Dissecting Causal Associations of Diet-Derived Circulating Antioxidants with Six Major Mental Disorders: A Mendelian Randomization Study

Although observational studies have suggested associations between circulating antioxidants and many mental disorders, causal inferences have not been confirmed. Mendelian randomization (MR) analyses were conducted using summary-level statistics from genome-wide association studies (GWASs) to explore whether genetically determined absolute circulating antioxidants (i.e., ascorbate, retinol, β-carotene, and lycopene) and metabolites (i.e., α- and γ-tocopherol, ascorbate, and retinol) were causally associated with the risk of six major mental disorders, including anxiety disorders (AD), major depressive disorder (MDD), bipolar disorder (BIP), schizophrenia (SCZ), post-traumatic stress disorder (PTSD), and obsessive–compulsive disorder (OCD). MR analyses were performed per specific-outcome databases, including the largest GWAS published to date (from 9725 for OCD to 413,466 for BIP participants), UK Biobank (over 370,000 participants), and FinnGen (over 270,000 participants), followed by meta-analyses. We found no significant evidence that genetically determined diet-derived circulating antioxidants were significantly causally associated with the risk of the six above-mentioned major mental disorders. For absolute antioxidant levels, the odds ratios (ORs) ranged from 0.91 (95% CI, 0.67–1.23) for the effect of β-carotene on OCD to 1.18 (95% CI, 0.90–1.54) for the effect of ascorbate on OCD. Similarly, for antioxidant metabolites, ORs ranged from 0.87 (95% CI, 0.55–1.38) for the effect of ascorbate on MDD to 1.08 (95% CI, 0.88–1.33) for the effect of ascorbate on OCD. Our study does not support significant causal associations of genetically determined diet-derived circulating antioxidants with the risk of major mental disorders.


Introduction
According to the latest data on the global burden of disease, from 1990 to 2019, the disability-adjusted life years caused by mental disorders increased from 80.8 million to 125.3 million, and the percentage contributed by mental disorders increased from 3.1% to 4.9% [1]. Moreover, due to the COVID-19 pandemic, people are facing unprecedented centrations in plasma or serum, either or both. The instrumental variables needed to satisfy three assumptions: the relevance assumption, the independence assumption, and the exclusion restriction assumption [30]. For each mental disorder as an outcome, MR analyses were performed in each of three European databases (the largest GWAS published to date, UK Biobank, and the FinnGen study), and these were subsequently meta-analyzed to approximate the average genetically influenced effect on each specific outcome. This study was reported according to the Strengthening the Reporting of Observational Studies in Epidemiology Using Mendelian Randomization (STROBE-MR) checklist. The schematic overview and framework of the present study design are shown in Figure 1.

Determination of Exposures
For instrumental variables on absolute circulating antioxidants, genetically determined α-tocopherol, ascorbate, retinol, and β-carotene were identified in the recent large-scale GWAS (p < 5 × 10 −8 , linkage disequilibrium [LD]: r 2 = 0.001 and clump distance = 10,000 kb). Three independent single nucleotide polymorphisms (SNPs) associated with α-tocopherol were identified from a GWAS on 4014 individuals of European ancestry [31]; however, these three SNPs were reported to be related to lipid metabolism or regulation and therefore were not considered for MR analysis due to possible pleiotropic bias. Eleven SNPs associated with ascorbate were identified from a recently published GWAS on 52,018 individuals of European ancestry [32]. One (rs7740812) of the 11 SNPs was removed for subsequent MR analysis due to the LD with other variants (r 2 > 0.001) or absence from the LD reference panel using the EUR population reference. Two independent SNPs associated with retinol were identified from a GWAS of 5006 Caucasian individuals in two cohort studies [33]. Two independent SNPs associated with β-carotene were identified from a GWAS of 2344 participants in the Nurses' Health Study [34]. Five independent SNPs associated with lycopene were identified from a GWAS on 441 Caucasian participants under the relaxed threshold criteria (p < 5 × 10 −6 ) [35].
Based on PhenoScanner, the SNPs that did not reach a significant association with the confounders were retained in the filtered genetic instrument [38]. The F statistic for each SNP was calculated by the formula Beta 2 /SE 2 , and the F statistics of >10 for each SNP was recommended for subsequent MR analysis to avoid employing weak genetic instruments.

