Copy number variation analysis detects novel candidate genes involved in follicular growth and oocyte maturation in a cohort of premature ovarian failure cases

Abstract

STUDY QUESTION

Can spontaneous premature ovarian failure (POF) patients derived from population-based biobanks reveal the association between copy number variations (CNVs) and POF?

SUMMARY ANSWER

CNVs can hamper the functional capacity of ovaries by disrupting key genes and pathways essential for proper ovarian function.

WHAT IS KNOWN ALREADY

POF is defined as the cessation of ovarian function before the age of 40 years. POF is a major reason for female infertility, although its cause remains largely unknown.

STUDY DESIGN, SIZE, DURATION

The current retrospective CNV study included 301 spontaneous POF patients and 3188 control individuals registered between 2003 and 2014 at Estonian Genome Center at the University of Tartu (EGCUT) biobank.

PARTICIPANTS/MATERIALS, SETTING, METHODS

DNA samples from 301 spontaneous POF patients were genotyped by Illumina HumanCoreExome (258 samples) and HumanOmniExpress (43 samples) BeadChip arrays. Genotype and phenotype information was drawn from the EGCUT for the 3188 control population samples, previously genotyped with HumanCNV370 and HumanOmniExpress BeadChip arrays. All identified CNVs were subjected to functional enrichment studies for highlighting the POF pathogenesis. Real-time quantitative PCR was used to validate a subset of CNVs. Whole-exome sequencing was performed on six patients carrying hemizygous deletions that encompass genes essential for meiosis or folliculogenesis.

MAIN RESULTS AND THE ROLE OF CHANCE

Eleven novel microdeletions and microduplications that encompass genes relevant to POF were identified. For example, FMN2 (1q43) and SGOL2 (2q33.1) are essential for meiotic progression, while TBP (6q27), SCARB1 (12q24.31), BNC1 (15q25) and ARFGAP3 (22q13.2) are involved in follicular growth and oocyte maturation. The importance of recently discovered hemizygous microdeletions of meiotic genes SYCE1 (10q26.3) and CPEB1 (15q25.2) in POF patients was also corroborated.

LIMITATIONS, REASONS FOR CAUTION

This is a descriptive analysis and no functional studies were performed. Anamnestic data obtained from population-based biobank lacked clinical, biological (hormone levels) or ultrasonographical data, and spontaneous POF was predicted retrospectively by excluding known extraovarian causes for premature menopause.

WIDER IMPLICATIONS OF THE FINDINGS

The present study, with high number of spontaneous POF cases, provides novel data on associations between the genomic aberrations and premature menopause of ovarian cause and demonstrates that population-based biobanks are powerful source of biological samples and clinical data to reveal novel genetic lesions associated with human reproductive health and disease, including POF.

STUDY FUNDING/COMPETING INTEREST

This study was supported by the Estonian Ministry of Education and Research (IUT20-43, IUT20-60, IUT34-16, SF0180027s10 and 9205), Enterprise Estonia (EU30020 and EU48695), Eureka's EUROSTARS programme (NOTED, EU41564), grants from European Union's FP7 Marie Curie Industry-Academia Partnerships and Pathways (IAPP, SARM, |EU324509) and Horizon 2020 innovation programme (WIDENLIFE, 692065), Academy of Finland and the Sigrid Juselius Foundation.

Introduction

Menopause is a normal part of the female aging process. The median age of natural menopause is approximately 51 years, although variance in age is strongly determined by genetic (de Bruin et al., 2001) and environmental factors such as smoking, parity and diet (Lund, 2008). However, ∼1% of women experience premature menopause with cessation of ovarian function before the age of 40 (Haller-Kikkatalo et al., 2015), a condition commonly known as premature ovarian failure (POF). Owing to estrogen deficiency, POF patients suffer from symptoms similar to menopause, but also present a higher risk of developing osteoporosis, cardiovascular diseases and neurodegenerative disorders (Goswami and Conway, 2005). However, the most widely recognized effects of POF concern the reproductive system. The first symptoms that usually precede POF are irregular menstrual cycles (Haller-Kikkatalo et al., 2015) and female infertility or poor outcome of infertility treatment (de Boer et al., 2003).

The designation of POF has been under discussion for a long time, but it is still preferred among many alternatives, including ovarian insufficiency, dysfunction or insult (Shelling, 2010). POF is highly heterogeneous and can be associated with iatrogenic factors such as oophorectomy, ovarian resection or cystectomy reducing ovarian reserve (Busacca et al., 2006) or chemotherapy, and systemic diseases (Group, 2002). Furthermore, to minimize misinterpretation (Shah and Nagarajan, 2014) we use the term ‘spontaneous POF’ to define retrospectively ovarian quiescence in patients younger than 40 who do not present with any other known extraovarian causes for premature menopause. Therefore, spontaneous POF can arise as a result of noniatrogenic conditions affecting primarily ovaries, and include autoimmune reactions (Monnier-Barbarino et al., 2005) and genetic factors (Pouresmaeili and Fazeli, 2014).

Studies of familial spontaneous POF cases showed that genetic background strongly influences the onset of POF with familial incidences ranging from around 4% up to 30% of cases (Cramer et al., 1995; Conway et al., 1996; Vegetti et al., 1998). Disorders of the X chromosome are considered to be one of the main causes of POF, and conventional karyotyping has identified several X chromosome regions containing genes required for normal ovarian function (Beke et al., 2013). Mutations in X-linked genes have been described and include FMR1 (MIM*309550), BMP15 (MIM*300247) and DACH2 (MIM*300608). In addition, some autosomal genes are associated with POF, such as GDF9 (MIM*601918), FSHR (MIM*136435), LHR (MIM*152790) and NR5A1 (MIM*184757) (reviewed in Persani et al., 2010). The association between mutations in meiotic genes and ovarian decline has also been shown in POF patients (Lacombe et al., 2006; Caburet et al., 2014) and women with 46,XX gonadal dysgenesis (Zangen et al., 2011), highlighting the importance of meiotic genes in the regulation of ovarian function. Nevertheless, the cause of premature menopause remains unknown in many cases.

