Genome-Wide Gene-Environment Study Identifies Glutamate Receptor Gene GRIN2A as a Parkinson's Disease Modifier Gene via Interaction with Coffee

Article (PDF Available)inPLoS Genetics 7(8):e1002237 · August 2011with49 Reads
DOI: 10.1371/journal.pgen.1002237 · Source: PubMed
Abstract
Our aim was to identify genes that influence the inverse association of coffee with the risk of developing Parkinson's disease (PD). We used genome-wide genotype data and lifetime caffeinated-coffee-consumption data on 1,458 persons with PD and 931 without PD from the NeuroGenetics Research Consortium (NGRC), and we performed a genome-wide association and interaction study (GWAIS), testing each SNP's main-effect plus its interaction with coffee, adjusting for sex, age, and two principal components. We then stratified subjects as heavy or light coffee-drinkers and performed genome-wide association study (GWAS) in each group. We replicated the most significant SNP. Finally, we imputed the NGRC dataset, increasing genomic coverage to examine the region of interest in detail. The primary analyses (GWAIS, GWAS, Replication) were performed using genotyped data. In GWAIS, the most significant signal came from rs4998386 and the neighboring SNPs in GRIN2A. GRIN2A encodes an NMDA-glutamate-receptor subunit and regulates excitatory neurotransmission in the brain. Achieving P(2df) = 10(-6), GRIN2A surpassed all known PD susceptibility genes in significance in the GWAIS. In stratified GWAS, the GRIN2A signal was present in heavy coffee-drinkers (OR = 0.43; P = 6×10(-7)) but not in light coffee-drinkers. The a priori Replication hypothesis that "Among heavy coffee-drinkers, rs4998386_T carriers have lower PD risk than rs4998386_CC carriers" was confirmed: OR(Replication) = 0.59, P(Replication) = 10(-3); OR(Pooled) = 0.51, P(Pooled) = 7×10(-8). Compared to light coffee-drinkers with rs4998386_CC genotype, heavy coffee-drinkers with rs4998386_CC genotype had 18% lower risk (P = 3×10(-3)), whereas heavy coffee-drinkers with rs4998386_TC genotype had 59% lower risk (P = 6×10(-13)). Imputation revealed a block of SNPs that achieved P(2df)
Genome-Wide Gene-Environment Study Identifies
Glutamate Receptor Gene
GRIN2A
as a Parkinson’s
Disease Modifier Gene via Interaction with Coffee
Taye H. Hamza
1
, Honglei Chen
2
, Erin M. Hill-Burns
1
, Shannon L. Rhodes
3
, Jennifer Montimurro
1
,
Denise M. Kay
1
, Albert Tenesa
4
, Victoria I. Kusel
1
, Patricia Sheehan
1
, Muthukrishnan Eaaswarkhanth
1
,
Dora Yearout
1,5
, Ali Samii
5
, John W. Roberts
6
, Pinky Agarwal
7
, Yvette Bordelon
8
, Yikyung Park
9
, Liyong
Wang
10
, Jianjun Gao
2
, Jeffery M. Vance
10
, Kenneth S. Kendler
11
, Silviu-Alin Bacanu
11
, William K. Scott
10
,
Beate Ritz
3,8,12
, John Nutt
13
, Stewart A. Factor
14
, Cyrus P. Zabetian
5
, Haydeh Payami
1
*
1 New York State Department of Health Wadsworth Center, Albany, New York, United States of America, 2 Epidemiology Branch, National Institute of Environmental
Health Sciences, Research Triangle Park, North Carolina, United States of America, 3 Department of Epidemiology, School of Public Health, University of California Los
Angeles, Los Angeles, California, United States of America, 4 Institute of Genetics and Molecular Medicine and The Roslin Institute, University of Edinburgh, Edinburgh,
United Kingdom, 5 VA Puget Sound Health Care System and Department of Neurology, University of Washington, Seattle, Washington, United States of America, 6 Virginia
Mason Medical Center, Seattle, Washington, United States of America, 7 Booth Gardner Parkinson’s Care Center, Evergreen Hospital Medical Center, Kirkland, Washington,
United States of America, 8 Department of Neurology, School of Medicine, University of California Los Angeles, Los Angeles, California, United States of America,
9 Nutritional Epidemiology Branch, Divisions of Cancer Epidemiology and Genetics, National Cancer Institute, Bethesda, Maryland, United States of America, 10 Dr. John T.
Macdonald Foundation Department of Human Genetics and John P. Hussman Institute for Human Genomics, University of Miami Miller School of Medicine, Miami, Florida,
United States of America, 11 Virginia Institute for Psychiatric and Behavioral Genetics, Department of Psychiatry, Virginia Commonwealth University, Richmond, Virginia,
United States of America, 12 Department of Environmental Health Sciences, Center for Occupational and Environmental Health, School of Public Health, University of
California Los Angeles, Los Angeles, California, United States of America, 13 Department of Neurology, Oregon Health and Sciences University, Portland, Oregon, United
States of America, 14 Department of Neurology, Emory University School of Medicine, Atlanta, Georgia, United States of America
Abstract
Our aim was to identify genes that influence the inverse association of coffee with the risk of developing Parkinson’s disease (PD).
We used genome-wide genotype data and lifetime caffeinated-coffee-consumption data on 1,458 persons with PD and 931
without PD from the NeuroGenetics Research Consortium (NGRC), and we performed a genome-wide association and interaction
study (GWAIS), testing each SNP’s main-effect plus its interaction with coffee, adjusting for sex, age, and two principal
components. We then stratified subjects as heavy or light coffee-drinkers and performed genome-wide association study (GWAS)
in each group. We replicated the most significant SNP. Finally, we imputed the NGRC dataset, increasing genomic coverage to
examine the region of interest in detail. The primary analyses (GWAIS, GWAS, Replication) were performed using genotyped data.
In GWAIS, the most significant signal came from rs4998386 and the neighboring SNPs in GRIN2A. GRIN2A encodes an NMDA-
glutamate-receptor subunit and regulates excitatory neurotransmission in the brain. Achieving P
2df
=10
26
, GRIN2A surpassed all
known PD susceptibility genes in significance in the GWAIS. In stratified GWAS, the GRIN2A signal was present in heavy coffee-
drinkers (OR = 0.43; P = 6610
27
) but not in light coffee-drinkers. The aprioriReplication hypothesis that ‘‘Among heavy coffee-
drinkers, rs4998386_T carriers have lower PD risk than rs4998386_CC carriers’’ was confirmed: OR
Replication
= 0.59,
P
Replication
=10
23
;OR
Pooled
=0.51, P
Pooled
=7610
28
. Compared to light coffee-drinkers with rs4998386_CC genotype, heavy
coffee-drinkers with rs4998386_CC genotype had 18% lower risk (P = 3610
23
), whereas heavy coffee-drinkers with rs4998386_TC
genotype had 59% lower risk (P = 6610
213
). Imputation revealed a block of SNPs that achieved P
2df
,5610
28
in GWAIS, and
OR = 0.41, P = 3610
28
in heavy coffee-drinkers. This study is proof of concept that inclusion of environmental factors can help
identify genes that are missed in GWAS. Both adenosine antagonists (caffeine-like) and glutamate antagonists (GRIN2A-related)
are being tested in clinical trials for treatment of PD. GRIN2A may be a useful pharmacogenetic marker for subdividing individuals
in clinical trials to determine which medications might work best for which patients.
Citation: Hamza TH, Chen H, Hill-Burns EM, Rhodes SL, Montimurro J, et al. (2011) Genome-Wide Gene-Environment Study Identifies Glutamate Receptor Gene
GRIN2A as a Parkinson’s Disease Modifier Gene via Interaction with Coffee. PLoS Genet 7(8): e1002237. doi:10.1371/journal.pgen.1002237
Editor: Jonathan Flint, The Wellcome Trust Centre for Human Genetics, University of Oxford, United Kingdom
Received April 9, 2011; Accepted June 24, 2011; Published August 18, 2011
This is an open-access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for
any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication.
Funding: NGRC was supported by grants from the National Institute of Neurological Disorders and Stroke (NINDS) R01 NS36960 and R01 NS067469 (http://www.
ninds.nih.gov/) and from the Michael J. Fox Foundation (Edmond J. Safra Global Genetics Consortia) (http://www.michaeljfox.org/). Additional support for subject
recruitment was provided by the Department of Veterans Affairs (1I01BX000531) and The Close to a Cure Foundation: A Fund for Parkinson’s Research of
Foundation for the Carolinas. Genome-wide genotyping was performed at CIDR and funded by the National Institutes of Health (HHSN268200782096C). The PEG
study was funded by Nati onal Institute of Environmental Health Sciences (NIEHS R01 ES010544 and P01 ES016732), UCLA Center for Neurodegeneration Science,
and NINDS (P50 NS038367). The PAGE study was supported by NIEHS (Z01-ES-101986) and the National Cancer Institute (Z01 CP010196-02). HIHG was supported
by NINDS (P50 NS039764). Some of the HIHG samples were collected while JMV and WKS were faculty members at Duke University. The content is solely the
responsibility of the authors and does not necessarily represent the official views of the funding agencies. The funders had no role in study design, data collection
and analysis, decision to publish, or preparation of the manuscript.
Competing Interests: The authors have declared that no competing interests exist.
* E-mail: hpayami@wadsworth.org
PLoS Genetics | www.plosgenetics.org 1 August 2011 | Volume 7 | Issue 8 | e1002237
Introduction
Common disorders are thought to have both genetic and
environmental components. Genome-wide association studies
(GWAS) have successfully identified numerous susceptibility loci
for many common disorders ranging from behavioral traits such as
addiction and substance abuse to infectious and immune-related
disorders, age-related neurodegenerative disorders like Alzheimer’s,
Parkinson’s and macular degeneration, metabolic disorders, psychi-
atric disorders, and many more (for the list and results of over 800
published GWAS see http://www.genome.gov/gwastudies). De-
spite the success of GWAS, the heritability of common disorders
cannot be fully explained by the genes that have been discovered [1].
GWAS are built on the notion that common alleles predispose to
common disorders. Rare variants, which are probably responsible
for some of the missing heritability, would not have been detected by
GWAS. Sequencing the genome and novel analytical methods will
help identify the rare variants. Another hiding place for the missing
heritability is in interactions. Genes that impact disease through
interactions with other genes or environmental factors are not
detected by GWAS if their main effects are small. GWAS can only
identify genes that exhibit significant main effects; genes that require
the interacting factor to be included in the study to show their
association with disease are missed. Inclusion of key environmental
factors in genome-wide studies is anticipated to be an important next
step for deciphering the genetic structure of common multifactorial
disorders. Amassing sufficient analytic power for gene-environment
studies, however, is a challenge. Power decreases dramatically as a
function of frequency of exposure, number of parameters being
estimated and sample size. Interaction studies require at least four
times the sample size that standard GWAS would require to detect
an effect of similar magnitude (reviewed in [2]). Yet, there are fewer
datasets with both DNA and environmental exposure data than
those with DNA alone, and their sample sizes are often smaller.
Parkinson’s disease (PD) is a classic example of a common
multifactorial disorder. PD is characterized by neurodegeneration
in the substantia nigra that manifests initially as a movement
disorder but often leads to cognitive and psychiatric problems as
well. PD is progressive and there is no treatment currently
available that could prevent or slow disease progression. PD is the
second most common neurodegenerative disease after Alzheimer’s
disease; it affects about 5 million individuals in the 10 most
populous nations and is expected to double in frequency by 2030
[3]. Until the 1990’s PD was thought to be purely environmental
with no genetic component. In the last decade, numerous genes
have been identified, some of which can cause PD [4] and others
that are susceptibility loci [5–10]. There are also compelling data
from epidemiology that cigarette smoking and caffeinated-coffee
consumption are associated with reduced risk of developing PD
[11,12] and that exposure to environmental neurotoxins is
associated with increased risk of developing PD [13]. Thus PD
is a strong candidate for studying gene-environment interactions
[14].
