the absence of somatic DNA, and no significant difference was found between the two groups (Fig. 1b, Additional file 1: Table S3).

Fig. 1
figure 1

Testicular germ cell samples selection for whole genome bisulfite sequencing. a Schematic representation of the experimental design. b Box plot showing the methylation values of H19, MEST, DDX4 and XIST measured using deep bisulfite sequencing (DBS) of the supernatant fraction at day 3–4 of culture for the normal controls (CTR, teal, n = 4) and the cryptozoospermic (CZ, purple, n = 4) samples. No significant difference was found in the methylation values of the four genes between the two groups. c Box plot showing the results of the ploidy analysis of the day 0 single-cell suspension of the normal controls (CTR, teal, n = 4) and the cryptozoospermic (CZ, purple, n = 4) samples used for whole genome bisulfite sequencing (WGBS). d Stacked bar plots showing the percentages of tubules containing germ cells (most advanced germ cell type shown), only Sertoli cells, or tubular shadows in each biopsy from which the samples for WGBS were prepared. e Box plot showing the results of the ploidy analysis of the supernatant fraction at day 3–4 of culture for the normal controls (CTR, teal, n = 4) and the cryptozoospermic (CZ, purple, n = 4) samples. No significant difference was found in the cellular composition of the supernatant fraction between the two groups

Full size image
Table 1 Clinical parameters
Full size table

Ploidy analyses of the selected samples showed an enrichment of haploid cells in the CTR samples compared to the CZ samples in the initial single-cell suspension (Fig. 1c). Histological analysis of the testicular biopsies corroborates this result, showing a median of 78.5% and 23.5% of tubules containing elongated spermatids in patients with obstructive azoospermia and cryptozoospermia, respectively (Fig. 1d). Importantly, following 3–4 days of culture, the CTR and the CZ samples had a similar proportion of haploid, diploid, and tetraploid cells in the germ cell fraction (SN), as demonstrated by ploidy analysis (Fig. 1e).

Whole genome bisulfite sequencing of testicular germ cell DNA from patients with normal and impaired spermatogenesis

Following the coverage recommendations by Ziller et al. [27], we sequenced the eight TGC samples at ~ 14 × coverage each (Additional file 1: Table S4). To further investigate the absence of somatic cell DNA contamination in these TGC samples, we analysed the methylation levels at the 50 known maternally and paternally methylated imprinted control regions (ICR) [28]. As previously shown for uncontaminated sperm samples [16], the eight TGCs showed unmethylated oocyte DMRs and methylated sperm DMRs (Additional file 1: Table S5, Additional file 2: Fig. S2AB), consistent with the absence of somatic DNA contamination.

Comparison of methylomes from normal TGCs and related cell types

To understand how the methylomes of testicular germ cells compare to the methylomes of related cell types, we first compared the CTR-TGC methylomes with those of normal control sperm samples (n = 5) previously generated by our group [16]. Although the same library preparation method and sequencing platform were employed for all methylomes, mean methylation levels were slightly lower in the TGCs than in sperm, both globally and in all genomic features analysed (Additional file 2: Fig. S3AB). Next, we performed a principal component analysis (PCA) using WGBS data from (i) human embryonic stem cell lines (ESCs) (n = 2, H1 and H9), (ii) PGC samples collected from 7 to 19 weeks-of-gestation human male embryos (n = 8), a developmental time frame at which methylation has been almost completely erased [5], (iii) SSEA4+ spermatogonial stem cells (SSCs) isolated from testicular cells by magnetic activated cell sorting (n = 2) [29], (iv) CTR-TGC samples from this study (n = 4) and (v) sperm samples mentioned above (n = 5). The PCA revealed three major clusters: 1. SSCs, TGCs and sperm (upper left hand corner), 2. PGCs (upper right hand corner) and 3. ESCs (lower left hand corner) (Additional file 2: Fig. S3C). Given that the first principal component (PC1) explains 88% of the variability, SSCs, TGCs and sperm have methylation levels much closer to ESCs than to globally demethylated PGCs. Cluster 1 can be further divided into three subclusters corresponding to the different cell types. The SSCs are some distance away from the TGCs and sperm samples, and by investigating the methylation levels of the 34 imprinting control regions (ICRs), which are paternally unmethylated and maternally methylated, we found that the SSC samples from Guo et al. [29], despite having been purified by marker-based sorting, likely contained residual somatic cell DNA (Additional file 2: Fig. S2C). Therefore, we did not consider the SSC datasets for further analyses. The TGCs are slightly separated from sperm, which is in line with their slightly lower global methylation level (see above).

