Comprehensive Assessment of Genetic Sequence Variants in the Antioxidant ‘Master Regulator’ Nrf2 in Idiopathic Parkinson’s Disease

Parkinson’s disease (PD) is a complex neurodegenerative disorder influenced by a combination of genetic and environmental factors. The molecular mechanisms that underlie PD are unknown; however, oxidative stress and impairment of antioxidant defence mechanisms have been implicated as major contributors to disease pathogenesis. Previously, we have reported a PD patient-derived cellular model generated from biopsies of the olfactory mucosa, termed hONS cells, in which the NRF2-mediated antioxidant response pathway genes were among the most differentially-expressed. To date, few studies have examined the role of the NRF2 encoding gene, NFE2L2, and PD. In this study, we comprehensibly assessed whether rare and common NFE2L2 genetic variations modify susceptibility to PD using a large Australian case-control sample (PD=1338, controls=1379). We employed a haplotype-tagging approach that identified an association with the tagging SNP rs2364725 and PD (OR = 0.849 (0.760-0.948), P = 0.004). Further genetic screening in hONS cell lines produced no obvious pathogenic variants in the coding regions of NFE2L2. Finally, we investigated the relationship between xenobiotic exposures and NRF2 function, through gene-environment interactions, between NFE2L2 SNPs and smoking or pesticide exposure. Our results demonstrated a significant interaction between rs2706110 and pesticide exposure (OR = 0.597 (0.393-0.900), P = 0.014). In addition, we were able to identify some age-at-onset modifying SNPs and replicate an ‘early-onset’ haplotype that contains a previously identified ‘functional promoter’ SNP (rs6721961). Our results suggest a role of NFE2L2 genetic variants in modifying PD susceptibility and onset. Our findings also support the utility of testing gene-environment interactions in genetic studies of PD.


Introduction
Parkinson's disease (PD) is a complex neurodegenerative disorder influenced by a combination of genetic and environmental factors. The molecular mechanisms that underlie neurodegeneration in PD are unknown; however, recent studies have highlighted the significant role that oxidative stress and impairment of antioxidant defence mechanisms play as major contributors to disease pathogenesis [1][2][3].
Previously, we have reported a PD patient-derived cellular model generated from biopsies of the olfactory mucosa (termed human olfactory neurosphere-derived (hONS) cells) that demonstrate disease-specific differences in gene expression, protein function and metabolic activity, when compared to healthy controls [4]. Subsequent bioinformatics pathway analysis highlighted the "NRF2-mediated antioxidant response" pathway as one of the most altered in PD hONS cells.
Nuclear factor erythroid-2-related factor 2 (protein: NRF2; gene: NFE2L2) is a transcription factor in the phase II antioxidant and xenobiotic response pathway and is termed a 'master regulator' of expression for many antioxidant and detoxification pathway genes [5]. At basal levels, NRF2 is constitutively degraded in the cytoplasm by its antagonist, KEAP1 ('Kelch-like erythroid-cell-derived protein with CNC homology (ECH)-associated protein 1'). KEAP1 forms a complex with CUL3-RBX1 and regulates NRF2 through targeted ubiquitination and subsequent degradation. The mode of action for NRF2 begins upon exposure to oxidative stress, xenobiotics, or electrophilic compounds. This causes the modification of cysteine-151 of KEAP1 and subsequent stabilisation and translocation of NRF2 to the nucleus where it binds to the antioxidant response element (ARE) of its target genes [5,6].
Recent in vitro and in vivo studies have demonstrated a potential neuroprotective role of NRF2, attenuating neurotoxicity in 6-hydroxydopamine, hydrogen peroxide, and 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP) induced models of PD [7][8][9]. NRF2 activity and expression also significantly fall with age, the most common predisposing factor for PD [10]. Subtle accumulative changes in antioxidant response mechanisms may contribute to the oxidative damage previously identified in the post-mortem midbrain of PD patients [11]. In addition, other molecules directly or indirectly regulated by NRF2 have been strongly linked with PD; these include glutathione [12], heme oxygenase-1 (HO-1) [13,14], and NAD(P)H:quinone oxidoreductase 1 (NQO1) [15]. Functional experiments performed using our PD hONS cell lines have further demonstrated reductions in associated metabolic function, including reduced levels of glutathione, suggesting deficiencies in NRF2 function [16]. Moreover, these deficiencies were ameliorated after induction of the NRF2-mediated antioxidant response pathway with l-sulforaphane, an NRF2 activating agent.
Despite this putative neuroprotective role of NRF2, there have been few studies examining associations between NFE2L2 and PD. A candidate gene approach was undertaken in a Taiwanese cohort and two independent European (Swedish and Polish) case-control groups [17,18]. Of the 11 single nucleotide polymorphisms (SNPs) investigated, none were significantly associated with PD in either cohort. However, associations between a specific haplotype and PD were reported in both European case-control groups. Interestingly, this haplotype includes a previously described 'functional variant' of the promoter [19] and was associated with delayed age at onset (AAO) in the Swedish material and reduced risk of PD in the Polish material [17]. Both case-control groups used in this study were modestly sized and not adequately powered to unequivocally detect the reported associations.
We decided to investigate whether genetic variations in and around NFE2L2 modify susceptibility to PD using a large case-control sample recruited via the Queensland Parkinson's Project. We employed a haplotype-tagging approach to comprehensively determine if common SNPs around NFE2L2 are associated with PD. In addition, we also screened the coding regions of NFE2L2 to determine if rare coding variants could be responsible for the functional alterations observed in our PD hONS cell lines. Finally, because of the relationship between xenobiotic exposures and NRF2 function, we investigated possible gene-environment interactions between NFE2L2 SNPs and smoking or pesticide exposure.

