ARTICLE https://doi.org/10.1038/s41467-021-26280-1 OPEN Finding genetically-supported drug targets for Parkinson’s disease using Mendelian randomization of the druggable genome Catherine S. Storm 1, Demis A. Kia1, Mona M. Almramhi1,2, Sara Bandres-Ciga3, Chris Finan 4,5,6, International Parkinson’s Disease Genomics Consortium (IPDGC)*, Aroon D. Hingorani 4,5,7 & Nicholas W. Wood 1 ✉ 1234567890():,; Parkinson’s disease is a neurodegenerative movement disorder that currently has no disease- modifying treatment, partly owing to inefficiencies in drug target identification and validation. We use Mendelian randomization to investigate over 3,000 genes that encode druggable proteins and predict their efficacy as drug targets for Parkinson’s disease. We use expression and protein quantitative trait loci to mimic exposure to medications, and we examine the causal effect on Parkinson’s disease risk (in two large cohorts), age at onset and progression. We propose 23 drug-targeting mechanisms for Parkinson’s disease, including four possible drug repurposing opportunities and two drugs which may increase Parkinson’s disease risk. Of these, we put forward six drug targets with the strongest Mendelian randomization evidence. There is remarkably little overlap between our drug targets to reduce Parkinson’s disease risk versus progression, suggesting different molecular mechanisms. Drugs with genetic support are considerably more likely to succeed in clinical trials, and we provide compelling genetic evidence and an analysis pipeline to prioritise Parkinson’s disease drug development. 1 Department of Clinical and Movement Neurosciences, University College London Queen Square Institute of Neurology, London, UK. 2 Department of Medical Laboratory Technology, Faculty of Applied Medical Sciences, King Abdulaziz University, Jeddah, Kingdom of Saudi Arabia. 3 Laboratory of Neurogenetics, National Institute on Aging, Bethesda, MD, USA. 4 Institute of Cardiovascular Science, Faculty of Population Health, University College London, London WC1E 6BT, UK. 5 University College London British Heart Foundation Research Accelerator Centre, New Delhi, India. 6 Department of Cardiology, Division Heart and Lungs, University Medical Center Utrecht, Heidelberglaan 100, 3584 CX Utrecht, the Netherlands. 7 Health Data Research UK, 222 Euston Road, London, UK. *A list of authors and their affiliations appears at the end of the paper. ✉email: n.wood@ucl.ac.uk NATURE COMMUNICATIONS | (2021)12:7342 | https://doi.org/10.1038/s41467-021-26280-1 | www.nature.com/naturecommunications 1 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-26280-1 P arkinson’s disease (PD) is a neurodegenerative movement As such, eQTLs were available for 2786 and 2448 druggable disorder that currently has no disease-modifying treatment. genes in blood and brain tissue, respectively, and these were Despite efforts, around 90% of drugs that enter clinical clumped at r2 = 0.2. Compared to a lower clumping threshold, trials fail, mostly due to insufficient efficacy or safety1–3. This this increases the number of SNPs available per gene, which in contributes to the staggering $1.3 billion mean price of bringing a turn improves the power to detect an effect and makes it possible new drug to the market1. to test for biases in the MR estimate (as discussed later). When Incorporating genetics in drug development could be one of clumping at r2 = 0.2, SNPs are not strictly independent. We the most efficient ways to improve the process, because drugs therefore used MR methods that incorporate a linkage disequili- with genetic support are considerably more likely to succeed in brium matrix based on the 1000 genomes EUR reference panel in clinical trials4–6. “Druggable” genes encode proteins that have the MR analysis, which accounts for correlation between been targeted by medications or are possible to target with a small SNPs19,20. These methods therefore take linkage disequilibrium molecule or monoclonal antibody7,8. While genome-wide into account. association studies (GWAS) have effectively identified single nucleotide polymorphisms (SNPs) linked to PD risk and Discovery phase identifies 31 potential drug targets to prevent progression9–11, the GWAS design cannot reliably pinpoint PD. The largest GWAS available for a PD trait studied disease causal genes and directly inform drug development. risk in European-ancestry individuals, which we obtained from Mendelian randomization (MR) is a genetic technique that can the International PD Genomics Consortium (IPDGC)9. Our predict the efficacy of a drug by mimicking a randomized con- discovery cohort consisted of 13,708 PD patients and 95,282 trolled trial12–15. SNPs associated with expression levels of a gene controls collected for a 2014 GWAS meta-analysis21. The MR (expression quantitative trait loci, eQTLs) may be analogous to effect estimate for each SNP (Wald ratio) was calculated, and lifelong exposure to a medication targeting the encoded where >1 eQTL was available per gene after clumping, Wald protein8,16. The association between the same genetic variants ratios were meta-analysed, weighted by inverse-variance (IVW). and a disease (the outcome) can then be extracted from a GWAS Genetically-determined expression of 31 genes (11 in blood only; for the outcome (Fig. 1a). The SNP-gene-expression and SNP- 17 in brain tissue only; three in both blood and brain tissue) was disease associations can be combined using MR to infer the causal significantly associated with PD risk in the discovery cohort at effect of the exposure on the outcome. Since the exposure and FDR < 0.05. All remained significant when clumping at r2 = 0.001 outcome can be measured in two independent cohorts, openly (Supplementary Data 1). available data from two large-scale GWASs can be used for one well-powered MR study. Because of Mendel’s law of independent assortment, individuals are “randomized” at conception to have 15 potential preventative agents replicate in an independent genetically higher or lower expression levels of the druggable gene PD case-control cohort. We sought to replicate all genes that (Fig. 1b). Individuals are generally unaware of their genotype, so reached significance in the discovery phase using the Wald ratio the MR study is effectively blinded. or IVW method in an independent PD case-control cohort In this study, we use eQTLs in blood and brain tissue to predict (Fig. 1). The replication population consisted of 8036 PD patients the efficacy of over 3000 drug-targeting mechanisms in two and 5803 controls (no overlap with the discovery cohort)9. The independent PD case-control cohorts and examine several PD MR methods were identical to those used in the discovery phase. progression markers (Fig. 1c). Where possible, we repeat the Genetically-predicted expression of 15 genes (four in blood analysis using SNPs associated with circulating levels of the only; nine in brain tissue only; two in both tissues) replicated encoded proteins. Using large-scale, openly available data and using the Wald ratio or IVW method (Fig. 2 and Supplementary MR techniques, we propose a list of genetically-supported drug Data 1). BST1, CD38, CHRNB1, CTSB, GPNMB, LGALS3, MAPT, targets for PD, including repurposing opportunities of already- MMRN1, NDUFAF2, PIGF, VKORC1 and WNT3 reached licensed or clinical-phase drugs. FDR < 0.05; ACVR2A, HSD3B7 and MAP3K12 reached nominal significance. GPNMB and HSD3B7 reached significance in both blood and brain tissue. Of these 15 potential drug targets to Results prevent PD, nine were not nominated by the PD risk GWAS Mimicking medications with expression quantitative trait loci. meta-analysis9. The druggable genome encompasses human genes that encode Three replicated genes encode targets of approved or clinical- drug targets, including proteins targeted by approved and phase drugs with an appropriate direction of effect for PD clinical-phase drugs, proteins similar to approved drug targets protection, presenting a possible repurposing opportunity: and proteins accessible to monoclonal antibodies or drug-like CHRNB1, NDUFAF2 and VKORC1 (Table 1 and Supplementary small molecules in vivo7. The most comprehensive version to date Data 1). The GPNMB protein is a receptor targeted by includes 4863 genes, and we sought to identify openly available glematumumab, an antibody-drug conjugate that is being eQTL data for these genes to mimic exposure to the corre- evaluated for several types of cancer22. After binding to GPNMB, sponding medications7. Although the transcript level is biologi- the drug is internalised by the cell and is cytotoxic. Since this cally a step before the protein level, expression GWASs cover mechanism of action does not reflect a change in GPNMB levels, many genes across the genome and provide tissue specificity. we do not consider glematumumab a potential candidate for Gene expression GWAS data thus provide a very good resource repurposing. We find that CD38-inhibitors such as daratumu- for high-level screens to develop drug-targeting hypotheses. mab, licensed to treat multiple myeloma, and MAP3K12- We used eQTL data from blood (31,684 mostly European- inhibitors such as CEP-1347 may increase PD risk. Interestingly, ancestry individuals)17 and brain tissue (1387 prefrontal cortex CEP-1347 failed to modify PD progression in a phase 3 clinical samples of mostly European ancestry, including 679 healthy trial23, and our data may provide a genetic explanation why CEP- controls, 