Figure 1 delineates the methodology of our two-sample MR analysis, exploring the link between T2DM and iRBD. Initially, we employed LDSC to assess the genetic correlation between these two conditions. Following this, with T2DM as the exposure factor and iRBD as the outcome factor, we conducted a univariate Mendelian Randomization analysis (UVMR) to investigate their gene-driven causal relationship. To further corroborate this result, UVMR was again utilized, to examine the connection between three glucose metabolic traits (blood glucose, glycated hemoglobin, and fasting insulin levels) and iRBD. The next phase of our study involved a multivariable MR (MVMR) method, integrating confounders like Body Mass Index (BMI), smoking habits, and educational level, to validate the robustness and consistency of the T2DM-iRBD association. Lastly, our analysis extended to a network MR approach with cytokines, immune-related circulating protein levels, and immune cell characteristics as potential mediators, delving into the underlying mechanisms of T2DM leading to iRBD. Our methodology adheres rigorously to the STROBE-MR guidelines (28).

To strengthen causality inference and reduce bias from small sample sizes, we carefully selected the most comprehensive GWAS datasets for T2DM and iRBD available from the GWAS Catalogue (https://www.ebi.ac.uk/gwas/), focusing on individuals of European descent. The T2DM GWAS dataset, specifically, was derived from a meta-analysis conducted by Angli Xue’s team, combining data from MAGIC, DIAGRAM, and GERA, encompassing a total of 659,316 participants (29). For the outcome data, we chose a GWAS dataset of an iRBD cohort (1,061 cases and 8,386 controls) compiled by the International RBD Study Group (16). The iRBD cases were diagnosed using the International Classification of Sleep Disorders (2nd or 3rd Edition), inclusive of video polysomnography assessments. To ensure better comparability between the case and control groups, principal components analysis was used to adjust for population substructure, accounting for variables like gender and age. Data concerning metabolic traits related to blood sugar, confounders (BMI, education years, smoking status) and potential mediators (91 plasma cytokines, 106 immune-related circulating proteins, 731 immune cell characteristics) were sourced from the IEU online database (https://gwas.mrcieu.ac.uk/) and the GWAS Catalogue database ( Table 1 ). All the aforementioned GWAS data are publicly accessible. Importantly, we retrieved exposure and outcome data from independent samples to avoid bias stemming from sample overlap.

In our study, we employed single-variate LDSC to estimate the heritability of both T2DM and iRBD. Additionally, to clarify the genetic correlations between these two conditions, we utilized the bivariate LDSC method. (30–34).

Genetic variants were selected as instrumental variables based on strong associations with the exposure (P < 5 × 10^-8). To mitigate reverse causation bias, we excluded SNPs highly associated with the outcome ( P   <   5 × 10 − 5 ) (35). To quantify the proportion of variance in exposure, the R 2   clue of single nucleotide polymorphisms (SNPs) was estimated by effect estimates (beta) and allele frequencies (EAF): R 2 =   2   ×   E A F   ×   ( 1   −   E A F ) ×   b e t a 2 (36). The instrument strength was estimated using the F-statistic: F   = R 2 ( 1 − R 2 ) × ( N   −   k   −   1 ) k , where N represents the sample size and k is the number of instrumental variables (37). Only the SNP of F-statistic > 10 could be included in this study. A minor allele frequency (MAF) ≥0.01. We clumped independent SNPs based on European ancestry reference data (1000 Genomes Project, r²>0.001, genomic region=10,000 kb). Summary statistics were harmonized on alleles positively associated with exposures. Ambiguous palindromic SNPs (A/T, C/G) with a MAF>0.3 were excluded (38). To fulfill the independence assumption for MR analysis, SNPs associated with various factors, including BMI, smoking, alcohol consumption, years of schooling, occupation, head injury history, olfactory dysfunction, antidepressant drug usage, and antipsychotics usage (16–18, 39–41), were excluded based on the PhenoScanner query results. SNPs associated with confounding factors were displayed in Supplementary Table S4 . The statistical power of the Mendelian randomization analysis was computed using the tool available at https://shiny.cnsgenomics.com/mRnd/ (42).

