The effect of Interdependencies between CYP2A6 variant and smoking, and ADH1B variant and alcohol, on the risk of head and neck cancers
Smoking and alcohol consumption are well-established risk factors for squamous cell carcinomas of the head and neck (SCCHN) (1). Single nucleotide polymorphisms (SNP) in genes encoding tobacco and alcohol metabolising enzymes [Cytochrome P450 2A6 (CYP2A6) and Alcohol dehydrogenase 1B (ADH1B), respectively] have been implicated in the risk for SCCHN among smokers and alcohol consumers, respectively (2-4). However, the underlying potential casual pathways that may or may not involve these risk behaviours have not yet been quantified.
CYP2A6 is the primary enzyme responsible for the oxidation of 80-90% of nicotine entering the body. Carriers of the slow metabolizing CYP2A6*2 variant (A allele) metabolize nicotine at slower rates, display decreased nicotine clearance, higher plasma nicotine levels and consequently smoke at lower intensities (to maintain optimal nicotine levels) relative to wildtype (T allele) carriers, lowering the risk for tobacco related cancers (5-7). Studies on CYP2A6 SNPs with similar functional consequences as CYP2A6*2 support this hypothesis for SCCHN risk (2). Similarly, the ADH1B*2 polymorphism (A allele) encodes a version of the ADH1B enzyme that converts ethanol to acetaldehyde 50-100 times faster compared to wildtype (G allele) carriers (8). Individuals who lack this SNP do not exhibit aversive physiological reactions (alcohol induced flushing) associated with prompt build-up of acetaldehyde, documented among carriers, and consequently are associated with increased likelihood of heavy alcohol consumption (9-11). Consequently, non-carriers of these variants may have a higher risk for SCCHN relative to carriers (4, 12). This raises the possibility of an indirect causal pathway, whereby absence of CYP2A6*2 and ADH1B*2 variants lead to higher intensities of smoking and alcohol consumption, which may mediate the risk for SCCHN. Alternatively, there is also evidence that relative to carriers, individuals who lack the CYP2A6 and ADH1B SNPs have increased susceptibility to the carcinogenic effects of tobacco smoking and alcohol consumption, respectively suggesting interaction (2, 4, 6, 13).
The combination of these two hypotheses, that is, that the risk behaviour (smoking or drinking) may mediate and/or interact with the associated genetic exposure, results in 4 possible causal pathways for SCCHN risk: (i) no mediation or interaction between the genetic exposure and risk behaviour: SNPs and risk behaviours are independent; (ii) interaction alone: the effect of SNP is modified in the presence of the risk behaviour, but the SNP itself is not required for the risk behaviour to be present; (iii) both interaction and mediation: the effect of the SNP is in the presence of the risk behaviour and also, the SNP is required for the behaviour to be present; (iv) mediation alone: the effect of SNP on the outcome occurs only through the risk behaviour, where the risk behaviour is affected by the SNP itself. Deciphering these 4 pathways is important to understand the direct and indirect effects of tobacco/alcohol metabolizing genes and smoking/alcohol consumption on risk of SCCHN (14).
In case-control studies, one can estimate the proportions of the excess risk for SCCHN attributable to the four potential causal pathways using a 4-way decomposition technique based on a counterfactual causal framework (14). Using this causal framework, we estimated the extent to which the effects of two functional SNPs in CYP2A6 and ADH1B on SCCHN risk are mediated by heavy smoking and alcohol consumption, respectively, in a case-control sample of Canadian Caucasians. In addition, we use this data to demonstrate the estimation of proportions of excess risk attributable to the four possible pathways combining mediation and interaction hypotheses between each of these variants and associated risk behaviour.
Population, study design and data collection
The data were drawn from the Canadian site of an international hospital-based case‐control study, Head and Neck Cancer (HeNCe) Life, investigating the aetiology of head and neck cancers in relation to social, behavioural, lifestyle, biologic and genetic risk factors. Adult participants (N= 918) were recruited from the outpatient clinics of four major referral hospitals in Montreal between 2005 and 2013. Participant eligibility criteria for the study were: (I) born in Canada, (II) aged ≥18 years, (III) English or French speaking; (IV) living within 50 Km from the hospitals, (V) without history of cancer, and (VI) without mental or immune suppression disorders. Cases (N=460) included consecutive, incident, histologically confirmed, stage I to IV squamous cell carcinomas of mouth, oropharynx and larynx (C01‐C06, C09, C10, C12‐ C14, and C32, under the International Statistical Classification of Diseases, 10th Revision). Cancer‐free controls (N=458), frequency matched to each identified case by 5-year age group and sex, were randomly selected from 10 outpatient clinics in the same hospitals from a list of non-chronic diseases not strongly associated with tobacco and alcohol consumption (with no single diagnostic group contributing to more than 20% of the total). Ethical approval was obtained from review boards of McGill University, Institut National de la Recherche Scientifique, and all participating hospitals. All participants signed an informed consent form prior to enrolment in the study.
Face-to face interviews using a questionnaire with a life-grid technique (15) were used to collect data on several domains of exposure including sociodemographic factors, lifetime history of tobacco smoking and alcohol consumption.
