Respiratory chain protein turnover rates in mice are highly heterogeneous but strikingly conserved across tissues, ages, and treatments

Abstract

The mitochondrial respiratory chain (RC) produces most of the cellular ATP and requires strict quality-control mechanisms. To examine RC subunit proteostasis in vivo, we measured RC protein half-lives (HLs) in mice by liquid chromatography-tandem mass spectrometry with metabolic [²H₃]-leucine heavy isotope labeling under divergent conditions. We studied 7 tissues/fractions of young and old mice on control diet or one of 2 diet regimens (caloric restriction or rapamycin) that altered protein turnover (42 conditions in total). We observed a 6.5-fold difference in mean HL across tissues and an 11.5-fold difference across all conditions. Normalization to the mean HL of each condition showed that relative HLs were conserved across conditions (Spearman’s ρ = 0.57; P < 10^–4), but were highly heterogeneous between subunits, with a 7.3-fold mean range overall, and a 2.2- to 4.6-fold range within each complex. To identify factors regulating this conserved distribution, we performed statistical analyses to study the correlation of HLs to the properties of the subunits. HLs significantly correlated with localization within the mitochondria, evolutionary origin, location of protein-encoding, and ubiquitination levels. These findings challenge the notion that all subunits in a complex turnover at comparable rates and suggest that there are common rules governing the differential proteolysis of RC protein subunits under divergent cellular conditions.—Karunadharma, P. P., Basisty, N., Chiao, Y. A., Dai, D.-F., Drake, R., Levy, N., Koh, W. J., Emond, M. J., Kruse, S., Marcinek, D., Maccoss, M. J., Rabinovitch, P. S. Respiratory chain protein turnover rates in mice are highly heterogeneous but strikingly conserved across tissues, ages, and treatments.

A primary function of the mitochondrion is to maintain the electron transport chain (ETC), complexes I–IV (CI–IV), and an ATP synthase, CV, which generates the currency of ATP cellular energetics. Together, these complexes form the mitochondrial (mt) respiratory chain (RC). Because of electron leakage from the ETC during oxidative phosphorylation, mitochondria are also the major source of reactive oxygen species (ROS) in the cell (1) and therefore require a network of quality-control systems to maintain protein homeostasis. Continuous fission and fusion serve to protect the integrity of the mt network, with damaged mt regions being subject to mitophagy (2, 3). Mitophagy is generally perceived as a bulk removal process, although there is some debate as to whether damaged proteins can be preferentially segregated from the rest of the mitochondrion for removal by mitophagy (4). In addition, proteases and chaperones within each mt compartment oversee proper folding or degradation of damaged proteins (5). Accumulating evidence indicates that the ubiquitin proteasome system (UPS) also plays a role in mt protein degradation (6). However, in the event of mt protein damage that exceeds the organelle’s quality-control capacity, a collective cellular response termed mt unfolded protein response is initiated: a mitochondria-to-nucleus signal transduction pathway that up-regulates genes encoding mt chaperones and proteases (7). Even though these multiple quality control mechanisms limit mt damage, failure to maintain mt protein quality control is believed to underlie many pathologic events and aging (8). Understanding mt proteome dynamics can provide a framework for the investigation of protein quality control in the mitochondria (9), but there is a paucity of understanding of the mechanisms of mt dynamics and quality control in the study of mt pathologies and aging. In addition, researchers have not attempted to identify common underlying components of mt protein dynamics across the various tissues and metabolic conditions.

Early studies investigating mt protein turnover reported that it is a unitary process involving the whole organelle, suggesting that all mt proteins turn over at similar rates (10–12). Consistent with this view, 2 recent reports have shown that proteins localized to specific organelles or belonging to complexes often have similar turnover rates (13, 14). In particular, it has been suggested that RC complexes and supercomplexes are maintained in unitary solid states (15, 16). However, other investigators have found less homogeneity in mt turnover rates within respiratory complexes and challenge the notion of unitary turnover. Kim and colleagues (17) reported wide differences in mt protein HLs within heart and liver tissues, suggesting a complexity in regulation of the mt RC proteome. In contrast to previous reports, the findings in their study also showed that subunits of RC complexes turn over with different kinetics, albeit over a narrow range. A recent study of Arabidopsis mt proteins also reported a wide (30-fold) range of degradation rates within the RC (18), supporting a more protein-specific turnover process than a unitary one.