497 schizophrenia, 172 bipolar disorder, 31 autism 1347 was unsuccessful. spectrum disorder, 8 affective disorder patients)18. We kept eQTLs with false discovery rate (FDR) < 0.05 and located MR quality control suggests that CD38, CTSB, GPNMB within 5 kb of the associated gene to increase the specificity of and MAP3K12 have the most robust MR evidence for PD risk. the eQTL. We completed a series of quality control steps to prioritise 2 NATURE COMMUNICATIONS | (2021)12:7342 | https://doi.org/10.1038/s41467-021-26280-1 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-26280-1 ARTICLE Fig. 1 Overview of MR and our study. a Genetic variants associated with the expression of a gene are called eQTLs, and they mimic life-long exposure to higher or lower levels of gene expression (the exposure). These variants affect PD (the outcome) through the exposure only, i.e. there is no horizontal pleiotropy. b MR is analogous to a randomized controlled trial, where individuals are randomly allocated to a genotype according to Mendel’s law of independent assortment14. c Workflow and summarized results of our MR study. eQTL expression quantitate trait locus, MR Mendelian randomization, PD Parkinson’s disease, pQTL protein quantitative trait locus. the replicated genes. The direction of effect was consistent consistent between the discovery and replication phase. between the discovery and replication phases for all 15 replicated Although this may suggest opposing effects between tissues, genes (Supplementary Data 2). Previous eQTL-based MR stu- there was only one eQTL available for HSD3B7 in brain and dies have reported heterogeneity in magnitude and direction of two eQTLs in blood (discovery phase). Results based on one effect between tissues8,24, and we found that raised HSD3B7 or two SNPs should be interpreted with caution, because it expression was associated with raised PD risk in blood and is not possible to perform the additional quality control reduced PD risk in brain tissue (Fig. 2). This pattern was discussed below. NATURE COMMUNICATIONS | (2021)12:7342 | https://doi.org/10.1038/s41467-021-26280-1 | www.nature.com/naturecommunications 3 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-26280-1 Fig. 2 Fifteen potential preventative drug targets reach significance in two independent PD case-control cohorts. Forest plots showing the discovery- phase results for the 15 replicated genes. The centre of the error bars represents the PD odds ratio per 1-standard-deviation increase in gene expression, calculated using the Wald ratio (if 1 SNP) or IVW (if >1 SNP) and corrected for the number of genes tested. Results are colour-coded according to the tissue (red = blood, blue = brain tissue). 95% CI 95% confidence interval, FDR false discovery rate, OR odds ratio, PD Parkinson’s disease. Table 1 Four potential drug-targeting mechanisms for PD may constitute repurposing opportunities for existing drugs. Gene Outcome (tissue) Drug name Clinical phase Indications/Uses CHRNB1 Risk (brain) Rocuronium Approved Muscle relaxant in anaesthesia NDUFAF2 Risk (brain) Metformin Approved Type 2 diabetes mellitus, polycystic ovarian syndrome RHD Dyskinesia (brain) Roledumab Phase 2 Prevent alloimmunisation in Rhesus negative mothers carrying a Rhesus positive child VKORC1 Risk (blood) Warfarin Approved Prophylactic anticoagulation *high risk of falls in Parkinson’s disease These drugs are either approved or in clinical trial phase, and the mechanism of action is consistent with the direction of our MR effect estimate. The second column displays the potential effect on PD and target tissue. Clinical phase and drug indication based on https://clinicaltrials.gov/ and the British National Formulary. Direction of effect was confirmed using https://www.drugbank.ca or https:// www.ebi.ac.uk/chembl/ databases. The IVW method assumes that (1) the genetic variant(s) must nominal significance using the maximum likelihood method be associated with the exposure, (2) the genetic variant(s) must (uncorrected p < 0.05). The magnitude and direction of effect not be associated with any confounders, and (3) the genetic were largely consistent between methods, except for BST1. For variant(s) must not be associated directly with the outcome. This BST1, the MR-Egger estimate was in the opposite direction to the means that the SNP should affect the outcome (PD risk) through IVW and maximum likelihood results (Supplementary Data 1). the exposure (gene expression) only, so the y-intercept of the IVW All genes with >2 SNPs available passed the MR-Egger intercept regression is fixed at zero25. This assumption is violated if there is test except BST1, explaining the deviant MR-Egger estimate for genetic pleiotropy, where a SNP affects the outcome through an this gene (Supplementary Data 2). alternative pathway. This kind of pleiotropy may arise due to Nevertheless, if SNPs for a gene are pleiotropic in opposing measured and unmeasured confounders, for example if the SNP is directions, the MR-Egger y-intercept will still be zero. The an eQTL for another gene that is not tested in this MR study. If Cochran’s Q and I2 tests usefully assess overall heterogeneity pleiotropy pushes the effect in one direction, the IVW method yield between Wald ratios. NDUFAF2, WNT3 and VKORC1 did not a biased effect estimate. The MR-Egger method relaxes this pass the Cochran’s Q (p < 0.05) nor I2 (I2 > 0.50) tests assumption by not constraining the y-intercept. If the MR-Egger (Supplementary Data 2). This means that there is significant y-intercept significantly deviates from zero, this suggests that there heterogeneity in the MR result for these genes, and such is directional pleiotropy. This method assumes that any pleiotropic heterogeneity among Wald ratios can for example happen if at effects are independent of the gene-exposure association26. least one SNP for the gene is pleiotropic30. If several meta-analysis methods yield a similar result, such as We repeated the analysis in the discovery outcome data using the MR-Egger and maximum likelihood methods, we consider only SNPs that were specifically associated with our replicated the MR result more robust25,27,28. The latter allows more genes. In other words, we removed any SNPs associated with the uncertainty in the SNP-exposure and SNP-outcome expression of any other gene in the original eQTL dataset. All associations29. These methods are only possible if >2 SNPs are replicated genes remained significant in this analysis (Supple- available per gene, and all genes with >2 SNPs reached at least mentary Data 8 and 9). 4 NATURE COMMUNICATIONS | (2021)12:7342 | https://doi.org/10.1038/s41467-021-26280-1 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-26280-1 ARTICLE In addition, a spurious MR result may arise from a locus where (Supplementary Data 8 and 9). Of the four genes, we had suffi- the SNP-exposure and SNP-outcome associations are rooted in cient power ðPPH3 þ PPH4 ≥ 0:8Þ to perform a colocalization two distinct causal SNPs in close linkage disequilibrium30. When analysis for BST1, and we found strong evidence of colocalization the SNP is significantly associated with both exposure and (PPH4 ≥ 0:75; Supplementary Data 7). outcome, this can be probed using colocalization analysis31. If increased expression of a gene predicts reduced PD risk, this There is evidence that proteins with both MR and colocalization gene should be associated with a delayed age at onset. This was evidence are more likely to be successful drug targets32; this may consistently the case for all four genes that reached significance simply reinforce that GWAS-nominated drug targets are more for age at onset. Overall, these data suggest that there may be likely to reach approval4. Using the discovery outcome data, we some shared molecular mechanisms driving PD risk and age at had sufficient power ðPPH3 þ PPH4 ≥ 0:8Þ to perform a onset, yet this overlap may be incomplete. colocalization analysis for 13 genes (see ‘Methods’ and Supple- mentary Data 7). Of these, ACVR2A, BST1, CHRNB1, CTSB, GPNMB, HSD3B7, LGALS3, MAPT, MMRN1 and VKORC1 had There is little overlap between drug targets to prevent PD and strong evidence of colocalization ðPPH4 ≥ 0:75Þ. All genes with reduce PD progression. The PD risk GWAS data afford large sufficient power colocalized in the replication data (BST1, CD38, discovery and replication cohorts, which is a great advantage in GPNMB, HSD3B7, MAPT, MMRN1, VKORC1 and WNT3). MR. Nevertheless, it is currently not possible to reliably predict Similarly, Kia et al. recently found that eQTLs in brain tissue for PD, limiting the immediate usefulness of a drug to prevent or CD38 and GPNMB based on a different eQTL dataset colocalize delay this condition. Many clinical trials for PD use progression with PD risk loci33, strengthening the evidence for the encoded markers such as the Unified PD Rating Scale (UPDRS) to evaluate proteins as drug targets for PD. drug efficacy, and it remains unclear how the molecular mechanisms driving PD risk relate to clinical progression. We used MR to probe whether expression of any of the 4863 drug- Four potential targets for preventative drugs may also affect gable genes is significantly associated with PD progression, PD age at onset. Pharmacologically delaying the age of onset of a measured by the UPDRS (total and parts 1 to 4), mini-mental debilitating disease may have a considerable impact on both state examination (MMSE), Montreal cognitive assessment socioeconomic burden of disease and quality of life by providing (MOCA), modified Schwab and England activities of daily living disability-free years to people at risk. Evidence from polygenic scale (SEADL), Hoehn and Yahr stage, dementia, depression, and risk score analyses suggests that genetic risk of PD is correlated dyskinesia. The MR pipeline for each progression marker was with PD age at onset11,34–36. We therefore asked whether identical to the discovery phase for PD risk (Fig. 1). expression of the genes reaching significance in our MR discovery We used openly available summary statistics from a GWAS for phase for PD risk also predict PD age of onset. We sourced these PD progression markers in 4093 European PD patients, openly available summary statistics from a PD age of onset followed over a median of 2.97 years10. 