Introduction
There has been longstanding interest and debate on a possible relationship of allergic disease (e.g., asthma, hay fever, and eczema) with cancer [
1,
2]. In general, two distinct and contradictory theories have been raised. The immune surveillance hypothesis proposes that the presence of allergies reflect an enhanced ability of the immune system to detect and destroy malignant cells; thus, having an allergic disorder may decrease a person’s cancer risk. In contrast, the antigen stimulation hypothesis suggests that chronic stimulation of the immune system by allergies may lead to increased levels of random pro-oncogenic mutations, repeated tissue inflammation, damage, and repair in favor of cancer onset [
3]. Understanding the potential role of allergy in carcinogenesis is important as it may shed new light on the biological mechanisms underpinning intrinsic immunity and cancer and indicate therapeutic implications.
Epidemiologic evidence on associations of allergic disease with risk of breast and prostate cancer, the two most common hormone-driven malignancies among women and men, remains inconclusive [
4‐
8]. Methodological concerns include small sample size, lack of proper control for confounding and bias. Meta-analyses provide a valuable way to evaluate the association combining results across studies. Vojtechova et al
. aggregated data across 16 observational studies and found no evidence that allergy or asthma are associated with cancers of the breast (
ncase = 10,736; relative risk (RR) [95% confidence interval (CI)] 1.01 [0.94–1.08] for any allergy; 0.93 [0.73–1.19] for asthma) or prostate (
ncase = 7,890; 1.01 [0.87–1.17] for any allergy; 0.93 [0.76–1.15] for asthma) [
9]. Similarly, Zhu et al
. combined data from 20 observational studies and found no association of allergy or asthma with prostate cancer (
ncase = 19,450; 0.96 [0.86–1.06] for any allergy; 1.04 [0.92–1.17] for asthma) [
10]. Such findings, however, could have results due to attenuation to the null caused by measurement error or masking of a causal effect by confounders. For example, individuals with asthma or other allergic diseases might follow a different lifestyle than those without the diseases. They are likely to smoke less or refrain from certain foods to avoid irritation of diseases; they may pay more attention to environmental stimuli that could trigger their symptoms; they are less likely to engage with occupations that are exposed to high levels of chemical hazards. These confounders could possibly alter their risk of cancer onset.
Mendelian randomization (MR) overcomes the limitations of observational approaches by using genetic variants (mainly single nucleotide polymorphisms, SNPs) as instrumental variables (IVs) to evaluate the potential causal effect of an exposure on an outcome [
11]. MR is based on the fact that: (i) SNPs (genotypes) are randomly assigned at conception, mirroring the randomization process in controlled trials and limiting the effect of confounding, and (ii) SNPs always precede disease onset, precluding reverse causality. Under certain assumptions, namely that the selected IVs are associated with the exposure, but not associated with any confounder of the exposure–outcome relationship, nor associated with the outcome via pathways other than through the exposure, an unconfounded estimate of causal effect of an exposure can be obtained using the observed IV-exposure and IV-outcome associations [
12].
The rapidly increasing sample sizes of genome-wide association studies (GWAS) and the release of summary-level data, provides an unprecedented opportunity to conduct well-powered MR analyses. In this study, we conducted the first and also the largest two-sample MR analysis (with IV-exposure and IV-outcome associations from two different sets of participants) using genetic variants associated with a broad allergic disease phenotype including asthma, hay fever, or/and eczema as IVs, to appraise the causal relevance of allergic disease in cancers.
Results
Under the current sample size (Table
1), our study had 80% power to detect a causal association of a relative 10.2% change in breast cancer risk and 12.8% for prostate cancer; corresponding estimates for ER+ breast cancer, ER− breast cancer, and advanced prostate cancer were 12.2%, 18.5%, and 12.1% relative changes. We also presented power estimations for a range of proportions of phenotype variation explained by the genetic variants (e.g., from 1 to 5%).
