Inflammatory Bowel Disease Risk Variants Are Associated with an Increased Risk of Skin Cancer

Abstract

Background

Inflammatory bowel disease is associated with an increased risk of skin cancer. The aims of this study were to determine whether IBD susceptibility variants are also associated with skin cancer susceptibility and if such risk is augmented by use of immune-suppressive therapy.

Methods

The discovery cohort included participants in the UK Biobank. The validation cohort included participants in the Michigan Genomics Initiative. The primary outcome of interest was skin cancer, subgrouped into nonmelanoma skin cancers (NMSC) and melanoma skin cancers (MSC). Multivariable logistic regression with matched controls (3 controls:1 case) was performed to identify genomic predictors of skin malignancy in the discovery cohort. Variants with P < .05 were tested for replication in the validation cohort. Validated Single nucleotide polymorphisms were then evaluated for effect modification by immune-suppressive medications.

Results

The discovery cohort included 10,247 cases of NMSC and 1883 cases of MSC. The validation cohort included 7334 cases of NMSC and 3304 cases of MSC. Twenty-nine variants were associated with risk of NMSC in the discovery cohort, of which 5 replicated in the validation cohort (increased risk, rs7773324-A [DUSP22; IRF4], rs2476601-G [PTPN22], rs1847472-C [BACH2], rs72810983-A [CPEB4]; decreased risk, rs6088765-G [PROCR; MMP24]). Twelve variants were associated with risk of MSC in the discovery cohort, of which 4 were replicated in the validation cohort (increased risk, rs61839660-T [IL2RA]; decreased risk, rs17391694-C [GIPC2; MGC27382], rs6088765-G [PROCR; MMP24], and rs1728785-C [ZFP90]). No effect modification was observed.

Conclusions

The results of this study highlight shared genetic susceptibility across IBD and skin cancer, with increased risk of NMSC in those who carry risk variants in IRF4, PTPN22, CPEB4, and BACH2 and increased risk of MSC in those who carry a risk variant in IL2RA.

Introduction

Inflammatory bowel disease (IBD) is a chronic immune-mediated disease of the gastrointestinal tract. Patients with IBD are at higher risk for both intestinal and extraintestinal malignancies.¹ Traditionally, intestinal malignancies have been attributed to the secondary effects of disease-related chronic inflammation.^2–4 Alternatively, extraintestinal malignancies have been attributed to the effects of immune suppressive therapy. However, recent epidemiologic data suggest that there may be an increased risk of extraintestinal malignancies in IBD that are mediated via medication-independent effects.^5,6 This work underscores the importance of further investigation into the pleiotropic effects of IBD-risk variants. In this study, we sought to investigate whether genomic variants that increase the risk of IBD also increase the risk of a common extraintestinal malignancy, skin cancer.

Increased risks of both nonmelanoma skin cancers (NMSC) and melanoma skin cancers (MSC) have been described among patients with IBD.⁷ However, the exact mechanisms by which skin cancers occur in the IBD population remain unclear. Nonmelanoma skin cancer includes both squamous cell and basal cell carcinomas. In 2012, there were estimated to be more than 5 million cases of NMSC in the United States, with a relatively equal breakdown between squamous and basal cell carcinomas.⁸ The incidence rate of NMSC in IBD has been shown to be increased above the general population (incidence rate ratio 1.64; 95% CI, 1.51-1.78).⁹ This increased risk of NSMC has been linked with the use of immune suppressive medications. In a study using a national claims data set, both recent (odds ratio, [OR] 3.56; 95% CI, 2.81-4.50) and persistent (OR, 4.27; 95% CI, 3.08-5.92) thiopurine use were associated with increased risk of NMSC in IBD.⁹ The increased risk of NMSC with thiopurine use has been confirmed in several studies including a large prospective observational cohort from France (hazard ratio [HR], 5.9; 95% CI, 2.1-16.4)¹⁰ and a separate national claims data set (OR, 1.85; 95% CI, 1.66–2.05).⁷ In addition, a recent systematic review highlighted that a majority of studies have shown a higher risk of NMSC with thiopurine use.¹¹

Significant associations between antitumor necrosis factor (anti-TNF) therapies and NMSC have also been observed in several studies across many autoimmune diseases. First, in the nested case-control study referenced mentioned previously, the authors showed that recent (OR, 2.07; 95% CI, 1.28-3.33) and persistent (OR, 2.18; 95% CI, 1.07-4.46) anti-TNF use were associated with an increased risk of NMSC in Crohn’s disease.⁹ Second, in a meta-analysis of 6 studies with 123,031 individuals, anti-TNF use was associated with an increased risk of NMSC in rheumatoid arthritis (relative risk [RR], 1.28; 95% CI, 1.19-1.38).¹² Third, in a study using data from 13 clinical trials, anti-TNF use was associated with an increased risk of NMSC in psoriasis (standardized incidence rate, 1.76; 95% CI, 1.26-2.39).¹³ Therefore, both thiopurines and anti-TNFs are considered to be risk factors for the development of NMSC.

