Common Variants within Oxidative Phosphorylation Genes Influence Risk of Ischemic Stroke and Intracerebral Hemorrhage

Abstract

Background and Purpose

Prior studies demonstrated association between mitochondrial DNA variants and ischemic stroke (IS). We investigated whether variants within a larger set of oxidative phosphorylation (OXPHOS) genes encoded by both autosomal and mitochondrial DNA were associated with risk of IS and, based on our results, extended our investigation to intracerebral hemorrhage (ICH).

Methods

This association study employed a discovery cohort of 1643 individuals, a validation cohort of 2432 individuals for IS, and an extension cohort of 1476 individuals for ICH. Gene-set enrichment analysis (GSEA) was performed on all structural OXPHOS genes, as well as genes contributing to individual respiratory complexes. Gene-sets passing GSEA were tested by constructing genetic scores using common variants residing within each gene. Associations between each variant and IS that emerged in the discovery cohort were examined in validation and extension cohorts.

Results

IS was associated with genetic risk scores in OXPHOS as a whole (odds ratio (OR)=1.17, p=0.008) and Complex I (OR=1.06, p=0.050). Among IS subtypes, small vessel (SV) stroke showed association with OXPHOS (OR=1.16, p=0.007), Complex I (OR=1.13, p=0.027) and Complex IV (OR 1.14, p=0.018). To further explore this SV association, we extended our analysis to ICH, revealing association between deep hemispheric ICH and Complex IV (OR=1.08, p=0.008).

Conclusions

This pathway analysis demonstrates association between common genetic variants within OXPHOS genes and stroke. The associations for SV stroke and deep ICH suggest that genetic variation in OXPHOS influences small vessel pathobiology. Further studies are needed to identify culprit genetic variants and assess their functional consequences.

INTRODUCTION

Despite remarkable advances in prevention, diagnosis, and treatment, stroke remains the world’s second-leading cause of death and a leading cause of disability. While many modifiable environmental factors contribute to stroke risk, there are ample data demonstrating a genetic risk component as well ¹. Recent genome-wide association studies (GWAS) have demonstrated that common DNA variants influence risk of ischemic stroke ^2-4.

A prior study demonstrated that common mitochondrial variants influence risk of ischemic stroke ⁴. The mitochondrial genome is vital to the assembly of the oxidative phosphorylation (OXPHOS) apparatus, but the majority of OXPHOS structural proteins are encoded within the autosomal genome ⁵. The OXPHOS apparatus consists of five complexes that are necessary to maintain aerobic homeostasis and preserve reduction/oxidation (redox) balance in the cellular environment. Multiple rare disorders are caused by mutations of OXPHOS genes, many of which result in neurodegenerative or stroke-like phenotypes, including seizures, metabolic infarcts, and encephalomyopathies ⁶. Additionally, OXPHOS fitness plays a role in the neuronal response to and recovery after oxidative stress ⁷.

We hypothesized that common genetic variants in OXPHOS genes, both within the autosomal and mitochondrial genome, influence the risk of stroke. To test this hypothesis, we performed a pathway-based genetic association analysis interrogating genetic variants within OXPHOS genes. We initially performed a cumulative test of all common genetic variation within OXPHOS loci by employing a gene set enrichment analysis (GSEA) technique. This allowed us to investigate whether the OXPHOS pathway was enriched for association with stroke risk. Based on this analysis, we sought to ascertain and quantify the role of these variants by calculating a genetic risk score from OXPHOS genes in the Massachusetts General Hospital (MGH) Ischemic Stroke GWAS. We then replicated this risk score association in a separate dataset comprised of individuals from the Ischemic Stroke Genetics Study (ISGS) and Siblings with Ischemic Stroke Study (SWISS). We then tested the same risk score in intracerebral hemorrhage (ICH) using individuals from the International Stroke Genetics Consortium ICH GWAS (ISGC ICH).