Table 1Number of cancer cases and controls and statistical power in Mendelian randomization study of allergic disease and risk of breast and prostate cancer
Breast cancer |
Overall | 122,977 | 105,974 | 228,951 | 0.54 | 0.89/1.12 | 0.92/1.09 | 0.94/1.07 | 0.94/1.06 | 0.95/1.05 |
ER-positive | 69,501 | 95,042 | 164,543 | 0.42 | 0.87/1.15 | 0.91/1.10 | 0.92/1.09 | 0.93/1.07 | 0.94/1.07 |
ER-negative | 21,468 | 100,594 | 122,062 | 0.18 | 0.80/1.25 | 0.86/1.17 | 0.88/1.13 | 0.90/1.11 | 0.91/1.10 |
Prostate cancer |
Overall | 79,148 | 61,106 | 140,254 | 0.56 | 0.86/1.16 | 0.90/1.11 | 0.92/1.09 | 0.93/1.08 | 0.94/1.07 |
Advanced | 15,167 | 58,308 | 73,475 | 0.21 | 0.79/1.26 | 0.85/1.18 | 0.87/1.15 | 0.88/1.13 | 0.90/1.11 |
We did not find any evidence to support a causal role of allergic disease in overall (random-effects IVW OR [95% CI] 1.00 [0.96–1.04]), ER + (0.99 [0.95–1.04]), or ER− (1.05 [0.99–1.10]) breast cancer. Similarly, there was little evidence to support a causal role in overall prostate cancer (random-effects IVW OR [95% CI] 1.00 [0.94–1.05]) or its advanced subtype (0.97 [0.90–1.05]) (Table
2). Maximum likelihood methods provided similar results. However, we detected heterogeneity among the estimates of 132 index variants (all
pheterogeneity < 0.001 for overall breast and prostate cancer, and their subtypes), indicating the existence of SNP-specific horizontal pleiotropy (scatter plots of effect sizes on exposure
vs. outcome are shown in Supplementary Fig 1).
Table 2Mendelian randomization estimates for the causal effect of self-reported allergic disease on cancer risk using multiple allergy GWAS-identified variants
Overall breast cancer |
Inverse-variance weighted | 1.00 | (0.96–1.04) | 0.95 | < 0.001 |
Maximum likelihood | 1.00 | (0.96–1.04) | 0.95 | NA |
MR-Egger | 0.99 | (0.89–1.11) | 0.92 | 0.93 |
MR-Egger intercept | | 0.00 (− 0.005 – 0.006) | |
Weighted median | 1.03 | (0.99–1.07) | 0.17 | NA |
ER+ subset |
Inverse-variance weighted | 0.99 | (0.95–1.04) | 0.71 | < 0.001 |
aximum likelihood | 0.99 | (0.95–1.04) | 0.71 | NA |
MR-Egger | 0.99 | (0.88–1.12) | 0.88 | 1.00 |
MR-Egger intercept | | 0.00 (− 0.006 – 0.006) | |
Weighted median | 1.00 | (0.95–1.05) | 0.97 | NA |
ER− subset |
Inverse-variance weighted | 1.05 | (0.99–1.10) | 0.11 | < 0.001 |
Maximum likelihood | 1.05 | (0.99–1.10) | 0.11 | NA |
MR-Egger | 1.03 | (0.90–1.19) | 0.66 | 0.85 |
MR-Egger intercept | | 0.00 (− 0.007 – 0.008) | |
Weighted median | 1.05 | (0.97–1.13) | 0.23 | NA |
Overall prostate cancer |
Inverse-variance weighted | 1.00 | (0.94–1.05) | 0.93 | < 0.001 |
Maximum likelihood | 1.00 | (0.94–1.06) | 0.93 | NA |
MR-Egger | 1.00 | (0.87–1.16) | 0.99 | 0.96 |
MR-Egger intercept | | 0.00 (− 0.014 – 0.014) | |
Weighted median | 0.95 | (0.90–1.01) | 0.10 | NA |
Advanced prostate cancer |
Inverse-variance weighted | 0.97 | (0.90–1.05) | 0.46 | < 0.001 |
Maximum likelihood | 0.97 | (0.90–1.05) | 0.46 | NA |
MR-Egger | 0.98 | (0.81–1.20) | 0.87 | 0.89 |
MR-Egger intercept | | 0.00 (− 0.014 – 0.014) | |
Weighted median | 0.95 | (0.87–1.05) | 0.32 | NA |
Estimates from MR-Egger and weighted median approach did not provide any evidence of an effect of allergic disease on breast or prostate cancer. We did not observe aggregated directional pleiotropy using MR-Egger (
ppleiotropy for overall breast cancer, ER+ and ER− : 0.93, 1.00 and 0.85;
ppleiotropy for overall prostate cancer and its advanced subtype: 0.96 and 0.89. Intercepts and 95% CIs are shown in Table
2). We further examined pleiotropy using MR-PRESSO. Although there appeared to be a few significant outliers in overall breast cancer and ER+ subset based on global test (none of the outliers were associated with known confounders), the outlier-corrected results were very similar to those from IVW for both breast (OR [95% CI] overall 1.02 [0.98–1.05]; ER+ 1.01 [0.97–1.05]; ER− 1.05 [0.99–1.10]) and prostate cancer (OR [95% CI] overall 1.00 [0.95–1.04]; advanced subtype 0.96 [0.90–1.04]).
