Cancer risks in patients with vitiligo: a Mendelian randomization study

Abstract

Purpose

Few studies have investigated the relationship between vitiligo and risks of various types of cancers, especially those other than skin cancer. Conventional observational studies are susceptible to potential confounders and inverse causation. With a Mendelian randomization approach, we were able to evaluate the causality between vitiligo and different cancer risks.

Methods

37 vitiligo-related single-nucleotide polymorphisms identified by the published genome-wide association studies were used as instrumental variables in our study. Summary data of individual-level genetic information were obtained from corresponding studies and cancer consortia. A total of 246,706 cases and 1,021,154 controls were included. The inverse variance-weighted method was applied to estimate the causation between vitiligo and different cancers.

Results

The results revealed that vitiligo patients were at lower risks of lung cancer [odds ratio (OR) 0.9513; 95% confidence interval (CI) 0.9174–0.9864; p = 0.0070], breast cancer (OR 0.9827; 95% CI 0.9659–0.9997; p = 0.0468), ovarian cancer (OR 0.9474; 95% CI 0.9271–0.9682; p < 0.001), melanoma (OR 0.9983; 95% CI 0.9976–0.9990; p < 0.001), non-melanoma skin cancer (OR 0.9997; 95% CI 0.9995–0.9999; p < 0.001), kidney cancer (OR 0.9998; 95% CI 0.9996–1.0000; p = 0.0212), and liver cancer (OR 0.9999; 95% CI 0.9999–1.0000; p = 0.0441), while no correlation was observed for other cancer types.

Conclusions

Vitiligo was causally associated with reduced risks of several cancers, suggesting that vitiligo-associated autoimmune process might play a role in the suppression of cancer.

Electronic supplementary material

The online version of this article (10.1007/s00432-020-03245-3) contains supplementary material, which is available to authorized users.

Introduction

Vitiligo is a kind of autoimmune disease characterized by selective destruction of melanocytes, which leads to depigmented skin (Jin et al. 2016). It affects 0.5–1% of the world’s population, making it the most common depigmenting disorder worldwide (Ezzedine et al. 2015).

The association between vitiligo and skin cancer, including both melanoma and non-melanoma skin cancer (NMSC), has been discussed in previous studies with conflicting results (Beral et al. 1983; Paradisi et al. 2014; Schallreuter et al. 2002; Sharquie et al. 2016; Teulings et al. 2013). The latest meta-analysis (Ban et al. 2018) demonstrated that patients with vitiligo were at a lower risk of NMSC, while no significant correlation was found between vitiligo and melanoma risk. In contrast, the correlation between vitiligo and the risks of other cancers is less of a concern. To date, only two retrospective cohort studies (Bae et al. 2019; Li et al. 2018) have been performed to evaluate the relationship between them. Using a sample of 12,391 patients from the National Health Insurance Research Database of Taiwan, Li et al. (2018) illustrated that standard incidence ratios (SIRs) of prostate cancer and bladder cancer increased in male patients while the SIRs of thyroid cancer, breast cancer, and bladder cancer increased in female patients. Interestingly, another study (Bae et al. 2019) from Korea revealed that vitiligo was associated with decreased risks of overall cancer, colorectal cancer, ovarian cancer, and lung cancer and an increased risk of thyroid cancer, with a sample of 101,078 cases and 202,156 controls. Both studies indicated the association between vitiligo and an increased risk of thyroid cancer in female patients. However, it is noted that most results of these two studies were inconsistent.

Due to the limitation of observational design, both studies could be influenced by potential confounding factors and reverse causation, and resulted in different conclusions. For example, it is hard to rule out the interference of the treatments for vitiligo and some important lifestyle discrepancies, such as smoking status and alcohol use. Consequently, it remains unclear if there is a true causal relationship between vitiligo itself and cancer risks.

