Genetic Risk Score is Associated With Prevalence of Advanced Neoplasms in a Colorectal Cancer Screening Population

Abstract

Background & Aims

The presence of specific single nucleotide polymorphisms (SNPs) can be used to calculate an individual’s risk for colorectal cancer (CRC), called a genetic risk score (GRS). We investigated whether GRS can identify individuals with clinically relevant neoplasms in a screening colonoscopy population.

Methods

We derived a GRS based on 48 SNPs associated with CRC, identified in a comprehensive literature search. We obtained genetic data from 1043 participants (50–79 years old) in a screening colonoscopy study in Germany, recruited from 2005 through 2013 (294 with advanced neoplasms, 249 with non-advanced adenomas, and 500 without neoplasms). Each participant was assigned a GRS by aggregating their risk alleles (0, 1, or 2). Risk of advanced neoplasms and non-advanced adenoma according to GRS was calculated by multiple logistic regression. Risk advancement periods were calculated. We replicated our findings using data from a subset of the Tennessee Colorectal Polyp Study.

Results

An increased GRS was associated with higher prevalence of advanced neoplasms, but not non-advanced adenomas. Participants in the middle and upper tertile of GRSs had a 2.2-fold and 2.7-fold increase in risk, respectively, of advanced neoplasms compared to those in the lower tertile. Adjusted odds ratios (ORs) were 1.09 (95% CI, 0.76–1.57) for non-advanced adenoma in the middle tertile and 1.05 (95% CI, 0.70–1.55) for non-advanced adenoma in the upper tertile. The ORs were largest for proximal advanced neoplasms for participants in the middle tertile (OR, 3.55; 95% CI 1.85–6.82) and the upper tertile (OR, 3.61; 95% CI 1.84–7.10). The risk advancement period for medium vs low GRS was 13.4 years (95% CI 4.8–22.0) and for high vs low GRS was 17.5 years (95% CI, 7.8–27.3).

Conclusions

In a genetic analysis of participants in a CRC screening study in Germany, an increased GRS (based on CRC-associated SNPs) was associated with increased prevalence of advanced neoplasms. These findings might be used in defining risk-adapted screening ages.

Graphical abstract

Background

Colorectal cancer (CRC) is the third most common cancer and the fourth most common cancer cause of death globally. Estimated numbers of incident cases and deaths in 2012 amounted to almost 1.4 million and 700,000, respectively¹. Most CRCs slowly develop over many years through the adenoma-carcinoma sequence², which provides unique opportunities for prevention by early detection and endoscopic removal of adenomas.

Randomized controlled trials (RCTs) have consistently demonstrated high effectiveness in prevention of CRC by flexible sigmoidoscopy (FS) through detection and removal of adenomas^{3, 4}. Although results of RCTs on effectiveness of colonoscopy in preventing CRC are not available yet, evidence from epidemiological studies suggests even stronger protection⁵. However, adherence rates for both procedures are low and only a minority of participants, i.e. those in whom clinically relevant neoplasms are detected and removed, benefit from the procedure. Risk stratification that would enable tailoring offers of screening endoscopy according to risk of such neoplasms would therefore be highly welcome.

In recent years, genome-wide association studies (GWAS) have identified an increasing number of single nucleotide polymorphisms (SNPs) that are associated with increased CRC risk^6–18. The combination of a large number of such SNPs in genetic risk scores (GRS) has been demonstrated to enable relevant risk stratification^19–21. However, previous evidence is almost exclusively restricted to risk of CRC, whereas evidence on risk stratification for relevant CRC precursors sought for and removed at screening endoscopy is sparse.

In this study, we aimed to evaluate the potential of a GRS to predict presence of CRC and its major precursors, i.e. advanced adenomas, in a large population of participants of screening colonoscopy.

Material and Methods

Study population

Data for the analyses were drawn from the BLITZ study (Begleitende Evaluierung innovativer Testverfahren zur Darmkrebsfrüherkennung) which has been described in detail before^{22, 23}. In brief, unselected participants of the German screening colonoscopy program are recruited by gastroenterology practices in southern Germany in this multicenter study. Participants of every age undergoing screening colonoscopy, which is reimbursed in the German health-care system since October 2002 to individuals aged ≥55 years, are informed about the study at a preparatory visit, usually one week prior to colonoscopy. Patients are invited to provide stool and blood samples prior to colonoscopy and to complete a standardized questionnaire. The ethics committees of the Medical Faculty Heidelberg of the University of Heidelberg and of the state physicians’ chambers of Baden-Württemberg, Rhineland-Palatinate and Hessia approved the study. Written informed consent is obtained from every participant. Since the end of 2005, more than 8,000 participants were recruited for the BLITZ study, which is still ongoing. This current analysis is based on a subset of patients aged 50–79 years recruited between 2005 and 2013 for whom both questionnaire and genomic data were available.