Using a newly developed mass spectrometry (MS) method that accounts for precursor pool enrichment (19), we examined the in vivo turnover rates of more than 250 mt proteins, including 84 individual RC proteins in mouse heart, liver, skeletal muscle, and the brain cortex. To examine conditions that modify cellular homeostasis, we investigated the effect of aging, caloric restriction (CR) and the CR mimetic rapamycin (RP) (20, 21) on mt proteome dynamics. CR and RP affect the mammalian target of rapamycin complex 1 (mTORC1) to alter protein synthesis and degradation rates (22), making them useful modulators of cellular proteostasis.

MATERIALS AND METHODS

Animals

C57BL/6 female mice aged 3 and 25 mo (Fig. 1A) were obtained from the National Institute on Aging Charles River Colony (Wilmington, MA, USA). Female mice were used, because the effects of RP in murine aging are much stronger in mice of that sex (20). The mice were housed at 20°C with a 12-h light and dark cycle in an Association for Assessment and Accreditation of Laboratory Animal Care–accredited facility under University of Washington Institutional Animal Care and Use Committee supervision.

Figure 1.

A) Summary of the experimental design showing young (3 mo) and old (25 mo) C57BL/6 female mice maintained on a control diet or fed ad libitum, an encapsulated RP (14 ppm)-containing diet, or a CR (40% restricted) diet for 10 wk. Mice were subsequently switched to the same basic diet but with leucine replaced with [³H₂]-leucine and euthanized 3, 7, 12, or 17 d thereafter. Heart, liver, brain cortex, and the skeletal muscles EDL and soleus were harvested, and LC-MS/MS was performed on tryptic peptides of total lysates or purified mitochondria. B) The protein turnover rate was determined by fitting the fractional syntheses of peptide isotopomers of a protein into an exponential curve of first-order kinetics over the labeling period. Individual data points represent all peptide isotopomers that matched a single protein.

Diet regimens and feeding

One week after arrival, all mice were started on a synthetic diet (Teklad diet TD.99366; Harlan Laboratories, Madison, WI, USA) to facilitate the subsequent substitution of a heavy-labeled leucine diet to enable protein turnover measurements. After 3 wk, the mice were individually housed and randomly assigned to 3 groups: 1) maintained on an ad libitum synthetic food regimen (control); 2) fed an RP-containing synthetic diet; and 3) fed a CR diet, as detailed below. They were maintained on these 3 regimens for 10 wk. There was an average of 7.4% body weight loss in those animals on the synthetic diet that stabilized 1 wk after the start of feeding.

Mice in the CR group received a vitamin- and mineral-adjusted diet (diet TD.10943; Harlan Laboratories) that provided the same essential nutrient levels as the control group diet. CR rations per mouse were calculated based on the age-matched ad libitum mouse intake normalized to each CR animal’s starting body weight. The CR mice received 10% less than the control mice consumed during wk 1, 25% less during wk 2, and 40% less from wk 3 onward. The young CR cohort lost, on average, ∼20% of their stabilized body weight compared with those on the 3 wk progressive CR diet. The weight loss of the old CR mice averaged ∼25%. After this initial loss, the CR mice maintained a stable weight. If any CR mouse lost 30% or more of its stabilized body weight, food rations were increased by 5–10% until the animal stabilized at 70% of its initial body weight, a limit set by the University of Washington Institutional Animal Care and Use Committee. The old control and RP mice maintained a stable weight over the study period. For details on body weights see the Supplemental Information in Karunadharma et al. (23). Microencapsulated RP was purchased from the University of Texas Health Science Center at San Antonio and administered at 14 mg/kg of food (2.24 mg RP/kg body weight/d). RP blood levels were measured at 4 wk and at the time of euthanasia. The mean concentration of RP in young and old mice was 76 ± 8 and 44 ± 4 ng/ml, respectively (23).

Stable isotope labeling

After 10 wk of the diet regimens, all mice were started on a leucine-deficient synthetic diet (Teklad TD.09846; Harlan Laboratories) with the light leucine fully replaced by 11 g/kg of deuterated [5,5,5-²H₃]-l-leucine (Cambridge Isotope Laboratory, Cambridge, MA, USA), with CR and RP cohort conditions continued as above. Four mice per cohort were euthanized by cervical dislocation at 4 time points: d 3, 7, 12, and 17 after they were switched to the [²H₃]-leucine diet (Fig. 1A).