We conducted a genome-wide association and interaction study
(GWAIS) using the joint test [15] for each SNP’s marginal
association and its interaction with coffee consumption on PD risk,
followed by stratified GWAS in heavy and light coffee drinkers (see
Analytic Strategy in Materials and Methods section). Our aim was
to identify genes that enhance or diminish the protective effect of
caffeinated-coffee for use as biomarkers for pharmacogenetic
prevention and treatment. Caffeine is an adenosine-receptor
antagonist. In animal models of PD, where administration of
neurotoxins is used to destroy dopaminergic neurons mimicking
PD, caffeine and selective A
2A
-antagonists have been shown to be
neuroprotective and attenuate dopamine loss [16]. Selective A
2A
-
antagonists have been studied in human clinical trials and found to
be safe, well tolerated and to provide symptomatic benefit for
persons with PD [17,18]; however, efficacy has not been high
enough in the first generation of the drugs to meet regulatory
approval for use as PD drugs. We posit that subsets of patients with
certain genotypes may respond well to a given treatment and
others may not. When they are combined the average efficacy may
be insufficient for regulatory approval, while a subgroup of
patients with certain genotype might still benefit substantially. If
our prediction is correct, incorporating genetics in clinical trials of
PD could revolutionize PD drug development. By examining the
interaction of caffeinated-coffee with 811,597 SNPs in a
hypothesis-free genome-wide study, we discovered GRIN2A as a
novel PD modifier gene. GRIN2A encodes a subunit of the
NMDA-glutamate-receptor which is well known for regulating
excitatory neurotransmission in the brain and for controlling
movement and behavior.
Materials and Methods
Human subjects
Human Subject Committees of the participating institutions
approved the study. The Discovery dataset was nested in the
NeuroGenetics Research Consortium (NGRC) GWAS which
successfully identified known PD genes as well as a novel association
with HLA [5] which has been widely replicated [10,19]. For the
present GWAIS, Replication samples were provided by PEG [20]
(Parkinson, Environment, and Gene), PAGE [21] (Parkinson’s,
Genes, and Environment from the prospective NIH-AARP Diet
and Health Study cohort), and HIHG [9] (Hussman Institute for
Human Genomics). Persons with PD had been diagnosed by
neurologists using standard criteria [22], control subjects self-
reported as not having PD. Cases and controls were all unrelated,
non-Hispanic Caucasian, from United States. The NGRC cohort
was clinic-based sequentially ascertained patients, PEG and PAGE
Author Summary
Parkinson’s disease (PD), like most common disorders,
involves interactions between genetic make-up and
environmental exposures that are unique to each individ-
ual. Caffeinated-coffee consumption may protect some
people from developing PD, although not all benefit
equally. In a genome-wide search, we discovered that
variations in the glutamate-receptor gene GRIN2A modu-
late the risk of developing PD in heavy coffee drinkers. The
study was hypothesis-free, that is, we cast a net across the
entire genome allowing statistical significance to point us
to a genetic variant, regardless of whether it fell in a
genomic desert or an important gene. Fortuitously, the
most significant finding was in a well-known gene, GRIN2A,
which regulates brain signals that control movement and
behavior. Our finding is important for three reasons: First,
it is a proof of concept that studying genes and
environment on the whole-genome scale is feasible, and
this approach can identify important genes that are missed
when environmental exposures are ignored. Second, the
knowledge of interaction between GRIN2A, which is
involved in neurotransmission in the brain, and caffeine,
which is an adenosine-A
2A
-receptor antagonist, will
stimulate new research towards understanding the cause
and progression of PD. Third, the results may lead to
personalized prevention of and treatment for PD.
GRIN2A, Coffee, and Parkinson’s Disease
PLoS Genetics | www.plosgenetics.org 2 August 2011 | Volume 7 | Issue 8 | e1002237
were community-based incident cases, HIHG was clinic-based and
self-referral cases. The numbers of cases/controls with genotype,
coffee/caffeine and key clinical and demographic data were
NGRC = 1458/931, PEG = 280/310, PAGE = 525/1474, HIHG
= 209/133 (Table S1).
Coffee/caffeine
NGRC, PEG and HIHG had collected lifetime caffeinated-
coffee consumption data, measured as cups per day multiplied by
the number of years of consumption (ccy) [12,23]. PAGE had daily
mg caffeine intake from all caffeine-containing drinks and foods
for 12 months prior to enrollment (1995–1996) and only incident
PD cases diagnosed after 1997 were included in the analysis [24].
Despite the variation in data collection, results were consistent
across studies, corroborating robustness of the interaction between
coffee/caffeine and GRIN2A. We could not, and did not, attempt
to distinguish the bioactive ingredient in caffeinated-coffee.
Although caffeine has been shown to be neuroprotective, there
may be other ingredients in caffeinated-coffee that may affect
disease pathogenesis. To classify coffee/caffeine intake, each
dataset was treated separately according to the measurements
available. The median ccy or mg was determined for controls
within each dataset (excluding those with zero intake) and used as
the cut-off for heavy drinkers (.median) vs. light drinkers (0 to
#median). The median was 67.5 ccy for NGRC, 74.0 ccy for
PEG, 70.0 ccy for HIHG, and 237.8 mg/day for PAGE. For
coffee dose, quartiles were defined for each dataset using the full
range from zero to maximum intake in controls. Results shown for
NGRC, PEG and HIHG are based on lifetime caffeinated-coffee
consumption. Truncating coffee use at age-at-onset or age-at-
diagnosis in patients did not affect the results. To assess the effects
of caffeinated tea and soda, we performed sensitivity analysis in
NGRC dataset. Caffeinated soda and tea were commonly and
equally consumed by heavy and light coffee drinkers (soda: 80% in
both heavy and light drinkers; caffeinated tea: 66% in heavy coffee
drinkers and 61% in light coffee drinkers). We repeated GWAIS
and stratified GWAS with caffeinated soda and tea as covariates.
We also explored association of caffeinated tea and soda with PD
expecting an inverse association if caffeine were the bioactive
ingredient in coffee.
Genotyping
The source of DNA was whole blood for NGRC and HIHG,
saliva for PAGE, and whole blood (all PD and half of controls) or
saliva (half of controls) for PEG. NGRC was genome-wide
genotyped using Illumina HumanOmni1-Quad_v1-0_B array
and achieved 99.92% call rate and 99.99% reproducibility.
GWAS genotyping and statistical quality control (QC) have been
published [5]. 811,597 SNPs (excluding Y chromosome SNPs
because they are not amenable to sex adjustment) passed GWAS
QC and were included in GWAIS. Replication groups genotyped
GRIN2A_rs4998386. Only one SNP was genotyped for replication;
we have no other undisclosed replication results. PEG and HIHG
used ABI TaqMan assay-by-design (C__28018721_20), PAGE
used Sequenom and all achieved call rates of 96%–99%.
Analytic strategy
The first step was to test the hypothesis that the effect of coffee
on PD risk is affected by a gene; ie, test statistical interaction
between SNPs and coffee genome-wide. Theoretically, a test of
SNP*coffee interaction would have been suitable; however, a pure
test of interaction has low power; reportedly, it requires more than
four times the sample size that GWAS would require to detect a
main effect of similar size (reviewed in [2]). We chose the joint test
of SNP main effect and its interaction with coffee as proposed by
Kraft et al [15]. We call the test GWAIS for genome-wide
association and interaction study. The main advantage of the joint test
is that it does test for interaction and it has more power than pure
interaction test when there is a modest SNP marginal effect. Next
we performed stratified GWAS in heavy and light drinkers to gain
insight to where the interaction signal was coming from and to
formulate a hypothesis for replication. We then replicated the top
signal and performed pooled analysis. Methods for meta-analysis
of the joint test are available [25,26]; however, since we had
individual level data we pooled the datasets.
Statistical analysis
Quality control for GWAIS and stratified GWAS in
Discovery (NGRC).
The genome-wide genotypes for NGRC
had been cleaned previously for GWAS using standard rigorous
measures [5]. We had identified two significant principal
components (PC1, PC2) marking Jewish/non-Jewish ancestry and
European countries of origin [5]. Sex was a significant variable,
because PD affects more men than women and our data has a
significant gender disparity (Table S1). Controls were older than
patients at age at onset, which was by design to minimize the
chances that controls were too young to have developed the disease.
Nevertheless, we controlled for age at enrolment both for patients
and controls to avoid confounding by age-related factors. We
examined coffee consumption and the most significant SNP for
potential variation by disease related variables, recruitment sites,
and ethnic and geographic origins of subjects (Table S2). Smoking
was a potential confounder because it is correlated with coffee use
and is an independent inverse risk factor of PD. Thus we repeated
all analyses with smoking included as a covariate in the model
(Table S3 without smoking as covariate, Table S4 with smoking as
covariate). We also repeated analyses with caffeinated tea and soda
in the model (Table S5). For details on how the data on tea, soda
and smoking were collected in NGRC, see [12].
GWAIS in Discovery. 811,597 SNP genotypes [5] and
lifetime caffeinated-coffee consumption data [12] from 1458
persons with PD and 931 controls from NGRC were analyzed.
We tested the following models: [SNP+coffee+SNP*coffee+
covariate vs. coffee+covariate] henceforth referred to as
[SNP+SNP*coffee] joint test [15]. Critically, the main effect of
coffee on PD risk was present in both models being compared thus
we controlled for coffee in the test. This model conducts a 2
degrees of freedom (df) joint test of SNP marginal effect and its
interaction with coffee on PD risk [15]. Sex, age, PC1 and PC2
were included as covariates. We used likelihood ratio test statistics
as implemented in PLINK [27], and tested the Dominant,
Additive and Recessive modes of inheritance. GWAIS was
repeated once with the addition of smoking as a covariate, and
again by addition of caffeinated tea and soda as covariates.
Stratified GWAS in Discovery. There were 512 cases and
387 controls in the heavy coffee drinking group and 946 cases and
544 controls in the light coffee drinking group. We tested
association of 811,597 SNPs with PD in each group using
standard GWAS with 1 df. We used PLINK [27] and adjusted for
age, sex, PC1 and PC2. Stratified GWAS were repeated with
smoking, caffeinated soda and caffeinated tea added as covariates
(Tables S4 and S5).
Replication. Based on the main finding in Discovery, we
specified the replication hypothesis a-priori as follows: ‘‘Among
heavy coffee drinkers, carriers of rs4998386_T allele have a lower
risk of PD than carriers of rs4998386_CC genotype’’. Note that we
were using the GWAIS as a means to identify the genes that might
enhance the inverse association of coffee with PD with the goal of
GRIN2A, Coffee, and Parkinson’s Disease
PLoS Genetics | www.plosgenetics.org 3 August 2011 | Volume 7 | Issue 8 | e1002237
carrying the discovery forward to pharmacogenetic studies.
Hence, the replication hypothesis was framed as specified. We
used three datasets for replication PEG [20], PAGE [21], and
HIHG [9]. We tested between-study heterogeneity using Breslow-
Day test statistics. There was no heterogeneity in coffee use, in
rs4998386_CC or in rs4998386_TC genotype frequencies, but
rs4998386_TT frequency, which is quite rare, varied significantly
across studies. There were a total of 26 cases and 26 controls with
rs4998386_TT genotype in Discovery and Replication combined.
