Intake of Dietary Fruit, Vegetables, and Fiber and Risk of Colorectal Cancer According to Molecular Subtypes: A Pooled Analysis of 9 Studies

Colorectal cancer is the third most commonly diagnosed cancer worldwide, and the number of cases is predicted to increase to 2.2 million new cases per year by 2030 (1). To date, various dietary factors have been investigated in relation to colorectal cancer risk in many epidemiologic studies. Meta-analyses report that higher intake of fruit (2), vegetables (2), and fiber (3) is associated with decreased colorectal cancer risk. However, the World Cancer Research Fund and the American Institute for Cancer Research (WCRF/AICR) concluded that there is no “convincing strong evidence” for these dietary factors. High intake of foods containing dietary fiber only has “probable strong evidence” for decreasing the risk of colorectal cancer, and low intake of fruit and vegetables only has “limited-suggestive evidence” for increasing risk (https://www.wcrf.org/dietandcancer/colorectal-cancer; ref. 4) because these preventive associations have been shown in many, but not all epidemiologic studies (5–7). One possible reason for study heterogeneity is differential effects of diet on distinct colorectal cancer molecular subtypes.

Four molecular markers in particular have been well-studied with regard to colorectal cancer heterogeneity: microsatellite instability (MSI), the CpG island methylator phenotype (CIMP), and oncogenic mutations in BRAF and KRAS genes (8). Increasing evidence showed that some lifestyle risk factors can be differentially associated with these molecular markers (9). Probably the evidence is currently strongest for smoking, which is more strongly associated with MSI-high, CIMP-positive, and BRAF-mutated tumors (10, 11). Furthermore, there is some evidence that body mass index and hormone replacement therapy are associated with MSI status (12). These findings provide a strong rationale to investigate if dietary risk factors may also be differentially associated with colorectal cancer molecular markers. Although several epidemiologic studies have evaluated the association of fruit (13–19), vegetables (13–20), and fiber (13, 16–22) intake with risk of colorectal cancer molecular subtypes, results from these studies have been inconsistent. In addition, there has been no study to this point that has investigated all four characteristic molecular markers together despite the fact that these can point to differential pathways. Specifically, three different pathways to colorectal cancer development may be affected by the combination of molecular subtypes: (i) a serrated pathway, (ii) an alternate pathway, and (iii) a traditional pathway (23, 24). Only one meta-analysis reported the association of fiber intake with MSI status of colorectal cancer (12). However, this meta-analysis of three case–control studies used heterogeneous fiber definitions across studies (12). The direction of the association between fiber intake and MSI status in each study was different because one study found higher fiber intake was associated with decreased risk of MSS tumors (13), but another study reported decreased risk of both MSI-high and MSS/MSI-low tumors (16). On the contrary, one study found increased risk of MSS/MSI-low tumors with low fiber intake (17). To better address the hypothesis that these dietary factors may affect risk of colorectal cancer molecular subtypes differently, here we pooled nine population-based studies with individual-level data harmonized in a consistent manner.

We analyzed data from 9,592 colorectal cancer cases and 7,869 controls within the Genetics and Epidemiology of Colorectal Cancer Consortium (GECCO) and the Colon Cancer Family Registry (CCFR) to assess the association of fruit, vegetables, and fiber intake with colorectal cancer risk by molecular subtypes using data on MSI, CIMP, BRAF mutation, and KRAS mutation status. Identification of associations with specific molecular subtypes may point to specific diet–colorectal cancer associations and help inform underlying biological mechanisms relevant to colorectal cancer risk.

This study population consisted of 8,783 colorectal cancer cases and 7,869 controls from 9 observational studies within GECCO and the CCFR with available tumor marker, fruit, fiber, vegetables, and total energy intake data. In addition, 809 population-based colorectal cancer cases from the Mayo Clinic CCFR, which did not enroll population-based controls, were included in case-only analyses. Among study participants, 3,258 colorectal cancer cases and 3,984 controls were from cohort studies, and 6,334 colorectal cancer cases and 3,885 controls were from case–control studies. Descriptive characteristics of each study and mean intake of quartile cut points of fruits, vegetables, and fiber in each study are shown in Supplementary Tables S1 and S2. In addition, details regarding each observational study are described in Supplementary Text. All colorectal cancer cases were defined by colorectal adenocarcinoma and confirmed by pathological records, medical records, and/or death certificate information. All study participants provided written-informed consent, and each study was approved by their relevant research ethics committee or Institutional Review Board.

