Genetic variants in the regulatory T cell related pathway and colorectal cancer prognosis

Abstract

Background:

High numbers of lymphocytes in tumor tissue, including T regulatory cells (Treg), have been associated with better colorectal cancer (CRC) survival. Tregs, a subset of CD4⁺ T lymphocytes, are mediators of immunosuppression in cancer and therefore variants in genes related to Treg differentiation and function could be associated with CRC prognosis.

Methods:

In a prospective German cohort of 3 593 CRC patients, we assessed the association of 771 SNPs in 58 T-reg related genes with overall and CRC-specific survival using Cox regression models. Effect modification by microsatellite instability (MSI) status was also investigated since tumors with MSI show greater lymphocytic infiltration and have been associated with better prognosis. Replication of significant results was attempted in 2 047 CRC patients of the International Survival Analysis in Colorectal cancer Consortium (ISACC).

Results:

A significant association of the TGFBR3 SNP rs7524066 with more favorable CRC-specific survival (hazard ratio (HR) per minor allele: 0.83, 95% confidence interval (CI): 0.74–0.94, p-value: 0.0033) was replicated in ISACC (HR: 0.82, 95% CI 0.68–0.98, p-value: 0.03). Suggestive evidence for association was found with two IL7 SNPs, rs16906568 and rs7845577. Thirteen SNPs with differential associations with overall survival according to MSI in the discovery analysis were not confirmed.

Conclusions:

Common genetic variation in the Treg pathway implicating genes such as TGFBR3 and IL7 was shown to be associated with prognosis of CRC patients.

Impact:

The implicated genes warrant further investigation.

INTRODUCTION

Colorectal cancer (CRC) is the third most common cancer in the world ¹. Implementation of population screening and the availability of new treatments have led to a decline in CRC-specific mortality and an increase in the 5-year survival ^2–5. CRC prognosis is very heterogeneous and dependent on several factors. The main prognostic factors that have been identified include TNM stage, tumor grade, presence of metastases (especially in the liver), baseline alkaline phosphatase levels, baseline C-reactive protein and albumin levels (Glasgow prognostic score), as well as neutrophil lymphocyte ratio and pre-operative carcinoembryonic antigen (CEA) levels ^6–10. In addition, the tumor microenvironment has been increasingly recognized to influence the prognosis of cancer including CRC ¹¹. T regulatory cells (Tregs), a subset of CD4⁺ T-lymphocytes expressing the transcription factor FOXP3, are heterogeneous cell types, which play a central role in the maintenance of self-tolerance and immune homeostasis by suppressing the activation, proliferation, and function of numerous immune cells ^{12, 13}. In tumor tissue, Tregs are able to suppress anti-tumor immune response and contribute to the development of an immunosuppressive tumor microenvironment. The presence of high numbers of Tregs has a negative prognostic effect on many cancer types such as breast cancer, melanoma or cervical cancer ¹⁴. In contrast, for CRC patients, the presence of a high number of tumor-infiltrating lymphocytes, including Tregs, in the tumor microenvironment has been associated with a more favorable survival ^14–22. Tumors with microsatellite instability (MSI) are frequently characterized by inflammatory lymphocytic infiltration and tend to be associated with a better survival than non-MSI-high CRCs ^23–25. This may, in part, be due to more effective immune responses involving Tregs ^{26, 27}.

Genetic variation in inflammatory genes could play a role in the survival of patients after CRC diagnosis ²⁸. To gain further insight into the biological mechanisms underlying Treg pathway and survival after CRC, we investigated common, inherited single nucleotide polymorphisms (SNPs) affecting genes involved in the regulation of Treg functions.

No studies have so far investigated a possible influence of genetic variants in Treg-related genes on the prognosis of CRC patients. Therefore, our aim was to investigate the association between 771 germline variants in 58 Treg-related genes and the overall disease-specific survival of CRC patients, and to assess possible effect modification by MSI status.

