SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods and Procedures
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Objective

Although genome-wide association studies (GWAS) have substantially contributed to understanding the genetic architecture, unidentified variants for complex traits remain an issue. One of the efficient approaches is the improvement of the power of GWAS scan by weighting P values with prior linkage signals. Our objective was to identify the novel candidates for obesity in Asian populations by using genemapping strategies that combine linkage and association analyses.

Design and Methods

To obtain linkage information for body mass index (BMI) and waist circumference (WC), we performed a multipoint genome-wide linkage study in an isolated Mongolian sample of 1,049 individuals from 74 families. Next, a family-based GWAS, which integrates within- and between-family components, was performed using the genotype data of 756 individuals of the Mongolian sample, and P values for association were weighted using linkage information obtained previously.

Results

For both BMI (LOD = 3.3) and WC (LOD = 2.6), the highest linkage peak was discovered at chromosome 10q11.22. In family-based GWAS combined with linkage information, six single-nucleotide polymorphisms (SNPs) for BMI and five SNPs for WC reached a significant level of association (linkage weighted P < 1 × 10-5). Of these, only one of the SNPs associated with WC (rs1704198) was replicated in 327 Korean families comprising 1,301 individuals. This SNP was located in the proximity of the prosperorelated homeobox 1 (PROX1) gene, the function of which was validated previously in a mouse model.

Conclusion

Our powerful strategic analysis enabled the discovery of a novel candidate gene, PROX1, associated with WC in an Asian population.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods and Procedures
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Obesity is one of the most serious global health problems, as it increases the risk of many diseases, such as cardiovascular disease (CVD), type 2 diabetes (T2D), and certain cancers [1]. Both overall and central obesity, which are commonly measured by body mass index (BMI) and waist circumference (WC), respectively, are well-documented risk factors for many diseases. Moreover, the effects of central abdominal fat are independent of overall adiposity ([2]-[4]). In particular, several studies have reported that WC is more closely associated with cardiovascular disease risk factors than BMI [2, 4].

In the past few years, genome-wide association studies (GWAS) have become a popular approach to identify causal variants for common human complex diseases and traits, including obesity. Human obesity is considerably heritable, with heritability ranging from 45 to 85%. GWAS have led to substantial discoveries of genetic susceptibility loci for obesity. Nevertheless, common variants identified via GWAS only explain part of the heritability, emphasizing the importance of new genetic variants [5]. Many strategies to discover the remaining genetic variations for complex traits have been discussed [5, 6]. One of the effective and promising approaches to explain the remaining heritability is the use of extended family data, which may help detect rare variants, parent-of-origin-specific effects, and inheritance patterns. In particular, linkage information facilitates the interpretation of unexplained genetic variation, improving the power of GWAS by up-weighting P values based on linkage signals [5]. Structural variations, including copy number variations (CNVs), may also account for part of the unidentified heritability [5, 6].

The genetic heterogeneity of disease-causing variants among different ethnic groups has been well documented [7]. However, the majority of previous GWAS have focused on samples with European ancestry, and gene-mapping results in Asian populations remain rare. Manolio et al. emphasized the need to expand GWAS to non-European populations to find interesting new variants [5]. The use of isolated and extended families in family-based genetic studies has several advantages, such as the homogeneity of the genetic background, less environmental differences, and enhanced power based on multi-generation pedigree data [8, 9].

The objective of this study was to discover new variants associated with BMI and WC phenotypes in isolated Asian populations by using more powerful gene-mapping strategies that combine linkage and association analyses. We obtained genome-wide linkage information for a sample of 74 Mongolian families comprising 1,049 individuals recruited for the Gene Discovery for Complex Traits in Isolated Large Families of Asians of the Northeast (GENDISCAN) project. Among them, 756 samples were genotyped and analyzed for 567,072 SNPs by using family-based association results combined with linkage information. SNPs showing significant association with BMI or WC in the Mongolian sample were validated in an independent sample from the Korean Healthy Twin study, which consisted of 327 families comprising 1,301 individuals.