The slightly higher methylation levels of the maternal ICRs in PGCs (Additional file 2: Fig. S2C) are due to the samples collected from 7- and 10-week-old embryos (Additional file 2: Fig. S2D). At this stage, demethylation is probably not yet complete [6], although the presence of somatic cell DNA cannot be excluded.

Comparison of TGC methylomes from controls and cryptozoospermic men at the global level

We observed that CTR and CZ methylomes do not differ in their global mean methylation values nor in their overall patterns of methylation in various genomic features (Additional file 1: Table S4, Additional file 2: Fig. S4A, B). Moreover, a PCA of the eight methylomes showed that samples do not cluster according to the diagnosis (Fig. 2a).

Fig. 2
figure 2

CTR-CZ DMRs are associated with 132 genes. a Global comparison of methylomes from control (CTR) and cryptozoospermic (CZ) testicular germ cells. PCA generated for ca. 20 million CpG loci where all samples show methylation values. Only loci with minimum coverage of five in all samples and minimum mapping quality of 10 are considered. CTR testicular germ cell samples in teal, CZ samples in purple. b Cluster analysis of the methylation values for the 271 CTR-CZ DMRs in the eight TGC samples. c Enrichment/depletion of DMRs for functional genomic regions. LMR, low-methylated region; UMR, unmethylated regions; CGI, CpG islands. d DMRs are associated with 132 genes by overlapping genes, promoters and/or “double-elite” enhancers (GeneHancer)

Full size image

For comparing the number and size of proximal and distal regulatory elements in the CTR and CZ genomes, we used MethylSeekR [30] to segment the methylomes into unmethylated regions (UMRs; proximal regulatory regions) and low-methylated regions (LMRs; distal regulatory regions) [31]. We identified on average 36,000 UMRs and 22,000 LMRs per methylome, with CTR and CZ methylomes showing a similar number and total genomic size for each class of regions (Additional file 2: Fig. S4C). Of these, 35,593 UMRs and 14,041 LMRs were found in at least three of the CTR methylomes and were thus considered to be of high confidence (Additional file 2: Tables S6 and S7). Due to the lack of regulatory region annotations for testicular germ cells (unavailability of e.g. histone ChIP-seq data), we used the high-confident UMRs and LMRs as a proxy for putative TGC promoters and enhancers.

Identification of differentially methylated regions in testicular germ cells from controls and cryptozoospermic men

For identifying group-specific methylation differences between the CTR and CZ samples at the regional level (differentially methylated regions, DMRs), we used three different DMR calling algorithms (metilene, bsmooth, and camel), which typically provide different, but overlapping lists of DMRs. We only considered CpGs that were covered by at least five reads. For metilene, we used q > 0.05 as a threshold. For bsmooth and camel, which use t-statistics for identifying differentially methylated CpGs, but do not provide q-values for DMRs, we set a threshold of four CpGs as the minimum DMR length and 0.3 as the minimum methylation difference. Using these filters, we identified 1329 different DMRs (Additional file 2: Fig. S5A, Additional file 1: Table S8). For several of these DMRs, the range of methylation values within each group of samples was very high (Additional file 2: Fig. S5B—left). This was also evident in a cluster analysis of the methylation levels for the 1329 DMRs that showed methylated and unmethylated samples of the same group for some DMRs (Additional file 2: Fig. S5C). In addition, the cluster dendrogram shows a CZ sample clustering together with the CTR group. For each DMR, a high methylation range within a group is likely to result from chance or from differences in genetic background, where a genetic variant determining the methylation state of nearby CpGs can lead to low, medium, or high methylation levels, depending on the presence of one of three genotypes in an individual [32]. To reduce this confounder, we limited the range of methylation values to 0.3 within both groups, thus keeping 271 DMRs for which the two diagnosis groups are clearly separated in the dendrogram (Additional file 2: Fig. S5AB—right, Additional file 1: Table S8). Of these DMRs, 238 are hypermethylated in CZ (binominal test, p < 2.2 × 10–16).

