A genome-wide association study of alcohol dependence

Abstract

Excessive alcohol consumption is one of the leading causes of preventable death in the United States. Approximately 14% of those who use alcohol meet criteria during their lifetime for alcohol dependence, which is characterized by tolerance, withdrawal, inability to stop drinking, and continued drinking despite serious psychological or physiological problems. We explored genetic influences on alcohol dependence among 1,897 European-American and African-American subjects with alcohol dependence compared with 1,932 unrelated, alcohol-exposed, nondependent controls. Constitutional DNA of each subject was genotyped using the Illumina 1M beadchip. Fifteen SNPs yielded P < 10⁻⁵, but in two independent replication series, no SNP passed a replication threshold of P < 0.05. Candidate gene GABRA2, which encodes the GABA receptor α2 subunit, was evaluated independently. Five SNPs at GABRA2 yielded nominal (uncorrected) P < 0.05, with odds ratios between 1.11 and 1.16. Further dissection of the alcoholism phenotype, to disentangle the influence of comorbid substance-use disorders, will be a next step in identifying genetic variants associated with alcohol dependence.

Excessive alcohol use is the third leading cause of preventable death in the United States (1). Although normative alcohol use is ubiquitous, alcohol dependence is a serious medical illness (2) that is experienced by ≈14% of alcohol users (3). Alcohol dependence constitutes a substantial health and economic burden, costing an estimated $184 billion in expenditures stemming from alcohol-related morbidity, accidents, lost productivity, and incarceration (4). These challenges underscore the importance of clarifying the etiology of alcohol dependence as a key public health priority.

Liability to alcohol dependence has both genetic and environmental influences, which act independently and in concert. First-degree relatives of affected individuals are at a 2- to 8-fold increased risk for alcohol dependence (5, 6). Adoption studies and twin studies have clarified that this familial clustering of alcohol dependence is attributable largely to genetic factors (7–11). In most recent studies, these heritable influences explain ≈50–80% of the individual differences in liability to alcohol dependence (12, 13).

In an effort to unmask specific genomic influences on alcohol dependence, scientists have brought a vast genomic toolkit to bear on this problem. Large-scale genome-wide association studies (GWAS) offer considerable promise. By genotyping a dense set of SNPs throughout the genome, investigators have the potential to identify with considerable precision genes that may lead to unknown biological pathways involved in alcohol dependence.

Candidate gene strategies frequently have identified significant associations between SNPs in the gene encoding the alpha2 subunit of the γ-aminobutyric acid A receptor (GABRA2), a major inhibitory neurotransmitter in the human nervous system that is involved in the behavioral effects of alcohol (14). Although some exceptions exist (15–17), there are multiple positive reports of association between SNPs in GABRA2 and alcohol- and other substance-related phenotypes (14, 18–22 and reviewed in 23).

We report on a large, well-characterized sample of 1,897 Diagnostic and Statistical Manual of Mental Disorders, edition 4 (DSM-IV) alcohol-dependent cases and 1,932 alcohol-exposed, nondependent controls from the Study of Addiction: Genetics and Environment (SAGE) analyzed at 948,658 SNPs that span the genome. First, using a hypothesis-free, genome-wide association strategy, we nominate SNPs associated with vulnerability to alcohol dependence. Second, we specifically examine the important role of SNPs in GABRA2 using a targeted, hypothesis-driven approach.

Results

The characteristics of the study participants are listed in Table 1, and further details are provided in Table S1. Based on self-report of race, the sample is 69% European descent and 31% African-American. A small number of subjects (3%) reported Hispanic ethnicity. Over 60% of alcohol-dependent cases are male. Comorbid drug dependence is common, with almost half of the alcohol-dependent sample diagnosed with comorbid marijuana or cocaine dependence.

Table 1.

