Genetic Colorectal Cancer and Adenoma Risk Variants Are Associated with Increasing Cumulative Adenoma Counts

This article is featured in Highlights of This Issue, p. 2105

Studies suggest that individuals with high cumulative adenoma counts may be at significantly increased risk for poor clinical outcomes, including colorectal cancer, need for colectomy, and extra-intestinal malignancies (1, 2). As such, colorectal cancer screening guidelines now suggest that individuals who are found to have ≥10 cumulative lifetime adenomas be considered for genetic counseling and testing, as they are thought to have an increased risk for actionable hereditary colorectal cancer syndromes (3–5).

However, current genetic testing recommendations are based on limited data from referral populations. These studies found that highly-penetrant gene mutations associated with hereditary colonic adenomatous polyposis syndromes were actually present in only a small fraction of individuals who would be typically encountered during routine colorectal cancer screening and surveillance (i.e., those with 10–99 cumulative adenomas; refs. 6, 7). Therefore, the genetic basis for most individuals with high cumulative adenoma counts remains unexplained, and referring everyone with ≥10 cumulative adenomas for genetic testing will likely be costly and uninformative.

Although ongoing work continues to identify additional genes associated with this high-risk colonic adenomatous polyposis phenotype (8–11), these mutations remain rare in the general population. Rather, it is likely that many genetic changes responsible for this phenotype are low-penetrant single-nucleotide polymorphisms (SNP) that also have an association with colorectal cancer development (12, 13). Large meta-analyses of numerous genome-wide association studies have identified over 100 SNPs associated with the development of colorectal cancer (i.e., “colorectal cancer–risk SNPs”; refs. 13–17), but few studies have evaluated whether these colorectal cancer–risk SNPs are also associated with the development of multiple precancerous adenomas (18–23). Furthermore, these prior studies investigating SNPs associated with multiple adenomas (“adenoma-risk SNPs”) were performed in heterogeneous populations mostly with prior adenomas or those referred for genetic testing, and so there is a high likelihood of selection bias which limits generalizability. Additional investigations are needed into the low-penetrant SNPs that may be associated with the colonic adenomatous polyposis phenotype, either singly or in combination, to help further inform colorectal cancer clinical and genetic screening guidelines.

The goal of our analysis was to evaluate associations between previously identified colorectal cancer–risk SNPs and cumulative adenoma counts in a colorectal cancer screening population. We also investigated whether a polygenic risk score (PRS), comprised of multiple prespecified adenoma-risk SNPs, would be associated with higher cumulative adenoma counts.

In 1993, the U.S. Department of Veterans Affairs (VA) Cooperative Studies Program (CSP) initiated CSP #380, a prospective screening study of asymptomatic individuals in which the primary aim was to characterize the distribution of adenomas on baseline colonoscopy (24). The design and outcomes of CSP #380 have been described previously (25). In brief, this cohort is comprised of asymptomatic veterans age 50 to 75 from 13 VA Medical Centers who underwent a baseline screening colonoscopy from 1994 to 1997. Although participants were oversampled for a family history of colorectal cancer in first-degree relatives in the initial protocol, there was no association found between a first-degree family history of colorectal cancer and the development of ≥10 cumulative adenomas in the overall CSP #380 cohort. Following the completion of a randomized component comparing different surveillance intervals within the first 5 years based on index findings, a longitudinal study now with at least 10 years of follow-up was conducted to identify risk factors, including polyp number and histology, associated with the development of advanced neoplasia. Colonoscopy and pathology records were ascertained through the VA electronic medical record, which included procedures and pathology performed outside of the VA. The protocol was approved by Institutional Review Boards at each participating center.

In this analysis, our outcome of cumulative adenoma counts was characterized in two clinically relevant ways. The first outcome is the number of cumulative adenomas per individual, including advanced adenomas and invasive adenocarcinoma, summed across all available colonoscopies and associated pathology reports over at least 10 years of follow-up. The second outcome is the binary finding of ≥10 cumulative adenomas compared to those with zero cumulative adenomas, to allow for as much discrimination as possible between cases–controls. Importantly, no instances of sessile serrated lesions (SSL) were identified, as these exams were performed prior to an awareness of SSLs. However, there were no serrated polyps with adenomatous change or dysplasia in this cohort, therefore the finding of “no adenomas” includes both hyperplastic polyps or no polyps.