Case-control ascertainment
Patients with neurologist diagnosed idiopathic Parkinson's disease (n = 1338) and neurologically normal, healthy, community-dwelling control subjects (n = 1379) were recruited as a part of the Queensland Parkinson's Project, a collaborative research study register of over 4000 community-dwelling Queenslanders, recruited since 2005, who have agreed to participate in research into Parkinson's disease and related disorders [20]. All subjects were of European Caucasian background and were recruited by experienced neurologists from specialist movement disorder clinics in Brisbane, Australia and the Australian electoral roll, and detailed epidemiological data has been obtained from these individuals. Study participants provided blood samples from which DNA was extracted using standard methods. A subset of the sample (67 individuals) also provided olfactory mucosa biopsies [4]. Study demographic data is presented in Table 1.

Ethics Statement
Written informed consent was obtained from all participating subjects in this study via a signed subject consent form, which is held in hard copy by the authors. The study and the consent procedures were approved by the Human Research Ethics Committee of Griffith University under clearance numbers: ESK/04/11/HREC "The Queensland Parkinson's Project-uncovering genetic risk factors for Parkinson's Disease"; and ESK/04/07/HREC "Olfactory biopsies in Parkinson's disease and related disorders". All participants had the capacity to provide individual consent at the time of recruitment.

High Resolution Melt analysis
High Resolution Melt (HRM) analysis was employed to screen the NFE2L2 coding regions, including the 5' and 3' UTR of DNA derived from 67 PD and control hONS cell lines. Assays were designed to amplify sequences no more than 200bp long, with final amplicon lengths between 145 and 183bp (S1 Table). NFE2L2 exons were sequenced in a randomly selected control to serve as reference samples. HRM analysis was performed using the Corbett Rotor-Gene 6000 Real-Time PCR system (Qiagen, Doncaster, Australia) and the Sensimix HRM Master Mix (Bioline, Alexandria, Australia). Assays were carried out as per manufacturer's instructions and standard laboratory protocols.

SNP and haplotype association analysis
We utilised linkage disequilibrium-based mapping approaches to "tag" all known common variants ± 5kb (minor allele frequencies > 10%, r 2 > 0.90) of NFE2L2 (S1 Fig). NFE2L2 SNPs previously genotyped in published case-control association studies of PD [17] were also genotyped here and, where possible, were used to complement or substitute tagging SNPs derived from HapMap data. This approach produced 11 tagging SNPs and was derived from the Hap-Map project CEU population (data rel 28/phase II+III, August10, on NCBI B36 Assembly). Assays for the 11 tagging SNPs-rs13035806, rs2706110, rs10183914, rs2001350, rs1806649, rs2364722, rs2886161, rs2364725, rs7557529, rs16865105 and rs6726395 were designed and executed using the MassARRAY genotyping platform (Sequenom, San Diego, USA). In total, 11 tagging SNPs were screened in 1338 PD and 1379 control subjects. Three NFE2L2 promoter SNPs (rs6721961, rs6706649, rs35652124) previously genotyped in published age-of-onset studies of PD [17] were also genotyped in 502 PD and 189 control subjects. SNP genotypes were analysed in PLINK version 1.07 (http://pngu.mgh.harvard.edu/~purcell/ plink/). Hardy-Weinberg equilibrium was calculated for all SNPs to assess genotyping reliability. Odds ratios (ORs) for genotype and haplotype associations (additive model) were calculated using logistic regression with adjustment for age and gender. Testing for multiple hypotheses was adjusted using a strict bonferroni correction. This study has 90% power to detect ORs of at least 1.41. NFE2L2 haplotype windows were phased to include haplotypes >1% in order to confidently replicate previously published age-at-onset (AAO) haplotype associations. AAO is defined as patient-reported 'age at first primary motor symptom'. Associations between individual SNPs and AAO were evaluated using linear regression models adjusted for gender. Pvalues were adjusted for multiple hypotheses testing using a strict Bonferroni correction. An overview of SNPs and haplotype definitions can be found on Table 2.