Characteristics of alcohol-dependent cases and nondependent controls

Characteristic	Cases n = 1,897	Controls n = 1,932	Total n = 3,829
Sex, N (%)
Males	1,155 (60.9)	606 (31.4)*	1,761 (46.0)
Females	742 (39.1)	1,326 (68.6)	2,068 (54.0)
Age, years
Mean ± SD	39.0 ± 9.3	39.3 ± 9.1	39.2 ± 9.2
Range	18.0–77.0	18.0–65.0	18.0–77.0
Self-reported race, n (%)
European-American	1,235 (65.1)	1433 (74.2)*	2,668 (69.5)
African-American	662 (34.9)	499 (25.8)	1,161 (30.3)
Self-reported ethnicity, n (%)
Hispanic	76 (4.0)	56 (2.8)	132 (3.4)
Alcohol dependence
Diagnosis, n (%)	1,897 (100.0)	0 (0.0)*	1,897 (49.5)
Number of symptoms, ± s	5.2 ± 1.5	0.6 ± 0.9*	2.9 ± 2.6
Comorbid diagnoses, n (%)
Marijuana dependence	663 (34.9)	0 (0.0)*	663 (17.3)
Cocaine dependence	916 (48.2)	0 (0.0)*	916 (23.9)
Opioid dependence	263 (13.8)	0 (0.0)*	263 (6.8)
Other dependence	469 (24.7)	0 (0.0)*	469 (12.2)
Smoked 100+ cigarettes	1,707 (89.9)	1,139 (53.4)*	2,846 (74.3)
Nicotine dependence†	1,159 (61.0)	95 (4.9)*	1,254 (32.7)

*Difference between cases and controls, P < 0.0001.

^†Nicotine dependence defined by a score of 4 or greater on the Fagerström Test for Nicotine Dependence.

Primary genome-wide association analyses identified 15 SNPs with P < 10⁻⁵ (Table 2 and Figs. S1 and S2). Secondary analyses stratified by race demonstrated similar odds ratios (OR) in populations of European and African descent, although the allele frequencies are markedly different across the two groups (Table S2).

Table 2.

Genetic association testing for alcohol dependence: Adjusted odds ratios and confidence intervals for SNPs with P ≤ 1 × 10⁻⁶

					Frequency of test allele				Overall SAGE Sample (n = 3,829)
SNP	Chr	Position (hg18)	Locus and context	Test allele	EA cases	EA controls	AA cases	AA controls	Adjusted OR (95% CI)	P	FDR
rs10893366	11	124,683,613	PKNOX2, intronic	T	0.19	0.15	0.31	0.26	1.39 (1.23–1.58)	1.93E-07	0.178
rs2039617	10	97,749,193	CC2D2B, intronic	T	0.06	0.10	0.33	0.40	0.69 (0.60–0.80)	5.95E-07	0.274
rs9302534	16	17,956,211	Intergenic	C	0.35	0.39	0.73	0.76	0.78 (0.70–0.86)	2.73E-06	0.582
rs1318937	3	15,270,368	SH3BP5, intronic	G	0.15	0.11	0.51	0.45	1.35 (1.19–1.52)	3.54E-06	0.582
rs2700648	3	100,590,373	Intergenic	A	0.29	0.24	0.45	0.40	1.29 (1.16–1.44)	3.99E-06	0.582
rs10803574	2	139,427,333	Intergenic	A	0.45	0.40	0.17	0.14	1.28 (1.15–1.43)	4.41E-06	0.582
rs6483362	11	88,052,099	GRM5, intronic	A	0.09	0.07	0.30	0.24	1.42 (1.22–1.66)	4.53E-06	0.582
rs2722650	19	49,593,366	ZNF285A, intronic	A	0.14	0.17	0.57	0.63	0.76 (0.67–0.86)	7.14E-06	0.582
rs10893365	11	124,681,647	PKNOX2, intronic	C	0.19	0.15	0.44	0.41	1.31 (1.17–1.48)	7.20E-06	0.582
rs1386449	11	19,444,659	Intergenic	T	0.00	0.00	0.14	0.07	1.98 (1.47–2.67)	7.29E-06	0.582
rs750338	11	124,677,803	PKNOX2, intronic	C	0.26	0.20	0.47	0.45	1.28 (1.15–1.43)	7.61E-06	0.582
rs1505846	2	139,494,752	Intergenic	A	0.45	0.40	0.17	0.14	1.28 (1.15–1.42)	8.01E-06	0.582
rs9636231	2	139,388,425	Intergenic	A	0.32	0.26	0.14	0.12	1.30 (1.16–1.46)	9.13E-06	0.582
rs1363605	5	172,949,693	Intergenic	A	0.15	0.17	0.36	0.44	0.76 (0.68–0.86)	9.62E-06	0.582
rs10224675	7	143,923,695	TPK1, intronic	G	0.00	0.00	0.06	0.10	0.46 (0.33–0.65)	9.75E-06	0.582

AA, African-American; Chr, chromosome; EA, European-American; FDR, false discovery rate.