We found no trend in rs4998386_TT subject characteristics that
could help pinpoint the source of heterogeneity (Table S6). Given
the unanticipated heterogeneity in rs4998386_TT, we performed
genotype-specific analysis (comparing TC to CC, excluding TT) as
well as Dominant and Additive models which included TT.
Categorical data were analyzed using logistic regression in SAS
(version 9.2) and were adjusted for age and sex, and for source of
data when data were pooled. Age at onset was analyzed as a
continuous variable using linear regression in SAS.
Significance. P values were two-sided for Discovery, one-sided
for Replication given the clear directional prior hypothesis [28], and
two sided when Discovery and Replication were pooled. There is no
agreed-upon significance threshold for GWAIS. The Bonferroni
corrected threshold for all 811,597 SNPs on the array is
P,6.4610
28
. However, not all 811,597 SNPs are independent
due to linkage disequilibrium (LD). SimpleM [29] provides a sound
Bonferroni-based multiple testing correction method for GWAS
based on the estimated number of independent tests, allowing for
marker-to-marker LD. It was shown to be the best approximation
for permutation, which is computationally prohibitive for GWAS.
Using simpleM we calculated the number of independent SNPs
genome-wide for NGRC as M
eff
= 430,151; thus the Bonferroni
corrected threshold for independent tests was P,1.16610
27
.
Imputation. We used IMPUTE v2 [30] with HapMap and
1000 Genomes genotypes combined as reference data to infer
genotypes for SNPs that were not originally included on the
Illumina OMNI-1 array and thus not genotyped in the NGRC
dataset. 2,710,971 SNPs were imputed with high reliability
(information score $ 0.95) and had MAF.0.01, increasing the
genomic coverage to 3,522,568 SNPs total (genotyped and
imputed). We performed GWAIS and stratified GWAS for the
GRIN2A region (Chromosome 16, 97 Mb–102 Mb) using
genotype probability data (dose 2-0) in R software http://www.
r-project.org/.
Linkage disequilibrium. Linkage disequilibrium and haplo-
type blocks were estimated using the Haploview software [31].
Haplotype analysis was performed using hapstat adjusting for sex
and age [32].
Copy number variations (CNV). We used Golden Helix
SNP Variation Suite version 7.2.3 (http://www.goldenhelix.com/)
and PennCNV [33] to explore for deletions or duplications in the
GRIN2A region. Golden Helix found no CNVs; PennCNV
identified two controls with CNVs, which even if confirmed to
be real, would not affect the results of the study.
Data access. NGRC genome-wide genotypes, phenotype
and environmental data are available at www.ncbi.nlm.nih.gov/
gap study accession number phs000196.v1.p1.
Results
GWAIS in Discovery
The most significant result was the novel appearance, on the
Manhattan plot (Figure 1A, Figure S1), of a block of linked SNPs
which map to the GRIN2A gene on chromosome 16 (Figure S2).
This locus had not been detected in PD GWAS previously because
its main effect is modest. However, when considered in the context
of interaction with coffee, GRIN2A surpassed all known PD-
associated genes in significance including SNCA which has been
the strongest association with PD in GWAS. The signal for known
PD genes were driven only by their main effects with no evidence
for interaction (P
interaction
= 0.5–0.7); whereas the signals for PD-
associated SNPs in GRIN2A were enhanced by SNP*coffee
interaction (P
interaction
,10
23
). The quantile-quantile (QQ) plot
of the expected vs. observed genome-wide P values (Figure 1B) is
also evidence for the impact of GRIN2A on PD risk.
GWAIS results described above were obtained from a test that
measures the combined significance of the SNP and its
interaction with coffee on risk of PD [15] . The test has 2 df;
hence when interaction is absent, GWAIS is less powerful than
GWAS which has only 1 df. Further more, the sample size was
smaller in GWAIS because it required not only genotypes but
also coffee data, which was available for 2/3 of NGRC. Under
these conditions, GWAIS produced P
2df
.10
26
(Figure 1A) for
the top SNP in SNCA which had reached P = 3610
211
in NGRC
GWAS [5]. This drop in significance demonstrates the dramatic
loss of power in GWAIS as compared to GWAS. Under these
conditions, GWAIS yielded P
2df
=1610
26
for rs4998386 in
GRIN2A (as compared to P
2df
=3610
26
for SNCA and
P
2df
=7610
25
for MAPT). Dominant and Additive models
produced nearly identical res ults for GRIN2A SNPs (Table 1).
Recessive model had no notable signal (Figure S1).
GWAS in heavy and light coffee-drinkers in Discovery
samples
With one goal being pharmacogenetic applications, we were
interested in genes that modulate risk in people who consume
caffeine, thus we stratified the subjects as heavy drinkers or light
drinkers (light includes non-drinkers) and performed GWAS in each
group (SNP-PD test, 1 df). The sample size was now further reduced
to only 512 cases and 387 controls who drank more than the median
(heavy drinkers) and 946 cases and 544 control subjects who drank
less than the median (light drinkers). As expected due to interaction,
which suggests different association patterns across categories, most
of the signals seen in GWAIS (Figure 1A) appeared within either
heavy drinkers (Figure 2, Table 2, Figure S1) or light drinkers
(Figure 3, Table 2, Figure S1). In heavy drinkers, the focus of this
study, the most significant result was GRIN2A_rs4998386
(P = 6610
27
) and 11 neighboring SNPs (P = 10
25
to 10
26
, Table 2).
The QQ plots for stratified GWAS also demonstrate clearly that
GRIN2A is the single primary PD associated locus in heavy coffee
drinkers (Figure 2): exclusion of SNCA, HLA and MAPT did not
have an impact in heavy drinkers, whereas exclusion of GRIN2A
nearly abolished the extreme P values of 10
25
–10
26
. No clear
signals were detected in light coffee drinkers (Figure 3).
The 12 GRIN2A SNPs that were associated with PD via heavy
coffee consumption had similar minor allele frequencies
(MAF = 0.13–0.16 in controls) and odds ratios (OR = 0.43–0.51)
and were in strong LD (Figure S3). Haplotype analysis did not
strengthen the signal. Within this gene varying CNV software tools
called either no CNVs or just two CNVs in controls. Thus, CNVs
are unlikely to explain a large fraction of the phenotype variability.
We therefore selected only the SNP with the lowest P value for
replication (GRIN2A_rs4998386).
Genotype-specific association of coffee with PD in
Discovery
Testing the association of coffee with PD in NGRC, when
calculated irrespective of genotype, showed an average of 34%
GRIN2A, Coffee, and Parkinson’s Disease
PLoS Genetics | www.plosgenetics.org 4 August 2011 | Volume 7 | Issue 8 | e1002237
lower PD risk in heavy coffee drinkers than in light drinkers
(OR = 0.66, P = 6610
26
, Table 3, Coffee irrespective of geno-
type). GRIN2A, irrespective of coffee, had a modest main effect on
PD in NGRC (Table 3, GRIN2A rs4998386 genotype irrespective
of coffee). A key question was if, and to what degree, GRIN
2A_rs4998386 genotype modifies the effect of coffee on PD risk
(Table 3): Within heavy drinkers, PD risk was 58% lower
(OR = 0.42, P = 2610
26
) for rs4998386_TC, and 81% lower
(OR = 0.19, P = 0.05) for rs4998386_TT genotype than
rs4998386_CC; whereas in light drinkers genotype had no effect
on risk. Similar results were obtained for Additive and Dominant
models (Table S3). The joint effect comparing rs4998386_TC
genotype and heavy coffee vs. rs4998386_CC genotype and light
coffee was most dramatic, suggesting a highly significant 68% risk
reduction (OR = 0.32, P = 7610
211
) in NGRC (Table 3, Joint
effects of GRIN2A rs4998386 and coffee).
Hypothesis for replication
We used GWAIS as a means to identify genes that might
enhance the inverse association of coffee with PD with the goal of
carrying the discovery forward as a genetic marker for use in
pharmacogenetic studies. Hence, the replication hypothesis was
specified a-priori, based on results of NGRC, as follows: ‘‘Among
heavy coffee drinkers, carriers of rs4998386_T allele have lower
risk of PD than carriers of rs4998386_CC genotype’’. Although
this test does not reflect our most significant results, it is the test
that has the clearest interpretation because it keeps the effect of
coffee constant. For example, comparing TC+heavy vs. CC+light
gave larger effect size and the P value was 3-orders of magnitude
lower than the specified hypothesis, however, unlike our
hypothesis, the test included coffee, which would have made it
difficult to draw firm conclusions about the effect of genotype on
coffee’s inverse association with PD.
Figure 1. Manhattan Plot and QQ Plot of GWAIS. Panel A depicts the Manhattan plot for the GWAIS (joint test of association and interaction
with coffee, 2df, adjusted for sex, age, PC1 and PC2). The novel spike on chromosome 16 corresponds to 12 GRIN2A SNPs that were genotyped.
Imputed SNPs achieved P
2df
,5610
28
(see Table 4). Additive model is shown here, Dominant and Recessive are in Figure S1. Dominant and Additive
models yielded similar results for top hits (see Table 1). Panel B is the QQ plot where the observed P values (red line) are plotted against the expected
P values under no association (straight black line). The plots were made first by including all genotyped SNPs (red), then excluding those in the SNCA,
HLA and MAPT regions (green) and finally by excluding GRIN2A (blue).
doi:10.1371/journal.pgen.1002237.g001
GRIN2A, Coffee, and Parkinson’s Disease
PLoS Genetics | www.plosgenetics.org 5 August 2011 | Volume 7 | Issue 8 | e1002237
Table 1. GRIN2A was the most significant signal in GWAIS.
