INTRODUCTION

Ethanol, absorbed through the gastrointestinal tract, enters the liver via the portal vein. After being oxidized in the liver, ethanol metabolites and any unmetabolized ethanol enter the general circulation. Two kinds of enzymes are primarily responsible for the oxidation of alcohol. The first enzyme family is the alcohol dehydrogenases (ADH), which convert alcohol to acetaldehyde (which is toxic). The second enzyme family is the aldehyde dehydrogenases (ALDH), which quickly convert acetaldehyde into acetate. Acetate is then oxidized into CO2 and H2O via the tricarboxylic acid cycle.

The development of alcohol dependence is related to an individual's ethanol consumption, which is influenced by ADH activity. Increased ADH activity can increase the rate of alcohol metabolism, thereby increasing the production of acetaldehyde. If the acetaldehyde cannot be metabolized quickly, the accumulation of acetaldehyde will result in an aversive ‘flushing reaction,’ characterized by facial flushing, headache, increased skin temperature, dizziness, anxiety, nausea, vomiting, decreased blood pressure with resultant tachycardia and, rarely, severe cardiovascular complications. The aversive effects of acetaldehyde can limit alcohol consumption, and, it is hypothesized, thereby reduce the risk of alcohol dependence (Thomasson et al, 1994; Muramatsu et al, 1995; Hasin et al, 2002).

Active ADH enzyme is composed of two subunits, which can be, but are not necessarily, identical. Humans express at least nine possible different ADH subunits, each with slightly different properties. These isoforms include α, β, γ, π, χ, ADH6 isoenzyme, μ, and σ. A unique gene encodes each isoform. A human ADH gene cluster maps to chromosome 4 (4q21–25), extending a total of 364 128 base pairs (bp) and including genes encoding five classes, in the following order: IV (ADH7), I-γ (ADH1C), I-β (ADH1B), I-α (ADH1A), V (ADH6), II (ADH4), and III (ADH5).

The human π subunit is encoded by the ADH4 gene. In adults, the π subunit mainly contributes to liver ADH activity, catalyzing the oxidation or reduction of a wide range of substrates at moderate and high concentrations, including ethanol, primary long-chain aliphatic alcohols, and aromatic aldehydes (Li et al, 1977; Li and Bosron, 1987; Edenberg et al, 1999). Variation in ADH4 could change the function of the π subunit and affect the risk for alcohol dependence by modulating alcohol metabolism. For example, the −75A allele has promoter activity that is more than twice that of the −75C allele; this −75A/C substitution (SNP6: rs1800759) (see Figure 1 for single nucleotide polymorphism (SNP) definitions and locations) at the promoter could significantly affect π subunit expression (Edenberg et al, 1999). Considering the important role of the ADH4 gene in alcohol metabolism, it is a functional candidate for alcohol dependence.

Figure 1
figure 1

Map of markers at the ADH4 gene.

Full size image

Several linkage studies, including the Collaborative Study on the Genetics of Alcoholism (COGA) (Reich et al, 1998) and a study by investigators at the National Institute on Alcohol Abuse and Alcoholism (NIAAA) (Long et al, 1998), have demonstrated that the ADH gene cluster is coincident with a possible linkage signal for alcohol dependence. Reich et al (1998) performed a genome-wide screen by examining 291 markers in 987 individuals from 105 affected families and found suggestive evidence of a protective locus for alcohol dependence on chromosome 4 near the ADH gene cluster. Zinn-Justin and Abel (1999) performed an autosome-wide scan in the COGA data using the method of weighted pairwise correlation incorporating identity-by-descent (WPC-IBD), and located a risk locus for alcohol dependence at a 50 cM region spanning from GABRB1 to D4S1651, a region harboring the ADH genes. Williams et al (1999) applied a variance-component method to analyze COGA data and located a risk locus for alcohol dependence at a chromosome 4 region near the ADH gene cluster. Saccone et al (2000) carried out Haseman–Elston-based regression analyses on the COGA data set and mapped susceptibility genes for a quantitative alcoholism-related phenotype, that is, ‘maximum number of drinks consumed in a 24-h period,’ which is correlated with the alcohol dependence diagnosis, to the region of the ADH gene cluster. In the study from NIAAA, Long et al (1998) performed a whole-autosomal genome scan with 517 microsatellite markers in several extended pedigrees and found that three loci in the ADH gene cluster showed evidence for linkage with alcohol dependence. In summary, the ADH gene cluster is in a region that has been positionally implicated in alcohol dependence. Since the ADH4 gene maps within this ADH gene cluster, ADH4 gene is also a positional candidate for alcohol dependence.

