Genome-wide association analysis identifies TXNRD2, ATXN2 and FOXC1 as susceptibility loci for primary open-angle glaucoma.

Primary open-angle glaucoma (POAG) is a leading cause of blindness worldwide. To identify new susceptibility loci, we performed meta-analysis on genome-wide association study (GWAS) results from eight independent studies from the United States (3,853 cases and 33,480 controls) and investigated the most significantly associated SNPs in two Australian studies (1,252 cases and 2,592 controls), three European studies (875 cases and 4,107 controls) and a Singaporean Chinese study (1,037 cases and 2,543 controls). A meta-analysis of the top SNPs identified three new associated loci: rs35934224[T] in TXNRD2 (odds ratio (OR) = 0.78, P = 4.05 × 10(-11)) encoding a mitochondrial protein required for redox homeostasis; rs7137828[T] in ATXN2 (OR = 1.17, P = 8.73 × 10(-10)); and rs2745572[A] upstream of FOXC1 (OR = 1.17, P = 1.76 × 10(-10)). Using RT-PCR and immunohistochemistry, we show TXNRD2 and ATXN2 expression in retinal ganglion cells and the optic nerve head. These results identify new pathways underlying POAG susceptibility and suggest new targets for preventative therapies.

Glaucoma is a clinically and genetically complex disease that is the leading cause of irreversible blindness worldwide[1,2]. Primary open-angle glaucoma (POAG), the most common form of the disease in most populations[3], is characterized by retinal ganglion cell apoptosis and progressive optic nerve damage[4]. While recent genome-wide association studies (GWAS) have identified interesting POAG risk loci[5-9], these account for only a fraction of disease heritability. To identify new POAG loci, we have completed a meta-analysis of GWAS summary findings of individuals of European descent from the United States with replication in an Australian study (ANZRAG) and further evaluation in a second Australian study (BMES), 3 European studies and a Singaporean Chinese dataset.

For stage 1 (discovery) we meta-analyzed summary data from 8 independent datasets (3,853 cases and 33,480 controls; Supplementary Table 1) with European ancestry from the United States collectively referred to as the National Eye Institute Glaucoma Human Genetics Collaboration Heritable Overall Operational Database (NEIGHBORHOOD). For all 8 NEIGHBORHOOD studies cases were primarily defined as at least 1 reliable visual field showing loss consistent with glaucoma, without a secondary cause, or CDR (cup-to-disc ratio) ≥ 0.7 or CDR asymmetry ≥ 0.2 or documented progression of optic nerve degeneration (in the Ocular Hypertension Treatment Study [OHTS])[10]. Controls had CDR <0.7. Additionally, for all datasets except OHTS, controls had intraocular pressure (IOP) of < 21 mmHg (Supplementary Table 2). For each dataset, site-specific quality control (sample and genotype call rates ≥ 95%), principal components analysis (EIGENSTRAT[11]), and imputation (IMPUTE2[12] or MACH[13,14]) were completed using the 1000 Genomes Project reference panel (March 2012) (Supplementary Note, Supplementary Table 3). Imputed variants with minor allele frequencies <5% or imputation quality scores (r2) <0.7 were removed prior to analysis. Dosage data, in the form of estimated genotypic probabilities, were analyzed in ProbABEL[15] for each dataset using logistic regression models, adjusting for age, sex, any significant eigenvectors and study-specific covariates. Genomic inflation was less than 1.05 (λ-value) for each individual dataset (Supplementary Figure 1). Estimated genotypic probabilities for 6,425,680 variants were meta-analyzed in METAL[16] using the inverse variance weighted method. To confirm that the results were not skewed by a particular dataset we completed a sensitivity analysis by selectively removing each dataset and meta-analyzing the remaining 7. The ORs from each grouping of 7 datasets were highly correlated with the results obtained from all 8 datasets (Supplementary Figure 2).

The stage 1 genome-wide association results are shown in Supplementary Figure 3, and the association results for all SNPs with P < 1×10−5 are shown in Supplementary Table 4. One SNP (rs2745572[A]) located in a novel region on 6p 50Kb 5′ of FOXC1 reached genome-wide significance (OR = 1.25, P = 2.36×10−9) in stage 1 (Table 1). Additionally, 873 SNPs including SNPs located in regions not previously associated with POAG on 1p, 2p, 2q, 5p, 6p, 6q, 10q, 12q, 20p, and 22p had P< 1×10−5 (Supplementary Table 4).