MATERIALS AND METHODS

Subjects

Genetic data and phenotypic information were contributed by the MGH Ischemic Stroke GWAS ⁸, ISGS ⁹, SWISS ¹⁰, and the ISGC ICH GWAS ¹¹ [TABLE 1]. Additional control individuals for the MGH dataset were contributed by the MIGen Consortium, a case-control study of genetic risk for myocardial infarction ¹². Hospital-based ischemic stroke case and control recruitment and phenotype ascertainment were performed according to protocols described previously, and stroke subtypes were assigned by TOAST criteria ^8-10. In cases where ischemic stroke subtype data was unavailable, individuals were dropped from subtype analyses but were allowed to remain in all-cause ischemic stroke analyses (n=124 in MGH, n=387 in ISGS/SWISS). Multi-center hospital-based ICH case and control recruitment and phenotype ascertainment were performed according to protocols described previously ¹¹. Location of ICH was assigned by stroke neurologists, based on standard criteria with central adjudication ^{11, 13}. Institutional review boards from all participating centers approved the study, and all participants gave informed consent for data collection, genotyping, and analysis of genetic data.

Table 1.

Study populations

	MGH/MIGen Cases	MGH/MIGen Controls	ISGS/SWISS Cases	ISGS/SWISS Controls	ISGC ICH Cases	ISGC ICH Controls
n	484	1159	1048	1384	928	909
- Small vessel	55	---	197	---	---	---
- Large artery	114	---	223	---	---	---
- Cardioembolic	191	---	241	---	---	---
- Deep ICH	---	---	---	---	430	---
- Lobar ICH	---	---	---	---	409	---
Sex (% female)	0.39	0.41	0.43	0.52	0.52	0.51
Age at enrollment (mean, SD)	66.5 (14.6)	47.5 (8.5)	64.8 (13.6)	66.5 (12.6)	72.4 (11.5)	73.0 (8.4)
Hypertension (%)	0.62	0.55	0.64	0.34	0.71	0.55
DM 2 (%)	0.21	0.18	0.19	0.11	0.20	0.09
Atrial fibrillation (%)	0.12	0.12	0.08	0.04	0.22	0.20
Current smoker (%)	0.20	0.18	0.16	0.06	0.17	0.20
Warfarin use (%)	---	---	---	---	0.08	0.03

ICH = intracerebral hemorrhage, n = number of cases/controls, SD = standard deviation, DM 2 = type 2 diabetes mellitus, --- = not applicable

Genotyping and Imputation

Blood samples from MGH/MIGen were processed and genotyped using the Affymetrix 6.0 platform, while ISGS and SWISS were assayed with the Illumina 660W and 1M platforms according to previously-published protocols ^8-10. Blood samples for the ISGC ICH cases and controls were genotyped on Illumina 660W ¹¹. For harmonization across platforms, all datasets had additional genotypes imputed using PLINK v1.07 (http://pngu.mgh.harvard.edu/~purcell/plink) and the International HapMap Project phase 3 reference dataset (http://www.hapmap.org). Captured mitochondrial variants from the genotyping arrays were extracted ¹⁴, and raw intensity files were inspected visually by C.D.A. and A.B. to confirm accuracy of genotype calls.

Genomic Quality Control

For all analyzed cohorts, quality control of genotyped individuals included gender-sex discordance, filtering for missingness by individual > 0.1, missingness by SNP > 0.05, and minor allele frequency (MAF) < 0.01. Individuals displaying cryptic relatedness (p-hat ≥ 0.125) and genotypes with significant departure from Hardy-Weinberg Equilibrium (p<10E-5) were excluded from analysis ¹⁵. Autosomal imputation was performed using PLINK v1.07 after quality control filtering. Mitochondrial imputation was performed using a haplotype-based approach with reference datasets from GenBank and Mitokor after additional mtDNA-specific quality control measures [Supplementary Appendix I] ¹⁶. Post-imputation, SNPs were excluded with MAF < 0.01 or RSQR quality index < 0.3.