Methods and Procedures

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods and Procedures
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Subjects and phenotypes

In 2006, 2,008 samples were collected in Dashbalbar, Dornod Province, Mongolia, as part of the GENDISCAN project [10, 11]. Briefly, the purpose of the GENDISCAN project is to discover complex trait loci in isolated large Mongolian families that are suitable for gene-mapping studies. This population is geographically isolated and family structures are very complicated, with many generations, full and half siblings, cousins, avuncular, and spouses. In this study, we selected 74 extended families comprising 1,049 individuals with phenotypic information. Complicated pedigree information was obtained via personal interviews and was confirmed by genotyping data. DNA (for genotyping) was extracted from venous blood samples from study participants according to standard protocols. Informed consent was obtained from all subjects and the study was approved by the Institutional Review Board of Seoul National University (approval number, H-0307-105-002). Anthropometric data, such as height, weight, and WC (for obesity traits) were measured with individuals wearing light clothing and without shoes. Height and weight were measured using a wall stadiometer and an electronic balance, respectively, and were recorded by well-trained examiners. BMI was calculated by dividing the weight (kg) by the height2 (m2). Because of the non-normal distribution of phenotypes, we applied a z-transformation. In addition, descriptive information, such as age and sex, was obtained from questionnaires.

Familial correlations and heritability

Prior to the linkage analysis, we evaluated the familial correlation and narrow-sense heritability (i.e., the proportion of phenotype variance attributable to additive genetic variance) with residuals of phenotype after adjusting for age and sex. The FCOR option of the Statistical Analysis for Genetic Epidemiology (S.A.G.E) version 6.0.1 software was used to estimate familial correlation,

  • display math

by using a set of two random variables inline image and arbitrary weights {wi}. Heritability was estimated via a maximum likelihood procedure using the Sequential Oligogenic Linkage Analysis Routines (SOLAR) software package.

Genome-wide linkage scan

Seventy-four families comprising 1,049 family members were genotyped for 1,039 short tandem repeat (STR) markers. The methods used for genotyping and error checking are described in detail in a previous study [10]. Briefly, we checked the Mendelian and non-Mendelian genotype errors and calculated the MIBD using the LOKI package. To localize the QTL associated with obesity, we performed multipoint linkage analysis across the 22 autosomes using SOLAR. To calculate the empirical P value of logarithm of odds (LOD) scores, we implemented 10,000 permutation simulations. The linkage results were adjusted using covariates, such as age and sex.

Genome-wide association study

We also performed a family-based GWAS to discover variants associated with obesity. Of the families used for the linkage study, 55 extended families comprising 756 individuals were genotyped for SNP data using Illumina Human 610-Quad Beadchip. The Mendelian and double-recombination genotype errors were removed using PEDCHECK and MERLIN, respectively [11]. The genotyping quality of SNPs was validated by considering the SNP call rate (>99%) and marker error rate (<1%). In addition, we excluded SNPs that had a minor allele frequency (MAF) < 5%. After quality assessments, the final number of SNPs used in this study was 567,072. The PBAT tool of the HelixTree software, version 6.4 (GoldenHelix, Bozeman, MT), was used for family-based association analyses. The family-based association tests were derived under the null hypothesis of “no linkage and no association” and the assumption of additive genetic effects of individual alleles. Family-based association tests (FBATs) based only on transmission disequilibrium test (TDT) can lead to the loss of information on population distribution. For this reason, we calculated powerful and robust screen P values by combining independent family based association test (FBAT) and screening statistics [12]. The association results were adjusted using covariates, such as age and sex.

P value weighting using linkage information

We combined linkage scan information with the association results and obtained new weighted P values for SNPs based on the method suggested by Roeder et al. (2006). The new approach can enhance the power of association studies by incorporating linkage scan results as prior information [13]. We applied this new approach to obtain the weighted P value of each SNP. The weight was calculated as inline image, where inline image, B = 1 (exponential weighting), N is the number of SNPs, and Zi is the corresponding linkage score at the position of each SNPi. The linkage signal value at each SNP base-pair position was interpolated using linkage scores at nearby STR markers. Subsequently, the weighted P value was obtained by the division of the GWAS P value by the weight value of each SNP position [14]. We used the SFDR software version 1.6 to compute weighted P values (http://www.utstat.utoronto.ca/sun/Software/SFDR/index.html).