Table 1

Association and meta-analyses of the NEIGHBORHOOD and ANZRAG cohorts for the top-ranked loci.

						NEIGHBORHOOD(discovery, stage 1)		ANZRAG(replication, stage 2)		Meta-analysis NEIGHBORHOOD + ANZRAG
Chr	SNP	Position	A1	A2	Gene	OR	P	OR	P	OR	P	Het I2	Het P
1	rs7518099	165736880	t	c	TMCO1	0.70	3.12E-13	0.71	8.02E-06	0.70	6.35E-18	4.1	0.40
4	rs11732100	7924690	t	c	AFAP1	0.85	3.93E-06	0.78	6.77E-06	0.83	1.98E-10	24.1	0.23
6	rs2745572	1548369	a	g	FOXC1	1.25	2.36E-09	1.18	6.46E-03	1.23	6.50E-11	0	0.58
9	rs7866783	22056359	a	g	CDKN2B-AS1	0.70	1.04E-23	0.67	2.92E-12	0.69	1.22E-34	39.2	0.11
9	rs2472493	107695848	a	g	ABCA1	0.83	1.24E-07	0.70	2.08E-10	0.79	2.44E-15	34	0.15
12	rs7137828	111932800	t	c	ATXN2	1.17	6.53E-06	1.22	4.36E-04	1.18	9.20E-09	3.1	0.41
14	rs33912345	60976537	a	c	SIX6	0.76	8.94E-15	0.78	6.21E-06	0.76	1.71E-19	0	0.87
17	rs9897123	10020501	t	c	GAS7	0.85	6.86E-06	0.79	1.45E-05	0.83	5.85E-10	0	0.69
22	rs35934224	19872645	t	c	TXNRD2	0.79	1.39E-06	0.74	2.01E-04	0.77	1.08E-09	0	0.83

Association results for the SNPs reaching genome-wide significance in the discovery cohort (rs2745572) as well as other top-ranked loci showing replication. Genomic position is based on build 37. A1 is the effect allele for both cohorts. Novel loci are highlighted in bold font. Abbreviations: NEIGHBORHOOD, National Eye Institute Glaucoma Human Genetics Collaboration Heritable Overall Operational Database; ANZRAG, Australian and New Zealand Registry of Advanced Glaucoma; Chr, chromosome; OR, odds ratio; Het I2, heterogeneity I2 index; Het P, p-for-heterogeneity; N/A, not available. Loci not previously reported are denoted in bold text.

Next we investigated the associations of the most significant stage 1 SNPs (P< 1×10−5) in a replication dataset of European Caucasians from Australia (ANZRAG, Australian and New Zealand Registry of Advanced Glaucoma; 1,155 cases and 1,992 controls) (Supplementary Note), and performed a meta-analysis of these SNPs in the NEIGHBORHOOD and ANZRAG datasets using the effect sizes and their standard errors (stage 2). In the meta-analysis, SNPs in novel regions 50kb 5′ of FOXC1 [top SNP rs2745572[[A], OR = 1.23, P = 6.5×10−11], within intron 14 of ATXN2 [top SNP rs7137828 [T], OR = 1.18, P = 9.2×10−9] and within intron 11 of TXNRD2 [top SNP rs35934224[T], OR = 0.77, P = 1.8×10−9] reached genome-wide significance (Table 1, Supplementary Table 5). The regional association results for these SNPs are shown in Figure 1.

Figure 1

Association results for the regions reaching genome-wide significance after stage 2

These plots show the regional association and recombination rates for the top SNPs in the discovery cohort (NEIGHBORHOOD, 3,853 cases and 33,480 controls) after meta-analysis with data for these SNPs from ANZRAG (1,155 cases and 1,992 controls). In each plot, the solid diamond indicates the top-ranked SNP in the region based on two-sided P values. The colored box at the right or left corner of each plot indicates the pairwise correlation (r2) between the top SNP and the other SNPs in the region. The blue spikes show the estimated recombination rates. The box underneath each plot shows the gene annotations in the region. Each plot was created using LocusZoom for the top-ranked SNP in each region with a 400 kb region surrounding it. (a) The top SNP for this plot is rs2745572 on chromosome 6 upstream of FOXC1 with P = 6.50×10−11. (b) The top SNP for this plot is rs7137828 on chromosome 12 within ATXN2 with P = 9.20×10−9. (c) The top SNP for this plot is rs35934224 on chromosome 22 within TXNRD2 with P = 1.08×10−9.

