Genome-wide gene-environment interaction analyses to understand the relationship between red meat and processed meat intake and colorectal cancer risk

Abstract

Background:

High red meat and/or processed meat consumption are established colorectal cancer (CRC) risk factors. We conducted a genome-wide gene-environment (GxE) interaction analysis to identify genetic variants that may modify these associations.

Methods:

A pooled sample of 29,842 CRC cases and 39,635 controls of European ancestry from 27 studies were included. Quantiles for red meat and processed meat intake were constructed from harmonized questionnaire data. Genotyping arrays were imputed to the Haplotype Reference Consortium. Two-step EDGE and joint tests of GxE interaction were utilized in our genome-wide scan.

Results:

Meta-analyses confirmed positive associations between increased consumption of red meat and processed meat with CRC risk (per quartile red meat OR = 1.30; 95%CI = 1.21–1.41; processed meat OR = 1.40; 95%CI = 1.20–1.63). Two significant genome-wide GxE interactions for red meat consumption were found. Joint GxE tests revealed the rs4871179 SNP in chromosome 8 (downstream of HAS2); greater than median of consumption ORs = 1.38 (95%CI = 1.29–1.46), 1.20 (95%CI = 1.12 −1.27), and 1.07 (95%CI = 0.95 – 1.19) for CC, CG and GG, respectively. The two-step EDGE method identified the rs35352860 SNP in chromosome 18 (SMAD7 intron); greater than median of consumption ORs = 1.18 (95%CI = 1.11–1.24), 1.35 (95%CI = 1.26–1.44), and 1.46 (95%CI = 1.26–1.69) for CC, CT, and TT, respectively.

Conclusions:

We propose two novel biomarkers that support the role of meat consumption with an increased risk of CRC.

Impact:

The reported GxE interactions may explain the increased risk of CRC in certain population subgroups.

INTRODUCTION

Colorectal cancer (CRC) is currently the third most common cancer worldwide, and second leading cause of cancer death (1). It is estimated that at least ~50% of CRC cases and CRC deaths could be attributed to modifiable lifestyle factors (2–4). The main established modifiable CRC risk factors are high consumption of processed meat and/or red meat, consumption of alcoholic drinks, smoking, being overweight or obese, low consumption of foods containing dietary fiber, low consumption of whole grains, and low physical activity (5–7). Based on the existing literature, the World Cancer Research Fund has concluded that there is strong evidence that red meat and/or processed meat consumption increases the risk of CRC (5). Moreover, based on epidemiological, animal, and mechanistic data, the International Agency for Research on Cancer (IARC) classified consumption of processed meat as a group 1 carcinogen (i.e., an established cause of CRC), and red meat as a group 2a carcinogen (i.e., a probable cancer agent) (6,8). These classifications were in great part based on the evidence for CRC.

There are several mechanisms that have been proposed to explain the relationship between consumption of red or processed meat and CRC risk. Among them is the presence of carcinogenic N-nitroso compounds (NOC) in processed meat that can also be formed endogenously in the gut after consumption of red meat (9,10). The abundance of heme iron in red meat, combined with the presence of gut bacteria can facilitate this carcinogenic process. NOC primarily produced through bacterial decarboxylation of amino acids in the presence of a nitrosating agent may cause damage and inflammation to the gut lining. Furthermore, the existence of a persistent intestinal dysbiosis might exacerbate the carcinogenic process, as it has been linked to chronic gut inflammation (11–14). Additional carcinogens linked to red meat and/or processed meat are heterocyclic amines (HCAs) (15) and polycyclic aromatic hydrocarbons (PAHs) (16), which can be formed by different cooking methods (17,18).

