Copy Number Variation on Chromosome 10q26.3 for Obesity Identified by a Genome-Wide Study

Abstract

Background:

Obesity is a highly heritable disease defined by high body mass index (BMI). However, a large proportion of the heritability of obesity remains unexplained. Copy number variations (CNVs) might contribute to the missing heritability of obesity.

Methods:

We conducted genome-wide CNV analyses on obesity phenotypes, including BMI and body fat mass in a discovery sample of 2215 unrelated white subjects. After quality control, 314 CNVs were used for association tests. For significant CNVs identified, follow-up replication analyses were performed in three independent samples, including an unrelated sample of 1000 white subjects (OM sample), a family-based sample of 8385 white subjects (FHS sample), and an African-American sample of 1479 obesity cases and 1575 lean controls (AA sample).

Results:

Genome-wide CNV analyses detected that a CNV located at 10q26.3, which, even after multiple testing corrections, showed a strong association with both BMI (P = 2.30 × 10⁻⁴, β = 2.164) and body fat mass (P = 6.76 × 10⁻⁵, β = 4.126). This CNV was successfully replicated in the three replication samples (OM sample: P = 0.0465 for BMI, 0.0435 for fat mass; FHS sample: P = 0.0038 for BMI; AA sample: P = 0.0023 for obesity). Quantitative PCR validated this CNV, which covers a gene, CYP2E1. The protein encoded by CYP2E1 involves the synthesis of cholesterol, steroids and other lipids, which may have a potential impact on obesity.

Conclusion:

Our findings suggest the significant contribution of CNV10q26.3 to the pathogenesis of obesity.

Obesity is a major public health problem that is associated with an increased risk of many common diseases, such as hypertension, type 2 diabetes, and coronary heart disease (1). Body mass index (BMI) is the most widely used quantitative measure of obesity, and adults with a BMI 30 kg/m² or greater are termed obese.

Obesity is a heritable complex disease, with heritability estimates of BMI ranging from 40 to 70% (2). Genome-wide association studies (GWAS) of BMI based on single-nucleotide polymorphisms (SNPs) have identified a number of genes (3), which collectively explain only less than 10% of the genetic variation in obesity. In addition to SNPs, copy number variation (CNV), as another common type of genomic variability, has been suggested as a possible source for explaining the missing heritability (4). For example, several CNVs have been reported for obesity (5–10), including 16p11.2, 10q11.22, 11q11, 16p12.3, NEGR1, et al.

The goal of this study was to perform a genome-wide CNV association analysis to identify common CNVs for obesity. The discovery sample included 2215 U.S. white subjects. Follow-up replication analyses were performed to validate our major findings in three independent samples of white and African-American populations involving 12,439 subjects.

Materials and Methods

Subjects

The study was approved by the required institutional review board or research administration of all participating institutions. Signed informed-consent documents were obtained from all participants before entering the study. The study was initially conducted with a discovery stage for detection of CNVs in a GWAS sample of 2215 unrelated white subjects with BMI and body fat mass phenotypes (KC sample). A two-step replication design was subsequently used. CNVs with a P < 0.05 for association with obesity phenotypes in the discovery stage were selected for the first-step replication in another sample of 1000 unrelated white subjects with BMI and body fat mass phenotypes (OM sample). In the second-step replication stage, CNVs with a P < 0.05 in the OM sample were further validated into two additional samples [family-based sample (FHS) and African-American (AA) samples]. The FHS sample included 8385 white subjects from 1180 families with BMI. Among these subjects, 2791 subjects had body fat mass measurements. The AA sample included 1479 African American obese children (BMI with the 95th or greater percentile), and 1575 AA lean controls (BMI 50th or less percentile). The basic characteristics of all study samples are summarized in Supplemental Table 1, published on The Endocrine Society's Journals Online web site at http://jcem.endojournals.org, with detailed descriptions in Supplemental Methods.

Genotyping

The discovery sample was genotyped using the Affymetrix genome-wide human SNP Array 6.0 (Santa Clara, CA) (Supplemental Methods). Genotyping methods for the replication samples were also included in Supplemental Methods.

