Studied Subjects

The Diet, Cancer and Health Study is an ongoing Danish cohort study designed to investigate the relation between diet, lifestyle and cancer risk [19]. The cohort consists of 57,053 persons, recruited between December 1993 and May 1997. All the subjects were born in Denmark, and the individuals were 50 to 64 years of age and had no previous cancers at study entry. Blood samples and questionnaire data on diet and lifestyle were collected at study entry.

Follow-up and endpoints

Follow-up was based on population-based cancer registries. Between 1994 and 31th December 2009, nine hundred and seventy CRC cases were diagnosed. A subcohort of 1897 persons was randomly selected within the cohort. Of these, 108 with missing genotype data were excluded. All information on genotypes and diet and lifestyle factors was available for nine hundred and seventy CRC cases and 1789 subcohort members.

Dietary and lifestyle questionnaire

Information on diet, lifestyle, weight, height, medical treatment, environmental exposures, and other socio-economic factors were collected at enrolment using questionnaires and interviews. In the food-frequency questionnaire, diet consumption was assessed in 12 categories of predefined responses, ranking from ’never’ to ’eight times or more per day’. The daily intake was then calculated by using FoodCalc [19]; this program uses population specific standardized recipes and portion sizes. Intake of red meat in grams per day was calculated by adding up intake of beef, veal, pork, lamb and offal. Intake of processed meat in grams per day was calculated by adding up intake of processed red meat, including bacon, smoked ham, salami, frankfurter, Cumberland sausage, cold cuts and liver pâté. Dietary fibre intake was based on country-specific food composition tables, which were reviewed to ensure comparability to the association of official analytical chemists (AOAC) fibre definition, which includes lignin and resistant starch [20]. Fibre intake is calculated by multiplying the frequency of consumption of relevant foods by their fibre content as determined from national databases of food content [21].

Contributing food items to the food group ‘cereals’ included wholegrain foods (wholegrain bread, rye bread, wholegrain flour, oatmeal, corncobs, müsli, and crispbread) and refined grain foods (white wheat bread, wheat flour, rice flour, potato flour, corn flour/starch, pasta, wheat) and was measured in grams per day [22]. Intake of ‘fish’ in grams per day was calculated by adding up intake of fresh and processed fish. For fruit, only intake of fresh fruit (as indicated on the FFQ) was examined, while vegetable intake also included estimated contributions from recipes. The questionnaire was tested in a validation study preceding the Diet, Cancer and Health study. Pearson correlation coefficients (adjusted for total energy intake) illustrating the comparison of nutrient scores estimated from the food-frequency questionnaire and from weighed diet records were 0.39 and 0.53 for dietary fibre and 0.37 and 0.14 for meat for men and women, respectively [23,24].

Smoking status was classified as never, past or current. Persons smoking at least 1 cigarette daily during the last year were classified as smokers.

The lifestyle questionnaire included this question regarding use of NSAID: “Have you taken more than one pain relieving pill per month during the last year?” If the answer was yes, the participant was asked to record how frequently they took each of the following medications: “Aspirin”, “Paracetamol”, “Ibuprofen”, or “Other pain relievers”. The latter category included NSAID preparations other than aspirin and ibuprofen. Based on all records, we classified study subjects according to use of “any NSAID” (≥ 2 pills per month during one year) at baseline.

Genotyping

Buffy coat preparations were stored at minus 150°C until use. DNA was extracted as described [25]. The DNA was genotyped by KBioscience (KBioscience, Hoddesdon, United Kingdom) by PCR-based KASP™ genotyping assay. (http://www.lgcgenomics.com/). One marker polymorphism, the IL10 C-rs3024505-T, and 8 functional polymorphisms were selected; IL10 C-592A (rs1800872), IL1B C-3737T (rs4848306), G-1464C (rs1143623), T-31C (rs1143627) and PTGS2 (encoding COX-2) A-1195G (rs689466), G-765C (rs20417), and T8473C (rs5275). IL1B T-31C (rs1143627) [18], PTGS2 (encoding COX-2) A-1195G (rs689466), G-765C (rs20417), and T8473C (rs5275) [3] were determined and reported previously for a subset of the study group. Furthermore, all the three SNPs in IL1B were determined for the whole comparison group independently of the present work and reported previously [26]. To confirm reproducibility, genotyping was repeated for 10 % of the samples yielding 100% identity.