There are also case–control association data implicating the same region of chromosome 4 (but different specific loci) by showing associations of ADH genes to alcohol dependence (eg, Thomasson et al, 1994; Hasin et al, 2002; Konishi et al, 2003). Foroud et al (2003) analyzed 58 markers at the ADH gene cluster in COGA alcoholic families using the pedigree disequilibrium test and primarily identified eight variants (published abstract only; specific marker names were unavailable to us) in the ADH4 gene associated with the risk for alcohol dependence in European-Americans (EAs).

The present study aimed to extend these findings by examining the relationship between ADH4 gene variants and alcohol dependence in several populations, including EAs and African-Americans (AAs). We focused on the ADH4 gene in this study because of the functional importance of its protein product, remarkable for its activity at high ethanol concentrations that might be of particular relevance in the context of alcohol dependence risk (Li et al, 1977); because of the preliminary COGA results (Foroud et al, 2003); and because our preliminary case–control survey of variants at many ADH loci immediately identified this as a gene of interest (Luo et al, unpublished data).

Drug dependence, which mainly includes cocaine dependence and opioid dependence in our sample, is one of the most common phenotypes comorbid with alcohol dependence (Gossop, 2004). Many studies have demonstrated that drug dependence has a number of features in common with alcohol dependence, including symptomatology (Glatt, 1967; Heil et al, 2001), neuropsychological impairments (Robinson et al, 1999), pathogenetic mechanisms (Anokhina et al, 1990; O'Neill et al, 2001), and response to specific treatments, especially the effects of disulfiram, an ALDH blocker, which though approved for the treatment of alcohol dependence, has also been shown to be efficacious in the treatment of cocaine dependence (Carroll et al, 1993, 1998, 2000, 2004; George et al, 2000; Petrakis et al, 2000). Drug dependence may share susceptibility genes with alcohol dependence (reviewed by Kreek et al, 2004), for example, OPRM1 gene variation moderated susceptibility to alcohol dependence and/or drug dependence (Hoehe et al, 2000; Szeto et al, 2001; Schinka et al, 2002; Luo et al, 2003). To answer the question of whether ADH4 variants influence risk for alcohol dependence, drug dependence, or both of these traits, we examined the association of ADH4 gene with drug dependence diagnoses in this study.

MATERIALS AND METHODS

Subjects

A total of 926 unrelated subjects (557 male and 369 female subjects), including 400 EA cases, 317 EA healthy controls, 161 AA cases, and 48 AA controls, were included in this study. The cases met lifetime DSM-IV criteria (American Psychiatric Association, 1994) for alcohol dependence, drug (cocaine or opioid) dependence, or a combination of these disorders. Among the cases, 436 subjects had a diagnosis of alcohol dependence (334 EAs, 102 AAs) and 356 had a diagnosis of drug dependence (207 EAs, 149 AAs), including 318 with cocaine dependence (178 EAs, 140 AAs) and 152 with opioid dependence (95 EAs, 57 AAs). The control subjects were screened to exclude major Axis I mental disorders, including alcohol or drug abuse or dependence, psychotic disorders (including schizophrenia or schizophrenia-like disorders), mood disorders, and anxiety disorders. Male subjects constituted 73.2% of the cases and 40.0% of the controls. The average ages were 39.4±9.2 years for cases and 28.1±9.1 years for controls.