Testing for MSI, CIMP, and mutations in the BRAF and KRAS gene was conducted previously by each study and according to individual study protocols. Details regarding marker testing in each study are described in Supplementary Text. Especially, Supplementary Table S3 shows study-specific markers used to assess MSI and definition of MSI status. In addition, Supplementary Table S4 also shows study-specific panels used to assess CIMP status. We defined marker combinations for molecular subtypes, consistent with previously suggested classifications (23, 25): Types 1–10. Details of the defined marker combinations are given in Fig. 1 and Supplementary Text. Subtype classifications with fewer than 50 cases were excluded from analyses. With regard to colorectal carcinogenic pathways, previous studies reported that three different pathways to colorectal cancer development may be affected by the combination of molecular subtypes: (i) a serrated pathway (Type 1: MSI-high, CIMP-positive, BRAF-mutated, KRAS-wildtype and Type 2: MSS/MSI-low, CIMP-positive, BRAF-mutated, KRAS-wildtype), (ii) an alternate pathway (Type 3: MSS/MSI-low, CIMP-negative, BRAF-wildtype, KRAS-mutated), and (iii) a traditional pathway (Type 4: MSS/MSI-low, CIMP-negative, BRAF-wildtype, KRAS-wildtype and Type 5: MSI-high, CIMP-negative, BRAF-wildtype, KRAS-wildtype; refs. 23, 24).

In each study, demographic and lifestyle risk factor information was assessed via in-person interviews or structured self-administered questionnaires, as described in the Supplementary Text. Dietary variables were ascertained using food frequency questionnaires or diet history. Data were collected at study entry, at blood draw, or 1 to 2 years prior to sample ascertainment. A multistep, iterative data harmonization procedure was applied, reconciling each study's unique protocols and data collection instruments (26). Multiple quality-control checks were performed, and outlying values of variables were truncated to the minimum or maximum value of an established range for each variable. Variables were combined into a single dataset using common definition, standardized coding, and standardized permissible values.

In our study analysis, we selected dietary intake variables for fruit (servings/day), vegetables (servings/day), and fiber (measured as g/day). Sex- and study-specific quartiles for dietary variables were created based on the distribution in the controls. Data harmonization was performed using SAS and T-SQL.

We used the χ² and Mann–Whitney t tests to compare baseline characteristics between cases and controls. To test if the association of fruit, vegetables, and fiber intake with each molecular subtype differed, we conducted case-only logistic regression analysis. We also estimated the strength of the association between the dietary variables and risk for the specific molecular subtypes using polytomous logistic regression models with two case groups and one control set. The ORs and their corresponding 95% confidential intervals (CI) for the quartiles of dietary variables were compared with the lowest intake category as a reference. P values for linear trends were calculated by treating quartiles of intake as a continuous variable. Participants with missing values of each dietary intake for fruit (n = 1,933), vegetables (n = 1,775), and fiber (n = 5,422) were excluded from the analyses. In our primary analyses, minimally adjusted models (minimally adjusted OR) included study site, age at diagnosis, total energy consumption (kcal/day), and sex as covariates. To determine confounding factors for multivariate-adjusted models (multivariate-adjusted OR), we evaluated the association of the following colorectal cancer–related parameters: tobacco smoking, alcohol, body mass index, physical activity, history of diabetes mellitus, red meat intake, processed meat intake, and aspirin/nonsteroidal anti-inflammatory drugs use, with dietary factors on the risk of colorectal cancer overall. We included those factors that changed the beta estimate of the dietary factors by more than 10% when compared with the minimally adjusted models (27). Based on this, we included tobacco smoking, red meat intake, and processed meat intake in the multivariate adjusted models. Missing covariates were imputed by sex-specific mean for each study for age at diagnosis (n = 8 missing), total energy consumption (n = 5,473 missing), tobacco smoking (n = 1,375 missing), red meat intake (n = 1,709 missing), and processed meat intake (n = 6,565 missing). Processed meat was not imputed for NFCCR because it was missing for all subjects (n = 1,045 missing). We conducted a subgroup analysis stratified by study design (case–control or cohort study) as sensitivity analyses to evaluate differences in effects between study designs.

