Studies of Metabolic Phenotypic Correlates of 15 Obesity Associated Gene Variants

Aims Genome-wide association studies have identified novel BMI/obesity associated susceptibility loci. The purpose of this study is to determine associations with overweight, obesity, morbid obesity and/or general adiposity in a Danish population. Moreover, we want to investigate if these loci associate with type 2 diabetes and to elucidate potential underlying metabolic mechanisms. Methods 15 gene variants in 14 loci including TMEM18 (rs7561317), SH2B1 (rs7498665), KCTD15 (rs29941), NEGR1 (rs2568958), ETV5 (rs7647305), BDNF (rs4923461, rs925946), SEC16B (rs10913469), FAIM2 (rs7138803), GNPDA2 (rs10938397), MTCH2 (rs10838738), BAT2 (rs2260000), NPC1 (rs1805081), MAF (rs1424233), and PTER (rs10508503) were genotyped in 18,014 middle-aged Danes. Results Five of the 15 gene variants associated with overweight, obesity and/or morbid obesity. Per allele ORs ranged from 1.15–1.20 for overweight, 1.10–1.25 for obesity, and 1.41–1.46 for morbid obesity. Five of the 15 variants moreover associated with increased measures of adiposity. BDNF rs4923461 displayed a borderline BMI-dependent protective effect on type 2 diabetes (0.87 (0.78–0.96, p = 0.008)), whereas SH2B1 rs7498665 associated with nominally BMI-independent increased risk of type 2 diabetes (1.16 (1.07–1.27, p = 7.8×10−4)). Conclusions Associations with overweight and/or obesity and measures of obesity were confirmed for seven out of the 15 gene variants. The obesity risk allele of BDNF rs4923461 protected against type 2 diabetes, which could suggest neuronal and peripheral distinctive ways of actions for the protein. SH2B1 rs7498665 associated with type 2 diabetes independently of BMI.


Introduction
Several studies have established that a genetic contribution to the pathogenesis of obesity exists [1][2][3]. Since the prevalence of obesity is increasing at alarming rates worldwide, great efforts have been made to identify the genetic components predisposing some individuals to accumulate more body fat. Especially searches for disease-associated single nucleotide polymorphisms (SNPs) have been prioritised, but without noticeable success prior to 2007 where genome-wide association studies (GWAS) were introduced. Several GWAS of body mass index (BMI) and/or obesity have been performed, and the first wave resulted in the suggestion of SNPs in/ near the INSIG2 [4], FTO [5], PFKP [6], and CTNNBL1 [7] genes, however, only SNPs in the FTO locus were convincingly replicated in other GWAS [6,8] and independent replication studies [9][10][11]. At the end of the first obesity GWAS wave, SNPs near the MC4R gene were identified, both in an independent GWAS [12] and through meta-analysis of several GWAS [13]. These associations have subsequently been confirmed in independent replication studies [14,15]. At the same time, another obesity gene, PCSK1, was suggested using re-sequencing of linkage peaks [16], and some replication studies have established week associations between this gene and obesity related measures [17,18].
In 2009, the second wave of BMI/obesity GWAS were performed, identifying 15 novel susceptibility loci. A GWAS of BMI and weight in almost 32,000 individuals of primarily European descendant validated the associations with FTO and MC4R. Moreover, genome-wide significant associations for variants in/ near two genes formerly suggested as biological candidate genes for obesity, BDNF and SH2B1 were observed, together with the identification of seven new BMI/weight loci, TMEM18, KCTD15, NEGR1, SEC16B, ETV5, FAIM2 and BAT2 [19]. A meta-analysis of GWAS on BMI comprising more than 32,000 individuals also confirmed the associations of FTO and MC4R, and furthermore identified six new loci; four of them overlapping with the other BMI/weight GWAS; SH2B1, TMEM18, KCTD15 and NEGR1 and two novel loci GNPDA2 and MTCH2 [20]. The last GWAS of the second wave identified three novel loci, NPC1, MAF and PTER in individuals with early-onset morbid adult obesity [21].
Given the hypothesis-free approach used in GWAS no obvious candidacy could be explained for most of the suggested loci. However, a predominance of the genes nearest the identified SNPs were expressed in the brain, and several particularly in the hypothalamus, suggesting putative roles in the regulation of appetite and energy expenditure [19,20,22].
Since there in the first GWAS wave were identified six loci, but only three FTO, MC4R and PCSK1 were confirmed in independent GWAS or replication studies, the importance of independent replication studies to distinguish between true associations and ''winners curse'' observations is underlined. Therefore, we attempt to confirm reported GWAS findings of association with obesity in Danish study samples. Furthermore, obesity is a strong risk factor of type 2 diabetes and other key metabolic traits, and therefore, the aim of the present study is also to elucidate whether 15 SNPs in/ near the 14 loci from the second obesity GWAS wave associate with type 2 diabetes and other key metabolic correlates.