In 2011, there were more than 65,000 cases of MSC in the United States with over 9000 melanoma-related deaths.¹⁴ The risk of MSC has been found to be independently associated with IBD, with conflicting evidence regarding the impact of anti-TNF use. The independent association between MSC and IBD was reported in a systematic review and meta-analysis, which utilized 179 cases across 12 studies.⁵ In this manuscript, authors reported an increased risk of MSC in IBD patients compared with the general population (RR, 1.37; 95% CI, 1.10-1.70). This risk persisted when including only those studies which were completed prior to introduction of anti-TNFs (RR, 1.52; 95% CI, 1.02–2.25) and was not related to thiopurine use (RR, 1.10; 95% CI, 0.73-1.66), suggesting that the increased MSC risk in IBD patients was a medication-independent effect. There has also been data supporting a medication-related increase in risk, specifically with anti-TNFs. In a nested case-control study of claims data from the LifeLink Health Plan, each IBD patient with MSC was matched to 4 IBD patients without MSC. The results of the study showed that anti-TNF therapies were associated with an increased risk of MSC (OR, 1.88; 95% CI, 1.08-3.29).⁷ However, subsequent data has been conflicting.^5,15

The medication-independent association between MSC and IBD raises the possibility that there may be biologic mechanisms inherent to IBD, which also predispose patients to skin cancer. Interestingly, it was recently shown that patients with pediatric-onset Crohn’s disease are at a higher risk of developing malignancies later in life compared with controls, independent of thiopurine or anti-TNF use.⁶ These data suggest that mechanisms inherent to the pathogenesis of IBD may also be important in cancer pathogenesis. This is not surprising given the critical role of the immune system in both autoimmune disease and tumor surveillance. Furthermore, the skin and the intestine are highly similar organs, both constantly being exposed to environmental pathogens, necessitating a level of immune homeostasis (ie, a balance between immune tolerance to nonpathogenic exposures and immune activation against pathogenic exposures). Therefore, shared immune perturbations that confer risk of both intestinal inflammation and skin malignancy are biologically plausible.

As the use of immune suppressive therapies continues to expand, a deeper understanding of the impact of IBD risk variants on skin malignancy would help to better direct therapeutic choices. In this study, we aimed to determine (1) if the genetic variants that increase the risk of IBD also increase the risk of skin cancer, independent of immune suppressive medication use, and (2) if there are gene-medication interactions that confer multiplicative risk.

Materials and Methods

Study Design

This study was conducted using a retrospective case-control study design, with 2 prospectively recruited cohorts (discovery cohort, UK Biobank; validation cohort, Michigan Genomics Initiative).

Study Cohorts

UK Biobank

The UK Biobank (UKBB) is a prospective population-based cohort of middle-aged individuals (40-69 years) who were recruited between the years 2006 and 2010 at over 20 centers to capture ethnic, socioeconomic, and geographic diversity.¹⁶ Comprehensive data including completion of an electronic consent and self-reported questionnaire and collection of biospecimens were obtained at the index visit. Details of the study protocol and interview process can be found on the UKBB website (https://www.ukbiobank.ac.uk/). The initial ~50,000 participants were genotyped using the Affymetrix UK BiLEVE Axiom array.¹⁷ The remainder of participants were genotyped using the Affymetrix UK Biobank Axiom array. There is a high degree of overlap between the 2 arrays, with over 95% commonality. For the UK Biobank, genotypes were imputed from the Haplotype Reference Consortium and the merged UK10K/1000 Genomes phase 3 reference panel using the IMPUTE4 program.¹⁷

There were 487,411 individuals with available genotype data. A total of 109,929 non-White or related White individuals (22.6%) were excluded from the study. The analyses in this study were conducted through the UK BioBank Resource Project 18120 (awarded to author E.K.S.).

Michigan Genomics Initiative

The Michigan Genomics Initiative (MGI) is an ongoing University of Michigan (single-center) biorepository (2013-current) that provides approved investigators access to genotype and phenotype data. Informed consent and genotyping were completed through the MGI. Clinical data, including demographics, medical diagnoses, and medication use, were extracted from the electronic medical record. DNA was genotyped using the Illumina Human Core Exome Array. Genotypes were imputed from the Haplotype Reference Consortium r1.1 using software Minimac4 (v1.0.0) from the Michigan Imputation Server. Poorly imputed variants with R2 <0.3 and/or minor allele frequency (MAF) <0.01% were filtered, resulting in more than 31 million imputed variants available after quality control and filtering.

There were 56,984 individuals with available genotype data. A total of 5496 non-White or related White individuals (9.8%) were excluded from the study. This study was conducted with approval from the University of Michigan Institutional Review Board (HUM00159951).

IBD Risk Variant Selection

First, all established and novel SNPs from the largest meta-analysis of IBD susceptibility to date were included.¹⁸ This study cohort included 42,992 cases and 53,536 controls. Because this study included a diverse population and our analysis was restricted to European ancestry only, we chose only those SNPs that associated with IBD susceptibility in the European cohort (n = 232). We concurrently selected SNPs from a recent study in which authors fine mapped IBD susceptibility loci to single-variant resolution.¹⁹ Specifically, the authors identified 18 single causal variants with greater than 95% certainty and 27 variants with greater than 50% certainty (n = 45). Therefore, a total of 277 SNPs were selected for inclusion in the study.

Among these 277 SNPs, we identified independent loci using a distance-based criteria (>1 mega base). For groups of variants located within 1 mega base of each other, a ranking algorithm was applied to preferentially select the representative SNP from the region. Highest priority group was given to those variants which were mapped to single-variant resolution in the fine mapping study. Second priority was given to those SNPs that were identified as lead variants where signals mapped to 2-50 variants, also in the fine mapping study. Final priority was based on the reported P value from the meta-analysis data. A total of 188 variants remained for subsequent analyses.

