A meta-analysis of genome-wide association studies identifies 17 new Parkinson's disease risk loci

Abstract

Common variant genome-wide association studies (GWASs) have, to date, identified >24 risk loci for Parkinson's disease (PD). To discover additional loci, we carried out a GWAS comparing 6,476 PD cases with 302,042 controls, followed by a meta-analysis with a recent study of over 13,000 PD cases and 95,000 controls at 9,830 overlapping variants. We then tested 35 loci (P < 1 × 10(-6)) in a replication cohort of 5,851 cases and 5,866 controls. We identified 17 novel risk loci (P < 5 × 10(-8)) in a joint analysis of 26,035 cases and 403,190 controls. We used a neurocentric strategy to assign candidate risk genes to the loci. We identified protein-altering or cis-expression quantitative trait locus (cis-eQTL) variants in linkage disequilibrium with the index variant in 29 of the 41 PD loci. These results indicate a key role for autophagy and lysosomal biology in PD risk, and suggest potential new drug targets for PD.

Full text

A meta-analysis of genome-wide association studies identifies
17 new Parkinson’s disease risk loci
Diana Chang1, Mike A Nalls2,3, Ingileif B Hallgrímsdóttir4,6, Julie Hunkapiller1, Marcel van
der Brug1,6, Fang Cai1, International Parkinson’s Disease Genomics Consortium5,
23andMe ResearchTeam5, Geoffrey A Kerchner1, Gai Ayalon1, Baris Bingol1, Morgan
Sheng1, David Hinds4, Timothy W Behrens1, Andrew B Singleton2, Tushar R Bhangale1,7,
and Robert R Graham1,7,iD
1Genentech, Inc., South San Francisco, California, USA
2Laboratory of Neurogenetics, National Institute on Aging, US National Institutes of Health,
Bethesda, Maryland, USA
3Data Tecnica International, Glen Echo, Maryland, USA
423andMe Inc., Mountain View, California, USA
Abstract
Common variant genome-wide association studies (GWASs) have, to date, identified >24 risk loci
for Parkinson’s disease (PD). To discover additional loci, we carried out a GWAS comparing
6,476 PD cases with 302,042 controls, followed by a meta-analysis with a recent study of over
13,000 PD cases and 95,000 controls at 9,830 overlapping variants. We then tested 35 loci (
P
< 1 ×
10−6) in a replication cohort of 5,851 cases and 5,866 controls. We identified 17 novel risk loci (
P
< 5 × 10−8) in a joint analysis of 26,035 cases and 403,190 controls. We used a neurocentric
strategy to assign candidate risk genes to the loci. We identified protein-altering or
cis
–expression
quantitative trait locus (
cis
-eQTL) variants in linkage disequilibrium with the index variant in 29
of the 41 PD loci. These results indicate a key role for autophagy and lysosomal biology in PD
risk, and suggest potential new drug targets for PD.
Reprints and permissions information is available online at http://www.nature.com/reprints/index.html.
Correspondence should be addressed to R.R.G. (graham.robert@gene.com).
5A list of members appears in Supplementary Note 1
6Present addresses: Amgen, South San Francisco, California, USA (I.B.H.); E-Scape Bio, South San Francisco, California, USA
(M.v.d.B.).
7These authors contributed equally to this work
Robert R Graham http://orcid.org/0000-0001-7151-4277
URLs. PDGene, http://pdgene.org/; LDScore, https://github.com/bulik/ldsc; INRICH, https://atgu.mgh.harvard.edu/inrich/; GWAS
catalog, https://www.ebi.ac.uk/gwas/; GTEx portal, http://gtexportal.org/; STRING, http://string-db.org/.
Note: Any Supplementary Information and Source Data files are available in the online version of the paper
AUTHOR CONTRIBUTIONS
D.C., M.A.N., I.B.H., the 23andMe Research Team, G.A.K., B.B., M.S., D.H., T.W.B., A.B.S., T.R.B., and R.R.G. contributed to the
study design. D.C., M.A.N., I.B.H., T.R.B., and D.H. contributed to analysis and methods. D.C., M.A.N., T.W.B., A.B.S., T.R.B., and
R.R.G. wrote the manuscript. D.C., M.A.N., I.B.H., J.H., M.v.d.B., F.C., the International Parkinson’s Disease Genomics Consortium
(IPDGC), the 23andMe Research Team, G.A.K., G.A., B.B., M.S., D.H., T.W.B., A.B.S., T.R.B., and R.R.G. reviewed the manuscript.
M.v.d.B., F.C., IPDGC and the 23andMe Research Team provided samples or data.
COMPETING FINANCIAL INTERESTS
The authors declare competing financial interests: details are available in the online version of the paper.
HHS Public Access
Author manuscript
Nat Genet
. Author manuscript; available in PMC 2018 February 14.
Published in final edited form as:
Nat Genet
. 2017 October ; 49(10): 1511–1516. doi:10.1038/ng.3955.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

PD is the second most common neurodegenerative disorder1,2, with a prevalence of 3–4% in
individuals over 80 years of age3. PD is characterized by the loss of dopaminergic neurons
in the substantia nigra and the presence of Lewy bodies1,2. These neuropathologies manifest
in affected individuals primarily as motor-related symptoms, but the involvement of other
brain regions can lead to nonmotor symptoms4.
Early-onset, familial PD (onset at <60 years of age) accounts for a small fraction of cases5,
but the identified associated genes, including
LRRK2
,
GBA
, and
SNCA
, provide insight into
disease pathogenesis6,7. For the later-onset, common form of PD, at least 24 loci have been
associated at a genome-wide significant level with disease risk in individuals of European
ancestry8. The narrow-sense heritability (
h
2) explained by the confirmed PD risk loci is low
(0.033)9; however, the heritability explained by common variants is estimated at 0.227 (s.d.:
0.08)9, which suggests that additional loci with smaller effect sizes remain to be discovered.
We carried out a GWAS of 6,476 subjects from a 23andMe PD cohort (PDWBS (Web-Based
Study of Parkinson’s Disease)) and 302,042 controls genotyped on custom Illumina arrays
(Fig. 1). The 6,476 PD cases of European ancestry were independent from those previously
reported8 but met the same inclusion criteria, except that carriers of the
LRRK2
G2019S
mutation were not removed8,10. The 302,042 controls did not report having PD and were of
similar ancestry as the cases. The data were imputed with Minimac2 using 1000 Genomes
phase 1 haplotypes11,12. Single-nucleotide polymorphisms (SNPs) with low imputation
quality or that failed general quality control metrics were removed (Online Methods). After
correcting for age, sex, and the top principal components (Online Methods), we observed
minimal inflation for
P
values genome-wide (λgc = 1.057; λ1000 = 1.004; Supplementary
Fig. 1).
A total of 12 loci had
P
< 5 × 10−8 in the PDWBS analysis, including 11 of the loci that
were reported in a previous GWAS in individuals of European ancestry8 (Table 1). For the
remaining 13 previously reported loci, we observed
P
< 0.05 for 11 loci, with no significant
evidence for association observed in the PDWBS sample for
CHMP2B
(rs115185635) or
TMEM229B
(rs155399). The remaining novel locus in the PDWBS analysis, rs9468199 (
P
= 1.77 × 10−9), is more than 4 Mb from the nearest PD association in the HLA class II
region and is independent of rs9275326 (
P
conditional = 2.64 × 10−9).
Using genome-wide summary statistics from the PDWBS analysis, we estimated the
h
2
value for PD explained by common variants as 0.209 (95% confidence interval (CI): 0.148–
0.271, assuming a prevalence of 0.01), which is similar to the
h
2 value reported
previously9,10. Regions contributing to PD heritability were significantly enriched for
acetylation of histone H3 at lysine 27 (
P
= 0.001; Supplementary Table 1), a mark of active
regulatory regions. PD heritability was also enriched for histone marks in central nervous
system, adrenal, and pancreatic cell types (Supplementary Table 2), in agreement with a
previous study13.
We next carried out a meta-analysis between the PDWBS GWAS and results for the top
10,000 variants available from a large-scale meta-analysis for PD with over 13,000 cases and
95,000 controls8 (PDGene) (Fig. 1). For the 9,830 overlapping SNPs between the PDWBS
Chang et al. Page 2
Nat Genet
. Author manuscript; available in PMC 2018 February 14.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