The subjects were recruited at the University of Connecticut Health Center or the VA Connecticut Healthcare System-West Haven Campus. All subjects gave informed consent before participating in the study, which was approved by the Institutional Review Board at the respective institutions.

Genotyping

Marker inclusion (see Figure 1)

Seven known SNPs (all of those available from the public Applied Biosystems (Foster City, CA) website (http://www.appliedbiosystems.com/) at the time of genotyping; designated by hcv# or rs#) were included in this study (see Figure 1). These SNPs were selected based on the following findings or hypotheses: (1) the promoter SNP (SNP6: rs1800759) is known to be a functional polymorphism, altering promoter activity (Edenberg et al, 1999); (2) SNP2 (rs1042363) at exon 9 and SNP3 (rs1126671) at exon 7, which result in the amino-acid substitutions Gly → Arg and Val → Ile, respectively, might also affect the function of the π subunit; and (3) some ‘silent’ SNPs such as SNP1 (hcv2033010) at the 3′ end, SNP4 at exon 6, SNP5 (rs7694646) at intron 4, and SNP7 (rs1984362) at the 5′ end might influence π subunit expression as well, either directly (eg, through effects on regulation or mRNA stability) or indirectly via linkage disequilibrium (LD) with other functional variants such as SNP2, SNP3, or SNP6.

The allele frequencies and PCR conditions for these markers have been validated by ABI. The heterozygosity of each of these markers in the ABI database is >0.375 in an EA population, which corresponds to a minor allele frequency >0.25. These markers, spanning the full length of ADH4 gene, cover a range of 28 752 bp, with an average spacing of 4792 bp.

Genotyping

Genomic DNA was extracted from peripheral blood by standard methods. The seven ADH4 SNPs were genotyped with a fluorogenic 5′ nuclease assay method, that is, the TaqMan technique (Shi et al, 1999), using the ABI PRISM® 7900 Sequence Detection System (ABI, Foster City, CA, USA). PCR was performed in a final volume of 3 μl, including 1.5 μl 2 × TaqMan Universal PCR Master Mix (ABI), 0.075 μl 20 × Assays-on-Demand or 0.0375 μl 40 × Assays-by-Design (ABI), 0.03 μl 100 × BSA (New England Biolabs Inc., Beverly, MA), and 1 ng DNA. All genotyping was performed in duplicate and compared to ensure validity of the data. Mismatched genotypes, which constituted <0.5% of the total number of duplicate genotypes performed, were discarded.

Statistical Analysis

Linkage disequilibrium analysis

Different populations have different generations of ancestries and different degrees of admixture, which may result in different LD decays from the initial generations to the current generations. Different LD decay is one factor that results in different LD block sizes between populations. Thus, in the present study, the pairwise LD between any two ADH4 markers was analyzed separately for EAs and AAs. The D′ value for each LD pair was calculated and visualized through the program Haploview (Barrett et al, 2005) (see Figure 2a and b).

Figure 2
figure 2

LDs among seven ADH4 markers in (a) unrelated EA subjects and (b) unrelated AA subjects. D′=1.00 in the blank red squares; the numbers inside the darker squares are D′ × 100%; the lighter squares represent low r2 values.

Full size image

Hardy–Weinberg equilibrium test

Deviation from Hardy–Weinberg equilibrium (HWE) expectations, that is, Hardy–Weinberg disequilibrium (HWD), can indicate a valid gene–phenotype association (Feder et al, 1996; Nielsen et al, 1998; Jiang et al, 2001; Hoh et al, 2001; Lee, 2003; Hao et al, 2004). The HWD test as an association method in cases only has been reported to be more powerful than a case–control study, especially for a recessive model (Nielsen et al, 1998). HWE was tested within populations, and separately for cases and controls. The test in a mixed population was conducted as an exploratory evaluation of population stratification effects. HWE tests in the groups with mixed affection status (ie, those resulting from combining cases and controls) were used to determine whether the markers were in HWE within each population. HWE for each marker was tested using goodness-of-fit implemented in the program PowerMarker (Liu and Muse, 2004). Genotype frequency distributions of the seven markers within each phenotype group were also considered together and tested by an omnibus HWD test via the program ‘min_p_test’ (see below).

