Although our analysis was underpowered to detect associations for individual mitochondrial variants, our data provide evidence for an aggregate effect of common mitochondrial variation on risk of ischemic stroke. The score-based approach used to demonstrate this effect has been applied successfully to resolve genetic associations for lipid profiles, myocardial infarction, and blood pressure, as well as other complex traits [20
]. While in these conditions the score was comprised of SNPs with genome-wide significance, in schizophrenia a score built from multiple non-significant SNPs with small effect sizes has been validated in multiple independent cohorts [26
]. Our approach parallels that taken by the schizophrenia researchers, although we are not able to present replication in an independent data set. Our results were derived from meta-analysis of four independent cohorts, the results of which were all quite consistent with one another and without significant heterogeneity. Although this does not constitute formal replication, the correspondence in effect size and lack of between-study heterogeneity in these independent cohorts is suggestive of generalizability. In addition, the genetic score derived from these cohorts was also robust to two different forms of permutation, in which our results were tested repeatedly against the null hypothesis. Finally, we have applied our ischemic stroke genetic score to WMHV in an independent case-only sample, showing that this score comprised of mitochondrial variants that influence stroke risk also has an effect on the burden of white matter disease in individuals with ischemic stroke. This biologic extension to a related phenotype of ischemic stroke further strengthens the robustness of our findings.
We did not identify any SNP of large effect (OR > 2), despite sufficient power to detect such variants. Our analysis was underpowered to detect variants of small or moderate effects. Because we have not identified an association between ischemic stroke and any specific common variants, mitochondrial genetic testing for common alleles is not warranted at this time. Additionally, because our haplogroup-based analysis does not demonstrate a robust association between individual haplogroups and ischemic stroke, our results do not support haplogroup assignment for clinical prediction. However, future DNA sequencing studies building on our demonstration that common variants have an aggregate influence on risk of ischemic stroke may lead to discovery of individual rare variants with clinical implications for risk stratification in stroke.
This analysis provides concrete examples of the challenges particular to GWAS of mitochondrial variants. Autosomal GWAS studies often assume MAFs > 0.05 for common variants, while our observed median mitochondrial MAF was 0.032, with some variants displaying MAF as low as 0.005. These low observed MAFs greatly limit power to detect differences between groups, because larger sample sizes must be assembled to recruit rarer risk alleles. Furthermore, the mitochondrial genome is haploid, making it impossible for homozygous risk alleles to contribute additional signal in an additive genetic model. Finally, the low effect sizes seen in meta-analysis also limit power to detect novel SNP associations. These constraints altered our power estimates greatly in post hoc analysis.
The majority of the known genes in the mitochondrial genome subserve respiratory chain functions [7
]. It has been demonstrated that multiple rare mitochondrial variants can work in concert to cause disease [32
], possibly through collective impairment of the respiratory chain [33
]. This provides a strong biologic rationale for the application of a genetic-score approach, as this technique allows assessment of the aggregate effect of all mitochondrial SNPs. Under the genetic score paradigm, SNPs that do not contribute to disease association only introduce random associations to the model, favoring the null hypothesis. For this reason, and because our genetic score was robust to two independent forms of permutation, our data provide strong support for a role of common mitochondrial variants in ischemic stroke. Of note, our genetic score was comprised of all available genetic variants, and was therefore an unbiased approach to disease association. The selection of individual candidates for inclusion in a genetic score could introduce biases that limit replicability.
Our mtGWAS was performed using an exhaustive panel of SNPs, encompassing all variation with MAF > 1% in individuals of European ancestry. While mitochondrial SNPs are available on commercially-available genotyping platforms, they do not capture all known common variation. Even with imputation, these platforms cannot achieve full coverage of all common variants, limiting the utility of pooling data from the exhaustive mitochondrial panel and standard GWAS platforms for analysis. Future mitochondrial association studies will require continued use of targeted exhaustive panels [16
] to minimize risk of false negative results due to insufficient mitochondrial genome coverage.
Our study has limitations. Despite amassing a large, multi-center cohort of ischemic stroke cases and controls, we did not have power to detect individual SNP associations for the effect sizes and MAF observed in meta-analysis. To be adequately powered to identify associations with individual SNPs would require a cohort in excess of 80,000 cases with matching controls. Even introducing additional controls to our study to reach a 4:1 (control:case) ratio would only increase our power from 0.002 to 0.004. Our results cannot exclude a stroke subtype-specific effect for mitochondrial variants. Restricting sample size in sub-analysis of stroke subtypes would further erode power, decreasing the probability of detecting positive associations even for substantial subtype specificity. This study focused solely on inherited common variation in SNPs outside the D-loop. Mutations within the D-loop or insertion/deletion mutations would require a different study design to detect. We did not model heteroplasmy in our analysis, but this phenomenon would theoretically only erode power to detect associations for common inherited variants. Most importantly, we do not have an additional cohort of individuals in which to replicate our findings, and only after replication can our results be considered robust. While the fact that our results are robust to permutation supports our conclusions, our analysis is still based on a single data-set. However, the observation that the genetic score derived from ischemic stroke is also associated with WMHV in a nested case-only cohort represents a biologic extension of our results in what must be considered an independent phenotype assessed only in stroke cases. It is important to note that this result is insufficient to constitute a demonstration of an effect of specific common mitochondrial variants on WMHV. Rather, it shows that the common variants that increase the risk of ischemic stroke seem to also play some as-yet undefined role in the genetics of WMHV. Further research will be needed to independently assess the role of common mitochondrial variants in the phenotype of WMHV.
This large, multi-center mitochondrial GWAS of ischemic stroke was underpowered to detect individual variants. However, we have shown that some association does exist between common mitochondrial variants and stroke risk. Further studies are necessary to replicate our findings, and substantially larger sample sizes will be required to identify specific variants. It is possible that common variants in the mitochondrial genome conferring a small increase in stroke risk are in linkage with rare variants conferring much larger risk. Detection of these rare variants, through sequencing of the mitochondrial genome, may yield important new information on the role of mitochondria on ischemic stroke and other neurologic diseases.