The primary results did not substantially alter in our subsequent sensitivity analysis where we excluded SNPs from the HLA region (chr6: 25–35 Mb) and restricted to only known (and thus validated) SNPs (
nSNP = 76; IVW OR [95% CI] for overall breast cancer 0.98 [0.93–1.04]; ER+ 0.98 [0.92–1.04]; ER− 1.04 [0.98–1.11]; for overall prostate cancer 0.99 [0.93–1.06]; advanced subtype 0.99 [0.90–1.09])
(Table
3). Consistent findings were observed when we excluded SNPs associated with potential confounders (
nSNP = 107; IVW OR [95% CI] for overall breast cancer 0.98 [0.94–1.03]; ER+ 0.98 [0.93–1.03]; ER− 0.96 [0.90–1.02]; for overall prostate cancer 1.01 [0.95–1.08]; advanced subtype 0.99 [0.91–1.08]). Similar results were observed in the leave-one-out analysis where we iteratively removed one SNP each time and performed IVW using the remaining 131 SNPs (Supplementary Table 2). We did not find any causal effect of doctor-diagnosed asthma or other allergic disease on cancers across various MR approaches (Table
4).
Table 3Mendelian randomization estimates for the causal effect of self-reported allergic disease on cancer risk using known allergy-associated variants, as well as excluding SNPs associated with potential confounders
Overall breast cancer | | | | |
Inverse-variance weighted | 0.98 | (0.93–1.04) | 0.56 | < 0.001 | 0.98 | (0.94–1.03) | 0.41 | < 0.001 |
Maximum likelihood | 0.98 | (0.93–1.04) | 0.57 | NA | 0.98 | (0.93–1.03) | 0.42 | NA |
MR-Egger | 1.04 | (0.91–1.19) | 0.60 | 0.42 | 1.08 | (0.96–1.23) | 0.19 | 0.08 |
Weighted median | 1.04 | (0.99–1.10) | 0.10 | NA | 0.97 | (0.92–1.02) | 0.20 | NA |
ER+ subset | | | | |
Inverse-variance weighted | 0.98 | (0.92–1.04) | 0.46 | < 0.001 | 0.98 | (0.93–1.03) | 0.44 | < 0.001 |
Maximum likelihood | 0.98 | (0.92–1.04) | 0.46 | NA | 0.98 | (0.93–1.03) | 0.44 | NA |
MR-Egger | 1.06 | (0.91–1.23) | 0.47 | 0.27 | 1.11 | (0.97–1.27) | 0.14 | 0.05 |
Weighted median | 1.00 | (0.94–1.06) | 1.00 | NA | 0.99 | (0.93–1.04) | 0.61 | NA |
ER− subset | | | | |
Inverse-variance weighted | 1.04 | (0.98–1.11) | 0.23 | 0.04 | 0.96 | (0.90–1.02) | 0.16 | 0.01 |
Maximum likelihood | 1.04 | (0.98–1.11) | 0.23 | NA | 0.96 | (0.90–1.02) | 0.16 | NA |
MR-Egger | 1.07 | (0.91–1.27) | 0.40 | 0.67 | 0.95 | (0.81–1.12) | 0.54 | 0.94 |
Weighted median | 1.05 | (0.96–1.14) | 0.31 | NA | 0.93 | (0.85–1.01) | 0.07 | NA |
Overall prostate cancer | | | | |
Inverse-variance weighted | 0.99 | (0.93–1.06) | 0.79 | < 0.001 | 1.01 | (0.95–1.08) | 0.76 | < 0.001 |
Maximum likelihood | 0.99 | (0.92–1.06) | 0.79 | NA | 1.01 | (0.94–1.08) | 0.76 | NA |
MR-Egger | 0.93 | (0.78–1.11) | 0.40 | 0.43 | 1.06 | (0.88–1.26) | 0.55 | 0.60 |
Weighted median | 0.94 | (0.88–1.00) | 0.06 | NA | 0.96 | (0.90–1.02) | 0.20 | NA |
Advanced prostate cancer | | | | |
