Participants with suspected mitochondrial disorders were recruited to the 100 000 Genomes Project (main programme) between 2015 and 2018 with an unexplained multisystem progressive disorder usually involving the central nervous system, the neuromuscular system, or both. All participants provided written informed consent. Eligibility criteria stated that mtDNA and common nuclear genetic causes (eg, POLG mutations) should have been excluded (see supplementary methods for full inclusion criteria). All participants recruited under the category “suspected mitochondrial disorder” who had tiering data available in data release v8_2019-11-28 were included in the study. In 2015-18, genetic testing in the UK was arranged through 20 regional genetics laboratories, and there were three NHS highly specialised services for rare mitochondrial disorders. Testing of POLG and common mtDNA mutations (m.3243A >G associated with mitochondrial encephalomyopathy with lactic acidosis and stroke-like episodes and maternally inherited diabetes and deafness, m.8344A >G associated with myoclonic epilepsy with ragged red fibres, and m.8993T >G/C associated with Leigh syndrome and neurogenic muscle weakness ataxia and retinitis pigmentosa) in DNA extracted from blood was available through the highly specialised service laboratories. Further testing (such as gene panel testing) was available after discussion with the highly specialised services depending on the patient’s clinical features (supplementary table A).

The referring clinician provided clinical information against a standardised list of clinical features as yes/no answers (supplementary table B) or by using a standardised vocabulary for describing clinical features encountered in human disease (the Human Phenotype Ontology (HPO)).23 HPO classifies clinical features by organ system by using a branching tree incorporating increasing levels of detail. We used modified HPO terms to score participants using the Nijmegen Mitochondrial Disease Criteria2425(supplementary table C), which use clinical features (muscular, central nervous system, and multisystem), magnetic resonance imaging and biochemical features, and muscle biopsy results to give a score out of 12. We classified patients with a total score of 0-1 as being unlikely to have a mitochondrial disorder, those with score 2-4 as a possible mitochondrial disorder, those with score 5-7 as a probable mitochondrial disorder, and those with score 8 or more as definite mitochondrial disorder. One investigator scored all the participants. A second investigator also scored 100 participants independently to develop the modifications for using the Nijmegen Mitochondrial Disease Criteria with HPO terms. For each HPO term, the HPO system was determined by following the branching tree back to the subtype of “phenotypic abnormality.”

We extracted DNA from peripheral blood and quantified and sequenced it according to a national specification (Illumina TruSeq, HiSeq 2500, and HiSeq X).2627 We aligned reads to the Genome Reference Consortium human genome build 37 (GRCh37) for the earlier participants recruited and GRCh38 for later participants by using Isaac Genome alignment software. Family based variant calling of single nucleotide variants and insertion-deletions for chromosomes 1-22 and X and the mtDNA used the Platypus variant caller,28 allowing joint variant calling for all family members and considering the sequence alignments from all family members together. Figure 1 shows the analysis workflow.

We analysed genomes in families and classified variants into four “tier” groups according to the probability of the variant being causative.21 Tier 1 included loss of function variants (nonsense variants, essential splice donor variants, and essential splice acceptor variants) and de novo missense or splice region variants in genes on the panels applied. Tier 2 included missense and splice region variants in genes on the panels applied. Tier 3 included other rare variants, and a final group of unclassified variants (tier 4) had higher population frequency or the segregation pattern in the family was not consistent with the family history. We chose virtual gene panels according to each participant’s phenotypes, using curated “PanelApp” gene lists, which include causative genes for each disorder generated through crowdsourcing.29 This allowed the prioritisation of variants likely to be causative and minimised the reporting of abnormal genetic results that are unrelated to the reason for testing—for example, cancer predisposition genes (referred to as incidental findings). All participants had the mitochondrial panel applied and further panels depended on the phenotypes. Tier 1-3 variants were accessed from the Main Programme v8_2019-11-28. All tiered variants had passed in-house Genomics England quality control.30 We also prioritised variants by using Exomiser,31 an application that prioritises variants in exome or genome data by using protein interaction networks, clinical relevance, and cross species phenotype comparisons, as well as computational filters for variant frequency, predicted pathogenicity, and pedigree information. We extracted variants classified as pathogenic, likely pathogenic, or pathogenic/likely pathogenic from Clinvar32 (3.3.2020) for GRCh38 and GRCh37 and used bedtools intersect (https://bedtools.readthedocs.io) to compare them against tier 1-3 variants to identify previously reported pathogenic and likely pathogenic variants.