Replication

We used the samples of the Korean Healthy Twin study (data collected from 2005 to 2006) as a replication sample. This cohort is part of the Korean Genomic Cohort Study and includes Korean twins older than 30 years, as well as their adult family members. The samples were genotyped using the Affymetrix Genome-Wide Human SNP Array 6.0. The samples genotyped included 327 families comprising 1,301 family members with phenotypic information for BMI and WC. The mean phenotypic values between monozygotic twins were used in the association test. We identified the SNPs available in the replication data within linkage disequilibrium (LD) blocks containing the significant SNPs in the discovery step using Haploview software version 4.1, because of the difference in genotype platforms between the two studies. Four SNPs for BMI and three SNPs for WC were available in the genotype data of the replication sample. A total of 26 and 28 SNPs for BMI and WC, respectively were tested for replication test. We performed a family-based association test and obtained the FBAT P values for the replication sample. In addition, the combined P values for overall association of GWAS and replication were calculated. Bonferroni correction for multiple testing assumes that all comparisons are independent. However, most of tested SNPs here are highly correlated and are not absolutely independent from each other. In this situation, using the Bonferroni adjustment based on the number of tested SNPs may lead to overly strict correction. Therefore, we applied an alternative significance threshold based on the effective number of independent tests suggested by Cheverud [15]. A total of five and six effective tests for BMI and WC were respectively performed in replication test, suggesting a threshold P value of 0.01 and 0.008 for significance of BMI and WC.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods and Procedures
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Study characteristics

Table 1 presents the main descriptive characteristics of the individuals included in this study. We performed a genome-wide linkage scan and family-based GWAS using samples from the GENDISCAN project. A total of 1,049 individuals from 74 families were included in the linkage study. Among these subjects, 756 individuals from 55 families were selected for the association test. The mean BMI and WC were 23.6 and 78.5, respectively. The linkage and association samples had consistent distributions of characteristics. As a replication sample, 1,301 individuals from 327 nuclear families were collected from the Korean Healthy Twin cohort. The age of the subjects in this cohort was higher than that of the GENDISCAN cohort because of the restricted recruitment of individuals over the age of 30 in the latter. The mean values of BMI and WC in the replication cohort were 24.0 and 81.8, respectively, which were slightly higher than those of the discovery set.

Table 1. Characteristics of study participants
CharacteristicsGenome-wide linkage scanFamily-based GWASReplication study
  1. BMI, body mass index; WC, waist circumference; GWAS, genome-wide association study.

  2. aData are provided as means (standard deviations) for continuous variables.

StudyGENDISCANGENDISCANKorean Healthy Twin
Family (n)7455327
Sample (n)10497561301
Women, n (%)544 (51.9)400 (52.9)790 (60.7)
Age (years)a32.1 (16.5)31.4 (16.2)47.1 (13.7)
Weight (kg)a58.0 (15.7)57.9 (15.4)62.9 (11.0)
BMI (kg/m2)a23.6 (5.0)23.6 (5.0)24.0 (3.3)
WC (cm)a78.5 (13.6)78.0 (13.9)81.8 (9.0)

Estimation of the genetic component

We evaluated the relative contribution of genetic factors to obesity by estimating correlation coefficients in familial pairs and by estimating heritability. The adjusted familial correlations and heritability estimates for obesity are summarized in Table 2. Familial correlations between unrelated spouse pairs were not significant (P > 0.05), whereas familial correlations between related pairs, such as parent–offspring and sibling pairs, were significant (P < 0.05). The familial correlations in sibling pairs (BMI = 0.32; WC = 0.24) were higher than those observed for parent–offspring pairs (BMI = 0.11; WC = 0.11). In particular, brother–brother pairs had the largest correlation coefficients (BMI = 0.42; WC = 0.42). Moreover, the narrow-sense heritabilities of BMI and WC in the Mongolian sample were estimated to be 34 and 31%, respectively (P < 0.05). Based on these results, we identified significant familial resemblance and heritable genetic components of the obesity phenotype in the Mongolian population.

Table 2. Familial correlation and heritability for obesity
RelationshipnBMIWC
Correlation (SE)P valueCorrelation (SE)P value
  1. The results were calculated as residuals after adjusting for age and sex. BMI, body mass index; WC, waist circumference; SE, standard error.

