Genome-Wide Copy Number Variation Association Analyses for Age at Menarche

Abstract

Context:

Menarche is a significant physiological event for women. Age at menarche (AAM) is a heritable trait associated with many common female diseases. The genetic basis and the mechanism for AAM are largely unknown. Copy number variation (CNV) is a common type of genetic variation underlying human complex traits. The importance of CNV to AAM variation is unclear.

Objective:

The objective of the study was to identify CNV important to AAM variation.

Design:

We performed the first genome-wide CNV study of AAM in 1654 Caucasian females using Affymetrix human single-nucleotide polymorphism 6.0 array. We also replicated our findings in another Chinese cohort containing 752 women.

Results:

We identified a CNV, variation_38399, in the 2q14.2 region, for association with AAM (P = 1.03 × 10⁻³). The CNV has two variants (one copy and two copy), with a mean AAM of 14.00 yr and 12.90 yr, respectively. Interestingly, in a Chinese sample containing 752 women, this CNV has been replicated both with a marginally significant P = 0.090 and with a same direction of effect (a lower copy number for a later AAM). The CNV is located approximately 75 kb upstream of the diazepam binding inhibitor (DBI), a gene known to regulate estrogen levels, a key factor for menarche.

Conclusion:

Our findings for the first time identified a novel CNV and suggested the DBI-mediated endocrinological pathway as a potential mechanism for AAM regulation.

Menarche is an important milestone in a female's physiological development. Age at menarche (AAM) has a significant impact on a woman's health later in life. For example, an early AAM is associated with breast and endometrial cancers (1, 2) and a late AAM increases the risk of Alzheimer's disease (3) and osteoporosis (4, 5). Overall, an early AAM appears to be more harmful to women's health than a late AAM. In a cohort containing more than 61,000 Norwegian women, it was found that with each year's decrease in AAM, there was an average increase of 2.4% in mortality rate (6). The significant health implications of AAM make it an interesting and important trait to study. Understanding the determining factors of AAM may shed light on the etiology of AAM-associated diseases and women's health in general.

Genetic factors play a dominant role in determination of AAM. From 50 to 70% variation of AAM can be explained by genetic factors (7–9). However, specific genes underlying AAM are still largely unknown. A few candidate genes were suggested to influence AAM, e.g. the estrogen receptor-α (ER-α) and -β (ER-β) genes (10–12), the SHBG gene (13), the androgen receptor gene (14), the IGF-I (IGF-1) gene (15), the chemokine (C-C-motif) receptor 3 (CCR3) gene (16), and the genes of the cytochrome P450 family (17–19). To date, only three genome-wide linkage studies on AAM were published (9, 20, 21), including one by our own group (21). The studies identified several genomic regions (e.g. 22q11, 22q13, 16q12, 16q21, and 12q) that may harbor quantitative trait loci underlying AAM.

A promising strategy to facilitate identification of AAM genes is a genome-wide association study (GWAS). So far, six single-nucleotide polymorphism (SNP)-based GWAS on AAM were reported (22–27), including one from our group (27). Several of the GWAS supported the LIN28B gene on the 6q21 region for AAM (22, 24–26). Our GWAS suggested that another gene, the SPOCK gene, may also be an important genetic factor underlying AAM variation (27). A more recent large-scale meta-analysis (23) identified 30 new menarche loci, including four loci previously associated with body mass index (FTO, SEC16B, TRA2B, and TMEM18), three implicated in energy homeostasis (BSX, CRTC1, and MCHR2), and three implicated in hormonal regulation (INHBA, PCSK2, and RXRG).

Copy number variation (CNV) is a type of DNA variation, in which a DNA segment ranging from one kilobase to several megabases varies in number of copies. A CNV may change gene dosage, disrupt coding sequences, or have long-range positional effects on gene expression outside the CNV region, consequently leading to phenotype variation (28, 29). CNV have been implicated in genetic basis of many human complex diseases, e.g. susceptibility to autoimmune diseases (30), glomerulonephritis (31), and Crohn disease (32). Our group also identified that CNV involving two genes, UGT2B17 and VPS13B, were associated with osteoporosis risk (33, 34). However, to date, no CNV-based study was reported for AAM.