For each of the 3 novel regions reaching genome-wide significance after stage 2, we further examined their association with POAG in: a second Australian dataset (BMES, Blue Mountains Eye Study) (107 cases and 600 controls); 3 European datasets (875 cases and 4,107 controls in total); and a study of Singaporean Chinese (1,037 cases and 2,543 controls); (stage 3). Meta-analysis of all datasets exceeded genome-wide significance for all three top SNPs (Figure 2, Supplementary Figure 4): TXNRD2 rs35934224[T], combined P = 4.05×10−11, OR = 0.78; ATXN2 rs7137828[T], combined P = 4.40×10−10, OR = 1.17; and FOXC1 rs2745572[A], combined P = 1.76×10−10, OR = 1.17 (Supplementary Tables 6 and 7). The ATXN2 top SNP (rs7137828) is very rare in the Singaporean Chinese population and thus could not be evaluated.

Figure 2

Meta-analysis Results

Forest plots showing effect estimates for participating studies, as well as for the replication effort. Pooled estimates for odds ratios and 95% confidence intervals were calculated by fixed effects, inverse variance weighting meta-analysis. Individual dataset results are indicated by blue squares and summary values are indicated by black diamonds. (Top) Association results for rs2745572 (FOXC1 region top SNP). (Middle) Association results for rs7137828 (ATXN2 region top SNP). (Bottom) Association results for rs35934224 (TXNRD2 region top SNP). For the overall NEIGHBORHOOD (NBH), the summary value for the population cohorts (POP; NHS/HPFS/WGHS) are presented separately from the case/control cohorts (CC; Iowa, OHTS (Ocular Hypertension Treatment Study), Marshfield, MEEI, NEIGHBOR). Results for the individual NBH datasets are shown in Supplementary Figure 4. Individual and summary results for Stage 2 (ANZ and ANZ+NBH) and Stage 3 cohorts (EPIC, GER, UK, BMES, SC) and summary points for all European ancestry (EU) datasets and all datasets (EU + SC) are shown. For rs7137828 replication could not be completed in SC due to rare minor allele frequency. Total sample size for rs2745572 and rs35934224 is 7,027 cases and 42,772 controls, and for rs7137828, 5,990 cases and 40,179 controls. Abbreviations: ANZ, ANZRAG;EPIC, European Prospective Investigation into Cancer-Norfolk Eye Study; GER, Germany; UK, United Kingdom; SC, Singapore Chinese; EU, European ancestry.

SNPs in the GAS7 region, previously associated with intraocular pressure (IOP), a quantitative trait that, when elevated, is a risk factor for glaucoma[17-19], were significantly associated with POAG after stage 2 (top SNP rs9897123[T], OR = 0.83, P = 5.85×10−10) (Table 1). Other POAG loci identified in recent studies[5-9] were also confirmed, including TMCO1, CDKN2BAS, SIX6, ABCA1, and AFAP1 (Table 1, Supplementary Table 8). PMM2 SNPs recently identified in Chinese POAG[9] were nominally associated with POAG (top SNP rs12444233[T], OR = 1.13, P = 0.0016).