For the UKBB, all 188 SNPs were available for all samples included in the study. For the MGI cohort, only 1 SNP (rs7711427) was not available. The additional 187 SNPs were available for all samples.

Outcomes of Interest and Covariates

UKBB

The primary outcome of interest was skin cancer. Skin cancers were grouped into NMSC and MSC. Nonmelanoma skin cancer was defined using the ICD-10 code, “C44 Other malignant neoplasms of skin.” Melanoma skin cancer was defined using the ICD-10 code, “C43 Malignant melanoma of skin.” Additional covariates of interest included age, gender, body mass index (BMI),²⁰ and thiopurine or anti-TNF use.^7,9,10,21 Body mass index was calculated as the mean of all recorded BMIs in the data set. All BMI values <10 or >100 were excluded from analyses. Medication data was obtained according to UKBB protocol and reflects patients who reported taking the medication at the index visit with a minority having a repeat assessment. Medications of interest included thiopurines (ie, 6-mercaptopurine, azathioprine) and anti-TNFs (regardless of indication). Anti-TNFs were limited to the use of adalimumab, as no medication data were available for infliximab, certolizumab, or golimumab. There was also no medication data available for newer IBD therapies such as vedolizumab, ustekinumab, and tofacitinib.

MGI

The primary outcome of interest was skin cancer. Skin cancers were grouped into NMSC and MSC. Nonmelanoma skin cancer was defined using the ICD-9 code, “173.x” and the ICD-10 code “C44.x.” Melanoma skin cancer was defined using the ICD-9 code, “172.x” and the ICD-10 code “C43.x.” Additional covariates included age, gender, BMI, and medication use. Body mass index was classified as the most recent measurement at the time of data extraction. All BMI values <10 or >100 were excluded from analyses. Medication data were defined as “ever vs never exposed” and limited to prescriptions recorded in the institutional medical record. Medications of interest included thiopurines (ie, 6-mercaptopurine, azathioprine) and anti-TNFs (infliximab, adalimumab, certolizumab, and/or golimumab).

Statistical Analyses

Identification of clinical risk factors

Multivariable logistic regression was performed to evaluate for clinical predictors of NMSC and MSC in the discovery cohort. The model included age (years), gender (reference category, female), BMI, thiopurine use, and anti-TNF use. Variables that had a P value < .05 were considered significant and tested for replication in the validation cohort, again using multivariable logistic regression controlling for age, gender, BMI, thiopurine use, and anti-TNF.

Identification of genomic risk variants

Multivariable logistic regression was performed to evaluate for genomic predictors of NMSC and MSC in the discovery cohort. Each SNP (n = 188) was tested in an independent model, controlling for age (years), gender (reference category, female), BMI, population stratification (genetic principal components 1-10), thiopurine use, and anti-TNF use. Matched controls were randomly selected to make a balanced case:control ratio (3 times the case number). Variants that had a P value < .05 were considered to have medication-independent associations with skin cancer.

These variants were then tested for replication in the validation cohort. Multivariable logistic regression model was performed, controlling for age, gender, BMI, population stratification (genetic principal components 1-10), and medication use (thiopurine, anti-TNF). Variants that had a P value < .05 and a concordant direction of effect across both cohorts were considered to have replicated. To further confirm findings, a meta-analysis including both cohorts (UKBB and MGI) was completed using the program METAL.²² Variants with a P value < .05 were considered significant.

The direction of effect was aligned with the Crohn’s disease risk-increasing allele.¹⁸ All analyses were performed using R, version 3.6.0.²³

Effect modification

Single nucleotide polymorphisms that replicated in the validation cohort were evaluated for effect modification by immune suppressive therapy. We restricted interaction analyses to thiopurines/genotype and NMSC risk, given the low absolute number of medication exposures in other subgroups. A multivariable logistic regression model including main effects (SNP, thiopurine exposure) and their interaction terms, controlling for age, gender, and BMI, was completed. Interaction terms with a P value < .05 were considered significant. All analyses were again performed using R, version 3.6.0.²³

Results

Study Cohorts

The discovery cohort (UKBB) included 10,247 cases of NMSC, with 30,741 randomly selected controls (Table 1). The mean age was 71 years in the NMSC group and 67 years in the control group (P < 1 × 10^-324). Males were more common in the NMSC group (53%) than in the control group (46%, P = 1.03 × 10^-34). There were 165 individuals with IBD in the NMSC group (1.6%) compared with 322 individuals with IBD in the control group (1%, P = 6.78 × 10^-6). Thiopurine medications were recorded in 65 NMSC cases (0.6%) and 58 controls (0.2%, P = 1.94 × 10^-12). Anti-TNF medications use was recorded in 10 NMSC cases (0.1%) and 13 controls (<0.1%, P = .07). There were 1883 cases of MSC, with 5649 randomly selected controls (Table 2). The mean age was 69 years for the MSC group and 67 years for control group (P = 2.41 × 10^-28). Males were more common in the control group (47%) compared with the MSC group (45%, P = .07). The frequency of IBD was not statistically significant different across groups (1.1% MSC vs 1.5% controls, P = .21). Thiopurine exposure was recorded in 5 (0.3%) cases of MSC and 24 (0.4%, P = .45) controls. Anti-TNFs were recorded in 0 cases of MSC and 2 (<0.1%, P = 1) controls.

Table 1.

Demographic and clinical characteristics of the UKBB and MGI nonmelanoma skin cancer study cohorts.

