Lack of Causal Effects or Genetic Correlation between Restless Legs Syndrome and Parkinson's Disease

Background: Epidemiological studies have reported association between Parkinson's disease (PD) and restless legs syndrome (RLS). Objectives: We aimed to use genetic data to study whether these two disorders are causally linked or share genetic architecture. Methods: We performed two-sample Mendelian randomization (MR) and linkage disequilibrium score regression (LDSC) using summary statistics from recent genome-wide meta-analyses of PD and RLS. Results: We found no evidence for a causal relationship between RLS (as the exposure) and PD (as the outcome, inverse variance-weighted; b=-0.003, se=0.031, p=0.916, F-statistic=217.5). Reverse MR also did not demonstrate any causal effect of PD on RLS (inverse variance-weighted; b=-0.012, se=0.023, p=0.592, F-statistic=191.7). LDSC analysis demonstrated lack of genetic correlation between RLS and PD (rg=-0.028, se=0.042, p=0.507). Conclusions: There was no evidence for a causal relationship or genetic correlation between RLS and PD. The associations observed in epidemiological studies could be, in part, attributed to confounding or non-genetic determinants.


Introduction
Restless legs syndrome (RLS) and Parkinson's disease (PD) are common neurological disorders with a prevalence of 1.9-4.6% and 0.1-2.9% in Europeans, respectively. 1,2 Epidemiological studies suggest that RLS is more common than expected in PD patients, and PD affects RLS patients more frequently than matched controls or the general population. 3 Some studies suggest that RLS may be an early clinical manifestation of PD, [4][5][6] whereas other studies found no association between RLS and PD. 3 A recent meta-analyses showed a higher odds for RLS in PD patients compared to controls. 3 In this study, the previous contradictory results were explained by different inclusion and diagnostic criteria and differences in sex distribution. 3 However, there are major differences between RLS and PD including clinical, ultrasonographic, functional and neuroimaging aspects, which do not support an association between RLS and PD. [7][8][9][10] Therefore, the true nature of the association between RLS and PD remains unclear.
Mendelian Randomization (MR) may help mitigate some of the bias introduced by reverse causation and confounding in traditional observational studies. 11 In addition, genetic correlation using linkage disequilibrium (LD) score regression (LDSC) may help determine whether different traits have overlapping genetic background, which may explain some of the observed associations between traits.
Here, we used bidirectional MR and LDSC to seek evidence for a causal relationship and/or shared genetic architecture between RLS and PD.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Study population and genetic data
To perform MR and LDSC, we used summary statistics from two recent genome-wide association study (GWAS) meta-analyses of RLS and PD. 12,13 The RLS summary statistics included data from 15,126 patients and 95,725 controls, 12 and the PD summary statistics included data from 33,674 cases (15,056 PD patients and 18,618 proxy-cases), and 449,056 controls. 13 A subset of data (23andMe data) was not included in the PD summary statistics to avoid potential overlap with the RLS data which included 23andMe data. 23andMe participants provided informed consent and participated in the research under a protocol approved by the external AAHRPP-accreted IRB, Ethical & Independent Review Services (E&I Review). The full GWAS summary statistics for the 23andMe discovery data set will be made available through 23andMe to qualified researchers under an agreement with 23andMe that protects the privacy of the 23andMe participants. Please visit https://research.23andme.com/collaborate/#dataset-access/ for more information and to apply to access the data. Information on recruitment procedures and diagnostic criteria is detailed in the original publications. 12,13 All cases and controls in this study were of European ancestry.

Power calculation
Power was calculated for detecting an effect size of odds ratio of 1.2 on RLS and PD risk, using online sample size and power calculator for Mendelian randomization with a binary outcome (https://sb452.shinyapps.io/power/). 14 For all analyses the power was estimated at >80%.

Mendelian randomization
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) We performed bidirectional MR, i.e. examining whether RLS is a causal risk factor (exposure) for PD (outcome) and if PD is a causal risk factor for RLS. For each MR analysis, we constructed multi-variant instruments from the independent ("index") GWAS significant SNPs (p<=5e-08) from the exposure GWAS. In brief, index SNPs were obtained by clumping all GWAS significant SNPs within each LD block using an R 2 threshold of 0.001 or a distance of 10,000 kb from the index SNP. This process increased the independence of each index SNP based on the above parameters. Additional details regarding the instrument construction and the code used for the analysis are available at https://github.com/gan-orlab/MR_LDSC_RLS-PD.
To calculate the proportion of variability in the exposure explained by the SNPs and to test the strength of the instrument variables (IVs), we used the statistical power for MR analyses (the coefficient of determination, R 2 ) and F-statistics tests, as previously described. 15 In order to perform MR, an estimate of the individual effect of SNPs on the exposure and outcome (RLS and PD, interchangeably) was used to calculate the Wald ratio. Then, the effect estimates were combined using the Inverse-variance weighted (IVW) method, which is a weighted mean of the Wald ratio estimates obtained from each individual SNP separately. 16