MATERIALS AND METHODS

The study sample consisted of CRC patients recruited into the ongoing population-based case-control study DACHS (Darmkrebs: Chancen der Verhütung durch Screening) conducted in the Rhine-Neckar Odenwald region in southwestern Germany ^{29, 30}. Cases diagnosed between January 2003 and December 2013 were included if they were older than 30 years of age (with no upper limit), were able to communicate in German, were able to participate in a personal interview of around one hour and were a resident of the Rhine-Neckar Odenwald region. Only histologically confirmed cases who were diagnosed with their first primary CRC (ICD-10:C18-C20) were included. All patients gave their written informed consent. The study was approved by the relevant ethical committees of the University of Heidelberg and the State Medical Boards of Baden-Württemberg and Rhineland-Palatinate, Germany and was conducted in agreement with the Declaration of Helsinki.

At baseline, trained interviewers collected information on the patient’s demographics, anthropometric indices, medical history including reproductive history, and lifestyle factors. A blood sample was requested at baseline and for a minority of patients who refused to provide blood, a mouthwash sample was collected instead (0.9% of participants). After five years of follow-up, information on treatment and disease course was collected from the treating physician. Vital status was collected from the population registries and cause of death was verified by death certificates from health authorities.

GENOTYPE DATA

The Flexigene kit was used to extract DNA from EDTA blood and mouthwash samples of the DACHS patients and quantification of the DNA was performed using Q6 Quanti-iT picoGreen dsDNA reagent and kit (Invitrogen/Life technologies, Darmstadt, Germany).

Genotyping was performed in collaboration with the Genetics and Epidemiology of Colorectal Cancer Consortium (GECCO); details have been previously described ³¹. DACHS samples were genotyped using the whole-genome Illumina CytoSNP assay (Illumina, San Diego, CA, USA) for patients recruited 2003–2006, the Illumina HumanOmniExpress BeadChip Kit for patients recruited 2007–2010 and the Illumina HumanOmniExpress BeadChip kit or the Illumina Infinium OncoArray-500K BeadChip for those recruited 2011–2013. For quality control, genotyped variants were excluded based on call rate (< 98%), lack of Hardy–Weinberg Equilibrium in controls (HWE, P < 1 × 10− 4), and low minor allele frequency (MAF < 0.05) as described elsewhere ^31–34. Samples were imputed using as reference panel the cosmopolitan haplotypes from Phase 1 of the 1,000 Genome Project (for patients recruited between 2003 and 2010) or the Haplotype Reference Consortium (for patients between 2011 and 2013) ³⁵ using the University of Michigan Imputation Server ³⁶. Before imputation, Shapeit2 was used to phase the GWAS data ³⁷.

SNP SELECTION

Through extensive literature research, the most important genes in the Treg pathway were selected. Tagging SNPs were selected to represent genetic variation across the genes. SNPs in these genes as well as SNPs in the flanking regions (e.g. +/− 10 kb) were considered, after which Haploview 4.2 (Broad Institute, Cambridge, MA, USA) was used for the selection of tagging SNPs, with a pairwise tagging approach based on reference data from the HapMap project (Utah resident with Northern and Western Europe ancestry (CEU population), Phase II/Release 24. In total, 771 SNPs in 58 genes were selected for this analysis (see Supplementary Table 1).

MSI DATA

Formalin-fixed paraffin-embedded (FFPE) tumor samples were used to determine MSI status. FFPE samples were collected from the different pathology departments at the cooperating hospitals and were stored at the tissue bank at the National Center for Tumor Diseases (NCT) in Heidelberg. The area with the highest tumor cell concentration was identified microscopically and this section was then isolated using the DNeasy kit from Qiagen. A mononucleotide marker panel including BAT25, BAT26, and CAT25 was used to determine MSI status. Tumors showing amplifications in two or more markers were classified as MSI-high ³⁸. These markers have a sensitivity of 98.2% and a specificity of 100% to differentiate MSI-high from non–MSI-high tumors ^{38, 39}.