Discovery Cohort	No NMSC	NMSC	P
Male	14,219/30,741 (46)%	5458/10,247 (53)%	1.03E-34
Age	66.77 (7.94)	71.36 (6.38)	P < 1 × 10-324
BMI	27.42 (4.76)	27.14 (4.48)	5.35E-05
IBD	322/30,741 (1.0)%	165/10,247 (1.6)%	6.78E-06
Thiopurine use	58/30,741 (0.2)%	65/10,247 (0.6)%	1.94E-12
Anti-TNF use	13/30,741 (<0.1)%	10/10,247 (<0.1)%	0.070872326
Validation Cohort	No NMSC	NMSC	P
Male	20,168/44,071 (46)%	4122/7334 (56)%	1.12E-61
Age	55.61 (16.24)	68.92 (12.36)	P < 1 × 10-324
BMI	30.11 (7.25)	28.77 (6.31)	3.17E-43
IBD	1113/44,071 (2.5)%	178/7334 (2.4)%	0.65
Thiopurine use	668/44,071 (1.5)%	113/7334 (1.5)%	0.911831936
Anti-TNF use	661/44,071 (1.5)%	101/7334 (1.4)%	0.451469688

Data are reported as mean (+/- SD) or percentage (n/N, %).

Table 2.

Demographic and clinical characteristics of the UKBB and MGI melanoma skin cancer study cohorts.

Discovery Cohort	No MSC	MSC	P
Male	2671/5649 (47)%	845/1883 (45)%	0.073968424
Age	67.01 (7.88)	69.26 (7.48)	2.41E-28
BMI	27.50 (4.79)	27.43 (4.47)	0.901896776
IBD	87/5649 (1.5)%	21/1883 (1.1)%	0.218292867
Thiopurine use	24/5649 (0.4)%	5/1883 (0.3)%	0.452094526
Anti-TNF use	2/5649 (<0.1)%	0/1883 (0)%	1
Validation Cohort	No MSC	MSC	P
Male	22,383/48,101 (47)%	1907/3304 (58)%	1.61E-35
Age	56.90 (16.40)	66.25 (14.02)	2.57E-216
BMI	29.93 (7.18)	29.82 (6.46)	0.511065639
IBD	1223/48,101 (2.5%)	68/3304 (2.1%)	0.10
Thiopurine use	770/48,101 (1.6)%	11/3304 (0.3)%	1.27E-08
Anti-TNF use	718/48,101 (1.5)%	44/3304 (1.3)%	0.505255304

Data are reported as mean (+/- SD) or percentage (n/N, %).

The validation cohort (MGI) included 7334 cases of NMSC, with 44,071 controls (Table 1). Males were more common in the NMSC group (56%) than controls (46%, P = 1.12 × 10^-61). There were 178 individuals with IBD in the NMSC group (2.4%) compared with 1113 individuals with IBD in the control group (2.5%, P = .65). Thiopurine use was recorded in 113 NMSC cases (1.5%) and 668 controls (1.5%, P = .91). Anti-TNF use was recorded in 101 NMSC cases (1.4%) and 661 controls (1.5%, P = .45). There were 3304 cases of MSC, with 48,101 controls (Table 2). Males were more common in the MSC group (58%) compared with controls (47%, P = 1.61 × 10^-35). Patients in the MSC group were also older (66 years) compared with the control group (57 years, P = 2.57 × 10^-216). There were 68 individuals with IBD in the MSC group (2.1%) compared with 1223 individuals with IBD in the control group (2.5%, P = .10). Thiopurine use occurred in 11 cases of MSC (0.3%) and 770 (1.6%, P = 1.27 × 10^-8) controls. Anti-TNFs were recorded in 44 cases of MSC (1.3%) and 718 controls (1.5%, P = .50).

Clinical Predictors of Skin Cancer

In the discovery cohort, age (OR, 1.09; 95% CI, 1.09-1.10; P < 2 × 10^-16), gender (OR, 1.27; 95% CI, 1.21-1.33; P < 2 × 10^-16), BMI (OR, 0.98; 95% CI, 0.97-0.98; P < 2 × 10^-16), thiopurine use (OR, 3.69; 95% CI, 2.54-5.38; P = 7.27 × 10^-12), and anti-TNF use (OR, 2.54; 95% CI, 1.05-6.03; P = .03) were all significant predictors of NMSC. In the validation cohort, age (OR, 1.06; 95% CI, 1.06-1.06; P < 2 × 10^-16), gender (OR, 1.26; 95% CI, 1.19-1.33; P < 2 × 10^-16), BMI (OR, 0.97; 95% CI, 0.97-0.98; P < 2 × 10^-16), and thiopurine use (OR, 1.37; 95% CI, 1.09-1.71; P = .005) were significant predictors of NMSC. Anti-TNF use was not a significant predictor of NMSC (P = .10), however. Inclusion of an interaction term between thiopurine use and anti-TNF use in the model revealed a significant association between the 2 variables in predicting risk of NMSC (P = .0002). With this interaction term in the model, there was a significant association between anti-TNF use and NMSC (OR, 1.56; 95% CI, 1.20-2.01).