Parent–offspring9490.11 (0.04)0.00690.11 (0.04)0.0062
Father–son1730.13 (0.09)0.14520.15 (0.09)0.1034
Mother–son2750.10 (0.07)0.14770.10 (0.07)0.1912
Father–daughter1770.03 (0.08)0.73930.05 (0.08)0.5654
Mother–daughter3240.15 (0.06)0.01490.14 (0.06)0.0133
Sibling7340.32 (0.05)<0.00010.24 (0.05)<0.0001
Brother–brother1670.42 (0.09)<0.00010.42 (0.09)<0.0001
Sister–brother3690.32 (0.06)<0.00010.22 (0.06)0.0010
Sister–sister1980.27 (0.08)0.00140.15 (0.08)0.0639
Half-sibling3950.19 (0.07)0.00550.18 (0.06)0.0061
Avuncular8880.06 (0.06)0.27980.04 (0.05)0.3758
Cousin5980.05 (0.07)0.49090.04 (0.06)0.5662
Spouse1380.15 (0.08)0.07270.16 (0.08)0.0685
Heritability1,0490.34 (0.05)<0.00010.31 (0.06)<0.0001

Genome-wide linkage scan

The genome-wide multipoint linkage study was performed using 1,039 STR genotype markers. Linkage analysis for obesity revealed two candidate loci with suggestive evidence of linkage, i.e., LOD > 2 (Table 3). The highest LOD scores for both BMI and WC (BMI = 3.3; WC = 2.6) were observed on chromosome 10q11.22, and the nearest marker was D10S1772. Another modest linkage signal for BMI was observed on chromosome 17p12, with a LOD score of 2.3 and D17S922 was the nearest marker. We obtained all LOD score information for each locus of the 22 autosomes (Figure 1).

Table 3. Genome-wide linkage scan results for obesity (LOD score > 2)
PhenotypeChromosome (location, cM)Maximum LOD scoreNearest markerLocusSupport intervala (cM)Empirical P value
  1. The LOD results were estimated using residuals after adjusting for age and sex. LOD, logarithm of odds; cM, centimorgan.

  2. aA support interval is defined from the maximum to 1.5 LOD score.

BMI10 (73)3.3D10S177210q11.2228-760.0002
 17 (42)2.3D17S92217p1232-430.0008
WC10 (73)2.6D10S177210q11.2231-760.0005
image

Figure 1. Genome-wide linkage results adjusted for age and sex across 22 autosomes in GENDISCAN samples. Multipoint LOD scores for (a) BMI and (b) WC.

Download figure to PowerPoint

Combination of GWAS results with linkage analysis

To identify potential candidate variants for obesity in this isolated Asian population, we applied a method that combined the GWAS results with linkage scan results, which had been obtained previously. First, we performed a family-based GWAS for two obesity phenotypes (BMI and WC) in Mongolian extended families; then, we obtained P values from FBATs for within-family information and screen statistics for between-family information using the PBAT tool. To enhance the power of traditional family-based GWAS, which use only the within-family component, we estimated new screen P values by integrating a between-family component. Finally, this result was combined with our linkage LOD score. Figure 2 shows quantile–quantile plots and Manhattan plots of P values from the GWAS combined with linkage information. More detailed association results for obesity are summarized in Table 4. Six SNPs for BMI and five SNPs for WC reached our significance level of <1 × 10−5 for linkage weighted P values. Two of the SNPs located on chromosome 5q14.1 (listed in Table 4) were significant for both BMI and WC: rs259102 (P for BMI = 3.3 × 10−7; P for WC = 4.79 × 10−6) and rs259067 (P for BMI = 1.25 × 10−6; P for WC = 6.05 × 10−6). The gene located nearest to these two SNPs was CMYA5 (rs259102 was located in an intron and rs259067 was located at a distance of 2.9 kb). The rs499953 SNP on chromosome 7q22.1 was associated with BMI and was located within one of the introns of the Reelin gene (RELN). In addition, we identified two significant loci on chromosomes 7q31 (MDFIC) and 10q11.22 (WDFY4). rs12237835, which is located in an intergenic region, was the SNP with the strongest P (2.27 × 10−6) for WC. The protein tyrosine phosphatase receptor type D (PTPRD) gene was the gene located nearest to this SNP (142.2 kb away). We also found two loci associated with WC on chromosomes 3q26.33 and 1q32.3. In particular, rs1704198 (P = 9.44 × 10−6) located on chromosome 1q32.3 reached the significant threshold after weighting using linkage information. This SNP was located 251 kb away from the prospero-related homeobox 1 (PROX1) gene. Figure 3 indicates the genomic localization of this SNP, near PROX1. As shown in Figure 3, we identified two SNPs that were in strong LD with rs1704198 (rs1704199: r2 = 0.87, D′ = 1; rs6680799: r2 = 0.81, D′ = 1).

image

Figure 2. Genome-wide association results for obesity in GENDISCAN samples. Quantile–quantile plots and Manhattan plots for (a) BMI and (b) WC.

