Restless legs syndrome (RLS) is a sleep-related sensorimotor disorder characterized by an urge to move the legs during periods of rest, especially in the evening and night1. The high narrow-sense heritability (~70%) of RLS estimated by twin studies, as well as the strong familial aggregation, have motivated genetic studies of this condition2,3. The latest genome-wide association studies (GWAS) meta-analysis identified a total of 19 RLS risk loci that explained 60% of the estimated SNP-based heritability of 19.6%4. Using the summary statistics of the most recent RLS-GWAS cohort of European ancestry, various post-GWAS approaches including gene annotation, pathway and gene-set enrichment analyses were applied to prioritize genes in associated loci and identify related biological mechanisms. However, the candidate causal genes at these loci have yet to be completely clarified.

Unlike the previous methods that often associate loci with the nearest gene or focus on individual significant SNP and eQTL associations, transcriptome-wide association studies (TWASs) focus on whole expression and trait associations rather than only top eQTL associations5. As a complementary to the previous GWAS, this study aimed to identify novel genes associated with RLS that are not well explained by individual SNP tagging. Through leveraging available transcriptomic imputation approaches, FUSION5 and S-PrediXcan6, we sought to integrate eQTL analyses with summary-level GWAS data to determine more detailed information on, and discover novel genes, underlying the pathology of RLS.


Study overview

TWAS was performed using the summary statistics of the largest GWAS meta-analysis at the time of the analysis that contained 15,126 RLS patients and 95,725 control participants in the discovery stage (Fig. 1). Two different approaches, FUSION5 and S-PrediXcan6, were applied for transcriptomic imputation using 17 tissue panels and 67,156 gene models (Supplementary Table 1) (see “Methods”).

Fig. 1: Transriptome-wide association study workflow for restless legs syndrome.
figure 1

A total of 15,126 cases and 95,725 controls of European ancestry from EU-RLS-GENE, INTERVAL, and 23andMe cohorts were used in the discovery stage of the meta-analysis. Transcriptomic imputation was performed using S-PrediXcan and FUSION. A fine-mapping approach was conducted using FOCUS. Gene-set enrichment and phenome-wide association studies were done using EnrichR, GWASAtlas and TWAShub, respectively.

TWAS genes for RLS

The top TWAS genes with the strongest associations with a |Z-score| > 3.50, P < 5 × 10−4 are listed in Supplementary Table 2 (with the results from the multi-tissue TWAS presented in Fig. 2). The majority of the identified associations encompass the genes that are in the 1p13.3 (S100A16, S100A2, S100A3, and S100A3), 2p14 (MEIS1 and PPP3R1) and 15q23 (SKOR1, MAP2K5, IQCH, IQCH-AS1, AAGAB, RP11-34F13.2, and TMEM87A) (Fig. 2).

Fig. 2: Transcriptome-wide association study genes whose differential expression in relevant tissues was associated with restless legs syndrome.
figure 2

Ribbons link the genes to the segments that represent the relevant tissues. Gene groups that belong to the different GWAS loci and their relevant tissues were coloured differently. S100A16, S100A2, S100A3, and S100A3: yellow; DOCK7: light blue; MEIS1 and PPP3R1: dark red; CCDC148: brown; DLX1: dark blue; CMTR1 and RP1-153P14.5: green; TAC1: olive; AF131215.2: gold; KCTD9: purple; GSTO: pink; AP000462.3: orange; CTD-2292M16.8: violet; RP5-1021I20.1 and RP5-1021I20.2: red; SKOR1, MAP2K5, IQCH, IQCH-AS1, AAGAB, RP11-34F13.2, and TMEM87A: pale pink; RP11-645C24.5: khaki; SKAP1: goldenrod; ZGPAT and RGS19: green; JAM2: grey. Gene models were visualized using the online version of Circos (

Transcriptome-wide significant hits

A total of seven gene-level models (consisting of five unique genes) reached transcriptome-wide significance (P = 7.45 × 10−7) after Bonferroni correction for 67,156 total tests (Fig. 3). In addition, six significant splicing events were identified using CommonMind Consortium (CMC) dorsolateral prefrontal cortex RNA-seq data (Supplementary Table 3). Among the total 13 significant associations, six of them were not reported in the previous GWAS. These are SKAP1 and the splicing events that were detected in the SLC36A1, CCDC57, FN3KRP, and NCOA6/TRPC4AP genes.