DATA ANALYSIS

Cox regression models were used to test the individual SNP associations with overall survival and CRC-specific survival. Hazard ratios (HRs) and 95% confidence intervals (CIs) were calculated for each SNP. Survival time was calculated from date of diagnosis until date of death by any cause, death by CRC or date of last contact. Median follow-up time was calculated using the reverse Kaplan-Meier method ⁴⁰. Age, sex, and TNM stage were included in the model as relevant prognostic factors. Additional covariates were determined using backward elimination of a set of variables including grade (1, 2 vs 3, 4), family history of CRC in first degree relatives (no vs. yes), smoking, body mass index (18.5–25, 25–30, 30+ kg/m²), alcohol intake (0 and quartiles in subjects with alcohol intake >0g/day), and physical activity (0 and quartiles). The variables body mass index (at diagnosis) and alcohol intake were retained in the final model as they were significantly associated with overall survival. Heterogeneity in the associations of the Treg gene polymorphisms with overall survival according to MSI status was assessed statistically using interaction terms between MSI status (high, non-high) and the polymorphisms and was evaluated using the likelihood ratio test. The SNP association according to MSI status was also estimated in subgroup analysis. The proportional hazards assumption was tested according to Grambsch and Therneau ⁴¹. The statistical analysis was carried out using SAS version 9.3 (SAS Institute, Cary, NC, USA) and R version 3.1.0 (www.r-project.org).

REPLICATION SET

For SNPs that showed a significant association with survival or an interaction with MSI (p-value ˂ 0.01), replication was performed using studies participating in the International Survival Analysis in Colorectal cancer Consortium (ISACC), a consortium aimed at investigating demographic, environmental, and genetic risk factors in association with CRC survival ⁴². For replication of the SNP associations with overall survival and CRC-specific survival, 1821 CRC patients from six studies were included: Diet And Lifestyle Study (DALS), Health Professionals Follow-up study (HPFS), Nurses` Health Study (NHS), Colon Cancer Family Registry (CCFR), Cancer Prevention Study II (CPS II), and the Melbourne Collaborative Cohort Study (MCCS). All studies were genotyped on Illumina GWAS platforms and imputed to the Haplotype Reference Consortium (HRC) panel using the University of Michigan Imputation Server ³⁵. Prior to imputation, Shapeit2 was used to phase the GWAS data ³⁷. Details of genotyping and QC for studies included in the validation are described elsewhere^43,32–34. Patients included in ISACC studies are of European ancestry.

For replication of results on effect modification according to MSI status, 1 554 CRC patients from five studies (DALS, HPFS, NHS, CPS II, MCCS) with available data on MSI were included. For DALS MSI status was determined using 12 markers, a panel of ten tetranucleotide repeats and the two mononucleotide repeats (BAT-26 and TGFBR2) ⁴⁴. For HPFS and NHS, MSI status was assessed based on the same ten tetranucleotide repeats ⁴⁵. For MCCS, MSI status was determined using a 10-loci panel in tumor DNA and matched normal tissue DNA (BAT25, BAT26, BAT40, MYCL, D5S346, D17S250, ACTC, D18S55, D10S197 and BAT34C4)⁴⁶. For CPS II, determination of MSI status was based on the Bethesda Consensus Panel⁴⁷. Classification was based on ≥5 interpretable markers (unless all four markers were unstable, in which case the tumor was classified as MSI-high). For these studies, MSI-high tumors were defined when ≥30% of the markers showed instability.

Analyses were performed by pooling the samples from all studies and adjusting for study. We used the same statistical methods as for the discovery phase. To account for the overrepresentation of patients with a positive family history of CRC in the DALS Minnesota samples, we included family history as an additional covariate in the MSI interaction analysis given the correlation between positive family history and MSI-high tumors. Missing data on environmental factors was imputed using single imputation.

FUNCTIONAL ANNOTATION

To add functional information to significant variants, we used the National Cancer Institute’s ‘LDlink’ web tool (https://ldlink.nci.nih.gov) to find all variants in linkage disequilibrium (LD; R² ≥ 0.4 in Phase 3 1000 Genomes ‘EUR’ population) with the significant variant. Subsequently, we used the Variant Effect Predictor (VEP) tool of the Ensembl webpage (https://uswest.ensembl.org/info/docs/tools/vep/index.html) to show annotations ⁴⁸.