The primary analysis utilized the inverse-variance weighted (IVW) method (43), benchmarked against other MR methods robust to pleiotropy, including the maximum likelihood method (44), weighted median (45), weighted mode (46), MR-Egger regression (47) simple mode (48) and the MR-PRESSO (49). We initially investigated the genetically driven causal relationship between T2DM as the exposure and iRBD as the outcome. Subsequently, to validate our findings, we delved into the causal connections between three key blood glucose metabolic traits—blood glucose levels, glycated hemoglobin levels, and fasting insulin levels—and iRBD. Although MR effectively reduces confounding bias, it does not completely eliminate its influence. If the instrumental variables used in the analysis are associated with known risk or protective factors for the outcome, they violate the second assumption of MR, leading to biased results. Therefore, we employed MVMR to adjust for instrumental variables closely related to BMI, years of education, and smoking habits (16–18, 39–41), in order to assess the robustness of the relationship between T2DM and iRBD.

To investigate the potential immune-inflammatory mechanisms by which T2DM affects iRBD, we selected 91 plasma cytokines, 106 immune-related circulating proteins, and 731 immune cell characteristics as candidate mediators and conducted a network MR analysis—an MR-based mediation analysis. We sequentially calculated the total effect of T2DM on iRBD, the direct effect of T2DM on the mediators, and the direct effect of the mediators on iRBD. The magnitude of the mediation effect was estimated using the product-of-coefficients method, and the confidence interval was determined using the Delta method (50).

Cochran’s Q statistic assessed heterogeneity, serving as a potential indicator of pleiotropy in IVW estimators. Horizontal pleiotropy was identified through the p-value for the intercept in MR-Egger (47) and the global test in MR-PRESSO (49). MR-PRESSO also pinpointed outliers, prompting a reanalysis of all MR methods after outlier SNP exclusion. Leave-one-out analyses determined if a single SNP unduly influenced estimates.

All statistical tests were two-sided, a P value of less than 0.05 was considered statistically significant. All above statistical analyses were conducted by using the R-statistical software (version 4.3.1) with related R packages.

Supplementary Table S2 details the selected genetic instruments for our MR analyses. These genetic variants were strongly associated with their respective exposures, explaining a variance that ranged from 0.26% to 21.81%. Importantly, all instrumental variables had F-statistics exceeding 10, indicating that our results are unlikely to be affected by weak instrument bias. In terms of statistical power within our MR analysis, except for the exposure factor of fasting insulin with a power value of 0.86, all other exposures had power values of 1, signifying robust statistical power to detect potential causal relationships.

Bivariable LDSC analysis indicated a genetic correlation between T2DM and iRBD (rg = 0.306, P = 0.029;, Table 2 ; Supplementary Table S1 ). The UVMR analysis corroborated a gene-driven causal relationship, with an odds ratio (OR) of 1.19 (95% CI: 1.03-1.37; P = 0.017, Figures 2A, B ). Subsequent secondary analyses using various methods supported these findings: maximum likelihood yielded an OR of 1.19 (95% CI: 1.03, 1.38, P = 0.016), weighted median produced an OR of 1.27 (95% CI: 1.02, 1.57, P = 0.031), simple mode estimated an OR of 1.76 (95% CI: 1.05, 2.96, P = 0.035), and MR-PRESSO calculated an OR of 1.19 (95% CI: 1.04, 1.27, P = 0.016). Notably, all analyses excluded the instrumental variable ‘rs13239186’ due to its strong association with smoking as identified in the PhenoScanner database, to avoid confounding bias ( Supplementary Tables S2 , 4 ). Sensitivity analyses, using Cochran’s Q statistic, showed no heterogeneity in the MR results. Furthermore, MR-Egger’s intercept and MR-PRESSO’s global test indicated no evidence of horizontal pleiotropy. MR-PRESSO also detected no outlier effects ( Supplementary Table S5 ). Leave-one-out analysis suggested that the results were not driven by any single SNP ( Figure 2C ; Supplementary Table S6 ). Reverse Mendelian randomization did not support the existence of reverse causation ( Supplementary Table S7 ).