Dominant Additive
MAF MAF HWE SNP Interaction 2df SNP Interaction 2df
CHR Gene SNP BP
Minor/
Major
Allele Case Control P OR (SE) P OR (SE) P P OR (SE) P OR (SE) P P
Novel PD-association identified via interaction with coffee
16 GRIN2A rs4998386 9978046 T/C 0.08 0.12 0.54 0.82 (0.12) 0.17 0.49 (0.11) 2610
23
1610
26
0.84 (0.11) 0.19 0.50 (0.11) 1610
23
1610
26
16 GRIN2A rs17569693 9987686 G/A 0.08 0.12 0.54 0.79 (0.11) 0.10 0.51 (0.12) 4610
23
1610
26
0.80 (0.10) 0.09 0.54 (0.12) 5610
23
2610
26
16 GRIN2A rs8043728 10003004 T/C 0.09 0.12 0.88 0.82 (0.12) 0.16 0.53 (0.12) 5610
23
8610
26
0.82 (0.11) 0.13 0.57 (0.12) 0.01 8610
26
16 GRIN2A rs8056683 10052710 T/C 0.09 0.13 1.00 0.82 (0.12) 0.17 0.52 (0.12) 3610
23
4610
26
0.83 (0.11) 0.14 0.55 (0.11) 4610
23
4610
26
16 GRIN2A rs9927926 10057405 C/T 0.09 0.13 1.00 0.82 (0.12) 0.17 0.52 (0.12) 3610
23
4610
26
0.83 (0.11) 0.14 0.55 (0.11) 4610
23
4610
26
16 GRIN2A rs17671033 10068727 A/G 0.09 0.13 1.00 0.85 (0.12) 0.24 0.50 (0.11) 2610
23
3610
26
0.85 (0.11) 0.20 0.53 (0.11) 2610
23
4610
26
16 GRIN2A rs9933111 10072100 G/A 0.09 0.13 0.88 0.85 (0.12) 0.24 0.49 (0.11) 2610
23
2610
26
0.85 (0.11) 0.20 0.53 (0.11) 2610
23
2610
26
16 GRIN2A rs13331465 10077968 T/C 0.09 0.13 0.88 0.85 (0.12) 0.23 0.49 (0.11) 2610
23
2610
26
0.85 (0.11) 0.19 0.53 (0.11) 2610
23
2610
26
16 GRIN2A rs13336632 10078155 C/A 0.09 0.13 0.88 0.85 (0.12) 0.23 0.49 (0.11) 2610
23
2610
26
0.85 (0.11) 0.19 0.52 (0.11) 2610
23
2610
26
16 GRIN2A rs1448270 10082819 T/G 0.09 0.13 0.77 0.86 (0.12) 0.29 0.49 (0.11) 2610
23
5610
26
0.86 (0.11) 0.24 0.53 (0.11) 2610
23
5610
26
16 GRIN2A rs11866570 10113676 C/T 0.11 0.15 0.18 0.89 (0.12) 0.35 0.51 (0.11) 1610
23
8610
26
0.92 (0.11) 0.46 0.54 (0.11) 2610
23
3610
25
16 GRIN2A rs1448253 10128367 C/T 0.09 0.13 0.67 0.83 (0.11) 0.18 0.52 (0.12) 3610
23
4610
26
0.84 (0.11) 0.18 0.55 (0.11) 3610
23
5610
26
Well-established PD-associated genes identified via their main effect
4 SNCA rs356220 90860363 T/C 0.43 0.35 0.77 1.37 (0.16) 0.01 1.25 (0.23) 0.23 3610
25
1.36 (0.11) 2610
24
1.05 (0.14) 0.69 3610
26
4 SNCA rs356168 90893454 G/A 0.51 0.44 0.89 1.33 (0.17) 0.03 1.33 (0.26) 0.16 1610
24
1.28 (0.10) 3610
23
1.09 (0.14) 0.51 5610
25
17 MAPT rs199533 42184098 T/C 0.17 0.22 0.15 0.65 (0.08) 2610
24
1.10 (0.21) 0.63 7610
25
0.68 (0.07) 2610
24
1.10 (0.18) 0.57 7610
25
6 HLA rs3129882 32517508 G/A 0.46 0.40 0.78 1.25 (0.15) 0.07 1.26 (0.24) 0.22 2610
23
1.24 (0.10) 0.01 1.08 (0.14) 0.54 5610
24
Also see Figure 1. GWAIS analysis was [SNP+SNP*coffee] test with 2 df adjusting for sex, age, PC1 and PC2. The test examines the significance of the SNP main effect and its interaction with coffee, without introducing the
significant effect of coffee on PD. Results for GRIN2A were equally significant under Dominant and Additive models. Recessive model had no clear signal (see Figure S1). Also shown are the results obtained with the same dataset
and under the same analytic model for the known PD genes SNCA, MAPT and HLA. SNCA and HLA had reached P,5610
28
in our GWAS. The fall in significance in GWAIS is due in part to 1/3 reduction in sample size due to
unavailability of coffee data, and also the penalty imposed by the added degree of freedom. GRIN2A did not have a strong main effect to be noticed in GWAS, but in GWAIS, the inclusion of coffee and interaction placed GRIN2A
higher than SNCA, HLA and MAPT.
doi:10.1371/journal.pgen.1002237.t001
GRIN2A, Coffee, and Parkinson’s Disease
PLoS Genetics | www.plosgenetics.org 6 August 2011 | Volume 7 | Issue 8 | e1002237
Potential confounders
Before attempting re plication, the following analyses were
conducted to identify potential confounders (Table S2). We
tested the frequency of rs4998386 and coffee use acro ss disease-
specific strata and population structure. There was no evidence
for heterogeneity by presence or absence of family history of
PD, age at onset, or recruitment site. rs49 98386 frequency was
different between Ashkenazi-Jewish and non-Jewish individuals
(P = 0.02) and across the European countries of ancestral origin
(P = 3610
23
) in cases, but not in c ontrols, which, PD being
heterogeneous, may indicate differen t ethnic-specific clusters of
disease subtypes as has been noted for LRRK2-associated PD
[34]. Not surprisingly, heavy coffee use was associated with
smoking (P,10
24
), which itself is inversely associated with PD
risk indepen dently of coffee [12]. Ad justing for smoking, in
addition to other covaria tes, did not change the results (Table
S4). We also repeat ed the analyses adjusting for caffeinated
soda and caffeinated tea consumption and found the results to
be robust (Table S5). Some reportssuggestpersonswithPDare
more likely to avoid sensation seeking and addi ctive behaviors
[35] and GRIN2A polymorph isms have been implicated in
predisposition to heroin addiction [36] and smoking [37]
raising the concern that our results could have been
confounded if the GRIN2A SNPs identified here were associated
with habitual coffee drinking. However, there was no evidence
for association between any of t he GRIN2A S NPs and hea vy vs.
light coffee consumption in cases and controls combined
(OR = 0.95–1.01, P = 0.61–0.94).
Figure 2. GWAS in heavy coffee-drinkers. Panel A depicts GWAS in heavy coffee drinkers with GRIN2A achieving the lowest P values. The P
values in stratified GWAS are for genotyped SNP’s main effect on PD risk, adjusted for sex, age, PC1 and PC2. Imputed SNPs (not shown here)
achieved P = 3610
28
(see Table 4). Additive model is shown here; see Figure S1 for Dominant model. Dominant and additive models yielded similar
results for top hits (see Table 2). Panel B is the QQ plot for heavy coffee drinkers where the observed P values (red line) are plotted against the
expected P values under no association (straight black line). The plots were made first by including all SNPs (red), then excluding SNCA, HLA and MAPT
(green) and finally by excluding GRIN2A (blue). Unlike the QQ plot for GWAIS, the effects of SNCA, HLA and MAPT are unnoticeable. The only deviation
is seen at the extreme ,10
25
which is primarily due to GRIN2A.
doi:10.1371/journal.pgen.1002237.g002
GRIN2A, Coffee, and Parkinson’s Disease
PLoS Genetics | www.plosgenetics.org 7 August 2011 | Volume 7 | Issue 8 | e1002237
Table 2. GRIN2A was the most significant result in GWAS in heavy coffee-drinkers.
GWAS in Heavy Coffee Drinkers GWAS in Light Coffee Drinkers
MAF MAF Dominant Additive MAF MAF Dominant Additive
CHR GENE SNP BP
Minor/
Major
Allele Case Control OR (SE) P OR (SE) P Case Control OR (SE) P OR (SE) P
Novel PD-association identified via interaction with coffee
16 GRIN2A rs4998386 9978046 T/C 0.07 0.14 0.41 (0.07) 8610
27
0.43 (0.07) 6610
27
0.09 0.11 0.82 (0.12) 0.18 0.84 (0.11) 0.19
16 GRIN2A rs17569693 9987686 G/A 0.07 0.13 0.41 (0.08) 1610
26
0.44 (0.08) 3610
26
0.09 0.11 0.79 (0.11) 0.10 0.84 (0.11) 0.09
16 GRIN2A rs8043728 10003004 T/C 0.08 0.14 0.41 (0.08) 5610
26
0.44 (0.08) 7610
26
0.09 0.11 0.82 (0.12) 0.17 0.84 (0.11) 0.13
16 GRIN2A rs8056683 10052710 T/C 0.08 0.14 0.43 (0.08) 2610
26
0.44 (0.08) 3610
26
0.10 0.12 0.82 (0.12) 0.17 0.84 (0.11) 0.14
16 GRIN2A rs9927926 10057405 C/T 0.08 0.14 0.43 (0.08) 2610
26
0.44 (0.08) 3610
26
0.10 0.12 0.82 (0.12) 0.17 0.84 (0.11) 0.14
16 GRIN2A rs17671033 10068727 A/G 0.08 0.14 0.43 (0.08) 2610
26
0.44 (0.08) 3610
26
0.10 0.11 0.82 (0.12) 0.25 0.85 (0.11) 0.20
16 GRIN2A rs9933111 10072100 G/A 0.08 0.15 0.42 (0.08) 1610
26
0.45 (0.07) 1610
26
0.10 0.12 0.82 (0.12) 0.24 0.85 (0.11) 0.20
16 GRIN2A rs13331465 10077968 T/C 0.08 0.15 0.42 (0.08) 1610
26
0.45 (0.07) 1610
26
0.10 0.12 0.82 (0.12) 0.24 0.85 (0.11) 0.20
16 GRIN2A rs13336632 10078155 C/A 0.08 0.15 0.42 (0.08) 1610
26
0.45 (0.07) 1610
26
0.10 0.12 0.82 (0.12) 0.24 0.85 (0.11) 0.20
16 GRIN2A rs1448270 10082819 T/G 0.08 0.14 0.43 (0.08) 3610
26
0.47 (0.08) 3610
26
0.10 0.12 0.86 (0.12) 0.29 0.85 (0.11) 0.24
16 GRIN2A rs11866570 10113676 C/T 0.09 0.16 0.46 (0.08) 3610
26
0.51 (0.08) 1610
25
0.12 0.13 0.88 (0.12) 0.35 0.85 (0.11) 0.45
16 GRIN2A rs1448253 10128367 C/T 0.08 0.15 0.44 (0.08) 2610
26
0.47 (0.08) 4610
26
0.10 0.12 0.83 (0.11) 0.18 0.84 (0.11) 0.18
Well-established PD-associated genes identified via their main effect
4 SNCA rs356220 90860363 T/C 0.42 0.34 1.71 (0.25) 2610
24
1.43 (0.15) 6610
24
0.43 0.36 1.37 (0.16) 0.01 1.36 (0.11) 2610
24
4 SNCA rs356168 90893454 G/A 0.50 0.42 1.75 (0.27) 3610
24
1.38 (0.14) 1610
23
0.51 0.45 1.33 (0.17) 0.03 1.28 (0.10) 3610
23
17 MAPT rs199533 42184098 T/C 0.16 0.21 0.71 (0.10) 0.02 0.75 (0.10) 0.02 0.17 0.23 0.65 (0.08) 3610
24
0.69 (0.07) 3610
24
6 HLA rs3129882 32517508 G/A 0.47 0.40 1.57 (0.24) 2610
23
1.34 (0.13) 4610
23
0.45 0.40 1.25 (0.15) 0.07 1.24 (0.10) 0.01
See also Figure 2. Standard GWAS (PD-SNP association, no interaction) was conducted among heavy and light coffee drinkers separately. GRIN2A was most notable only among heavy coffee drinkers. Odds ratio (OR) of 0.41–0.46
suggests that among heavy coffee drinkers, who are known to be at reduced risk for PD, GRIN2A genotypes further modifies risk by over two-fold. As expected due to interaction, GRIN2A did not have a significant effect in light
coffee drinkers. This is in contrast to known PD genes which exhibited their effects on PD risk regardless of coffee consumption.
doi:10.1371/journal.pgen.1002237.t002
GRIN2A, Coffee, and Parkinson’s Disease
PLoS Genetics | www.plosgenetics.org 8 August 2011 | Volume 7 | Issue 8 | e1002237
Replication
See Table 3, Table S3. The a-priori hypothesis for replication
that among heavy drinkers GRIN2A_rs4998386_T carriers had a
lower risk of PD than GRIN2A_rs4998386_CC was replicated
comparing TC to CC (excluding rare heterogeneous TT
genotype): OR = 0.59, P = 10
23
; under Additive model (TT vs.
TC vs. CC): OR = 0.77, P = 0.04; and Dominant model (TT+TC
vs. CC): OR = 0.70, P = 0.01. Note that the Additive and
Dominant models included the TT genotype which is rare and
its frequency varied significantly across datasets (Table 3, Table
S6). The TC vs. CC comparison is more robust for this reason;
Additive and Dominant model are shown for completeness. As
seen in NGRC data, genotype had no effect on risk of PD among
light coffee drinkers in Replication or combined data (OR = 1.0,
P = 0.99).