Fig. 3: Manhattan plot of restless legs syndrome (RLS) transcriptome-wide association study (TWAS) gene models.
figure 3

Green points represent signals from FUSION, pink represents signals from S-PrediXcan, and blue represents TWAS genes that were prioritized as putative candidates for RLS via statistical fine mapping. A significance threshold of P = 7.45 × 10−7 was used.

Fine-mapping prioritized putatively candidate genes for RLS

A posterior probability of causality for each gene was assigned using the Fine-mapping Of CaUsal gene Sets (FOCUS) software7. CMTR1, RP1-153P14.5, PPP3R1, and PRPF6 were prioritized as putatively candidate genes for RLS with posterior probabilities of 1.00, 0.83, 0.63 and 0.28, respectively (Table 1).

Table 1 Causal posterior probabilities for genes in 90%-credible sets for restless legs syndrome transcriptome-wide association study signals.

Gene-set enrichment and pathway analysis

To identify known biological pathways associated with the top TWAS genes (|Z-score| > 3.50, P < 5E−04), a gene-set enrichment analysis approach was performed using Reactome ( and Gene ontology (GO; A variety of relevant gene-sets were found to be overrepresented, such as calcium ion binding, metal ion binding, as well as receptor activity pathways (Supplementary Table 4). In addition, putatively causal genes were evaluated for evidence of small molecule druggability or known drugs based on queries of the Drug Gene Interaction database. Potential target molecules were found for PPP3R1 and UCKL1 (Supplementary Table 5).

Phenome-wide association study

We performed a phenome-wide association study (PheWAS) and identified the risk of other phenotypes for RLS-associated genes, which included metabolic traits such as triglycerides, cholesterol, impedance measures, as well as psychiatric traits such as worrier/anxious feelings, neuroticism, and nervous feelings (Supplementary Data 1 and 2).


This study identified 13 associations at the transcriptome-wide significant level, of which six were not implicated in the previous GWAS. Among these, seven gene-level models (consisting of five unique genes) and six splicing events were found to be associated using CMC dorsolateral prefrontal cortex RNA-seq data. Associations for splicing events highlight an important contribution of variation in RLS risk and suggest an avenue for further functional follow-up studies. However, the direction of the effect should be interpreted with caution since alternatively spliced exons are usually negatively correlated with the risk of a disease8.

Consistent with previous findings9,10, expression of three previous RLS GWAS genes, MEIS1 in the dorsolateral prefrontal cortex and tibial nerve, SKOR1 in frontal cortex and pituitary, and MAP2K5 in the dorsolateral prefrontal cortex were associated with RLS. Two additional genes, IQCH and SKAP1, which were not reported in the previous GWAS, were significantly associated with the RLS.

In the previous RLS GWAS, MEIS1 was identified as the most significant genetic risk factor, motivating subsequent functional studies of this gene9,10. However, fine-mapping of the corresponding genomic locus prioritized PPP3R1 with a posterior probability of 0.63 in the 90%-credible gene set. PPP3R1 encodes for calcineurin subunit B type 1 protein, which functions downstream of dopaminergic pathways11. Calcineurin contains iron and zinc in its active site and is regulated by oxidation of an iron cofactor12,13. Although the role of altered dopaminergic function and brain-iron homoeostasis in RLS has been well characterized14,15, none of the previously identified candidate genes were directly implicated as key molecules in dopaminergic neuro-transmission15. Being implicated in iron homeostasis16, as well as expressed mostly in dopamine receptor-positive neurons11, suggests PPP3R1 as a candidate gene for the altered brain iron homeostasis and disrupted dopaminergic function leading to the manifestation of RLS.