Data collection

Blood samples are provided by the patients at the preparatory visit. Information about previous colonoscopies, family history of CRC and other known CRC risk factors is obtained by a self-administered questionnaire. Specifically, participants are asked whether their mother, father or sibling was diagnosed with CRC and if so, at which age. Colonoscopy and histology reports are collected and information from these reports is extracted independently in a standardized way by two trained investigators, who are blinded with respect to questionnaire and genotype data and who resolve discrepancies by consensus after further review and discussion. Based on colonoscopy reports, participants are categorized with respect to the most advanced lesion: CRC, advanced adenoma (AA), non-advanced adenoma (NAA), hyperplastic polyp, or undefined polyp. Advanced adenomas are defined as adenomas ≥1 cm or adenomas with cellular or structural atypia.

Genotyping

Genotyping within the BLITZ study was conducted for a subset (n=1098) of participants who were recruited between 2005 and 2013. In a sex-matched case-control design, participants were randomly selected within subgroups defined by the most advanced finding at screening colonoscopy, with higher sampling fractions for the much smaller subgroups of participants with more advanced findings in order to increase the power of between-group comparisons (Figure 1).

Figure 1. Flow diagram of inclusion of study participants.

Abbreviations: AA, advanced adenoma; AN, advanced neoplasm; CRC, colorectal cancer; GRS, genetic risk score; HP, hyperplastic polyp; NAA, non-advanced adenoma; QC, quality check

Extracted DNA from blood cells was genotyped using the Illumina OncoArray-500k V1.0 BeadChip (Illumina, San Diego, CA, USA). General genotyping quality control assessment was conducted as previously described²⁴. A total of 1098 samples fulfilled all criteria and qualified for imputation. Genotypes for common variants across the genome were imputed using data from UK10K-1000 Genomes Project (phase 3, Oct. 2014) with IMPUTE2 v2.3.2 after pre-phasing with SHAPEIT software v2.12. We set thresholds for imputation quality to retain both potential common and rare variants for validation. Sufficient quality of imputation was assumed if imputed SNPs reached an information metric I of at least >0.30. All genomic locations are given in NCBI Build 37/UCSC hg19 coordinates. All SNPs having a MAF<1% were excluded. After imputation, the SNP set consisted of 92580817 genotyped and imputed SNPs. PLINK v1.90 was then used to extract SNPs for the required regions of interest.

Identification and selection of common risk variants

Previously published and validated risk variants of CRC for persons of European descent which were identified through literature research were considered for inclusion in the GRS. A comprehensive overview of the previously discovered risk variants considered and included for these analyses (n=55) is provided in Supplementary Table 1. One SNP (rs6066825) was excluded from further analyses as its imputation resulted in missing values in more than 10% of our sample. Of SNPs in linkage disequilibrium (D′≥0.95 and r²>0.5) only those were included in the final GRS which were most significantly associated with the risk for advanced neoplasms in our study sample. Thus, six SNPs were excluded from further analyses (rs10505477, rs1862748, rs7014346, rs7259371, rs7837328, rs961253). We conducted sensitivity analyses with a more conservative cut-off for D′ (0.5), which resulted in excluding nine additional SNPs from the GRS.

Statistical analyses

First, main characteristics of the BLITZ study population were described (Table 1). GRS for all eligible study participants were calculated by aggregating their risk alleles (0, 1 or 2) for SNPs as reported in the discovery studies. For sensitivity analyses, we also generated a weighted GRS with weights equal to the log odds ratios (OR) of the respective SNPs reported in the discovery studies. For easier interpretation we report results from the unweighted GRS in the main text. Tertiles of the GRS were calculated according to the distribution of risk alleles in “healthy” participants without any finding at screening colonoscopy. Based on those tertiles, all participants were then categorized in three risk groups (low, medium, and high).

Table 1.

Selected characteristics of the study population, stratified by most advanced finding at screening colonoscopy