Functional annotation of DMRs

For functionally annotating the DMRs, we determined their overlap with common genomic features such as genes and exons as well as with high-confidence UMRs and LMRs (Additional file 1: Table S9, Additional file 2: Fig. S5D). Of the 271 DMRs, 128 (47%) overlap a gene and 65 (24%) overlap a TGC regulatory region as determined by segmentation: 34 overlap a UMR corresponding to a proximal regulatory site and 31 overlap a LMR corresponding to a distal regulatory region. Monte Carlo simulations for DMR enrichment analysis revealed that DMRs are significantly enriched for intergenic regions and LMRs (p = 1.0 × 10–6) (Fig. 2c). Some DMRs (11%) overlap GeneHancer DoubleElite regulatory elements (Additional file 1: Table S9), which comprise high-confident human enhancers/promoters with high confident association to their gene targets [33].

Based on the intersection of DMRs with genes and/or regulatory elements, we found that 134 DMRs (49%) are associated with 132 different genes (Fig. 2d, Additional file 1: Table S10). Most of the DMR-associated genes are protein-coding (82%), and the vast majority of the DMRs associated with these genes (70%) overlapped an intron. Smaller fractions of the DMR-associated genes arise from DMRs overlapping promoters (11%) and/or for being GeneHancer associated targets (14%) (Fig. 2d, Additional file 1: Table S10).

Expression of the DMR associated genes during spermatogenesis

To ascertain the relevance of the DMR-associated genes, we made use of the available scRNA-seq data obtained by our group from three patients with normal spermatogenesis (CTR) and three patients with cryptozoospermia (CZ) [26]. We subset the germ cells and obtained 14,098 and 5,939 cells from the CTR and CZ groups, respectively (Fig. 3a). We aligned the cells along the latent time, setting the undifferentiated spermatogonia as starting point, and found that it recapitulated the spermatogenic process with the elongated spermatids at the end (Fig. 3b). We then checked the expression of the 132 genes in the normal dataset and found that 55 (41.7%) of them were highly expressed. To identify genes with a similar expression pattern, we performed clustering analysis of the 55 genes. This resulted in the identification of 5 clusters covering 43 of these genes, while 12 genes remained unclustered (Fig. 3c). While the genes in cluster 1 showed their expression peak in spermatogonia, the genes in clusters 2 and 3 had highest expression levels in spermatocytes. Finally, genes in clusters 4 and 5 reached their peak at the spermatid stage (Fig. 3d). We also analysed scRNA-seq data from two other scRNA-seq studies and found that 42 of the 132 genes (32%) are differentially expressed at various stages of spermatogenesis, with 34 reported by Hermann et al. [23] and 40 by Guo et al. [22]. Overall, 61 genes were found in at least one of the three datasets and 29 genes were shared between all three datasets (Additional file 1: Table S10). The 61 spermatogenesis-regulated genes are associated with 67 DMRs, 94% of which are hypermethylated in CZ (63/67, binomial test p = 1.107 × 10–14) (Fig. 4). The majority of the DMRs are intronic (79%). The other DMRs are located at promoters (13%), exons (16%), and GeneHancer DoubleElite regulatory regions (15%) (Fig. 4), with possible impact of the DMRs on gene expression levels and/or mRNA isoform regulation.