3455 druggable genes had GWAS, including 17,996 patients (Fig. 1c). Based on the same an eQTL available for MR using a PD progression marker (2752 analysis pipeline as the replication step for PD risk, expression of in blood, 2353 in brain tissue), and eight genes reached four genes predicted PD age of onset at p < 0.05: BST1 in blood, significance across five progression outcomes (Fig. 4 and CD38 in brain tissue, CTSB in brain tissue and MMRN1 in brain Supplementary Data 1). One of these, RHD, encodes the target tissue (Fig. 3 and Supplementary Data 3). CD38 and MMRN1 of a clinical-phase medication with an appropriate direction of remained significant when clumping at r2 = 0.001. There were >2 effect, possibly representing a repurposing opportunity (Table 1). SNPs available for BST1, CD38 and CTSB, and the IVW, max- IRAK3 expression in blood was significantly associated with imum likelihood and MR-Egger methods yielded a consistent UPDRS parts 2 and 4, depression, and dyskinesias. LMAN1 direction of effect for these genes (Supplementary Data 1). All expression in blood reached significance for dyskinesias and three genes passed the MR-Egger intercept (p > 0:05), Cochran’s UPDRS part 2. Reaching significance for several progression Q (p > 0:05), and I2 tests (ðI 2 < 0:50Þ. BST1 and MMRN1 markers strengthens the evidence for these two genes. No genes remained significant when removing SNPs associated with reached significance for both PD risk and progression. Since age expression of any other gene in the original eQTL dataset at motor symptom onset may be considered an early marker of Fig. 3 Four potential preventative drugs may also affect PD age at onset. Forest plot; the centre of the error bars represents the standard-deviation change in PD age at onset per 1-standard-deviation increase in gene expression, calculated using the Wald ratio (if 1 SNP) or IVW (if >1 SNP) and colour- coded by tissue (red = blood, blue = brain tissue). A negative beta corresponds to a younger age at onset, and a positive beta corresponds to an older age at onset. 95% CI 95% confidence interval, PD Parkinson’s disease. NATURE COMMUNICATIONS | (2021)12:7342 | https://doi.org/10.1038/s41467-021-26280-1 | www.nature.com/naturecommunications 5 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-26280-1 Fig. 4 Genetically-predicted expression of eight genes in blood or brain tissue is associated with PD progression markers. Forest plot; the centre of the error bars show the standard-deviation change in each progression marker, per 1-standard-deviation increase in gene expression, calculated using the Wald ratio (if 1 SNP) or IVW (if >1 SNP). Results are colour-coded by tissue (red = blood, blue = brain tissue) and corrected for the number of genes tested. 95% CI 95% confidence interval, DEPR depression, FDR false discovery rate, HY Hoehn and Yahr, DEPR depression, UPDRS2-4 Unified Parkinson’s Disease Rating Scale parts 2 to 4. PD progression, we also used our MR approach to assess whether Our MR analysis found that BST1, CTSB and LGALS3 levels genes discovered by our progression analysis causally predict age were consistently associated with PD risk (p < 0.05; Fig. 5 and at onset. Of the genes that reached significance for a progression Supplementary Data 5). The result for GPNMB (risk) and PYGL marker, none reached nominal significance using the age at onset (UPDRS part 4) lost significance when using data from different data (Supplementary Data 3 and 4). CD177, IRAK3, RHD and pQTL studies. The direction of effect was consistent between the STK4 remained significant when removing SNPs associated with pQTL and eQTL results for all genes except BST1, and the MR- expression of any other gene in the original eQTL dataset Egger intercept, Cochran’s Q and I2 tests suggest that the BST1 (Supplementary Data 8 and 9). It was not possible to perform a results may be biased by genetic pleiotropy (Supplementary reliable colocalization analysis in our progression study, since the Data 6). This illustrates the importance of MR quality control— discovered genes did not have sufficient power to do so (i.e. maximizing the number of SNPs available per drug target and PPH3 þ PPH4 < 0:8). validation with different data types and independent replication The direction of effect was consistent between the IVW, cohorts is essential for a reliable effect estimate. maximum likelihood and MR-Egger methods for all genes except RHD, where the MR-Egger method opposed the direction of the IVW and maximum likelihood methods. CD177 (depression), Discussion RHD (dyskinesia), PYGL (UPDRS part 4) and STK4 (Hoehn and This work explicitly seeks to identify new drug targets for PD, and Yahr) reached significance when clumping at r 2 ¼ 0:001. we provide genetic evidence in favour of 23 potential disease- ADAM32 (Hoehn and Yahr), IRAK3 (dyskinesia), LMAN1 modifying drug targets. Tables 2 and 3 summarize the evidence (UPDRS part 2), and RHD (dyskinesia) passed MR-Egger supporting these genes. The genes were prioritised using several intercept, Cochran’s Q and I2 tests (Supplementary Data 2). meta-analysis methods (IVW, MR-Egger and maximum like- Taken together, these five genes have the most robust MR lihood), the MR-Egger intercept test, Cochran’s Q test, the I2 test, evidence for modifying a PD progression marker. a pQTL study, colocalization analysis and previously published MR and colocalization evidence. This allowed us to look for pleiotropy due to both measured and unmeasured confounders25. Protein quantitative trait locus data provide further genetic We propose six drug targets with the strongest MR evidence: evidence. Most clinically-used drugs target proteins, not gene CTSB, GPNMB, CD38, RHD, IRAK3 and LMAN1. expression, and genetic variants associated with protein levels We identifed four genes encoding targets for existing drugs (protein quantitative trait loci, pQTLs) may model drug target warranting further discussion (Table 1). NDUFAF2 encodes a effects more accurately than eQTLs8. Even with high throughput subunit of a target of metformin, an approved medication for type protein assays, however, the spectrum of reliable, well-powered 2 diabetes mellitus. There is extensive evidence for a relationship GWAS data on protein targets is limited. Many genetic studies on between diabetes mellitus and PD41, and several rodent studies protein levels are based on plasma and lack tissue diversity37–39. have investigated the potential of metformin as a neuroprotective Of the 23 proposed targets, we found pQTLs for BST1, CD38, agent41–43. We found significant heterogeneity in the MR result CTSB, GPNMB and LGALS3 for PD risk, as well as PYGL and for this gene. Although this may be because at least one SNP for QDPR for UPDRS part 437–40. this gene is pleiotropic, we speculate that this could occur if the 6 NATURE COMMUNICATIONS | (2021)12:7342 | https://doi.org/10.1038/s41467-021-26280-1 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-26280-1 ARTICLE Fig. 5 Protein quantitative trait loci in blood provide further genetic evidence. Forest plots showing the results for all proteins and outcomes where a pQTL was available. The centre of the error bars show the (a) PD odds ratio and (b) standard-deviation change in UPDRS part 4 score, per 1-standard- deviation increase in circulating protein levels, calculated using the Wald ratio (if 1 SNP) or IVW (if >1 SNP). The “pQTL Source” column indicates which pQTL study the SNPs were derived from. 95% CI 95% confidence interval, OR odds ratio, PD Parkinson’s disease, pQTL protein quantitative trait locus, UPDRS Unified Parkinson’s Disease Rating Scale. effect is driven by a subset of PD patients. This, however, remains which is inhibited by muscle relaxants used during surgical a subject for future research, because the GWAS data used in this anaesthesia. VKORC1 encodes the catalytic subunit of the vitamin study are not stratified by any kind of PD subtype. Epidemiolo- K epoxide reductase, and this enzyme is targeted by the oral gical studies on the relationship between long-term medication anticoagulant warfarin. The key adverse effect of warfarin treat- use and incidence of a disease are an invaluable contribution to ment is haemorrhage, and since PD is a movement disorder evaluating preventative agents for PD. A retrospective cohort where patients experience frequent falls, any potential benefit of study of over 6000 patients with type 2 diabetes mellitus found warfarin treatment would likely be outweighed by the added risk that more than four years of metformin use maybe associated of haemorrhagic strokes and complications of bleeding. with a reduced PD