Examination of All Type 2 Diabetes GWAS Loci Reveals HHEX-IDE as a Locus Influencing Pediatric BMI

Our study cohort consisted of 7,184 singleton children of European ancestry with systematically recorded height and weight. All subjects were consecutively and randomly recruited from the greater metropolitan area of Philadelphia from 2006 to 2009 at The Children's Hospital of Philadelphia; i.e., participants were not specifically targeted for obesity-related traits. The study was approved by the institutional review board of The Children's Hospital of Philadelphia. Parental informed consent was given for each study participant for both the blood collection and subsequent genotyping.

We performed high throughput genome-wide SNP genotyping using the Illumina Infinium II HumanHap550 or Human 610 BeadChip technology (Illumina, San Diego, CA) at The Children's Hospital of Philadelphia's Center for Applied Genomics as described previously (15). The overall genomic control value was 1.036. The SNPs analyzed survived the filtering of the genome-wide dataset for SNPs with call rates <95%, minor allele frequency <1%, missing rate per person >2%, and Hardy-Weinberg equilibrium P < 10⁻⁵.

Most loci described from GWAS published to date have been found using either the Affymetrix or Illumina platform. In the event a locus was reported using both the Illumina and Affymetrix arrays, we used the SNPs present on the Illumina array. In the event of a signal only being described on the Affymetrix array, we either already had the SNP on our Illumina array or identified and used the best surrogate SNP available based on the CEPH (Centre d'Etude du Polymorphisme Humain) from Utah (CEU) HapMap (supplemental Table 1, which can be found in an online appendix at http://diabetes.diabetesjournals.org/cgi/content/full/db09-0972/DC1). We used two SNPs at the CDKAL1 (rs4712523 and rs7756992; r² = 0.677) and HHEX-IDE (rs1111875 and rs7923837; r² = 0.698) loci as the association with type 2 diabetes from various GWAS reported different SNPs, which were in imperfect linkage disequilibrium (LD) with each other. rs3751812 at FTO was included as a positive control as we have previously reported the association with this SNP and both pediatric obesity and pediatric BMI (16,17).

BMI percentiles were defined using the standard Centers for Disease Control (CDC) growth chart z scores that take into account age and sex. All subjects were biologically unrelated and were between 2 and 18 years of age. All subjects were between ±3 SDs of CDC corrected BMI; i.e., outliers (n = 356) were excluded to avoid the consequences of potential measurement error or Mendelian causes of extreme obesity.

We queried the data for the SNPs of interest in our pediatric sample. All statistical analyses were carried out using the software package PLINK (version 1.05) (18). We applied PLINK to the generation of genome-wide identical by state estimates between all subjects and then generated multidimensional scaling (MDS) plots for visual examination of population outliers. To help interpret the population genetic analysis, we included 924 HapMap3 individuals from 11 populations as positive control subjects into the MDS analysis. The individuals of European ancestry were selected by the principal component one of >0.04 and principal component two of >0.01. Comparing self-identified ancestry with the MDS-inferred ancestry confirmed the reliability of MDS to identify genetically inferred individuals of European ancestry.

By treating the normalized BMI z score as a quantitative trait, association analysis for each SNP was carried out using linear regression (additive model) with the SNP included as an independent variable (coded as 0, 1, and 2). With 3,592 subjects in the discovery cohort, the powers to detect 0.2, 0.3, 0.4, 0.5, 0.6, 0.8, and 1% variation at the α = 0.0025 level were 27.0, 49.0, 68.2, 82.0, 90.6, 97.9, and 99.6%, respectively.

In our analysis, 20 SNPs corresponding to the 18 type 2 diabetes loci previously discovered in GWAS of the disorder were investigated, namely, ADAMTS9, CDC123-CAMK1D, CDKAL1, CDKN2A/B, EXT2, FTO, HHEX-IDE, IGF2BP2, the intragenic region on 11p12, JAZF1, KCNQ1, LOC387761, MTNR1B, NOTCH2, SLC30A8, TCF7L2, THADA, and TSPAN8-LGR5 (Table 1).

We randomly partitioned our cohort exactly in half in order to have a discovery cohort (n = 3,592) and a replication cohort (n = 3,592). Five of these 20 SNPs yielded at least nominally significant association with BMI (P < 0.05) in the discovery cohort, representing four different independent loci.

Of these four loci, the minor allele of rs3751812 at the FTO locus yielded the strongest association with P = 3.81 × 10⁻⁵ and tracked with higher BMI. The direction of effect was also readily replicated in the additional cohort (P = 5.56 × 10⁻⁶), yielding a combined P = 1.05 × 10⁻⁹.

The major type 2 diabetes–conferring G allele of rs7923837 at the HHEX-IDE locus was associated with higher pediatric BMI in both the discovery (unadjusted P = 0.0013; Bonferroni correction for 20 variants threshold P ≤ 0.0025) and replication (unadjusted P = 0.023) sets (combined unadjusted P = 1.01 × 10⁻⁴). The major C allele of rs1111875 at the same locus was also trending with higher pediatric BMI but did not survive the Bonferroni correction for multiple testing in the discovery cohort.

The other two nominally significant loci in the discovery cohort, rs4402960 at IGF2BP2 (P = 0.05) and rs11257622 at CDC123-CAMK1D (P = 0.024) failed to replicate in the additional cohort. Association was not detected at all with any of the other type 2 diabetes loci uncovered to date through GWAS.