OXPHOS SNP Selection

Genes encoding proteins directly involved in the OXPHOS respiratory chain were selected based on published criteria from a chemical dissection of mitochondrial function, yielding a total of 95 genes in the autosomal genome and 13 genes in the mitochondrial genome ⁵. SNPs falling within these genes +/− 100 kilobases and passing quality control filtering were extracted from the MGH and ISGS/SWISS datasets after imputation and included in the final analysis. Sub-analyses were performed for genes grouped according to each OXPHOS respiratory complex, classified according to annotation in the Ensembl Genome Browser (http://www.ensembl.org) [Supplementary Table S1 and S2].

Population Structure and Control

Only individuals of European ancestry were analyzed in the present study. Population structures for autosomal and mitochondrial variants were assessed independently due to their significantly different inheritance patterns, using principal component analysis ¹⁶. Autosomal principal components (PCs) 1-5 were extracted for each individual, and were added in association testing of autosomal SNPs until no additional reduction in genomic inflation factor (GIF) could be achieved (PC1-2 in all analyses). Mitochondrial PCs 1-10 were extracted for each individual, and were similarly added in association testing of mitochondrial SNPs until mitochondrial GIF was minimized (PC1-5 in all analyses).

Gene-set Enrichment Analysis

Testing for cumulative OXPHOS pathway associations with ischemic stroke risk was performed using the GSEA method ¹⁷ as implemented in the GenGen v.2010Apr29 software package ¹⁸. GSEA was implemented in this study as a preliminary screen of the OXPHOS pathway prior to generation of genetic scores, as a means to minimize the possibility of any false positive associations. The GSEA method determines whether variants within a pre-defined biological pathway contain more associations with the chosen phenotype than would be expected by chance alone. For ischemic stroke, GSEA testing was performed in the ISGS/SWISS cohort as a preliminary analysis prior to genetic score generation. GSEA was performed in the ISGS/SWISS replication cohort rather than the MGH/MIGen discovery cohort to prevent any chance enrichment of OXPHOS association in the MGH/MIGen cohort from influencing gene sets chosen for genetic score testing. A separate GSEA was performed in the ISGC ICH cohort. Results are reported as a permutation-derived empiric p-value (100,000 gene-set permutations) for gene set association with the ischemic stroke or ICH risk, with the null hypothesis derived from random sampling of an equal number of variants of similar MAFs chosen from genes not within the OXPHOS pathway. Using this technique, the significance threshold for our GSEA was set at empiric p < 0.05.

Genetic Score Generation in Ischemic Stroke

Combined effects of all autosomal and mitochondrial OXPHOS SNPs were evaluated using a score-based method previously described ^{4, 12} [FIGURE 1]. Briefly, each OXPHOS SNP was tested for association with all-cause ischemic stroke risk in the MGH/MIGen discovery case-control dataset. The results of this analysis, expressed as a beta coefficient for the risk allele at each SNP, were then clumped according to linkage disequilibrium (LD) using the “clump” function in PLINK v1.07. Only the SNP with the highest significance value was retained in regions where LD > 0.6 between SNPs. No additional pruning or thresholding was performed in an attempt to “optimize” the SNPs included in the genetic score. These beta coefficients were then applied to the corresponding OXPHOS SNPs in the ISGS/SWISS validation dataset. All subsequent analyses in replication and extension were based on this single set of beta coefficients from the MGH/MIGen discovery cohort in association with all-cause ischemic stroke [Supplementary Table S3]. Following beta coefficient extraction, the MGH/MIGen discovery cohort was not included in any further analyses. A risk score was generated for each individual by summing the beta coefficients associated with each risk allele present in the individual. Informed by GSEA results, scores were developed for all OXPHOS complexes, Complex I, and Complex IV. Because the risk score distributions failed testing for normality by Shapiro-Wilk, the score was divided into quintiles in an unsupervised fashion using the “cut” command in STATA v10.0 (www.stata.com) for association testing.

