We genotyped 15 SNPs (average inter-marker interval of 20,530 bp, range 6,598–39,849 bp) extending 287,400 bp in the GABRG1-GABRA2
region of chromosome 4 in a sample of 1,634 self-identified EA and 248 AA subjects, , comprising AD and control subjects from Connecticut (CT) and a sample of AD subjects from Project MATCH, an NIAAA-sponsored, multi-center clinical trial conducted at 10 sites throughout the US. That study sought to identify predictors of response to three psychotherapeutic treatments for alcoholism (Project MATCH Research Group. 1998
). lists the primers, MGB-probes and annealing-extension temperatures used to examine the 15 GABRG1-GABRA2
region SNPs. The trivial names for the SNPs examined in GABRA2
(SNPs A-H) are the same as were used previously (Covault et al. 2004
); the additional 7 SNPs in the intergenic and 5′-region of GABRG1
are numbered 1–7.
GABRA2 SNPs examined in samples of alcohol-dependent and control subjects
Two main haplotype blocks were observed in EA controls (). One was a 94 kb region that included SNP markers 2–6 and extended from GABRG1
intron 2 to 62,000 bp 5′ of the GABRG1
transcript start site. The second haplotype block extended 137 kb and included SNP marker 7 (50,000 bp 3′ from the 3′UTR of the GABRA2
transcript) and the 7 adjacent SNP markers A-G (ending in GABRA2
intron 3), which we had previously identified as being in high LD with one another (Covault et al. 2004
). The LD plot and haplotype block structure for AA subjects was similar to EA subjects regarding the presence of a region of increased recombination in the 14 kb interval between SNPs 6 and 7 (). Among AAs, SNPs 3–6 were in high LD with one another in the GABRG1
block and in the GABRA2
haplotype block region SNPs 7 and A-E showed high LD. The LD plot and haplotype block structure for AD subjects paralleled those for the controls of their respective racial group (data not shown).
Figure 1 LD plot from Haploview 3.32 for EA control subjects (Panel A) compared with that for AA control subjects (Panel B). Pair-wise SNP |D′| values (x100) of linkage are shown together with haplotype blocks identified using the four-gamete rule (Wang (more ...)
GABRG1 allele and genotype frequencies for the three EA groups and the two AA groups are shown in . Among EAs, all SNPs in the GABRG1 haplotype block (SNPs 2–6) showed significant association (nominal p<0.05) with AD for both allele and genotype frequencies in both the CT and Project MATCH samples. The strongest association (p=0.001) was observed for SNPs 4 and 5 in the 5′-upstream region of the gene. GABRA2 allele and genotype frequencies (SNP7 and SNPs A-H) are shown in . For the GABRA2 haplotype block region, the 3 SNPs closest to GABRG1 (i.e., SNPs 7, A, and B) showed modest allelic and genotypic association with AD in the CT EA sample (p=0.009–0.016 for allelic association), while SNPs D-G showed less evidence of association (p=0.021–0.032). In parallel with the CT sample, MATCH EA alcoholics showed a non-significantly higher prevalence than controls of the minor allele for each SNP in the 3′ GABRA2 haplotype block (p=0.083 – 0.167). None of the 15 markers showed allelic association with AD in the AA sample. Allele frequencies in the controls differed significantly by racial population for 13 of the 15 SNPs ( and , right-hand column).
GABRG1 SNP allele and genotype frequencies for self-identified EA and AA subjects.
GABRA2 SNPs allele and genotype frequencies for self-identified EA and AA subjects.
Haplotype frequencies for the two major haplotype blocks in the GABRG1-GABRA2 interval were estimated using markers defining the slightly shorter core of each block based on the haplotype structure for AAs (). Four markers (SNPs 3–6) were used to define the GABRG1 haplotype block and six markers (SNPs 7 and A-E) for the GABRA2 haplotype block. Estimated haplotype frequencies were compared to the sum of all other haplotypes for alcoholics vs. controls and for AA vs. EA controls using a series of 2X2 contingency tables ( and ). Among EAs, two haplotypes comprised >90% of chromosomes for the GABRG1 haplotype block and >95% of chromosomes for GABRA2. AA subjects had a third common haplotype for both haplotype blocks. Haplotype frequencies differed significantly between EA and AA controls. For GABRG1, there was a significantly greater frequency of the ATCC haplotype in AD than controls in both EA groups (Δ=0.078 in CT EA cases, p<0.001; 0.064 in MATCH EA cases, p=0.002), but this did not reach significance in AA cases (Δ=0.059). For the 6-SNP GABRA2 3′-region haplotype, there was a non-significantly greater frequency of the AGACTC minor haplotype in EA alcoholics than controls (Δ=0.046 in CT cases, p=0.052 and 0.028 in MATCH cases, p=0.20). The frequency of the GABRA2 AGACTC minor haplotype did not differ by diagnosis among AAs.