Liquid chromatography-tandem mass spectrometry

Tissues were removed immediately, rinsed in cold saline, and homogenized in cold isolation buffer [250 mM sucrose, 1 mM EGTA, 10 mM [4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid]; HEPES), and 10 mM Tris-HCl (pH 7.4)]. Lysates were centrifuged at 800 g for 10 min to remove the debris. Portions of liver, muscle, and heart tissues were processed for mt fractions as described previously (24). All samples were trypsin digested, and nanoscale liquid chromatography tandem mass spectrometry (LC-MS/MS) analysis was performed with a nanoAcquity Ultra Performance Liquid Chromatography (Waters, Milford, MA, USA) and an LTQ Orbitrap Velos (Thermo Scientific, Waltham, MA, USA).

MS data were processed with the Hardklör (v. 1.33) and Bullseye (v. 1.25) algorithms (both developed at the University of Washington), to refine precursor mass measurements (25, 26), followed by a database search against all mouse entries of the UniProt database (UniProt release 2013_02; European Bioinformatics Institute, Hinxton, United Kingdom) with the SEQUEST algorithm (v. UW2012.01.7; The Scripps Research Institute), searching a total of 74,888 protein entries that were designated Mus musculus. A dynamic modification of 3.0188325 for leucine was set to account for [5,5,5-²H₃]-leucine and a static modification of 57.021461 for cysteine was set for carbamidomethyl modifications. The precursor monoisotopic mass tolerance was set to ±10 ppm and the fragment mass tolerance window was set to 0.36 mass-to-charge ratio (m/z). Enzyme specificity was set to semitryptic, allowing for up to 2 missed cleavage sites per peptide. The false-discovery rate for spectrum matches was determined by the Percolator algorithm (v. 2.04; University of Washington), with a reversed copy of the UniProt database used as a decoy (27). Only results with q < 0.01 were kept for further analysis.

LC-MS/MS analysis

Topograph (v. 1.1.0.297; University of Washington) software was used for the deconvolution and measurement of peptide isotopologue abundances from the LC-MS/MS chromatograms (Fig. 1B) and the calculation of peptide turnover rates (19). (http://proteome.gs.washington.edu/software/topograph/). Relative peptide abundances were determined by integrating MS1 peaks. For peptides that were identified in one sample, the regression of the identified peptide’s MS/MS scan number was used to estimate a window for the same peptide in the other samples, and a matching chromatographic peak was identified within that time range. This method enables peaks areas to be measured, even in samples in which they are low and otherwise difficult to identify. For 2 given LC-MS/MS chromatograms, the MS/MS scan numbers for peptides identified in both samples were plotted against each other in a scatter plot. LOESS (local regression) was used to find the best-fit line through the data points (19, 28).

Topograph enables the measurement of the proportion of the amino acid precursor pool that is labeled, which varies over time and condition. This information enabled the correct calculation of the percentage of each peptide that was newly synthesized, which, when plotted for 12 biologic replicates over 4 time points, generated an exponential curve following first-order kinetics. Using a logarithmic transformation, we determined the first-order protein turnover rate (slope) by linear regression.

Only peptides that uniquely mapped to a single UniProt protein accession were used for quantification of abundance and turnover. In the cases where a protein consisted of more than 1 peptide, statistical models were modified to account appropriately for the multiple peptides by using a blocking factor. For each protein, we applied nonlinear regression fits of first-order exponential curves to the percentage of newly synthesized protein: y = 100 + β1^eαt. ANCOVA was used to identify significantly different turnover rate (slope), α, between the experimental groups. For details see the Supplemental Methods in Hsieh et al. (19).

This turnover calculation relies on the assumption that proteins are in a condition of steady state at the time of sampling, in which protein synthesis and degradation rates are approximately equal. To reasonably account for steady state, we acclimated the mice to experimental diets until they reached stable weights, which remained stable over the course of heavy labeling. Furthermore, the abundance of all peptides identified did not change significantly over the labeling period (regression slopes centered over zero), confirming that the proteins were at steady state at the time of labeling. To reduce the complexity of each regression, we derived a single x-intercept (time of first appearance of heavy label) as an average for each age-treatment group and subsequently fixed the x-intercept for regression of individual proteins within each group to this value. These intercepts for liver and heart have been published as Supplemental Information in Karunadharma et al. and Dai et al., respectively (23, 29).