and PDGene studies, we used an inverse-variance weighted method to combine association
statistics for meta-analysis14. The odds ratios and
P
values for the 9,830 overlapping SNPs
in the PDWBS and PDGene studies were correlated (ρ−log10(
P
value) = 0.85, ρOR = 0.58).
Furthermore, quantile–quantile (Q-Q) plots indicated an increase in the number of variants
with low
P
values (Supplementary Fig. 1), even after the exclusion of variants in regions
previously reported as associated with PD risk at a genome-wide significant level
(Supplementary Table 3).
The meta-analysis identified 35 loci associated at
P
< 1 × 10−6, including 15 loci with
P
< 5
× 10−8 (Fig. 2, Supplementary Figs. 2 and 3, Supplementary Table 4). Only two of the
previously reported loci (
BCKDK:
rs14235 and
MAPT:
rs17649553) and 2 of the 20
suggestive loci (
FCGR2A:
rs4657041,
ITGA2B:
rs5910) were in linkage disequilibrium (LD)
(
r
2 > 0.8) with variants associated (at
P
< 5 × 10−8) with any phenotype in the NHGRI
GWAS catalog15. Significant pleiotropy of PD risk loci with other complex diseases has not
been identified16, but this pleiotropy landscape may change as more modest effects are
uncovered.
We next sought validation of these 35 candidate loci in an independent cohort of 5,851 cases
and 5,866 controls of European ancestry genotyped with the semi-customized NeuroX
Illumina array8,17 (Fig. 1). Twenty-nine of the 35 loci either were directly genotyped on the
NeuroX array or had suitable proxies (
r
2 > 0.9 with the original SNP; Supplementary Table
5). Weaker proxies at four additional SNPs (
r
2 > 0.5) were available but were not used for
validation in this study (Supplementary Table 5). In a replication-phase joint analysis of
these 29 loci (meta-analysis of PDGene, PDWBS, and NeuroX), 16 had
P
< 5 × 10−8 (Table
2). Of these 16, all but 3 (rs4073221, rs10906923, and rs9468199) were also nominally
associated in the NeuroX study (one-sided
P
< 0.05). A genetic risk score8,18,19 defined by
these 16 loci, in addition to the previously reported loci, had a non-negligible ability to
predict PD case status (area under the curve, 0.6518; 95% CI, 0.6419–0.6616). This
represents a significant improvement over the predictive power of risk scores defined by
previously reported loci alone (
P
= 6 × 10−8) (Supplementary Note 1). In sum, we identified
16 independent PD risk loci with a joint
P
< 5 × 10−8 and 1 locus (rs601999) with
P
< 5 ×
10−8 in the discovery cohort with no suitable proxy for replication in the NeuroX cohort
(Table 2).
Overall, 11 of 17 novel loci were in high LD (
r
2 > 0.8) with at least one variant predicted to
affect transcription factor binding (Supplementary Table 6). Of the 17 novel loci and 24
previously reported loci, 10 contained residual associations with
P
< 1 × 10−3 after
conditioning on each region’s most significant SNP in the PDWBS data (Supplementary
Table 7). These regions included three of the four independent secondary signals reported by
Nalls
et al.
8, as well as one variant previously reported at a non-genome-wide significant
level (
P
= 5.15 × 10−7)20.
We note that the HLA region association with PD is particularly complex. Two candidate
genes from the HLA region were nominated on the basis of support from either a protein-
coding variant or an eQTL (Fig. 3). This is in line with a previous study that suggested that
the PD association in the HLA region may point to multiple HLA factors, including
Chang et al. Page 3
Nat Genet
. Author manuscript; available in PMC 2018 February 14.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

independent regulatory factors21. The association pattern observed at this locus may be
reminiscent of the HLA association observed in schizophrenia and linked to C4 copy
number22.
The identification of the causal variants and genes underlying regions associated with
common, complex disease is a major challenge23. Several statistical methods have been
proposed for the fine-mapping of causal variants23–25. Alternatively, some studies have
narrowed down lists of candidate genes by combining multiple levels of evidence with
scoring-based strategies26,27. Here we implemented a neurocentric strategy to nominate
candidate genes for PD-associated loci.
We incorporated seven sources of data to annotate the index variant and linked variants from
PD-associated loci (including eQTLs and expression data from GTEx28, as well as
expression data from brain cell types in mice29; a full list is provided in the Online
Methods). We used a two-stage approach to assign candidate genes to each locus (see the
Online Methods for further details, and Supplementary Fig. 4 for a graphical visualization).
In the first stage, we assigned a gene to a locus if (i) the index SNP or linked variants (
r
2 >
0.6) altered the protein sequence or (ii) the index variant was a
cis
-eQTL for the gene. When
no candidate genes were identified by the first stage, we ranked neighboring gene(s) on the
basis of neurologically related phenotypes and expression and assigned the gene with the
highest score to the locus (Online Methods).
With this strategy we identified a single candidate gene for 28 loci, and multiple candidate
genes with similar levels of supporting evidence for 13 loci (Fig. 3, Supplementary Figs. 5
and 6). The candidate-gene nomination strategy confirmed several known PD risk genes,
including
GBA
,
LRRK2
,
SNCA
, and
MAPT
. Among the 41 PD risk loci, a total of 29 loci
(71%) had either a protein-altering or a
cis
-eQTL variant linked to the index SNP
(Supplementary Tables 8 and 9). In addition, we carried out a colocalization analysis to
determine whether the GWAS signal and the eQTL signal pointed to the same causal
variant30 (Supplementary Note 1). Seven candidate genes also had evidence for protein–
protein interaction (Online Methods, Supplementary Table 10). Further studies are needed to
experimentally determine the causal genes in the PD risk loci; however, the identification of
candidate genes provides testable hypotheses for functional studies.
To gain insight into the biology, we tested the identified candidate genes in the 41 PD risk
loci for association with any pathways or gene sets compared with a background gene list
(Online Methods). We investigated whether candidate genes were enriched for pathways
previously implicated in PD: autophagy, lysosomal, and mitochondrial biology1. PD-
associated signals were enriched (at a threshold of
P
< 0.05/3 = 0.017) for lysosomal and
autophagy genes (
P
= 3.35 × 10−6 and
P
= 5.71 × 10−3, respectively). The addition of
candidate genes more than doubled the number of lysosomal genes observed in PD loci and
improved the enrichment significance (
P
all_loci = 3.35 × 10−6,
P
novel_loci = 3.64 × 10−5). We
also observed that one previously identified gene (
MCCC1
) and two novel candidate genes
(
COQ7
and
ALAS1
) mapped to the mitochondrial gene set (Supplementary Table 11).
Chang et al. Page 4
Nat Genet
. Author manuscript; available in PMC 2018 February 14.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