In addition to modifiable risk factors, there are common genetic variants that have been linked to CRC. To date, there are over 200 genetic variants that were identified with genome-wide association studies (GWAS) that altogether explain ~20% of the heritability in CRC risk (19–21). Gene-environment (GxE) interactions refers to the phenomenon in which the effects of genetic variations at an individual level are modified by environmental factors and vice versa. In other words, it is the concept that environmental factors can impact the expression/function of genes, and that genetic factors can influence a person’s sensitivity or reaction to environmental factors. It has been speculated that these GxE interactions may explain part of the large fraction of missing heritability (22). There have been multiple studies that have explored the role of common genetic variants as potential modifiers of the effect of processed meat or red meat (23), most of them focused on candidate genes and single nucleotide polymorphisms (SNPs). There have only been two genome-wide GxE scans for red and processed meat (24,25), and only one of these reported a significant GxE interaction with processed meat (24). A more recent review and evaluation of all available data assigned this finding a moderate plausibility score and overall concluded that most studies of GxE in CRC to date have been underpowered to detect GxE interactions (23).

In this study, we applied powerful methods in a genome-wide GxE scan to identify possible interactions between common variants and red meat and/or processed meat intake and CRC risk, using a large CRC pooled dataset.

MATERIALS and METHODS

Study participants

A total of 27 studies (17 prospective cohorts; 10 case-controls) were included in this study from three CRC genetic consortia: the Genetics and Epidemiology of Colorectal Cancer Consortium (GECCO), the Colorectal Cancer Transdisciplinary Study (CORECT) and the Colon Cancer Family Registry (CCFR) (19,20,26) (Supplementary Table 1). For cohort studies, nested case-control sets were assembled via risk-set sampling, while population-based controls were used for case-control studies. Controls were mainly matched on age (study-specific range), sex, race, and enrollment date/trial group, when applicable (for the SELECT trial). Cases were defined as colorectal adenocarcinoma (the most common type of CRC) and were confirmed by medical records, pathology reports, or death certificate information. All participants gave written informed consent and studies were approved by their respective institutional review boards. Analyses were limited to individuals of European ancestry, based on self-reported race and ethnicity, and clustering of principal components with the 1000 Genomes Europeans (EUR) superpopulation. We excluded individuals based on cryptic relatedness or duplicates, genotyping/imputation errors, and age outliers. Individuals missing both red meat and processed meat intake or any of the adjustment covariates we considered were excluded. We further excluded studies of individuals diagnosed with advanced adenomas only. The final pooled sample included 29,842 CRC cases and 39,635 controls.

Environmental exposure data

Lifestyle and environmental risk factors were collected by in-person interviews, phone interviews or structured self-administered questionnaires. Harmonized quality-control checks were performed in each study following similar criteria (27). Common data elements (CDEs) were defined a priori and through an iterative process, and responses from study questionnaires and data dictionaries were mapped to these CDEs. All definitions, permissible values, and standardized coding were tracked in a database via SAS and T-SQL. Data checks were performed to identify outliers and other errors. We developed a pooled variable that tracked consumption of red meat intake (beef, pork, lamb), and another one that tracked consumption of processed meat (bacon, sausages, luncheon/deli meats, hot dogs), each expressed as servings per day. Many of the red meat variables in the pooled studies had included processed meats as part of the red meat definition. Of the total study population, 64,152 individuals had both red and processed meat consumption information; 5,325 had information only on red meat intake. We evaluated these two exposures using sex- and study-specific quartiles in which study-specific quartiles were computed and then the median of intake within each quartile was obtained. Body Mass Index (BMI) was calculated as the weight of each participant (in kg) divided by the square of height (in m²). A BMI variable was used that captured 5 kg/m² increments. In addition, the World Health Organization (WHO) pre-defined BMI cut-points were used: normal weight (18.5–<25 kg/m²), overweight (>=25.0–<30 kg/m²), and obesity (>=30 kg/m²) (28). Total energy intake was calculated from food frequency questionnaires. For studies with partially missing data, total caloric intake was imputed using the mean value in the study. Studies with missing energy intake estimates were set to zero.