POAG, like many complex human diseases, displays clinical sub-phenotypes[20,21]. In particular, optic nerve degeneration in POAG can occur without elevation of IOP, a clinical subtype defined as normal-tension glaucoma (NTG)[22]. The NEIGHBORHOOD POAG dataset included 725 NTG cases (maximum IOP ≤ 21 mm Hg) and 1,868 high tension glaucoma (HTG) (maximum IOP > 21 mm Hg) cases (pretreatment IOP was not available for 1260 cases). The meta-analysis of NTG cases (using all the controls from the datasets with NTG cases) revealed one novel locus on chromosome 12q (rs2041895 [C], OR= 1.48, P = 2.41×10−8) in stage 1 (Supplementary Figure 5, Supplementary Table 9). The direction of effect was consistent (OR=1.15) in the ANZRAG NTG dataset, but did not reach significance (P=0.11) and the combined association result (NEIGHBORHOOD + ANZRAG) fell just short of genome-wide significance (P= 8.01×10−8), possibly due to a smaller number of NTG cases in the ANZRAG dataset (N=363). In the NEIGHBORHOOD discovery dataset we confirmed previous NTG associations on 9p[7] (CDKN2BAS top SNP rs1333037[T], OR = 1.67, P=1.35×10−12) and 8q22[7] (top SNP, rs284491[T], OR = 0.66, P=2.30×10−8) and in the HTG subgroup, (1,868 cases) confirmed associations with TMCO1[6,17], and SIX6[7,23] (Supplementary Figure 6, Supplementary Table 10). The FOXC1 region SNPs associated with POAG overall were also significant in the NEIGHBORHOOD HTG subgroup (most significant SNP rs2317961, OR =0.76, P = 2.58×10−8).

To assess the possible functional effects of SNPs at the three newly identified POAG loci, we accessed and applied data from ENCODE[24], SCAN (eQTL)[25], GENEVAR (eQTL)[26], GTEx (eQTL)[27] and RegulomeDBv2[28,29]. After stage 2, seven SNPs reached genome-wide significance in the FOXC1 region (Supplementary Table 5) and all seven of these are located 50Kb 5′ to FOXC1 in a region annotated by ENCODE as regulatory (Supplementary Figure 7) and are associated with enhancers in several cell types (P=0.01, RegulomeDBv2). The most significant SNP (rs2745572) is evolutionarily conserved (GERP = 1.8) and alters a Barhl1 transcription factor binding site (RegulomeDBv2). In zebrafish Barhl1 is expressed in distinct retinal cell lineages and is differentially regulated by Atoh7[30], a retinal-specific transcription factor that has been previously associated with optic nerve area[31] and glaucoma[32]. In the NEIGHBORHOOD meta-analysis we found nominal evidence for association with ATOH7 and POAG (top SNP rs1867567[A] P = 0.042, OR, 1.07).

Four SNPs in the ATXN2 region were significantly associated with POAG after stage 2 (Supplementary Table 5). These SNPs are located in genomic regions enriched for enhancers (P = 0.01) in lymphoid cells (RegulomeDBv2). The most significant SNP (rs7137828) is located in a SP1 transcription factor binding site, and another associated SNP, rs653178 which is in linkage disequilibrium with rs7137828 (R2>0.8, Caucasian European Ancestry) is located in an Esr2 (Estrogen receptor 2) binding site. Both SP1 and Esr2 are expressed in retinal cells[33, 34] and could influence expression of ATXN2[35].

The TXNRD2 region that includes 22 associated SNPs at the genome-wide level after stage 2 (Supplementary Table 5) is significantly enriched for enhancers (P=1×10−6, RegulomeDBv2) and DNaseI hypersensitivity sites (P = 4.3×10−5, RegulomeDBv2) in multiple cell types. Additionally, 6 of the TXNRD2 SNPs are cis eQTLs significantly affecting TXNRD2 transcript levels in MuTHer (Multiple Tissue Human Expression Resource) Twins[36] in lymphoblasts and skin (P=1×10−8, GENEVAR; Supplementary Figure 8). The top SNP (rs35934224) is also an eQTL in skin using RNA seq and 1000 Genomes imputation (P=2.32×10−13)[37]. All 22 TXNRD2 SNPs are significant cis eQTLs (P<1×10−4) in the GTEx database[27] in thyroid tissue and 19 are significant eQTLs in tibial nerve tissue (Supplementary Figure 9). The most significant TXNRD2 SNP (rs35934224) is located in a binding site for NRSF (neuron-restrictive silencer factor, also known as REST, (repressor element 1-silencing transcription factor), a transcription factor that potently protects neurons from oxidative stress[38].