All P values are two-sided. A P value <0.05 was considered statistically significant for association with colorectal cancer risk. For assessing the heterogeneous association with molecular subtypes, a Bonferroni-corrected P value <0.05/10 was considered statistically significant to account for the 10 subtypes being tested. All statistical analyses were performed using R version 3.6.0.

Cases were more likely to be men, younger, past or current smokers, have a higher intake of red meat, processed meat, and energy, and have a lower intake of fruits and vegetables (Table 1). In a case–control analysis of all colorectal cancer cases combined, we observed an inverse association between fiber intake and overall colorectal cancer risk [multivariate-adjusted OR _{4th vs. 1st quartile} = 0.85 (95% CI, 0.76–0.97); P_trend = 6.2e-03], but no statistically significant association between colorectal cancer risk with fruit intake [multivariate-adjusted OR _{4th vs. 1st quartile} = 1.04 (95% CI, 0.93–1.15); P_trend = 0.99] and vegetable intake [multivariate-adjusted OR _{4th vs. 1st quartile} = 0.92 (95% CI, 0.82–1.03); P_trend = 0.09; Supplementary Table S5]. Among colorectal cancer cases, 15.6% were MSI-high, 18.9% CIMP-positive, 12.8% BRAF-mutated, and 32.8% KRAS-mutated.

When we analyzed the association of dietary variables with risk of each molecular colorectal cancer subtype in a case-only analysis (Table 2), we observed that fruit intake was differentially associated with BRAF-mutated compared with BRAF-wildtype tumors [multivariate-adjusted OR _{4th vs. 1st quartile} = 0.75 (95% CI, 0.60–0.94), P difference = 0.02]. In addition, a polytomous logistic regression analysis (Table 3) comparing BRAF-mutated and BRAF-wildtype tumors with controls showed that a trend toward inverse association between fruit intake and colorectal cancer risk was limited to BRAF-mutated tumors [multivariate-adjusted OR _{4th vs. 1st quartile} = 0.82 (95% CI, 0.65–1.04), P_trend = 0.06] as compared with controls, whereas there was no association between fruit intake and BRAF-wildtype tumors [multivariate-adjusted OR _{4th vs. 1st quartile} = 1.09 (95% CI, 0.97–1.22), P_trend = 0.54] as compared with controls. In subgroup analyses, these associations were found statistically significant in case–control studies, but not in cohort studies (Table 4 and Supplementary Table S6). Fruit intake did not show a statistically significant differential association with other molecular subtypes. Neither vegetable nor fiber intake showed significant differential associations for any of the molecular markers as shown by the case-only analysis (Table 2). However, in the subgroup analyses, we found that fiber intake was differentially associated with CIMP-negative tumors as compared with CIMP-positive tumors in cohort studies (Table 4).

When we combined markers to define subtypes in polytomous logistic regression analyses (Fig. 1; Supplementary Table S7), we observed a linear trend for increased fiber intake in relation to colorectal cancer risk among Type 4 tumors, MSS/MSI-low, CIMP-negative, BRAF-wildtype, and KRAS-wildtype [multivariate-adjusted OR = 0.94 (95% CI, 0.88–0.99), P_trend = 0.03)] compared with controls, although the difference between Type 4 and each of the other Types was not statistically significant. This finding was consistent with results in the single-marker analyses (Table 3), where the same trend was seen for MSS/MSI-low tumors (P_trend = 3.4e-03), CIMP-negative tumors (P_trend = 9.0e-03), BRAF-wildtype tumors (P_trend = 3.5e-03), and KRAS-wildtype tumors (P_trend = 0.03) compared with controls.