Download figure to PowerPoint

image

Figure 3. Regional association plot for PROX1 on chromosome 1. The blue diamond indicates the strongest SNP in each gene region. The colors of the circles represent LD structures with the strongest SNP (white, r2 < 0.2; yellow, 0.2 ≤ r2 < 0.5; orange, 0.5 < r2 < 0.8; red, r2 ≥ 0.8). SNP locations are based on build 36 of the human genome.

Download figure to PowerPoint

Table 4. GWAS results combined with linkage information for obesity (linkage-weighted P < 1 × 10–5)
 ChrDiscoveryReplicationCombined (P)Nearest genes (distance, kb)
SNPLocusPositionMA (MAF)FBAT (P)aScreen (P)bLinkage-weighted (P)cFBAT (P)a
  1. Significantly replicated results are indicated in bold.

  2. Chr, chromosome; FBAT, family-based association test; NA, not available; MA, minor allele; MAF, minor allele frequency.

  3. aFBAT (P) indicates the P value for only the within-family component.

  4. bScreen (P) indicates the integrating P of the within- and between-family information.

  5. cLinkage-weighted (P) was calculated as screen(P)/weight.

BMI5rs2591025q14.179128126A (0.47)1.96 × 10−63.11 × 10−73.32 × 10−7NACMYA5(–)
 5rs2590675q14.179134754A (0.47)1.03 × 10−51.17 × 10−61.25 × 10−64.32 × 10−29.58 × 10−7CMYA5(2.9)
 7Rs4999537q22.1103184902T (0.31)2.48 × 10−36.04 × 10−63.86 × 10−6NARELN (–)
 7Rs132259067q31.1114355340G (0.12)5.16 × 10−31.11 × 10−54.46 × 10−63.26 × 10−12.10 × 10−5MDFIC (–)
 7Rs77844477q31.2114416521A (0.11)1.85 × 10−31.34 × 10−55.37 × 10−65.19 × 10−24.49 × 10−6MDFIC (–)
 10rs73339410q11.2249618247C (0.45)1.79 × 10−35.27 × 10−57.15 × 10−66.03 × 10−15.76 × 10−5WDFY4 (–)
WC9Rs122378359p24.18162046A (0.20)1.02 × 10−21.80 × 10−62.27 × 10−6NAPTPRD (142.2)
 5rs2591025q14.179128126A (0.47)1.66 × 10−45.73 × 10−64.79 × 10−6NACMYA5 (–)
 3rs14034763q26.33183965551T (0.49)2.37 × 10−55.06 × 10−65.26 × 10−69.16 × 10−16.38 × 10−5ATP11B (28.4)
 5rs2590675q14.179134754A (0.47)1.86 × 10−47.24 × 10−66.05 × 10−63.54 × 10−13.01 × 10−5CMYA5 (2.9)
 1rs17041981q32.3211977117A (0.19)1.41 × 10−31.16 × 10−59.44 × 10−62.34 × 10−34.11 × 10−7PROX1 (251.4)

Replication in a cohort of Korean families

To confirm our GWAS findings in the Mongolian population, we performed a replication study using samples from Korean families from the Korean Healthy Twin cohort. The LD blocks including the positions of the three SNPs, rs259102, rs499953, and rs12237835, which were associated with obesity in the Mongolian cohort, were not found in the Korean population. Therefore, we could not include these three SNPs in the replication study. The remaining six SNPs were tested for family-based association with obesity in the Korean sample (Table 4), and the combined P values of GWAS and replication were calculated. In replication step, only SNP rs1704198 located near the PROX1 gene exhibited a significant association with WC in the Korean sample (P = 2.34 × 10−3). This SNP showed the lowest overall P value and reached the less strict genome-wide significance threshold of 5 × 10−7.