In our UVMR analysis examining the relationship between three blood sugar metabolism factors and iRBD, we discovered that elevated blood glucose levels increase the risk of iRBD, with an OR of 1.55 (95% CI: 1.04, 2.30, P = 0.03). Secondary analyses using Maximum likelihood and MR-PRESSO corroborated these findings, with both methods indicating an OR of 1.55 (95% CI: 1.04, 2.31, P = 0.03 for Maximum likelihood; 1.55, 95% CI: 1.04, 2.30, P = 0.03 for MR-PRESSO). Significantly, our analyses systematically excluded the instrumental variables ‘rs8047587’ and ‘rs34402524’, due to their strong associations with alcohol consumption and years of education, respectively ( Supplementary Table S4 ). There was no evidence to suggest a causal link between HbA1c or fasting insulin levels and iRBD. Sensitivity analysis indicated heterogeneity in the relationship between HbA1c levels and iRBD (Q-value=0.01), and potential horizontal pleiotropy (Global P-value =0.01). The remaining analyses showed no disturbances from heterogeneity, horizontal pleiotropy, or outliers ( Table 3 ; Supplementary Tables S8 , 9 ).

Study conclusions were supported by weighted-median estimation, weighted-mode, and MR-Egger methods ( Supplementary Table S2 ). Cochran’s Q statistic indicated no significant heterogeneity in SNP effects (p > 0.05). No evidence of potential horizontal pleiotropy was detected for the eight factors identified (p > 0. 05). To further assess the robustness of the results, we conducted MR-PRESSO tests on the included SNP loci. In addition, a leave-one-out sensitivity analysis was conducted to assess the influence of each SNP on the overall causal relationship. The results demonstrated that systematically removing individual SNPs and repeating the MR analysis did not reveal significant differences in the observed causal relationships ( Supplementary Tables S3 – 5 ).

The results of the MVMR suggest that T2DM remains an independent risk factor for iRBD, even after adjusting for potential confounders such as years of education, smoking status, and BMI. This was demonstrated through various analytical methods: IVW-OR at 1.230 (95% CI: 1.031, 1.467, P = 0.022), Lasso-OR at 1.176 (95% CI: 1.003, 1.379, P = 0.046), MR-Egger-OR at 1.208 (95% CI: 1.011, 1.442, P = 0.037), and Weighted Median-OR at 1.278 (95% CI: 1.008, 1.619, P = 0.043) as shown in Figure 3 and Supplementary Table S10 . Sensitivity analysis revealed no horizontal pleiotropy in the multivariable MR analyses, as indicated by MR-Egger regression (MR Egger intercept P value = 0.092).

In our network MR analysis ( Table 4 ), we identified significant associations: T2DM was observed to decrease plasma levels of Stromal Cell-Derived Factor 2 (SDF-2) (P=8.03E-03), with lower SDF-2 levels elevating the risk of iRBD (P=1.75E-03). The mediation effect of SDF-2 in the T2DM-iRBD relationship accounted for 7.03% (95% CI 0.11%–7.70%). Furthermore, T2DM was associated with an increase in BAFF-receptor (BAFF-R) on IgD- CD38- B cells (P = 3.08E-03), a characteristic that also heightened the risk of iRBD (P = 7.43E-04). The mediation effect of this immune characteristic in the T2DM-iRBD causal pathway was 11.17% (95% CI 6.22%–11.65%). Notably, none of the network MR analyses indicated the presence of heterogeneity and horizontal pleiotropy ( Supplementary Tables S11 , 12 ).