As previously shown, the relative standard errors (RSEs) of the HL estimates were small (<2.5%), as shown for liver and heart. There was an increase in the RSEs at the short (<3 d) or very long (>12 d in liver, >45 d in heart) HL ranges in both tissues, possibly because of the reduced precision in HL estimates that were beyond our chosen span of harvest times (3–17 d) (23, 29).

Between the total and mt fractions, liver HLs differed by 6% (P < 0.0001) and heart HLs differed by 16% (P < 0.0001), on average, among the age-treatment groups (Table 1); we tentatively ascribed these differences to nonrepresentative extraction or survival of mitochondria during purification or contamination by microsomal fractions.

TABLE 1.

Summary of ETC protein mean HLs across tissues and treatments

	Average HL of ETC
Age/treatment	HT	HM	LT	LM	BT	EM	SM	Treatment average	% of YCL
YCL	24.1 ± 1.2	20.3 ± 0.8	4.5 ± 0.2	4.8 ± 0.2	23.7 ± 1.0	27.6 ± 1.7	24.5 ± 1.6	18.5a	100
YCR	38.7 ± 1.8	32.2 ± 1.5	7.6 ± 0.3	8.0 ± 0.3	29.1 ± 1.6	45.2 ± 2.6	45.8 ± 2.2	29.5b	159.5
YRP	26.9 ± 1.2	23.8 ± 0.9	4.9 ± 0.2	5.2 ± 0.2	24.5 ± 1.1	33.9 ± 2.1	34.8 ± 2.3	22.0c	118.9
OCL	25.6 ± 1.1	21.5 ± 0.9	4.0 ± 0.2	4.3 ± 0.2	25.5 ± 1.0	28.2 ± 1.4	26.0 ± 1.8	19.3a	104.5
OCR	32.1 ± 1.3	29.1 ± 1.2	6.2 ± 0.2	6.2 ± 0.2	33.0 ± 1.6	36.9 ± 2.2	35.1 ± 1.8	25.5d	138.0
ORP	30.2 ± 1.4	25.1 ± 1.0	4.1 ± 0.2	4.6 ± 0.2	25.1 ± 1.2	31.3 ± 1.7	29.9 ± 2.2	21.5c	116.0
Tissue average	29.6A	25.3B	5.2C	5.5D	26.8B	33.9E	32.9E
% of HT	100.0	85.5	17.6	18.5	90.5	114.3	110.4

Data are expressed as days ± sem. Tissues: HT, heart total; HM, heart mitochondria; LT, liver total; LM, liver mitochondria; BT, brain cortex total; EM, EDL mitochondria; SM, soleus mt proteins. Treatments: YCL, young control; YCR, young CR; YRP, young RP; OCL, old control; OCR, old CR; ORP, old RP. The geometric mean HL ratio was tested for a difference from 1 in each pair-wise comparison of treatments and tissues, adjusting each for the other, using ANOVA on proteins in common to all groups. ^A–D,a–dAverage HLs with different letters are significantly different from each other. P < 10⁻⁵, except LM vs. LT, P < 0.0003; SM vs. HT, P < 0.02.

Ubiquitin enrichment and analysis

A portion of liver tissues was enriched for polyubiquitinated proteins by antibody pull down (30). To purify polyubiquitinated proteins, we prepared total liver lysates in a simple 50 mM ammonium bicarbonate buffer containing 8 M urea and then split them into 2 fractions. The first fraction was immunopurified with the agarose-conjugated polyubiquitin-specific antibody FK2 (D058-8; MBL, Nagoya, Japan), and the second fraction from each portion of liver was enriched with an agarose-conjugated isotype negative control antibody (M194-3; MLB). Using low-molecular-weight exclusion filter spin columns, we washed all samples 2 times with an 8 M urea washing buffer, followed by 2 washes with 50 mM ammonium bicarbonate. Antibody-bound proteins were eluted with 100 mM glycine (pH 2.8) and precipitated by a standard methanol chloroform extraction. The insoluble proteins were then pelleted, decanted, and resuspended in an MS-compatible mt isolation buffer [250 mM sucrose, 1 mM EGTA, 10 mM HEPES, and 10 mM Tris-HCl (pH 7.4)], then prepared for MS by a method identical to that used for whole-cell extracts, described above. Isotype control–enriched abundances were subtracted from FK2-enriched sample abundances to remove nonspecific contamination.

