Recurrent spontaneous miscarriages from sperm after ABVD chemotherapy in a patient with Hodgkin’s lymphoma: sperm DNA and methylation profiling

Abstract

Lymphomas represent one of the most common malignant diseases in young men and an important issue is how treatments will affect their reproductive health. It has been hypothesized that chemotherapies, similarly to environmental chemicals, may alter the spermatogenic epigenome. Here, we report the genomic and epigenomic profiling of the sperm DNA from a 31-year-old Hodgkin lymphoma patient who faced recurrent spontaneous miscarriages in his couple 11–26 months after receiving chemotherapy with adriamycin, bleomycin, vinblastine, and dacarbazine (ABVD). In order to capture the potential deleterious impact of the ABVD treatment on mutational and methylation changes, we compared sperm DNA before and 26 months after chemotherapy with whole-genome sequencing (WGS) and reduced representation bisulfite sequencing (RRBS). The WGS analysis identified 403 variants following ABVD treatment, including 28 linked to genes crucial for embryogenesis. However, none were found in coding regions, indicating no impact of chemotherapy on protein function. The RRBS analysis identified 99 high-quality differentially methylated regions (hqDMRs) for which methylation status changed upon chemotherapy. Those hqDRMs were associated with 87 differentially methylated genes, among which 14 are known to be important or expressed during embryo development. While no variants were detected in coding regions, promoter regions of several genes potentially important for embryo development contained variants or displayed an altered methylated status. These might in turn modify the corresponding gene expression and thus affect their function during key stages of embryogenesis, leading to potential developmental disorders or miscarriages.

INTRODUCTION

Lymphomas represent the second most common malignant disease in young men of reproductive age after testicular cancer. Two-thirds of the reported cases correspond to lymphoid hematological diseases, i.e., Hodgkin’s lymphoma (HL) and non-Hodgkin’s lymphoma (NHL). Age-standardized incidence rates of HL range between 2.3 and 2.7 per 100 000 in industrialized countries.1 In 2018, the number of men suffering from hematological malignancies in metropolitan France was estimated at 25 000.2 Fortunately, treatments have been developed in these last decades, and now, 80% of HL patients can be treated with chemotherapy, radiotherapy,3,4 and monoclonal antibodies,5 according to the stage and biological type of the disease. Treatment allows to improve the patient survival, making it one of the most treatable cancers with a 5-year survival rate of 88%.6,7 For patients who relapse, high-dose chemotherapy followed by autologous stem cell transplant is the standard of care. Chemotherapy ABVD (composed of four agents including adriamycin, bleomycin, vinblastine, and dacarbazine) is a very effective treatment for HL. However, it is composed of genotoxic agents that can cause direct or indirect damages to DNA of tumor cells but also of normal cells: adriamycin is an intercalating agent; bleomycin induces single- and double-strand DNA breaks due to free oxygen radical generation and directs DNA intercalation; vinblastine influences mitotic spindles; and dacarbazine is an alkylating agent.8 It has been shown that serum from patients with HL treated with ABVD presents increased free radical levels and decreased antioxidant levels.9 In addition, dacarbazine displays mutagenic, teratogenic, and carcinogenic effects in animal models.10 The bleomycin, etoposide, adriamycin, cyclophosphamide, oncovin, procarbazine, and prednisone (BEACOPP) treatment protocol can also be applied to HL cases, and although both ABVD and BEACOPP are gonadotoxic for HL men,11 strong evidence indicate that the ABVD protocol is the least gonadotoxic option of the lymphoma chemotherapy regimens.11,12,13

Over the past decades, major advances in therapeutic management have led to improvement in the prognoses and quality of life of cancer survivors. Hence, many young patients wish to procreate although cancer therapy might affect their reproductive functions. Sperm cryopreservation is thus recommended14,15 and systematically proposed before cancer treatment. In France, a unique national public standardized sperm bank network, the Center for Study and Conservation of Human Eggs and Sperm (CECOS), provides help to preserve fertility and counsels to these cancer patients. In this context, an important issue for young men suffering from HL is whether treatments will affect, transiently or permanently, their future fertility.

Several studies reported that HL alone (i.e., before treatment) can alter sperm characteristics, such as sperm count, motility, and aneuploidy.16,17,18,19,20,21 These alterations could be due to various risk factors for spermatogenesis, such as stress,22,23 fever,24,25 or inflammatory immune-mediated mechanisms.26 Gonadal functions could also be drastically affected after treatment and the intensity of spermatogenesis alterations may be related to the type of drugs used, the cumulative doses, the radiation received, and the time between the last treatment and sperm analysis.27,28,29 Indeed, some studies have described late adverse effects of chemotherapy or radiotherapy on sperm characteristics after chemotherapy or radiotherapy.13,30,31,32 While sperm DNA integrity after cancer treatment has been investigated,30,33,34 these studies are debated since they are mostly retrospective, and, consequently, the dynamics of sperm recovery are not precisely described. One prospective study reported an increase in the rate of sperm aneuploidy in the 3 months after ABVD chemotherapy, with a return to normal levels after 6–12 months. An increased rate of sperm aneuploidy was then observed 12 months after treatment with alkylants.12 The impact of cancer treatment on the sperm epigenome is a topical issue that surprisingly has been little studied so far. Indeed, epigenetic modifications to sperm cells may have deleterious effects on the offspring.35,36,37 A case report in a patient treated by temozolomide (alkylating agent used to treat serious brain cancers) for an anaplastic oligodendroglioma described a drop in sperm methylation levels.38 Shnorhavorian et al.39 reported modifications on DNA methylation regions in 18 sarcoma patients treated with cisplatin or with a combination of cisplatin and ifosfamides, but the transversal design of this study limited results interpretation. Although many epigenetic studies have been performed in HL,40,41,42 only two recent landmark studies used high-throughput technologies to describe the impact of cancer treatments on sperm DNA methylation in patients with HL and testicular cancer.43,44 Both publications found that some DNA methylation defects in sperm persist for up to 2 years after treatment, including genes important for development. To our knowledge, few studies have described genome-wide analysis of sperm from cancer patients who have experienced recurrent spontaneous miscarriages (RSMs) after treatment.

In this context, we report the genomic and epigenomic profiling of frozen spermatozoa from a HL patient before and after ABVD chemotherapy. We focused on this case since following treatment, the couple experienced four spontaneous miscarriages and was unable to conceive a child naturally, whereas the use of pre-treatment cryopreserved spermatozoa resulted in a pregnancy. Classical exploration of RSMs was performed and did not reveal any etiology. Therefore, we hypothesized that ABVD chemotherapy had affected the sperm DNA genome and/or epigenome characteristics. Here, we describe genomic and epigenomic changes in sperm DNA after chemotherapy in a patient with HL that could be related to the embryo losses.

PATIENTS AND METHODS

Ethical considerations and patient samples

Written informed consent of the patient was obtained for the study of his sperm DNA. Before chemotherapy, the patient collected semen one time by masturbation and 16 straws containing semen for fertility preservation being stored in liquid nitrogen (LN2). An aliquot of the fresh sample was used for semen analysis. Two straws stored in LN2 were used for later genetic studies and were referred to as prechemotherapy (pre-CT) semen samples. Twenty-six months after CT (September 2011, six cycles of ABVD), the patient remained able to ejaculate. An aliquot of a fresh sample was used for semen analysis and DNA integrity examination, and 1.5 ml were allocated into two straws, which were then stored in LN2 for later genetic studies (post-CT semen samples). Briefly, sperm cryopreservation was the method currently used in the laboratory to preserve fertility prior to cancer treatment. We used Sperm Freeze (JCD SA, La Mulatière, France) as cryoprotective media and high security straws (Cryo Bio System, Saint Ouen sur Iton, France). The straws were sealed using MAPI automate (Cryo Bio System) and freezing was performed using Freezal automate (Air Liquide, Paris, France). Each straw contained 26.7 × 10⁶ spermatozoa in the first freezing before treatment and 10.6 × 10⁶ spermatozoa in the second freezing. Sperm samples were conserved for the study on the GERMETHEQUE national Biobank (No. BB-0033-00081) censored by the Institutional Review Board (IRB) of the South-West Human Research Ethics Committee (Bordeaux, France).

The semen analysis was performed according to the 2010 guidelines of the World Health Organization (WHO).45 To investigate sperm DNA fragmentation and sperm chromatin condensation, we used commonly used techniques, as we have done previously,46 such as terminal uridine nick end labeling (TUNEL), sperm chromatin structure assay (SCSA), and aniline blue staining. The TUNEL assay, which is a useful and sensitive method for assessing DNA strand breaks by binding terminal deoxynucleotidyl transferase (TdT) to exposed 3’-OH ends of DNA fragments, was performed according to a previously published method and the percentage of normal value was less than 20%.47 SCSA detects abnormalities in sperm chromatin compaction and the cut-off value was 30%.48 Acidic aniline blue stains human spermatozoa with defects in chromatin condensation and the cut-off value was 80%.49

Whole-genome sequencing (WGS)

Straws from semen before and after treatment were thawed at 37°C for 10 min. Samples were washed twice with phosphate-buffered saline (PBS). In order to eliminate somatic cells, samples were submitted to a somatic cell lysis method according to a study of Goodrich et al.50 Samples were incubated at 70°C during 30 min in 200 µl PBS buffer (Sigma-Aldrich, Burlington, VT, USA) containing 40 µl of proteinase K (20 mg ml⁻¹; Thermo Fisher Scientific, Waltham, MA, USA) and 20 µl of 1 mol l^-1 dithiothreitol (DTT; Sigma-Aldrich) before DNA sperm extraction using the high pure PCR template preparation kit (Roche Diagnostics, Rotkreuz, Switzerland).