Data Sources of Major Mental Disorders
Summary-level statistics for each mental disorder were obtained from three large databases, including the largest GWAS published by European-ancestry to date, UK Biobank, and the FinnGen study. Neither the UK Biobank nor the FinnGen study were major parts of the largest GWAS study, preventing the inclusion of overlapping samples. For SNPs of the instrument that were not available in the outcome GWAS, the LDlink tool was used to identify proxy SNPs of European ancestry [39]. SNPs missing in the outcome GWAS without appropriate proxy SNPs available were then excluded. The largest GWAS summary statistics for AD, MDD, BIP, SCZ, PTSD, and OCD were extracted from the Psychiatric Genomics Consortium (PGC) website (https://www. med.unc.edu/pgc/results-and-downloads/ (accessed on 25 June 2022)). The PGC is the largest consortium in the history of psychiatry, which has conducted the most influential meta-and mega-analysis of genome-wide genomic data for mental disorders. The GWAS summary datasets of European-ancestry, for AD (5712 cases and 11,598 controls) from the Anxiety NeuroGenetics STudy (ANGST) Consortium in 2016 [40], MDD (59,851 cases and 113,154 controls) from PGC in 2018 [41], BIP (41,917 cases and 371,549 controls) from PGC in 2021 [42], SCZ (52,017 cases and 75,889 controls) from PGC in 2022 [43], PTSD (23,212 cases and 151,447 controls) from PGC in 2019 [44], and OCD (2688 cases and 7037 controls) from the International Obsessive-Compulsive Disorder Foundation Genetics Collaborative (IOCDF-GC) and the OCD Collaborative Genetics Association Study (OCGAS) in 2017 [45] were obtained from genome-wide meta-or mega-analyses.
The UK Biobank was a population-based prospective cohort study with deep genetic, physical, and health data collected on about 500,000 individuals, aged 40-69 years, across the UK from 2006 to 2010. The second-round analysis of UK Biobank data from the Pan-UK Biobank project (https://pan.ukbb.broadinstitute.org/ (accessed on 25 June 2022)) was used in the present study. Based on the code, there were 9705 cases of AD, 478 cases of MDD, 1257 cases of BIP, 701 cases of SCZ, 187 cases of PTSD, 193 cases of OCD, and 369,930 controls. The FinnGen summary statistics were from release 7 in 2022 (https: //finngen.gitbook.io/documentation/data-download (accessed on 25 June 2022)). There were 18,358 cases of AD, 5763 cases of BIP, 6050 cases of SCZ, 1639 cases of PTSD, 1466 cases of OCD, and over 270,000 controls.

Statistical Analysis
The selection process of the main MR analysis methods is shown in Figure 1. When the MR-Egger intercept test found that there may be potential horizontal pleiotropy, MR-Egger regression was performed with pleiotropy-robust causal estimates by bootstrapped standard errors [46]. Notably, the existence of significant horizontal pleiotropy was extremely rare due to the rigorous screening of instrumental variables. The inverse-variance weighted (IVW) meta-analysis assumes that either all the instruments are valid or any horizontal pleiotropy is balanced. Heterogeneity was detected using Cochran's Q-statistics test. If there was significant heterogeneity, the random-effect IVW model was used as the main analysis; otherwise, the fixed-effect IVW model was used as the main analysis [47].
For each specific exposure and each specific outcome, MR analyses were performed separately in each of the specific-outcome databases from different sources and then were subsequently meta-analyzed to generate the pooled estimates for each specific exposure on the risk of each specific outcome. Through Cochran's Q-statistics test, we calculated I 2 statistics to quantify heterogeneity between estimates from different databases and corresponding p values. If there was no significant heterogeneity across different databases, fixed-effect model meta-analyses were used to pool estimates across the databases from different sources for each specific exposure. If there was significant heterogeneity after excluding the obvious clinical heterogeneity, the random effect model meta-analyses were used to pool estimates.