This is the largest analysis using individual level of dietary data to investigate their association with well-described molecular subtypes for colorectal cancer. Although higher fruit intake was not associated with overall colorectal cancer risk, we observed that fruit intake was statistically significantly associated with a decreased risk for BRAF-mutated tumors but not BRAF-wildtype tumors. This finding was observed only among case–control studies. In addition, we observed in single-marker polytomous logistic regression analyses that higher fiber intake was associated with a decreased risk for MSS/MSI-low, CIMP-negative, BRAF-wildtype, and KRAS-wildtype molecular subtypes compared with controls. The association in the combined-marker analysis also showed a trend toward a negative association. These negative associations were stronger than they were for MSI-high, CIMP-positive, BRAF-mutated, or KRAS-mutated tumors, but the differences were not statistically significant. Increased fiber intake was differentially associated with a decreased risk of CIMP-negative compared with CIMP-positive tumors in cohort studies, but not in case–control studies. We did not identify any differences in vegetable intake with colorectal cancer risk in subtype analyses examining MSI, CIMP, and KRAS and BRAF mutations separately or in combination.

Colorectal cancer development is caused by different etiological pathways underlying different genetic and epigenetic aberrations, which have been defined by specific molecular subtypes associated with distinct development trajectories. If dietary risk factors for colorectal cancer, such as fruit, vegetable, and fiber intake, affect specific etiologic pathways, then we can expect that the specific dietary factors are differentially associated with these molecular subtypes. For specific molecular subtypes, the mutation in the KRAS oncogene has been widely known as an acting driver of colorectal cancer development (28). In addition, mutation of the BRAF oncogene induces proliferation and inhibits normal apoptosis of colonic epithelial cells (29). Both KRAS and BRAF are key players of the MAPK pathway (24). Another driving force of colorectal cancer development is CIMP. Widespread methylation of numerous promoter CpG island loci is responsible for inactivation of tumor-suppressor genes and other tumor-related genes (28). CIMP is also strongly related to the serrated pathway. Silencing of tumor-suppressor genes such as p16INK4a and IGFBP7 via the synergistic effects of mutation of BRAF and CIMP resulting from hypermethylation could facilitate progression to serrated colorectal polyps (24). MSI is recognized by high frequency of genetic alterations in repeated microsatellite sequences of DNA resulting from a DNA mismatch repair deficiency (30).

Fruit intake was associated in BRAF-mutated tumors, which is of interest given that BRAF-mutated tumors have particularly poor survival (31). Differences in findings for fruit intake by study design may be due to differences in sample size, given that approximately two thirds of cases were from case–control studies. To evaluate study-specific differences among case–control studies, we explored consistency of the case–control study findings. The meta-analysis OR from case–controls studies was similar to the pooled estimate from all studies, and we did not observe substantial heterogeneity among the case–control studies [OR_{case–control meta-analysis} = 0.85 (95% CI, 0.78–0.93), P heterogeneity = 0.52]. However, selection and recall biases, which are more likely to occur in case–control studies, may also explain observed differences due to study design. Selection bias could occur when colorectal cancer cases and/or controls are not representative of the target population. Although it could be possible that controls with a healthy diet are more willing to participate in studies, this is unlikely to explain the observed differential effect of fruit intake specifically on BRAF-mutated versus BRAF-wildtype tumors in the case-only analysis. For this case-only finding to be explained by selection bias, the participation of BRAF-mutated versus BRAF-wildtype colorectal cancer cases would need to differ by fruit intake. This may be possible if survival after colorectal cancer diagnosis affects participation and differs by BRAF mutational status and by fruit intake. However, we recently showed in a pooled analysis that BRAF-mutational status is not associated with survival after colorectal cancer diagnosis (32), and previous studies do not show strong evidence for the impact of fruit intake on survival (33, 34). For recall bias to explain the case-only finding, cases with and without BRAF mutation would need to recall their fruit intake differently. If treatment with BRAF inhibitors could have an impact is unknown; however, the majority of colorectal cancer cases included in this study were diagnosed before treatment with BRAF inhibitors were introduced (35). As BRAF mutation is measured by standard methods, it is probably also less likely that measurement error could explain the observed difference in the case-only analysis for BRAF-mutated versus wildtype tumors.