DATA AVAILABILITY

Genotyping data of the GECCO studies are available at the database of Genotypes and Phenotypes (dbGaP) for download at the accession number: phs001078.v1.p1.

RESULTS

Table 1 shows the characteristics of the DACHS study participants. The majority of patients were aged between 60–80 years at diagnosis with a median of 69 years and 60.5% were male. Over 60% of the patients were diagnosed with TNM stage II or III disease and 60.2% had a tumor in the colon. The median follow-up time was 63.0 months and, in total, 1 100 patients died during study follow-up. See the flow diagram in Figure 1 for inclusion of patients in the discovery stage. In the DACHS study, 10.7% of patients with data on MSI status were MSI-high and 89.3% were non-MSI-high.

Table 1).

Basic characteristics for 3593 CRC DACHS patients with complete follow-up information and for 1965 CRC patients according to MSI status

	All	MSI-high	Non-MSI-high
N (number of events)	3593 (1100)	211 (50)	1754 (590)
	n (%)	n (%)	n (%)
Sex
Female	1420 (39.5)	110 (52.1)	712 (40.6)
Male	2173 (60.5)	101 (47.9)	1042 (59.4)
Age
<60 years	744 (20.7)	36 (17.1)	333 (19.0)
60-<70 years	1123 (31.3)	59 (28.0)	661 (31.4)
70-<80 years	1189 (33.1)	61 (28.9)	607 (34.6)
≥80 years	537 (14.9)	55 (26.1)	263 (15.0)
Median (Interquartile range)	69 (61–76)	71 (64–80)	69 (62–76)
TNM Stage
1	820 (22.8)	30 (14.2)	325 (18.7)
2	1088 (30.3)	113 (53.6)	554 (31.6)
3	1188 (32.9)	62 (28.4)	612 (34.9)
4	502 (14.0)	6 (2.8)	260 (14.8)
Site
Colon	2164 (60.2)	200 (94.8)	1036 (59.1)
Rectum	1429 (39.8)	11 (5.2)	718 (40.9)
CRC-specific death
No	2811 (78.2)	191 (90.5)	1334 (76.1)
Yes	718 (20.0)	19 (9.0)	404 (23.0)
Missing	64 (1.8)	1 (0.5)	16 (0.9)
Recurrence
No	2576 (71.7)	185 (87.7)	1206 (68.8)
Yes	983 (27.4)	25 (11.8)	542 (30.9)
Missing	34 (0.9)	1 (0.5)	6 (0.3)
BMI category
Normal weight	1390 (38.7)	65 (30.8)	704 (40.1)
Overweight	1527 (42.5)	91 (43.1)	728 (41.5)
Obese	676 (18.8)	55 (26.1)	322 (18.4)
Current alcohol intake (g/day)
no alcohol	1093 (30.4)	74 (35.1)	524 (29.9)
0.1–6.1	705 (19.6)	52 (124.6)	331 (18.9)
6.1–15.6	624 (17.4)	34 (16.1)	303 (17.3)
15.6–32.6	607 (16.9)	31 (14.7)	306 (17.4)
>= 32.6	564 (15.7)	20 (9.5)	290 (16.5)
Family history of CRC
No	3086 (85.9)	174 (82.5)	1505 (85.8)
Yes	507 (14.1)	37 (17.5)	249 (14.2)

Abbreviations: BMI, body mass index; CRC, colorectal cancer; MSI, microsatellite instabilitiy

Figure 1.

Flow diagram on patient’s participation in the discovery stage

Three SNPs showed an association with overall survival (nominal p-value < 0.01) in the single SNP analysis (Table 2; see supplementary table 2 for results of all SNPs). The minor alleles of two genetic variants were associated with an increased risk of dying, rs2290065 (CCR7) with a HR of 1.31 per allele (95% CI: 1.07–1.61) and rs10815237 (CD274) with a HR of 1.13 per minor allele (95% CI 1.04–1.23). The minor allele of rs2421826 (CD44) was associated with lower overall survival (HR: 0.89, 95% CI: 0.82–0.97).