Sample collection and analysis: Genetic polymorphisms and Human papillomavirus
Oral epithelial samples for genetic and human papillomavirus (HPV) analysis were collected using brush biopsy and oral rinse following standardized protocols (16, 17). To identify genetic polymorphisms, genotyping was performed on DNA isolated from the samples using real-time Taqman PCR expression assays. Briefly, reactions were set up using 5 µl of 2X Genotyping Master Mix (Applied Biosystems, Foster City, CA) combined with assay-specific concentrations of primers and probes, and 10 ng of sample DNA. The reactions were then spun down at 2000 rpm for 2 min, and run in the 7500FAST real-time PCR thermocycler in Genotyping mode under default settings. 7500FAST v2.0.1 (updated to v2.0.6) software was used for allelic discrimination. The 2 SNPs selected for this analysis were CYP2A6*2 (rs1801272) and ADH1B*2 (rs1229984). HPV DNA detection was performed using a standardized PCR protocol (18, 19) as described in Laprise et al (in press).
Genetic exposure definition: CYP2A6*2 and ADH1B*2
CYP2A6*2 was genotyped as TT, AA and AT (A = minor allele). Relative to carriers of the variant (AT or AA genotypes), the non-carriers (TT genotype) have decreased nicotine levels in plasma, higher tobacco smoking frequencies, higher pro-carcinogen to carcinogen activation rates, and higher risk for SCCHN (6, 13). ADH1B*2 was genotyped as GG, AA and AG (A = minor allele). Relative to carriers of the variant (AA or AG genotypes), non-carriers (GG genotype) consume alcohol at higher intensities due to lower acetaldehyde build up, and have an increased risk for SCCHN (4, 9, 10). Furthermore, the proportion of participants homozygous for the minor allele of both variants were low. Hence, we used the genetic exposures as binary. Carriers of CYP2A6*2 (AT/AA) and non-carriers (TT) were coded as 0 and 1 respectively. For ADH1B*2, carriers (AG/AA) and non-carriers (GG) were coded as 0 and 1 respectively.
Risk behaviour- mediator- definition: Tobacco smoking, alcohol consumption
CY2A6*2 and ADH1B*2 SNPs are associated with intensity of smoking and alcohol consumption respectively. Therefore, we used frequency of smoking and alcohol use as measures for these risk behaviours (6, 20). For tobacco smoking, we collected detailed information on commercial cigarettes, hand rolled cigarettes, cigars and pipes [duration (age of cessation minus age of initiation), frequency (how many per day)] during multiple stable smoking periods over an individual’s life. All tobacco types were then converted to the commercial cigarette equivalent based on nicotine content (1/9 cigar = 1/3.5 pipe=1/2 hand rolled cigarettes= 1 commercial cigarette) (21). From the total duration and frequency of commercial cigarettes used, we calculated average number of commercial cigarettes smoked per day over the lifetime. A non-linear dose-response relationship was identified between cigarettes smoked per day and SCCHN risk with risk increasing till 35 cigarettes per day and plateauing thereafter. Using a parametric outcome-based approach (22), we identified 18 cigarettes per day as the optimal cut point and used this threshold to dichotomize participants into moderate smokers (>0 to 18 cigarettes per day, coded 0) and heavy smokers (>18 cigarettes per day, coded 1).
We collected similar information for alcohol consumption: multiple beverages [type (wine/cider, beer, hard liquor, aperitif), duration (age of cessation minus age of initiation), quantity (small glass-50ml, medium glass-100ml, big glass-250 ml, half bottle-330 ml, bottle-700-750ml), and frequency of consumption (how many per day or per week or per month)] for multiple time periods of stable consumption across life. Each beverage was converted to ethanol equivalents (10% ethanol in wine and aperitif, 5% in beer/cider, and 50% in hard liquor) (23). Similar to tobacco intensity, this information was used to calculate the average amount of ethanol (in millilitres) consumed per day over the lifetime. The risk for SCCHN increased till approximately 100ml of ethanol per day and then plateaued. An amount of 25ml of ethanol per day was identified as the optimal cut point using the parametric outcome based approach (22), and participants were then grouped into moderate drinkers (>0 to 25ml per day, coded 0) and heavy drinkers (>25ml per day, coded 1).
Mediation and 4-way-decomposition
The mediation and 4-way decomposition models used in this analysis were based on the counterfactual framework for causal inference in case-control studies (24). Under this framework, the average total effect (TE) of the genetic exposure on SCCHN in the population can be decomposed into the product of overall direct (NDE) and indirect effects (NIE) on the relative risk scale (25, 26). In our scenarios, TE reflects the change in risk of SCCHN for an overall change in the exposure in the population from AT/AA to TT genotype for CYP2A6*2, and AG/AA to GG genotype for ADH1B*2 SNPs. The NDE is the estimated effect of TT and GG genotypes on SCCHN risk operating through pathways other than heavy smoking and heavy alcohol intensities, respectively. By contrast, the NIE estimated the effects of TT and GG genotypes through heavy smoking and alcohol intensities, respectively.