Characteristic	Group	No neoplasm n=500 (47.9)	NAA n=249 (23.9)	p-valuea	AN n=294 (28.2)	p-valuea
Sex	Female	190 (38.0)	106 (42.6)	0.2281	103 (35.0)	0.4030
	Male	310 (62.0)	143 (57.4)		191 (65.0)
Age	50–59	222 (44.4)	85 (34.1)	0.0072	92 (31.3)	<0.0001
	60–69	206 (41.2)	110 (44.2)		132 (44.9)
	70–79	72 (14.4)	54 (21.7)		70 (23.8)
Genetic risk score (tertile)b	Low	173 (34.6)	82 (32.9)	0.8734	55 (18.7)	<0.0001
	Medium	188 (37.6)	98 (39.4)		124 (42.2)
	High	139 (27.8)	69 (27.7)		115 (39.1)
Family history of colorectal cancer in a 1st-degree relative		61 (12.2)	33 (13.3)	0.6149	44 (15.0)	0.2521
Smoking	Never	240 (48.0)	137 (55.0)	0.0892	127 (43.2)	0.0973
	Former	204 (40.8)	82 (32.9)		123 (41.8)
	Current	49 (9.8)	29 (11.7)		43 (14.6)
Body mass index	≤18.5 kg/m2	0 (0)	0 (0)	0.0052	2 (0.7)	0.0427
	≤25 kg/m2	177 (35.4)	60 (24.1)		84 (28.6)
	≤30 kg/m2	210 (42.0)	128 (51.4)		127 (43.2)
	>30 kg/m2	101 (20.2)	57 (22.9)		74 (25.2)
High alcohol intakec		160 (32.0)	81 (32.5)	0.8837	115 (39.1)	0.0419
High red meat intaked		58 (11.6)	23 (9.2)	0.3266	23 (7.8)	0.0895
Low physical activitye		44 (8.8)	23 (9.2)	0.8435	25 (8.5)	0.8861
Hormone-replacement therapyf		99 (52.1)	52 (49.1)	0.5526	43 (41.8)	0.0931
Education	≤ 9 yrs	282 (56.4)	151 (60.6)	0.5268	187 (64.7)	0.0867
	10–11 yrs	111 (22.2)	51 (20.5)		50 (17.3)
	≥ 12 yrs	103 (20.6)	45 (18.4)		52 (18.0)
Regular use of nonsteroidal anti-inflammatory drugs		89 (17.8)	46 (18.5)	0.8212	51 (17.4)	0.8715
Personal history of colonoscopy		102 (20.4)	47 (18.9)	0.6406	49 (16.7)	0.1955

Note: values are expressed as n (%)

Abbreviations: AN, advanced neoplasm; NAA, non-advanced adenoma

^aCompared to group without neoplasm

^bNumber of risk alleles in different categories: low (21–32), medium (33–37), high (38–47)

^cHigh alcohol intake defined as drinking alcoholic beverages on average on (1) more than five days per week or (2) more than seven standard alcohol glasses per week for women or more than 14 standard alcohol glasses per week for men

^dHigh red meat intake defined as eating either processed meat or red meat more than once per day or eating both processed meat and red meat at least once per day

^eLow physical activity defined as being physically active for less than 30 minutes per day

Missing values for No finding/NAA/AN: family history 9/9/7, smoking 7/1/1, body mass index 12/4/7, hormone replacement therapy 3/-/2, education 4/2/5

^fProportions refer to women only

Chi²-tests were conducted to test for associations of the GRS with sex, age and other covariates among controls (see Table 1 for description of covariates). Using multiple logistic regression models, ORs with 95% confidence intervals (CI) and p-values for trend were calculated in order to quantify associations between GRS and risk for advanced neoplasms (AN, defined as either CRC or AA) and NAA. All models were calculated with two sets of adjustment: model 1 was adjusted for sex and age, and model 2 was additionally adjusted for factors that were found to be statistically associated with GRS categories in neoplasm-free study participants in bivariate analyses (Chi²-tests, p<0.2). C-statistics, continuous net reclassification improvement (NRI) and integrated discrimination improvement (IDI) were obtained. Dose-response analyses on the risk of AN according to the number of risk alleles were conducted for model 2 using cubic spline functions with the median number of risk alleles in neoplasm-free participants serving as the reference value and knots at the 5^th, 50^th, and 95^th percentiles of the GRS. Furthermore, we assessed the joint association of the age of the participants and GRS level (low, medium, high risk) from a multiple logistic regression model with the reference value of a 60-year old participant with a medium GRS. Finally, from the same regression model, we estimated risk advancement periods (RAPs), the difference in age at which participants in the different GRS categories reach the same level of risk. RAPs were estimated by the ratios of the regression coefficients for the respective GRS categories and the regression coefficient for an increase in age by one year, as described in detail by Brenner et al.²⁵.

SNP selection was conducted with Plink^{26, 27}. All statistical analyses were conducted two-sided on a 0.05 significance level using the SAS software, version 9.4 (SAS institute, Cary, NC).

Replication and meta-analyses

The main analyses were replicated in a subset of participants of the Tennessee Colorectal Polyp Study (TCPS), which has been described in more detail elsewhere²⁸. Genotyping in this subset was conducted with the Affymetrix Genome-Wide Human SNP Array 5.0, and imputation of data was performed in a similar manner as for the BLITZ participants. Only 35 SNPs of the 48 SNPs used in the main analyses passed the imputation quality control thresholds and were included in the consecutive analyses. Interaction tests were employed to investigate potential differences between studies, and results of both studies were summarized by meta-analyses.