Confirmation of associations with overweight and obesity
The BDNF rs4923461 A-allele associated with increased risk of overweight, with a per allele OR of 1.15 (1.07-1.24, p = 2.5610 24 ), and borderline with obesity with a per allele OR of 1.14 (1.05-1.23, p = 0.002) but not with morbid obesity ( A different pattern was observed for the TMEM18 rs7561317 G-allele, which strongly associated with obesity with an OR of 1.25 (1.14-1.37, p = 2.1610 26 ) and with morbid obesity with an OR of 1.46 (1.17-1.82, p = 8.3610 24 ) per allele, but not with overweight.

Association with type 2 diabetes
Only the SH2B1 rs7498665 G-allele strongly associated with increased risk of type 2 diabetes when adjusting for age and sex with an allelic OR of 1.18 (1.09-1.28, p = 3.0610 25 ) and when additionally adjusting for BMI a nominal association sustained with an OR of 1.16 (1.07-1.27, p = 7.8610 24 ).
The BDNF rs4923461 A-allele showed a tendency towards a reduced risk of type 2 diabetes, however, the association was merely nominally, with an OR of 0.87 (0.78-0.96, p = 0.008) per allele when adjusting for BMI (Table 1, Table S4), however, when omitting adjustments for BMI the nominally tendency disappeared. The contradictory risk altering patterns of obesity and type 2 diabetes could suggest an interaction between the risk allele and BMI, however, no such interaction was found (data not shown).