Alternatively, the four-way decomposition involves segregating the total excess relative risk (i.e., TE-1) for SCCHN among those exposed into four non-overlapping components on the excess relative risk scale (14). These included: (i) the controlled direct effect (CDE); the portion of the effect of genetic exposure on SCCHN risk when the associated risk behaviour intensity is set to moderate/mild levels (i.e., component of excess risk attributed to neither mediation nor interaction with heavy intensity of risk behaviour); (ii) the reference interaction (INTref); the portion of the effect of the genetic exposure that requires the joint presence of high intensity of the associated risk behaviour (interaction alone), with the high intensity behaviour arising independently of the associated genetic exposure; (iii) the mediated interaction (INTmed); the portion of the effect of genetic exposure that requires the joint presence of associated heavy intensity behaviour, with the heavy intensity behaviour arising as a consequence of the associated genetic exposure (both interaction and mediation), and (iv) the pure indirect effect (PIE); the portion of the effect of genetic exposure that is due to genetic exposure-induced high intensity behaviour (mediation alone). The overall proportion of the effect of genetic exposure on SCCHN risk mediated (PM) by associated heavy intensity risk behaviour can be calculated as the sum of PIE and INTmed components, divided by the excess relative risk. The proportion of the effect attributable to interaction (PAI) between the genetic variant and associated heavy intensity of risk behaviour is given by sum of INTref and INTmed components, divided by the excess relative risk. Proportion eliminated (PE) is the proportion of effect of the genetic variant on SCCHN risk that can be eliminated in the population if the level of the associated risk behaviour was decreased to that of moderate/mild intensity in the population. This is given by the sum of INTref, INTmed and PIE, divided by the excess relative risk.
Assumptions for causal interpretation and potential confounders
Causal interpretation of our results through the counterfactual framework rely on four no-confounding assumptions as well as correct model specifications (14): no unmeasured confounding of the effects of (i) genetic exposure on SCCHN risk, (ii) genetic exposure on associated risk behaviour, and (iii) risk behaviour on SCCHN risk, and (iv) none of the risk behaviour-SCCHN confounders are affected by the associated genetic exposures. We addressed assumptions (i) and (ii) by restricting our analysis to Caucasians, thus mitigating confounding due to population stratification (27). Regarding assumption (iii), we adjusted for potential confounders of the relationship between risk behaviours and SCCHN risk. For the smoking intensity-SCCHN association, we identified duration and time since cessation (continuous, mean centred, current smokers recoded to zero) of smoking, and intensity of alcohol (continuous, adjusted for restricted cubic spline) as confounders. For the alcohol intensity-SCCHN association, time since cessation of use (continuous, mean centred, current users recoded to zero) of alcohol, and pack-years of commercial cigarette equivalence (continuous, adjusted for restricted cubic spline, 20 commercial cigarettes = 4 hand-rolled cigarettes = 4 cigars = 5 pipes = 1 pack of commercial cigarettes) were identified (28). Additionally, we adjusted for age (continuous), sex, number of years of education (continuous) and HPV risk types (4 categories: HPV negative; low risk HPV; HPV 18, 31, 33, 35, 39, 51; HPV 16 alone or combined) for both associations. These variables are not known to be affected by the associated genetic exposures which may potentially address assumption (iv).
The CYP2A6*2-smoking-SCCHN and ADH1B*2-alcohol-SCCHN analyses were performed only among smokers and alcohol consumers, respectively as no association has been documented between these genetic variants and SCCHN among non-consumers. The direct and indirect effects and decomposition estimates were obtained by fitting logistic regression models on the binary outcome and mediator (14). For CYP2A6*2-smoking-SCCHN, SCCHN was regressed on the CYP2A6*2, cigarettes per day, their product term (denoting interaction) and associated potential confounders (outcome model). Next, cigarettes per day was regressed on CYP2A6*2 and potential confounders (mediator model, fit only among controls). For ADH1B*2-alcohol-SCCHN, the outcome model was fit on ADH1B*2, ethanol per day, their product term and associated potential confounders. For the mediator model, ethanol per day was fit on ADH1B*2 and potential confounders among controls. An indicator variable for ex-smokers and ex-alcohol consumers was added in the smoking and alcohol related models, respectively, to account for time since cessation of use of the respective products (29). For both scenarios, the mediator model in the full sample (i.e., in both cases and controls) weighted on the sampling fraction (30) gave quantitatively similar estimates as the model fit among controls (14). Effect estimates and associated proportions were obtained by combining parameters from these two models according to their corresponding analytical equations (14, 31). The 95% Confidence Intervals (CI) for the estimates were obtained using bootstrapping (2000 replications). All analyses were carried out using Stata, version 13SE (StataCorp. 2013, College Station, TX). The 4-way decomposition codes for Stata were specifically developed for this study using mathematical equations for the binary outcome and the binary mediator scenario provided by VanderWeele (personal communication)(32). Annotated Stata codes are provided in the Supplementary material, eAppendix.
Of the total 918 participants, 818 were genotyped on CYP2A6*2 and ADH1B*2. Of these, 633 and 678 were Caucasian smokers and alcohol consumers among whom 32 (13 controls and 19 cases) and 3 (2 controls and 1 case) had data missing on CYP2A6*2 and ADH1B*2, respectively. Therefore, we present the CYP2A6*2-smoking-SCCHN analysis on a sub-sample of 601 participants (only smokers), and ADH1B*2-alcohol-SCCHN on 675 participants (only alcohol consumers).