Sensitivity Analysis
To further assess the robustness of our findings, a series of sensitivity analyses was performed. First, complementary MR analyses with different assumptions were applied to help verify causal inference. The likelihood-based MR can effectively estimate the log-linear association between exposure and outcome risk (No. SNPs > 1) [48]. The weighted median can provide valid estimates if at least 50% of the weight comes from valid instrumental variables (No. SNPs > 2) [49]. The MR Pleiotropy RESidual Sum and Outlier (MR-PRESSO) was performed, which detects and corrects the effects from outliers (No. SNPs > 3) [50]. Second, we also performed scatter, forest, funnel, and leave-one-out plots to detect high influence points. Third, if MR-PRESSO detected outliers, the meta-analysis was performed again for each specific exposure and each specific outcome by removing the outliers. Finally, as only depression and no MDD phenotypes were found in the FinnGen database, we considered depression as a secondary outcome and estimated the effect of circulating antioxidants on broader depression through MR analysis in two GWAS analyses, the PGC and UK Biobank study (170,756 cases and 329,443 controls) [51], and the FinnGen study (33,812 cases and 271,380 controls).
The prior statistical power was calculated using the mRnd power calculation online tool [52]. Given a type 1 error of 5%, we had sufficient power (>80%) to detect the minimum detectable OR. To account for multiple testing in our analyses, we used a Bonferronicorrected threshold of p < 0.001 (α = 0.05/48) as significant evidence of associations, and a p-value between 0.05 and 0.001 was considered suggestive evidence of associations. All statistical analyses were performed using R version 4.1.0.

Strength of Genetic Instruments
The summary information on instrumental variables for absolute circulating antioxidants and antioxidant metabolites is presented in Table 1. The information on the cohorts contributing to the GWAS of absolute antioxidant levels is given in Table S1. In our study, the F statistic of each SNP was greater than 10, indicating that the instrumental variables of antioxidants could better avoid the bias of potentially weak instrumental variables. The summary information of GWAS for six major mental disorders is shown in Table S2. Although MDD was not found in the FinnGen study, the other five outcomes used three GWAS databases from different sources. As shown in Table S3, the minimum detectable effect sizes were reasonable at sufficient power, especially for the largest GWAS source databases. Therefore, most of our studies had enough power given the current parameters. The raw data information on the effect estimation for the associations of selected SNPs with antioxidants and with major mental disorders is given in Tables S4 and S5.