As part of the initial CSP #380 study, a biorepository was developed that included selected individuals with large (≥10 mm) polyps or colorectal cancer identified at baseline colonoscopy as well as age and gender-matched individuals with small (<10 mm) polyps or without neoplasia. The biorepository is comprised of 815 out of 3,121 (26.1%) total CSP #380 participants. A total of 612 individuals had sufficient DNA to allow genetic analysis and passed genome-wide association study (GWAS) quality control measures, including 176 (28.8%) individuals with large polyps or colorectal cancer and 436 (71.2%) with small polyps or no polyps. A CONSORT diagram is included as Fig. 1. A comparison of individuals with analyzable DNA from the biorepository and those not represented is included as Supplementary Table S1. This biorepository includes 612 individuals without known genetic syndromes.

Genotyping was completed on participants' blood using the Illumina Infinium Omni2.5–8 v1.3 GWAS Beadchip; data analysis showed excellent genotyping quality. For this analysis, we evaluated a set of 43 prespecified colorectal cancer–risk SNPs (Supplementary Table S2), which were obtained from prior literature in similar populations (13, 16, 17). Two SNPs with a minor allele frequency in CSP #380 of less than 0.05 were excluded (rs12603526 and rs17879961), resulting in a total of 41 colorectal cancer–risk SNPs included for evaluation. These 41 colorectal cancer–risk SNPs include a subset of eight SNPs previously shown to be associated with multiple adenomas, so these eight specific SNPs are also termed “adenoma-risk SNPs” (Table 1; ref. 18). Otherwise, age at last colonoscopy was recorded since participants had different amounts of follow-up and at different ages. A principal component (PC) analysis on a set of independent SNPs was performed to develop genotype-based PCs to allow for adjustment based on genetic ancestry.

We constructed a PRS similar to prior studies (16, 17, 20, 21) using the eight prespecified adenoma-risk SNPs for this analysis (Table 1). A PRS was assigned for each individual based on the total number of risk alleles present per adenoma-risk SNP (regardless if the allele is major or minor), with a range per SNP between 0 and 2 points. A specific adenoma-risk SNP received a score of 0 if the risk allele was not present, 1 if heterozygous, and 2 if homozygous. The score was summed across all adenoma-risk alleles present in each individual for the unweighted PRS. The weighted PRS was calculated by adjusting each SNP's effect size by the published risk allele log odds ratio (OR) for multiple adenomas (18), and summing together all effect size adjusted adenoma-risk alleles present in each individual.

Descriptive statistics of the dataset are reported using proportions and means with standard deviations (SD), and included in Table 2 and Supplementary Table S1. We fit a Poisson regression model to the cumulative adenoma count. Descriptive histograms of our data (Fig. 2) helped determine these corrections, which included use of a zero-inflated, negative binomial model. Individual colorectal cancer–risk SNP associations with increasing cumulative adenoma counts were modeled as rate ratios [RR with 95% confidence intervals (CI)] by corrected Poisson regression, adjusting for age at last colonoscopy (since participants have varying degrees of follow-up), gender, and PCs. The same corrected Poisson regression model was used to analyze the association between the PRS and higher cumulative adenoma counts, while controlling for the aforementioned covariates. Logistic regression was used to generate ORs for associations between the colorectal cancer–risk SNPs and the binary outcome of ≥10 cumulative adenomas or zero cumulative adenomas, again controlling for age at last colonoscopy, gender, and PCs. Wilcoxon rank sum tests were used to compare the mean adenoma PRS (both unweighted and weighted by log OR for effect size) between individuals with ≥10 cumulative adenomas versus those with no adenomas. Finally, stratified analyses were performed by sex (male vs. female), race (European vs. non-European), and age at last colonoscopy (age 50–64 vs. age 65+). All statistical analyses were performed using R (version 3.6.1). Given that we tested only prespecified colorectal cancer or adenoma-risk SNPs identified in independent data sets, all tests of significance were performed at an α level of 0.05.

The overall CSP #380 cohort includes 3,121 participants, of whom 612 (19.6%) had genotyped blood samples in the biorepository, and thus are included in this analysis. Table 2 describes the baseline demographic and clinical characteristics of all included individuals, including the distribution of cumulative adenoma counts. The mean age was 64.1 (SD 6.7) years, 97.1% were male, 81.7% were white, and 22.2% had a family history of colorectal cancer in first-degree relatives. At baseline, 174 (28.4%) were found to have advanced neoplasia, including 20 (3.2%) with colorectal cancer. Over 10 years, most individuals were found to have zero cumulative adenomas (40.2%), whereas 31.4% had one or two cumulative adenomas, 24.7% had three to nine cumulative adenomas, and 3.8% had ≥10 cumulative adenomas (Fig. 2). The maximum individual cumulative adenoma count was 22.