Descriptive characteristics of smoker and alcohol consumer sub-samples are given in Tables 1. Briefly, both samples had similar sociodemographic characteristics. Among the smoker sub-sample, cases had higher proportions of TT genotype (CYP2A6*2 non-carriers) and of heavy smokers compared to controls. Similarly, in the alcohol user sub-sample, cases had a higher proportion of GG genotype and of heavy drinkers compared to controls. The estimates (risk ratio, RR) for genetic exposure-SCCHN, mediator-SCCHN, and genetic exposure-mediator (among controls) associations in both samples were all above 1 (Tables 1 and 2).
The results of the standard mediation analysis (2-way decomposition) are given in Table 3. The average TE estimate for SCCHN for a change from AT/AA to TT genotype in the sample of smokers was RR=1.28 (95% CI: 0.46, 3.59) which was composed of a direct effect estimate (NDE) of 1.22 (95% CI: 0.45, 3.33) and an indirect effect estimate (NIE) through smoking of 1.05 (95% CI: 0.94, 1.17). The TE estimate amounted to an excess RR estimate of 0.28 (95% CI: -1.8, 2.37). Among the sample of alcohol consumers, the average TE estimate for a change from AG/AA to GG genotype was RR= 2.37 (95% CI: 1.12, 4.25) with an excess RR estimate of 1.37 (95% CI: 1.62, 4.38). The TE estimate decomposed into a NDE estimate of 2.24 (95% CI: 0.88, 5.71), and NIE estimate of 1.06 (95% CI: 0.97, 1.16).
Table 3 displays the results for the 4-way decomposition. Among smokers, 65% of the excess RR estimate for SCCHN due to TT genotype (CYP2A6*2 non-carriers) was attributable to the CDE component, 14% to INTref, 11% to PIE, and 10% to INTmed. The overall proportion of risk due to TT genotype mediated by heavy smoking was 21% and the proportion attributable to interaction with heavy smoking was 24%. The overall proportion eliminated was estimated at 35%. Among alcohol consumers, approximately 84% of the excess RR for SCCHN due to GG genotype (ADH1B*2 non-carriers) was attributable to the CDE component. Proportions attributable to the other 3 components were about 5% each. The proportion of risk due to GG genotype mediated by heavy alcohol use was 10%, that attributable to interaction was 11% and the proportion eliminated was approximately 16%.
In this study, we aimed to quantify the causal pathways between two functional SNPs in CYP2A6 and ADH1B genes, and SCCHN risk which may or may not involve mediation by heavy smoking and alcohol consumption behaviours respectively. We further demonstrated theF potential for existence of 4 causal pathways between these genetic exposures and SCCHN risk combining mediation and interaction hypotheses. Albite imprecise, the risk estimates seem to indicate that effects of TT genotype (CYP2A6) and GG genotype (ADH1B) on SCCHN risk were mainly through pathways not mediated by heavy smoking or alcohol intensities, respectively. However, results were suggestive of the possibility that these risk behaviours may interact with and mediate the effect of respective genetic variants on SCCHN risk.
Before interpreting the results, it is important to consider the limitations of this study. Firstly, our analysis was limited by sample size and confidence intervals of most estimates were wide. This limits our capability to assert that inference based on these estimates are true of the population parameters. Nevertheless, in this work, we intended to demonstrate the technique of decomposition analysis with respect to these genetic variants, respective risk behaviours and SCCHN, that has not been explored in the oral health literature. These analyses were performed based on the positive and monotonic estimates for the exposure-outcome, mediator-outcome and exposure-mediator associations, whose directions were as documented in the literature based on underlying biological mechanisms. In addition, our estimates for interaction between the genotypes and heavy intensities of respective risk behaviours (Supplementary material, etable 1), were also in the expected direction. Secondly, it is possible that the effect of ADH1B*2 on SCCHN documented in this study is a reflection of it being in linkage disequilibrium with the variants in ADH1C gene (with similar functional consequences). We did not have information on the ADH1C variant to adjust for in the models. However, studies in both Caucasian and Asian populations suggest that the associations between ADHIB*2, intensity of alcohol consumption and SCCHN risk are independent of variants in the ADH1C gene, and were strongest among all alcohol dehydrogenase (ADH) related genes studied (3, 4, 33). Also, the magnitude and direction of estimates for the association between ADH1B*2, alcohol consumption and SCCHN risk were similar to what has been documented before. Thirdly, since the CYP2A6*2-smoking-SCCHN analysis, and ADH1B*2-alcohol-SCCHN was restricted to smokers and alcohol consumers respectively, there is possibility of a small collider stratification bias, as selection into the study is affected by both exposure and outcome. This may have led to an underestimation of true causal effects (34). However, not restricting may lead to a higher variance because a large proportion of the controls have no direct exposure effect (the non-smokers and non-alcohol consumers). Hence, restriction was performed assuming a small bias vs large variance in estimates. Simulation studies are required to quantify this bias-variance trade-off. Lastly, although the possibility of confounding by population stratification within a single ethnicity has been documented, limiting the analysis to a single ethnicity, as we have done, can mitigate this bias.