Fig. 3
figure 3

Characterization of the expression of the DMR associated genes in the scRNA-seq datasets. a Right panel: uniform manifold approximation and projection (UMAP) plot showing the germ cell subset of the scRNAseq data published in [26]. Left panel: UMAP plots showing the cells obtained from three normal controls (CTR, n = 14,098) and from the three cryptozoospermic patient samples (CZ, n = 5939). The cells are color coded according to their identity defined by the assignment published in [26]. b UMAP plot showing the integrated CTR-CZ germ cell dataset aligned along the latent time. The cells are color-coded according to their progression along the latent time. The undifferentiated spermatogonia cluster was set as starting point of the differentiation process. c Heatmap showing the normalized expression of the 55 DMR-associated genes with more than 500 counts in the CTR dataset. The cells are plotted along the latent time with the undifferentiated spermatogonia as starting point on the left side. The clustering analysis identified 5 clusters and left 12 genes unclustered. d Line plots showing the normalized expression along the latent time of the 55 DMR-associated genes with high expression grouped according to the belonging cluster in the CTR (teal) dataset

Full size image
Fig. 4
figure 4

DNA methylation levels and functional characterization of the DMRs associated with genes showing spermatogenesis-regulated expression. Box plots show the distribution of the methylation values for CTR (teal, n = 4) and CZ (purple, n = 4) (Additional file 1: Table S8). The overlaps of each DMR with specific genomic features are shown: exons (yellow), introns (light-blue), lncRNAs (grey), promoters (orange), UMRs (unmethylated regions, green), LMRs (low-methylated regions, dark-blue) and GeneHancer “double-elite” regulatory regions (pink). The 13 genes in bold were shown to be differentially expressed between CTR and CZ (see Fig. 5)

Full size image

Differential expression of the DMR associated genes between normal and cryptozoospermic patients

To assess whether DNA methylation changes are associated with changes in gene expression levels, we used our recently published differential gene expression analysis of CTR and CZ testicular samples [26]. For this, germ cells were divided into three knot groups (Additional file 2: Fig. S6A) and tradeSeq was used to find trajectory-based differential expression [34]. Of the 132 DMR-associated genes, 11 were differentially expressed (log2 fold change > 1 and FDR < 0.01) (Fig. 5a, Additional file 1: Tables S10 and S11). Ten of them were associated with a hypermethylated DMR (Fig. 4, Additional file 2: Fig. S7) and downregulated in CZ (Fig. 5).

Fig. 5
figure 5

Characterization of the 13 differentially expressed DMR-associated genes between CTR and CZ patients. a Line plots showing the expression along the latent time of the 11 DMR associated differentially expressed genes in CTR (teal) and CZ (purple) dataset determined by tradeSeq analysis. The dashed lines mark the knots dividing the three knot groups. The grey areas identify the knot groups in which statistical significance is reached for each gene. Values can be found in Additional file 1: Table S11. b Box plots showing the average expression values of the DMR-associated genes that resulted to be differentially expressed using MAST while comparing the CTR (teal, n = 3) and CZ scRNA-seq datasets (purple, n = 3). Values can be found in Additional file 1: Table S12

Full size image

To identify in which specific germ cell type the DMR associated genes were differentially expressed, we used the previously published list of differentially expressed genes obtained comparing each CTR germ cell cluster with its respective in the CZ samples [26]. The comparison revealed two additional differentially expressed genes and allowed us to narrow down the differential expression detected by tradeSeq to specific cell type(s) in 6 of the 11 genes (Fig. 5b, Additional file 1: Table S12, Additional file 2: Fig. S6B). Interestingly, with this analysis, no DMR-associated gene was differentially expressed in spermatogonia, but all were differentially expressed during the meiotic stages and/or spermiogenesis. Similarly, the tradeSeq analysis showed all but one gene with expression changes between CTR and CZ testicular samples in the knot group containing cells at later stages of the meiotic divisions and cells undergoing spermiogenesis (Additional file 2: Fig. S6B).

Discussion

To identify DNA methylation differences associated with impaired spermatogenesis, we generated whole genome methylomes of testicular germ cells (TGCs) of men with cryptozoospermia (CZ) and of men with obstructive azoospermia who had normal spermatogenesis and served as controls (CTR). To the best of our knowledge, there is only one published study focusing on the methylomes of testicular cells from men with reproductive issues and it compared men with obstructive and non-obstructive azoospermia [35]. This study is based on the analysis of testicular biopsies, which comprise both germ cells and somatic cells. The presence of somatic DNA, however, is a major confounder of methylation studies in germ cells [16]. In our study, we separated germ cells from somatic cells by a short-term cell culture and checked the purity of the germ cells by determining the methylation levels of four loci (MEST, H19, XIST, and DDX4). In the CZ group, we obtained pure germ cell fractions in only 50% of the cases compared to the 87.5% success rate in the CTR group. This discrepancy is likely due to the increased ratio of somatic cells in the CZ patients and highlights the necessity of a careful isolation of TGC populations especially in patients with impaired spermatogenesis. The analysis of 50 known maternally and paternally methylated imprinted control regions in the WGBS data further confirmed the absence of somatic DNA in the testicular germ cell samples.