FOXC1 is a member of the forkhead family of transcription factors and rare coding sequence mutations (missense, nonsense, and CNVs) are known to cause anterior segment dysgenesis and early-onset glaucoma with dominant inheritance[39,40]. FOXC1 has not been previously implicated in common adult-onset forms of glaucoma including POAG or HTG. Interestingly, association over GMDS, located 3′ to FOXC1, has been identified in a study using some of the same samples used here[8]. In our study we found genome-wide significant association adjacent to FOXC1 in the 5′ regulatory region and less significant association in GMDS (top SNP rs9378638, OR = 0.83, P = 7.50×10−6). The top SNPs in the two regions are approximately 400kb apart and are not in linkage disequilibrium. Conditional analysis confirmed that the odds ratio and P-value for the significantly associated SNPs 5′ to FOXC1 are unchanged by conditioning on the GMDS peak SNP, suggesting that these are independent associations (Supplementary Figure 10). The 5′ regulatory SNPs associated with POAG and HTG identified by this study could be involved in regulation of FOXC1 expression.

The ATXN2 and TXNRD2 genomic regions have not been previously associated with POAG or with any glaucoma-related quantitative traits such as optic nerve parameters or IOP. Expansions of an ATXN2 CAG repeat can cause spinocerebellar ataxia 2 with optic atrophy and intermediate expansions can contribute to development of amyotrophic lateral sclerosis (ALS)[41]. Interestingly, very recently two other genes known to be responsible for Mendelian forms of NTG have also been shown to contribute to ALS (amyotrophic lateral sclerosis)[42,43]. The ATXN2-SH3 region has been associated with retinal venular caliber in Caucasians with European ancestry[44]. We analyzed the expression of ATXN2 mRNA in normal human ocular tissues by RT-PCR and found expression in the cornea, trabecular meshwork, ciliary body, retina and optic nerve (Supplementary Figure 11). Immuno-labeling of sections of a normal mouse eye showed evidence of Atxn2 in the retinal ganglion cells and optic nerve (Figure 3).

Figure 3

ATXN2 and TXNRD2 are expressed in the retina and optic nerve head

(A) Representative images of immunofluorescence using an anti-ATXN2 antibody shows ATXN2 (green) present in cells in the ganglion cell layer (arrows, upper panels) as well as punctate staining in the inner plexiform layer (arrowhead, right most upper panel). Only a low level of punctate staining was observed in the optic nerve head (arrowhead, lower panels). (B) Representative images of immunofluorescence using an anti-TXNRD2 antibody shows TXNRD2 (green) present in cells in the ganglion cell layer (arrows, upper panels) as well as significant punctate staining in the inner plexiform layer (arrowheads, right most panel). Significant staining was also observed in cells in the optic nerve head (lower panels) indicative of astrocytes that form pial columns (arrows, right most panel). Punctate staining was also observed in the optic nerve head (arrowheads, lower panels). For each antibody, at least 3 sections from 6 eyes were assessed. No staining (not even punctate staining) was observed in the no primary control tissue (data not shown). Blue=DAPI. In all rows, right most panels are boxed regions in center panels. Scale bars: Upper left and center panels in A and B = 20 μm; Lower left and center panels in A = 15 μm; Lower left center and panels in B = 25 μm; Right most panels in A and B = 5 μm.

TXNRD2 codes for thioredoxin reductase 2, a mitochondrial protein necessary for reducing damaging reactive oxygen species generated by oxidative phosphorylation and other mitochondrial functions[45]. Cellular oxidative stress has been hypothesized as a cause of retinal ganglion cell dysfunction in glaucoma[46] and over-expression of thioredoxin 2, the substrate of thioredoxin reductase 2 (encoded by TXNRD2), increased retinal ganglion cell survival in an experimental glaucoma model[47]. We confirmed by RT-PCR that TXNRD2 is expressed in normal human ocular tissue (Supplementary Figure 11) including the retina and optic nerve. Immuno-labeling in mice showed strong staining in retinal ganglion cells as well as in the optic nerve (Figure 3). These data suggest that reduction of reactive oxygen species by TXNRD2 could prevent mitochondrial dysfunction and retinal ganglion cell apoptosis in glaucoma. TXNRD2 is the first mitochondrial protein associated with glaucoma risk.

