www.nature.com/npjparkd ARTICLE OPEN Moving beyond neurons: the role of cell type-specific gene regulation in Parkinson’s disease heritability Regina H. Reynolds 1, Juan Botía1,2, Mike A. Nalls3,4, International Parkinson’s Disease Genomics Consortium (IPDGC), System Genomics of Parkinson’s Disease (SGPD), John Hardy1,5, Sarah A. Gagliano Taliun6 and Mina Ryten1 Parkinson’s disease (PD), with its characteristic loss of nigrostriatal dopaminergic neurons and deposition of α-synuclein in neurons, is often considered a neuronal disorder. However, in recent years substantial evidence has emerged to implicate glial cell types, such as astrocytes and microglia. In this study, we used stratified LD score regression and expression-weighted cell-type enrichment together with several brain-related and cell-type-specific genomic annotations to connect human genomic PD findings to specific brain cell types. We found that PD heritability attributable to common variation does not enrich in global and regional brain annotations or brain-related cell-type-specific annotations. Likewise, we found no enrichment of PD susceptibility genes in brain- related cell types. In contrast, we demonstrated a significant enrichment of PD heritability in a curated lysosomal gene set highly expressed in astrocytic, microglial, and oligodendrocyte subtypes, and in LoF-intolerant genes, which were found highly expressed in almost all tested cellular subtypes. Our results suggest that PD risk loci do not lie in specific cell types or individual brain regions, but rather in global cellular processes detectable across several cell types. npj Parkinson’s Disease (2019)5:6 ; https://doi.org/10.1038/s41531-019-0076-6 INTRODUCTION related pathways and, by implication, neurons in both sporadic Late-onset sporadic forms of neurodegenerative diseases are and Mendelian forms of PD. Despite this neuronal focus, there is devastating conditions imposing an increasing burden on health- also growing evidence to suggest the involvement of other cell care systems worldwide. Currently, 2–3% of the population over 65 types in PD pathogenesis. In particular, astrocytes and microglia years of age are living with Parkinson’s disease (PD), making this have been highlighted;10,11 for instance, with a recent study disorder the most prevalent late-onset neurodegenerative disorder demonstrating that blocking the microglial-mediated conversion worldwide after Alzheimer’s disease.1 This progressive condition is of astrocytes to an A1 neurotoxic phenotype was neuroprotective characterised by the loss of dopaminergic neurons in the in mouse models of sporadic and familial α-synucleinopathy.12 substantia nigra pars compacta manifesting clinically as a tremor In previous work, we applied stratified LD score regression and at rest, muscle rigidity and bradykinesia.1,2 Existing symptomatic gene-set enrichment methods to determine if particular functional treatments do not alter the course of the disease and their marks for regulatory activity and gene-set lists were enriched for effectiveness declines with time, which makes the identification of sporadic PD genetic heritability.13 We did not observe enrichment potential therapeutic targets of key importance. for the various brain annotations assessed (this did not include The primary focus of PD research to date has been on neurons brain-relevant cell types) and in fact found further evidence for the and, more specifically, nigrostriatal dopaminergic neurons. This importance of the adaptive and innate immune system. focus is driven in part because the death of dopaminergic neurons The increasing power of GWASs (with the most recently is primarily responsible for the motor features of PD, but also published PD GWAS including 37.7K cases, 18.6K ‘proxy-cases’ because the most prominent and distinctive neuropathological and 1.4M controls, resulting in 90 associated loci14) coupled with findings in PD are the presence of neuronal inclusions, termed the increased availability of cell-specific gene expression data Lewy bodies.1,2 The findings that alpha-synuclein (encoded by the provides a new opportunity to address the potential cellular gene SNCA) is predominantly expressed in neurons,2,3 is the major specificity of disease heritability, as was elegantly demonstrated component of Lewy bodies,3,4 and mutations in SNCA give rise to for schizophrenia in a study by Skene et al.15 Brain regions contain autosomal dominant PD,5–8 provide a key link between SNCA a mixture of cell types, such as neurons, microglia and astrocytes, function, neurons and disease pathogenesis. Furthermore, the which may exhibit their own specific regulatory features that identification of risk single nucleotide polymorphisms (SNPs) at could be masked when averaging features across cell types. the SNCA locus through genome-wide association studies (GWAS) Resolving this question has become increasingly important; with of sporadic PD9 provides support for the importance of SNCA- the advent of induced pluripotent stem cell models of disease, 1 Department of Neurodegenerative Disease, University College London (UCL) Institute of Neurology, London, UK; 2Departamento de Ingeniería de la Información y las Comunicaciones, Universidad de Murcia, Murcia, Spain; 3Laboratory of Neurogenetics, National Institute on Aging, US National Institutes of Health, Bethesda, Maryland, USA; 4 Data Tecnica International, Glen Echo, Maryland, USA; 5UK Dementia Research Institute at University College London (UCL), London, UK and 6Center for Statistical Genetics and Department of Biostatistics, University of Michigan, Ann Arbor, Michigan, USA Correspondence: Mina Ryten (mina.ryten@ucl.ac.uk) A list of consortium members appears before the Acknowledgements. These authors contributed equally: Sarah A. Gagliano Taliun, Mina Ryten Received: 14 November 2018 Accepted: 28 February 2019 Published in partnership with the Parkinson’s Foundation R.H. Reynolds et al. 2 modelling PD in vitro is now possible, and this implies some relative to other brain regions for SCZ, but saw no enrichments for decision about the cell type of interest. In this study, we addressed PD (Fig. 2b, Supplementary Table 1). cellular heterogeneity through the analysis of genomic regions We also compared the PD and SCZ GWAS results to sets of overlapping regulatory marks or gene expression from cell types blood- and brain-specific eQTLs derived from GTEx. We demon- within the brain, including neurons. We focus on PD GWAS strated an enrichment of SCZ heritability in brain-specific eQTLs datasets and use schizophrenia (SCZ) GWAS datasets for and blood-specific eQTLs, but no enrichment of PD heritability in comparisons purposes. either eQTL annotation (Fig. 3a, Supplementary Table 2). A comparison of eQTLs specific to each brain region revealed no preferential enrichment of disease heritability in one region RESULTS relative to the others (Fig. 3b, Supplementary Table 2). In Overview of methods summary, these analyses revealed no enrichment of PD To study the cellular specificity of the heritability of sporadic PD, heritability in brain annotations, while in contrast, SCZ heritability we compiled brain-related genomic annotations denoting tissue- was highly enriched in both global and specific regional brain and cell-type-specific markers of activity. We used several annotations. approaches to capture the expression profiles of human brain- related cell types (Fig. 1, see Methods). This was because no single PD heritability in brain-related cell-type annotations data set had all the desirable properties; namely, data that was Given the lack of enrichment of PD heritability in global and human in origin, covered multiple brain regions, had high cellular regional brain annotations, we wondered whether cellular detail and was derived from large numbers of individuals. Using heterogeneity may be masking signals, and provided more cell- the largest publicly available GWASs of PD16 and SCZ,17 we type-specific information the enrichment would become more applied stratified LD score regression (LDSC) to assess enrichment apparent. Thus, to address the relative importance of brain cell of the common-SNP heritability of PD and SCZ, respectively, for types in PD and SCZ, we generated cell-type-specific annotations each annotation category. SCZ heritability has been previously from three types of brain-related cell-type-specific data: bulk RNA- shown to be enriched in genes expressed within the central sequencing from the Barres group of immunopanned cell types 1234567890():,; nervous system (CNS) and, more specifically, neuronal cell from human temporal lobe cortex;21 single-cell RNA-sequencing types,15,18 and was therefore included as a measure of robustness. from the Linnarsson group of the adolescent mouse nervous For all stratified-LDSC analyses, we report coefficient p-values, system;22 and finally, cell-type modules inferred from human which test whether the regression coefficient of an annotation tissue-level co-expression networks.23 Genes were assigned to cell category positively contributes to trait heritability, conditional types by fold enrichment (i.e. mean expression in one cell type upon the LDSC baseline model, which accounts for underlying divided by the mean expression in all other cell types) or module genetic architecture. membership in the case of co-expression (module membership is a measure of how correlated a gene’s expression is with respect to PD heritability in human brain a module’s eigengene). Each of these datasets came with advantages and disadvan- It is well recognised that regional differences in gene expression tages, which motivated