Mitochondrial disorders are the most common group of metabolic disorders, with an estimated prevalence of 1 in 5,000 1. Mitochondrial disorders may present with any symptom, in any organ, at any age and with any mode of inheritance 2. These often devastating disorders are clinically characterized by multi-system involvement with primarily progressive neurologic disease and myopathy, including both skeletal and cardiac muscle. The variability in clinical presentation and underlying causative mutations make the diagnosis very challenging, involving extensive clinical and specialized laboratory evaluation 3. However, no reliable diagnostic screening or biomarker is available that is both sensitive and specific in all cases of mitochondrial disorders 4. In current clinical practice, the diagnosis of mitochondrial disease relies heavily on identifying deficient activity of one or more of the mitochondrial respiratory chain enzymes, but in many cases the enzyme activity is found to be only moderately decreased or even normal, which makes the interpretation very difficult. Additionally, there are intrinsic problems in the biochemical characterization of mitochondrial disorders, such as variability in tissue manifestation, difficulty in establishing realistic normal reference ranges, inability of enzyme assays to detect some functional defects, variations in assay protocols, no uniform or standardized guidelines, and lack of widely accepted diagnostic criteria and a quality assurance scheme 5. For these reasons, some patients may remain undiagnosed and even die of untreated disease. Early and definitive diagnosis is crucial for permitting appropriate management and accurate counseling 3. Thus, a new simplified and reliable approach for the diagnosis of mitochondrial disorders with better accuracy and precision has been strongly advocated.

Mitochondria are cellular organelles with numerous essential functions, such as production of energy, metabolism of amino acids, fatty acids, and cofactors, and cell signaling. Their biogenesis and function is under the genetic control of mitochondrial DNA (mtDNA) and nuclear DNA. The number of mitochondrial proteins encoded by nuclear genes is estimated to be around 1,500 6, constituting 99% of mitochondrial proteins 7. mtDNA contains 37 genes encoding 13 respiratory chain subunits, 2 rRNAs and 22 tRNAs 8. Because of this dual genetic control, mitochondrial disorders can originate from mutations in either mtDNA or nuclear genes that encode the organelle proteins 8.

Mutations have been found in approximately 170 nuclear genes in patients with mitochondrial disorders 689. However, many nuclear genes causing disease are still unknown 8. It is expected that most mitochondrial disorders are caused by mutations in nuclear genes. The nuclear genes encode the subunits of the complexes involved in oxidative phosphorylation and relative assembly factors, proteins controlling the synthesis and stability of mtDNA, mitochondria transcription and translation, biogenesis, metabolism, and signaling. While substantial progress has been made in recent years in identifying nuclear genes that are mutated in mitochondrial disorders, the key clinical challenge lies in determining which one of hundreds of genes is responsible for the disease in any given patient. Comprehensive sequencing of all nuclear genes known to be involved in mitochondrial disease would be cost-prohibitive and time consuming using traditional DNA sequencing technology. Not surprisingly, clinical tests are available for only a limited number of nuclear genes for a few conditions in which the causative genes can be predicted by the clinical phenotypes 8.

Pathogenic mutations in mtDNA are found in at least one in 5,000 affected individuals 1 and appear to be very common in the general population (>1 in 200 live births) even though these mutations show different penetrance during the lifetime of the carriers. Many of these mutations are primarily responsible for adult-onset mitochondrial disorders 1. Mutations in nuclear genes are likely the major cause of mitochondrial disease particularly in pediatric cases 3. Nevertheless, the pediatric prevalence of mtDNA mutations may have been underestimated because mtDNA testing is typically performed by targeted mutation analysis. This strategy cannot identify mutations beyond those targeted 10. While mtDNA (16,569 bp) could easily be sequenced by traditional Sanger methods, this technology is inadequate to detect some mtDNA mutations that occur in a small fraction of the total mtDNA molecules. Indeed, mutated molecules of mtDNA coexist with normal mtDNA (heteroplasmy) and can be below the limit of detection of Sanger sequencing, especially in DNA extracted from blood 11.

Within the last few years, next-generation sequencing has been tested for whole genome or targeted resequencing with promising results. The new platforms allow sequencing hundreds of genes in parallel and detection of mutations or alterations with a dramatically reduced cost. Over the coming years, next-generation sequencing is highly anticipated to transition from basic research applications into clinical diagnostics 121314. One such opportunity is the rapid identification of mutations in diseases that can be caused by one of several genes, as with mitochondrial disorders.

Despite the fact that whole genome or exome sequencing is now possible, it is still desirable to limit the analysis to genes responsible for a certain condition for both cost benefit and time saved for analysis and interpretation. Given that most pathogenic mutations are typically located in coding regions or at intron-exon boundaries, and that it is not practical to use PCR to enrich a large number of exons, several methodologies have been developed to enrich exons of target genes as a preliminary step to next-generation sequencing 151617181920. Compared with whole genome/exome sequencing, this enables a major reduction in cost and allows higher sequence coverage over the areas of interest.