DNA sequencing was performed at the GeT-PlaGe core facility, INRAE Toulouse (Toulouse, France). Libraries were prepared using the Bioo Scientific (Austin, TX, USA) polymerase chain reaction (PCR) free Library Prep Kit according to the manufacturer’s protocol. An amount of 3 µg of total DNA was used for library construction. Briefly, DNA was fragmented by sonication, size selection was performed using CleanPCR beads and adaptators (GC Biotech, Waddinxveen, the Netherlands) were ligated. Library quality was assessed using an Advanced Analytical Fragment Analyser (Agilent Technologies Inc., Santa Clara, CA, USA). Libraries were quantified by quantitative PCR (qPCR)/real-time PCR using the Kapa Library Quantification Kit (Roche Diagnostics). Sequencing was performed on an Illumina HiSeq3000 using a 2× 150 bp paired-end protocol (San Diego, CA, USA).

WGS data pre-processing and analysis

A workflow was designed to process WGS data and to identify variants potentially induced by the ABVD treatment (Supplementary Figure 1 ^{(773.2KB, tif)} ). This workflow consists of several steps including read mapping, variant calling, quality filtration, variant annotation, variant filtration, detection of copy number variations, and mutational signatures.

After data quality control with FastQC,51 reads were mapped on the human genome (hg38) with Burrows-Wheeler Aligner-mem (BWA; release version 0.17.0).52

The detection of single-nucleotide polymorphisms (SNPs) and deletion/insertion variants (INDELs) was performed using a strategy combining three variant callers (Supplementary Figure 1a ^{(773.2KB, tif)} ): VarScan (version 2.4.4)53 and two other tools implemented in GATK (version 4.1.7), HaplotypeCaller54,55 and Mutect255. Briefly, variant detection was performed with VarScan (using default parameters) based on the bam files produced by BWA. Regarding HaplotypeCaller and Mutect2, bam files were first preprocessed according to the GATK Best Practice Workflows (GATK version 4.1.7; available from: https://gatk.broadinstitute.org/hc/en-us/sections/360007226651-Best-Practices-Workflows). First MarksDuplicates and ReorderSam functions implemented in Picard (version 2.18.2)56 were used to mark duplicated reads and reorder bam files. Next, the BaseRecalibrator function implemented in GATK was used for base quality recalibration using known SNPs and INDELs from dbSNP (version 155)57 and 1000 Genome58 respectively. Mutect2 enables the detection of somatic mutations between two samples. In this context, the pre-CT sample was used as a “normal tissue” and the post-CT sample as a “tumor tissue” into the CreateSomaticPanelOfNormals function.

With the exception of the VCF file generated by VarScan, hard filtration with recommended GATK Best Practice Workflows parameters was used. Variants were annotated with variant effect predictor (VEP; version 99)59 in order to determine variants effect on gene, transcript, protein, and regulatory regions (Supplementary Figure 1b ^{(773.2KB, tif)} ). Variants located in coding regions can affect protein function, whereas variants located in intergenic regions can potentially be associated with regulatory elements. VEP predicts regions in the human genome that could regulate gene expression using data from projects such as ENCODE, IHEC, and Blueprint.59 The distance of the annotated variants to the transcription start site (TSS) of their related genes is calculated by using the genomic coordinates of the canonical transcripts (gencode version 33 GTF file). Variants associated to regulatory elements but not to a gene following the VEP procedure were annotated by using the Genomic Regions Enrichment of Annotations Tool (GREAT; version 4.0.4).60

Variants with a coverage <8 reads were filtered out. Next, only variants with a ΔFrequency ≥50% between pre- and post-treatment samples were selected for subsequent analysis. To address this issue, we used the pysamstat library on the resulting BAM file before treatment for each variant position detected in the VCF file after treatment. Variant information (such as the numbers of A, C, T, G, insertions, and deletions) was extracted and the frequency of each variant into the pre-CT and post-CT samples was estimated. To reduce the number of false-positive calls originating from sequencing bias, mapping errors, and errors near repeat regions,61 only variants detected by VarScan and at least one of the two other variant callers were selected. This approach was chosen because GATK appears to generate more false-positive variants than VarScan.62

In order to exclude variants that are often detected in the general population, we removed variants present in more than 0.01% of the general population.

Copy number variation (CNVs) were detected in both samples with the cn.MOPS R package (version 1.4).63 A mutational signature analysis was performed with SigProfilerExtractor (version 3.1)64 by calculating the relative mutation frequency of the 96 possible triplets, with a triplet being defined as a given mutated nucleotide and its two flanking nucleotides. The signature was performed independently on variants detected by at least two tools including VarScan in the pre-CT and post-CT samples with ≥8 read coverage in both samples. Then, the mutational signature was performed on the variants showing ≥50% of frequency difference between pre-CT and post-CT samples.

Reduced representation bisulfite sequencing (RRBS)

Straws from semen before and after treatment were thawed and subjected to density gradient centrifugation (830g for 30 min, 25°C) on a two-layer 80%/40% Percoll (GE Heathcare, Chicago, IL, USA). Samples were divided into two fractions: the “motile” fraction of spermatozoa (corresponding to the 80% fraction), and the rest of the sample referred as “other”. The “motile” fraction was washed twice for 15 min each with 10 ml wash buffer (150 mmol l⁻¹ NaCl and 10 mmol l⁻¹ pH8 ethylenediaminetetraacetic acid [EDTA; Sigma-Aldrich]) at 1600g (Centrifuge 5810R; Eppendorf, Hamburg, Germany). Both the “motile” and “other” fractions have been considered for further analysis, representing a total of four samples (pre-CT_motile, pre-CT_other, post-CT_motile and post-CT_other). Cell lysis was performed in Low-Binding tubes and incubated in a thermomixer (Eppendorf) overnight at 55°C with 500 µl of digestion buffer (50 mmol l⁻¹ pH 8 Tris, 100 mmol l⁻¹ NaCl, 10 mmol l⁻¹ EDTA, 1% sodium dodecyl sulfate [SDS; Sigma-Aldrich], 21 µl of 1 mol l⁻¹ DTT [Sigma-Aldrich], 20 µl 20 mg ml⁻¹ proteinase K [Thermo Fisher Scientific], and 2.5 µl Triton X-100 [Sigma-Aldrich]).

Cell debris were pelleted down by centrifugation at 14 000g (Centrifuge MiniSpin; Eppendorf), and the supernatant was subjected to a 10 mg ml⁻¹ RNAse A (Sigma-Aldrich) digestion for 30 min at 37°C in a thermomixer. DNA was extracted with an equal volume of phenol:chlorophorm:isoamyl alcohol (Sigma-Aldrich) following by a second extraction with chloroform:isoamyl alcohol (Sigma-Aldrich). The aqueous phase containing DNA was purified by a standard precipitation with 300 mmol l⁻¹ NaCl per 2.5 v/v alcohol and 1 µl GlycoBlue (Thermo Fisher Scientific). Final DNA quantification was performed using the dsDNA QuantiFluor Dye System (Promega, Southampton, UK).

The RRBS method provides methylation information of regions of DNA with a high density of cytosine-phosphate-guanine (CpG) dinucleotides, also called CpG-rich regions.65 Methylation profiles of the samples were studied using the Premium RRBS Kit according to the manufacturer’s instructions (Diagenode, Seraing, Belgium). Briefly, 100 ng of DNA was digested using the methylation-sensitive restriction enzyme, MspI, followed by end repair and A-tailing. Adaptor ligation was performed and followed by a size selection (40–220 bp) with the AMPure XP Beads (Beckman Coulter, Pasadena, CA, USA). Quantification for sample pooling was evaluated by a qPCR analysis. Then, DNA samples with adjusted quantity underwent bisulfite conversion followed by desulphonation and purification on column. The converted DNA was used for PCR amplification and clean-up with the AMPure XP Beads. Finally, quality of the RRBS libraries was measured using an Agilent 2100 bioanalyzer prior to sequencing. The RRBS libraries were sequenced on an Illumina Hiseq 4000 sequencer as single-end 50-base reads following Illumina’s instructions. Image analysis and base calling were performed with RTA 2.7.3 and bcl2fastq. Following the sequencing, quality control has been verified for each sample with FastQC.51 Briefly, we ensured that a percentage higher than 95% of bases above Q30 (always ≥Q27) was obtained for each sample. The proportion of each nucleotide along the read has also been verified, certifying the fact that reads start with CGG due to MspI digestion cutting between cytosines at CCGG sites of the genome. Following a general tendency across samples, the nucleotide proportions on the first base of reads converge to a 50:50 ratio between cytosines and thymines (slightly higher for cytosines). The cytosines correspond to bisulfite unconverted methylated cytosines, while thymines represent bisulfite converted unmethylated cytosines. This observation, paired with a high ratio of guanine (95%) on the second and third base of reads, retranscribe MspI cutting sites.

