Genetic susceptibility loci, environmental exposures, and Parkinson’s disease: a case-control study of gene-environment interactions

Abstract

Background

Prior studies causally linked mutations in SNCA, MAPT, and LRRK2 genes with familial parkinsonism. Genome-wide association studies have demonstrated association of single nucleotide polymorphisms (SNPs) in those three genes with sporadic Parkinson’s disease (PD) susceptibility worldwide. Here we investigated the interactions between SNPs in those three susceptibility genes and environmental exposures (pesticides application, tobacco smoking, coffee drinking, and alcohol drinking) also associated with PD susceptibility.

Methods

Pairwise interactions between environmental exposures and 18 variants (16 SNPs and two variable number tandem repeats, or “VNTRs”) in SNCA, MAPT and LRRK2, were investigated using data from 1,098 PD cases from the upper Midwest, USA and 1,098 matched controls. Environmental exposures were assessed using a validated telephone interview script.

Results

Five pairwise interactions had uncorrected P-values < 0.05. These included pairings of pesticides x SNCA rs3775423 or MAPT rs4792891, coffee drinking x MAPT H1/H2 haplotype or MAPT rs16940806, and alcohol drinking x MAPT rs2435211. None of these interactions remained significant after Bonferroni correction. Secondary analyses in strata defined by type of control (sibling or unrelated), sex, or age at onset of the case also did not identify significant interactions after Bonferroni correction.

Conclusions

This study documented limited pairwise interactions between established genetic and environmental risk factors for PD; however, the associations were not significant after correction for multiple testing.

Introduction

The causes of Parkinson’s disease (PD) are largely unknown. Both genetic and environmental factors have been implicated. Genetic loci that have been causally linked with familial parkinsonism and reproducibly associated with PD susceptibility worldwide include α–synuclein (SNCA), microtubule associated protein tau (MAPT), and leucine rich repeat kinase 2 (LRRK2) [1–7]. Immunohistochemical studies demonstrated that Lewy bodies, the neuropathological hallmark of PD, contain not only α–synuclein, but also MAPT [8] and LRRK2 [9] proteins. However, as individual factors or combined, these genetic susceptibility loci account for only a small fraction of PD. Environmental exposures that have been reproducibly associated with PD include pesticides application (“pesticides”), tobacco smoking (“smoking”), coffee drinking (“coffee”), and in some studies alcohol drinking (“alcohol”) [10–12]. However, as individual factors or combined, these environmental exposures also account for only a small fraction of PD.

Since genetic susceptibility loci and environmental exposures independently account for only a small fraction of PD, it has been postulated that interactions between genetic and environmental factors may be associated with much greater risk. For example, pesticides accelerate the rate of α–synuclein fibril formation in vitro and exacerbate the pathology associated with causal SNCA mutations in transgenic mice [13,14]. We previously reported that SNCA genotypes and herbicides had independent effects on PD risk, with no significant pairwise interactions [15]. However, in another study of interactions, while analyses of interactions were limited by small sample sizes, risk due to SNCA variations seemed to vary with pesticide exposure and smoking, especially in younger onset cases, suggesting an age-of-onset effect. [16]. Here we expand the scope of our previous studies of genetic susceptibility loci (main effects and gene-gene interactions analyses) [17,18], to also include environmental factors (gene-environment interaction analyses), focusing on the genetic and environmental factors that have been reproducibly associated with PD.

Methods

Study subjects

All subjects were recruited as part of a National Institutes of Health funded study of the molecular epidemiology of PD (2R01ES10751). The enrollment of matched cases and controls has been previously described [15,17]. PD cases were referred sequentially to the Department of Neurology of Mayo Clinic in Rochester, MN, from June 1, 1996 through June 30, 2007. Controls consisted of unaffected siblings of PD cases or matched unrelated controls. Cases were matched to a single participating sibling first by sex (when possible) and then by closest age. Cases without an available sibling were matched to unrelated controls of the same sex, age (year of birth ± 2 years), and residential region (Minnesota, Wisconsin, Iowa, or North and South Dakota pooled together). Unrelated controls of ages 65 and older were randomly selected from the Centers for Medicare and Medicaid Services (CMS) lists. Unrelated controls younger than 65 years were selected using random digit dialing, according to standard techniques [19]. Initially 1,103 cases and 1,103 matched controls were enrolled in the study [15,17]. Genomic DNA was collected, extracted, and stored as previously described [15]. Five cases were excluded subsequently because of indeterminate diagnosis. Thus 1,098 cases and 1,098 matched controls were used in subsequent analyses. The Institutional Review Board of the Mayo Clinic approved the study, and all 2,196 subjects provided written informed consent.