Inverse-variance weighted | 0.99 | (0.90–1.09) | 0.86 | < 0.001 | 0.99 | (0.91–1.08) | 0.81 | < 0.001 |
Maximum likelihood | 0.99 | (0.90–1.09) | 0.86 | NA | 0.99 | (0.91–1.08) | 0.81 | NA |
MR-Egger | 0.91 | (0.71–1.18) | 0.49 | 0.50 | 1.07 | (0.85–1.35) | 0.57 | 0.48 |
Weighted median | 0.95 | (0.85–1.05) | 0.31 | NA | 1.00 | (0.90–1.11) | 0.98 | NA |
Table 4Mendelian randomization estimates for the causal effect of doctor-diagnosed asthma, doctor-diagnosed allergic disease, on cancer risk using multiple GWAS-identified variants
Overall breast cancer |
Inverse-variance weighted | 1.01 | (0.97–1.05) | 0.54 | < 0.001 | 1.00 | (0.94–1.07) | 0.97 | < 0.001 |
Maximum likelihood | 1.01 | (0.97–1.06) | 0.53 | NA | 1.00 | (0.94–1.07) | 0.97 | NA |
MR-Egger | 0.99 | (0.86–1.15) | 0.93 | 0.79 | 0.90 | (0.71–1.12) | 0.33 | 0.30 |
Weighted median | 1.04 | (1.00–1.08) | 0.04 | NA | 1.10 | (1.00–1.10) | 0.03 | NA |
ER+ subset |
Inverse-variance weighted | 1.00 | (0.96–1.04) | 0.99 | < 0.001 | 1.00 | (0.93–1.07) | 1.00 | < 0.001 |
Maximum likelihood | 1.00 | (0.96–1.05) | 0.99 | NA | 1.00 | (0.93–1.07) | 1.00 | NA |
MR-Egger | 0.95 | (0.81–1.11) | 0.54 | 0.52 | 0.90 | (0.69–1.12) | 0.29 | 0.27 |
Weighted median | 0.99 | (0.95–1.04) | 0.77 | NA | 1.00 | (0.97–1.09) | 0.30 | NA |
ER− subset |
Inverse-variance weighted | 1.06 | (1.01–1.11) | 0.03 | 0.06 | 1.10 | (0.97–1.14) | 0.21 | < 0.001 |
Maximum likelihood | 1.06 | (1.01–1.12) | 0.03 | NA | 1.10 | (0.97–1.14) | 0.20 | NA |
MR-Egger | 1.15 | (0.95–1.38) | 0.15 | 0.39 | 1.00 | (0.72–1.25) | 0.69 | 0.44 |
Weighted median | 1.07 | (1.00–1.14) | 0.04 | NA | 1.10 | (1.00–1.18) | 0.06 | NA |
Overall prostate cancer |
Inverse-variance weighted | 1.03 | (0.96–1.09) | 0.43 | < 0.001 | 1.01 | (0.94–1.09) | 0.75 | < 0.001 |
Maximum likelihood | 1.03 | (0.96–1.10) | 0.42 | NA | 1.01 | (0.94–1.10) | 0.75 | NA |
MR-Egger | 0.92 | (0.73–1.17) | 0.50 | 0.35 | 1.00 | (0.75–1.34) | 0.98 | 0.95 |
Weighted median | 1.03 | (0.98–1.08) | 0.27 | NA | 1.06 | (1.00–1.13) | 0.05 | NA |
Advanced prostate cancer |
Inverse-variance weighted | 1.02 | (0.94–1.10) | 0.71 | < 0.001 | 0.99 | (0.90–1.09) | 0.83 | < 0.001 |
Maximum likelihood | 1.02 | (0.94–1.10) | 0.70 | NA | 0.99 | (0.90–1.09) | 0.83 | NA |
MR-Egger | 0.84 | (0.62–1.13) | 0.24 | 0.18 | 0.91 | (0.63–1.32) | 0.62 | 0.64 |
Weighted median | 1.00 | (0.92–1.09) | 0.98 | NA | 0.99 | (0.89–1.09) | 0.78 | NA |
Discussion
In this study, we used a strong instrumental variable based on 132 SNPs and capitalized on summary statistics of the largest meta-GWAS conducted for breast and prostate cancers in European populations. We aimed to determine whether the relationship between allergic disease and risk of two cancers was causal by using two-sample MR. In general, none of our analyses suggested a causal relationship between allergic disease, asthma, and breast or prostate cancer risk.
