We performed MitoExome sequencing of the entire mtDNA and exons of 1034 nuclear genes encoding mitochondrial proteins, including all 77 nuclear OXPHOS disease-related genes that were reviewed recently (8
). We applied MitoExome sequencing to 42 unrelated patients with a spectrum of early-onset OXPHOS disorders who lacked molecular diagnoses. Of note, we did not know what percent of unsolved OXPHOS cases would be expected to be due to mutations in established loci, since to our knowledge no study had sequenced even the 77 known disease-related genes in a collection of cases. We found that only 24% of unsolved cases were due to mutations in the known disease loci (including 1 mtDNA deletion and 9 nuclear gene defects shown in ), highlighting the locus heterogeneity of OXPHOS disease. A further 31% of patients harbored rare, protein-modifying, recessive variants in candidate genes not previously linked to OXPHOS disease. Since such variants exhibit five-fold enrichment in cases over background, they are likely enriched for truly causal alleles. We performed complementation experiments to firmly establish pathogenicity for NDUFB3
in complex I deficiency, and based on independent mutations in P41 and P42, we suggest a novel link between the gene AGK
and myopathic mtDNA depletion through an unknown mechanism.
Recessive mutations in 13 additional genes, never previously linked to mitochondrial OXPHOS disease, were identified. Formal proof of pathogenicity will be established by cDNA complementation experiments in cellular models or patient fibroblasts (13
), when such cells exhibit an OXPHOS defect, or by detecting independent mutations in individuals with similar phenotype. We estimate that six of these 13 genes will prove to be bona fide
disease genes. These candidates are particularly exciting since until recently such discoveries generally required familial forms of disease to narrow down the search region. Some of the candidate genes have established roles in OXPHOS biology consistent with the enzyme defect found in the relevant patient (e.g. UQCR10, LYRM4, EARS2
), while the majority of candidates encode poorly characterized proteins that have never before been linked to OXPHOS and will reveal fundamentally novel biochemical insights (e.g. C1orf31
, C6orf125, AKR1B15
Importantly, this pilot study showed that approximately half of the 42 sequenced patients lacked “smoking gun” prioritized variants. We can envision at least five possible explanations. First, we may have missed the causal variants due to a technical lack of sensitivity. Although we detected over 90% of SNVs present in the MitoExome, there is an unknown sensitivity for indels and exon deletions due to lack of training data. Second, the causal variant may have been located in a gene that we did not target with MitoExome sequencing. While possible, this explanation seems less likely since unbiased linkage and homozygosity mapping strategies to date have found that 94% of causal genes encode mitochondrial proteins. Third, the causal variant may be located in a non-targeted intronic or regulatory region. While these are very likely to exist, no robust methods are available yet for interpreting such variants. Fourth, and perhaps most likely, our stringent filters may have rejected the truly causal variant. For example, de novo
dominant alleles (39
), acting through gain-of-function or haploinsufficiency, were not prioritized since without parental DNA these alleles are difficult to distinguish from the high heterozygote burden of apparently deleterious alleles in healthy individuals. Similarly, it is difficult to distinguish benign from pathogenic mtDNA variants (40
). By applying MitoExome sequencing to parental DNA, such de novo
alleles may be identified in the future (41
). Finally, and potentially most interesting, our assumptions on the genetic architecture of OXPHOS disease may be inaccurate. Nearly all of the nuclear genes discovered to date correspond to Mendelian syndromes, characterized by strong, highly penetrant alleles. It is possible that many of the sporadic cases of OXPHOS disease in our cohort are due to the combined action of multiple “weak” alleles, each with incomplete penetrance.
Our study underscores the need for clinical standards for interpretation of genetic variants to evolve as NGS is applied more widely. First, current guidelines for interpreting clinical genetic tests, such as the American College of Medical Genetics (ACMG) guidelines (42
), are deliberately restricted to gene loci with established roles in disease. Second, it is notable that many variants purported to be causal for disease may, in some cases, be benign polymorphisms (43
). For example, we detected 44 alleles previously reported as pathogenic, but only six actually appear to explain the phenotype while the remainder were heterozygous or present in patients with unrelated phenotypes (Supplementary Methods
We anticipate that the success rate for establishing molecular diagnosis in unselected cases of infantile OXPHOS disease using NGS will be higher than that observed in this study. Indeed, our 42 patients were refractory to molecular diagnosis using traditional methods as most patients had been screened for common mutations in mtDNA or in relevant genes suggested by phenotype (e.g. POLG
) and were not from informative pedigrees. Analysis of a representative cohort of 291 unrelated infantile patients with “definite” OXPHOS disease from the Murdoch Childrens Research Institute, of which 124 cases had previous molecular diagnosis, suggests that MitoExome sequencing could enable diagnosis in roughly 47% of all infantile patients and prioritize candidates in a further 20% (see Supplementary Materials
In the coming years, we anticipate that three advances will greatly aid interpretation of DNA variation for routine clinical diagnosis. First, NGS studies of Mendelian families will rapidly expand the set of validated OXPHOS disease-related genes from 77 to perhaps 200 nuclear loci. Second, NGS studies of ethnically diverse, healthy individuals will generate databases of allele frequencies that are necessary for filtering out common variants unlikely to cause severe disease, as was recently shown for interpretation of carrier screening (43
) and has long been appreciated in the interpretation of mtDNA variation (45
). Third, while costs currently support MitoExome sequencing (roughly one-third the cost of exome sequencing), future cost reductions will enable sequencing of the entire exome or genome, as well as simultaneous analysis of parental DNA in order to phase compound heterozygous variants and detect de novo
The subset of patients for whom a clear molecular etiology was possible spotlights the immediate promise of NGS in clinical diagnosis. For these patients, MitoExome sequencing, requiring a single blood sample, can accelerate clinical diagnosis and enable genetic counseling where appropriate. Genetic diagnosis will also enable rational subclassing of disease, which may help predict clinical course and severity, and lead to patient grouping for targeted therapeutics. However, we anticipate that even with improvements afforded by broader catalogs of genomic variation, interpretation of most sequence variants will be most useful when integrated with the broader clinical and biochemical presentation.