In the discovery cohort, age (OR, 1.04; 95% CI, 1.03-1.05; P < 2 × 10^-16) and gender (OR, .88; 95% CI, 0.79-0.98; P = .02) were significant predictors of MSC. Body mass index (P = .41), thiopurine use (P = .43), and anti-TNF use (P = .94) were not significant predictors. In the validation cohort, age (OR, 1.04; 95% CI, 1.04-1.04; P < 2 × 10^-16), gender (OR, 1.37; 95% CI, 1.28-1.47; P < 2 × 10^-16), thiopurine use (OR, 0.22; 95% CI, 0.11-0.39; P = 1.2 × 10^-6), and anti-TNF use (OR, 1.54; 95% CI, 1.10-2.09; P = .009) use were all significant predictors of MSC. Body mass index was not a significant predictor of MSC (P = .70), however. Inclusion of an interaction term between anti-TNF use and thiopurine use revealed no significant interaction (P = .38).

Genomic Predictors of Skin Cancer

In the discovery cohort, 29 variants were associated with the risk of NMSC, all of which were tested for replication in the validation cohort (Supplementary Table 1). Of the 29 variants identified in the discovery cohort, 5 replicated at a P value < .05 with concordant directions of effect: rs7773324-A (DUSP22; IRF4: OR, 1.14; 95% CI, 1.09-1.19; P = 5.51e-08), rs2476601-G (PTPN22: OR, 1.13; 95% CI, 1.06-1.21; P = .0001), rs72810983-A (CPEB4: OR, 1.07; 95% CI, 1.02-1.11; P = .003), rs1847472-C (BACH2: OR, 1.05; 95% CI, 1.01-1.09; P = .02), and rs6088765-G (PROCR; MMP24: OR, 0.95; 95% CI, 0.92-0.99; P = .008; Table 3). Variants in DUSP2/IRF4, PTPN22, CPEB4, and BACH2 were associated with an increased risk of NMSC, whereas the variant in PROCR/MMP24 was associated with a decreased risk of NMSC.

Table 3.

Validated genomic predictors of nonmelanoma skin cancer.

Variant	Chromosome: Position	Effect Allele	Other Allele	Gene	UKBB Effect Allele Frequency	UKBB OR (95% CI)	UKBB P	MGI Effect Allele Frequency	MGI OR (95% CI)	MGI P
rs7773324	6:382559	A	G	DUSP22; IRF4 (inter)	0.63	1.13 (1.09-1.17)	4.90e-12	0.58	1.14 (1.09-1.19)	5.51e-8
rs1847472	6:90973159	C	A	BACH2 (i)	0.66	1.07 (1.03-1.11)	1.54e-4	0.66	1.05 (1.01-1.09)	0.02
rs2476601	1:114377568	G	A	PTPN22 (e)	0.90	1.10 (1.05-1.17)	3.94e-4	0.90	1.13 (1.06-1.21)	1.23e-4
rs72810983	5:173318254	A	G	CPEB4 (i)	0.71	1.05 (1.01,-1.08)	0.01	0.70	1.07 (1.02-1.11)	0.003
rs6088765	20:33799280	G	T	PROCR; MMP24 (inter)	0.43	0.96 (0.93-0.99)	0.009	0.44	0.95 (0.92-0.99)	0.008

Study cohorts: discovery cohort (UKBB) and validation cohort (MGI). Effect allele: Crohn’s disease risk-increasing allele. Gene annotations are (e), exonic variant; (i), intronic variant; (inter), intergenic variant. Controlling for age, gender, BMI, medication use, and population stratification.

Twelve variants were found to be significantly associated with MSC in the discovery cohort (Supplementary Table 2). Of the 12 variants in the discovery cohort, 4 were associated with risk of MSC and had concordant directions of effect: rs61839660-T (IL2RA: OR, 1.11; 95% CI, 1.02-1.21; P = .015), rs6088765-G (PROCR; MMP24: OR, 0.92; 95% CI, 0.87-0.97; P = .001), rs17391694-C (GIPC2; MGC27382: OR, 0.92; 95% CI, 0.86-0.997; P = .04), and rs1728785-C (ZFP90: OR, 0.94; 95% CI, 0.88-0.99; P = .03; Table 4). The variant in IL2RA was associated with an increased risk of MSC, whereas variants in PROCR/MMP24, GIPC2/MGC27382, and ZFP90 were associated with a decreased risk of MSC.

Table 4.

Validated genomic predictors of melanoma skin cancer.

Variant	Chromosome: Position	Effect Allele	Other Allele	Gene	UKBB Effect Allele Frequency	UKBB OR (95% CI)	UKBB P	MGI Effect Allele Frequency	MGI OR (95% CI)	MGI P
rs6088765	20:33799280	G	T	PROCR; MMP24 (inter)	0.43	0.87 (0.80-0.94)	2.19e-4	0.44	0.92(0.87-0.97)	0.001
rs61839660	10:6094697	T	C	IL2RA (i)	0.09	1.17 (1.03-1.32)	0.01	0.09	1.11(1.02-1.21)	0.015
rs17391694	1:78623626	C	T	GIPC2; MGC27382 (inter)	0.87	0.88 (0.79-0.98)	0.015	0.88	0.92 (0.86-0.997)	0.04
rs1728785	16: 68591230	C	A	ZFP90 (i)	0.77	0.91 (0.83-0.99)	0.03	0.78	0.94 (0.88-0.99)	0.03

Study cohorts: discovery cohort (UKBB) and validation cohort (MGI). Effect allele: Crohn’s disease risk-increasing allele. Gene annotations are (e), exonic variant; (i), intronic variant; (inter), intergenic variant. Controlling for age, gender, BMI, medication use, and population stratification.

