Study Population

The Radiation Effects Research Foundation (RERF) conducted a cohort study within the AHS of approximately 20,000 atomic bomb survivors. The AHS is a clinical research program based on biennial health examinations established in July, 1957 as the clinical subcohort of the Life Span Study (LSS) cohort of atomic bomb survivors in Hiroshima and Nagasaki, Japan. AHS subjects were selected from the LSS cohort stratified according to distance from the hypocenter at the time of bombing and presence or absence of acute symptoms. Subjects for a broad immunology study including the genome study were selected from the AHS subjects because blood samples are not available from other members of the LSS. The clinical data collected during these examinations facilitate long-term follow-up studies of disease incidence and changes in physiological and biochemical endpoints, which benefit participants and contribute to health management of the atomic bomb survivors. Between 1981–2002, we obtained blood samples from 7,131 AHS participants who visited the clinic for examinations as part of the broader immunology study. After excluding subjects who had a history of first primary cancer at the time of blood collection, whose radiation dose could not be estimated, who were exposed in utero (organ doses not estimable), who were aged 80 or older at the time of blood collection or who refused to provide informed consent (84 subjects), 4,690 subjects remained for analysis [3,175 subjects provided informed consent (1,515 subjects who were deceased after blood collection were approved for this study by the RERF Ethical Committee)]: we call this the “cancer and immuno-genome (IMG) cohort”. As a part of the LSS study, incident cancer cases were detected by the Hiroshima Tumor and Tissue Registries and the Nagasaki Cancer Registries, which also provided data on histological classification. Histological classification of gastric cancer was based on the Japanese Research Society for gastric cancer classification until 1986, and subsequently on the WHO coding system (ICD-O, ICD-O-2 and ICD-O-3), which was converted into Lauren's classification as reported elsewhere (32, 33). Baseline for follow-up was defined as the date of the first blood sample for the IMG study (collection began in 1981). The end of follow-up was December 31, 2001, the latest date of complete cancer ascertainment as of the time the data analysis was initiated. Follow-up of individual subjects ended on the date of first primary cancer onset, date of death, or the end of cohort follow-up, whichever occurred first. During the follow-up period (maximum 21 years), 200 gastric cancer cases (93 intestinal, 96 diffuse and 11 other types) were identified.

Baseline characteristics of cases and cohort subjects are shown in Table 1. We analyzed linkage disequilibrium (LD) with 300 controls, performed association analysis between IL-10 haplotypes and plasma IL-10 levels in 644 noncancer cohort members, and conducted risk estimation for radiation and IL-10 haplotyping using 200 cases amidst the 4,690 IMG cohort subjects.

Ethical Consideration

This study was approved by the Ethical Committee (Human Investigation Committee) and by the Ethics Committee for Genome Research at the RERF.

Measurement of Plasma IL-10 Levels

We measured plasma IL-10 levels in the IMG cohort members using a highly sensitive enzyme-linked immunosorbent assay kit (Quantikine HS, R&D systems, Minneapolis, MN). The minimum detectable dose of IL-10 was 0.5 pg/mL. Six hundred forty-four subjects who did not have a cancer history were randomly selected from the IMG cohort members to exclude an effect of cancer history on the relationship between IL-10 haplotype and plasma IL-10 levels.