Table 2).

SNPs in Treg related genes showing associations with overall survival in 3593 CRC DACHS patients at nominal p-value <0.01 and replication in independent 2047 CRC patients

		DACHS		ISACC
SNP	Gene	Hazard Ratio per minor allele (95% CI)	p-value	Hazard Ratio per minor allele (95% CI)	p-value
Overall survival
rs10815237	CD274	1.13 (1.04– 1.23)	0.0064	1.06 (0.94–1.19)	0.38
rs2421826	CD44	0.89 (0.82– 0.97)	0.0090	0.98 (0.87–1.10)	0.71
rs2290065	CCR7	1.31 (1.07– 1.61)	0.0095	0.92 (0.72–1.19)	0.87
CRC-specific survival
rs10815237	CD274	1.19 (1.07– 1.32)	0.0018	1.14 (0.97–1.35)	0.16
rs7524066	TGFBR3	0.83 (0.74– 0.94)	0.0033	0.82 (0.68–0.98)	0.03
rs17571088	TGFBR3	0.82 (0.71– 0.94)	0.0050	0.95 (0.78–1.17)	0.65
rs1495578	TGFBR2	1.17 (1.05– 1.30)	0.0046	1.09 (0.94–1.27)	0.25
rs17623772	TGFBR2	1.16 (1.04– 1.29)	0.0081	1.07 (0.91–1.25)	0.41
rs4252328	TGFB3	0.84 (0.73– 0.95)	0.0070	1.06 (0.90–1.27)	0.48
rs16906568	IL7	1.18 (1.05– 1.33)	0.0050	1.20 (1.00–1.43)	0.05
rs7845577	IL7	1.22 (1.05– 1.42)	0.0084	1.23 (0.98–1.55)	0.07
rs2421826	CD44	0.87 (0.78– 0.97)	0.0098	0.92 (0.79–1.08)	0.31

Abbreviations: BMI, body mass index; CI, confidence intervall; CRC, colorectal cancer; DACHS, Darmkrebs: Chancen der Verhütung durch Screening; ISACC, International Survival Analysis in Colorectal cancer Consortium; SNP, single nucleotide polymorphism.Hazards ratios, confidence intervals and p-values estimated from Cox Proportional Hazards regression analysisModels were adjusted by age, sex, and TNM stage, BMI at diagnosis, current alcohol intake

For CRC-specific survival, nine SNPs were associated at p-value<0.01. Here, the minor alleles of two SNPs in each of the genes IL7 (rs7845577 and rs16906568; LD r2=0.35) and TGFBR2 (rs1495578 and rs17623772; LD r2=0.8) as well as rs10815237 in gene CD274 were associated with poorer CRC-specific survival. The latter SNP rs10815237 was also associated with overall survival. The minor alleles of the other four SNPs, two in TGFBR3 (rs7524066 and rs17571088; LD r2=0.36), and one each in TGFB3 (rs4252328) and CD44 (rs2421826) were associated with better CRC-specific survival (Table 2, see supplementary Table 3 for all SNPs).

DIFFERENTIAL ASSOCIATIONS BY MSI STATUS

The results of the effect modifications by MSI status (211 MSI-high tumors, 1754 non-MSI-high) are shown for all SNPs in Supplementary Table 4. In the replication sample (ISAAC) 18.3% of samples were MSI-high and 81.7% non-MSI-high. Thirteen SNPs showed statistically significant interaction (nominal p-value <0.01) with MSI status (Table 3). Three of the SNPs lie in gene CD4 (rs7957426, rs10774451, rs10849524), the minor alleles of two of which were associated with decreased survival in MSI-high tumors (rs7957426, rs10774451) and of one SNP rs10849524 was associated with increased survival in MSI-high tumors. These SNPs were not associated with survival in non-MSI-high tumors. Five of the SNPs that showed significant heterogeneity were annotated to gene HLA-DRA, of which the minor alleles of four SNPs were also associated with decreased survival (rs3129848, rs17496549, rs3135392, rs9268644) in MSI high tumors. None of these five SNPs were associated with survival in non-MSI-high tumors. Two SNPs in IL15RA (rs2228059 and rs1998521) also showed associations with overall survival although in different directions for the minor alleles, and only in MSI high tumors but not in non-MSI-high tumors. The remaining SNPs, rs2069772 (IL2), rs11165376 (TGFBR3) and rs7135373 (IFNG), were also found associated with decreased survival solely in MSI high tumors.