Replication Studies.

The top associated SNPs were tested for replication in two independent datasets (Table S3). The first replication sample is the family-based study from the Collaborative Study on the Genetics of Alcoholism (COGA). In our family-based association analyses, none of the SNPs demonstrated association with a P value < 0.05; however, rs1386449 and rs10224675, which in our primary analyses are associated with alcohol dependence only in African-Americans, have a P value < 0.10 in the family-based analysis with a small number of African-American families. The top results also were examined in alcohol-dependent case and community-based comparison subjects of German ancestry (24). Of the seven SNPs that were genotyped and tested in the sample, none reached a significance level of P < 0.05.

We also show association results (Table S4) from our study for the SNPs recently reported in the independent GWAS of alcohol-dependent men by Treutlein and colleagues (24). Only one SNP, rs13160562, shows modest evidence of replication [0.88; 95% confidence interval (CI) 0.78–0.99, P = 0.03]. In a meta-analysis, this SNP did not reach genome-wide significance (meta-analysis OR = 0.83, 95% CI 0.77–0.90, P = 2.74 × 10⁻⁶). None of the other SNPs reported by Treutlein and colleagues were associated with alcohol dependence (P > 0.05).

Candidate Gene Findings.

The findings for SNPs genotyped in GABRA2 that overlap with SNPs reported by Edenberg et al. (14) are displayed in Table 3. Results for all GABRA SNPs in the entire sample and stratified by self-reported race are shown in Table S5 and Table S6, respectively. This analysis confirms the modest association of alcohol dependence with variants in GABRA2.

Table 3.

SAGE association results for GABRA2 SNPs also genotyped in the family-based COGA sample

			SAGE
				Frequency of risk allele
SNP	Position	COGA*P	Risk allele	Cases	Controls	Adjusted odds ratio (95% CI)	P
rs572227	45,946,150	3.80E-02	A	0.376	0.366	1.15 (1.04–1.27)	8.89E-03
rs548583	45,958,101	1.20E-02	T	0.404	0.385	1.14 (1.03–1.26)	1.05E-02
rs279858	46,009,350	8.70E-03	G	0.375	0.366	1.16 (1.05–1.28)	5.04E-03
rs279843	46,019,961	4.90E-02	T	0.444	0.421	1.11 (1.00–1.22)	4.24E-02
rs279841	46,035,520	3.80E-02	A	0.368	0.364	1.13 (1.02–1.25)	2.29E-02

*COGA family based association from ref. 14.

Discussion

Our study of a relatively large sample of alcohol-dependent cases and nondependent controls used a two-pronged approach to investigate the genetics of alcohol dependence: a GWAS with testing of previously identified genome-wide significant results supplemented by a targeted gene study of GABRA2. Advantages of the genome-wide design include its hypothesis-free strategy and its suitability for the discovery of novel genetic contributors to disease. However, the genome-wide examination requires correction for multiple testing, and the threshold for significance of GWAS findings is high. In contrast, targeted gene studies test specific hypotheses to provide validation of previously reported findings and therefore require a much lower threshold for significance.

In the GWAS arm of the study, we identified 15 SNPs associated with alcohol dependence using a significance threshold of P < 10⁻⁵. In two independent samples, one a large family-based study of 258 families with more than 2,000 genotyped individuals and the second a study of alcohol-dependent men and community-based comparison subjects of German descent, none of the association findings replicated using a significance threshold of P < 0.05. Two of the top SNPs identified in SAGE are common in African-American populations and are rare (minor allele frequency < 1%) in subjects of European origin. These SNPs trended toward significance (P < 0.10) in our family-based association tests. Overall, no newly identified variants were associated with alcohol dependence at the genome-wide significance threshold; however, this interpretation of our results is conservative.