our decision to use all three. The Barres within human brain and related co-expression modules are driven data was based on the analysis of human tissue; however, it by differences in the type and density of specific cell types.19 covered only one brain region, was derived from a small number Therefore, we first used regional data as a means of capturing of individuals (n = 14) who all had an underlying neurological major cellular profiles. This information, comprising sets of human disorder (epilepsy, stroke and glioma), and lacked cellular detail. tissue-specific genes generated by Finucane et al. with GTEx gene While the cell-type-specific data provided by the Linnarsson group expression,18,20 had the advantage of being comprehensive in covered both the central and peripheral nervous system, and terms of sampling across the human CNS, and being robust in that contained remarkable cellular detail, it was mouse in origin. Cell- greater than 63 independent samples contributed to the type modules also covered several brain regions, were based on generation of each profile. We confirmed that SCZ heritability large sample sizes, and importantly, were human in origin. was significantly enriched in all 13 brain regions relative to all Nevertheless, they were inferred cell types, the definition of other tissues, as previously demonstrated by Finucane et al.18 which was strongly dependent on the quality of the cell-type using the 2014 SCZ GWAS (Fig. 2a, Supplementary Table 1). In markers used to identify them. contrast, no tissues were enriched for PD heritability, although Using immunopanning data, we identified a neuronal enrich- spinal cord and substantia nigra approached the Bonferroni ment for SCZ heritability, but no cell-type enrichment for PD significance threshold (threshold p-value = 4.72 × 10−4; spinal (Fig. 4a, Supplementary Table 3). We questioned whether this lack cord (cervical c-1), p-value = 1.36 × 10−3; substantia nigra, of cell-type enrichment in PD may result from sampling a tissue p-value = 8.96 × 10−4). Our comparison of PD and SCZ GWAS which is typically affected only in the later stages of sporadic PD.2 iterations across the years revealed the robust nature of the CNS Thus, we analysed a subset of mouse single-cell data representing enrichment in SCZ, which was apparent in the first and smallest tissues affected in earlier stages of sporadic PD, including the SCZ GWAS (Supplementary Figure 1). Furthermore, increasing enteric nervous system, the substantia nigra and the basal ganglia. GWAS sample sizes were associated with coefficient p-values Once again, we found no cell-type enrichment for PD heritability becoming more significant, particularly for CNS-related tissues. (Fig. 4b, Supplementary Table 3). Conversely, we demonstrated a Interestingly, we also observed an ordering of tissues, with brain significant enrichment of SCZ heritability in three types of regions of greater relevance to disease pathology demonstrating GABAergic medium spiny neurons (MSNs): MSN2, MSN3 and the most significant coefficient p-values in the largest GWAS MSN5. This is consistent with the findings reported by Skene iterations (e.g. substantia nigra in PD and frontal cortex in SCZ). et al.15 Common to all three types of MSN is that they express the However, due to the way these annotations were constructed, D2 dopamine receptor, a common target of antipsychotic drugs related tissues (e.g. brain regions) have overlapping gene sets and used in SCZ therapy.24 therefore may appear enriched as a group. To differentiate among To our knowledge, there is currently no single-cell RNA- brain regions, we used fine-scale brain expression data generated sequencing data for human striatum or substantia nigra, so we by Finucane et al. from a brain-only analysis of the 13 GTEx brain sought to validate our findings using cell-type modules inferred regions.18 We confirmed significant enrichments in the cortex from co-expression networks constructed from human tissue-level npj Parkinson’s Disease (2019) 6 Published in partnership with the Parkinson’s Foundation R.H. Reynolds et al. 3 Fig. 1 Overview of approach and datasets used. This study compiled several brain-related genomic annotations reflecting tissue- and cell- type-specific activity, using data generated by the GTEx project,20 the Barres group21 and the Linnarsson group.22 These annotations, each of which varied in their cellular resolution, included: tissue-specific eQTLs (reflecting the effect of genetic variation on gene expression); tissue- specific co-expression networks (reflecting the connectivity of a gene to all other expressed genes in the tissue), and tissue- and cell-type- specific gene expression. All annotations were constructed in a binary format (1 if the SNP is present within the annotation and 0 if not). For annotations where the primary input was a gene, all SNPs with a minor allele frequency > 5% within ± 100 kb of the transcription start and end site were assigned a value of 1. For more details of how each individual annotation was generated see Methods. Stratified LDSC was then used to test whether an annotation was significantly enriched for the common-SNP heritability of PD or SCZ expression data of the frontal cortex, putamen and substantia Supplementary Table 4). All of these modules were enriched for nigra. We observed no significant enrichments for PD heritability markers of pyramidal S1 neurons, which have previously been in any modules, while SCZ heritability was enriched in several associated with SCZ.15 Furthermore, some modules (brown, neuronal modules, including: brown and turquoise modules in the turquoise, dark magenta and cyan) were enriched for markers of frontal cortex; blue and dark magenta modules in the putamen; interneurons and dopaminergic neurons, both of which are and cyan and darkgrey modules in the substantia nigra (Fig. 5, implicated in SCZ.15,24 Published in partnership with the Parkinson’s Foundation npj Parkinson’s Disease (2019) 6 R.H. Reynolds et al. 4 PD susceptibility genes in brain-related cell types weighted cell-type enrichment (EWCE). This method statistically To ensure rigour, we attempted to identify cell types of evaluates whether a set of genes has higher expression in one cell importance in PD in a separate analysis using expression- type than expected by chance. Using the same subset of clusters npj Parkinson’s Disease (2019) 6 Published in partnership with the Parkinson’s Foundation R.H. Reynolds et al. 5 Fig. 2 Enrichment of PD and SCZ common-SNP heritability in tissue-specific gene expression annotations as used in Finucane et al.18 a Stratified LDSC analyses showed significant enrichment of SCZ heritability in all GTEx brain regions but no enrichment of PD heritability. GTEx tissue annotations represent the top 10% most upregulated genes in each tissue with respect to the remaining tissues, excluding those from a similar tissue category. b Stratified LDSC analyses showed significant enrichment of SCZ heritability in cortical brain regions, but no enrichment for PD heritability. GTEx brain-only annotations represent the top 10% most upregulated genes in each brain region with respect to the remaining regions. Tissues were ordered within each tissue category by the coefficient p-value obtained for SCZ. The black dashed lines indicate the cut-off for Bonferroni significance (a, p < 0.05/(2 × 53); b, p < 0.05/(2 × 13)). Bonferroni-significant results are marked with black borders. The proportion of SNPs accounted for by each annotation (compared to the baseline model), the regression coefficient calculated for the latest PD and SCZ GWASs, and the coefficient p-values for previous iterations of the PD and SCZ GWASs are displayed in Supplementary Figs 1–4. Numerical results are reported in Supplementary Table 1 from the Linnarsson single-cell RNA-sequencing, cell-type speci- subtypes. Autophagy and lysosomal gene sets were significantly ficity values were computed for each gene (i.e. proportion of enriched in a limited number of major cell-type classes, with expression of a gene in a given cell type), and cell-type autophagy enriched in oligodendrocytes and cholinergic/mono- enrichments of PD susceptibility genes implicated by common- aminergic neurons, and lysosomal enriched in microglia (Fig. 7b, variant studies were estimated (Fig. 6a, see Methods). Suscept- Supplementary Table 6). The mitochondrial gene set, on the other ibility genes were derived using MAGMA,25 which estimates the hand, was significantly enriched in almost all cell-type classes, association between each gene and a phenotype while account- including astrocytes, oligodendrocytes, oligodendrocyte precursor ing for LD, and from a study by Li et al.26 using TWAS27 and cells, cholinergic/monoaminergic neurons and telencephalon coloc,28 which evaluate the association between eQTLs and GWAS projecting inhibitory neurons. Likewise, the LoF-intolerant gene risk loci. We found no significant enrichment of PD susceptibility set was found significantly enriched in all cell-type classes. As genes in any of the major cell-type classes (Fig. 6b, Supplementary expected, analyses performed on cell subtypes predominantly Table 5) or their cell subtypes (Fig. 6c, Supplementary Table 5). reflected those performed on the overarching cell-type classes, In summary, our EWCE and stratified LDSC analyses would with significant pathway enrichments observed in cell subtypes suggest that PD heritability/susceptibility cannot be attributed to associated with the pathway-enriched cell-type classes (Fig. 7c, a specific cell type (amongst those tested), unlike what has been Supplementary Table 