Genome-wide interaction study with smoking for colorectal cancer risk identifies novel genetic loci related to tumor suppression, inflammation and immune response

Abstract

Background:

Tobacco smoking is an established risk factor for colorectal cancer (CRC). However, genetically-defined population subgroups may have increased susceptibility to smoking-related effects on CRC.

Methods:

A genome-wide interaction scan was performed including 33,756 CRC cases and 44,346 controls from three genetic consortia.

Results:

Evidence of an interaction was observed between smoking status (ever vs never smokers) and a locus on 3p12.1 (rs9880919, p=4.58x10⁻⁸), with higher associated risk in subjects carrying the GG genotype (OR 1.25, 95%CI 1.20-1.30) compared with the other genotypes (OR <1.17 for GA and AA). Among ever smokers, we observed interactions between smoking intensity (increase in 10 cigarettes smoked per day) and two loci on 6p21.33 (rs4151657, p=1.72x10⁻⁸) and 8q24.23 (rs7005722, p=2.88x10⁻⁸). Subjects carrying the rs4151657 TT genotype showed higher risk (OR 1.12, 95%CI 1.09-1.16) compared with the other genotypes (OR <1.06 for TC and CC). Similarly, higher risk was observed among subjects carrying the rs7005722 AA genotype (OR 1.17, 95%CI 1.07-1.28) compared with the other genotypes (OR <1.13 for AC and CC). Functional annotation revealed that SNPs in 3p12.1 and 6p21.33 loci were located in regulatory regions, and were associated with expression levels of nearby genes. Genetic models predicting gene expression revealed that smoking parameters were associated with lower CRC risk with higher expression levels of CADM2 (3p12.1) and ATF6B (6p21.33).

Conclusions:

Our study identified novel genetic loci that may modulate the risk for CRC of smoking status and intensity, linked to tumor suppression and immune response.

Impact:

These findings can guide potential prevention treatments.

Introduction

Colorectal cancer (CRC) is one of the most commonly diagnosed malignancies globally, being the 2^nd and 3^rd most frequent cancer among women and men, respectively (1).

Germ-line genetics play an important role in the etiology of this complex disease. Genome-wide association studies (GWAS) to date have identified approximately 150 independent loci associated with CRC risk (2-6). However, the identified GWAS risk loci explain less than 20% of the heritability of CRC, leaving a large fraction unexplained (7). Gene-environment (GxE) interactions have been postulated to identify novel genetic risk loci and to explain some of this missing heritability (8).

Among modifiable cancer risk factors, tobacco smoking is the single strongest cause of cancer worldwide. It has been observed that between 8 to 35% of cancer deaths are attributable to smoking in the United States (9); while recent cohort and literature-based studies of wide-world populations estimated that around 10% of CRC cases can be attributable to cigarette smoking (10,11). There is sufficient evidence to consider cigarette smoking as a causal risk factor for CRC, as reported in the last Monograph of the International Agency for Research on Cancer (IARC) (12). Ever smokers showed a risk increase of 18-20% for CRC as compared with never smokers. This risk was higher for rectal cancer, and for CRC tumors with CpG island methylator phenotype (CIMP), microsatellite instability (MSI), and presence of BRAF somatic mutations, features often seen in tumors derived through the serrated pathway (13,14). Additionally, CRC risk increases with smoking intensity and duration in a dose-dependent manner (14,15). However, a recent study using an instrumental approach found no causal association between ever having smoked regularly and CRC risk; while positive association between lifetime amount of smoking and CRC risk was observed (16). The dichotomization of the continuous phenotype (amount of smoking) to the binary exposure (ever having smoked) was done at 100 cigarettes over the course of life. In an instrumental setting, the causal estimates for a binary exposure assume that the causal effect is a stepwise function at the point of dichotomization (17), which could not reflect the relationship between smoking and CRC risk.