In pooled Replication (without Discovery), the [SNP+SNP*cof-
fee] joint test yielded P
2df
= 2.3610
23
comparing TC to CC
(excluding rare heterogeneous TT genotype); P
2df
= 0.12 for the
Additive model, P
2df
= 0.02 for the Dominant model. The pooled
analysis of Replication and Discovery with the [SNP+SNP*coffee]
joint test yielded, P
2df
= 1.9610
27
comparing TC to CC
(excluding rare heterogeneous TT genotype), P
2df
= 1.4610
25
for the Additive model, and P
2df
= 8.6610
27
for the Dominant
model.
In pooled data, compared to the light coffee drinkers with
GRIN2A_rs4998386_CC genotype (the group with highest risk),
heavy coffee use (with CC genotype) reduced risk by 18%
(OR = 0.82, P = 3610
23
), having GRIN2A_rs4998386_T allele
(light coffee) had no effect on risk (OR = 1.0, P = 0.99), but the
combination of heavy coffee use and GRIN2A_rs4998386_TC
genotype was associated with a highly significant 59% risk
reduction (OR = 0.41, P = 6610
213
) (Table 3, Joint effects of
GRIN2A rs4998386 and coffee).
Imputation
See Table 4, Table S7. The array used in the study, Illumina
OMNI-1 had nearly a million SNPs, which is a relatively dense
coverage, but which could be further improved by imputing the
Figure 3. GWAS in light coffee-drinkers. Neither the Manhattan plot (Panel A) nor the QQ plot (Panel B) exhibit evidence of association between
GRIN2A and PD among individuals who drink little or no coffee.
doi:10.1371/journal.pgen.1002237.g003
GRIN2A, Coffee, and Parkinson’s Disease
PLoS Genetics | www.plosgenetics.org 9 August 2011 | Volume 7 | Issue 8 | e1002237
Table 3. PD risk conditioned on GRIN2A genotype and coffee use.
GRIN2A
Coffee
NGRC (Discovery) Pooled Replications Pooled NGRC
+
Replications
N
Case
N
Control
OR
(SE) P
N
Case
N
Control
OR
(SE) P
N
Case
N
Control
OR
(SE) P
Coffee irrespective of genotype
- Light 946 544 Ref 621 1012 Ref 1567 1556 Ref
- Heavy 512 387 0.66
(0.06)
6610
26
393 905 0.79
(0.07)
2610
23
905 1292 0.73
(0.04)
3610
27
GRIN2A
rs4998386 genotype irrespective of coffee
CC - 1227 716 Ref 837 1558 Ref 2064 2274 Ref
TC 219 204 0.62
(0.07)
2610
25
163 344 0.89
(0.10)
0.14 382 548 0.75
(0.06)
2610
24
TT 12 11 0.53
(0.23)
0.15 14 15 26 26 Heterogeneity
P = 0.06*
CC Heavy 441 283 Ref 330 706 Ref. 771 989 Ref
TC Heavy 69 99 0.42
(0.08)
2610
26
54 192 0.59
(0.10)
1610
23
123 291 0.51
(0.06)
7610
28
TT Heavy 2 5 0.19
(0.16)
0.05 9 7 11 12 Heterogeneity P = 0.04*
CC Light 786 433 Ref 507 852 Ref. 1293 1285 Ref
TC Light 150 105 0.81
(0.12)
0.16 109 152 1.24
(0.18)
0.93 259 257 1.00
(0.10)
0.99
TT Light 10 6 0.81
(0.44)
0.70 5 8 15 14
Joint effects of
GRIN2A
rs4998386 and coffee
CC Light 786 433 Ref. 507 852 Ref. 1293 1285 Ref
CC Heavy 441 283 0.75
(0.08)
6610
23
330 706 0.88
(0.08)
0.08 771 989 0.82
(0.06)
3610
23
TC Light 150 105 0.81
(0.12)
0.15 109 152 1.24
(0.18)
0.07 259 257 1.00
(0.10)
0.99
TC Heavy 69 99 0.32
(0.06)
7610
211
54 192 0.52
(0.09)
5610
25
123 291 0.41
(0.05)
6610
213
TT Light 10 6 0.81
(0.44)
0.70 5 8 15 14
TT Heavy 2 5 0.14
(0.12)
0.02 9 7 11 12
Interaction of
GRIN2A
rs4998386 genotypes and coffee consumption
1446 920 0.52
(0.12)
4610
23
1000 1902 0.48
(0.11)
5610
24
2446 2822 0.51
(0.08)
3610
25
Genotype specific dose-dependent effect of coffee
CC #25% 334 189 Ref 117 98 Ref. 451 287 Ref
25%–#50% 344 178 1.03
(0.14)
0.84 120 92 1.11
(0.12)
0.70 464 270 1.06
(0.12)
0.61
50%–#75% 366 203 0.91
(0.12)
0.47 91 87 0.83
(0.17)
0.19 457 290 0.89
(0.10)
0.30
.75% 183 146 0.58
(0.09)
3610
24
75 82 0.68
(0.15)
0.04 258 228 0.61
(0.08)
6610
25
TC #25% 69 55 Ref 27 18 Ref. 96 73 Ref
25%–#50% 65 41 1.31
(0.37)
0.34 22 16 0.89
(0.41)
0.40 87 57 1.24
(0.30)
0.36
50%–#75% 59 56 0.71
(0.20)
0.21 16 22 0.40
(0.19)
0.03 75 78 0.63
(0.15)
0.05
.75% 26 52 0.31
(0.10)
2610
24
13 21 0.37
(0.18)
0.02 39 73 0.34
(0.09)
5610
25
TT #25% 6 0 2 2 8 2
25%–#50% 2 3 0 3 2 6
50%–#75% 2 4 3 1 5 5
.75% 2 4 3 1 5 5
GRIN2A, Coffee, and Parkinson’s Disease
PLoS Genetics | www.plosgenetics.org 10 August 2011 | Volume 7 | Issue 8 | e1002237
SNPs that were not on the array using 1000 Genomes and
HapMap data, a practice that has successfully aided many
projects. After QC, we had over 3.5 million imputed and
genotyped SNPs per individual in NGRC, each with information
score $0.95 (measure of imputation certainty), and each passing
standard GWAS QC. Imputation could only be applied to NGRC
(Discovery) because only NGRC had genome-wide data. GWAIS
and GWAS analysis of the GRIN2A region with imputed SNPs
uncovered a block of densely linked SNPs embedded amongst the
genotyped GRIN2A, six of which achieved P
2df
#5610
28
in
GWAIS (Table 4). The interaction term was OR = 0.44,
P=4610
25
(Table 4). In GWAS conducted in heavy coffee
drinkers, 12 SNPs achieved P = 3610
28
to 5610
28
with
OR = 0.41–0.42 (Table 4).
Discussion
In a genome-wide gene-environment study we identified
GRIN2A as a genetic modifier of the inverse association of coffee
with the risk of developing PD. The discovery was made in
NGRC, and replicated in independent data. Risk reduction by
heavy coffee use, which was estimated to be 27% on average, was
genotype-specific and varied according to GRIN2A genotype from
18% (P = 3610
23
) for individuals with rs4998386_CC genotype to
59% (P = 6610
213
) for those with rs4998386_TC genotype. When
coffee intake was categorized in four doses, the dose trend was
more prominent in individuals with rs4998386_T allele than those
with rs4998386_CC genotype, with the 3
rd
and 4
th
quartiles
exhibiting only 11% and 39% risk reduction for rs4998386_CC
carriers, vs. 37% and 66% for rs4998386_T carriers. With
imputation we uncovered a block of GRIN2A SNPs not included
on the genotyping array, which achieved P = 3610
28
to 5610
28
.
We propose GRIN2A as a new modifier gene for PD, and posit that
if coffee-consumption is considered, GRIN2A may prove to be one
of the most important PD-associated genes to have emerged from
genome-wide studies. We base this suggestion on statistics, biology
and the potential for immediate translation to clinical medicine, as
we discuss below.
GRIN2A had not previously been tested as a candidate gene for
PD, and was not detected in PD GWAS which have all been
C genotype was associated with reduced risk consistently across studies. rs4998386_TT frequency varied significantly across studies. The a-priori hypothesis for
replication that among heavy drinkers GRIN2A_rs4998386_T carriers had a lower risk of PD than GRIN2A_rs4998386_CC was replicated under three conditions:
comparing TC to CC (excluding rare and variable TT genotype) shown here, Dominant model (TT+TC vs. CC) and Additive model (TT vs. TC vs CC) shown in Table S3. As
predicted from the Discovery phase, genotype had no effect on risk of PD among light coffee drinkers. The joint effects of genotype and coffee showed a significant
59% drop in PD risk in people who had the rs4998386_TC genotype and were heavy drinkers, but little or no effect in other combinations. A formal interaction test
demonstrated that the effects of coffee and genotype are dependent on each other. By definition, statistical interaction exists if the joint effect of gene (g) and exposure
(e) is significantly different from the product of their individual effects. Interaction OR is the ratio of the OR of disease when g and e are present, divided by the product
of the individual OR; i.e., OR
interaction
=OR
g+e
/(OR
g
6OR
e
). (F) Dose-dependent risk reduction by coffee was clear and strong for rs4998386_TC genotype. Analyses were
repeated with smoking (Table S4) or caffeinated soda/tea (Table S5) as additional covariates, results were unchanged.
*Heterogeneity P: Breslow-Day test statistics to assess between-study heterogeneity conducted for coffee and genotypes and found to be significant only for TT
genotype. Analyses were adjusted for sex and age at interview in each dataset, and also for study in the poole d analyses. Expanded analysis including results for
individual replication data sets are shown in Table S3.
doi:10.1371/journal.pgen.1002237.t003
Table 3. Cont.
Table 4. GWAIS and GWAS results on combined genotyped and imputed data.
SNP BP
GWAIS in all NGRC subjects GWAS In NGRC heavy coffee drinkers
MAF MAF SNP Interaction 2DF MAF MAF OR
P
Minor/
Major
Allele
Impute
Info
Score Case Control OR (SE) P OR (SE) P P Case Control (SE)
16-10105921 10105921 T/C 0.98 0.11 0.15 0.91 (0.11) 0.45 0.44 (0.09) 4610
25
5610
28
0.09 0.17 0.41 (0.07) 3610
28
16-10103787 10103787 G/A 0.98 0.11 0.15 0.91 (0.11) 0.45 0.44 (0.09) 4610
25
5610
28
0.09 0.17 0.41 (0.07) 3610
28
16-10102229 10102229 T/C 0.98 0.11 0.15 0.91 (0.11) 0.45 0.44 (0.09) 4610
25
5610
28
0.09 0.17 0.41 (0.07) 3610
28
16-10102124 10102124 T/C 0.98 0.11 0.15 0.91 (0.11) 0.45 0.44 (0.09) 4610
25
5610
28
0.09 0.17 0.41 (0.07) 3610
28
rs56275045 10108893 A/C 0.99 0.11 0.15 0.91 (0.11) 0.41 0.45 (0.09) 5610
25
5610
28
0.09 0.17 0.42 (0.07) 3610
28
16-10109203 10109203 A/T 0.99 0.11 0.15 0.91 (0.11) 0.41 0.45 (0.09) 5610
25
5610
28
0.09 0.17 0.42 (0.07) 4610
28
16-10110896 10110896 C/T 0.98 0.11 0.15 0.91 (0.11) 0.41 0.45 (0.09) 6610
25
6610
28
0.09 0.17 0.42 (0.07) 4610
28
16-10101465 10101465 A/G 0.98 0.11 0.15 0.91 (0.11) 0.42 0.45 (0.09) 6610
25
7610
28
0.09 0.17 0.41 (0.07) 5610
28
16-10092692 10092692 T/C 0.98 0.11 0.15 0.91 (0.11) 0.40 0.46 (0.09) 7610
25
8610
28
0.09 0.17 0.42 (0.07) 5610
28
16-10093997 10093997 T/C 0.98 0.11 0.15 0.91 (0.11) 0.39 0.46 (0.09) 7610
25
8610
28
0.09 0.17 0.42 (0.07) 5610
28
16-10094528 10094528 G/A 0.98 0.11 0.15 0.91 (0.11) 0.39 0.46 (0.09) 7610
25
8610
28
0.09 0.17 0.42 (0.07) 5610
28
rs17671178 10094708 G/A 0.98 0.11 0.15 0.90 (0.11) 0.39 0.46 (0.09) 7610
25
8610
28
0.09 0.17 0.42 (0.07) 5610
28
GRIN2A was the most significant area in both the GWA IS and the GWAS in heavy coffee users. This table shows results that achieved P#5610
28
; for a comp lete list of all
SNPs that achieved P,10
25
see Table S7.
doi:10.1371/journal.pgen.1002237.t004
GRIN2A, Coffee, and Parkinson’s Disease
PLoS Genetics | www.plosgenetics.org 11 August 2011 | Volume 7 | Issue 8 | e1002237
examining gene main effects without considering interactions with
relevant environmental exposures. The most significant and
consistently replicated main effects detected to date are for SNCA,
MAPT and HLA. Here we added, for the first time, a common and
relevant environmental exposure (coffee) to a genome-wide study.