Results

Of 6438 recruited study participants aged 50–79 years until 2013, a total of 1098 were selected for genotyping (all of CRC cases, and random samples using different sampling fractions of participants with other findings at colonoscopy), of whom 55 had to be excluded due to a lack of sufficient DNA or fail in quality control of genotyped or imputed data (Figure 1).

Main characteristics of the study population after quality control are shown in Table 1. The sample included 294 participants with AN, and 249 with NAA as the most advanced finding, 500 participants had no neoplasm at screening colonoscopy. About 60% of all participants in each group were male. Median age was 60, 63 and 65 in participants without neoplasm, with NAA and with AN, respectively. The GRS ranged from 21 to 49, with cut-offs for the tertiles in participants without finding being 32 and 37. The distribution of the GRS was very similar among participants with NAA and participants without neoplasms (Supplementary Figure 1). By contrast, much higher and lower proportions of participants with AN had a GRS in the highest and lowest tertile, respectively, compared to persons without neoplasm (p<0.0001).

The distribution of potential confounders among study participants without finding at screening colonoscopy according to GRS categories is listed in Table 2. Neither sex nor age was unequally distributed between the different GRS categories. Only for a personal history of colonoscopy and low physical activity some modest association with the distribution of GRS was seen (p<0.2), and these covariates were therefore selected for inclusion in multivariable adjustments.

Table 2.

Distribution of covariates according to GRS among study participants without neoplasm at screening colonoscopy

		Tertile of GRS
Characteristic	Group	Low 173 (34.6)	Medium 188 (37.6)	High 139 (27.8)	p-valuea
Sex	Female	64 (37.0)	69 (36.7)	57 (41.0)	0.6900
	Male	109 (63.0)	119 (63.3)	82 (59.0)
Age	50–59	75 (43.4)	89 (47.3)	58 (41.7)	0.2335
	60–69	73 (42.2)	67 (35.6)	66 (47.5)
	70–79	25 (14.5)	32 (17.0)	15 (10.8)
Family history in a 1st-degree relative		20 (11.6)	25 (13.3)	16 (11.5)	0.8497
Smoking	Never	81 (46.8)	92 (48.9)	67 (48.2)	0.8728
	Former	73 (42.2)	73 (38.8)	58 (43.7)
	Current	17 (9.8)	21 (11.2)	11 (7.9)
BMI	≤18.5 kg/m2	0 (0)	0 (0)	0 (0)	0.5633
	≤25 kg/m2	55 (31.8)	69 (36.7)	53 (38.1)
	≤30 kg/m2	74 (42.8)	75 (39.9)	61 (43.9)
	>30 kg/m2	41 (23.7)	36 (19.2)	24 (17.3)
High alcohol intakeb		56 (32.4)	63 (33.5)	41 (29.5)	0.7377
High red meat intakec		20 (11.6)	21 (11.2)	17 (12.2)	0.9570
Low physical activityd		21 (12.1)	14 (7.5)	9 (6.5)	0.1520
Hormone-replacement therapye		33 (53.2)	34 (50.0)	56 (56.1)	0.7897
Education	≤ 9 years	100 (57.8)	98 (52.1)	84 (60.4)	0.5929
	10–11 years	35 (20.2)	47 (25.0)	29 (20.9)
	≥ 12 years	37 (21.4)	41 (21.8)	25 (18.0)
Regular use of nonsteroidal anti-inflammatory drugs		25 (14.5)	40 (21.3)	24 (17.3)	0.2338
Personal history of colonoscopy		23 (13.3)	45 (23.9)	34 (24.5)	0.0163

Note: values are expressed as n (%)

Abbreviations: GRS, genetic risk score

^aComparison between the three GRS groups, using Chi²-test

^bHigh alcohol intake defined as drinking alcoholic beverages on average on (1) more than five days per week or (2) more than seven standard alcohol glasses per week for women or more than 14 standard alcohol glasses per week for men

^cHigh red meat intake defined as eating either processed meat or red meat more than once per day or eating both processed meat and red meat at least once per day

^dLow physical activity defined as being physically active for less than 30 minutes per day

^eProportions refer to women only

Missing values Low/Medium/High GRS: family history 2/3/4, smoking 2/2/3, body mass index 3/8/1, hormone-replacement therapy 2/1/0, education 1/2/1

Having a GRS in the medium and high tertile was associated with a 2.1-fold and 2.6-fold risk, respectively, for AN in sex- and age-adjusted analyses (ORs 2.07, 95% CI 1.41–3.04 and 2.64, 95% CI 1.77–3.92, Table 3). Additional adjustment for personal history of colonoscopy and low physical activity increased the estimates slightly (ORs 2.16, 95% CI 1.46–3.19 and 2.74, 95% CI 1.84–4.09, for medium and high GRS, respectively). The risk for NAA was not significantly associated with GRS in either model (OR 1.04, 95% CI 0.70–1.55 for high GRS in sex- and age-adjusted model).