First, we compared the methylomes of human testicular germ cells obtained from men with normal spermatogenesis with that of previously published sperm of control individuals [16] confirming that there are no major changes of DNA methylation during spermatogenesis. This was previously described by Guo et al. [29], who showed that the DNA methylation profiles of human spermatogonial stem cells and mature sperm were nearly identical at promoters, putative enhancer sites and imprinted loci. Our comparison of control sperm and testicular germ cell methylomes revealed slightly lower mean methylation levels in testicular germ cells. Although the same library preparation method and sequencing platform were employed in the nine methylomes, the libraries were sequenced in separate runs, which might have introduced a bias. Other possibilities are that the difference is due to a lag in maintenance of DNA methylation by DNMT1 in the newly synthesized daughter DNA strand in dividing TGCs [36], or that the period in culture has an impact on the TGC DNA methylome.

Next, we compared the methylomes of testicular germ cells obtained from controls and cryptozoospermic men. Since unrelated individuals differ in genetic background, which is a major confounder of epigenetic case–control studies in humans, we selected against sequence-based DMRs by applying a range filter for methylation values between samples of the same group. This led to the identification of 271 differentially methylated regions (DMRs). Functional annotation of the 271 DMRs showed that they are enriched for putative distal regulatory regions (LMRs, p = 1.0 × 10–6) and associated with 132 genes. The analysis of scRNA-seq data on testicular tissues from men with normal spermatogenesis [22, 23, 26] revealed that almost half of them (n = 61) are regulated during spermatogenesis and relevant at several stages.

Multiple mechanisms have been suggested on how DNA methylation modulates gene expression. Our analysis has shown that many of the DMRs associated with the 61 spermatogenesis relevant genes overlap known or putative gene regulatory regions. It has long been established that gene promoter methylation is typically related to a down-regulation of gene transcription. We detected nine DMRs overlapping gene promoters, which may be directly linked to differences in gene expression levels between CTR and CZ TGCs or to changes in transcription initiation from an alternative promoter. Furthermore, DNA methylation at enhancers is established by and/or affects the binding of transcription factors (TF), as in the case of CTCF and NRF1 [37]. A few of the DMRs (n = 10) coincide with elite enhancers from the GeneHancer DoubleElite track [33] and six of them have been reported as super-enhancers, clusters of transcriptional enhancers driving cell-type-specific gene expression programs, and fundamental to cell identity [38]. Many of the gene-associated DMRs overlap intronic regions and some coincide with putative distal regulatory regions (LMRs), which suggests they might be as-of-yet unknown, possibly germ cell-specific, enhancer regions. Whether DMRs overlapping elite-enhancers, introns, or LMRs actually act as enhancers regulating gene expression in TGCs remains to be determined. DNA methylation has also been shown to regulate splicing of about 22% of alternative exons (reviewed in [39]), and we identified 11 DMRs overlapping exons, which may have an impact on gene isoform regulation.

To address whether some of these DMRs affect gene expression, we performed differential gene expression analysis of scRNA-seq data and found 13 genes showing differences in expression patterns between CTR and CZ TGCs. It is possible that this number is underestimated, since scRNA-seq cannot accurately quantify lowly expressed genes. Furthermore, the variability of isoforms could not be addressed, since a UMI-based scRNA-seq approach was used, which is particularly suited to quantify expression at the gene level, but prevents isoform discrimination when differences are not located in the small sequenced fraction of the transcript [40].