Effect of Absolute Circulating Antioxidants on the Risk of Major Mental Disorders
The primary results of the MR estimate for absolute circulating antioxidants are presented in Figure 2. For instrumental variables with three or more SNPs, the MR-Egger intercept test found no significant horizontal pleiotropy for all outcomes; therefore, the IVW method was used as the primary analysis results. Genetically determined absolute ascorbate levels were not associated with the risk of each mental disorder in any database, with the pooled ORs per 1 µmol/L of ascorbate ranging from 0.93 (95% CI, 0.81-1.06) for SCZ to 1.18 (95% CI, 0.90-1.54) for OCD. Genetically determined absolute retinol levels were also not associated with the risk of each mental disorder in any database, with the pooled ORs per 0.1 ln-transformed retinol ranging from 0.97 (95% CI, 0.92-1.03) for AD to 1.04 (95% CI, 0.97-1.11) for PTSD. Similarly, genetically determined absolute β-carotene levels were not associated with the risk of each mental disorder in any database, with the pooled ORs per ln-transformed β-carotene ranging from 0.91 (95% CI, 0.67-1.23) for OCD to 1.07 (95% CI, 0.97-1.19) for AD. Except for the suggestive association of absolute lycopene with the risk of AD found only in the FinnGen study, we observed no evidence that absolute lycopene levels were associated with the risk of each mental disorder. Notably, the pooled estimates were not significantly associated between lycopene and the risk of any mental disorder, with the pooled ORs per 1 µg/dL of lycopene ranging from 0.97 (95% CI, 0.87-1.09) for OCD to 1.04 (95% CI, 0.98-1.11) for MDD. The supplementary MR analyses showed that the results of the likelihood-based MR method were in good agreement with the IVW results, and the results of the MR-Egger and weighted median method for ascorbate and lycopene were also comparable to the IVW results, even though their estimates were more conservative (Tables S6-S9). The MR-PRESSO method identified one outlier SNP for ascorbate on BIP in PGC and FinnGen and for ascorbate on SCZ in PGC. MR analyses after removing this outlier showed that the OR in the corresponding database and combined OR estimates did not change significantly ( Figure S1). No outlier SNPs were identified in the MR-PRESSO analysis for the other outcomes, and the leave-one-out analyses also found no significant outliers. The combined estimates also found no significant causal association between absolute antioxidant levels and depression (as a secondary outcome), which was consistent with the results of MDD ( Figure S2). The supplementary MR analyses showed that the results of the likelihood-based MR method were in good agreement with the IVW results, and the results of the MR-Egger and weighted median method for ascorbate and lycopene were also comparable to the IVW results, even though their estimates were more conservative (Tables S6-S9). The MR-PRESSO method identified one outlier SNP for ascorbate on BIP in PGC and FinnGen and for ascorbate on SCZ in PGC. MR analyses after removing this outlier showed that the OR in the corresponding database and combined OR estimates did not change significantly ( Figure S1). No outlier SNPs were identified in the MR-PRESSO analysis for the other outcomes, and the leave-one-out analyses also found no significant outliers. The combined estimates also found no significant causal association between absolute antioxidant levels and depression (as a secondary outcome), which was consistent with the results of MDD ( Figure S2). Figure 3 shows the primary results of MR estimates for circulating antioxidant metabolites. We used the IVW method as the primary analysis outcome, except γ-tocopherol for AD in the UK Biobank database and ascorbate for OCD in the IOCDF-GC and OCGAS database, using the MR-Egger method due to possible potential pleiotropy (Tables S10-S13). Although some suggestive associations were found in some single databases, such as αtocopherol for SCZ in the FinnGen database, γ-tocopherol for MDD in the PGC and UK Biobank databases, γ-tocopherol for PTSD in the UK Biobank database, ascorbate for MDD in the UK Biobank database, and retinol for AD and BIP in the FinnGen database, the combined estimates showed no significant causal association between any of the four antioxidant metabolites and the risk of any of the six mental disorders. For per 0.1 increase in log-transformed α-tocopherol, the pooled ORs ranged from 0.96 (95% CI, 0.88-1.04) for OCD to 1.02 (95% CI, 0.97-1.07) for PTSD. Similarly, for per 0.1 increase in log-transformed γ-tocopherol, the pooled ORs ranged from 0.94 (95% CI, 0.85-1.05) for MDD to 1.02 (95% CI, 0.99-1.04) for PTSD. For per increase in log-transformed ascorbate, the pooled ORs ranged from 0.87 (95% CI, 0.55-1.38) for MDD to 1.08 (95% CI, 0.88-1.33) for OCD. For per increase in log-transformed retinol, the pooled ORs ranged from 0.95 (95% CI, 0.90-1.01) for OCD to 0.99 (95% CI, 0.97-1.02) for SCZ.

Effect of Circulating Antioxidant Metabolites on the Risk of Major Mental Disorders
Similarly, the supplementary MR analyses showed that the results obtained by the likelihood ratio method, weighted median method, MR-Egger method, and IVW method were comparable (Tables S10-S13). The MR-PRESSO method identified 1 outlier SNP for α-tocopherol on AD in FinnGen, for γ-tocopherol on AD in PGC, and for retinol on AD in PGC. MR analyses after removing this outlier showed that the OR in the corresponding database and combined OR estimates did also not change significantly ( Figure S3). No outlier SNPs were identified in the MR-PRESSO analysis for the other outcomes, and the leave-one-out analyses also found no significant outliers. The combined estimates also found no significant causal association between antioxidant metabolites and depression (as a secondary outcome), which was consistent with the results of MDD ( Figure S4).

Discussion
This study investigated the causal associations between diet-derived circulating antioxidants and the risk of six major mental disorders based on MR analyses. The genetic variation of circulating antioxidants was evaluated as authentic absolute blood levels and metabolite concentrations as instrumental variables, and comparable results were obtained. Although several suggestive associations were found in a single database, these were considered to be chance findings because no significant causal association between any antioxidants and any mental disorders was found in the combined database estimates. Therefore, we found no evidence to support a significant causal relationship between genetically determined diet-derived antioxidants and the risk of six major mental disorders.