Table 3.

Risk of non-advanced adenoma and advanced neoplasm according to genetic risk score (GRS)

GRS category	No neoplasm	Non-advanced adenoma			Advanced neoplasm
	N (%)	N (%)	OR (95% CI)		N (%)	OR (95% CI)
			Model 1a	Model 2b		Model 1a	Model 2b
Low	173 (34.6)	82 (32.9)	Ref.	Ref.	55 (18.7)	Ref.	Ref.
Medium	188 (37.6)	98 (39.4)	1.08 (0.75–1.55)	1.09 (0.76–1.57)	124 (42.2)	2.07 (1.41–3.04)	2.16 (1.46–3.19)
High	139 (27.8)	69 (27.7)	1.04 (0.70–1.55)	1.05 (0.70–1.55)	115 (39.1)	2.64 (1.77–3.92)	2.74 (1.84–4.09)
			Ptrend=0.8158	Ptrend=0.8060		Ptrend<0.0001	Ptrend<0.0001

Abbreviations: CI, confidence interval; GRS, genetic risk score; OR, odds ratio; Ref., reference

^aAdjusted for sex and age

^bAdjusted for sex, age, previous colonoscopy, physical activity; missing values, n=1

Supplementary Table 2 shows the risk for AN by subgroup of features defining the AN status, namely carcinoma, or size (>10mm), histology (cellular or structural atypia) or both size and histology of advanced adenomas. Overall, similar associations were seen in all analyses, even though some of the subgroup specific associations for the smaller groups, in particular participants with CRC, did not reach statistical significance.

C-Statistics, NRI and IDI are presented in Supplementary Table 3. For models with the outcome advanced neoplasm, the GRS improved the predictive value in all adjustment settings. No improvement of the c-statistics could be observed for predicting non-advanced adenoma when additionally considering the GRS next to traditional risk factors.

The results of cubic spline modelling for AN risk are portrayed in Supplementary Figure 2. Compared to the median number of risk alleles in neoplasm-free participants (n=34), ORs at the knots (5^th and 95^th percentile) were 0.40 (95% CI 0.26–0.63) and 1.84 (95% CI 1.19–2.85), respectively. The analyses also show the risk increase according to the number of risk alleles to be monotonous over the full range of the GRS assessed.

Figure 2 depicts the joint relationship of age and GRS with the risk for AN. Due to the strong variation of risk by GRS categories, people with a low GRS reach the same level of risk as those with a medium GRS at much higher ages only. For example, the risk of the 60-year old medium GRS group is reached in the low GRS group at age 73 only. Conversely, the same risk level is reached at age 56 already in the high GRS group.

Figure 2.

Dose-response association between age in years and risk for advanced neoplasm, stratified by genetic risk score (GRS) tertiles, plotted with a joint reference (medium GRS, 60 years)

The RAPs for comparisons between the GRS categories and their 95% CIs are shown in Table 4. The RAPs for having a medium or a high GRS compared to a low GRS were 13.4 (95% CI 4.8–22.0) years and 17.5 (95% CI 7.8–27.3) years, respectively. Persons with a high GRS had a RAP of 4.1 (95% CI −2.0–10.3) years compared to persons with a medium GRS.

Table 4.

Risk advancement periods (RAP) for comparisons between genetic risk scores (GRS) categories

GRS comparison	OR (95% CI)a	RAP (95% CI) [years]
Medium vs. low	2.16 (1.46–3.19)	13.4 (4.8–22.0)
High vs. low	2.74 (1.84–4.09)	17.5 (7.8–27.3)
High vs. medium	1.27 (0.90–1.79)	4.1 (−2.0–10.3)

Abbreviations: CI, confidence interval; GRS, genetic risk score; OR, odds ratio; RAP, risk advancement periods

^aAdjusted for sex, age, previous colonoscopy, physical activity; missing values, n=1

Supplementary Table 4 shows the risk of AN according to the GRS, stratified by age, which was divided into two groups. Associations of the GRS with the risk of AN were tentatively stronger in age group 65–79 years than in age group 50–64 years, but confidence intervals were overlapping, and sensitivity analyses showed that interaction by age was not statistically significant (p=0.7758).