Tobacco use has been observed to interact with germ-line genetics, providing novel risk loci for some cancers, such as lung (18), pancreatic (19) and bladder cancers (20). Furthermore, our group estimated that a proportion of CRC risk heritability is explained by interactions between smoking and common genetic variants (21). However, our previous genome-wide interaction study (GWIS) of smoking and CRC risk did not identify any interaction that reached genome-wide significance (22). At that time, the study included 11,219 cases and 11,382 controls, and the null results were possibly due to the limited sample size.

The standard interaction test generally needs a sample size at least 4 times larger than that required to detect a main effect of comparable magnitude (23). However, novel statistical approaches enable an increase in power to detect GxE interactions. For instance, joint tests can detect novel susceptibility loci by accounting for both main and interaction effects, while two-step methods prioritize SNPs to decrease multiple testing burden by applying a filtering step prior to GxE testing (24).

In this study, we applied a set of interaction tests in a combined large dataset from three CRC consortia to identify novel genetic risk loci among: a) nearly 35,000 cases and 45,000 controls with smoking status data, and b) over 13,000 cases and 16,000 controls among current or former smokers with smoking intensity and duration data. Significant results were stratified by colon subsite (proximal colon, distal colon and rectum) and, in a subset of cases, by tumor molecular markers including CIMP, MSI, and BRAF and KRAS mutations. Additionally, significant lead and correlated genetic variants were evaluated for chromatin accessibility and associations with gene expression of nearby genes. For the identified genes, genetic models predicting expression were developed, and interaction between predicted gene expression and smoking was tested on the CRC consortia datasets.

Materials and Methods.

Study participants and smoking habits

A total of 34 studies (cohort and case-control) comprising individuals of European ancestry 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) (2). 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 matched on age, sex, race, and enrollment date/trial group, when applicable. Cases were defined as colorectal adenocarcinoma or advanced adenomas, and were confirmed by medical records, pathological reports, or death certificate information. For the small subset of advanced adenoma cases, matched controls displayed polyp-free sigmoidoscopy or colonoscopy at the time of adenoma selection. All participants gave written informed consent and studies were approved by their respective Institutional Review Boards.

Lifestyle and environmental risk factors were collected by telephone or in-person interviews and structured self-administrated questionnaires. Harmonized quality-control checks were performed in each study following similar criteria (25,26). Participants with any available present or past smoking habits comprised a total of 33,756 CRC cases (14,975 never vs 18,781 ever cigarette smokers) and 44,346 controls (21,976 never vs 22,370 ever cigarette smokers). Among ever smokers, 13,320 cases and 16,176 controls also had indicators of smoking intensity (cigarettes smoked per day), and 12,531 cases and 15,271 controls had data on duration of cigarette smoking (packs of cigarettes smoked in years of habit; pack-years) (Supplementary Table S1). For the analysis of ever versus never smoking variable, we excluded 184 participants from the Alpha-Tocopherol, Beta Carotene Cancer Prevention Study (ATBC) since study participants only included ever smokers. Never smokers were excluded from analyses of smoking intensity and duration. To improve the interpretation of the results, smoking intensity and duration parameters were rescaled to units of 10 cigarettes smoked per day and 20 pack-years, respectively, in order to approximate one unit of the rescaled measures to one standard deviation of the raw measures.

Genotyping and imputation

Details on quality control and genotyping were previously published (2,27), and the genotyping arrays used are summarized in Table S1. Briefly, exclusion criteria included: single nucleotide polymorphisms (SNPs) with missing call rate >2-5%, departure from Hardy-Weinberg equilibrium (HWE) (p<1x10⁻⁴), inconsistencies between self-reported and genotypic sex, and discordant genotype calls within duplicate samples. Genotypes were imputed to the Haplotype Reference Consortium (HRC) panel (28) (39.1 million variants) using the University of Michigan Imputation Server (29), 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 over 7.2 million SNPs were selected for subsequent 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 of imputation quality score. We observed that the first three PCs explained the 72% of the variation and that the additional PCs explained a small and decreasing amount of the variation.

