Genetic risk factors modulate the association between physical activity and colorectal cancer

Abstract

Background

Physical activity (PA) is an established protective factor for colorectal cancer (CRC), but it is unclear if genetic variants modify this effect. To investigate this possibility, we conducted a genome-wide gene–PA interaction analysis.

Methods

Using logistic regression and two-step and joint tests, we analyzed interactions between common genetic variants across the genome and PA in relation to CRC risk. Self-reported PA levels were categorized as active (≥ 8.75 MET-h/wk) vs. inactive (< 8.75 MET-h/wk) and as study- and sex-specific quartiles of activity.

Results

PA had an overall protective effect on CRC (OR [active vs. inactive] = 0.85; 95%CI = 0.81–0.90). The two-step GxE method identified an interaction between rs4779584, an intergenic variant near the GREM1 and SCG5 genes, and PA for CRC risk (p-interaction = 2.6×10^{− 8}). Stratification by genotype at this locus showed a significant reduction in CRC risk by 20% in active vs. inactive participants with the CC genotype (OR = 0.80; 95%CI = 0.75–0.85), but no significant PA–CRC association among CT or TT carriers. When PA was modeled as quartiles, the 1-d.f. GxE test identified that rs56906466, an intergenic variant near the KCNG1 gene, modified the association between PA and CRC (p-interaction = 3.5×10^{− 8}). Stratification at this locus showed that increase in PA (highest vs. lowest quartile) was associated with a lower CRC risk solely among TT carriers (OR = 0.77; 95%CI = 0.72–0.82).

Conclusions

In summary, we identified two genetic variants that modified the association between PA and CRC risk. One of them, related to GREM1 and SCG5, suggests that the bone morphogenetic protein (BMP)-related, inflammatory, and/or insulin signaling pathways may be associated with the protective influence of PA on colorectal carcinogenesis.

BACKGROUND

Colorectal cancer (CRC) is a major global cause of morbidity and mortality. It is the third most commonly diagnosed cancer and second leading cause of death in the world, with more than 1.9 million incident cases and 0.9 million deaths in 2020 [1]. It is predicted that there will be 2.2 million and 3.2 million new CRC cases by 2030 [2] and 2040 [3], respectively, confirming CRC as a major continuing public health burden. The underlying etiology of CRC is multifactorial with a combination of genetic and environmental factors increasing the likelihood of developing CRC [4]. Among these risk factors, physical activity, a lifestyle factor, is an established protective factor against CRC [5–9].

Multiple observational studies and several systematic reviews have shown that regular physical activity (occupational or leisure time) is a modifiable factor associated with lower CRC risk [10–13]. In particular, the World Cancer Research Fund/American Institute for Cancer Research (WCRF/AICR) Continuous Update Project reported lower CRC risk with increased physical activity and classified the evidence linking physical activity to lower CRC risk as ““strong” [5]. Despite the beneficial health effects of physical activity, a recent study reported that more than a quarter of all adults globally were not getting sufficient physical activity [14].

There is substantial understanding of the mechanisms underlying the protective association of physical activity with CRC risk, for example, physical activity is known to have beneficial effects on skeletal muscle mass, immune function, sleep, and mental health [7, 15–21]. Physical activity also reduces obesity (fat mass), which has a beneficial effect on CRC through a reduction in insulin resistance and inflammation, both of which have been associated with CRC development [7, 22–24]. More recently, physical activity has been linked to improved gut microbiome diversity [25]. Further, non-modifiable genetic factors may play a role between physical activity and CRC. However, only a few gene-environment (GxE) interaction studies to date have investigated the association of physical activity with CRC risk according to genetic variants [26–29], all of which were limited by small sample size or restricted to candidate genes/pathways.

Understanding the genetic factors that may influence the relationship between physical activity and CRC risk can offer novel insights into potential biological mechanisms of colorectal carcinogenesis, as well as better inform efforts to promote physical activity and potentially identify individualized physical activity prescriptions. We conducted the largest genome-wide GxE analysis to date, aiming to identify novel genetic variants that may modify the protective association between self-reported physical activity and CRC risk in order to obtain insight into potential mechanisms behind this association.

METHODS

Study participants

The study included individual level genomic and epidemiologic data from three CRC consortia: the multi-centered Colon Cancer Family Registry (CCFR), the Genetics and Epidemiology of Colorectal Cancer Consortium (GECCO), and the Colorectal Cancer Transdisciplinary Study (CORECT), which have been previously described [30–35]. Nested case-control sets were assembled from cohort studies. Control participants were matched on age, sex, and enrollment date/trial group, when applicable. CRC cases were defined as invasive colon or rectal tumors and were confirmed via multiple sources including electronic medical records, pathology reports, state or provincial cancer registries, and/or death certificates. For the small subset of advanced adenomas (7–8%), matched controls were polyp-free and were confirmed by sigmoidoscopy or colonoscopy at the time of adenoma diagnosis. Each study was approved by relevant ethics committees or review boards from respective institutions. All participants provided written informed consent at recruitment.

Data harmonization

Data were collected and centralized at the GECCO consortium coordinating center at the Fred Hutchinson Cancer Center [34]. Briefly, data harmonization consisted of a multi-step procedure, in which common data elements (CDEs) were defined a priori for data harmonization. Study questionnaires and data dictionaries were examined and, through an iterative process of communication with data contributors, elements were mapped to these CDEs. Definitions, permissible values, and standardized coding were implemented into a single database via SAS and T-SQL. Resulting data were checked for errors and outlying values within and between studies [36].