Identification and Genotyping of SNPs

The Celera Genomic database including Asian populations (34, 35) was used to screen haplotype-tagging (ht) SNPs in the IL-10 gene region, along with the detection of novel SNPs over the region using the NCBI database. We selected the 19 htSNPs with allele frequency >5% among Japanese. After examining allele frequency in the study population, we found that ten of the 19 htSNPs showed variant allele frequencies >5% in our study population. We therefore selected these ten htSNPs: IL-10 −92952C>T (IL10-1, rs1400986), −64871T>C (IL10-2, rs2073186), −32510A>G (IL10-3, rs4347211), −23055T>C (IL10-4, rs11583394), −14378T>C (IL10-5, rs880790), −1082T>C (IL10-6, rs1800896), −819G>A (IL10-7, rs1800871), −592T>G (IL10-8, rs1800872), +1177T>C (IL10-9, rs1518111), and +1589A>G (IL10-10, rs1554286). To reduce time and labor, 300 noncancer cohort members from the nonexposed group in the IMG cohort were analyzed to identify IL-10 haplotypes for these 10 htSNPs. Primers and probes for these htSNPs were designed using Primer Express software, version 2.1 (Applied Biosystems, Foster City, CA). The TaqMan-Allelic Discrimination method was used for the detection of SNPs. All of the assays were conducted in 384-well PCR plates. The principle of TaqMan Real-Time PCR assay system using fluorogenic probes and the 5′ nuclease is described by Livak (36). Amplification reactions (5 μl) were carried out in duplicate with 10 ng of template DNA, 1× TaqMan Universal Master Mix buffer (Applied Biosystems), 300 nM of each primer and 200 nM of each fluorogenic probe. Thermal cycling was initiated with 2 min incubation at 50°C, followed by a first denaturation step of 10 min at 95°C, and then by 40 cycles of 15 s at 95°C and of 1 min at 60°C. After PCR was completed, the plates were brought to room temperature and read in an ABI PRISM 7900 Sequence Detection System (Applied Biosystems). Results were analyzed using the Allelic Discrimination software (SNPAlyze, Dynacom, Yokohama, Japan).

Haplotype and Radiation Risk Analysis

Rate ratios or excess rate ratios for gastric cancer incidence were estimated using general risk models extending the standard proportional hazards time-to-event analysis for cohort follow-up data, with attained age as the underlying time axis and left truncation on age at the time of blood collection (37, 38). Analyses were performed using the Cox regression module (Peanuts) of Epicure (version 1.81, Hirosoft Inc., Seattle, WA). All statistical models were stratified on gender and included adjustment for year of birth, city (Hiroshima or Nagasaki) and smoking intensity (cigarettes per day). City was included in the background in all analyses to adjust for any potential confounders, since IL-10 haplotype-specific risk for radiation may possibly differ by city due to potential involvement of other gene polymorphisms that are mechanistically associated with IL-10 functions and differ by city. Information on smoking was obtained by interview at the time of blood collection. IL-10 haplotypes were determined as described below. Atomic bomb radiation dose in weighted Gray (wGy) was estimated using the DS02 dosimetry system (39), based on weighted skin dose computed as the gamma dose plus 10 times the neutron dose. A radiation dose <0.005 Gy was called nonexposed when performing analyses based on dose group.

Regression analysis of gastric cancer incidence was based on histological subtype – intestinal or diffuse. Cases in this study were primary gastric cancers, which included 17 synchronous multiple cancer cases that experienced primary cancer(s) of other organs within one year before or after gastric cancer diagnosis. Censoring was imposed at the time of other primary cancer diagnosis, death or the end of cancer follow-up (December 31, 2001). An excess relative risk (ERR: excess rate ratio) model was used for radiation risk estimation with adjustment for gender using stratification and adjustment for smoking intensity and year of birth using a log-linear model.

For IL-10 haplotypes, indicator variables for the three categories (major homozygote, heterozygote and minor homozygote) were used to estimate either the rate ratio (RR, in the case of multiplicative models using the standard log-linear Cox model for IL-10 haplotypes) or ERR (in the case of additive models for the joint effect of haplotyping and radiation), with the major homozygotes treated as the reference category. In this study, we considered only heterozygotes involving the most frequent (major) and second-most frequent (minor) alleles among the numerous heterozygotes that appeared in the process of LD analysis. Four genomic models for IL-10 haplotype effects were considered: arbitrary (no structure among the three IL-10 haplotypes; the “categorical” genomic model); “linear” (a codominant model where the effect in heterozygotes is assumed to be one-half of the effect in minor homozygotes); “dominant” (the presence of one or two minor haplotype alleles confers the same risk); and “recessive” (risk occurs only in minor homozygotes). Statistical interaction (effect estimate modification) between radiation and IL-10 haplotypes was tested on both multiplicative and additive scales (40). The multiplicative joint risk model is r(g,d) = e^δg[1 + βd] and the additive joint risk model is r(g,d) = 1 + φg + γd, with g being the haplotype (a two- or three-level factor) and d the radiation dose (a continuous variable). Both models specify a rate ratio r(g,d) that multiplies the background incidence (incidence among nonexposed persons).