Discussion
This study investigated the causal associations between diet-derived circulating antioxidants and the risk of six major mental disorders based on MR analyses. The genetic variation of circulating antioxidants was evaluated as authentic absolute blood levels and metabolite concentrations as instrumental variables, and comparable results were obtained. Although several suggestive associations were found in a single database, these were considered to be chance findings because no significant causal association between any antioxidants and any mental disorders was found in the combined database estimates. Therefore, we found no evidence to support a significant causal relationship between genetically determined diet-derived antioxidants and the risk of six major mental disorders.
Two retrospective reviews of the role of vitamin C in human mental disorders found current clinical evidence to be limited and inconclusive [53,54]. Two systematic reviews involving antioxidant nutritional supplements for depression and bipolar disorder found no significant effect of vitamin C on reducing depressive symptoms and preventing or treating bipolar disorder [23,55]. The World Federation of Societies of Biological Psychiatry guidelines also reported no evidence to support the treatment of mental disorders with antioxidants such as vitamin C [56]. A study based on the National Health and Nutrition Examination Survey also found no differences in the intake of antioxidants, including vitamins A, C, and E, between patients with SCZ and healthy controls [24]. These results are consistent with our findings. However, a meta-analysis of the effects of α-tocopherol on depression and anxiety showed that vitamin E supplementation could not improve anxiety but could improve depressive symptoms, but this meta-analysis had substantial heterogeneity (89-95%) due to the small sample size (less than 50 samples in half of the included studies), multiple interventions (mixed with components such as omega-3 fatty acids), and inconsistent outcome measurements [19], with the result being that this finding should be interpreted cautiously.
The current RCTs focus more on the role of antioxidants in the treatment of mental disorders, and there is a lack of large RCTs on whether the supplementation of antioxidants can reduce the risk of mental disorders. For example, in a study of Alzheimer's disease, also a neurological disorder, an RCT of about 4000 people found no statistically significant effect of antioxidant supplementation on cognitive function [57], and a recent MR study of circulating antioxidants and Alzheimer's disease also showed that higher levels of ascorbate, β-carotene, and retinol exposure did not reduce the risk of Alzheimer's disease [58]. This also suggests that MR studies based on instrumental variables proxying circulating antioxidants from a genetic perspective have obtained comparable and consistent conclusions with empirical studies from large RCTs. Unfortunately, no MR studies have assessed the causal relationship between circulating antioxidants and the risk of major mental disorders to date. A previous MR study of circulating antioxidants has demonstrated that the effect of genetic variants on circulating antioxidant levels is generally comparable with those that would be achieved by dietary supplementation [59]. Considering the robust and broadly consistent null results in this study that neither absolute blood antioxidant levels nor metabolites measured by high-throughput techniques were causally associated with major mental disorders, dietary supplements that increase blood antioxidant concentrations may not reduce the risk of major mental disorders in healthy people.
Oxidative stress is an imbalance between oxidants and antioxidants in favor of the oxidants, leading to a disruption of redox signaling and control and/or molecular damage [60]. Our genetic findings do not contradict the hypothesis that oxidative stress plays an important role in the pathogenesis of mental disorders. The results may be related to the fact that the circulating antioxidant levels do not necessarily correspond to antioxidant nutritional intake [12]. Moreover, circulating levels of vitamin E do not provide complete insight in its antioxidant capacity [61]. Therefore, for healthy adults without nutritional deficiency and pathological damage, genetic evidence does not support the significant protective effect of dietary supplements that increase the concentration of blood antioxidants on the prevention of mental disorders. In addition, mental disorders may be associated with oxidative stress, and even if genetically determined diet-derived circulating antioxidants are not significantly associated with the risk of mental disorders, patients may benefit from antioxidant supplementation to reduce damages due to mental disorders. In the future, more real-world data on oxidative stress assessment will be needed to guide public health and clinical practice in mental disorders.
There are three main strengths in the present study. First, the MR design of two independent samples based on instrumental variables reduces the possibility of subjects being exposed to unnecessary risks and hazards in clinical trials and expands the genetic theoretical basis of dietary antioxidants and major mental disorders. Second, we used two independent sets of genetic instrument variables as proxies for dietary-derived antioxi-dants for ascorbate and retinol, including the absolute circulating antioxidants and their metabolites. Third, for each outcome except MDD, we conducted a separate analysis in three databases followed by a meta-analysis, which added to the robustness of our results.
However, some limitations should also be considered concerning the interpretation of the results. First, based on published summary data, we were unable to test nonlinear causal relationships between antioxidant levels and risk for select mental disorders. Second, the number of SNPs currently identified as instrumental variables for antioxidants is limited, but their ideal representation has been widely used in previous MR studies [59,62]. Also limited by the number of SNPs of instrumental variables, the MR-Egger, weighted median, and MR-PRESSO methods could not be performed for absolute retinol and β-carotene [33,34], but these instruments are mapped in genes critical to antioxidant metabolism and are not associated with any other risk factors for mental disorders in the PhenoScanner database [38], indicating that horizontal pleiotropy is unlikely to exist. In addition, the SNPs related to β-carotene were obtained from the Nurses' Health study in the female population. In the future, it will be necessary to further discover more relevant sites through more large GWASs of antioxidants to improve the instrument variable strength further. Third, due to the rarity of some people with a mental health condition in the UK Biobank or FinnGen databases, it is also evident from our power calculations that there is not enough power in the database with a low proportion of patients, which may explain several associations that are thought to be chance. However, importantly, our findings on mental disorders based on different analysis methods are robust, and the results of the largest GWAS and meta-analyses to date are generally consistent. Fourth, we found that the meta-GWAS analysis for ascorbate by Zheng et al. [32] showed a considerable variation in detection concentrations in different cohorts, and it is necessary to standardize homogeneous detection techniques in the future. Finally, we could not explore these associations in nutritionally deficient populations that might be more promising with antioxidant supplementation or to test the effects of antioxidants in combination with other treatments. Furthermore, due to the availability of data, this study focused on populations of European ancestry, and the associations in other populations need further validation.