Inclusion of coffee allowed GRIN2A to rise to the top. In the gene-
environment (GWAIS) model, GRIN2A surpassed SNCA, MAPT
and HLA in statistical significance. Among heavy coffee drinkers,
the impact of GRIN2A on PD risk (measured as OR) was 50%
greater, and 2 to 5 orders of magnitude more significant (measured
as P value) than the strongest associations reported for SNCA,
MAPT or HLA. This study is proof of concept that inclusion of
environmental factors can help identify disease-associated genes
that are missed in SNP-only GWAS.
GRIN2A is an important gene for the central nervous system.
Accelerated evolution of GRIN2A in primates is said to have
contributed to the dramatic increase in the size and complexity of
the human brain which defines human evolution [38]. GRIN2A
encodes a subunit of the N-methyl-D-aspartate-2A (NMDA)
glutamate receptor. It is central to excitatory neurotransmission
and the control of movement and behavior [39–41]. The literature
suggest imbalances in NMDA-dependent neurotransmission
contribute to neurodegeneration in PD, possibly through massive
influx of calcium and impaired mitochondrial function leading to
apoptosis; and/or disruption of glutamate-mediated autophagy
which is implicated in degradation and removal of proteins like a-
synuclein (see [42] for review). The portion of intron 3 containing
SNPs with the most significant associations (from base pair
9978046 to base pair 10128367, Table 1, Table 2, and Table 4)
includes numerous transcription factor binding sites and two peaks
of enhanced histone H3K4 mono-methylation (http://genome.
ucsc.edu) [43]. Polymorphisms throughout this region could
therefore disrupt regulatory elements, potentially leading to
variation in levels of GRIN2A transcript. GRIN2A is expressed at
high levels in the brain, most notably in the subthalamic nucleus
(STN) [44]. Pharmacologic inhibition of STN with an NMDA-
antagonist reduces nigral neuron loss in a rodent model of PD
[45]. Deep-brain-stimulation, which also targets STN, is an
effective surgical symptomatic therapy for PD.
The other piece of this finding is coffee/caffeine. Our study was
not designed to distinguish the active ingredients in coffee.
However, we note that the largest replication study (PAGE)
measured specifically the caffeine intake in mg from all food sources
(drink, food, and chocolate) and replicated our hypothesis and
interaction robustly. We also found trends for inverse association of
tea and soda with PD, and interestingly, the varied effect size and
strength of association was consistent with the relative amount of
caffeine in each drink (Table S5). Thus, our data are consistent with
experimental observations that caffeine is neuroprotective. Caffeine
is an adenosine A
2A
-receptor antagonist. A
2A
-receptor enhances
calcium influx via NMDA [41] and A
2A
-receptor antagonists are
neuroprotective in animal models of PD; they attenuate excitotox-
icity by reducing extracellular glutamate levels in the striatum
[46,47]. Thus interaction between coffee/caffeine and GRIN2A is
biologically plausible, and can help formulate testable hypotheses
towards a better understanding of the disease pathogenesis.
GRIN2A genotyping may be useful for pharmacogenetic studies.
Genetics has not yet entered drug development for PD but the
time is here. We now have several susceptibility loci (SNCA, MAPT,
HLA, BST1, PARK16, GAK [5–10]) that can help identify
individuals who are at moderately increased risk of developing
PD. We also have at least one neuroprotective compound (coffee/
caffeine) which can be pharmacologically modified to alleviate its
undesirable side effects. GRIN2A genotyping might also inform
treatments for people who already have PD. L-DOPA, the
primary PD drug for 40 years, does not slow disease progression
and has serious side effects. Clinical trials for new PD drugs have
not found drugs that surpass the symptomatic benefits of L-
DOPA. There have been numerous drug trials for glutamate-
receptor blockers as well as for selective A
2A
-receptor antagonists.
Most were shown to be safe, well tolerated and beneficial
[17,18,48]; however, the majority did not reach the regulatory
threshold for efficacy to be approved as PD drugs. We wonder if
some of these clinical trials will succeed if patients are subdivided
by GRIN2A genotype. We acknowledge the distinction that the
present study examined risk of developing PD; whereas clinical
trials have thus far aimed for symptomatic improvements in
patients. Nonetheless, there are sufficient parallels to suggest that
GRIN2A genotype might also influence efficacy of glutamate-
receptor antagonists and A
2A
-receptor antagonists. This is a simple
and inexpensive hypothesis that can be tested in future, ongoing
and even closed clinical trials that have banked DNA.
Common non-coding variants in GRIN2A have been associated
with Huntington disease (HD) [49,50] and schizophrenia [51], and
rare mutations have been described in patients with neurodevel-
opmental phenotypes [52]. Schizophrenia is associated with a
(GT)n repeat in the GRIN2A promoter that may increase disease
risk by suppressing gene expression [51]. Three GRIN2A SNPs
have been associated with onset-age of HD; they are conserved
and reportedly tag a binding site for CCAAT/enhancer-binding
protein [49,50]. HD and PD are both neurodegenerative
movement disorders, thus the possibility of a common genetic
element was of interest. The reported HD-associated GRIN2A
SNPs, rs1969060, rs8057394 and rs2650427, were not on the
genotyping array but were imputed with high fidelity (information
score .0.99). They map within the 150 kb region identified here
for PD, they are in strong LD with PD-associated SNPs defined by
D’ (0.48–1.0) but not by r
2
(0–0.33) (Figure S4). We tested the HD
SNPs for association with onset age and risk of PD in NGRC while
conditioning on the neighboring top PD SNP (rs4998386). One
HD SNP, rs8057394, yielded OR = 0.85, P = 0.02 for PD overall;
OR = 0.79, P = 0.04 for heavy coffee drinkers; and OR = 0.90,
P = 0.24 for light coffee drinkers. We found no other evidence for
association of HD SNPs with PD, including when we jointly tested
HD SNPs and possible interaction with coffee [SNP+SNP*coffee]
on risk or onset of PD. Conversely, we retested, in NGRC, the
association of top genotyped PD SNP (rs4998386) with PD,
conditioning on HD SNP (rs8057394) and found it to be robust
(P
2df
=8610
26
).
Unlike GWAS, which is now a fully standardized practice, there
is no established protocol for testing gene*environment interaction
on a whole-genome scale. Our strategy of starting with the joint
test (GWAIS) and following up with GWAS in subgroups stratified
by exposure was driven by the aims of our study. In Table S7 we
present a side-by-side comparison of the results for the top
GRIN2A SNPs (P,10
25
), when analyzed for main effect (GWAS),
for interaction, with Kraft’s joint test, and in GWAS stratified by
exposure. Amassing a large enough sample size for GWAIS is
challenging. GWAIS requires larger sample sizes than GWAS yet
there exist fewer samples that have data on relevant environmental
exposures in addition to DNA and phenotype. To our knowledge,
NGRC is the largest genetic study of PD that has collected
exposure data. No other publically available PD GWAS has coffee
data, eliminating the possibility of in-silico replication. We were
able to identify and get access to only 3 datasets that had DNA and
coffee, giving us a total sample size of 393 cases and 905 controls
to replicate the GRIN2A effect in heavy coffee drinkers. In contrast,
replications and meta-analyses for gene-only GWAS now have
GRIN2A, Coffee, and Parkinson’s Disease
PLoS Genetics | www.plosgenetics.org 12 August 2011 | Volume 7 | Issue 8 | e1002237
over 17,000 PD cases and controls [10]. We detected the known
and confirmed PD-associated genes (SNCA, MAPT and HLA)in
GWAIS but at much lower significance levels than in GWAS
because of the smaller sample size with coffee data and the added
degree of freedom in GWAIS. It is noteworthy, however, that at
P
2df
=10
26
, GRIN2A surpassed all known PD loci in significance.
With the aid of imputation, we achieved P = 3610
28
for a 2.4-fold
difference in genotype specific effect of coffee on risk of PD.
Importantly, we were able to replicate the hypothesis that we set
out a-priori based on discovery.
Supporting Information
Figure S1 GWAIS and stratified GWAS for Dominant and
Recessive Models. Panel A is the Manhattan Plot of GWAIS for
the Dominant model, and Panel B is for the Recessive model.
Additive model is shown in Figure 1. We tested 811,597 SNPs in
combination with coffee consumption for association with PD.
The model was [SNP+SNP*coffee] test with 2 df, adjusted for sex,
age, PC1 and PC2. Dominant and Additive models yielded similar
results for the top hits (see Table 1 of main text). Panel C shows the
GWAS in heavy coffee drinkers and Panel D is GWAS in light
coffee drinkers, both for the Dominant model. Additive model is
shown in Figure 2 (heavy drinkers) and Figure 3 (light drinkers).
The P values in stratified GWAS (Panels C and D) are for SNP
main effect on PD risk, adjusted for sex, age, PC1 and PC2.
Dominant and Additive models yielded similar results for top hits
(see Table 2 in main text). Genotyped SNPs only (imputed SNPs
not included).
(PDF)
Figure S2 Map of GRIN2A. Panel A: Chromosomal location and
gene structure of GRIN2A. Numbers 1–14 denote exons. Panels B
and C: LD map of all the genotyped SNPs that are located in the
GRIN2A gene or within 50 kb upstream or downstream of the
gene. LD is measured as r
2
(shades of grey) in Panel B and as D’
(shades of red) in Panel C. The intensity of the color depicts
strength of LD and the numbers in the grids are the values of r
2
and D’ in percentage.
(PDF)
Figure S3 LD among the PD-associated SNPs. SNPs marked in
red boxes were genotyped and achieved P,10
25
in either 2 df
GWAIS or GWAS in heavy coffee drinkers. SNPs not in red boxes
were imputed and achieved P#5610
28
in either 2 df GWAIS or
GWAS in heavy coffee drinkers. Panel A is r2, Panel B is D’.