In summary, by combining the results of two different multi-tissue transcriptomic imputation approaches we have not only helped to clarify candidate genes for associations identified by previous GWAS but also identified several novel genes that might be involved in RLS biology and pathogenesis. This study is the first to impute the expression data and search for associations between gene expression and RLS from summary-level GWAS data. In addition to newly identified RLS-associated genes, several pathways and associated phenotypes that were not reported in the previous post-GWAS analysis were identified using imputed expression data. Although we employed the largest RLS-GWAS cohort of European ancestry, a follow-up study in additional populations would add evidence to support these findings. This study will also provide impetus for further functional research to analyse the consequences of altered gene expression in RLS.


GWAS summary statistics and gene expression data

Summary statistics of a GWAS meta-analysis consisting of EU-RLS-GENE, INTERVAL17, and 23andMe cohorts were obtained from Schormair et al.4 (Fig. 1). A total of 15,126 cases and 95,725 controls of European ancestry: 6228 and 10,992 from EU-RLS-GENE; 3065 and 24,923 from INTERVAL, and 5833 and 59,810 from 23andMe cohorts were used in the discovery stage of the meta-analysis. The relevant tissue panels from GTEx 53 v7, and the CMC dorsolateral prefrontal cortex were downloaded from the FUSION and PrediXcan websites (Supplementary Table 1). Expression weights of CMC for PrediXcan were obtained from the GitHub ( It was shown that brain iron homeostasis and dysfunction in the dopaminergic system are involved in the pathogenesis of RLS18. Therefore, we used relevant tissue panels of brain and nervous system from GTEx 53 v7, and the CMC dorsolateral prefrontal cortex.

Transcriptomic imputation

Transcriptomic imputation was performed using S-PrediXcan and FUSION models applied to genotype data to predict expression values as described in FUSION and S-PrediXcan websites. Both FUSION and S-PrediXcan were performed with default parameters using available expression weights. The European 1000 Genomes v3 LD panel was used for the TWAS. To account for the large number of hypotheses tested, a strict Bonferroni corrected p-value threshold of 7.45 × 10−7 was used (α = 0.05/67,156 predictive models). To search for whether the significant TWAS associations were estimated from the same or independent GWAS loci, especially for the associations on the same chromosomes, we checked the linkage disequilibrium between the most significant GWAS SNPs in the identified loci using LDpair tool (


To prioritize genes for each TWAS hits, we performed a gene-based fine-mapping approach using FOCUS. RLS GWAS summary statistics along with eQTL weights were used to assign a posterior probability of causality and estimate a credible set of genes in related tissues. The same 1000 Genomes reference LD panel for FUSION and S-PrediXcan was used as in the analysis described above. A combined eQTL reference panel weight database consisting of GTExv7 weights from PrediXcan and CMC weights from FUSION software was used, as recommended for FOCUS. Finally, the genes in the 90%-credible set with a higher posterior probability were prioritized as putatively causal genes.

Gene-set enrichment analysis

A gene-set enrichment analysis approach was conducted using public datasets containing GO ( and Reactome ( pathways in EnrichR19,20.

Drug targets

Genes identified as potentially causal using were interrogated against the gene-drug interactions table of the Drug-Gene Interactions Database ( Drugs were mapped to CHEMBL IDs.

Phenome-wide association studies

To identify phenotypes associated with the RLS genes identified via TWAS, a phenome-wide association study was conducted using publicly available data provided by GWAS Atlas ( and TWAS Hub ( Only the top phenotypes (i.e., those with a p-value less than 1 × 10−10 in GWAS Atlas or an average chi-square ratio higher than 10 in TWAS Hub) are reported.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.