Gene-environment interaction analysis
Gene-environment interactions were tested using an additive model for NFE2L2 SNPs by the interactive terms: smoking and pesticide exposure. Smoking terms were dichotomised according to median pack years (<17 pk/yrs & !17 pk/yrs) and regular pesticide exposure denote a self-reported exposure to herbicides, insecticides or fungicides at least once weekly for a period of six months or more across their life [21]. Pesticide exposure was dichotomised according to regular pesticide exposure and non-regular pesticide exposure. Representative groups were created to reflect the combined effects on risk of PD: (1) common homozygote, non-exposed, (2) common homozygote, exposed, (3) heterozygote, non-exposed, (4) heterozygote, exposed, (5) alternate homozygote, non-exposed and (6) alternate homozygote, exposed. The common homozygote, non-exposed (1) group served as the reference in an additive interactive model. Odds ratios were calculated using binary logistic regression with adjustment for age and gender.

Exploring AAO effects using Moving Average Plots
Moving average plots (MAPs) of the differences in allele frequency between cases and controls can provide useful information about the true nature of the impact of genotype on the AAO of disease in case control studies [22]. MAPs graphically portray how the differences in allele frequency, between cases and controls, move with the age of the participants in each group. Such an analysis may provide a means to distinguish between risk factors and AAO modifiers. To do this, we ranked all cases and controls for age, and grouped the participants into 10 percentile ranges. We then plotted the frequency distribution of cases and controls as a function of age. MAPs were constructed for all SNPs. SNPs with significant interactive effects and associations are presented.

Cell culture
hONS cell lines were derived from biopsies of the olfactory mucosa from donor subjects and established as previously described [4]. Aliquots of hONS cell lines from PD patients and healthy controls were grown in Dulbecco's modified eagles medium/ham's F-12 (Gibco, Invitrogen) containing 10% foetal calf serum (Gibco, Invitrogen) and incubated at 37°C with 5% CO 2 .

Gene expression
Genome-wide expression data previously generated for investigating transcriptional differences between PD-and control-derived hONS cell lines was obtained and processed as previously described [4]. Briefly, total RNA was purified from 56 hONS cell lines (30 PDs & 26 We next conducted an analysis to determine whether the normalized gene expression of NFE2L2, KEAP1, and NQO1 was dependent on genotypes at the NFE2L2 tagging and promoter SNPs. As the gene-expression data was normally distributed for KEAP1 and NQO1, ANOVA tests were performed. NFE2L2 gene-expression data was not normally distributed; therefore a Kruskal Wallis test was performed to look for genotype dependant differences. Genotypes were dichotomised according to possession of 'major allele' (G/G) and 'minor allele' (G/T or T/T).

Cell viability assays
CyQUANT Cell Proliferation Assays (Life Technologies) were used to determine the cellular response of hONS cell lines treated with 50nM rotenone or DMSO for 24, 48, 72, 96 and 120 hours. CyQUANT Cell Proliferation Assays measure the cellular DNA content via fluorescent dye incorporation. The fluorescence intensity of each sample was measured using the Synergy2 plate reader, set with an excitation~485nm and emission detection at~530nm. T-tests were performed using normalised cell viability data and genotype data of previously described NFE2L2 tagging and promoter SNPs.

High resolution melt analysis
Thirty-two assays were designed that spanned the entirety of the protein-coding regions of

Genotype associations
Significant associations were observed between two NFE2L2 tagging SNPs and PD ( Table 3). The frequency of the alternative alleles of rs2364725 and rs7557529 were significantly higher in controls, although we noted that these SNPs were highly correlated (r 2 = 0.98; S1 Fig). In addition, the P-value for the association of rs2364725 with PD remained statistically significant following a Bonferroni correction for multiple hypothesis testing (P<0.05). All genotyping data in the control subjects conformed to Hardy-Weinberg equilibrium.