Our findings are in line with previous reports. Despite a few observational studies suggesting a possible association of allergic disease or asthma with breast or prostate cancer [
26-
28], such evidence has been scattered, restricted by small number of cancer cases (some studies included just a couple of hundred cases), and not supported by prospective epidemiological studies. For example, in the meta-analysis conducted by Vojtechova et al
., no association was found for any allergy (RR [95% CI] 1.01 [0.94–1.08]) or asthma (0.93 [0.73–1.19]) with breast cancer, pooling results from both retrospective (
ncase = 3,544) and prospective studies (
ncase = 7,192). There was no difference in the effect estimates when cohort studies were analyzed separately [
9]. Similarly, in the meta-analysis conducted by Zhu et al
., restricting studies to those with a cohort design (
ncase = 10,769), no association was observed for any allergy (1.06 [0.84–1.33]) or asthma (1.02 [0.91–1.15]) with prostate cancer [
10]. Consistent with these previous findings, we did not observe evidence in support for an association between allergy, asthma, and risk of breast or prostate cancer.
The null findings of allergy with breast and prostate cancer are perhaps not surprising. Carcinogenesis is a multi-stage event where different defense mechanisms exist in different stages, including but not limited to, detoxification of metabolites originated from environmental carcinogens, decomposition to reactive oxygen species, DNA repair enzymes, and natural inhibitors of proliferating initiated cell [
29]. The immune system, which acts as maybe the last line of host defense against cancer development, can be easily influenced by other existing defense mechanisms, leading to either surveillance or tolerance. In addition, the relation between allergy and cancer has been demonstrated complex and site-specific. While firm conclusions cannot be drawn for cancer at many sites, strong inverse associations have been reported for pancreatic cancer [
30], glioma [
31], and lymphoma [
32], whereas positive association has been revealed for lung cancer [
33] (although most such meta-analysis or pooled studies did not analyze prospective studies separately). Several shared biological mechanisms have been involved in the asthma–lung cancer association, including elevated levels of free radicals and reduced levels of antioxidants in the respiratory tract, persistent stimulation of cell regeneration to repair lung damage, and a heightened sensitivity to carcinogens [
6]. While the global and individual mechanisms involved in other cancer sites might be more complicated and less straightforward than in the case of lung cancer, current evidence indicates an organ specific immunological effect elicited by allergic disease.
