Severe obesity is highly familial, often begins early in life, and is associated with greatly increased morbidity and mortality (1, 2). Because genetic influences are thought to play a greater role in severe obesity than more moderate obesity, identification of obesity genes may be more easily detected in this group. The glutamate decarboxylase 2 (GAD2)1 gene encoding the glutamic acid decarboxylase enzyme regulates γ-aminobutyric acid, a central nervous system neurotransmitter (3). Despite an early suggestion that GAD2 might be associated with type 1 diabetes (4), a subsequent study with much greater power did not find evidence that GAD2 was involved with diabetes (5), and a small study showed no association with GAD2 antibody levels (6).
GAD2 is located on chromosome 10 (7) in a region that has been linked to obesity in family studies (8, 9, 10). Three polymorphisms in this gene have been associated with severe obesity in a large French cohort of 575 severely obese subjects, although the polymorphisms did not explain a large proportion of the strong linkage signal found in the family study (11). A case-control analysis in that study replicated the family-based association. One of the polymorphisms tested, A-243G, may have functional consequences on promoter expression (11) and on lower birth weight, lower insulinogenic index, and greater obesity in children (12). Another polymorphism, T83897A, was associated with insulin levels and secretion. These polymorphisms were also related to higher hunger and disinhibition scores (11). Using a large cohort of 855 severely obese subjects in Utah and two control groups (Table 1), the three polymorphisms in the GAD2 gene were tested for association to see whether these findings could be replicated.
|Variable||Morbidly obese||Random normal weight||Volunteer normal weight|
|Age (years)||43.7 ± 11.4||51.7 ± 7.7||50.6 ± 9.7|
|BMI (kg/m2)||48.3 ± 6.8||22.1 ± 1.8||22.4 ± 1.7|
|Glucose (mg/dL)||103.1 ± 25.9||85.6 ± 9.3||83.5 ± 6.7|
|Insulin (μU/mL)||18.1 ± 15.0||9.3 ± 5.8|
|HOMA-IR||4.6 ± 4.0||2.0 ± 1.4|
|HOMA-B%||216 ± 242||165 ± 139|
Genotypes were first obtained for single nucleotide polymorphisms (SNPs) A-243G, C61450A, and T83897A in a series of 462 subjects who were randomly sampled from the Utah population. Mean age of the total random sample was 54 ± 7, and mean BMI was 28 ± 6 kg/m2. Allele frequencies of the A allele in the Utah population were estimated as 30.5% for C61450A, 19.3% for T83897A, and 81.2% for A-243G. After deleting the random subjects with a fasting glucose ≥ 126 mg/dL, there were no significant differences in the frequencies of the three genotypes between the normal-weight (BMI < 25 kg/m2; N = 130) and severely obese (BMI ≥ 35 kg/m2; N = 64) randomly ascertained subjects (p = 0.59, p = 0.33, and p = 0.49) or between the obese (BMI ≥ 30 kg/m2; N = 302) and non-obese (BMI < 30 kg/m2; N = 295) subjects (p = 0.69, p = 0.48, and p = 0.78) for SNPs A-243G, C61450A, and T83897A, respectively. There were too few (N = 20) severely obese subjects in the random group for comparison with the normal-weight group.
The genotype and allele frequency distributions of the subset of 130 normoglycemic, normal-weight subjects were compared with 855 severely obese subjects (BMI ≥ 40 kg/m2) examined before gastric bypass surgery (13). There were no significant differences for any of the three SNPs (Table 2). The associations were reexamined using logistic regression to control for age and sex differences between groups. None of the individual genotype or minor allele odds ratios (ORs) was significant for frequency differences between the severely obese and normal-weight groups for any SNP (Table 3). Adjustments of the p values for multiple comparisons were not made because this study was trying to replicate previous findings; however, multiple comparison adjustment of the p values would make all results even less significant. Additional logistic models were fit assuming that the G allele of A-243G was either dominant or recessive. Neither model [dominant OR = 1.30 (0.87, 1.93); recessive OR = 0.35 (0.08, 1.54)] showed significant associations of the SNP with severe obesity. Mean levels of glucose, insulin, homeostasis model assessment (HOMA)-insulin resistance (IR), and HOMA-beta cell function percentage (B%) in the normal-weight, normoglycemic controls did not differ among genotypes for any of the three SNPs (Table 4).