Site-specific analyses are presented in Table 5. Having a higher GRS was associated with higher risks for AN in all bowel sites. The association was most pronounced for proximal neoplasms (ORs for GRS in medium and highest tertile compared to lowest tertile in the fully adjusted model: 3.55, 95% CI 1.85–6.82 and 3.61, 95% CI 1.84–7.10, respectively), with statistically significant differences in heterogeneity analyses (pairwise differences between low vs. medium/high GRS in the three sites) between the proximal and rectal location (p=0.0437). Additional adjustment for history of colonoscopy and physical activity yielded little change in all risk estimates.

Table 5.

Risk of advanced neoplasm according to genetic risk score (GRS), stratified by site of neoplasm

Site of neoplasma	GRS	No finding	Advanced neoplasm
		N (%)	N (%)	OR (95% CI)
				Model 1b	Model 2c
Proximal	Low	173 (34.6)	13 (13.1)	Ref.	Ref.
	Medium	188 (37.6)	49 (49.5)	3.48 (1.82–6.65)	3.55 (1.85–6.82)
	High	139 (27.8)	37 (37.4)	3.55 (1.81–6.96)	3.61 (1.84–7.10)
			ptrend=	0.0004	0.0003
Distal	Low	173 (34.6)	24 (18.9)	Ref.	Ref.
	Medium	188 (37.6)	52 (40.9)	1.96 (1.15–3.33)	2.06 (1.21–3.52)
	High	139 (27.8)	51 (40.2)	2.68 (1.57–4.60)	2.86 (1.66–4.93)
			ptrend=	0.0003	0.0002
Rectum	Low	173 (34.6)	18 (26.5)	Ref.	Ref.
	Medium	188 (37.6)	23 (33.8)	1.16 (0.60–2.24)	1.22 (0.63–2.37)
	High	139 (27.8)	27 (39.7)	1.87 (0.98–3.58)	2.04 (1.06–3.92)
			ptrend=	0.0542	0.0318

Abbreviations: CI, confidence interval; GRS, genetic risk score; OR, odds ratio; Ref., reference

^aNeoplasms in more than one location were allocated to the most proximal location (n=31)

^bAdjusted for sex and age

^cAdjusted for sex, age, previous colonoscopy, physical activity

Sensitivity analyses using a lower threshold for linkage disequilibrium (D′≤0.5) in the construction of the GRS, which resulted in including 39 SNPs only, and sensitivity analyses using a weighted GRS and a unweighted GRS only applied in persons without previous colonoscopies all yielded similar patterns (Supplementary Table 5, 6 and 7). ORs for AN were somewhat lower using the weighted GRS, but confidence intervals were overlapping, and very consistent null results were found for NAA both with the weighted and the unweighted GRS.

Supplementary Table 8 shows results of the replication analyses within the TCPS with a subset of 35 SNPs of the original GRS that were measured with sufficient quality. For comparability, analyses within the BLITZ population were repeated with the same 35 SNPs. Altogether, 957 persons from the TCPS could be used for replication analyses. Thereof, 140 had advanced adenoma, 355 had non-advanced adenoma and 462 had no neoplasm. Similar results were obtained in both studies for most analyses, even though ORs were overall somewhat lower in both studies than in our main analyses given the lower number of SNPs involved. ORs for AN were 1.58 (95% CI 0.92–2.71) and 2.21 (95% CI 1.33–3.68, p_trend=0.002) in TCPS compared to 1.56 (1.08–2.26) and 2.01 (1.41–2.88, p_trend<0.0001) in BLITZ for medium and high GRS, respectively. Non-advanced adenoma were not associated with GRS in the BLITZ study (ORs 0.96, 95% CI 0.67–1.39 and 0.89, 95% CI 0.61–1.29 for medium and high GRS, respectively), but a small risk increase was seen in the TCPS for the highest GRS (ORs 1.08, 95% CI 0.75–1.56 and 1.49, 95% CI 1.06–2.10 for medium and high GRS, respectively). A meta-analysis of both studies yielded ORs of 1.57 (95% CI 1.15–2.13) and 2.08 (95% CI 1.55–2.79, ptrend<0.0001) for medium and high GRS for AN, and ORs of 1.02 (95% CI 0.79–1.32) and 1.18 (95% CI 0.91–1.52, p_trend=0.5671) for medium and high GRS for NAA. RAPs for the TCPS were comparable to the BLITZ population (Supplementary Table 9), as were the analyses regarding the dose-response associations between age, GRS and the risk for advanced neoplasms (Supplementary Figure 3).