Our study has several strengths. We took a series of steps to guarantee the validity of estimates. We selected the most significant independent SNPs identified by the largest allergic disease GWAS, so all were robustly associated with the exposure of interest. These SNPs constitute a strong instrument with an F-statistic of 32.6. Secondly, none of the instrumental variables used in our analysis were cited by the NHGRI-EBI GWAS Catalog as associated with known strong confounders of cancer risk, such as BMI, smoking, or mammographic density at α = 10–8 level (a full list of confounders for asthma–breast cancer association includes weight, diet, exercise, alcohol consumption, smoking, stress and anxiety, levels of female hormone estrogen; while a full list of confounders is less clear for asthma–prostate cancer association, it probably includes diet, obesity, smoking, and chemical exposure). Nevertheless, some variants were indeed associated with autoimmune diseases, psychiatric traits and alcohol consumption; however, sensitivity analysis excluding those potentially confounding SNPs provided similar results as the primary analysis. Finally, we employed sensitivity analyses to control for pleiotropy and obtained highly consistent results.
Our current results also provide implications and directions for future research. While there is no strong evidence in support for a putative causal relationship between allergic disease, asthma, and the risk of cancers on breast and prostate, it is likely that the presence of allergies, cytokines, inflammatory, and immune responses it brings could influence the prognosis and mortality of cancer. For example, in a prospective cohort study of 1,102,247 cancer-free individuals at baseline, significant inverse associations between a history of both asthma and hay fever and overall cancer mortality (RR [95%CI] 0.88 [0.83–0.93]) as well as colorectal cancer mortality (0.76 [0.64–0.91]) were found. A history of hay fever only was associated with a significantly lowered risk of pancreatic cancer mortality, and a history of asthma only was associated with a significantly lowered risk of leukemia mortality [
34]. Another study consisting of 475 incident pancreatic cancer cases found allergy to be associated with improved prognosis (longer survival among those with self-reported allergies than those without: 13.3 vs. 10.4 months) [
35]. Future MR studies may be conducted to focus on understanding the causal role of allergic disease in cancer prognosis or drug responses (e.g., immunotherapy), rather than incidence. In addition, allergic disease belongs to type-I hypersensitivity, which is characterized by an atypical Th2-dominated immune response to innocuous environmental agents, leading to a large amount of IgE antibody production [
36]. Most autoimmune inflammatory diseases such as systemic lupus erythematosus, rheumatoid arthritis and multiple sclerosis involve different immunological mechanisms such as the antigen–antibody complex triggered IgG involved type III hypersensitivity or cell-mediated type IV hypersensitivity. It is possible that these immune responses may be related to cancer development, and this therefore warrants further investigations.
Our study had also several caveats. Despite using the largest ever GWAS data on both the exposure and outcome, the statistical power in detecting a weak effect is poor—if the true causal effect of allergic disease on cancer was less than 5%, a magnitude that is probably of limited clinical importance—we only had power of 24% for breast cancer and 17% for prostate cancer, with our current sample size. Although we have involved clinically meaningful disease subtypes such as ER+ /− breast cancer and aggressive prostate cancer, we could not examine breast cancer based on menopause status (pre- and post-menopause; 85% of breast cancer cases in our samples are postmenopausal). In addition, for two-sample MR to be valid, the two samples have to be from the same underlying population, which is not the case for our sex-specific cancers. While our outcome data are breast cancer in female and prostate cancer in male, we could not examine the exposure trait stratified on sex; neither can we test the association of genetic instruments with other effect modifiers such as personal history of allergies, physical activity, and hormonal and lifestyle factors, due to lack of data. Violation of the exclusion restriction assumption may arise with a binary exposure in MR studies, if it is a dichotomization of a continuous risk factor (e.g., allergy could be a dichotomization of a continuous spectrum of subclinical problems), then the IVs can influence outcome via the continuous risk factor even if the binary exposure stays the same. In such cases, calculating a causal estimate can be difficult [
37]. It may be helpful to test relevant continuous variables such as severity of allergy, and immunoglobulin E levels when GWAS of such phenotypes become available.
In conclusion, our MR analysis, despite its good overall statistical power and well-designed analytical protocol, provides no evidence in support of a causal relationship between allergic disease, asthma, and the risk of breast or prostate cancer.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.