MR is an approach that uses genetic variants as instrumental variables to investigate the causal relationship between exposures and outcomes, which should satisfy three principal assumptions: 1) the instrumental variables should be significantly associated with exposure; 2) the instrumental variables should not be associated with any confounders; and 3) the instrumental variables should affect the risk of the outcome only by the exposure. The second and third assumptions are also considered independent of pleiotropy. In this study, MR is based on the publicly available GWAS summary datasets of TBVMs (Smith et al., 2021) and SCZ (Schizophrenia Working Group of the Psychiatric Genomics Consortium, 2014), and all subjects provided informed consent in the original studies. Specifically, the genetic variants that were significantly associated with TBVMs were used as instrumental variables to examine the causal influence of TBVMs on SCZ risk (Figure 1).

The GWAS summary data of TBVMs, including TGMV, TWMV, TCSFV, and TBV, were downloaded from an open resource named the Oxford Brain Imaging Genetics (BIG40) web server (https://open.win.ox.ac.uk/ukbiobank/big40/), which included GWAS summary statistics with 33,224 individuals in the United Kingdom Biobank (Smith et al., 2021). The genome-wide significance threshold was set at p < 5 × 10⁻⁸ in the discovery cohort (N = 22,138) and p < 0.05 in the replication cohort (N = 11,086). Only SNPs that met the significance level in both cohorts were used as instrumental variables in MR analyses, and these SNPs were independent and had no linkage disequilibrium, as described in the original studies (Elliott et al., 2018; Smith et al., 2021). Detailed information about the instrumental variables of TGMV, TWMV, TCSFV, and TBV is shown in Supplementary Tables S1-S4.

GWAS summary data regarding SCZ were downloaded from a meta-analysis provided by the Schizophrenia Working Group of Psychiatric Genomics Consortium (https://www.med.unc.edu/pgc/pgc-workgroups/schizophrenia/), including 36,989 cases and 113,075 controls of predominantly European ancestry without population stratification. In total, 128 significant associations in 108 genetic loci were identified (Schizophrenia Working Group of the Psychiatric Genomics Consortium, 2014).

Comprehensive pleiotropy analyses were performed to assure that instrumental variables met the MR assumptions. First, an MR-Egger intercept test was performed to evaluate the potential pleiotropic associations of the instrumental variables with known and unknown confounders (Bowden et al., 2015; Bowden et al., 2016; Burgess and Thompson, 2017). Second, an MR pleiotropy residual sum and outlier (MR-PRESSO) analysis was carried out to detect horizontal pleiotropy (i.e., MR-PRESSO global test) (Verbanck et al., 2018). Heterogeneity across instrumental variables is also an indicator of pleiotropy. Thus, Cochran’s Q test and I ² statistic were calculated to estimate the heterogeneity (Sun et al., 2021). Specifically, Cochran’s Q test is a conventional test for heterogeneity and approximately follows a chi-square distribution with n-1 degrees of freedom (here, n is the number of instrumental variables). The I ² index is another measure to quantify heterogeneity, which divides the difference between the Q statistic and its degrees of freedom by the Q statistic itself and then multiplies by 100. The value of the I ² index ranges from 0 to 100%, with 0%–25%, 25%–50%, 50%–75%, and 75%–100% representing low, moderate, large, and extreme heterogeneity, respectively (Liu et al., 2013; He et al., 2020). The significance threshold of all the MR-Egger intercept, MR-PRESSO, and Cochran’s Q tests was set at p < 0.05.

The effect alleles of the instrumental variables were adjusted to be associated with increased TBVMs (i.e., the effect estimates of SNPs were larger than zero). Subsequently, the effect alleles of these genetic variants were aligned to be consistent with the effect alleles in the SCZ GWAS dataset. If the instrumental SNPs were not available in the outcome dataset, a proxy SNP that was in high linkage disequilibrium (r ² > 0.8) with the requested SNP was searched instead with the online tool SNiPA (https://snipa.helmholtz-muenchen.de/snipa3/index.php) (Arnold et al., 2015).

The inverse variance weighted (IVW) method was employed to estimate the causal effects of TGMV, TWMV, TCSFV, and TBV on SCZ risk. Specifically, for each TBVM, the effect estimates of each instrumental variable on TBVMs and SCZ were extracted, and Wald estimates and their standard errors were then calculated (Burgess et al., 2017b). The Wald estimates of all the instrumental variables were combined with a weighted mean using inverse variance weights. The significance threshold of the associations between exposures and outcomes was set at p < 0.05.

For each TBVM, the proportion of variance explained by each instrumental variable ( R 2 ) was calculated using the following formula: R 2 = 2 × M A F × ( 1 − M A F ) × β 2 2 × M A F × ( 1 − M A F ) × β 2 + 2 × M A F × ( 1 − M A F ) × N × s e ( β ) 2 where MAF represents the minor allele frequency for a given SNP, β represents the effect size associated with the TBVM for a given SNP, s e ( β ) represents the standard error of the effect size associated with the exposure for a given SNP, and N represents the sample size of the exposure GWAS data.