Mendelian randomization (MR) analysis is a novel epidemiological method to evaluate the causation between an exposure and an outcome, with less susceptibility to potential confounders and reverse causation due to its use of genetic variants as instrumental variables (IVs). Based on Mendel’s second law, genetic variations are randomly distributed at conception, which are generally independent of environmental risk factors and precede risk factors and the development of diseases (Lawlor et al. 2008). With the summary data from the published genome-wide association studies (GWASs), a two-sample MR analysis is able to comprehensively evaluate the causation and increase the statistical power (Burgess et al. 2015; Pierce and Burgess 2013). Furthermore, we evaluated the risks of several site-specific cancers from respective studies and specific consortia, aiming to provide relatively adequate sample sizes and improve the statistical power, which was generally limited by the low incidence rate of site-specific cancer in previous cohort studies (Li et al. 2018). Using a two-sample MR method, we have evaluated the causal relationship between vitiligo and cancer risks.

Methods

Genetic instruments for vitiligo

In 2010, Jin et al. (2010) published a discovery of 50 susceptibility loci from European ancestry in generalized vitiligo. An updated study performed by Jin et al. (2016) in 2016 highlighted another 23 new significant risk-associated loci for vitiligo among European patients in a larger population sample GWAS. The outcomes of both studies were combined and selected at the genome-wide significance threshold of p < 5 × 10^–8 to build up a total of 53 loci as original instrumental variants sample. Using linkage disequilibrium (LD) analysis, we performed an exclusion once mutual LD shared larger p value conjugately and surpassed the limited value (R² < 0.001). To eliminate the genetical bias produced by palindrome with intermediate allele frequencies, several SNPs were further excluded using R package “TwoSampleMR” (Hemani et al. 2018). Eventually, 37 SNPs were brought into the final IV set as the genetic instruments of our study, which explained approximately 17.4% of the variation of vitiligo collectively (Jin et al. 2016). Considering the variation of sample size in individual-level participants of different cancer types, the F-statistics among our study ranged from 577.96 to 97,535.79 for different cancers, which suggested strong instruments (F > 100) for our MR analyses (Table 1).

Table 1.

Number of cancer cases and controls and statistical power in Mendelian randomization analyses of vitiligo and risks of different types of cancers

Trait	Consortium	Number of cases	Number of controls	Population	Sex	Power	F-statistics
Lung cancer	ILCCO	11,348	15,861	European	Males and females	1.00	5732.68
Lung adenocarcinoma	ILCCO	3442	14,894	European	Males and females
Squamous cell lung cancer	ILCCO	3275	15,038	European	Males and females
Ovarian cancer	OCAC	25,509	40,941	European	Females	1.00	577.96
Breast cancer	BCAC	122,977	105,974	European	Females	1.00	2444.18
ER+ breast cancer	BCAC	69,501	105,974	European	Females
ER− breast cancer	BCAC	21,468	105,974	European	Females
Malignant melanoma	Neale Lab	2677	334,482	European	Males and Females	1.00	71,024.81
Non-melanoma skin cancer	MRC-IEU	672	462,261	European	Males and females	1.00	97,378.01
Rectal cancer	MRC-IEU	328	462,605	European	Males and females	0.82	97,519.57
Prostate cancer	PRACTICAL	79,148	61,106	European	Males	0.83	29,546.03
Colon cancer	MRC-IEU	1494	461,439	European	Males and females	1.00	97,519.57
Kidney cancer	MRC-IEU	1114	461,896	European	Males and females	0.96	97,118.70
Non-Hodgkin's lymphoma	MRC-IEU	293	462,717	European	Males and females	0.05	97,118.70
Oral cancer	MRC-IEU	103	462,830	European	Males and females	0.22	97,519.57
Pancreas cancer	MRC-IEU	233	462,777	European	Males and females	0.35	97,535.79
Liver cancer	MRC-IEU	169	462,764	European	Males and females	0.13	97,519.57
Esophageal cancer	MRC-IEU	201	462,732	European	Males and females	0.58	97,519.57
Hodgkin’s lymphoma	MRC-IEU	440	462,493	European	Males and females	0.99	97,519.57