6). For example, all three microglial subtypes observed by us and others for SCZ,15 wherein a limited set of (MGL1-3, representing one baseline and two activated microglial neuronal cell types have been implicated. subtypes), were enriched for lysosomal genes. Interestingly, the subtype analyses also revealed a significant enrichment of the lysosomal gene set in two astrocytic subtypes, ACNT1 and ACNT2 PD heritability in PD-relevant genes sets (both non-telencephalon astrocytes, but protoplasmic and fibrous, Risk loci can operate in several manners, including: a cell-type-/ respectively), and in two oligodendrocyte subtypes, MFOL1 and tissue-specific manner, which is only detectable if measured in the MOL1 (myelin forming oligodendrocytes and mature oligoden- “correct” cell type/tissue, or in a pathway-specific manner, which drocytes, respectively), which was not reflected when using the one might expect to be detectable across more than one cell type/ major cell-type classes, suggesting that analyses performed with tissue. Given our inability to implicate a cell type in PD, we cellular subtypes provide greater resolution. It is worth noting that wondered whether the latter scenario of pathway-specific risk despite higher than expected expression of the lysosomal gene might be applicable in PD. set in astrocytic, microglial and oligodendrocyte subtypes, the To address this question, we applied stratified LDSC to gene lysosomal gene set is ubiquitously expressed across all cellular sets implicated in PD by Mendelian forms of PD, functional assays subtypes (Supplementary Figure 19). In other words, higher than performed in the context of PD-associated mutations, such as the expected expression is not necessarily equivalent to exclusive A53T missense mutation in SNCA, and rare-variant studies of expression in a cell type. sporadic PD.29–34 In particular, we focused on gene sets associated Taken together these findings provide support for the view that with autophagy,30,31 the lysosomal system32 and mitochondrial in contrast to the genetic structure of SCZ, PD risk loci operate in a function.33,34 Our gene sets were derived either from Gene more global manner enriching within pathways and gene sets, Ontology terms (autophagy) or curated gene databases (lysoso- which are highly expressed across several cell subtypes. mal, hLGDB; mitochondrial, MitoCarta 2.0; see Methods), devel- oped using literature curation (with a focus on unbiased proteomic studies) and experimental approaches. As with DISCUSSION previous stratified LDSC analyses, we included SCZ for comparison One of the most striking features of PD is the specificity of its purposes. In addition, we used a gene set comprising loss-of- neuropathology and clinical symptoms, which has implicated α- function (LoF)-intolerant genes, as defined by the Exome synuclein biology in dopaminergic neurons of the substantia nigra Aggregation Consortium (ExAC)35 using their gene-level constraint pars compacta as a key component of the disease.1,2 This stands in metric (pLI ≥ 0.9), which has been previously shown enriched for stark contrast to SCZ, which has a very heterogeneous clinical SCZ heritability17 and thus would serve as a positive control. The phenotype and lacks a characteristic neuropathology,36,37 with a overlap between these gene sets was relatively low (Supplemen- notable absence of pathological lesions and no reported overall tary Figure 15). We identified a significant enrichment of PD neuronal loss.38 The apparent cellular specificity of PD has heritability in the lysosomal and LoF-intolerant gene set, while SCZ encouraged researchers to hypothesise that selective vulnerability heritability was only found enriched in LoF-intolerant genes, as is prompted by the action of risk loci in specific cell types; in other expected (Fig. 7a, Supplementary Table 6). We ran these analyses words, it is the nature of the cell type itself, which renders it with and without a category accounting for “all genes” and found vulnerable. However, given the interrelated nature of brain little difference between the estimates provided by stratified LDSC regions, apparently specific and reproducible patterns of abnorm- (Supplementary Figure 16), thus we report those results that do ality could also be the result of a more global effect that exposes not account for “all genes”. functional systems (e.g. neural networks) at different times along a Using the same gene sets together with EWCE, we also disease’s natural history, a view now put forward by several evaluated whether these PD-implicated gene sets were highly independent groups.39–41 That is, risk loci may not necessarily lie expressed in any of the Linnarsson cell-type classes and their cell in cellular subtypes or individual brain regions, but in global Published in partnership with the Parkinson’s Foundation npj Parkinson’s Disease (2019) 6 R.H. Reynolds et al. 6 Fig. 3 Enrichment of PD and SCZ common-SNP heritability in tissue-specific eQTL annotations. a Stratified LDSC analyses showed significant enrichment of SCZ heritability in brain-specific and blood-specific GTEx eQTLs. b A within-brain analysis of GTEx eQTLs showed no significant enrichment of PD and SCZ heritability in one region relative to others. In both analyses, eQTLs were assigned to a tissue/brain region based on their effect size (i.e. the absolute value of the linear regression slope). Tissues were ordered within each tissue category by the coefficient p- value obtained for SCZ. The black dashed lines indicate the cut-off for Bonferroni significance (a, p < 0.05/(2 × 2); b, p < 0.05/(2 × 11)). Bonferroni-significant results are marked with black borders. The proportion of SNPs accounted for by each annotation (compared to the baseline model), the regression coefficient calculated for the latest PD and SCZ GWASs, and the coefficient p-values for previous iterations of the PD and SCZ GWASs are displayed in Supplementary Figures 5–7. Numerical results are reported in Supplementary Table 2 cellular processes to which cellular subtypes have varying GWAS, specifically in the inferior temporal lobe, and they observed vulnerability. enrichment of fetal brain regulatory features for age at Addressing the question of cellular specificity in sporadic PD in menarche.47 a meaningful manner is now possible due to increasing GWAS Considering our inability to attribute PD heritability/suscept- sample sizes, increased availability of cell-type-specific gene ibility to a specific brain-related cell type, we also applied stratified expression data, and the recent development of robust meth- LDSC and EWCE to gene sets implicated in PD (autophagy, odologies. In this study, we used stratified LDSC and EWCE lysosomal and mitochondrial gene sets) and SCZ (LoF-intolerant together with several brain-related genomic annotations to genes), all of which can be considered global pathways/gene sets. connect common-variant genetic findings for PD to specific brain Here we show a significant enrichment of PD heritability in the cell types, with SCZ included for comparison purposes. We show lysosomal and LoF-intolerant gene set, with the former highly that PD heritability does not enrich in global brain annotations or expressed in astrocytic, microglial, and oligodendrocyte subtypes brain-related cell-type-specific annotations, as one might expect if and the latter highly expressed in almost all tested cellular cellular heterogeneity was masking the signal. In contrast, SCZ subtypes, providing support for the view that PD is a disorder of heritability significantly enriches in global and regional brain global pathways working across various cell types, as opposed to annotations and in select neuronal cell types, in line with previous specific cell types themselves driving disease risk. results.15,18 With these results in mind, it is tempting to speculate that PD One might argue that the lack of PD heritability enrichment in presents genetically as more of a systemic disorder, with a bias to any cell-type-specific categories could be due to PD having a brain pathology, as opposed to a primary brain disorder. In relatively low estimated total heritability; PD heritability estimates support of this view, PD-associated risk variants have been found range between 20 and 27%9,42,43 (by comparison, SCZ heritability associated with monocytes and the innate immune system,13,26,48 estimates range between 25 and 80%, depending on whether it is in addition to lymphocytes, mesendoderm, liver- and fat-cells.49 estimated from common SNPs44 or twin studies.45,46) However, we Recent work has also demonstrated a causal relationship between suggest that this is not a complete explanation as significant BMI and PD,50 which together with the re-purposing of exenatide enrichments have been observed in other GWASs with relatively (a glucagon-like peptide-1 receptor agonist currently licensed for low overall heritability estimates. For example, in the original the treatment of type 2 diabetes) for the potential treatment of stratified LDSC paper they observe enrichment of genomic PD,51 highlights the need to look beyond the brain and selective overlap of histone modifications for the CNS in the ever-smoked neuronal vulnerability. npj Parkinson’s Disease (2019) 6 Published in partnership with the Parkinson’s Foundation R.H. Reynolds et al. 7 Fig. 4 Enrichment of PD and SCZ common-SNP heritability in brain- related cell-type-specific gene expression annotations. Stratified LDSC analyses using cell-type-specific annotations derived from bulk RNA-sequencing of immunopanned cell types from human temporal lobe cortex (a) and single-cell