Discussion

In this study among participants of screening colonoscopy, a GRS consisting of 48 SNPs that were previously identified to be related to CRC in large scale GWAS, was highly and selectively predictive on the presence of advanced neoplasms (which are advanced adenomas in most cases in screening populations), whereas no association was seen with the presence of non-advanced adenomas. Strong associations were seen in both younger and older participants, and associations were most pronounced for advanced neoplasms located in the proximal colon. A low genetic risk goes along with much later (more than 10 years) attainment of risk for advanced neoplasms compared to persons with medium or high genetic risk.

The consistency of associations of GRS with CRC and AA is in agreement with suggestions that the vast majority of CRC develop from advanced adenomas through the adenoma-carcinoma sequence². Although multiple studies have meanwhile confirmed the usefulness of GRS to predict CRC risk^19–21, data on their association with presence of advanced adenomas have been very sparse. So far, two studies examined this association and they found significant associations of GRS with risk for advanced adenomas indeed^{28, 29}, but both studies included only a subset of the currently known common genetic variants. Other researchers found meaningful associations between single SNPs and the risk for AA in different study populations^30–32. Nevertheless, one of those previous studies already found a stronger association of GRS with proximal neoplasms than distal neoplasms and hypothesized that – equivalent to germline mutations causing Lynch syndrome – many common genetic variants are also more closely related to neoplasms in the proximal colon compared to the distal colon²⁸, a finding fully in line with our results.

Given that prevention of CRC through detection and removal of AA (apart from early detection of CRC) is a key component of CRC screening, the strong association of the GRS with presence of AN may have important implications for risk stratification. In particular, the GRS could complement other factors commonly used for identification of high risk groups, such as a family history of CRC, by allowing to identify both high and low risk groups in whom screening recommendations different from those for the average risk population might be considered. For example, our data showed that persons with a low genetic risk reach the same AN risk as those with a medium genetic risk approximately 13 years later, which suggests that one might consider a substantially higher age for screening initiation in people with low GRS.

Previous research indicated that the majority of published common risk variants is associated with a higher risk for adenomas²⁸. They contribute to CRC promotion as they are involved in early carcinogenesis in the adenoma-carcinoma pathway^{29, 33}. It is assumed that the genetic risk for CRC is likely to be mediated through susceptibility to adenomas³⁴. One would therefore expect to see associations also between GRS and the risk for non-advanced adenoma. However, no such associations were observed in our study (and although a minor risk increase for the highest GRS tertile was seen for non-advanced adenoma in TCPS, combined analyses of both studies yielded no association, neither in single GRS categories nor in trend analyses), which is in line with previous literature, which also concluded that less SNPs are associated with the risk for NAA than with the risk for AA and associations of a GRS were at least much stronger for advanced neoplasms than for non-advanced adenoma^{28, 33}. One possible explanation for these results is that development of non-advanced adenomas might be rather driven by environmental and lifestyle factors (or epigenetic factors), whereas a possible transition to advanced adenomas and subsequent potential promotion to carcinomas might primarily be related to genetic variants.

Our results may have important clinical implications. For example, estimates of RAPs such as those derived from our analyses may help to refine personalized recommendations regarding age at screening initiation which so far are rather crude in most existing guidelines as they often use a positive family history in a first-degree relative as the only or main criterion for this purpose. Although the “family history criterion” allows to identify a small proportion of high risk individuals (typically around 10%), it does not allow any differentiation of risk among the vast majority of people without a family history. Even though risk scores based on environmental and lifestyle factors may provide such differentiation, they may vary during life course due to variation of risk factors over time. Furthermore, for many of these risk factors cumulative exposure over lifetime seems to be more important than current exposure. For example, smoking exposure cumulating over life beyond the common starting age of screening would be of little help to define a starting age for screening. By contrast, the GRS provides a constant value that could be measured once at a relatively young age, such as 40 years, and then be used for personalized advice on when to start screening. The selective association with advanced neoplasms would be a further advantage of the GRS over “environmental” risk scores, as the latter would typically lead to similar “enrichment” of people with low-risk NAA in the apparent high risk group. Most NAA would never develop into cancer during lifetime but their detection at screening might lead to unnecessary concern and burden of surveillance. Nevertheless, the best use of environmental and genetic risk scores would probably be a combination of both rather than relying on each type of score individually.

It would be of great interest to further explore to what extent the GRS would also lend itself for personalized recommendations on screening and surveillance intervals. While no direct evidence for the latter purpose can be drawn from our study, evidence for this highly plausible possibility should be sought for in further studies specifically focusing on participants of surveillance colonoscopies.

Our study has several strengths and limitations. A major strength of the BLITZ study is its conduction in a true screening setting, in which every participant undergoes the gold standard examination – colonoscopy – for detecting CRC and its precursors. Our results therefore pertain to the very target population of CRC screening, and include major targets of CRC screening, i.e. not only CRC, but also AA whose detection and removal at screening enables effective prevention of CRC.