In the meta-analysis, all significant variants remained significant. For NMSC, rs7773324 (DUSP22; IRF4, P = 1.45e-18), rs2476601 (PTPN22, P = 2.09e-07), rs72810983 (CPEB4, P = .0001), rs1847472 (BACH2, P = 1.40e-05), and rs6088765 (PROCR, MMP24, P = .0002) were significant (Supplementary Table 3). For MSC, rs61839660 (IL2RA, P = 6e-04), rs6088765 (PROCR, MMP24, P = 1.79e-06), rs17391694 (GIPC2, MGC27382, P = 0.002), and rs1728785 (ZFP90, P = .003) were again significant (Supplementary Table 4).

Gene-medication Interactions

We next evaluated for interaction effects between immune suppressive medication use and replicated variants. We restricted analyses to NMSC risk and thiopurine-genotype interactions, given the small sample sizes in other groups. The variant rs7773324 was the only genotype that interacted with thiopurines to alter the risk of NMSC in the discovery cohort (P = .05; Table 5). However, this finding was not replicated in the validation cohort (P = .25). Therefore, there were no reproducible genotype-thiopurine interactions identified.

Table 5.

Genotype-thiopurine interactions effect and significance in the risk of NMSC.

Interaction Term	UKBB OR	UKBB 95% CI	UKBB P	MGI OR	MGI 95% CI	MGI P
rs72810983-A	0.92	(0.50-1.68)	0.78	1.15	(0.82-1.62)	0.41
rs7773324-A	0.56	(0.31-1)	0.05	1.25	(0.85-1.84)	0.25
rs6088765-G	0.96	(0.54-1.72)	0.90	1.14	(0.83-1.55)	0.42
rs2476601-G	1.09	(0.49-2.4)	0.84	0.55	(0.34-0.88)	0.01
rs1847472-C	0.999	(0.57-1.76)	0.997	0.88	(0.64-1.22)	0.44

Interaction terms refer to the odds ratio associated with thiopurine use multiplied by dosage of the risk allele.

Discussion

Epidemiological studies have demonstrated an increased risk of skin cancer with immune suppressive therapies.^7,9,10 However, the risk of melanoma has also been associated with IBD independent of medication use, suggesting that the relationship between IBD and skin cancer may be more complex than previously thought.⁵ In this study, we sought to determine whether IBD and skin cancer have shared genetic determinants of susceptibility. To answer this question, we performed a retrospective case-control candidate gene study testing known IBD susceptibility variants in 2 well-established genomic biobanks with linkable data on skin cancer diagnoses. The results of this study highlight 5 IBD susceptibility SNPs that associate with the risk of IBD and NMSC, in addition to 4 susceptibility variants that associate with the risk of IBD and MSC.

The variants that were associated with the risk of NMSC annotated to the genes DUSP22/IRF4, PTPN22, CPEB4, BACH2, and PROCR/MMP24. Of the identified variants, the most significant association was observed with rs7773324-A, which increased the risk of NMSC by 14%. The rs7773324 locus is located between DUSP22 and IRF4, with closer proximity to the latter. The IRF4 gene encodes interferon regulatory factor 4, which is a transcription factor expressed in a myriad of immune cells and is involved in regulating interferon responses to viral infections.²⁴ Interestingly, rs12203592-T in IRF4 has been identified on multiple genome-wide association studies (GWASs) to be a strong predictor of squamous cell carcinoma. The first genome-wide association study was published in 2016 and used the Kaiser Permanente cohort. In this study, the T allele of rs12203592 was associated with a 56% increased risk of squamous cell carcinoma (OR, 1.56; 95% CI, 1.49-1.62).²⁵ The second GWAS, utilizing cohorts from 23andme and the Nurses’ Health Study/Health Professionals Follow-Up Study, replicated this finding at a similar magnitude of effect (OR, 1.62; P = 2.9 × 10^-111).²⁶ The IRF4 rs12203592 T allele has also been found to be associated with an increased risk of invasive squamous cell carcinoma (OR, 1.71; 95% CI, 1.63-1.80) compared with in situ squamous cell carcinoma, highlighting the role of this gene not only in disease susceptibility but also disease behavior.²⁷

The functional effects of IRF4 in the skin and intestinal tissue have been investigated in previous studies. In the intestine, IRF4 expression is increased in patients with IBD and mediates Th17-dependent intestinal inflammation.²⁸ In epidermal skin, the intronic IRF4 variant rs12203592 was found to influence gene expression through interaction with the promoter via a chromatin loop.²⁹ The T allele, which is associated with NMSC risk, led to decreased expression of IRF4. The variable expression of interferon regulatory factor across tissue and disease type may reflect the diverse roles of IRF4 in the immune response and would benefit from further investigation.

Interestingly, BACH2 was also found to associate with NMSC. The gene BACH2 encodes a transcription factor, which is located in T and B cells³⁰ and can directly influence the binding of IRF4 to DNA.³¹ The protein BACH2 has an important role in maintaining T-cell homeostasis by preventing T-cell subset differentiation and promoting T regulatory cell expansion.^30,31 In our study, the BACH2 variant (effect allele, C) increased the risk of NMSC by 5% to 7%. Comparatively, the C allele increases risk of Crohn’s disease by approximately 11%.¹⁸ The rs1847472 locus in BACH2 has also been linked with risk of postoperative Crohn’s disease recurrence (OR, 1.54; 95% CI, 1.00-2.36).³² The findings of both IRF4 and BACH2 as risk loci in IBD and NMSC highlight the importance of further investigation into this signaling network, which may represent an important target for treatment development.

