Associations between Variants of the 8q24 Chromosome and Nine Smoking-Related Cancer Sites

Tobacco smoking is responsible for >20% of all cancer deaths worldwide (1) and is a known cause of lung, aerodigestive, urinary tract, and cervical cancers. Recent IARC reviews have found growing epidemiologic evidence supporting associations between tobacco smoking with liver and stomach cancer as well (2, 3). Among developed countries, smoking cessation can decrease cancer risk (4, 5); however, in developing countries, particularly China, smoking prevalence continues to increase (6). It was estimated from twin cohort studies that heritable factor may attribute to 26% in lung cancer and 31% in bladder cancer development (7). Thus, understanding the role of genetics within smoking-related cancers continues to be of importance.

The 8q24 chromosome region has been of increasing research interest in cancer development and epidemiology. Amplification within the 8q24 loci has been observed within a diverse group of cancers (8-15). Recent genome-wide association studies identified associations between genetic variants or single nucleotide polymorphisms (SNPs), DG8S737, rs1447295, rs16901979, and rs6983267, along the 8q24 region and prostate cancer among multiple study populations: Icelandic, Swedish, European American, African American, and the Multiethnic Cohort (16-19). Haiman et al., using fine-mapping markers, designated 8q24 into three “regions” and identified SNPs that showed the strongest single association in two “regions” (“region 2”: rs16901979 and “region 3”: rs6983267; ref. 20). DG8S737 and rs1447295 variants from “region 1” were observed previously to have strong associations with prostate cancer (16). These SNPs and additional 8q24 variants have been subsequently confirmed by genetic association studies (21-27). Furthermore, studies have investigated the associations between variants of 8q24 region and cancers of the breast (23), colon (25, 28-30), endometrium (31), and testes (32). It is still unknown whether SNPs at 8q24 region are associated with tobacco smoking-related cancer sites.

Some epidemiologic studies have suggested that tobacco smoking may be associated with colorectal cancers (33). Additionally, several studies observed associations between colorectal cancer and SNPs rs6983267 (25, 28-30); therefore, we hypothesize that 8q24 SNPs may be associated with smoking-related cancers. To test this hypothesis, we selected one SNP from each “region” to investigate their potential associations with nine smoking-related cancer sites (lung, oropharynx, larynx, esophagus, stomach, liver, bladder, and pilot studies on nasopharynx and kidney) using data from three case-control studies: Los Angeles County (the LA study), Memorial Sloan Kettering Cancer Center (the MSKCC study), and China Taixing City study.

Details of this population-based case-control study have been described previously (34, 35). Study participation criteria included the following: (a) all subjects were residents of Los Angeles County at the time of recruitment (for controls) or diagnosis (for cases), (b) during the study period were ages 18 to 65 years, and (c) were able to speak either English or Spanish. Newly diagnosed pathologically confirmed cases were identified using the rapid ascertainment system of the Cancer Surveillance Program for Los Angeles County (34). Lung cancer cases (n = 611) and the upper aerodigestive tract (UADT) cancer cases (n = 601; oropharynx, larynx, nasopharynx, esophagus, and others) were interviewed from 1999 to 2004. Population-based controls (n = 1040) who were lung and UADT cancer free were identified through a formal algorithm providing a list of households within the neighborhood of each individual case. Recruitment rates were 39% for eligible lung cancer cases, 46% for eligible UADT cancer cases, and 79% for contacted eligible controls. Cases and controls were matched by age (10-year categories) and gender. Informed consent approved by the institutional review boards of University of California at Los Angeles and University of Southern California were obtained from all study participants.

Specific details regarding this study population were reported previously (36-38). In brief, this was a population-based case-control study conducted in Taixing City, Jiangsu Province, People's Republic of China. Eligible cases were residents of Taixing City (living in Taixing for ≥10 years), ages ≥20 years, and newly diagnosed with esophagus, stomach, or liver cancer from June 1 to December 30, 2000. All cases were pathologically or clinically confirmed and reported to the Taixing Tumor Registry at the Taixing Center for Disease Prevention and Control. A total of 206 stomach cancer cases, 204 liver cancer cases, 218 esophageal cancer cases, and 464 population-based healthy controls were interviewed using an epidemiologic questionnaire. Control groups were randomly selected from a generated list of residents, frequency-matched with cases on gender, age group (5 years), and residential village (or residential block in the city). In the 6-month study period, the recruitment rates were 89.4% for controls, 65% for stomach, 57% for liver, and 67% for esophageal cancer cases.