RRBS data preprocessing and analysis

A four-step workflow was designed to preprocess RRBS data to analyze the CpG methylation data and, finally to identify and annotate differentially methylated regions (DMRs; Supplementary Figure 2 ^{(545.3KB, tif)} ). This workflow includes several steps including read mapping, CpG detection and methylation level, detection of high-quality differentially methylated regions (hqDMRs), and gene association.

Reads were mapped on the human genome (hg38) with the Bismark tool (release version 0.17.0).66 After data quality control, methylation calls were extracted by Bismark and a CpG coverage file containing the average methylation level of each covered CpG site was generated for each sample.

The clusterSites function implemented in BiSeq (version 1.19.0) was used to identify CpG islands for each sample.67 It takes into account spatial correlation of the methylation of nearby CpG sites, constraining the analysis on CpG sites among CpG clusters. CpG islands were defined as regions containing at least 20 frequently covered CpG sites, corresponding to those covered in 100% of the samples – closer to each other than a maximum distance of 100 bp. It is important to note that frequently covered CpG sites are considered to define the CpG cluster boundaries only. For downstream analysis, all CpG methylation levels within these CpG clusters were used. Methylation levels being strongly spatially correlated,68 and as suggested by Hansen et al.,69 the raw methylation data were smoothed to reduce the required sequencing coverage data.70 In addition, DMR detection methods without smoothing data often discard lowly covered CpG sites for further analyses. Methylation data smoothing was performed with the predictMeth function implemented in BiSeq (with a bandwidth of h = 80 bp). Within each CpG cluster and for each sample, a smoothing function is modeled on the local raw methylation data. The resulting methylation levels range from 0 to 1. Methylation has been called with MethGO71 on all samples resulting in methylation levels of genomic features described in the human reference transcriptome (promoters, genes, exons, introns, and intergenic regions).71 Methylation levels of CpG clusters are averaged to take into account the associated CpG methylation levels. In order to include a CpG within a feature, a minimum depth of four reads is required, the size of the promoter is set to 1000 bp and a 200-bp long sliding window is used to browse the genome.

The detection of DMRs relies on finding genomic regions showing smoothed CpG methylation level differences between two experimental conditions. The compareTwoSample function implemented in BiSeq was used to identify DMRs for each pairwise comparison between samples. DMRs are assembled with CpGs closer than a maximum distance of 100 bp having a methylation change greater than 30%. DMR extension is stopped if one of these rules is broken, establishing the DMR boundaries which can thereby be constituted of a unique differentially methylated CpG. It is worth remembering that one of the earliest studies of DNA methylation impacting transcription factor binding sites considered a single 6 bp region and the impact of one methylated cytosine within that region.72 The resulting nonredundant DMRs are then all aggregated in a single DMR file. Each detected DMR is then quantified in each sample by calculating the median smoothed methylation level of all CpG sites within this genomic region. Finally, we further refined the detected DMRs by identifying a subset of hqDMRs containing at least 3 CpG sites, a minimum 50 bp length, separated by a maximum of 100 bp and with a methylation change of 30% or higher.

We first assembled a unique set of human reference transcripts. To address this issue, Ensembl73 and RefSeq74,75 transcript annotations of the hg38 release of the human genome were downloaded from the University of California Santa Cruz (UCSC) genome browser website76 on November 11, 2017. Both transcript annotation files (GTF and GFF formats) were subsequently merged into a combined set of non-redundant human reference transcripts with Cuffcompare.77 Based on this reference transcriptome, we further associated DMRs with their adjacent human genes on the human genome (less than 10 kb away).

Functional analysis

The enrichment analysis module implemented in the AMEN suite.78 was used to identify gene ontology terms significantly over-represented in each gene group by calculating Fisher’s exact probability using the Gaussian hypergeometric function (false discovery rate [FDR]-adjusted P ≤ 0.01, number of genes in a given group associated with a given annotation term ≥ 5).

Gene sets of interest

To identify genes implicated in early embryo development process, we integrated microarray data (Affymetrix Human Genome U133 Plus 2.0 Array) published by Xie et al.79 available on the NCBI Gene Expression Omnibus (GEO) repository under the accession number GSE18290. This dataset includes the transcriptome of 6 embryonic stages (18 samples): 1-, 2-, 4- and 8-cell stages as well as morula and blastocysts. Raw data (cel files) were normalized and background corrected with the RMA method based on the Brainarray custom CDF environment for directly mapping Affymetrix gene to Entrez gene identifiers (hta20_Hs_ENTREZG version 22.0.0; http://brainarray.mbni.med.umich.edu).80 Briefly, genes showing a signal higher or equal to a given background cut-off (median of the normalized dataset, cut-off: 5.04) and at least a 3-fold change in at least one pairwise comparison were selected. To define the set of 4264 differentially expressed genes displaying significant statistical changes across early embryo development stages, the linear models for microarray data (LIMMA) package was used (F-value adjusted with the FDR method p[BH]≤0.01)81 implemented in AMEN.78

Genes associated with embryo development

A list of 969 human genes known to be associated with the embryo development process were selected on the basis of their association with the “embryo development” Gene Ontology term (GO: 0009790) based on the “gene2go” file downloaded from the NCBI website. A list of 91 human maternally and paternally imprinted genes has been gathered from a web-based interface at http://www.geneimprint.com/site/genes-by-species.Homo+sapiens.82

RESULTS

Patient and sperm collection

The 31-year-old Caucasian patient consulted the CECOS-Reproductive Medicine Department (Toulouse University Hospital [CHU], Toulouse, France) in September 2011 after the couple tried to conceive naturally and obtained four pregnancies resulting in four spontaneous miscarriages. The patient’s clinical history (Figure 1) reported a diagnosis of Hodgkin’s disease in December 2008. After the cancer extension assessment and sperm cryopreservation, six cycles of ABVD chemotherapy were administrated to the patient until June 2009. Four pregnancies were obtained naturally in May, August and December 2010 and in September 2011 but all pregnancies resulted in spontaneous miscarriages at 3 amenorrhea weeks (AW), 3 AW, 9 AW and 4 AW, respectively. Following the last miscarriage, the couple consulted the reproductive center. Patient clinical examination, serologic and endocrinologic testing, karyotypes, female endovaginal ultrasound and histerosonography, and the classical RSM exploration, were normal. Semen parameters, sperm aniline blue staining, DNA fragmentation index (DFI) and TUNEL of the fresh semen samples before and after CT (near the last miscarriage, at 26 months after CT) were also within normal ranges (Table 1). Intrauterine insemination (IUI), using semen cryopreserved before CT, was proposed to the couple. The second IUI attempt resulted on a normal pregnancy, with an eutocic delivery of a healthy female baby.

Figure 1.

Description of patient pattern. The Hodgkin lymphoma patient cryopreserved sperm before 6 cycles of ABVD chemotherapy in 2008. The chemotherapy protocol ended in June 2009. The couple obtained four natural pregnancies between May 2010 and September 2011, but all ended as miscarriages. Semen parameters were normal in September 2011, but due to the miscarriages an assisted reproduction method (intra-uterine-insemination [IUI] of 8.4 × 10⁶ motile sperm from cryopreserved sperm straws before chemotherapy) was used and allowed to obtain a healthy child in 2013. CT: chemotherapy; ABVD: adriamycin, bleomycin, vinblastine, and dacarbazine; WGS: whole genome sequencing; RRBS: reduced representation of bisulfite sequencing; HL: Hodgkin lymphoma; GWAS: Genome Wide Association Study.

Table 1.

Sperm characteristics before (pre-CT) and 26 months after chemotherapy (post-CT)

Sperm samples	Date	Abstinence time	Semen volume (ml)	Sperm count (×106 ml−1)	Sperm progressive motility (%)	Sperm vitality (%)	Normal sperm morphology (%)	MAI
Pre-CT	December 17, 2008	ND	6	152	65	76	44	1.46
Post-CT	September 19, 2011	ND	4.5	88	55	64	21	1.73
Lower reference limits (WHO 2010)	-	-	1.5	15	32	58	4	-

MAI: multiple anomalies index; ND: not done; -: no value; CT: chemotherapy; WHO: World Health Organization

The results of sperm condensation and sperm DNA fragmentation from the sperm sample after chemotherapy, performed very close to the time of the 4^th embryo loss, were in the normal ranges in our laboratory, aniline blue staining: 7% (normal <20%), DFI: 20% (normal <30%) and TUNEL: 11.4% (normal <20%). Since several miscarriages occurred with natural conception after treatment (Figure 1), we hypothesized that the causal alteration(s) responsible for the four RSMs, if they (it) exist(s), should be detected in a large proportion of the sperm cell population after the chemotherapy and not (or in a lower fraction) before chemotherapy. Consequently, WGS and a genome-wide methylation analysis (RRBS) were conducted to identify genetic and/or epigenetic alterations induced by the ABVD chemotherapeutic treatment which could be responsible for RSMs.

