We designed a two-sample MR study to investigate the causal effect of plasma Hcy level on the risk of MPDs, including SCZ, MDD, and BD. We selected the single nucleotide polymorphisms (SNPs) as instrumental variables (IVs) for plasma Hcy level based on the following three basic assumptions (Figure 1): (1) the selected SNPs should be significantly associated with plasma Hcy level; (2) SNPs must be independent of confounders; (3) SNPs were only associated with the risk of MPDs by plasma Hcy levels.

We obtained the genetic variants associated with plasma Hcy level from a large-scale GWAS meta-analysis, with up to 44,147 individuals of European ancestry (20). We selected SNPs associated with plasma Hcy levels at the genome-wide significance threshold (P < 5 × 10⁻⁸) from the plasma Hcy GWAS. We then extracted the associations between IVs and MPDs from GWAS datasets of the Psychiatric Genomics Consortium (PGC) among individuals of European ancestry. Genetic associations for SCZ based on a large-scale PGC3 meta-analysis included 67,390 cases and 94,015 controls of European ancestry (24). Genetic associations with MDD were also obtained from a GWAS based on 16,823 cases and 25,632 controls of European descent (25). The associations between the genetic instrument and BD for common variants (minor allele frequency [MAF] >1%) came from GWAS of BD from the Psychiatric Genomics Consortium (PGC) among individuals of European ancestry. Summary statistics for BD were obtained from a recent GWAS meta-analysis including 41,917 cases and 371,549 controls from the BD working group of the PGC (26). For BD subtypes, BD-I GWAS meta-analyses consisted of 25,060 cases and 449,978 controls, and BD-II GWAS included 6,781 cases and 364,075 controls. The basic characteristics of GWAS samples are listed in Table 1. All the summary statistics of MPDs are available on the PGC website (https://www.med.unc.edu/pgc/results-and-downloads). Descriptions of exposure and outcome sources, such as the number of controls and cases, population structure, and dataset source, are presented in Table 1.

To examine the causal relationship between Hcy and MPDs, we analyzed the direction of causal effects using Generalized Summary-data based on Mendelian Randomization (GSMR; http://cnsgenomics.com/software/gsmr/) (27). For the genetic variants associated with plasma Hcy level, we selected linkage disequilibrium (LD)-independent lead SNPs (r² < 0.01 and windows: 1,000 kb) with a genome-wide significant level (P < 5 × 10⁻⁸) as instrumental variables in the GSMR analyses (27). Then, we performed the HEIDI-outlier test implemented in GSMR software to test for horizontal pleiotropy (P_HEIDI < 0.01) and remove SNPs with pleiotropic effects on plasma Hcy level and MPDs (27). This study included analyses of 3 MPDs and 2 BD subtypes. A Bonferroni corrected threshold of P = 0.01 was considered to be significant. To complement the GSMR analysis, we also conducted the MR analyses using the inverse-variance-weighted (IVW) regression method, implemented via the 'TwoSampleMR' R package (28, 29). We applied the IVW method to examine the effects of plasma homocysteine level on the risks of MPDs (30), and compare whether the effect size estimates from the IVW regression method were consistent with those calculated using the GSMR method. The inverse-variance weighted (IVW) method estimates the causal effect by combining the casual estimates of each SNP in a fixed-effect meta-analysis model (30). To further investigate horizontal pleiotropy, we used the “TwoSampleMR” package to perform the MR-Egger regression to detect and correct horizontal pleiotropy (28, 29, 31). The leave-one-out sensitivity analyses were performed to determine whether the overall estimate was disproportionately affected by a specific SNP (28, 29).