Study participants of cancers

We aimed to evaluate the potential causal relation between vitiligo and several types of cancers with commonly high morbidity rates and wide range of impact in both male and female. Therefore, the incorporation of different populations according to cancer types originated from six cancer consortia: International Lung Cancer Consortium (ILCCO) (Wang et al. 2014) for lung cancer (11,348 cases and 15,861 controls, including both adenocarcinoma and squamous cell cancer), Ovarian Cancer Association Consortium (OCAC) (Phelan et al. 2017) for ovarian cancer (25,509 cases and 40,901 controls), Breast Cancer Association Consortium (BCAC) (Michailidou et al. 2017) for breast cancer (122,977 cases and 105,974 controls, including both ER-positive and ER-negative breast cancer), Neale Lab for malignant melanoma (2677 cases and 334,482 controls), Prostate Cancer Association Group to Investigate Cancer Associated Alterations in the Genome (PRACTICAL) (Schumacher et al. 2018) for prostate cancer (79,148 cases and 61,106 controls), and MRC-IEU for multiple cancer types (672 cases of non-melanoma skin cancer, 328 cases of rectal cancer, 1 494 cases of colon cancer, 1114 cases of kidney cancer, 293 cases of non-Hodgkin's lymphoma, 440 cases of Hodgkin's lymphoma, 103 cases of oral cancer, 233 cases of pancreas cancer, 169 cases of liver cancer, 201 cases of esophageal cancer, and 462,830 controls) (Table 1).

Statistical analysis

The existence of causal interpretations is reasonable in MR estimates only if the instrumental variable assumptions of the method are valid (VanderWeele et al. 2014), which form the basis of the statistical analysis tool: (i) the IVs are strongly associated with vitiligo; (ii) the IVs affect cancers only through their effect on vitiligo, and (iii) the IVs are independent of any confounding factors.

To determine estimates of vitiligo for different cancers, we conducted a random-effects inverse variance-weighted (IVW) meta-analysis of the Wald ratio for individual SNPs. Besides, the weighted median and MR-Egger method were conducted for sensitivity analysis and testing the second assumption indirectly. The intercept of MR-Egger was obtained by means of Egger regression to access global pleiotropic effects. For testing the third assumption, we retrieved the previously published GWASs of our included SNPs to see whether they were also related to any confounder of both vitiligo and cancer. Moreover, leave-one-out analysis was conducted as well to assess whether a separate SNP could determine or bias the estimation of each MR analysis. Our MR analysis was performed in R (version 3.6.2) using the package TwoSampleMR (version 0.5.0) (Hemani et al. 2018).

Results

Power calculation

According to the approach of Brion et al. (2013), power calculations were performed ahead of MR analyses to evaluate whether our sample sizes were sufficient in detecting moderate-effect sizes on cancer outcomes. Given the previously reported odds ratio (OR) of each site-specific cancer in patients with vitiligo (Ban et al. 2018; Bae et al. 2019) and the variation of vitiligo explained by the final IV set, our study had adequate statistical power (> 80%) to detect the previously estimated causal effect size between vitiligo and most cancer types at a significance level of 0.05, including lung cancer, ovarian cancer, breast cancer, melanoma, non-melanoma skin cancer, rectal cancer, prostate cancer, colon cancer, kidney cancer, and Hodgkin’s lymphoma. However, it was also worth noting that the statistical power was not sufficient (< 80%) for non-Hodgkin’s lymphoma, oral cancer, pancreas cancer, liver cancer, and esophageal cancer (Table 1).

Association between individual SNP and cancer

Associations between each included SNP and risk of breast, ovarian, lung, skin, rectum, prostrate, colon, kidney, pancreas, liver, esophageal, oral cancer, and lymphoma, and their subtypes with data from ILCCO, OCAC, BCAC, PRACTICAL, Neale Lab, and MRC-IEU were examined. For 37 SNPs in our final IVs set, only 5 SNPs were not significantly associated with risk of any types of cancer introduced in our study, namely rs1031034, rs1043101, rs117744081, rs2247314, and rs78037977 (Supplementary Table 5).