Genotyping

Single nucleotide polymorphisms (SNPs) in species-conserved regions and tag SNPs for the SNCA, MAPT, and LRRK2 loci were selected for genotyping as previously described [17,18]. In total, 19 SNPs in SNCA, 35 in MAPT, and 65 in LRRK2 were successfully genotyped using a bead array platform (Illumina GoldenGate). In addition, two variable number tandem repeats (VNTRs; SNCA REP1 and MAPT H1/H2 haplotype) that have been shown to be associated with PD worldwide via regularly updated meta-analysis (www.pdgene.org), were genotyped using a sequencing platform (Applied Biosystems). In total, 121 variants in the three susceptibility gene loci were successfully genotyped.

Selection of SNPs for gene-environment interaction analyses

Variants with minor allele frequency < 0.05 or showing departures from Hardy-Weinberg equilibrium (P < 0.001) were excluded from the analyses. We limited the gene-environment interaction analyses to SNPs with at least marginal evidence of association with PD (P < 0.1 in a univariate test of SNP main effect under the assumption of log-additive allele effects). We further applied a tag-SNP selection strategy to the resulting SNP list, using the pairwise Tagger algorithm with r² = 0.9 implemented in Haploview 4.2 [20]. This procedure resulted in the selection of 8 SNPs in SNCA, 6 SNPs in MAPT, and 2 SNPs in LRRK2 that had P-values < 0.1 in a trend test for association with PD. In addition, SNCA REP1 (coded based on the number of 259 bp alleles, the number of 263 bp alleles, or as the previously described REP1 score [15]) and the MAPT VNTR that distinguishes the H1/H2 haplotype were included in the analyses, as these variables also showed marginal evidence of association at the P < 0.1 level. The 18 SNPs and VNTRs used for the interaction analysis are listed in Table 1 [18].

Table 1.

Genetic variants included in the interaction analyses

Gene	Variants	SNP location or known function	Minor allele frequency	P-value3
SNCA	rs1372520	Intron	0.1848	0.0056
	rs2572324	Intron; associated with the extent of neurofibrillary pathology	0.3226	0.0090
	rs2583959	Intron	0.3013	0.0506
	rs2736990	Intron	0.4902	0.0017
	rs356186	Intron	0.1686	0.0119
	rs356218	3′ downstream	0.3545	0.0419
	rs3775423	Intron	0.0903	0.0090
	rs3775439	Intron	0.1290	0.0716
	REP1-2591	Regulates SNCA gene expression	0.2413	0.0345
	REP1-2632	Regulates SNCA gene expression	0.0749	0.0465
	REP1 score	Regulates SNCA gene expression	n.a.	0.0118
MAPT	rs16940758	Intron; regulates MAPT gene expression	0.1745	0.0844
	rs16940806	3′ UTR	0.1800	0.0059
	rs2435200	Intron	0.3968	0.0001
	rs2435211	Intron; regulates MAPT gene expression	0.3012	0.0856
	rs4792891	Intron	0.3016	0.0036
	rs8079215	Intron	0.2945	0.0660
	MAPT H1/H2	Regulates MAPT gene expression	0.1793	0.0042
	rs17484286	Intron	0.0927	0.0128
LRRK2	rs2404835	Intron	0.3255	0.0872

¹REP1-259 denotes the SNCA VNTR REP1 coded as the number of 259 bp alleles

²REP1-263 denotes the SNCA VNTR REP1 coded as the number of 263 bp alleles

³P-value for trend test for association with PD

Ascertainment of environmental exposures

All exposures were ascertained via telephone by direct or proxy (for subjects who had died or incapacitated subjects) interviews using a structured risk factors questionnaire administered by specifically trained study assistants. We previously described the ascertainment of pesticides exposures, the coding of types of pesticides (herbicides, insecticides, or fungicides), and the reliability of the exposures ascertainment [15, 21]. Subjects were also asked to provide information on smoking (ever or never) and number of cigarettes per day and number of years (with details about periods of life with different amounts). Subjects were also queried about coffee drinking (ever or never), number of cups (quantified as “small cups”, i.e., regular coffee cup of 6 ounces, or one shot of espresso; or as “medium or large cups”, i.e., coffee mug of 12 ounces, or two shots of espresso) per day and number of years (with details about periods of life with different amounts) [22]. Subjects were also asked to provide information about alcohol drinking (ever or never), the number of drinks per day and number of years (with details about periods of life with different amounts) [23]. For this study, pesticide, alcohol, and tobacco use were coded as yes (ever) and no (never). Coffee consumption was coded as low (0–3 cups/day) and high (4+ cups/day) [10]. In addition, tobacco use was quantified as pack years.