Advanced DNA microarray technologies have highlighted the importance of submicroscopic copy number variations (CNVs) in the pathogenesis of rare and complex diseases (Vulto-van Silfhout et al., 2013). CNVs represent deletions, duplications and insertions that are >1 kb in size and are present in variable copy number compared with the reference human genome (Feuk et al., 2006). Associations between CNVs and a certain disease may reveal novel candidate regions and genes involved in the pathogenesis of the disease. A limited number of studies have already demonstrated a possible association between the POF phenotype and CNVs. However, most of these studies used low-resolution array comparative genome hybridization (aCGH) platforms with a mean spatial resolution of 0.7 Mb or average probe spacing of 21.7 kb (Aboura et al., 2009; Ledig et al., 2010), or targeted the X chromosome only (Quilter et al., 2010; Knauff et al., 2011).

Until now, only two CNV studies in POF patients used high-resolution array platforms, either whole-genome SNP array (McGuire et al., 2011) or a custom-designed aCGH with average probe spacing of 2.2 kb that targeted genes implicated in gender development (Norling et al., 2014). Using high-resolution arrays, these studies were able to identify novel microdeletions and microduplications previously not associated with POF. However, the cohort sizes of these studies remained relatively modest, and involved 89 and 26 patients, respectively. Thus, in the current study we used high-resolution SNP genotyping of 301 spontaneous POF patients derived from Estonian Genome Center at the University of Tartu (EGCUT) biobank to reveal novel CNV regions and candidate genes related to POF phenotype. In addition, whole-exome sequencing (WES) was performed on genomic DNA of six women with spontaneous POF carrying hemizygous deletions to determine whether these deletions can result in unmasking of recessive mutations in POF candidate genes in undeleted regions of the homologous chromosomes, or reveal other POF-associated variants in the exome. The data reported here reveal new associations between rare CNVs and susceptibility to spontaneous POF. Furthermore, our study indicates that using samples from population-based biobanks may be a useful and efficient way to recruit patients for large-scale retrospective CNV and disease association studies.

Materials and Methods

Ethical approval

The Ethics Review Committee on Human Research of the University of Tartu and Scientific Committee of the EGCUT approved this study.

Population-depictive cohort samples from EGCUT

EGCUT is the population-based biobank (www.geenivaramu.ee, 6 June 2016, date last accessed) formed according to the Estonian Gene Research Act and all participants have signed the broad informed consent. The cohort of EGCUT currently includes >51 000 gene donors over 18 years of age. EGCUT represents 5% of Estonian population and closely reflects the age, gender and geographical distribution of the Estonian population. All subjects were recruited randomly and voluntarily by general practitioners and physicians in hospitals. A Computer Assisted Personal Interview (CAPI) was completed at a doctor's office and included personal, genealogical, educational, occupational history and lifestyle data. Anthropometric measurements, blood pressure and resting heart rate were measured, and venous blood was taken during the visit. Medical history, including parameters of reproductive health, was collected retrospectively during the recruitment. All past and current medical conditions were recorded according to the International Statistical Classification of Diseases (ICD-10) codes.

Women with spontaneous POF

Participants of the study were selected from 34 041 women at the EGCUT according to phenotype. The ovarian cause for spontaneous premature menopause was considered if secondary amenorrhea occurred before the age of 40 years, but no exclusive criteria were reported by participants or recorded by family physicians. The exclusive criteria were: iatrogenic manipulation (surgical hysterectomy, bilateral oophorectomy or unilateral oophorectomy diminishing ovarian reserve, use of oral contraceptives or other medications causing amenorrhea), comorbidities affecting menstrual cycle (including obesity with body mass index (BMI) ≥30 kg/m², hyperprolactinemia, renal failure and HIV infection), sex chromosome abnormalities (including Turner syndrome, 46,XY or 47,XXX), developmental disorders (uterine agenesis), as well as primary amenorrhea. As a result of the selection, this study included a total of 301 women with spontaneous POF (Table I). The average age at the time of the study was 57.8 ± 12.5 years (mean ± SD) and on average, 20.7 years had passed since the onset of menopause, which started at an average age of 37.2 years. The average BMI value by the time of study was 28.9 kg/m², indicating overweight among participants (BMI 25.1–30 kg/m² indicates overweight and BMI ≥ 30 kg/m² stands for obesity). Also, the waist-to-hip ratio of study participants exceeded 0.8, indicating a moderately increased risk for being overweight (0.8–0.85 indicates moderate risk, while >0.85 indicates substantial risk).

Table I.

Anamnestic data of 301 women with spontaneous premature ovarian failure (POF).

Age at study (years)	57.8 ± 12.5
Duration of amenorrhea (years)	20.7 ± 11.6
Smoking status (n)
Never smoked	206 (68.4%, 62.8–73.6)
Former smoker	27 (9.0%, 6.1–12.9)
Current smoker	68 (22.6%, 18.1–27.8)
Height (cm)	163.0 ± 6.5
BMI (kg/m2)	28.9 ± 6.7
Waist-to-hip ratio	0.84 ± 0.08
Age at menarche (years)	13.8 ± 1.7
Length of menstrual cycle at age 25–35
25–29 days	174 (57.8%, 52.0–63.4)
≤20 days	8 (2.7%, 1.2–5.4)
21–24 days	57 (18.9%, 14.8–23.9)
30–35 days	12 (4.0%, 2.2–7.0)
>35 days	1 (0.3%, 0–2.1)
Irregular	25 (8.3%, 5.6–12.2)
Do not know	12 (4.0%, 2.2–7.0)
Duration of fertility (years)	23.6 ± 5.1
Age of 1st pregnancy (years)	22.5 ± 3.6
No. of pregnancies	3.6 ± 2.0
No. of live births	2.0 ± 1.1
Age at menopause (years)	37.2 ± 4.4

Data are means ± SD or count (percentage, 95% CI of percentage).

Since the EGCUT cohort is a volunteer-based longitudinal population biobank, additional kinship analysis using the web-based PLINK v1.07 toolset (http://pngu.mgh.harvard.edu/purcell/plink/, 6 June 2016, date last accessed) was performed (Purcell et al., 2007) to identify possible familial cases among the POF patients.