Confirmation of Obesity Genes Reported Previously

We also verified the genes related to obesity phenotypes that had been reported previously in the HuGE Navigator literature database (www.hugenavigator.net). Our results for BMI and WC were evaluated using a less stringent P value (< 1 × 10−3) and are summarized in Tables S1 and S2 (Supporting Information), respectively. We replicated the four obesity genes identified in previous GWAS (EDIL3, FTO, MSRA, and TNKS) in our Mongolian sample. In addition, 19 and 21 genes that were reported for BMI and WC, respectively, based on functional or other association studies, were confirmed in our Mongolian cohort. These results provide evidence that replicates the previous GWAS findings in the Mongolian population. In addition, we identified the association of INSIG2 which has shown inconsistent replications ([16]-[18]). We sought to the SNPs within 300 kb up- or downstream of INSIG2, because the previously known SNP, rs7566605 was not included in our GWAS data. We observed that none of the SNPs achieved our significance level (Data not shown).

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods and Procedures
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

The aim of the present study was to discover new candidate loci for obesity in an isolated Asian population by applying several powerful gene-mapping methodologies. Using a weighted analysis of association signals by incorporating linkage signal information, we identified a novel locus at 1q32, near the PROX1 gene, that was associated with obesity (WC) in the GENDISCAN sample. This locus was confirmed using the Korean Healthy Twin cohort.

In addition to the 1q32 locus, our linkage analysis also identified an independent obesity susceptibility locus at chromosome 10q11.22. Our strong linkage signal on chromosome 10, with a support interval ranging from 28 to 76 cM, supports several previous findings. In 1998, Hager et al. reported a major locus for human obesity on chromosome 10 [19], which was replicated independently in 2000 using German families [20]. Another linkage study has also emphasized the importance of linkage to obesity at chromosome 10 [21]. Our results revealed that one SNP (rs733394) located in this region was associated with BMI in the Mongolian sample; however, we have failed to replicate this association in the Korean sample. This may be explained by the hypothesis that the genetic variants related to obesity at this locus are not single variants but structural variants, including CNVs. Interestingly, in 2009, one study based on the Han Chinese population discovered that a CNV at 10q11.22 was significantly associated with BMI, contributing to 1.6% of BMI variation [22]. Recently, an association between obesity and a common deletion at chromosome 10q11.22 was replicated using family-based and case-control GWAS samples [23]. The authors suggested that this CNV may account for some of the “missing heritability” for obesity as variants that contribute substantially to the genetic basis of obesity [23]. Currently, the major effects of single variants, including SNPs, located in this region have not been well documented in previous GWAS. In our study, the failure to replicate the SNP at 10q11.22, despite the strong linkage signal, may be due to the presence of CNVs in this region.

The 1q32 locus was reported by a previous linkage study based on an American-Indian diabetes sample, which yielded significant linkage evidence with a maximum LOD score of 3.7 as weight [24]. Our results showed that one SNP (rs1704198) was significantly associated with obesity in both the GENDISCAN and Korean Healthy Twin cohorts. This variant is located in an intergenic region located near PROX1. Several hypothetical PROX1 functions have been proposed in previous reports. PROX1 regulates cell differentiation and organogenesis in various tissues, including the liver, pancreas, lymphatic vessels, and retina ([25]-[28]). PROX1 expressed in endothelial cells acts as a regulator of the development of the lymphatic system [25], and loss of PROX1 is closely related to the formation of a smaller liver [26]. PROX1 also controls the exit of progenitor cells from the cell cycle in the retina [27] and is essential for the regulation of early pancreas organogenesis [28]. Notably, Harvey et al. reported that PROX1 heterozygosity in a mouse model led to adult-onset obesity and heavier adipocytes with a greater circumference compared with wild-type animals. In addition, PROX1 heterozygous mice had not only high levels of insulin and leptin but also abnormal hepatic lipid accumulation [29]. These authors proposed that the disruption of lymphatic vessels promotes fat accumulation in lymphatic areas because of increased levels of lipids in adipocytes. A recent GWAS for fasting glucose homeostasis revealed that the PROX1 locus was associated with both fasting glucose and T2D [30]. We propose PROX1 as an interesting candidate gene for obesity based on its validation in a mouse model and its functional relation to this trait.