In addition to the use of the HWD test to evaluate associations between genes and diseases, the HWD test must be applied to determine what analysis methods (software programs) are valid for haplotype reconstruction (see below), because some methods are valid only in the presence of HWE.

Two HWD measures (ie, F and J) can be used to fine-map the disease risk locus. These measures are calculated using the following equations: F=Δ/He (Feder et al, 1996) and J=Δ/2He′ (Jiang et al, 2001), where Δ=He−Ho, He and Ho are the expected and the observed frequencies, respectively, of the heterozygotes in cases (assuming HWE), and He′ is the expected frequency of the non-disease-associated homozygotes in cases. F and J are decreasing functions of θ, where θ is the recombination rate between the marker and the disease susceptibility locus. Higher values of F and J reflect greater LD between marker and disease locus. The marker locus with either the highest F or J value was taken as the marker closest to the disease susceptibility locus. The value of F depends not only on the genetic distance between the marker locus of interest and the disease susceptibility locus, but also on the allele frequencies at the marker locus. J does not hinge on the allele frequencies at the marker locus.

Case–control comparisons for allele and genotype frequencies

The allele and genotype frequencies of the ADH4 markers among EAs and AAs are shown in Tables 1 and 2. Associations between the alleles, genotypes, and phenotypes were tested by comparing the allele and genotype frequency distributions between cases and controls with the exact tests in the program PowerMarker.

Table 1 Genotype and Allele Frequencies of Seven ADH4 Markers in EAs
Full size table
Table 2 Genotype and Allele Frequencies of Seven ADH4 Markers in AAs and HWD Tests in AAs, EAs and EAs+AAs
Full size table

To avoid type I error due to multiple comparisons without information loss due to the overly conservative Bonferroni correction, the genotypes of these seven markers were considered together in an omnibus test using the program ‘min_p_test’ (Hao et al, 2004; personal communication), to obtain adjusted global p-values and potentially to identify real susceptibility SNPs among these markers. ‘min_p_test’ provides three p-values based on three different statistics including the Pearson χ2 test statistic (PE) for genotypewise analysis, the Cochran–Armitage χ2 trend test statistic (CA) for allelewise analysis, and the HWD statistic for HWD test.

Haplotype Frequency Analysis

Multilocus haplotypes usually are more informative than any single marker. However, the ADH4 markers are in HWD in cases, violating the assumption of the expectation–maximization (EM) algorithm upon which many haplotype-reconstruction programs are based. The Bayesian approach and the Partition Ligation algorithm may still be applied to reconstruct haplotypes even when HWE is violated (Niu et al, 2002). The program PHASE 2.0.2, based on a combination of the Bayesian statistical method and the so-called ‘divide-and-conquer’ (ie, Partition Ligation) strategy, which is claimed to be more accurate in reconstructing haplotypes than the EM algorithm, was used for this purpose (Stephens et al, 2001; Stephens and Donnelly, 2003). PHASE not only provides the overall haplotype frequency distribution in the whole sample, but also directly compares the haplotype frequency distributions between cases and controls.

The haplotype frequency distribution depends on the LD among markers, which varies by population. In order to infer the haplotypes more accurately, haplotypes were reconstructed separately for EAs and AAs.