Gene-environment interaction analysis
In our QPP study sample, PD cases were more likely to be males and reported fewer pack-years of cigarette smoking compared to controls. The cases were also more likely to have reported regular exposure to pesticides (Table 1). Given the role of NRF2 in antioxidant response, we stratified our study sample based on cigarette smoking and regular pesticide exposure to look for interactive effects. No significant interactions were observed for SNPs rs2364725 (which showed a main effect on disease status) or rs10183914, rs35652124, rs6726395, and rs1806649 (which showed main effects on AAO) (S3 Table). However, a significant additive interaction was observed between rs2706110 and pesticide exposure (Table 4), this effect did not survive multiple hypotheses testing. Additionally, no interactive effect was observed for the previously described 'functional promoter SNP' rs6721961 (S3 Table).  Effects of genotype on cell viability in rotenone exposed hONS cells

Haplotype associations
Haplotype definitions were retained from a previous study [17] to investigate possible influence of susceptibility of PD within our sample. We also examined associations between haplotypes comprised of all SNPs and PD for completeness (Table 5). Our 11 tagging SNPs produced nine haplotypes, of which the haplotype designated 'T1' was modestly associated with PD (OR = 1.167 (1.03-1.32), P = 0.0149), although this did not survive a correction for multiple hypothesis testing. Haplotype age-of-onset analysis Our 11 tagging SNPs generated 9 haplotypes, of which, T2 was associated with a delayed AAO (P = 0.002, 2.2 years/allele) (Fig 1). Four phased haplotypes were generated from the three promoter SNPs (S4 Table). We found that haplotypes P2 and P3 were significantly associated with differences in AAO. A reduction of 4.6 years per copy was observed with haplotype P2, and conversely haplotype P3 was associated with an increase in AAO of 3.4 years per copy.