Table 3).

SNPs in Treg related genes showing differential association with overall survival according to MSI status in 1965 CRC patients (at nominal p-value for interaction <0.01) and replication in independent 1307 CRC patients

		DACHS					ISACC
		MSI-High (211 CRC patients)		Non-MSI-high (1754 CRC patients)			MSI-High (239 CRC patients)		Non-MSI-high (1068 CRC patients)
SNP	Gene	Hazard Ratio per minor allele (95% CI)	p-value	Hazard Ratio per minor allele (95% CI)	p-value	p-value interaction	Hazard Ratio per minor allele (95% CI)	p-value	Hazard Ratio per minor allele (95% CI)	p-value	p-value interaction
Overall mortality
rs7957426	CD4	2.13 (1.39– 3.26)	0.0005	1.00 (0.89– 1.13)	0.9754	0.0035	1.28 (0.81–2.02)	0.2888	0.94 (0.82–1.07)	0.3493	0.3445
rs3129848	CD4	2.00 (1.28– 3.12)	0.0024	0.97 (0.86– 1.09)	0.5737	0.0045	1.58 (0.99–2.53)	0.0573	0.92 (0.81–1.05)	0.2209	0.3800
rs10774451	CD4	0.51 (0.32– 0.80)	0.0031	1.02 (0.90– 1.14)	0.7973	0.0052	1.59 (1.00–2.54)	0.0501	0.91 (0.81–1.04)	0.1711	0.3195
rs17496549	HLA-DRA	2.09 (1.20– 3.64)	0.0088	1.16 (0.97– 1.39)	0.0995	0.0052	0.49 (0.25–0.96)	0.0389	1.34 (1.11–1.62)	0.0026	0.2190
rs10849524	HLA-DRA	1.75 (1.10– 2.78)	0.0179	0.97 (0.85– 1.10)	0.6514	0.0041	0.50 (0.30–0.84)	0.0080	1.15 (1.00–1.32)	0.0513	0.3972
rs3135392	HLA-DRA	1.50 (1.00– 2.26)	0.0512	0.93 (0.82– 1.05)	0.2387	0.0067	0.59 (0.37–0.94)	0.0269	0.99 (0.87–1.14)	0.9220	0.1468
rs2069772	HLA-DRA	1.51 (0.99– 2.31)	0.0554	0.96 (0.85– 1.07)	0.4497	0.0097	2.30 (1.38–3.81)	0.0013	0.82 (0.72–0.93)	0.0027	0.8181
rs6911419	HLA-DRA	0.72 (0.48– 1.09)	0.1165	1.10 (0.98– 1.23)	0.0939	0.0075	2.28 (1.38–3.75)	0.0012	0.90 (0.79–1.03)	0.1388	0.5347
rs2228059	IL15RA	1.80 (1.15– 2.82)	0.0107	0.92 (0.82– 1.03)	0.1455	0.0075	0.98 (0.64–1.49)	0.9113	0.96 (0.84–1.10)	0.5549	0.8178
rs1998521	IL15RA	0.60 (0.38– 0.94)	0.0267	1.03 (0.92– 1.16)	0.6011	0.0076	0.86 (0.55–1.34)	0.5042	1.00 (0.87–1.14)	0.9774	0.6695
rs11165376	TGFBR3	0.54 (0.31– 0.93)	0.0275	0.98 (0.86– 1.12)	0.8021	0.0082	1.13 (0.69–1.85)	0.6343	0.92 (0.81–1.06)	0.2581	0.5994
rs7135373	IFNG	0.47 (0.27– 0.82)	0.0079	0.99 (0.87– 1.12)	0.8357	0.0089	1.07 (0.65–1.77)	0.7889	0.92 (0.80–1.05)	0.2149	0.7597
rs9268644	IL2	0.56 (0.33– 0.95)	0.0326	0.98 (0.86– 1.11)	0.7510	0.0075	1.51 (0.95–2.39)	0.0830	0.95 (0.82–1.09)	0.4507	0.6975