Reference data

The Database of Genomic Variants (DGV; http://dgv.tcag.ca/, 6 June 2016, date last accessed) was used as a reference dataset to exclude benign polymorphisms. DGV holds information on common CNVs found in more than 20 000 healthy control samples and serves as a catalog of control data for correlating genomic variation with phenotypic data (MacDonald et al., 2013). In addition, genotype and phenotype information was drawn for the Estonian general population samples (n = 3188) from the EGCUT, previously subjected to SNP genotyping with HumanCNV370 (n = 489) and HumanOmniExpress (n = 2699) BeadChip arrays (Illumina, Inc., San Diego, CA, USA) to determine and exclude benign population-specific CNV regions. The derived EGCUT dataset represented control group of women with normal age at menopause, as the group included >41 years old pre- and post-menopausal women, but excluded potential POF cases.

SNP genotyping and CNV calling

The study design is represented schematically in Fig. 1. Infinium^® II whole-genome genotyping assay with HumanOmniExpress (n = 43) or HumanCoreExome (n = 258) BeadChip arrays (Illumina, Inc.) were used for whole-genome SNP genotyping. The HumanOmniExpress BeadChip contains >715 000 SNP markers with median marker spacing of 2.1 kb, while HumanCoreExome BeadChip has 547 644 markers, with 265 919 of them covering exonic regions, and a median marker spacing of 1.9 kb. SNP genotyping wet-lab experiments were performed according to the manufacturer's protocols.

Figure 1.

Schematic representation of study design. A total of 301 spontaneous premature ovarian failure (POF) patients were enrolled in the study. After single nucleotide polymorphism (SNP) genotyping and copy number variation (CNV) calling, genome-wide CNV load analysis was performed per individual to estimate the effect of CNV load in the genome on time of menopause. Subsequently, all the detected CNVs were compared against the EGCUT control population (n = 3188) to identify nonoverlapping CNV regions (CNVRs) (n = 220), for which functional enrichment and association studies were performed. After the following interpretation of each CNVR, six individuals with hemizygous deletions were also selected for whole-exome sequencing (WES) to determine whether identified deletions can result in unmasking recessive mutations or other POF-associated variants in the exome.

Genotypes were called by GenomeStudio software Genotyping Module v.3.1 (Illumina, Inc.). A call rate of >98% was accepted as the primary quality control for each sample on both arrays. Log R Ratio and B Allele Frequency values generated by the GenomeStudio software were applied in further CNV calling using two independent Hidden Markov Model-based algorithms: QuantiSNP v.2.1 (Colella et al., 2007) and PennCNV (Wang et al., 2007). Parameters suggested by the authors were applied to each algorithm. The genomic size threshold for CNVs was set to 50 kb for the HumanOmniExpress array and to 100 kb for the HumanCoreExome array, and CNV contained at least 10 consecutive SNP markers. To minimize the number of false-positive findings, initial CNV calls from the two algorithms were merged and only CNVs called by both algorithms and visually confirmed in GenomeStudio Genome Viewer were considered for further data interpretation. All CNVs identified in this study were submitted to ClinVar database (http://www.ncbi.nlm.nih.gov/clinvar/, 6 June 2016, date last accessed; accession numbers SCV000212280 to SCV000212499).

For the EGCUT control group, the HumanCNV370 and HumanOmniExpress microarray data were processed in the same manner by performing independent CNV calling using QuantiSNP and PennCNV and by considering only CNV regions that were called by both algorithms.

Interpretation of CNV results

The genomic context of aberrant regions was studied using the UCSC Genome Browser database (http://genome.ucsc.edu/) along with the OMIM database (www.omim.org, 6 June 2016, date last accessed) to evaluate the potential clinical relevance of a particular CNV. The interpretation of CNVs was performed according to the ACMG Guidelines (Kearney et al., 2011). Briefly, chromosomal aberrations were deemed clinically significant if they overlapped with a genomic region associated with a well-established syndrome or contained a gene or a part of a gene implicated in a known disorder. Findings were considered likely benign if they were present in healthy individuals (e.g. >1% of Estonian population or DGV) and/or they were gene deserts or gene-poor regions and/or did not encompass any previously known disease-associated genes. All remaining findings were categorized as variants of unknown clinical significance.

Functional analysis of results

Functional annotation and enrichment analysis was carried out using g:Profiler gGOSt web-based software (http://biit.cs.ut.ee/gprofiler/, 6 June 2016, date last accessed) (Reimand et al., 2007, 2011). The g:Profiler public web server provides a comprehensive set of functional annotation tools for clustering functionally related genes according to different criteria such as Gene Ontology terms and biological pathways. Enrichment for functional terms was considered to be statistically significant, if the multiple testing corrected P-value was <0.05. A 20 Mb hemizygous deletion on the X chromosome, found in one of the POF patients, was excluded from the analysis to avoid bias.

CNV validation

Confirmation of a subset of CNVs (n = 14) was done using SYBR-Green based real-time quantitative PCR (qPCR). CNVs were selected for experimental confirmation if they met the following criteria: (i) there were no data reported in the DGV and/or CNV was present in <1% of Estonian population, and (ii) CNV harbored previously reported POF regions and genes, or contained genes that may potentially impact the age at menopause based on their biological function. Primers were custom-designed in a web-based freeware Primer3web (http://primer3.ut.ee/, 6 June 2016, date last accessed) (Untergasser et al., 2012). A full list of primers used for qPCR analysis is available upon request. The analysis was performed on a 7900 HT Real-Time PCR system (Applied Biosystems, Carlsbad, CA, USA). These data were acquired using SDS 2.2.2 software (Applied Biosystems) and further processed using qBase+ (Biogazelle, Ghent, Belgium). Analysis was performed as relative quantification using the Pfaffl method of calculation while taking into account the amplification efficiencies of each primer pair (Pfaffl, 2001). All selected CNVs have been successfully validated.