incidence44. Our MR study thus provides The two-sample MR design allowed us to explore different further evidence in favour of repurposing anti-diabetic drugs tissues and PD traits, and we identified different candidates to for PD. prevent, delay onset, and slow progression of PD (Figs. 2, 3 Other medications may not be as suitable for repurposing. To and 4). Although we found that four of the drug targets for PD our knowledge, there is no evidence linking PD and the drug risk may also affect PD age at onset, we found very different roledumab, which is currently in a phase II clinical trial to pre- candidates for progression. Age at motor symptom onset can be vent alloimmunisation in Rhesus negative mothers carrying a considered an early sign of PD progression, and it is striking that Rhesus positive child (NCT02287896). Our evidence suggests that none of the genes that reached significance for a progression RHD expression in brain tissue, rather than blood, is associated outcome reached significance in the age at onset data. These with PD dyskinesia. Next, CHRNB1 encodes the beta subunit of results are in line with the GWAS data, finding little overlap the muscle acetylcholine receptor at the neuromuscular junction, between loci associated with PD risk, age at onset and progression NATURE COMMUNICATIONS | (2021)12:7342 | https://doi.org/10.1038/s41467-021-26280-1 | www.nature.com/naturecommunications 7 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-26280-1 Table 2 Evidence supporting druggable genes whose expression was significantly associated with PD risk or age at onset using the Wald ratio or IVW method. Gene PD Tissue Replication Sign. with Sign. with MR-Egger Cochran’s I2 test pQTL Coloc Previously outcome max. lik. MR-Egger intercept test Q test evidence published MR or coloc support GPNMB Risk Blood ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ MR9 Risk Brain ✓ ✓ ✓ ✓ ✓ ✓ ✓ MR24; coloc33 CTSB Risk Brain ✓ ✓ x ✓ ✓ ✓ ✓ ✓ MR9 Age Brain ✓ x ✓ ✓ ✓ at onset VKORC1 Risk Blood ✓ ✓ x ✓ ✓ ✓ ✓ CD38 Risk Brain ✓ ✓ x ✓ ✓ ✓ x x Coloc33 Age Brain ✓ x ✓ ✓ ✓ at onset MAP3K12 Risk Blood ✓ ✓ x ✓ ✓ ✓ x NDUFAF2 Risk Brain ✓ ✓ x ✓ ✓ ✓ x BST1 Risk Blood ✓ ✓ x x ✓ ✓ ✓ ✓ MR9 Age Blood ✓ x ✓ ✓ ✓ ✓ at onset HSD3B7 Risk Blood ✓ ✓ ✓ ✓ x Risk Brain ✓ ✓ MR24 LGALS3 Risk Brain ✓ ✓ ✓ ACVR2A Risk Brain ✓ ✓ CHRNB1 Risk Brain ✓ ✓ MR9 MAPT Risk Brain ✓ ✓ MMRN1 Risk Brain ✓ ✓ Age Brain at onset PIGF Risk Brain ✓ WNT3 Risk Blood ✓ ✓ x ✓ x x x MR9 ✓ pass, x fail, blank not possible to test, coloc colocalization, max. lik. maximum likelihood, MR Mendelian randomization, PD Parkinson’s disease, pQTL protein quantitative trait locus, sign significant. Table 3 Evidence supporting druggable genes whose expression was significantly associated with a PD progression trait using the Wald ratio or IVW method. Gene PD outcome Tissue Sign. with Sign. with MR-Egger Cochran’s Q test I2 test pQTL Coloc max. lik. MR-Egger intercept test evidence RHD Dyskinesia Brain ✓ x ✓ ✓ ✓ IRAK3 Dyskinesia Blood ✓ ✓ ✓ ✓ ✓ Depression Blood ✓ ✓ x ✓ ✓ UPDRS part 2 Blood ✓ x x ✓ ✓ UPDRS part 4 Blood ✓ ✓ x ✓ ✓ ADAM32 Hoehn and Yahr Brain ✓ x ✓ ✓ ✓ LMAN1 UPDRS part 2 Blood ✓ x ✓ ✓ ✓ Dyskinesia Blood ✓ ✓ x ✓ ✓ PYGL UPDRS part 4 Brain ✓ CD177 Depression Brain QDPR UPDRS part 4 Blood ✓ ✓ ✓ x x x STK4 Hoehn and Yahr Brain x x x ✓ pass, x fail, blank not possible to test, coloc colocalization, max. lik. maximum likelihood, MR Mendelian randomization, PD Parkinson’s disease, pQTL protein quantitative trait locus, sign significant, UPDRS Unified Parkinson’s Disease Rating Scale. markers9–11. This may reflect the limited sample size of current used here are from the largest openly available progression PD progression GWAS data. Nevertheless, this raises questions GWAS to date. We did not find any non-overlapping PD pro- about what drives PD susceptibility versus progression, painting a gression GWAS with sufficient power for a replication step in our yet unclear picture of partially overlapping molecular progression analysis, which would have to measure progression in mechanisms. a similar way to the study used here. As such, the preventative list Our candidates to slow PD progression may be of most carries more robust evidence, because each gene reached sig- immediate relevance, because currently PD cannot be accurately nificance in two large, independent cohorts. Replication is critical predicted. A preventative agent would need to be highly tolerable to validating scientific findings and eliminating false positives, and have a very safe side effect profile, and our approach is not and this has been an crucial lesson for genetic research45–47. well suited for systematically evaluating the safety aspects of our Replication is not common practice in MR yet48, and it is a key proposed candidates in this study. To our knowledge, the data strength of our study. Although including all samples available in 8 NATURE COMMUNICATIONS | (2021)12:7342 | https://doi.org/10.1038/s41467-021-26280-1 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-26280-1 ARTICLE one analysis would maximise statistical power46,49, using inde- capture as many genes as possible and explore two potential tissue pendent discovery and replication cohorts allowed us to validate sites of action, but we note that it is difficult to prioritise genes our proposed drug targets. Since our overarching intention was to based on which tissue(s) they reached significance in. provide genetic evidence to improve success rates in clinical trials, Furthermore, the sample size of our blood eQTL data we made this decision to reduce the number of false positives. (n = 31,684) is larger than that for brain tissue eQTLs (n = 1387) Our study has valuable advantages compared to previous and the blood pQTL study (n = 750–4137). A larger sample size MR projects studying PD using QTL data. In the latest GWAS allows greater power to detect QTLs, meaning there are more meta-analysis for PD risk in Europeans, Nalls et al. selected SNPs SNPs per gene. Nevertheless, it is unclear how well QTL data associated with PD risk and used MR to identify whether any of mimic medications that modulate activity levels of the protein. these loci alter expression or methylation of genes within 1 Mb of We are encouraged that five of the seven proteins we were able to the SNP9. This contrasts with our exposure-centred MR analysis, probe using both eQTL and pQTL data were successfully vali- where we chose SNPs associated with the expression of a drug- dated, adding to existing evidence that regulatory variants may be gable gene, rather than the disease outcome. More recently, Baird used for robust causal inferences in drug target MR8. Never- and colleagues conducted a transcriptome-wide MR study for a theless, this MR study does not provide functional evidence for series of brain diseases and found six genes whose expression in the proposed drug targets, and the MR process does not replace brain tissue was significantly associated with PD risk24. Two of pre-clinical evaluation of drug targets in vitro and in vivo. these were also discovered in our study: GPNMB and HSD3B7. Genomic approaches serve as adjuncts thereto, promising to The remaining four were either not part of the druggable genome, better prioritise drug targets carried forward to clinical phase rendering the encoded proteins less actionable drug targets, or did trials. not reach significance in our discovery or replication cohorts, A 9.6% vs. 13.8% success rate for drugs from phase 1 trials to illustrating the importance of replication. Furthermore, our MR approval may mean a $480 million difference in the median study is the first to study druggable genes in the context of PD age research and development cost of bringing a new drug to the at onset and progression. market1. The druggable genome resource has opened up new Nevertheless, progression and age at onset studies are parti- avenues for drug target identification using existing genetic cularly affected by collider bias50–52. For example, if expression of data7,56,57, and if genetic evidence increases success rates even by a a gene and depression are both associated with disease risk, that few percent, this could have a substantial effect on drug develop- gene’s expression will be artificially associated with depression in ment costs4,5. As such, MR a highly compelling, time- and cost- a cohort containing only cases. In a progression study, genetic effective adjunct to the randomized controlled trial. We have made variants that cause disease will thus be associated with other risk our code openly available for use beyond PD research (https:// factors for disease. The druggable genes we identified in our github.com/catherinestorm/mr_druggable_genome_pd/)58, and we progression study did not reach significance in our risk study, so have demonstrated ways to prioritise drug targets based on genetic this kind of collider bias is less likely to have occurred for our data. We have provided human genetic evidence of drug efficacy for candidate genes. The age at onset analysis was comparably more PD, and we hope that these data will serve as a useful resource for affected, since we tested genes that reached significance for PD prioritising drug development efforts. risk. Overall, this emphasises the importance of MR quality control methods (including replication) for identifying reliable causal effects, representative sampling in GWAS, as well as Methods All DNA positions are based on the human