Epidemiologic and lifestyle data collection

Information on demographic, lifestyle, and environmental factors as well as potential risk factors such as age at diagnosis or enrollment, sex, education level, smoking status, total energy consumption (kcal/day), and self-reported or measured weight and height were collected via in-person interviews or through structured self-administered questionnaires in each study. Total energy consumption was derived from the Food Frequency Questionnaires, with missing values imputed by study-sex-specific means. Body mass index (BMI) was calculated using the weight (kg) and height (m) of each participant.

Physical activity exposure measure

Information on physical activity was obtained from structured questionnaires, such as the International Physical Activity Questionnaire (IPAQ) short form [37], European Prospective Investigation into Cancer and Nutrition (EPIC) physical activity questionnaire, and Nurses’ Health Study physical activity questionnaire, among others. Physical activity was estimated in metabolic equivalent tasks hours per week (MET-h/wk), which was derived for each participant, to determine the approximate average amount of time per week that the individual spent in leisure activities or all activities if leisure was not specified.

Moderate activity was defined as 3.5 to 6 MET-h/wk and vigorous activities as ≥ 6 MET-h/wk [38]. Thus, at least 8.75 MET-h/wk approximately corresponds to the current physical activity guidelines of a minimum of 150 minutes (= 2.5 hours) of moderate or 75 minutes of vigorous activity per week as recommended for individuals with cancer or for cancer prevention [39–42]. Based on these guidelines and previously published literature in CRC [43–45], the participants in the present study were categorized into two groups: active (≥ 8.75 MET-h/wk) vs. inactive (< 8.75 MET-h/wk; reference category). Because the majority of the participants were active, we also calculated study- and sex-specific quartiles for physical activity as a secondary variable, where the quartile groups were coded as 1, 2, 3, or 4, respectively. This variable was treated as continuous (change in one quartile) when assessing the association between physical activity and CRC, and as categorical (1st quartile as reference group) in the genome-wide scans.

Genotyping, quality control, and imputation

Detailed information on genotyping, imputation, and quality control have been described previously [30, 32]. In brief, genotyped single nucleotide polymorphisms (SNPs) were excluded based on deviation from Hardy-Weinberg Equilibrium (p < 1×10^{− 4}), low call rate (< 95–98%), discrepancies between reported and genotypic sex, and discordant calls between duplicates. Autosomal SNPs in all studies were imputed to the Haplotype Reference Consortium (HRC) r1.1 (2016) panel using the University of Michigan Imputation Server [46] and treated as dosage for data management and analyses using R package BinaryDosage [47]. Imputed common SNPs were excluded if they had low imputation quality (R² < 0.8) and pooled minor allele frequency (MAF) ≤ 1%. After quality control, a total of over > 7.2 million SNPs were used for the gene-environment interaction analysis, noticeably with high redundancy due to linkage disequilibrium (LD).

Sample size

Analyses were limited to individuals of European ancestry, based on self-reported race and clustering of principal components (PCs) with 1000 Genomes EUR superpopulations [48]. Participants were excluded based on cryptic relatedness or duplicates (prioritizing cases and/or individuals genotyped on the better platform), and genotyping/imputation errors. We also excluded studies that did not collect physical activity data. The pooled sample size for the study- and sex-specific quartile physical activity variable was 42,602 participants from 31 studies (71% prospective cohort studies). For the dichotomous active-inactive physical activity variable, with 8.75 MET-h/wk as the cutoff value, the final pooled sample size was 39,992 participants from 27 studies (74% prospective cohort studies) (Supplementary Table 1).

Statistical Analyses

To evaluate the main effects of physical activity on CRC risk, logistic regression models were conducted for each study, with adjustment for age at diagnosis or enrollment, sex, and total energy consumption (when available). Models with genetic variables were further adjusted for the first three PCs of genetic ancestry to account for potential population substructure. The study-specific results were combined using random-effects meta-analysis methods (Hartung-Knapp) to obtain summary odds ratios (ORs) and 95% confidence intervals (CIs) [49]. The heterogeneity p-values were calculated using Cochran’s Q statistics [50], while funnel plots identified studies with outlying ORs for potential exclusion and sensitivity analyses. Additional models were fitted, stratified by study design (case-control vs. cohort), sex, and tumor site (proximal colon, distal colon, rectal). All meta-analyses were performed using the R package Meta [51].

Genome-wide interaction scans of common markers were conducted in the overall study population to maximize power. For the purposes of this study, E indicates physical activity, G indicates a particular SNP, D indicates CRC disease status, and C refers to a set of adjustment covariables. We utilized not only the traditional logistic regression test of GxE (1-degree of freedom test; 1-d.f.), but also the more powerful joint 3-d.f. test [52, 53] and two-step EDGE method [54–56]. The R package GxEScanR [57] was used to perform these analyses.

For the 1-d.f. test, we examined multiplicative interactions by fitting a traditional logistic regression model including an interaction term in the form: logit (Pr (D = 1|G)) = β. ₀ + β _GG + β _EE + β _GxEGxE + β _CC, where H0 : β _GxE = 0 tests potential departures from multiplicative associations of E and G on D.

We also performed a joint test of association, which can improve power to detect disease susceptibility loci in a wider range of circumstances by accounting for GxE interactions, e.g., in circumstances where susceptibility loci affect only individuals with certain environmental exposure profiles [53, 58]. For this we used the 3-d.f. test of the joint null hypothesis H0 = β _G = β _GxE = γ_G = 0, where β _G and β _GxE are the main and interaction effects from the logistic model above and γ _G represents the association between G and E in the combined case-control sample [53, 59].