Statistical analyses

The association of smoking exposure variables with CRC risk was assessed by meta-analysis of study-specific estimates, adjusted by sex, age, study, and three principal genetic components to account for population stratification. Between-study heterogeneity was investigated using the Chi-squared test, and inconsistency was measured using the I² statistic, which represents the proportion of total variation attributable to between-study variance. No outlier studies were identified, by estimating the posterior probability (>0.99) of having outliers based on mixture random effects using the “outlierProbs” function of metaplus R package. We also assessed smoking habits with CRC risk stratified by sex and by tumor site. Between-sex and between-site heterogeneity were investigated using the chi-square test, and inconsistency was measured using the I² statistic.

To identify novel CRC interaction loci, 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. All the analyses were adjusted for sex, age, study, and three principal genetic components to account for population stratification. These covariates did not lead to any missing values. No single GxE test is universally most powerful and with that in mind, we applied three different testing procedures to maximize the chance of discovering novel loci related to CRC. These include the standard one-degree-of-freedom (1df) test of GxE interaction, a 3df joint test, and an efficient two-step test.

All three of the abovementioned tests utilize a logistic regression model of the form: logit(Pr(D = 1∣G)) = β₀ + β_GG + β_EE + β_GxEGxE + β_CC, where D is CRC, G is a particular SNP, E is smoking, and C are the adjustment covariates. The standard 1df GxE test is based on a likelihood ratio test of the null hypothesis H0: β_GxE = 0. The 3df test (30) extends the well-known 2df joint test (31) of H0: β_G = β_GxE = 0 to test H0: β_G = β_GxE = δ_G = 0, where δ_G represents the association between G and E in the combined case-control sample (32). The component sources of information for the 3df test have been shown to be statistically independent (33), thus guaranteeing that the overall Type I error rate for this test is preserved. The two-step procedure weights GxE interaction tests (step 2, based on β_GxE) using ranks of a filtering statistic (step 1, based on β_G and δ_G) (34) under the weighted hypothesis testing framework (35,36) (Supplementary Fig. S1). The two-step procedure can decrease multiple testing burden and improve power to detect interaction loci. A full description of this two-step methodology can be found in Supplementary data.

Only SNPs not previously associated with CRC risk or smoking phenotypes, and not in linkage disequilibrium (LD) with variants associated with these phenotypes were considered. The likelihood of confounding is substantially smaller when testing for interactions as a confounder needs to be correlated with both the main effects and the interaction (37). However, since it has been observed that obesity is causally influencing smoking habits using an instrumental setting (38), SNPs that reached genome-wide significance (5x10⁻⁸) in the scans were further adjusted by body-mass index. In addition, significant SNPs were further assessed by stratifying by sex, and cases by tumor location (proximal colon, distal colon and rectum) or tumor molecular markers, when available, including CIMP, MSI or microsatellite stable (MSS), and BRAF and KRAS somatic mutation status.

Functional follow up

Regional plots for all significant findings were generated, which enables inspection of strength of association, the extent of association signal and LD, and position of findings relative to genes in the region. Plots were generated using the software LocusZoom v1.3 (39). Measures of LD were estimated using European populations of the 1000 Genomes Project.

The putative functional role of these SNPs and those in LD (R²>0.5) at 500 kb flanking regions were investigated with relation to their potential contribution to regulate gene expression in two ways: first, by their physical location in regions of chromatin accessibility or histone modifications, and second, through their direct association with expression of nearby genes (expression quantitative trait loci: eQTLs). To evaluate the physical location we obtained regions containing active enhancer elements in tissue from healthy colon and from tumor colon tissue samples from previously analysed ATAC-seq, DNaseI Hypersensitivity (DHS)-seq, and H3K27ac histone ChIP-seq datasets (40). To identify eQTLs, we used the following databases: colon transverse tissue sample from GTEx v8 dataset (41), and Colon Transcriptome Explorer (CoTrEx 2.0; https://barcuvaseq.org/cotrex/, accessed on May 2021) of the University of Barcelona and University of Virginia genotyping and RNA sequencing (BarcUVa-Seq) project dataset, which is comprised of 445 epithelium-enriched healthy colon biopsies from ascending, transverse and descending colon (42). We reported on all genes for which expression was associated with the lead and correlated SNPs. In addition, for loci at which gene expression was associated with lead and correlated SNPs, eQTL models predicting gene expression were developed through an elastic net regularized regression (43) using SNPs with MAF>0.01 and imputation R²>0.7 of the BarcUVa-Seq project dataset, and validated in the colon transverse tissue sample from GTEx v8 dataset (34). These predicted gene expression levels were converted into quartiles and tested for interaction with smoking on CRC risk using the previously described logistic regression models.

