Systems proteomics of liver mitochondria function

Abstract

Recent improvements in quantitative proteomics approaches, including Sequential Window Acquisition of all Theoretical Mass Spectra (SWATH-MS), permit reproducible large-scale protein measurements across diverse cohorts. Together with genomics, transcriptomics, and other technologies, transomic data sets can be generated that permit detailed analyses across broad molecular interaction networks. Here, we examine mitochondrial links to liver metabolism through the genome, transcriptome, proteome, and metabolome of 386 individuals in the BXD mouse reference population. Several links were validated between genetic variants toward transcripts, proteins, metabolites, and phenotypes. Among these, sequence variants in Cox7a2l alter its protein's activity, which in turn leads to downstream differences in mitochondrial supercomplex formation. This data set demonstrates that the proteome can now be quantified comprehensively, serving as a key complement to transcriptomics, genomics, and metabolomics--a combination moving us forward in complex trait analysis.

Fig. 1.. Overview and validation of omics layers.

(A) General model of the multilayered approach. Arrows indicate causality between metabolic layers. HFD should not modify DNA, although other environmental factors can (i.e., mutagens). (B) Phenotyping pipeline for all individuals. (See the methods section for details on each experiment.) (C) Body weight in two strains of BXD for both diets over the full phenotyping experiment. (D) Area under the curve (AUC) of glucose excursion during a 3-hour oral glucose tolerance test for all cohorts. Bars represent mean ±SEM. (E) Heritability for several phenotypes, calculated by one-way (CD/HFD) or two-way (Mixed) analysis of variance. Some traits are affected by diet (weight and fasted glucose), others are not (heart rate and body temperature), and G×E contributions vary. (F) Volcano plot of diet effect on clinical phenotypes. (G) Volcano plot of diet effect on all transcripts. (H) Dot plot showing two example hepatic metabolites affected by diet. (I) An enriched Spearman correlation transcript network using the cholesterol biosynthesis and SREBF targets gene set. Edges indicate P ≤ 0.001. All correlations are positive. (J) Error in SWATH measurements due to different factors: technical (median CV = 6.5%), biological (CV = 17.0%), across strain (within diet) (CV = 29.6% HFD, 31.4% CD), and across all measurements (CV = 30.8%). Reported P values between diets (panels C–D, F–H) are all for Welch’s t-test.

Fig. 2.. Multilayer analysis of associations and causality.

(A) Histogram of 2600 transcript-protein pair Spearman correlations in CD. ρ = 0.32 corresponds to a nominal P < 0.05. ρ = 0.65 corresponds to Bonferroni-corrected significance. (B) Correlation plot of transcript-peptide Spearman correlation coefficients in CD against HFD. (C) Transcript-protein correlation prevalence in CD cohorts, binned by transcript variation. Among the top 10% most variable transcripts (260 pairs), 56% of pairs correlate, in contrast to only 20% of pairs in the lowest bin. Nominal significance cutoffs are used, so ~5% of matches in each bin are false positives. (D) The transcript Pura correlates significantly with its protein in CD but not in HFD. (E) Malate and fumarate, two adjacent metabolites in the TCA cycle, correlate strongly. Several other cross-layer correlations are observed between metabolites and their adjacent enzymes in major metabolic pathways. (F) Venn diagram and count of all cis- and trans-eQTLs across diets for the 2600 transcripts with matching protein measurements. (G) Venn diagram and count of all cis- and trans-pQTLs for the same 2600 proteins. (H) Overlap between cis-eQTLs and cis-pQTLs in both diets. Fifty-nine genes map to cis-QTLs in all four data sets (intersection not shown). (I) Venn diagram of all mQTLs and cQTLs in both diets. In red for cQTLs: overlapping cQTLs that are genome-wide significant in one diet (LRS ≥ 18) and locally significant in the other (LRS ≥ 12).

Fig. 3.. Identifying the QTGs and causal mechanisms driving QTLs.

(A) Combined QTL map of Bckdha transcript and protein in both diets. Significant trans-pQTLs map to Bckdhb (yellow triangle) on chromosome 9, whereas no cis-QTLs map to Bckdha on chromosome 7 (red triangle). (B) Spearman correlation matrices of the four subunits of the BCKDC at the transcript or protein level in both diets. (C) Homeostatic model assessment for insulin resistance (HOMA-IR) is significantly increased in HFD cohorts compared with CD (P = 2 × 10⁻⁶, Welch’s t test), but no association is seen between Bckdhb allele and HOMA-IR in either dietary cohort. (D) D2HG maps significantly to chromosome 1 in the HFD cohort. (E) This locus contains 56 genes, of which 16 have a major genetic variants variable, including D2hgdh. (F) Composite eQTL and pQTL map for D2hgdh. The protein maps as a cis-pQTL in both diets to the same chromosome 1 locus, whereas only the HFD transcript levels map to a cis-eQTL. (G) D2hgdh drives one of several pathways generating α-ketoglutarate. (H) D2HG is positively associated with heart rate in both diets in a Pearson correlation. (I) Eci2 exhibits no cis-eQTLs, but yields significant cis-pQTLs in both diets. (J) Peptide sequence analysis of ECI2, with the nine measured peptides and the single missense mutation highlighted. (K) ECI2 Western blots show two distinct molecular weight bands depending on the BXD genotype.