CYP2A6*2, smoking and SCCHN risk
Genetic studies have hypothesised that the effect of variants in CYP2A6 on the risk of tobacco related cancers (specifically of squamous cell origin) may be due to interaction, mediation, both interaction and mediation, or independent of smoking (5, 7, 13, 35, 36). However, these potential pathways have not been quantified yet. In our study, among smokers, the total effect estimate and positive excess risk suggest that TT genotype could confer a higher risk for SCCHN risk. Our results also indicate that this total effect could be composed of a large direct effect and a small indirect effect.
Analysis among controls indicated that the TT genotype had a positive, albeit imprecise, excess risk for being heavy smokers and smoked 6 cigarettes more per day on average relative to AT/AA genotype (17±9 cigarettes per day vs 23±15 cigarettes per day) (Supplementary material, etable 2).This is consistent with other studies conducted among North American Caucasian smokers including Canadians (7, 37) and is based on the mechanism that, relative to AT/AA genotype, the TT genotype metabolizes nicotine faster, increasing the need for smoking more cigarettes to maintain optimal nicotine levels in plasma. Based on this mechanism, a small proportion of the risk for SCCHN due to TT genotype being mediated by heavy smoking (PM = 21%) is a possibility. As explained in other clinical and biological settings (38), this overall proportion mediated could comprise of two distinct components; a) the increase in risk for SCCHN due to the total pool of carcinogens supplied by the excess number of cigarettes smoked per day as a consequence of CYP2A6 enzyme activity among TT genotype i.e., pure mediation (PIE), and b) the increase in risk for SCCHN due higher levels of carcinogenic products resulting from metabolism of pro-carcinogenic substrates specific to CYP2A6 enzyme, among the total pool of procarcinogens/carcinogens supplied by the excess number of cigarettes smoked per day due to TT genotype, i.e., both interaction and mediation (INTmed). The overall PM in our study was half attributable to PIE and half attributable to INTmed.
The majority of the effect of TT genotype seemed to be through a pure direct effect (NDE) which is a combination of controlled direct effect (CDE) and reference interaction (INTref) components. The CYP2A6 enzyme, expressed in both hepatic and extra hepatic tissues (e.g., upper aerodigestive tract), is mainly associated with the activation of tobacco specific nitrosamines and other nitrosamine procarcinogens (e.g., NNK, NNN, NDMA, NDEA, NDBA) to carcinogens (36). This gives rise to the possibility of interaction between CYP2A6 variants such as CYP2A6*2 and smoking which is a major source of these pro-carcinogenic substrates. Relative to AT/AA genotype, the rate of conversion of these pro-carcinogens to carcinogens is faster for individuals with TT genotype. The strength of association between CYP2A6 variants and cancers is stronger among heavy smokers, and a lack of association among non-smokers has been reported (35, 39). These suggest potential interaction between non-carriers of the variant and heavy smoking (that need not be induced by the CYP2A6 variant itself, i.e., INTref) Out of about 79% of excess risk due to NDE, 14% was attributed to INTref; the effect of TT genotype operating only in the presence of heavy smoking (not induced by the variant). Interaction results of our study may lend support to this finding (Supplementary material, etable 1); we estimated the overall proportion attributable to interaction (INTref + INTmed) at 24%.
Research on biologic mechanisms majorly support the view that the effect of TT genotype on SCCHN may involve mediation or interaction pathways with smoking intensity. However, approximately 65% of the excess risk for SCCHN due to TT genotype seemed to be attributed to CDE, i.e., TT has an effect on SCCHN risk even without the presence of heavy smoking, and without changing smoking intensity. However, given the wider confidence intervals and lack of studies exploring other mechanistic pathways through which CYP2A6 variants could lead to SCCHN risk (e.g., interaction/mediation with other sources of nitrosamines such as diet, environmental pollutants, gene-gene interactions), it may be speculative to interpret these results further.
ADH1B, alcohol and SCCHN
There is strong evidence for the significant impact of ADH1B*2 on intensity of alcohol consumed, as well as SCCHN risk among alcohol consumers. This is based on its involvement in the metabolism of ethanol to acetaldehyde (8). Alcohol consumers of Caucasian descent with GG genotype have been associated with 1.5 to 2 times the risk of being heavy alcohol consumers (including increased frequency) relative to AG/AA genotype (9-11). Among controls in our study, GG and AG/AA genotypes consumed 51 and 40 ml ethanol per day on average respectively (Supplementary material, etable 2). Although small, we documented an indirect effect between ADH1B*2 and SCCHN risk (constituting about 10% of the total excess risk) supporting the mediation hypothesis. Half of this indirect excess risk was attributable to excess amount of ethanol consumed due to GG genotype (PIE), and half due to interaction between GG genotype and excess amount of ethanol consumed due to GG genotype (INTmed=5%). In our study, and as documented by a pooled study among Australian twins (9), the average difference in frequency of ethanol consumed between GG genotype and AG/AA genotype is approximately 10 ml per day. Ethanol at this small dose may be of limited biologic relevance for SCCHN and could probably explain the relatively small magnitude of the indirect effect.