Abbreviations: CI, confidence interval; CRC, colorectal cancer; MSI, microsatellite instability; SNP, single nucleotide polymorphism.

Hazards ratios, confidence intervals and corresponding p-values estimated from Cox Proportional Hazards regression analysis

Interaction p-values calculated using likelihood ratio tests comparing the model with and without interaction term.

REPLICATION ANALYSIS

The characteristics of the participants from the ISACC studies included in the replication analysis are shown in Supplementary Table 5. None of the SNPs associated with overall survival in the DACHS discovery set showed significant association with overall survival in the ISACC replication sample (see Table 2). One of the nine associations with improved CRC-specific survival in DACHS, for rs7524066 (TGFBR3), was confirmed in the replication analysis, with a similar magnitude of association (HR: 0.82, 95% CI: 0.68–0.98). Two more SNPs (rs16906568 and rs7845577) in gene IL7, associated with poorer survival, were replicated with similar effect sizes and borderline significance (p-values of 0.05 and 0.07, respectively). Although not reaching statistical significance most SNPs showed the same direction of association in the replication sample except for rs4252328 (TGFB3). None of the SNPs that showed differential association by MSI status in DACHS were confirmed in ISACC.

FUNCTIONAL GENOMIC ANNOTATION

The SNP rs7524066 (TGFBR3), for which the association with disease-specific survival was replicated was further investigated to add functional information. We found 33 SNPs to be in LD with the investigated variant (R² > 0.4) (Supplementary Table 6). We further assessed the function of these variants using the variant effect predictor (VEP) tool of the Ensembl webpage (Supplementary Table 7). We found that 15 of the LD SNPs are located in regulatory regions of the gene. Two of the SNPs are located in transcription factor binding sites (Supplementary Table 7).

DISCUSSION

We investigated the association of 771 Treg-related genetic variants with overall and CRC-specific survival in a large cohort of CRC patients and performed replication of top findings in an independent cohort of CRC patients from the ISACC consortium. While none of the SNPs associated with overall survival were confirmed in the independent dataset, one of nine SNPs associated with CRC-specific survival in the discovery dataset, rs7524066 (TGFBR3), was confirmed in the independent replication dataset. The minor allele was similarly associated with better CRC-specific survival in the discovery sample (HR: 0.83, 95% CI: 0.74–0.94) and the replication sample (HR: 0.82, 95% CI: 0.68–0.98).

Based on previous observations of differential T-cell infiltration of colorectal tumors according to MSI status ^{26, 49, 50}, we also evaluated SNPs association with overall survival by MSI status. We found thirteen SNPs in six different genes that showed interactions with MSI in the discovery set (p<0.01) but were not able to replicate any of these interactions.

The SNP rs7524066, for which the association was replicated in the primary analysis of CRC-specific survival, is annotated to gene TGFBR3 (transforming growth factor β type III receptor), encoding one of the transforming growth factor β (TGF-β) receptors, which is also known as betaglycan ⁵¹. TGF-β is an important growth factor for normal development and homeostasis of all cells in the human body and can have both tumor suppressor as well as tumor promoting functions depending on context ^{52, 53}. In contrast to the other two TGF-β receptors, TGFBR1 and TGFBR2, TGFBR3 does not have kinase activity ⁵³. Still, it is not only a co-receptor but seems to act as a tumor suppressor for several cancer types ^54,55. TGFBR3 appears to suppress WNT/CTNNB1 (β-catenin) signaling ⁵⁶, which is linked to CRC development and progression. Our in silico functional analyses indicated that several SNPs in LD with rs7524066 lie in regulatory regions of the gene TGFBR3 and therefore might modify gene regulation. These mechanisms support the plausibility of TGFBR3 being associated with CRC-specific survival.