Statistical Analysis

Deviation from Hardy-Weinberg equilibrium was assessed using a Chi square test.

Incidence rate ratios (IRR) and 95% Confidence Interval (95%CI) were calculated according to the principles for analysis of case-cohort studies using an un-weighted approach [27]. Age was used as the time scale in the Cox regression models. Tests and confidence intervals were based on Wald’s tests using the robust estimate of the variance-covariance matrix for the regression parameters in the Cox regression models [28] as previously described [3,5,17,18,29–34].

All models were adjusted for baseline values of suspected risk factors for colorectal cancer such as body mass index (BMI) (kg/m², continuous), NSAID (yes/no), use of hormone replacement therapy (HRT) (never/past/current, among women), smoking status (never/past/current), intake of dietary fibre (g/day, continuous), and red meat and processed meat (g/day, continuous). Cereals, fibre, fruit and vegetables were also entered linearly. All analyses were stratified by gender, so that the basic (underlying) hazards were gender specific. For all the polymorphisms, IRR was calculated separately for heterozygous and homozygous variant allele carriers. For all the SNPs except for PTGS2 A-1195G, all variant allele carriers were subsequently grouped for interaction analyses since no recessive effects were observed. For PTGS2 A-1195G, a recessive mode was used in the subsequent analyses.

Haplotypes of PTGS2 and IL1B were inferred manually as done previously [3,35,36].

For the different genes, we investigated possible interactions between the polymorphisms and intake of meat, dietary fibre, cereals, fish, fruit and vegetables, smoking status and NSAID use using the likelihood ratio test [3,14,32,35–37].

In another set of interaction analyses between the polymorphisms and the dietary intake subdivided in tertiles, dietary intake was entered as a categorical variable. Tertile cut-points were based on the empirical distribution among cases. The possible interactions were investigated using the likelihood ratio test.

All analyses were performed using R version 2.15-1 (R Core Team, 2013) [38]. A p<0.05 was considered to be significant.

Ethics Statement

All participants gave verbal and written informed consent. The Diet, Cancer and Health study was approved by the National Committee on Health Research Ethics (journal nr. (KF) 01-345/93) and the Danish Data Protection Agency.

Associations between polymorphisms and CRC

IL1B C-3737T variant allele carriers were at lowered risk of CRC and the G-1464C variant allele carriers were at higher risk of CRC compared to the homozygous wildtype genotype carriers (p=0.02 and p=0.02, respectively) (Table 2). Haplotype analyses revealed that the IL1B haplotype combinations which included the CCC haplotype (C-3737T, G-1464C, T-31C) were associated with increased risk of CRC compared to the reference TGT/TGT haplotype (Table S1). However, only the haplotype combination CCC/CGT were statistically significantly associated with risk of CRC (p=0.02). Carriers of one copy of the haplotype CCC had an IRR of 1.20 (p=0.04) and carriers of two CCC haplotypes had an IRR of 1.29 (p=0.12) (reference group: no CCC haplotype) (Table 3). Carriers of one copy of the TGT haplotype had an IRR of 0.82 (p=0.04) and carriers of two TGT copies had an IRR of 0.79 (p=0.05) (Table 3).

Carriers of the high COX-2 activity PTGS2 T8473C variant allele were at lower risk of CRC and homozygous carriers of the low COX-2 activity PTGS2 A-1195G variant G-allele were at marginally higher risk of CRC than the homozygous wildtype genotype (p=0.02 and p=0.07, respectively) (Table 2). Furthermore, carriers of the haplotype combination which included both copies of the A-1195G variant alleles (GGT/GGT), were at increased risk of CRC (p=0.09) compared to the reference PTGS2 AGT/AGT (A-1195G, G-765C, T8473C) haplotype combination (Table S1). In a separate analysis, carriers of one GGT copy had an IRR of 1.06 (p=0.51) whereas carriers of two copies had an IRR of 1.62 compared to all non-carriers of the haplotype (p=0.02) (Table 3).

Where no recessive effects were observed, variant genotypes were combined in the interaction analysis to maximize the statistical power. A recessive effect was found for PTGS2 A-1195G and consequently, AA and AG carriers were grouped versus GG carriers.