The present study is based on a subset of participants, for whom genotyped data was obtained. Participants with relevant findings at screening colonoscopy were selected at higher proportions to enhance precision and power of the risk estimates and between group comparisons. Despite selecting all CRC patients with usable genotype data from an overall very large screening population, their numbers were still rather low leading to relatively wide confidence intervals for the risk estimates in this subgroup. Our study did not include participants with sessile serrated polyps as a specific risk group, as these lesions were not yet routinely ascertained in a standardized manner in routine screening practice during the earlier years of recruitment. Sample size limitations also hindered more refined categorization of risk strata. As illustrated in our dose-response analyses, risk stratification should be possible along the continuum of the GRS. Only SNPs identified in previous GWAS were included in the GRS. Further enhancements of the GRS are to be expected by the steady discovery of new risk alleles, or other genetic or epigenetic risk markers, mostly by large international GWAS consortia. It should also be kept in mind that our results pertain to a study population of (almost exclusively) European ancestry only. Nevertheless, similarly predictive GRS, albeit based on a partly different set of risk alleles, would be expected for populations with different ancestry. In contrast to the results in the BLITZ study, we saw a slightly increased risk for non-advanced adenoma in the TCPS cohort for the highest GRS tertile. This suggests that there might be room for improvement on the question of which SNPs to include in the GRS when aiming to selectively identify people at increased risk of advanced neoplasms only. The selection of SNPs that were included in our GRS was not driven by the association of each SNP with advanced neoplasms in our study, but by previously identified associations with CRC, and our GRS was therefore not yet optimized for selective identification of people at increased risk of advanced neoplasms only. A GRS based on SNPs explicitly associated with advanced neoplasms only might perform even better than our GRS.

In summary, despite these limitations, we were able to show that a genetic risk score based on published risk alleles might already now enable more precise risk stratification not only for CRC, but also for any advanced neoplasm. The selective (BLITZ study) or at least much stronger (TCPS) association with advanced neoplasms, the most important targets for CRC screening, compared to non-advanced adenomas, could be particularly useful in this context. Further research should aim to replicate and expand our results in larger and ethnically diverse populations and include participants of both screening and surveillance colonoscopy in order to expand the evidence base for more refined, personalized CRC prevention strategies. Further research should also explore both the possibilities to increase the potential for risk stratification of the GRS including additional informative SNPs or for achieving essentially the same potential for risk stratification at lower costs with a lower number of SNPs by excluding redundant or less informative ones. Finally, further research is needed to address the acceptance, feasibility and cost-effectiveness of GRS based risk stratification in CRC screening.

Supplementary Material

supplement

Supplementary Table 1. Previously identified GWAS findings for colorectal cancer in persons of European descent

Supplementary Table 2. Association of the genetic risk score with specific features of advanced neoplasms (AN)

Supplementary Table 3. C-Statistics, net reclassification improvement (NRI) and integrated discrimination improvement (IDI) of models including a genetic risk score (GRS) and easy collectable environmental data

Supplementary Table 4. Risk of advanced neoplasm according to genetic risk score (GRS), stratified by age of person at risk

Supplementary Table 5. Risk of non-advanced adenoma and advanced neoplasm according to genetic risk score (GRS) based on 39 SNPs, LD cut-off=0.5

Supplementary Table 6. Risk of non-advanced adenoma and advanced neoplasm according to weighted genetic risk score (GRS) based on 48 SNPs

Supplementary Table 7. Risk of non-advanced adenoma and advanced neoplasm according to genetic risk score (GRS), only including participants with no history of previous colonoscopy

Supplementary Table 8. Risk of non-advanced adenoma and advanced neoplasm according to genetic risk score (GRS) based on 35 SNPs available in both BLITZ and TCPS in high quality

Supplementary Table 9. Risk advancement periods (RAP) for comparisons between genetic risk scores (GRS) categories (based on 35 SNPs) within the TCPS

Supplementary Figure 1. Distribution of risk alleles in participants of the BLITZ study

Supplementary Figure 2. Risk of any advanced neoplasm according to sum of risk alleles

Supplementary Figure 3. Dose-response association between age in years and risk for advanced neoplasm, stratified by genetic risk score (GRS) tertiles, plotted with a joint reference (medium GRS, 60 years), using TCPS data

Abbreviations

AA

advanced adenoma

AN

advanced neoplasm

BMI

body-mass-index

CI

confidence interval

CRC

colorectal cancer

GRS

genetic risk score

GWAS

genome-wide association study

NAA

non-advanced adenoma

OR

odds ratio

RCT

randomized controlled trial

RAP

risk advancement periods