GABRG1 4-SNP haplotype (SNPs 3–6; rs1391166, 7654165, 10033451, 4280776) frequencies for self-identified EA and AA subjects.
GABRA2 6-SNP haplotype (snps 7, A-E; rs1440133, 567926, 534459, 529826, 279869, 279858) frequencies for self-identified EA and AA subjects.
There was a moderate degree of LD between markers in the two haplotype blocks (r = 0.51–0.59 for EA subjects and r=0.00–0.53 for AA subjects). To examine whether variation in one or the other of the two adjacent genes accounted for the association to AD, we examined the phase and pattern of linkage of the adjacent GABRG1
haplotype blocks by estimating the frequency of extended haplotypes defined by the 10 SNPs from these two blocks. lists the frequency of the most common 10-SNP haplotypes covering the larger 208,000 bp interval. The most common extended haplotype among EAs (TCTT-GAGTGT) was under represented in both of the EA alcoholic samples (Δ=0.098 in CT cases and Δ=0.059 in MATCH cases; p<0.001 and 0.002, respectively). Chromosomes with the GABRG1
risk haplotype (ATCC) were over-represented in EA alcoholics, irrespective of the GABRA2
haplotype in both EA samples. The distortion was greatest when the less common GABRG1
risk haplotype was paired with the major GABRA2
haplotype (GAGTGT) (Δ=0.042 in CT cases, and Δ=0.044 in MATCH cases; p=0.04 and p=0.005, respectively). In contrast, the GABRA2
haplotype defined by the minor allele at each marker (AGACTC), which we previously found to be associated with AD (Covault et al. 2004
), when paired with the GABRG1
non-risk haplotype (TCTT), did not differ in frequency between EA alcoholics and controls in the present analysis. A more formal restatement of this analysis is to examine the ratio in cases vs. controls of the two putative risk haplotypes with each conditional on the presence of the other. Both haplotype ratios, f(AB)/f(Ab) and f(aB)/f(ab), are expected not to differ in cases and controls if the “A vs. a” component of the haplotype identifies a disease-associated marker and the “B vs. b” site represents a neutral marker, irrespective of genetic mode of inheritance (Valdes&Thomson 1997
). The χ2
test of the null hypothesis that these ratios are equivalent is rejected (p < .01) in EA AD samples, if the GABRA2
AD-associated haplotype is considered the risk marker and the GABRG1
haplotype is considered the neutral marker. In this context, markers in the GABRA2
haplotype block do not capture all of the disease risk associated with this chromosomal region. In contrast, the null hypothesis that the ratios are equivalent is sustained (p > 0.2) when the GABRG1
AD-associated haplotype block is treated as the conditional risk marker. Equivalent results were obtained examining 2-SNP haplotypes using a single representative SNP for the GABRG1
blocks (e.g., SNPs 4 and A). These observations suggest that the allelic and haplotypic association of GABRA2
SNPs with AD in our CT sample in both this and our prior study (Covault et al. 2004
), may in part be secondary to LD of these markers with risk-related variants in the adjacent GABRG1
5′-region. In AAs, although not statistically significant, the extended 10-SNP haplotype pattern was qualitatively similar to the pattern in EAs, in that there was a greater frequency of chromosomes with an extended haplotype containing the GABRG1
GABRG1-A2 10-SNP extended haplotype (SNPs 3–7 & A-E; rs1391166, 7654165, 10033451, 4280776, 1440133, 567926, 534459, 529826, 279869, 279858) frequencies for self-identified EA and AA subjects.