In addition to PROX1, PTPRD, and RELN are potential candidate genes for obesity, although we did not confirm the signals found at the discovery step because of the unavailability of the variants of these genes in the replication sample. PTPRD is expressed in many tissues, such as the brain, kidney, skeletal muscle, and thyroid; in particular, PTPRD-deficient mice exhibit impaired learning and memory [31]. In humans, PTPRD was reported previously as a susceptibility gene for T2D in the Han Chinese [32] and for both obesity and asthma in children in a white population [33]. Similar to PTPRD, RELN is expressed mainly in the brain and plays a critical role in the central nervous system [34]. This gene has been associated with schizophrenia [35] and major depression [36]. In 2010, a report demonstrated an association between obesity and depression using a bivariate model approach, indicating the presence of 12% shared effects of genetic components [37]. However, replication of these genes needs to be achieved before they can be identified as causal genes for obesity.

Some genes suggested in previous GWAS, such as FTO, EDIL3, MSRA, and TNKS, were also moderately replicated in this study. In particular, MSRA is well documented in the European population as a candidate locus for central obesity, which is measured by WC and waist–hip ratio [38, 39]. Our results support the replication of the genes identified for obesity in European populations in a Mongolian population, although the association signal was moderate. Besides, we failed to replicate the association of INSIG2 which was previously reported by family based GWAS from Framingham Heart Study [16]. Similar to ours, one study based on the Korean population showed the no significant association of rs7566605 with obesity related phenotypes [18]. Many studies tried to validate this finding, but it has shown inconsistent replication results ([16]-[18]). These facts may be explained by ethnic difference in effects of INSIG2 variants.

In this study, several strategies were applied to improve the power of GWAS by using family data, which is robust against genetic heterogeneity and population stratification. First, we calculated the screen P values, integrating the mutually independent within-family component of FBAT and the between-family component with screening statistics. In general, traditional family-based association tests, such as the TDT and FBAT, use only the within-family component. However, this method leads to the loss of between-family information (population-based information) because it considers only the transmission of genotypes within each family. The integrating method used in this study is more powerful than FBAT alone, as it provides a robust combination of mutually independent FBAT P values and screening statistics [12]. In addition, we estimated linkage-weighted P values to enhance the power of GWAS results. Roeder et al. reported that, if the linkage study is informative, i.e., the region of association is supported by a strong linkage signal, this approach can improve the power of association results considerably [13]. The PROX1 SNP rs1704198 reached our significance level of P value < 1 × 10−5 after recalculation of the P values for the entire region using weights from the linkage signal. The methods used in this study can enhance the power of family-based genome-wide association analysis, while maintaining the same validity as a “naive” association analysis, and may be helpful to find out the additional novel variants for obesity [5]. In our study, none of SNPs achieved a genome-wide significance of 5 × 10−8. This threshold corresponds to the Bonferroni correction for one million independent markers, which might be too conservative considering the correlations of SNPs in LD [15]. Therefore, we relaxed the screening criteria for replication study using 1 × 10−5 threshold. Selecting top ranking SNPs in the first stage of GWAS is a commonly used practice and methodologically proven [40]. With our relaxed threshold, we selected total of 28 SNPs for replication in WC and this number is quiet small compared to usual ranking based strategies.

In summary, we identified a novel locus for obesity at 1q32, near PROX1, in a sample of isolated Mongolian families by incorporating multiple strategies to improve the power of the GWAS. The significance of this locus located near PROX1 was confirmed in a replication sample of Korean families. Our candidate gene, PROX1, was validated in a previous functional study based on a mouse model, suggesting that PROX1 heterozygous mice are a new model of adult-onset obesity [29]. To the best of our knowledge, we have reported a novel candidate locus located near PROX1 that is associated with WC in an Asian population by applying several powerful strategic analyses. However, additional studies are needed to confirm the causality and functionality of PROX1 in obesity.

Acknowledgments

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods and Procedures
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

We thank Sungho Won for providing the R script for combining within- and between-families components in the PBAT tool.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods and Procedures
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods and Procedures
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Additional Supporting Information may be found in the online version of this article.

FilenameFormatSizeDescription
oby20153-sup-0001-SuppInfo.doc103KSupporting Information

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.