RNA-sequencing of the adolescent mouse nervous system (b) demonstrated an enrichment of SCZ heritability in neuronal cell types (in particular, medium spiny neurons), but no cell-type enrichment for PD. All cell-type annotations were generated using the top 10% of enriched genes within a cell type compared to all others. Cell types were ordered alphabetically within each overarching cell type category. The black dashed lines indicate the cut-off for Bonferroni significance (a, p < 0.05/(2 × 6); b, p < 0.05/(2 × 30)). Bonferroni-significant results are marked with black borders. The proportion of SNPs accounted for by each annotation (compared to the baseline model), the regression coefficient calculated for the latest PD and SCZ GWASs, and the coefficient p-values for previous iterations of the PD and SCZ GWASs are displayed in Supplementary Figs 8–10. Numerical results and cell-type abbreviations are reported in Supplementary Table 3 It is entirely possible that with growing sample sizes, our understanding of the contribution of tissue-specific and brain- region-specific eQTL annotations to PD and SCZ heritability may change. Likewise, given that both TWAS27 and coloc28 are dependent upon the accuracy and integrity of eQTL datasets, it is likely that growing samples sizes will alter the list of PD susceptibility genes, potentially resulting in a different cellular expression profile. It is worth noting that there are several tools which evaluate associations between eQTLs and GWAS risk loci to identify susceptibility genes (including coloc, TWAS and others like eCAVIAR52). These tools differ in their underlying algorithms and assumptions (e.g. one versus multiple causal variants at a locus), thus motivating the use of multiple methods followed by integration across the results, as performed in the study by Li et al.26 used here. Second, our strategy for creating cell-type-specific profiles primarily involved gene expression data and assumed disease relevance only if disease heritability enriched for SNPs within genes with high specific expression. This approach together with the use of the GTEx eQTL dataset (which is likely to be enriched only for eQTLs with larger effect sizes due to power limitations) should capture regulatory SNPs in close proximity to genes of interest. However, as demonstrated in a recent study from Hormozdiari et al. how one chooses to construct an eQTL annotation is fraught with challenges53 and we recognise that our approach may have produced conservative enrichment estimates. Perhaps more importantly though, our strategy for creating cell-type-specific profiles does not account for the effect of regulatory SNPs that function at longer distances to impact upon gene expression. At present, our ability to address this issue is limited since detecting trans-acting eQTLs has proven to be There are several caveats to this study aside from the limitations challenging,54 especially in human brain. posed by the current PD GWAS, which does not capture the Third, our approach accounts only for cell type and pathway impact of rare variation or some forms of structural variation. and, moreover, builds on the assumption that cellular diversity can These caveats include the quality of our annotations, the be sufficiently described by discrete cells classes, which a recent strategies employed to generate them and, perhaps most single-cell RNA-sequencing study of the hippocampal CA1 area critically, the annotations we cannot account for. has called into question.55 In this study, it was suggested that First, the quality of our annotations is especially pertinent in the characterisation of cells requires continuous modes of variation in case of the gene sets used to reflect various PD-implicated addition to discrete cell classes; that is, some cell classes exist on a pathways. While lysosomal and mitochondrial gene sets were common genetic continuum. Inherent within this spectrum is derived from rigorously curated gene databases, with a focus on cellular state, which reflects the physiological condition of a given unbiased proteomic and localisation studies, the autophagy list cell, whether it be the degree of differentiation or activation in stemmed from Gene Ontology, which has not undergone the response to a stimulus. There may be cellular states that we have same meticulous curation. The noise introduced by potentially not assayed or captured which harbour PD heritability enrich- inaccurate annotation could affect our ability to detect heritability ments. Furthermore, one would expect preferential enrichment of enrichments. Furthermore, the quality of eQTL datasets, reflected pathways in specific cell types/subtypes to vary dependent on in their power to detect an eQTL, is dependent upon sample size. their physiological profile. In view of increasing evidence for the Published in partnership with the Parkinson’s Foundation npj Parkinson’s Disease (2019) 6 R.H. Reynolds et al. 8 but instead in global cellular processes, with effects in a range of cellular subtypes. This view has significant implications for disease modelling, with a choice of model perhaps based upon the cell type, which best reflects the process of interest, as opposed to the cell type which demonstrates the highest burden of α-synuclein aggregates. Likewise, viewing PD as a systemic disorder may have implications for potential drug re-purposing, as in the case of exenatide. Thus, our work here may have wider implications in terms of understanding neurodegenerative disorders more gen- erally as disorders of key cellular processes rather than disorders driven solely by specific cell types. METHODS Stratified LD score regression (LDSC) We applied stratified LDSC47 (see URLs, Supplementary information) to determine if various categories of genomic annotations (marking tissue- or cell-type-specific activity, as summarised in Annotation datasets) were enriched for heritability of various GWASs (see GWAS datasets below). LDSC exploits the expected relationships between true association signals and surrounding local linkage disequilibrium (LD) to correct out confounding biases, such as cryptic relatedness and population stratifica- tion, and arrive at unbiased estimates of genetic heritability within a given set of SNPs (here stratified according to whether they were located within genomic annotation regions). Following the procedure employed by Finucane et al.,47 we added annotation categories individually to the baseline model (version 1.1, see URLs, Supplementary information). We used HapMap Project Phase 3 (HapMap3)56 SNPs for the regression, and 1000 Genomes Project57 Phase 3 European population SNPs for the LD reference panel. We only partitioned the heritability of SNPs with minor allele frequency >5%, and we excluded the MHC region from analysis due to the complex and long-range LD patterns in this region. To map SNPs to genes, we used the SNPlocs.Hsapiens.dbSNP144.GRCh37 R package (dbSNP build 144 and GRCh37 coordinates).58 For all stratified LDSC analyses, we report a one-tailed p-value (coefficient p-value) based on the coefficient z-score outputted by stratified LDSC. A one-tailed test was used as we were only interested in annotation categories with a significantly positive regression coefficient (i.e. the annotation positively contributed to trait heritability, conditional upon the baseline model, which accounts for the underlying genetic architecture). We looked at three versions of PD GWAS summary statistics and four versions of SCZ, and for each set of analyses we corrected for multiple testing of the GWASs across the number of annotation categories, resulting in Bonferroni significance thresholds for each set of analyses. Annotation datasets Fig. 5 Enrichment of PD and SCZ common-SNP heritability in cell- Tissue-specific gene expression. Annotation files were generated by type modules inferred from human tissue-level co-expression Finucane et al.,18 using GTEx V6P gene expression,20 and obtained from networks. Stratified LDSC analyses using cell-type-specific co- Alkes Price’s group data repository (see URLs, Supplementary information). expression modules from frontal cortex (a), putamen (b), and Briefly, for each GTEx tissue, genes were ranked by a computed t-statistic substantia nigra (c) demonstrated significant enrichment of SCZ reflecting their specific expression within that tissue versus all other heritability in certain neuronal modules across all three tissues, but tissues, excluding those that were from a similar tissue category (e.g. no enrichment for PD heritability. Genes were assigned to cell-type expression in cortex samples was compared to expression in all other modules by module membership. Cell-type-specific modules were tissues except other brain regions; see Supplementary Table 2 from ordered alphabetically within each overarching cell type category. Finucane et al.18 for t-statistic tissue categories). The top 10% of expressed The black dashed lines indicate the cut-off for Bonferroni genes from each tissue was selected and a 100-kb window was added significance (a, p < 0.05/(2 × 5); b, p < 0.05/(2 × 15); c, p < 0.05/(2 × around their transcribed regions to obtain a tissue-specific gene expression 11)). Bonferroni-significant results are marked with black borders. annotation. As described by Finucane et al.,18 the two parameters The proportion of SNPs accounted for by each annotation (proportion of genes selected and window size around each gene) were (compared to the baseline model), the regression coefficient selected following testing of six different parameter settings, which calculated for the latest PD and SCZ GWASs, and