The variant rs2476601-G, located in the gene PTPN22, was also associated with an increased risk of NMSC. The gene PTPN22 encodes protein tyrosine phosphatase nonreceptor type 22, which regulates T-cell activation.³³ The missense mutation rs2476601 (1858 C>T) substitutes a tryptophan for arginine in the first proline-rich C terminal domain.³³ For ease of reading or referring to our findings, the corresponding base pairs of this mutation will be used going forward (ie, G>A). The rs2476601-A allele is associated with an increased risk of several autoimmune diseases including but not limited to rheumatoid arthritis, type 1 diabetes mellitus, and psoriatic arthritis.³³ Paradoxically, the rs2476601-A allele is associated with a reduced risk of Crohn’s disease (OR, 0.81; 95% CI, 0.75-0.89).³⁴ The reason behind these discrepant effects remains unclear. In our study, we found that the G allele of rs2476601, which is associated with increased risk of Crohn’s disease,¹⁸ also increased the risk of NMSC. We found no literature describing an association between PTPN22 and NMSC; therefore, this is a novel finding.

The rs72810983 A allele, located in CPEB4 (cytoplasmic polyadenylation element-binding protein 4), was also associated with an increased risk of NMSC. The protein CPEB4 is involved in adenylation/de-adenylation of the poly A tail and has been implicated in the risk of numerous cancer phenotypes such as colorectal cancer,³⁵ melanoma,³⁶ and within microbial diversity of the intestine.³⁷ Interestingly, we did not see an association between CPEB4 genotype and MSC risk in our study.

In MSC, the rs61839660 variant (effect allele, T) in IL2RA was found to increase risk by 11% to 20%. The IL2RA gene encodes for the alpha subunit of the interleukin (IL)-2 receptor, which mediates IL-2 signaling. Interleukin-2 is responsible for T cell homeostasis with effects on both regulatory and effector T cell populations.³⁸ Interestingly, high dose IL-2 therapy was found to have antitumor effects in animal models,³⁹ which led to clinical trials studying the efficacy of this therapy for melanoma. Of 270 patients exposed to high dose IL-2 therapy, response was observed in 43 patients (16%; 95% CI, 12% to 21%).⁴⁰ High dose IL-2 therapy was therefore approved for the treatment of melanoma in the 1990s, but its use has been limited by side effects and newer, more effective immunotherapies.⁴¹ To our knowledge, a genomic association between IL-2 receptor genes and MSC has not previously been demonstrated. In a recent meta-analysis with 30,134 cases of melanoma and 375,188 controls, an association between melanoma and IL2RA was not described.⁴² Given the established role of IL-2 in the treatment of melanoma, the finding in this study warrants further investigation. In IBD, rs61839660-T is associated with an increased risk of Crohn’s disease (OR, 1.28; posterior probability 0.999) but not ulcerative colitis.¹⁹

Of the 5 Crohn’s disease risk-increasing alleles associated with NMSC, 1 was associated with a reduced risk of NMSC: rs6088765 (PROCR; MMP24). Similarly, of the 4 Crohn’s disease risk-increasing alleles associated with MSC, 3 were associated with a reduced risk of MSC: rs6088765 (PROCR; MMP24), rs17391694 (GIPC2; MGC27382), and rs1728785 (ZFP90). The rs6088765 variant (effect allele, G), which decreased risk of both MSC and NMSC, is located between the genes PROCR and MMP24. The G allele is associated with a small increase in risk of Crohn’s disease (<1%).¹⁸ An association between MSC and PROCR (P = 7.43 × 10^-10) has been demonstrated using expression quantitative trait (eQTL)-based analyses in melanocytes.⁴² In addition, eQTL analyses with other skin data sets (skin unexposed, skin not sun-exposed, and transformed skin fibroblasts) highlighted associations with both PROCR (not sun-exposed, P = 9.02 × 10^-21) and MMP24 (not sun-exposed, P = 2.82 × 10^-7; sun-exposed, 2.31 × 10^-15).⁴² The gene PROCR is also referred to as EPCR or endothelial cell protein C receptor. Endothelial cell protein receptor C is expressed on keratinocytes,⁴³ innate immune cells,⁴⁴ and intestinal epithelial cells.⁴⁵ The substrate of endothelial cell protein receptor C is activated protein C, which has anti-inflammatory and wound-healing effects.⁴³ In active IBD, expression of the endothelial cell protein receptor C has been shown to be significantly reduced on epithelial cells compared with controls.⁴⁵ Thus, the endothelial cell protein receptor C appears to play an important anti-inflammatory role in both the skin and intestine.

Congruent with prior literature, thiopurine medications increased the risk of NMSC in both cohorts and should be considered a risk factor for NMSC. Interestingly, thiopurines displayed no association with MSC in the discovery cohort but displayed a protective effect in the validation cohort. The reasons behind the protective association in the validation cohort remain unclear but, as this effect was not seen consistently across both cohorts and the sample sizes are low, no definitive conclusion can be drawn.