We applied the mitochondrial disorders panel in all participants. Other panels applied included undiagnosed metabolic disorders (148 participants), intellectual disability (139), congenital myopathy (77), and hereditary ataxia (60) (see supplementary table D). The mean number of panels applied per participant was 4.7 (range 1-18). A total of 93 different panels were applied to between one and 345 participants.

We detected copy number variant calls by using Canvas software,33 on the basis of sequence coverage and both nucleotide and insertion-deletion variant calling. We included only copy number variant calls that had PASS filter status assigned by Canvas and overlapped any gene in PanelApp. We used gencode v29 to annotate copy number variant calls. We evaluated copy number variants that interrupted exons of a PanelApp gene. We manually confirmed all the copy number variants reported as a diagnosis on Integrative Genomics Viewer.

We used the ExpansionHunter version 3.2.2 software package for short tandem repeat expansion genotyping.34 We assessed 13 loci (HTT, AR, ATN1, ATXN1, ATXN2, ATXN3, ATXN7, CACNA1A, ​TBP, ​C9orf72, FXN, FMR1, and DMPK) by using the coordinates listed in supplementary table E. We visually inspected potentially causative repeat expansions above an established threshold for each locus (supplementary table F) by using pileup plots and re-classified them on the basis of the quality of the reads.34

We used an in-house pipeline to call mtDNA single nucleotide variants above an established detection threshold of 1% variant allele frequency (or percentage heteroplasmy level),35 after excluding likely errors.27 We compared these with a manually curated list of pathogenic mutations with functional evidence supporting pathogenicity.36 Our pipeline does not detect large scale mtDNA rearrangements, which usually require targeted mtDNA analysis in DNA extracted from skeletal muscle.

A clinical geneticist reviewed tier 1 and tier 2 variants, the top 10 prioritised variants by Exomiser, ClinVar pathogenic/likely pathogenic variants, and the mtDNA variants (tier 1-3 and from the in-house pipeline) and classified them using internationally accepted criteria for pathogenic variants, likely pathogenic variants, variants of uncertain clinical significance, and likely benign or benign variants (American College of Medical Genetics criteria37), incorporating information from gnomAD,38 ensembl,39 VarSome,40 OMIM, and a review of the literature (see supplementary methods for a description of the online resources used). Variant quality and variant allele frequency were checked using Integrative Genomics Viewer.41

We reviewed copy number variants that overlapped at least one exon in a PanelApp gene in DECIPHER. We reviewed information about the type of variant (copy number loss, copy number gain, or loss of heterozygosity), the size of the variant, and the gene content (in particular, haploinsufficient genes and OMIM morbid genes). We also made a comparison with previously observed copy number variants in the general population or in affected individuals.

We reviewed participants with repeat expansions to see whether these were causative, considering the size of the expansion (whether it was a “pre-mutation” or “full mutation”) and the participant’s age and phenotypic features.

The Genomic Medicine Centre laboratories were commissioned to analyse the tier 1 and tier 2 variants. They filled out an “exit questionnaire” for each family through an online questionnaire. For each family, they fed back whether genetic cause had been identified, with options of “solved,” “partial,” “uncertain,” or “no,” and whether the result was already known through clinical testing. For variants that the clinical scientists had identified as potentially causative, they fed back their variant classification according to the American College of Medical Genetics criteria and whether or not the variant had been confirmed by Sanger sequencing and fed back to the clinician. Feedback from the Genomic Medicine Centre laboratories was available in 277/319 (87%) families (data release v12_21_05_06). We successfully contacted clinicians for further information in 55 (17%) families.

We described the molecular diagnosis as “definite,” “probable,” or “possible” on the basis of our overall clinical assessment of whether the variant(s) explained the clinical features, taking into account the American College of Medical Genetics classification of the variant(s), the inheritance pattern, and the clinical fit between the patient’s HPO terms and the reported clinical phenotypes for the gene or variant. For example, we used “probable” for compound heterozygotes for which the phenotype fitted with the disorder, but we classified only one variant as pathogenic/likely pathogenic and the other was a variant of uncertain clinical significance. We described the contribution as “full” or “partial” depending on whether the whole phenotype or only one aspect could be explained by the variant(s). We classified nuclear genes as “mitochondrial” on the basis of the curated PanelApp list,29 plus recently discovered genes known to have a direct effect on oxidative phosphorylation.