Four of 13 CTR-CZ differentially expressed genes (C2orf92, C10orf120, PPP3CC, and PPP1R36) have hypermethylated promoters in cryptozoospermic patients, which correlates with a lower expression levels in these patients. For the remaining nine genes, eight of which are downregulated, expression changes may be driven by methylation at putative germ cell specific enhancers (Table 2). The 13 genes are differentially expressed during the different steps of meiosis and/or spermiogenesis, which suggests a potential correlation between their dysregulation and spermatogenic failure. Indeed, this patient cohort shows a drastic reduction in germ cell number after pachytene stage [26]. Moreover, the great majority of these 13 genes have been shown to have a role in spermatogenesis or to have enriched expression in the testis (Table 2).

Table 2 DMRs relevance in CTR-CZ differentially expressed genes
Full size table

Conclusions

Based on our findings, we conclude that TGCs from men with impaired spermatogenesis differ from control TGCs in DNA methylation levels at defined genomic regions, many of which appear to be gene-regulatory elements. Imprinting control regions were not affected, suggesting that testicular sperm extraction (TESE) does not further increase the risk of imprinting defects which is associated with intracytoplasmatic sperm injection (ICSI) and other assisted reproduction techniques. We do not know the cause of the aberrant methylation patterns in CZ men, but one possibility is that the observed methylation changes mediate or reflect gene expression changes involving gene-regulatory circuits at the cellular level. The fact that most of the DMRs are hypermethylated in CZ points to a failure of upregulating important genes during spermatogenesis. As we have not observed recurrent epigenetic defects in sperm of a large cohort of infertile men [16], the methylation changes probably contribute to premature abortion of spermatogenesis, and therefore do not appear in mature sperm. As we considered the motile sperm following swim-up procedure, we cannot exclude that aberrantly methylated sperm is present in the discarded semen fraction. Another possibility is that the methylation changes reflect differences in the relative proportion of different types of germ cells, although they were grossly similar by ploidy analysis. The two possibilities are not mutually exclusive, and future improvements in single-cell methylation analysis may resolve this issue. At present, methods for single-cell methylome analysis are not yet efficient enough for case–control studies. Be that as it may, the disruption of cellular events in CZ germ cells (whatever its cause) has allowed us to highlight the importance of DNA methylation in testicular germ cells and to identify gene regulatory regions that undergo DNA methylation changes during spermatogenesis.

Methods

Clinical characterization and selection of testicular biopsies

Prior to surgery all patients underwent full phenotyping by physical examination, hormonal analysis (including luteinizing hormone (LH), follicle stimulating hormone (FSH) and testosterone (T)) [16], semen analysis [50] and genetic analyses (karyotype and screening for azoospermia factor (AZF) deletions). Known genetic causes of infertility, acute infections and tumours were exclusion criteria. Control patients were selected from those diagnosed with obstructive azoospermia due to congenital bilateral absence of the vas deferens (CBAVD) or from those undergoing vasectomy reversals. These patients had no sperm in their ejaculate, normal testicular volumes and normal FSH levels (Additional file 1: Table S1). Cryptozoospermic patients had a sperm concentration of less than 0.1 million/ml (cryptozoospermia) in the ejaculate, a reduced testicular volume and in six of the ten cases elevated FSH levels indicating testicular failure. Detailed clinical information about all the patients are available in Additional file 1: Table S1. Following this selection strategy, we obtained biopsies with qualitatively and quantitatively normal spermatogenesis (n = 24; controls, CTR), and with idiopathic reduced spermatogenesis (n = 10; cryptozoospermic, CZ) between January 2018 and February 2020. For a selection of these samples (n = 3 for CTR and CZ each) scRNA-seq was performed and previously published [26].

Histological analysis of testicular tissues

For routine diagnostic purposes two testicular biopsies per testis were fixed in Bouin’s solution, paraffin embedded and sectioned at 5 µm. Two independent sections per biopsy were subjected to periodic acid-Schiff/hematoxylin (PAS) staining as previously described [26]. The spermatogenic status was evaluated using the Bergmann and Kliesch score [51].