reference genome build hg19 continued development of methods to formally test for collider (GRCh37). Data processing was completed using R software version 3.6.359. bias53,54. Another key limitation of this study is that MR cannot fully recapitulate a clinical trial. MR mimics lifelong, low-dose expo- Exposure data. Tissue-specific eQTL data were obtained from the eQTLGen sure to a drug and assumes a linear relationship between exposure (https://eqtlgen.org/) and PsychENCODE consortia (http:// resource.psychencode.org/); full descriptions of the data are available in the original and outcome. This differs from a clinical trial, which typically publications17,18. Briefly, the eQTLGen data consisted of cis-eQTLs for 16,987 investigates comparably high doses of drug over a much shorter genes and 31,684 blood samples, of which most were healthy European-ancestry timeframe. The MR result may therefore not directly correspond individuals. We downloaded the full significant cis-eQTL results (FDR < 0.05) and to the effect size in practice and does not perfectly predict the allele frequency information from the eQTLGen consortium on 13 May 2020. The PsychENCODE data included 1387 prefrontal cortex, primarily-European effect of a drug. samples (679 healthy controls, 497 schizophrenia, 172 bipolar disorder, 31 autism In addition, the eQTL cohorts contained some non-European spectrum disorder and 8 affective disorder patients). We downloaded all significant individuals17,18, three of the pQTL studies sourced were based on eQTLs (FDR < 0.05) for genes with expression >0.1 fragments per kilobase per Icelandic, Scottish and German cohorts38–40, and the PD popu- million mapped fragments (FPKM) in at least ten samples and all SNP information, accessed on 13 May 2020. lations were comprised of European individuals only9–11. Linkage We obtained an updated version of the druggable genome containing 4863 disequilibrium patterns differ between populations, which may genes from the authors of the original publication7, double-checking the compromise how well our QTLs mimic drug action in the PD druggability level for all genes marked as approved or in clinical trials cohorts and introduce bias to the MR effect estimate25. (“druggability tier 1”). We removed non-autosomal genes, leaving 4560 druggable It is difficult to interpret which tissue would be the most genes. We filtered both eQTL datasets to include SNPs 5 kb upstream of the target druggable gene start or 5 kb downstream of the target druggable gene end position. appropriate site of action. Whereas the genes that reached sig- We sought freely available pQTL data from blood or brain tissue for all nificance in both blood and brain tissue may have stronger MR druggable genes that reached significance for any PD outcome in our study. Out of evidence, targeting the protein of a widely expressed gene may 23 pQTL studies identified, four studies (1) reported significant pQTLs in lead to systemic side-effects. Brain tissue may be more biologi- individuals of European descent for any of the druggable proteins proposed by our eQTL analysis, (2) provided all the SNP information required for MR and (3) cally relevant for neurodegeneration, but a drug acting in the reported SNPs that were available in our PD outcome data37–40. Sun and colleagues blood stream may not need to cross the blood–brain barrier to measured 3622 proteins in 3301 healthy European blood donors from the exert its effect. A limitation of using brain tissue is that gene INTERVAL study and identified 1927 pQTLs for 1478 proteins. Emilsson and expression is quantified post-mortem, and measured expression colleagues measured 4137 proteins in the serum of 5457 Icelanders from AGES Reykjavik study. Effect alleles and effect allele frequencies were obtained from the levels are influenced by RNA degradation occurring after cell authors. Suhre and colleagues measured 1124 proteins in 1000 blood samples from death as well as transcriptional changes occurring in response to a German population. Hillary and colleagues measured 92 proteins in the blood of death55. We included both blood and brain tissue eQTLs to 750 healthy Scottish controls. NATURE COMMUNICATIONS | (2021)12:7342 | https://doi.org/10.1038/s41467-021-26280-1 | www.nature.com/naturecommunications 9 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-26280-1 In total, we found pQTLs that were available in the appropriate PD outcome posterior probabilities corresponding to one of five hypotheses: PPH0, no asso- data for seven of our druggable proteins of interest: BST1, CD38, CTSB, GPNMB, ciation with either trait; PPH1, association with expression of the gene, but not the LGALS3, PYGL and QDPR. All pQTLs included in our analysis had p < 5e−6 in PD trait; PPH2, association with the PD trait, but not expression of the gene; the original pQTL study. All pQTLs were found on the same chromosome as the PPH3, association with the PD trait and expression of the gene, with distinct causal associated gene except for: rs62143198 for PYGL, rs62143197 for QDPR, rs4253282 variants; PPH4, association with the PD trait and expression of the gene, with a for GPNMB, rs2731674 for GPNMB37. These latter four SNPs are therefore acting shared causal variant31. A low PPH3 and PPH4 in combination with a high PPH0, in trans. PPH1 and/or PPH2 indicates limited power in the colocalization analysis31. We therefore restricted our analysis to genes reaching PPH3 þ PPH4 ≥ 0:8. Outcome data. All PD data were obtained from the IPDGC, and details on recruitment and quality control are available in the original publications9–11,21. In Reporting summary. Further information on research design is available in the Nature the discovery phase for PD risk, we used openly available summary statistics from a Research Reporting Summary linked to this article. 2014 case-control GWAS meta-analysis, which included 13,708 PD patients and 95,282 controls21. In the replication phase for PD risk, we obtained summary statistics from 11 Data availability The data generated in this study and a data dictionary (Supplementary Information) are case-control GWAS studies included in the most recent PD risk GWAS meta- provided in the Supplementary Information. Tissue-specific eQTL data were obtained analysis from the authors9. The 11 studies, as named and described in the PD from the eQTLGen (https://eqtlgen.org/) and PsychENCODE consortia (http:// GWAS meta-analysis, were Spanish Parkinson’s, Baylor College of Medicine/ University of Maryland, McGill Parkinson’s, Oslo Parkinson’s Disease Study, resource.psychencode.org/). The druggable genome data were obtained via from the Parkinson’s Progression Markers Initiative (PPMI), Finnish Parkinson’s, Harvard original authors; an immediately accessible version of the druggable genome is available Biomarker Study (HBS), UK PDMED (CouragePD), Parkinson’s Disease in the supplementary materials of the original publication7. Subsets of the most recent Biomarker’s Program (PDBP), Tübingen Parkinson’s Disease cohort (CouragePD) PD risk GWAS were obtained from the original authors9. GWAS summary statistics for and Vance (dbGap phs000394). These yielded a total of 8036 PD cases and 5803 PD age at onset are available on the IPDGC website (http://pdgenetics.org/resources). controls. We meta-analysed the data using METAL (version 2011-03-25) using Parkinson’s progression GWAS data can be found here: https://pdgenetics.shinyapps.io/ default settings, weighting by sample size60. The overall genomic inflation factor pdprogmetagwasbrowser/. was λ ¼ 1:116, and when scaled to 1000 cases and 1000 controls λ1000 ¼ 1:017. Based on genomic inflation factors and quantile–quantile plots of the original Code availability GWASs9,21, we considered our quantile–quantile plot to show adequate agreement The code used for this study is openly available on GitHub, accompanied by instructions with the expected null distribution (Supplementary Fig. 1). for use and required tools (https://github.com/catherinestorm/ For the progression marker analyses, we used summary statistics from the mr_druggable_genome_pd). Data processing was completed using R software (version largest publicly available GWAS meta-analyses for PD age at onset and clinical 3.6.3), with packages TwoSampleMR (version 0.5.4), MendelianRandomization (version progression10,11. For age at onset, this includeed 17,996 PD cases, and age at onset 0.4.2), coloc (version 4.0.4), dplyr (version 1.0.0), readr (version 1.3.1), stringr (version was defined as self-reported age at motor symptom onset or PD diagnosis. The 1.4.0), tidyverse (version 1.3.0), forestplot (version 1.9), plyr (version 1.8.6), devtools authors reported a high correlation between age at diagnosis and age at onset. The progression GWAS meta-analysis included 4093 PD patients from 12 (version 2.3.0), remotes (version 2.1.1). The GWAS meta-analysis was completed using cohorts, followed over a median of 2.97 years (mean visits per individual over the METAL (version 2011-03-25). study period: 5.44). We downloaded summary statistics for nine continuous outcomes and four binomial outcomes (https://pdgenetics.shinyapps.io/ Received: 21 October 2020; Accepted: 14 September 2021; pdprogmetagwasbrowser/). Continuous outcomes included Hoehn and Yahr stage (PD progression rating scale), total UPDRS/Movement Disorder Society revised version total (PD progression rating scale), UPDRS parts 1 to 4 (1 = non-motor symptoms, 2 = motor symptoms, 3 = motor examination, 4 = motor complications), MOCA (cognitive impairment), MMSE (cognitive impairment) and SEADL (activities of daily living and independence). The binomial outcomes we used were dementia, depression, dyskinesia, as well as reaching Hoehn and References Yahr stage 3 or more. 1. Wouters, O. J., McKee, M. & Luyten, J. Estimated research and development investment needed to bring a new medicine to market, 2009–2018. JAMA - J. Am. Med. Assoc. 