Detailed information of this study population was reported previously (39, 40). Briefly, this was a hospital-based case-control study conducted at MSKCC. Eligible cases of bladder and kidney cancer were seen at MSKCC from August 1, 1993 to June 30, 1997. Cases were recruited according to the following criteria: had a pathologically confirmed diagnosis, lived in the U.S. for ≥1 year, and were in stable medical condition. All cases were either newly diagnosed or undergoing surgical procedure for their relevant cancer. A total of 233 cases with bladder cancer and 34 cases with kidney cancer were interviewed. Controls were recruited based on the following criteria: consented in writing to participate in the study, resided in the United States for at least 1 year, and were in stable medical condition. During the 4-year study period, 178 controls were recruited from the MSKCC blood bank or were patients with a negative diagnosis for cancers at MSKCC. This study was approved by the Institutional Review Board on Human Subjects of MSKCC, and all study participants signed informed consents.

Epidemiologic data were collected by trained interviewers using study-specific standardized questionnaires. The detailed standard questionnaires of all three studies included the following information: (a) demographic factors; (b) personal habits: cigarette smoking, passive smoking, alcohol consumption, coffee and tea consumption, etc.; (c) history of occupational and environmental exposures; (d) family history of cancer; (e) dietary factors (food frequency questionnaire); (f) medical history; and (g) questions regarding environmental exposures that were specific to each of these nine cancer sites. The personal interview process took ∼40 min to 1 h.

For the LA study, buccal cells were collected from both cases and controls using the brushing of buccal mucosa and rinsing with mouthwash method (41). Response rates for interviewed participants providing buccal cells were 89% for controls and 89%, 68%, 88%, and 90% for lung, oropharyngeal/nasopharyngeal, laryngeal, and esophageal cancer cases, respectively. In the Taixing study, peripheral blood samples were collected from interviewed participants with response rates of 97.5% for controls, 95% for stomach and liver cancer cases, and 94% for esophageal cancer cases, respectively. Lastly, for the MSKCC study, peripheral blood samples were collected from both cases and controls, and normal and tumor tissue samples from cases who had undergone radical cystectomy. Biological specimens were available for 166 healthy controls, 174 bladder cancer cases, and 20 kidney cancer cases. Biological specimens were transported and stored in freezers of -70°C of the Molecular Epidemiology Laboratory, University of California at Los Angeles School of Public Health.

DNA samples were isolated from biological specimens using a modified phenol-chloroform method and assayed for purity and concentration by spectrometry (41). We selected from each “region” the strongest single association SNPs, “region 1”: rs1447295, “region 2”: rs16901979, and “region 3”: rs6983267. SNP genotyping was done using the TaqMan allelic discrimination method with the ABI 7900HT Real-time PCR System (TaqMan; Applied Biosystems). Aliquots of DNA from cases and controls were randomized onto PCR plates, into which a reaction mix containing Applied Biosystems TaqMan universal master mix and a probe for either SNP (Applied Biosystems) were added. Specific primers and probes were custom-designed by the ABI TaqMan system. Modified from the protocols of ABI TaqMan manual, after holding the plates at 92°C for 10 min, they underwent 60 thermocycles of denaturing at 92°C for 15 s and annealing at 62°C for 80 s. Following PCR amplification, end-point fluorescence was read using the ABI Primer 7900HT instrument and genotypes were scored using SDS 2.3 Allelic Discrimination Software from Applied Biosystems. For quality control, we genotyped 5% duplicated samples randomly selected to evaluate reproducibility and concordance rate was >99%. The automatic call rates were >96% for all three SNPs. Furthermore, as quality-control process, all laboratory researchers were blinded to the case or control statuses and to the identity of quality control samples.