Statistical analyses were performed in R. We used Fisher’s exact test to compare the number of participants with a genetic diagnosis between children and adults and between singletons and trios/quads. We used Student’s t test to compare the mean number of HPO terms in individuals with and without a diagnosis and the number of HPO systems affected. We used Fisher’s exact test to compare the proportion of families with each inheritance pattern between nuclear-mitochondrial diagnoses and non-mitochondrial disorders. We compared the proportion of participants with each HPO system affected between participants with definite mitochondrial diagnoses and definite non-mitochondrial diagnoses, using Fisher’s exact test. We compared the mean Mitochondrial Diagnostic Criteria score between participants with mitochondrial diagnoses, non-mitochondrial diagnoses, and no diagnosis by using a one way analysis of variance and post-hoc Tukey testing.

The 100 000 Genomes Project has a Participant Panel made up of participants and parents or carers of people involved in the project, which was established in 2016. The panel meets with senior staff from Genomics England and NHS England four times a year. They are asked about project design and help to ensure that participants’ health data are looked after with respect and used in the best interests of the participants. Panel members sit on other committees including the Access Review Committee and the Ethics Advisory Committee. Patients and carers were involved in developing the consent literature and the infographics and information for the public.

Three hundred and forty five affected individuals (186 females, 159 males) were referred with a suspected mitochondrial disorder after exclusion of common causes. They were from 319 families of different reported ethnicities across England (fig 2). Genomic data were available for more than one affected member in 25 families (15 sibling pairs, eight mother and child pairs, one father and child pair, and one mother with two affected children). The median age at enrolment for probands was 25 (interquartile range 10-54; range 0-92) years; 143 (41%) were aged ≤18 at enrolment. No participants withdrew or were lost to follow-up.

Phenotypic data were available for 341 participants (missing in four participants). A total of 3095 HPO terms were recorded, with a median of 7 (range 1-39) HPO terms per participant; 806 different HPO terms were used. Figure 3 shows the most common clinical terms, investigation result terms, and HPO systems affected. A median of 4 (range 1-13) HPO systems were affected per participant. Application of the HPO modified Nijmegen score gave a mean total score of 4.30 (range 0-10), with 24/345 (7%) participants classified as being unlikely to have mitochondrial disease (score 0 or 1), 193/345 (56%) as Nijmegen possible (score 2-4), 95/345 (28%) as Nijmegen probable (score 5-7), and 33/345 (10%) as Nijmegen definite (score of ≥8).

We identified the definite or probable genetic diagnosis in 98/319 (31%) families, with a possible diagnosis in an additional six (2%) (table 1; table 2; fig 1; fig 4). A definite genetic diagnosis was reached in 28% (89/319), including 14 (4%) genetic diagnosis that provided only a partial explanation for their clinical features. Nine (3%) families had a probable diagnosis, and six (2%) had a possible diagnosis. One participant was diagnosed as having two disorders that together fully explained the combination of phenotypes. Most diagnoses (69/104; 66%) came from single nucleotide variants or small insertion-deletion variants on the panels applied (tier 1 or tier 2 variants). An additional 19/104 (18%) diagnoses were made through other analyses of the single nucleotide variants and small insertion-deletion variants (from the Genomic Medicine Centre laboratory reports, comparison with ClinVar, and Exomiser). The copy number variant analysis added eight (8%) diagnoses, and the short tandem repeat analysis added three (3%) diagnoses. Five (5%) diagnoses were made through the mtDNA analysis. Further details of the variant analysis are shown in supplementary table G.

The overall diagnostic rate was 59/186 (32%) in females and 51/159 (32%) in males. Sixty four (45%) of 143 participants recruited aged 18 or under received a diagnosis (any type), compared with 46/202 (23%) participants over the age of 18 (P<0.001). Considering definite diagnoses only, 50/143 (35%) participants under 18 received a definite molecular diagnosis compared with 30/202 (15%) adults over the age of 18 (P<0.001). Figure 5 shows the age profile for diagnosis. Although diagnoses were more frequent in younger people, diagnoses were still being made in participants who were in their 70s or 80s at enrolment.

The diagnostic yield per family (all diagnoses) was 23% (23/102) for singletons, 29% (12/42) for duos with a parent, 42% (62/148) for trios/quads, and 26% (7/27) for other family structures. The diagnostic rate in trios/quads was higher than in singletons (P=0.005). The mean age of singletons was 56 years, compared with a mean age of 12 years for probands in trios/quads.

The mean number of HPOs in participants with any diagnosis (10.0) was slightly higher than in patients with no diagnosis (8.48) (P=0.03). The mean number of HPO systems affected in patients with a diagnosis (4.67) was not significantly different from the mean number of HPO systems affected in those with no diagnosis (4.61) (P=0.82).