The majority of the total effect of GG genotype on SCCHN risk seems to be direct, as supported by both INTref and CDE proportions. Acetaldehyde, the primary metabolite of ethanol, is a human carcinogen (40-42). Contrary to what is expected among alcohol consumers with AG/AA genotype in whom acetaldehyde builds up promptly, studies in various ethnicities show higher risk for SCCHN among GG genotype carriers. It is hypothesised that AG/AA genotype exhibit alternative mechanisms to clear off acetaldehyde peak formed following alcohol ingestion (3). On the contrary, lower acetaldehyde peaks in GG genotype results in excess alcohol consumption. This may result in higher local exposure to ethanol in the head and neck region which is acted up on my ADH1B enzyme present in this region as well as oral microflora, leading to slower but increased build-up of acetaldehyde (4, 33, 43, 44). These findings suggest possibility for strong interaction between GG genotype and heavy alcohol consumption for SCCHN risk. In our results, only 11% of the total effect was attributed to interaction of which only about half was due to interaction with higher alcohol intensity not due to GG genotype (INTref=6%). Although, secondary analysis was suggestive of positive additive interaction, confidence intervals were wide (Supplementary material, etable 2).
Out of 90% of excess risk on SCCHN due to GG genotype through direct effect, more than 80% seem to be independent of heavy alcohol consumption, i.e., controlled direct effect. Ethanol being the only documented substrate for ADH1B enzyme, and sample size limitation, prevents interpretation of this result. It is to be noted, however, that we used ethanol per day as the measure of mediator, and it is possible that other aspects of alcohol consumption behaviour such as overall cumulative alcohol consumption and alcohol dependence (with which ADH1B*2 is also strongly associated) could potentially be important mediators of GG genotype on SCCHN risk.
A few strengths of this study are worth mentioning. Our work is one of the first to apply the 4-way decomposition causal analytical technique in a case-control setting, which provides the maximum insight into the interrelationship between two specific exposures and their effect on an outcome (14). This analytical strategy is based on the counterfactual framework, which allows for mediation and 4-way decomposition analysis in the presence of exposure-mediator interaction, and within a case-control design. This is an advantage over other methods proposed in the literature (45, 46). In a case-control study, the total effect of the variants on SCCHN risk cannot be decomposed into four non-overlapping components unlike cohort data that provides risk estimates on absolute scale. However, 4-way decomposition method does provide valid estimates of proportions of total excess risk, attributable to the 4 pathways in a case-control study. Although the possibility of indirect effect of variants in CYP2A6 gene on SCCHN risk through smoking behaviour has been hypothesised, they have not been previously explored. Also, it has been proposed that any effect of the ADH1B*2 variant on SCCHN risk is likely due to interaction alone. Our study encompasses these possibilities in the relationship between these genetic variants and SCCHN risk. Furthermore, based on our estimates, the potential for the existence of four pathways which may or may not include interaction, mediation or both with associated heavy intensity risk behaviours may not be ruled out.
Our study provides indication for the potential for existence of four distinct pathways underlying the relationship between the variants under study and SCCHN risk. Although the study was limited by sample size, our study is one of the first illustrations of segregation of direct and indirect effects into four non-overlapping components within a case-control study. The approach not only has the potential for deciphering mechanistically relevant pathways for rare disease outcomes, but also quantify measures of policy relevance (47). For example, the majority of the Caucasian population carries the TT (CYP2A6*2) and GG (ADH1B*2) genotypes. Direct modification of these variants to reduce their effect may not be possible nor economical. However, their effect on SCCHN can be modified by changing the level of modifiable risk behaviours such as smoking and alcohol consumption. For example, if the effect estimates in this study were indeed valid, approximately 37% of the effect of TT genotype on SCCHN risk could be eliminated, if the level of smoking was brought down to that of moderate smokers (i.e., under a packet of cigarettes per day). And among GG carriers, about 16% risk could be eliminated if the alcohol consumption level was brought down to that of mild drinkers (<25ml ethanol per day) (48). The overall proportion eliminated is higher than proportion mediated because of the additional risk for SCCHN due to interaction between the exposure and the mediator (31). Future studies with large sample size should explore these findings further.
1. Hashibe M, Brennan P, Chuang SC, Boccia S, Castellsague X, Chen C, et al. Interaction between tobacco and alcohol use and the risk of head and neck cancer: pooled analysis in the International Head and Neck Cancer Epidemiology Consortium. Cancer Epidemiol Biomarkers Prev. 2009;18(2):541-50.
2. Ruwali M, Pant MC, Shah PP, Mishra BN, Parmar D. Polymorphism in cytochrome P450 2A6 and glutathione S-transferase P1 modifies head and neck cancer risk and treatment outcome. Mutation Research/Fundamental and Molecular Mechanisms of Mutagenesis. 2009;669(1–2):36-41.
3. Hashibe M, Boffetta P, Zaridze D, Shangina O, Szeszenia-Dabrowska N, Mates D, et al. Evidence for an important role of alcohol- and aldehyde-metabolizing genes in cancers of the upper aerodigestive tract. Cancer Epidemiol Biomarkers Prev. 2006;15(4):696-703.
4. Hashibe M, McKay JD, Curado MP, Oliveira JC, Koifman S, Koifman R, et al. Multiple ADH genes are associated with upper aerodigestive cancers. Nat Genet. 2008;40(6):707-9.