To our knowledge, only one study has published findings on fruit intake and colorectal cancer risk by BRAF mutation status (15). This cohort study of 186 cases found no clear differences in risk. Several meta-analyses of studies not accounting for tumor markers have been published. A meta-analysis published in 2003 reported statistically significant inverse associations between fruit intake and colorectal cancer risk in case–control studies, but the relative risk was null when restricted to cohort studies (36). In addition, a previous pooled analysis of 14 cohort studies published in 2007 also showed no overall association (37). However, a more recent meta-analysis of 19 cohort studies published in 2018 reported statistically significant inverse associations between fruit intake and colorectal cancer risk, and reported nonlinear relationships between fruit intake and colorectal cancer risk (2). It is possible as sample sizes of meta-analyses increase that the power becomes sufficient to capture a significant association in a subset of the cancers (in this case BRAF-mutated tumors). Additional studies ideally conducting in cohorts are needed to replicate our finding.

Fruit intake may plausibly play a role in inhibiting events of BRAF-mutated tumors. Fruits are rich in many nutrients and biologically active compounds, such as vitamins, carotenoids, and folic acid, which may be cancer-preventive (38). Accordingly, a previous laboratory study showed that high levels of vitamin C specifically kill BRAF-mutated colorectal cancer cells, but not BRAF-wildtype cells (39). This effect is due to uptake of dehydroascorbate, which is the oxidized form of vitamin C. Increased intracellular dehydroascorbate uptake accumulates cellular reactive oxygen species and inactivates GAPDH. Inhibition of GAPDH in BRAF-mutated colorectal cancer cells leads to an energetic crisis and cell death through inhibiting glycolysis and depleting ATP (39). However, two randomized, double-blind, placebo-controlled trials with a mean follow-up period ranging from 8.0 to 9.4 years did not find that 500 mg of vitamin C supplementation daily was associated with a decreased colorectal cancer incidence in both men and women (40, 41). Fruit intake may also affect BRAF-mutated colorectal cancer development through affecting the MAPK pathway. It has been previously reported that Lycium barbarum fruit, also known as Chinese Wolfberry, inhibits cancer cell growth by cell-cycle arrest and apoptosis through regulating the activation of MAPK-signaling pathway (42). Furthermore, previous studies have shown that fisetin, a dietary flavonoid found in apples, strawberries, kiwi, and other fruits, inhibits cell invasion by targeting MAPK signaling pathway (43) and reduces the anti-invasive and antimetastatic effects of BRAF-mutated cells (44). Accordingly, fruit intake may reduce BRAF-mutated colorectal cancer through multiple nutrients and biologically active compounds.

With regard to colorectal carcinogenic pathways, we investigated three different pathways: (i) a serrated pathway (Types 1 and 2), (ii) an alternate pathway (Type 3), and (iii) a traditional pathway (Types 4 and 5; refs. 23, 24). Because this colorectal carcinogenic hypothesis only shows the predominant pathways, there is overlap between them (24). We observed that higher intake of fiber was associated with a decreased risk of combined colorectal cancer subtypes that were MSS/MSI-low, CIMP-negative, BRAF-wildtype, and KRAS-wildtype (Types 4 and 5: traditional pathway). Moreover, higher fiber intake was associated with a decreased risk for CIMP-negative tumors, but not for CIMP-positive tumors in cohort studies. Higher dietary fiber intake has “probable strong evidence” for decreasing the risk of colorectal cancer as defined by WCRF/AICR (4), which is in line with our findings given that the traditional pathway is the predominate pathway for colorectal cancer development. A previous meta-analysis of seven observational studies investigating how lifestyle exposures may increase or decrease the risk of serrated colorectal polyps presented a suggestive trend toward an inverse association between fiber intake and risk of serrated colorectal polyps, but findings were not statistically significant (45), which is not consistent with our finding. Currently functional data are missing that could explain why the association with fiber intake is restricted to a subset of molecularly-defined colorectal cancer tumors.

A strength of our pooled analysis from 9 observational studies with up to 9,592 colorectal cancer cases and 7,869 controls is that we were well powered to detect associations of fruit, vegetables, and fiber intake with risk of major colorectal cancer molecular subtypes. Due to this sizable dataset, we could further assess the association of colorectal cancer molecular subtype markers in combination. Furthermore, we used a consistent approach to harmonize all dietary variables across the studies to enable a pooled analysis of individual level data. In addition, we have robust data on other factors including age at diagnosis, total energy consumption, and sex as covariates.