Fig. 4.. Network analysis.

(A) Spearman correlation network showing mixed genes involved in fat metabolism at the transcript and protein level, along with key metabolites and phenotypes. The dashed circle represents the core enriched gene set. Edges are significant at P < 0.001 for a positive (blue) or negative (red) correlation. (B) Diet-dependent expression of key genes and metabolites involved in fat metabolism; P values are for Welch’s t test. (C) A Spearman correlation network of 74 transcripts taken at random from the list of 2600 genes measured at the transcript and protein level, using the same network analysis. Edge counts correspond to the level expected from noise. (D) (Left) Hmgcs1 and Srebf1, as well as other transcripts and proteins in the cholesterol biosynthesis pathway, are highly variable in the BXDs. (Right) PCA of a set of eight cholesterol biosynthesis genes shows that their variances are highly explained by a single factor. (Bottom) Two candidate cholesterol genes, Mmab and Echdc1, which correlate with PC1 in both diets. (E) In vitro validation of HMGCS1, along with two proteins not known to be involved in cholesterol metabolism—MMAB and ECHDC1—which clustered with known cholesterol genes. MMAB and ECHDC1 both respond like HMGCS1 to lipid-deficient serum and statin treatment or to knockdown of LDLR, SREBF2, or SREBF1/2, suggesting that they are indeed involved in cholesterol metabolism. (F) Unbiased Spearman correlation matrices of the first PC in CD (bottom left) and HFD (top right) conditions with other genes turned up many known cholesterol-regulatory genes (orange) as well as new candidates (green). (G) Transcript and protein networks for the 73 genes with paired transcript-protein data in the cytosolic ribosome complex. Both were highly enriched, although with tighter coregulation at the transcriptional level. Edges represent Spearman correlations with P < 0.0001.

Fig. 5.. Variable mitochondrial phenotypes in the BXDs.

(A) The oxphos protein Spearman correlation network is somewhat more tightly coregulated than the transcript network. In particular, CI proteins cluster more tightly than CI transcripts (black nodes). (B) Circos plot of 67 ETC genes. Green bar ring: effect of diet, relative change between medians. Light green: transcript; dark green: protein. Purple bar ring: correlation between transcript and protein in CD (light purple) or HFD (dark purple). Red bar ring: LRS of peak pQTLs in CD (light red) and HFD (dark red). Blue bar ring: LRS of peak eQTLs in CD (light blue) and HFD (dark blue). Inside: drawing of significant cis-QTLs (LRS ≥ 12). Significant trans-QTLs (LRS ≥ 18) are not drawn. (C) Diet-consistent cis-pQTLs were observed only for COX7A2L, which does not map to significant cis-eQTLs. (D) (Top) The Cox7a2l transcript is affected by diet, whereas both transcript and protein are highly variable across genotype. (Bottom) Expression is consistent across diets within the transcript and protein level, despite the presence of dietary effect in mRNA and its absence in protein. (E) BN-PAGE for four strains with three biological replicates. Individual complexes are labeled. Several distinct upper SC bands are observed, labeled initially as 1 through 6. (F) Upper SCs for all CD cohorts (several independent gels are aligned and spliced together). SCs were quantified in binary fashion by presence (+1) or absence (0) of a particular band.

Fig. 6.. Tissue variance in SC formation.

(A) SC bands 4 and 5 mapped significantly as cQTLs to a locus on chromosome 17. (B) In-gel activity assays were performed in the liver tissues to determine SC’s composition and relation to COX7A2L. Bands 2 to 5 could be identified confidently as CI + CIII₂ + variable numbers of CIV (0 to 3). (C) In-gel activity assays from livers of six additional BXD strains—three with the B6 allele of Cox7a2l (BXD73, BXD80, and BXD100) and three with the D2 allele (BXD43, BXD61, and BXD96). COX7A2L is present in bands 4 and 5 for strains with the D2 allele. (D) In-gel activity assays from hearts of the same individuals as above. COX7A2L is absent in bands 4 and 5 and III₂+IV₁ in strains with the B6 allele and present in strains with the D2 allele. Unlike liver, bands 4 and 5 are observed in all strains, albeit at lower levels in strains with the B6 allele of Cox7a2l, indicating tissue-specific differences in SC formation.