5. Pianezza ML, Sellers EM, Tyndale RF. Nicotine metabolism defect reduces smoking. Nature. 1998;393(6687):750.
6. Tyndale RF, Sellers EM. Variable CYP2A6-mediated nicotine metabolism alters smoking behavior and risk. Drug metabolism and disposition: the biological fate of chemicals. 2001;29(4 Pt 2):548-52.
7. Rao Y, Hoffmann E, Zia M, Bodin L, Zeman M, Sellers EM, et al. Duplications and defects in the CYP2A6 gene: identification, genotyping, and in vivo effects on smoking. Mol Pharmacol. 2000;58(4):747-55.
8. Brennan P, Lewis S, Hashibe M, Bell DA, Boffetta P, Bouchardy C, et al. Pooled analysis of alcohol dehydrogenase genotypes and head and neck cancer: a HuGE review. Am J Epidemiol. 2004;159(1):1-16.
9. Macgregor S, Lind PA, Bucholz KK, Hansell NK, Madden PA, Richter MM, et al. Associations of ADH and ALDH2 gene variation with self report alcohol reactions, consumption and dependence: an integrated analysis. Human molecular genetics. 2009;18(3):580-93.
10. Li D, Zhao H, Gelernter J. Strong association of the alcohol dehydrogenase 1B gene (ADH1B) with alcohol dependence and alcohol-induced medical diseases. Biol Psychiatry. 2011;70(6):504-12.
11. Bierut LJ, Goate AM, Breslau N, Johnson EO, Bertelsen S, Fox L, et al. ADH1B is associated with alcohol dependence and alcohol consumption in populations of European and African ancestry. Molecular psychiatry. 2012;17(4):445-50.
12. Guo H, Zhang G, Mai R. Alcohol Dehydrogenase-1B Arg47His Polymorphism and Upper Aerodigestive Tract Cancer Risk: A Meta-Analysis Including 24,252 Subjects. Alcoholism: Clinical and Experimental Research. 2012;36(2):272-8.
13. Raunio H, Rautio A, Gullstén H, Pelkonen O. Polymorphisms of CYP2A6 and its practical consequences. British Journal of Clinical Pharmacology. 2001;52(4):357-63.
14. VanderWeele TJ. A unification of mediation and interaction. Explanation in causal inference: Methods for mediation and interaction. USA: Oxford university press; 2015. p. 371-96.
15. Berney L, Blane DB, Soc Sci M, Blane B. Collecting retrospective data: accuracy of recall after 50 years judged against historical records. Social Science & Medicine. 1997;45(10):1519-25.
16. Egan KM, Abruzzo J, Cytobrush B, Newcomb PA, Titus-ernstoff L, Franklin T, et al. Collection of Genomic DNA from Adults in Epidemiological Studies by Buccal Cytobrush and Mouthwash. 2001:687-96.
17. Souza GD, Sugar E, Ruby W, Gillison M, Gravitt P. Analysis of the Effect of DNA Purification on Detection of Human Papillomavirus in Oral Rinse Samples by PCR Analysis of the Effect of DNA Purification on Detection of Human Papillomavirus in Oral Rinse Samples by PCR. 2005.
18. Coutlée F, Mayrand MH, Provencher D, Franco E. The future of HPV testing in clinical laboratories and applied virology research. Clinical and diagnostic virology. 1997;8(2):123-41.
19. Kornegay JR, Roger M, Davies PO, Shepard P, Guerrero NA, Lloveras B, et al. International Proficiency Study of a Consensus L1 PCR Assay for the Detection and Typing of Human Papillomavirus DNA : Evaluation of Accuracy and Intralaboratory and Interlaboratory Agreement International Proficiency Study of a Consensus L1 PCR Assay for. 2003.
20. Hakenewerth AM, Millikan RC, Rusyn I, Herring AH, North KE, Barnholtz-Sloan JS, et al. Joint effects of alcohol consumption and polymorphisms in alcohol and oxidative stress metabolism genes on risk of head and neck cancer. Cancer Epidemiol Biomarkers Prev. 2011;20(11):2438-49.
21. Hoffmann D, Hoffmann I. Chemistry and toxicology.In: US Department of Health and Human Services. Cigars: health effects and trends (Smoking and Tobacco Control Monograph 9). DHHS (Publ No. NIH 98-4302), 1998:55–104.
22. Williams BA, Madrekar JN, Madrekar SJ, Cha SS, Furth AF. Finding Optimal Cutpoints for Continuous Covariates with Binary and Time-to-Event Outcomes Mayo Clinic, Rochester, Minnesota Division of Biostatistics DoHSR; June 2006. Contract No.: 10027230.
23. Schlecht NF, Franco EL, Pintos J, Kowalski LP. Effect of smoking cessation and tobacco type on the risk of cancers of the upper aero-digestive tract in Brazil. Epidemiology (Cambridge, Mass). 1999;10(4):412-8.
24. Brady, Henry. “Models of Causal Inference: Going Beyond the Neyman-Rubin-Holland Theory.” In Janet M. Box-Steffensmeier, Henry E. Brady, and David Collier, editors, The Oxford Handbook of Political Methodology New York: Oxford University Press, 2008.