Genotyping and imputation

Details on quality control and genotyping metrics were previously published (19,26). Several genotyping arrays were used and are summarized in Supplementary Table 1. Briefly, exclusion criteria included: single nucleotide polymorphisms (SNPs) with a missing call rate of >2-to-5%, departure from Hardy-Weinberg equilibrium (HWE) (P<1×10−4), inconsistencies between self-reported and genotyped sex, and discordant genotype calls within duplicate samples. Genotypes were imputed to the Haplotype Reference Consortium (HRC) panel (39.1 million variants) using the University of Michigan Imputation Server (29,30), and converted into a binary format for data management and analyses using the BinaryDosage R package (https://cran.r-project.org/web/packages/BinaryDosage). Imputed SNPs were restricted based on a pooled minor allele frequency (MAF) ≥1% and imputation accuracy (R²>0.8). After imputation and quality control, a total of approximately 7.2 million SNPs were selected for analyses. Principal component analysis (PCA) for population stratification assessment was performed using PLINK1.9 on 30,000 randomly sampled imputed SNPs with MAF > 5% and R²>0.99.

Statistical analyses

The associations of red meat or processed meat intake variables with CRC risk were assessed by meta-analysis of study-specific estimates, adjusted by sex, age and total energy intake. Between-sex, between-tumor site, between-study design as well as between-study heterogeneity and inconsistency were investigated using the heterogeneity Chi-squared and I² statistic (31). Between-study heterogeneity represents the proportion of total variation in effect estimates attributable to between-study variance. Potential outlier studies were assessed by estimating the posterior probability of outliers based on mixture random effects using the “outlierProbs” function of metaplus R package (32). No outlier studies (posterior probability >0.99) were identified. We also assessed the relationship between red meat or processed meat with CRC risk stratified by sex, tumor localization (proximal colon, distal colon, or rectum) and study design (case-control or cohort study).

To identify novel GxE interactions for CRC, we performed genome-wide scans using the GxEScanR R package (https://cran.r-project.org/web/packages/GxEScanR), which implements several interaction testing methods. Imputed allelic dosages were modelled as continuous variables. In addition to the standard 1 degree-of-freedom (1-df) test of GxE interaction based on logistic regression, we utilized the more powerful two-step EDGE method (33) and the 3-degree-of-freedom (3-df) joint test (34). All models were adjusted for age, sex, study, total energy intake (kcal/day), and the first three principal components to account for ancestry. We considered a P value of <5×10⁻⁸ statistically significant. We also calculated odds ratios (OR) for red meat intake stratified by genotype and for genotype stratified by meat intake to examine patterns of sub-group-specific associations. A detailed description of the notation and methods used in the GxE analysis has been previously published (33–34).

Functional annotation plots for GxE findings and regional plots were also generated. Regional plots enable inspection of the strength and extent of association signal, linkage disequilibrium (LD), and position of findings relative to genes in the region. Regional plots were generated using the software LocusZoom v1.3.32. Measures of LD were estimated using 1000G EUR study population controls.

Data availability

The data generated in this study are available upon request from the corresponding author.

RESULTS

Red meat and processed meat consumption and CRC risk

CRC cases were slightly older, were more likely living with obesity, had a higher total energy intake (2011kca/day ± 730.8 versus 1920kcal/day ± 706.6, p<0.001), and consumed more servings of red (0.60 ± 0.46 versus 0.54 ± 0.44, p<0.001) and/or processed meat (0.36 ± 0.35 versus 0.28 ± 0.35, p<0.001) per day when compared to controls (Table 1).

Table 1.

Summary statistics for demographic and CRC related risk factors, by case/control status