We further implemented the two-step EDGE method that assesses GxE interaction tests (step 2) based on ranks of an independent filtering or ranking statistic (step 1) [56]. The two-step method can decrease the multiple testing burden and improve power to detect interaction loci [56, 59, 60], provided that steps 1 and 2 are independent. The original approach uses step 1 ranks to prioritize and partition SNPs into exponentially larger bins of fixed sizes and increasingly more stringent step- 2 significance thresholds. However, when analyzing imputed SNPs, highly correlated markers from the same loci fill the top bins, thereby diminishing statistical power. To address this issue, the original weighted hypothesis-testing framework [61] was modified to accommodate bins of varying sizes while appropriately controlling for type I error [55]. In particular, SNPs were partitioned into bins based on step 1 p-value thresholds in expectation, which were calculated using the original predetermined bin sizes (initial bin size of 5 and overall alpha = 0.05) with assumed uniform distribution of 1 million independent tests. For step 2 GxE testing, the influx of correlated markers into each bin was accounted for by correcting for the effective number of tests, which was estimated using principal component analysis (PCA) performed on bin-specific genotype correlation matrices [54, 55, 62]. This modification reduces multiple testing burden and improves statistical power, while preserving the overall type I error rate at 5%. For any SNP achieving significance at the overall type I error rate, we computed its corresponding SNP-specific p-value accounting for both steps 1 and 2 of the EDGE procedure, to allow direct comparison to the standard GWAS threshold of 5 × 10^{− 8} [62].

To follow-up statistically significant interactions, we estimated stratified ORs by modeling physical activity in relation to CRC within genotypic groups and the per-allele increase in genotype in relation to CRC stratified by physical activity. We also assessed the extent of genomic inflation by creating quantile-quantile (Q-Q) plots and calculating the genomic inflation factor (lambda). Additionally, we calculated lambda₁₀₀₀, which scales the genomic inflation factor to an equivalent study of 1000 cases and 1000 controls, since as lambda scales according to the sample size [63, 64].

To explore variation in GxE effect strengths of association, we also conducted stratified analyses for novel findings by study design, sex, and tumor site. We conducted a sensitivity analysis including the interaction terms GxBMI and E(= physical activity)xBMI in the model, because BMI it is a potential confounder in the physical activity–CRC association [65].

Functional follow-up

Regional plots for all statistically significant findings were generated using the command- line version (standalone) of LocusZoom v1.3 [66] to examine, in depth, the magnitudes of association, the extent of association signal due to LD, and chromosomal position of findings relative to genes in the given region. Measures of LD were estimated using study population controls. The putative functional role of these SNPs and those in LD (R² > 0.5) at 500 kb flanking regions were examined relative to their potential contribution to regulate gene expression by their: i) direct association with expression of nearby genes (expression quantitative trait loci (eQTLs); and ii) physical location in regions of chromatin accessibility or histone modifications (variant enhancer loci).

Possible eQTL relationships were explored using: i) the Genotype-Tissue Expression (GTEx v8); and ii) the University of Barcelona and University of Virginia genotyping and RNA sequencing project (BarcUVa-Seq) dataset, which includes normal colon tissue samples from 445 healthy individuals [67]. In addition, the BarcUVa-Seq project has data on physical activity in 352 (79%) participants, which we also used to test both specific eQTLs for physical activity status (active vs. inactive; study- and sex-specific quartile variable) and interactions between SNPs and physical activity on gene expression. The BarcUVA-Seq models were adjusted for age (years), sex, sequencing batch (one to four), and tissue location (left, right, transverse, missing). The putative functional role of SNPs and those in LD (r2 > 0.2) and MAF > 0.01 at 500kb flanking regions were investigated relative to their potential contribution to regulate gene expression by their physical location in regions of chromatin accessibility or histone modifications (variant enhancer loci). We annotated only suggestive eQTLs, i.e., those having a nominal p-value < 0.05.

Details of the functional- annotations analyses have been previously published [68, 69]. Briefly, we used an assay for transposase-accessible chromatin with sequencing (ATAC-seq), DNaseI Hypersensitivity (DHS)-seq, H3K27ac histone ChIP-seq, H3K4me1 histone ChIP-seq datasets of primary tissue from healthy colon and primary-tumor primary tissue samples containing active enhancer elements from Scacheri et al. [70], as well as from three CRC cell lines (SW480, HCT116, COLO205). These datasets were processed through ENCODE ATAC-seq/DNASE-seq [71] and histone ChIP-seq pipelines [72] to perform alignment and peak calling.

GxE analyses for rare variants

To assess the potential contribution of rare SNPs, we also performed a gene-set-based aggregate tests only for rare SNPs using the Mixed effects Score Test for Interactions (MiSTi) approach [73] as a secondary analysis, as the power for rare SNPs testing usually is low. We examined the interactions of physical activity and aggregated rare SNP sets at the gene and enhancer level using MiSTi (MiSTi R package). We used a Fisher’s combination approach under MiSTi (fMiSTi) to discover GxE interactions [73], after adjusting for age, sex, study, and the first three PCs. Because 25,000 gene regions were tested and this was a secondary analysis, interactions with p < 2×10^{− 6} were considered statistically significant, while whereas those with p < 1×10^{− 4} were considered suggestive.

RESULTS

Study population characteristics

The total sample size was n = 39,992 (16,383 CRC cases and 23,609 controls), with 76% classified as active (i.e., ≥ 8.75 MET-h/wk). Detailed descriptive characteristics of the study population are presented in Table 1. Compared to controls, CRC cases were more likely to be older, female, ever smokers, have a higher BMI and total energy consumption, and have a lower education level (each p < 0.001). Descriptive characteristics of the study population for the secondary physical activity variable assessed as study- and sex-specific quartiles are provided in Supplementary Table 2.

Table 1.