the coefficient p- identified 10% and 100 kb as the settings producing the most significant P- values for previous iterations of the PD and SCZ GWASs are values for identifying enrichment in disease-relevant tissues. For the displayed in Supplementary Figures 11–14. Numerical results and within-brain analysis, tissues were restricted to the 13 brain regions found module descriptions are reported in Supplementary Table 4. FCTX, in GTEx, including: amygdala, anterior cingulate cortex (BA24), caudate, frontal cortex; PUTM, putamen; SNIG, substantia nigra cerebellar hemisphere, cerebellum, cortex, frontal cortex (BA9), hippocam- pus, hypothalamus, nucleus accumbens, putamen, spinal cord (cervical c-1), substantia nigra. association between PD and the innate immune system,13,26,48 we think that cellular state is likely to be an important factor, which Tissue-specific eQTLs. From the GTEx Portal (V7, accessed 04/16/18, see we cannot fully assess at this stage. URLs, Supplementary information), we downloaded all SNP-gene (expres- In conclusion, our results add to a growing body of evidence in sion quantitative trait loci, eQTL) association tests (including non- support of the view that PD risk loci may not lie entirely in those significant tests) for blood (to allow for a blood-brain comparison) and cell types that display the disease’s characteristic neuropathology, 11 of the 13 available brain regions.20 To reduce redundancy across the npj Parkinson’s Disease (2019) 6 Published in partnership with the Parkinson’s Foundation R.H. Reynolds et al. 9 Fig. 6 PD susceptibility genes do not enrich in brain-related cell types. a PD susceptibility genes were derived from MAGMA analyses and a study attempting to prioritise genes in PD using TWAS and colocalisation analyses.26 Genes overlapping between the two sets were removed, resulting in a list of 89 genes. Bootstrapping tests performed using the EWCE method revealed no enrichment of PD susceptibility genes in the major cell-type classes (b) or their cell subtypes (c) from the Linnarsson single-cell RNA-sequencing dataset. Gene lists and numerical results are available in Supplementary Table 5 brain regions, we excluded cortex and cerebellum, and instead included generated using bulk RNA-sequencing and single-cell RNA-sequencing, frontal cortex, anterior cingulate cortex and cerebellar hemisphere. We respectively. Due to the disparate nature of the RNA-sequencing methods, performed an FDR correction for each tissue and included all SNP-gene each dataset was analysed separately. Common to both analyses was the associations that passed FDR <5% in our downstream analyses. For the calculation of an enrichment value for each gene in each cell type. blood-brain comparison, eQTLs from all 11 brain regions were combined to Enrichment was calculated as: gene expression in one cell type divided by form one brain category. eQTLs that replicated across brain regions were the average gene expression across all other cell types. We thereafter collapsed into one entry and allocated an effect size (i.e. the absolute value selected the top 10% of genes enriched within each cell type and added a of the linear regression slope) equal to that of the maximum effect size 100 kb window to reflect the approach used by Finucane et al.18 When observed across the brain regions. Finally, eQTLs were assigned to either using the Barres data, we averaged gene expression across samples of the blood or brain by their effect size. A similar approach was used for the same cell type, filtered genes on the basis of an FPKM ≥1 in at least one within-brain analysis, where eQTLs were assigned to one of the 11 brain cell type (this equates to ~66% of all genes with FPKM >0.1, which was set regions based on effect size. by Zhang et al.21 as the threshold for minimum gene expression), and then calculated gene enrichment. Our detection threshold of FPKM ≥1 was Cell-type-specific gene expression. Cell-type-specific annotations were employed on the basis that smaller thresholds tend to produce large and constructed using gene expression data from the Barres group21 and the misleading enrichments.59 The Linnarsson data was available with gene Linnarsson group22 (see URLs, Supplementary information), which was expression aggregated by sub-cell type/cluster. Genes were filtered on the Published in partnership with the Parkinson’s Foundation npj Parkinson’s Disease (2019) 6 R.H. Reynolds et al. 10 Fig. 7 PD heritability enriches in lysosomal and LoF-intolerant gene sets which are ubiquitously expressed. a Stratified LDSC analyses using gene sets implicated in PD demonstrated a significant enrichment of PD heritability in the lysosomal and LoF-intolerant gene sets. The black dashed lines indicate the cut-off for Bonferroni significance (p < 0.05/(2 × 4)). Bonferroni-significant results are marked with black borders. The proportion of SNPs accounted for by each annotation (compared to the baseline model), the regression coefficient calculated for the latest PD and SCZ GWASs, and the coefficient p-values for previous iterations of the PD and SCZ GWASs are displayed in Supplementary Figures 17 and 18. Bootstrapping tests performed using the EWCE method demonstrated enrichment of autophagy, lysosomal and mitochondrial gene sets in specific cell-type classes (b) and their cell subtypes (c) from the Linnarsson single-cell RNA-sequencing dataset. Asterisks denote significance at p < 0.05 after correcting for multiple testing with the Benjamini-Hochberg method over all gene sets and cell types tested. Gene lists and numerical results are reported in Supplementary Table 6 basis of expression >0, enrichment was calculated and a subset of the 265 et al.23 Modules were assigned to cell types using the userListEnrichment R identified clusters were used as annotations. Mouse genes were converted function implemented in the WGCNA R package, which measures to human orthologs using Biomart (of the 23,368 genes with expression >0 enrichment between module-assigned genes and defined brain-related in at least one cluster, 16,025 genes were converted to human orthologs. A lists19,62–65 using a hypergeometric test. Genes assigned to modules list of those genes that did not convert can be found in Supplementary significantly enriched for brain-related cell-type markers of predominantly Table 3). one cell type, with a module membership of ≥0.5, were allocated a cell- type “label” of neuron, microglia, astrocyte or oligodendrocyte and Cell-type-specific co-expression modules. Co-expression networks for fron- considered cell-type specific. Module membership values range between tal cortex, putamen and substantia nigra were constructed using GTEx V6 0 and 1, with 1 indicating that a gene’s expression is highly correlated with gene expression,20 the WGCNA R package60 and post-processing with the module eigengene. An eigengene is defined as the first principal k-means61 (see URLs, Supplementary information), as described by Botia component of a given module and can be considered representative of the npj Parkinson’s Disease (2019) 6 Published in partnership with the Parkinson’s Foundation R.H. Reynolds et al. 11 Table 1. Summary of GWAS datasets Disease First author, Year N cases N controls PMID Reference 71 PD IPDGC Consortium, 2011 5,333 12,019 21738488 14 PD Nalls, 2014 13,708 95,282 25064009 a 16 PD Nalls, 2018 (excluding 23andMe contributions) 33,674 (18,618 proxy cases from UK Biobank) 449,037 72 SCZ SCZ Consortium, 2011 9,394 12,462 21926974 73 SCZ Ripke, 2013 14,395 18,705 23974872 74 SCZ SCZ Consortium, 2014 (EUR subset) 33,640 43,456 25056061 17 SCZ Pardiñas, 2018 40,675 64,643 29483656 a Access to PD 2018 summary statistics (excluding 23andMe contributions) was provided by Mike A Nalls, with permissions from IPDGC and SGPD gene expression profiles within the module, as it summarises the largest transcription start site to transcription stop site. In addition, we added a amount of variance in expression. As with tissue-specific and cell-type- window of 35 kb upstream and 10 kb downstream of each gene. This specific gene expression annotations, we added a 100 kb window around choice was based on (1) most transcriptional regulatory elements fall each gene. To ensure we were using consistent and reproducible modules, within this interval,75 (2) a fraction of GWAS risk loci lie outside gene we evaluated preservation of GTEx modules with modules derived from boundaries and may regulate gene expression76 and (3) previous work in co-expression networks constructed for frontal cortex, putamen and the field of pathway analysis.17,77,78 Furthermore, the MHC region on substantia nigra using gene expression data from the United Kingdom chromosome 6 (chr6: 25500000–33500000, human genome assembly Brain Expression Consortium (UKBEC).61,66 Preservation values were GRCh37) was excluded. The gene p-value was computed based on the calculated using the modulePreservation function from WGCNA. Preserva- mean association statistic of SNPs within a gene, with genome-wide tion is a measure of how well the connectivity and correlation structures within the genes of a network´s module are preserved in another tissue significance set to p < 2.82 × 10−6, and LD was estimated from the expression dataset. The final preservation score we report is Z-summary- European subset of 1000 Genomes Phase 357. press, which is an aggregation of a number of measures (see Langfelder et al.67 for a detailed explanation). Values under 2 denote no preservation whereas values greater than 2 suggest preservation, and values over Evaluating enrichment of PD-associated genes and gene sets 10 suggest strong preservation