Mendelian randomization shows depression increase the risk of type 2 diabetes

doi:10.21203/rs.3.rs-2437434/v1

Type 2 diabetes (T2D) is associated with severe mental illnesses (SMIs), such as schizophrenia, bipolar disorder, and depression. However, causal relationships between SMIs and T2D remain unclear owing to potential bias in observational studies. We aimed to characterize the causal effect of SMI liability on T2D using two-sample Mendelian randomization (MR). The causality between liability to SMI and T2D was investigated using the inverse-variance weighted (IVW), MR-Egger, MR-Egger with a simulation extrapolation, weighted median, and the MR pleiotropy residual sum and outlier method. Similarly, we performed additional MR which can detect the reverse causation effect by switching exposure and outcome for T2D liability for SMI. To further consider pleiotropic effects between SMIs, multivariable MR analysis was performed after accounting for the other traits. In the univariable IVW method, depression showed a causal effect on T2D (odds ratio [OR]: 1.128, 95% confidence interval [CI]: 1.024–1.245, P = 0.014). Multinomial MR more strongly supported these results (IVW OR: 1.197, 95% CI: 1.069, 1.340, P = 0.002; MR-Egger OR: 1.198, 95% CI: 1.062, 1.349, P = 0.003). Bidirectional MR showed reverse-causality absence between depression and T2D. However, causal relationship of bipolar and schizophrenia on T2D was not detected. Careful attention is needed for patients with depression regarding T2D prevention and treatment.

Health sciences/Diseases/Psychiatric disorders/Depression

Health sciences/Diseases/Psychiatric disorders/Bipolar disorder

Health sciences/Diseases/Psychiatric disorders/Schizophrenia

Bipolar disorder

Schizophrenia

Diabetes mellitus

type 2

Mendelian randomization analysis

Depression

Causality

To date, several epidemiologic studies have suggested a link between type 2 diabetes (T2D) and severe mental illnesses (SMIs), including bipolar disorder (BPD), schizophrenia (SCZ), and depression [1]. The prevalence of T2D among individuals with BPD and SCZ has been estimated to be 8–17% and 16–25%, respectively; furthermore, some studies have reported that adults with depression have a 37% increased risk of developing T2D [2, 3, 4]. The side effects of the medications used for treating SMI, unhealthy lifestyle behaviors of patients with SMI, and hypothalamic–pituitary–adrenal axis dysregulation could contribute to the association of SMI with T2D [5, 6, 7].

In contrast, T2D also affects mental health. Some studies have reported that the risk of depression increases in people with T2D [8], while the prevalence of SCZ is higher in patients with T2D than in the general population [9]. Furthermore, T2D and prediabetes may be risk factors in patients with BPD [10], and diabetes affects mental health [11]. Furthermore, it is also well known that many diseases have downstream effects on mental health as part of a prodromal phase. However, owing to potential biases, such as confounding and reverse causation in these studies, the true causality between liability to SMI and T2D remains unclear. Moreover, the difference between the age at onset of T2D and SMI makes it extremely challenging to infer a causal relationship between them. In general, T2D affects middle-aged adults (aged > 40 years) with low heritability (10–15%), while SMI usually occurs in young adulthood (20s–30s) with high heritability (80–85% for BPD, 80% for SCZ, and 31–42% for depression) [12, 13]. Accordingly, both directions warrant investigation.

Two-sample Mendelian randomization (MR) studies, widely adopted using genetic variants, such as single nucleotide polymorphisms (SNPs) for instrumental variable (IV) analysis, should pinpoint the causal relationship between liability to SMI and T2D [14]. Anti-psychiatric treatments may cause bias, and some SNPs that are used as potential IVs may be strongly associated with side effects; if this is the case, genetically determined depression and anti-depressive efficacy may be responsible for the estimated causal effect of SMIs on T2D [15, 16, 17]. Therefore, in this study, which aimed to clarify the causal relationship between mental illnesses and T2D, it is important to clearly specify the roles on shared genes between two traits, and pleiotropy. For these reasons, in our MR analysis, the pleiotropic effects were considered from multiple aspects, and various sensitivity MR analyses were applied [18, 19, 20]. Consequently, we investigated the causality between SMI (BPD, SCZ, and depression) liability and T2D using genome-wide summary statistics in a meta-analysis of a large population of European individuals through a two-sample bidirectional MR study.

Two-sample MR analysis was conducted, and Fig. 1 displays a flowchart describing the overall procedure.

Exposure datasets: summary statistics of genetic association analyses for BPD, depression, and

