A mitochondrial protein compendium elucidates complex I disease biology

Abstract

Mitochondria are complex organelles whose dysfunction underlies a broad spectrum of human diseases. Identifying all of the proteins resident in this organelle and understanding how they integrate into pathways represent major challenges in cell biology. Toward this goal, we performed mass spectrometry, GFP tagging, and machine learning to create a mitochondrial compendium of 1098 genes and their protein expression across 14 mouse tissues. We link poorly characterized proteins in this inventory to known mitochondrial pathways by virtue of shared evolutionary history. Using this approach, we predict 19 proteins to be important for the function of complex I (CI) of the electron transport chain. We validate a subset of these predictions using RNAi, including C8orf38, which we further show harbors an inherited mutation in a lethal, infantile CI deficiency. Our results have important implications for understanding CI function and pathogenesis and, more generally, illustrate how our compendium can serve as a foundation for systematic investigations of mitochondria.

Figures

Figure 1. Building a Compendium of Mitochondrial Proteins

MitoCarta is a compendium of 1098 genes encoding proteins with strong support of mitochondrial localization. Each protein was determined to be mitochondrial by one or more of the following approaches: 1) an integrated analysis of seven genome-scale data sets, including in-depth proteomics of isolated mitochondria from 14 mouse tissues (gray circle), 2) large-scale GFP-tagging/microscopy (green circle), and 3) prior experimental support from focused studies (red circle). The union of genes from each approach comprises the MitoCarta compendium.

Figure 2. Discovery Proteomics and Subtractive Proteomics of Isolated Mitochondria

(A) Purification of mitochondria from 14 mouse tissues. Mitochondrial enrichment was tracked by the ratio of an ER protein (calreticulin) to mitochondrial proteins (VDAC and CI 8kDa subunit) at three stages of isolation (W, whole tissue lysate; C, crude mitochondrial extracts; P, purified mitochondrial extracts). Electron micrographs show intactness of the purified organelles. (B) Saturation of protein identifications by discovery MS/MS is plotted for previously known mitochondrial proteins (Tmito), abundant proteins (>25% coverage), and all proteins. (C) Gene Ontology annotations of proteins enriched in pure (red) or crude (black) mitochondrial samples based on subtractive MS/MS experiments. Inset: schematic overview of subtractive MS/MS method. (D) Likelihood ratio of a protein being truly mitochondrial based on detection in discovery and subtractive MS/MS experiments.

Figure 3. Data Integration and Validation by Microscopy

(A) Eight genome-wide methods for predicting mitochondrial localization, with sensitivity and corrected false discovery rates (cFDR) calculated from large training sets at predefined thresholds (Experimental Procedures). Rightmost columns show each method's log-likelihood score for a selection of mouse genes, which are summed to produce the Maestro log-likelihood of mitochondrial localization. (B) Fluorescence microscopy images of 10 GFP-fusion constructs with clear mitochondrial localization, corresponding to examples in panel A. Images for all 131 constructs showing mitochondrial localization are available at www.broad.mit.edu/publications/MitoCarta.

Figure 4. Mitochondrial Protein Expression Across 14 Mouse Tissues

(A) Heatmap of protein abundance, measured by log10 (total MS peak intensity), for 1098 MitoCarta genes across 14 tissues. Genes are ordered by number of tissues and total intensity. White background indicates genes whose protein product was not detected by MS/MS, but are mitochondrial based on prior annotation, computation, or microscopy. (B) Tissue-distribution of proteins within selected pathways. Tick marks indicate locations of corresponding proteins within (A), and gray shading indicates the total number of tissues in which the protein was detected (0-14). (C) Correlation matrix of MitoCarta proteins detected by MS/MS in each tissue, clustered hierarchically. Counts on diagonal indicate number of MitoCarta proteins identified by MS/MS. (D) Mitochondrial quantity per tissue, assessed by ELISA measurements of cytochrome c from whole tissue lysates.

Figure 5. Ancestry of Mitochondrial Proteins

(A) Presence/absence matrix for the 1098 MitoCarta proteins across 500 fully sequenced organisms. Blue squares indicate homology of the mouse protein (row) to a protein within a target species (column). (B) Ancestry of MitoCarta proteins from selected groups. Tick marks indicate location of proteins within (A). (C) Comparison of MitoCarta protein ancestry to all mouse proteins, considering only best-bidirectional hits. P values based on hypergeometric distribution with Bonferroni multiple hypothesis correction: *p = 6e-64, **p = 4e-78, ***p = 2e-232.

Figure 6. Identification of Complex I Associated Proteins through Phylogenetic Profiling

(A) Presence/absence matrix for 44 respiratory chain CI subunits and 3 assembly factors across 42 eukaryotic species. Blue squares indicate homology of the mouse protein (row) to a protein in a target species (column). (B) MitoCarta proteins matching the phylogenetic profile of the subset of CI subunits lost independently at least four times in evolution. Asterisks indicate candidates tested by RNAi in (D-F). (C) Reconstructed phylogenetic eukaryotic tree, with red text indicating species that have lost CI. (D) Effect of candidate knockdown on CI levels in human fibroblasts. Immunoblots of actin and a CI subunit from whole cell lysates were performed following lentiviral-mediated delivery of an empty vector or hairpins targeted against GFP (negative control), NDUFAF1 (known CI assembly factor) and four CI candidates. (E) Percent knockdown of mRNA expression achieved for controls (gray bars) or CI candidates (blue bars) as measured by real-time qPCR. (F) CI activity assays from fibroblast lysates (as in D) for controls (gray bars) and four candidates (blue bars). Error bars represent the range of duplicate assays.

Figure 7. Discovery of a C8orf38 Mutation in an Inherited Complex I Deficiency

(A) Pedigree from a consanguineous Lebanese family with two children affected by Leigh syndrome and complex I deficiency. Letters beneath each family member represent the genotype for a c.296A>G mutation in C8orf38. Proband indicated by arrow. (B) Respiratory chain enzyme activities, standardized against the mitochondrial matrix marker enzyme citrate synthase, expressed as percentages of the mean value (normal ranges in parentheses). The final column lists citrate synthase activities (relative to total protein) as % of normal control mean (see Experimental Procedures). (C) Results of homozygosity mapping using DNA from family members in (A). Eight intervals of homozygosity shared by the affected siblings but not the parents or unaffected sibling are listed along with the number of genes in various categories for each interval (CI, known complex I genes; COPP, Complex One Phylogenetic Profiling candidates). (D) Sequence traces of C8orf38 from each family member in (A) and one healthy control demonstrating homozygosity for a c.296A>G mutation in both affected siblings.