Statistical analysis

Tissue age-treatment groups (listed and defined in Table 1), CI–V, and localization [membrane, matrix, intermembrane space (IMS)] log-HL interactions were examined by least-squares regression, whereas ancestral age, assembly order, susceptibility to 20S proteasome degradation and ubiquitin enrichment were fitted as continuous linear variables. RC protein subunits with membrane-spanning domains were assigned “membrane” as their location for statistical purposes. Factors for each protein were also fitted as precision variables for tests of tissue, age-treatment, and complex. These protein effects did not appreciably alter the estimated effects, but did result in greater statistical power. A likelihood ratio test was used to examine the overall effects of tissue, age-treatment, and complex, as well as linear trend tests for the continuous variables. The Wald test was used for the pair-wise comparisons reported in Results. A pair-wise comparison to test for a nonzero difference in the mean log HL between 2 groups is equivalent to testing that the geometric mean of the HL ratios between the 2 groups is different from 1. After large and significant tissue and age-treatment group effects were established, all other tests were adjusted for these 2 variables. Because of imbalances on ancestral group across the different complexes that preclude proper adjustment for confounding by complex, a second test for ancestral group effect was performed on a balanced subset of the data and adjusted for complex effects. The subset was restricted to ancestral groups that were present in all complexes, resulting in restriction to bacteria (group 0), LECA (group 1), and opisthokonts (group 3) in CI, -III, -IV, and -V. Eighty-four different RC proteins were used in the analyses. The protein succinate dehydrogenase complex, subunit D (SDHD) was excluded from statistical and turnover analyses because of the high variability and inconsistency across tissues. Statistical significance was set at P < 0.05. Statistical computations were done in R (The R Project; www.r-project.org).

Raw data repository and other files

Raw files and datasets from the LC-MS/MS analysis can be viewed and downloaded online on the Chorus Project data repository (https://chorusproject.org/pages/blog.html#/352). In order access the data, a free account should be registered at https://chorusproject.org/pages/register.html#/. For R-scripts and other data, please contact the corresponding author to receive copies.

RESULTS

Experimental design and protein HLs of RC proteins

The experimental design is summarized in Fig. 1A. All mice were maintained on 3 diet regimens for 10 wk; their stabilized weights varied less than 1.1% in the final 5 wk. Total proteins were extracted from liver, heart, and brain. Mitochondria were also purified from heart and liver and 2 skeletal muscles, the extensor digitorum longus (EDL) and soleus, before protein extraction. Trypsin-digested peptides from lysates were analyzed by shotgun nanoscale LC-MS/MS followed by protein synthesis measurements with Topograph software (19). We measured turnover rates for up to 84 proteins of the RC (of 90 known RC proteins) for each of 6 age-treatment groups in each of 7 tissues or tissue fractions. This coverage is the highest for RC protein HL measurement to date. Average HLs of the RC proteins in each tissue and age or treatment group are shown in Table 1. Mean HL in the different tissues ranged from 5.2 d (liver) to 33.9 d (EDL), a 6.5-fold difference. In comparison to heart, the 2 muscle types had 5–10% longer average HLs, whereas the liver proteins had ∼80% shorter HLs (Table 1). There was no significant change in HLs with age except after CR. Both the CR and RP treatments resulted in significantly altered HLs (P < 10⁻⁵), with CR having the strongest effect: 38–60% longer HLs than young controls vs. 16–19% longer HLs after RP. The substantial shifts in distribution of HLs between tissues are further illustrated by the histograms in Supplemental Fig. 1A. All RC protein HL values are listed in Supplemental Dataset 1.