SCZ

The first exposure dataset is the summary statistics of the genetic association analyses for BPD with 20,352 cases and 31,358 controls [21]. The characteristics of the data is shown in Table 1. Cases were required to meet international consensus criteria (Diagnostic and Statistical Manual of Mental Disorders [DSM-IV] and International Classification of Diseases, Ninth or Tenth Revision [ICD-9 or ICD-10]) for a BPD lifetime diagnosis. To select an appropriate instrument, “ld_clump” function of R package “ieugwasr” was used. Among the genome-wide (P < 5\(\text{×}{\text{10}}^{\text{-8}}\)) significant SNPs, some variants were removed due to LD with other variants or absence from the LD reference panel. Thus, 16 genome-wide SNPs associated with BPD liability were identified after LD pruning (distance < 10,000 kb or LD \({\text{r}}^{\text{2}}\)<0.001). The above series of processes will be referred to as "IV QC (IV quality control)" in the remainder of this article. After the removal of SNPs nominally associated with T2D (P < 0.05), 11 BPD-associated SNPs were available. The second exposure dataset is the summary statistics of the genetic association analyses for depression with 246,363 cases and 561,190 controls (Table 1) [22]. The definition of depression was different in each cohort. In the UK Biobank, the following three depression phenotypes were used: 1) self-reported help-seeking for problems with nerves, anxiety, tension, or depression, 2) self-reported depressive symptoms, and 3) depression identified from hospital admission records. In the case of 23andMe, Inc., a self-reported clinical diagnosis of depression was used. Finally, cases were required to meet various international consensus criteria (DSM-IV, ICD-9, or ICD-10) in the Psychiatric Genomics Consortium. A total of 50 were selected as IV candidates after IV QC. We checked the pleiotropic effect of those SNPs, and eight SNPs associated with T2D (P < 0.05) were eliminated. Thus, 42 SNPs were associated with depression liability. The third exposure dataset is the summary statistics of the genetic association analyses for SCZ with 33,640 cases and 43,456 controls (Table 1) [23]. From this genome-wide association study (GWAS), we exclusively considered the European population analysis to satisfy the two-sample MR assumption of having the same underlying population in both exposure and outcome GWAS. Cases with clinical diagnoses (not self-reported) of SCZ or schizoaffective disorder were included in our study. A total of 83 SNPs was associated with SCZ after IV QC. Eleven SNPs were associated with T2D (P < 0.05), and the remaining 72 SCZ liability-associated SNPs were used for our MR analyses.

Outcome datasets: summary statistics of genetic association analyses for T2D

To avoid inflated type-1 error rates and false-positive findings due to sample overlap in two-sample MR [20], we obtained summary statistics for the association of SNP with T2D from the DIAbetes Genetics Replication And Meta-analysis Consortium stage 1 meta-analyses with 26,676 cases and 132,532 controls (Table 1) [24]. T2D diagnosis was based on diagnostic fasting glucose (≥ 7 mmol/L) or hemoglobin A1c levels (≥ 6.5%), hospital discharge diagnosis, use of oral anti-diabetic medication, or self-reports. Summary statistics of genome-wide association analyses for 12.1 million SNPs were available and were considered for our MR analyses. When T2D status is considered as an exposure in reverse MR analysis, a total of 35, 39 and 43 SNPs were finally used as instrument variables for BPD, depression and SCZ, respectively.

MR analysis

Two-sample MR requires three strong basic assumptions as follows: (ⅰ) a strong association between IVs and intermediate exposure, (ⅱ) IVs independent of confounders, and (a) IVs that do not directly affect the outcome. Additionally, sensitivity MR requires the “NO Measurement Error” (NOME) assumption and InSIDE assumption (INstrument Strength Independent of Direct Effect) [20]. F-statistics provided an indication of instrument strength, and F > 10 indicated that the analysis was unlikely to suffer from weak instrument bias [25]. A degree of violation of the NOME assumption was quantified using \({\text{I}}^{\text{2}}\) statistics, while \({\text{I}}^{\text{2}}\)>90 indicated lower estimate dilution in MR analysis [26]. To detect pleiotropic outlier SNPs, we used Cochran’s Q-test in the inverse-variance weighted (IVW) method and Rucker Q’ statistics in the MR-Egger [27]. We further conducted an MR- Pleiotropy RESidual Sum and Outlier (PRESSO) test as an indicator of no violations of MR assumptions in the final instrumental variable sets [28]. Given that no weak instrument bias (F > 10) was observed and the three tests (Cochran’s Q-test, Rucker Q’ test, and MR-PRESSO test) indicated no directional pleiotropic bias, the IVW method, which is robust when all SNPs are valid instruments, was applied [20]. If pleiotropy and outlier SNPs were detected from MR-PRESSO, since IVW is not recommended, several sensitivity analyses were considered to minimize bias [28]. It shows the results of removing outliers in case of the presence of the horizontal pleiotropy effect. The other sensitivity method is the “weighted median method,” which provides valid causal estimates unless > 50% of the instruments are invalid [20]. The median is not affected by outliers; therefore, the weighted median estimate is not sensitive to a pleiotropic genetic variant. Causal effects obtained from the weighted median of the ratio estimates in genetic instruments indicate which smaller standard error receives more weight. The sensitivity MR method used for estimating the causal effect considering the pleiotropic effect is called the MR-Egger method. This method can estimate appropriate causal effects in the presence of pleiotropy effects even if all SNPs are invalid [20]. When Cochran’s Q-test is rejected or both Cochran’s Q and Rucker’s Q' tests are rejected, the MR-Egger method is recommended [20]. If the InSIDE assumption holds, then the slope of the MR-Egger regression provides a causal effect. However, when the \({I}^{2}\) statistic quantifying the strength of NOME violation for IVs is low (\({\text{I}}^{\text{2}}\)<90) for the MR-Egger method, regression dilution will occur. In cases wherein the NOME assumption was violated, the simulation extrapolation (SIMEX) method was applied to correct attenuation bias [26]. However, there is no uniformly powerful and robust MR methods, and each method have advantages for different circumstance. Therefore, according to the IV assumptions, the most recommended method is indicated in bold in the Tables 3 and 4.

To facilitate the understanding of the methods and analytical procedures, additional figures (scatter, funnel, forest, and leave-one-out sensitivity analysis plot) are shown. The associations of the variants with exposures and outcomes are shown in a scatter plot with several MR-fitted lines. The generated funnel plot shows symmetry, indicating heterogeneity due to horizontal pleiotropy. In addition, the forest plot shows individual estimate between IVs and the risk for T2D, and leave-one-out variant analysis is illustrated in the form of a forest plot. The MR results were rescaled to represent the odds of T2D per doubling of genetic liability to each exposure through multiplying the log causal estimate and 95% confidence interval (CI) by 0.693 and then exponentiating [28]. A statistical significance threshold of Bonferroni-corrected P = 1.67\(\times {10}^{-2}\) (0.05/3) was used considering bi-directional MR. All statistical analysis was performed using the R package “TwoSampleMR”.