Analysis was done using SAS v9.2 software. Tests for Hardy-Weinberg equilibrium (HWE) and differences in minor allele frequencies were evaluated for all three SNPs using the χ² test. Unconditional logistic regression models were employed to determine crude and adjusted odds ratios (OR) and 95% confidence intervals (95% CI) for the association between SNPs and each cancer site. For the LA study, we adjusted for the following variables: age, gender, ethnicity, educational level, and tobacco smoking. Education level and tobacco smoking were treated as continuous variables. Age was adjusted for in fine categories (<34, 35-36, 37-38, 39-40, 41-42, 43-44, 45-46, 47-48, 49-50, 51-52, 53-54, 55-56, 57-58, and 59-62), and controls who were more than 3 years younger than the youngest case or 3 years older than the oldest case were excluded from the analysis. This resulted in 11 excluded controls for lung cancer and 1 excluded control for UADT cancers. For all UADT cancers, the variable alcohol drinking was also adjusted for in the model. All models for the Taixing study were adjusted for age, gender, smoking pack-years, and alcohol drinking. For stomach cancer, we also adjusted for Helicobacter pylori infection status; for liver cancer, we adjusted for HBsAg status. Age and smoking variables were continuous, whereas the remaining variables were categorical. Alcohol drinking was defined in categories of never, occasionally, often, or everyday in the Taixing study. Lastly, for the MSKCC study of bladder cancer, we adjusted for the following factors: gender, age (<55, 55 to <60, 60 to <65, and ≥65), race (White versus non-White), and smoking (never versus ever). We first analyzed SNP genotypes (CC, CA, AA or TT, TG, GG) as a continuous variable (additive models) and as dummy variables for each cancer site. These results were used to identify the appropriateness of the use of a dominant or recessive model. For each site, changes in the OR for these three SNPs across levels of tobacco smoking were evaluated using unconditional logistic regression adjusting for previous mentioned confounding factors and OR for interaction were estimated by including smoking (never or ever), SNP genotypes (0 and 1) according to the dominant or recessive model, and product terms of smoking by each SNP.

To account for false-positive findings due to multiple testing, we calculated the false-positive report probability (FPRP) using an approach presented in Wacholder et al. (42). We set the FPRP threshold at 0.5, because this is an initial study investigating the associations between three 8q24 SNPs and smoking-related tumor sites. Due to the overwhelming evidence of associations between 8q24 variants with prostate and colon cancer, we assigned a prior probability range of 0.01 to 0.1 to detect an OR of 1.5 or 0.67.

The baseline characteristics of each study's population were reported previously (34-39) and a short summary for study populations can be found on Supplementary Table S1. The LA study consisted of a multiethnic population, non-Hispanic Whites (59%), African American (12%), Hispanic (17%), and others (12%, predominantly Asian). In the MSKCC study, the majority of participants were Whites (92.1% of cases and 96.8% of controls). Table 1 presents the distribution of genotypes of controls stratified by study sites and ethnicities. The distributions of 8q24 SNPs were consistent with the HWE (P > 0.05) among Whites (the LA and MSKCC studies), Mexican (the LA study), and Asian American (the LA study), African American (the LA study, two SNPs), and Chinese (the Taixing study, two SNPs). However, the distribution of rs16901979 in African Americans in the LA study and of rs6983267 in the Chinese population did not meet HWE (P = 0.021 and 0.028, respectively). There were notable variations in the distribution of minor allele frequencies in both African American (the LA study) and Chinese (the Taixing study) when compared with Whites (the LA study) for all SNPs (rs1447295: P < 0.0001 and = 0.014; rs16901979: P < 0.0001 and P < 0.0001; rs6983267: P < 0.0001 and P = 0.0016, respectively).

Table 2 presents ORs and 95% CIs for rs1447295, rs16901979, and rs6983267. After initial analyses by genotyping of each SNP, we determined the dominant model was appropriate for rs1447295 and rs16901979 in all cancer sites, whereas for rs6983267 the recessive model was appropriate for all sites, except nasopharynx, stomach and liver cancers, where the dominant model was employed. Using a recessive model and adjusting for potential confounding factors, rs6983267 (region 3) was positively associated with UADT cancers (OR_adj, 1.69; 95% CI, 1.28-2.24). When stratified by tumor site, rs6983267 was associated with cancers of the oropharynx (OR_adj, 1.80; 95% CI, 1.30-2.49) and larynx (OR_adj, 2.04; 95% CI, 1.12-3.72). Using the dominant model, there was a suggestive positive association between rs6983267 and liver cancer (OR_adj, 1.51; 95% CI, 0.99-2.31). Lastly, in a pilot study, using the dominant model, we observed an inverse association of rs16901979 (region 2) with kidney cancer (OR, 0.48; 95% CI, 0.23-1.00; data not shown). No obvious associations were observed between rs1447295 (region 1) and each smoking-related cancer.