Our top 15 association signals were not replicated in the independent datasets, nor did the present study replicate the two genome-wide significant results reported by Treutlein and colleagues (24), who examined alcohol-dependent men in treatment and compared them with individuals from the community. Their top two findings, rs7590720 and rs1344694, were not associated with alcohol dependence in our study in our primary analysis or in secondary analyses stratified by ancestry and sex. In comparing our findings for the top 15 signals reported by Treutlein et al. (24), our results modestly support one finding, rs13160562 in the endoplasmic reticulum aminopeptidase 1 (ERAP1) gene (meta-analysis OR = 0.83, 95% CI 0.77–0.90, P = 2.74 × 10⁻⁶).

Different analytic strategies can highlight various findings as the top results. We decided a priori that our primary analysis would examine men and women of European and African origin together to maximize our sample size and power to detect robust association. Although a substantial number of variants across the genome differ in frequency between these populations, this strategy posits that the underlying biologic contributions to disease risk act in a similar manner across populations. Likewise, although the prevalence of a disorder such as alcohol dependence differs between men and women, this strategy assumes that at least part of the underlying biologic risk is the same. This analytic strategy by design highlights SNPs that have similar effects across both populations and genders. However, we recognize that this strategy can decrease power if there is genetic heterogeneity and variants act in one group but not another. Results of secondary analyses performed separately in subjects of European and African descent are presented in Tables S7 and S8. None of the top SNPs overlap in these different analyses.

In our hypothesis-based approach to test association, we examined the GABRA2 gene, the genetic finding most consistently associated with alcohol dependence in the literature. We did see modest evidence of association (with P values comparable to the original reports) in GABRA2, albeit with an OR ∼1.1 (P ∼0.01). This modest genetic risk is consistent with a model whereby multiple genes of small effect contribute to the vulnerability to alcohol dependence. Similar levels of genetic risk have been identified by meta-analyses of other complex diseases such as diabetes (25) and traits such as height (26).

Discrepancies in findings between our results and other samples may be attributed to these inherent differences in study design, sampling strategies, gender, and ethnicity. The design of our study introduces certain caveats into the interpretation of findings. Subjects were drawn from three studies that were individually ascertained for alcohol (COGA), nicotine (Collaborative Genetic Study of Nicotine Dependence, COGEND), and cocaine (Family Study of Cocaine Dependence, FSCD) dependence. Although alcohol-dependent cases and nondependent controls were uniformly screened and defined in this study, the potential introduction of genetic heterogeneity exists because of the three distinct ascertainment designs. In addition to the possibility that some of our top signals are false positives, the high levels of comorbid substance-use disorders may have increased the likelihood that we would identify association to genes contributing broadly to addiction, potentially limiting our ability to replicate these associations in samples ascertained solely for alcohol dependence. Given a substantial genetic epidemiological literature (27–29) supporting considerable overlap of genetic influences of alcohol dependence and substance-use disorders, it is unlikely that the excess comorbidity has biased the findings.

Finally, power is an important consideration in GWAS. This study included more than 3,800 subjects and had 80% power to detect a genetic variant with an OR of 1.3 and greater (for a risk locus with 30% minor allele frequency) at a genome-wide significance threshold of 5 × 10⁻⁸. A genetic effect of this magnitude can be considered strong, given that previous studies of GABRA2 report genetic risks in the 1.2 range. Significance levels are related to sample size and strength of the genetic effect. To increase the power to detect significant results, two strategies are possible: enlarge the sample or refine the phenotype to increase the detectable genetic effect. Increasing the sample size has been a common strategy to detect robust association results and has been used to identify association in diabetes (25) and schizophrenia (30). To this end, meta-analysis efforts are currently in progress. As noted by Zeggini and Ioannidis (31), single GWAS studies rarely have been successful at achieving genome-wide significance, and careful meta-analysis provides an avenue for systematically augmenting power to detect modest effects. However, the phenotypic precision often is reduced in these larger studies, and the potential for the introduction of genetic heterogeneity exists.