We also analyzed male and female subjects separately, but the effect of the G allele rs7923837 at the HHEX-IDE locus on pediatric BMI did not vary by sex (supplemental Table 2). However, we did look at different age bins and found that the variant was associated with higher pediatric BMI most strongly in the 2- to 6-year-old age bin (supplemental Table 3). By further breaking down the ages into individual years, nominally significant association for this HHEX-IDE variant in the same direction was observed at ages 3, 7, 14, and 16 years (supplemental Table 4). However, we did not observe an overall statistical interaction with age, with the interaction P values for rs1111875 and rs7923837 being 0.2507 and 0.1076, respectively.

If a genomic variant is well established to be associated with a trait that is the consequence of a defect of recognition of insulin by the body or by a fault in the amount of insulin released for the pancreatic islets (i.e., type 2 diabetes), then if these defects are operating at all in childhood, one might expect there to be an impact on BMI in childhood.

With this notion in mind, we queried the existing dataset from our ongoing GWAS of pediatric BMI if any of the type 2 diabetes loci uncovered in GWAS to date played a role in our trait of interest; it should be noted that PPARG, KCNJ11, and WFS1 were not included as their discovery with respect to being type 2 diabetes loci predates GWAS and thus have already been more extensively investigated. Our data in fact do show that the same genetic HHEX-IDE variant that is significantly associated with type 2 diabetes from previous studies also influences pediatric BMI. Indeed, the major G allele of rs7923837 at the HHEX-IDE locus was associated with higher pediatric BMI in both the discovery and replication cohorts, which is the same allele that has been reported to confer risk of type 2 diabetes. This mirrors very well what has been seen with the much more established FTO gene reported here and in other studies.

SNP rs7923837 yielded the fourth strongest association with type 2 diabetes in a Canadian/French GWAS carried out on the Illumina HumanHap platform (1). SNPs rs1111875 and rs7923837 yielded the strongest association at the HHEX-IDE locus, but it should be noted that they are far from being in perfect LD with each other (r² = 0.698), and thus both are included in the current study. However, despite the lack of complete concordance and the large sample size, we were unable to separate the effects of these SNPs as they cannot be considered to be totally independent signals either.

One hypothesis could be that the fetal genotype for rs7923837 is primarily associated with birth weight given that reduced birth weight is often reported to be associated with increased BMI and type 2 diabetes later in life. However, this does not appear to be the case as we have already investigated and reported the role of these type 2 diabetes loci in the context of birth weight in our cohort. Although we have agreed with previous studies that CDKAL1 is a birth weight-associated gene, we have not observed such an association with HHEX-IDE (19). Further, although there is no CDC categorization for the under 2-year-old age-group, we do not observe association between rs7923837 and BMI in this age category following our own normalization (data not shown). The correlation between birth weight and BMI in later childhood is less correlated than in earlier stages, suggesting that the HHEX-IDE variant exerts its physiological influence directly rather than as a consequence of a knock-on effect from a primary impact on birth weight. However, we do acknowledge that of the age bins studied, the strongest effect was observed in the 2- to 6-year-old age bin (effect size [SE] = 0.12 [± 0.04]) (supplemental Table 3). But this is not the whole story because at the individual age level, although more limited in terms of power, the impact continues to be observed into the mid-teens (supplemental Table 4).

The assumption in this study is that deficient insulin secretion mediates the effect on childhood BMI, but it is also possible that higher childhood BMI results in impaired insulin secretion later in life. There could indeed be pleiotropic associations from multiple independent mechanisms; however, we were not able to address this as we do not have insulin secretion/sensitivity measures in our study.

From our analysis, apart from FTO it is clear that only one of the loci previously reported from type 2 diabetes GWAS plays a role in our phenotype of interest, i.e., pediatric BMI. While this recently discovered locus unveils a new biomolecular pathway not previously studied in the context of type 2 diabetes and obesity, it is also important to note that this and other genetic associations with childhood obesity explain very little of the genetic risk for the pathogenesis of the trait (17); indeed, an estimate of the explained variance of the HHEX-IDE and FTO loci combined is only 0.98%, suggesting the existence of additional loci whose number and effect size remain mainly unknown. Current knowledge concerning the impact of genetic factors in the determination of pediatric BMI may still be very limited due to both the lack of availability of large pediatric cohorts with GWAS data and methodological difficulties in the analysis of the phenotype that changes with age and depends on many other contributing factors. Once our GWAS is complete, we will have the opportunity to look for other variants in the genome associated with BMI in childhood.

The study was supported in part by a Research Development Award from the Cotswold Foundation (to H.H. and S.F.A.G.) and National Institutes of Health Grant 1R01HD056465-01A1. This research was financially supported by The Children's Hospital of Philadelphia.

No potential conflicts of interest relevant to this article were reported.

Parts of this study were presented in abstract form at the 59th annual meeting of the American Society of Human Genetics, Honolulu, Hawaii, 20–24 October 2009.

We thank all participating subjects and families. Elvira Dabaghyan, Hope Thomas, Kisha Harden, Andrew Hill, Kenya Fain, Crystal Johnson-Honesty, Cynthia Drummond, Shanell Harrison, and Sarah Wildrick provided expert assistance with genotyping or data collection and management. We also thank Smari Kristinsson, Larus Arni Hermannsson, and Asbjörn Krisbjörnsson of Raförninn ehf for their extensive software design and contributions.