Descriptive characteristics of all study participants by colorectal cancer case-control status with available physical activity data.

Characteristics	Cases (N = 16,383)	Controls (N = 23,609)	P-value
Age (median imputed) a
Mean (SD)	65.0 (±9.4)	63.4 (± 8.3)	<0.001
Sex
Female	8,677 (53%)	12,005 (51%)	<0.001
Male	7,706 (47%)	11,604 (49%)
Total energy consumption (kcal/day; mean imputed) b,c
Mean (SD)	1,967 (±713)	1,910 (±680)	<0.001
BMI (kg/m2) c
Mean (SD)	27.2 (± 4.7)	26.9 (± 4.5)	<0.001
Family history of colorectal cancer c
No	10,430 (64%)	12,945 (55%)	0.06
Yes	2,295 (14%)	2,685 (11%)
Education level (highest completed) c
Less than High School	3,070 (19%)	3,488 (15%)	<0.001
High School/GED	3,366 (21%)	3,161 (13%)
Some College	3,476 (21%)	5,783 (24%)
College/Graduate School	5,601 (34%)	8,488 (36%)
Ever smoker c
No	7,050 (43%)	11,479 (49%)	<0.001
Yes	9,086 (55%)	11,862 (50%)

NOTE: Data might not add to 100% because of rounding.

Abbreviations: SD, standard deviation; BMI, Body-Mass-Index; GED, General Educational Development Test.

Physical activity categorized as active (≥ 8.75 MET-h/wk) vs. inactive (< 8.75 MET-h/wk; reference category) dichotomous variable.

^aAge was assessed at diagnosis or enrollment.

^bCalculations exclude individuals with missing total energy intake information.

^cMissing values not shown.

P-values < 0.05 are statistically significant.

Physical activity and CRC risk

We observed that being active (≥ 8.75 MET-h/wk) vs. inactive (< 8.75 MET-h/wk) was associated with a 15% risk reduction in CRC in the overall meta-analysis (OR = 0.85; 95% CI = 0.81–0.90; Supplementary Fig. 1A; Supplementary Table 3). Sensitivity analyses showed even greater risk reduction for case-control studies (OR = 0.75; 95% CI = 0.66–0.85) compared to cohort-based studies (OR = 0.88; 95% CI = 0.83–0.93). No evidence for heterogeneity was observed across all studies (P_het=0.64; I² = 0%;) or among case-control (P_het=0.36; I² = 9%) or cohort-based studies (P_het=0.91; I² = 0%). Further, analysis stratified by sex showed a risk reduction in both men (OR = 0.83; 95% CI = 0.76–0.90; P_het=0.56; I² = 0%) and women (OR = 0.87; 95% CI = 0.81–0.94; P_het=0.86; I² = 0%) when comparing active vs. inactive participants. For tumor site, the strongest inverse associations were observed for distal colon (OR = 0.77, 95% CI = 0.71–0.84; P_het=0.64; I² = 0%) and proximal colon (OR = 0.84, 95% CI = 0.81–0.90; P_het=0.46; I² = 0%), but not for rectal cancer (OR = 0.94, 95% CI = 0.85–1.04; P_het=0.27; I² = 15%) comparing active vs. inactive participants. For physical activity measured as study- and sex-specific quartiles (treated as a continuous variable), we observed similar risk reductions for the overall meta-analysis as well as for stratified analysis by sex (Supplementary Fig. 1B; Supplementary Table 4). In dose-response (per-quartile) analyses, inverse associations were also observed for rectal cancer (per quartile OR = 0.95; 95% CI = 0.92–0.98; P_het<0.001; I² = 54%) as well as for distal and proximal colon, with some inter-study heterogeneity observed for case-control studies (P_het<0.001; I² = 74%). As we found statistically significant associations between physical activity and CRC for the overall population without significant evidence for heterogeneity, we conducted genome-wide GxE testing in the overall study population to maximize power.

Genome-wide physical activity-interaction scans for CRC risk

The quantile-quantile (Q-Q) plot for the traditional gene-physical activity interactions for CRC risk using 1-d.f. analysis did not show p-value inflation for either primary and or secondary physical activity variables (Supplementary Fig. 2).

Table 2 summarizes the statistically significant gene-physical activity interactions identified. Using the two-step EDGE method and the dichotomous physical activity variable (active vs. inactive), we identified statistically significant interactions for 5 SNPs, all of them in LD, on chromosome 15q13.3 located in the intergenic region between Gremlin 1 (GREM1) and Secretogranin V (SCG5) genes.[74] Among these SNPs with statistically significant interactions, we report only on the interaction of SNP rs4779584 with physical activity in this study (two-step p-value = 2.6×10^{− 8}; Table 2), as this SNP was supported by prior evidence on the association with CRC as main effect (per T allele OR: active = 1.20; 95% CI = 1.10–1.20 vs. inactive = 1.00; 95% CI = 0.93–1.10; Table 3).[75] This result was robust in a sensitivity analysis that further accounted for BMI and interactions with BMI, as well as age, sex, study type, total energy consumption, and the first three PCs of genetic ancestry. Specifically, these additional adjustments caused less than a 2% change in the GxPA interaction estimates. Analysis stratified by rs4779584 genotype showed that participants who were physically active vs. inactive had 20% lower CRC risk among those who were carriers of CC (OR = 0.80; 95% CI = 0.75–0.85; p = 1.6x×10^{− 11}), while this risk reduction was diminished among those carrying the CT (OR = 0.92; 95% CI = 0.84–1.00) and TT (OR = 1.30; 95% CI = 1.00–1.70;) genotypes (Fig. 1; Table 3). We observed similar interaction effects when analyses were stratified for study type, sex, or tumor site (Supplementary Table 5).