Figure 1.

Flowchart describing the MGH/MIGen discovery, ISGS/SWISS validation, and ISGC ICH extension cohorts. ICH = intracerebral hemorrhage, OXPHOS = all structural proteins directly contributing to oxidative phosphorylation complex function, SNP = single nucleotide polymorphism.

Genetic Score Association Testing in Ischemic Stroke

The ischemic stroke risk score quintiles were used as the independent variable in an ordinal logistic regression model for ischemic stroke risk, using age and sex as pre-specified covariates. Results reported represent risk (as expressed by odds ratio) per unit increase in score quintile. SNPs that were present in the MGH discovery cohort but were absent from the ISGS/SWISS validation cohort were dropped from the analysis (n=30). Additional covariates of hypertension, diabetes, and atrial fibrillation were also tested. Results of genetic score testing in the replication and extension datasets represent independent tests. Therefore p < 0.05 in the replication and extension datasets is considered statistically significant. All regression analyses were performed using STATA v10.0.

Extension of Genetic Score Analysis to Intracerebral Hemorrhage

The same beta coefficients of association with ischemic stroke from the MGH/MIGen discovery cohort were applied to individuals within the ISGC ICH cohort, again resulting in a risk score for each individual. Risk score quintiles were then used as the independent variable in a logistic regression for ICH risk, using age, sex, hypertension, and warfarin exposure as pre-specified covariates. For ICH, scores were developed only for all OXPHOS and Complex IV. Separate analyses were performed for deep and lobar ICH subtypes. Cerebellar and multi-compartment ICH were included in the all ICH analysis, but were dropped from deep and lobar analyses.

Post-hoc Power Calculation

Power for discovery of association between individual variants within OXPHOS genes and ischemic stroke was computed using the Genetic Power Calculator ¹⁹, with calculated ORs of 1.10, 1.20 and 1.40 and MAF of 0.10, 0.20 and 0.30. For this analysis, alpha was set at 6 × 10E-5 (842 independent tests for autosomal and mitochondrial variants within OXPHOS genes).

Power to detect an association between the genetic risk score and stroke phenotypes was computed using the expected proportion of variance explained, assuming that the overall information content of the score would account for 0.5%, 1% or 5% of variance between cases and controls. Power calculations were performed for ischemic stroke, ischemic stroke subtypes, ICH, and ICH deep and lobar subgroups.

RESULTS

Genotyping Quality Control and Imputation Results

Implementation of quality control and imputation methods left 1843 individuals and 707 autosomal SNPs in the MGH/MIGen cohort, 2632 individuals and 677 autosomal SNPs in the ISGS/SWISS cohort, and 1837 individuals and 707 autosomal SNPs in the ISGC ICH cohort [FIGURE 1]. 135 mitochondrial SNPs were retained in all three cohorts after extraction and haplotype-based imputation [Supplementary Appendix II].

Gene Set Enrichment Analysis in Ischemic Stroke

GSEA was performed using the ISGS/SWISS cohort, testing the OXPHOS gene-set 100,000 times against randomly-assigned gene sets of equal size [TABLE 2]. GSEA testing for the full OXPHOS gene-set and those within each respiratory complex demonstrated associations between ischemic stroke and the full OXPHOS gene-set (p=0.012) as well as OXPHOS Complexes I and IV (p=0.024 in both). Among ischemic stroke subtypes, GSEA revealed significant association between SV stroke and OXPHOS Complex I and IV (p = 0.008 and p = 0.005, respectively), although there was only a trend toward association between SV stroke and the full OXPHOS gene-set (p=0.091).

Table 2.