Statistical analysis

Pairwise interactions between variants in three genes (SNCA, MAPT, and LRRK2) and environmental exposures (smoking, coffee, pesticides, and alcohol) were assessed using conditional logistic regression analyses. For SNPs, a log-additive genotype coding scheme was used, while for SNCA REP1 genotypes were converted to scores ranging from 0 to 4 as previously described [15]. REP1 genotypes coded as the number of 259 alleles and the number of 263 alleles were also used in the interaction analyses. A total of 20 pairwise interactions were tested for each environmental factor (primarily defined as ever/never for pesticides use, cigarette smoking, coffee drinking, or alcohol drinking). All analyses were adjusted for age at study and sex. For each genetic variant and environmental exposure we calculated the odds ratio (OR), 95% confidence interval (CI), and P value for the univariate effect in the conditional logistic regression model. In addition, the coefficient for the multiplicative interaction term and the associated P value were calculated.

We performed similar analyses of gene-environment interactions in subgroups, restricting either to case-unaffected sibling or case-unrelated control pairs, to men–men or women–women pairs, or to younger or older pairs as defined by median age at onset in the cases.

Although the expected effect sizes of the interaction terms were unknown, we used the assumed/observed marginal effects to estimate power for a range of values of the interaction term. The power calculations were performed using the software Quanto (http://hydra.usc.edu/GxE) [24], with an alpha level of 0.003, which takes into account the multiple genetic variants that were tested for interaction with each environmental factor. As an example, we estimated power to detect an interaction between coffee consumption and the genetic variants of primary interest (SNCA REP1 and MAPT H1/H2 variation). The observed allele frequencies were used in the power calculations, and dominant effects of genetic variants were assumed. We considered models with marginal effect sizes consistent with those observed in our data (with both the 259 REP1 allele and the MAPT H2 haplotype having protective effects, with marginal odds ratios of approximately 0.80 and 0.75, respectively, and with coffee having a protective effect with an odds ratio of 0.84). Under these assumptions the power was estimated to be 80% to detect an interaction odds ratio of 2.0 for the 259 REP1 allele-coffee interaction and 2.1 for the MAPT H1/H2-coffee interaction.

The statistical packages SAS (version 9.2; SAS Institute Inc., Cary, NC) and R were used for all analyses [25]. In addition to the uncorrected P values, a Bonferroni correction was applied to correct the P values for the number of tests performed.

Results

The demographic characteristics of the 1,098 cases and 1,098 matched controls are summarized in Table 2. Patients with PD were more often men (64.2%) than women (35.8%). The subjects were primarily Caucasian of European origin.

Table 2.

Demographic characteristics of Parkinson’s disease (PD) cases, siblings, and unrelated controls

General characteristics	PD Case-Sibling Pairs		PD Case-Unrelated Control Pairs		All PD Case-Control Pairs
	PD Cases	Sibling Controls	PD Cases	Unrelated Controls	PD Cases	All Controls
Total sample, n	653	653	445	445	1,098	1,098
Men, n (%)	417 (63.8)	329 (50.4)	288 (64.7)	288 (64.7)	705 (64.2)	617 (56.2)
Women, n (%)	236 (36.1)	324 (49.6)	157 (35.3)	157 (35.3)	393 (35.8)	481 (43.8)
Age at onset of PD, median (range)	60.4 (28.2–86.9)	--	64.6 (23.3–88.0)	--	62.2 (23.3–88.0)	--
Age at study, median (range)a	66.3 (30.8–91.4)	65.1 (32.0–90.4)	70.3 (44.5–90.4)	71.8 (44.9–92.8)	68.0 (30.8–91.4)	67.6 (32.0–92.8)
Ethnicity:
Both parents of European origin, n (%)	570 (87.3)	557 (85.3)	357 (80.2)	388 (87.2)	927 (84.4)	945 (86.1)
Only one parent of European origin, n (%)	52 (8.0)	60 (9.2)	60 (13.5)	41 (9.2)	112 (10.2)	101 (9.2)
Family History of PD
First degree family history of PD, n (%)	106 (16.5)	--	73 (16.4)	--	179 (16.3)	--
At least third degree family history of PD, n (%)	193 (30.1)	--	122 (27.4)	--	315 (28.7)	--

^aAge at blood draw

For the overall sample at the uncorrected significance level (P < 0.05), five interaction pairs between the environmental exposures and SNCA, MAPT, and LRRK2 gene variants were associated with PD susceptibility (Table 3). However, none of the pairwise interactions were significant after Bonferroni correction for multiple comparisons. The MAPT SNP rs2435200 that had a significant main effect on PD susceptibility in our previous study had no significant interaction with environmental exposures in this study [17]. Secondary analyses (in strata defined by type of control, sex, or age at onset of the case) showed several pairwise interactions (Supplementary Table 1), but none remained significant after accounting for multiple testing.