Gene-environment analyses

Associations between polymorphisms and CRC

We now extend our previous studies of IL10, IL1B and PTGS2 polymorphisms in relation to diet and colorectal carcinogenesis in a study group of three hundred and seventy-eight CRC cases and 775 participants in a randomly selected comparison group [3,17,18] studies to a larger cohort with more than twice the number of cases and members of the comparison group and include more dietary factors. In contrast to our previous findings [3,18], the studied functional polymorphisms in IL1B and PTGS2 were associated with risk of CRC in the present larger study group probably reflecting the increased statistical power. We reproduced the previous findings that the studied IL10 polymorphisms were not associated with risk of CRC per see 17.

In the present study, the IL1B CCC (C-3737T, G-1464C, T-31C) haplotype was associated with increased risk of CRC. This haplotype gives high transcription levels in studies of the promoter using transient transfections [39]. Conversely, the TGT haplotype was associated with low CRC risk. This haplotype leads to low IL1B transcription and encompasses the variant allele of C-3737T. The IL1B C-3737T polymorphism abolishes a binding site for the anti-inflammatory NF-κB subunit p50 [39]. Thus, our results, both in terms of SNP analyses and haplotype analyses consistently indicate that genetically determined high IL1B levels are associated with increased risk of CRC and genetically low IL1B levels leads to lowered risk of CRC. Furthermore, our result may suggest involvement of the anti-inflammatory p50 subunit of NF-κB. We found no statistically significant interaction between the studied IL1B polymorphisms and any of the studied diet variables. However, in the tertile analysis of vegetable intake (table S2) the association between the IL1B polymorphisms and risk of CRC was strongest in the lowest tertile. Thus, carriers of the variant allele of IL1B G-1464C and T-31C were at 1.40 (95%CI: 1.05-1.86) and 1.47 (95%CI: 1.11-1.95) fold increased risk of CRC, respectively, whereas carriers of the variant allele of C-3737T were at reduced risk of CRC (IRR=0.66, 95% CI: 0.49-0.89). This may suggest that the association between IL1B polymorphisms and risk of CRC can only be detected in populations with relatively low vegetable intake such as the Danish population [40].

The PTGS2 A-1195G variant allele leads to low transcription levels of COX-2 [41] whereas PTGS2 T8473C gives high mRNA levels [42]. In Danes including the present study group, the variant allele of G-765C almost exclusively co-segregates with the variant allele of T8473C (Table S1). This haplotype has been shown to be associated with highly elevated COX-2 activity [43]. In the present study, the PTGS2 GGT (A-1195G, G-765C, T8473C) haplotype was associated with increased risk of CRC (P=0.024). The PTGS2 A-1195G homozygous variant genotype was marginally associated with increased risk of CRC (P=0.07) and PTGS2 T8473C variant carriers with genetically determined high COX-2 activity were at lowered risk of CRC (P=0.02). In accordance with the present study, genetically low COX-2 activity was found to predispose to inflammatory bowel disease, a risk factor for CRC [44].

Gene-environment analyses

The intake of meat in the Danish population is among the highest intakes world-wide and we have previously identified interactions between meat and genes [3,5]. We found interaction between intake of meat and PTGS2 G-765C. Thus, among variant allele carriers, daily intake of meat was associated with 8% increased risk of CRC pr 25g meat, whereas homozygous wildtype allele carriers were not at increased risk (P_int=0.006). The result is in accordance with the finding of a statistically significant association between PTGS2 G-765C variant genotypes and CRC among subjects with high n-6 PUFA intake in a prospective, population-based cohort of 310 Singapore Chinese cases [45]. N-6 PUFA is present in meat. We also found interaction between meat intake and IL10 rs3024505. A similar interaction was found for fish intake. However, it is difficult to interpret the functional implications as IL10 rs3024505 is a marker SNP with no known function. The lack of interaction with the functional promoter polymorphism C-592A may suggest that the detected interaction may be related to other genes than IL10, but on the other hand, IL10 rs3024505 is located very far away from other genes [46]. In summary, the interactions between meat intake on one hand and genetic variation in PTGS2 and IL10 on the other hand, suggest that inflammation plays a role in meat related carcinogenesis. In support of this, we have also found interaction between the functional promoter polymorphism NFKB1 -94ins/del and meat intake in relation to CRC [5].