323, 844–853 (2020). Mendelian randomization. MR analyses were completed using the R package 2. Smietana, K., Siatkowski, M. & Møller, M. Trends in clinical success rates. Nat. “TwoSampleMR” (version 0.5.4)61, unless stated otherwise. The exposure and Rev. Drug Discov. 15, 379–380 (2016). outcome data were loaded and harmonized using in-built functions. SNPs were 3. Harrison, R. K. Phase II and phase III failures: 2013–2015. Nat. Rev. Drug then clumped at r 2 < 0:2 using European samples from the 1000 Genomes Discov. 15, 817–818 (2016). Project20,61. Steiger filtering was used to remove genes where SNPs explained a 4. Nelson, M. R. et al. The support of human genetic evidence for approved drug greater proportion of variation in the outcome (PD trait) than variation in the indications. Nat. Genet. 47, 856–860 (2015). exposure (gene expression). For the eQTL analysis, the Steiger filtering excluded 5. King, E. A., Wade Davis, J. & Degner, J. F. Are drug targets with genetic 0–403 genes per outcome tested in a tissue, representing 0–15% of all genes studied support twice as likely to be approved? Revised estimates of the impact of per outcome tested in a tissue. genetic support for drug mechanisms on the probability of drug approval. Wald ratios were calculated for all SNPs. These were meta-analysed using the PLoS Genet. 15, 1–20 (2019). IVW, MR-Egger and maximum likelihood methods, including a linkage 6. Hingorani, A. D. et al. Improving the odds of drug development success disequilibrium matrix to account for correlation between SNPs; this function uses the R package “MendelianRandomization” version 0.4.229. Forest plots were through human genomics: modelling study. Sci. Rep. 9, 1–25 (2019). produced using the R package “forestplot”. 7. Finan, C. et al. The druggable genome and support for target Where >2 SNPs were available per exposure, we assessed whether the MR-Egger identification and validation in drug development. Sci. Transl. Med. 9, intercept significantly deviated from zero, as well as Cochran’s Q and I2 methods to eaag1166 (2017). test for heterogeneity between Wald ratios62. FDR-corrected p-values were 8. Schmidt, A. F. et al. Genetic drug target validation using Mendelian calculated within each exposure-outcome combination to correct for multiple randomisation. Nat. Commun.https://doi.org/10.1038/s41467-020-16969-0 testing. In the discovery study for PD risk and the PD progression studies, we (2020). considered FDR < 0.05 significant. In the replication studies for PD risk and age at 9. Nalls, M. A. et al. Identification of novel risk loci, causal insights, and heritable onset, as well as the pQTL study, we considered nominal p < 0.05 significant. risk for Parkinson’s disease: a meta-analysis of genome-wide association For genes which reached significance using the IVW method (>1 SNP studies. Lancet Neurol. 18, 1091–1102 (2019). available), we carried out another MR analysis, clumping at r 2 < 0:001. If >1–2 10. Iwaki, H. et al. Genetic risk of Parkinson disease and progression: an analysis SNPs were available at this clumping threshold, Wald ratios were meta-analysed of 13 longitudinal cohorts. Neurol. Genet. 5, e354 (2019). using the IVW, MR-Egger, weighted mode and weighted median methods. 11. Blauwendraat, C. et al. Parkinson’s disease age at onset genome-wide association study: defining heritability, genetic loci, and α-synuclein mechanisms. Mov. Disord. 1–10, https://doi.org/10.1002/mds.27659 (2019). Colocalization. We carried out a colocalization analysis for PD risk, age at onset 12. Katan, M. B. Apoliporotein E isoforms, serum cholesterol, and cancer. Lancet and progression outcomes using the R package “coloc”31. We harmonized exposure 1, 507–508 (1986). and outcome datasets using the “TwoSampleMR” package. We used default priors: 13. Davey Smith, G. & Ebrahim, S. ‘Mendelian randomization’: can genetic p1 ¼ 104 , p2 ¼ 104 , p12 ¼ 105 . p1, p2 and p12 are the prior probabilities that a epidemiology contribute to understanding environmental determinants of SNP in the tested region is significantly associated with expression of the tested disease? Int. J. Epidemiol. 32, 1–22 (2003). gene, the tested PD outcome, or both, respectively. The colocalization yields 10 NATURE COMMUNICATIONS | (2021)12:7342 | https://doi.org/10.1038/s41467-021-26280-1 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-26280-1 ARTICLE 14. Hingorani, A. & Humphries, S. Nature’s randomised trials. Lancet 366, 44. Shi, Q., Liu, S., Fonseca, V. A., Thethi, T. K. & Shi, L. Effect of metformin on 1906–1908 (2005). neurodegenerative disease among elderly adult US veterans with type 2 15. Holmes, M. V., Ala-Korpela, M. & Davey Smith, G. Mendelian randomization diabetes mellitus. BMJ Open 9, e024954 (2019). in cardiometabolic disease: challenges in evaluating causality. Nat. Rev. 45. Hirschhorn, J. N., Lohmueller, K., Byrne, E. & Hirschhorn, K. A comprehensive Cardiol. 14, 577–590 (2017). review of genetic association studies. Genet. Med. 4, 45–61 (2002). 16. Zhu, Z. et al. Integration of summary data from GWAS and eQTL studies 46. Chanock, S. J. et al. Replicating genotype-phenotype associations. predicts complex trait gene targets. Nat. Genet. 48, 481–487 (2016). Naturehttps://doi.org/10.1038/447655a (2007). 17. Võsa, U. et al. Large-scale cis- and trans-eQTL analyses identify thousands of 47. Marigorta, U. M., Rodríguez, J. A., Gibson, G. & Navarro, A. Replicability and genetic loci and polygenic scores that regulate blood gene expression. Nat. prediction: lessons and challenges from GWAS. Trends Genet. 34, 504–517 Genet. 53, 1300–1310 (2021). (2018). 18. Wang, D. et al. Comprehensive functional genomic resource and integrative 48. Burgess, S., Foley, C. N. & Zuber, V. Inferring causal relationships between model for the human brain. Science 362, eaat8464 (2018). risk factors and outcomes from genome-wide association study data stephen. 19. Burgess, S., Dudbridge, F. & Thompson, S. G. Combining information on Annu. Rev. Genom. Hum. Genet. 303–27, https://doi.org/10.1146/annurev- multiple instrumental variables in Mendelian randomization: Comparison of genom-083117-021731-021731 (2018). allele score and summarized data methods. Stat. Med. 35, 1880–1906 (2016). 49. Huffman, J. E. Examining the current standards for genetic discovery and 20. The 1000 Genomes Project Consortium. An integrated map of genetic replication in the era of mega-biobanks. Nat. Commun. 9, 1–4 (2018). variation from 1,092 human genomes. Nature 491, 56–65 (2012). 50. Paternoster, L., Tilling, K. & Smith, G. D. Genetic epidemiology and 21. Nalls, M. A. et al. Large-scale meta-analysis of genome-wide association data Mendelian randomization for informing disease therapeutics: conceptual and identifies six new risk loci for Parkinson’s disease. Nat. Genet. 46, 989–993 methodological challenges. 1–10, https://doi.org/10.1371/ (2014). journal.pgen.1006944 (2017). 22. Rose, A. A. N., Biondini, M., Curiel, R. & Siegel, P. M. Targeting GPNMB with 51. Munafò, M. R., Tilling, K., Taylor, A. E., Evans, D. M. & Davey Smith, G. glembatumumab vedotin: current developments and future opportunities for Collider scope: when selection bias can substantially influence observed the treatment of cancer. Pharmacol. Ther. 179, 127–141 (2017). associations. Int. J. Epidemiol. 47, 226–235 (2018). 23. Wang, L. H. & Johnson, E. M. Mixed lineage kinase inhibitor CEP-1347 fails 52. Bartlett, J. W., Harel, O. & Carpenter, J. R. Asymptotically unbiased estimation to delay disability in early Parkinson’s disease. Neurology 71, 462–463 (2008). of exposure odds ratios in complete records logistic regression. Am. J. 24. Baird, D. A. et al. Identifying drug targets for neurological and psychiatric Epidemiol. 182, 730–736 (2015). disease via genetics and the brain transcriptome. PLoS Genet. 17, 1–26 (2021). 53. Gkatzionis, A. & Burgess, S. Contextualizing selection bias in Mendelian 25. Burgess, S. et al. Guidelines for performing Mendelian randomization randomization: How bad is it likely to be? Int. J. Epidemiol. 48, 691–701 investigations. Wellcome Open Res. 4, 186 (2020). (2019). 26. Bowden, J. et al. Assessing the suitability of summary data for two-sample 54. Griffith, G. J. et al. Collider bias undermines our understanding of COVID-19 Mendelian randomization analyses using MR-Egger regression: the role of the disease risk and severity. Nat. Commun. 11, 1–12 (2020). I2 statistic. Int. J. Epidemiol. 45, 1961–1974 (2016). 55. Ferreira, P. G. et al. The effects of death and post-mortem cold ischemia on 27. Haycock, P. C. et al. Statistical Commentary Best (but oft-forgotten) practices: human tissue transcriptomes. Nat. Commun. 9, 490 (2018). the design, analysis, and interpretation of Mendelian randomization studies 1. 56. Hegvik, T. A. et al. Druggable genome in attention deficit/hyperactivity Am. J. Clin. Nutr. 103, 965–978 (2016). disorder and its co-morbid conditions. New avenues for treatment. Mol. 28. Slob, E. A. W. & Burgess, S. A comparison of robust Mendelian Psychiatryhttps://doi.org/10.1038/s41380-019-0540-z (2019). randomization methods using summary data. Genet. Epidemiol. 