Lysosomal biology and its role in the degradation of protein aggregates emerged as a highly
significant pathway in PD risk. Among the five candidate genes linked to lysosomal biology,
two were previously identified candidate genes (
GBA
(glucocerebrosidase) and
TMEM175
(transmembrane protein 175)), and three were newly identified candidate genes (
CTSB
(cathepsin B),
ATP6V0A1
(ATPase H+ transporting V0 subunit a1), and
GALC
(galactosylceramidase)). Glucocerebrosidase is required for normal lysosomal activity and
α-synuclein degradation. In addition,
GBA
loss-of-function alleles are a common PD risk
factor31.
TMEM175
was recently shown to encode a potassium channel that can regulate
lysosomal function32, and the missense variant
TMEM175
M393T is strongly linked to the
index variant in the region (Supplementary Table 8).
CTSB
is a lysosomal cysteine protease.
A PD risk allele is linked to a
cis
-eQTL for
CTSB
in multiple tissues (Supplementary Table
9), where the risk allele is associated with reduced levels of
CTSB
mRNA. Double-knockout
mice for
Ctsb
and
Ctsl
(cathepsin L) show a tremor phenotype with cerebral and cerebellar
atrophy33. CTSB is also capable of degrading membrane-bound and soluble α-synuclein in
mice34.
Autophagy is the catabolic process that targets long-lived proteins and dysfunctional
organelles for lysosomal degradation. Autophagy and lysosomal degradation have been
implicated in PD by rare familial and common GWAS-associated
GBA
variants. We note
that a strong
cis
-eQTL for lysine acetyltransferase 8 (
KAT8
) is associated with PD risk, with
lower levels of
KAT8
mRNA linked to increased PD risk. Inhibition of KAT8 was recently
shown to decrease autophagic flux35.
Next, we used INRICH36 to investigate whether PD-associated regions were enriched for
gene sets in an unbiased fashion. Once again, we found significant enrichment of the
lysosomal pathway (
P
adjusted = 0.02) (Supplementary Table 12). We further examined the
expression of the PD candidate genes in a brain-specific cell-type expression data set in
mice29; however, we observed broad expression across the major brain cell types, and no
clear cell-type-specific pattern was evident (Supplementary Fig. 7).
Among the candidate genes newly identified in this study is
SH3GL2
(SH3 domain-
containing GRB2-like 2, endophilin A1), a gene recently demonstrated to be phosphorylated
by LRRK2 and which may have a role in clathrin-mediated endocytosis of synaptic
vesicles37. Dysregulation of
Elovl7
(elongation of very long chain fatty acids protein 7) in
mice results in several neurological phenotypes, including inflammatory astrocytosis and
microgliosis in the brain, and neuronal degeneration38. Upregulation of the candidate gene
SCN3A
(sodium voltage-gated channel α-subunit 3) enhances neuronal excitability and is
associated with epilepsy in both humans and animal models39.
The new loci also encode three transcription factors: SATB1, ZNF184, and TOX3. TOX3
has been implicated in neuronal survival40, and SATB1 has been associated with T cell
function, particularly the development of regulatory T cells41.
Several of the PD candidate genes are within the ‘druggable’ genome42, including the
previously identified serine/threonine kinase 39 (
STK39
) and the novel candidate gene
inositol 1,4,5-trisphosphate kinase B (
ITPKB
). An in-frame deletion of
ITPKB
Chang et al. Page 5
Nat Genet
. Author manuscript; available in PMC 2018 February 14.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

(rs147889095) is linked to a PD-associated variant, and complete loss of ITPKB was
reported in a patient with common-variable immunodeficiency43. STK39 is a kinase linked
to hypertension44, regulation of K+ levels, and the cellular stress response.
In summary, this study presents what to our knowledge is the largest meta-analysis of PD so
far, involving a total of 26,035 cases and 403,190 controls. We identified 17 novel PD loci
and, using a neurocentric candidate-gene nomination pipeline, found that several of the
newly identified PD risk genes have a role in lysosomal biology and autophagy. The
identification of these candidate genes allows for the prioritization of functional studies to
determine causal genes for PD and possible therapeutic targets.
ONLINE METHODS
PDWBS GWAS
The PDWBS is a genome-wide analysis of 6,476 PD cases and 302,042 control subjects, all
of whom were customers of 23andMe Inc. and consented to participate in research. The
study protocol was approved by the external AAHRPP-accredited institutional review board,
Ethical and Independent Review Services (E&I Review). Cases and controls were
designated on the basis of surveys10. Controls were selected from 23andMe Inc. research
participants who did not self-report as having been diagnosed with PD. Although the use of
self-reported controls can result in a reduction of power, the effect of this on the current
study was probably minimal (Supplementary Note 1). Any samples present in the PDGene
study8 were removed from the PDWBS analysis. The average age of cases and controls was
67.6 and 50.8 years, respectively. The study also included 147 cases (2.3%) and 554 controls
(0.18%) that were
LRRK2
G2019S carriers. Removing
LRRK2
G2019S carriers from the
analysis removed genome-wide significant associations at the
LRRK2
locus.
DNA extraction and genotyping were performed on saliva samples by CLIA-certified CAP-
accredited clinical laboratories of the Laboratory Corporation of America. Samples were
genotyped on one of the following four platforms: V1 and V2, two variants of the Illumina
HumanHap550+ BeadChip, with ~25,000 custom SNPs and ~950,000 total SNPs; V3,
Illumina OmniExpress+BeadChip with custom SNPs to increase overlap with the V2 chip,
with a total of ~950,000 SNPs; and V4, a custom chip that included SNPs overlapping V2
and V3 chips, low-frequency coding variants and ~570,000 SNPs. Samples with a call rate
lower than 98.5% were reanalyzed, and research participants with samples that failed
repeatedly were re-contacted and asked to provide additional samples.
Research participants were restricted to those of mainly (>97%) European ancestry10,45. All
research participants in the study were also required to share <700 cM identity by descent
(IBD) (estimated by a segmental IBD estimation algorithm46), corresponding approximately
to the sharing expected between first cousins. We additionally excluded individuals who
shared >700 cM IBD with any 23andMe research participant whose data was used in the
PDGene GWAS. Data were imputed on 1000 Genomes phase 1 haplotypes (September 2013
release) with Minimac2 on default settings11,47. Imputation was run separately on data from
each genotyping platform.
Chang et al. Page 6
Nat Genet
. Author manuscript; available in PMC 2018 February 14.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