Identification of Haplotype Blocks

The ten htSNPs (IL-10-1 to −10) are located on one gene region spanning about 95 kb in length (Fig. 1A). When we examined all combinations between the ten htSNPs for LD, four of ten htSNPs formed one haplotype block with r² values greater than 0.9 (Fig. 1B) and generated two major haplotype alleles, wild allele IL-10 –ATTA and variant one IL-10 -GGCG (IL-10-7, −8, −9 and −10). These two haplotype alleles constituted 98.1% of all haplotype alleles available in this block, 65.9% for wild IL-10 -ATTA and 32.2% for variant IL-10 -GGCG. Posterior probabilities for these two haplotype alleles exceeded 0.99, so these haplotypes were treated as known.

IL-10 Haplotypes and Plasma IL-10 Levels

As these htSNPs are located in untranslated regions including the promoter regions, we examined the relationship between IL-10 haplotypes and plasma IL-10 levels in 664 cancer-free IMG cohort members. Mean plasma IL-10 protein levels were 2.79 pg/ml (SD 1.85, CV 0.68) among persons with the major homozygote IL-10 haplotype, 3.02 pg/ml (SD 2.36, CV 0.78) among heterozygotes and 3.69 pg/ml (SD 2.97, CV 0.81) among minor homozygotes. Using log transformed IL-10 protein levels, regression analysis revealed no difference in IL-10 protein levels between major homozygotes and heterozygotes (P = 0.39), but minor homozygotes had significantly higher levels (P = 0.030). Using a linear model for IL-10 haplotype, there was a significant trend in number of minor haplotyope alleles (P = 0.034). Radiation dose was related to log plasma IL-10 protein levels with a relative change in actual protein levels (not log transformed) of e^0.1278 = 1.14 (14% increase) at 1 Gy (P < 0.001). These results suggest that plasma IL-10 levels varied not only by genetic factors but also by radiation exposure, and might therefore be closely associated with gastric cancer risk.

The incidence rate ratio of all gastric cancer combined for women compared with men was 0.41 [95% confidence interval (CI) 0.28, 0.58; P < 0.001]. Incidence increased with smoking intensity (RR = 1.24 among smokers of 10 cigarettes/day compared with nonsmokers; 95% CI 1.07, 1.43; P < 0.001) and year of birth (rate increased with more recent years of birth; P < 0.001). With those adjustments, crude RRs for IL-10 haplotypes were: 1.26 (ATTA/GGCG heterozygote; 95% CI 0.91, 1.75; P = 0.17) and 1.59 (minor homozygote; 95% CI 0.98, 2.50; P = 0.062), with heterogeneity test P = 0.13. The linear genomic model was statistically significant: risk for the minor homozygote was 1.60 (95% CI 1.03, 2.48; P = 0.038), with the risk for the ATTA/GGCG heterozygote being one-half of the risk for minor homozygote on the log scale (i.e., RR = exp{[ln(1.60)]/2} = 1.26). The other two types of one-parameter genomic models (dominant and recessive) did not produce significant effects (P = 0.066 with the dominant model, P = 0.13 with the recessive model).

There was no evidence of interaction between gender and IL-10 haplotypes (P = 0.23). Crude (adjusted only for gender, smoking and birth year, but not for IL-10 haplotypes) ERR of all gastric cancers for radiation was 0.11 (95% CI −0.05, 0.34; P = 0.22) and after adjustment for IL-10 haplotypes using the categorical genomic model the ERR for radiation was 0.12 (95% CI −0.05, 0.36; P = 0.19). The main effect of IL-10 haplotypes after adjustment for radiation was essentially the same as the crude risk with each genomic model. The best fitting genomic model, and only statistically significant parameter, was with the linear genomic model. However, the apparent effects of radiation and IL-10 haplotypes with all gastric cancers combined reflect large differences in their effects according to subtypes (see below), hence results for interaction between these two factors with all gastric cancers combined are not reported.