Sensitivity analyses
To explore whether IVW results might be biased due to violations of MR assumptions and to evaluate the robustness of the results, we used weighted median and MR Egger 16 estimators as sensitivity analyses. The weighted median estimate provides a reliable pooled estimate assuming that at least half of the weight of the SNPs in the instrument are valid. MR Egger assesses directional pleiotropy similarly to the IVW approach except that the regression slope y-intercept is not constrained to pass through the origin. For each approach, we constructed funnel plots to detect outliers. We evaluated the heterogeneity statistics Q for IVW and Q′ for MR-Egger.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)  17 was used to examine outlier SNPs which might occur in the presence of horizontal pleiotropy and correct pooled estimates. Steiger filtering was used to discard SNPs that explain more variance in the outcome than in the exposure. To find all the SNPs that are in LD with the index SNP, the LDmatrix module on the LDlink web tool was used. 18

Genetic correlation analyses
To assess the genetic correlation between RLS and PD, we performed LDSC after computing zscores and formatting data from the two GWASs as previously described. 19,20 In brief, LDSC calculates genetic correlation between two traits by incorporating LD scores (the more variants in LD with a SNP, the higher the LD score) and GWAS summary statistics (z scores) in a regression model.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Results
In total, 20 and 55 index SNPs for RLS and PD, respectively, were initially used as IVs for exposure. These IVs explain 3.5% and 2.1% of the risk in RLS and PD. All SNPs were strong instruments for MR analysis as measured by F-statistics (RLS F-statistics=217.5; PD F-statistics=191.7). There was no overlap between the genes where the clumped SNPs are located in both meta-analyses (Supplementary Table 1).
We then performed MR analyses to assess the bidirectional causal relationship between RLS and PD. RLS, as the exposure, was not causally associated with PD (IVW; b=-0.051, se=0.037, p=0.172). However, the p values of IVW-Q and MR Egger-Q′ tests were 0.034 and 0.025, respectively, raising the possibility of pleiotropic SNP(s) in our dataset, which violates MR assumptions. MR-PRESSO 17 was applied and a pleiotropic index SNP, rs11860769 (p=0.02) was identified when RLS was used as exposure. This SNP has an opposite effect on risk of RLS and PD as was previously shown. 21 After removing the pleiotropic index SNP (rs11860769), 19 index SNPs were used as IVs for RLS, respectively. Again, there was no causal effect of RLS on PD (b=-0.003, se=0.031, p=0.916) or of PD on RLS (b=-0.012, se=0.023, p=0.592) with 55 IVs, and the results of sensitivity analyses suggested that there were no additional deviations from the MR assumptions (Table 1, Figure 1, Supplementary Figure 1,2).
We then sought to examine whether there is genetic correlation between RLS and PD that may explain the overrepresentation of these disorders in one another. There was no genetic correlation between RLS and PD (rg=-0.028, se=0.042, p=0.507).
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Discussion
Our findings suggest lack of causal relationship between RLS and PD, and lack of genetic correlation. One locus on chromosome 16, including the genes TOX3 and CASC16, is pleiotropic with opposite direction of effect, as SNPs associated with increased risk of RLS are associated with reduced risk of PD, as previously reported. 21 Therefore, this locus also cannot explain the observed increased frequency of PD in RLS and of RLS in PD.
Although RLS and PD co-occur at a rate higher than expected and share several traits with PINK1 mutations in the same cohort did not show RLS features. 29 In a study of 258 RLS patients vs 235 healthy controls, the authors reported that the SNCA Rep1 allele was associated with reduced risk of RLS. 30 However, this association was not replicated by the much larger RLS GWAS. 12 Overall, genetic studies, including the current study, do not support a genetic overlap between RLS and PD.
Our study has some limitations. We could not exclude PD patients with RLS and RLS patients with PD in the datasets used for this analysis, which would have made the results cleaner, since these data was not available. In addition, this study focused on individuals of European ancestry. Studies from multiple ethnicities are required to further study PD, RLS and the association between them. It is possible that rare or structural variants outside of what can be detected with current GWAS technologies are contributing to a shared genetic etiology.
In light of the current and previous findings, it is likely that confounding factors such as treatment, closer neurological follow up and others may have contributed to the observed epidemiological association between RLS and PD. While additional studies are required to identify these potential confounders, the observed association between RLS and PD should not be considered causal on current evidence.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)  Supplementary Tables and Figures   Supplementary Table 1 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted February 20, 2021. ; https://doi.org/10.1101/2021.02.16.21251687 doi: medRxiv preprint