1–17, https:// 57. Jacobs, B. M. et al. Summary-data-based Mendelian randomization prioritizes doi.org/10.1002/gepi.22295 (2020). potential druggable targets for multiple sclerosis. Brain Commun. 2, 1–17 29. Yavorska, O. O. & Burgess, S. MendelianRandomization: an R package for (2020). performing Mendelian randomization analyses using summarized data. Int. J. 58. Storm, C. S. Finding genetically-supported drug targets for Parkinson’s disease Epidemiol. 46, 1734–1739 (2017). using Mendelian randomization of the druggable genome. GitHub Repos. 30. Hemani, G., Bowden, J. & Davey Smith, G. Evaluating the potential role of (2020). pleiotropy in Mendelian randomization studies. Hum. Mol. Genet. 27, 59. R Core Team. R: A Language and Environment for Statistical Computing R195–R208 (2018). (2019). 31. Giambartolomei, C. et al. Bayesian test for colocalisation between pairs of 60. Willer, C. J., Li, Y. & Abecasis, G. R. METAL: fast and efficient meta-analysis genetic association studies using summary statistics. PLoS Genet. 10, e1004383 of genomewide association scans. Bioinformatics 26, 2190–2191 (2010). (2014). 61. Hemani, G. et al. The MR-Base platform supports systematic causal inference 32. Zheng, J. et al. Phenome-wide Mendelian randomization mapping the across the human phenome. Elife 7, e34408 (2018). influence of the plasma proteome on complex diseases. Nat. Genet. 52, 62. Del Greco M, F., Minelli, C., Sheehan, N. A. & Thompson, J. R. Detecting 1122–1131 (2020). pleiotropy in Mendelian randomisation studies with summary data and a 33. Kia, D. A. et al. Integration of eQTL and Parkinson’s disease GWAS data continuous outcome. Stat. Med. 34, 2926–2940 (2015). implicates 11 disease genes. JAMA Neurol.https://doi.org/10.1101/627216 (2020). 34. Escott-Price, V. et al. Polygenic risk of Parkinson disease is correlated with Acknowledgements disease age at onset. Ann. Neurol. 77, 582–591 (2015). C.S.S. would like to thank Dr. Vishal Rawji for his invaluable support and insightful ideas 35. Ibanez, L. et al. Parkinson disease polygenic risk score is associated with about the clinical implications and communication of this study. C.S.S. is funded by Parkinson disease status and age at onset but not with alpha-synuclein Rosetrees Trust, John Black Charitable Foundation and the University College London cerebrospinal fluid levels. BMC Neurol. 17, 1–9 (2017). MBPhD Programme. D.A.K. is supported by an MBPhD Award from the International 36. Nalls, M. A. et al. Genetic risk and age in Parkinson’s disease: continuum not Journal of Experimental Pathology. M.A. is funded by the Faculty of Applied Medical stratum. Mov. Disord. 30, 850–854 (2015). Sciences, King Abdulaziz University, Jeddah, Saudi Arabia. N.W.W. and A.D.H. are 37. Sun, B. B. et al. Genomic atlas of the human plasma proteome. Nature 558, National Institute for Health Research senior investigators. N.W.W., A.D.H. and C.F. 73–79 (2018). receive support from the National Institute for Health Research University College 38. Emilsson, V. et al. Co-regulatory networks of human serum proteins link London Hospitals Biomedical Research Centre. We would like to thank all members of genetics to disease. Science 361, 769–773 (2018). the International Parkinson Disease Genomics Consortium (IPDGC) and the authors of 39. Suhre, K. et al. Connecting genetic risk to disease end points through the QTL projects referenced here, who make their data openly available. We thank all the human blood plasma proteome. Nat. Commun. 8, 14357 (2017). patients and families whose decision to donate tissue samples make our research possible. 40. Hillary, R. F. et al. Genome and epigenome wide studies of neurological protein biomarkers in the Lothian Birth Cohort 1936. Nat. Commun. 10, 3160 (2019). Author contributions 41. Foltynie, T. & Athauda, D. Repurposing anti-diabetic drugs for the treatment of Conceptualization: C.S.S., D.A.K., M.M.A. and N.W.W.; methodology: C.S.S., D.A.K., Parkinson’s disease: Rationale and clinical experience. Progress in Brain M.M.A. and N.W.W.; investigation, formal analysis, visualization: C.S.S.; resources: Research 252 (Elsevier B.V., 2020). S.B.C., C.F., A.D.H. and IPDGC; writing—original draft: CSS; writing—review & editing: 42. Rotermund, C., Machetanz, G. & Fitzgerald, J. C. The therapeutic potential of C.S.S., D.A.K., M.M.A., N.W.W., S.B.C., C.F., A.D.H. and IPDGC. metformin in neurodegenerative diseases. Front. Endocrinol. 9, 1–26 (2018). 43. Mor, D. E. et al. Metformin rescues Parkinson’s disease phenotypes caused by hyperactive mitochondria. PNAS 117, 26438–26447 (2020). Competing interests The authors declare no competing interests. NATURE COMMUNICATIONS | (2021)12:7342 | https://doi.org/10.1038/s41467-021-26280-1 | www.nature.com/naturecommunications 11 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-26280-1 Additional information Open Access This article is licensed under a Creative Commons Supplementary information The online version contains supplementary material Attribution 4.0 International License, which permits use, sharing, available at https://doi.org/10.1038/s41467-021-26280-1. adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Correspondence and requests for materials should be addressed to Nicholas W. Wood. Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless Peer review information Nature Communications thanks Neil Davies, Haydeh Payami indicated otherwise in a credit line to the material. If material is not included in the and Heiko Runz for their contribution to the peer review of this work. Peer reviewer article’s Creative Commons license and your intended use is not permitted by statutory reports are available. regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/ Reprints and permission information is available at http://www.nature.com/reprints licenses/by/4.0/. Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. © The Author(s) 2021 International Parkinson’s Disease Genomics Consortium (IPDGC) Alastair J. Noyce8,9, Rauan Kaiyrzhanov10, Ben Middlehurst11, Demis A. Kia1, Manuela Tan12, Henry Houlden10, Catherine S. Storm 1, Huw R. Morris12, Helene Plun-Favreau10, Peter Holmans13,14, John Hardy10, Daniah Trabzuni10, John Quinn11, Vivien Bubb11, Kin Y. Mok10, Kerri J. Kinghorn15, Nicholas W. Wood 1 ✉, Patrick Lewis16, Sebastian R. Schreglmann10, Ruth Lovering17, Lea R’Bibo10, Claudia Manzoni16, Mie Rizig10, Mina Ryten10, Sebastian Guelfi10, Valentina Escott-Price18, Viorica Chelban10, Thomas Foltynie19, Nigel Williams20, Karen E. Morrison21, Carl Clarke22,23, Kirsten Harvey24, Benjamin M. Jacobs8, Alexis Brice25, Fabrice Danjou25, Suzanne Lesage25, Jean-Christophe Corvol26, Maria Martinez27,28, Claudia Schulte29,30, Kathrin Brockmann29,30, Javier Simón-Sánchez29,30, Peter Heutink31, Patrizia Rizzu32, Manu Sharma33, Thomas Gasser29,30, Susanne A. Schneider34, Mark R. Cookson3, Sara Bandres-Ciga3, Cornelis Blauwendraat3, David W. Craig35, Kimberley Billingsley3, Mary B. Makarious3, Derek P. Narendra36, Faraz Faghri3,37, J. Raphael Gibbs38, Dena G. Hernandez3, Kendall Van Keuren-Jensen39, Joshua M. Shulman40,41, Hirotaka Iwaki3, Hampton L. Leonard3, Mike A. Nalls3,42, Laurie Robak43, Jose Bras44, Rita Guerreiro44, Steven Lubbe45, Timothy Troycoco46, Steven Finkbeiner47,48,49, Niccolo E. Mencacci50, Codrin Lungu51, Andrew B. Singleton3, Sonja W. Scholz52, Xylena Reed3, Ryan J. Uitti53, Owen A. Ross54, Francis P. Grenn3, Anni Moore3, Roy N. Alcalay55,56, Zbigniew K. Wszolek53, Ziv Gan-Or57, Guy A. Rouleau57, Lynne Krohn57, Kheireddin Mufti57, Jacobus J. van Hilten58, Johan Marinus58, Astrid D. Adarmes-Gómez59, Miquel Aguilar60, Ignacio Alvarez60, Victoria Alvarez61, Francisco Javier Barrero62, Jesús Alberto Bergareche Yarza63, Inmaculada Bernal-Bernal59, Marta Blazquez61, Marta Bonilla-Toribio59, Juan A. Botía64, María Teresa Boungiorno60, Dolores Buiza-Rueda59, Ana Cámara65, Fátima Carrillo59, Mario Carrión-Claro59, Debora Cerdan66, Jordi Clarimón67,68, Yaroslau Compta65, Monica Diez-Fairen60, Oriol Dols-Icardo67,68, Jacinto Duarte66, Raquel Duran69, Francisco Escamilla-Sevilla70, Mario Ezquerra65, Cici Feliz71, Manel Fernández65, Rubén Fernández-Santiago65, Ciara Garcia61, Pedro García-Ruiz72, Pilar Gómez-Garre59, Maria Jose Gomez Heredia73, Isabel Gonzalez-Aramburu74, Ana Gorostidi Pagola63, Janet Hoenicka75, Jon Infante68,76, Silvia Jesús59, Adriano Jimenez-Escrig77, Jaime Kulisevsky68,78, Miguel A. Labrador-Espinosa59, Jose Luis Lopez-Sendon77, Adolfo López de Munain Arregui63, Daniel Macias59, Irene Martínez Torres79, Juan Marín68,78, Maria Jose Marti80, Juan Carlos Martínez-Castrillo81, Carlota Méndez-del-Barrio59, Manuel Menéndez González61, Marina Mata82, Adolfo Mínguez82, Pablo Mir59, Elisabet Mondragon Rezola63, Esteban Muñoz80, Javier Pagonabarraga78,83, Pau Pastor60, Francisco Perez Errazquin73, Teresa Periñán-Tocino59, Javier Ruiz-Martínez84, Clara Ruz69, Antonio Sanchez Rodriguez74, María Sierra74, Esther Suarez-Sanmartin61, Cesar Tabernero66, 12 NATURE COMMUNICATIONS | (2021)12:7342 | https://doi.org/10.1038/s41467-021-26280-1 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-26280-1 ARTICLE Juan Pablo Tartari60, Cristina Tejera-Parrado59, Eduard Tolosa80, Francesc Valldeoriola80, Laura Vargas-González59, Lydia Vela85, Francisco Vives69, Alexander Zimprich86, Lasse Pihlstrom87, Mathias Toft88, Pille Taba89, Sulev Koks90,91, Sharon Hassin-Baer92,93, Kari Majamaa94,95, Ari