The combined search yielded 1024 references. After discarding the overlapping references and those which clearly did not meet the criteria, 91 studies remained. These studies were then filtered to ensure conformity with the inclusion criteria (13 studies(13
) were excluded (Supplement 1
)). In the end, 78 case-control studies (Table S1 in Supplement 1
) met our criteria for inclusion. These studies included 48 studies (5
) of Asian populations; 21 studies(43
) of European populations; four studies(79
) of Mexican Americans; one(76
) of African Americans; and four studies(76
) investigating the Cys allele (Native Americans, African Americans, and Trinidadians). Among these studies, two(48
) investigated heroin dependence or abuse; three(76
) investigated multi-drug dependence, and the other 73 studies investigated AD or AD and alcohol abuse. These 73 studies included 18 studies(28
) in which the alcoholic patients were affected by alcoholic liver disease, cirrhosis, and (or) pancreatitis (10 of the 18 studies also included data for the patients without any alcohol-induced diseases). These studies included 9,638 cases and 9,517 controls. The results are detailed below.
Based on all these samples, the frequency of the protective ADH1B His allele varied widely across the populations: high in the Asian normal populations 69% (19% – 91%) and affected subjects, 51% (9% – 93%); low in the European normal populations 5.5% (1% – 43%) and affecteds, 6.9% (0% – 51%); and rare in the Mexican normal populations 3% (2% – 7%) and affecteds, 4% (2% – 8%). On average, the frequencies of the His allele were 34% and 45% in the combined patients and controls, respectively. The ADH1B Cys allele was found in African Americans, native Americans, and Trinidadians with frequencies of 11% (1% – 30%) in controls and 7% (0% – 18%) in affecteds. In the 48 Asian studies, 45 studies showed lower frequency in cases than in controls; in the 21 European studies, 16 showed lower frequency in cases than in controls; in the four Mexican American studies, two showed lower frequency; and all the four studies of the Cys allele, that allele showed lower frequency in cases than in controls. Only the studies of AD and alcohol abuse were included in the following meta-analysis.
All the combined studies of AD and alcohol abuse, in particular, the Asian studies, showed strong association, with allelic (Arg vs. His) P values of 1×10−36 (OR = 2.06 (1.84, 2.31)) and 7×10−42 (OR = 2.24 (1.99, 2.51)), respectively (). The strong association was also found under both dominant (ArgArg + ArgHis vs. HisHis) and recessive (ArgArg vs. ArgHis + HisHis) models (P < 9×10−23 in all the populations and P < 2×10−31 in Asians) with much lower heterogeneity under the dominant model (P > 0.01). Strong association was also revealed in the combined Asian and European-ancestry studies (allelic P = 2×10−36), however, it was moderate in the European (P = 0.0002) and non-Asian studies (P = 2×10−5). The strict random effects model was applied when evidence for significant heterogeneity between studies was found throughout this meta-analysis.
Results of the overall and sub-grouped studies
In some studies the AD subjects had alcoholic liver disease, cirrhosis, or pancreatitis (designated as “induced diseases” in ). Meta-analysis of these studies showed significant evidence of association (P = 4×10−12 and OR = 1.76 (1.5, 2.07)). The association was stronger in the Asian populations (P = 3×10−12 and OR = 1.97 (1.54, 2.52)) but was not significant in the European populations (). The patients with only alcoholic liver disease also produced significant association (P = 0.005). For the subjects only with cirrhosis, the significant association was found with P = 9×10−8, which was also significant in Asians (P = 4×10−8) but not in Europeans.
In order to understand whether these strong associations were due only to the alcohol-induced medical diseases, we also analyzed the samples without any of these diseases. The results showed that there was not any decrease on the level of significance compared with the results with these induced diseases (e.g., P = 9×10−33 and OR = 2.1 (1.86, 2.37), ).
Two studies investigated both AD and alcohol abuse; seven studies had no explicit description of the patients as alcohol dependent (the possibility that subjects with a diagnosis of alcohol abuse could not be excluded); the other studies clearly described the patients as alcohol dependent (these “definite” alcohol dependent patients without any alcohol-induced diseases were designated as “AD” in ). The meta-analysis based on these “definite” alcohol dependent subjects also showed that there was no major change on the significance level of the association. Strong association was still detected in all the combined populations, in particular, in the Asian populations and in the combined Asians and Europeans for both allelic and genotypic analyses. For instance, the allelic P values were 2×10−29 (OR = 2.12 (1.86, 2.42)) and 1×10−33 (OR = 2.28 (1.99, 2.6)) in the combined populations and the Asian populations, respectively ().
For the Cys allele, evidence of significant association was detected with P value of 7×10−5 and OR of 2.74 (1.67, 4.51) under the recessive model (ArgArg vs. ArgCys + CysCys). Statistical significance was also identified in allelic analysis (). The studies of AD and alcohol abuse combined with those from the Cys allele showed stronger association with allelic P value of 1×10−38 (OR = 2.08 (1.86, 2.32)).
The four studies with the Cys allele, two studies of heroin dependence and abuse, and three studies of multi-drug dependence were analyzed separately. However, they were also combined with AD and alcohol abuse considering that alcohol and drug dependence have been reported to share common genetic risk(84
). The results showed that there was still strong evidence of association with ADH1B
His and Cys in both allelic (Arg vs. His and Cys) and genotypic analyses (). The overall P
value was 4×10−31
and OR was 1.98 (1.76, 2.22) for the allelic analysis, which was stronger under the dominant model. The Asian studies produced an allelic P
value of 6×10−32
(OR = 2.17 (1.9, 2.46)), and significant results were found consistently under the dominant and recessive models (P
, respectively). Significant association was further detected in the combined Asian and European studies and non-Asian studies (P
, respectively). In contrast, the European studies showed moderate evidence of association with an allelic P
value of 0.0007. The Mexican studies showed weak association (P
= 0.034) possibly due to limited sample size. In addition, significant association was also revealed in the combined Mexican Americans and African Americans (including the studies investigating the Cys allele), and the P
values were 0.0001 and 4×10−15
for the allelic analysis and recessive model, respectively.