Data availability

The summary statistics data that support the findings of this study are available on request from the corresponding authors.

Results

Smoking habits and CRC risk

Ever smoking was more prevalent among CRC cases (56%) than controls (50%, p<0.001) (Supplementary Table S2). Among ever smokers, CRC cases were heavier smokers than controls (18.6 ±11.5 vs 18.0 ±11.4 cigarettes smoked per day p<0.001; and 26.8 ±22.4 vs 24.6 ±21.6 pack-years, p<0.001) (Supplementary Table S1).

We observed that ever being a cigarette smoker was associated with increased CRC risk (odds ratio (OR) 1.25, 95% confidence interval (CI) 1.20-1.30; p_het=0.07; I²=23%) (Fig. 1). Sensitivity analyses showed a slightly lower estimate for cohort-based studies (OR 1.22, 95%CI 1.16-1.29) compared with case-control studies (OR 1.28, 95%CI 1.19-1.37). No evidence for heterogeneity was observed among cohort-based studies (p_het=0.33; I²=9%), while results among case-control studies showed moderate heterogeneity (p_het=0.05; I²=34%) (Supplementary Table S3). Analyses stratified by sex showed higher risk for men (OR 1.32, 95%CI 1.24-1.40) than for women (OR 1.19, 95%CI 1.12-1.26) (p_{sex difference}=0.01) (Fig. 1). The association between smoking and CRC varied by tumor site, with the strongest association observed in rectum (OR 1.34, 95%CI 1.24-1.45), compared with distal colon (OR 1.23, 95%CI 1.15-1.31) and proximal colon (OR 1.18, 95%CI 1.13-1.24) (p_{site difference}=0.03) (Fig. 1). Among smokers, smoking intensity and duration parameters also showed a positive association with CRC risk (OR 1.06, 95%CI 1.03-1.09, per an increase in 10 cigarettes smoked per day; and OR 1.11, 95%CI 1.08-1.15, per an increase in 20 pack-years). Similar patterns for smoking intensity and duration were observed in analyses stratified by sex and tumor site (Fig. 1). Given that these results showed CRC-smoking associations overall and across all subsets, we focused our genome-wide GxE testing on the overall study population to maximize power. However, to explore variation in GxE effect estimates, we also conducted stratified analyses for novel findings.

Figure 1-.

Association parameters of smoking habits for colorectal cancer risk in the overall sample and stratified by sex and tumour site.

Genome-wide smoking-interaction scans for CRC risk

Genomic control inflation and quantile-quantile (QQ) plots for the SNP-smoking interactions for risk of CRC at the genome-wide level did not show evidence for residual population stratification (Supplementary Fig. S2).

The initial conventional test did not identify any genome-wide significant loci interacting with smoking status on the risk of CRC (Fig. 2). However, based on the 3-df test, we identified a novel susceptibility locus at chromosome 3p12.1 (p<=4.58x10⁻⁸) (Table 1). The most significant SNP in this locus is rs9880919, located in an intron of cell adhesion molecule 2 (CADM2, 3p12.1) gene (Supplementary Table S4; Supplementary Fig. S3). This SNP showed evidence of an interaction with smoking status on CRC risk, as well as direct associations with both CRC and smoking status (Table 1). When stratified by genotype for this SNP, smoking status showed higher risk in those carrying the GG genotype (OR 1.25, 95%CI 1.20-1.30, p=4.80x10⁻²⁷) compared with those carrying the GA genotype (OR 1.17, 95%CI 1.11-1.23, p=2.40x10⁻⁹), and the AA genotype (OR 1.12, 95%CI 0.99-1.27, p=0.08) (Fig. 3; Supplementary Table S5). We observed similar interaction effects when analyses were stratified for study type, sex or tumor site (Supplementary Table S4). The two-step approach did not identify any significant interactions (Supplementary Fig. S4).