Siitonen94,95, Pentti Tienari96,97, Njideka U. Okubadejo98, Oluwadamilola O. Ojo98, Chingiz Shashkin99, Nazira Zharkinbekova100, Vadim Akhmetzhanov101, Gulnaz Kaishybayeva102, Altynay Karimova102, Talgat Khaibullin103 & Timothy L. Lynch104,105 8 Preventive Neurology Unit, Wolfson Institute of Preventive Medicine, QMUL, London, UK. 9Department of Molecular Neuroscience, UCL, London, UK. 10Department of Molecular Neuroscience, UCL Institute of Neurology, London, UK. 11Institute of Translational Medicine, University of Liverpool, Liverpool, UK. 12Department of Clinical Neuroscience, University College London, London, UK. 13Biostatistics and Bioinformatics Unit, Institute of Psychological Medicine and Clinical Neuroscience, Cardiff, UK. 14MRC Centre for Neuropsychiatric Genetics & Genomics, Cardiff, UK. 15 Institute of Healthy Ageing, Research Department of Genetics, Evolution and Environment, University College London, London, UK. 16University of Reading, Reading, UK. 17University College London, London, UK. 18MRC Centre for Neuropsychiatric Genetics and Genomics, Cardiff University School of Medicine, Cardiff, UK. 19UCL Institute of Neurology, London, UK. 20MRC Centre for Neuropsychiatric Genetics and Genomics, Cardiff, UK. 21 Faculty of Medicine, University of Southampton, Southampton, UK. 22University of Birmingham, Birmingham, UK. 23Sandwell and West Birmingham Hospitals NHS Trust, Birmingham, UK. 24UCL School of Pharmacy, London, UK. 25Institut du Cerveau et de la Moelle épinière, ICM, Inserm U 1127, CNRS, UMR 7225, Sorbonne Universités, UPMC University Paris 06, UMR S 1127, AP-HP, Pitié-Salpêtrière Hospital, Paris, France. 26 Institut du Cerveau et de la Moelle épinière, ICM, Inserm U 1127, CNRS, UMR 7225, Sorbonne Universités, UPMC University Paris 06, UMR S 1127, Centre d’Investigation Clinique Pitié Neurosciences CIC-1422, AP-HP, Pitié-Salpêtrière Hospital, Paris, France. 27INSERM UMR 1220, Toulouse, France. 28Paul Sabatier University, Toulouse, France. 29Department for Neurodegenerative Diseases, Hertie Institute for Clinical Brain Research, University of Tübingen, Tübingen, Germany. 30DZNE, German Center for Neurodegenerative Diseases, Tübingen, Germany. 31DZNE, German Center for Neurodegenerative Diseases and Department for Neurodegenerative Diseases, Hertie Institute for Clinical Brain Research, University of Tübingen, Tübingen, Germany. 32DZNE, German Center for Neurodegenerative Diseases, Bonn, Germany. 33Centre for Genetic Epidemiology, Institute for Clinical Epidemiology and Applied Biometry, University of Tubingen, Tübingen, Germany. 34Department of Neurology, Ludwig-Maximilians-University Munich, München, Germany. 35Department of Translational Genomics, Keck School of Medicine, University of Southern California, Los Angeles, CA, USA. 36Inherited Movement Disorders Unit, National Institute of Neurological Disorders and Stroke, Bethesda, MD, USA. 37Department of Computer Science, University of Illinois at Urbana-Champaign, Urbana, IL, USA. 38Laboratory of Neurogenetics, National Institute on Aging, National Institutes of Health, Bethesda, MD, USA. 39Neurogenomics Division, TGen, Phoenix, AZ, USA. 40 Departments of Neurology, Neuroscience, and Molecular & Human Genetics, Baylor College of Medicine, Houston, TX, USA. 41Jan and Dan Duncan Neurological Research Institute, Texas Children’s Hospital, Houston, TX, USA. 42CEO/Consultant Data Tecnica International, Glen Echo, MD, USA. 43Baylor College of Medicine, Houston, TX, USA. 44Center for Neurodegenerative Science, Van Andel Research Institute, Grand Rapids, MI, USA. 45Ken and Ruth Davee Department of Neurology and Simpson Querrey Center for Neurogenetics, Northwestern University Feinberg School of Medicine, Chicago, IL, USA. 46National Institutes of Health, Bethesda, MD, USA. 47Departments of Neurology and Physiology, University of California, San Francisco, CA, USA. 48Gladstone Institute of Neurological Disease, San Francisco, CA, USA. 49Taube/Koret Center for Neurodegenerative Disease Research, San Francisco, CA, USA. 50Northwestern University Feinberg School of Medicine, Chicago, IL, USA. 51 National Institutes of Health Division of Clinical Research, NINDS, National Institutes of Health, Bethesda, MD, USA. 52Neurodegenerative Diseases Research Unit, National Institute of Neurological Disorders and Stroke, Bethesda, MD, USA. 53Department of Neurology, Mayo Clinic, Jacksonville, FL, USA. 54Departments of Neuroscience & Clinical Genomics, Mayo Clinic, Jacksonville, FL, USA. 55Department of Neurology, College of Physicians and Surgeons, Columbia University Medical Center, Taub Institute for Research on Alzheimer’s Disease, New York, NY, USA. 56The Aging Brain, College of Physicians and Surgeons, Columbia University Medical Center, New York, NY, USA. 57Montreal Neurological Institute and Hospital, Department of Neurology & Neurosurgery, Department of Human Genetics, McGill University, Montréal, QC H3A 0G4, Canada. 58 Department of Neurology, Leiden University Medical Center, Leiden, Netherlands. 59Instituto de Biomedicina de Sevilla IBiS Hospital Universitario Virgen del Rocío/CSIC/Universidad de Sevilla, Seville, Spain. 60Fundació Docència i Recerca Mútua de Terrassa and Movement Disorders Unit, Department of Neurology, University Hospital Mutua de Terrassa, Terrassa, Barcelona, Spain. 61Hospital Universitario Central de Asturias, Oviedo, Spain. 62Hospital Universitario San Cecilio de Granada, Universidad de Granada, Granada, Spain. 63Instituto de Investigación Sanitaria Biodonostia, San Sebastián, Spain. 64Universidad de Murcia, Murcia, Spain. 65Hospital Clinic de Barcelona, Barcelona, Spain. 66Hospital General de Segovia, Segovia, Spain. 67Memory Unit, Department of Neurology, IIB Sant Pau, Hospital de la Santa Creu i Sant Pau, Universitat Autònoma de Barcelona, Barcelona, Spain. 68Centro de Investigación Biomédica en Red en Enfermedades Neurodegenerativas (CIBERNED), Madrid, Spain. 69 Centro de Investigacion Biomedica, Universidad de Granada, Granada, Spain. 70Hospital Universitario Virgen de las Nieves, Instituto de Investigación Biosanitaria de Granada, Granada, Spain. 71Departmento de Neurologia, Instituto de Investigación Sanitaria Fundación Jiménez Díaz, Madrid, Spain. 72Instituto de Investigación Sanitaria Fundación Jiménez Díaz, Madrid, Spain. 73Hospital Universitario Virgen de la Victoria, Malaga, Spain. 74Hospital Universitario Marqués de Valdecilla-IDIVAL, Santander, Spain. 75Institut de Recerca Sant Joan de Déu, Barcelona, Spain. 76 Hospital Universitario Marqués de Valdecilla-IDIVAL and University of Cantabria, Santander, Spain. 77Hospital Universitario Ramón y Cajal, Madrid, Spain. 78Movement Disorders Unit, Department of Neurology, IIB Sant Pau, Hospital de la Santa Creu i Sant Pau, Universitat Autònoma de Barcelona, Barcelona, Spain. 79Department of Neurology, Instituto de Investigación Sanitaria La Fe, Hospital Universitario y Politécnico La Fe, Valencia, Spain. 80Hospital Clinic Barcelona, Barcelona, Spain. 81Instituto Ramón y Cajal de Investigación Sanitaria, Hospital Universitario Ramón y Cajal, Madrid, Spain. 82Hospital Universitario Virgen de las Nieves, Granada, Instituto de Investigación Biosanitaria de, Granada, Spain. 83 Department of Neurology, Hospital Universitario Infanta Sofía, Madrid, Spain. 84Hospital Universitario Donostia, Instituto de Investigación Sanitaria Biodonostia, San Sebastián, Spain. 85Department of Neurology, Hospital Universitario Fundación Alcorcón, Madrid, Spain. 86Department of Neurology, Medical University of Vienna, Vienna, Austria. 87Department of Neurology, Oslo University Hospital, Oslo, Norway. 88Department of Neurology and Institute of Clinical Medicine, Oslo University Hospital, Oslo, Norway. 89Department of Neurology and Neurosurgery, University of Tartu, Tartu, Estonia. 90Centre for Molecular Medicine and Innovative Therapeutics, Murdoch University, Murdoch, 6150 Perth, Western Australia, Australia. 91The Perron Institute for Neurological and Translational Science, Nedlands, 6009 Perth, Western Australia, Australia. 92The Movement Disorders Institute, Department of Neurology and Sagol Neuroscience Center, Chaim Sheba Medical Center, Tel-Hashomer, 5262101 Ramat NATURE COMMUNICATIONS | (2021)12:7342 | https://doi.org/10.1038/s41467-021-26280-1 | www.nature.com/naturecommunications 13 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-26280-1 Gan, Israel. 93Sackler Faculty of Medicine, Tel Aviv University, Tel Aviv, Israel. 94Institute of Clinical Medicine, Department of Neurology, University of Oulu, Oulu, Finland. 95Department of Neurology and Medical Research Center, Oulu University Hospital, Oulu, Finland. 96Clinical Neurosciences, Neurology, University of Helsinki, Helsinki, Finland. 97Helsinki University Hospital, Helsinki, Finland. 98University of Lagos, Lagos State, Nigeria. 99 Kazakh National Medical University named after Asfendiyarov, Almaty, Kazakhstan. 100South Kazakhstan Medical Academy, Shymkent, Kazakhstan. 101Astana Medical University, Astana, Kazakhstan. 102Scientific and Practical Center “Institute of Neurology named after Smagul Kaishibayev”, Almaty, Kazakhstan. 103Semey Medical University, Semey, Kazakhstan. 104School of Medicine and Medical Science, University College Dublin, Dublin, Ireland. 105The Dublin Neurological Institute at the Mater Misericordiae University Hospital, Dublin, Ireland. 14 NATURE COMMUNICATIONS | (2021)12:7342 | https://doi.org/10.1038/s41467-021-26280-1 | www.nature.com/naturecommunications