Figure 1 Forest plots of ln(OR) with 95% CI for the allelic analysis. Black squares indicate the ln(OR) (ln(OR) can be better fitted than OR), with the size of the square inversely proportional to its variance, and horizontal lines represent the 95% CIs. The pooled (more ...)
Other Heterogeneity Analyses
Heterogeneity Q tests were also performed for differences in OR between the studies using the World Health Organization s International Statistical Classification ofDiseases and Related Health Problems (ICD)(87
), the American Psychiatric Association s Diagnostic and Statistical Manual of Mental Disorders (DSM)(88
) system, or other identified criteria and studies with no description of diagnosis criteria; between the English-language publications and Chinese-language publications; and between the studies of China-mainland and Taiwan and those of other countries or regions. The results showed that there was only evidence of marginal heterogeneity regarding diagnosis criteria using the recessive model (P
(Q) = 0.04), but no heterogeneity using the allelic analysis and dominant model; and weak heterogeneity regarding publication languages using the allelic analysis (P
(Q) = 0.01), but no heterogeneity under the dominant and recessive models. The weak heterogeneity between the languages may be due to the heterogeneity between the Asian studies because all the Chinese-language publications were investigating Chinese populations. The results are shown in Table S3 in Supplement 1
Publication Bias Analyses
Publication bias is an important issue in meta-analysis. In the present study, no evidence of significant publication bias was found in the studies of AD and alcohol abuse with European samples (P
(T) > 0.05) for either allelic or genotypic analysis. However, evidence of publication bias was found in those studies with Asian samples as well as when all the populations were combined as described below. For the allelic analysis, in the studies of AD and alcohol abuse with Asian samples, the Egger's regression P
value (1-tailed) was 0.0002, and Kendall's tau (Begg and Mazumdar rank correlation)(89
value (1-tailed) was 0.005; the analysis of Duval and Tweedie's trim and fill showed that there might potentially be eight missing studies, and the adjusted overall effect size was 2.50(2.21, 2.83) under the random effects model; in all the populations the Egger's regression P
value was 0.0001, and Kendall's tau P
value was 0.04; the trim and fill analysis showed that there might be six potential missing studies, and the adjusted effect size was 2.17(1.93, 2.43) under the random effects model. For the dominant model, in the Asian samples the P
(T) was 0.032, there might be four potential missing studies, and the adjusted effect size was 2.24(1.96, 2.57). However, it should be noted that all the three adjusted values of effect size were larger
than the corresponding observed values, which implied stronger associations in the adjusted studies compared with the observed studies. That is, the imputed missing studies were positive.
The classic fail-safe analysis showed that for the allelic analysis at least 8,512 assumed non-significant studies could bring the overall P
(Z) value to > 0.05 for all the studies of AD and alcohol abuse (6,606 for the Asian studies and 62 for the European studies); for the dominant model at least 2,839 assumed non-significant studies could bring the P
value to > 0.05 (2,466 for the Asian studies); and for the recessive model it needed at least 5,882 assumed non-significant studies (4,337 for the Asian studies and 64 for the European studies). The results further supported the strong associations detected in this meta-analysis. The funnel plots are shown for the allelic and genotypic analyses of all the studies of AD and alcohol abuse in Figures S3-S5 in Supplement 1
. Figure S3 in Supplement 1
indicates the increase of effect size from the observed to adjusted values.
Sensitivity and Retrospective Analyses
The sensitivity analyses showed that no individual study among the 78 that were included biased the findings to the extent that it could account for the strong observed associations. For example, the studies of AD and alcohol abuse showed strong consistency, regardless of the data set removed, with the allelic P
values always between 3×10−39
among the 70 studies; For the dominant and recessive models, the results were also strong and consistent, regardless of the data set removed, with the P
values never > 3×10−31
among the 57 studies and never > 3×10−21
among the 66 studies, respectively. The results are shown for the allelic analysis, dominant model, and recessive model in Tables S4–S6
(see Supplement 1
The asymptote lines of the analyses in retrospect based on 21 publication years showed that the cumulative synthesis tended towards stability in recent years, in line with the overall results of this meta-analysis. The results of the allelic analysis are shown in ; and the results of the genotypic analyses are shown in Figures S6 and S7 in Supplement 1
. The P
(Z) and P
(Q) values are shown in Tables S7-S9 in Supplement 1
Retrospective analysis for the allelic analysis. Analysis in retrospect was based on publication year since 1990.
LD and Haplotype Structure Analyses
The genes encoding alcohol dehydrogenase alpha, beta, and gamma subunits are organized as a gene cluster on chromosome 4q. Strong LD was found in the region of the gene cluster of the ADH6
, and ADH7
genes (Figure S8 in Supplement 1
), which were consistent with the studies by colleagues(76
). The first four genes were in a strong LD structure (large triangle in dark red and blue that was composed of multiple haplotype blocks), and ADH1A
were in a same haplotype block, which indicated that the contribution of the gene cluster to the association effect on alcohol and drug dependence was not independent. Since ADH1B
haplotypes have not been fully evaluated, it will be necessary for subsequent studies to investigate the roles of other polymorphisms in the same haplotype block (e.g., the nonsynonymous SNPs shown on the LD plot or in Table S10 in Supplement 1
) or the polymorphisms on other genes within the strong LD structure. The LD plots are shown for the Asian and European populations in Figures S8 and S9
(see Supplement 1