Whole-exome sequencing and data analysis

Genomic libraries were prepared according to the Illumina Nextera Rapid Capture Exome protocol (Illumina, Inc.). The captured libraries were sequenced on HiSeq2500 platform (Illumina, Inc.) with 2 × 101 bp paired-end reads, with 92.7% of bases sequenced above the quality of Q30. Demultiplexing was done with CASAVA 1.8.2. (Illumina, Inc.) allowing 1 mismatch in 8 bp index read. Number of reads varied between samples from 45 to 56 M.

Sequenced reads were aligned to the human reference genome (hg19, GRCh37) with the Burrows-Wheeler Aligner (BWA, version 0.6.1). SAMtools (version 0.1.18) was used to filter out reads marked as PCR duplicates and reads not in a proper read pair. Single-nucleotide substitutions and small indel variants were then called with GATK, after which variant sites more than 50 bp away from the nearest exome sequence capture target or with quality score <20 were filtered out. Variants were annotated using in-house developed scripts based on Ensembl API release 75, 1000 Genomes, dbSNP138, ExAC Browser, BIOBASE's Genome Trax™ database, and in-house whole-exome and whole-genome databases.

Statistical analysis

The R3.1.0 a Language and Environment (Free Software Foundation, Boston, MA, USA) was used for statistical analysis. The Welch two-sample t-test and the Wilcoxon rank sum test with continuity correction were used to assess differences in number and size for CNV calling between two SNP-arrays. Multivariate linear regression models adjusted for confounders were used to assess the associations between the number and size of CNVs and anamnestic data. A P-value of <0.05 was considered statistically significant.

Results

The association between CNV load and anamnestic data

After applying stringent filtering criteria of initial CNV calls, a total of 346 CNVs were detected on both the X chromosome and autosomal chromosomes in women with spontaneous POF. CNVs identified by both algorithms were spread over the entire genome and included 132 copy number losses (38.2%) and 214 copy number gains (61.8%).

To establish the association between the CNV load and general reproductive fitness, all identified CNVs in POF patients and women from the control group who have reached menopause (n = 1735) were subjected to a global genome-wide CNV load analysis per individual. Linear regression model adjusted by the age of menopause and genotyping array attributes did not reveal any difference in the number of CNVs detected between POF patients and controls (adjusted regression coefficient (ad r) = −0.012, P = 0.329). Similarly, there were no significant difference in the cumulative size of CNVs detected among POF patients compared with the controls (ad r = −77 658, P = 0.477). Linear regression model adjusted by genotyping array attributes revealed a negative association between the number of CNVs in the genome and the age of menopause (ad r = −0.04 CNVs per 1 year difference in menopause, P = 0.010) in a group of POF patients. When the same model was used for the control population, no association between the number of CNVs in the genome and the age of menopause was observed (ad r = −0.005 CNVs per 1 year difference in menopause, P = 0.709). Also, since the age at menopause determined the duration of fertility (r = 0.77, P < 0.00001 for POF patients and r = 0.84, P < 0.00001 for the control group), the number of CNVs was also negatively associated with the duration of fertility in POF patients (ad r = −0.04, P = 0.0008), but not in the control group (ad r = −0.016, P = 0.216), when the model was controlled by genotyping array attributes. Linear regression models were not able to detect any relationships between the number of CNVs and the length or the regularity of menstrual cycles, the number of pregnancies or live births, or parameters of body structure. The cumulative size of CNVs was positively associated with the age at first pregnancy if confounders affecting the age of pregnancy were considered in the statistical analysis (linear regression adjusted by age of menarche, number of live births and age at study, adjusted r = 80.1 kb per one additional year of the time to first pregnancy; P = 0.037). The mean age ± SD of first pregnancy was 22.5 ± 3.6 years and ranged from 15 to 36 years. The association was the same if the CNV detection genotyping array attributes were also considered (data not shown). This result suggests that overall CNV load can affect reproductive fitness in terms of the time of the first pregnancy. However, since confounding factors were not available for the control population, we were not able to perform similar association study between the cumulative sizes of the CNVs per genome with the age at first pregnancy.

Functional profiling of genes within CNV regions identifies enrichment in pathways related to the immune system functions

All identified CNVs were clustered into 95 deleted and 125 duplicated nonoverlapping copy number variation regions (CNVRs), respectively (Supplementary data, Table SI). Functional enrichment analysis on CNVRs was performed using the g:Profiler web-based software. Results of Gene Ontology (GO), KEGG pathway and Reactome (REAC) datasets with up to third hierarchical level are provided in Table II. The analysis revealed statistically significant networks related to immunological processes. In particular, a notable number of genes were involved in chemotaxis of lymphocytes (20.0% of genes, P = 2.83 × 10⁻²), monocytes (20.4% of genes, P = 8.49 × 10⁻³) and eosinophils (43.5% of genes, P = 7.68 × 10⁻⁶). Alterations in KEGG pathway of ‘chemokine signaling’ (P = 4.11 × 10⁻²), and REAC pathways of ‘innate immune system’ (P = 2.11 × 10⁻³), ‘adaptive immune system’ (P = 7.48 × 10⁻⁵) and ‘scavenging of heme from plasma’ (P = 2.41 × 10⁻³⁰) were observed in our spontaneous POF cases. None of these functional pathways were affected by chromosomal structural rearrangements in women with normal age at menopause, when a similar analysis was performed for EGCUT control individuals. These results may underlie the dysregulation of immune system in patients with spontaneous POF, as numerous interactions between the immune system and ovaries are involved in the regulation of the hypothalamic–pituitary–ovarian axis and maintaining growth and regression of both follicles and corpus luteum (Haller-Kikkatalo et al., 2012).

Table II.

Significant networks identified by pathway analysis in spontaneous premature ovarian failure (POF) patients.