Supplementary Table S1 compares baseline characteristics between participants included and not included in the CSP #380 biorepository. When compared with the CSP #380 participants not included in the biorepository, the individuals represented in the biorepository are slightly older (64.1 vs. 62.6, P < 0.0001), more likely to have a first-degree family history of colorectal cancer (22.2% vs. 11.9%, P < 0.0001), and had a higher prevalence of advanced neoplasia (including colorectal cancer) on baseline screening colonoscopy (28.1% vs. 6.3%, P < 0.0001). All of these factors are expected and related to the initial design of the biorepository.

Supplementary Table S3 includes results for all 41 prespecified colorectal cancer–risk SNPs and their association with cumulative adenoma counts. Four of these colorectal cancer–risk SNPs were statistically significantly associated with increasing cumulative adenoma counts: rs12241008 (risk allele RR = 1.31; 95% CI, 1.00–1.71), rs2423279 (RR = 1.29; 95% CI, 1.07–1.55), rs3184504 (RR = 1.24; 95% CI, 1.04–1.46), and rs961253 (RR = 1.23; 95% CI, 1.04–1.47). Thus, for these four colorectal cancer–risk SNPs, mean cumulative adenoma counts were 1.2 to 1.3× higher for each additional risk allele present. These four SNPs are associated with the genes VTI1A (13, 17), HAO1/BMP2 (4, 19), SH2B3 (15–18), and BMP2 (6, 8, 9, 16, 17), respectively. Furthermore, these risk alleles are the same as reported in prior studies for colorectal cancer, with published risk allele RRs for colorectal cancer of 1.12, 1.07, 1.09, 1.12, respectively (13). Of the subset of eight colorectal cancer–risk SNPs previously found to be associated with multiple adenomas (i.e., “adenoma-risk SNPs”), only one (rs961253) was significant in our dataset, with the same literature risk allele for multiple adenomas (published RR of 1.18, Table 1; ref. 18). When aggregating all eight adenoma-risk SNPs into the PRS, the median unweighted PRS was 8 (7–9 IQR) and the median weighted PRS was 1.16 (0.97–1.38 IQR) for all included participants. Only the weighted PRS was associated with higher cumulative adenoma counts (unweighted RR = 1.03, P = 0.39; weighted RR = 1.57, P = 0.03) in our dataset. In other words, for each additional weighted PRS point, mean cumulative adenoma counts were increased by approximately 1.6x per individual. Figure 3 shows the average unweighted and weighted PRS by cumulative adenoma count. The results from sex, race, and age stratified analyses are available in Supplementary Table S4. In the age stratified analysis, both the unweighted and weighted PRS were significantly associated with higher cumulative counts in those age 50 to 64 at the time of last colonoscopy (unweighted RR = 1.17, P = 0.05; weighted RR = 2.48, P = 0.04), but not those age 65+ (unweighted RR = 1.04, P = 0.25; weighted RR = 21.39, P = 0.14).

Three of the prespecified colorectal cancer–risk SNPs were found to be associated with the finding of ≥10 cumulative adenomas, when compared with those with no cumulative adenomas: rs3184504 (risk allele OR = 2.09; 95% CI, 1.05–4.17), rs961253 (OR = 2.30; 95% CI, 1.17–4.49), and rs3217901 (OR = 1.94; 95% CI, 1.02–3.67, gene CCND2; ref. 16). Thus, for each additional risk allele of any of these three SNPs, the odds of developing ≥10 cumulative adenomas were about two times higher. These are the same risk alleles found in the literature, and compare to published risk allele RRs for colorectal cancer of 1.09, 1.12, and 1.10, respectively (13, 16). The mean adenoma PRS was slightly higher in those with ≥10 cumulative adenomas compared with those without, however these differences were not statistically significant: unweighted PRS of 8.43 versus 7.82, respectively (P = 0.13), and weighted PRS of 1.24 versus 1.13, respectively (P = 0.10). Results were similar in the sex, race, and age stratified analysis (Supplementary Table S5).

