From 1363 genomes of cardiomyopathy patients, only 172 had been previously solved by the standard/clinical approach. As described in more detail elsewhere [4], this approach involved the following steps: WGS (Illumina TruSeq, HiSeq 2500) via the 100KGP. Human Phenotype Ontology (HPO) terms extracted from notes and submitted at recruitment. Based on these, gene panels were applied, constructed via “PanelApp”, which classifies genes as “green” (diagnostic), “amber” (borderline evidence), and “red” (insufficient evidence). Only “green” genes were included in the clinical analysis. For further prioritisation, a clinical scientist reviewed all tier one (loss-of-function variants and de novo protein altering variants in virtual panels applied) and tier two (non-loss-of-function protein altering variants in virtual panels applied) variants. “Likely pathogenic” or “pathogenic” variants were confirmed with Sanger and reported; otherwise a report stating there were “no primary findings” was released.

For our analysis, the bioinformatics pipeline, MitoHPC (https://github.com/ArkingLab/MitoHPC) was used to call mtDNA variants in 1363 genomes of cardiomyopathy patients. MitoHPC’s key feature is that it constructs a consensus mitochondrial sequence for each sample, and calls heteroplasmies against an individual unique mitochondrial genome, which greatly improves heteroplasmy estimates. MitoHPC performs two iterations of variant calling using mutect2; firstly to identify major alleles, or homoplasmies, which are used to construct the consensus mitochondrial sequence, and secondly to call heteroplasmic variants. As MitoHPC also provides various quality metrics at both sample and variant levels (e.g. mitochondrial copy number estimation, haplogroup determination, sequencing coverage statistics), we have limited our analysis to samples that are not labelled as suspicious by MitoHPC. Each reported variant passed the variant quality filter. Resultant mtDNA variants were reviewed by a clinician and only those with a heteroplasmy level of > 10% and a plausible link to the phenotype were analysed further.

For the control population, we have retrieved via the Genomics England LabKey application all Rare Disease participants with Participant Type ‘rd_relative’ and Affection Status ‘Unaffected’. We then removed participants recruited as relatives of subjects from the cardiomyopathy cohort. We randomly selected from the resulting table (using ‘head’ function in R on the unsorted table with unique participants) samples to match in numbers our cardiomyopathy cohort. From the resulting dataset, we have further removed those with HPO terms “mitral valve prolapse”, “palpitations”, ”syncope” and “prolonged QTc interval”. IDs of all cases and controls used are accessible to registered users of Genomics England Research environment, and can be found in directory: /re_gecip/shared_all_GeCIPs/4300/.

This produced a set of 1329 controls. We have additionally compared the frequency of detected variants with published UK Biobank (UKB) data [5].

Pathogenicity annotations were assigned based on the MitoMap database (http://www.mitomap.org/), which curates mtDNA variants based on their link to disease.