25. Robins JM, Greenland S. Identifiability and exchangeability for direct and indirect effects. Epidemiology (Cambridge, Mass). 1992;3(2):143-55.
26. Pearl J. Direct and indirect effects. Proceedings of the Seventeenth conference on Uncertainty in artificial intelligence; Seattle, Washington. 2074073: Morgan Kaufmann Publishers Inc.; 2001. p. 411-20.
27. Wacholder S, Rothman N, Caporaso N. Population stratification in epidemiologic studies of common genetic variants and cancer: quantification of bias. Journal of the National Cancer Institute. 2000;92(14):1151-8.
28. Schlecht NF, Franco EL, Pintos J, Negassa A, Kowalski LP, Oliveira BV, et al. Interaction between Tobacco and Alcohol Consumption and the Risk of Cancers of the Upper Aero-Digestive Tract in Brazil. American Journal of Epidemiology. 1999;150(11):1129-37.
29. Leffondre K. Modeling Smoking History: A Comparison of Different Approaches. American Journal of Epidemiology. 2002;156(9):813-23.
30. Vanderweele TJ, Vansteelandt S. Odds ratios for mediation analysis for a dichotomous outcome. American journal of epidemiology. 2010;172(12):1339-48.
31. VanderWeele TJ. Explanation in Causal inference: Methods for mediation and Interaction. Press OU, editor. USA2015. 706 p.
32. Erratum: A Unification of Mediation and Interaction: A 4-Way Decomposition. Epidemiology (Cambridge, Mass). 2016;27(5):e36.
33. Asakage T, Yokoyama A, Haneda T, Yamazaki M, Muto M, Yokoyama T, et al. Genetic polymorphisms of alcohol and aldehyde dehydrogenases, and drinking, smoking and diet in Japanese men with oral and pharyngeal squamous cell carcinoma. Carcinogenesis. 2006;28(4):865-74.
34. Hernan MA, Hernandez-Diaz S, Robins JM. A structural approach to selection bias. Epidemiology (Cambridge, Mass). 2004;15(5):615-25.
35. Kamataki T, Fujieda M, Kiyotani K, Iwano S, Kunitoh H. Genetic polymorphism of CYP2A6 as one of the potential determinants of tobacco-related cancer risk. Biochemical and Biophysical Research Communications. 2005;338(1):306-10.
36. Rossini A, de Almeida Simao T, Albano RM, Pinto LF. CYP2A6 polymorphisms and risk for tobacco-related cancers. Pharmacogenomics. 2008;9(11):1737-52.
37. Schoedel Ka, Hoffmann EB, Rao Y, Sellers EM, Tyndale RF. Ethnic variation in CYP2A6 and association of genetically slow nicotine metabolism and smoking in adult Caucasians. Pharmacogenetics. 2004;14(9):615-26.
38. Ikram MA, VanderWeele TJ. A proposed clinical and biological interpretation of mediated interaction. European Journal of Epidemiology. 2015;30(10):1115-8.
39. Canova C, Richiardi L, Merletti F, Pentenero M, Gervasio C, Tanturri G, et al. Alcohol, tobacco and genetic susceptibility in relation to cancers of the upper aerodigestive tract in northern Italy. Tumori. 2010;96(1):1-10.
40. Baan R, Straif K, Grosse Y, Secretan B, El Ghissassi F, Bouvard V, et al. Carcinogenicity of alcoholic beverages. The Lancet Oncology. 2007;8(4):292-3.
41. Secretan B, Straif K, Baan R, Grosse Y, El Ghissassi F, Bouvard V, et al. A review of human carcinogens—Part E: tobacco, areca nut, alcohol, coal smoke, and salted fish. The Lancet Oncology. 2009;10(11):1033-4.
42. Salaspuro M. Acetaldehyde as a common denominator and cumulative carcinogen in digestive tract cancers. Scand J Gastroenterol. 2009;44(8):912-25.
43. Chang JS, Straif K, Guha N. The role of alcohol dehydrogenase genes in head and neck cancers: a systematic review and meta-analysis of ADH1B and ADH1C. Mutagenesis. 2012;27(3):275-86.
44. Homann N, Jousimies-Somer H, Jokelainen K, Heine R, Salaspuro M. High acetaldehyde levels in saliva after ethanol consumption: methodological aspects and pathogenetic implications. Carcinogenesis. 1997;18(9):1739-43.
45. Baron RM, Kenny DA. The moderator-mediator variable distinction in social psychological research: conceptual, strategic, and statistical considerations. J Pers Soc Psychol. 1986;51(6):1173-82.
46. Hayes AF. Beyond Baron and Kenny : Statistical Mediation Analysis in the New Millennium Beyond Baron and Kenny : Statistical Mediation Analysis in the New Millennium. 2009(July 2014):37-41.
47. VanderWeele TJ. A unification of mediation and interaction: a 4-way decomposition. Epidemiology (Cambridge, Mass). 2014;25(5):749-61.
48. VanderWeele TJ. Policy-relevant proportions for direct effects. Epidemiology (Cambridge, Mass). 2013;24(1):175-6.
Cite This Work
To export a reference to this article please select a referencing stye below:
Related ServicesView all
DMCA / Removal Request
If you are the original writer of this dissertation and no longer wish to have your work published on the UKDiss.com website then please: