A two-sample univariable bidirectional MR analysis was initially performed to investigate the causal effects between PDs and OSA. Recognizing that BMI, smoking, and alcohol consumption are significant risk factors for OSA development, multivariable MR (MVMR) analyses were subsequently conducted for outcomes causally linked to OSA. These analyses aimed to ascertain whether the exposures have independent causal effects on OSA. To ensure the reliability of MR results, three critical assumptions must be met: (i) the IVs must be strongly correlated with the exposures, (ii) the IVs should not be significantly associated with confounding factors, and (iii) the IVs must influence the outcomes solely through the exposures (25). The research design is depicted in Figure 1 , created using Figdraw (www.figdraw.com).

Genome-wide association studies (GWAS) aim to identify relationships between differences in DNA sequences and specific traits by examining individuals with diverse phenotypes and analyzing their genotypes at several single-nucleotide polymorphisms (SNPs) locations (26). As data-sharing has advanced, GWAS has discovered several significant variants related to PDs and OSA. These variations provide valuable IVs for conducting a robust MR analysis to investigate the causal relationship between PDs and OSA. To mitigate the risk of pleiotropic bias in cross-lineage cases, this study incorporated publicly available GWAS results of five PDs, and OSA conducted within a European population.

The most recent GWAS data for OSA was obtained from the FinnGen (27), which included a total of 38,998 patients and 336,659 controls. The patients are identified using the Finnish National Hospital Discharge Registry and the Causes of Death Registry. OSA was diagnosed depending on the International Statistical Classification of Diseases and Related Health Problems, 9th edition (ICD-9: 3472A) and 10th edition (ICD-10: G47.3), which are based on subjective symptoms, clinical examination, and sleep registration, utilizing the apnea-hypopnea index of five per hour or the respiratory event index of five per hour. All association tests for all sources were adjusted for principal variables, including age and sex. The Psychiatric Genomics Consortium (PGC) provided summarized GWAS data on the five PDs:

The GWAS data for MDD was obtained from Howard DM et al.’s GWAS meta-analysis (28). It consists of PGC, UK Biobank, and 23andMe, three large-scale GWAS. We included the complete summary statistics from 2 cohorts to conduct bi-directional MR analysis because, among these three GWAS, the summary statistics for all examined genetic variants were only publically available for UK Biobank and PGC (which contains 170,756 cases and 344,901 controls). Howard DM et al. employed the inclusive criteria for MDD within UK Biobank, which was determined based on participants’ answers to the questions ‘Have you ever sought medical help from a general practitioner or psychiatrist for nerves, anxiety, tension, or depression?’ MDD was diagnosed in PGC cohorts using internationally recognized criteria, such as DSM-IV, ICD-9, or ICD-10.

The GWAS data for ANX were obtained from the GWAS meta-analysis conducted by Otowa T et al. (29), which included 7 distinct studies with a combined sample size of 5580 cases and 11730 controls. The identification of ANX primarily relied on the Diagnostic and Statistical Manual of Mental Disorders (DSM) and utilized standardized evaluation tools (30).

The GWAS data for BIP were obtained from a meta-analysis of 57 cohorts from Europe, North America, and Australia undertaken by Mullins N et al. (31). These cohorts included a total of 41,917 cases and 371,549 controls in European heritage. Cases that satisfied the global consensus criteria (DSM-IV, ICD-9, or ICD-10) for BIP were determined through formal diagnostic interviews, clinician checklists, or medical records reviews.

The largest GWAS meta-analysis, carried out by Trubetskoy V et al. (32), yielded the GWAS data for SCZ. The SCZ samples were from several SCZ studies. The DSM-IV or ICD-10 diagnostic criteria were used to make the diagnoses. For the purpose of performing a bi-directional MR study, we only included the SCZ meta-analysis in participants of European ancestry. This analysis included 53,386 cases and 77,258 controls.

The GWAS data for PTSD were obtained from the GWAS meta-analysis of an ancestrally diverse group from 60 different PTSD studies carried out by Nievergelt CM et al. (33), ranging from large cohorts with self-reported PTSD symptoms to clinically deeply characterized small patient groups. The diagnosis of PTSD was determined using a variety of instruments and DSM versions (DSM-III-R, DSM-IV, DSM-5). To conduct a bi-directional MR analysis, we exclusively included the PTSD meta-analysis in participants of European ancestry, which included 23,212 cases and 151,447 controls.

