Although over 90 independent risk variants have been identified for Parkinson’s disease using genome-wide association studies, most studies have been performed in just one population at a time. Here we performed a large-scale multi-ancestry meta-analysis of Parkinson’s disease with 49,049 cases, 18,785 proxy cases and 2,458,063 controls including individuals of European, East Asian, Latin American and African ancestry. In a meta-analysis, we identified 78 independent genome-wide significant loci, including 12 potentially novel loci (MTF2, PIK3CA, ADD1, SYBU, IRS2, USP8, PIGL, FASN, MYLK2, USP25, EP300 and PPP6R2) and fine-mapped 6 putative causal variants at 6 known PD loci. By combining our results with publicly available eQTL data, we identified 25 putative risk genes in these novel loci whose expression is associated with PD risk. This work lays the groundwork for future efforts aimed at identifying PD loci in non-European populations.

Parkinson’s disease is a genetically complex disorder. Multiple genes have been shown to contribute to the risk of Parkinson’s disease, and currently 90 independent risk variants have been identified by genome-wide association studies. Thus far, a number of genes (including SNCA, LRRK2, and GBA) have been shown to contain variability across a spectrum of frequency and effect, from rare, highly penetrant variants to common risk alleles with small effect sizes. Variants in GBA, encoding the enzyme glucocerebrosidase, are associated with Lewy body diseases such as Parkinson’s disease and Lewy body dementia. These variants, which reduce or abolish enzymatic activity, confer a spectrum of disease risk, from 1.4- to >10-fold. An outstanding question in the field is what other genetic factors that influence GBA-associated risk for disease, and whether these overlap with known Parkinson’s disease risk variants. Using multiple, large case-control datasets, totalling 217 165 individuals (22 757 Parkinson’s disease cases, 13 431 Parkinson’s disease proxy cases, 622 Lewy body dementia cases and 180 355 controls), we identified 1691 Parkinson’s disease cases, 81 Lewy body dementia cases, 711 proxy cases and 7624 controls with a GBA variant (p.E326K, p.T369M or p.N370S). We performed a genome-wide association study and analysed the most recent Parkinson’s disease-associated genetic risk score to detect genetic influences on GBA risk and age at onset. We attempted to replicate our findings in two independent datasets, including the personal genetics company 23andMe, Inc. and whole-genome sequencing data. Our analysis showed that the overall Parkinson’s disease genetic risk score modifies risk for disease and decreases age at onset in carriers of GBA variants. Notably, this effect was consistent across all tested GBA risk variants. Dissecting this signal demonstrated that variants in close proximity to SNCA and CTSB (encoding cathepsin B) are the most significant contributors. Risk variants in the CTSB locus were identified to decrease mRNA expression of CTSB. Additional analyses suggest a possible genetic interaction between GBA and CTSB and GBA p.N370S induced pluripotent cell-derived neurons were shown to have decreased cathepsin B expression compared to controls. These data provide a genetic basis for modification of GBA-associated Parkinson’s disease risk and age at onset, although the total contribution of common genetics variants is not large. We further demonstrate that common variability at genes implicated in lysosomal function exerts the largest effect on GBA associated risk for disease. Further, these results have implications for selection of GBA carriers for therapeutic interventions.

Personalized medicine promises individualized disease prediction and treatment. The convergence of machine learning (ML) and available multimodal data is key moving forward. We build upon previous work to deliver multimodal predictions of Parkinson’s disease (PD) risk and systematically develop a model using GenoML, an automated ML package, to make improved multi-omic predictions of PD, validated in an external cohort. We investigated top features, constructed hypothesis-free disease-relevant networks, and investigated drug–gene interactions. We performed automated ML on multimodal data from the Parkinson’s progression marker initiative (PPMI). After selecting the best performing algorithm, all PPMI data was used to tune the selected model. The model was validated in the Parkinson’s Disease Biomarker Program (PDBP) dataset. Our initial model showed an area under the curve (AUC) of 89.72% for the diagnosis of PD. The tuned model was then tested for validation on external data (PDBP, AUC 85.03%). Optimizing thresholds for classification increased the diagnosis prediction accuracy and other metrics. Finally, networks were built to identify gene communities specific to PD. Combining data modalities outperforms the single biomarker paradigm. UPSIT and PRS contributed most to the predictive power of the model, but the accuracy of these are supplemented by many smaller effect transcripts and risk SNPs. Our model is best suited to identifying large groups of individuals to monitor within a health registry or biobank to prioritize for further testing. This approach allows complex predictive models to be reproducible and accessible to the community, with the package, code, and results publicly available.