RESULTS

(1) Pairwise analysis in different populations showed that the ADH4 markers were in strong LD. In EAs, the seven ADH4 markers were in strong LD (D′>0.9) (see Figure 2a). In AAs, six ADH4 markers were in strong LD (D′>0.9), with SNP1 in weaker LD with the others (D′>0.79) (Figure 2b). These seven markers can be taken as representing a single haplotype block (Gabriel et al, 2002). There were no significant differences in LD among the ADH4 markers between cases and controls (within each population).

(2) The genotype frequency distributions of the seven ADH4 markers were in HWD in the EA patients with alcohol dependence or cocaine dependence, but were in HWE in the EA healthy controls (see Table 3).

Table 3 p-Values for Allelewise and Genotypewise Case–Control Comparisons and for HWD Tests in EAs
Full size table

In EAs, the genotype frequency distributions of all seven ADH4 markers were in HWD in patients with alcohol dependence or cocaine dependence, but were in HWE in patients with opioid dependence and in controls. After Bonferroni correction, where α was set at 0.007 (=0.05/7; 7 is the number of markers), five markers (SNP2–6) and seven markers remained in HWD in patients with alcohol dependence and cocaine dependence, respectively. SNP2 was in HWD in subjects with drug dependence without comorbid alcohol dependence, even after Bonferroni correction. As mentioned above, the Bonferroni correction is very conservative, as the markers were in LD and their genotypes were correlated. The omnibus HWD test, which obviates the Bonferroni correction, showed that the genotype frequency distributions of ADH4 markers were in HWD in all affected groups of EAs (see global p-values for HWD tests in Table 3).

Among these markers, SNP6 and SNP2 were in most significant HWD in cases with alcohol dependence and cocaine dependence, respectively (see Table 3). A goodness-of-fit test showed that these deviations from HWE were attributable to the excess of the observed homozygote frequency (fo) (see Table 1) over the expected homozygote frequency (fe) (data not shown).

According to the values of F, SNP6 and SNP2 were the markers closest to the risk loci for alcohol dependence and drug dependence (including cocaine and opioid dependence), respectively. But after correcting for the effects of allele frequencies of markers on HWD, that is, as reflected in the values of J, SNP6 is the marker closest to the risk loci for both alcohol dependence and drug dependence (J values not shown).

In AAs, all of the markers were in HWE in patients and controls. Although SNP5 and SNP7 were nominally in HWD in patients with opioid dependence (p=0.020 and 0.029, respectively), the effects were not significant after Bonferroni correction (see Table 2, p-values not listed).

Within EAs and AAs (mixing cases and controls; see second-to-last and third-to-last columns in Table 2), most ADH4 markers were in HWE, except SNP6 in EAs (p=0.015). When mixing EAs and AAs (see the last column in Table 2), over half of these markers were in HWD, which was expected given that the population was not stratified.

(3) The genotypes of ADH4 markers were associated with alcohol dependence and drug dependence in EAs (p-values listed in Table 3).

The allele and genotype frequencies of the seven ADH4 markers are listed in Tables 1 and 2. In EAs, there were nominally significant differences in genotype, but not in any allele, frequency distributions between controls and subjects with alcohol dependence, cocaine dependence, opioid dependence, or any drug dependence without comorbid alcohol dependence (see Table 3). All of these associations resulted from the higher frequency of homozygotes of each marker in cases than in controls (see Table 1). Before Bonferroni correction, the genotypes of all seven SNPs were associated with alcohol dependence and cocaine dependence (p<0.05); three SNPs, that is, SNP2, SNP4, and SNP5, were associated with opioid dependence; and three SNPs, that is, SNP2, SNP5, and SNP7, were associated with drug dependence without comorbid alcohol dependence. After Bonferroni correction (α=0.05/7=0.007), four SNPs (ie, SNP3–6) were associated with alcohol dependence, all seven SNPs were associated with cocaine dependence, and one SNP (ie, SNP2) was associated with drug dependence without comorbid alcohol dependence.