MR estimates for multi-polymorphism scores

Lung cancer

The result of the IVW method indicated that the risk of lung cancer decreased in patients with vitiligo [odds ratio (OR) 0.9513; 95% confidence interval (CI) 0.9174–0.9864; p = 0.0070]. Similar association was also observed for squamous cell lung cancer (OR 0.9318; 95% CI 0.8893–0.9763; p = 0.0030], while no significant causal relationship between vitiligo and the risk of adenocarcinoma was found (OR 0.9589; 95% CI 0.9047–1.0165; p = 0.1586) (Table 2).

Table 2.

Mendelian randomization estimates of the associations between genetically predicted vitiligo and risks of different types of cancers

Cancer type	IVW		MR-Egger		Weighted median
	OR (95% CI)	p value	OR (95% CI)	p value	OR (95% CI)	p value
Lung cancer
Lung cancer overall	0.9513 (0.9174, 0.9864)	0.007	0.9968 (0.8882, 1.1188)	0.958	0.9631 (0.9203, 1.0079)	0.105
Lung adenocarcinoma	0.9589 (0.9047, 1.0165)	0.159	1.0076 (0.8367, 1.2134)	0.937	0.9746 (0.9076, 1.0464)	0.478
Squamous cell lung cancer	0.9318 (0.8893, 0.9763)	0.003	1.0715 (0.9276, 1.2377)	0.355	0.9500 (0.8880, 1.0163)	0.136
Breast cancer
Breast cancer overall	0.9827 (0.9659, 0.9997)	0.047	1.0146 (0.9651, 1.0668)	0.574	0.9694 (0.9517, 0.9875)	0.001
ER+ breast cancer	0.9890 (0.9706, 1.0078)	0.249	1.0364 (0.9824, 1.0933)	0.199	0.9776 (0.9569, 0.9988)	0.039
ER− breast cancer	0.9681 (0.9481, 0.9885)	0.002	0.9985 (0.9391, 1.0617)	0.963	0.9871 (0.9598, 1.0152)	0.365
Ovarian cancer	0.9474 (0.9271, 0.9682)	< 0.001	0.9356 (0.8780, 0.9970)	0.048	0.9372 (0.9088, 0.9665)	< 0.001
Skin cancer
Malignant melanoma (skin)	0.9983 (0.9976, 0.9990)	< 0.001	0.9962 (0.9943, 0.9981)	< 0.001	0.9990 (0.9985, 0.9996)	0.001
Non-melanoma skin cancer	0.9997 (0.9995, 0.9999)	< 0.001	0.9995 (0.9990, 0.9999)	0.032	0.9996 (0.9995, 0.9998)	< 0.001
Colon cancer	1.0000 (0.9998, 1.0002)	0.878	1.0000 (0.9995, 1.0006)	0.879	1.0001 (0.9998, 1.0004)	0.535
Kidney cancer	0.9998 (0.9996, 1.0000)	0.021	0.9998 (0.9993, 1.0003)	0.349	0.9998 (0.9996, 1.0000)	0.349
Liver cancer	0.9999 (0.9999, 1.0000)	0.044	0.9998 (0.9996, 1.0000)	0.044	0.9999 (0.9998, 1.0000)	0.093
Esophageal cancer	1.0000 (0.9999, 1.0000)	0.293	0.9998 (0.9996, 1.0000)	0.112	0.9999 (0.9998, 1.0000)	0.108
Oral cancer	1.0000 (0.9999, 1.0000)	0.259	1.0000 (0.9998, 1.0001)	0.751	1.0000 (0.9999, 1.0000)	0.168
Lymphoma
Hodgkin's lymphoma	1.0000 (0.9999, 1.0001)	0.748	1.0000 (0.9997, 1.0003)	0.920	1.0000 (0.9999, 1.0001)	0.937
Non-Hodgkin's lymphoma	1.0000 (0.9999, 1.0001)	0.472	1.0001 (0.9998, 1.0004)	0.471	1.0000 (0.9999, 1.0001)	0.864
Pancreas cancer	1.0000 (0.9999, 1.0001)	0.985	0.9999 (0.9996, 1.0001)	0.277	1.0000 (0.9999, 1.0001)	0.994
Prostate cancer	1.0104 (0.9903, 1.0310)	0.313	1.0198 (0.9587, 1.0849)	0.538	0.9952 (0.9742, 1.0167)	0.659
Rectal cancer	1.0000 (0.9999, 1.0001)	0.924	0.9999 (0.9996, 1.0002)	0.499	1.0000 (0.9998, 1.0001)	0.620