Figure 2-.

Manhattan plots for the standard interaction tests of smoking habits for colorectal cancer risk. A: Smoking status (ever vs never smokers). B: Cigarettes smoked per day. C: Pack-years.

Table 1-. Association parameters of identified genetic variants interacting with smoking habits on colorectal cancer risk.

rs number: single nucleotide polymorphism (SNP) code, Chr: chromosome, Bp: base pair position in hg19, Ref Al: reference allele, Eff Al: effect allele, Eff Al Freq: reference allele frequency, P1: P value of genetic component for CRC risk (Ho: βG=0), P2: P value of interaction component for CRC risk (Ho: βGxE=0), P3: P value of genetic component for the exposure (Ho:δG=0), P3df; P value of 3 df test (Ho:βG=βGxE=δG=0). Allele frequencies calculated in 1000G EUR population.

Smoking habit	rs number	Chr	Bp	Locus	Ref Al	Eff Al	Eff Al Freq	Gene	P1 (βG)	P2 (βGxE)	P3 (δG)	P3df
Status (ever vs never)	rs9880919	3	85,461,302	3p12.1	G	A	0.243	CADM2	1.86x10−6	0.01	4.20x10−3	4.58x10−8
Cigarettes smoked per day	rs4151657	6	31,917,540	6p21.33	T	C	0.365	CFB	0.21	1.72x10−8	0.04	3.52x10−8
	rs7005722	8	138,788,813	8q24.23	A	C	0.743	Intergenic	0.51	2.84x10−8	0.66	6.81x10−7

Figure 3-.

Association parameters of smoking habits for colorectal cancer risk stratified by genotypes.

Among smokers, we identified two genome-wide significant loci interacting with smoking intensity (cigarettes smoked per day) on the risk of CRC (Fig. 2). The first locus was on chromosome 6p21.33, with rs4151657 being the most significant interacting SNP (1df GxE p=1.72x10⁻⁸, 3df test p=3.52x10⁻⁸, Table 1) located in the intron of complement factor B (CFB) gene (Supplementary Table S4; Supplementary Fig. S5). When risk for CRC of smoking intensity was stratified by rs4151657 genotype, we observed that the association between smoking intensity and CRC risk was stronger in those carrying the TT genotype (OR 1.12, 95%CI 1.09-1.16, p=1.30x10⁻¹², per an increase in 10 cigarettes smoked per day) compared with those carrying the TC genotype (OR 1.06, 95%CI 1.03-1.10, p=1.10x10⁻⁴) or the CC genotype (OR 0.94, 95%CI 0.89-0.99, p=0.03) (Fig. 3; Supplementary Table S6). The second locus was on chromosome 8q24.23 with rs7005722 being the most significant interaction SNP (1df GxE p=2.88x10⁻⁸, Table 1) located in an intergenic region between an uncharacterized non-coding RNA LOC101927915 and family with sequence similarity 135 member B (FAM135B) gene (Supplementary Table S4 and Supplementary Fig. S5). When stratified by genotype, we observed that smoking intensity was only associated with CRC risk in those with genotypes carrying the A allele (OR 1.17, 95%CI 1.07-1.28, p=6.70x10⁻⁵; and OR 1.13, 95%CI 1.10-1.16, p=3.40x10⁻¹³; among individuals with AA and AC genotypes, respectively, for each 10 extra cigarettes smoked per day) but not in those carrying CC genotype (OR 1.01, 95%CI 0.98-1.04, p=0.48) (Fig. 3; Supplementary Table S6). Similar interaction patterns were observed when analyses were stratified by study type, sex or tumor colon site for both rs4151657 and rs7005722 (Supplementary Table S4). We did not identify novel susceptibility loci interacting with cigarettes per day using 3-df joint test or two-step test (Supplementary Fig. S4) and did not observe interactions with pack-years by any method (Fig. 2, and Supplementary Fig. S4).