For genotyped SNPs, SNPs were removed if they were genotyped on only the V1 and/or the
V2 chip, if they failed a parent–offspring transmission test on trio data, if they were not in
Hardy–Weinberg equilibrium (
P
< 10−20), or if they had a call rate < 0.90. For imputed
SNPs, SNPs were removed if they had an average
r
2 < 0.5 or minimum
r
2 < 0.3 in any
imputation batch, or failed a test for imputation batch effect (testing imputation dosage with
imputation batch;
P
< 10−50).
We applied logistic regression assuming an additive model to test for association between
case/control status and either genotypes or imputed dosages (for imputed SNPs). Only SNPs
with minor allele frequency (MAF) > 0.1% were analyzed. Covariates were added to adjust
for age, sex, the first five principal components, and genotyping platform version. A total of
12,896,220 variants (11,933,700 SNPs) were analyzed. The genomic inflation factor was
calculated from the median
P
value of analyzed variants. Scaling of the genomic inflation
factor by sample size was carried out as described previously for 1,000 cases and 1,000
controls48.
Meta-analysis of PD GWASs
Summary odds ratios, 95% CIs, and
P
values of the 10,000 most significant GWAS meta-
analysis results were obtained from PDGene (“URLs”). Cohort descriptions, quality control,
and meta-analysis for this study have been described previously8. SNP s.e. was derived from
the reported
P
values and odds ratios. More specifically, the
z
-statistic was calculated as the
square root of the inverse χ-square transformation of the
P
value, and the s.e. was calculated
as follows: s.e.m. = ln(odds ratio)/absolute(
z
-score).
There were 9,830 SNPs in common between the PDGene and the PDWBS data sets. A
fixed-effects model based on inverse-variance weighting, as implemented in METAL, was
used to combine summary statistics from the two studies14. Heterogeneity values (
I
2 and
Q
)
were obtained with PLINK49. Novel signals of association were defined as genome-wide
significant associations in the meta-analysis that did not overlap loci associated with PD at
genome-wide significant thresholds in the PDGene data (35 loci with
P
< 1 × 10−6).
Joint analysis with NeuroX
The NeuroX cohort was previously described8,17. Briefly, 5,851 cases and 5,866 controls of
European ancestry were genotyped on a semi-custom NeuroX array. A logistic regression
was carried out to test for association, with covariates to adjust for age, sex, and population
ancestry (the first five principal components). Twenty-five of the 35 novel loci were directly
genotyped on the chip, and four additional SNPs had suitable proxies (
r
2 > 0.9). At these 29
SNPs, we carried out a fixed-effects inverse-variance weighted meta-analysis14 for all three
studies (PDGene, PDWBS, and NeuroX) as described above.
Conditional analysis
Conditional analysis was run on all 17 loci that were significantly associated with PD in the
joint meta-analysis (
P
< 5 × 10−8) and the 24 previously reported PD loci using the PDWBS
study. For each locus, SNPs within 500 kb of the index SNP (the SNP with the most
Chang et al. Page 7
Nat Genet
. Author manuscript; available in PMC 2018 February 14.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

significant
P
value) were tested for association by the same methods as described above for
the PDWBS GWAS with the index SNP added as an additional covariate.
Heritability estimates
We used LD score regression (LDSC)50,51 to compute the narrow-sense heritability (
h
2)
estimates of PD in the PDWBS GWAS data (described above). Several methods exist for
estimating
h
2 with GWAS data50–52. We used LDSC to estimate
h
2 in this study because it
requires only summary-level data and is more computationally efficient for larger data sets.
Reference LD scores were computed with the European ancestry subset of the 1000
Genomes data for SNPs within 500 kb of the SNP to be scored. Strict filtering was applied
to ensure the robustness of heritability estimates as recommended50,51. After filtering,
Z
-
scores for 7,629,099 SNPs from the 23andMe study were used as input to LDSC. We further
used the stratified LD-score regression approach to partition heritability into 24 different
cell-type-agnostic annotation categories including conserved regions, histone marks, DNase
I hypersensitivity sites, ENCODE chromatin states, and enhancers51, as well as 10 different
cell-type-specific histone annotations. Significant enrichment was assessed at a strict
Bonferroni threshold of 0.0021 (0.05/24) for the 24 general categories, and 0.005 for the
cell-type-specific enrichment.
Pleiotropy analysis: overlap with EBI-NHGRI GWAS catalog
Data were downloaded from the EBI-NHGRI catalog15 (version available on 17 April
2016). If a variant in the meta-analysis was within 500 kb and in LD (
r
2 > 0.8) with an
association (
P
< 5 × 10−8) in the catalog, the meta-analysis signal was considered to be
overlapping the reported signal.
A neurocentric strategy to identify candidate causal variants and genes
Associated index SNPs were paired to candidate genes on the basis of two broad levels of
evidence: variant-level support and gene-level support (see Supplementary Fig. 4 for a
graphic representation). In the former category, index SNPs were paired with candidate
genes if there was evidence that the index SNP or an SNP in LD (
r
2 > 0.6) with the index
SNP was annotated with a putative high-impact variant (chromosome number variation,
exon loss variant, frame-shift variant, rare amino acid variant, splice donor or acceptor
variant, start-lost, stop-gained or stop-lost, and transcript ablation) or moderate-impact
variant (3′ or 5′ UTR truncation and exon loss, coding sequence variant, disruptive in-
frame deletion or insertion, in-frame deletion or insertion, missense variant, regulatory
region ablation, splice region variant, and transcription factor binding-site ablation). We
obtained variant annotations by running SnpEff53 on dbSNP build 142. A second source of
variant-level support consisted of
cis
-eQTL evidence.
Cis
-eQTLs as pre-computed by GTEx
(v6)28 were downloaded directly from the GTEx portal (“URLs”). Although eQTL results
were available for 46 tissues, including ten regions from the brain, our search for eQTLs was
limited by the sampled tissues and cell types, and therefore we might have missed any
eQTLs that are cell-type or tissue specific, in addition to eQTLs that are present only under
certain stimuli (for example, ‘response’ eQTLs). The index SNP was tested for significant
association with any gene where the TSS was within 250 kb of the index SNP. As roughly
90% of eQTLs are within 250 kb of a gene28, it is likely that we captured the majority of
Chang et al. Page 8
Nat Genet
. Author manuscript; available in PMC 2018 February 14.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

eQTLs while missing rarer, more distal events. For brain eQTLs, a strict Bonferroni
correction was applied to the raw eQTL
P
values to adjust for multiple testing of genes
within 250 kb of the index SNP. For other tissues, only eQTLs with a false discovery rate of
<0.05 as determined by GTEx28 were considered. We weighted brain and non-brain eQTLs
equally.
Gene-level support was used when an index SNP had no candidate genes supported by
variant-level data as described above. A list of genes within 250 kb of the index SNP was
obtained (gene models used by GTEx were downloaded from the GTEx portal), and each
gene was scored for neurological relevant features or annotations. Genes were first weighted
for neurological relevant phenotypic annotations. Genes were (i) scored for being
differentially expressed between PD patients and healthy controls (see Supplementary Note
1 for further details) (311 genes total genome-wide); (ii) annotated with ‘neuro’-associated
phenotypes in FlyBase54 (1,521 genes); (iii) scored for behavioral, neurological, and
olfactory phenotypes annotated in MGI55 (3,890 genes); and (iv) annotated with any
phenotypes related to neurological disorders or the brain in OMIM56 (521 genes). Lastly,
genes were scored for being expressed (median expression across samples > 2 reads per
kilobase per million mapped reads) in any cohort of GTEx brain region samples (15,197
genes) or in at least one brain cell type in the mouse expression data set29 (astrocyte,
microglia, neuron, or oligodendrocyte) (12,092 genes). For the gene-level support, we used a
tiered scoring scheme to weight phenotypic annotations more heavily than expression in the
brain (scores demarked in Supplementary Fig. 4) to enrich for genes with demonstrated
neurological related roles. At each locus, the gene (or tied genes) with the highest score was
nominated as the candidate gene for the region.
Protein–protein and coexpression analysis
All protein-coding genes within 250 kb of PD-associated loci (Supplementary Table 13)
were used as input to STRING57. Gene pairs that were either coexpressed or involved in
experimentally validated protein–protein interactions with a medium score or higher (score ≥
0.4) are reported in Supplementary Table 10.
Pathway enrichment analysis
Previously reported PD loci and novel PD associations were tested for enrichment in
particular pathways or gene sets. First, the nominated candidate genes for these PD-
associated loci were tested for enrichment in several targeted gene sets by a hypergeometric
test. The background list of genes for comparison was matched to the neurological-centric
candidate-gene nomination pipeline. The background list thus consisted of genes that had
mouse knockout phenotypes, had fly mutant phenotypes, had OMIM-related phenotype
annotations, had nominally significant
cis
-eQTLs in GTEx, were differentially expressed in
PD patients versus controls, and were expressed in GTEx brain tissue or mouse brain cell
types in the Barres data set.
We obtained mitochondrial genes from MitoMiner58 using the MitoCarta59,60 reference set
after excluding genes that mapped to the mitochondria (genes that map to the mitochondria
were not included in this metaanalysis). Lysosomal genes were obtained from the hlGDB61
Chang et al. Page 9
Nat Genet
. Author manuscript; available in PMC 2018 February 14.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