Using the HEIDI-outlier method, we removed 3 SNPs (rs2251468, rs548987, and rs838133; P_HEIDI < 0.01; Supplementary Table 3) showed pleiotropic effects on plasma Hcy levels and SCZ and significantly deviated from causal models. Then, using the 10 remaining genetic instruments, we performed GSMR analysis, and our results showed that plasma Hcy level had a risk effect on SCZ (OR = 1.12, se = 0.035, P_GSMR = 1.73 × 10⁻³; Figure 2 and Table 2). An ORvalue of 1.12 indicates that individuals whose plasma Hcy level is increased by one Standard Deviation will have an increased risk of SCZ by 1.12 times compared with the population prevalence. Additionally, the effect size estimates from the IVW regression method (OR = 1.110, se = 0.039, P_IVW = 2.74 × 10⁻³; Table 2) were consistent with those calculated using the GSMR method. Moreover, our MR-Egger analysis suggested no evidence of bias from horizontal pleiotropy (intercept = −0.004, se = 0.01, P = 0.71). The leave-one-out analysis showed that no specific SNP affected the overall estimate (Supplementary Figure 1).

Our GSMR results suggested no evidence of a causal effect of plasma Hcy levels on MDD (OR = 0.951, SE = 0.034, P_GSMR = 0.139; Figure 2 and Table 2). Moreover, the IVW analysis also suggested no evidence of an association between genetically predicted plasma Hcy level and MDD (OR = 0.948, SE = 0.032, P_IVW = 0.115; Table 2). The estimates from MR-Egger regression (intercept = −0.001, se = 0.01, P = 0.86) analyses indicate that no horizontal pleiotropy exists. We detected no specific SNP affected the MR estimate in leave-one-out analyses (Supplementary Figure 2).

Using the HEIDI test, we removed two SNPs (rs2251468 and rs548987; P_HEIDI < 0.01), showing pleiotropic effects on plasma Hcy level and BD risk. Then, we conducted GSMR analysis using 11 retained SNPs and identified suggestive significant risk effects of plasma Hcy level on BD (OR = 1.088, SE = 0.04, P_GSMR = 0.037; Figure 2 and Table 2). Our IVW results also suggested the same direction of effect size estimates (OR = 1.081, se = 0.041, P_IVW = 0.05; Table 2), consistent with the GSMR results. However, the results were not significant after the Bonferroni correction (P > 0.01).

Then, we further examined the associations between plasma Hcy level and two BD subtypes, including BD-I and BD-II. We identified a significant risk effect of plasma Hcy level against BD-I (OR = 1.144, SE = 0.048, P_GSMR = 5.23 × 10⁻³; Figure 2 and Table 2), surviving after Bonferroni correction (P < 0.01). However, we found no evidence for a causal relationship between plasma Hcy level against BD-II (OR = 0.977, SE = 0.082, P_GSMR = 0.783; Figure 2 and Table 2). In addition, our IVW analysis also showed that plasma Hcy levels have a significant positive effect on BD-I risk (OR = 1.133, SE = 0.048, P_IVW = 9.44 × 10⁻³; Table 2), and no association between Hcy and BD-II (OR = 0.977, SE = 0.082, P_IVW = 0.773; Table 2). The associations were consistent in the MR-Egger regression analysis (BD-I: intercept = −0.006, se = 0.014, P = 0.68; BD-II: intercept = −0.020, se = 0.017, P = 0.26). Based on the leave-one-out sensitivity analyses, there was no evidence of obvious associations between plasma Hcy level with BD and its subtypes (Supplementary Figures 3–5).

Reports suggest Hcy may cause neurotoxicity by directly or indirectly activating glutamate receptors (32, 33). Based on the fact that Hcy and its oxidative metabolite, homocysteic acid, are N-methyl-D-aspartic-acid (NMDA) receptor agonists (32, 33), Hyperhomocysteinemia may cause long-term activation of NMDA receptors and exert neurotoxic and vasculotoxic effects (34). However, the definitive mechanism of elevated plasma Hcy levels in patients with SCZ or BD is still unclear. Our study provides further evidence for the causal effects of plasma Hcy levels on SCZ or BD-I risk. The accumulated evidence substantiates increased Hcy plasma levels are associated with SCZ or BD risk. A recent MR analysis in the Japanese population identified that increased plasma Hcy level was significantly associated with the risk of SCZ (35). Several observational studies found that the prevalence of hyperhomocysteinemia in patients with SCZ or BD was significantly higher than in normal controls (5, 36). Additionally, a meta-analysis provided evidence that plasma Hcy level is elevated in individuals with BD compared with healthy controls (4). Furthermore, some studies found that that Hcy was positively associated with the severity of SCZ, including negative symptoms (37) and cognitive functions (38–40). However, previous observational and epidemiological studies may suffer from confounding factors and limited sample size, leading to fallacious findings (16). In this regard, the MR method can ingeniously reduce the weaknesses of traditional research, providing a complementary method regarding etiology. Our MR results indicate significantly higher plasma Hcy levels were associated with SCZ or BD-I risk, which is in line with previous studies. Additionally, for the analysis of the association between Hcy and BD, most of the observational studies did not focus on BD subtypes. Herein, we interestingly found an increased level of plasma Hcy was associated with risk of BD-I, not with BD-II, suggesting that the one-carbon metabolism may be promising for clinical classification of BD. Notably, further randomized controlled trials need to be designed to verify the causal relationship between plasma Hcy level and MPDs risks.

