Aldehyde Dehydrogenase 2 and Head and Neck Cancer: A Meta-analysis Implementing a Mendelian Randomization Approach

Alcohol drinking at high doses is a risk factor for head and neck cancer, and exposure to acetaldehyde, the principle metabolite of alcohol, is supposed to account for the increased risk. Individuals homozygous for the *2 variant allele of aldehyde dehydrogenase 2 (ALDH2) are unable to metabolize acetaldehyde, which prevents them from alcohol drinking, whereas *1*2 have 6-fold higher blood acetaldehyde concentration postalcohol consumption with respect to *1*1. According to the concept of Mendelian randomization, because this polymorphism is distributed randomly during gamete formation, its association with head and neck cancer should be not confounded by smoking. We carried out a meta-analysis of ALDH2 and head and neck cancer searching for relevant studies on Medline and Embase up to January 31, 2008, and investigated the consistency between the expected odds ratio (OR) among drinkers from the largest pooled analysis among never smokers and the observed OR from this meta-analysis by an interaction test. Six studies were selected (945 cases, 2,917 controls). The OR of head and neck cancer among *2*2 was 0.53 [95% confidence interval (95% CI), 0.28-1.00] relative to *1*1 and 1.83 (95% CI, 1.21-2.77) among *1*2. The expected OR for head and neck cancer due to alcohol intake among *1*1 was 1.38 (95% CI, 0.88-2.17) and the observed OR among *1*1 compared with 2*2 from this meta-analysis was 1.88 (95% CI, 1.00-3.57; P for interaction = 0.43). Besides showing the effectiveness of the Mendelian randomization approach, these findings support the theory that alcohol increases head and neck cancer risk through the carcinogenic action of acetaldehyde. (Cancer Epidemiol Biomarkers Prev 2009;18(1):248–54)

Head and neck cancers account for almost 600,000 new cancer cases worldwide each year, and at least 75% of them are attributable to a combination of cigarette smoking and alcohol drinking (1, 2). These two risk factors are strongly correlated with each other (3); their individual contribution to head and neck cancer etiology was recently addressed by a combined analysis of 15 case-control studies reporting data on cigarette smokers among never drinkers and on alcohol drinkers among never smokers. The results showed that cigarette smoking is a strong risk factor for head and neck cancer among never drinkers, whereas in the absence of tobacco use alcohol is a weak risk factor, apparent only at high doses (2).

The functional polymorphisms of genes related to alcohol metabolism can be used to investigate the association between alcohol consumption and head and neck cancer as surrogate for directly measuring such exposure. According to the principle of “Mendelian randomization”, the distribution of alleles in a population is expected to be unrelated to the confounders that may distort observational epidemiologic studies because of the random assignment of alleles at the time of gamete formation (4). As such, if a functional genetic variant, or a variant in strong linkage disequilibriun with it, is strongly associated with a modifiable exposure, it can be used to retrieve an unbiased estimate of the association of such exposure (e.g., alcohol) with a disease (e.g., head and neck cancer). In analogy with the fact that the random allocation of a treatment in a randomized controlled trial should prevent from known and unknown confounding factors, the Mendelian randomization concept suggests that a genetic variant of interest can be used to infer causality because it is not associated with confounders. As recently shown by Davey Smith et al., behavioral, socioeconomic, and physiologic factors are strongly interrelated, unlike genetic variants, which are not associated with each other, neither with the previously mentioned covariates (5). Moreover, because genotypes are a fixed characteristic, regression-dilution bias, a type of measurement error that leads to an underestimation of the association between the exposure and the disease under study in observational studies, is unlikely (4). Hence, using a gene that influences alcohol drinking propensity as a proxy of alcohol exposure in assessing the association between alcohol and head and neck cancer, should provide an estimate unconfounded by smoking. In fact, smoking and other behavioral or environmental factors should be independent of the distribution of genetic variants. As such, the Mendelian randomization approach may prove a valuable tool to infer causality in cancer epidemiology, although there are some limitations that should be acknowledged as later discussed.

Alcohol in humans is oxidized to acetaldehyde, which in turn is oxidized by aldehyde dehydrogenases (ALDH) to acetate (6). ALDH2 is the major enzyme responsible for acetaldehyde elimination, and its polymorphic variants determine blood acetaldehyde concentrations after drinking. A single point mutation in the ALDH2 gene results in the ALDH2*2 allele (rs number 671), bearing a lysine amino acid at residue 487 instead of glutamic acid and characterized by a reduced ability to metabolize acetaldehyde (7). Half of the Japanese are heterozygotes or homozygotes for the *2 allele of ALDH2, showing respectively peak blood acetaldehyde concentrations postalcohol consumption 6- and 19-fold higher than homozygous wild-type individuals (8). As a result, ALDH2*2*2 homozygotes show facial flushing and nausea after alcohol consumption that prevent them from alcohol drinking, whereas heterozygotes exhibit less severe reactions (8). It would therefore be expected that the ALDH2 genotype influences diseases known to be related to alcohol consumption because of its strong influence on the propensity to alcohol drinking, and in fact the ALDH2*2*2 genotype was found to lower the risk of liver cirrhosis and esophageal cancer (9, 10).