CNV detection

CNV detection and quality control in discovery sample

For the discovery sample, common CNVs were identified using the CANARY algorithm implemented in the Birdsuite (11). PennCNV algorithm (12) was also used to validate the results from CANARY (Supplemental Methods).

To generate results with a high confidence from CANARY, we conducted quality control at both the sample and CNV levels (Supplemental Methods). After the quality control, 2215 subjects were successfully genotyped by the CANARY and a total of 314 CNVs were consistently called by both PennCNV and CANARY and therefore were available for subsequent analyses.

CNV detection in replication samples

Because CANARY cannot handle data from chips other than the Affymetrix SNP 6.0, we used the PennCNV (12) to call CNVs in replication samples.

Statistical analyses

Association analyses for all study samples are described in Supplemental Methods in detail.

Quantitative PCR for CNV validation

We conducted quantitative PCR (qPCR) in the discovery sample to validate the significant CNVs identified and fine map the specific genes involved using TaqMan copy number assays (Supplemental Methods).

Results

Discovery stage

In the discovery stage, genome-wide association results for all the 314 tested CNVs with BMI and body fat mass are presented in Supplemental Fig. 1. We detected 16 CNVs that showed suggestive evidence for association with BMI or body fat mass at P < 0.05 (Table 1). The top two significant CNVs, CNV1670 and CNV1669, achieved the P values of 2.30 × 10⁻⁴ and 2.35 × 10⁻⁴ for BMI and 6.76 × 10⁻⁵ and 9.37 × 10⁻⁵ for body fat mass, respectively. The minor allele frequencies for these two CNVs were 0.040. After adjustment for multiple testing, these two CNVs remained significant (Table 1).

Table 1.

Major association results in the GWAS discovery sample (KC sample)

CNV ID	Frequency	Chr	Start (bp)a	End (bp)	Probe no.b	BMI			Body fat mass
						P valuec	Pc valued	βe	P valuec	Pc valued	βe
CNV1670	0.040	10q26.3	135178653	135227268	45	2.30 × 10−4	0.037	2.164	6.76 × 10−5	0.015	4.126
CNV1669	0.040	10q26.3	135092863	135146259	37	2.35 × 10−4	0.037	2.119	9.37 × 10−5	0.015	4.115
CNV124	0.228	1q25.1	173063179	173068463	16	1.45 × 10−3	0.152	−0.818	2.97 × 10−3	0.233	−1.371
CNV703	0.581	4q32.1	156022455	156022528	3	2.35 × 10−3	0.184	−0.548	4.70 × 10−3	0.246	−0.912
CNV10204	0.050	1q32.3	211071635	211077059	7	2.79 × 10−3	0.175	1.650	2.94 × 10−3	0.308	2.947
CNV10815	0.021	4q31.3	153209736	153212191	5	6.76 × 10−3	0.354	−2.281	0.0323	1.000	−3.238
CNV337	0.333	2q33.3	208063423	208067581	16	0.0116	0.521	−0.555	9.08 × 10−3	0.407	−1.030
CNV2150	0.271	16p13.3	19853151	19874863	30	0.0150	0.590	0.593	4.07 × 10−3	0.255	1.257
CNV128	0.133	1q25.2	177597151	177599938	4	0.0276	0.963	0.738	0.0527	1.000	1.165
CNV10279	0.019	2p21	42300861	42306117	6	0.0297	0.932	1.941	0.0189	0.741	3.762
CNV2200	0.026	16q23.1	74115584	74133500	18	0.0372	1.000	−1.262	0.0647	1.000	−2.009
CNV10798	0.021	4q28.3	138401883	138411420	51	0.0381	1.000	1.773	0.0862	1.000	2.635
CNV877	0.333	5q33.2	155409350	155415307	14	0.0430	1.000	−0.448	0.0927	1.000	−0.667
CNV1834	0.100	12p11.21	31264657	31293961	37	0.0555	1.000	−0.748	0.0444	1.000	−1.407
CNV977	0.467	6q14.1	79025784	79091904	109	0.0653	1.000	0.360	0.0446	1.000	0.704
CNV12144	0.052	14q11.2	22164681	22169049	7	0.0797	1.000	−0.919	0.0495	1.000	−1.850

Chr, Chromosome; Freq, frequency for subjects with abnormal copy numbers.