Significant racial population differences were observed in the allele frequencies for 13 of the 15 SNPs, as well as for the most common haplotypes for the GABRG1
blocks (–). To evaluate population genetic stratification, we used the program STRUCTURE v2.1 (Pritchard et al. 2000
; Falush et al. 2003
) to generate estimates of the proportion of African vs. European genetic ancestry using genotype results from a panel of 34 ancestry informative markers (Stein et al. 2004
; Luo et al. 2005
; Yang et al. 2005
). We found no differences in the degree of EA genetic ancestry in the three EA samples [controls = 0.980 ± 0.043, CT alcoholics = 0.980 ± 0.043, and MATCH alcoholics = 0.983 ± 0.029; F(2,1631)=1.56, p=0.21]. Consequently, population stratification is unlikely to be an explanation for the observed allele frequency differences between EA alcoholics and controls. Similarly, for AAs, there was no significant difference in estimated European genetic admixture for alcoholics and controls [0.084 ± 0.149 and 0.054 ± 0.143, respectively; F(1,246)=1.61, p=0.11].
Quantitative estimates of genetic ancestry proportion from STRUCTURE were also used as a covariate in binary logistic regression analysis (together with age and sex, which differed by diagnosis). This yielded corrected odds ratios (ORs) for AD as a function of GABRG1 or GABRA2 AD-associated haplotype copy number as determined using PHASE. Dominant, recessive, and additive models were compared using dummy coding: 0, 1, 1; 0, 0, 1 and 0, 0.5, 1, respectively, for 0, 1 or 2 copies of the GABRG1 or GABRA2 AD-associated haplotype (i.e., ATCC or AGACTC, respectively). For EAs (combined MATCH and CT alcoholics), an additive genetic model provided the best fit for the GABRG1 ATCC haplotype (), with an OR=1.26 (95% CI=1.06–1.50) for one copy (i.e. ATCC/x) and OR=1.60 (95% CI=1.13–2.25) for two copies of the AD-associated haplotype (i.e. ATCC/ATCC). Qualitatively similar results were seen when age, sex and genetic ancestry were omitted from the model. For the GABRA2 AGACTC haplotype, a dominant genetic model provided the best fit and an OR=1.39 (95% CI=1.08–1.80) for carriers. These models did not show interactive effects of haplotype with either gender or genetic ancestry proportion. Due to the co-linearity of the two risk haplotypes, we were unable simultaneously to examine the independent effects on AD risk of the GABRG1 and GABRA2 haplotype blocks. There was no evidence that the association of AD with GABRG1 or GABRA2 markers in the CT sample was due to co-morbid drug dependence (relevant data were not available for Project MATCH). Considering the impact of drug dependence, ORs for the AD risk haplotypes were marginally greater (GABRG1) or unchanged (GABRA2) for subjects in the CT EA sample with no co-morbid lifetime diagnosis of cocaine, opioid or cannabis dependence compared with all CT EA alcoholics [GABRG1 additive model OR=1.35 (95% CI=1.05–1.74) vs. 1.25 (95% CI=0.99–1.58) for one copy and OR=1.82 (95% CI=1.09–3.04) vs. 1.56 (95% CI=0.98–2.48) for two copies of the GABRG1 risk haplotype; GABRA2 dominant model OR=1.45 (95% CI=0.99–2.14) vs. 1.40 (95% CI=0.98–1.99) for carriers].
Binary logistic regression analysis of GABRG1 and GABRA2 AD-risk haplotypes as a function of the genetic model (including age, sex and % EA genetic heritage as covariates).
Finally, based on the apparent difference in genetic mode of action of risk elements potentially represented in the two haplotype blocks, we used binary logistic regression analysis to derive corrected measures for association of individual SNPs with AD in the combined EA sample using variable coding for both additive and dominant genetic models. illustrates the log10 (p-value) from binary logistic regression analyses in which age, sex, and genetic ancestry were used as covariates. The strongest association with AD was seen for SNPs 4 and 5 (rs7654165 and rs10033451) in the 5′-upstream region of GABRG1 assuming an additive genetic model followed by SNPs 7, A, and B (rs1440133, rs567926, and rs534459) in the 3′ region of GABRA2 assuming a dominant genetic model. This difference in the best-fit genetic model for markers in the two haplotype block regions suggests that there may be separate contributions to risk for AD by GABRG1 and GABRA2.
Figure 2 Negative log10 of binary logistic regression significance (p-value) for allelic association with AD after correction for age, sex, and proportion of EA genetic ancestry among all EA subjects (1099 AD and 535 controls) as a function of GABRG1 and GABRA2 (more ...)