A second strategy is to narrow the phenotype and to analyze a more homogeneous sample. A GWAS analysis was performed on the COGA subset of European-American subjects (996 subjects overlap with this report) (32). This approach focuses the analysis on subjects recruited under a single ascertainment protocol for alcohol dependence, and the severity of illness in these subjects is high. No finding in this analysis reached a genome-wide statistically significant level of genetic risk, and different top SNPs were nominated in this approach.

Nonetheless, our results underscore the important contribution of GWAS by nominating genes that may play a role in the etiology of alcohol dependence. Continued efforts aimed at gene identification using complementary approaches and coupled with refinement of the phenotypes will be pivotal in illuminating the complex biological and environmental substrate in which alcohol dependence develops.

Methods

The Study of Addiction: Genetics and Environment (SAGE) is funded as part of the Gene Environment Association Studies (GENEVA) initiative supported by the National Human Genome Research Institute (dbGaP study accession phs000092.v1.p1). Alcohol-dependent cases and nondependent control subjects were selected from three large, complementary datasets, COGA, FSCD, and COGEND. Across all studies, case subjects were identified as having a lifetime history of alcohol dependence using DSM-IV criteria (2). Control subjects were required to report a history of drinking because alcohol use is required to develop alcohol dependence. Control subjects had no significant alcohol-dependence symptoms. Because of the likely genetic overlap between alcohol and drug dependence, a diagnosis of drug dependence was an exclusionary criterion for control subjects.

The Institutional Review Board at each contributing institution reviewed and approved the protocols for genetic studies under which all subjects were recruited. Subjects provided written informed consent for genetic studies and agreed to have their DNA and phenotypic information available to qualified investigators through National Institutes of Health repositories. Additional description of the studies is available at http://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs000092.v1.p1.

The datasets used for the analyses described in this paper can be obtained from the database of Genotypes and Phenotypes (dbGaP) at http://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs000092.v1.p1 through dbGaP accession number phs000092.v1.p1.

Collaborative Study on the Genetics of Alcoholism.

A case-control series of unrelated individuals was selected from more than 8,000 subjects who participated in the genetic arm of COGA. COGA systematically recruited families with multiple members affected with alcohol dependence and community-based comparison families from participating centers across the United States. COGA contributed 612 alcohol-dependent cases and 412 control subjects of European descent and 287 alcohol-dependent cases and 97 controls of African-American descent. Individuals in this case-control sample were independent from the COGA family linkage association sample that was genotyped previously (33, 34).

Family Study of Cocaine Dependence.

Cocaine-dependent subjects were recruited systematically from chemical dependency treatment units in the greater St. Louis, MO, metropolitan area (35). Community-based comparison subjects were identified through the Missouri Family Registry and matched by age, race, gender, and residential zip code. This study contributed 280 alcohol-dependent cases and 247 controls of European descent and 268 alcohol-dependent cases and 249 controls who self-identified as African-American. Because of the study design, alcohol-dependent case subjects also met criteria for cocaine dependence.

Collaborative Genetic Study of Nicotine Dependence.

COGEND was designed as a community-based study of nicotine dependence. Subjects were recruited from Detroit, MI, and St. Louis, MO. More than 53,000 subjects were screened by telephone, more than 2,800 were personally interviewed, and nearly 2,700 donated blood samples for genetic studies (36, 37). COGEND contributed 343 alcohol-dependent cases and 774 controls of European descent and 107 African-American alcohol-dependent cases and 153 African-American controls.

Source of DNA.

All subjects deposited a blood sample in the Rutgers University Cell and DNA Repository (RUCDR), a central biologic repository for the National Institute on Alcohol Abuse and Alcoholism and National Institute on Drug Abuse (http://www.rucdr.org). DNA was extracted from the blood sample, and cell lines were developed as an additional DNA source.

Assessment.

A common assessment was performed across all three studies and was based on the Semi-Structured Assessment for the Genetics of Alcoholism (SSAGA) (38). This shared methodology of interview administration, question format, and queried domains allowed harmonization of phenotypic data across all studies.

Genotyping and Data Cleaning.