	Cases	Controls
	N = 29842	N = 39635	P-value
Red meat intake (servings/day)a
Mean (SD)	0.60 (± 0.46)	0.54 (± 0.44)	<0.001
Missing	70 (0.2%)	69 (0.2%)
Processed meat intake (servings/day)a
Mean (SD)	0.36 (± 0.35)	0.28 (± 0.29)	<0.001
Missing	2930 (9.8%)	2395 (6.0%)
Ageb
Mean (SD)	64.2 (± 10.8)	63.5 (± 9.42)	<0.001
Sex
Female	14569 (48.88%)	19615 (49.52%)	0.082
Male	15273 (51.2%)	20020 (50.48%)
Height (cm)
Mean (SD)	169 (± 9.56)	169 (± 9.53)	0.314
Missing	411 (1.4%)	282 (0.7%)
BMI
Mean (SD)	27.4 (± 4.90)	27.0 (± 4.62)	<0.001
Missing	1492 (5.0%)	1459 (3.7%)
BMI
Normal (18.5 to <25 kg/m2)	9286 (31.1%)	13570 (34.2%)	<0.001
Overweight (>=25 to <30 kg/m2)	11915 (39.9%)	16263 (41%)
Obese (>=30 kg/m2)	6903 (23.1%)	8071 (20.4%)
Missing	1738 (5.8%)	1731 (4.4%)
Total Energy Intake (kcal/day)b
Mean (SD)	2011 (± 730.8)	1920 (± 706.6)	<0.001
Missing	7475 (25%)	14998 (37.8%)

^a1 serving is equivalent to 70.9 grams or 2.5 ounces.

^bMean imputed for partially missing data. Studies with missing variable not used for mean estimation.

Meta-analyses of exposure main effects adjusting for age, sex, and total energy intake, showed associations between intake of red meat (per quartile increase servings/day OR=1.30; 95%CI 1.21–1.41) and intake of processed meat and CRC risk (per quartile increase OR=1.40; 95%CI 1.20–1.63) (Figure 1). We observed significant between study heterogeneity that was largely limited to case-control studies (Phet = 0.001, I² = 65% & Phet = 0.008, I² = 61% for red meat and processed meat, respectively) (Supplemental Figure 1). For both exposures, estimates of association were greater in magnitude in meta-analyses of case-control studies (red meat meta-OR = 1.45; 95% CI = 1.22–1.72; processed meat meta-OR = 1.56; 1.22–2.00) than cohort studies (red meat meta-OR = 1.21; 95% CI = 1.08–1.35; processed meat meta-OR = 1.16; 0.98–1.37) (Supplemental Figure 1). Analysis stratified by sex suggested a slightly higher association for consumption of processed meat and CRC risk among men (OR = 1.51; 95%CI 1.25–1.82) compared to women (OR = 1.33; 95% CI = 0.99–1.80) (Figure 1). There were no substantial sex differences for the association between CRC and red meat intake. The association between red meat consumption and CRC risk was slightly higher for distal colon compared to proximal or rectal; whereas for processed meat consumption the association was higher for both distal colon and rectal localization compared to proximal location, albeit overlap in the confidence intervals was observed and these differences are not statistically significant (Figure 1). We obtained similar estimates when performing pooled analyses of associations further adjusted by the first three ancestry principal components. Whereas there was evidence for heterogeneity across studies (Supplemental Figure 1), when considering cohort and case-control studies separately, we decided to conduct GxE testing across all studies combined given that a positive association was reported in both study types.

Figure 1.

Results from overall association between colorectal cancer and A) red meat intake and B) processed meat intake, overall and stratified by sex and tumor site. Models are adjusted for age, sex, and total energy intake. Meat intake servings per day were coded as median of sex/study specific quartiles, modeled as a continuous variable.

Genome-wide interaction scan results

Evaluation of red meat consumption interaction using the two-step EDGE method identified a statistically significant interaction of red meat with rs35352860 (p-value = 2 × 10⁻⁸) which maps to chromosome 18 in an intronic region of the SMAD7 (SMAD family member 7) gene (Figure 2, Table 2).

Figure 2.

Results from EDGE two-step GxE testing procedure using expectation-based SNP partitioning with principal components approach for effective number of tests adjustment (59). Numbers reflect total number of SNPs assigned to each partition, with the number of effective tests listed below. Only the first 7 bins were plotted; significant hit corresponds to rs35352860 (SMAD7 region).

Table 2.