The omnibus tests using the ‘min_p_test’ program showed compelling evidence that (1) all seven SNPs were susceptibility markers for both alcohol dependence (PPE=0.007) and drug dependence (PPE=0.0004) (highly significant in cocaine dependence (PPE<10−6) and modestly significant in opioid dependence (PPE=0.044)), and (2) one of the SNPs (SNP2) was a susceptibility marker for drug dependence without comorbid alcohol dependence (PPE=0.008), an association identified despite the small sample size available for this subgroup.

In AAs, no significant differences were found in either allele or genotype frequencies between cases and controls.

(4) The program PHASE was used to compute estimated haplotype frequency distributions and directly compare these distributions between cases and controls. There were no significant differences in the seven-variant haplotype frequency distributions between cases and controls, in either EAs or AAs (data not shown, all p>0.05).

DISCUSSION

The present study provides strong evidence supporting the hypothesis that ADH4 is a susceptibility gene for alcohol dependence, and presents the first evidence that it is also a susceptibility gene for drug dependence. These findings were obtained using the HWD test in a case-only design and were confirmed using a case–control design.

In this study, the average age for controls was 28.1 years, more than 10 years younger than that (39.4 years) for cases. Many healthy controls have not completely passed the risk age for manifesting alcohol dependence or drug dependence. The healthy controls have a probability (≈prevalence of disease) to develop disease in their lifetime, and this probability increases when the prevalence is moderate (eg, alcohol dependence or major depression), so that the case–control association design becomes less powerful than a case-only study. That is, some associations that can be discovered by the use of a case-only study might not be detected using a case–control design. This is due to the fact that there are likely to be some control subjects who, although presently unaffected, have the genetic liability for affection and may be affected in the future. Therefore, under these circumstances, the case-only study design is more sensitive for the detection of allelic association between susceptibility alleles and diseases. The HWD test is an association method for case-only study, which has been demonstrated to be more powerful than case–control study in some circumstances. This seems to be particularly true for recessive diseases (Nielsen et al, 1998). Because, to our knowledge, this method has not been previously employed in psychiatric genetics (except in one previous study, Luo et al, 2005), it was necessary to confirm the results obtained with this method using a conventional case–control association study as a positive control.

A disease has the potential to exert strong selection on the genome, which can result in HWD for some specific loci. This deviation from HWE could be very significant if it occurs right at the disease risk locus (Nielsen et al, 1998; Wittke-Thompson et al, 2005). The markers in LD with the risk locus may be in HWD as well; alternatively, HWD may indicate that the markers in HWD are in close proximity to the disease risk locus. The detection of HWD has recently been shown to be useful as a method for detecting gene–phenotype association (Feder et al, 1996; Nielsen et al, 1998; Jiang et al, 2001; Hoh et al, 2001; Lee, 2003; Hao et al, 2004; Wittke-Thompson et al, 2005; Luo et al, 2005). The present study provides an additional illustration of this approach. We found that the ADH4 markers were in HWE in EA healthy controls, but were in HWD in EA cases with alcohol dependence or drug (ie, cocaine or opioid) dependence, which strongly suggests that the HWD reflects a gene–phenotype association. When considering all the markers together in an omnibus test (using the program ‘min_p_test’) to avoid multiple comparisons, we obtained global p-values that ranged from 10−6 to 0.022 for HWD tests in all phenotype groups (see PHWD in Table 3), which is a strong indication that ADH4 is in LD with the putative disease risk locus (or may be a risk locus itself). In other words, there are strong associations between diseases (including alcohol dependence and drug dependence) and ADH4 in EAs.