Table 2.

Results of genome-wide interaction analyses with physical activity for colorectal cancer risk.

Physical Activity Variable	SNP	Chr	BP Position	Locus	Closest Gene	Reference Allele	Alternate Allele	Alternate Allele frequency	Type	Statistical Method	P-value GxEb
Active / inactivea	rs4779584	15	32994756	15q13.3	GREM1 and SCG5	C	T	0.20	intergenic variant	Two-step EDGE	2.6×10−8
Quartilesc	rs56906466	20	49693755	20q4.5	KCNG1	T	C	0.06	intron	1-d.f. test	3.5×10−8

Abbreviations: SNP, single nucleotide polymorphism; Chr, chromosome; BP Position, base pair position based on NCBI Build 37; 1-d.f., 1-degree of freedom.

^aPhysical activity categorized as active (≥ 8.75 MET-h/wk) vs. inactive (< 8.75 MET-h/wk; reference category).

^bP-value corresponds to the interaction between genetic variants (G) and physical activity (E) on risk of colorectal cancer in the combined case-control population based on the indicated statistical method.

^cPhysical activity assessed as study- and sex-specific quartiles.

P-values that are statistically significant are indicated in bold text.

Notes: Directly genotyped SNPs were coded as 0, 1, or 2 copies of the count allele. Imputed SNPs were coded as expected gene dosage. Multiplicative interaction terms were modelled as the product of PA and each SNP of interest.

Table 3.

Associations between physical activity for colorectal cancer risk stratified by genotypes of SNPs of interest.

SNP	Physical Activity	Homozygous non-carriers			Heterozygous			Homozygous carries of the alternate/minor allele			Per alternative allele within strata of Physical Activity cetegories
		N (Ca/Co)	OR (95% Cl)	P-value	N (Ca/Co)	OR (95% Cl)	P-value	N (Ca/Co)	OR (95% Cl)	P-value	OR (95% Cl)	P-value
		CC			CT			TT
rs4779584	Inactivea	2,537/3,642	1.00 (Ref.)	-	1,304/1,806	1.00 (0.95–1.10)	0.40	137/228	0.87 (0.69–1.10)	0.23	1.00 (0.93–1.10)	0.98
	Activea	7,701/11,960	0.80 (0.75–0.85)	1.6×10−11	4,155/5,372	0.95 (0.89–1.00)	0.19	549/601	1.10 (0.99–1.30)	0.08	1.20 (1.10–1.20)	2.0×10−15
	Active vs. inactive (by genotype)		0.80 (0.75–0.85)	1.6×10−11	-	0.92 (0.84–1.00)	0.05	-	1.30 (1.00–1.70)	0.04
		TT			TC			CC
rs56906466	Q1b	4,168/5,290	1.00 (Ref.)	-	443/715	0.77 (0.67–0.87)	8.0×10−5	20/19	1.10 (0.56–2.10)	0.81	0.77 (0.68–0.88)	1.4×10−4
	Q2b	4,085/5,745	0.91 (0.85–0.96)	0.002	481/710	0.87 (0.76–0.99)	0.03	13/25	0.61 (0.3–1.20)	0.16	0.93 (0.82–1.10)	0.28
	Q3b	3,792/5,896	0.81 (0.76–0.86)	6.8×10−12	469/669	0.87 (0.77–1.00)	0.047	21/26	1.00 (0.56–1.90)	0.96	1.10 (0.99–1.30)	0.08
	Q4b	3,342/5,564	0.77 (0.72–0.82)	1.1×10−16	442/637	0.94 (0.82–1.10)	0.35	17/13	1.90 (0.91–4.20)	0.09	1.30 (1.10–1.50)	5.7×10−4
	Q2 vs. Q1 (by genotype)a	-	0.91 (0.85–0.96)	0.002	-	1.10 (0.95–1.30)	0.16	-	0.56 (0.21–1.50)	0.23
	Q3 vs. Q1 (by genotype)a		0.81 (0.76–0.86)	6.8×10−12		1.10 (0.96–1.40)	0.14		0.93 (0.38–2.30)	0.88
	Q4 vs. Q1 (by genotype)a		0.77 (0.72–0.82)	1.1×10−16		1.20 (1.00–1.50)	0.03		1.80 (0.65–4.90)	0.26

Abbreviations: SNP, single nucleotide polymorphism; PA, physical activity; N, number; Ca/Co, case/control; OR, odds ratio; 95% CI, 95% confidence interval. Case/control counts were calculated by imputed genotype probabilities.

^aPhysical activity categorized as active (≥ 8.75 MET-h/wk) vs. inactive (< 8.75 MET-h/wk; reference category)

^bPhysical activity, assessed as study- and sex-specific quartiles.

P-values that are statistically significant are indicated in bold text.

Figure 1.

Association between physical activity and colorectal cancer risk stratified by genotype of SNP rs4779584. Physical activity is categorized as active (≥8.75 MET-h/wk) vs. inactive (<8.75 MET-h/wk; reference category).