We observed strong interaction between the marker IL10 rs3024505 and intake of fibre and vegetables. In both cases, homozygous wildtype allele carriers benefited from high intake, whereas variant allele carriers had no risk reduction when eating fibre or vegetables in relation to CRC risk. However, since variant allele carriers in the tertile with the lowest fibre intake were at marginally lowered risk of CRC (IRR: 0.73, 95% CI: 0.54-1.01) the results suggest that wildtype allele carriers experience a risk reduction by fibre intake that variant allele carriers already have. The results support and extend our previously finding of interaction between fibre and IL10 in a subcohort of the present study cohort [17]. Similarly, we observed interactions between dietary fibre, dietary cereals and fruit on one hand and genetic variation in PTGS2 on the other. The interactions are quite consistent and suggest that subjects with genetically low PTGS2 activity benefit the most from high intake of fibres, fruit, and cereals. Furthermore, tertile analyses showed that those with the genetically determined lowest COX-2 activity, namely homozygous carriers of the variant allele of PTGS2 A-1195G, were at high risk of CRC in the group with the lowest intake of fibres (IRR=3.08 (95%CI:1.51-6.28), and fruits (IRR=2.11, 95%CI: 1.03-4.33), whereas those with the genetically determined high COX-2 activity, carriers of the variant allele of PTGS2 G-765C, were at low CRC risk even in the tertile with the lowest fibre intake (IRR=0.69, 95% CI: 0.50-0.96). Thus, COX-2 seems intimately implicated in the biological mechanism underlying the protective effect of fibres in relation to CRC.

We found interaction between NSAID use and IL1B C-3737T in relation to development of CRC suggesting that NSAID intake reduce the risk of CRC among those with high risk of CRC due to genetically determined high IL1B level. We found no statistically significant interaction between NSAID and COX-2 in relation to CRC. A non-statistically significant tendency towards protection by NSAID use among those with genetically low COX-2 activity was found. Long-term intake of aspirin (a COX-1 inhibitor) has been found to confer protection against CRC including the presently used Diet, Cancer and Health cohort [47]. It was not possible to assess the effects of specific COX-2 inhibitors due to late introduction to the Danish market and low frequency of use in the follow-up period [47,48].

The biological interpretation of our results is supported by other findings. IL-1B, IL-10, and COX-2 are part of the same inflammatory pathways. IL-1 has been found to induce the synthesis of COX-2 through activation of the pro-inflammatory p65 unit of nuclear factor κB (NF-κB) [49]. Furthermore, IL-10 has been found to block IL-1-induced NF-κB activation in intestinal cells (by inhibiting IκB phosphorylation) and to reduce COX-2 induction in intestinal cells [49], The latter is in accordance with the finding that cox-2 expression is high in il-10 deficient mice [7]. Therefore, diet such as fibre may modify IL10 which act as an inflammatory “gate-keeper” and thereby affect inflammation.

Furthermore, our results suggest that those with genetically determined low COX-2 activity are at high risk of CRC by smoking and meat intake and, furthermore, protected by fibre intake. Thus PTGS2 polymorphisms may have differential impact on CRC risk dependent on environmental factors. However, once carcinogenesis has been initiated, a high COX-2 enzyme activity seems to be a risk factor for further progression [7,50,51].

Taken together, our interaction analyses suggest that diet modify intestinal carcinogenesis through impact on inflammatory response and furthermore suggest that the effect may differ among various populations depending on gene-environment interactions. Our findings should be explored in other well-characterized prospective cohorts.

This study used a nested prospective case-cohort design and has the major advantage that data and samples were collected before diagnosis thus minimizing the risk of differential misclassification between cases and comparison group. The risk estimates were adjusted for known confounding factors affecting risk of CRC in this cohort including dietary factors, body mass index (BMI), alcohol, smoking status and NSAID use. A main strength of the study is the large sample size. The genes were carefully selected based on their role in the inflammatory pathway and the polymorphisms were mainly selected based on their functional effects in order to allow interpretation of the involved biological pathways in colorectal carcinogenesis. Only the interactions between fibre and IL10, fibre and PTGS2, vegetables and IL10, and fruit and PTGS2 withstood correction for multiple analyses. However, as our hypothesis was biologically based we did not correct for multiple analyses [52]. In the light of the number of statistical tests performed, we would expect that some of the findings may be due to chance, but the number of statistically significant findings exceed the number expected by chance.