Gene set enrichment analysis (p-values)

	Cases/Controls	Full OXPHOS	Complex I	Complex II/III	Complex IV	Complex V
All ischemic stroke	1048/1384	0.012	0.024	> 0.20	0.024	0.11
- CE	241/1384	> 0.20	> 0.20	> 0.20	> 0.20	> 0.20
- LA	223/1384	> 0.20	> 0.20	> 0.20	> 0.20	> 0.20
- SV	197/1384	0.091	0.008	> 0.20	0.005	> 0.20
All ICH	928/548	>0.20	0.12	---	0.035	---
- Lobar ICH	409/548	>0.20	>0.20	---	>0.20	---
- Deep ICH	430/548	>0.20	0.16	---	0.009	---

Association between gene sets and ischemic stroke (ISGS/SWISS) and ICH (ISGC ICH), with p-values reported from 100,000 permutations against the null.

CE = cardioembolic stroke, ICH = intracerebral hemorrhage, LA = large artery stroke, SV = small vessel stroke, OXPHOS = all structural proteins directly contributing to oxidative phosphorylation complex function, --- = analysis not performed.

Gene Set Enrichment Analysis in Intracerebral Hemorrhage

As with our ischemic stroke analysis, GSEA was performed in the ISGC ICH cohort in order to determine whether OXPHOS gene-sets were enriched for association with ICH risk [TABLE 2]. Based on our results from ischemic stroke GSEA, only the full OXPHOS, Complex I, and Complex IV gene-sets were carried over for testing in ICH. An association was found between all ICH and Complex IV (p=0.035). After restricting cases to deep and lobar subgroups, deep ICH retained association with Complex IV (p=0.008).

Genetic Score Analysis in Ischemic Stroke

Based on GSEA results, only SNPs in the full OXPHOS gene-set as well as Complexes I and IV were used to calculate genetic scores. Similarly, only all-cause ischemic stroke and the SV subtype were carried forward for genetic score analysis [TABLE 3]. Application of beta coefficient-based scores in the ISGS/SWISS cohort demonstrated associations between a score comprised of the full OXPHOS gene-set and ischemic stroke (odds ratio (OR)=1.17, 95% confidence interval (CI)=1.03–1.33). This full OXPHOS score was also associated with the SV stroke subtype (OR=1.16, 95% CI=1.04–1.29). Of note, these genetic score results were largely driven by autosomal variants, with results deviating less than 20% when mitochondrial variants were excluded [Supplementary Table S6].

Table 3.

Genetic score analysis

	Cases/Controls	Full OXPHOS		Complex I		Complex IV
		OR (95% CI)	p	OR (95% CI)	p	OR (95% CI)	p
All ischemic stroke	1048/1384	1.17 (1.03- 1.33)	0.008	1.06 (1.00-1.12)	0.050	1.05 (0.99-1.12)	0.075
SV only	197/1384	1.16 (1.04- 1.29)	0.007	1.13 (1.01-1.26)	0.027	1.14 (1.02-1.27)	0.018
All ICH	928/548	---	---	---	---	1.08 (1.01-1.17)	0.039
Deep only	430/548	---	---	---	---	1.14 (1.03-1.25)	0.008

Genetic score analysis results in the ISGS/SWISS validation cohort and ISGC ICH extension cohort, performed only on gene sets and subgroups with significant p-values in gene set enrichment analysis. Odds ratios represent risk per unit increase in risk score quintile.

CI = confidence interval, OR = odds ratio, OXPHOS = all structural proteins directly contributing to oxidative phosphorylation complex function, SV = small vessel stroke, --- = analysis not performed.

Nominal p-values are reported for association with ischemic stroke and ICH in the validation and extension cohorts.

In analysis of our Complex I score, ischemic stroke (OR=1.06, 95% CI=1.00–1.12) and the SV stroke subtype (OR=1.13, 95% CI=1.01–1.26) demonstrated significant association. For our Complex IV score, there was a trending association for all-cause stroke, and a significant association for SV stroke (OR=1.14, 95% CI=1.02-1.27).