Main results from genome-wide interaction scans of red meat consumption

Meat intake variable	SNP	Chr	BP Position	Gene	Ref	Alt	Alt Allele Freq (1000G EUR)	Method	P-value G	P-value GxE	P-value 3-df
Red meat	rs4871179	8	122247679	Downstream HAS2	C	G	0.33	Joint test (3-df)	3.99E-04	1.00E-06	3.05E-08
Red meat	rs35352860	18	46453754	SMAD7	C	T	0.26	Two-step	4.75E-21	2.0E-8	7.16E-23

SNP, single nucleotide polymorphism; Chr, chromosome; df, degrees of freedom; BP Position, base pair position based on NCBI Build37.

Notes: Imputed SNPs were coded as expected gene dosage. Multiplicative interaction terms were modelled as the product of red meat and each SNP of interest. All statistical tests were two-sided. Models were adjusted for age, sex, study, total energy intake (kcal/day), and the first three principal components to account for ancestry.

One additional SNP for red meat intake on chromosome 8 was identified based on the joint 3-df test (Table 2). This SNP, rs4871179, maps downstream of HAS2 (hyaluronan synthase 2). No statistically significant interactions were identified for processed meat. The less powerful standard 1-df GxE analysis revealed no evidence of interaction with either red meat or processed meat and genome-wide statistically significant loci (Supplemental Figure 2). No other statistically significant GxE interactions (p-value <10⁻⁸) that were also genome-wide significant GWAS loci were found.

To further explore the significant interactions, we constructed a dichotomous variable for red meat intake with median of consumption as cutoff point and evaluated the association between red meat intake and CRC risk, stratifying by genotypes of the identified loci (Table 3). For the SNP on chromosome 18 identified by the two-step method (rs4871179), red meat was associated with CRC risk within each genotype group, but the magnitude increased with every copy of the major T allele. Specifically, the red meat in relation to CRC odds ratio was OR = 1.18 (95% CI = 1.11–1.24) for those with genotype CC, OR = 1.35 (95% CI 1.26–1.44) for CT, and OR = 1.46 (1.26–1.69) for TT (Table 3; Supplemental Figure 3). For the SNP in chromosome 8, the association between red meat intake and CRC risk was significant among homozygous carriers of the more common allele (OR for CC = 1.38; 95% CI = 1.29–1.46) and heterozygous (OR for CG = 1.2; 95% CI = 1.12–1.27) with a non-statistically significant association among homozygous carriers of the minor G allele (OR for GG = 1.07; 95% CI = 0.95–1.19) (Table 3; Supplemental Figure 3).

Table 3.

Odds ratio (95% confidence interval) showing association between red meat intake^a and CRC risk, stratified by genotypes of loci identified using joint testing and two-step methods

	rs4871179 (8:122247679)
	CC	CG	GG
red meat intake = 0	1 (Reference)	1.03 (0.98–1.09)	1.07 (0.98–1.16)
red meat intake = 1	1.38 (1.29–1.46)	1.24 (1.16–1.31)	1.14 (1.05–1.24)
red meat intake by genotype	1.38 (1.29–1.46)	1.2 (1.12–1.27)	1.07 (0.95–1.19)
	rs35352860 (18:46453754)
	CC	CT	TT
red meat intake = 0	1.0 (Ref)	0.83 (0.79–0.88)	0.65 (0.58–0.73)
red meat intake = 1	1.17 (1.11–1.24)	1.13 (1.06–1.19)	0.94 (0.85–1.04)
red meat intake by genotype	1.18 (1.11–1.24)	1.35 (1.26–1.44)	1.46 (1.26–1.69)

^aRed meat intake variable was dichotomized with median of servings/day from study specific-quartiles as cutoff point.

Notes: Models were adjusted for age, sex, study, total energy intake (kcal/day), and the first three principal components to account for ancestry.

DISCUSSION

In this large-scale genome-wide GxE analysis of over 69,000 individuals from 27 studies, we found evidence of a statistical interaction between 2 SNPs (rs35352860, rs4871179) and red meat consumption in relation to CRC risk. We did not find evidence of GxE interactions when considering consumption of processed meat. A prior publication reported a G x processed-meat interaction for the rs4143094 SNP (10p14, near GATA3). Analysis of this SNP in our current sample revealed some suggestion of an interaction (p=0.0065) but it did not achieve genome-wide significance in any of our analyses (24).