Discussion
Previously, we have reported disease-related differences in the NRF2-mediated antioxidant response pathway in hONS cell lines derived from patients with PD [16]. In the current study we attempted to comprehensively investigate whether coding region genetic variations in the NRF2 encoding gene could account for these differences. While we observed two rare NFE2L2 variants of unknown pathogenicity in 67 hONS cell lines, our results indicate that the NRF2-pathway differences previously observed in hONS cell lines are not likely driven by these variants. We also tested for genetic associations between SNPs that comprehensively tag all common genetic variability in and around the NFE2L2 gene locus and investigated their influence on PD susceptibility using the large Queensland Parkinson's Project case-control sample. To date, this is the largest focused candidate gene case-control study to examine the association of NFE2L2 with PD, and the first to examine NFE2L2 coding variants in PD patient cell lines. It is also the first to examine gene-environment interactions between NFE2L2 SNPs and PD susceptibility. Our association analysis reported two highly correlated NFE2L2 SNPs that are significantly associated with a reduced risk of PD. One of these variants (rs2364725) has previously been reported to interact with increased exposure to particulate air pollutants, in a population of nonsmoking myocardial infarction survivors, to prolong the QTc interval measured by an electrocardiogram [23]. This study retrospectively hypothesised that defence against oxidative stress is diminished in individuals with at least one minor allele. Paradoxically, in our study, the alternate allele of rs2364725 is protective for PD. However, no significant AAO, gene-environment interaction, gene expression, or cell viability differences for rs2364725 was observed in our study. It is currently unclear as to how this non-coding SNP in the 5'-region of the locus could impact mechanistically on NRF2 pathway function.
It is widely accepted that oxidative stress plays an important role in the onset of PD. However, the mechanism is difficult to explore in isolation due to the multifactorial nature of the disease. Previous literature on PD risk factors has emphasised the importance of considering gene-environment interactions when exploring genetic factors involving biological pathways impacted upon by external exposures [24]. This is particularly important for NFE2L2 variants, given the functional relevance of NRF2 in detoxification pathways.
To assess for possible gene-environment interactions between NFE2L2 SNPs and xenobiotic exposures commonly associated with PD (smoking and pesticide exposure) we examined the joint effects of genotype and exposure on PD risk using logistic regression models. We found preliminary evidence that the alternate allele at rs2706110 may reduce PD risk in people regularly exposed to pesticides; however, this did not survive a correction for multiple hypotheses. In addition, baseline expression levels of KEAP1 were found to be significantly lower in cell lines (n = 26) carrying the alternate allele of rs2706110 compared to cell lines (n = 26) containing the common allele. Reduced KEAP1 expression may correspond with a reduction in the targeted post-translational degradation of NRF2 by KEAP1 and its subsequent stabilisation and translocation to the nucleus. We observed no differences in cell viability between common and alternate alleles for this SNP in rotenone treated cell lines. Further investigation is required to determine how the alternate allele of rs2706110, located in the non-coding 3' region flanking NFE2L2, modulates NFE2L2 transcriptional response to environmental pesticide exposure.
Our investigation into disease modifying genotypes identified a number of AAO-modifying SNPs; rs10183914 (P<0.001, 2.6 years/allele), rs6726395 (P<0.002, 1.9 years/allele), rs1806649 (P<0.001, 2.4 years/allele), and rs7557529 (P = 0.024, 1.6 years/allele). Larger haplotype windows containing these alternate alleles also exhibit the age-modifying effect of these SNPs. This may indicate that any age-modifying effect observed in larger haplotypes may be driven by these alternate alleles.
MAPs graphically portray how the differences in allele frequency, between cases and controls, move with the age of the participants in each group. In our analysis we were able to distinguish between risk factors and age-at-onset modifiers using this method. The results of our MAPs are in concordance with our association and AAO study.
There have been several isolated reports of genetic associations, particularly haplotypes around NFE2L2 with PD. Our data has allowed a re-examination of some of these previous reports. None of the haplotypes generated from our tagging SNPs were associated with PD. In addition, none of the haplotypes associated with PD in previous studies were found to have a significant effect on disease susceptibility in our population (haplotypes VO1-7; Table 5).
For complete haplotype assessment, we investigated AAO effects in haplotypes generated from our promoter SNPs (P), tagging SNPs (T), replication haplotypes (VO), and combination haplotypes (promoter haplotypes in phases of the VO windows (VOP) and T windows (TP)) (S4 Table). We were able to replicate an 'early-onset' haplotype, P2 (TCT), which contains a previously identified 'functional promoter' SNP (rs6721961). Possession of the common allele, P3 (GCT), was conversely associated with a 'disease-delaying' AAO. This dichotomy of AAO was observed in all larger haplotype windows containing these alleles. Interestingly, rs6721961 was not found to independently modify AAO significantly (S2 Table). Haplotype P2 has previously been associated with significantly reduced NFE2L2 promoter activity in vitro [19], and possession of this AAO-modifying allele may reflect changes in transcriptional activity. hONS cell lines containing the alternate allele were found to significantly overexpress NQO1 compared to those containing the common allele (image C S4 Fig). Furthermore, a reduction in cell death was observed in rotenone treated cell lines, over time, containing the alternate allele (S5 Fig). Increased antioxidant capacity, in the form of NQO1, may correspond with an ability to mitigate xenobiotic insult. These findings contradict previous studies that suggest that the alternate allele at rs6721961 is associated with decreased transcriptional activity. No changes in NFE2L2 and KEAP1 expression were observed for this SNP. We did not examine this SNP in the larger case-control association study as this SNP was in the set selected for AAO investigation. It would be interesting to see whether this SNP is associated directly with disease risk, generally. Unfortunately, rs6721961 is not in LD with any of the examined SNPs in this study, therefore future experiments would be required to test this. It is most likely that our observation is a spurious finding, but further data is required to disprove this.
A recent large-scale meta-analysis of all PD GWAS data failed to identify SNPs in NFE2L2 that are associated with PD at a genome-wide significance level [25]. This suggests that independent main effects of NFE2L2 genotype may not be strong genetic risk-factors for PD. However, these GWAS studies do not consider the potentially important interactions between environmental exposures and genotypes, which might exist in specific populations or in specific geographical contexts. Thus more focused studies of such interactions are warranted, especially considering that meta-analyses of NFE2L2 SNPs from published data (as appears in PD Gene) suggest that there may, in certain populations, be some relationship between these genes and PD risk [25].

Conclusion
Overall, we report a comprehensive genetic assessment of NFE2L2 variants in PD. While we identified two rare NFE2L2 coding variants from our hONS cell lines, it appears unlikely that they are pathogenic and are not likely the cause of NRF2-mediated pathway differences in PD hONS cell lines. The results we present here suggest that common NFE2L2 variants may reduce PD susceptibility in certain conditions, such as regular exposure to pesticides. While our haplotype replication analysis did not confirm a previous disease association study, we did find significant AAO modification with a reported functional promoter SNP. Furthermore, we identified a number of SNPs associated with a significantly later AAO. These finding remained statistically significant following the stringent Bonferroni correction for multiple comparisons. Further functional analysis regarding the role of the disease-associated NFE2L2 SNPs on NRF2 pathway expression and activity in hONS cell lines is warranted to elucidate their potential role in modulating PD susceptibility.