Discussion
In the second obesity GWAS wave 15 variants in/near 14 loci were identified, and several independent studies have already attempted replication of these findings. Here in our Danish study population we aim to validate these associations, and to elucidate whether some of the variants associate with other clinically relevant phenotypes. Our analyses show the same trends towards increased obesity risk, for all variants except rs10508503 near PTER. We report significant associations with overweight and/or obesity for the risk variants in/near BDNF, TMEM18, ETV5 and GNPDA2. In extended follow-up analyses of anthropometric and metabolic traits in the population-based Inter99 cohort we identified associations with risk alleles in/near BDNF, TMEM18, GNPDA2, SEC16B and FAIM2 when no adjustments for BMI were made. The effect sizes in our Danish study population were generally somewhat higher than for the discovery study (e.g. 0.52 kg/m 2 vs. 0.26 kg/m 2 for TMEM18 rs7561317 and 0.28 kg/m 2 vs. 0.19 kg/m 2 for GNPDA2 rs10938397 [20]), but the pattern of TMEM18 yielding the highest effect sizes is true both in our study and the discovery studies [19,20]. FAIM2 rs7138803 on the other hand show the second highest effect size on BMI in our study, whereas it exerts a more modest effect in the discovery study [19].
As obesity is a strong risk factor of type 2 diabetes we also examined associations with risk of type 2 diabetes, and unexpectedly established a BMI-dependent borderline association with the BDNF rs4923461 obesity risk allele and reduced risk of type 2 diabetes. An additional association with risk of type 2 diabetes, though increasing the risk as expected, was established for the variant in SH2B1. This latter association was nominally independent of BMI adjustments.
Other studies have also confirmed the associations between quantitative measures of obesity, i.e. BMI, weight and waist circumference, and/or the risk of obesity for variants in/near TMEM18 [23][24][25][26][27], SEC16B [27], NEGR1 [24,[27][28][29], SH2B1 [26,28,30], MTCH2 [24,28,29], GNPDA2 [25,28,30], FAIM2 [25,27,30], BDNF [25,27,29], and KCTD15 [29,30], however, no convincing pattern exist between the verified variants in the performed studies, which could be caused by low power for detecting modest effect sizes, and therefore, as the number of independent replication studies increases, a meta-analyses may be needed to determine true positive associations. The risk of type 2 diabetes has only been addressed in two other studies, and here association between TMEM18 [28], GNPDA2 [28,30], ETV5, FAIM2 and SH2B1 [30] and a BMI-dependent increased risk of type 2 diabetes have been reported. Hence, we are the first to report a tendency towards an increased risk of type 2 diabetes independent of BMI for SH2B1, and rs7498665 could therefore be suggested as a type 2 diabetes variant if this association is replicated in other independent studies. Particularly surprising from our analyses was the finding that the obesity risk allele of BDNF (rs4923461) was borderline associated with a reduced risk of type 2 diabetes when adjusted for BMI. However, these divergent associations were not explained by an interaction between the variant and BMI.
Interestingly, both SH2B1 and BDNF are obvious candidate genes for metabolic disorders based on the biological roles of the encoded proteins. That is, the SH2B1 adaptor protein is involved in several signal transduction processes, including the signalling mediated by the binding of insulin and leptin [31]. The association with type 2 diabetes is in line with previous studies showing insulin resistance in knockout mice [32][33][34][35]. Furthermore, a recent study demonstrated that TgKO mice, that only express SH2B1 in the brain and thus have loss of peripheral SH2B1, have impaired insulin sensitivity independent of body weight [35]. However, none of the examined measures of glucose homeostasis supported a possible reduction in insulin sensitivity in our study.    BDNF, brain-derived neurotrophic factor, is implicated in the regulation of body weight [36]. Yet, our results could suggest more ways for action of the BDNF protein. This is also supported by two studies of db/db mice. One where exogenous BDNF treatment reduces blood glucose concentrations independently of the hypophagic effect [37] and another where subcutaneous BDNF resulted in reduced blood glucose levels in BDNF administered mice, whereas pair-fed control mice displayed unchanged levels, despite that plasma insulin levels were significantly reduced in both groups [38].
Consistent with animal models, plasma BDNF levels have previously been reported to be inversely correlated with fasting plasma glucose among type 2 diabetes patients and to be associated with the severity of insulin resistance [39]. This suggests that BDNF regulates blood glucose homeostasis and insulin sensitivity peripherally. Hence, tissue specific up-regulation of endogenous BDNF levels in peripheral tissues could explain this BMI-dependent protective effect on type 2 diabetes for the reported BDNF obesity risk allele (rs4923461).
Contradictory, quantitative trait analyses show that the other risk allele near BDNF (rs925946) associated with elevated and not decreased fasting plasma glucose levels in our study population. Serum leptin and the inflammatory marker CRP were also found to be significantly elevated among rs925946 T-allele carriers in the present study. However, these associations omitted when adjusting for BMI and is therefore most likely mediated through the increased body fat accumulation, rather than being the cause of it.
Different neuronal and peripheral regulatory mechanisms of BDNF could be explained by different isoforms of the protein, and in fact tissue specific alternative splicing has been reported for BDNF in humans [40,41]. Furthermore, the BDNFOS gene, which is a non-protein-coding natural antisense transcript positioned downstream of BDNF in reverse orientation, has been suggested to have an important role in tissue specific regulation of BDNF expression through the formation of dsRNA duplexes [41]. In fact both gene variants are positioned within the BDNFOS locus. Thus, it could be hypothesized that the risk alleles could result in a muscle specific impairment of BDNFOS transcription and splicing, which would lead to a reduction in complementary BDNF RNAs and consequently increase the level of BDNF. Therefore physiological and experimental studies to illuminate the tissue specific and functional role of BDNF and BDNFOS will be of great interest.
Of the remaining associated obesity risk loci the potential functional role is less obvious as little is known of the encoded proteins. Furthermore, some of the identified risk variants are next to additional loci, in which the risk variants may influence as well. However, most of the identified loci are highly expressed in the brain, particularly in the hypothalamic region, indicating roles in appetite regulation and energy expenditure [19,20,22].
Many analyses were made in this study and some should therefore be regarded as hypothesis-generating with confirmation in independent studies as an important next step, but these analyses in the population-based Inter99 cohort still contribute to the follow-up of the reported findings of the second obesity GWAS wave.
It should moreover be noted that individuals from the Inter99 cohort (Note S1) were included in the follow-up of the top 43 variants in the BMI GWAS performed by Thorleifsson et al. However, since these studies were pooled with other ethnicities, we found it necessary to elucidate whether the associations with type 2 diabetes and obesity were present in an ethnically homogeneous study to avoid confounding by population stratification. In view of that, little overlap is observed in associations with type 2 diabetes suggesting that the effect of the BMI risk variants on type 2 diabetes risk is dependent on geographic origin. Moreover, we have in the present study examined the underlying metabolic phenotypes profoundly in order to shed light on the possible metabolic mechanisms causing the reported associations. The Inter99 cohort is in addition used in the replication part of the  newest approach in identifying BMI associated variants, the metaanalyses of ,125,000 individuals and independent replication in the same number, performed by the GIANT (Genomewide Investigation of ANThropometric measures) consortium, revealing 18 new BMI associated loci [42].
In conclusion, of the variants found to associate with obesity and related traits in the second GWAS wave, we were able to report association with obesity and/or measures of adiposity for variants in/near BDNF, TMEM18, ETV5, GNPDA2, SEC16B and FAIM2. Moreover, we found that SH2B1 rs7498665 strongly associated with type 2 diabetes in a BMI-independent manner. Our analyses also suggested that although rs4923461 in BDNF increase the risk of obesity, it conversely protect against type 2 diabetes, which could be through different neuronal and peripheral mechanisms.