The rs35352860 SNP which maps to the SMAD7 gene was found to have the strongest evidence to support effect modification of the red meat intake and CRC risk association as reported by the p-values in the GxE scan (Table 2). The SMAD7 gene codes for an intracellular protein, traditionally considered as a negative regulator of TGF-B1 (35) by interfering with TGF-B1 signaling. Several SMAD7 polymorphisms have been previously associated with CRC risk in several GWAS studies for CRC in European and Asian individuals (36–38). Moreover, a regional plot for rs35352860 shows that the SNP aligns with multiple hits in LD previously reported by Broderick et.al (39) (Supplementary Figure 4A). Additionally, an earlier study from this consortium reported a statistically significant GxE interaction between rs4939827 and BMI (40). We conducted a sensitivity analysis after adjustment of main effects model for BMI. Results did not differ from main findings obtained in genome-wide interaction scans of red meat consumption (Supplementary Table 2).

There are no reports of a mechanism for a potential GxE interaction for the rs35352860 polymorphism and red meat. Previous genome-wide GxE interaction analyses including red meat did not report associations of genome-wide significance for SMAD7 polymorphisms (27,41). However, SMAD7 deletion has been previously associated with a protective dose-effect in overall survival and disease-free survival in tumor biopsies of CRC patients, with a greater effect on increased CRC-related death with every additional copy (42), and with overexpression in colonic adenocarcinoma cells inducing tumorigenesis (43).

Key components of red meat that have been linked to CRC risk include established carcinogens, such as heterocyclic amines and nitrosamines, as well as heme iron (6). The porphyrin structure of heme iron stimulates the production of pro-carcinogenic endogenous N-nitroso compounds as well as lipid oxidation products that can cause DNA damage (6,13,44). It is still unclear if the effects of red meat in colorectal carcinogenesis are only mediated through either of these compounds, or potential synergism between all of these. Given our observation of an effect modification of the association of red meat and CRC risk by genetic variants in SMAD7, we propose a mechanism that involves the SMAD7 protein via regulation of circulating levels of hepcidin, which is a liver-derived peptide that closely influences erythrocyte production and is the main molecule in charge of regulating systemic iron homeostasis (45). Systemically, hepcidin binds and degrades the membrane transporter ferroportin; whose function is to excrete the previously intracellular stored iron in enterocytes from a pool that contains both ferrous iron (Fe²⁺) from the heme pathway as well as non-heme, into the bloodstream (46). Alterations in hepcidin concentration that reduce circulating levels of hepcidin lead to increase in duodenal iron absorption and clinical iron overload, and the SMAD7 protein has been reported to function as a regulator of iron homoeostasis by reducing hepcidin expression (47). Knock-out mice of the hepcidin protein modulator were reported to have an increased risk of CRC (48). Hepcidin is canonically regulated via the bone morphogenetic protein 6 (BMP6)-SMAD1/5/8 pathway (49). However, murine models with iron overload have also proposed that the SMAD6 and SMAD7 proteins are co-regulated with hepcidin levels (50), and that suppression of SMAD7 can suppress hepcidin (50).

The SMAD7 rs35352860 SNP and several SNPs in LD align with open chromatin regions in the functional annotation plot due to the high density of mapped DNase type I cleavages (Supplemental Figure 5A). These regions correspond to DNase I hypersensitive sites, canonical epigenetic markers (H3K27ac & H3K4me) and cell lines evaluated in normal colon histological samples, tumor samples and CRC cell lines.