Additionally, summary-level data for BMI, smoking, and alcohol intake were acquired from corresponding publicly available GWAS. The Genetic Investigation of Anthropometric Traits (GIANT) Consortium, which includes 681,275 samples with European ancestry, provided the GWAS data for BMI (34). The GWAS data on alcohol consumption and smoking initiation were taken from the GWAS meta-analysis conducted by the Sequencing Consortium of Alcohol and Nicotine Use (GSCAN) and the GWAS, which included up to 1.2 million people with European ancestry (35). The binary phenotype of smoking initiation indicates whether a person has ever smoked regularly throughout their life. The average number of drinks a person reported having each week, regardless of the type of alcohol, was used to characterize alcohol consumption. As all original studies were public, anonymized, and de-identified, ethical approval did not apply to this study. Table 1 provides further details about the GWAS dataset included in this analysis.

A specific methodology was employed to select IVs to adhere to the three core assumptions of MR analysis. Genome-wide significant SNPs (P<5×10⁻⁸) were initially extracted from exposure GWAS. For the GWAS of ANX and PTSD, suggestively significant SNPs (P <5×10⁻⁶) were also used as IVs to ensure comprehensive results. Subsequently, linkage disequilibrium (LD) clumping (R² < 0.001, clumping distance = 10,000 kb), based on the European 1,000 Genomes Project reference panel, was performed to select independent significant SNPs. The harmonization process involved removing palindrome SNPs with intermediate or incompatible allele frequencies. The statistical strength of each SNP was then evaluated by calculating the F-statistic (beta²/se²) (36, 37), with an F-statistic > 10 indicating the absence of bias due to weak IVs. Additionally, MR pleiotropy residual sum and outlier (MR-PRESSO) test analysis were conducted to eliminate outlier IVs. A stringent filtering step was adopted, discarding SNPs with a P-value < 1 in the outlier test and repeating this process until no outliers remained (38). Finally, the Steiger test was applied to each SNP to ascertain if the R² of the exposure exceeded that of the outcome, excluding SNPs where the test indicated a ‘FALSE’ direction.

This study used the inverse-variance weighted (IVW) as the primary analytical approach to assess the causal relationship between PDs and OSA. This method aggregates Wald ratio estimates for each instrumental SNP using a meta-analysis-like model to derive overall effect estimates (39). Given that IVW yields unbiased causal estimates only when all IVs are valid, additional methods were employed to ensure accuracy and stability in cases of assumption violations. These methods included weighted mode, weighted median, and MR-Egger. The weighted median method can offer a consistent estimate even if up to 50% of the IVs are invalid, albeit with reduced statistical power (40). Under the Instrument Strength Independent of Direct Effect (InSIDE) assumption, MR-Egger regression allows all genetic variants to be potentially invalid IVs. It provides an intercept term for estimating average pleiotropic effects across genetic variants (41). Despite stringent control measures, the MR-PRESSO method might still detect horizontal pleiotropy in the global test, potentially due to InSIDE assumption violations (42). Therefore, if the MR-PRESSO international test detected pleiotropy or if the MR-Egger intercept’s P-value was below 0.05, results from the weighted median method were prioritized. The weighted mode approach is applicable even when most IVs do not fulfill the causal inference prerequisites in MR as long as most IVs yield consistent causal estimations (41).

A series of essential sensitivity assessments were conducted to evaluate the results’ reliability and stability. The Cochran Q statistic (MR-IVW) was employed to assess heterogeneity among the effect sizes derived from the selected genetic IVs (43). A P-value greater than 0.05 was indicative of negligible heterogeneity. The MR-PRESSO method and MR-Egger intercept were then applied to detect horizontal pleiotropy (42). Furthermore, a leave-one-out analysis addressed potential biases due to horizontal pleiotropy arising from any SNP. For multiple testing adjustments, a Bonferroni-corrected P-value threshold was set at 0.05/5, while a P < 0.05 was considered nominally significant. Additionally, reverse MR analysis was conducted to explore the possibility of reverse causality, precisely the effect of OSA on PDs.