In this study, common variants near FOXC1, ATXN2 and TXNRD2 were identified as new risk loci for POAG. These genes suggest novel pathways that may contribute to glaucoma development including abnormal ocular development (FOXC1), neuro-degeneration (ATXN2) and mitochondrial dysfunction secondary to accumulating reactive oxygen species (TXNRD2). Targeting these pathways could lead to effective and potentially preventative glaucoma therapies.

ENCODE, http://www.genome.gov/encode/ and http://genome.ucsc.edu/ENCODE/

GENEVAR, http://www.sanger.ac.uk/resources/software/genevar/

GTEx, http://www.gtexportal.org/home/

IMPUTE, http://mathgen.stats.ox.ac.uk/impute/impute.html

LocusZoom, (http://csg.sph.umich.edu/locuszoom/)

MACH, http://csg.sph.umich.edu/abecasis/MACH/download/

METAL, http://csg.sph.umich.edu/abecasis/metal/index.html

NEIGHBORHOOD, http://glaucomagenetics.org

ProbABEL, http://www.genabel.org/

RegulomeDBv2, http://www.broadinstitute.org/mammals/haploreg/haploreg.php

SCAN, http://www.scandb.org/

SHAPEIT, https://mathgen.stats.ox.ac.uk/genetics_software/shapeit/shapeit.html

1000 Genomes Project, http://www.1000genomes.org/

URLS

Data access

Summary data for the NEIGHBORHOOD POAG meta-analysis is available on the NEIGHBORHOOD website (URL listed above, see ‘Publications’).

Online Methods

Study design

Imputed genotypes (1000 Genomes panel, March 2012) for 3,853 cases and 33,480 controls from 8 independent datasets were used as the discovery cohort for this genome-wide association study for Primary Open Angle Glaucoma (POAG) (stage 1). The association results for the top SNPs from the discovery cohort were replicated in 1,155 cases and 1,992 controls from an Australian POAG study of Caucasians of European ancestry (stage 2) followed by further replication (stage 3) in a second Australian study, BMES (Blue Mountains Eye Study) and 3 European studies (EPIC (European Prospective Investigation into Cancer-Norfolk Eye Study), GER (Germany), UK (United Kingdom); (982 cases and 4,707 controls total), and a Singaporean Chinese datasest of 1,037 cases and 2,543 controls. The details for all datasets including genotyping platforms, quality control, imputation methods and diagnostic criteria are listed in the Supplementary Notes.

Meta-analysis (Discovery, stage 1)

Quality-control was performed for each data set as described in the Supplementary Note. Overall sample and genotype call rates were ≥ 95% for each site. Samples with Log R ratio (LRR) and B allele frequency (BAF) values suggestive of copy number variants were removed prior to analysis. Principal components (eigenvectors) were computed for all participants using EIGENSTRAT[11]. For each dataset logistic regression was performed in ProbABEL[15] for all analyses (POAG overall, HTG, NTG), controlling for age, sex, and study-specific covariates including study-specific eigenvectors. Each analysis was evaluated separately for overall genomic inflation (implementing the R package GenABEL) (λ-value ≤ 1.05 for each dataset) (Supplementary Figure 1). Results were meta-analyzed in METAL[16] implementing the inverse variance weighted method and applying genomic control correction.

Replication (Stage 2 and 3)

Loci of interest in the discovery cohort (NEIGHBORHOOD; P<1×10−5) were evaluated in the first replication cohort (ANZRAG) and meta-analyzed with the NEIGHBORHOOD results (stage 2). The top SNPs for the three novel regions were evaluated in 5 additional datasets, one Australian (BMES), 3 European (EPIC, GER, UK) and a Singaporean Chinese dataset.

Power calculations

Power calculations were done as described[48]. For the stage 1 discovery analysis, power calculations using disease prevalence of 2%[49] indicated that there was 96% power of detecting loci at P < 1.0 × 10−5 (the threshold for carrying over to stage 2) at minor allele frequencies as low as 30% with per-allele odds ratios of 1.17. The entire sample set (stages 1, 2 and 3) had 99% power to detect loci at P < 5.0 × 10−8 at minor allele frequencies as low as 30% with per-allele odds ratios as low as 1.17.