Bold indicates statistically significant: P < 0.05

Breast cancer

From the result of IVW method, patients with vitiligo were at a mildly lower risk of breast cancer (OR 0.9827; 95% CI 0.9659–0.9997; p = 0.0468). Additionally, a reduced risk for ER-negative breast cancer was also observed in patients with vitiligo (OR 0.9681; 95% CI 0.9481–0.9885; p = 0.0023). However, such association was not observed for the risk of ER-positive breast cancer (OR 0.9890; 95% CI 0.9706–1.0078; p = 0.2494). Therefore, the existence of a causal relationship between vitiligo and breast cancer was mainly driven by the ER-negative subtype (Table 2).

Ovarian cancer

We found a significant causal association for ovarian cancer and the result of IVW method showed that patients with vitiligo were causally associated with a decreased risk of ovarian cancer (OR 0.9474; 95% CI 0.9271–0.9682; p < 0.001). Therefore, a protective effect was unequivocal on vitiligo to the development of ovarian cancer (Table 2).

Skin cancer

Skin cancer was stratified into melanoma and NMSC in our study, since the pathogenic procedures are in sharp contrast in both carcinomas. While, according to our study, the illustration of IVW method revealed that the risks of both melanoma (OR 0.9983; 95% CI 0.9976–0.9990; p < 0.001) and NMSC (OR 0.9997; 95% CI 0.9995–0.9999; p < 0.001) decreased slightly in patients with vitiligo, which may indicate some sort of relationship between both skin cancer types (Table 2).

Other cancer types

The outcomes of IVW method illustrated that vitiligo was also causally associated with reduced risks of kidney cancer [OR 0.9998; 95% CI 0.9996–1.0000; p = 0.0212] and liver cancer [OR 0.9999; 95% CI 0.9999–1.0000; p = 0.0441], even though the effect remained extremely slight.

In other cancer types assessed in our study, no correlation was found between vitiligo and colon cancer [OR 1.0000; 95% CI 0.9998–1.0002; p = 0.8783], Hodgkin’s lymphoma [OR 1.0000; 95% CI 0.9999–1.0001; p = 0.7484], non-Hodgkin’s lymphoma [OR 1.0000; 95% CI 0.9999–1.0001; p = 0.4719], esophageal cancer [OR 1.0000; 95% CI 0.9999–1.0000; p = 0.2928], oral cancer [OR 1.0000; 95% CI 0.9999–1.0000; p = 0.2588], pancreas cancer [OR 1.0000; 95% CI 0.9999–1.0001; p = 0.9846], prostate cancer [OR 1.0104; 95% CI 0.9903–1.0310; p = 0.3130], and rectal cancer [OR 1.0000; 95% CI 0.9999–1.0001; p = 0.9241] (Table 2).