Of the 75 families with a definite diagnosis that fully explains the phenotype, 28 (37%) were in genes known to cause primary mitochondrial disease, including four mtDNA variants and 24 diagnoses in nuclear-mitochondrial genes, whereas 47 (63%) were in non-mitochondrial genes. For families with any genetic diagnosis (including probable or possible diagnoses and partial explanations), 39/104 (38%) were in genes known to cause primary mitochondrial disease, including six mtDNA variants and 30 nuclear-mitochondrial diagnoses, and 65 (63%) were in non-mitochondrial genes. These non-mitochondrial diagnoses included a wide variety of disorders, including developmental disorders with intellectual disability, metabolic disorders, epileptic encephalopathies, Bardet-Biedl syndrome, cardiomyopathies, MYH2-related myopathy, and amyloidosis.

Potentially treatable disorders were identified in six participants with a mitochondrial disorder and nine participants with a non-mitochondrial disorder (shown in supplementary table H). However, the evidence base is weak for several of these rare disorders.42

One hundred and seventeen (34%) participants had HPO terms relating to muscle biopsy abnormalities, and 73 (21%) had abnormal mitochondrial complex activities. Of patients with definite mitochondrial diagnoses, 15/28 (54%) had muscle biopsy abnormalities, 11/28 (39%) had abnormal respiratory chain complex activities, and two had abnormal pyruvate dehydrogenase complex activity. Of patients with definite non-mitochondrial diagnoses, 14/47 (30%) had muscle biopsy abnormalities and 12/47 (26%) had abnormal respiratory chain complex activities. The patients with non-mitochondrial diagnoses and abnormal respiratory chain complex activities had pathogenic variants in ASXL3, CACNA1E, CTBP1, EXOSC3, HK1, KCNT1, NPHP1, P4HTM, PPP2R5D and SCN2A, SOS1, and TANGO2.

Our study had several strengths, including the large sample size and the fact that patients were recruited nationally from both secondary and tertiary care, meaning that the findings have a wider relevance than those of studies focused on highly selected groups identified by specialist centres. In addition, the use of HPO terms allowed us to analyse phenotype data in a systematic way, and we contacted clinicians for detailed phenotypic information in selected participants. Our analysis of the nuclear genes included copy number variants and short tandem repeat expansions in addition to single nucleotide variants and small insertion-deletion variants, and we also studied mtDNA variants with a heteroplasmy level >1%. The advantage of using whole genome sequencing rather than whole exome sequencing is that it is easier to pick up copy number variations, repeat expansions, and lower level heteroplasmies. Weaknesses are that we have not explored novel disease genes or variants in the non-coding regions, other than previously published variants, and tracing all family members was not possible.

The previously published study using whole genome sequencing in patients with suspected mitochondrial disease looked at a cohort of 40 children recruited from four paediatric genetic metabolic centres in Australia.19 Of these, 34/40 had a probable or definite mitochondrial disorder according to the modified Nijmegen Mitochondrial Disease Criteria score (between 5 and 12) and 28 had abnormal respiratory chain enzyme activities. Analysis of nuclear and mtDNA enabled a definitive genetic diagnosis for 55% of patients, and a likely molecular diagnosis in 67%, with 18% having a non-mitochondrial disorder. The higher diagnostic yield in this study likely reflects the section criteria, which focused on children within trios identified by national specialist clinics,19 compared with our more inclusive recruitment criteria whereby around two thirds of our probands were adults and only 46% of families were recruited as trios.

Several studies using whole exome sequencing in patients with suspected mitochondrial disorders have been reviewed recently.46 Most of these studies were conducted in highly selected patients seen in specialist centres,101216 focusing on childhood onset disorders (apart from Wortmann and colleagues’ study,14 which included patients up to 27 years of age). Their stringent recruitment criteria led to higher diagnostic yields (57-68%) and lower numbers of non-mitochondrial diagnoses compared with our study. Previous exome studies of larger cohorts have also tended to have a lower diagnostic yield (35-39%),1415 which is thought to be more reflective of everyday clinical practice.46 Non-mitochondrial disorders have been found in some of the previous studies, but in much smaller numbers than in our study. The recruitment criteria in our study were broad, which is reflective of mainstreaming of genomic medicine, and we studied both adults and children. Most participants (64%) did not have muscle biopsy results available at the time of referral, whereas most participants in previous studies had evidence of mitochondrial dysfunction. The more inclusive eligibility criteria have led to us finding a wide range of different genetic diagnoses. The finding that a large number of patients had non-mitochondrial disorders is important because these diagnoses would have been missed if the participants had been investigated only for mitochondrial disorders through muscle biopsy, a specific mitochondrial gene panel, or both. This would have led to missed opportunities for treatment, surveillance, and reproductive management. Our findings highlight the difficulty of diagnosing these rare multisystem disorders clinically and the need to keep an open mind about the differential diagnosis.