Study populations
The 15 SNPs from the second obesity GWAS wave were genotyped in 18,014 individuals ascertained from four different study groups; 1) the Inter99 cohort, which is a population-based, randomised, non-pharmacological intervention study of middleaged individuals for the prevention of ischemic heart disease (n = 6,514), conducted at the Research Centre for Prevention and Health in Glostrup, Copenhagen (ClinicalTrials.gov ID-no: NCT00289237) [43]; 2) the ADDITION Denmark screening study cohort (Anglo-Danish-Dutch Study of Intensive Treatment in People with Screen-Detected Diabetes in Primary Care) (ClinicalTrials.gov ID-no: NCT00237548) [44], which is a population-based, high-risk screening and intervention study for type 2 diabetes in general practice (n = 8,664); 3) a populationbased group of unrelated middle-aged individuals (n = 680) examined at Steno Diabetes Center; and 4) unrelated type 2 diabetic patients (n = 2,158) sampled through the out-patient clinic at Steno Diabetes Center. In BMI stratified case-control analyses individuals from study group 2 with BMI,25 kg/m 2 were excluded. Hence, in the combined study sample 3,512 were normal weight, 7,458 were overweight, 5,044 were obese, and 340 were morbidly obese. Study groups 1 and 3 underwent a standard 75 g oral glucose tolerance test. Association with type 2 diabetes was evaluated in the combined study sample of which 5,302 were glucose-tolerant and 3,778 were type 2 diabetes patients. Definitions of overweight, obesity, morbid obesity and type 2 diabetes were according to WHO criteria, i.e. 25-29. Of note, a total of 5,586 individuals from study group 1 and 5,450 individuals from study group 2 were included in the GWAS performed by Thorleifsson et al. [19] (Note S1).
All study participants were Danes by self report, and informed written consent was obtained from all individuals before participation. The studies were approved by the regional Ethical Committees (The Scientific Ethics Committee of the Capital Region of Denmark for study group 1, 3 and 4 and The Scientific Ethics Committee of the Central region of Denmark for study group 2) and were in accordance with the principles of the Helsinki Declaration. More details of the study groups are given in Table S1.