Term ID	Term name	Level	Adjusted P-valuea
GO:0048247	Lymphocyte chemotaxis	1	2.83 × 10−2
GO:0002548	Monocyte chemotaxis	1	8.49 × 10−3
GO:0072677	Eosinophil migration	1	4.96 × 10−5
GO:0048245	Eosinophil chemotaxis	1.1	7.68 × 10−6
KEGG:04062	Chemokine signaling pathway	1	4.11 × 10−2
REAC:168249	Innate immune system	1	2.11 × 10−3
REAC:2454202	Fc epsilon receptor (FCERI) signaling	1.1	5.76 × 10−7
REAC:2029480	Fc gamma receptor (FCGR) dependent phagocytosis	1.2	2.00 × 10−19
REAC:166658	Complement cascade	1.3	2.41 × 10−27
REAC:5653656	Vesicle-mediated transport	1	5.80 × 10−10
REAC:2173782	Binding and uptake of ligands by scavenger receptors	1.1	3.68 × 10−23
REAC:2168880	Scavenging of heme from plasma	1.1.1	2.41 × 10−30
REAC:1280218	Adaptive immune system	1	7.48 × 10−5
REAC:198933	Immunoregulatory interactions between a lymphoid and a nonlymphoid cell	1.1	2.05 × 10−15

Functional enrichment analysis of genes within rearranged regions was performed using the web-based g:Profiler software. Gene Ontology (GO), KEGG pathway and Reactome (REAC) datasets with adjusted P-value < 0.05 were considered to be statistically significant.

^aMultiple testing corrected enrichment P-value.

Detection of potentially relevant regions for spontaneous POF

Subsequently, clustered CNVRs were screened against the DGV database to distinguish benign copy number changes from rare POF-associated CNVs. However, the exclusion of benign polymorphisms was done with caution, as genetic imbalances affecting gender development can depend on the sex of the individual, and this information is not always available for individuals included in the DGV. In addition, the Estonian population-based control group was used to exclude common variants in the Estonian population. Therefore, in subsequent case-by-case analyses, 93 out of 125 microduplications and 72 out of 95 microdeletions were classified as benign polymorphisms, and the remaining 55 CNVRs were considered to be rare copy number changes, as they were not reported in DGV nor present in the EGCUT control group (Supplementary data, Table SII). Emphasis for selecting CNVs for subsequent validation by qPCR was placed on unique aberrations that harbored previously reported POF candidate genes, or contained genes that may potentially impact the age at menopause based on their biological function. Using these criteria, 14 CNVs were selected as relevant in the susceptibility to POF, and were successfully validated by qPCR. Data from these CNVs, as well as the 20 Mb hemizygous X chromosome deletion, are summarized in Table III, while phenotypic data of CNV carriers are given in Table IV. Overlapping aberrations previously reported in POF patients were identified in five cases: two patients had duplications in the Xp22.31 region (Case14 and Case15) (Quilter et al., 2010), two patients had a 10q26.3 hemizygous deletion (Case7 and Case8) with identical breakpoints (based on microarray data), and one patient had a 15q25.2 hemizygous deletion (Case10) (McGuire et al., 2011). All of these women were unrelated according to kinship analysis. The remaining rare 11 CNVs identified in our study were considered to be novel.

Table III.

Summary of validated copy number variations (CNVs) in spontaneous premature ovarian failure (POF) cases.

Locus	Position (hg19)	Length (kb)	CN	Probe count	Genes within CNV	Cases (n = 301)	Controls (n = 3188)	CNV carriers among cases (%)	CNV carriers among controls (%)	Case ID
1q43	chr1: 240341241–240561727	189.9	1	37	FMN2	1	1	0.33	0.03	Case1
2p13.11	chr2: 73828493–73900329	71.8	1	13	ALMS1, NAT8	1	0	0.33	0	Case2
2q33.1	chr2: 200250898–201845999	1595.1	3	451	SATB2, FTCDNL1, C2orf69, TYW5, C2orf47, SPATS2L, KCTD18, SGOL2, AOX1, AOX2P, BZW1, CLK1, PPIL3, NIF3L1, ORC2, FAM126B	1	1	0.33	0.03	Case3
6q27	chr6: 170713690–170890384	176.7	3	70	FAM120B, PSMB1, TBP, PDCD2	1	1	0.33	0.03	Case4
7p14.3	chr7: 33639870–33730376	90.5	1	30	BBS9	1	10	0.33	0.31	Case5
9q22.31	chr9: 95063947–95179836	115.9	1	70	NOL8, CENPP, OGN, OMD	1	1	0.33	0.03	Case6
10q26.3c	chr10: 135256762–135379710	122.9	1	70	CYP2E1, SYCE1	2	3	0.66	0.09	Case7 Case8
12q24.31	chr12: 125260645–125321461	60.8	3	13	SCARB1	1	1	0.33	0.03	Case9
15q25.2c	chr15: 83213963–84811815	1597.8	1	325	CPEB1, AP3B2, FSD2, WHAMM, HOMER2, FAM103A1, C15orf40, BTBD1, TM6SF1, HDGFRP3, BNC1, SH3GL3, ADAMTSL3, EFTUD1P1	1	0	0.33	0	Case10
16p13.12	chr16: 13889247–14163635	274.3	3	35	ERCC4	1	0	0.33	0	Case11
17q12	chr17: ×–36220373	1404.8	3	327	ZNHIT3, MYO19, PIGW, GGNBP2, DHRS11, MRM1, LHX1, AATF, ACACA, C17orf78, TADA2A, DUSP14, SYNRG, DDX52, HNF1B	1	8	0.33	0.25	Case12
22q13.2	chr22: 43122720–43500212	377.5	1	54	ARFGAP3, PACSIN2, TTLL1	1	3	0.33	0.09	Case13
Xp22.31d	chrX: 6516735–8138080	1618.3	3	111	HDHD1, STS, VCX, PNPLA4	2	21b	0.66	0.66	Case14 Case15
Xq12	chrX: 66905875–67475065	569.2	3	20	AR, OPHN1	1	1	0.33	0.03	Case16
Xq22.1-q24	chrX: 99931059–120328627	20397.6	1	2001	SYTL4…GLUD2a	1	0	0.33	0	Case17

All the discrete regions have been validated by qPCR, except for 20 Mb Xq22.1-q24 deletion; del, deletion; dup, duplication; CN, copy number.

^a>100 genes in total, first and last genes in the region are indicated.

^bSTS gene is not covered in any of these 21 control individuals with overlapping Xp22.31 duplications.

^cFirst reported by McGuire et al (2010).