(PDF)
Figure S4 LD among the PD-associated and HD-associated
SNPs. SNPs marked in red boxes were genotyped and achieved
P,10
25
in either 2 df GWAIS or GWAS in heavy coffee drinkers
with PD. SNPs not in boxes were imputed and achieved
P#5610
28
in either 2 df GWAIS or GWAS in heavy coffee
drinkers. SNPs in blue boxes are reported as being associated with
HD [50,51]. Panel A is r2, Panel B is D’.
(PDF)
Table S1 Summary statistics on subject characteristics. NA: not
available * Non-smoker: ,100 cigarettes in lifetime. Smoker:
$100 cigarettes.
(DOC)
Table S2 Frequency of GRIN2A rs4998386_T and heavy coffee
use in NGRC by disease strata and population structure. Family
history: Patients who had at least one first or second degree
relative with PD were classified as familial. All others were
classified as non-familial (sporadic) Age at onset: The higher coffee
consumption in late-onset PD (.50 years) is because they are older
than patients who have early-onset PD and therefore have had
higher cumulative lifetime coffee use over the years. Smoking:
having smoked $100 cigarettes in the lifetime qualified as smoker
(a standard criterion the literature). Coffee: Number of cups of
caffeinated coffee drank per day multiplied by the number of years
of consumption (ccy); heavy and light divided at the median in
controls (67.5 ccy). Jewish/Non-Jewish clusters: The core of the
Jewish cluster was defined within 0.04#PC1#0.055 and
0.001#PC2#0.013. A core within non-Jewish Caucasian cluster
was defined within 20.0075#PC1#0.0025 & 20.005#PC2
#0.003. See Hamza et al. [5]. Recruitment site: US states where
subjects were recruited from. Paternal & maternal ancestry:
Subjects whose both paternal and maternal ancestors came from
the same country. Paternal or maternal ancestry: Since having
only one lineage tracing back to a country was sufficient for this
classification, an individual may fall in more than one group.
* Adjusted for age.
(DOC)
Table S3 PD risk conditioned on GRIN2A genotype and coffee
use (Expanded version of Table 3). (A) Heavy coffee use was
associated with 27% risk reduction (1-OR) in the pooled data. (B)
GRIN2A rs4998386_TC genotype was associated with reduced risk
consistently across studies. rs4998386_TT frequency varied signif-
icantly across studies. (C) The a-priori hypothesis for replication that
among heavy drinkers GRIN2A_rs4998386_T carriers had a lower
risk of PD than GRIN2A_rs4998386_CC was replicated under three
conditions: comparing TC to CC (excluding rare and variable TT
genotype), Dominant model (TT+TC vs. CC) and Additive model
(TT vs. TC vs CC). As predicted from Discovery phase, genotype
had no effect on risk of PD among light coffee drinkers. (D) The
joint effects of genotype and coffee showed a significant 59% drop in
PD risk in people who had the rs4998386_TC genotype and were
heavy drinkers, but little or no effect in other combinations. (E) A
formal interaction test demonstrated that effects of coffee and
genotype are dependent on each other. By definition, statistical
interaction exists if the joint effect of gene (G) and exposure (E) is
significantly different from the product of their individual effects.
Interaction OR is the ratio of the OR of disease when g and e are
present, divided by the product of the individual OR; i.e.,
OR
interaction
=OR
g+e
/(OR
g
6OR
e
). (F) Dose-dependent risk reduc-
tion by coffee was clear and strong for rs4998386_TC genotype.
Analyses were repeated with smoking added as covariate, results
were unchanged (Table S4). OR: odds ratio. P: statistical
significance, two sided for NGRC and pooled analysis, one-sided
for replication studies. *Heterogeneity P: Breslow-Day test statistics
to assess between-study heterogeneity conducted for coffee and
genotypes and found to be significant only for TT genotype.
Analyses were adjusted for sex and age at interview in each dataset,
and also for study in the pooled analyses.
(DOC)
Table S4 Smoking does not alter the results. GWAIS adjusted
for smoking, sex, age, PC1, PC2 for [SNP+SNP*coffee] model
gave P
2df
=2610
26
,P
interaction
=10
23
. GWAS in heavy coffee-
drinkers yielded OR = 0.44, P = 10
26
. PD risk conditioned on
GRIN2A_rs4998386 genotype and coffee use, adjusted for smoking
as well as sex and age are given in Table S4.
(DOC)
Table S5 Caffeinated soda and tea do not alter the results.
GWAIS and stratified GWAS results were robust when caffeinated
tea and soda were included as additional covariates, along with
sex, age, PC1 and PC2.
(DOC)
GRIN2A, Coffee, and Parkinson’s Disease
PLoS Genetics | www.plosgenetics.org 13 August 2011 | Volume 7 | Issue 8 | e1002237
Table S6 Exploring characteristics of individuals with the rare
TT genotype in search of the source of heterogeneity. There was
no trend for any of the PD-relevant characteristics that would
explain the heterogeneity in TT genotype across studies. Due to
the low frequency of TT and the small number of subjects of
Jewish heritage, the N = 0 for the Jewish subgroup is expected.
(DOC)
Table S7 Side-by-side comparison of results from GWAS,
Interaction, GWAIS and stratified GWAS analyses for the top
GRIN2A SNPs. Genotyped and imputed SNPs (info score $95%)
that reached P,10
25
in GWAIS are shown in the order of base
pair position (BP).
(DOC)
Acknowledgments
We thank the persons with Parkinson’s disease and the volunteers who
participate in research. We acknowledge Kimberley Doheny and the team
at the Center for Inherited Disease for their significant contribution to
generating the genotype data. We acknowledge the team at Golden Helix
for their services.
Author Contributions
Conceived and designed the experiments: HP. Performed the experiments:
HP THH EMH-B. Analyzed the data: HP THH EMH-B PS ME.
Contributed reagents/materials/analysis tools: HP SAF JN CPZ PA AS
JWR HC SLR YB YP WKS LW JG JMV BR. Wrote the paper: HP. DNA
and phenotype preparations and database operations: JM DY DMK VIK.
Consulted on choice of analytical method: AT KSK S-AB.
References
1. Manolio TA, Collins FS, Cox NJ, Goldstein DB, Hindorff LA, et al. (2009)
Finding the missing heritability of complex diseases. Nature 461: 747–753.
2. Thomas D (2010) Gene–environment-wide association studies: emerging
approaches. Nat Rev Genet 11: 259–272.
3. Dorsey ER, Constantinescu R, Thompson JP, Biglan KM, Holloway RG, et al.
(2007) Projected number of people with Parkinson disease in the most populous
nations, 2005 through 2030. Neurology 68: 384–386.
4. Hardy J (2010) Genetic analysis of pathways to Parkinson disease. Neuron 68:
201–206.
5. Hamza TH, Zabetian CP, Tenesa A, Laederach A, Montimurro J, et al. (2010)
Common genetic variation in the HLA region is associated with late-onset
sporadic Parkinson’s disease. Nat Genet 42: 781–785.
6. Satake W, Nakabayashi Y, Mizuta I, Hirota Y, Ito C, et al. (2009) Genome-wide
association study identifies common variants at four loci as genetic risk factors for
Parkinson’s disease. Nat Genet 41: 1303–1307.
7. Simon-Sanchez J, Schulte C, Bras JM, Sharma M, Gibbs JR, et al. (2009)
Genome-wide association study reveals genetic risk underlying Parkinson’s
disease. Nat Genet 41: 1308–1312.
8. Pankrat z N, Wilk JB, Latourelle JC, DeStefano AL, Halter C, et al. (2009)
Genomewide association study for susceptibility genes contributing to familial
Parkinson disease. Hum Genet 124: 593–605.
9. Edwards TL, Scott WK, Almonte C, Burt A, Powell EH, et al. (2010) Genome-
Wide Association Study Confirms SNPs in SNCA and the MAPT Region as
Common Risk Factors for Parkinson Disease. Ann Hum Genet 74: 97–109.
10. Nalls MA, Plagnol V, Hernandez DG, Sharma M, Sheerin UM, et al. (2011)
Imputation of sequence variants for identification of genetic risks for Parkinson’s
disease: a meta-analysis of genome-wide association studies. Lancet 377:
641–649.
11. Hernan MA, Takkouche B, Caamano-Isorna F, Gestal-Otero JJ (2002) A meta-
analysis of coffee drinking, cigarette smoking, and the risk of Parkinson’s disease.
Ann Neurol 52: 276–284.
12. Pow ers K, Kay D, Factor S, Zabetian C, Higgins D, et al. (2008) Combined
effects of smoking, coffee and NSAIDs on Parkinson’s disease risk. Mov Disord
23: 88–95.
13. Costello S, Cockburn M, Bronstein J, Zhang X, Ritz B (2009) Parkinson’s
disease and residential exposure to maneb and paraquat from agricul tural
applications in the central valley of California. Am J Epidemiol 169: 919–926.
14. McCulloch CC, Kay DM, Factor SA, Samii A, Nutt JG, et al. (2008) Exploring
gene-environment interactions in Parkinson’s disease. Hum Genet 123:
257–265.
15. Kraft P, Yen YC, Stram DO, Morrison J, Gauderman WJ (2007) Exploiting
gene-environment interaction to detect genetic associations. Hum Hered 63:
111–119.
16. Chen JF, Xu K, Petzer JP, Staal R, Xu YH, et al. (2001) Neuroprotection by
caffeine and A(2A) adenosine receptor inactivation in a model of Parkinson’s
disease. J Neurosci 21: RC143.
17. LeWitt PA, Guttman M, Tetrud JW, Tuite PJ, Mori A, et al. (2008) Adenosine
A2A receptor antagonist istradefylline (KW-6002) reduces ‘‘off’’ time in
Parkinson’s disease: a double-blind, randomized, multicenter clinical trial
(6002-US-005). Ann Neurol 63: 295–302.
18. Fernandez HH, Greeley DR, Zweig RM, Wojcieszek J, Mori A, et al. (2010)
Istradefylline as monotherapy for Parkinson disease: results of the 6002-US-051
trial. Parkinsonism Relat Disord 16: 16–20.
19. Simon-Sanchez J, Scholz S, Matarin Mdel M, Fung HC, Hernandez D, et al.
(2008) Genomewide SNP assay reveals mutations underlying Parkinson disease.
Hum Mutat 29: 315–322.
20. Ritz BR, Manthripragada AD, Costello S, Lincoln SJ, Farrer MJ, et al. (2009)
Dopamine transporter genetic variants and pesticides in Parkinson’s disease.
Environ Health Perspect 117: 964–969.
21. Chen H, Huang X, Guo X, Mailman RB, Park Y, et al. (2010) Smoking
duration, intensity, and risk of Parkinson disease. Neurology 74: 878–884.
22. Hughes AJ, Daniel SE, Ben-Shlomo Y, Lees AJ (2002) The accuracy of diagnosis
of parkinsonian syndromes in a specialist movement disorder service. Brain 125:
861–870.
23. Hancock DB, Martin ER, Stajich JM, Jewett R, Stacy MA, et al. (2007)
Smoking, caffeine, and nonste roidal anti-inflammatory drugs in families with
Parkinson disease. Arch Neurol 64: 576–580.
24. Thompson FE, Kipnis V, Midthune D, Freedman LS, Carroll RJ, et al. (2008)
Performance of a food-frequency questionnaire in the US NIH-AARP (National
Institutes of Health-American Association of Retired Persons) Diet and Health
Study. Public Health Nutr 11: 183–195.
25. Manning AK, LaValley M, Liu CT, Rice K, An P, et al. (2010) Meta-analysis of
gene-environment interaction: joint estimation of SNP and SNP6environment
regression coefficients. Genet Epidemiol 35: 11–18.
26. Aschard H, Hancock DB, London SJ, Kraft P (2011) Genome-wide meta-
analysis of joint tests for genetic and gene-environment interaction effects. Hum
Hered 70: 292–300.