of the module in the tissue tested. We Expression-weighted cell-type enrichment (EWCE, see URLs, Supplemen- included only GTEx modules with a preservation value >2. Preservation tary information)79 was used to determine whether PD-associated genes or values and module descriptions are reported in Supplementary Table 4; gene sets have higher expression within a particular cell type than the latter can also be viewed at https://snca.atica.um.es/coexp/Run/ expected by chance. As our input, we used the same subset of clusters Catalog/. from the Linnarsson single-cell RNA-sequencing dataset used in stratified LDSC, in addition to a target gene list. For each gene in the Linnarsson Gene sets. We investigated three gene sets with previous biological dataset, we estimated its cell-type specificity i.e. the proportion of total support for involvement in PD: autophagy, lysosomal and mitochon- expression of a gene found in one cell type compared to all cell types, drial.29–34 The autophagy gene set included all genes associated with the using the ‘generate.celltype.data’ function of the EWCE package. EWCE Gene Ontology terms: GO:0006914 (“autophagy”) and GO:0005776 (“autophagosome”), as derived from the GO C5 collection of the MSigDB with the target list was run with 100,000 bootstrap lists. We controlled for database (v5.2). Lysosomal genes were downloaded from the Human transcript length and GC-content biases by selecting bootstrap lists with Lysosome Gene Database (hLGDB, see URLs, Supplementary informa- comparable properties to the target list. P-values were corrected for tion).68 All genes reported lysosomal by any of the listed sources (9 of the multiple testing using the Benjamini-Hochberg method overall cell types 16 were unbiased proteomic studies) were used. Mitochondrial genes were and gene lists tested. We performed the analysis with major cell-type obtained from Human MitoCarta 2.0, an inventory of human genes with classes (e.g. “astrocyte”, “microglia”, “enteric neurons”, etc.) and subtypes of the strong support of mitochondrial localisation based on literature these classes (e.g. ACNT1 [“Non-telencephalon astrocytes, protoplasmic”], curation, proteomic analyses and epitope tagging/microscopy (see URLs, ACNT2 [“Non-telencephalon astrocytes, fibrous”], etc.). Data are displayed Supplementary information).69 In addition, we used a gene set comprising as standard deviations from the mean, and any values < 0, which reflect a loss-of-function (LoF)-intolerant genes, as defined by the Exome Aggrega- depletion of expression, are displayed as 0. tion Consortium (ExAC)35 using their gene-level constraint metric (pLI ≥0.9), and constructed an “all genes” gene set. The “all genes” gene set was constructed by extracting all genes from BioMart (using the homo sapiens PD susceptibility genes. PD susceptibility genes were derived from our GRCh37 library, as with all other annotations). For stratified LDSC analyses, own MAGMA analyses and a study attempting to prioritise genes in PD an additional window of 100 kb was added around genes. The genes using TWAS and colocalisation analyses (Supplementary Table 1 in ref. 26). comprising these lists are available in Supplementary Table 6. Overlap The genes comprising these lists are available in Supplementary Table 5. In between gene sets was determined using Intervene, a command line tool the case of MAGMA, only those genes passing genome-wide significance and web application that computes and visualises intersections of gene (p < 2.82 × 10-6) were used. In the case of TWAS/coloc, only those eQTL- sets (see URLs, Supplementary information).70 gene associations found within dorsolateral prefrontal cortex tissue, which were both TWAS and coloc hits (as defined in ref. 26) were used. GWAS datasets Table 114,16,17,71–74 Reporting Summary Further information on experimental design is available in the Nature MAGMA: assessing gene-level enrichment Research Reporting Summary linked to this article. Gene-level p-values were calculated with the Nalls et al.16 PD GWAS (excluding 23andMe contributions) using MAGMA v1.06 (see URLs, Supplementary information),25 which tests the joint association of all SNPs in a gene with the phenotype while accounting for LD between SNPs. SNPs DATA AVAILABILITY were mapped to genes using NCBI definitions (GRCh37, annotation release Datasets analysed in this study are all derived from publicly available resources (see 105); only genes in which at least one SNP mapped were included in URLs, Supplementary information). All data generated during this study are included downstream analyses. Gene boundaries were defined as the region from in this published article’s supplementary information. Published in partnership with the Parkinson’s Foundation npj Parkinson’s Disease (2019) 6 R.H. Reynolds et al. 12 CODE AVAILABILITY Hospital, Departments of Neurology and Neurosurgery, Department of Human Open source software is available on GitHub for all tools used (see URLs, Genetics, McGill University, Montréal, QC H3A 0G4, Canada; 33Department of Human Supplementary information). Genetics, McGill University, Montréal, QC H3A 0G4, Canada; 34Department of Neurology, Leiden University Medical Center, Leiden, Netherlands; 35Instituto de Biomedicina de Sevilla (IBiS), Hospital Universitario Virgen del Rocío/CSIC/Universidad INTERNATIONAL PARKINSON’S DISEASE GENOMICS de Sevilla, Seville, Spain; 36Fundació Docència i Recerca Mútua de Terrassa and Movement Disorders Unit, Department of Neurology, University Hospital Mutua de CONSORTIUM (IPDGC) Terrassa, Terrassa, Barcelona, Spain; 37Hospital Universitario Central de Asturias, Alastair J. Noyce1,7, Aude Nicolas3, Mark R. Cookson3, Sara Bandres-Ciga3, J. Raphael Oviedo, Spain; 38Hospital Universitario Parque Tecnologico de la Salud, Granada, Gibbs3, Dena G. Hernandez3, Andrew B. Singleton3, Xylena Reed3, Hampton Leonard3, Spain; 39Instituto de Investigación Sanitaria Biodonostia, San Sebastián, Spain; Cornelis Blauwendraat3,8, Faraz Faghri3,9, Jose Bras5,10, Rita Guerreiro5,10, Arianna 40 Hospital General de Segovia, Segovia, Spain; 41Memory Unit, Department of Tucci10, Demis A. Kia10, Henry Houlden10, Helene Plun-Favreau10, Kin Y. Mok10, Neurology, IIB Sant Pau, Hospital de la Santa Creu i Sant Pau, Universitat Autònoma Nicholas W. Wood10, Ruth Lovering10, Lea R’Bibo10, Mie Rizig10, Viorica Chelban10, de Barcelona, Barcelona, Spain; 42Centro de Investigación Biomédica en Red en Daniah Trabzuni10,11, Manuela Tan12, Huw R. Morris12, Ben Middlehurst13, John Enfermedades Neurodegenerativas (CIBERNED), Madrid, Spain; 43Hospital Universi- Quinn13, Kimberley Billingsley13, Peter Holmans14, Kerri J. Kinghorn15, Patrick Lewis16, tario Marqués de Valdecilla-IDIVAL, Santander, Spain; 44University of Cantabria, Valentina Escott-Price17, Nigel Williams17, Thomas Foltynie18, Alexis Brice19, Fabrice Santander, Spain; 45Movement Disorders Unit, Department of Neurology, IIB Sant Pau, Danjou19, Suzanne Lesage19, Jean-Christophe Corvol19, Maria Martinez19,20, Anamika Hospital de la Santa Creu i Sant Pau, Universitat Autònoma de Barcelona, Barcelona, Giri21,22, Claudia Schulte21,22, Kathrin Brockmann21,22, Javier Simón-Sánchez21,22, Peter Spain; 46Centro de Investigacion Biomedica, Universidad de Granada, Granada, Spain; Heutink21,22, Thomas Gasser21,22, Patrizia Rizzu22, Manu Sharma23, Joshua M. 47 Hospital Universitario Virgen de las Nieves, Instituto de Investigación Biosanitaria de Shulman24,25, Laurie Robak24, Steven Lubbe26, Niccolo E. Mencacci27, Steven Granada, Granada, Spain; 48Hospital Clinic de Barcelona, Barcelona, Spain; 49Instituto Finkbeiner28,29, Codrin Lungu30, Sonja W. Scholz31, Ziv Gan-Or32, Guy A. Rouleau32, de Investigación Sanitaria Fundación Jiménez Díaz, Madrid, Spain; 50Hospital Lynne Krohan33, Jacobus J. van Hilten34, Johan Marinus34, Astrid D. Adarmes- Universitario Virgen de la Victoria, Malaga, Spain; 51Institut de Recerca Sant Joan Gómez35, Inmaculada Bernal-Bernal35, Marta Bonilla-Toribio35, Dolores Buiza-Rueda35, de Déu, Barcelona, Spain; 52Hospital Universitario Ramón y Cajal, Madrid, Spain; Fátima Carrillo35, Mario Carrión-Claro35, Pablo Mir35, Pilar Gómez-Garre35, Silvia 53 Department of Neurology, Instituto de Investigación Sanitaria La Fe, Hospital Jesús35, Miguel A. Labrador-Espinosa35, Daniel Macias35, Laura Vargas-González35, Universitario y Politécnico La Fe, Valencia, Spain; 54Hospital General de Segovia, Carlota Méndez-del-Barrio35, Teresa Periñán-Tocino35, Cristina Tejera-Parrado35, Segovia, Spain; 55Department of Neurology, Hospital Universitario Fundación Monica Diez-Fairen36, Miquel Aguilar36, Ignacio Alvarez36, María Teresa Boungiorno36, Alcorcón, Madrid, Spain; 56Department of Neurology, Medical University of Vienna, Maria Carcel36, Pau Pastor36, Juan Pablo Tartari36, Victoria Alvarez37, Manuel Vienna, Austria; 57Department of Neurology, Oslo University Hospital, Oslo, Norway; Menéndez González37, Marta Blazquez37, Ciara Garcia37, Esther Suarez-Sanmartin37, 58 Department of Pathophysiology, University of Tartu, Tartu, Estonia; 59Department of Francisco Javier Barrero38, Elisabet Mondragon Rezola39, Jesús Alberto Bergareche Reproductive Biology, Estonian University of Life Sciences, Tartu, Estonia; 60Perron Yarza39, Ana Gorostidi Pagola39, Adolfo López de Munain Arregui39, Javier Ruiz- Institute for Neurological and Translational Science, Perth, WA, Australia; 61Depart- Martínez39, Debora Cerdan40, Jacinto Duarte40, Jordi Clarimón41,42, Oriol Dols- ments of Neurology and Neurosurgery, University of Tartu, Tartu, Estonia; 62Institute Icardo41,42, Jon Infante42,43,44, Juan Marín42,45, Jaime Kulisevsky42,45, Javier Pagona- of