MitoPhen database: a human phenotype ontology-based approach to identify mitochondrial DNA diseases

Abstract

Diagnosing mitochondrial disorders remains challenging. This is partly because the clinical phenotypes of patients overlap with those of other sporadic and inherited disorders. Although the widespread availability of genetic testing has increased the rate of diagnosis, the combination of phenotypic and genetic heterogeneity still makes it difficult to reach a timely molecular diagnosis with confidence. An objective, systematic method for describing the phenotypic spectra for each variant provides a potential solution to this problem. We curated the clinical phenotypes of 6688 published individuals with 89 pathogenic mitochondrial DNA (mtDNA) mutations, collating 26 348 human phenotype ontology (HPO) terms to establish the MitoPhen database. This enabled a hypothesis-free definition of mtDNA clinical syndromes, an overview of heteroplasmy-phenotype relationships, the identification of under-recognized phenotypes, and provides a publicly available reference dataset for objective clinical comparison with new patients using the HPO. Studying 77 patients with independently confirmed positive mtDNA diagnoses and 1083 confirmed rare disease cases with a non-mitochondrial nuclear genetic diagnosis, we show that HPO-based phenotype similarity scores can distinguish these two classes of rare disease patients with a false discovery rate <10% at a sensitivity of 80%. Enriching the MitoPhen database with more patients will improve predictions for increasingly rare variants.

Figure 1.

Confirmed pathogenic variants in the mitochondrial genome. Eighty-nine pathogenic mtDNA variants and the corresponding mtDNA gene positions. The number of published affected individuals for each variant is represented by outward-facing grey bars, with the number shown in brackets, and the number of publications reporting these individuals are represented as inward-facing grey bars. The mtDNA genes are colored according to coding category: pink represents coding genes, purple represents tRNA genes, green represents rRNA genes and the D-loop is shown in blue.

Figure 2.

Human Phenotype Ontology terms curated from 6689 individuals harboring 89 pathogenic mtDNA variants. (A) Histogram of the number of non-redundant HPO terms assigned to the probands. (B) Phenotypes of the probands affected by the 25 mtDNA variants carried by the greatest number of probands. Each box represents a different variant. Within each box, individuals are represented by vertical arrangements of colors corresponding to top level HPO terms that they have been assigned. (C) Heatmap of the fraction of probands carrying each of the 25 mtDNA variants who have abnormalities in each of the top level HPO terms. The histograms show the number of probands carrying each variant and having each phenotypic abnormality. (D) Heatmap of the phenotype similarity score comparing sets of probands carrying each of the 25 mtDNA variants. The rows and columns have been arranged by hierarchical clustering. For each of the three apparent clusters, the most common HPO term and the most specific terms in at least 50% of the probands in the cluster are listed.

Figure 3.

Heteroplasmy levels and associated phenotypes. (A) Scattergrams of heteroplasmy levels in muscle and blood. Variants are shown when there are at least three probands with data available. The Spearman correlation coefficient ρ is shown in each scattergram. Only eight out of 179 individuals had a lower heteroplasmy level in muscle than in blood. (B) Fitted logistic regression curves showing correlations between specific HPO terms and variant heteroplasmy levels (%) in muscle (red) or blood (blue) (in each case, P < 0.01). By permutation, we expected 1.86 P-values less than 0.01 under the null, yielding an expected false discovery rate of 15.5%. Note that two significant associations with more general HPO terms than those shown in the panel have been omitted (between m.3243A > G in muscle and both ‘Ophthalmoplegia’ and ‘External ophthalmoplegia’).

Figure 4.

Variant-specific enrichment of HPO terms. HPO terms exhibiting strong evidence of enrichment in probands carrying particular variants are shown. Each arrow leads from the frequency (as a proportion of 1) of the term amongst all probands in the MitoPhen database who do not carry the corresponding variant (arrow tail) to the frequency amongst those carrying the variant (arrow head). For example, for m.11778A>G, nearly 100% of individuals in the MitoPhen database carrying this variant have 'Leber Optic Atrophy' listed as a phenotype, and 25% of individuals in MitoPhen who do not harbour m.11778A>G also have 'Leber Optic Atrophy' listed, reflecting the other mtDNA variants known to cause this disorder that are included in the database. Asterisks are used to denote under-recognized terms, specifically those that are not listed by Orphanet as associated with the syndrome caused by the variant.

Figure 5.

Predicting the cause of rare disease in patients using phenotypic similarity to MitoPhen. (A) Distribution of causal variants in the independent mtDNA disease cohort of 77 individuals. (B) Distribution of phenotypic similarity scores of each cohort, where the non-mitochondrial disease cohort was partitioned into neurodevelopmental diseases and other diseases caused by nuclear genetic mutations. (C) False discovery rate for predicting mtDNA disease, thresholding phenotypic similarity to achieve a given sensitivity. FDR was estimated separately for the neurodevelopmental and non-neurodevelopmental sections of the non-mitochondrial disease cohort with known nuclear genetic diseases. Bootstrap sampling was used to achieve a mixture of 10% mitochondrial DNA disease for each estimated FDR. The vertical grey line indicates a sensitivity of 0.8.