To confirm the findings we obtained using HWD tests, we also compared cases with controls on the allele, genotype, and haplotype frequency distributions within EA and AA populations. We found that (1) the allelewise analysis (using the program PowerMarker) did not detect any significant difference in the allele frequency distribution for any marker between cases and controls within EAs or AAs. Considering all of the markers together in an omnibus test to avoid multiple tests, we obtained global p-values >0.05 for all allelewise analyses (see PCA in Table 3). These negative results may be attributable to HWD, because under HWD, the marker alleles are not independent of one another, which leads to the case that the effects of disease-predisposing alleles may be ‘masked’ by other non-disease-predisposing alleles (Cordell, 2002). This reduces the power for the allelewise analysis (Sasieni, 1997; Nielsen et al, 1998; Luo et al, 2005) and may even invalidate it. (2) A multilocus haplotype is actually the subset of every single-locus allele; both allele and haplotype reflect the features of chromosomes in the population. Similar to the allelewise analysis, the haplotypewise analysis (using the program PHASE) in this study showed significant differences in the haplotype frequency distributions between cases and controls neither in EAs nor in AAs. Although these negative results (directly derived from the program PHASE) might be partly attributable to the low power of PHASE for case–control comparison, we speculate that they more likely result from HWD, because under HWD, as with the allelewise analysis, the haplotypes are not independent of one another. Consequently, disease-predisposing haplotypes may be ‘masked’ by other non-disease-predisposing haplotypes, which thus reduces the power for, or even invalidates, the haplotypewise analysis. We have also demonstrated that the haplotypewise analysis is potentially less powerful than the diplotypewise analysis (Luo et al, 2005). A diplotype (ie, a haplotype pair) is the subset of every single-locus genotype; both genotype and diplotype represent the types of chromosome pairs in each individual. Diplotype is predicted by PHASE too, but diplotypewise analysis is exempt from the HWE requirement and thus preserves power. (3) Genotypewise analysis is valid regardless of the presence of HWE and thus should preserve statistical power even in the presence of strong HWD. Very interestingly, as predicted, the genotypewise analysis showed that there were significant differences in genotype frequency distributions for all seven ADH4 markers between cases and controls in EAs (see Table 3). After Bonferroni correction, the genotypes of four ADH4 markers were significantly associated with alcohol dependence; seven markers were associated with drug dependence (mainly cocaine dependence); and one marker was associated with drug dependence without comorbid alcohol dependence. Bonferroni correction treats every association test as an independent test, but since the seven ADH4 markers are in strong LD, this approach is clearly too conservative for setting the significance level, which could result in information loss. To avoid multiple comparisons and the concomitant inflation of type I error, we considered all of the markers together in an omnibus test using the program ‘min_p_test’; this approach identified seven susceptibility markers for alcohol dependence (global PPE=0.007) and drug dependence (cocaine dependence, global PPE<10−6; opioid dependence, global PPE=0.044) and one susceptibility marker for drug dependence without alcohol dependence (global PPE=0.008). These associations are population specific, that is, they were seen only in EAs, not in AAs. (We did not exclude the possibility that other unknown ADH4 markers might be associated with alcohol dependence or drug dependence in AAs, and did not exclude the possibility that negative results in AAs are attributable to small sample size for that population.) Therefore, in the present study, the allele and haplotype analyses were considered exploratory, and we emphasized the genotypewise analysis, whence we drew conclusions.

Comparing the genotypewise analyses and the HWD tests, we find that (1) the findings from the genotypewise analyses basically coincide with those obtained from the HWD tests. That is, the findings in the case-only study are confirmed by the case–control association study, which is post hoc evidence that the HWD test is a valid association method, and which is also post hoc evidence that makes other explanations (except for this indication of association) for HWD highly unlikely. (2) Most associations from the HWD tests are more significant than those from the case–control comparisons (with lower p-values—compare the p-values between two tests or compare PHWD with PPE in Table 3). In other words, the HWD method can be more powerful than the case–control association method, supporting the conclusion by Nielsen et al (1998).