Regression analyses for ischemic stroke were performed with and without the inclusion of vascular risk factors as covariates in logistic regression (hypertension, diabetes mellitus, and atrial fibrillation). These regressors did not demonstrate significant association with the genetic scores, and did not alter the results of the regression analysis (p = NS, data not shown).

Extension of Genetic Score Analysis to Intracerebral Hemorrhage

We constructed genetic scores in the ISGC ICH cohort based on beta coefficients from the MGH/MIGen ischemic stroke cohort [TABLE 3]. This analysis revealed association between a genetic score from Complex IV genes and all ICH (OR=1.08, 95% CI=1.01-1.17, p=0.039) as well as deep ICH (OR=1.14, 95% CI=1.03-1.25, p=0.008).

Post-hoc Power Calculation

Power calculations for discovery of individual OXPHOS genetic variants in association with ischemic stroke revealed a maximum power of 73% to detect variants conferring an OR of 1.4 in association with ischemic stroke risk, at an MAF of 0.30 [Supplementary Table S3]. SNPs conferring lower OR and lower MAF substantially limited study power to detect individual variants, as did restriction of samples to the SV subtype. We performed power calculations for our genetic score analyses based on percentage of variance explained by the genetic score, ranging from 0.5% to 5%. The genetic score analysis was powered at 21% to explain 0.5% of variance in ischemic stroke risk for all-cause strokes, and 3% to explain 0.5% of variance in the SV subtype. In application of this genetic score to ICH, power was 16% to explain 0.5% of variance in all ICH, and 8% to explain 0.5% of variance in deep ICH [Supplementary Table S4]. As a reference, percentages of variance explained in our logistic regression models incorporating genetic scores ranged from 0.5-1% in most analyses.

DISCUSSION

Our pathway-based analysis demonstrates that common genetic variants in OXPHOS genes are associated with risk of both ischemic stroke and ICH. These associations are robust, having passed GSEA and replication in independent cohorts. Stratifying ischemic stroke and ICH by subtype and OXPHOS genes by mitochondrial Complex, we reveal associations for Complex I and IV in SV stroke, and Complex IV in deep ICH. These sub-analyses retain significance despite a substantial restriction in sample size and SNP counts.

While ample evidence exists for rare mutations leading to severe OXPHOS dysfunction in a variety of familial mitochondrial syndromes with stroke phenotypes, our analysis provides evidence of a role for common genetic variants within OXPHOS in sporadic ischemic stroke and ICH. These results contribute to a growing body of evidence linking OXPHOS genetic and functional variation to common neurologic diseases, including Alzheimer Disease, Amyotrophic Lateral Sclerosis, and Parkinson Disease, to name a few ^20-23.

Our sub-analyses restricted to variants within Complexes I and IV genes reveal additional parallels to rare mitochondrial syndromes. Mutations within Complex I account for up to 1/3 of the known respiratory chain diseases, and represent a major determinant of the redox state of the cell ^{24, 25}. Complex IV is the final electron donor in the pathway, receiving electrons from Cytochrome C and passing them to oxygen. Complex IV dysfunction, in addition to causing early-life mitochondrial diseases such as Leigh’s Disease and encephalomyopathies, has also been implicated in neurodegenerative diseases ²⁰. While Complex I is far larger than Complex IV (50 vs. 23 gene products), our demonstrated positive associations for both Complexes suggests that statistical power alone did not determine our results, and the correlations with existing knowledge of mitochondrial disease supports a possible role for these Complexes in sporadic human disease. Neither Complex I nor Complex IV dysfunction have effective treatments, although administration of cofactors has been reported to improve function in some instances ²⁶.