WGS mutational signature reveals a large number of C>T and T>C transition at CpG sites

The WGS analysis yielded 267 million and 220 million reads for pre-CT and post-CT samples, respectively. Reads were mapped on the human genome leading to a mapping rate of 82.4% for the pre-CT sample and 86.7% for the post-CT sample (Supplementary Table 1). At least 97% of the genome had a minimal coverage of 8 reads (Supplementary Figure 3a ^{(451.5KB, tif)} ). CNV analysis based on the coverage of both samples did not detect any significant CNVs.

Table S1.

Sperm DNA: Overview of read numbers and mapping statistics for each sample

Sample	Total read numbers	Alignement	Mapping efficiency
PreCT	267,154,697	220,235,888	82.43 %
PostCT	255,978,902	221,843,659	86.68 %

A total of 7 859 108 unique SNPs and 1 484 218 INDELs were detected in post-CT sample (Figure 2a) by VarScan (3 456 068 SNPs and 457 704 INDELs), HaplotypeCaller (3 636 176 SNPs and 594 552 INDELs), and Mutect2 (640 683 SNPs and 432 116 INDELs) allowed to detect new somatic variants compared to the pre-CT sample, with the vast majority having a frequency difference of less than 50% between the two conditions (Supplementary Figure 3b ^{(451.5KB, tif)} ). To test whether ABVD treatment has a genotoxic effect and could induce random DNA damages,83 mutational signature analysis was performed on the resulting SNPs and INDELs in pre-CT and post-CT samples (Supplementary Figure 4 ^{(1.3MB, tif)} –6 ^{(434.9KB, tif)} ). This analysis showed no difference between the pre-CT and post-CT samples. To reduce the number of false positives, we performed an additional mutational signature analysis on the 1845 SNPs and 822 INDELs detected by at least two tools (including VarScan), with a read coverage ≥8 and a frequency difference between pre-CT and post-CT samples ≥50%. This analysis revealed that the most common single base substitutions (SBS) corresponded to transition from C to T (34.3% in pre-CT, 34.4% in post-CT, and 34.2% in the differential variants) and from T to C (pre-CT = 32.8%, post-CT = 34.4%, differential variants = 31.6%; Supplementary Figure 4 ^{(1.3MB, tif)} ). The highest mutability was observed at CpG sites representing one fourth (pre-CT = 24.1%, post-CT = 24.2%, differential variants = 27.4%) of the SBSs (Supplementary Figure 4 ^{(1.3MB, tif)} ). Almost a third of the double base substitutions (DBSs) corresponded to the transition from TG to NN (any nucleotides) in pre-CT (29.2%) and post-CT (30.0%) samples, whereas it represents almost half (49.7%) of DBSs in differential variants (Supplementary Figure 5 ^{(1.3MB, tif)} ). Most of the INDELs were associated with deletions (pre-CT = 18.1%, post-CT = 18.3%, differential variants = 18.3%) or insertions (pre-CT = 18.1%, post-CT = 20.0%, differential variants = 20.4%) of multiple T (Supplementary Figure 6 ^{(434.9KB, tif)} ). Together these results did not show major changes between pre-CT and post-CT samples with the exception of DBS showing a slight increase of TG into NN after chemotherapy.

Figure 2.

Filtering strategy to select variants and functional annotation. (a) The numbers of selected single-nucleotide polymorphisms (SNPs) and insertions/deletion variants (INDELs) at each step and for each tool are indicated. This strategy is based on (1) read coverage; (2) difference of variant frequency ≥50% between the sample before chemotherapy (pre-CT) and the sample after chemotherapy (post-CT) samples; (3) variants must be detected by at least 2 tools including VarScan; and (4) variant frequency in the general population ≤0.1%. (b) Variant annotation to genomic regions. The association to genes important for embryogenesis is indicated. (c) Variant annotation to regulatory elements and distance to transcription start site (TSS) in kilobases (kb) are indicated. TF: transcription factor; CDS: coding DNA sequence; UTR: untranslated regions; NMD: nonsense-mediated mRNA decay.

Some variants are located in regulatory regions of genes important for embryogenesis

To identify the potential causal variants that may be responsible for the patient’s RSMs, all variants (SNPs and INDELs) with a frequency ≥0.1% into the general population were filtered out (Figure 2a). This strategy yielded the selection of 403 variants (136 SNPs and 267 INDELs) out of which 233 were detected by the three tools (Supplementary Table 2 ^{(1.5MB, pdf)} ). The variant annotation analysis revealed that 234 were located in the vicinity of 227 genes although none of them were located in coding regions (CDS). Most of the detected variants were associated with: 1) intergenic regions (38.7%); 2) transcripts of non-coding RNAs (24.5%); 3) up/down-stream regions of the gene (12.1%); 4) intronic regions (13.1%); and 5) untranslated region (UTR) regions of protein-coding genes (0.7%), as shown in Figure 2b. Among the 56 variants located in regulatory regions (Supplementary Table 2 ^{(1.5MB, pdf)} ), ten were associated with proximal promoter regions (≤5 kb, 18.0%) and the vast majority (82.0%) were associated with distal regions (5–500 kb) corresponding to transcription factor binding sites, promoters, enhancers, and open chromatin regions (Figure 2c).

It is noteworthy that the functional analysis revealed that the set of 227 variant-associated genes were significantly enriched in genes coding for proteins on the cell periphery (GO: 0071944, 62 genes, adjusted P=0.018), and more specifically in transporter complex (GO: 1990351, 11 genes, adjusted P=0.01). We next wondered whether some of those genes might be relevant for embryogenesis (Supplementary Table 3 and Supplementary Results). To address this issue, a list of 4264 genes differentially expressed during early embryogenesis was identified.79 and a set of 969 genes known to be associated with embryo development based on their associated GO terms was also considered. This analysis revealed that out of the 227 variant-associated genes, five were known to be involved in embryonic development and 25 were differentially expressed during early embryogenesis (Supplementary Table 2 ^{(1.5MB, pdf)} and 4). Among those, five genes (zinc finger protein 217 [ZNF217], solute carrier family 2 member 3 [SLC2A3], C-type lectin domain family 4 member C [CLEC4C], cadherin 3 [CDH3], and MOB kinase activator 3B [MOB3B]) contained variants in their regulatory regions including CLEC4C for which a mutation (chr12_7749685_T/A, pre-CT =15.8% and post-CT = 69.2%) was located at 212 bases from its TSS (Supplementary Figure 7 ^{(661.3KB, tif)} ). Based on the GREAT predictions, 13 other variants located into regulatory elements were also associated with genes presumably important for embryogenesis. Most of them are far from TSS of genes except for trinucleotide repeat containing adaptor 6A (TNRC6A) and STE20 related adaptor alpha (STRADA; <10 kb; data not shown).

Table S3.

Detection of high-quality differentially methylated regions (hqDMRs) and associated differentially methylated genes (DMGs).

Contrasts	All	PostCT motile vs PreCT motile		PostCT other vs PreCT other		PreCT motile vs PreCT other		PostCT motile vs PostCT other
Methylation status	All	Hyper	Hypo	Hyper	Hypo	Hyper	Hypo	Hyper	Hypo
Number of hqDMRs	273	30	99	30	71	40	76	33	86
Number of DMGs	165	18	69	23	61	35	50	22	49
Expr. in early embryo	14	WDR70, PTBP1	SSTR5-AS1,	CLUH, JARID2	CENPX, ATG2A,	ETFB, JARID2,	WDR70, PTBP1,	WDR70	SSTR5-AS1,
dev. (4264 genes)			CENPX, EPS8L1,		SH2D4B, WDR70,	SH2D4B, DGKZ	KMT2C, AHNAK		CLUH, HSPBAP1
			JARID2, SH2D4B,		AHNAK
			DGKZ
Associated with embryo	11	ARHGAP35,	FOXH1, HPN,	XAB2	ARHGAP35, TBX4,	None	SATB2,	ARHGAP35, TBX4,	RPGRIP1L, XAB2
dev. (969)		NCOR2, SATB2	XAB2		NCOR2		RPGRIP1L, GBX2,	NCOR2
							CECR2,
							ARHGAP35, XAB2,
							BCR
Imprinted genes (91)	None	None	None	None	None	None	None	None	None

Table S4.

RRBS: Overview of read numbers and mapping statistics for each sample.