Short-term culture for purification of testicular germ cells

A differential plating strategy was applied to separate the testicular germ cells from the somatic cells. For this purpose, testicular biopsies were digested into a single-cell suspension using a two-step enzymatic digestion protocol as previously published [52]. Briefly, testicular tissues were minced with sterilized blades into ~ 1 mm3 pieces and incubated in MEMα (22561021, Gibco, Thermo Fisher scientific, Waltham, MA USA) with 1 mg/ml collagenase IA (C9891, Sigma-Aldrich, Merck KGaA, Darmstadt, Germany) at 37 °C for 10 min. The digestion was stopped by addition of MEMα supplemented with 10% foetal bovine serum (FBS) (S0615 Merck KGaA, Darmstadt, Germany) and 1% Penicillin/Streptomycin (P/S) (15140–148, Gibco, Thermo Fisher scientific, Waltham, MA USA). After centrifugation the cells were incubated in Hank’s balanced salt solution (14175053, Gibco, Thermo Fisher scientific, Waltham, MA USA) containing 4 mg/ml of trypsin (27250018, Gibco, Thermo Fisher scientific, Waltham, MA USA) and 2.2 mg/ml of DNase I (DN25, Sigma-Aldrich, Merck KGaA, Darmstadt, Germany) at 37 °C for 10 min. Finally, a single-cell suspension was obtained by strong pipetting. The reaction was stopped as outlined above and cells were washed three times with MEMα containing 10% FBS and 1% P/S. The erythrocytes were removed with a three-minute incubation in haemolysis buffer (0.83% NH4Cl solution).The reaction was stopped as outlined above. At the end of the procedure the cells were passed through a 70 µm Sterile CellTrics® filter (Sysmex) and counted using the trypan blue exclusion method. 20,000 cells of the single-cell suspension were stored at −80 °C for ploidy analysis, 12,000 cells were used for scRNA-seq [26] while the rest was plated at a density of < 50.000 cells/cm2 onto uncoated cell culture dishes and cultured in MEMα containing 10% FBS and 1% P/S at 35 °C and 5% CO2. After 1 day of culture the cells from the supernatant (SN, Germ cells) were separated from the attached cells (AT, Somatic cells) to obtain a pure germ cell fraction [52, 53]. After 3–4 days of culture 20,000 cells of the germ cell fraction were stored at −80 °C for ploidy analysis. The remaining cells were stored at -80 °C for subsequent DNA isolation and DNA methylation analyses.

Ploidy analysis for analysis of cellular composition

Ploidy analyses of single-cell suspensions obtained immediately after tissue digestion and after 3–4 days of culture were performed to evaluate the cellular composition of each sample. 20,000 cells per sample were processed and analysed as previously published [26].

DNA isolation and targeted deep bisulfite sequencing for purity screening

To screen the collected germ cell fractions for presence of somatic cell DNA, the following approach was applied. DNA was purified from cultured germ cell fractions using the MasterPure DNA purification kit (MC85200, Epicentre Biotechnologies, Madison, WI, USA) according to the manufacturer’s protocol. DNA concentration was measured using a fluorescence plate reader (FLUOstar Omega, BMG Labtech, Germany). The bisulfite conversion and the deep bisulfite sequencing for MEST, H19 (CTCF6 region), XIST and DDX4 were performed as previously described [16]. The primer pairs and PCR conditions are described in Additional file 1: Table S2 and were based on previous publications [54,55,56]. CTR (n = 21) and CZ (n = 5) samples resulted to have pure germ cell fractions.

Whole genome bisulfite sequencing

WGBS libraries were prepared from CTR and CZ samples (n = 4 each) using 10 ng testicular germ cell DNA supplemented with 1% unmethylated lambda-DNA (Promega, Madison, USA) according to a previously described tagmentation-based method [16], which was based on the protocols described by Wang et al. [57] and Souren et al. [58]. The libraries were sequenced in HiSeq4000 100-bp paired end runs (Illumina, San Diego, USA) using one lane per sample.

WGBS data analysis

We used wg-blimp v0.9.4 to process WGBS data [59]. In brief, wg-blimp integrates established algorithms for processing WGBS data. These include algorithms for alignment, quality control, DMR calling, methylome segmentation and annotation based on USCS and Ensembl databases [60, 61]. All data were aligned and annotated against human reference hg38. We used R 3.6.0 to import wg-blimp methylation reports and perform PCA analysis on CpG loci where all samples showed at least 5 × coverage.