using only the proteomics and literature resources. Finally, we obtained autophagy genes
from the Human Autophagy Database62, as well as ten additional genes reported in a recent
siRNA screen of autophagic flux modulators35. A list of all genes in each pathway is
provided in Supplementary Note 1. The minimum
P
value per gene (for genes that an SNP
within the 9,830 variants assayed in this meta-analysis mapped to) is provided in
Supplementary Tables 14–16.
Second, we applied a non-targeted gene-set enrichment approach using INRICH36 to assess
whether regions associated with PD were enriched for genes in KEGG63 and Gene Ontology
(GO)64 gene sets. The 24 previously reported PD index variants and the 17 novel PD-
associated variants reported in this study were used as input into PLINK’s65 “show-tags”
function. The European 1000 Genomes12 samples were used for reference LD patterns. An
interval for each PD-associated variant was defined as the region from the leftmost tag
variant to the rightmost tag variant in the 1000 Genomes data.
We ran INRICH on these 41 intervals with the default settings, with the exception of
increasing the number of replicates and bootstraps to 5,000 (-r 5000 --q 5000) and setting
the pre-compute feature to false for software stability (-c). Enrichment for KEGG and GO
gene sets was assessed separately.
Data availability
A Life Sciences Reporting Summary for this paper is available. Summary statistics for the
9,830 variants presented in the discovery phase meta-analysis are available at http://research-
pub.gene.com/chang_et_al_2017. The full GWAS summary statistics for PDWBS will be
made available through 23andMe and Genentech to qualified researchers under an
agreement with 23andMe that protects the privacy of the 23andMe participants and an
agreement with Genentech for data sharing. Please contact D.H. (dhinds23andme.com) for
more information and to apply to access the data.
Supplementary Material
Refer to Web version on PubMed Central for supplementary material.
Acknowledgments
We thank all of the subjects who donated their time and biological samples to be a part of this study. Funding
details and additional acknowledgments are provided in Supplementary Note 1.
References
1. Corti O, Lesage S, Brice A. What genetics tells us about the causes and mechanisms of Parkinson’s
disease. Physiol. Rev. 2011; 91:1161–1218. [PubMed: 22013209]
2. Verstraeten A, Theuns J, Van Broeckhoven C. Progress in unraveling the genetic etiology of
Parkinson disease in a genomic era. Trends Genet. 2015; 31:140–149. [PubMed: 25703649]
3. Nussbaum RL, Ellis CE. Alzheimer’s disease and Parkinson’s disease. N. Engl. J. Med. 2003;
348:1356–1364. [PubMed: 12672864]
4. Shulman JM, De Jager PL, Feany MB. Parkinson’s disease: genetics and pathogenesis. Annu. Rev.
Pathol. 2011; 6:193–222. [PubMed: 21034221]
Chang et al. Page 10
Nat Genet
. Author manuscript; available in PMC 2018 February 14.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

5. Klein C, Westenberger A. Genetics of Parkinson’s disease. Cold Spring Harb. Perspect. Med. 2012;
2:a008888. [PubMed: 22315721]
6. Hardy J. Genetic analysis of pathways to Parkinson disease. Neuron. 2010; 68:201–206. [PubMed:
20955928]
7. Singleton AB, Farrer MJ, Bonifati V. The genetics of Parkinson’s disease: progress and therapeutic
implications. Mov. Disord. 2013; 28:14–23. [PubMed: 23389780]
8. Nalls MA, et al. Large-scale meta-analysis of genome-wide association data identifies six new risk
loci for Parkinson’s disease. Nat. Genet. 2014; 46:989–993. [PubMed: 25064009]
9. Keller MF, et al. Using genome-wide complex trait analysis to quantify ‘missing heritability’ in
Parkinson’s disease. Hum. Mol. Genet. 2012; 21:4996–5009. [PubMed: 22892372]
10. Do CB, et al. Web-based genome-wide association study identifies two novel loci and a substantial
genetic component for Parkinson’s disease. PLoS Genet. 2011; 7:e1002141. [PubMed: 21738487]
11. Fuchsberger C, Abecasis GR, Hinds DA. minimac2: faster genotype imputation. Bioinformatics.
2015; 31:782–784. [PubMed: 25338720]
12. Abecasis GR, et al. An integrated map of genetic variation from 1,092 human genomes. Nature.
2012; 491:56–65. [PubMed: 23128226]
13. Gagliano SA, et al. Genomics implicates adaptive and innate immunity in Alzheimer’s and
Parkinson’s diseases. Ann. Clin. Transl. Neurol. 2016; 3:924–933. [PubMed: 28097204]
14. Willer CJ, Li Y, Abecasis GR. METAL: fast and efficient meta-analysis of genomewide association
scans. Bioinformatics. 2010; 26:2190–2191. [PubMed: 20616382]
15. Welter D, et al. The NHGRI GWAS Catalog, a curated resource of SNP-trait associations. Nucleic
Acids Res. 2014; 42:D1001–D1006. [PubMed: 24316577]
16. Pickrell JK, et al. Detection and interpretation of shared genetic influences on 42 human traits. Nat.
Genet. 2016; 48:709–717. [PubMed: 27182965]
17. Nalls MA, et al. NeuroX, a fast and efficient genotyping platform for investigation of
neurodegenerative diseases. Neurobiol. Aging. 2015; 36:1605.e7–1605.e12.
18. Nalls MA, et al. Imputation of sequence variants for identification of genetic risks for Parkinson’s
disease: a meta-analysis of genome-wide association studies. Lancet. 2011; 377:641–649.
[PubMed: 21292315]
19. International Parkinson’s Disease Genomics Consortium & Wellcome Trust Case Control
Consortium 2. A two-stage meta-analysis identifies several new loci for Parkinson’s disease. PLoS
Genet. 2011; 7:e1002142. [PubMed: 21738488]
20. Pankratz N, et al. Meta-analysis of Parkinson’s disease: identification of a novel locus. RIT2. Ann.
Neurol. 2012; 71:370–384. [PubMed: 22451204]
21. Wissemann WT, et al. Association of Parkinson disease with structural and regulatory variants in
the HLA region. Am. J. Hum. Genet. 2013; 93:984–993. [PubMed: 24183452]
22. Sekar A, et al. Schizophrenia risk from complex variation of complement component 4. Nature.
2016; 530:177–183. [PubMed: 26814963]
23. Kichaev G, et al. Integrating functional data to prioritize causal variants in statistical fine-mapping
studies. PLoS Genet. 2014; 10:e1004722. [PubMed: 25357204]
24. Maller JB, et al. Bayesian refinement of association signals for 14 loci in 3 common diseases. Nat.
Genet. 2012; 44:1294–1301. [PubMed: 23104008]
25. Chen W, et al. Fine mapping causal variants with an approximate Bayesian method using marginal
test statistics. Genetics. 2015; 200:719–736. [PubMed: 25948564]
26. Okada Y, et al. Genetics of rheumatoid arthritis contributes to biology and drug discovery. Nature.
2014; 506:376–381. [PubMed: 24390342]
27. Bentham J, et al. Genetic association analyses implicate aberrant regulation of innate and adaptive
immunity genes in the pathogenesis of systemic lupus erythematosus. Nat. Genet. 2015; 47:1457–
1464. [PubMed: 26502338]
28. GTEx Consortium. The Genotype-Tissue Expression (GTEx) pilot analysis: multitissue gene
regulation in humans. Science. 2015; 348:648–660. [PubMed: 25954001]
29. Zhang Y, et al. An RNA-sequencing transcriptome and splicing database of glia, neurons, and
vascular cells of the cerebral cortex. J. Neurosci. 2014; 34:11929–11947. [PubMed: 25186741]
Chang et al. Page 11
Nat Genet
. Author manuscript; available in PMC 2018 February 14.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