The aim of this study was to quantify the association between the ALDH2 genotype and head and neck cancer via a meta-analysis of published studies, and to compare the results with those of the pooled analysis previously mentioned (2). This will imply an assessment of the performance of the Mendelian randomization approach. According to our prior hypotheses, (a) ALDH2*2*2 should have a reduced head and neck cancer risk predicted by their very low alcohol consumption; (b) the relative risk for head and neck cancer among ALDH2*1*1 versus *2*2 (virtually all never drinkers) should be comparable with the estimate from Hashibe et al. (2) given that both are not confounded by the effect of smoking; and (c) if acetaldehyde plays a carcinogenic role in head and neck cancer it is expected that ALDH2*1*2 heterozygotes are at increased risk of head and neck cancer compared with ALDH2*1*1, given similar level of alcohol intake.

Identification of relevant studies was carried out through a search of Medline and Embase up to January 31, 2008 using the following terms: “aldehyde dehydrogenase” or “ALDH” and “polymorphism” and “head and neck neoplasms” (MeSH), without any restriction on language. The search produced 31 articles. A cited reference search of the retrieved articles was carried out, and publications were also identified by reviewing the bibliographies of the retrieved articles. Eligible studies were those reporting the frequency of the ALDH2 polymorphism among head and neck cancer cases and controls according to the three variant genotypes (*2*2, *2*1, and *1*1). If more than one article was published from the same case series, we included the paper where the most individuals were reported in the analysis. Of the 31 articles retrieved, 12 studies were eligible for the meta-analysis (11-22). Six reports (17-22) were later excluded because they were subjects included in an expanded series (13). Finally, six case-control studies conducted on Japanese, and comprising a total of 945 cases and 2,917 controls were retained for the analysis (11-16). A description of the studies is given in Table 1.

Two of the authors (P.G and E.DF) extracted the data from each article using a structured sheet and entered them into a database. The followings items were considered: year and location of the study, characteristics of the case and control group, and number of cases and controls homozygous and heterozygous for the ALDH2 variant alleles. In order to check for a possible confounding effect of age, gender, and smoking status on the relation between ALDH2 and cancer, the authors of the eligible studies were invited to provide data on the ALDH2 genotype stratified by these covariates. As expected, there was no evidence of any confounding effect at P ≤ 0.10; unadjusted odds ratios (OR) for head and neck cancer associated with the ALDH2*2*2 and *1*2 genotypes were estimated using the homozygous wild-type genotype (*1*1) as a reference group in each study. In carrying out the meta-analyses, random effect models were used to take into account the possibility of heterogeneity between studies, which was tested by the Q statistic (23, 24). Statistical analyses were carried out using the STATA software package v.9.0 (Stata Corporation). To determine the deviation from Hardy-Weinberg equilibrium in each included study, we used a publicly available program.¹⁰

Publication bias was not investigated because of the low sensitivity of the qualitative and quantitative tests when the number of studies is lower than ten (25). Because our a priori hypothesis was that the homozygous ALDH2 variant genotype protects against head and neck cancer because individuals with the variant genotype avoid alcohol consumption, in the meta-analysis of ALDH2*2*2 versus *1*1 genotype we excluded a study that specifically selected cases and controls on the basis of alcohol intake (13).

In order to compare the result of our meta-analysis of *2*2 versus *1*1 with that of the pooled analysis of Hashibe et al. (2), we calculated the expected OR of head and neck cancer given alcohol intake in *1*1 never smokers by using the relative risks associated with different levels of alcohol consumption from Hashibe et al. (2). The following equation for relative risk (RR) was used, as suggested by Lewis and Smith (10): RR = ∑ RRi [95% confidence interval (CI) lb, ub] × Pi, with i denoting the drinking category. RRi is the relative risk in the ith drinking category as estimated in the pooled analysis of Hashibe et al. (ref. 2; namely ORs, 1.04, 1.30, 1.82, and 2.81 for the following drinking categories, <12.32 g/d, 12.32-36.97 g/d, 36.98-61.62 g/d, and ≥61.63 g/d ethanol, respectively), lb and ub are the lower bound and upper bound risks from the 95% CI estimations in the pooled analysis (2), and Pi is the proportion of ith drinking category among *1*1 controls, namely 20.08%, 31.58%, 20.65%, and 13.61%, respectively, resulting from the summing of the absolute numbers of *1*1 control individuals falling on the ith drinking categories as a single population, as we obtained from all the authors of the papers included in this meta-analysis, except Yokoyama et al. (13), which enrolled only alcoholics, and Hashimoto et al. (14). Because the reference group in the pooled analysis of Hashibe et al. (2) is never drinkers, equivalent to ALDH2*2*2 individuals as virtually all of these individuals are never drinkers, the expected OR resulting from this computation was then compared with the inverse of the OR obtained from our meta-analysis of ALDH2*2*2 versus *1*1. Consistency between the expected and observed OR was assessed by an interaction test (26). Lastly, even though the previous equation uses RR (10), the pooled ORs from the study of Hashibe et al. (3) should be a fairly good estimate of the RRs because the incidence of head and neck cancer is rare among never smokers.