Two distinct methodologies were implemented to mitigate the influence of potential confounders. The first step involved the identification and exclusion of pleiotropic SNPs associated with known confounders like BMI, smoking, and alcohol consumption. This identification was conducted using the PhenoScanner V2 database (44). After removing SNPs linked to these confounding variables, a reanalysis was carried out using the remaining IVs. Additionally, MVMR analysis was used. An improved version of MR named MVMR enables the simultaneous analysis of several genetic variations linked to various exposures (45). This method is advantageous in discerning whether the exposures under study have independent effects on the outcome, in this case, OSA. In the second approach, MVMR analysis was explicitly used to determine if these exposures had independent causal effects on OSA. These analyses were conducted using various packages in R Software 4.3.0, including ‘TwoSampleMR’ (version 0.5.7), ‘MendelianRandomization’ (version 0.8.0), and ‘MRPRESSO’ (version 1.0).

A thorough quality control process led to the selection of 41, 6, 38, 137, and 20 SNPs as IVs for MDD, ANX, BIP, SCZ, and PTSD, respectively, to evaluate their causal effect on OSA. Conversely, in the reverse MR analysis, 11, 14, 14, 16, and 18 SNPs were chosen as IVs for OSA to assess its causal impact on the five PDs. The F-statistic for each selected SNP was greater than 10, indicating the absence of weak instrument bias. Detailed information about the IVs ultimately included in our MR analysis can be found in Supplementary Tables S1–S10 .

The univariable MR analysis revealed that genetically predicted MDD was associated with an increased risk of OSA (IVW OR: 1.377, 95% CI: 1.242–1.526, P = 1.05×10^-9). This association was corroborated by the weighted median method (OR: 1.301, 95% CI: 1.135–1.491, P = 1.576×10^-4). Although MR-Egger and weighted mode methods provided consistent estimates, they were not statistically significant. No heterogeneity was observed in Cochran’s Q test (P = 0.084), and both the MR-Egger intercept (P = 0.480) and MR-PRESSO test (P = 0.114) indicated an absence of significant directional pleiotropy for the IVs. Leave-one-out analyses further confirmed that individual SNPs did not influence the causal relationship. For SCZ, despite stringent controls and no outliers detected pre-MR analysis, the MR-PRESSO global test indicated pleiotropy among IVs (P = 0.037), suggesting unreliability of causation estimates from IVW and MR-Egger due to violation of the InSIDE assumption (38). However, estimates from the weighted median method were also non-significant (OR: 0.997, 95% CI: 0.964–1.032, P = 0.884). MR analysis indicated no causal effect of genetically predicted ANX, BIP, and PTSD on OSA. Cochran’s Q test detected no heterogeneities except for PTSD (P = 0.042), and no notable directional pleiotropies were observed in the MR-Egger intercept and MR-PRESSO test. Detailed results of the causal effects of PDs on OSA, including outcomes of the four causal estimation methods, pleiotropy tests, and heterogeneity tests, are presented in Table 2 . The scatter, funnel, and forest plots are shown in Supplementary Figure S1 .

To elucidate the precise causal relationship between PDs and OSA, a reverse MR analysis was conducted, positioning OSA as the exposure and PDs as the outcomes. However, no significant causal effects of OSA were found on five PDs across all MR methods. Additionally, no heterogeneities or directional pleiotropies were observed in the reverse MR analysis. Detailed results of the causal effects of OSA on PDs, encompassing outcomes from the four causal estimation methods, pleiotropy tests, and heterogeneity tests, are provided in Supplementary Table S11 . The associated scatter and forest plots are illustrated in Supplementary Figure S2 .

Employing the online tool PhenoScanner V2, the study identified that rs10235664, rs17641524, rs2568958, rs30266, and rs9831648 correlated with BMI, while rs2232423 and rs4799949 linked to smoking, and rs4936276 and rs66511648 were associated with alcohol consumption. Upon excluding these SNPs from the IVs, the significant causal relationship between MDD and OSA remained intact (OR: 1.373, 95%CI: 1.230–1.532, P = 1.60×10^-8). No considerable heterogeneity or pleiotropy was observed ( Supplementary Table S12 ). Subsequently, an MVMR analysis was performed to investigate MDD’s direct effect on OSA, adjusting for BMI, smoking, and alcohol consumption. The MVMR analysis demonstrated that MDD still exerted an immediate impact on OSA. Detailed results of the MVMR analysis are presented in Table 3 .