30. Giambartolomei C, et al. Bayesian test for colocalisation between pairs of genetic association
studies using summary statistics. PLoS Genet. 2014; 10:e1004383. [PubMed: 24830394]
31. Sidransky E, et al. Multicenter analysis of glucocerebrosidase mutations in Parkinson’s disease. N.
Engl. J. Med. 2009; 361:1651–1661. [PubMed: 19846850]
32. Cang C, Aranda K, Seo YJ, Gasnier B, Ren D. TMEM175 is an organelle K+ channel regulating
lysosomal function. Cell. 2015; 162:1101–1112. [PubMed: 26317472]
33. Felbor U, et al. Neuronal loss and brain atrophy in mice lacking cathepsins B and L. Proc. Natl.
Acad. Sci. USA. 2002; 99:7883–7888. [PubMed: 12048238]
34. McGlinchey RP, Lee JC. Cysteine cathepsins are essential in lysosomal degradation of α-
synuclein. Proc. Natl. Acad. Sci. USA. 2015; 112:9322–9327. [PubMed: 26170293]
35. Hale CM, et al. Identification of modulators of autophagic flux in an image-based high content
siRNA screen. Autophagy. 2016; 12:713–726. [PubMed: 27050463]
36. Lee PH, O’Dushlaine C, Thomas B, Purcell SM. INRICH: interval-based enrichment analysis for
genome-wide association studies. Bioinformatics. 2012; 28:1797–1799. [PubMed: 22513993]
37. Arranz AM, et al. LRRK2 functions in synaptic vesicle endocytosis through a kinase-dependent
mechanism. J. Cell Sci. 2015; 128:541–552. [PubMed: 25501810]
38. Shin D, Shin JY, McManus MT, Ptácek LJ, Fu YH. Dicer ablation in oligodendrocytes provokes
neuronal impairment in mice. Ann. Neurol. 2009; 66:843–857. [PubMed: 20035504]
39. Tan NN, et al. Epigenetic downregulation of
Scn3a
expression by valproate: a possible role in its
anticonvulsant activity. Mol. Neurobiol. 2016; 54:2831–2842. [PubMed: 27013471]
40. Dittmer S, et al. TOX3 is a neuronal survival factor that induces transcription depending on the
presence of CITED1 or phosphorylated CREB in the transcriptionally active complex. J. Cell Sci.
2011; 124:252–260. [PubMed: 21172805]
41. Kondo M, et al. SATB1 plays a critical role in establishment of immune tolerance. J. Immunol.
2016; 196:563–572. [PubMed: 26667169]
42. Hopkins AL, Groom CR. The druggable genome. Nat. Rev. Drug Discov. 2002; 1:727–730.
[PubMed: 12209152]
43. Louis AG, Yel L, Cao JN, Agrawal S, Gupta S. Common variable immunodeficiency associated
with microdeletion of chromosome 1q42.1–q42.3 and inositol 1,4,5-trisphosphate kinase B
(ITPKB) deficiency. Clin. Transl. Immunology. 2016; 5:e59. [PubMed: 26900472]
44. Wang Y, et al. Whole-genome association study identifies
STK39
as a hypertension susceptibility
gene. Proc. Natl. Acad. Sci. USA. 2009; 106:226–231. [PubMed: 19114657]
45. Durand, EY., Do, CB., Mountain, JL., Macpherson, JM. Ancestry composition: a novel, efficient
pipeline for ancestry deconvolution. bioRxiv. 2014. Preprint at http://www.biorxiv.org/content/
early/2014/10/18/010512
46. Henn BM, et al. Cryptic distant relatives are common in both isolated and cosmopolitan genetic
samples. PLoS One. 2012; 7:e34267. [PubMed: 22509285]
47. Howie B, Fuchsberger C, Stephens M, Marchini J, Abecasis GR. Fast and accurate genotype
imputation in genome-wide association studies through prephasing. Nat. Genet. 2012; 44:955–959.
[PubMed: 22820512]
48. de Bakker PI, et al. Practical aspects of imputation-driven meta-analysis of genome-wide
association studies. Hum. Mol. Genet. 2008; 17:R122–R128. [PubMed: 18852200]
49. Purcell S, et al. PLINK: a tool set for whole-genome association and population-based linkage
analyses. Am. J. Hum. Genet. 2007; 81:559–575. [PubMed: 17701901]
50. Bulik-Sullivan BK, et al. LD score regression distinguishes confounding from polygenicity in
genome-wide association studies. Nat. Genet. 2015; 47:291–295. [PubMed: 25642630]
51. Finucane HK, et al. Partitioning heritability by functional annotation using genome-wide
association summary statistics. Nat. Genet. 2015; 47:1228–1235. [PubMed: 26414678]
52. Lee SH, Yang J, Goddard ME, Visscher PM, Wray NR. Estimation of pleiotropy between complex
diseases using single-nucleotide polymorphism-derived genomic relationships and restricted
maximum likelihood. Bioinformatics. 2012; 28:2540–2542. [PubMed: 22843982]
Chang et al. Page 12
Nat Genet
. Author manuscript; available in PMC 2018 February 14.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

53. Cingolani P, et al. A program for annotating and predicting the effects of single nucleotide
polymorphisms, SnpEff: SNPs in the genome of
Drosophila melanogaster
strain w1118; iso-2;
iso-3. Fly (Austin). 2012; 6:80–92. [PubMed: 22728672]
54. dos Santos G, et al. FlyBase: introduction of the
Drosophila melanogaster
Release 6 reference
genome assembly and large-scale migration of genome annotations. Nucleic Acids Res. 2015;
43:D690–D697. [PubMed: 25398896]
55. Eppig JT, Blake JA, Bult CJ, Kadin JA, Richardson JE. The Mouse Genome Database (MGD):
facilitating mouse as a model for human biology and disease. Nucleic Acids Res. 2015; 43:D726–
D736. [PubMed: 25348401]
56. McKusick VA. MENDELIAN Inheritance in Man and its online version, OMIM. Am. J. Hum.
Genet. 2007; 80:588–604. [PubMed: 17357067]
57. Szklarczyk D, et al. The STRING database in 2017: quality-controlled protein-protein association
networks, made broadly accessible. Nucleic Acids Res. 2017; 45:D362–D368. [PubMed:
27924014]
58. Smith AC, Robinson AJ. MitoMiner v3.1, an update on the mitochondrial proteomics database.
Nucleic Acids Res. 2016; 44:D1258–D1261. [PubMed: 26432830]
59. Calvo SE, Clauser KR, Mootha VK. MitoCarta2.0: an updated inventory of mammalian
mitochondrial proteins. Nucleic Acids Res. 2016; 44:D1251–D1257. [PubMed: 26450961]
60. Pagliarini DJ, et al. A mitochondrial protein compendium elucidates complex I disease biology.
Cell. 2008; 134:112–123. [PubMed: 18614015]
61. Brozzi A, Urbanelli L, Germain PL, Magini A, Emiliani C. hLGDB: a database of human
lysosomal genes and their regulation. Database (Oxford). 2013; 2013:bat024. [PubMed:
23584836]
62. Moussay E, et al. The acquisition of resistance to TNFα in breast cancer cells is associated with
constitutive activation of autophagy as revealed by a transcriptome analysis using a custom
microarray. Autophagy. 2011; 7:760–770. [PubMed: 21490427]
63. Kanehisa M, Goto S. KEGG: Kyoto Encyclopedia of Genes and Genomes. Nucleic Acids Res.
2000; 28:27–30. [PubMed: 10592173]
64. Gene Ontology Consortium. Gene Ontology Consortium: going forward. Nucleic Acids Res. 2015;
43:D1049–D1056. [PubMed: 25428369]
65. Chang CC, et al. Second-generation PLINK: rising to the challenge of larger and richer datasets.
Gigascience. 2015; 4:7. [PubMed: 25722852]
Chang et al. Page 13
Nat Genet
. Author manuscript; available in PMC 2018 February 14.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

