Background
Depression is the most common mental illness worldwide. The incidence of depression worldwide increased by 49.86% from 1990 to 2017 [
1]. As estimated by World Health Organization (WHO), depression has been affecting over 300 million people by 2015, which accounts for 4.4% of the global population [
2]. WHO predicted that depression would rank first among all the causes of burden of disease worldwide by 2030. As depression impairs both mental and physical health, it has become an important public health problem and a leading cause of disability worldwide nowadays [
3].
Depression has been reported to be associated with many physical diseases, such as cardiovascular disease [
4]. For a long time, depression had been recognized as a comorbidity of cancer, rather than a risk factor of cancer. In recent years, the causal relationships between depression and cancer risk have been widely explored in many observational studies. However, their results were controversial. Some observational studies suggested that causal relationship exists between depression and cancer risk [
5‐
8], while others did not [
9‐
11]. Meanwhile, a recent meta-analysis reported a small and positive association between depression and risk of overall cancer [
12], as well as risk of lung cancer and liver cancer, while a previous meta-analysis did not [
13]. The reasons of these controversies might be that the settings in these studies vary greatly, including the types of cancer and the controlled confounding factors. Since the inference of causal relationships in observational studies is usually confronted with the challenge of potential confounding bias and reverse causality, the association between depression and cancer remained to be elucidated. Although the best approach for causal inference is the randomized controlled trial, it is not feasible in the causal inference for depression as exposure, because depression cannot be randomized to different groups of individuals.
Mendelian randomization (MR) analysis is a promising tool for causal inference under the background of rapid development of large-scale GWAS [
14]. It utilizes the genetic variants strongly associated with exposure as instrumental variables to explore the causal relationship between exposures and outcomes. The MR analysis depends on the natural randomized assortment of genetic variants. According to the principle of mendelian inheritance, each parent randomly contributes one allele for each gene to its offspring. This process is independent of confounders. Thus, the MR analysis provides an analogue for randomized controlled trials. A genetic variable is valid in the MR analysis if it meets the following 3 assumptions: i) the genetic variants are associated with exposure; ii) the genetic variants are independent of confounders between exposure and outcomes; iii) the genetic variants only influence outcome via exposure [
15]. The last assumption is also known as the no-pleiotropy assumptions or exclusion-restriction principle, which means that the genetic variants cannot act on outcome via other alternative causal pathways that exclude exposure. A two-sample MR analysis refers to the MR analysis which included a pair of exposure and outcome from different or non-overlapping populations, and a bi-directional MR analysis tries to explore the reverse causality.
In this study, we performed the two-sample bi-directional MR with publicly available GWAS summary statistics to explore the causal relationship between depression and the risk of multiple types of cancers, including ovarian cancer, breast cancer, lung cancer, glioma, pancreatic cancer, lymphoma, colorectal cancer, thyroid cancer, bladder cancer, and kidney cancer. The selection of the types of cancer for analysis depends on the public availability of their GWAS data. The illustration of the causal relationships between depression and cancer will contribute to the prevention and treatment of these diseases.
Discussion
Depression has been reported to be associated with certain types of cancer in observational studies [
5‐
8]. We performed the two-sample bi-directional MR with publicly available GWAS summary statistics to investigate the causal relationship between the genetically predicted depression and the risk of multiple types of cancers, including ovarian cancer, breast cancer, lung cancer, glioma, lymphoma, colorectal cancer, thyroid cancer, bladder cancer, and kidney cancer. We detected suggestive evidence for the causality of the genetically predicted depression on breast cancer (OR = 1.09, 95% CI: 1.03–1.15,
P = 0.0022), but no evidence for reverse causality.
A previous observational study from the Taiwan National Health Insurance Research Database found that depression was independently associated with a 1.62-fold (95% CI: 1.12–2.34) overall increased risk of subsequent cancer during five-year follow-up [
7]. Our study reveals that depression leaded to a higher risk of developing breast cancer (OR = 1.09, 95% CI: 1.03–1.15,
P = 0.0022), and the causal estimate remained similar after correction for heterogeneity (OR = 1.09, 95% CI: 1.02–1.17,
P = 0.0176) and pleiotropy (OR = 1.10, 95% CI: 1.03–1.16,
P = 0.0072). The causal effect of depression on breast cancer estimated by weighted median method or estimated in the 3-cohort replication set (OR = 1.09, 95% CI: 1.03–1.15,
P = 0.0037) also demonstrated consistency in direction and magnitude. Moreover, their causal relationship showed correct Steiger direction orientating from depression to breast cancer (
P < 0.0001). These evidence suggest a robust causal effect of depression on breast cancer. A meta-analysis found that evidence from epidemiological studies was insufficient to support the causal effect of depression on breast cancer [
34]. For lung cancer, the causal relationship between depression and lung cancer turned negative after Steiger filtering in our study, which is consistent with the results of the Baltimore Epidemiologic Catchment Area Study [
8]. For ovarian cancer, a 19-year prospective cohort study did not find association between depression and ovarian cancer [
35], while a pooled analysis of two large prospective cohort studies, Nurses’ Health Study I and II, demonstrated that depression diagnosed prior to cancer diagnosis was associated increased risk of ovarian cancer [
5]. Our study did not identify the causal effect of depression on different types of ovarian cancer after Steiger filtering. The inconsistency between the results of MR and observational studies and the controversies among the results of observational studies might result from the confounding factors and bias in the real-world epidemiological studies. MR, an analogue for randomized controlled trials, is a more effective tool for causal inference as it is less likely to be affected by confounding [
14].