Our study has some limitations. First, it is difficult to exclude potential selection bias. Although molecular subtypes are ideally available for all cases, there may be tissue retrieval biases with tissue availability potentially associated with tumor size and stage (46). However, some of the included studies previously have shown that there were no differences in age, diet, or other lifestyle characteristics between cases with and without available tumor tissue (47, 48). Second, fruit, vegetables, and fiber intake are likely measured with error because they were assessed via in-person interviews or self-administered questionnaires, resulting in exposure misclassification. To best account for differences between dietary assessment methods, we harmonized data as study- and sex-specific quartiles rather than the continuous intake value, as it is commonly done for pooled dietary analyses (37). As exposure misclassification would likely be nondifferential, it would result in an underestimation of the association of these dietary variables with colorectal cancer risk. Third, we analyzed the colorectal cancer risk only using the lifestyle information measured at a single point in time, whereas lifestyle habits of study participants might change during the relevant etiologic time window. However, as such changes are not differential between cases and controls, this would have led to an attenuation of the associations. Fourth, some studies had a sizable number of missing covariate data; however, the missing covariate data were imputed by sex-specific mean for each study to ensure no samples were missed in the analysis due to missing confounders.

In summary, we found in our large pooled analysis for well-defined molecular subtypes that higher fruit intake was associated with a decreased risk of BRAF-mutated colorectal cancer but not with BRAF-wildtype colorectal cancer. In additionally, higher fiber intake may be associated with a stronger decreased risk of colorectal cancer developing by the traditional adenoma-colorectal cancer pathway that is MSS/MSI-low, CIMP-negative, BRAF-wildtype, and KRAS-wildtype tumors. These results potentially explain in part the inconsistent findings between fruit or fiber intake and overall colorectal cancer risk that have previously been reported.

L.C. Sakoda reports grants from National Cancer Institute during the conduct of the study. M. Giannakis reports grants from Bristol Myers-Squibb and grants from Merck outside the submitted work. A.E. Toland reports grants from National Institutes of Health during the conduct of the study. A.T. Chan reports grants and personal fees from Bayer Pharma AG, personal fees from Pfizer Inc., personal fees from Janssen Pharmaceuticals, and personal fees from Boehringer Ingelheim outside the submitted work. H. Hampel reports other aid from Myriad Genetic Laboratories, Inc. (free genetic testing for a subset of patients on this study) during the conduct of the study; personal fees from Invitae Genetics (Scientific Advisory Board), personal fees from Promega (Medical Advisory Board), personal fees from Genome Medical (Scientific Advisory Board), and personal fees from 23andMe (consulting) outside the submitted work. M.A. Jenkins reports grants from NCI (funding to my institution) outside the submitted work. V. Moreno reports grants from Agency for Management of University and Research Grants (AGAUR) of the Catalan Government and grants from Instituto de Salud Carlos III during the conduct of the study. R. Nishihara reports personal fees from Pfizer (current employer) outside the submitted work. S. Ogino reports grants from National Institutes of Health (grant number R35 CA197735) and grants from National Institutes of Health (grant number R01 CA151993) during the conduct of the study. P.S. Parfrey reports grants from Canadian Institutes Health Research during the conduct of the study. B. Van Guelpen reports grants from the Swedish Research Council, grants from the Swedish Cancer Society, grants from the Knut and Alice Wallenberg Foundation, grants from the Lion's Cancer Research Foundation at Umeå University, grants from the Cancer Research Foundation in Northern Sweden, and grants from Region Västerbotten during the conduct of the study. No potential conflicts of interest were disclosed by the other authors.