Both SV stroke and deep ICH result from disease of cerebral small vessels, and share common risk factors such as diabetes and hypertension ^{27, 28}. Our findings suggest a possible shared genetic contribution to small vessel pathobiology underlying SV stroke and deep ICH, which could be mediated through disruption in oxidative function at the tissue level, or through modification of upstream systemic or endothelial risk factors shared by the two diseases. We have previously reported an association between mitochondrial common variants and white matter hyperintensity volume (WMHV) ⁴, a phenotype to which SV stroke and deep ICH have been linked ^{29, 30}. These new data provide additional support for the role of energy metabolism in small vessel disease. However, given that OXPHOS dysfunction can result in numerous physiologic derangements, including ATP depletion, reactive oxygen species generation, defects in cell signaling, and alteration in apoptotic thresholds, our demonstrated associations cannot directly inform the underlying pathobiology of this small vessel link. Functional studies to identify the mechanisms of bioenergetic dysfunction will be needed to build on these results.

APOE allele status has been demonstrated to affect the risk and severity of lobar intracerebral hemorrhage, presumably due to a strong relationship between cerebral amyloid angiopathy and the lobar ICH subtype ^{11, 13}. The present study suggests a relationship between OXPHOS variants and deep ICH only, contributing to growing evidence that deep and lobar ICH represent genetically distinct entities. Genetic approaches appears to be useful tools to explore the differences between these ICH subtypes, and hopefully can lead to a more comprehensive understanding of the pathogenesis of these similar but unique disease subtypes.

Limitations render our results preliminary. The magnitude of effect sizes for OXPHOS genetic scores in stroke risk in our analyses are small, but are in-line with the results from other GWAS efforts in ischemic and hemorrhagic stroke ^{2-4, 8, 11, 13}. Our GSEA did not find associations for the large artery or cardioembolic stroke subtypes in ischemic stroke, or the lobar subtype in ICH. Given our power calculations, it is possible that the restriction in sample size for subtype-stratified analyses led to a false negative for these subtypes. We cannot, therefore, definitively demonstrate that the effect of OXPHOS genetic variants on ischemic or hemorrhagic stroke is isolated to SV ischemic or deep ICH subtypes. Many subjects in the MGH/MIGen and ISGS/SWISS datasets were previously used in prior study of mitochondrial variants in ischemic stroke, although the method of analysis differed between these studies ⁴. These datasets theoretically could be particularly enriched with OXPHOS associations, although the positive extension to the ISGC ICH cohort would not be predicted if this were the case.

Genetic score analysis, although effective in aggregating signals to detect association, cannot identify individual causative variants. As a result, we are unable to determine the particular genetic loci conferring risk in the present study. GWAS platform-based SNPs were used in this analysis, which are not highly enriched for functional variants likely to cause missense, nonsense, or splice-site mutations. It is possible that other common or rare genetic variants in the OXPHOS pathway lie in linkage disequilibrium with our included variants, exerting a more substantial effect in affected individuals. Given the small aggregate effect sizes of the genetic scores in our analysis, prohibitively large sample sizes would be required to achieve sufficient power to detect individual OXPHOS variants. The significance thresholds in the current study were set according to established techniques in GSEA and genetic score analysis, and can be considered robust due to the use of permutation in the case of GSEA, and the use of separate discovery and replication cohorts in the case of the genetic score analysis. Because the majority of genes encoding OXPHOS proteins are autosomal, we cannot determine whether the low risk contribution (≤20%) of mitochondrial variants to the risk scores for ischemic stroke and ICH in our analysis is due to an imbalance in SNP contributions to the genetic score or a true difference in risk proportion. Finally, we cannot determine whether the involvement of the OXPHOS pathway in ischemic stroke and ICH is mediated at the brain tissue level, or possibly through modification of systemic vascular or metabolic risk factors. Follow-up analyses will be required to address OXPHOS function in different tissue types.

CONCLUSION

Through a pathway-based analysis, we have demonstrated that genetic variation within genes involved in the OXPHOS apparatus associates with ischemic stroke risk, particularly the SV stroke subtype. Extension to ICH reveals retained association with OXPHOS Complex IV in deep ICH. Further studies will be necessary to clarify the functional impact of these variants on OXPHOS function.