Statistical power analyses

Observed power calculations were performed using an online tool (https://sb452.shinyapps.io/power/) [29]. Proportions of variance in the exposure explained by genetic variants (\({\text{r}}^{\text{2}}\)) were required for MR power analysis, and 0.08 (BPD), 0.03 (depression), 0.03 (SCZ), and 0.06 (T2D) were used for \({\text{r}}^{\text{2}}\) [21, 22, 23, 24].

Reverse MR analysis

We conducted a bidirectional MR to investigate the presence of reverse-causality between liability to T2D and SMI risk. Among GWAS for 12.1 million SNPs for T2D liability, a total of 35, 43, and 39 SNPs genome-wide, which were significantly associated with T2D liability, had summary GWAS results for BPD, SCZ, and depression, respectively, after LD pruning. The analysis process described in Fig. 1, which involved examining the assumption of the instruments and calculating power, was similarly applied to the bidirectional MR analysis.

Multivariable MR analysis

There is substantial evidence for partial overlap of genetic influences on SMIs [30, 31]. Therefore, the multivariable MR method would be useful to disentangle effects in situations where the three traits are highly related [32]. A total of 143 SNPs associated with at least one exposure was used for multivariable MR. To assess instrument strength and heterogeneity, the conditional F-statistic and Cochrane’s Q statistic were calculated [33]. All analysis was performed using the R package “MVMR” and “MendelianRandomization”. If Cochran's Q test P less than 0.05, we used the random-effects multivariable IVW method. In addition, to consider potential pleiotropy, an extension of the MR-Egger method (i.e., multivariable MR-Egger) were applied to identify the causal effect of BPD, depression, SCZ on T2D [34]. Multivariable MR-Egger intercept indicate pleiotropy (intercept P < 0.05) and it can also provide a corrected estimate when such pleiotropy exists.

Effect of liability to BPD on T2D risk

Eleven SNPs associated with BPD, rather than T2D, were used as IVs. All SNP to the exposure and SNP-outcome effects are presented in Supplementary Table 2. We found no evidence of weak instrument bias (F-statistic = 32.9), heterogeneity, or outlier pleiotropy (Q-test, P = 0.718; Q’-test, P = 0.821; MR-PRESSO global test, P = 0.695) (Table 2). Additionally, the MR-Egger test indicated no directional pleiotropic bias (intercept P = 0.165) or violation of the NOME assumption (\({\text{I}}^{\text{2}}\)=96.9%) (Table 2). Since all IV assumptions were satisfied, the inverse-variance weighted (IVW) method was considered the most appropriate method to provide unbiased estimates [20, 27]. There was no evidence of a causal effect of BPD liability on T2D (odds ratio [OR]: 1.004, 95% CI: 0.942, 1.070, P = 0.892). All other sensitivity analysis also has the OR with CIs crossing 1 (Table 3). In the MR for the reverse causation of T2D liability on BPD, there was no weak instrument bias (F-statistic = 35.8) or violation of the NOME assumption (\({\text{I}}^{\text{2}}=\)97.3%). No evidence of heterogeneity was found in the Q-test (P=0.684), Q’-test (P = 0.639), and MR-PRESSO global test (P = 0.598) (Table 2). The most appropriate method, IVW, showed no reverse-causal effect (OR: 1.021, 95% CI: 0.979, 1.064, P = 0.313) and all other sensitivity analysis showed similar results (Table 4). Given the sample size and \({\text{r}}^{\text{2}}\), a power of 98% and 50.8% was estimated to detect a true OR of 1.100 for BPD on T2D MR and bidirectional MR, respectively. Visualizations for bi-directional MR analysis (scatter, funnel, forest, and leave-one-out sensitivity analysis plots) are shown in Supplementary Figs. 1 and 2. In the multivariable MR controlling for depression and SCZ, weak instrument bias is not expected (conditional F-statistic=30.56), and heterogeneity was detected (Cochran's Q test P<0.05). Random-effects multivariable IVW and multivariable MR-Egger results showed that the effect of BPD liability did not have a causal effect on T2D (Table 5).

Effect of liability to depression on T2D risk

Forty-two independent SNPs associated with depression liability, rather than T2D, were used as IVs. All SNP-exposure and SNP-outcome effects are presented in Supplementary Table 3. Weak instrument assumption was guaranteed using F-statistic (37.9), and there was no measurement error of estimates from the MR study (\({\text{I}}^{\text{2}}\text{=}\)97.3%). No evidence of heterogeneity or pleiotropy was confirmed through the Q-test (P=0.808), Q’-test (P = 0.788), and MR-PRESSO global test (P = 0.733) (Table 2). Since all IV assumptions were satisfied, the robust IVW method was selected to obtain promising results (OR: 1.128, 95% CI: 1.024, 1.245, P = 0.014) (Table 3). It was also significant even when Bonferroni correction (P<0.017) controlling the family-wise error rate strictly was applied. All other sensitivity analyses showed no significant causal relationship (Table 4). In the MR for the reverse causation for T2D liability on depression, we observed that all assumptions for MR analyses were preserved (F-statistic=36.0; Q-test, P=0.714; Q’-test, P = 0.683; MR-PRESSO global test, P = 0.711; \({\text{I}}^{\text{2}}=\)97.3%) (Table 2). Therefore, the IVW method was considered the most appropriate method and no reverse-causal effect was observed (OR: 1.001, 95% CI: 0.989, 1.013, P = 0.793). All other sensitivity analyses showed no significant reverse-causal relationship (Table 4). Given the sample size and \({\text{r}}^{\text{2}}\), a power of 69% and 100% was estimated to detect a true OR of 1.100 for depression on T2D MR and bidirectional MR, respectively. Visualizations for bi-directional MR analysis are shown in Supplementary Figs. 3 and 4. In the multivariable MR controlling for BPD and SCZ, weak instrument bias is not expected (conditional F-statistic=27.33), and heterogeneity was detected (Cochran's Q test P<0.05). Random-effects multivariable IVW (OR: 1.197, 95% CI: 1.069, 1.340, P = 0.002) and multivariable MR-Egger (OR: 1.198, 95% CI: 1.062, 1.349, P=0.003) provide a corrected causal estimate when such pleiotropy exists (Table 5).