^dFirst reported by Quilter et al (2010); loci indicated in bold were chosen for subsequent whole-exome sequencing in corresponding CNV carriers.

Table IV.

Phenotype data of premature ovarian failure (POF) patients carrying aberrations with potentially clinical significance.

Case ID	Locus	CN	Phenotype
			Age at study (years)	Age of menarche (years)	Age at menopause (yr)	No. of pregnancies	No. of live births	Smoking status	Concomitant diseases (age of diagnose)
Case1	1q43	1	71	18	40a	4	2	Never smoked	Hypothyroidism (24 yr); primary hypertension (67 yr)
Case2	2p13.11	1	53	11	39	3	2	Never smoked	Adiposity (42 yr); hypertension with heart failure (50 yr)
Case3	2q33.1	3	41	11	33	5	2	Never smoked	Unspecified arthrosis (35 yr)
Case4	6q27	3	39	12	24	2	1 (IVF)	Current smoker	Arthritis (24 yr); asthma (29 yr)
Case5	7p14.3	1	58	15	32	4	2	Never smoked	Allergic contact dermatitis (45 yr); coxarthrosis and unspecified rheumatism (52 yr); post-menopausal osteoporosis (53 yr); primary hypertension (55 yr)
Case6	9q22.31	1	68	15	38	2	1	Never smoked	–
Case7	10q26.3	1	56	13	38	6	3	Never smoked	Allergic contact dermatitis (45 yr); hypertension with heart failure (50 yr); unspecified polyarthritis (57 yr);
Case8	10q26.3	1	69	18	30	0	0	Never smoked	Primary bilateral gonarthrosis (49 yr); hypertension with heart failure (52 yr); familial hypercholesterolemia (62 yr); Type 2 diabetes (67 yr)
Case9	12q24.31	3	51	16	40	1	1	Former smoker	–
Case10	15q25.2	1	55	13	38	3	2	Never smoked	–
Case11	16p13.12	3	50	15	39	5	3	Current smoker	Adiposity (41 yr)
Case12	17q12	3	74	15	39	2	2	Never smoked	Hypertension with heart failure (73 yr)
Case13	22q13.2	1	47	13	37	0	0	Current smoker	–
Case14	Xp22.31	3	52	13	40	1	1	Current smoker	Spondylosis with radiculopathy of unspecified location (33 yr)
Case15	Xp22.31	3	37	17	37	3	2	Current smoker	–
Case16	Xq12	3	31	13	25	3	3	Current smoker	–
Case17	Xq22.1-q24	1	52	12	40	3	2	Never smoked	–

Subsequent whole-exome sequencing was performed for cases indicated in bold; yr, year.

^aAge of 39 years and 6–11 months was rounded to 40 years.

Whole-exome sequencing data of six patients carrying hemizygous deletions

WES was performed on six female carriers of hemizygous deletions (Tables III and IV) that encompass genes essential for meiosis or folliculogenesis. The average read depth for sequenced exome was 82×. WES analysis did not reveal any rare POF-associated homozygous variants in genes within the CNV regions based on ExAC database and in-house whole-exome and whole-genome reference databases (Supplementary data, Table SIII). Additionally, WES identified heterozygous missense mutation in GDF9 (growth differentiation factor 9 at 5q31.1, rs61754582, c.1360C > T; p.Arg454Cys) in Case13, when analyzing whole exome data outside of CNV regions. The prevalence of GDF9 mutations in POF patients is estimated to be ∼1–4%, (Dixit et al., 2005; Laissue et al., 2006; Kovanci et al., 2007; Zhao et al., 2007); however, this particular variant is present in 0.4% of European population, thus most likely representing a nonpathogenic variant. In addition, secondary findings were identified by WES that might contribute to comorbid phenotypic features diagnosed in POF patients (Table V).

Table V.

Secondary findings identified by whole-exome sequencing in six premature ovarian failure (POF) patients.

Case ID	Gene	Transcript	Chr	Position (hg19)	rsID	Zygosity	cDNA	Protein	ClinVar accession	Associated phenotype	Variant involved in patient's phenotypea
Case1	SCN9A	NM_002977.3	2	167136962	rs182650126	Het	c.2215A>G	p.Ile739Val	SCV000191928.1	Small fiber neuropathy	Not likely
Case1	TMEM43	NM_024334.2	3	14180731	rs113449357	Het	c.934C>T	p.Arg312Trp	SCV000051602.1	Cardiomyopathy	Not likely
Case6	SPINK1	NM_003122.4	5	147207583	rs148954387	Het	c.194+2T>C	p.(-)	SCV000253884.1	Hereditary pancreatitis	Likely
Case6	NEBL	NM_006393.2	10	21157673	rs137973321	Het	c.604G>A	p.Gly202Arg	SCV000062382.3	Cardiomyopathy	Uncertain significance
Case6	MLH3	NM_001040108.1	14	75514138	rs28756990	Het	c.2221G>T	p.Val741Phe	SCV000026082.1	Endometrial carcinoma	Uncertain significance
Case6	DSC2	NM_004949.4	18	28672114	rs144799937	Het	c.304G>A	p.Glu102Lys	SCV000063116.3	Cardiomyopathy	Uncertain significance
Case6	GPR101	NM_054021.1	X	136112910	rs73637412	Het	c.924G>C	p.Glu308Asp	SCV000203835.3	Pituitary adenoma	Not likely
Case7	CX3CR1	NM_001171174.1	3	39307162	rs3732378	Het	c.839C>T	p.Thr280Met	SCV000028838.3	Age-related macular degeneration	Likely
Case7	JAK2	NM_004972.3	9	5073770	rs77375493	Het	c.1849G>T	p.Val617Phe	NA	Thrombocythemia	Likely
Case7	GPR101	NM_054021.1	X	136112910	rs73637412	Het	c.924G>C	p.Glu308Asp	SCV000203835.3	Pituitary adenoma	Not likely
Case8	TNNC1	NM_003280.2	3	52485426	rs267607124	Het	c.435C>A	p.Asp145Glu	SCV000209137.1	Cardiomyopathy	Likely
Case8	ANK2	NM_001127493.1	4	114294462	rs121912706	Het	c.5434C>T	p.Arg1812Trp	SCV000223217.2	Arrhythmia	Uncertain significance
Case10	APOE	NM_000041.2	19	45412079	rs7412	Het	c.526C>T	p.Arg176Cys	NA	Hyperlipoproteinemia	Likely
Case13	BCO1	NM_017429.2	16	81298282	rs119478057	Het	c.509C>T	p.Thr170Met	SCV000025214.2	Hypercarotenemia and vitamin A deficiency	Likely

All data is available upon request. het, heterozygous; NA, not available.