Using the DNA from blood in selected participants from the CSP #380 screening cohort, we identified four known colorectal cancer–risk SNPs to also be associated with increasing cumulative adenoma counts and three known colorectal cancer–risk SNPs to be associated with the finding of ≥10 cumulative adenomas. Of the subset of colorectal cancer–risk SNPs previously shown to be associated with adenoma-risk (i.e., “adenoma-risk SNPs”), one was also statistically significant in our dataset (rs961253). When combining these eight known adenoma-risk SNPs into an aggregate PRS, we found that an increased burden of these adenoma-risk SNPs per individual was associated with higher cumulative adenoma counts when taking into account the adenoma-risk SNPs' published effect sizes (RR = 1.57; P = 0.03). This work suggests a role for developing precancerous “adenoma PRSs” (as opposed to colorectal cancer PRSs) using larger genomic datasets to provide enhanced risk stratification for the prevention of colorectal cancer.

Prior studies have also found similar germline genetic variants between individuals with colorectal cancer and those with multiple adenomas. A study by Carvajal-Carmona and colleagues found that eight of 18 previously identified colorectal cancer–risk SNPs were overrepresented in colorectal cancer–free individuals with adenomas when compared with controls (18). Another study by Hes and colleagues showed that two of 16 colorectal cancer–risk SNPs were associated with the finding of >10 adenomas, with ORs of 1.6 and 1.5 (19). These findings, along with three additional studies, are summarized in Supplementary Table S6. Our findings extend previous research, as we additionally found four known colorectal cancer–risk SNPs to be also associated with increasing cumulative adenoma counts, as well as three colorectal cancer–risk SNPs associated with ≥10 cumulative adenomas. Furthermore, two of our significant colorectal cancer–risk SNPs (rs961253, rs2423279) have been previously shown to be associated with adenoma development (18, 21, 26). We were unable to support growing evidence implicating an association between the development of multiple adenomas and the SNPs rs3802842, rs6983267, rs10505477, rs10795668, and rs4779584 (18–20, 22, 23, 27). However, the overall consistency of these results suggest common underlying risk genes between colorectal cancer and adenomatous polyposis.

Although many genetic risk scores have been tested to predict colorectal cancer risk in various study populations (28), few tools have been developed to predict precancerous adenoma risk. A recent study by Weigl and colleagues did not find an association between a PRS of 48 colorectal cancer–risk SNPs and the finding of nonadvanced adenomas (16). However, a study by Abuli and colleagues found that an increased burden of 14 colorectal cancer–risk SNPs was associated with a higher risk of advanced adenomas and/or ≥3 adenomas (21). However, this PRS incorporated only the SNPs found to be individually associated with this outcome in their population, thus strongly biasing the results. Finally, three other studies have suggested that a higher weighted PRS comprised of different subsets of published colorectal cancer–risk SNPs were associated with varying degrees of increased adenoma counts (20, 23, 27) in differing populations. Our study is unique and builds on prior literature, in that it utilized a PRS created solely from unbiased precancerous adenoma-risk SNPs and found that an increasing burden of adenoma-risk SNPs present per individual are associated with higher cumulative adenoma counts. Thus, a precancerous PRS comprised of germline variants may have the potential to earlier identify individuals at risk for colonic adenomatous polyposis, particularly those with a young-onset phenotype. Such information could inform genetic testing guidelines (7) or recommendations regarding age to start colorectal cancer screening (17).

The colorectal cancer–risk SNPs in this analysis found to be associated with increasing cumulative adenoma counts have been mapped to nearby genes of VTI1A (rs12241008), BMP2 (rs2423279, rs961253), SH2B3 (rs3184504), and CCND2 (rs3217901). The functional relevance of these genes on the outcome of adenoma or colorectal cancer formation have been theorized to disrupt proper tissue architecture through varying effects on cell adhesion, differentiation, and migration via dysregulation of the Wnt (VTI1A), TGF-β/BMP (BMP2), or MAPK (SH2B3, CCND2) signaling pathways (13). Only SH2B3 is identified as a missense mutation, which increases the likelihood that this is a candidate driver gene in the adenoma–carcinoma sequence. However, although several of these SNPs are located in intergenic regions (rs12241008, rs961253), it is possible that their impact on splicing may affect gene function through disruption of key regulatory elements or other potential mechanisms (29). Additional genetic and functional studies will be needed to validate the impact of these genetic variants on colonic adenomatous polyposis and/or colorectal cancer risk. Clarifying potential novel tumorigenesis or protein pathways may identify new therapeutic targets, which may be ideal for developing vaccines other chemopreventative agents to prevent colorectal cancer (30, 31).