Effect of liability to SCZ on T2D risk

Seventy-two independent SNPs associated with SCZ liability, rather than T2D, were found to have strong instrument strength (F-statistic = 41.5) and no heterogeneity and outlier pleiotropy (Q-test, P = 0.319; Q’-test, P = 0.290; MR-PRESSO global test, P = 0.704) (Table 2). All SNP-exposure and SNP-outcome effects are shown in Supplementary Table 3. In addition, there was no dilution bias from violation of the NOME assumption (\({\text{I}}^{\text{2}}=\)97.2%). The IVW method was the most powerful method, and it showed no causal effect of SCZ liability on T2D (OR: 1.011, 95% CI: 0.981, 1.040, P=0.463). Results of sensitivity MR analyses showed no evidence about the SCZ liability on T2D with wide CIs (Table 3). In the MR for the reverse causation for T2D liability on SCZ, F-statistic showed no weak instrument bias (F-statistic=35.8) or violation of the NOME assumption (\({\text{I}}^{\text{2}}\text{=}\)97.2) (Table 2). The heterogeneity test showed substantial evidence of outlier pleiotropy using the Q-test (P<0.001), Q’-test (P<0.001), and MR-PRESSO global test (P < 0.001) (Table 2). Since all the three tests were rejected, the MR-PRESSO method was adopted [20] after excluding two outlier SNPs (OR: 0.999, 95% CI: 0.993, 1.006, P=0.987), suggesting no causal effect of T2D liability on SCZ. Given the sample size and \({\text{r}}^{\text{2}}\), we were 69% and 89.5% powered to detect a true OR of 1.100 for both SCZ on T2D MR and bidirectional MR. Visualizations for bi-directional MR analysis are shown in Supplementary Figs. 5 and 6. In the multivariable MR controlling for BPD and depression, weak instrument bias is not expected (conditional F-statistic=11.78), and heterogeneity was detected (Cochran's Q test P<0.05). Random-effects multivariable IVW and multivariable MR-Egger results showed that the effect of SCZ liability did not have a causal effect on T2D (Table 5).

In this study, two-sample univariable and multivariable MR results provided some evidence in support of the hypothesis that depression liability increases the risk of T2D, whereas there was no evidence to suggest that liability to BPD and SCZ are risk factors for T2D. Additionally, a bidirectional MR study found no reverse causality between SMI and T2D, which supports the hypothesis that depression is one of the causal factors influencing T2D. However, observational studies have provided contradictory and controversial findings. One systematic review demonstrated that depression is associated with a 60% increased risk of T2D, while the evidence is also compatible with the high prevalence rates of depression among individuals with T2D [35]. A large meta-analysis showed that T2D is associated with only a modestly increased risk of depression [36]. Depression is difficult to detect in older adults, which may partially explain the utter modesty of this association [37].

Our finding, which relates to the causal role of depression liability in an increased risk of T2D, could be explained by the pathophysiological mechanisms underlying the two diseases. Two major molecular mechanisms have been suggested to explain the causal pathway between them. First, the hypothalamic-pituitary-adrenal axis, a central stress response system, is commonly activated in patients with depression suffering from emotional stressors leading to a rise in the levels of glucocorticoids, primarily cortisol [38]. High cortisol level induces and aggravates insulin resistance in a vicious cycle [39]. Second, sympathetic nervous system (SNS) activity is also elevated in depression [40]. The SNS axis interacts with the hypothalamic-pituitary-adrenal axis to maintain homeostasis during stress, resulting in an increased release of cortisol and other glucocorticoids, catecholamines, growth hormone, and glucagon. Indeed, catecholamines have marked metabolic effects, particularly on glucose metabolism [41].

However, our findings are inconsistent with an observational study suggesting a causal role of liability to BPD and SCZ in the risk of T2D and that liability to T2D predicts the development of depression [3]. Such associations may have been driven by residual confounders, and several suggestive pieces of evidence can act as confounders. First, a sedentary lifestyle, demonstrated to be strongly associated with SMI, may play a role as a potential confounder [1]. A large meta-analysis of general population studies reported that sedentary behavior is independently associated with an increased risk of T2D [42]. Additionally, the side effects of medication could be another important potential confounder. A systematic review of cross-sectional and prospective studies indicated that the use of antipsychotics, antidepressants, and mood stabilizers could contribute to an increased body mass index, which is a major risk factor for T2D [43]. Furthermore, the highest prevalence of daily cigarette smoking was observed among patients with SCZ, followed by patients with BPD and those with depression, compared with the general population. The association with smoking is stronger in SCZ and BPD than in depression [44]. The evidence that nicotine addiction begins before any of these SMIs develop suggests the involvement of shared genes associated with nicotine addiction and SMI [45]. In contrast, in MR analysis, genetic variants (i.e., SNPs) used as IVs are innately random, and are assumed to be independent of confounding factors such as age, gender, and race.