Candidate genes and functional effects

Genes of interest in the associated region were identified using Ensembl[50], UCSC genome Bioinformatics[51], and Genecards[52]. To predict functional effects of the top POAG associated SNPs, we used the ENCODE project data[24], HaploReg v2[28] and RegulomeDB[29]. We used SCAN[25], Genevar[26] and GTEx[27] and a study of UK twins using RNA seq and 1000 genomes imputation[37] to investigate expression quantitative trait loci within genomic regions of interest.

Statistical analyses

Conditional analyses were done using the top SNPs in the FOXC1, ATXN2, and TXNRD2 regions as well as the top SNP in the previously reported GMDS region[8] conditioning on the risk allele in the region of interest. Conditional analyses were performed using GCTA (Genome Complex Trait Analysis)[53].

Forest plots to visualize the effect sizes of top SNPs in each region by dataset were created using the rmeta package in R. The odds ratios and 95% confidence intervals for each displayed SNP were plotted and the P-values listed for each analysis (Figure 2) and each NEIGHBORHOOD dataset (Supplementary Figure 4).

Sensitivity analysis using the leave-one-out method was done by excluding each NEIGHBORHOOD dataset from a meta-analysis of the other 7 datasets. We compared the odds ratios from these analyses by calculating the Pearson’s product-moment correlation coefficient between each leave-one-out analysis and the overall meta-analysis of eight NEIGHBORHOOD datasets (Stage 1), as shown in Supplementary Figure 2. Correlations were calculated in R using the corrplot package and ellipse option.

Expression analysis of genes at associated loci in ocular tissues

Total RNA was extracted from dissected tissues from normal human donor eyes as previously described[54,55] using an RNA isolation kit from Life Technologies (Carlsbad, CA, USA). Reverse Transcriptional reactions were completed using SuperScript III reverse transcriptase from Life Technologies. Primer sequences were designed to specifically amplify TXNRD2 and ATXN2. PCR reactions were performed using the recommended conditions with Platinum Taq DNA polymerase (Life Technologies, Carlsbad, CA, USA) using a Touch Down program. Amplified PCR products were visualized by gel electrophoresis with 2% agarose gel.

Immunohistochemistry

C57BL/6J mice (males and females) were maintained on a 12/12 hours light/dark cycle. All experiments were approved by the Animal Care and Use Committee at The Jackson Laboratory. Eyes from 2–4 months old C57BL/6J mice were enucleated and fixed in 4% Paraformaldehyde for 2 hours, rinsed in 0.1M Phosphate buffer, immersed in 30% sucrose overnight and frozen in OCT. 15 mm sections were placed on Fisherbrand Superfrost Plus Slides and stored at −70°C until required. Sections were incubated overnight at 4°C in the following primary antibodies: rabbit anti TXNRD2 (1:50 Acris); rabbit anti-ATXN2 (1:50, Acris). All antibodies were diluted in PBT (1×PBS, 1% TritonX-100). Sections were blocked in 2.5% chicken serum (in PBT) for 1 hour, then incubated overnight at 4oC. After primary incubation, sections were washed 3 times in PBT and incubated with the secondary antibody (goat anti-rabbit IgG) for 4hrs at 4°C. All sections were then counterstained with DAPI and mounted with Aqua PolyMount. Images were collected on a Leica SP5 Confocal microscope. For each antibody, at least 3 sections from 6 eyes were assessed. Antibodies were obtained from Acris: Ataxin 2, Catalogue number: 21776-1 AP; Immunogen: Ag16470; Genebank ID (clone info): BC114546; Purification method: Antifen affinity purification; Txnrd2: Catalogue number: 16360-1-AP; Immunogen: Ag8367; Genebank ID (clone info): BC007489; Purification method: Antifen affinity purification.

All images in Figure 3 were taken on a Leica SP5 confocal microscope. Images in left and center panels were taken with a 20× glycerol objective, right panels were taken with a 63× glycerol objective. Excitation was performed using a 405 Diode laser (DAPI) and Argon laser (ATXN2 or TXNRD2). Collection was performed using sequential scanning: scan 1 = PMT 1 (gain 966) for DAPI, scan 2 = PMT 2 (gain 1013) for ATXN2 or TXNRD2.