The analysis of physical activity assessed as study- and sex-specific quartiles revealed an interaction with one SNP (rs56906466) on chromosome 20q4.5 located near the Potassium Voltage-Gated Channel Modifier Subfamily G Member 1 (KCNG1) gene, using the traditional 1-d.f. test (GxE p-value = 3.5×10^{− 8}; Table 2; Supplementary Fig. 3B). This result was still consistent in a sensitivity analysis that also considered BMI and interactions with BMI along with age, sex, study type, total energy consumption, and the first three PCs of genetic ancestry. As in the previous sensitivity analysis, these adjustments resulted in less than a 2% variation in the GxPA interaction estimates. Analysis stratified by rs56906466 genotype showed statistically significantly lower CRC risk with increases in physical activity, especially when comparing the highest quartile (Q4) to the lowest quartile (Q1), among those who were carriers of TT (OR = 0.77; 95% CI = 0.72–0.82; p = 1.1×10^{− 16}). The corresponding inverse associations were not observed for those with TC (Q4 vs. Q1: OR = 1.20; 95% CI = 1.–1.50; p = 0.03) and CC (Q4 vs. Q1: OR = 1.80; 95% CI = 0.65–4.90; p = 0.26) genotypes (Table 3). Similar interactions were observed when analyses were stratified by study type, sex, or tumor site (Supplementary Table 5). No other statistically significant interactions were observed (data not shown). Additionally, the GxE analyses for rare variants did not identify any statistically significant interactions. There was also no significant LD-based correlation between rs4779584 and rs56906466 (correlation coefficient, r² = 0.001).

Functional follow-up

Functional annotation analyses around rs4779584 and rs56906466 showed enhanced activities. The SNP rs4779584 and correlated SNPs showed peaks in both normal (i.e., ATAC-seq, H3K4me1) and colon tumor samples (i.e., tumor DHS, tumor H3K27ac) as well as in cancer cell lines (i.e., H3K27ac, H3K4me1). The SNP rs56906466, although not correlated with other SNPs, was identified as a variant enhancer for tumor DHS and cell line DHS (Supplementary Figs. 4–5).

Two independent sources of eQTLs analyses were used to expand on the regulatory roles of SNPs rs4779584 and rs56906466. The SNP rs4779584 was observed to be an eQTL in the GTEx v8 compendium as it modified the expression of GREM1 in liver and pancreas, SCG5 in liver, and RP11- 758N13.1 in brain, cultured fibroblast, liver, and pancreas tissues. We did not observe any statistically significant eQTL findings for SNP rs56906466.

In relation to the BarcUVa-Seq dataset, which provides colon-specific eQTLs, the SNP in the 15q13.3 region did not modify the expression of FNM1, GREM1, SCG5, or other genes in the region (Supplementary Fig. 4). Likewise, the models tested in this dataset on the interaction with physical activity measured in the subjects did not reach statistical significance. The same approach was used to assess whether the SNP rs56906466 and the interaction term had eQTL effects on gene expression, but no statistically significant results were observed.

DISCUSSION

To our knowledge, this is the largest genome-wide study conducted to date to investigate the interactions between variants across the genome and self-reported, harmonized physical activity data. Consistent with previous studies and the WCRF, we observed a statistically significant 15% risk reduction in CRC due to physical activity, similar in magnitude to that previously observed [5, 10–13]. Our analyses identified two novel, statistically significant GxE interactions for physical activity – SNPs rs4779584 and rs56906466 significantly modified the association between physical activity and CRC risk.

The SNP rs4779584, located in the 15q33.3 region, lies between the GREM1 and SCG5 genes and has been previously found to contribute to CRC susceptibility [31, 74, 76–79]. Carrying the T allele in rs4779584 has been reported to be associated with an increased CRC risk of 1.26 (95% CI = 1.19–1.34) as compared to the C allele [80]. In our study, we found that physical activity was significantly associated with a lower risk of CRC only among those with the C allele. GREM1 encodes gremlin 1, which is a signaling protein involved in several pathways relevant to CRC, including the transforming growth factor-β (TGF-β) pathway which has been implicated in tumor invasion and metastasis [81]. GREM1 is also a proangiogenic factor, suggesting a possible role in cancer development when upregulated [82]. Additionally, Gremlin 1 is an insulin antagonist with elevated levels in type 2 diabetes [83], and has been linked to bone morphogenetic proteins (BMPs) signaling imbalance, which accelerates tumor cell proliferation [84], and is associated with inflammatory processes independently of BMPs [85, 86], SCG5 encodes secretogranin V (also named 7B2 protein or SGNE1), an essential neuroendocrine signaling molecule that plays a role in cellular proliferation [87, 88]. Although SCG5 is associated with polyposis syndromes which is linked with CRC risk [89], its direct role in CRC is not as well characterized as compared to GREM1’s role in CRC [90]. Further, some studies have also reported a role of SCG5 in BMI modulation [91, 92]. The identified interactions suggest that the CRC risk reduction due to physical activity may be related to one or several more of these above-mentioned pathways.

There are only a small number of GWAS studies that have identified genetic loci associated with physical activity [93, 94], with one preclinical study suggesting that exercise training epigenetically reprograms GREM1 expression [95]. However, to our knowledge, no prior studies have reported an interaction between rs4779584 and physical activity on CRC risk. The epidemiologic evidence indicating the beneficial effect of physical activity on CRC risk is extensive, and several biological mechanisms have been identified or proposed, including in some intervention studies, such as physical activity’s effect on immune system, systemic inflammatory markers, energy regulation, hormones levels, insulin resistance, and gut microbial composition [7, 96–98]. Related to our findings, a randomized trial conducted in obese patients who followed different resistance training protocols observed significant reductions in plasma gremlin 1 and C-reactive protein levels compared to a control group [99]. Additionally, myokines (i.e., cytokines), such as myostatin (member of the TGF-β family) or interleukin-6, are secreted by the skeletal muscle in response to intensity training [100, 101]. The effect of regular exercise on SCG5, the other gene close to the SNP rs4779584 that showed interactions with physical activity on CRC risk, has been investigated in experimental studies using animal models. However, the results were inconclusive, with one study reported non-significantly decreased SCG5 expression, while the other study reported significantly increased expression levels [102, 103]. Future studies are warranted to describe the plausible biological mechanism by which SNP rs4779584 interacts with physical activity and modifies CRC risk, but on the basis of our findings, genetic markers in this region showed enhanced activity in both normal and tumor samples suggesting a potential regulatory role on transcription of adjacent genes. Consistent with this, we observed that SNP rs4779584 modified the expression of GREM1 and SCG5 in pancreas and liver, but not in colon tissue.