MR studies on the association between SMI liability and T2D are scarce, with no studies on liability to BPD comorbid with T2D. To investigate the potential causal relationship of T2D with depression, MR analysis was performed with a large Chinese longitudinal cohort from 2011 to 2013 [46]. In their studies, effect of depression on T2D was not significant, which is inconsistent with our finding. There are multiple reasons about such inconsistency. First, there may be a racial difference between non-Hispanic whites and Chinses. Second, we considered two-stage methods and Xuan et al considered one-sample Mendelian randomizations. Both methods require several assumptions to extend the analysis results to the causality of depression on T2D, and if they are not satisfied, causality cannot be guaranteed. For instance, some of assumptions such as horizontal pleiotropy can be violated. Furthermore, the fitted values from the first-stage regression are correlated with the outcome in finite samples even, and there can be a finite-sample bias in a one-sample setting [37]. Regarding the MR studies of SCZ and T2D, two-sample MR was performed using the IVW and MR-Egger methods in European, East Asian, and trans-ancestry groups [47]. No evidence of a causal effect on T2D for SCZ was observed in any analyses, consistent with our findings; however, they did not perform a bidirectional analysis to investigate the causal effect of SCZ on T2D. Unlike epidemiological studies, previous and present MR studies could not consider the multi-episode status of the disease, which may have led to the non-causal effect of SCZ and BPD. This could be because multi‐episode (versus first‐episode) patients with SMI were more likely to have T2D than matched controls in the meta‐analysis of observational studies [1].

Our study has some limitations. First, our research has a potential limitation for “winner’s curse” in a two-sample MR framework using SNPs as instruments from discovery GWASs, which can cause bias. Second, there are different clinical subtypes of depression (melancholic, psychotic, atypical, or undifferentiated), BPD (type 1 or 2), and mood states (manic, depressive, mixed, or euthymic); however, a large category of diseases was analyzed without distinction. A mixture of classifications is problematic because the effect of the subtype disease liability on T2D may differ even if they are included in the same SMI category. Especially in the case of BPD or SCZ liability, the causal effect on T2D may have been annulled depending on the diseases’ subtype. Third, although we conducted bidirectional MR studies, the sample size of GWASs for BPD (< 100,000) was relatively small, which could lead to low analysis power (50.8%) with a true OR of less than 1.100. The reliability of the analysis result is low, and further MR studies with large sample size are required. Firth, we only included a European population; hence, it is difficult to apply the same clinical interpretation to other populations. Finally, horizontal pleiotropy, a natural flaw of MR design, can occurs when a genetic variant affects the outcome variable without mediating the exposure variable [48].

Nevertheless, this study provided evidence for depression liability having a causal effect on T2D, which is supported by previously reported biological mechanisms. Therefore, it is imperative to consider screening for diabetes and metabolic abnormalities in patients with depression or probable depression.

BPD

bipolar disorder

CI

confidence interval

GWAS

genome-wide association study

ICD

International Classification of Diseases

IVW

inverse-variance weighted

LD

linkage disequilibrium

MR

Mendelian randomization

MR-PRESSO

Mendelian Randomization-Pleiotropy RESidual Sum and Outlier

NOME

NO Measurement Error

OR

odds ratio

SCZ

schizophrenia

SMI

severe mental illness

SIMEX

simulation extrapolation

SNP

single nucleotide polymorphism

SNS

sympathetic nervous system

T2D

type 2 diabetes.

Ethics approval and consent to participate

The GWAS summary statistics for all traits were extracted from the public domain. Therefore, no ethical approval and consent was required for this study. The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.

Consent for publication

Not applicable.

Availability of data and materials

We have used publicly available data for this work. The datasets generated and/or analyzed during the current study are available in the DIAbetes Genetics Replication And Meta-analysis Consortium repository, https://www.diagram-consortium.org/ and Psychiatric Genomics Consortium repository, https://www.med.unc.edu/pgc/.

Competing interests

The authors declare that they have no competing interests.

Funding

This work was supported by the Industrial Core Technology Development Program (20000134) funded by the Ministry of Trade, Industry and Energy (MOTIE, South Korea); and the National Research Foundation of Korea (2017M3A9F3046543).

Authors’ contributions

H.J. analyzed and interpreted the results and wrote the manuscript. J.L and S.O contributed to discussion. S.W. and S.L. designed the study and equally contributed. All authors revised this paper critically for important intellectual content. S.W. acts as the guarantor and agrees to take responsibility for the integrity and veracity of this paper, and for the work and research it represents. S.W. accepts full responsibility for the work and/or the conduct of the study, had access to the data, and controlled the decision to publish.

Acknowledgments

We would like to thank Editage (www.editage.co.kr) for English editing and reviewing.