A. Hidaka: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing-original draft, writing-review and editing. T.A. Harrison: Conceptualization, formal analysis, supervision, investigation, methodology, writing-original draft, project administration, writing-review and editing. Y. Cao: Writing-review and editing. L.C. Sakoda: Resources, writing-review and editing. R. Barfield: Methodology, writing-review and editing. M. Giannakis: Writing-review and editing. M. Song: Writing-review and editing. A.I. Phipps: Resources, writing-review and editing. J.C. Figueiredo: Writing-review and editing. S.H. Zaidi: Writing-review and editing. A.E. Toland: Writing-review and editing. E.L. Amitay: Writing-review and editing. S.I. Berndt: Resources, writing-review and editing. I. Borozan: Writing-review and editing. A.T. Chan: Resources, writing-review and editing. S. Gallinger: Resources, writing-review and editing. M.J. Gunter: Resources, writing-review and editing. M.A. Guinter: Writing-review and editing. S. Harlid: Writing-review and editing. H. Hampel: Resources, writing-review and editing. M.A. Jenkins: Resources, writing-review and editing. Y. Lin: Data curation, software, writing-review and editing. V. Moreno: Resources, writing-review and editing. P.A. Newcomb: Resources, writing-review and editing. R. Nishihara: Writing-review and editing. S. Ogino: Resources, writing-review and editing. M. Obón-Santacana: Writing-review and editing. P.S. Parfrey: Resources, writing-review and editing. J.D. Potter: Resources, writing- and editing. M.L. Slattery: Resources, writing-review and editing. R.S. Steinfelder: Data curation, writing-review and editing. C.Y. Um: Writing-review and editing. X. Wang: Writing-review and editing. M.O. Woods: Resources, writing-review and editing. B. Van Guelpen: Resources, writing-review and editing. S.N. Thibodeau: Resources, writing-review and editing. M. Hoffmeister: Resources, writing-review and editing. W. Sun: Formal analysis, writing-review and editing. L. Hsu: Formal analysis, methodology, writing-review and editing. D.D. Buchanan: Resources, writing-review and editing. P.T. Campbell: Resources, writing-review and editing. U. Peters: Conceptualization, supervision, funding acquisition, methodology, project administration, writing-review and editing.

This work was supported by the Uehara Memorial Foundation. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the article.

Genetics and Epidemiology of Colorectal Cancer Consortium (GECCO): National Cancer Institute, National Institutes of Health, U.S. Department of Health and Human Services (U01 CA164930, U01 CA137088, R01 CA059045, U01 CA164930, R21 CA191312, and R01 CA201407).

Genotyping/Sequencing services were provided by the Center for Inherited Disease Research (CIDR; X01-HG008596 and X-01-HG007585). CIDR is fully funded through a federal contract from the NIH to The Johns Hopkins University, contract number HHSN268201200008I. This research was funded in part through the NIH/NCI Cancer Center Support Grant P30 CA015704.

The Colon Cancer Family Registry (CCFR, www.coloncfr.org) was supported in part by funding from the NCI, NIH (award U01 CA167551) and through U01/U24 cooperative agreements from NCI with the following CCFR centers: Australasian (CA074778 and CA097735), Ontario (OFCCR,CA074783), Seattle (SFCCR, CA074794 and R01 CA076366 to P.A. Newcomb), and the Mayo Clinic (CA074800). Support for case ascertainment was provided in part from the Surveillance, Epidemiology, and End Results (SEER) Program, the Minnesota Cancer Surveillance System (MCSS), the Victoria Cancer Registry (Australia), and the Ontario Cancer Registry (Canada).The content of this article does not necessarily reflect the views or policies of the NIH or any of the collaborating centers in the CCFR, nor does mention of trade names, commercial products, or organizations imply endorsement by the US Government, any cancer registry, or the CCFR.

CPS-II: The American Cancer Society funds the creation, maintenance, and updating of the Cancer Prevention Study-II (CPS-II) cohort. This study was conducted with Institutional Review Board approval.

DALS: NIH (R01 CA48998 to M.L. Slattery).