Figure 1.
A flow chart of the two-stage meta-analysis design. In stage 1, we carried out a meta-
analysis of 9,830 SNPs between the PDWBS and PDGene studies. Thirty-five loci with
P
<
1 × 10−6 were carried forward into the replication-phase meta-analysis. In stage 2, we
carried out a meta-analysis between the two discovery-phase studies and the NeuroX study
for these 35 loci. Of these loci, 16 of the 29 available in NeuroX and 1 locus without
replication data were carried forward for downstream analyses (see the main text for further
details).
Chang et al. Page 14
Nat Genet
. Author manuscript; available in PMC 2018 February 14.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

Figure 2.
Results of the Parkinson’s disease discovery-phase meta-analysis. The top SNPs in
associated regions are indicated by pink symbols. Candidate genes for previously associated
loci are labeled in black (
P
< 5 × 10−8 in the discovery phase) or gray text (
P
> 5 × 10−8 in
the discovery phase); candidate genes for newly identified loci are labeled in red. The
y
-axis
shows the two-sided unadjusted −log10(
P
) values for association with PD. SNPs with
P
< 1
× 10−25 are indicated by triangles.
Chang et al. Page 15
Nat Genet
. Author manuscript; available in PMC 2018 February 14.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

Figure 3.
The candidate genes for regions associated with Parkinson’s disease. The most likely
candidate gene is annotated for each region that was significantly associated with PD in the
final joint analysis. Black or gray text indicates previously reported loci that had
P
values
less than or greater than 5 × 10−8 in the discovery phase, respectively. Red text indicates
newly identified loci that were significantly associated with PD in the final joint analysis.
Gray lines at the outer edge spanning multiple genes indicate candidate genes within a single
locus. Chromosome numbers are shown in the gray shaded ring, and support for candidate
genes is indicated by color-coding in the inner rings. The innermost ring indicates
expression of the gene in brain cell types (in a mouse expression data set) or in human brain
regions (in GTEx), or differential expression between PD brains and healthy control brains.
Chang et al. Page 16
Nat Genet
. Author manuscript; available in PMC 2018 February 14.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript

Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Chang et al. Page 17
Table 1
Parkinson’s disease risk loci previously reported at genome-wide significance levels
CHR:BPaSNP Candidate geneb
Effect allele/
alternate
allele
EAF in
1000
Genomes EAFcases/controlscPPDGenedORPDGene PPDWBSeORPDWBS Pdiscovery ORdiscovery
ORdiscovery
95% CI
1:155135036 rs35749011
GBA
G/A 0.976 0.979/0.988 6.10 × 10−23 0.57 5.33 × 10−14 0.59 2.59 × 10−35 0.58 0.53–0.63
1:205723572 rs823118
NUCKS1
,
SLC41A1
C/T 0.467 0.419/0.443 1.96 × 10−16 0.89 8.78 × 10−9 0.90 1.12 × 10−23 0.89 0.87–0.91
1:232664611 rs10797576
SIPA1L2
T/C 0.137 0.145/0.135 1.76 × 10−10 1.13 7.4 × 810−4 1.10 8.41 × 10−13 1.12 1.09–1.15
2:135539967 rs6430538
TMEM163
,
CCNT2
T/C 0.488 0.426/0.450 3.35 × 10−19 0.88 1.5 × 410−6 0.91 8.24 × 10−24 0.89 0.87–0.91
2:169110394 rs1474055
STK39
C/T 0.881 0.855/0.874 7.11 × 10−16 0.82 1.11 × 10−11 0.83 5.68 × 10−26 0.83 0.80–0.86
3:87520857
f
rs115185635
CHMP2B
C/G 0.036 0.040/0.039 2.2 × 10−8 1.79 0.182 1.08 1.22 × 10−4 1.21 1.10–1.33
3:182762437 rs12637471
MCCC1
A/G 0.219 0.175/0.198 5.38 × 10−22 0.84 4.27 × 10−10 0.86 2.11 × 10−30 0.85 0.82–0.87
4:951947 rs34311866
TMEM175
,
DGKQ
C/T 0.199 0.212/0.184 6.00 × 10−41 1.26 2.48 × 10−12 1.18 1.47 × 10−50 1.23 1.20–1.27
4:15737101 rs11724635
FAM200B
,
CD38
C/A 0.437 0.437/0.452 4.26 × 10−17 0.89 1.0 × 410−4 0.93 1.22 × 10−19 0.90 0.88–0.92
4:77198986 rs6812193
g FAM47E
T/C 0.398 0.351/0.370 1.85 × 10−11 0.91 1.24 × 10−4 0.93 1.43 × 10−14 0.92 0.90–0.94
4:90626111 rs356182
SNCA
G/A 0.375 0.406/0.349 1.85 × 10−82 1.34 1.44 × 10−42 1.31 5.21 × 10−123 1.33 1.30–1.36
6:32666660 rs9275326
HLA-DRB6
,
HLA-DQA1
T/C 0.114 0.099/0.105 5.81 × 10−13 0.80 1.04 × 10−3 0.90 1.26 × 10−13 0.85 0.82–0.89
7:23293746 rs199347
KLHL7
,
NUPL2
,
GPNMB
G/A 0.368 0.389/0.412 5.62 × 10−14 0.90 8.66 × 10−6 0.92 3.51 × 10−18 0.91 0.89–0.93
8:16697091 rs591323
MICU3
A/G 0.293 0.258/0.274 3.17 × 10−8 0.91 1.61 × 10−4 0.92 2.38 × 10−11 0.91 0.89–0.94
10:121536327 rs117896735
BAG3
A/G 0.012 0.021/0.015 1.21 × 10−11 1.77 1.75 × 10−9 1.57 2.23 × 10−19 1.65 1.48–1.85
11:83544472 rs3793947
DLG2
A/G 0.463 0.431/0.442 2.59 × 10−8 0.91 8.92 × 10−3 0.95 3.72 × 10−9 0.93 0.91–0.95
11:133765367 rs329648
MIR4697
T/C 0.327 0.369/0.351 8.05 × 10−12 1.11 9.16 × 10−4 1.07 1.11 × 10−13 1.09 1.07–1.12
12:40614434 rs76904798h
LRRK2
T/C 0.132 0.152/0.137 4.86 × 10−14 1.16 4.10 × 10−7 1.14 1.21 × 10−19 1.15 1.12–1.19
12:123303586 rs11060180
OGFOD2
G/A 0.45 0.423/0.449 3.08 × 10−11 0.91 4.95 × 10−11 0.88 2.05 × 10−20 0.90 0.88–0.92
14:55348869 rs11158026
GCH1
T/C 0.307 0.309/0.331 2.88 × 10−10 0.91 2.65 × 10−7 0.90 4.30 × 10−16 0.91 0.89–0.93
14:67984370 rs1555399
TMEM229B
T/A 0.544 0.518/0.514 5.70 × 10−16 1.15 0.453 1.01 9.61 × 10−11 1.09 1.06–1.11
15:61994134 rs2414739
VPS13C
G/A 0.292 0.250/0.266 3.59 × 10−12 0.90 1.1 × 10−3 0.93 3.94 × 10−14 0.91 0.89–0.93
16:31121793 rs14235
ZNF646
,
KAT8
A/G 0.397 0.388/0.378 3.63 × 10−12 1.10 0.0339 1.04 5.44 × 10−12 1.08 1.06–1.10
17:43994648 rs17649553
ARHGAP27
,
CRHR1
,
SPPL2C
,
MAPT
,
STH
,
KANSL1
T/C 0.232 0.187/0.221 6.11 × 10−49 0.77 9.24 × 10−22 0.80 1.26 × 10−68 0.78 0.76–0.80
18:40673380 rs12456492
SYT4
G/A 0.332 0.336/0.315 2.15 × 10−11 1.10 5.13 × 10−6 1.10 5.56 × 10−16 1.10 1.07–1.12
Nat Genet
. Author manuscript; available in PMC 2018 February 14.

Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Chang et al. Page 18
CHR:BPaSNP Candidate geneb
Effect allele/
alternate
allele
EAF in
1000
Genomes EAFcases/controlscPPDGenedORPDGene PPDWBSeORPDWBS Pdiscovery ORdiscovery
ORdiscovery
95% CI
19:2363319
f
rs62120679 LSM7 t/C 0.324 0.314/0.310 2.52 × 10−9 1.14 0.24O 1.03 6.64 × 10−7 1.08 1.05–1.11
20:3168166
f
rs8118008 DDRGK1 A/G 0.596 0.615/0.609 2.32 × 10−8 1.11 0.283 1.02 1.99 × 10−6 1.07 1.04–1.09
Rows in bold text refer to loci that did not pass the genome-wide significance threshold (5 × 10−8) in the discovery-phase meta-analysis.
a
Chromosome and physical position according to Hg19.
b
Details regarding the assignment of candidate genes are provided in the Online Methods.
c
Effect allele frequency (EAF) measured in PDWBS controls or cases.
dP
value for SNP in the publicly available PDGene data (13,708 cases, 95,282 controls). Publicly available data for the following SNPs include an additional 5,450 cases and 5,798 controls genotyped on
NeuroX: rs115185635, rs35749011, rs117896735, rs62120679, rs9275326, rs3793947, rs1555399, rs1474055, and rs8118008.
eP
value for SNP in PDWBS (6,476 cases, 302,042 controls).
f
The alternate SNP is genome-wide significant (rs12651582;
P
= 3.51 × 10−8).
g
The alternate SNP is genome-wide significant (rs76904798;
P
= 4.45 × 10−75).
Nat Genet
. Author manuscript; available in PMC 2018 February 14.

Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Chang et al. Page 19
Table 2
Seventeen novel regions associated with Parkinson’s disease at genome-wide significance levels
CHR:BPaSNP Candidate
geneb
Effect allele/
alternate
allele EAF in 1000
Genomes Pdiscovery ORdiscovery PNeuroX ORNeuroX Pjoint ORJoint
ORJoint (95%
CI)
1:226916078 rs4653767
ITPKB
C/T 0.315 2.40 × 10−10 0.92 0.017 0.93 1.63 × 10−11 0.92 0.90–0.94
2:102413116 rs34043159
IL1R2
C/T 0.352 3.83 × 10−8 1.07 1.91 × 10−4 1.11 5.48 × 10−11 1.08 1.06–1.10
2:166133632 rs353116
SCN3A
T/C 0.385 9.73 × 10−7 0.94 8.98 × 10−3 0.93 2.98 × 10−8 0.94 0.92–0.96
3:18277488 rs4073221
SATB1
G/T 0.132 3.02 × 10−9 1.11 0.583 1.02 1.57 × 10−8 1.10 1.06–1.13
3:48748989 rs12497850
NCKIPSD
,
CDC71
G/T 0.347 6.80 × 10−8 0.93 0.040 0.94 9.16 × 10−9 0.93 0.91–0.96
3:52816840 rs143918452
ALAS1
,
TLR9
,
DNAH1
,
BAP1
,
PHF7
,
NISCH
,
STAB1
,
ITIH3
,
ITIH4
G/A 0.996 2.25 × 10−7 0.68 0.095 0.73 3.20 × 10−8 0.68 0.60–0.78
4:114360372 rs78738012
ANK2
,
CAMK2D
C/T 0.106 2.11 × 10−9 1.14 7.5 × 10−3 1.12 4.78 × 10−11 1.13 1.09–1.17
5:60273923 rs2694528
ELOVL7
C/A 0.115 1.69 × 10−11 1.15 6.25 × 10−5 1.19 4.84 × 10−15 1.15 1.11–1.20
6:27681215 rs9468199
ZNF184
A/G 0.172 3.44 × 10−13 1.12 0.302 1.04 1.46 × 10−12 1.11 1.08–1.14
8:11707174 rs2740594
c CTSB
A/G 0.753 9.54 × 10−11 1.10 7.95 × 10−3 1.08 5.91 × 10−12 1.09 1.07–1.12
8:22525980 rs2280104
SORBS3
,
PDLIM2
,
C8orf58
,
BIN3
T/C 0.367 9.06 × 10−7 1.06 7.87 × 10−3 1.08 2.53 × 10−8 1.07 1.04–1.09
9:17579690 rs13294100
SH3GL2
T/G 0.371 1.99 × 10−12 0.91 0.037 0.94 4.84 × 10−13 0.92 0.89–0.94
10:15569598 rs10906923
FAM171A1
C/A 0.306 2.37 × 10−8 0.93 0.133 0.96 1.35 × 10−8 0.93 0.91–0.96
14:88472612 rs8005172
GALC
T/C 0.424 1.20 × 10−9 1.08 0.022 1.06 8.77 × 10−11 1.08 1.05–1.10
16:19279464 rs11343
COQ7
T/G 0.454 1.46 × 10−9 1.07 0.019 1.06 9.13 × 10−11 1.07 1.05–1.10
16:52599188 rs4784227
TOX3
T/C 0.265 8.29 × 10−8 1.08 1.47 × 10−4 1.12 9.75 × 10−11 1.09 1.06–1.12
17:40698158 rs601999
ATP6V0A1
,
PSMC3IP
,
TUBG2
C/T 0.699 8.03 × 10−9 0.93 NA NA NA NA NA
Summary statistics are shown for the discovery cohort (PDWBS and PDGene), NeuroX (5,851 cases, 5,866 controls), and the joint meta-analysis of the discovery and NeuroX data.
Additional summary statistics for NeuroX and the joint meta-analysis are available in Supplementary Table 5. EAF, effect allele frequency.
a
Chromosome and physical position according to Hg19.
b
Details regarding the assignment of candidate genes are provided in the Online Methods.
c
NeuroX and joint statistics are shown for proxy SNP rs1293298.
Nat Genet
. Author manuscript; available in PMC 2018 February 14.