High accessibility to DNase Hypersensitive Sites increases the likelihood of transcriptional alterations due to cis-regulatory elements, namely: enhancers, promoters, silencers, insulators, and locus control regions. We hypothesize a mechanism in which over-expression of the SMAD7 molecule effectively inhibits hepcidin production. This in turn, may prevent hepcidin binding to ferroportin, thus preventing the internalization of the transmembrane transporter with increased output of iron through the basolateral membrane into the bloodstream. Free circulating iron can contribute to free radical formation, facilitating a pro-inflammatory, pro-carcinogenic state. Therefore, we speculate that among carriers of a SMAD7 variant that leads to overexpression of this protein, the effects of excess iron will be worsened by a diet with high consumption of red meat, due to an impaired inability to increase hepcidin production in response to heme iron intake levels.

The variant in the 8q24 region (rs4871179) resides downstream of the HAS2 gene (Supplementary Figure 4B). The functional annotation plot did not portray accessible chromatin regions (Supplementary Figure 5B). There is evidence that this gene may play a role in colorectal carcinogenesis. Specifically, the HAS2 gene is a member of the Glycosyltransferase family 2, which confers glycosylation profiles to molecules in the endoplasmic reticulum and Golgi apparatus. Alterations of these glycosylation profiles in cancer cells have been associated with carcinogenesis, and tumor progression in the past (51,52). The HAS2 gene, catalyzes the synthesis of Hyaluronan Acid (HA), one of the main extracellular matrix components, has been reported to be overexpressed in CRC (53) and has been associated with tumorigenesis and metastasis in breast cancer (54,55). Inhibition or reduced expression of HAS2 and/or HAS3 decreases metastatic colon carcinoma cell adhesion to laminin (55), and increased apoptosis and reduced metastasis in experimental models (54). Higher HA levels are associated with a worse prognosis in CRC patients, with an inverse correlation between percentage of HA-positive carcinoma cells and cancer-related survival rate and recurrence-free survival (56). There are no reports that link HAS2 to possible effects of red meat consumption, or report GxE interactions with red meat consumption and CRC risk.

Our study has several strengths, including our large sample size with uniformly harmonized data and systematic quality control across all pooled studies, which allowed for consideration of tumor anatomical localization, gender, and study design in the meta-analyses. Another key strength is the implementation of powerful approaches for GxE analyses, including the two-step methods that improve our ability to identify novel interactions that were not detected using the traditional 1-df GxE test. As well, we were able to consider previously reported confounders of both the exposure and the genotypes. Among the limitations of our study is the fact that consumption of red meat and processed meat was obtained via questionnaires, and that many of the included studies are case-control studies where the exposure was assessed after the cancer was diagnosed. However, case-control study participants were asked to report on intake typically 1 to 2 years before diagnosis/selection into the study. Therefore, we cannot exclude the possibility of misclassification and/or recall bias. For cohort studies, risk factors were assessed at the study-specific reference time, which aligns with the time of blood draw or buccal collection. A comprehensive description has been previously published (57). Moreover, four studies did not report total caloric intake which was set to zero in the analyses (i.e., ASTERISK, DACHS, PHS, UKB) (Supplementary Table 1). Nevertheless, sensitivity analyses confirmed that, since the interaction models included a fixed effect indicator for study, subjects in those four studies did not contribute information for estimating the energy effect. However, they did contribute to the estimation of all other effects including GxE interaction. Furthermore, confounding analyses revealed that GxE estimates for both our significant SNPs changed <0.04% on the Odds Ratio scale when an adjusted model was compared to an unadjusted model without total caloric intake. Additionally, study-specific quartiles were created to evaluate meat consumption which do not account for absolute differences in exposure variable. However, given differences in assessment tool, the use of study-specific quantiles is a valid commonly used approach in pooled analysis of nutritional exposures (58). As well, other behavioral patterns (e.g., exercise) were not considered as confounders in the analysis. Finally, we acknowledge that this study pooled data from most studies conducted among populations of European ancestry, and thus these findings may not apply to other racial and ethnic populations.

In summary, we report two novel GxE interactions for red meat consumption and colorectal cancer risk. Our strongest finding is in a SNP in the SMAD7 gene, which provides further supportive evidence for a role of heme iron in the carcinogenic pathway of red meat consumption and CRC development.