Sample	Total read numbers	Alignement w/ unique best hit	Mapping efficiency	C analysed	C (CpG) meth	C (CHG) meth	C (CHH) meth	C (CpG) unmeth	C (CHG) unmeth	C (CHH) unmeth
PreCT_Motile	48,001,678	29,064,659	60.50 %	324,081,670	25,913,037	737,519	4,508,122	23,753,856	72,010,044	197,159,092
PreCT_Other	56,835,636	33,525,087	59.00 %	385,134,424	33,666,371	888,493	4,880,499	30,382,184	86,880,740	228,436,137
PostCT_Motile	62,128,721	36,138,882	58.20 %	413,389,231	29,385,856	926,821	6,693,874	40,340,517	92,290,641	243,751,522
PostCT_Other	61,799,879	35,991,010	58.20 %	416,309,694	35,804,211	959,223	5,813,382	35,323,519	93,836,285	244,573,074

The variant analysis also pointed out three variants showing the highest frequency difference between pre-CT and post-CT samples (chr1_228633373_T/A, chrUn_KI270746v1_40552_T/A and chr2_132404733_-/TTC) with a ∆Frequency ≥80% (Supplementary Figure 8a ^{(1.3MB, tif)} and 8b ^{(1.3MB, tif)} ). The former is located in a promoter site (Supplementary Figure 8a ^{(1.3MB, tif)} ) and is upstream of the ribosomal gene RNA 5S ribosomal 13 (RNA5S13) at 3841 bp from the TSS. The latter is located into an intergenic region with several deletions in both pre-CT and post-CT samples surrounded by the noncoding RNA LOC10537955 (Supplementary Figure 8b ^{(1.3MB, tif)} ).

Chemotherapy induces a low but genome-wide demethylation of CpGs in DNA sperm

The RRBS method was used to identify the DNA methylation changes in the patient’s sperm at the single cytosine resolution and investigate the potential epigenomic impact of the ABVD treatment. To address this issue, pre-CT and post-CT samples were divided into two fractions: the “motile” sperm fraction, and the rest of the fraction referred as “other”.

A total of 48–62 million reads per sample were mapped on the converted human genome leading to a mapping rate ranging from 58.2% to 60.5% (Figure 3a and Supplementary Table 5). Among the 324–416 millions of cytosines analyzed, 50–71 millions (15%–17%) corresponded to CpGs (Figure 3b). The coverage of CpGs was limited to the 90% quantile for further analysis in order to reduce bias due to unusually high coverages. The median coverage of those CpGs ranged from 11–19 reads across the four samples (Figure 3c). It is important to note that only 1.0% of CHG and 2.1%–2.7% of CHH (where H corresponds to A, T or C) are methylated whereas 52.2% and 52.7% of the CpGs are methylated before the ABVD treatment in both “motile” and “other” fractions, respectively (Figure 3d). This methylation level was slightly lower in the “motile” fraction post-CT sample (42.1%) and, to a less extent, in the “other” fraction (50.3%), as shown in Figure 3d. Finally, the CpG island detection and smoothing of their methylation level led to the identification of 19 175 CpG islands corresponding to 883 188 CpGs.

Figure 3.

Read mapping and methylation statistics for the sample before (pre-CT) and after chemotherapy (post-CT). (a) Read mapping statistics. (b) Sample-wise numbers of CpG, CHG and CHH (H corresponds to A, T or C). (c) Sample-wise coverage distribution of CpG sites. Distribution of coverage across samples after coverage limitation (90% quantile). (d) Proportion of methylated CpG, CHG and CHH for each sample. (e) Distribution of methylated CpGs among genomic features. A minimum depth of four reads is required for each CpG. CpG methylation levels are averaged by feature to produce feature-associated methylation levels.

Table S5.

Genes important for embryogenesis associated with variants or differentially methylated regions between the preCT

symbol	gene description	A mutation frequency	Methylation status	Embryogenesis
ABCB10	ATP binding cassette subfamily B	chr1_229519820_-/TAAA (63.1%)		expressed during
	member 10
ARHGAP35	Rho GTPase activating protein 35		Hyper	associated with
ARL4C	ADP ribosylation factor like GTPase 4C	chr2 234578391 AG/- (54.7%)		expressed during
ATP8B1	ATPase phospholipid transporting 8B1	chr18_57764339_C/T (57.9%)		expressed during
BICC1	BicC family RNA binding protein 1	chr10 58814796 -/A (65.1%)		expressed during
C6	complement C6	chr5_41312686_-/TTCC (78.2%)		associated with
CADPS2	calcium dependent secretion activator 2	chr7 122706181 AT/- (76.4%)		expressed during
CDH3	cadherin 3	chr16_68704211_C/T (51.3%)		expressed during
CENPX	centromere protein X		Hypo	expressed during
CHD2	chromodomain helicase DNA binding pro	chr15_92809928_C/T (58.4%)		expressed during
CLEC4C	C-type lectin domain family 4 member C	chr12_7749685_T/A (53.4%)		expressed during
CPVL	carboxypeptidase vitellogenic like	chr7_29080558_A/- (70.5%)		expressed during
CTNNA2	catenin alpha 2	chr2 79202335 A/- (56.2%)		expressed during
CTSL	cathepsin L	chr9_87724383_AA/- (57.6%)		expressed during
DGKZ	diacylglycerol kinase zeta		Hypo	expressed during
EPS8L1	epidermal growth factor receptor		Hypo	expressed during
	pathway substrate 8 like 1
	family with sequence similarity 174	chr15_92809928_C/T (58.4%)		expressed during
FAM174B	member B
FN1	fibronectin 1	chr2_215401251_G/A (50%)		both
FOXH1	forkhead box H1		Hypo	associated with
GABARAPL1	GABA type A receptor associated protein	chr12_10181167_GI 1 1 1/- (61.3%)		expressed during
HES1	hairy and enhancer of split-1	chr3 193888954 A/T (63.3%)		associated with
HOMER1	homer scaffold protein 1	chr5_79506856_AT/- (79.4%)		expressed during
HPN	hepsin		Hypo	associated with
JARID2	jumonji and AT-rich interaction domain		Hypo	expressed during
	containing 2
KAT2B	lysine acetyltransferase 2B	chr3 20121731 -/TG (67.8%)		expressed during
	LIM homeobox transcription factor 1	chr9_126558101_C/- (75%)		associated with
LMX1B	beta
LRBA	LPS responsive beige-like anchor protein	chr4 150320421 T/- (54.1%)		expressed during
LY6E	lymphocyte antigen 6 family member E	chr8_143071297_T/A (51.7%)		associated with
MAP4K3	mitogen-activated protein kinase 3	chr2 39363993 -/A (76.4%)		expressed during
MED1	mediator complex subunit 1	chr17_39440018_GAAA/- (53.8%)		associated with
MOB3B	MOB kinase activator 3B	chr9_27357149_-/T (59%)		expressed during
NCOR2	nuclear receptor corepressor 2		Hyper	associated with
NEO1	neogenin 1	chr15 73164020 T/- (56.6%)		expressed during
OPA1	OPA1 mitochondrial dynamin like GTPasi	chr3_193888954_A/T (63.3%)		both
PBX1	PBX homeobox 1	chr1 164874008 -/AA (68.7%)		associated with
PLS1		chr3_142628336_TGTGTGCGCGCA		expressed during
		CATGTGCATGCAGTGTGCA/-
	plastin 1	(81.8%)
PTBP1	polypyrimidine tract binding protein 1		Hyper	expressed during
RASA3	RAS P21 protein activator 3	chr13_114075511_G/C (53.8%)		expressed during
RPF1	ribosome production factor 1 homolog	chr1_84492159_AA/- (56.6%)		expressed during
SATB2	SAT homeobox 2		Hyper	associated with
SGK1	serum/glucocorticoid regulated kinase 1	chr6_134235542_-/TTA (63.3%)		expressed during
SH2D4B	SH2 (Src Homology 2) domain		Hypo	expressed during
SH3BP4	SH3 domain binding protein 4	chr2 234578391 AG/- (54.7%)		expressed during
SLC1A5	solute carrier family 1 member 5	chr19_46773476_T/- (51.4%)		expressed during
SLC2A3	solute carrier family 2 member 3	chr12 7928832 T/- (55.5%)		expressed during
SLC5A11	solute carrier family 5 member 11	chr16_24730742_CCCC/- (86.6%)		expressed during
SLC7A2	solute carrier family 7 member 2	chr8 17571739 T/A (50%)		expressed during
SNX9	sorting nexin 9	chr6_157737284_T/- (52.4%)		expressed during
SSTR5-AS1	somatostatin receptor 5 antisense RNA 1		Hypo	expressed during
STRADA	STE20 related adaptor alpha	chr17_63740144_-/AC (55%)		expressed during
SUCLA2	succinate-CoA ligase ADP-forming	chr13_47924523_A/C (54.3%)		expressed during
	subunit beta
TAF4	TATA-box binding protein associated fact	chr20_62029487_-/CA (60%)		expressed during
TNRC6A	trinucleotide repeat containing adaptor 6A	chr16_24730742_CCCC/- (86.6%)		expressed during
TULP3	TUB like protein 3	chr12_2937105_-/A (67.8%)		both
UBR3	ubiquitin protein ligase E3 component N	-r chr2_169832935_A/- (54.1%)		associated with
WDR70	WD repeat domain 70		Hyper	expressed during
XAB2	xeroderma pigmentosum group A		Hypo	associated with
	binding protein 2
XPO1	exportin 1	chr2_61585965_ACCT/- (61.1%)		expressed during
ZNF217	zinc finger protein 217	chr20 53588595 TATATA/- (63.1%)		expressed during

As the CpG methylation status of the human genome has become increasingly exploited, it has also emerged that differential methylation is not only restricted to CpG islands, but also extends to CpG regions, for example, at enhancers84,85 as well as across gene bodies. In the current study, methylation calls have been performed using MethGO71 on all samples resulting in methylation levels of genomic features across the whole genome, promoters, genes, exons, introns, and intergenic regions. A slight but global decrease of CpG methylation is observed among all features in the post-CT sample, for both the “motile” and “other” fractions (Figure 3e).