Considering other studies of integration of whole genome sequencing into healthcare, Stranneheim and colleagues describe the results for 3219 rare disease patients recruited in Stockholm, Sweden, between 2015 and 2019.47 Their study involved much more direct collaboration between academia, healthcare, and their SciLifeLab (which provides whole genome sequencing), which were all located in the same city. This meant that potential diagnoses were discussed and specialist advice was fed back quickly into clinical practice. Specialist clinicians could also access and analyse the data together with clinical scientists. The 100 000 Genomes Project used a different model with the bioinformatics managed centrally and the clinical interpretation of variants done by genomic medicine centre laboratories. Researchers were able to access pseudonymised data. The limited ability of researchers and clinicians to discuss potential diagnoses together and the long turnaround time were disadvantages compared with the Swedish model.

The participants in this study had standard of care NHS genetic testing before enrolment. The use of whole genome sequencing as a first genetic test for suspected mitochondrial disorders has not been directly explored. The only mitochondrial diseases that cannot be diagnosed by whole genome sequencing of DNA extracted from blood are extremely rare muscle specific mtDNA mutations and large mtDNA deletions, which usually have a very specific clinical phenotype. On the basis of known epidemiology,148 these muscle specific mtDNA mutations and deletions account for approximately 11.5% of patients with a genetically confirmed mitochondrial disorder. Therefore, mitochondrial disorders can be diagnosed by whole genome sequencing of DNA extracted from blood in nearly 90% of patients.49 Additionally, whole genome sequencing can diagnose other monogenic disorders that have similar clinical features, so we would expect a very high diagnostic yield. Finally, by carrying out this study we have identified a cohort of patients with suspected mitochondrial disease in whom the likely diagnosis is contained within the whole genome sequencing data but cannot be distinguished from background sequence variation. This will form a useful resource for the discovery of future mitochondrial disease genes that are not in the coding space and thus are not detectable by whole exome sequencing. Large scale mtDNA rearrangements are not reliably detected in DNA extracted from blood in adults, meaning that patients with suspected progressive external ophthalmoplegia and Kearns-Sayre syndrome are investigated by muscle biopsy.20 Future research will determine whether deep mtDNA sequencing will detect these large scale mtDNA rearrangements in DNA extracted from blood or another accessible tissue such as urinary epithelium.

Our findings indicate that whole genome sequencing is a useful diagnostic test in patients with suspected mitochondrial disorders recruited from secondary and tertiary care settings. We recommend that whole genome sequencing should be offered early in the diagnostic pathway in a patient’s local secondary or tertiary care centre and before invasive tests such as a muscle biopsy. Exceptions to this would be patients whose clinical features are highly suggestive of a specific cause that can be confirmed by a single gene test or common mtDNA mutation testing, as well as patients with progressive external ophthalmoplegia, which is diagnosed in most patients by testing mtDNA from a muscle biopsy sample for large scale rearrangements. Referral to a specialised mitochondrial clinic should be considered if whole genome sequencing is uninformative. Further investigations likely to increase the diagnostic yield beyond whole genome sequencing include laboratory studies of mitochondrial function and other “omics” approaches including transcriptomics, which provided an additional diagnosis in 10% of patients with suspected mitochondrial disease in one study,50 proteomics, and metabolomics.

The integration of whole genome sequencing into healthcare also has wider policy implications. The relatively high number of patients with probable or possible diagnoses partly reflects the lack of capacity for the functional evaluation of variants of uncertain clinical significance—for example, through splicing assays or Western blotting. Resources should also be made available for the regular reanalysis of whole genome sequencing data, either at specified time intervals in patients without a diagnosis or on a clinician’s request. The reanalysis of exome data has been shown to significantly increase diagnostic yield, mainly owing to newly discovered disease genes.51 Finally, rapid trio whole genome sequencing should be offered in acutely unwell individuals with suspected mitochondrial disorders,17 so that the results can help to guide clinical management. In the UK, rapid trio exome sequencing is available only for acutely unwell children.