^aChromosomal positions are based on genome build hg 18.

^bThe probe number indicates the total number of SNPs' and CNVs' probes contained in the region.

^cThe P values are the raw P values from the association tests.

^dThe Pc values represent the corrected P values by false discovery rate adjustment. Significant Pc values are shown in bold.

^eFor β, the effect sizes were estimated in kilograms per square meter for BMI and kilograms for body fat mass.

Assessment of genome-wide findings

Replication in OM sample

Replication analyses were first performed for the top 16 CNVs (P < 0.05 in the discovery stage) in the OM sample. Five of the 16 CNVs were called in the OM sample (genotyped by Affymetrix Human Mapping 500K array), and the association results are presented in Table 2. CNV1670 and CNV1669 showed evidence for a directionally consistent association with BMI and body fat mass (P < 0.05).

Table 2.

Summary of association results in the three replication samples

First-step replication: OM sample
CNV	Chr	Freq	P BMI	βa	P body fat mass	βa
CNV1670	10q26.3	0.036	0.0465	1.620	0.0435	2.728
CNV1669	10q26.3	0.036	0.0465	1.620	0.0435	2.728
CNV1834	12p11.21	0.104	0.0510	−0.892	0.0640	−1.378
CNV12144	14q11.2	0.049	0.3220	−0.326	0.4840	0.048
CNV2200	16q23.1	0.012	0.1360	−1.550	0.2440	−1.785

Second-step replication for CNV10q26.3 (CNV1670 and CNV1669)
FHS sample			AA sample
Freq	P BMI	P body fat mass	Freq cases	Freq controls	P obesity	OR (95% CI)
0.046	0.0038	0.593	0.061	0.037	0.0023	1.69 (1.21–2.37)

All P values listed in Table 2 are two-sided raw P values from the association tests. Chr, Chromosome; Freq, frequency for subjects with abnormal copy numbers; OR, odds ratio; CI, confidence interval.

^aFor β, the effect sizes were estimated in kilograms per square meter for BMI and kilograms for body fat mass.

CNV validation and fine mapping by qPCR

For technical validation and further analyses, we reassessed CNV1670 and CNV1669 by qPCR in the discovery sample. The Affymetrix array analyses showed that the numbers of carriers with one, two, three, and four copies were two, 2128, 82, and three for CNV1670, and two, 2130, 80, and three for CNV1669, respectively. The genotypes of these two CNVs were nearly consistent, except two subjects with two copies in CNV1669, compared with three copies in CNV1670. The results from the qPCR validated these two CNVs and showed that the copy numbers of these two CNVs were completely consistent in all subjects, which were two, 2128, 82, and three for carriers with one, two, three, and four copies. Because qPCR is the gold standard and commonly used for CNVs validation, it indicated that the different results in Affymetrix array data were due to genotyping error within the tolerable range.

CNV1670 and CNV1669 are in high linkage disequilibrium (r² = 1). According to the CNV map identified by McCarroll et al. (13), both of CNV1670 and CNV1669 are located at chromosome 10, and there is only a 32,394-bp gap between them, suggesting that these two CNVs might belong to one CNV. To address this, we selected seven copy number assays to validate (Supplemental Fig. 2). After qPCR, we found that CNV1670 and CNV1669 belonged to one CNV, which located at 10q26.3 (physical position: 135,092,863–135,227,268 bp). The genotypes of CNV10q26.3 were exactly the same as those of CNV1670 in the discovery sample.

CNV10q26.3 and obesity

In the discovery sample, we detected 2128 subjects (96.07%) with two copies, two subjects (0.09%) with one copy, 82 subjects (3.70%) with three copies, and three subjects (0.14%) with four copies. The association P values of CNV10q26.3 were the same as CNV1670 (P = 2.30 × 10⁻⁴ for BMI and 6.76 × 10⁻⁵ for body fat mass). Higher copy numbers were associated with higher BMI (copy numbers one vs. two vs. three vs. four: 22.26 vs. 27.06 vs. 28.45 vs. 35.09 kg/m²) and body fat mass (19.54 vs. 24.02 vs. 27.21 vs. 42.11 kg), with the effect size (β-coefficient) estimated to be 2.164 (BMI in kilograms per square meter) and 4.126 (body fat mass in kilograms) for each copy number, respectively (Supplemental Fig. 3). In the OM sample, we found 96.4% with two copies and 3.6% with three copies. The effect was in the same direction as in the discovery sample, with the β-coefficient estimated to be 1.620 (BMI in kilograms per square meter) and 2.728 (body fat mass in kilograms) for each copy number, respectively (Supplemental Fig. 3).