Further adjustment of interaction scans with body-mass index did not modify observed interactions between genetics and smoking parameters on CRC risk (Supplementary Table S7).

SNP-smoking interactions by CRC tumor molecular markers

To investigate if the above observed interactions differed by tumor subtypes defined by CIMP, MSI, BRAF and KRAS mutation status, we estimated the interaction in a subset with available data on these molecular characteristics (Supplementary Table S8). Similar to results using the whole dataset, the identified SNPs showed some evidence of interaction in all tumor marker subtypes, with the exception of the SNP in 3p12.1. However, the strength and significance of the interaction were different across the subtypes (Supplementary Table S9). For both rs4151657 (6p21.33) and rs7005722 (8q24.23), the interaction was observed for CIMP-, MSS, BRAF non-mutated, and KRAS both mutated and non-mutated tumors (Supplementary Table S9).

Functional annotations of the genetic loci

Regarding functional annotation, SNPs correlated with rs9880919 (3p12.1) and rs4151657 (6p21.33) were located in accessible chromatin regions, and H3K27ac and DHS peaks in non-malignant colon samples, and in H3K27ac peaks in colon tumor samples (Table 2). This suggests a potential regulatory role of these genetic regions on transcription of nearby genes in non-malignant and tumor samples. The SNP rs7005722 (8q24.23) was not correlated with other SNPs, nor was it identified as a variant enhancer locus or eQTL of nearby genes (Table 2). Lead and correlated SNPs in the 3p12.1 region were associated with expression levels of CADM2, in both BarcUVa-Seq and GTEx colon transverse datasets (Table 2). For the SNPs in the 6p21.33 region, the long LD extension of this region provided association with 22 genes; of which activating transcription factor 6 beta (ATF6B) and epidermal growth factor like domain multiple 8 (EGFL8) genes were associated in both datasets (BarcUVa-Seq and GTEx colon transverse, Table 2).

Table 2-. Functional annotation of lead and correlated SNPs as variant enhancer loci (VEL) or expression quantitative trait loci (eQTL).

Characters in bold have the lead SNP as VEL or eQTL. eQTLs were assessed in the BarcUVA-Seq project data and GTEx colon transverse.

				N VELs for regulating elements					N eQTLs for genes
				Non-malignant tissue			Tumor tissue
Smoking habit	Locus	Lead SNP	SNPs in LD (R2>0.5)	ATAC- Seq	H3K27ac	DHS	H3K27ac	DHS	Genes (N in BarcUVa- Seq; N in GTEx colon transverse)
Status (ever vs never)	3p12.1	rs9880919	588	13	1	5	3	0	CADM2(460;43)
Cigarettes smoked per day	6p21.33	rs4151657	44	0	4	3	13	0	VARS2 (4;8); CCHCR1 (0;9); PSORS1C3 (0;4); HCG27 (0,25); SNORA38 (36;0); BAG6 (0;25); CSNK2B (36;0); LY6G5B (0;1); VARS1 (8;0); SKIV2L (2;0); C4A (1;0); CYP21A1P (0;36); TNXA (0;5); C4B (5;0); ATF6B (38;36); PPT2 (5;0); EGFL8 (2;27); RNF5 (1;1); PBX2 (3;0); HLA-DRB6 (34;0); HLA-DQA2 (12;6); HLA-DQB2 (5;0)
	8q24.23	rs7005722	0	0	0	0	0	0	-