Here we report the first CNV-based GWAS for AAM. Using a sample containing 1654 Caucasian females, we performed a GWAS using Affymetrix genome-wide human SNP array 6.0 (Affymetrix, Santa Clara, CA). We identified a CNV, variation_38399, which was associated with AAM in our sample, achieving a P = 1.03 × 10⁻³. Interestingly, in another Chinese cohort containing 752 women, the CNV's association with AAM was confirmed, both with a marginally significant P value (P = 0.090) and with the same direction of effect.

Materials and Methods

Subjects

The study was approved by institutional review board or research administration of the involved institutions. Signed informed consent documents were obtained from all study participants before entering the study. In total, our sample contains 1654 Caucasian female subjects, with the mean age of 51.6 yr.

The inclusion criteria for the above cohort included the following: 1) Caucasians of European origin; 2) healthy female subjects with regular menses or, if postmenopausal, with a history of regular menses throughout the years before menopause; and 3) those without diseases and conditions that may potentially affect regular menstrual cycles, as listed in the exclusion criteria.

The detailed exclusion criteria were published elsewhere (35, 36). Briefly, subjects with chronic diseases and conditions involving vital organs (heart, lung, liver, kidney, and brain) and severe endocrinological, metabolic, and nutritional diseases that might affect regular menstrual cycles were excluded from this study.

AAM data of all the female subjects in our study cohort were collected based on a same standard nurse-administered questionnaire, which included a detailed medical and female history. All the study subjects reported their AAM to the accuracy of 1 yr. Our Caucasian study subjects have a mean AAM of 12.9 yr with a sd of 1.6.

For replication purposes, we took advantage of the available genome-wide CNV data and analyzed association of the CNV, variation_38399, with AAM in a Chinese cohort containing 752 female subjects, with a mean age of 37.4 yr. These subjects were recruited from Chinese Han adults living in the City of Changsha and its vicinity, Hunan province, and from Chinese Han adults living in the City of Xi'an and its surrounding areas, Shaanxi province. Except for the subjects' ethnicity, the in/exclusion criteria and AAM ascertainment method for the Chinese subjects are the same for our Caucasian subjects. These subjects have a mean AAM of 13.9 yr with a sd of 1.6.

Genotyping

Genotyping for the Caucasian and Chinese subjects followed the same procedures as detailed below.

Genomic DNA was extracted from peripheral blood leukocytes using a commercial kit following the standard protocols. We used the Affymetrix genome-wide human SNP Array 6.0 (Affymetrix), which features 1.8 million genetic markers, including more than 906,600 SNP and more than 946,000 probes for the detection of CNV to genotype the study subjects according to the standard protocol recommended by the manufacturer. Fluorescence intensities were quantified using an Affymetrix array scanner 30007G. Data management and analyses were performed using the Affymetrix GeneChip Command Console Software. For sample quality control (QC), a contrast QC threshold was set at a default value of greater than 0.4. The final average contrast QC across the entire sample reached a high level of 2.76 for our Caucasian cohort and 2.62 for our Chinese cohort. Genotype calling, genotyping QC, and CNV identification were conducted by using the Birdsuite package (http://www.broadinstitute.org/science/programs/medical-and-population-genetics/birdsuite/birdsuite-0).

We first measured the copy number estimates for each chromosome and for the genome-wide average (sum of all chromosomes), reported by the Birdseye Hidden Markov Model (37). Then we measured the variability of CNV probe intensities according to each chromosome and genome-wide average (sum of all chromosomes). We removed the subjects with excessive variability in probe intensity (>3 sd) according to either genome-wide average or the estimate on more than two chromosomes. We removed 71 such subjects from the Caucasian cohort and 98 such subjects from the Chinese cohort. We kept the subjects who had only one or two chromosomes failing in copy number estimate QC and probe intensity QC and treated the CNV in the failed chromosomes of those subjects as missing data in further association analysis.

We excluded those CNV with greater than 5% of uncertain or missing copy call or with a minor variant/s frequency of less than 1%, which refers to the total proportion of the subjects with a copy number of less or more than two in the entire samples. Using the above QC criteria, a total of 405 CNV remained for the Caucasian sample and a total of 207 CNV remained for the Chinese sample for the downstream association analyses.

It should be noted that although SNP 6.0 arrays are capable of identifying both SNP and CNV, only the CNV data will be reported here (SNP data will be reported elsewhere).