Clinical Medicine, Department of Neurology, University of Oulu, Oulu, Finland; barraga42,45, Isabel Gonzalez-Aramburu43, Antonio Sanchez Rodriguez43, María 63 Department of Neurology and Medical Research Center, Oulu University Hospital, Sierra43, Raquel Duran46, Clara Ruz46, Francisco Vives46, Francisco Escamilla-Sevilla47, Oulu, Finland and 64University of Lagos, Lagos, Lagos State, Nigeria Adolfo Mínguez47, Ana Cámara48, Yaroslau Compta48, Mario Ezquerra48, Maria Jose Marti48, Manel Fernández48, Esteban Muñoz48, Rubén Fernández-Santiago48, Eduard Tolosa48, Francesc Valldeoriola48, Pedro García-Ruiz49, Maria Jose Gomez Heredia50, SYSTEM GENOMICS OF PARKINSON’S DISEASE (SGPD) Francisco Perez Errazquin50, Janet Hoenicka51, Adriano Jimenez-Escrig52, Juan Carlos Toni Pitcher65,66, Tim Anderson65,66, Steven Bentley67, Javed Fowdar67, George Martínez-Castrillo52, Jose Luis Lopez-Sendon52, Irene Martínez Torres53, Cesar Mellick67, John Dalrymple-Alford68, Anjali K. Henders69, Irfahan Kassam69, Grant Tabernero54, Lydia Vela55, Alexander Zimprich56, Lasse Pihlstrom57, Sulev Koks58,59,60, Montgomery69, Julia Sidorenko69, Futao Zhang69, Angli Xue69, Costanza L. Vallerga69, Pille Taba61, Kari Majamaa62,63, Ari Siitonen62,63, Njideka U. Okubadejo64, Oluwada- Leanne Wallace69, Naomi R. Wray69,70, Jian Yang69,70, Peter M. Visscher69,70, Jacob milola O. Ojo64 Gratten69,70, Peter A. Silburn70, Glenda Halliday71, Ian Hickie71, John Kwok71, Simon 7 Lewis71, Martin Kennedy72, John Pearson72 Preventive Neurology Unit, Wolfson Institute of Preventive Medicine, QMUL, London, UK; 8National Institute of Neurological Disorders and Stroke, Bethesda, MD, USA; 9 Department of Computer Science, University of Illinois at Urbana-Champaign, 65 Urbana, IL, USA; 10Department of Molecular Neuroscience, UCL, London, UK; New Zealand Brain Research Institute, Christchurch, New Zealand; 66Department of 11 Department of Genetics, King Faisal Specialist Hospital and Research Centre, Medicine, University of Otago, Christchurch, New Zealand; 67Griffith Institute for Drug Riyadh 11211, Saudi Arabia; 12Department of Clinical Neuroscience, University Discovery, Griffith University, Brisbane, Australia; 68Dept. Psychology, University of College London, London, UK; 13Institute of Translational Medicine, University of Canterbury, New Zealand Brain Research Institute, Christchurch, New Zealand; Liverpool, Liverpool, UK; 14Biostatistics & Bioinformatics Unit, Institute of Psycholo- 69 Institute for Molecular Bioscience, University of Queensland, Brisbane, Australia; 70 gical Medicine and Clinical Neuroscience, MRC Centre for Neuropsychiatric Genetics Queensland Brain Institute, University of Queensland, Brisbane, Australia; 71Brain & Genomics, Cardiff, UK; 15Institute of Healthy Ageing, University College London, and Mind Centre, Sydney Medical School, University of Sydney, Sydney, Australia and London, UK; 16University of Reading, Reading, UK; 17MRC Centre for Neuropsychiatric 72 Department of Pathology, University of Otago, Christchurch, New Zealand Genetics and Genomics, Cardiff University School of Medicine, Cardiff, UK; 18UCL Institute of Neurology, London, UK; 19Institut 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; 20Paul Sabatier ACKNOWLEDGEMENTS University, Toulouse, France; 21Department for Neurodegenerative Diseases, Hertie R.H.R. was supported through the award of a Leonard Wolfson Doctoral Training Institute for Clinical Brain Research, University of Tübingen, Tübingen, Germany; Fellowship in Neurodegeneration. M.A.N. was supported by a consulting contract 22 DZNE, German Center for Neurodegenerative Diseases, Tübingen, Germany; between Data Tecnica International and the National Institute on Aging, NIH, 23 Centre for Genetic Epidemiology, Institute for Clinical Epidemiology and Applied Bethesda, MD, USA. J.H. and M.R. were supported by the UK Medical Research Council Biometry, University of Tubingen, Tubingen, Germany; 24Departments of Neurology, (MRC), with J.H. supported by a grant (MR/N026004/) and M.R. through the award of a Neuroscience, and Molecular & Human Genetics, Baylor College of Medicine, Tenure-track Clinician Scientist Fellowship (MR/N008324/1). J.H. was also supported Houston, Texas, USA; 25Jan and Dan Duncan Neurological Research Institute, Texas by the UK Dementia Research Institute. Full consortia acknowledgements are Children’s Hospital, Houston, Texas, USA; 26Ken and Ruth Davee Department of available in the supplemental materials (Text S2). Neurology, Northwestern University Feinberg School of Medicine, Chicago, IL, USA; 27 Northwestern University Feinberg School of Medicine, Chicago, IL, USA; 28Depart- ments of Neurology and Physiology, University of California, San Francisco, CA, USA; AUTHOR CONTRIBUTIONS 29 Gladstone Institute of Neurological Disease, Taube/Koret Center for Neurodegen- R.H.R., J.H., S.A.G.T. and M.R. conceived and designed the study. R.H.R. analysed data, erative Disease Research, San Francisco, CA, USA; 30National Institutes of Health drafted the figures, and together with S.A.G.T. and M.R. wrote the first draft of the Division of Clinical Research, NINDS, National Institutes of Health, Bethesda, MD, USA; manuscript. J.B. constructed the co-expression networks. M.A.N., IPDGC and SGPD 31 Neurodegenerative Diseases Research Unit, National Institute of Neurological helped with the use of Parkinson’s disease GWAS data. All authors contributed to the Disorders and Stroke, Bethesda, MD, USA; 32Montreal Neurological Institute and critical analysis of the manuscript. npj Parkinson’s Disease (2019) 6 Published in partnership with the Parkinson’s Foundation R.H. Reynolds et al. 13 ADDITIONAL INFORMATION 28. Giambartolomei, C. et al. Bayesian Test for Colocalisation between Pairs of Genetic Supplementary information accompanies the paper on the npj Parkinson’s Disease Association Studies Using Summary Statistics. PLoS Genet. 10, e1004383 (2014). website (https://doi.org/10.1038/s41531-019-0076-6). 29. Hernandez, D. G., Reed, X. & Singleton, A. B. Genetics in Parkinson disease: Mendelian versus non-Mendelian inheritance. J. Neurochem. 139, 59–74 (2016). Competing Interests: As a possible conflict of interest M.A.N. also consults for 30. Manzoni, C. The LRRK2-macroautophagy axis and its relevance to Parkinson’s Illumina Inc, Lysosomal Therapeutics Inc, the Michael J. Fox Foundation and Vivid disease. Biochem. Soc. Trans. 45, 155–162 (2017). Genomics among others. The remaining authors declare no competing interests. 31. Denny, P. et al. Exploring autophagy with Gene Ontology. Autophagy 14, 419–436 (2018). 32. Robak, L. A. et al. Excessive burden of lysosomal storage disorder gene variants in Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims Parkinson’s disease. Brain 140, 3191–3203 (2017). in published maps and institutional affiliations. 33. Haelterman, N. A. et al. A mitocentric view of Parkinson’s disease. Annu. Rev. Neurosci. 37, 137–159 (2014). 34. Ryan, B. J., Hoek, S., Fon, E. A. & Wade-Martins, R. Mitochondrial dysfunction and REFERENCES mitophagy in Parkinson’s: From familial to sporadic disease. Trends Biochem. Sci. 1. Poewe, W. et al. Parkinson disease. Nat. Rev. Dis. Prim. 3, 17013 (2017). 40, 200–210 (2015). 2. Del Tredici, K. & Braak, H. Review: Sporadic Parkinson’s disease: development and 35. Lek, M. et al. Analysis of protein-coding genetic variation in 60,706 humans. distribution of α-synuclein pathology. Neuropathol. Appl. Neurobiol. 42, 33–50 Nature 536, 285–291 (2016). (2016). 36. Kahn, R. S. et al. Schizophr. Nat. Rev. Dis. Prim. 1, 15067 (2015). 3. Bendor, J. T., Logan, T. P. & Edwards, R. H. The function of α-synuclein. Neuron 79, 37. Birnbaum, R. & Weinberger, D. R. Genetic insights into the neurodevelopmental 1044–1066 (2013). origins of schizophrenia. Nat. Rev. Neurosci. 18, 727–740 (2017). 4. Spillantini, M. G. et al. Alpha-synuclein in Lewy bodies. Nature 388, 839–840 38. Harrison, P. J. Postmortem studies in schizophrenia. Dialog-. Clin. Neurosci. 2, (1997). 349–357 (2000). 5. Polymeropoulos, M. H. et al. Mutation in the alpha-synuclein gene identified in 39. Ilieva, H., Polymenidou, M. & Cleveland, D. W. Non-cell autonomous toxicity in families with Parkinson’s disease. Science 276, 2045–2047 (1997). neurodegenerative disorders: ALS and beyond. J. Cell Biol. 187, 761–772 (2009). 6. Krüger, R. et al. Ala30Pro mutation in the gene encoding alpha-synuclein in 40. Eisen, A. & Turner, M. R.Does variation in neurodegenerative disease susceptibility Parkinson’s disease. Nat. Genet. 18, 106–108 (1998). and phenotype reflect cerebral differences at the network level. Amyotroph. 7. Singleton, A. B. et al. alpha-Synuclein locus triplication causes Parkinson’s disease. Lateral Scler. Front. Degener. 14, 487–493 (2013). Science 302, 841 (2003). 41. Warren, J. D. et al. Molecular nexopathies: A new paradigm of neurodegenerative 8. Zarranz, J. J. et al. The new mutation, E46K, of alpha-synuclein causes Parkinson disease. Trends Neurosci. 36, 561–569 (2013). and Lewy body dementia. Ann. Neurol. 55, 164–173 (2004). 42. Do, C. B. et al. Web-based genome-wide association study identifies two novel 9. Chang, D. et al. A meta-analysis of genome-wide association studies identifies 17 loci and a substantial genetic component for parkinson’s disease. PLoS Genet. 7, new Parkinson’s disease risk loci. Nat. Genet. (2017). https://doi.org/10.1038/ e1002141 (2011). ng.3955 43. Keller, M. F. et al. Using genome-wide complex trait analysis to quantify ‘missing 10. Brück, D., Wenning, G. K., Stefanova, N. & Fellner, L. Glia and alpha-synuclein in heritability’ in Parkinson’s disease. Hum. Mol. Genet. 