Genetic models predicting gene expression were developed for CADM2 (3p12.1), and ATF6B and EGFL8 (6p21.33) genes (Supplementary Table S10). The lead SNPs for loci 3p12.1 (rs9880919), and 6p21.33 (rs4151657) were not selected in the genetic models, but they were strongly associated with predicted gene expression levels of corresponding nearby genes in the full case-control sample (p<1x10⁻¹⁶). In turn, a statistical interaction was observed between expression levels of CADM2 with smoking status on risk for CRC (p=0.034) (Supplementary Table S11). When stratified by quartiles of CADM2 expression levels, smoking status showed higher risk in those individuals in the first quartile (OR 1.28, 95%CI 1.20-1.36, p=3.67x10⁻¹⁴) compared with those in upper quartiles (4^th quartile, OR 1.16, 95%CI 1.09-1.23, p=3.00x10⁻⁶) (Fig. 4). In addition, an interaction was observed between expression levels of ATF6B with smoking intensity (cigarettes smoked per day) on risk for CRC (p=5x10⁻⁵) (Supplementary Table S10). When stratified by quartiles of ATF6B expression levels, we observed that the association between smoking intensity and CRC risk was stronger in those in the first quartile (OR 1.11, 95%CI 1.06-1.16, p=2.65x10⁻⁶, per an increase in 10 cigarettes smoked per day) compared with those in upper quartiles (4^th quartile, OR 0.99, 95%CI 0.95-1.03, p=0.5) (Fig. 4). Finally, the interaction between expression of EGFL8 and smoking intensity on CRC risk was not significant (p=0.052, Supplementary Table S11).

Figure 4-.

Association parameters of genetically predicted gene expression for colorectal cancer risk stratified by smoking status.

Discussion

Combining genome-wide genetic data with harmonized smoking habit data across 34 studies enabled us to conduct the largest genome-wide exploration of GxSmoking interactions for CRC risk to date. We discovered one novel CRC locus in our genome-wide GxE scan of smoking status (3p12.1), and two SNPs in our scan of smoking intensity among smokers. None of the three genetic loci we discovered have been associated with CRC or smoking previously. Additionally, two of the identified GxE interactions (3p12.1 and 6p21.33) showed differences in strength by CRC tumor marker subtypes as defined based on MSI status, CIMP, and BRAF and KRAS mutation status. In addition, for these two loci, functional annotations indicated that they are located in enhancer regions and are linked to differential gene expression, and the putative gene for each was identified by demonstrating interactions between smoking and predicted expression affecting CRC risk.

Consistent with previous studies, we observed a strong positive association between smoking habits with risk of CRC with similar magnitudes to those previously observed. As expected, results showed higher risk estimates in men compared with women, and in rectal compared with colon cancer. The different associations across colon subsites have been related to tumor molecular markers of the serrated pathway associated with cigarette smoking and with colon subsites in the same fashion (13).

An enhancer locus in the CADM2 gene was found interacting with smoking status on CRC risk. In addition, genetically predicted CADM2 gene expression was observed to decrease CRC risk among ever smokers, but not among never smokers. This is the first time that this locus and CADM2 expression have been associated with CRC risk. However, CADM2 expression has been reported to reduce risk in prostate cancer, ovarian cancer, liver cancer, kidney cancer, lymphoma, melanoma, and glioma (44-50). These functional data could indicate a tumor suppression role of CADM2 gene. Other functional roles could be related to this gene. This locus has been associated with several risk-taking traits, where smoking was included among other phenotypes; however risk-taking proneness was found mediating these associations (51,52). CADM2 risk-taking risk variants were associated with increased CADM2 expression levels in brain tissues.(53) These reported results could explain both the marginal evidence of association with smoking status in this study, and its interaction with smoking status on CRC risk as tumor suppressor gene.