Real-time PCR confirmation of CNV

We performed real-time PCR experiments using an ABI StepOnePlus system (Applied Biosystems, Foster City, CA) to confirm the CNV, variation_38399, in our subjects, focusing on those subjects with an abnormal copy number, i.e. one-copy and three-copy subjects, and selecting only a few two-copy subjects as positive controls. To assay the target genomic region containing the CNV, we used a TaqMan assay, Hs05842891_cn, which spans a 105-bp region starting from chromosome 2:120046669 (build 37; National Center for Biotechnology Information, Bethesda, MD). We also used the human ribonuclease P TaqMan Copy Number Reference Assay as an endogenous control that is present in two copies in human genome and hence served as a base for relative quantitation of the copy number target. We followed the TaqMan Copy Number Assay protocol (http://tools.invitrogen.com/content/sfs/manuals/cms_062368.pdf) to run the PCR. The reaction tube contains 10 μl 2 × TaqMan Genotyping Master Mix, mixed with 1 μl TaqMan assay (Hs05842891_cn), 1 μl ribonuclease P TaqMan Copy Number Reference Assay, and 4 μl nuclease-free water. The reaction was run for 10 min at 95 C, followed by 40 cycles of PCR (15 sec at 95 C followed by 60 sec at 60 C). The copy number calling was based on the difference between the cycle threshold value of the FAM dye signal for the target copy number assay and that value of the VIC dye signal for the reference assay. The raw data were exported and analyzed using CopyCaller Software version 1.0 for copy number calling of the variation_38399.

Statistical analysis

A stepwise regression model was used to filter significant covariates for each study cohort. To detect and control the effect of population stratification, we used EIGENSTRAT (http://genepath.med.harvard.edu/∼reich/EIGENSTRAT.htm) to perform a principal component analysis to correct for stratification in GWAS (38). Approximately 700,000 SNP were selected to calculate the principal components, and 10 default main eigenvectors were used as covariates to adjust raw AAM values. We performed association analyses between CNV and AAM using PLINK software package (version 1.07) (http://pngu.mgh.harvard.edu/∼purcell/plink/).

Results

In our Caucasian sample, we identified a CNV, variation_38399 (following the nomenclature by the Database of Genomic Variants, http://projects.tcag.ca/variation/?source=hg19), which achieved a P = 5.69 × 10⁻⁴ for association with AAM, the second most significant P value among all the ones that survived the QC. Frequency distribution for CNV in our Caucasian sample is 0.012 (n = 20) for the one-copy variants, 0.985 (n = 1630) for the two-copy variants, 0.002 (n = 4) for the three-copy variants. The raw AAM values for this CNV are 14.00 (sd 1.89) for the one-copy variants, 12.90 (sd 1.57) for the two-copy variants, and 12.25 (sd 0.96) for the three-copy variants (Table 1 and Fig. 1).

Table 1.

AAM values of study subjects classified by the copy number of variation_38399

	Variation_38399	One-copy subjects	Two-copy subjects
Caucasians	n	20	1630
	AAM mean (sd)	14.00 (1.89)	12.90 (1.57)
Chinese	n	18	732
	AAM mean (sd)	14.44 (1.69)	13.91 (1.63)

AAM values are raw values without covariate adjustment.

Fig. 1.

Raw AAM values of Caucasian and Chinese subjects with different copy number of variation_38399. Note: In the discovery stage of our study on Caucasian women, we identified three CNV that reach a significance level of less than 1E-3. These CNV are variation_10365 (position: chr6:165,727,817.165,731,928), variation_38399 (position: chr2:120,045,534.120,049,842), and variation_38475 (position: chr3:162,137,600.162,142,705). The P values achieved by these three CNV in our discovery Caucasian cohort are comparable, which are 4.07E-4, 5.69E-4, and 6.00E-4, respectively. Importantly, in our Chinese replication cohort, variation_10365 achieved a nonsignificant P value (P = 0.51) and variation _38399 achieved a marginally significant P value of 0.090. Another CNV, variation_38475, is not very polymorphic in our Chinese cohort, which has only one subject who is a single-copy variant (frequency = 0.001). In our study, we excluded from analysis those poorly polymorphic CNV with a minor variant/s frequency of less than1%. Therefore, we did not further analyze variation_38475 in our Chinese sample. Taken together, variation_38399 stands out among the three most significant CNV identified in the discovery stage due to the strong replication evidence in Chinese. Therefore, we report only variation_38399 (not variation_10365, even though it achieved the most significant P value in the discovery stage) because a CNV with replication evidence is more likely to be an authentic finding in a genome-wide screen, and this guideline is now well recognized in the GWAS field.