Our study has important strengths. We investigated a novel precancerous adenoma PRS to assess risk for specific adenoma counts in individuals from a well characterized colonoscopy cohort with associated genetic data and 10 years of clinical follow-up. These findings are preliminary analyses on a high-risk clinical phenotype for whom there are limited data to inform colorectal cancer surveillance and genetic testing guidelines. Germline variants that predispose to high adenoma counts may predispose to early onset colorectal cancer, increase risk for missed lesions, or promote progression to colorectal cancer between colonoscopies (32–34). Thus, it will likely be important to identify these variants early and develop risk stratification protocols for colorectal cancer prevention. Our work should help conceive similar analyses in larger genomic datasets which seek to risk stratify screening participants using a precancer, adenoma genetic risk score to prevent colorectal cancer development (35). Importantly, we feel our tool represents an unbiased risk score, which should allow comparison to other risk scores.

This study has several limitations. The CSP #380 biorepository is comprised predominantly of Caucasian male veterans, and so possible differences in sex- or race-specific outcomes may be missed. The generalizability of our results may also be limited due to selection of CSP #380 biorepository participants based on the presence of advanced neoplasia or not at baseline colonoscopy. As such, only 612 of the initial 3,121 CSP #380 participants are included in this analysis. The baseline characteristics of participants selected for the biorepository differed from the overall CSP #380 cohort in several important ways that may have introduced potential bias due to incomplete genomic data (Supplementary Table S1). The individuals in the biorepository are slightly older, are more likely to have a first-degree family history of colorectal cancer, and have a higher incidence of colorectal cancer. In addition, follow-up over 10 years was variable, and so adenomas may have been present but undetected in some participants due to varying intensity of follow-up colonoscopies. Furthermore, significant false-positive associations are a possibility, as we did not adjust for numbers of comparisons. To address this concern, we used prespecified SNPs chosen based on prior knowledge of association patterns with colorectal cancer and the multiple adenoma phenotype to justify the conservative significance threshold. Finally, although the PRS in our analysis only included eight adenoma-risk SNPs based on a single prior study with a similar population, other studies have shown additional SNPs to be associated with multiple adenomas in different populations (Supplementary Table S6). Overall, while the consistency of our results with the literature seem to support our findings, our analyses should be considered exploratory given the small sample size and few cases of individuals with ≥10 cumulative adenomas. Future work in larger genomic datasets with diverse populations will be important for developing a standardized precancerous adenoma PRS based on an optimal number of specific adenoma-risk SNPs with reliable performance in independent samples across various populations (28).

In summary, we found specific germline genetic variants known to be associated with colorectal cancer development are also associated with increasing cumulative lifetime adenoma counts. In addition, an increasing burden of adenoma-risk SNPs, as measured by a weighted PRS, was also associated with higher cumulative adenoma counts. This work will serve as the basis for planned investigations to test precancerous risk stratification tools that could more accurately identify individuals who would benefit from genetic testing, earlier colorectal cancer screening, or enhanced colorectal cancer surveillance. Future work investigating the genetic underpinnings of the colonic adenomatous polyposis phenotype may also help prevent colorectal cancer by recognizing and addressing important colorectal cancer gene–environment interactions, clarifying therapeutic targets in cancer progression pathways, and developing novel colorectal cancer screening modalities using precancer genomic biomarkers (36–38).

D. Lieberman reports grants from the U.S. Department of Veterans Affairs during the conduct of the study. No potential conflicts of interest were disclosed by the other authors.

The views expressed in this article are those of the authors and do not necessarily represent the views of the U.S. Department of Veterans Affairs or the government of the United States.

B.A. Sullivan: Conceptualization, methodology, writing–original draft, writing–review and editing. X. Qin: Data curation, formal analysis, methodology. T.S. Redding IV: Data curation, formal analysis. Z.F. Gellad: Writing–review and editing. A. Stone: Data curation. D. Weiss: Writing–review and editing. A.N. Madison: Project administration, writing–review and editing. K.J. Sims: Data curation, writing–review and editing. C.D. Williams: Writing–review and editing. D. Lieberman: Supervision, writing–review and editing. E.R. Hauser: Supervision, methodology, writing–review and editing. D. Provenzale: Supervision, writing–review and editing.

This work was supported by the U.S. Department of Veterans Affairs Cooperative Studies Program.

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.