Vancampfort D, Correll CU, Galling B, Probst M, De Hert M, Ward PB, et al. Diabetes mellitus in people with schizophrenia, bipolar disorder and major depressive disorder: a systematic review and large scale meta-analysis. World Psychiatry.2016;15:166-74.
Dixon L, Weiden P, Delahanty J, Goldberg R, Postrado L, Lucksted A, et al. Prevalence and correlates of diabetes in national schizophrenia samples. Schizophr Bull. 2000;26:903-12.
Regenold WT, Thapar RK, Marano C, Gavirneni S, Kondapavuluru PV. Increased prevalence of type 2 diabetes mellitus among psychiatric inpatients with bipolar I affective and schizoaffective disorders independent of psychotropic drug use. J Affect Disord. 2002;70:19-26.
Anderson RJ, Freedland KE, Clouse RE, Lustman PJ. The prevalence of comorbid depression in adults with diabetes: a meta-analysis. Diabetes Care. 2001;24:1069-78.
Vancampfort D, Mitchell AJ, De Hert M, Sienaert P, Probst M, Buys R, et al. Type 2 diabetes in patients with major depressive disorder: a meta-analysis of prevalence estimates and predictors. Depress Anxiety. 2015;32:763-73.
Enger C, Jones ME, Kryzhanovskaya L, Doherty M, McAfee AT. Risk of developing diabetes and dyslipidemia among adolescents with bipolar disorder or schizophrenia. Int J Adolesc Med Health. 2013;25:3-11.
Scott D, Happell B. The high prevalence of poor physical health and unhealthy lifestyle behaviours in individuals with severe mental illness. Issues Ment Health Nurs. 2011;32:589-97.
Nouwen A, Winkley K, Twisk J, Lloyd CE, Peyrot M, Ismail K, et al. Type 2 diabetes mellitus as a risk factor for the onset of depression: a systematic review and meta-analysis. Diabetologia. 2010;53:2480-6.
Huang CJ, Hsieh HM, Tu HP, Jiang HJ, Wang PW, Lin CH. Schizophrenia in type 2 diabetes mellitus: prevalence and clinical characteristics. Eur Psychiatry. 2018;54:102-8.
Hajek T, Calkin C, Blagdon R, Slaney C, Alda M. Type 2 diabetes mellitus: a potentially modifiable risk factor for neurochemical brain changes in bipolar disorders. Biol Psychiatry.2015;77:295-303.
Feng X, Astell-Burt T. Impact of a type 2 diabetes diagnosis on mental health, quality of life, and social contacts: a longitudinal study. BMJ Open Diabetes Res Care. 2017;5:e000198.
International Schizophrenia Consortium, Purcell SM, Wray NR, Stone JL, Visscher PM, O'Donovan MC, Sullivan PF, Sklar P. Common polygenic variation contributes to risk of schizophrenia and bipolar disorder. Nature 2009;460:748-52.
Sullivan PF, Neale MC, Kendler KS. Genetic epidemiology of major depression: review and meta-analysis. Am J Psychiatry. 2000;157:1552-62.
Burgess S, Dudbridge F, Thompson SG. Combining information on multiple instrumental variables in Mendelian randomization: comparison of allele score and summarized data methods. Stat Med. 2016;35:1880-906.
Zhang F, Baranova A, Zhou C, Cao H, Chen J, Zhang X, et al. Causal influences of neuroticism on mental health and cardiovascular disease. Hum Genet. 2021;140:1267-81.
Zhang F, Rao S, Cao H, Zhang X, Wang Q, Xu Y, et al. Genetic evidence suggests posttraumatic stress disorder as a subtype of major depressive disorder. J Clin Invest. 2021:145942.
Zhang F, Baranova A. Smoking quantitatively increases risk for COVID-19. Eur Respir J. 2021:2101273.
Wang X, Fang X, Zheng W, Zhou J, Song Z, Xu M, et al. Genetic support of a causal relationship between iron status and type 2 diabetes: a Mendelian randomization study. J Clin Endocrinol Metab. 2021:dgab454.
Hou L, Xu M, Yu Y, Sun X, Liu X, Liu L, et al. Exploring the causal pathway from ischemic stroke to atrial fibrillation: a network Mendelian randomization study. Mol Med. 2020;26:7.
Jin H, Lee S, Won S. Causal evaluation of laboratory markers in type 2 diabetes on cancer and vascular diseases using various Mendelian randomization tools. Front Genet. 2020;11:597420.
Stahl EA, Breen G, Forstner AJ, McQuillin A, Ripke S, Trubetskoy V, et al. Genome-wide association study identifies 30 loci associated with bipolar disorder. Nat Genet. 2019;51:793-803.
Howard DM, Adams MJ, Clarke TK, Hafferty JD, Gibson J, Shirali M, et al. Genome-wide meta-analysis of depression identifies 102 independent variants and highlights the importance of the prefrontal brain regions. Nat Neurosci. 2019;22:343-52.
Lam M, Chen CY, Li Z, Martin AR, Bryois J, Ma X, et al. Comparative genetic architectures of schizophrenia in East Asian and European populations. Nat Genet. 2019;51:1670-8.
Scott RA, Scott LJ, Mägi R, Marullo L, Gaulton KJ, Kaakinen M, et al. An expanded genome-wide association study of type 2 diabetes in Europeans. Diabetes. 2017;66:2888-902.
Burgess S, Thompson SG; CRP CHD Genetics Collaboration. Avoiding bias from weak instruments in Mendelian randomization studies. Int J Epidemiol. 2011;40:755-64.
Bowden J, Del Greco MF, Minelli C, Davey Smith G, Sheehan NA, Thompson JR. Assessing the suitability of summary data for two-sample Mendelian randomization analyses using MR-Egger regression: the role of the I2 statistic. Int J Epidemiol. 2016;45:1961-74.
Bowden J, Spiller W, Del Greco MF, Sheehan N, Thompson J, Minelli C, et al. Improving the visualization, interpretation and analysis of two-sample summary data Mendelian randomization via the Radial plot and Radial regression. Int J Epidemiol. 2018;47:1264-78.
Verbanck M, Chen CY, Neale B, Do R. Detection of widespread horizontal pleiotropy in causal relationships inferred from Mendelian randomization between complex traits and diseases. Nat Genet. 2018;50:693-8.
Burgess S, Davey Smith G, Davies NM, Dudbridge F, Gill D, Glymour MM, et al. Guidelines for performing Mendelian randomization investigations. Wellcome Open Res. 2020;4:186.
Cardno AG, Owen MJ. Genetic relationships between schizophrenia, bipolar disorder, and schizoaffective disorder. Schizophr Bull. 2014;40:504-15.
Schulze TG, Akula N, Breuer R, Steele J, Nalls MA, Singleton AB, et al. Molecular genetic overlap in bipolar disorder, schizophrenia, and major depressive disorder. World J Biol Psychiatry. 2014;15:200-8.
Burgess S, Thompson SG. Multivariable Mendelian randomization: the use of pleiotropic genetic variants to estimate causal effects. Am J Epidemiol. 2015;181:251-60.
Sanderson, E., Davey Smith, G., Windmeijer, F., & Bowden, J. An examination of multivariable Mendelian randomization in the single-sample and two-sample summary data settings. International journal of epidemiology, 2019;48.3:713-727.
Burgess S, and Simon G. Thompson. Multivariable Mendelian randomization: the use of pleiotropic genetic variants to estimate causal effects. American journal of epidemiology. 2015; 181.4: 251-260.
Lustman PJ, Griffith LS, Clouse RE, Cryer PE. Psychiatric illness in diabetes mellitus. Relationship to symptoms and glucose control. J Nerv Ment Dis. 1986;174:736-42.
Mezuk B, Eaton WW, Albrecht S, Golden SH. Depression and type 2 diabetes over the lifespan: a meta-analysis. Diabetes Care. 2008;31:2383-90.
Gallo JJ, Anthony JC, Muthén BO. Age differences in the symptoms of depression: a latent trait analysis. J Gerontol. 1994;49:P251-64.
Tsigos C, Chrousos GP. Hypothalamic-pituitary-adrenal axis, neuroendocrine factors and stress. J Psychosom Res. 2002;53:865-71.
Geer EB, Islam J, Buettner C. Mechanisms of glucocorticoid-induced insulin resistance: focus on adipose tissue function and lipid metabolism. Endocrinol Metab Clin North Am. 2014;43:75-102.
Tabák AG, Akbaraly TN, Batty GD, Kivimäki M. Depression and type 2 diabetes: a causal association? Lancet Diabetes Endocrinol. 2004;2:236-45.
Barth E, Albuszies G, Baumgart K, Matejovic M, Wachter U, Vogt J, et al. Glucose metabolism and catecholamines. Crit Care Med. 2007;35:S508-18.
Biswas A, Oh PI, Faulkner GE, Bajaj RR, Silver MA, Mitchell MS, et al. Sedentary time and its association with risk for disease incidence, mortality, and hospitalization in adults: a systematic review and meta-analysis. Ann Intern Med. 2015;162:123-32.
Correll CU, Detraux J, De Lepeleire J, De Hert M. Effects of antipsychotics, antidepressants and mood stabilizers on risk for physical diseases in people with schizophrenia, depression and bipolar disorder. World Psychiatry. 2015;14:119-36.
Diaz FJ, James D, Botts S, Maw L, Susce MT, de Leon J. Tobacco smoking behaviors in bipolar disorder: a comparison of the general population, schizophrenia, and major depression. Bipolar Disord. 2009;11:154-65.
Kendler KS, Neale MC, MacLean CJ, Heath AC, Eaves LJ, Kessler RC. Smoking and major depression. A causal analysis. Arch Gen Psychiatry. 1993;50:36-43.
Xuan L, Zhao Z, Jia X, Hou Y, Wang T, Li M, et al. Type 2 diabetes is causally associated with depression: a Mendelian randomization analysis. Front Med. 2018;12:678-87.
Burgess S, Davies NM, Thompson SG. Bias due to participant overlap in two-sample Mendelian randomization. Genet Epidemiol. 2016;40:597-608.
Chen, L., Yang, H., Li, H., He, C., Yang, L., & Lv, G. Insights into modifiable risk factors of cholelithiasis: A Mendelian randomization study. Hepatology, 2022;75.4: 785-796.