The fact that the ADH4 markers were in HWE in healthy controls but in HWD in cases not only indicates a gene–phenotype association, but is also consistent with a recessive mode of inheritance of these markers (Feder et al, 1996; Nielsen et al, 1998). The fact that the associations were seen between the genotypes and the phenotypes, but not between the alleles and the phenotypes, also suggests a recessive mode of inheritance of these susceptibility markers, because when a trait-associated marker acts via a recessive mode of inheritance and is under HWD, the effect of the recessive allele (ie, the disease-risk allele) can be ‘masked’ by the dominant allele (ie, the non-disease-risk allele).

Both the HWD test and the case–control design as association methods are potentially vulnerable to population stratification, including admixture effects, which could result in spurious associations. For example, when EAs and AAs are mixed, over half of the markers are observed to be in HWD (see the last column in Table 2), which results from population stratification. Thus, we guarded against this confound by testing for HWD and making comparisons only within each population. Additionally, in a second study (Luo et al, in press), we used the structured association method to stratify our sample, and obtained results consistent with those reported here. Similarly, in the follow-up study (Luo et al, in press), the associations from the present case-only and case–control studies were confirmed using the transmission disequilibrium test (TDT) in an independent sample of small nuclear families.

An outlier marker with extreme HWD usually results from genotyping error, so HWD has also been used to detect such errors (Hoh et al, 2001). However, the genotyping technique applied in this study has a low measured error rate (<0.5%) and all the genotyping was performed in duplicate. Furthermore, none of the seven markers is an apparent outlier according to the HWD statistics, so HWD resulting from genotyping error is not a likely explanation for the observed results.

HWD in cases can also be useful for fine-mapping a disease locus. According to the F values (Feder et al, 1996), which reflect the extent of HWD, SNP6 and SNP2 were the markers closest to the risk loci for alcohol dependence and drug dependence (including cocaine and opioid dependence), respectively. After correcting for the influence of marker allele frequency on HWD, that is, according to the J values (Jiang et al, 2001), which also reflect the extent of HWD and are equivalent to Pexcess in a case–control design (Devlin and Risch, 1995), SNP6 is the marker closest to the risk loci for both alcohol dependence and drug dependence.

Association between the ADH4 gene and alcohol dependence accords with the functional role of the π subunit in the development of alcohol dependence. However, we noted that drug dependence was also associated with ADH4 gene polymorphisms. Interestingly, both in the HWD test and in the case–control comparison, the association of ADH4 gene with drug dependence reached levels of statistical significance that were at least as great as those for alcohol dependence. This suggests two major possibilities: (1) the susceptibility to drug dependence that is attributable to ADH4 is shared with alcohol dependence susceptibility (ie, it is independently a risk factor for each of the two phenotypes); or (2) the associations between the ADH4 gene and drug dependence are driven by the actual associations between ADH4 and alcohol dependence, because many patients with drug dependence have comorbid alcohol dependence. In addressing this question, we noted that some of the less robust associations, which we presumed to be due to smaller sample size, were found between the ADH4 gene and drug dependence only (ie, without comorbid alcohol dependence). This supports the first explanation, although further study of this question with a larger sample of such individuals is warranted. This explanation may help to shed light on the pharmacological mechanism by which disulfiram, an ALDH blocker that is thought to exert its treatment effect in alcohol dependence by interfering with acetaldehyde metabolism, decreases the relapse rate among cocaine-dependent individuals, even in the absence of comorbid alcohol dependence (Carroll et al, 1993, 1998, 2000, 2004; George et al, 2000; Petrakis et al, 2000).

In summary, findings from this study show that the seven markers at ADH4 are associated with alcohol dependence and drug dependence in EAs. Our results suggest that, of these SNPs, the functional variant SNP2 (rs1042363) at exon 9 and SNP6 (rs1800759) in the promoter region are the most important and may affect the risk for both alcohol dependence and drug dependence. Other susceptibility SNPs are in strong LD with SNP2 or SNP6. If the disease risk locus is not SNP2 or SNP6 itself, the putative disease risk locus may reside within an LD block containing SNP2 or SNP6 at ADH4 gene.