Table 3 shows the adjusted ORs for all three SNPs and cancer sites with at least 75 cases stratified by smoking status (never versus ever). In assessing the relationship between rs6983267 (region3) and lung cancer stratified by smoking, we observed adjusted ORs of 1.45 (95% CI, 1.05-2.00) for ever-smokers and 1.00 (95% CI, 0.58-1.70) for never-smokers, suggesting possible interaction between smoking and the SNP rs6983267 on lung cancer. Associations between rs6983267 and UADT cancers were observed in both never-smokers (OR_adj, 1.56; 95% CI, 1.01-2.39) and ever-smokers (OR_adj, 1.79; 95% CI, 1.23-2.61), suggesting the SNP rs6983267 may be independent of tobacco smoking for UADT cancers. Among smokers, the SNP rs6983267 was observed to be positively associated with oropharyngeal cancer (OR_adj, 2.01; 95% CI, 1.29-3.11) and laryngeal cancer (OR_adj, 2.05; 95% CI, 1.09-3.85) and inversely associated with bladder cancer (OR_adj, 0.35; 95% CI, 0.14-0.83).

The SNP rs16901979 (region 2) was positively associated with UADT among never-smokers (OR_adj, 1.86; 95% CI, 1.06-3.28). When stratified by tumor site, rs16901979 was associated the cancer of the oropharynx (OR_adj, 2.28; 95% CI, 1.19-4.39). Among ever-smokers, no obvious association was observed between rs16901979 and all tumor sites listed in Table 3. For rs1447295 (region 1), when stratified by smoking, the only noteworthy change in OR was found in liver cancer (P = 0.025): OR_adj, 1.96 (95% CI, 1.07-3.59) among smokers and 0.90 (95% CI, 0.49-1.65) among never-smokers with an adjusted OR for interaction of 1.95 (95% CI, 1.09-3.51).

Table 4 shows the FPRP for the observed associations presented in Tables 2 and 3. Assuming a prior probability of 0.01, we find two of our observed associations below FPRP threshold of 50%: rs6983267 and UADT cancers has an 11% probability of being a false positive, and when stratified by tumor site, cancer of the oropharynx has a 22% probability of being a false positive. If we increase our prior probability to 0.1, the observed associations among ever-smokers and cancers of the lung and UADT as well as oropharynx also have a <50% probability of false positivity.

Positive associations were observed between rs6983267 and UADT cancers in this study. When analyzed by genotypes, the GG genotype was strongly associated with the UADT cancers after adjusting for potential confounders. There was a clear dose-response relationship between rs6983267 and the UADT cancers (p for trend = 0.0071). When stratified by tobacco smoking, the adjusted ORs were 1.56 (95% CI, 1.01-2.39) for never-smokers and 1.79 (95% CI, 1.23-2.61) for smokers. There is no clear indication that the rs6983267 modifies the association between tobacco smoking and UADT cancers, although the point estimate of the adjusted OR was slightly higher among smokers. Among UADT cancers when stratified by tumor site, both cancers of oropharynx and larynx were positively associated with the rs6983267. Similarly, no clear difference of the associations was found between smokers and nonsmokers for both tumor sites. Small sample sizes of esophageal and nasopharyngeal cancers did not enable us to estimate the associations precisely with rs6983267. Although no overall association was observed between rs6983267 and lung cancer, a positive association was found for smokers (OR_adj, 1.45; 95% CI, 1.05-2.00) and a null association for never-smokers, indicating possible effect modification of the rs6983267 on smoking and lung cancer. The SNP rs6983267 was inversely associated with bladder cancer (OR_adj, 0.52; 95% CI, 0.25-1.07). When stratified by tobacco smoking, the adjusted ORs were 0.35 (95% CI, 0.14-0.83) among smokers and 1.16 (95% CI, 0.28-4.77) among never-smokers, suggesting the possibility of effect modification.

Our observations that rs6983267 was positively associated with UADT cancers, independent of tobacco smoking, positively associated with lung cancer only among smokers, and inversely associated with bladder cancer dependent of tobacco smoking status implicates this SNP as an important candidate marker for smoking-related cancers with etiological heterogeneity (43, 44). Our observations suggest that the SNP rs6983267 may play an important role in tobacco-related carcinogenesis involving target specific carcinogens, and perhaps metabolic, DNA repair, and other related pathways. Among nonsmokers, we observed positive associations between rs16901979 (region 2) and UADT as well as oropharyngeal cancers, and between rs6983267 and UADT cancer. A higher proportion of UADT cancer cases are diagnosed among nonsmokers, which may be associated with HPV infection, alcohol drinking, and other factors such as genetic predisposition. Our results indicate that both rs16901979 and rs6983267 may play a role in non-smoking-related pathways of UADT cancers.