Table 1. Description of GWAS data for MR study

Variable	PMID	Consortium	Sample size (case/control)	Population	SNPs
BPD	31043756	NA	20,352/31,358	European, North america, Australia	9,372,253
Depression	30718901	UK Biobank, 23andMe, Inc, PGC	246,363/561,190	European	8,098,588
SCZ	31740837	PGC	33,640/43,456	European	13,942,226
T2D	28566273	DIAGRAM	26,676/132,532	European	12,100,000

Abbreviations: BPD: bipolar disorder; SCZ: schizophrenia; T2D: type 2 diabetes, PGC: Psychiatric Genomics Consortium; DIAGRAM: DIAbetes Genetics Replication And Meta-analysis Consortium; NA: Not available

Table 2. Assumption test for instrumental variable sets

No.	Exposure	Outcome	N	F-Stat	I²(%)	Q-Test	Q’-Test	MR-PRESSO Global Test
1	BPD	T2D	11	32.9	96.9	0.718	0.821	0.695
2	Depression		42	37.9	97.3	0.808	0.788	0.733
3	SCZ		72	41.5	97.6	0.319	0.290	0.704
4	T2D	BPD	35	35.8	97.3	0.684	0.639	0.598
5		depression	39	36.0	97.3	0.714	0.683	0.711
6		SCZ	43	35.8	97.2	<0.001	<0.001	<0.001

Abbreviations: BPD: bipolar disorder; F-stat: F statistics; MR-PRESSO: Mendelian Randomization Pleiotropy RESidual Sum and Outlier; MR-PRESSO global test: P-value for Mendelian Randomization Pleiotropy RESidual Sum and Outlier global test; N: number of instruments; Q-test: P-value for the Q-test from inverse-variance weighted; Q’-test: P-value for the Q’-test from Mendelian Randomization-Egger; SCZ: schizophrenia.; T2D: Type 2 diabetes.