RC protein HL relative differences are broadly conserved across tissues and treatments

To compare the relative distribution of RC HLs within each condition, we normalized individual HLs in all tissues and treatments to their mean HLs (Fig. 2A). The mean value for each protein (in rows) across all conditions (in columns) is shown in the right-most column. With the tissue/treatment differences accounted for in this way, the RC proteins spanned a 7.3-fold range between longest and shortest mean HLs. CI proteins had the broadest range of normalized HLs, within a 4.6-fold difference (n = 42), whereas CV had the smallest range, within a 2.2-fold difference (n = 13). Despite the substantial differences in tissue and treatment HLs (Table 1), the normalized heat map shows broad conservation of HLs across tissues and conditions. Subsets of proteins with relatively homogenous normalized HLs within complexes can be appreciated by the horizontal color bands. This effect, measured as the average correlation between mouse groups, was statistically significant (Spearman’s ρ = 0.57; P < 10⁻⁴), illustrating that the relative (normalized) differences in RC protein HLs are highly consistent across tissues and treatment groups, despite the large differences in average HLs that we observed between tissues and treatments (Table 1). The average normalized HL across all tissues, ages, and treatments is shown as a separate column to the right of the tissue groups on the heat map, and this same color is plotted onto schematics of the RC complexes and proteins (Fig. 2B), based on known location information (28–35). All mean normalized log₂ HLs are listed in Supplemental Dataset 2.

Figure 2.

Heat map of RC protein-relative HLs shown for 6 experimental groups in 7 tissue types. A) All protein HLs are normalized to the mean HL of their respective tissue and treatment group (columns). The 6 age-treatment groups are numbered (see key). Protein subunits in each complex are grouped according to their location in the mitochondrion IMS, membrane, and matrix. The relative log₂ HLs, ranging from −3.4 to +2.5 (58-fold) around the mean, are depicted in a color gradient from dark blue to dark red (see color key). Gray: values missing in the MS data. The rightmost 3 columns depict mean-adjusted HLs of orthologous proteins from fly heads (columns a, WT; b, parkin and c, Atg7 mutant flies) from Vincow et al. (39). B) The RC complexes, showing the subunit topology within each complex. The proteins are colored according to their average relative HL across all mouse treatments and tissues (the rightmost column of the mouse heat map).

Mouse RC protein turnover is evolutionarily conserved

The heterogeneity in RC protein-relative HLs may be evolutionarily conserved in lower organisms, as suggested by the conserved colors in 43 orthologous proteins from data obtained from flies by Vincow et al. (39). Data sets from wild-type and parkin and Atg7 mutant flies are shown in the 3 rightmost columns of Fig. 2A. The mouse protein average HLs correlated significantly with the fly protein average HLs (Spearman’s ρ = 0.58; P = 9.2 × 10⁻⁵; Supplemental Fig. 2B, C).

RC protein HLs correlate with cellular location

Proteins localized in the matrix of CI had 16% shorter HLs on average (faster turnover) than the CI membrane proteins, a difference that was statistically significant after adjustment for tissue and treatment effects (P = 1 × 10⁻¹⁹; Fig. 3A). A similar pattern has been reported for Arabidopsis thaliana RC proteins, with the exceptions of carbonic anhydrase (CA) and CA-like subunits, which are unique to plants (18). In our study, this association of shorter HLs in matrix proteins was also statistically significant for CII and CV (P = 1 × 10⁻¹⁹ and 0.027, respectively), with an average 41 and 5% shorter HLs in the matrix than in their corresponding inner membrane proteins, respectively. However, this relationship was opposite in CIII (9% longer HLs in the matrix; P = 0.03) and CIV (44% longer HLs in the matrix; P < 0.00001), with CIII having generally longer HLs (21% longer; P < 0.049) and CIV having shorter HLs (29% shorter; P < 0.01) than other complexes (Fig. 3A).

Figure 3.

Factors affecting RC protein turnover. A) Tissue-, age-, and treatment-adjusted (log₂) HLs of RC proteins plotted based on their location within each complex. Boxes: interquartile range; line within the box: median log₂ HL for each group; whiskers: ±1.5 times the interquartile range (distance between the 75th and 25th percentiles). *P < 0.05, **P < 2 × 10⁻⁵, differences between 2 locations within complexes. B) HL comparison of the 5 RC complexes, adjusted for tissue, treatment, and location effects. Complexes with different letters are significantly different from each other (P < 10⁻⁶). C) Log₂ HLs of CI, -III, -IV, and -V adjusted for tissue, treatment, and complex effects correlated with ancestral levels of bacterium, LECA, and opisthokonts. D) Correlation of normalized log₂ HLs of RC proteins with their resistance to degradation by the 20S proteasome. The degradation values are from Lau et al. (43). E) Correlation of RC protein HL and ubiquitin enrichment of the same proteins, as determined by polyubiquitinated protein antibody pull-down and LC-MS/MS.