In our subsequent real-time PCR analysis on the CNV, variation_38399, we achieved success in confirming all the one-copy variants. Examples of the one-copy variants confirmed by PCR are shown in Fig. 2. As shown in the figure, albeit some variation of copy number signals in the PCR analysis, copy number called by PCR is consistent with the copy number called by the Affymetrix array analysis for those one-copy variants as well as the two-copy normal subjects.

Fig. 2.

Real-time PCR confirmation of copy number for variation_38399. Note: Shown are examples for some samples subject to real-time PCR analysis for copy number for the variation_38399. Specifically, as shown in the x-axis, samples KC02223, KC02632, KC02750, KC02614, KC02655, KC02546, and KC02656 were called as one-copy variants (for the variation_38399) based on our Affymetrix array analysis, and samples KC02185 and KC03636 were called as two-copy normal subjects in the array analysis. The copy number for the CNV for these samples based on real-time PCR analysis is shown in the y-axis. The P value for comparison of copy number estimation by PCR between the two CNV groups called by Affymetrix array (i.e. the group of white bars vs. the group of black bars) is 5.29E-7.

However, based on the PCR analysis, we were unable to validate the three-copy variants. Therefore, we excluded those three-copy variant subjects (n = 4) in our Caucasian subjects and recalculated the P value for AAM association for variation_38399. Due to the small number of the three-copy variant subjects, the statistical significance was not affected substantially by excluding the three-copy variant subjects and we still achieved a P = 1.03 × 10⁻³ in the reanalysis including only the one- and two-copy variants.

The CNV, variation_38399, is located on chromosome 2 at the 2q14.2 region from 120,045,534 to 120,049,842 bp, according to the UCSC Genome Browser on Human February 2009 (GRCh37/hg19) Assembly, with a total length of approximately 4.3kb. There are two well-annotated genes located in the neighborhood of this CNV, with the STEAP3 (six-transmembrane epithelial antigen of prostate) located approximately 22 kb on the centromere side, and diazepam binding inhibitor (DBI) located approximately 75 kb on the telomere side. In addition, another poorly defined gene, C2orf76, is located approximately 10 kb on the same side with the DBI gene.

Using our Chinese sample, we performed replication analysis for the variation_38399 for its association with AAM. Frequency distribution for this CNV in our Chinese sample is 0.024 (n = 18) for the one-copy variants, 0.973 (n = 732) for the two-copy variants. The raw AAM values for this CNV in our Chinese sample are 14.44 (sd 1.69) for the one-copy variants, and 13.91 (sd 1.63) for the two-copy variants (Table 1 and Fig. 1). The P value for association with AAM is 0.090.

Figure 3 shows the distribution of AAM phenotypes in both our Caucasian and Chinese subjects, stratified by the copy number of variation_38399. As shown the figure, compared with the normal two-copy variant subjects, there is a skewed distribution to the late AAM (AAM ≥14 yr) for the one-copy variant subjects. This trend is evident in both Caucasians and Chinese but more pronounced in the former cohort.

Fig. 3.

AAM value distribution of subjects of different copy numbers of variation_38399.

Due to the reported CNV's moderate size (<10 kb), it is possible that some subjects having the deletion (one copy variant) may not be detected, as suggested by the simulation in the study by Korn et al. (37). However, to the best of our knowledge, our data set, containing both Caucasians and Chinese, is so far the largest one to report this CNV (variation_38399), and hence may be the most robust result achieved for this CNV's detection.

Discussion

This is the first CNV-based GWAS for AAM. Through this study, we identified a CNV, variation_38399, for association with AAM in our Caucasian sample. The CNV's association with AAM was also confirmed in an independent Chinese cohort. If taking into account the association signals in both our Caucasian discovery cohort and our Chinese replication cohort, among all the CNV tested genome wide in our Caucasian subjects, the CNV, variation_38399, has the strongest evidence for importance to AAM variation.

According to Affymetrix microarray and the subsequent real-time PCR analysis, the CNV variation_38399 has two variants (one copy and two copy variants), with the one-copy variants having a very low frequency (∼1–2%). They represent a typical example for rare CNV variants, which have been recently shown to play an important role in several human complex diseases/traits (39–41).