Many mechanisms of depression’s carcinogenic effect have been proposed. Depression was deemed as a type of chronic stress. The perception of stress activates the hypothalamic–pituitary–adrenal axis and induces the release of stress hormone, such as cortisol which is involved in stress-related signaling as well as the regulation of cell growth and cell cycle [
36]. Another possible pathway is the stress-induced impairment of immune function, in which the decreased cytotoxic T-cells and natural killer cells affect the immune surveillance [
37]. In ovarian cancer, depression might also affect the repairment of post-ovulatory wounds, the accumulation of which might lead to the carcinogenesis of ovary surface epithelium [
38]. In addition, depression may also result in the unhealthy behavior, including gluttony, anorexia, smoking, drinking, or lack of physical activity.
Reversely, depression is also a common comorbidity in cancer patients. Patients with cancer were reported to be at higher risk of depression [
39]. The prevalence of depression varies among different types of cancer. Depression is more likely to be associated with pancreatic, breast, lung and oropharyngeal cancers [
40]. A variety of factors were associated with increased risk of depression in patients with cancer, including poorly controlled pain, several types of medications, etc. However, the causal effect of different types of cancer on depression as well as their detailed mechanism have not been elucidated. In this MR analysis, no reverse causality was detected, but it does not prove the absence of the causal effect of cancer on depression. Among 17 types of cancer included in the MR analysis of reverse causality, all F-statistics were far more than 10, but the statistical power of the 11 types of cancer was only 5%, indicating the insufficiency of power to draw a reliable conclusion. Limited by the availability of full summary statistics, we only adopted the GWAS from a single consortium for each type of cancer respectively, rather than the pooled summary statistics from the meta-analysis. The prevalence of cancer is far less than depression, and the sample size of different types of cancer are also less than the sample size of depression. Thus, the number of adopted SNPs is extremely low for most types of cancer. 14 types of cancer adopted less than 20 strongly associated SNPs as genetic instruments for exposure in MR analysis.
There were several strengths in our study. First, as depression could not be randomly assigned as an intervention to different groups, observational study is the only feasible method for the clinical study of depression. In this study, we utilized randomly allocated genetic variants as instrumental variables, which can avoid the reverse causality and the bias introduced by cofounders. Second, we have covered up to 26 types of cancer for MR analysis, to provide a systematic investigation of the causal relationships. Third, the data source of each pair of exposure and outcome is non-overlapping, ensuring the basic requirements of the two-sample MR analysis. Fourth, the F-statistics for each exposure all exceeded 10, which indicates no weak instrument bias. Finally, in addition to the conventional fixed-effect IVW method, we used random-effect IVW, MR-Egger and MR-PRESSO method to correct for heterogeneity and horizontal pleiotropy respectively, and we also used weighted median method as a sensitivity analysis to ensure the validity of the conclusions drawn on the causality.
Some limitations of this study should also be pointed out. First, limited by the availability of the full summary statistics of depression GWAS from 23AndMe Inc, we only utilized the full summary statistics of depression GWAS from PGC and UK Biobank, so the results of our study might be underpowered. Second, as two-sample MR requires that both samples come from the same population, our study only includes participants of European ancestry, so the generalization of these results to other populations requires further studies. Third, the number of strongly associated SNPs selected in the MR analysis of reverse causality is relatively small. The null result could be due to low power and insufficient SNPs, which restrict our capability of drawing conclusions of true causal relationship. Fourth, the inclusion criteria of depression participants in PGC cohort and UK Biobank cohort differs greatly. The PGC cohort used clinical diagnosis international consensus criteria (DSM-IV, ICD-9, or ICD-10), while UK Biobank cohort used self-diagnosis. The inconsistency of inclusion criteria might affect the associations between SNPs and depression, but Howard et al. detected strong genetic correlations between them, which suggested that similar underlying genetic architecture was captured by different studies [
16]. Finally, the causal relationship might be confounded by hidden population structure [
41]. Although the populations in our study were all European ancestry, within-population structures were not considered. Concerns were also raised about the sexual disparities between the population of depression and breast cancer, as the between-sex difference in the prevalence of depression was reported in some epidemiological studies [
42]. The Major Depressive Disorder Working Group of the PGC investigated between-sex genetic heterogeneity in major depressive disorder with GWAS summary statistics from multiple cohorts including PGC, Kaiser Permanente GERA, UK Biobank, and the Danish iPSYCH studies, but did not detect convincing evidence for between-sexes genetic differences [
43]. As a complex multifactorial disease, the between-sex difference in the prevalence of depression might result from biological, psychological, environmental, or other factors [
44]. Given the controversies, a gender-stratified MR analysis might provide more confirmatory evidence. However, generalized analytic method has not been established considering the challenging and complex nature of population structure, and stratification correction might lead to overcorrection [
45]. In addition, these publicly available data did not provide sufficient information for further analysis of impact from population structure.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.