The average HLs of each of the 5 RC complexes, after adjustment for differences in subunit location, tissue, and treatment, were significantly different for all pair-wise comparisons (P < 10⁻⁶; Fig. 3B), except between CII and -IV, demonstrating complex-specific differences in turnover rates.

Ancestral origin correlates highly with RC protein HLs

To determine whether the evolutionary origin of RC subunits has an effect on protein turnover rates, we used a linear regression model to test for correlation between HLs and ancestral age. Each protein’s origin was assigned to 1 of 6 ancestral ages, including the ancestral bacterium, last eukaryotic common ancestor (LECA), unikonts, opisthokonts, metazoa, and vertebrata (40) (Supplemental Fig. 2A). CII represents only a single ancestral group and was therefore removed from the linear analysis. The analysis was also restricted to ancestral groups that were present in all 4 remaining complexes: bacteria (group 0), LECA (group 1), and opisthokonts (group 3). The analysis was also restricted to ancestral groups that were present in all 4 remaining complexes: bacteria (group 0), LECA (group 1), and opisthokonts (group 3). There was linear correlation of HLs with ancestry groups 0, 1, and 3 (P = 0.013), and this correlation strengthened further after adjustment for the complex differences noted above (Fig. 3C; P = 2 × 10⁻⁵). The data suggest that RC proteins with the earliest evolutionary origins—that is, bacterium—are likely to have shorter HLs (faster turnover). Bacterial ancestry proteins were 4% shorter lived than their LECAs and 11% shorter lived than those more recently evolved, the opisthokonts. One explanation for this could be their proximity to electron transport and the catalytic core.

The order of assembly does not correlate with CI turnover

We assigned CI subunits into 3 sequential stages of assembly (38, 41). We hypothesized that early assembled subunits would be less likely to turn over faster because they would be more stably associated with or internal to the complex structure. However, there was no significant (linear) correlation between assembly order and CI HLs [P = 0.635; confidence interval (CI) 0.98–1.04, for ratio of HLs between groups]. This result is inconsistent with the view that subunits that are incorporated last turn over faster than those that are incorporated at early stages and become protected by their internalization, as suggested by Kim et al. (17). It also suggests that subunits that are incorporated last are not subject to appreciably greater proteolysis in the free environment.

The origin of protein encoding correlate with RC protein HLs

Mitochondria contain multiple copies of a 16.5 kb circular genome that encodes 13 essential subunits of 4 of the 5 RC complexes: CI, -III, -IV, and -V. All 13 proteins are hydrophobic integral proteins located in the inner membrane. The rest of the ETC proteins are encoded by the nuclear genome and are imported into the mt by the translocase proteins of the outer and inner mt membranes (TIM/TOM) (42). To determine whether the location of protein synthesis has an effect on protein turnover, we compared nuclear (n)DNA-encoded protein HLs with those of mtDNA-encoded proteins, adjusted for tissue, age, treatment, and complex. Proteins encoded by the nuclear genome had 14% shorter HLs, on average than did mtDNA-encoded ETC proteins (P = 1 × 10⁻¹⁰). Because all mtDNA-encoded proteins reside in the membrane-spanning region, we also compared their HLs to those of only the membrane-integral RC proteins encoded by the nDNA. The difference was maintained with 16% shorter HLs in nDNA-encoded membrane subunits compared with mtDNA-encoded proteins (P = 9 × 10⁻¹⁰). Our data suggest that even within the same sub-mt location, nDNA-encoded proteins turn over faster than mtDNA-encoded proteins, perhaps because of the additional proteostatic processes during the transit of nDNA-encoded proteins to the RC; however, the faster turnover that we observed of the nDNA-encoded RC subunits did not differ between total and mt fractions (ratio of HLs 0.999; CI 0.88–1.13; P = 0.99). This result would not be as supportive of the concept that degradation of subunits before entry into the mitochondria appreciably affects turnover rates. Thus, RC proteins synthesized in the cytoplasm may instead be subject to additional proteostatic processes during their transit through the mt IMS.

RC protein ubiquitination correlates with HLs

To investigate the involvement of proteasomal degradation systems in RC protein turnover, we used recently reported measurements of cardiac proteome’s susceptibility to proteolytic degradation (43). The linear correlation between our HL measurements and the 20S proteasome degradation susceptibility of RC proteins was statistically significant (Fig. 3D; ρ = 0.44; P = 0.05). In contrast, correlation between HLs and the reported susceptibility to degradation by the endogenous mt proteases trended toward, but did not reach, significance (P = 0.089). These results are consistent with the observation of Lau et al. (40) that the 20S proteasome has a selective effect on the turnover of specific mt proteins, including RC subunits.