^aInvolvement of the variant on comorbid phenotypic features (not in POF syndrome) based on anamnestic data.

Discussion

The current study is the first retrospective genetic association study of spontaneous POF cases using population-based biobank samples. The main goal of this study was to investigate the role of CNVs in premature menopausal transition using phenotype and genotype data from the population-based biobank samples. We found that the overall CNV load might influence the age at which an individual transitions to menopause, probably due to an increased chance of disrupting key genes and pathways that are essential for normal ovarian function. The number of CNVs was negatively related with the age of menopause in POF cohort, meaning that in patients with premature menopause the transition occurred earlier if the genome contained more CNVs. In addition, large structural rearrangements on the X chromosome, and likely those on autosomes, may disrupt normal chromosomal pairing in meiosis, leading to germ cell death. Consequently, patients may develop POF due to the loss of germ cells (Schlessinger et al., 2002). Still, these associations were observed only in POF patients, but not in the controls, suggesting the disease-specific pattern. Therefore, more studies on healthy populations are warranted to address this question in greater detail.

In addition, it was observed that larger cumulative size of all CNVs per individual genome in POF patients was associated with later first pregnancies. The association was quite remarkable as each year of pregnancy delay was associated with 80 kb bigger size of cumulative CNV load. This analysis was controlled by confounders affecting the time of the first pregnancy. Among those, the menarche represented the potential starting time for pregnancies, the number of live births was considered to unify the choice for family size, while the age at study was considered to unify any differences in family planning between generations. As the result, these data suggest that the overall size of CNVs might be related to the fecundity of a woman, at least in the POF cohort.

Genetic profiling and functional enrichment analysis of genes within identified CNVRs of POF patients identified pathways related to immune system functions. A notable number of genes were associated with immune cells (e.g. monocytes, lymphocytes) that may influence the differentiation and maturation of germ and granulosa cells (Bukovsky et al., 1995, 2005; Koks et al., 2010). These findings are also consistent with other studies that implicate humoral and cellular autoimmune mechanisms in the etiology of POF (Hoek et al., 1997; Chernyshov et al., 2001). In addition, there was a significant alteration in the ‘scavenging of heme from plasma’ pathway (57.1% of genes, P = 2.41 × 10⁻³⁰). Free heme has tissue-damaging properties, and it represents a source of redox-active iron that generates reactive oxygen species (ROS) and oxidative stress (Yamada et al., 2011). The accumulation of damage induced by ROS is believed to be involved in more rapid ovarian aging due to the reduced function of oocytes and granulosa cells (Tatone et al., 2008).

In a subsequent case-by-case analysis, we identified three previously reported aberrations and 11 novel rare microdeletions and microduplications that may contribute to spontaneous POF. As a major result, our data reveal a number of novel candidate genes that are affected by chromosomal rearrangements and may be potentially involved in the genetic pathways leading to premature menopause (Supplementary data, Table SIV). Since spontaneous POF refers to a nonspecific syndrome with many putative risk-genes/regions, the common CNVs are not expected to be prevalent among POF patients. Indeed; we saw that detected CNVs are unique to individual POF cases rather than being common to POF patient population. Therefore, identified CNVs are rare and do not succeed in statistical comparisons and P-values (all being >0.05, data not shown), and thus do not represent any statistically significant differences in the prevalence between the cases and the controls. Still, detected CNVs can potentially contribute to the onset of premature menopause based on the biological function of the genes within CNV regions. In addition; we corroborated the importance of recently discovered POF genomic regions, including hemizygous microdeletions of SYCE1 (synaptonemal complex central element protein 1 at 10q26.3, MIM*611486) and CPEB1 (cytoplasmic polyadenylation element-binding protein 1 at 15q25.2, MIM*607342) (McGuire et al., 2011). SYCE1 and CPEB1 have a crucial role in meiosis, maintaining synaptonemal complexes formed between homologous chromosomes (Bolcun-Filas et al., 2009; Zheng et al., 2010). The involvement of these genes in reproductive failure is supported by knockout mice studies, showing that knockout female mice are infertile due to meiotic arrest in meiosis I (Tay and Richter, 2001; Bolcun-Filas et al., 2009). Importantly, the deleted region on chromosome 15 also encompassed BNC1 (basonuclin 1; MIM*601930) that may also be essential for oogenesis, as oocyte morphology is affected in knockout mice, which leads to female subfertility (Ma et al., 2006).

Among the genes found within the novel microdeletions, two have a strong implication in the reproductive biology: FMN2 (formin 2 at 1q43; MIM*606373) in Case1 and ARFGAP3 (ADP-ribosylation factor GTPase activating protein 3 at 22q13.2, MIM*612439) in Case13. FMN2 plays a crucial role in cytoskeletal organization and establishment of cell polarity during oogenesis, which is essential for meiotic maturation and maintenance of oocyte asymmetry (Montaville et al., 2014). Knockout studies in mice revealed that Fmn2-deficient female mice have decreased fertility due to abnormal metaphase spindle position and abnormal first polar body formation during meiosis I (Leader et al., 2002). Notably, FMN2 was the only gene disrupted by CNV, and importance of FMN2 in human reproduction was also discussed in a case–control study of women with unexplained infertility (Ryley et al., 2005). The second gene, ARFGAP3, is an androgen-targeted gene that is thought to be involved in intracellular trafficking of proteins and in vesicular transport (Liu et al., 2001). In bovines, ARFGAP3 contributes to follicular growth, ovulation and/or luteinization (Ndiaye et al., 2005), and it might have a similar role in human. It is worth noting that whole-exome sequencing did not reveal any rare homozygous mutations relevant to POF on the remaining alleles of these deleted regions. This supports the hypothesis that haploinsufficiency of meiotic genes may facilitate germ cell loss, resulting in premature depletion of the follicular pool. In addition, secondary findings were identified by WES, and some of the conditions associated with the detected variant were also diagnosed in our POF patients (Table V). POF has an adverse long-term effect on female wellbeing, especially bone health, cardiovascular health and neurological function (Shuster et al., 2010); however, it is hard to evaluate retrospectively whether the identified variants directly contribute to the disease or the disease developed as a consequence of premature menopause.