DMRs were required to cover at least 4 CpG sites with at least 30% methylation difference in the groups compared, minimum mean coverage of 5 × and a maximum q-value of 0.05 (where available). We merged DMRs using the GenomicRanges R package [62]. Different filtering strategies based on the ranges of methylation values in CTR and CZ groups were further applied.

DMR overlap with genomic features was compared to simulated DMR distributions to identify potential over- or under-representation of overlap. Specifically, we simulated 1,000,000 sets of DMRs to compare the number of overlaps with genomic features between random and actual DMRs. To create a random set of DMRs we randomly selected 1000 DMRs with the same number of covered CpGs for each of our actual DMRs. From these 1000 DMRs one is selected according to a log-normal distribution of covered nucleotides based on our actual DMR sizes to ensure matching size distributions of simulated and actual DMRs. These simulated sets of DMRs then allow computation of empirical p values and odds ratios for over- and under-representation of overlap.

Putative regulatory regions were identified by segmenting each methylome with MethylSeekR [30], which is part of the wg-blimp pipeline. First, we identified and masked regions of disordered methylation (partially methylated domains, PMDs) and then identified unmethylated regions (UMRs) and low-methylated regions (LMRs). High-confidence UMRs and LMRs were defined as the overlapped regions present in at least three CTR methylomes.

We used the GeneHancer DoubleElite track as a source of highly-ranked human regulatory elements and their inferred target genes, due to their high-likelihood enhancer definition and a strong enhancer-gene association [33]. Repeats refer to elements from RepeatMasker [63].

ScRNA-seq analysis

The pre-processing, quantification, integration, dimensional reduction, labelling and the MAST differential gene expression analysis were performed as described in the methods in Di Persio et al. [26]. The expression of genes in the scRNA-seq dataset was assessed after subsetting the germ cells out of the CTR and CZ datasets. The latent time was computed on the integrated dataset of all samples, which results in a common latent time trajectory for both sample groups. To reduce noise and increase statistical power of the tradeSeq differential expression test, we focused on strongly expressed genes with a total expression above 500, excluding 77 of the 132 DMR associated genes (58.3%).

Differential expression analysis along latent time with tradeSeq

We utilized the R package tradeSeq v1.1.18 [64] to perform a differential expression analysis along the determined gene-shared latent time between the spermatogonial cells of the CTR and CZ datasets. The same expression counts from the previous analysis with MAST v1.10.0 [65] were used for this step. For each gene and lineage, negative binomial generalized additive models were fitted between four time points (knots) of the latent time. These knots are equally distributed among the cell density along the trajectory, with the first and last knot representing the minimal and maximal latent time value, respectively. The four knots can be comprised to three knot groups, where the first knot group consists of all cells between knot one and two, etc. To ensure convergence of the generalized additive model fitting process, we increased the maximal number of iterations to 1000. We focused on identifying genes that only show differential expression in a singular knot group. For this, we adopted the stageR v1.6.0 [66] two-staged testing scheme, using a whole-trajectory patternTest() for the screening stage and a earlyDETest() for each knot group in the confirmation stage. All tests were performed against a log2 fold change of 1, the stageR correction procedure used an overall false discovery rate of 0.01 and a multiple testing correction using the holm method [67]. Additionally, the fitted distributions of the normal lineage for the remaining 55 DMR associated genes were clustered to reveal genes with a common expression pattern. The clustering used 100 cluster points and a minimal cluster size of 5, resulting in five distinct clusters as well as a set of unclustered genes.

Analyses of spermatogenesis-regulated genes

Spermatogenesis-regulated genes as identified by published scRNA-seq datasets of human testicular biopsies were retrieved from the available supplementary information [22, 23, 26]. Genes shared by these datasets and the DMR-associated genes were obtained after all gene names were converted to HGNC approved symbols [68].

Statistical analyses

Difference between two independent groups was assessed by Mann–Whitney U test followed by Bonferroni correction for multiple testing. Binomial tests were used to assess whether two categories are equally likely to occur. Statistical analysis and graphs plotting were performed using R 4.0.0 [69] and appropriate R packages, namely stats v4.0.0 [69], ggplot2 v3.3.0 [70] and gplots v3.0.3 [71].