About seventy genes show a chemotherapy-induced demethylated status in motile spermatozoa

A differential analysis allowed us to identify 1475 genomic regions showing a differentially methylated status or DMRs across all samples. The resulting set of DMRs was further refined by selecting only those composed of at least three CpG sites and covering at least 50 bp on the genome yielding to a list of 305 high-quality DMRs (hqDRMs). Correlation and principal component analyses of those hqDMRs revealed the proximity between the pre-CT and post-CT “other” fractions (fractions containing spermatozoa of bad quality and non-sperm cells). Conversely, the two “motile” fractions (fractions containing only spermatozoa probably of high quality) are very distant, illustrating the impact of the ABVD treatment on the methylation status of sperm DNA (Supplementary Figure 9 ^{(617.4KB, tif)} ).

To further characterize chemotherapy-associated differential methylation of sperm DNA, four pairwise comparisons were performed using BiSeq: pre-CT _motile vs post-CT_motile, pre-CT _other vs post-CT_other, pre-CT_motile vs pre-CT_other, and post-CT_motile vs post-CT_other (Figure 4a–4d, and Supplementary Table 3). The resulting hqDMRs from each comparison were pulled together to generate a list of 273 non-redundant hqDMRs associated with 165 differentially methylated genes (DMGs; Figure 4e and Supplementary Table 3). Among the 129 and 101 hqDMRs detected when comparing semen samples before and after chemotherapy in the “motile” and “other” fractions, respectively (Figure 4a and 4b), the vast majority was hypomethylated after treatment. Together, those demethylated hqDRMs were associated with 107 DMGs, out of which 69 were detected in the “motile” sperm fraction (Figure 4a).

Figure 4.

Filtration of high-quality differentially methylated regions (hqDMRs) between the sample before (pre-CT) and after chemotherapy (post-CT). Number of hypermtheliated and hypomethylated hqDMRs and related differentially methylated genes (DMGs): (a) in the post-CT_motile vs pre-CT_motile comparison; (b) in the post-CT_other vs pre-CT_other comparison; (c) in the pre-CT_motile vs pre-CT_other comparison; and (d) in the post-CT_motile vs post-CT_other comparisons. (e) A false-color heatmap of standardized methylation levels of the resulting 273 hqDMRs which have been sorted (dendrogram on the left and on the top) based on a hierarchical clustering applied to the standardized methylation level data. Each line corresponds to an hqDMR and each column to a sample. Standardized methylation levels are displayed according to a color code (bottom right) ranging from blue (hypomethylation) to red (hypermethylation).

Most chemotherapy-induced methylation changes differ between the “motile” and “other” semen fractions

We next focused the analysis on hqDRMs that were detected when comparing DNA sperm methylation before and after chemotherapy in the “motile” sperm fraction (the most representative fraction concerning sperm quality; Figure 5). The resulting 129 regions were further clustered into four hypo-(P1–4) and three hyper-methylated patterns (P5–7; Figure 5). Briefly, patterns P1 and P5 include 30 and 8 hqDMRs which, despite the hypo-(P1) or hyper-methylation (P5) of the “motile” fraction, are weakly methylated in the “other” fractions. Patterns P2 and P6 included 14 and 11 hqDMRs which had opposite methylation dynamics between the “motile” and the “other” fractions after chemotherapy. Patterns P4 and P7 were associated with hqDMRs which, despite the hypo- (P4, 26 hqDMRs) or hyper-methylated (P7, 11 hqDMRs) status in the “motile” fraction after chemotherapy, were highly methylated in the “other” fraction. Finally, only pattern P3 included 29 hqDMRs (22.5%, corresponding to 20 DMGs) which were demethylated after chemotherapy in both the “motile” and “other” fractions. These results indicate that most methylation changes potentially induced by the chemotherapy are different between the “motile” and the “other” fractions.

Figure 5.

Heatmap of the 129 hqDMRs showing altered methylation status after chemotherapy in mobile sperms (post-CT_motile). The numbers of high-quality differentially methylated regions (hqDMRs) and related differential methylated genes (DMGs) are given for each of the seven methylation patterns (P1–P7) on the right, while the overall number of hyper- and hypo-methylated hqDMRs is indicated on the left. Each line corresponds to an hqDMR and each column to a sample. Standardized methylation levels are displayed according to a color code (bottom right) ranging from blue (low methylation level) to red (high).

The methylation status of fourteen genes associated with embryogenesis is altered in motile spermatozoa after chemotherapy

We next checked whether the 18 hyper- and 69 hypo-methylated DMGs associated with hqDMRs detected in the “motile” fractions (Supplementary Table 5) were significantly enriched in specific biological processes. Only demethylated hqDMRs after chemotherapy in the “motile” fraction were found to be statistically associated with genes annotated to filamentous actin (GO: 0031941, adjusted P ≤ 6×10⁻⁴), including Rac family small GTPase 3 (RAC3), junctional adhesion molecule 3 (JAM3), PDZ and LIM domain 2 (PDLIM2), and PDZ and LIM domain 7 (PDLIM7).

The next step was to identify, among the DMGs after chemotherapy in the “motile” fraction those that could be relevant for embryogenesis (Supplementary Table 3 and 4, and Supplementary Results). We were able to identify eight DMGs (WD repeat domain 70 [WDR70], polypyrimidine tract binding protein 1 [PTBP1], SSTR5 antisense RNA 1 [SSTR5-AS1], centromere protein X [CENPX], Jumonji and AT-rich interaction domain containing 2 [JARID2], SH2 domain containing 4B [SH2D4B], diacylglycerol kinase zeta [DGKZ], and EPS8 signaling adaptor L1 [EPS8L1]) that were expressed during early embryonic stages (Supplementary Figure 10 ^{(1.3MB, tif)} ) and six that were associated with embryo development (SATB homeobox 2 [SATB2], Rho GTPase activating protein 35 [ARHGAP35], nuclear receptor corepressor 2 [NCOR2], hepsin [HPN], XPA binding protein 2 [XAB2], and forkhead box H1 [FOXH1]), as shown in Supplementary Figure 11 ^{(753.6KB, tif)} and Supplementary Table 3. The altered methylated status of those DMGs important for early development might modify their expression pattern during key stages of embryogenesis, leading to potential developmental disorders or miscarriages.

It is important to note that none of those DMGs corresponded to known imprinted genes. In addition, only one variant (chr20_45646223_C/T, Δmutation frequency ≥57%) identified by the WGS analysis was located into a hqDMR hypomethylated after chemotherapy into the “other” fraction (chr20: 45646143–45646227). This mutation was in an intergenic region, between WFDC10A and WFDC11.

DISCUSSION

Post-treatment quality of life is one of the most important criteria in the management of HL in young men, as fertility is often impaired after chemotherapy and radiation therapy. Based on follicle-stimulating hormone (FSH) levels, a study reported that the median time to fertility recovery is 19 months for the vast majority of patients.86 Another study showed that 50% of ABVD-treated patients recover normal sperm characteristics 12–18 months after chemotherapy and 57% after 24 months.11 We have previously found that the ABVD regimen was less toxic than CHOP (composed of an alkylant agent [cyclophosphamide], an intercalating agent [hydroxydaunorubicin], an alcaloid agent [oncovin], and one corticosteroid [prednisone]) or MOPP-ABV (composed of an alkylant agent [mechlorethamine], an alkaloid agent [oncovin], an alkylant agent [procarbazine], one corticosteroid [prednisone], an anthracycline [adriamycin], an intercalant agent [bleomycin], and an alkaloid agent [vinblastine]) and 90% of lymphoma patients will recover normal sperm count 24 months after ABVD chemotherapy.13 The evaluation of post-treatment paternity in HL tumor survivors is indeed regularly described in the literature, with infertility estimated to affect 0–8% of patients treated with ABVD.86 However, beyond the alteration of spermatogenesis and the recovery of semen characteristics following chemotherapy, the question of genome or epigenome modifications exists and the determination of reproductive safety period after treatment is important.87 Some studies have investigated the effects of lymphoma treatment on sperm DNA using technologies such as fluorescence in situ hybridization (FISH) to study sperm aneuploidy,8,12,30,88,89,90,91 or several tests to study sperm DNA fragmentation,13,30,33,34 with contradictory results being presented. However, these kinds of studies cannot uncover the mutations found in animal species.92