Then, the strength of instrument variables can be measured by F-statistics, which were calculated based on the following equation: F = R 2 × ( N − 2 ) 1 − R 2 where R ² is the proportion of the variance explained by each SNP, and N represents the sample size of the exposure GWAS data. To minimize weak instrument bias, SNPs with F-statistics > 10 were retained for subsequent analyses (Lawlor et al., 2008).

A series of sensitivity analyses were conducted to validate the robustness of the results. First, four different MR methods including MR-Egger, weighted median, simple mode, and weighted mode methods were performed to estimate the causal effect of TBVMs on SCZ risk. Specifically, the MR-Egger method allows all variants to have pleiotropic effects and can provide a consistent estimate of the causal effect under a weaker instrument strength independent of direct effects (InSIDE) assumption (Burgess and Thompson, 2017); the weighted median method can provide valid causal estimates even if up to 50% of instruments are not valid (Bowden et al., 2016); and the model-based methods (i.e., simple mode and weighted mode) use the causal effect estimates for individual SNPs to form clusters, and the causal effect is estimated in the largest cluster of SNPs (Hartwig et al., 2017b). Second, a leave-one-out sensitivity analysis was carried out to identify SNPs that could potentially bias the causal relationship. To this aim, by sequentially removing each SNP, we estimated the relationship between the remaining SNPs and the risk of SCZ using the IVW method. Finally, reverse causation bias may occur when the outcome variable is at an earlier time point (i.e., the risk of SCZ causally influences the changes of each TBVM). Therefore, we also tested the possibility of reverse causation by treating the risk of SCZ as an exposure and each TBVM as an outcome. Specifically, the instrumental variables were the significant genetic variants associated with SCZ risk, and the same procedures as the main analyses were used to perform reverse MR causality detection.

All statistical analyses were conducted using R version 4.0.4 (R Foundation for Statistical Computing, Vienna, Austria) using the packages of “TwoSampleMR” (Hemani et al., 2018b) and “MR-PRESSO” (Verbanck et al., 2018).

Only two genetic variants without linkage disequilibrium were found to be associated with TGMV, and their summary statistics were extracted from SCZ GWAS data for MR analyses (Supplementary Table S1). Of the five genetic variants associated with TWMV, rs742396 was a palindromic SNP. Thus, we deleted it in the subsequent MR analyses (Supplementary Table S2). Seven genetic variants were associated with TCSFV. All seven instrumental SNPs were located on different chromosomes and were not in linkage disequilibrium with each other. However, rs4843550 is a palindromic SNP and was removed from the subsequent MR analyses. The summary statistics for these TCSFV variants are shown in Supplementary Table S3. Of the five genetic variants associated with TBV, the summary statistics for the four variants could be extracted from the SCZ GWAS data. The SNP rs2732714 was not available in the SCZ GWAS data, therefore, we used the information of its proxy SNP rs113138968, which was in high linkage disequilibrium (r ² > 0.8), to perform the following analyses. All five instrumental SNPs were not in linkage disequilibrium with each other, and none of them were palindromic SNPs. Detailed information about these five instrumental SNPs is shown in Supplementary Table S4.

Both the MR-Egger intercept test and MR-PRESSO test showed no significant pleiotropy for the genetic variants of TBVMs (all ps > 0.05). Furthermore, Cochran’s Q test and I ² statistic revealed no significant heterogeneity for these SNPs (Supplementary Table S5).

We performed a two-sample MR analysis by using genetic variants from TGMV, TWMV, TCSFV, and TBV as instrumental variables. As shown in Table 1, we did not find any causal influence on the risk of SCZ with the IVW method (p > 0.05).

The explained variances (R ²) and F-statistics of each instrumental variable are shown in Supplementary Tables S1-S4, and the F-statistics of each instrumental variable were larger than 10, indicating no weak instrumental bias among these variables.

All other MR approaches, including the MR-Egger, weighted median, simple mode, and weighted mode methods, did not identify any significant causal effects of TGMV, TWMV, and TBV on the risk of SCZ (Table 1). Although TCSFV was found to be causally associated with the risk of SCZ when using the MR-Egger method (BETA = 0.646, SE = 0.220, p value = 0.042), this result was not validated by other methods. In leave-one-out sensitivity analyses, no genetic variants could significantly affect the MR estimates (Figure 2). For the reverse MR causality analysis, 111 leading SNPs associated with SCZ risk were extracted from the GWAS summary data of TBVMs. Among them, ten palindromic SNPs were removed, and the remaining 101 SNPs were retained for subsequent analyses (Supplementary Table S6). All the MR methods indicated that there was no causal influence of any TBVM on SCZ risk (Supplementary Table S7).