Our study found no evidence to substantiate the causal association of Hcy with MDD. Many previous studies found that increased plasma Hcy levels may be associated with an increased risk of MDD (41), which is inconsistent with our MR findings. Multiple factors may account for this inconsistency. For instance, the sample size of traditional observational studies was not significant enough. Moreover, due to the presence of unobserved confounders and the excessively large number of confounders in observational studies, regression methods may fail to provide unbiased estimates of the true association. Additionally, elevated plasma Hcy levels detected in observational studies were potentially influenced by MDD.

Our findings on the effects of Hcy on MPDs are compatible with known biology. Hcy may cause neurotoxicity by directly or indirectly activating glutamate receptors (32, 33). Since Hcy and its oxidative metabolite, homocysteic acid, are NMDA receptor agonists, hyperhomocysteinemia may cause long-term activation of NMDA receptors and exert neurotoxic and vasculotoxic effects (34). However, the definitive mechanism of elevated Hcy levels in patients with SCZ or BD is still unclear. It is speculated that poor nutrition, tobacco consumption, alcohol, coffee and polymorphisms in the enzymes of Hcy metabolism can all contribute to elevated Hcy levels (42, 43). Therefore, more emphasis should be placed on one-carbon metabolism in patients with SCZ or BD to improve the current understanding of disease pathogenesis.

Our findings provide the basis for potential clinical applications of plasma Hcy levels as a therapeutic target for SCZ or BD prevention. A large number of clinical studies have found that vitamin B12 and folic acid supplementation can reduce the plasma level of Hcy. With reduced plasma Hcy levels, the disease severity and cognitive function of patients with SCZ can significantly be improved (44, 45). Meanwhile, mood stabilizers (e.g., valproate and lamotrigine) used in treating BD can interfere in the folate and Hcy metabolism pathway (46). Recently, clinical research found that the vitamin B12 levels were significantly higher in responders with BD, suggesting that vitamin B12 supplements might be beneficial to treating patients with BD (47). Accordingly, more randomized clinical trials are warranted to validate whether lowering Hcy levels could alleviate the clinical symptoms of patients with SCZ or BD.

Our study has several limitations. First, our MR study identified the causal relationship between Hcy and SCZ/BD-I risk. However, the number of genetic instruments used in our MR analysis was limited. Further studies using more SNPs that are associated with plasma Hcy levels are needed. Second, our findings should be interpreted with caution since we could not completely reduce the effects of potential pleiotropy, which can be a source of biased estimates. However, the MR-Egger intercept test detected no evidence of pleiotropic effect in our study, and similar results were identified in sensitivity analyses using the leave-one-out method. Moreover, the SCZ or BD risk might be affected by exposure to higher Hcy levels at a specific period. Our MR analysis investigated lifelong elevated Hcy levels on SCZ/BD. However, genome-wide associations between SNPs and Hcy levels from adults were used. These associations in the early stage of the disease can be examined in the future. Furthermore, the examined GWAS were primarily conducted in individuals of European ancestry. Hence these results cannot be generalized to all populations. Besides, a reverse analysis could not be performed since the GWAS for Hcy is not publicly available. Therefore, the effects of the risk of MPDs on plasma homocysteine level should be investigated in the future.