Although the phenotypic consequences of genomic gains are usually not as straightforward as of deletions, they may impair meiosis or proper folliculogenesis by altering the gene dosage and expression level (Newman et al., 2015). Two genomic gains were detected on X chromosome, and included STS (steroid sulfatase at Xp22.31, MIM*300747) and androgen receptor (AR at Xq12, MIM*313700) genes. STS encodes steroid sulfatase that catalyses metabolic precursors for estrogens, androgens and cholesterol (Alperin and Shapiro, 1997), while androgens promote follicular growth at early stages and inhibit at final stages (Sen et al., 2014), similar to polycystic ovary syndrome (PCOS) (Azziz et al., 2009). Therefore, it is possible that duplications in these two regions may lead to imbalanced estrogen and/or androgen levels, subsequently leading to excess of follicular demise. Autosomal microduplications include meiotic gene SGOL2 (shugoshin-like 2 at 2q33.1, MIM*612425) that protects centromeric cohesion from premature separase-mediated cleavage in oocytes during meiosis I (Llano et al., 2008). Genomic gain at 6q27 encompasses TBP (TATA box binding protein, MIM*600075), and mouse studies have revealed that overexpression of TBP significantly lowers the rate of mouse oocyte progression to MII stage during meiosis, possibly due to a toxic effect on the cell (Sun et al., 2013). As a result of duplication in Case4, the overexpression of TBP can occur, causing the failure of oocyte maturation. The 6q27 region also includes PSMB1 (proteasome subunit, beta-type, 1, MIM*602017), associated with Sjögren's syndrome (Martinez-Gamboa et al., 2013), a chronic autoinflammatory disease, suggesting autoimmune reactions as the cause of this patient’s POF at the age of 24 years. SCARB1 (scavenger receptor class b, member 1 at 12q43; MIM*601040), detected in Case 9, mediates cholesterol transfer to and from high-density lipoprotein (HDL). It is noteworthy that this individual has been smoking 10 cigarettes per day for 30 years since the age of 20 years and is overweight (BMI = 30 kg/m²). SCARB1 knockout mice have abnormal HDLs, ovulation of dysfunctional oocytes and infertility (Miettinen et al., 2001). Since HDL is the only lipoprotein present in follicular fluid (Shalgi et al., 1973; Perret et al., 1985), it is possible that changes in HDL level due to SCARB1 overproduction, combined with smoking, may disturb oocyte maturation or function, and thus contribute to infertility. However, all the gene-dosage hypotheses presented here remain to be tested by future studies.

It is important to note that some CNVs may have gone undetected due to the set threshold and many high-resolution arrays do not fully cover the genome (Cooper et al., 2008). In addition to technological boundaries, our study was limited by the fact that sample selection from the EGCUT biobank was performed only based on the available anamnestic data, some of which were self-reported as opposed to data derived directly from medical records. For example, the age at menopause was self-reported on the average 20.7 years later (the duration of amenorrhea by the time of study), which may lead to reporting errors. Also, the average age at the time of the study was 57.8 ± 12.5 years; therefore, we lack detailed information on clinical, biological (hormone levels) or ultrasonographic data, retrospectively from the time of menopause. Moreover, it is not always possible to exclude every disease that could cause POF secondarily (as opposed to primary ovarian causes). The shortcomings of this study are diminished by the large cohort size for such a rare disease and the close examination of a wide range of phenotypic data in every patient.

In summary, our data provide novel insight into the association between chromosomal aberrations and premature menopause of ovarian cause. Our results highlighted the possible role of CNVs in the pathogenesis of POF. Both duplications and deletions detected in our study were associated with the POF phenotype and contained genes that are essential for reproductive function. Although functional studies are required to further delineate the contribution of identified CNVs in the genetic etiology of POF, our study confirms that DNA microarrays are a useful tool for evaluating genomic imbalances in POF patients as it offers a much higher resolution and therefore, a higher diagnostic yield compared with conventional cytogenetic methods. Finally, we conclude that using samples from population-based biobanks may be a useful and efficient way to recruit patients for large-scale CNV studies, and data collected from such studies represent a starting point for both characterization of the disease and guidance for further research.

Supplementary data

Supplementary data are available at http://humrep.oxfordjournals.org/.

Authors’ roles

A.S. and A.K.: study design; K.H.-K.: clinical characterization of samples and statistical analysis; A.S., A.K., O.T., O.Ž. and M.N.: methodology; M.N., R.M., M.K. and PA.K.: computational analysis; O.T., M.N. and O.Ž.: data interpretation; K.H.: laboratory assistance; O.T.: preparation of draft and original manuscript; K.H.-K., A.S., A.K., O.T., O.Ž. and M.N.: revision of the manuscript; A.S., A.K. and J.S.T.: Funding; A.S. and A.K.: supervision; all authors approved the original manuscript.

Funding

This study was supported by the Estonian Ministry of Education and Research (IUT20-43, IUT20-60, IUT34-16, SF0180027s10 and 9205), Enterprise Estonia (EU30020 and EU48695), Eureka's EUROSTARS Programme (NOTED, EU41564), grants from the European Union's FP7 Marie Curie Industry-Academia Partnerships and Pathways (IAPP, SARM, EU324509) and Horizon 2020 innovation programme (WIDENLIFE, 692065), Academy of Finland and the Sigrid Juselius Foundation. Funding to pay the Open Access publication charges for this article was provided by Horizon 2020 innovation programme WIDENLIFE grant #692065.

Conflict of interest

None declared.