We also aimed to test if the ALDH2*1*2 genotype increases head and neck cancer risk compared with the *1*1 genotype given similar level of alcohol intake. Therefore, the authors of the eligible studies were invited to provide genotype data stratified by alcohol consumption (moderate drinkers, 1-59 g/d ethanol; heavy drinkers, >59 g/d ethanol; cut-offs are chosen because the most commonly adopted from the papers included in the meta-analysis).

In none of the studies did the genotype frequencies among controls deviate from values predicted from Hardy-Weinberg equilibrium (P ≥ 0.05). With the exception of the study of Yokoyama et al. (13), which selected individuals according to alcohol intake, heavy drinking was more common among subjects with the *1*1 genotype compared with the *1*2, and there were no heavy drinkers with the 2*2 genotype (Table 2).

The overall OR from the meta-analysis was 0.53 (95% CI, 0.28-1.00) for the risk of head and neck cancer among *2*2 homozygotes compared with 1*1 homozygotes (909 cases and 2,070 controls; Fig. 1). There was no evidence of between-study heterogeneity (P = 0.29).

The expected OR for head and neck cancer due to alcohol intake among never-smoker *1*1 individuals was 1.38 (95% CI, 0.88-2.17) compared with never drinkers. The OR of head and neck cancer for *1*1 homozygotes versus *2*2 homozygotes in this meta-analysis was 1.88 (95% CI, 1.00-3.57), with a P = 0.43 from interaction test between the expected and the observed estimates.

The OR for heterozygotes compared with *1*1 homozygotes was 1.83 (95% CI, 1.21-2.77; 753 cases and 2,725 controls; Fig. 2), with evidence of between-study heterogeneity (P < 0.001). Among never drinkers, there was no evidence for increase in risk for *1*2 versus *1*1 individuals (OR, 1.03; 95% CI, 0.69-1.55; Fig. 2). Among heavy drinkers, however, there was a 3.57 increase in risk (OR, 3.57; 95% CI, 1.41-9.05), and among moderate drinkers the OR was 1.68 (95% CI, 1.27-2.22; Fig. 2). The heterogeneity among alcohol strata was reduced compared with the overall estimate, with P for heterogeneity = 0.52 and 0.59 among never and moderate drinkers, however, it persisted among heavy drinkers (P = 0.004).

The present meta-analysis showed that the ALDH2*2*2 genotype reduces the risk of head and neck cancer by almost 50%, and this is likely due to markedly lower levels of alcohol consumption among *2*2 homozygotes compared with *1*1 homozygotes (Table 2). This result is consistent with those of the meta-analysis of Lewis and Davey Smith (10), reporting an OR of 0.36 (95% CI, 0.16-0.80) for esophageal cancer among ALDH2*2*2 individuals compared with 1*1. Individuals bearing the ALDH2*2*2 genotype show severe reaction after the intake of even small amounts of alcohol; therefore this genotype protects against head and neck cancer to the extent predicted by the individuals avoiding alcohol consumption. This result supports the evidence from observational studies between alcohol intake and head and neck cancer risk. Hashibe et al. (2) recently assessed the independent contribution of alcohol on head and neck cancer by pooling individual level data among never users of tobacco. Results show that alcohol consumption increases the risk of cancer when consumed at high frequency (2). However, recall bias resulting in misclassification of alcohol exposure or smoking habits could affect the reported associations by introducing residual confounding. Conversely, the findings with respect to genotype are not subject to reverse causation (ill status influencing the alcohol measurement in cases resulting in differential misclassification of the exposure), neither to confounding by smoking. In fact, a large study conducted on 1,919 cancer-free Japanese men showed that smoking is not associated with the ALDH2 genotype (27), and this was also confirmed in three studies (11, 15, 16) included in the present meta-analysis (not shown in detail). It seems therefore that the ALDH2 genotype can be used as a surrogate for alcohol intake, which is not subject to confounding by smoking and can be used to test the findings of observational epidemiologic studies on alcohol and cancer. This is supported by the result of the present study, because the observed increased risk of head and neck cancer among ALDH2*1*1 versus *2*2 is close to the predicted from the largest published study relating head and neck cancer risk given a specific level of alcohol intake in never smokers (2). Although the predicted OR of head and neck cancer due to alcohol intake might be biased because it was computed without including Hashimoto et al. data (14), we suppose that this bias would affect the comparability of the two effect measures estimates to a very low extent.