Here, we propose to develop a comprehensive clinical diagnostic tool based on sequencing the entire mtDNA genome, and the exons of previously implicated and candidate nuclear genes (Table 1) using sequence capture technology coupled to next-generation sequencing.

Positive control patient samples were obtained as anonymous samples from Seattle Children's Hospital; these were leftover specimens after routine standard clinical testing. Mutations in these samples were previously identified by clinical tests using traditional sequencing. Patient 1 was Caucasian and Patient 2 was of Caucasian-Native American origin. Human DNA from one HapMap individual was obtained from Coriell Repositories (NA18517, Yoruba ancestry). We used 10 μg of DNA per individual for these studies.

Custom programmable arrays (Agilent Technologies Inc.) were designed with 60-mer oligonucleotide probes complementary to the sequences to be captured. The target consisted of the entire mtDNA genome and coding sequences within >3,500 exons of 362 nuclear genes for proteins involved in mitochondrial function, for an aggregate target size of approximately 0.6 Mb (Table 1) excluding repetitive regions. The Consensus CDS (CCDS) database was utilized to obtain the exon coordinates for probe design. Due to discrepancies between identifiers used by our group and CCDS or because CCDS is not comprehensive, a small number of genes were inadvertently excluded from the array design process. These included, for example, Polymerase gamma 1 (POLG1), a nuclear gene involved in mitochondrial disorders 21 (annotated as POLG in CCDS), and highlights the importance of careful review in the design process.

As there are approximately 244,000 programmable oligos on the custom arrays used here, the targeted sequences were 'tiled' at a very high density (that is, 40 probes per 100-bp interval; probe sequences are available upon request). To construct an in vitro shotgun sequencing library, genomic DNA was sheared by nebulization and universal adaptor oligonucleotides were ligated and then amplified using the Illumina protocol 22. After this step, in order to enrich for the specific target exons and mtDNA, the amplified shotgun libraries were hybridized to the capture array as described in 23. After washing to remove unhybridized material, captured molecules were recovered by heat-based elution and subjected to PCR amplification. The target-enriched shotgun libraries were quantified (NanoDrop Products, Wilmington, DE, USA), and then subjected to deep sequencing on an Illumina Genome Analyzer, GAII. One lane of the flow cell was used for each sample. Read-lengths of up to 36 bp were obtained with per-base accuracies on the order of 99%. The sequence reads were aligned to the human reference genome, first using the standard Illumina package (ELAND). After removal of all but one of the reads mapping with identical coordinates and orientation (potential PCR duplicates), the remaining reads were remapped using the MAQ software package 24. Consensus calls for variant identification were also carried with MAQ.

In order to assess the significance of new variants found in the study, we analyzed the non-synonymous single nucleotide substitutions with PolyPhen (Polymorphism Phenotyping), a tool that predicts the possible impact of an amino acid substitution on the structure and function of a human protein using physical and comparative considerations 25.

We have developed an assay to streamline the molecular diagnosis of mitochondrial disorders by simultaneous sequencing of the entire mtDNA genome and the exons of 362 nuclear genes for targeted mitochondrial proteins. The current list of targeted genes includes 104 nuclear genes for which the causative mutations were previously found in various symptomatic patients 68213637383940, the entire mtDNA genome, and 258 additional nuclear genes potentially involved in mitochondrial disorders but that were never reported in patients due to either no attempts to sequence them or lack of clinically available testing (Table 1). The known/candidate genes include all of the structural components of oxidative phosphorylation complexes, as well as other mitochondrial proteins of the following functional groups: respiratory complex assembly factors, transcription and translation factors, enzymes, and carrier proteins. Some of the genes causing secondary inhibition of the mitochondrial respiratory chain are also included in this panel. One criterion for inclusion in the list of candidate genes was that members of each group had already been implicated in mitochondrial disease. Some candidate genes were recently reported as components of mitochondrial respiratory complexes by proteomics 635 or identified as candidate genes by integrative genomics 641. Since we first compiled the list of putative genes, three were identified as causing mitochondrial disease in patients (C20orf7, CoQ9 and NDUFAF3 414243). This encourages us to interrogate candidate genes in suspected patients with unknown molecular defects.