Conclusions
In summary, our study does not support significant causal associations of genetically determined diet-derived circulating antioxidants of vitamins E and C, retinol, β-carotene, and lycopene, with the risk of six major mental disorders. Therefore, for healthy adults without nutritional deficiency and pathological damage, simply taking antioxidants to increase blood antioxidant levels may not have a significant protective effect on the prevention of major mental disorders. In the future, large-scale GWASs are needed to further validate our current findings by utilizing additional genetic variants and more samples.

Supplementary Materials:
The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/antiox12010162/s1, Table S1. The summary information of the studies used for instrumental variables extraction of absolute circulating antioxidants. Table S2. The summary statistics of the six major mental disorders. Table S3. Minimum detectable OR with sufficient power in MR analysis. Table S4. The causal effect estimates of the associations between instrumental variables for absolute blood antioxidants and the risk of six major psychiatric disorders. Table S5. The causal effect estimates of the associations between instrumental variables for antioxidant metabolite concentrations and the risk of six major psychiatric disorders. Table S6. The MR analysis results of the causal effects of absolute ascorbate levels on major mental disorders. Table S7. The MR analysis results of the causal effects of absolute retinol levels on major mental disorders. Table S8. The MR analysis results of the causal effects of absolute β-carotene levels on major mental disorders. Table S9. The MR analysis results of the causal effects of absolute lycopene levels on major mental disorders. Table S10. The MR analysis results of the causal effects of α-tocopherol metabolites on major mental disorders. Table S11. The MR analysis results of the causal effects of γ-tocopherol metabolites on major mental disorders. Table S12. The MR analysis results of the causal effects of ascorbate metabolites on major mental disorders. Table S13. The MR analysis results of the causal effects of retinol metabolites on major mental disorders. Figure S1. The MR analyses for absolute circulating antioxidant levels on the risk of major mental disorders after removing outliers. Figure S2. The causal effects of four absolute circulating antioxidant levels on the risk of depression. Figure S3. The MR analyses for circulating antioxidant metabolites on the risk of major mental disorders after removing outliers. Figure S4. The causal effects of four circulating antioxidant metabolites on the risk of depression. Reference [63] is cited in supplementary materials.