The results of this meta-analysis show that heterozygosity for the *2 allele is associated with 80% statistically significant increased risk of head and neck cancer (Fig. 2). This suggests that acetaldehyde may be the mechanism through which alcohol increases the risk of head and neck cancer, because there is a 6-fold increase in acetaldehyde levels among heterozygotes who drink alcohol compared with wild-type individuals (8). This observation is strengthened by the result that among never drinkers the risk is not increased, whereas among moderate and heavy drinkers the risk is substantially elevated. Overall results suggest that possession of the variant allele of ALDH2 does not increase the risk of head and neck cancer unless alcohol is consumed, which is consistent with the results of the study on esophageal cancer by Lewis and Davey Smith (10).

The main acknowledged limitation of the meta-analysis of Lewis and Davey Smith (10) was the lack of access to individual level data, which we were able to overcome by working together with the authors of the published studies on the ALDH2 genotype and head and neck cancer. In this way we were able to verify that tobacco smoking is unlikely to be a confounder of the association between ALDH2 and head and neck cancer in our meta-analyses; to investigate the consistency between the expected OR of head and neck cancer given alcohol intake from observational studies and the observed OR from the meta-analysis of ALDH2*1*1 versus*2*2; and to stratify the results of the meta-analysis of ALDH2*1*2 by drinking status using the same cut-offs for the drinking categories across studies. Compared with the results of the individual studies, our meta-analyses were able to detect the protective effect of ALDH2*2*2 genotype with a larger precision, and to confirm the increased risk for ALDH2*1*2 drinking individuals as suspected by Asakage et al. (15) and Hiraki et al. (16), by pooling together data from 753 cases and 2725 controls. It is undoubted that larger studies should be done for confirming these findings, because our results are based on a limited sample size. Nevertheless, the total number of subjects included comprise the largest sample size so far.

In summary, the main finding of this meta-analysis is that the ALDH2*2*2 genotype reduces the risk of head and neck cancer and this is likely to be due to marked lower levels of alcohol consumption among *2*2 homozygotes compared with *1*1 homozygotes, thus strengthening the potential of the Mendelian randomization concept of using ALDH2*2*2 as a surrogate for alcohol intake (28). Our approach has some theoretical limitations, such as the potential for a pleiotropic effect of the polymorphism under study (i.e., the gene of interest influences many phenotypes); in fact, the ALDH2 variant is also associated with lower adiposity and weight among Japanese men, but not in women (29). This situation, however, should not be a particular concern for head and neck cancer, for which currently there is no evidence for supporting a strong role of obesity in its etiology. In order to rule out this effect, however, future studies would explore the effect of ALDH2 on head and neck cancer among Japanese women, for whom genotype is not associated with adiposity. Nevertheless, in our situation pleiotropy would become a real problem if another unknown compound being metabolized by ALDH2 also affects the risk of head and neck cancer, with acetaldehyde being associated with it, resulting in confounding. Again, the possibility for such confounding is likely to be much more limited than those encountered with lifestyle or environmental exposures as later discussed. As an additional general limitation for the Mendelian randomization approach is canalization, the developmental buffering against the effect of a polymorphism during fetal development, although this is very unlikely in our situation because alcohol consumption is adopted during adolescence or adulthood. Another limitation might be the potential for genetic confounding due to an unknown linkage disequilibrium with a polymorphism influencing alcohol behavior and/or acetaldehyde levels, but this is also very unlikely (5, 28). Lastly, population stratification might represent an additional source of confounding, but the homogeneity of the ethnic group in our study should protect against this form of bias.

In our situation the main potential confounding effect due to smoking related to ALDH2 genotypes is also very unlikely. In fact, the OR for head and neck cancer conferred by the *1*1 genotype compared with *2*2 was similar to the expected OR of head and neck cancer estimated from the largest phenotype-disease study among never users of tobacco (2). Because of the random allocation of genotype in advance with disease development, this result also implies that the relationship between alcohol levels and head and neck cancer reported by Hashibe et al. (2) was not subject to substantial residual confounding or reverse causality bias. Thus, in addition to showing the effectiveness of the Mendelian randomization approach in this context, our study supports evidence for a role of alcohol in head and neck cancer pathogenesis, and supports the role of acetaldehyde as a carcinogen for head and neck cancer.

No potential conflicts of interest were disclosed.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.