Assessment of MR assumptions

The first assumption was met, since the chosen SNPs were selected at the genome-wide significance threshold of p < 5 × 10^–8 and the F-statistics ranged from 577.96 to 97,535.79 (F > 100). Statistical tests and sensitivity analyses were carried out to evaluate the potential violation of the second assumption. Sensitivity analyses were conducted using MR-Egger regression to test for global pleiotropic effect in each type of cancer, which suggested weak evidence for the existence of horizontal pleiotropy in each MR analysis, since all the intercepts were not statistically significant (p > 0.05) (Table 3). For the third assumption, evidence in the published GWAS (Jin et al. 2016) stated that some SNPs related to vitiligo in our study were also associated with other autoimmune diseases and several melanocyte regulators, indicating that potential violation might occur in our study. Hence, caution was required in interpreting the outcomes of our study.

Table 3.

MR-Egger pleiotropy test of the associations between genetically predicted vitiligo and risks of different types of cancers

Cancer type	MR-Egger regression
	Intercept	p value
Lung cancer overall	− 0.0117	0.409
Lung adenocarcinoma	− 0.0123	0.586
Squamous cell lung cancer	− 0.0347	0.054
Breast cancer overall	− 0.0083	0.192
ER+ breast cancer	− 0.0121	0.077
ER− breast cancer	− 0.0080	0.301
Ovarian cancer	0.0033	0.682
Malignant melanoma (skin)	0.0005	0.030
Non-melanoma skin cancer	0.0001	0.296
Colon cancer	0.0000	0.914
Kidney cancer	0.0000	0.844
Liver cancer	0.0000	0.142
Esophageal cancer	0.0000	0.183
Oral cancer	0.0000	0.954
Hodgkin's lymphoma	0.0000	0.826
Non-Hodgkin's lymphoma	0.0000	0.310
Pancreas cancer	0.0000	0.246
Prostate cancer	− 0.0024	0.758
Rectal cancer	0.0000	0.493

Discussion

With a two-sample MR approach, our study demonstrated that patients with vitiligo were causally associated with reduced risks of lung cancer, breast cancer, ovarian cancer, skin cancer, kidney cancer, and liver cancer, indicating that the autoimmune process of vitiligo might help suppress the occurrence and development of cancer through immune surveillance. However, this association was not observed in other sites, which also suggests that the effect of vitiligo-induced immune surveillance to tumor cells is site-specific or that it may vary across different systems and organs. To the very best of our knowledge, this is the first MR study to evaluate the causal relationship between vitiligo and the risks of common cancers, with less susceptibility to potential confounders and inverse causation.

Most studies (Beral et al. 1983; Paradisi et al. 2014; Schallreuter et al. 2002; Sharquie et al. 2016; Teulings et al. 2013; Harrist et al. 1984) have focused on the relationship between vitiligo and the risk of skin cancer, including both melanoma and NMSC. Even so, their association remains uncertain due to the conflicting outcomes and study design. As far as melanoma was concerned, Beral et al. (1983) initially denoted that patients with vitiligo had a higher risk of melanoma. However, both Paradisi et al. (2014) and Teulings et al. (2013) revealed that vitiligo was a protective factor for melanoma. In terms of NMSC, Teulings et al. (2013) and Schallreuter et al. (2002) demonstrated that vitiligo was associated with a decreased risk of NMSC, but this relationship was not observed by Sharquie et al. (2016). Most studies (Paradisi et al. 2014; Sharquie et al. 2016; Teulings et al. 2013) were designed as cross-sectional studies, which could be easily affected by reverse causality and are not recommended for etiological research. The only cohort study (Beral et al. 1983) had a small sample size, with 287 cases and 574 controls. Thus, though the latest meta-analysis of these studies (Ban et al. 2018) concluded that vitiligo was only related to a decreased risk of NMSC rather than melanoma, it is still insufficient to draw a clear conclusion on the causality between vitiligo and the risk of skin cancer. Our study showed that patients with vitiligo were at lower risks of both melanoma and NMSC, consistent with previous two cross-sectional studies (Paradisi et al. 2014; Teulings et al. 2013), supporting the role of vitiligo-related autoimmunity in the immune surveillance on tumor cells originating from both keratinocytes and melanocytes.