We also discovered a new locus rs56906466 located near KCNG1 that has not been previously associated with CRC, physical activity, or its interaction with physical activity on CRC risk. This gene encodes a member of the large gene family that instructs the building of potassium channels and is abundantly expressed in skeletal muscle. KCNG1 has been related to insulin secretion, muscle contraction, and neurotransmitter release regulation, among others [104]; however, its functions are not fully understood. Our findings showed that rs56906466 had statistically significant interactions with physical activity in modifying CRC risk. Furthermore, functional-annotations analyses demonstrated that some of the genetic variants interacting with physical activity were located in enhancers and were linked to differential gene expression. However, additional targeted studies will be necessary to further investigate the joint effects of these genes with physical activity on CRC risk.

There is increasing evidence that gene-physical activity interactions (including being physically active or inactive) have an effect on several health-related outcomes such as blood pressure, hypertension, BMI, and insulin metabolism [105]. However, few studies have evaluated the gene-physical activity interaction on CRC risk, and all previous studies followed a candidate-gene approach and included only a limited number of SNPs [26–29]. Two studies evaluated the mediating effects of physical activity on CRC risk via alterations in polymorphisms in the insulin-like growth factor-1 (IGF-1) gene, since physical activity is known to modulate IGF-1 serum levels, and observed statistically significant interactions [26, 106]. Khoury-Shakour et al. focused their analysis on the polymorphism rs2665802 at intron 4 of the growth hormone 1 (GH1) gene and observed that the minor allele A was associated with lower risk of CRC among inactive participants [26]. A recent study assessed the interaction between physical activity and CRC risk based on a polymorphism (rs647161) in the paired-like homeodomain 1 (PITX1) gene in a Korean population, and reported a higher risk of CRC among participants who exercised less and carried the minor allele [27]. PITX1 is considered a tumor suppressor gene [107], and is known to influence the expression of GH1, and is related to IGF-1 [108]. Song et al. assessed interactions between physical activity and 31 SNPs (including rs4779584) on CRC risk among 703 CRC cases and 1,406 healthy controls [28]. However, they observed statistically significant interactions only with rs4444235––with increased CRC risk among C carriers who exercised regularly––but not for rs4779584, which may be due to the small sample size. However, none of the above findings could be replicated in the present study (data not shown). Additionally, we observed no LD-based correlation between rs4779584 and rs4444235 (r² = 0.004). Given the smaller sample size and candidate gene approach in the study by Song et al., it is possible that these are chance findings.

A main strength of our study was a large, well-characterized study population, the largest ever to have examined gene-physical activity interactions. The use of several complementary statistical approaches was also a strength of this study as it allowed detection of specific loci within GREM1 and SCG5 and near KCNG1 genes. However, our findings may not be generalizable outside of European-descent populations as the participants in this study were limited to those with European descent and were far more active than the US general population. The consortium is actively striving to overcome this limitation by expanding our research to encompass other racial and ethnic groups, as well as by harmonizing epidemiological data, which will enable us to expand our future GxE analyses. Additionally, this study included self-report measures of physical activity which are prone to recall and response biases, but these are likely to attenuate ‘true’ associations with disease risk [109]. Lastly, our sample size did not allow us to identify genes, whose rare variants may interact with physical activity and contribute to CRC risk in the aggregate test. Additional functional studies are needed to verify the role of the identified SNPs interacting with physical activity for CRC risk.

CONCLUSIONS

In conclusion, we identified two novel genetic loci that interact with physical activity to influence CRC risk. Potential mechanisms behind the interaction of rs4779584 and physical activity in CRC risk may be linked in part to the BMP-related, inflammation pathways, and/or insulin signaling in response to physical activity. However, SNP rs56906466 that is near a potassium channel gene, has not been previously described in relation to physical activity or CRC, and additional investigations are required to elucidate the potential mechanisms through which it may be involved in colorectal carcinogenesis, especially in individuals who are not physically active.

Supplementary Files

This is a list of supplementary files associated with this preprint. Click to download.

• SupplementaryTable1GxEPAv6FINAL.xlsx

• SUPPLEMENTARYTABLESANDFIGURES.docx

Funding Statement

Genetics and Epidemiology of Colorectal Cancer Consortium (GECCO): National Cancer Institute, National Institutes of Health, U.S. Department of Health and Human Services (U01 CA137088, R01 CA059045, U01 CA164930, R21 CA191312, R01201407, R01CA488857, R01CA273198, R01CA244588). Genotyping/Sequencing services were provided by the Center for Inherited Disease Research (CIDR) contract number HHSN268201700006I and HHSN268201200008I. This research was funded in part through the NIH/NCI Cancer Center Support Grant P30 CA015704. Scientific Computing Infrastructure at Fred Hutch funded by ORIP grant S10OD028685. Statistical methodology and software development at USC funded by P01CA196569.