Table 3.

Environment-SNP interactions with limited evidence of association with susceptibility to Parkinson’s disease (uncorrected P < 0.05)

Environment exposureb	Individual effect		SNP	Gene	Individual effect		Environment-SNP interactions
	OR (95% CI)	P value			OR (95% CI)	P value	Coefficient	P valuea
Pesticides (ever/never)	1.04 (0.86–1.28)	0.670	rs3775423	SNCA	1.40 (1.08–1.82)	0.010	−0.51	0.034
Pesticides (ever/never)	1.04 (0.85–1.27)	0.674	rs4792891	MAPT	0.78 (0.67–0.92)	0.002	−0.31	0.038
Coffee (low/high)	0.85 (0.69–1.04)	0.115	H1/H2	MAPT	0.76 (0.62–0.93)	0.007	−0.41	0.040
Coffee (low/high)	0.84 (0.68–1.03)	0.097	rs16940806	MAPT	0.77 (0.63–0.95)	0.012	−0.41	0.043
Alcohol (ever/never)	0.78 (0.62–0.97)	0.025	rs2435211	MAPT	1.16 (0.99–1.37)	0.071	0.38	0.020

^aPesticides and alcohol use were coded as yes (ever) and no (never). Coffee consumption was coded as low (0–3 cups/day) and high (4+ cups/day) [10].

^bP-values are not corrected for multiple testing.

In the overall sample at the uncorrected level (P < 0.05), pesticides use (ever/never) showed interactions with the genetic variants SNCA rs3775423 (coefficient_interaction = −0.51; P value = 0.034) and MAPT rs4792891 (coefficient_interaction = −0.31; P = 0.038) (Table 3). None of these pairwise interactions were significant after Bonferroni correction for multiple comparisons. Secondary analyses in strata defined by type of control, sex, or age at onset of the case (Supplementary Table 1) found no significant gene-pesticide interactions after accounting for multiple testing. Sub-analyses according to the types of pesticides (herbicides, insecticides, fungicides) also showed no significant pairwise interactions.

Coffee drinking (0–3 cups/day vs. 4+ cups/day) analyzed in the overall sample revealed interactions with the gene variants MAPT H1/H2 haplotype (coefficient_interaction = −0.41; P value = 0.04) and MAPT rs16940806 (coefficient_interaction = −0.41; P = 0.043) at the uncorrected significance level of P < 0.05 (Table 3). However, none of the pairwise interactions were significant after Bonferroni correction for multiple comparisons. Secondary analyses in strata (Supplementary Table 1) also showed no significant interactions after accounting for multiple testing.

At the uncorrected significance level (P < 0.05), alcohol drinking (ever/never) showed interactions with MAPT rs2435211 (coefficient_interaction = 0.38; P = 0.02) in the overall sample. This interaction was not significant after Bonferroni correction for multiple comparisons. Similarly, secondary analyses in sub-strata (Supplementary Table 1) showed no significant interactions after accounting for multiple testing.

Finally, in the overall sample, tobacco smoking (ever/never) was strongly associated with PD (OR, 0.56; 95% CI, 0.45–0.68; P < 0.0001). However, none of the pairwise interaction analyses between smoking (ever/never) and the 20 SNCA, MAPT, and LRRK2 variants were significantly associated with PD susceptibility (Table 3). Dose-effect was considered using pack-years smoked; none of the pairwise interactions between pack-years and the 20 SNCA, MAPT, and LRRK2 variants were significant. Secondary analyses (in strata defined by type of control, sex, or age at onset of the case) showed several pairwise interactions at an uncorrected significance level (Supplementary Table 1), but none remained significant after accounting for multiple testing.

Discussion

We observed limited evidence (nominal significance, P < 0.05) for pairwise interactions between common variants in the SNCA, MAPT, or LRRK2 genes and four environmental exposures (pesticides, smoking, coffee, and alcohol) in PD. Our results are consistent with a recently published study that also did not find significant interactions of either smoking or caffeine intake with SNPs in SNCA, MAPT, or LRRK2 genes [26]. Although in our study none of the pairwise interactions remained significant after Bonferroni correction, several interactions were suggestive and further investigation in larger samples is needed.