For smoking intensity, we observed an interaction with rs4151657 at 6p21.33 located in the human leukocyte antigen (HLA) super-locus that encodes for many genes key for immune regulation and cellular processes. Genetic variants in this gene rich region have been linked to several different diseases including CRC. The SNP rs4151657 falls within an enhancer region in the intron of the CFB gene. The complement system plays a pivotal role in the innate immune response to pathogens. It also has an important homeostatic role in recognising damaged or altered “self” components (such as apoptotic and necrotic cells) to promote their elimination. Complement factor B amplifies the turnover of C3, the most abundant complement protein, activating the terminal pathway (54,55). Insufficient or excessive complement activation can promote infections, immune-related disease, or tissue damage. The C allele of rs4151657 SNP (which homozygous genotype cancels out the risk effect of smoking intensity on CRC) has been reported to be associated to increased risk for ulcerative colitis (56,57) and IgA nephropathy (58). We did not observe an association between SNPs in this locus and expression of CFB. When we expanded our eQTL analysis to nearby genes we observed associations with expression of ATF6B and EGFL8. In addition, predicted ATF6B gene expression was observed to increase CRC risk among light smokers, but not among heavy smokers. The protein encoded by ATF6B is a transcription factor in the unfolded protein response (UPR) pathway during endoplasmic reticulum (ER) stress (59), and genetic variants in this gene region have been associated with levels of complement C4 protein (60) as well as with inflammatory lung diseases (61,62). Downregulation of EGFL8 expression has been observed in colon tumor tissue (63). It has also been associated with a less favourable immune component in tumor tissue and with poor prognosis for patients with CRC, indicating that EGFL8 may act as a tumor suppressor (64). In summary, the observed interaction between smoking and genetic variants in the HLA region point to as strong candidate risk factor the differential expression of ATF6B gene linked to inflammation and immune function. Interestingly, smoking promotes cellular damage (65,66) and bacterial CRC-related infections (67,68), and cigarette smoking has been associated with higher CRC risk particularly in tumor samples with lower number of infiltrated T cells and macrophages (69,70). Therefore, the identified germline genetic variants related to inflammation and immune response, may modify or reduce the risk of the effects of tobacco smoking on CRC.

Finally, we observed an interaction between smoking intensity and the intergenic region at 8q24.23 upstream of LOC101927915 and downstream of FAM135B. In contrast with the other identified regions, genetic markers of this region did not appear associated with enhancing activity or gene expression of nearby genes. FAM135B has been reported to be associated with risk for esophageal squamous cell carcinoma (71). Accordingly, the functional support for this genetic locus interacting with smoking on colorectal cancer risk is limited. In our analysis of GxE interaction incorporating gene expression data from the BarcUVa sample, tested genes (CADM2, ATF6B and EGFL8) did not include the respective lead SNP discovered in the primary genome-wide GxE scan in the predictive model. However, each lead SNP was strongly associated with predicted gene expression of corresponding nearby genes in our full case-control sample. This suggests that these lead SNPs are likely not causal variants themselves, but rather that they are pointing to genes for which expression is modifying the effect of smoking on CRC risk.

This study generated results using a large dataset from three CRC consortia for gene-cigarette smoking interaction on CRC risk, including analyses stratified by tumor molecular markers, and functional annotations related to gene regulation. However, molecular marker analysis was limited on sample size, and we could not provide robust results on the pathways where the identified potential mechanisms are involved. In the findings for interaction with smoking intensity, genetic variants seem to contribute to CRC risk in the tumor molecular subtypes characteristic of the adenoma pathway. Thus, cigarette smoking would be related to CRC risk mainly in the cases derived from the serrated pathway, but interacting with genetic variants to increase CRC risk in the cases derived from the adenoma pathway. Further analyses are needed to evaluate this dual relationship between cigarette smoking and CRC risk.

Finally, the map of common genetic variants interacting with smoking habits and associated with CRC risk is not still saturated. Novel functional genetic regions can be predicted integrating strong functional data, for instance genomic single cell data, and thereby reducing the multiple comparison. In addition, further efforts are needed to enlarge the sample size of the analysis including wide-world populations, since most common germline genetic risk loci are have been replicated across racial and ethnic groups (72,73), reducing inequalities in research, and to provide independent datasets to estimate the heritability contribution.

In conclusion, we identified novel genetic loci that modulates the association between cigarette smoking status and intensity and CRC risk. The potential mechanism behind these associations could be linked, in part, to tumor suppression, inflammation and immune response to the effects of tobacco smoking. These findings can guide potential prevention treatments in addition to quitting smoking. Additional functional studies are needed to verify the role of the identified loci interacting with smoking for CRC risk.