Table 3. Univariable MR results for SMI liability on T2D

MR Methods	Parameter	N	OR	95% CI	P
1. Effect of BPD liability on T2D
IVW	Estimate	11	1.004	0.942, 1.070	0.892
MR-Egger	Intercept		0.038	-0.015, 0.090	0.165
	Slope		0.990	0.519, 1.128	0.178
MR-Egger (SIMEX)	Intercept		0.003	-0.544, 0.690	0.540
	Slope		0.996	0.935, 1.060	0.917
Weighted median	Estimate		0.987	0.906, 1.075	0.770
MR-PRESSO	Estimate			No outlier	-
2. Effect of depression liability on T2D
IVW	Estimate	42	1.128	1.024, 1.245	0.014
MR-Egger	Intercept		0.007	-0.018, 0.032	0.566
	Slope		0.953	0.531, 1.711	0.875
MR-Egger (SIMEX)	Intercept		0.004	0.000, 0.008	0.036
	Slope		1.109	1.016, 1.210	0.025
Weighted median	Estimate		1.132	0.988, 1.295	0.072
MR-PRESSO	Estimate			No outlier	-
3. Effect of SCZ liability on T2D
IVW	Estimate	72	1.011	0.981, 1.040	0.463
MR-Egger	Intercept		0.001	-0.013, 0.013	0.969
	Slope		1.008	0.894, 1.137	0.888
MR-Egger (SIMEX)	Intercept		0.001	-0.002, 0.004	0.427
	Slope		1.010	0.986, 1.041	0.511
Weighted median	Estimate		1.011	0.968, 1.055	0.615
MR-PRESSO	Estimate			No outlier	-

Abbreviations: BPD: bipolar disorder; CI: confidence interval; IVW: inverse-variance weighted; MR: Mendelian randomization; MR-PRESSO: Mendelian Randomization-Pleiotropy RESidual Sum and Outlier; N: number of instruments; OR: odds of the outcome per doubling in the odds of exposure in the population (intercept of MR-Egger and MR-Egger (SIMEX) is on log‐odds scale); SCZ: schizophrenia; SIMEX: simulation extrapolation; T2D: type 2 diabetes.

Table 4. Univariable reverse MR results for T2D liability on SMI

MR Methods	Parameter	N	OR	95% CI	P
1. Effect of T2D liability on BPD
IVW	Estimate	35	1.021	0.979, 1.064	0.313
MR-Egger	Intercept		-0.002	-0.026, 0.022	0.890
	Slope		1.035	0.851, 1.258	0.727
MR-Egger (SIMEX)	Intercept		-0.003	-0.011, 0.004	0.428
	Slope		0.995	0.936, 1.056	0.867
Weighted median	Estimate		1.028	0.969, 1.091	0.349
MR-PRESSO	Estimate			No outlier	-
2. Effect of T2D liability on depression
IVW	Estimate	39	1.001	0.989, 1.013	0.793
MR-Egger	Intercept		-0.001	-0.006, 0.005	0.852
	Slope		1.005	0.964, 1.048	0.798
MR-Egger (SIMEX)	Intercept		0.001	-0.001, 0.002	0.278
	Slope		1.001	0.989, 1.013	0.808
Weighted median	Estimate		1.004	0.987, 1.022	0.581
MR-PRESSO	Estimate			No outlier	-
3. Effect of T2D liability on SCZ
IVW	Estimate	43	1.004	0.957, 1.051	0.872
MR-Egger	Intercept		-0.003	-0.026, 0.019	0.783
	Slope		1.027	0.861, 1.226	0.758
MR-Egger (SIMEX)	Intercept		0.003	-0.003, 0.009	0.246
	Slope		1.006	0.958, 1.056	0.785
Weighted median	Estimate		0.986	0.939, 1.034	0.564
MR-PRESSO	Estimate	41	0.999	0.993, 1.006	0.987

Abbreviations: BPD: bipolar disorder; CI: confidence interval; IVW: inverse-variance weighted; MR: Mendelian randomization; MR-PRESSO: Mendelian Randomization-Pleiotropy RESidual Sum and Outlier; N: number of instruments; OR: odds of the outcome per doubling in the odds of exposure in the population (intercept of MR-Egger and MR-Egger (SIMEX) is on log‐odds scale); SCZ: schizophrenia; SIMEX: simulation extrapolation; T2D: type 2 diabetes.

Table 5. Multivariable MR results for SMI liability on T2D

Multivariable MR Methods	Exposure	F-Stat	C-Q P	Parameter	OR	95% CI	P
IVW
	BPD	30.56	-	Estimate	0.922	0.859, 0.990	0.026
	Depression	27.33	-	Estimate	1.197	1.069, 1.340	0.002
	SCZ	11.78	-	Estimate	1.030	0.980, 1.083	0.241
MR-Egger
				Intercept	0.999	0.995, 1.003	0.769
	BPD	30.56	<0.05	Slope	0.925	0.861, 0.995	0.036
	Depression	27.33	<0.05	Slope	1.198	1.062, 1.349	0.003
	SCZ	11.78	<0.05	Slope	1.037	0.967, 1.112	0.299

Abbreviations: BPD: bipolar disorder; CI: confidence interval; C-Q P: Cochran’s Q-test P; F-Stat: F statistics; IVW: inverse-variance weighted; MR: Mendelian randomization; OR: odds of the outcome per doubling in the odds of exposure in the population (intercept of MR-Egger and MR-Egger (SIMEX) is on log‐odds scale); SCZ: schizophrenia; SMI: severe mental illnesses

The authors have declared there is NO conflict of interest to disclose The authors declare that they have no competing interests.

Mendelian randomization shows depression increase the risk of type 2 diabetes

Status:

Version 1

Abstract

Figures

Background

Methods

Exposure datasets: summary statistics of genetic association analyses for BPD, depression, and

SCZ

Outcome datasets: summary statistics of genetic association analyses for T2D

MR analysis

Statistical power analyses

Reverse MR analysis

Multivariable MR analysis

Results

Effect of liability to BPD on T2D risk

Effect of liability to depression on T2D risk

Effect of liability to SCZ on T2D risk

Discussion

Conclusions

List Of Abbreviations

Declarations

References

Tables

Additional Declarations

Supplementary Files

Status:

Version 1