Colon Cancer Family Registry (CCFR): CCFR (www.coloncfr.org) is supported in part by funding from the National Cancer Institute (NCI), National Institutes of Health (NIH) (award U01 CA167551). Support for case ascertainment was provided in part from the Surveillance, Epidemiology, and End Results (SEER) Program and the following U.S. state cancer registries: AZ, CO, MN, NC, NH; and by the Victoria Cancer Registry (Australia) and Ontario Cancer Registry (Canada). The CCFR Set-1 (Illumina 1M/1M-Duo) and Set-2 (Illumina Omni1-Quad) scans were supported by NIH awards U01 CA122839 and R01 CA143247 (to GC). The CCFR Set-3 (Affymetrix Axiom CORECT Set array) was supported by NIH award U19 CA148107 and R01 CA81488 (to SBG). The CCFR Set-4 (Illumina OncoArray 600K SNP array) was supported by NIH award U19 CA148107 (to SBG) and by the Center for Inherited Disease Research (CIDR), which is funded by the NIH to the Johns Hopkins University, contract number HHSN268201200008I. Additional funding for the OFCCR/ARCTIC was through award GL201–043 from the Ontario Research Fund (to BWZ), award 112746 from the Canadian Institutes of Health Research (to TJH), through a Cancer Risk Evaluation (CaRE) Program grant from the Canadian Cancer Society (to SG), and through generous support from the Ontario Ministry of Research and Innovation. The SFCCR Illumina HumanCytoSNP array was supported in part through NCI/NIH awards U01/U24 CA074794 and R01 CA076366 (to PAN). The content of this manuscript does not necessarily reflect the views or policies of the NCI, NIH or any of the collaborating centers in the Colon Cancer Family Registry (CCFR), nor does mention of trade names, commercial products, or organizations imply endorsement by the US Government, any cancer registry, or the CCFR.

COLO2&3: National Institutes of Health (R01 CA060987).

Colorectal Cancer Transdisciplinary (CORECT) Study: The CORECT Study was supported by the National Cancer Institute, National Institutes of Health (NCI/NIH), U.S. Department of Health and Human Services (grant numbers U19 CA148107, R01 CA81488, P30 CA014089, R01 CA197350; P01 CA196569; R01 CA201407) and National Institutes of Environmental Health Sciences, National Institutes of Health (grant number T32 ES013678).

CPS-II: The American Cancer Society funds the creation, maintenance, and updating of the Cancer Prevention Study-II (CPS-II) cohort. The study protocol was approved by the institutional review boards of Emory University, and those of participating registries as required.

DACHS: This work was supported by the German Research Council (BR 1704/6–1, BR 1704/6–3, BR 1704/6–4, CH 117/1–1, HO 5117/2–1, HE 5998/2–1, KL 2354/3–1, RO 2270/8–1 and BR 1704/17–1), the Interdisciplinary Research Program of the National Center for Tumor Diseases (NCT), Germany, and the German Federal Ministry of Education and Research (01KH0404, 01ER0814, 01ER0815, 01ER1505A and 01ER1505B).

DALS: National Institutes of Health (R01 CA48998 to M. L. Slattery).

EDRN: This work is funded and supported by the NCI, EDRN Grant (U01 CA 84968–06).

EPIC: The coordination of EPIC is financially supported by International Agency for Research on Cancer (IARC) and also by the Department of Epidemiology and Biostatistics, School of Public Health, Imperial College London which has additional infrastructure support provided by the NIHR Imperial Biomedical Research Centre (BRC). 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), German Institute of Human Nutrition Potsdam- Rehbruecke (DIfE), Federal Ministry of Education and Research (BMBF) (Germany); Associazione Italiana per la Ricerca sul Cancro-AIRC-Italy, Compagnia di SanPaolo 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); Health Research Fund (FIS) - Instituto de Salud Carlos III (ISCIII), Regional Governments of Andalucía, Asturias, Basque Country, Murcia and Navarra, and the Catalan Institute of Oncology - ICO (Spain); Swedish Cancer Society, Swedish Research Council and County Councils of Skåne and Västerbotten (Sweden); Cancer Research UK (14136 to EPIC-Norfolk; C8221/A29017 to EPIC-Oxford), Medical Research Council (1000143 to EPIC-Norfolk; MR/M012190/1 to EPIC-Oxford). (United Kingdom).

Harvard cohorts (HPFS, NHS): HPFS is supported by the National Institutes of Health (P01 CA055075, UM1 CA167552, U01 CA167552, R01 CA137178, R01 CA151993, and R35 CA197735), and NHS by the National Institutes of Health (R01 CA137178, P01 CA087969, UM1 CA186107, R01 CA151993, and R35 CA197735).

Hawaii Adenoma Study: NCI grants R01 CA072520.

LCCS: The Leeds Colorectal Cancer Study was funded by the Food Standards Agency and Cancer Research UK Programme Award (C588/A19167).

MEC: National Institutes of Health (R37 CA054281, P01 CA033619, and R01 CA063464).

NCCCS I & II: We acknowledge funding support for this project from the National Institutes of Health, R01 CA66635 and P30 DK034987.

NFCCR: This work was supported by an Interdisciplinary Health Research Team award from the Canadian Institutes of Health Research (CRT 43821); the National Institutes of Health, U.S. Department of Health and Human Serivces (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.

Swedish Mammography Cohort and Cohort of Swedish Men: This work is supported by the Swedish Research Council /Infrastructure grant, the Swedish Cancer Foundation, and the Karolinska Institute’s Distinguished Professor Award to Alicja Wolk.

UK Biobank: This research has been conducted using the UK Biobank Resource under Application Number 8614.

VITAL: National Institutes of Health (K05 CA154337).

WHI: The WHI program is funded by the National Heart, Lung, and Blood Institute, National Institutes of Health, U.S. Department of Health and Human Services through contracts 75N92021D00001,75N92021D00002, 75N92021D00003, 75N92021D00004, 75N92021D00005.

Data availability

The dataset used in the current study may be available from the corresponding author on reasonable request for researchers who meet the criteria for access to confidential data.