As a potential mechanism underlying this CNV's association with AAM, the CNV might cause AAM variation through its regulatory effects, such as changing the gene expression, on a nearby gene/s. One of such regulated genes might be the DBI gene. For this CNV, the gene is the nearest well-annotated gene on the telomere side, with the CNV located approximately 75 kb of its upstream. The major function of DBI is to down-regulate the signal transduction at type A γ-aminobutyric acid (GABA) receptors located in brain synapses (42). Importantly, it was found that DBI can regulate estrogen levels in vivo through the GABA signaling at the brain (43). Given that fluctuated estrogen levels are one of the key factors involved in initiation of the first menses (the menarche), it might be reasonable to hypothesize that the CNV may contribute to AAM variation through modulating the DBI gene.

According to UCSC Genome Browser, there are several conserved binding sites located within the genomic location of the CNV, variation_38399, for transcription factors including paired box transcription factor 4 (binding location: chr2:120048088-120048098), E47 (transcription factor 3) and TAL-1 (T-cell acute lymphocytic leukemia 1; binding location: chr2:120048629-120048644), homeobox A3 (binding location: chr2:120049036-120049044), Forkhead box D1 (binding location: chr2:120049639-120049654), and Forkhead box J2 (binding location: chr2:120049638-120049655). These transcription factor binding sites may represent potential regulatory elements in the CNV region and may contribute to regulation of the nearby DBI gene.

Imperfect data collection exists more or less in studies of almost all human diseases/traits. In our study, the AAM data were collected through retrospective self-reporting. Statistically, there is no systematic bias in reporting AAM in studies of this trait. Therefore, inaccuracy of the AAM recall data may be factored into random noise in the data collected and may have a potential effect only to decrease the power for detecting the AAM genes. However, the inaccuracy, if not systematically biased, should not render false-positive findings. The overall precision of our AAM data are partially supported by a high heritability of approximately 0.60 for AAM as detected in our previous whole-genome linkage study of AAM (21), in which we used the same approach in AAM data collection as in the current study. Importantly, such a high heritability could not be achieved if significant errors and thus noise existed in the AAM data of our study subjects.

Unlike for other traits, recollection by subjects is a generally reliable measure for AAM data collection. This is because AAM is a most significant event in female puberty, which often has a major impact on a woman's life, both physically and psychologically. A study found a high correlation of approximately 0.80 between the original AAM and the AAM recalled even 30 yr later (44). Consistent with the finding, several other studies also indicated reliability of the retrospective method in AAM data acquisition (45–47). Therefore, our study followed this common practice in the field, which is feasible, convenient, and accurate to perform.

Although there were several SNPs on chromosome 2 reported in the large-scale meta-analysis paper by Elks et al. (23), none of the SNPs are located on the same chromosomal region, i.e. 2q14.2, as the CNV reported here. The reported CNV is also not located in the previously identified linkage regions (9, 20, 21). This result is not unexpected. CNV, as a genetic marker, are very different from SNP and microsatellite markers used in GWAS and genome-wide linkage scans. The genomic span of CNV is far larger than SNP and microsatellites, and a CNV is copy number change rather than the form change (polymorphism), as in the case of SNP. Therefore, CNV may contribute to human phenotypic variation from different, largely independent mechanisms than SNP and microsatellites. It is therefore reasonable to observe different findings from a CNV than GWAS or linkage studies. Importantly, previous GWAS on SNP or linkage studies on microsatellites did not take CNV into account, and the current CNV analysis also does not consider SNP or microsatellites. For example, a deletion variant of a CNV may also have deletion of thousands of SNP or hundreds of microsatellites, which reside on the CNV region. Such information was not accounted for in SNP association or microsatellite linkage analyses (e.g. a SNP on a deleted region is often called as missing data in 0 copy variants and homozygous in one copy variants), nor does current CNV analysis consider the SNP or microsatellites involved in a CNV region. The above reasons may contribute to the nonoverlap for the signals from a CNV study vs. the signals from a GWAS or linkage study.

In summary, we performed the first study to search for CNV at the genome-wide scale for AAM and identified a CNV, variation_38399, as an important genetic structural variant underlying AAM variation. Given the proximity of DBI with this CNV and the function of DBI in central regulation of estrogen levels, our findings support a mechanism that the CNV may modulate AAM through influencing DBI and its related GABA signaling pathway.