|SNP||Common homozygotes||Heterozygotes||Minor allele homozygotes||Minor allele||p|
|SNP||Heterozygotes||Minor allele homozygotes||Minor allele|
|C61450A||0.68 (0.45, 1.02)||1.11 (0.52, 2.33)||0.88 (0.66, 1.18)|
|T83897A||0.72 (0.48, 1.08)||1.66 (0.46, 5.97)||0.87 (0.62, 1.22)|
|A-243G||0.70 (0.47, 1.05)||2.55 (0.58, 11.28)||0.90 (0.64, 1.27)|
|SNP||Common homozygotes||Heterozygotes||Minor allele homozygotes||p|
|C61450A||86.6 ± 1.4||86.7 ± 1.2||85.3 ± 3.0||0.90|
|T83897A||86.8 ± 1.2||86.7 ± 1.4||81.4 ± 5.4||0.63|
|A-243G||87.0 ± 1.2||86.0 ± 1.4||85.5 ± 6.5||0.83|
|C61450A||10.5 ± 0.9||8.7 ± 0.8||6.6 ± 1.8||0.08|
|T83897A||10.0 ± 0.7||8.2 ± 0.9||8.1 ± 3.4||0.24|
|A-243G||10.0 ± 0.7||8.0 ± 0.9||9.6 ± 4.1||0.18|
|C61450A||2.3 ± 0.2||1.9 ± 0.2||1.4 ± 0.5||0.11|
|T83897A||2.2 ± 0.2||1.8 ± 0.2||1.6 ± 0.8||0.33|
|A-243G||2.2 ± 0.2||1.7 ± 0.2||2.0 ± 1.0||0.25|
|C61450A||187.3 ± 21.1||140.8 ± 18.3||105.9 ± 44.2||0.10|
|T83897A||173.4 ± 18.1||129.6 ± 20.8||152.4 ± 81.1||0.25|
|A-243G||171.5 ± 18.1||131.3 ± 20.8||162.7 ± 98.2||0.32|
Because the number of normal-weight controls was smaller than the number of severely obese cases and could have less precise frequency estimates and because the previously observed allelic associations were reversed in these data, we genotyped a volunteer sample of 134 healthy, normal-weight, normal-glucose subjects. There were no significant differences in frequencies between this volunteer group and the severely obese group. Combining the two control groups yielded ORs for the A-243G SNP of 0.73 (0.54, 1.00) and 0.98 (0.44, 2.18) for AG vs. AA and GG vs. AA, respectively.
Haplotypes were estimated from A-243G, C61450A, and T83897A for each subject and tested for differences between the severely obese and normal-weight subjects by logistic regression. There were three common haplotypes: ACT (68%), GAA (16%), and AAT (11%), with the remaining haplotypes being combined into a single haplotype (5%). Severely obese subjects had an ACT frequency of 68.0% vs. 66.5% for the normal-weight subjects compared with 65.3% and 71.2% in the French study, respectively. ORs for GAA (the three minor alleles) and AAT vs. ACT (the three major alleles) were 0.78 (0.55, 1.12) and 0.90 (0.59, 1.37), respectively, with the overall test for differences among the haplotypes being p = 0.10.
This study had over 95% power to detect the significant mean differences found in the French study for insulin and HOMA-B. Power to detect small differences in proportions is usually low and requires extremely large sample sizes. Therefore, neither the French study nor ours had 80% power to detect a difference in frequencies of only 4% (17.3% vs. 21.3% for A-243G). Our study had only 30% power to detect a significant effect at α = 0.05 and Nh = 407 (harmonic mean of 855 and 264), whereas the French study had ∼40% because their control group was larger (Nh = 592) (14). The allele frequencies for the minor allele of the three SNPs and comparisons with the other study groups in Utah and France are shown in Table 5. Our point estimates in the random controls were all in the opposite direction, and unless the estimates change with increasing sample size, they provide no evidence for replication. Table 3 suggested that the heterozygote risk for all three SNPs was not intermediate between the two homozygote risks, implying non-linearity even if the risks were significant. Consideration of dominant or recessive models did not improve the significance levels.
|Normal weight: random||32.7||20.4||19.2|
|Normal weight: volunteer||28.4||20.5||21.6|
A second test of the primary hypothesis was available using only the randomly ascertained subjects from the Utah population. The obese subjects in this series did not have significantly different allele frequencies from the non-obese random subjects, and the normal-weight subjects did not differ from the severely obese random subjects. Finally, in a second series of normal-weight subjects, despite its being a volunteer sample, fairly consistent allele frequency estimates with both the severely obese and random normal-weight subjects were seen. The allele frequency estimates for C61450A from the two control groups bracketed the estimates of the severely obese group, suggesting that larger sample sizes would not produce consistent differences between both control groups and the cases. Misclassification of the controls is unlikely because the random controls were 8 years older than the cases. It is rare for persons age 52 and normal weight to later become severely obese.
Despite having almost 50% more severely obese subjects than the French study, we were not able to confirm an association between GAD2 SNPs and severe obesity or IR or sensitivity. The study by Boutin et al (11) found association in their first series of severely obese subjects but could not confirm it in a second series, although combination of both sets retained the significance found in the first set. The normal variation of estimates among the various samples seen in these studies suggests a cautious interpretation of the reported association of the GAD2 gene with severe obesity (11) in all populations. However, the molecular, biochemical, and behavioral findings in the French subjects are consistent with a real effect in that population. Even for genes that have many positive associations with disease, replication is not always found. Reasons for lack of replication of the GAD2 gene between Utah and French subjects may include differences in diet and lifestyle. The prevalence of obesity in the U.S. is higher than in France, suggesting such population differences (15, 16). Future studies of association of these SNPs with biochemical pathways and abnormalities of intermediate phenotypes of obesity may more rigorously prove or disprove the suggested statistical association.