Recently, two retrospective cohorts (Bae et al. 2019; Li et al. 2018) investigated the association between vitiligo and the risks of other common cancers in Korea and China, respectively. Both studies indicated an increased risk of thyroid cancer in female patients. Nonetheless, as mentioned above, the majority of their results were contradictory. In China's study, vitiligo seemed to be a risk factor for cancer, while in Korea’s study, it seemed to be a protective factor. Considering the characteristic of their observational design, potential confounding factors might bias the results. In addition, though both cohorts had adequate sample sizes, the number of cancers at each site was relatively small, which limited the statistical power to evaluate the causation between vitiligo and the risk of each site-specific cancer. Our study specifically selected corresponding consortia of each site-specific cancer, with a large sample size for some of the most common cancers, such as lung and breast cancer, providing adequate statistical power to evaluate the causality between vitiligo and the risks of most cancers. The results were consistent with Korea’s study that vitiligo was associated with decreased risks of lung cancer and ovarian cancer, but an association with colorectal cancer and thyroid cancer was not supported by our study. Furthermore, our study indicated that vitiligo was a protective factor for breast cancer, kidney cancer, and liver cancer, which has not been reported previously. Generally, the outcomes support that immune surveillance induced by autoimmunity of vitiligo may not only respond to cancer derived from skin cells, but also attack tumor cells in other organs with different effects.

Possible mechanisms have been proposed to explain the causality between vitiligo and cancer risks. The previous studies have demonstrated that on account of the abnormal expression of tyrosinase-related protein stabilizing the melanosome membrane, melanocytes from patients with vitiligo are more susceptible to oxidative stress than the individuals without (Jimbow et al. 2001). Therefore, melanocyte abnormalities may be the key inducers of the whole inflammatory cascade (Boorn et al. 2011). The activated inflammation ultimately results in a release of inflammatory mediators into the extracellular environment, with subsequent activation of innate immune cells. For instance, it has been proved that the global activation of melanocyte-specific CD8+ cytotoxic T-lymphocytes (CTLs) plays a role in progressive vitiligo and they can be isolated from depigmented skin and blood of patients with vitiligo, with the ability to kill melanocytes in vitro (Boorn et al. 2009; Lili et al. 2012). Previous studies also suggested that CD8+ CTLs-induced autoimmune process may exert immune surveillance on skin cells and CD8+ CTLs maybe participate in anti-tumor immune responses by killing tumor cells (Paradisi et al. 2014; Teulings et al. 2013; Bae et al. 2019; Harris et al. 2012). Meanwhile, an increased expression of interferon gamma (IFN-γ) is observed in affected skin in patients with vitiligo, which plays the key role in both the destruction of melanocyte in vitiligo and the promotion of anti-tumor immune responses, leading to the death of cancer cells (Bae et al. 2019; Harris 2015; Rashighi et al. 2014; Byrne et al. 2014). IFN-γ is an effector cytokine produced by activated CD8+ and CD4+ T cells in the tumor microenvironment and its role of recruiting CD8+ T cells to the skin has also been suggested (Harris et al. 2012). Furthermore, in terms of the decreased risk of breast cancer, it has been reported that human breast cancer cells can express detectable levels of the melanocyte-related protein TYRP1, which is previously considered as the signatures of melanomas or melanocytic origin. Consequently, breast cancer cells can potentially be the anti-tumor target of immune cells (Montel et al. 2009; Kawakami et al. 1994). Overall, with an increased number of CD8+ T cells and increased expression of IFN-γ, the immune surveillance and corresponding anti-tumor immune responses in vitiligo patients may be strengthened and, hence, decrease the risks of several cancers. Our results also indicated that either the effect of anti-tumor responses differed across different organs or that this effect was actually site-specific, however, studies are still needed to further investigate how vitiligo-induced immune surveillance targets each system and organ and why the effect of immune surveillance differs among different organs as evidence is lacking.