21, 4996–5009 (2012). neurodegeneration: A complex interaction. Neurobiol. Dis. 85, 262–274 (2016). 44. Anttila, V. et al. Analysis of shared heritability in common disorders of the brain. 11. Booth, H. D. E., Hirst, W. D. & Wade-Martins, R. The Role of Astrocyte Dysfunction Sci. (80-.). 360, eaap8757 (2018). in Parkinson’s Disease Pathogenesis. Trends Neurosci. 40, 358–370 (2017). 45. Sullivan, P. F., Kendler, K. S. & Neale, M. C. Schizophrenia as a complex trait: 12. Yun, S. P. et al. Block of A1 astrocyte conversion by microglia is neuroprotective in evidence from a meta-analysis of twin studies. Arch. Gen. Psychiatry 60, models of Parkinson’s disease. Nat. Med. https://doi.org/10.1038/s41591-018- 1187–1192 (2003). 0051-5 (2018). 46. Hilker, R. et al. Heritability of schizophrenia and schizophrenia spectrum based on 13. Gagliano, S. A. et al. Genomics implicates adaptive and innate immunity in Alz- the nationwide Danish twin register. Biol. Psychiatry 83, 492–498 (2018). heimer’s and Parkinson’s diseases. Ann. Clin. Transl. Neurol. 3, 924–933 (2016). 47. Finucane, H. K. et al. Partitioning heritability by functional annotation using 14. Nalls, M. A. et al. Large-scale meta-analysis of genome-wide association data genome-wide association summary statistics. Nat. Genet. 47, 1228–1235 (2015). identifies six new risk loci for Parkinson’s disease. Nat. Genet. 46, 989–993 (2014). 48. Raj, T. et al. Polarization of the effects of autoimmune and neurodegenerative risk 15. Skene, N. G. et al. Genetic identification of brain cell types underlying schizo- alleles in leukocytes. Science 344, 519–523 (2014). phrenia. Nat. Genet. 50, 825–833 (2018). 49. Coetzee, S. G. et al. Enrichment of risk SNPs in regulatory regions implicate 16. Nalls, M. A. et al. Expanding Parkinson’s disease genetics: novel risk loci, genomic diverse tissues in Parkinson’s disease etiology. Sci. Rep. 6, 1–11 (2016). context, causal insights and heritable risk. bioRxiv (2019). https://doi.org/10.1101/ 50. Noyce, A. J. et al. Estimating the causal influence of body mass index on risk of 388165 Parkinson disease: A Mendelian randomisation study. PLoS Med. 14, e1002314 17. Pardiñas, A. F. et al. Common schizophrenia alleles are enriched in mutation- (2017). intolerant genes and in regions under strong background selection. Nat. Genet. 51. Athauda, D. et al. Exenatide once weekly versus placebo in Parkinson’s disease: a 50, 381–389 (2018). randomised, double-blind, placebo-controlled trial. Lancet (Lond., Engl.) 390, 18. Finucane, H. K. et al. Heritability enrichment of specifically expressed genes 1664–1675 (2017). identifies disease-relevant tissues and cell types. Nat. Genet. 50, 621–629 (2018). 52. Hormozdiari, F. et al. Colocalization of GWAS and eQTL Signals Detects Target 19. Oldham, M. C. et al. Functional organization of the transcriptome in human brain. Genes. Am. J. Hum. Genet. 99, 1245–1260 (2016). Nat. Neurosci. 11, 1271–1282 (2008). 53. Hormozdiari, F. et al. Leveraging molecular quantitative trait loci to understand 20. GTEx Consortium et al. Human genomics. The Genotype-Tissue Expression (GTEx) the genetic architecture of diseases and complex traits. Nat. Genet. (2018). pilot analysis: multitissue gene regulation in humans. Science 348, 648–660 https://doi.org/10.1038/s41588-018-0148-2 (2015). 54. Brynedal, B. et al. Large-scale trans-eQTLs affect hundreds of transcripts and 21. Zhang, Y. et al. Purification and Characterization of Progenitor and Mature mediate patterns of transcriptional co-regulation. Am. J. Hum. Genet. 100, Human Astrocytes Reveals Transcriptional and Functional Differences with 581–591 (2017). Mouse. Neuron 89, 37–53 (2016). 55. Harris, K. D. et al. Classes and continua of hippocampal CA1 inhibitory neurons 22. Zeisel, A. et al. Molecular architecture of the mouse nervous system. Cell 174, revealed by single-cell transcriptomics. PLoS Biol. 16, e2006387 (2018). 999–1014.e22 (2018). 56. International HapMap 3 Consortium et al. Integrating common and rare genetic 23. Botia, J. A. et al. G2P: Using machine learning to understand and predict genes variation in diverse human populations. Nature 467, 52–58 (2010). causing rare neurological disorders. bioRxiv (2018). https://doi.org/10.1101/288845 57. 1000 Genomes Project Consortium et al. An integrated map of genetic variation 24. Urs, N. M., Peterson, S. M. & Caron, M. G. New Concepts in Dopamine D2Receptor from 1,092 human genomes. Nature 491, 56–65 (2012). Biased Signaling and Implications for Schizophrenia Therapy. Biol. Psychiatry 81, 58. Pagès, H. SNPlocs.Hsapiens.dbSNP144.GRCh37: SNP locations for Homo sapiens 78–85 (2017). (dbSNP Build 144). R Packag. version 0.99.20 (2017). 25. de Leeuw, C. A., Mooij, J. M., Heskes, T. & Posthuma, D. MAGMA: Generalized 59. Sheng, Q. et al. Multi-perspective quality control of Illumina RNA sequencing data Gene-Set Analysis of GWAS Data. PLoS Comput. Biol. 11, 1–19 (2015). analysis. Brief. Funct. Genom. 16, 194–204 (2017). 26. Li, Y. I., Wong, G., Humphrey, J. & Raj, T. Prioritizing Parkinson’s Disease genes 60. Langfelder, P. & Horvath, S. WGCNA: An R package for weighted correlation using population-scale transcriptomic data. Nat. Commun. 10, 994 (2019). network analysis. BMC Bioinformatics 9, 559 (2008). 27. Gusev, A. et al. Integrative approaches for large-scale transcriptome-wide asso- 61. Botía, J. A. et al. An additional k-means clustering step improves the biological ciation studies. Nat. Genet. 48, 245–252 (2016). features of {WGCNA} gene co-expression networks. BMC Syst. Biol. 11, 47 (2017). Published in partnership with the Parkinson’s Foundation npj Parkinson’s Disease (2019) 6 R.H. Reynolds et al. 14 62. Lein, E. S. et al. Genome-wide atlas of gene expression in the adult mouse brain. 74. Schizophrenia Working Group of the Psychiatric Genomics Consortium. Biological Nature 445, 168–176 (2007). insights from 108 schizophrenia-associated genetic loci. Nature 511, 421–427 63. Cahoy, J. D. et al. A Transcriptome database for astrocytes, neurons, and oligo- (2014). dendrocytes: a new resource for understanding brain development and function. 75. Maston, G. A., Evans, S. K. & Green, M. R. Transcriptional Regulatory Elements in J. Neurosci. 28, 264–278 (2008). the Human Genome. Annu. Rev. Genom. Hum. Genet. 7, 29–59 (2006). 64. Winden, K. D. et al. The organization of the transcriptional network in specific 76. Nicolae, D. L. et al. Trait-associated SNPs are more likely to be eQTLs: Annotation neuronal classes. Mol. Syst. Biol. 5, 1–18 (2009). to enhance discovery from GWAS. PLoS Genet. 6, e1000888 (2010). 65. Miller, J. A., Horvath, S. & Geschwind, D. H. Divergence of human and mouse brain 77. Network and Pathway Analysis Subgroup of Psychiatric Genomics Consortium. transcriptome highlights Alzheimer disease pathways. Proc. Natl. Acad. Sci. 107, Psychiatric genome-wide association study analyses implicate neuronal, immune 12698–12703 (2010). and histone pathways. Nat. Neurosci. 18, 199–209 (2015). 66. Ramasamy, A. et al. Resolving the polymorphism-in-probe problem is critical for 78. Moss, D. J. H. et al. Identification of genetic variants associated with Huntington’s correct interpretation of expression QTL studies. Nucleic Acids Res. 41, e88–e88 disease progression: a genome-wide association study. Lancet Neurol. 16, (2013). 701–711 (2017). 67. Langfelder, P., Luo, R., Oldham, M. C. & Horvath, S. Is my network module pre- 79. Skene, N. G. & Grant, S. G. N. Identification of vulnerable cell types in major brain served and reproducible? PLoS Comput. Biol. 7, e1001057 (2011). disorders using single cell transcriptomes and expression weighted cell type 68. Brozzi, A., Urbanell, L., Germain, P. L., Magini, A. & Emiliani, C. hLGDB: A enrichment. Front. Neurosci. 10, 1–11 (2016). database of human lysosomal genes and their regulation. Database 2013, bat024 (2013). 69. Calvo, S. E., Clauser, K. R. & Mootha, V. K. MitoCarta2.0: An updated inventory of Open Access This article is licensed under a Creative Commons mammalian mitochondrial proteins. Nucleic Acids Res. 44, D1251–D1257 (2016). Attribution 4.0 International License, which permits use, sharing, 70. Khan, A. & Mathelier, A. Intervene: A tool for intersection and visualization of adaptation, distribution and reproduction in any medium or format, as long as you give multiple gene or genomic region sets. BMC Bioinforma. 18, 1–8 (2017). appropriate credit to the original author(s) and the source, provide a link to the Creative 71. International Parkinson Disease Genomics Consortium et al. Imputation of Commons license, and indicate if changes were made. The images or other third party sequence variants for identification of genetic risks for Parkinson’s disease: a material in this article are included in the article’s Creative Commons license, unless meta-analysis of genome-wide association studies. Lancet (Lond., Engl.) 377, indicated otherwise in a credit line to the material. If material is not included in the 641–649 (2011). article’s Creative Commons license and your intended use is not permitted by statutory 72. Schizophrenia Psychiatric Genome-Wide Association Study (GWAS) Consortium. regulation or exceeds the permitted use, you will need to obtain permission directly Genome-wide association study identifies five new schizophrenia loci. Nat. Genet. from the copyright holder. To view a copy of this license, visit http://creativecommons. 43, 969–976 (2011). org/licenses/by/4.0/. 73. Ripke, S. et al. Genome-wide association analysis identifies 13 new risk loci for schizophrenia. Nat. Genet. 45, 1150–1159 (2013). © The Author(s) 2019 npj Parkinson’s Disease (2019) 6 Published in partnership with the Parkinson’s Foundation