Samples were genotyped at the Johns Hopkins Center for Inherited Disease Research (CIDR). Data were released for 4,189 study samples. Study samples, including 49 study duplicates, were plated and genotyped together with 135 HapMap controls (86 CEU; 49 YRI). Genotyping was performed using Illumina Human1Mv1_C BeadChips and the Illumina Infinium II assay protocol (39). Allele cluster definitions for each SNP were determined using Illumina BeadStudio Genotyping Module version 3.1.14 and the combined intensity data from the samples. Strict quality-control standards were implemented, and genotypes were released by CIDR for 1,040,106 SNPs (99.15% of attempted). The mean non-Y SNP call rate and mean sample call rate was 99.7% for the released CIDR dataset. Study duplicate reproducibility was 99.98%. Further extensive cleaning was undertaken to ensure high-quality genotyping by examining batch effects, potential chromosomal anomalies, and Mendelian errors. Further details are provided in the comprehensive data cleaning report posted at dbGaP http://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/document.cgi?study_id=phs000092.v1.p1&phv=22928&phd=2274&pha=&pht=116&phvf=&phdf=20&phaf=&phtf=&dssp=1&consent=&temp=1.

SNPs with an allele frequency > 1% in either the European- or African-descent populations were analyzed (948,658 SNPs). A SNP call rate of 98% was required. Hardy–Weinberg equilibrium (HWE) was tested, and SNPs that deviated from HWE (P < 10⁻⁴) were excluded. The final number of subjects included in analyses was 3,829. Individuals were dropped if there was potential sample misidentification, sample relatedness, or other misspecification (n = 171).

Statistical Analyses.

We used the software package EIGENSTRAT (40) with all SNPs to calculate principal components reflecting continuous variation in allele frequencies representing ancestral differences in subjects. Two principal components were identified; the first distinguished African-American participants from European-American participants and the second distinguished Hispanic and non-Hispanic subjects. Each individual received scores on each principal component. These scores, representing continuous variation in race and ethnicity, can be used to control for effects of population stratification.

Genome-wide association analysis was conducted using logistic regressions in PLINK (41). Genotypes were coded log-additively (0, 1, 2 copies of the minor allele). Covariates represented sex, age [defined, using quartiles, as 34 years and younger (reference), 35–39 years, 40–44 years, and 45 years and older] and two principal components indexing continuous variation in race/ethnicity. We repeated analyses using self-reported race (European-American, African-American) as categorical variables. Similar results were seen with both analyses, and we present results using self-reported race. The false-discovery rate was calculated using the method of Storey and Tibshirani (42).

Replication Samples.

Family sample from COGA.

A set of 258 genetically informative, multiplex alcohol-dependence pedigrees in COGA have been studied previously in linkage- and association-based analyses (33, 34). Families report European and African ancestry (n = 219 European American, n = 35 African American, and n = 4 other ancestry). These pedigrees do not overlap with any of the case-control subjects used in the SAGE GWAS sample. Affected individuals were defined as those meeting criteria for DSM-IV alcohol dependence, and unaffected subjects reported no symptoms of alcohol dependence. The SNPs with the most significant evidence of association from SAGE were selected for genotyping in this family-based sample. SNPs had a genotyping call rate > 99%. All SNPs passed an HWE threshold of P > 0.05 as calculated independently in African-American and European-American samples. Pedigree errors were cleaned using PEDCHECK (43). Family-based association analyses were performed using a Family-Based Association Test (FBAT) (44), with alcohol-dependent and nondependent phenotypes adjusted for age, gender, and ethnicity. Because FBAT is robust to population stratification, analyses were not performed independently by ethnicity.

Case-control sample.

Replication analyses were performed in the GWAS sample of alcohol-dependent subjects reported by Treutlein et al. (24). Alcohol-dependent men (n = 487) were recruited from consecutive admissions to treatment facilities as part of the German Addiction Research Network (GARN; http://www.bw-suchtweb.de). Controls subjects (n = 1,358) were recruited through population-based epidemiologic studies. SNPs reported in this replication phase passed the standard quality-control measures, and analyses were performed in PLINK. See Treutlein et al. (24) for further details on the sample description, analyses, and methods.