Our study has several advantages. First, it is the first MR study conducted to assess the causality between vitiligo and cancer risks. As mentioned above, MR analysis is a reliable epidemiological method to assess the causality between an exposure and an outcome, by avoidance of residual confounding factors and inverse causation. Second, vitiligo-associated SNPs were obtained from the published data of GWASs; 37 SNPs,  which could explain approximately 17.4% of the variation of vitiligo, were finally included (Jin et al. 2016). Third, several site-specific analyses were based on large-scale consortia in our study (Table 1). In contrast, the low incidence rate of certain cancers in the previous cohort studies would limit the power of their statistical analyses. For example, the consortium of breast cancer in our study consisted of 122,977 cases and 105,974 controls, which were much larger than 938 cases and 65 cases in the two previous studies (Bae et al. 2019; Li et al. 2018). Combined with adequate robustly associated SNPs, our study was able to provide valid statistical power to evaluate the causality between vitiligo and cancer risks (Table 1).

Several limitations of our study exist. First, all three MR assumptions could not be fully tested and potential violations against the assumptions might occur. The second assumption could not be evaluated directly in our study. Instead, additional sensitivity analyses were implemented, and the results showed no horizontal pleiotropic effects existed, suggesting that the second MR assumption was not violated. Additionally, according to the previously published GWAS (Jin et al. 2016), some of our included SNPs were also related to other autoimmune diseases, as well as several melanocyte regulators, indicating that the third assumption in our study might be violated. Thus, caution should be used in interpretation of the outcomes of our study. Second, the sample size of certain cancers in our study was also not sufficient. For example, the consortium of non-Hodgkin's lymphoma was comprised of only 293 patients compared to its vast number of 462,717 controls. Consequently, the statistical power of such result was not able to evaluate their causality exactly and future large-scale studies of these cancers are required (Table 1). Third, both the choice of SNPs and cancer-related consortia were restricted to European populations. Therefore, it remains uncertain whether our conclusions are applicable to other populations and regions.

Overall, our study indicated that patients with vitiligo were causally associated with decreased risks of several cancers, suggesting that the autoimmunity of vitiligo might help suppress the occurrence and development of cancer through immune surveillance. Due to the limited statistical power of certain cancers in our study, future large-scale studies of these cancers are required to confirm our results. Additionally, more studies are warranted to investigate the interaction between vitiligo and different cancers, especially those other than skin cancer.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

CI

Confidence interval

GWASs

Genome-wide association studies

ILCCO

International Lung Cancer Consortium

IVW

Inverse variance-weighted

MR

Mendelian randomization

OR

Odds ratio

SNP

Single-nucleotide polymorphism

NMSC

Non-melanoma skin cancer

SIRs

Standard incidence ratios

IVs

Instrumental variables

LD

Linkage disequilibrium

OCAC

Ovarian Cancer Association Consortium

BCAC

Breast Cancer Association Consortium

PRACTICAL

Prostate Cancer Association Group to Investigate Cancer Associated Alterations in the Genome

CTLs

Cytotoxic T-lymphocytes

IFN-γ

Interferon gamma

Author contributions

YW and WL were responsible for the concept and design of the study, interpretation of data, and drafting and writing of the article. The other authors were responsible for interpretation of data and revision of the intellectual content. All authors participated in final approval of the article and agreed to be accountable for all aspects of the work.

Availability of data and material

The data sets generated and/or analyzed during the current study are not publicly available, but are available from the corresponding author on reasonable request.

Code availability

The computer code generated and/or analyzed during the current study is not publicly available, but is available from the corresponding author on reasonable request.

Compliance with ethical standards

Conflict of interest

All authors declare no conflicts of interest.

Ethical approval

There are no patients involved in our study design, recruitment or research conduction, and thus, there is no need for ethical approval. No patient was asked to make recommendations about the interpretation or writing of the results. There are no plans to disseminate the results of the study to study participants or relevant patient communities. Thus, there is no need for informed consent in our study.