Furthermore, the association with worse CRC-specific survival of the minor allele of two SNPs (rs16906568 and rs7845577) related to gene IL7 (LD r2=between the SNPs =0.35) is of interest. IL7 is a cytokine which is important for B and T-cell development. Expression of IL7 was higher in CRC patients compared to controls and was associated with metastatic disease ⁵⁷. Therefore, IL7 variants could have a role in survival after CRC diagnosis.

The CD274 variant rs10815237 was associated with both overall and CRC-specific survival and showed fairly similar magnitude of association particularly for CRC-specific survival in the replication sample albeit non-significant. Tumor CD274 (programmed cell death 1 ligand 1, PD-L1) expression has been associated inversely with Treg density in CRC ⁵⁸. Tumor CD274 expression may modify prognostic association of aspirin ⁵⁹. These data suggest that CD274 (PD-L1) may modify CRC behavior depending on other factors in the tumor immune microenvironment. It would be of interest to examine the interaction of the CD274 variant and Treg density (or aspirin use) in future prognostic studies.

One SNP (rs2421826) in the CD44 gene was also found associated with improved both overall as well as CRC-specific survival in the discovery analysis but was not statistically significantly associated in the replication set. CD44 is a multistructural and multifunctional cell surface adhesion molecule that is highly expressed in many cancers and involved in physiological processes. Through interaction with extracellular matrix ligands, it promotes the migration and invasion processes involved in metastases. Functionally active CD44 is associated with enhanced suppressor activity of Treg ⁶⁰. Expression of stem-like factors including CD44 has been associated with metastatic disease and poorer prognosis in CRC ⁶¹ Two SNPs in gene TGFBR2 were associated with worse CRC-specific survival, but were not replicated. One of the roles of TGFBR2 is Treg suppression ⁶² and its inactivation has been associated with the development of CRC ⁶³. These mechanisms make it plausible that genetic variants in TGFBR2 could be associated with CRC-specific survival.

We were not able to confirm any of the results stratified by MSI status, which could be partly due to the difference in the panels used for MSI characterization. The inability to confirm also many of the SNPs associated with overall- or CRC-specific survival in the discovery sample could be in part due to the limited power since the replication sample was smaller than the discovery sample. New association findings are generally biased upwards in discovery datasets so that larger study samples are required for replication. Measurement of Treg/FOXP3 cell expression within the tumour may improve future studies investigating Treg-related SNP associations with CRC survival.

Research into Tregs remains challenging as the definition of Tregs has changed over the past decade. The interplay with other T helper cells, expression of surface markers as well as expression of cytokines influencing functionality of T cell subtypes add to the complexity of this research field. Several studies have shown that different factors, including activated state of the immune cells, their location in the cellular matrix, ratio of different immune cells, can influence their involvement in tumor progression and subsequent CRC survival ^21,64. Two distinct subpopulations of Treg cells were recently identified and shown to have differential impact on CRC prognosis ⁶⁵. Genetic variation, which can be robustly measured, could help to provide further evidence for the prognostic impact of regulatory T cells on CRC prognosis.

This study is one of the first studies investigating genetic variations in the Treg pathway with respect to CRC survival. Replication was attempted in an independent sample of CRC patients. A large amount of genotype data was available, which enabled the comprehensive investigation of the Treg pathways. The quality control measures all showed that the genotype data was of high quality. There were no opportunities to perform functional analyses for the SNPs that were implicated through these analyses.

Although no strong associations were found, there is suggestive evidence based on these analyses particularly for the TGF beta receptors and the biological functions of the implicated genes to support further investigations of the aforementioned SNPs and genes with respect to CRC prognosis in large study samples.