The strengths of this study include the use of well-established genomic biobanks, the ability to control for medication confounders including thiopurines and anti-TNF therapies, and the replication of findings in an independent cohort. Furthermore, there is an important scientific and clinical novelty to these results. Specifically, these data suggest that there are shared genetic underpinnings between IBD and skin cancer. Extraintestinal manifestations such as pyoderma gangrenosum, erythema nodosum, primary sclerosing cholangitis, etc are classically described and considered to be linked to the underlying process of inflammatory bowel disease. Comparatively, malignancies are considered to be a separate process influenced by long-standing intestinal inflammation or exposure to immune suppressive therapies. Although the medication-skin cancer association remains a clinically important and consistently validated finding, our data also suggest that there may be a shared genetic architecture across IBD and skin cancer. Furthermore, this work stresses the importance of further investigation into the pleiotropic effects of IBD risk variants across other disease states including malignancies.

There are several limitations of this study. First, we did not limit individuals studied to only those with a diagnosis of IBD. If we had chosen to implement such a selection strategy, the power to detect small to moderate sized effects, which are common in polygenic diseases, would have been significantly diminished. Furthermore, the questions posed did not depend on a diagnosis of IBD; rather, we sought to test whether established IBD variants increase the risk of skin cancer at all and whether those effects were modified by medication use. However, it would be beneficial to replicate these findings in a well-powered IBD cohort to determine whether these genetic variants associate with skin cancer risk among IBD patients and whether these genetic variants alter the overall disease onset and course. Second, immune suppressive medications could have been given for any indication and were not restricted to IBD only. We focused on controlling for any immune suppressive medication exposure, as these medications are primarily believed to drive the skin cancer risk in autoimmune disease. However, future work evaluating the impact of immune-mediated diseases may be beneficial to provide more clarity to our results. Third, participants were restricted to those of European ancestry only. This strategy was specifically chosen due to variation in allele frequencies across different ethnicities, which may confound interpretation of findings. Additional studies in non-European cohorts are needed to determine if these variants predict risk across ethnicity or are European-specific. Fourth, the validation cohort used institutional-level data. Therefore, any prescriptions that were ordered outside of the electronic medical record would not have been captured in this study. Fifth, the medications examined were restricted to thiopurines and anti-TNFs due to lack of data on newer biologic therapies in the discovery cohort and small sample sizes in the validation cohort. Further studies controlling for anti-IL-12/23, anti-integrin, and JAK inhibitor therapy would be beneficial in understanding the independence of these risk variants. Furthermore, medications in the discovery cohort (UKBB) were recorded using patient surveys and, thus, may be subject to recall bias. Sixth, there was a lack of available data on skin cancer predictors (such as fair skin, history of sunburns, and family history of skin cancer), limiting our ability to control for these important variables across both analyses. In future studies, it would be beneficial to include these factors in the analytical approach. Finally, comprehensive investigation into gene-medication interactions were restricted due small absolute numbers of medication-exposed patients. Therefore, no definitive conclusion can be drawn about genotype-medication interactions.

In summary, this study demonstrates that there are shared genetic risk variants across IBD and skin cancer. It is well-recognized that there is a link between IBD and skin cancer. However, that link has been assumed to be secondary to immune-suppressive medication use. Although medication-related risks exist, these data support the presence of independent genetic risks that increase the risk of skin cancer in a subset of IBD patients. To further investigate and validate this finding, future studies should test the significance of these genetic associations in an IBD-only cohort while controlling for additional skin cancer predictors such as fair skin, history of sunburns, and family history of skin cancer. Furthermore, additional studies investigating tissue-level gene expression of IRF4, PTPN22, CPEB4, and BACH2 in IBD patients with concurrent NMSC and IL2RA expression in IBD patients with MSC would be enlightening. If our findings are validated in replication cohorts and functional studies, the results could significantly impact clinical practice. Specifically, it would highlight the importance of precision medicine in IBD with efforts to identify those IBD patients at highest genetic risk of skin cancer and tailoring their medication plan and skin surveillance accordingly. For example, a patient with a high risk of NMSC may be best served by annual skin examinations and initiation of a nonthiopurine medication for control of IBD disease activity. As we look to a future that involves individualized decision-making, these data offer initial steps to prognosticate comorbid disease risk in our IBD patients and highlight the importance of understanding the pleiotropic effects of IBD susceptibility variants on other diseases.

Author Contributions

K.C.: study design, data analysis, results interpretation, drafting of the manuscript, and critical review of the manuscript

X.D., Y.C., and E.S.: study design, data analysis, results interpretation, critical review of the manuscript

K.C.S., A.K., V.C., and P.H.: study design, results interpretation, critical review of the manuscript

J.M.K. and J.E.G.: results interpretation, critical review of the manuscript

B.V.: genotyping, data preparation, quality control, and imputation, critical review of the manuscript

Funding

K.C.C. reports departmental funds. J.M.K. reports the mentoring award K24-AR076975.

Conflicts of Interest

J.M.K.: grants from BMS, Jannsen, and Q32 bio, advisory boards for Astrazeneca, Eli Lilly, BMS, Boehringer Ingleheim, Admirex Pharmaceuticals, ProVention Bio, and Ventus Therapeutics.

J.E.G.: Advisory Board for Eli Lilly, Almirall, Novartis, BMS, Astra-Zeneca, Boehringer Ingelheim, Sanofi, Galderma, AnaptysBio. Research Grants from Almirall, Eli Lilly, Kyowa Kirin, BMS/Celgene.

P.H. reports personal fees from Abbvie, Eli Lilly, and Pfizer outside of the submitted work. All other authors have nothing to declare.

Data Availability

The data underlying this article are available in the article and in its online supplementary material.