For pesticides we found limited evidence for interactions with SNCA and MAPT genes. Pesticide exposures were associated with PD in younger subjects in our previous study [15]. In this study there was a pairwise interaction of pesticides with SNCA SNP rs3775423; that SNP maps to a 3′ block of SNPs in linkage disequilibrium that has been consistently associated with PD in several studies [27–29]. The SNCA SNP rs3775423 had a nominally significant main effect on PD susceptibility in our previous study, but not after Bonferroni correction [17]. SNP variants in the 3′ region may increase alpha-synuclein expression, and pesticides may increase α–synuclein protein aggregation. In this study there was also a pairwise interaction of pesticides with MAPT SNP rs4792891. The MAPT SNP rs4792891 had a nominally significant main effect on PD susceptibility in our previous study, but not after Bonferroni correction [17]. It is unknown whether pesticides alter MAPT protein aggregation.

For coffee we found limited evidence for interactions with the MAPT gene (H1/H2 haplotype and rs16940806). The MAPT H1/H2 haplotype and SNP rs16940806 had nominally significant main effect on PD susceptibility in our previous study [17]. Coffee drinking has been reported to be inversely associated with PD [10]. Recently, a genome-wide gene-environment study suggested that the GRIN2A gene influences the inverse association of coffee with the risk of PD [30]. However, that study did not highlight MAPT and coffee interactions. Independent studies are needed to replicate the reported interaction of coffee and GRIN2A, as well as our suggestive findings for the interaction of coffee and MAPT, in PD.

For alcohol we found limited evidence for interaction with the MAPT gene (SNP rs2435211). This SNP has no main effect on PD susceptibility in our previous study [17]. Alcohol drinking was not associated with PD in our previous study using part of this study sample [10]. There has been very limited prior study of the interaction between alcohol and genetic factors in PD [23,31]. For smoking we observed no evidence for interaction with neither of the three genes selected, and to date there have been no genome-wide studies that revealed significant interactions of smoking with genomic variants in PD.

The limited evidence for gene-environment interactions in our study may be due to a number of factors. Three of the exposures that we studied (smoking, coffee, alcohol), though inversely associated with PD, may not be neuroprotective factors but rather early premotor manifestations of PD (“reverse causation”) [10]. If non-smoking is a PD trait (dependent variable) rather than a protective factor (independent variable), then it would not be expected to interact with genetic risk factors (other independent variables) in our analyses. The same could be true for coffee and alcohol.

Strengths of our study include the selection of three genes (SNCA, MAPT, and LRRK2) that have been causally linked to PD in families and in some ethnic groups, that are associated with PD susceptibility worldwide, and that encode proteins in Lewy bodies. We expanded the scope of our prior studies of main effects [17] and gene-gene interactions [18] to this study of gene-environment interactions. In aggregate, we have performed the most comprehensive analyses of the SNCA, MAPT, and LRRK2 genes, their gene-gene and gene-environment interactions, and PD.

Our study also has limitations. First, although our sample was large (1,098 PD cases and 1,098 controls), the study may have been underpowered to detect interactions for rare alleles, or smaller interaction effects for common alleles. Replication of our suggestive findings within a global genetics consortium may yield more robust evidence of interactions. Second, our exposure assessments were interview-based and we can’t exclude recall bias. It is somewhat reassuring that we found good test-retest agreement in the recall of exposures, good agreement between direct and proxy interviews, and good agreement between interviews and medical records in our previous studies [15, 21–23, 32]. Third, we did not exclude cases with monogenic forms of PD. The frequency of LRRK2 gene mutations in our sample was previously reported to be low (<1%) [33]. Fourth, we note that our study only considered SNPs and VNTRs. Further studies of gene-environment interactions should also consider copy number variations and epigenetic variations. Fifth, we did not perform complex model building procedures. However, such methods tend to produce over-fitted models and are difficult to replicate.

Despite studying the most established genetic and environmental risk factors for PD, we found limited evidence for pairwise interactions. Additional genome-wide gene-environment studies, that include pesticides exposures specifically, may provide the long anticipated evidence for gene-environment interactions in PD.

Supplementary Material

01

Supplementary Table 1. Subgroup analyses of environment-SNP interactions with limited evidence of association with susceptibility to Parkinson’s disease (uncorrected P < 0.05)