Biochemical and anthropometric measures
In all four study groups weight and height were measured in light indoor clothes and without shoes. Waist circumference (cm) was measured in standing position midway between the iliac crest and the lower costal margin. For evaluation of quantitative traits all analysis are performed in the Inter99 cohort as this cohort represents the general middle-aged Danish population and extensive phenotypic characterisations are available in this cohort. Blood samples were drawn after a 12-hour overnight fast. Plasma glucose was analysed by glucose oxidase method (Granutest; Merck, Darmstadt, Germany) and serum insulin (excluding des (31,32) and intact proinsulin) was measured using the AutoDELFIA insulin kit (Perkin-Elmer, Wallac, Turku, Finland). Serum triglyceride, total cholesterol and HDL-cholesterol were determined using enzymatic colorimetric methods (GPO-PAP and CHOD-PAP; Roche Molecular Biochemicals, Mannheim, Germany). LDL-cholesterol was calculated as: ((total cholesterol -HDL-cholesterol -triglyceride/2.

Genotyping
SNPs for genotyping were selected as described in Note S2. The 10 variants in TMEM18, SH2B1, KCTD15, NEGR1, SEC16B, SFRS10, BDNF, FAIM2 and BAT2 were genotyped in study group 1 by deCODE genetics using the Centaurus platform [19]. For study group 2-4 these 10 variants were genotyped by KBioscience using the KasPARH SNP Genotyping method. The remaining 5 variants in GNPDA2, MTCH2, NPC1, MAF and PTER were genotyped by KBioscience also using the KasPARH SNP Genotyping method in all four study groups. When adjusting for the multiple tests performed, all SNPs obeyed Hardy Weinberg equilibrium (p.0.003). All 15 SNPs passed quality control with an average mismatch rate of 0.17% (max. 0.97%) and an average success rate of 97.8% (min. 96.1%).

Statistical analyses
Case-control analyses were performed using logistic regression. Type 2 diabetes studies included the full study sample, whereas BMI stratified case-control analyses excluded individuals from study group 2 with BMI,25 kg/m 2 as controls , since this is a population of high-risk individuals. General linear models were used to test quantitative metabolic traits for differences between genotype groups in 6,039 treatment-naïve individuals from the population-based Inter99 cohort. All analyses were performed assuming an additive genetic model and with adjustments for age and sex. Additionally adjustments for BMI were introduced in case-control studies of type 2 diabetes and in quantitative trait analyses.
Quantitative traits that did not follow a normal distribution were logarithmically transformed.
Statistical power in replication case-control settings was determined using the CaTS power calculator version 0.0.2. The lowest and the highest risk allele frequencies (RAF) of the examined SNPs were 8% and 84%, respectively, and the mean RAF was 50%. Using the population-based Inter99 cohort as reference, the prevalence of overweight, obesity, morbid obesity and type 2 diabetes in the Danish population was estimated to 39%, 17%, 1.3% and 8%, respectively. Estimated statistical power calculations in case-control settings are presented in Table S2.
Statistical power for the quantitative traits was estimated using simulations (n = 5,000), where variance across genotypes was drawn from phenotypes simulated to follow normal distribution using empirical variances. The variance for adjustment factors, estimated using residuals of linear models, was also included in the model, assuming independency of genotypes. Linear models were used both for simulating and testing data, assuming additive models and using a significance threshold of 0.05. The estimated statistical power for variants with a RAF of 8%, 84% and 50% respectively, were determined and are presented in Table S3. All analyses were performed in RGui version 2.8.0. Due to the large amount of tests performed in the present study, we used Bonferroni correction for multiple testing. For the primary hypothesis traits this resulted in a significance level of p,0.0004, whereas it for follow-up traits resulted in a significance level of p,0.0001.