SNPs of the 8q24 chromosome are notable for their associations in prostate cancer (16-20, 25-27) and increasing evidence with colorectal cancer (25, 29, 30, 45); however, this region is one with few recognized genes and known functionality. 8q24 chromosome is located upstream of c-Myc proto-oncogene and located close to the pseudogene POU5F1P1. To our current knowledge, 8q24 SNPs have not been investigated in all of the mentioned nine smoking-related cancer sites, except bladder cancer (52). However, the 8q24 chromosome has often been observed to be amplified in liver (9), lung (11), kidney (46), bladder (13, 47), and oral (10, 48) cancers, suggesting that our results may not be due to chance. The clear association of 8q24 with prostate cancer suggests a potential hormone-related or other carcinogenic pathways, which may be associated with expression of microRNAs in the 8q24 region (49). Our results and those of previous studies shows that SNPs of “region 3” are more often observed to be associated in cancer sites other than prostate, indicating that this specific “region” may be involved in other carcinogenic pathways, such as a tobacco-related carcinogenic pathway, or a combination of different pathways. Recent studies have observed SNPs between 128.47 and 129.54 Mb, “region 3” to be associated colorectal and ovarian cancers (25, 28-30, 50). Ghoussaini et al. reported this cancer-associated “region” may be narrower than believed previously, spanning only 128.47 to 128.50 Mb (50). Further research will be required to determine whether 8q24 loci, specifically “region 3,” are associated with smoking-related carcinogenesis. Studies of SNPs in LD with rs6983267, SNPs within “region 3,” and those between 128.47 and 128.50 Mb in relation to smoking-related cancers may also be useful to detect new markers and reveal possible underlying biological mechanisms. Lastly, we cannot exclude the possibility that SNPs beyond “region 3” may also be associated with tobacco-related carcinogenesis and that our results for rs6982267 were due to its high minor allele frequency, providing us with more precision to detect the observed associations. Thus, functional studies and studies with larger sample sizes should be conducted to further investigate the association of these SNPs with smoking-related cancers.

Two minor deviations in HWE were observed (rs16901979 genotype distributions in African Americans and rs6983267 genotype distribution in the Chinese population); however, the allelic proportions remained consistent with the previously published literature (20, 24). Chance finding, selection bias, or laboratory genotyping error may potentially lead to the HWE deviations. Because we observed a high quality-control concordance rate for all three SNPs (>99%) in our laboratory, the possibility of genotyping error is unlikely. Controls in both LA and Taixing studies were randomly selected from the population at risk using algorithms to capture an accurate representation of their respective areas (34, 36). After removing the African American population in our analysis of rs16901979, we observed similar associations. The association between rs6983267 and liver cancer needs investigation by other studies.

Multiple comparison issue may be of concern from multiple testing of tumor sites and SNPs involved in this study. We have done the FPRP analyses. Using our FPRP cutoff of 50% and a prior probability of 0.01, it is likely that the observed association between rs6983267 and UADT cancers (FPRP = 11%), specifically cancer of the oropharynx (FPRP = 22%), is not due to chance from multiple hypothesis testing.

The potential of selection bias may exist due to the poor survival of many of these cancers: liver, esophagus, stomach, and lung (globally, these sites have survival rates <30%; ref. 51). The relatively low case participation rate was due to death before they were interviewed. For instance, among eligible lung cancer cases, 25% died before we could contact them. If 8q24 SNPs played a role in the prognosis of smoking-related cancer sites, selective survival bias would have affected our observed associations. Because of the lack of studies investigating such effects on the prognosis of these cancers, we were unable to estimate whether such bias was present in this study. The sample sizes in the Taixing study and in the stratified analyses of the LA study may affect the precision of our measurements. As a result, the interval estimates from both the Chinese study (>200 cases for each site and >400 controls) and the MSKCC study (172 cases/157 controls) may be imprecise. Among esophageal cancer analyses, we combined LA and Taixing study sites to increase precision and observed no obvious associations after adjusting for potential confounding variables. Strengths of our study include a relatively large sample size in our lung and UADT cancer sites, diverse populations allowing us to investigate ethnic-specific genotype distributions along multiple cancers, and the potential to adjust for a variety of confounders.

In conclusion, our results support the hypothesis that the 8q24 variants, particularly rs6983267, play a role in smoking-related cancer sites, particularly in UADT cancers and lung cancer among smokers. Laboratory-based functional studies and large epidemiologic studies in multiple populations should be conducted to further investigate the association of 8q24 SNPs with smoking-related cancers.

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.

We thank Drs. Thomas M. Mack and Wendy Cozen for contributions for the Los Angeles population-based lung and UADT case-control study and the studies' participants for their time and dedication.