The present study investigated the impact of AVBD chemotherapy on sperm DNA and epigenome of an HL patient who faced several RSMs in his couple following chemotherapy. Indeed, we postulated that this treatment resulted in the arrests of early post-implantation embryonic development that occurred in the partner of this patient. In this study, we have evaluated the genetic and epigenetic status of sperm from the ejaculate collected before and 26 months after chemotherapy. It is noteworthy that the couple experienced four early miscarriages after the end of chemotherapy, although the couple examination and semen parameters were normal immediately after the last miscarriage. In addition, the results of sperm DNA fragmentation and chromatin integrity studies were in the normal range 26 months after ABVD chemotherapy when the fourth miscarriage occurred. Other studies have investigated DNA fragmentation before and after chemotherapy. Smit et al.33 found no difference in DFI level before and after chemotherapy in 15 HL patients as did Stalh et al.34 who investigated 5 HL patients. In a previous study, we show that DFI remained increased 24 months after chemotherapy compared to the control group, but the DFI level was decreased compared to pretreatment values.13 O’Flaherty et al.93 investigated HL patients and found no difference in DFI level between before and after 24 months of treatment, but reported a difference using a single-cell gel electrophoresis (COMET) assay, suggesting that the COMET assay is a more sensitive measure of sperm DNA damage than SCSA.30 In our patient, the normal levels of sperm DNA fragmentation and chromatin evaluation 26 months after chemotherapy, despite the miscarriages, prompted us to investigate the sperm genome and epigenome.

Sperm DNA analysis detected new potential somatic variants in post-CT compared to pre-CT samples. Therefore, we investigated the mutational signature, as some chemotherapeutic agents can target specific motifs,83 such as cisplatin, which induces dinucleotide substitutions, possibly due to cisplatin’s tendency to form intra- and interstrand crosslinks between purine bases in DNA.94 At a genome-wide level, the mutational signature analysis of post-CT but also of pre-CT revealed that most of the SBS corresponded to C>T and T>C transitions, especially at CpG sites. Such transitions are typical features of the SBS1 and SBS5 signatures, which are age-related in both cancer and normal cells as defined in the COSMIC database.95 C>T transitions correspond to spontaneous deamination of methylated CpG sites and are also found in normal cells.96 Identified INDELs mostly corresponded to T insertions or deletions, which is a typical signature of Insertion-Deletion 1 (ID1) and 2 (ID2), corresponding to small insertions and deletions signatures, but could result from slippage during DNA replication.95 Taken together, the mutational analysis highlighted signatures already identified in normal cells and observed in pre-CT, suggesting that ABVD treatment does not have a mutagenic effect on the patient’s sperm DNA. We next refined our selection (reducing false positives by selecting variants detected by two tools) and selected the most frequent new variants (> 50%) to identify a more “localized” origin that could explain these RSMs. Sperm DNA analysis was able to identify 403 variants between pre-CT and post-CT samples potentially related to the ABVD chemotherapy. Inclusion of additional samples would be necessary to further improve variant detection and variant frequency estimation. Few SNPs (n=136) and INDELs (n=267) displayed a major frequency difference (≥50%) after chemotherapy. The variant annotation analysis revealed that no variants were detected in CDS regions, which leads us to believe that the ABVD treatment did not critically affect protein-coding gene functions for this patient. No variants were found in 5’-UTR regions, only three in 3’-UTR regions and 56 in regulatory regions. We hypothesize that altering regulatory elements could modify the binding of transcription factors and thus affect the expression of genes potentially important for embryogenesis, resulting in developmental disorders and/or miscarriages. While the set of annotated variants was not significantly associated with specific biological processes, 28 were related to genes important for and/or expressed during embryo development among which five were located in potential regulatory regions. A large proportion of the variants (91.0%) associated with predicted regulatory elements corresponded to promoter regions, promoter flanking regions and CCCTC-binding factor (CTCF) binding sites (DNA binding sites of a repressor transcription factor). It is also important to note that CTCF is a ubiquitously expressed and highly conserved 11-zinc finger protein that regulates gene expression at developmentally-regulated loci and their binding sites are mutated in many cancer types.97 It is noteworthy that 104 variants are also located into intergenic regions including one variant (chrUn_KI270746v1_40552_T/A) that was detected with a ∆Frequency >80% between pre-CT and post-CT samples. Nevertheless, no functional implication can be easily deduced for these variants since they are not associated with coding or noncoding genes.

The epigenome analysis resulted in the identification of 129 regions (hqDMRs) showing an altered methylated status after chemotherapy in motile sperm. The vast majority (76.7%) of those hqDMRs corresponded to hypomethylated regions (99 hypo- vs 30 hyper-methylated regions). Those hqDRMs were associated with 87 DMGs showing a chemotherapy-induced alteration of their sperm DNA methylation including seven genes (PTBP1, WDR70, SSTR5-AS1, CENPX, JARID2, SH2D4B, and DGKZ) expressed during early embryonic stages and six associated with embryo development based on annotation (ARHGAP35, NCOR2, SATB2, FOXH1, HPN, and XAB2). We hypothesize that altering the methylation status of such genes could modify their expression pattern and thus affect embryo development, leading to potential miscarriages.

Genomic imprinting refers to the epigenetic mechanism that results in the parent-of-origin mono-allelic expression of autosomal genes. Indeed, failure to establish correct germline-specific DNA methylation patterns has serious consequences for post-fertilization development primarily due to the necessity for epigenetic marking of genomic imprints. It was reported that the promoters of many key genes involved in early embryonic development are hypomethylated in mature sperm.98 Paternal genome methylation pattern could thus influence embryonic development through regulation of developmental gene expression.99 Importantly, none of the detected DMGs corresponded to known imprinted genes.

It is critical to highlight the preliminary nature of the findings in this study, which are based on a single case. Therefore, a cautious interpretation of the presented data is necessary. As a national reference center for fertility preservation in cancer patients, we have been collecting sperm samples from cancer patients for over two decades. Throughout this period, no other patients experiencing RSMs following chemotherapy consulted our center, making this case exceptionally unique as the patient experienced four RSMs before seeking consultation. It is more common for patients to seek our expertise after completing their chemotherapy, either to use cryopreserved sperm as they approach the end of their treatment or following a period of infertility. Despite the inherent limitations in obtaining a broader set of samples, we have extracted as much relevant information as possible using rigorous methods. The presented results should be considered as an initial exploration rather than definitive conclusions. To avoid any subjective evaluations, we recommend conducting additional research to identify other similar cases. This will contribute to future studies that will include a larger group of patients in order to validate and possibly replicate our findings. Such studies would provide a more comprehensive insight into the underlying mechanisms at play. Our study acts as a stepping stone for broader investigations, underscoring the urgent need for expanded research efforts to thoroughly ascertain the impact of cancer treatments on the incidence of RSMs.

In another case report, a woman with NHL treated with R-EPOCH experienced recurrent miscarriages and was found to carry the methylenetetrahydrofolate reductase (MTHFR) C677T mutation in a homozygous state.100 This mutation affects the enzyme MTHFR, which plays a crucial role in folic acid metabolism and regulation of the methylation cycle. Although our patient does not carry this mutation, this case highlights how epigenetic alterations, such as those resulting from DNA methylation defects, can significantly impact fertility. In our patient, we identified a thymine deletion upstream of the dihydrofolate reductase pseudogene 2 (DHFRP2) gene, the pseudogene of dihydrofolate reductase (DHFR), which encodes one of the key enzymes involved in methylation cycle reactions. Furthermore, the case report demonstrated that a treatment providing the product of the defective enzyme successfully restored fertility. While we were unable to pinpoint the exact cause of infertility following ABVD treatment in our patient, a similar therapeutic approach could be considered if comparable alterations are identified. This could be particularly beneficial for young patients who did not have the opportunity to cryopreserve gametes.

Here we characterized the DNA and methylation status of frozen sperm from an infertile HL patient before and after ABVD chemotherapy. It is noteworthy that the use of cryopreserved sperm before chemotherapy led to the birth of a healthy child. In contrast, the four spontaneous pregnancies that occurred between 11 and 26 months after the end of chemotherapy, following natural intercourse, all resulted in miscarriages. This clinical history shows that the sperm was able to fertilize but did not enable normal post-implantation embryonic development. Our results suggest that the observed RSMs following ABVD chemotherapy could result from modifications to the sperm DNA and epigenome observed 26 months after chemotherapy while normal classical sperm parameters have recovered. While this experimental design should increase the identification of genomic and epigenomic changes induced by the treatment, it cannot be ruled out that some of the measured changes could result from the cancer evolution, or from the patient gonadal environment. Overall, we were able to detect variants and DMRs between the pre-CT and post-CT samples. While no variants were detected in coding regions, several genes potentially important for embryo development contained variants in their promoter region or showed an altered methylation status. Although our data suggest that ABVD treatment did not have a mutagenic effect on the patient’s sperm DNA and that it did not critically affect protein-coding gene functions, it could ultimately be hypothesized that alteration of regulatory elements may alter the binding of transcription factors and thus affect the expression of genes potentially important for embryogenesis, resulting in developmental disorders and/or miscarriages. We also suggest that altering the methylation status of such genes could modify their expression pattern and thus affect embryo development, leading again to potential miscarriages. Although these hypotheses are based on data from a single patient with a single sample before and after chemotherapy, we believe that the present work paves the way for further studies investigating genetic and epigenetic changes in the sperm of cancer patients who have experienced RSMs. We argue for the introduction of such explorations to understand the potential reproductive toxicity of therapeutic agents.