Results for Intestinal Type Gastric Cancer

Crude proportions of cohort members with intestinal type gastric cancer increased with number of variant (GGCG) haplotype allele: 1.5% among those with the major (ATTA) IL-10 homozygote, 2.0% with the ATTA/GGCG heterozygote, and 3.9% with the minor homozygote. Incidence was about twice as high in Hiroshima as in Nagasaki (RR for Nagasaki 0.41; 95% CI 0.22, 0.71; P = 0.001) and was significantly related to smoking intensity (RR for smoking 10 cigarettes/day 1.38; 95% CI 1.12, 1.66; P = 0.004).

The cohort distribution of haplotypes (major homozygote, ATTA/GGCG heterozygote, minor homozygote) was similar in the two cities: (0.45, 0.44, 0.11) in Hiroshima and (0.46, 0.43, 0.11) in Nagasaki. Given no apparent association between city and IL-10 haplotypes, the city difference would not be expected to confound the IL-10 risk, and indeed there was no noticeable difference in IL-10 risk parameters depending on whether or not city was included in the background model. There was no evidence of an effect of radiation: after adjustment of background incidence for gender, birth year, city and smoking (not adjusting for IL-10 haplotypes), radiation ERR/Gy was −0.06 (95% CI −0.19, 0.21; P > 0.5).

Tables 2 and 3 show the jointly estimated risks for radiation and IL-10 haplotypes with intestinal type gastric cancer. Main effects for radiation and haplotypes were similar with either the multiplicative or the additive model, so ERRs for radiation and haplotypes in the additive model are not shown. With intestinal type gastric cancer, there was evidence of a significant effect of IL-10 haplotypes after adjustment for radiation (Table 2). As with all gastric cancers combined, the fit of the categorical genomic model was similar to that of the linear genomic model, but the RRs were larger than with all gastric cancers combined. The linear genomic model had the lowest value of Akaike information criterion (AIC), and the categorical model produced risk estimates for the minor homozygote and the ATTA/GGCG heterozygote that were consistent with those from the linear model (compare exp{ln[2.16]/2} = 1.47 from the linear model with 1.55 for major heterozygotes).

Although the intestinal subtype evidenced no overall effect of radiation, it is possible that radiation effects might exist if there were biologically significant interaction between radiation and IL-10 haplotypes. It was not possible to estimate or test interaction between radiation and particular IL-10 haplotypes with the categorical genomic model, but none of the estimable radiation ERRs with interaction in the multiplicative model were very far from 0 (not shown) and none of the interaction tests for the one-parameter genomic models produced significant results (Table 2). With the additive model, again there were no statistically significant interactions (Table 3) and there was no evidence of trend in radiation ERRs with frequency of minor haplotype allele (radiation ERR was higher with the ATTA/GGCG heterozygote relative to the major homozygote but lower with the minor homozygote; Table 3).

Results for Diffuse Type Gastric Cancer

Crude proportions of cohort members with diffuse type gastric cancer evidenced no association with IL-10 haplotypes: 1.9% with the major (ATTA) homozygote, 2.3% with the ATTA/GGCG heterozygote and 2.0% with the minor (GGCG) homozygote. Incidence of diffuse type did not depend on city (P = 0.49) and there was no impact on rate ratios for IL-10 haplotypes depending on whether or not city was included in the background model. Smoking was not significantly related to diffuse type gastric cancer: RR for smoking 10 cigarettes/day compared with nonsmokers was 1.15 (95% CI 0.91, 1.40; P = 0.23). However, smoking was left in the analyses to avoid potential confounding. The crude ERR for radiation (after adjustment for gender, birth year, city and smoking but not for IL-10 haplotypes) was statistically significant (ERR/Gy = 0.33; 95% CI 0.03, 0.83; P = 0.027).

Tables 4 and 5 show the jointly estimated risks for radiation and haplotype for diffuse type gastric cancer. With diffuse type, there was no evidence of an effect of IL-10 haplotypes after adjustment for radiation (Table 4). There was no significant evidence of interaction between radiation and IL-10 on either the multiplicative (Table 4) or additive (Table 5) scale. However, the haplotype-specific ERR for radiation was statistically significant only in the major homozygote category of IL-10 (Table 5), being higher than the overall ERR for radiation, whereas estimated ERR for radiation with the minor IL-10 homozygote was close to 0 and nonsignificant. Thus, the minor IL-10 haplotype might act to reduce radiation-related risk of diffuse type gastric cancer.