27. Purcell S, Neale B, Todd-Brown K, Thomas L, Ferreira MA, et al. (2007)
PLINK: a tool set for whole-genome association and population-based linkage
analyses. Am J Hum Genet 81: 559–575.
28. Neter J, Kunter M, Nachtsheim C, Wasserman W (1996) Applied Linear
Statistical Models. . pp 52–53.
29. Gao X, Becker LC, Becker DM, Starmer JD, Province MA (2010) Avoiding the
high Bonferroni penalty in gen ome-wide association studies. Genet Epidemiol
34: 100–105.
30. Howie BN, Donnelly P, Marchini J (2009) A flexible and accurate genotype
imputation method for the next generation of genome-wide association studies.
PLoS Genet 5: e1000529. doi:10.1371/journal.pgen.1000529.
31. Barrett JC, Fry B, Maller J, Daly MJ (2005) Haploview: analysis and
visualization of LD and haplotype maps. Bioinformatics 21: 263–265.
32. Lin DY, Huang BE (2007) The use of inferred haplotypes in downstream
analyses. Am J Hum Genet 80: 577–579.
33. Wang K, Li M, Hadley D, Liu R, Glessner J, et al. (2007) PennCNV: an
integrated hidden Markov model designed for high-resolution copy number
variation detection in whole-genome SNP genotyping data. Genome Res 17:
1665–1674.
34. Zabetian CP, Hutter CM, Yearout D, Lopez AN, Factor SA, et al. (2006)
LRRK2 G2019S in families with Parkinson disease who originated from Europe
and the Middle East: evidence of two distinct founding events beginning two
millennia ago. Am J Hum Genet 79: 752–758.
35. Evans AH, Lawrence AD, Potts J, MacGregor L, Katzenschlager R, et al. (2006)
Relationship between impulsive sensation seeking traits, smoking, alcohol and
caffeine intake, and Parkinson’s disease. J Neurol Neurosurg Psychiatry 77:
317–321.
36. Levran O, Londono D, O’Hara K, Randesi M, Rotrosen J, et al. (2009) Heroin
addiction in African Americans: a hypothesis-driven association study. Genes
Brain Behav 8: 531–540.
37. Vink JM, Smit AB, de Geus EJ, Sullivan P, Willemsen G, et al. (2009) Genome-
wide association study of smoking initiation and current smoking. Am J Hum
Genet 84: 367–379.
38. Dorus S, Vallender EJ, Evans PD, Anderson JR, Gilbert SL, et al. (2004)
Accelerated evolution of nervous system genes in the origin of Homo sapiens.
Cell 119: 1027–1040.
39. Shen W, Flajolet M, Greeng ard P, Surm eie r DJ (2008) Di ch otomous
dopaminergic control of striatal synaptic plasticity. Science 321: 848–851.
40. Jin X, Costa RM (2010) Start/stop signals emerge in nigrostriatal circuits during
sequence learning. Nature 466: 457–462.
41. Higley MJ, Sabatini BL (2010) Competitive regulation of synaptic Ca2+ influx
by D2 dopamine and A2A aden osine receptors. Nat Neurosci 13: 958–966.
42. Caudle WM, Zhang J (2009) Glutamate, excitotoxicity, and programmed cell
death in Parkinson disease. Exp Neurol 220: 230–233.
GRIN2A, Coffee, and Parkinson’s Disease
PLoS Genetics | www.plosgenetics.org 14 August 2011 | Volume 7 | Issue 8 | e1002237
43. Birney E, Stamatoyannopoulos JA, Dutta A, Guigo R, Gingeras TR, et al.
(2007) Identification and analysis of functional elements in 1% of the human
genome by the ENCODE pilot project. Nature 447: 799–816.
44. Su AI, Wiltshire T, Batalov S, Lapp H, Ching KA, et al. (2004) A gene atlas of
the mouse and human protein-encod ing transcriptomes. Proc Natl Acad
Sci U S A 101: 6062–6067.
45. Blandini F, Nappi G, Greenamyre JT (2001) Subthalamic infusion of an NMDA
antagonist prevents basal ganglia metabolic changes and nigral degeneration in a
rodent model of Parkinson’s disease. Ann Neurol 49: 525–529.
46. Schwarzschild MA, Xu K, Oztas E, Petzer JP, Castagnoli K, et al. (2003)
Neuroprotection by caffeine and more specific A2A receptor antagonists in
animal models of Parkinson’s disease. Neurology 61: S55–61.
47. Tebano MT, Pintor A, Frank C, Domenici MR, Martire A, et al. (2004)
Adenosine A2A receptor blockade differentially influences excitotoxic mecha-
nisms at pre- and postsy naptic sites in the rat striatum. J Neurosci Res 77:
100–107.
48. Shoulson I, Penney J, McDermott M, Schwid S, Kayson E, et al. (2001) A
randomized, controlled trial of remacemide for motor fluctuations in Parkinson’s
disease. Neurology 56: 455–462.
49. Arning L, Saft C, Wieczorek S, Andrich J, Kraus PH, et al. (2007) NR2A and
NR2B receptor gene variations modify age at onset in Huntington disease in a
sex-specific manner. Hum Genet 122: 175–182.
50. Andresen JM, Gayan J, Cherny SS, Brocklebank D, Alkorta-Aranburu G, et al.
(2007) Replication of twelve association studies for Huntington’s disease residual
age of onset in large Venezuelan kindreds. J Med Genet 44: 44–50.
51. Itokawa M, Yamada K, Yoshitsugu K, Toyota T, Suga T, et al. (2003) A
microsatellite repeat in the promoter of the N-methyl-D-aspartate receptor 2A
subunit (GRIN2A) gene suppresses transcriptional activity and correlates with
chronic outcome in schizophrenia. Pharmacogenetics 13: 271–278.
52. Endele S, Rosenberger G, Geider K, Popp B, Tamer C, et al. Mutations in
GRIN2A and GRIN2B encoding regulatory subunits of NMDA recept ors cause
variable neurodevelopmental phenotypes. Nat Genet 42: 1021–1026.
GRIN2A, Coffee, and Parkinson’s Disease
PLoS Genetics | www.plosgenetics.org 15 August 2011 | Volume 7 | Issue 8 | e1002237
    • "To date, GEWIS have been used to study the etiology of a number of neurodevelopmental and neurological phenotypes. For example, it has been used to examine the effect of genetic moderators of the effect of coffee drinking on Parkinson's Disease (Hamza et al., 2011). Although this innovative approach is currently one of the many long-term aspirations for policymakers in the psychiatric genetics community (Psychiatric GWAS Consortium Steering Committee, 2009), it has yet to be applied to schizophrenia research. "
    Full-text · Dataset · Jan 2015 · BMC Genomics
    • "Excluding external factors that influence internal biological processes generates an incomplete system at best, likely an inaccurate understanding of the interactions between environment and genetic makeup, and from a practical standpoint, misses an opportunity to identify modifiable factors that influence health. This report details the design and conduct of a discovery-based pilot study that accounts for (1) the known genetic uniqueness of individual humans (Olson 2012), (2) the intra-individual variability in homeostatic measurements (Williams 1956; Illig et al. 2010; Suhre et al. 2011), and (3) the challenge of characterizing complex phenotypes resulting from small contributions of many genetic and environmental factors (Goldstein 2009). The participants in the Delta Vitamin Obesity intervention study were children and teens (age 6–14) enrolled in a summer day camp that was a component of a community-based participatory research (CBPR) program. "
    [Show abstract] [Hide abstract] ABSTRACT: The discovery of vitamins and clarification of their role in preventing frank essential nutrient deficiencies occurred in the early 1900s. Much vitamin research has understandably focused on public health and the effects of single nutrients to alleviate acute conditions. The physiological processes for maintaining health, however, are complex systems that depend upon interactions between multiple nutrients, environmental factors, and genetic makeup. To analyze the relationship between these factors and nutritional health, data were obtained from an observational, community-based participatory research program of children and teens (age 6-14) enrolled in a summer day camp in the Delta region of Arkansas. Assessments of erythrocyte S-adenosylmethionine (SAM) and S-adenosylhomocysteine (SAH), plasma homocysteine (Hcy) and 6 organic micronutrients (retinol, 25-hydroxy vitamin D3, pyridoxal, thiamin, riboflavin, and vitamin E), and 1,129 plasma proteins were performed at 3 time points in each of 2 years. Genetic makeup was analyzed with 1 M SNP genotyping arrays, and nutrient status was assessed with 24-h dietary intake questionnaires. A pattern of metabolites (met-PC1) that included the ratio of erythrocyte SAM/SAH, Hcy, and 5 vitamins were identified by principal component analysis. Met-PC1 levels were significantly associated with (1) single-nucleotide polymorphisms, (2) levels of plasma proteins, and (3) multilocus genotypes coding for gastrointestinal and immune functions, as identified in a global network of metabolic/protein-protein interactions. Subsequent mining of data from curated pathway, network, and genome-wide association studies identified genetic and functional relationships that may be explained by gene-nutrient interactions. The systems nutrition strategy described here has thus associated a multivariate metabolite pattern in blood with genes involved in immune and gastrointestinal functions.
    Full-text · Article · Jul 2014
    • "Idiopathic PD involves complex interactions between the genome and environmental exposures [6,7,17,18]. It is operationally assumed that the same set of susceptibility genes predispose to Familial and Sporadic-PD. "
    [Show abstract] [Hide abstract] ABSTRACT: Parkinson's disease (PD) is complex and heterogeneous. The numerous susceptibility loci that have been identified reaffirm the complexity of PD but do not fully explain it; e.g., it is not known if any given PD susceptibility gene is associated with all PD or a disease subtype. We also suspect that important disease genes may have escaped detection because of this heterogeneity. We used presence/absence of family history to subdivide the cases and performed genome-wide association studies (GWAS) in Sporadic-PD and Familial-PD separately. The aim was to uncover new genes and gain insight into the genetic architecture of PD. Employing GWAS on the NeuroGenetics Research Consortium (NGRC) dataset stratified by family history (1565 Sporadic-PD, 435 Familial-PD, 1986 controls), we identified a novel locus on chromosome 1p21 in Sporadic-PD (PNGRC = 4x10-8) and replicated the finding (PReplication = 6x10-3; PPooled = 4x10-10) in 1528 Sporadic-PD and 796 controls from the National Institutes of Neurologic Disease and Stroke (NINDS) Repository. This is the fifth PD locus to be mapped to the short of arm of chromosome 1. It is flanked by S1PR1 and OLFM3 genes, and is 200 kb from a multiple sclerosis susceptibility gene. The second aim of the study was to extend the stratified GWAS to the well-established PD genes. SNCA_ rs356220 was associated with both Sporadic-PD (OR = 1.37, P = 1x10-9) and Familial-PD (OR = 1.40, P = 2x10-5). HLA_rs3129882 was more strongly associated with Sporadic-PD (OR = 1.38, P = 5x10-10) than Familial-PD (OR = 1.12, P = 0.15). In the MAPT region, virtually every single nucleotide polymorphism (SNP) had a stronger effect-size and lower P-value in Familial-PD (peak P = 8x10-7) than in Sporadic-PD (peak P = 2x10-5). We discovered and replicated a new locus for Sporadic-PD which had escaped detection in un-stratified GWAS. This demonstrates that by stratifying on a key variable the power gained due to diminished heterogeneity can sometimes outweigh the power lost to reduced sample size. We also detected distinct patterns of disease associations for previously established PD susceptibility genes, which gives an insight to the genetic architecture of the disease and could aid in the selection of appropriate study population for future studies.
    Full-text · Article · Feb 2014
Show more