Further follow-up studies are needed to assess the reversibility of genomic and epigenomic changes over time and to establish when a patient who was unable to preserve gametes before chemotherapy might safely procreate with minimal risk. However, our results strongly emphasize the critical need for proposing cryopreservation to all patients before any gonadotoxic treatments.

AUTHOR CONTRIBUTIONS

FC and LB supervised the research and designed the study. GL, FC and LB wrote the manuscript. BE, NM, MHMA and JM prepared the samples. FC, GL, CG, CK, ADR, and NA prepared, analyzed, and interpreted data. LB and JM interpreted data. NDR, CR, ASN, MHMA, CK and JM contributed to the manuscript. All authors read and approved the final version of the manuscript.

COMPETING INTERESTS

All authors declare no competing interests.

Supplementary Figure 1

Whole Genome Sequencing (WGS) data preprocessing and analysis. The workflow includes: (a) read mapping on the human genome and quality control, pre-processing steps, variant calling based on three tools including VarScan, HaplotypeCaller and Mutect2; (b) filtration and annotation of variants.

Supplementary Figure 2

Reduced Representation Bisulfite Sequencing (RRBS) data preprocessing and analysis. The workflow includes four distinct steps: read mapping on the human genome and quality control (STEP1), CpG detection and methylation level (STEP2), detection of high-quality differentially methylated regions (hqDMRs) (STEP3) and, gene association (STEP4).

Supplementary Figure 3

Whole Genome Sequencing (WGS) coverage and number of detected variants. (a) Coverage of the WGS data before (pre-CT) and after chemotherapy (post-CT). Fraction of reads equal or superior to the depth for pre-CT and post-CT samples. (b) Fraction of variant equal or superior to the difference of variant frequency before (pre-CT) and after chemotherapy (post-CT).

Supplementary Figure 4

Mutational signature analysis of the single base substitution (SBSs). Mutational signature before (pre-CT) and after chemotherapy (post-CT) and mutational signature of differential variants between pre-CT and post-CT samples (having a difference of variant frequency ≥50%).

Supplementary Figure 5

Mutational signature analysis of the double base substitutions (DBSs). Mutational signature before (pre-CT) and after chemotherapy (post-CT) samples and mutational signature of differential variants between pre-CT and post-CT samples (having a difference of variant frequency ≥50%).

Supplementary Figure 6

Mutational signature analysis of insertion/deletion variants (INDELs). Mutational signature before (pre-CT) and after chemotherapy (post-CT) and mutational signature of differential variants between pre-CT and post-CT samples (having a difference of variant frequency ≥50%).

Supplementary Figure 7

IGV representation of one detected variant in a predicted regulatory feature located upstream of the CLEC4C gene. The variant is the transversion of a T into a A at the position 7 749 685 on chromosome 12.

Supplementary Figure 8

IGV representation of two detected variants with the highest difference of frequency between before and after chemotherapy. (a) Representation of the transversion T into A at the position 228633373 on chromosome 1. (b) Represensation of the transversion T into A at the position 40552 on the unknown chromosome KI270746v1.

Supplementary Figure 9

Principal Component Analysis (PCA) analysis of the methylation status of samples. The four samples (“motile” and “other” fractions, before [pre-CT] and after chemotherapy [post-CT]) were projected on the two first components of the PCA.

Supplementary Figure 10

Methylation status and expression pattern of genes expressed during early embryogenesis. The percentage of methylation (y-axis) of the hqDMR(s) (left of each panel) and the log2-transformed expression signals (y-axis) at 1-cell, 2-cell, 4-cell, 8-cell, blastocyst and morula embryonic stages (x-axis) (right of each panel) are displayed for (a) WDR70, (b) PTBP1, (c) SSTR5-AS1, (d) CENPX, (e) EPS8L1, (f) JARID2, (g) SH2E4B, and (h) DGKZ.

Supplementary Figure 11

Methylation status of five differentially methylated genes (DMGs) associated with embryonic development. The percentage of methylation (y-axis) of the hqDMR(s) are displayed for (a) ARHGAP35, (b) NCOR2, (c) SATB2, (d) FOXH1, (e) HPN and (f) XAB2.

Table S2

Variants detected after successive filtration and that have a difference of frequency of >50% between PreCT and PostCT. Related genes known to be involved in embryo development or expressed during early embryo development are indicated.

Supplementary Results

Variant-associated genes

One of the variants (chr12_7749685_T/A) is located in the promoter of CLEC4C which is known to be a marker of dendritic cells involved in immune response1 but it has never been associated with major issues during embryo development. Other detected variants were closely related to regulatory regions of genes which are known to be important for mitotic spindle such as myosin IXB (MYO9B; located in promoter), tubulin gamma complex component 3 (TUBGCP3; CTCF binding site) and solute carrier family 2 member 3 (SLC2A3). The latter is part of the glucose transporter family (GLUT) which is required for preimplantation embryo development in the mouse. Deletion of this gene is associated with embryonic lethality in the mouse.2 The chr1_228633373_T/A variant (∆Frequency=89%) is located in the promoter region of the RNA 5S ribosomal 13 (RNA5S13) ribosomal gene that includes multiple deletions detected in both pre-CT and post-CT samples. It is thus unlikely that the variant could have a consequence on gene transcription. Thirteen variants were detected in regulatory regions and were predicted to be associated with genes important for embryo development, several of them being known from the literature to be link to embryo death outcome (cadherin 2 [CDH2], RAS P21 protein activator 3 [RASA3], lymphocyte antigen 6 family member E [LY6E], OPA1 mitochondrial dynamin like GTPase [OPA1], trinucleotide repeat containing adaptor 6A [TNRC6A], STE20 related adaptor alpha [STRADA], and TATA-box binding protein associated factor 4 [TAF4]). CDH2 is part of the CHD family of chromatin remodelers and it has been shown that mutations in its DNA binding domain can lead to a general growth delay and death prior to birth in mice;3 RASA3 encodes for a GTPase activating protein which targets R-Ras and RAP1 and its inactivation or deletion leads to severe hemorrhages and embryonic lethality in mice;4 and, LY6E encodes for a plasma membrane protein and its depletion in mice has been shown to be associated with embryonic lethality at mid-gestation (embryonic day 15.5 = E15.5).5 It is noteworthy that the depletion of this receptor in SynT-I cells alters syncytiotrophoblast fusion and placental morphogenesis, causing embryonic lethality in mice.6 OPA1 encodes a mitochondrial dynamin-related GTPase and its deletion results in embryonic lethality at E9 in the mouse.7 The TNRC6a gene product is involved in post-transcriptional gene silencing and it has been shown that its deletion in the mouse is associated with lethality at mid-gestation.8 STRADA plays a role in the mTor pathway and Strada^−/− mice exhibit high rates of perinatal mortality.9 TAF4 is part of the TFIID core module, a multiprotein complex involved in the initiation of gene transcription. Its deletion results in lethality at E9.5 in mice.10 Only one variant (chr20_45646223_C/T) was located into an hqDMR identified in the “other” fraction and corresponded to a C>T transition which may explain the altered methylation status. This variant is associated to the WAP four-disulfide core domain 11 (WFDC11) gene which is mainly expressed in epididymis and testis11 and encodes for a protease inhibitor with an antimicrobial activity. This gene could be involved in infertility with a bad sperm maturation through epididymis but this is unlikely here since fecundations occurred.^12–16

hqDMR-associated genes

Among the 13 embryo-related DMGs, PTBP1 belongs to the subfamily of ubiquitously expressed heterogeneous nuclear ribonucleoproteins (hnRNPs). It is involved in (pre-)mRNA processing, metabolism and transport and has a function in neurogenesis.17 Disruption of this gene in the mouse results of resorbed/degraded embryos at mid-gestation and the study also showed that PTBP1 is important for post-implantation development.2 JARID2 is part of the jumonji demethylase protein family and plays a role in heart development and neural tube formation. It is known to be a regulator of embryonic development by modulating the methylation level of H3K27me3.18 The XPA binding protein 2 (XAB2) is involved in mRNA splicing and one study has demonstrated its importance for mouse embryogenesis as the homozygous disruption results in arrest at the morula stage and thus preimplantation lethality.19 The forkhead box H1 (FOXH1) gene encodes a protein which is able to bind to SMAD2 and to activate an activin response element. Its disruption in the mouse results in embryonic lethality and a range of defects.20