In order to build a cost-effective but comprehensive diagnostic approach, we performed multiplex capture of the regions of interest using patients' DNA followed by sequencing with an Illumina Genome Analyzer. Considering that the majority of pathogenic mutations are in coding regions or at intron-exon boundaries, we restricted capture and sequencing to these subsequences in genes of interest. The total target size is approximately 0.6 Mb for the exons of the 362 nuclear genes and 16.6 Kb for the entire mtDNA genome. This strategy allows circumventing the high costs of PCR and conventional sequencing for a large number of targets, while maintaining high sensitivity and specificity for detection of potentially pathogenic variants. Coverage of ≥8× and a consensus quality score ≥20, which in our experience allows reliable variant calling 23, was observed for 96%, 94% and 94% of target bases in the nuclear genome in the HapMap, patient 1 and 2 samples, respectively. Normal and patient DNA samples with known pathogenic mutations were tested blindly. All known mutations in two different genes in the patients' DNA samples were identified correctly. The common mutation R263G in the X-linked gene PDHA1, which encodes a subunit of the Pyruvate dehydrogenase complex, was identified in the patient 1 sample. The observed mutation in PDHA1 has been described in patients with Leigh syndrome 44, a condition characterized by extensive genetic heterogeneity, since it can be due to mutations in several genes (OMIM 256000) and is thus a paradigm for the utility of the proposed assay. The mutations in the patient 2 sample affect HADHA, also called long-chain hydroxyacyl-CoA dehydrogenase (LCHAD). Patient 2 is a compound heterozygote of a novel mutation affecting the G nucleotide of the conserved splicing acceptor site 26 at the 5' end of exon 5 and the common mutation E510Q 27. Several polymorphisms in mtDNA were identified and the depth of sequencing coverage was extremely high, indicating that it will be feasible to detect pathogenic mtDNA mutations in the presence of low level heteroplasmy undetectable with Sanger sequencing. While validation with a larger panel of positive controls for nuclear and mtDNA mutations is needed, this approach appears highly promising since approximately 95% of the targeted regions of 362 nuclear genes were sequenced and the results for known mutations were 100% concordant. The high sensitivity of this method as well as the power to identify gene-disease relationships has been well demonstrated in a whole exome sequencing study 23.

While the mutations in the analyzed samples were known at the offset, this study exemplifies the necessity to interpret and validate with traditional sequencing potentially pathogenic new variants identified in patients with unknown molecular defects. However, our results indicate that the number of new variants is not as high as we anticipated. Indeed, of the variants identified in the samples, 90 to 94% were present in dbSNP while 6 to 10% represented new variations. Most of the non-synonymous new variants were predicted to be benign when analyzed with PolyPhen. Only few of the variants were predicted as possibly or probably damaging. However, a review of the literature or alignment to orthologues indicates that these may be tolerated changes. Moreover, after filtering these variants with the new variants identified in normal samples by exome sequencing 45, only three variants in the patient 1 sample and two variants in the patient 2 sample would have required careful interpretation.

In summary, we anticipate that with more data on individual genomes/exomes, the panel of polymorphisms present in the population will grow, thus reducing the need to interpret new variants and the extent of traditional sequencing to confirm the variants. While the use of prediction tools and analysis of the literature on the affected proteins can provide a relatively easy way to assess the significance of the new variants, integrated bioinformatic support seems very important for the successful implementation of next-generation sequencing in the clinical arena.

A variable range of coverage was achieved across the targeted areas and, for this reason or because of the challenges of mapping short reads to the human genome, a small portion of targets (4 to 6%) was not sufficiently covered by sequence reads for variant calling. This aspect will definitely require further improvements that may be achieved with modifications to capture as well as other steps. However, given the number of target genes analyzed and the lack of clinical testing, our initial results are highly encouraging. While it is ideal to achieve coverage for all the targeted regions, this may not be realistic as some regions may be refractory to capture, amplification or sequencing. We plan to expand the target pool to include additional known and candidate genes 9 and also to test the performance of other capture systems 1646.

A final consideration concerns the applicability of the technique in a clinical setting based on ease of workflow and economic aspects. The sample preparation and set-up of the sequencing runs, while requiring expert handling, is fairly straightforward. Only one sequencing lane was utilized per sample and up to eight samples can be analyzed in one run. Several aspects of the procedure are rapidly improving, allowing increases in sequence output, sample multiplexing, and better data analysis, which will certainly enable a cost-effective approach to the diagnosis of several complex genetic diseases.

Our data demonstrate that the use of next-generation sequencing holds great promise as a tool for screening mitochondrial disorders in patients. The availability of a diagnostic test will provide opportunities to identify patients early in life, eliminate lengthy and often invasive procedures, and provide life-saving therapies, permitting prompt management and accurate genetic counseling. Furthermore, the ability to diagnose patients will stimulate the development of new targeted therapies based on the known genetic defect. We expect that the analysis of samples from patients with uncharacterized molecular defects will allow the discovery of novel mutations in the targeted candidate genes, thus expanding and redefining the spectrum of mitochondrial disorders.

CCDS: Consensus CDS; mtDNA: mitochondrial DNA.

The authors declare that they have no competing interests.

VV conceived of the study, carried out the sample preparation, analyzed data and drafted the manuscript. SBN and EHT designed reagents and protocols and performed experiments. JS analyzed the data and helped to draft the manuscript. SH conceived of the study, analyzed the data and helped to draft the manuscript. All authors read and approved the final manuscript.