EPIC: The coordination of EPIC is financially supported by the European Commission (DGSANCO) and the International Agency for Research on Cancer. The national cohorts are supported by Danish Cancer Society (Denmark); Ligue Contre le Cancer, Institut Gustave Roussy, Mutuelle Générale de l'Education Nationale, Institut National de la Santé et de la Recherche Médicale (INSERM, France); German Cancer Aid, German Cancer Research Center (DKFZ), Federal Ministry of Education and Research (BMBF), Deutsche Krebshilfe, Deutsches Krebsforschungszentrum and Federal Ministry of Education and Research (Germany); the Hellenic Health Foundation (Greece); Associazione Italiana per la Ricerca sul Cancro-AIRC Italy and National Research Council (Italy); Dutch Ministry of Public Health, Welfare and Sports (VWS), Netherlands Cancer Registry (NKR), LK Research Funds, Dutch Prevention Funds, Dutch ZON (Zorg Onderzoek Nederland), World Cancer Research Fund (WCRF), Statistics Netherlands (the Netherlands); ERC-2009-AdG 232997 and Nordforsk, Nordic Centre of Excellence programme on Food, Nutrition and Health (Norway); Health Research Fund (FIS), PI13/00061 to Granada, PI13/01162 to EPIC-Murcia, Regional Governments of Andalucía, Asturias, Basque Country, Murcia and Navarra, ISCIII RETIC (RD06/0020, Spain); Swedish Cancer Society, Swedish Research Council and County Councils of Skåne and Västerbotten (Sweden).

Harvard cohorts [HPFS, NHS: HPFS is supported by the NIH (P01 CA055075, UM1 CA167552, U01 CA167552, R01 CA137178, R01 CA151993, R35 CA197735, K07 CA190673, and P50 CA127003)], NHS by the NIH (R01 CA137178, P01 CA087969, UM1 CA186107, R01 CA151993, R35 CA197735, K07 CA190673, and P50 CA127003).

MCCS cohort recruitment was funded by VicHealth and Cancer Council Victoria. The MCCS was further supported by Australian NHMRC grants 509348, 209057, 251553, and 504711 and by infrastructure provided by Cancer Council Victoria. Cases and their vital status were ascertained through the Victorian Cancer Registry (VCR) and the Australian Institute of Health and Welfare (AIHW), including the National Death Index and the Australian Cancer Database.

NFCCR: This work was supported by an Interdisciplinary Health Research Team award from the Canadian Institutes of Health Research (CRT 43821), the NIH, U.S. Department of Health and Human Services (U01 CA74783), and National Cancer Institute of Canada grants (18223 and 18226). The authors wish to acknowledge the contribution of Alexandre Belisle and the genotyping team of the McGill University and Génome Québec Innovation Centre, Montréal, Canada, for genotyping the Sequenom panel in the NFCCR samples. Funding was provided to Michael O. Woods by the Canadian Cancer Society Research Institute.

NSHDS: Swedish Research Council, Swedish Cancer Society, Cutting-Edge Research Grant and other grants from Region Västerbotten, Knut and Alice Wallenberg Foundation, Lion's Cancer Research Foundation at Umeå University, the Cancer Research Foundation in Northern Sweden, and the Faculty of Medicine, Umeå University, Umeå, Sweden.

CCFR: The Colon CFR graciously thanks the generous contributions of their 42,505 study participants, dedication of study staff, and the financial support from the U.S. National Cancer Institute, without which this important registry would not exist.

CPS-II: The authors thank the CPS-II participants and Study Management Group for their invaluable contributions to this research. The authors would also like to acknowledge the contribution to this study from central cancer registries supported through the Centers for Disease Control and Prevention National Program of Cancer Registries, and cancer registries supported by the National Cancer Institute Surveillance Epidemiology and End Results program.

Harvard cohorts (HPFS, NHS, and PHS): The study protocol was approved by the Institutional Review Boards of the Brigham and Women's Hospital and Harvard T.H. Chan School of Public Health, and those of participating registries as required. We would like to thank the participants and staff of the HPFS, NHS, and PHS for their valuable contributions as well as the following state cancer registries for their help: AL, AZ, AR, CA, CO, CT, DE, FL, GA, ID, IL, IN, IA, KY, LA, ME, MD, MA, MI, NE, NH, NJ, NY, NC, ND, OH, OK, OR, PA, RI, SC, TN, TX, VA, WA, and WY. The authors assume full responsibility for analyses and interpretation of these data.

NSHDS investigators thank the Biobank Research Unit at Umeå University, the Västerbotten Intervention Programme, the Northern Sweden MONICA study, and Region Västerbotten for providing data and samples and acknowledge the contribution from Biobank Sweden, supported by the Swedish Research Council (VR 2017-00650).

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.