Replication of CNV10q26.3 in FHS and AA samples

We further performed replication analyses for CNV10q26.3 with obesity in another two independent samples (FHS and AA samples). In the FHS sample, we detected three subjects (0.04%) with a homozygous deletion, 52 subjects (0.62%) with a heterozygous deletion, 8002 subjects (95.43%) with two copies, 323 subjects (3.85%) with three copies, and five subjects (0.06%) with four copies. A significant association was successfully replicated between CNV10q26.3 and BMI (P = 0.0038). However, no significant association signal was found for body fat mass (Table 2). One possible interpretation of this different effect could be due to the genetic heterogeneity between the FHS sample and the KC or OM samples.

In the AA sample, CNV10q26.3 again showed evidence for a directionally consistent association with obesity (P = 0.0023). We observed 90 subjects (6.09%) with duplication in the obese group vs. 58 subjects (3.68%) in the lean group. The odds ratio was 1.69 (95% confidence interval 1.21–2.37) (Table 2).

Previously reported CNVs in our genome-wide analysis

We examined the associations in our GWAS discovery sample for CNVs that were previously identified for obesity. We confirmed the association of CNV16p12.3 with BMI (P = 0.014) and body fat mass (P = 0.0021) (9, 14). Beyond the focus of this study on common CNVs, we also explored selected regions harboring rare CNVs. Glessner et al. (10) reported several CNVs that were exclusive to obese children. Among these CNVs, we found two obese subjects (BMI 30.6 and 37.3 kg/m²) with a deletion on chromosome 5q14.3 and one obese subject (BMI 34.08 kg/m²) with a deletion on chromosome 7q31.1. Moreover, for CNVs reported for subjects with severe early-onset extreme obesity (5), we detected CNV3p11.2 in one obese subject (BMI 39.6 kg/m²), CNV8q24.3 in two obese subjects (BMI 33.6 and 30.0 kg/m²), and CNV16p11.2 in one obese subject (BMI 34.5 kg/m²), respectively.

Discussion

In this study, we discovered that CNV10q26.3 was significantly associated with the risk of obesity in both white and African populations. Most of the genetic studies have been performed in whites. Evaluating the associations in different ethnicities, particularly African-derived populations, is necessary and valuable. First, the prevalence of obesity in African-Americans (men: 37.3%; women: 49.6%) is higher than in white Americans (men: 31.9%; women: 33.0%) (16). Second, it may help determine the generality of the findings because the genomic variation is greater when compared across ethnicities. For example, Zhao et al. (17) identified three CNVs for obesity in African-Americans, which have not been previously reported in whites. Glessner et al. (10) found eight rare CNVs that were exclusive to both obese white and African-American children. Our consistent association results in whites and African-Americans indicate that CNV10q26.3 might be a common variant for obesity, even across different ethnicities.

A gene, CYP2E1 (cytochrome P450, family 2, subfamily E, polypeptide 1), is located within CNV10q26.3. Levels of CYP2E1 are affected under a variety of physiological and pathophysiological conditions, like a high-fat diet, fast, and hunger (18). Yoshinari et al. (18) found that the mechanisms of CYP2E1 in the rat liver were also functional in the rat adipose tissues, although in humans, Emery et al. (19) indicated that CYP2E1 induction was closely related to morbid obesity. Moreover, a 96-bp insertion mutation of CYP2E1 was associated with greater CYP2E1 metabolic ability in obese subjects (15). Our findings, combined with the above evidence, support the potential contribution of CYP2E1 to the pathogenesis of obesity.

In conclusion, using data from 14,654 individuals, we suggest that CNV10q26.3 contributes to the genetic susceptibility of obesity. Further functional studies will be needed for clarification of the potential mechanism.