We extended these observations by measuring the abundance of ubiquitinated RC proteins from the liver, using an anti-polyubiquitin pull-down assay corrected for nonspecific binding, and compared the results to the total abundance of each protein. RC protein HL showed a significant negative correlation with ubiquitin enrichment (Fig. 3E; ρ = −0.33; P = 6 × 10⁻⁸). Both these measures indicate a significant contribution by the 20S proteasome on RC protein dynamics. In fact, recent data highlight the involvement of the 20S proteasome in modulating mt proteome dynamics under oxidative stress conditions (40).

DISCUSSION

Knowledge of proteome dynamics is essential to understanding how cellular systems maintain quality and integrity in various biologic conditions and different molecular niches. By using a whole-animal metabolic labeling strategy, we examined the in vivo mt proteome dynamics in 5 different tissues, each under the effects of aging, CR, and RP. We used a novel software tool that allowed the accurate determination of fractional protein synthesis and protein HL, independent of variation in the precursor pool relative isotope abundance (19).

We observed a 7.3-fold range in normalized RC protein turnover rates, with up to a 4.6-fold range within a complex. Protein HLs in different tissues vary greatly, and CR and RP are known to affect protein synthesis (17, 44, 45). We observed short protein HLs in liver compared with those in other tissues, consistent with findings in a study that demonstrated fast mt turnover in liver (46). However, there is discrepancy among studies in the effects of CR and RP on liver protein turnover (44, 46–48). Our results are consistent with those of Price et al. (44), which showed reduced protein turnover with CR, and we also observed a reduction with RP. Examination of HLs in different tissues and with these interventions allow us to deduce whether the observed diversity in RC turnover is tissue specific and is affected differentially by changes in overall turnover rates. Our findings reveal that despite the large differences in HL between tissues and ages/treatments (Table 1), the pattern of diversity in turnover dynamics is conserved across the tissue, age, and treatment groups (Fig. 2A). The great heterogeneity in mt HLs cannot be readily explained by a macrodegradation process such as mitophagy (49). The evidence instead supports a process of individually targeted protein recycling of RC subunits, such as proteolysis by intramitochondrial proteases or the UPS. Several lines of evidence support the presence of a pathway that retrotranslocates damaged mt proteins to the outer mt membrane (OMM) for ubiquitination and targeting to the proteasome (50, 51). For example, proteasome inhibition was shown to lead to a significant accumulation of key respiratory chain proteins, both mtDNA and nDNA encoded (52), some by retrotranslocation to the OMM, where ubiquitination and degradation occur (43). Consistent with this, we observed a significant positive correlation between our HL measurements and the reported susceptibility to 20S proteasome degradation (Fig. 3D) and a highly significant correlation between HL and the relative ubiquitination levels of RC proteins (Fig. 3E). These results suggest a role for the UPS in RC protein turnover in mice; however, further studies with inhibitors or genetic manipulations to modulate the proteasome and other degradation pathways are needed to determine their contributions to RC protein turnover.

The broad variation in mt protein HLs that we observed (Fig. 2A) is consistent with data from Arabidopsis (18), Drosophila (39), and mouse (17), and the large differences between tissues (especially liver vs. others) was dramatic but consistent with these data. However, our observation that a specific pattern of heterogeneity is maintained within the RC, both within and between individual complexes, and is broadly conserved across tissue and treatments was unexpected. The possibility that this finding is evolutionarily highly conserved is suggested by the strong correlation observed between mouse RC HLs and the average of the orthologous fly protein HLs that have been reported (39) (Supplemental Fig. 2B, C). Our data support findings in studies that have shown that mt protein quality-control mechanisms (53) are highly conserved but add to the understanding that, to a large extent, common mechanisms govern RC protein quality control.

Multiprotein complexes are often responsible for major functional processes in the cell, such as protein translation (ribosome), degradation (proteasome), and oxidative phosphorylation (RC). A range of evidence in prior studies has suggested that the constituents of these protein complexes turn over at similar rates in a coordinated fashion (13, 14, 17, 54, 55). However, exceptions have also being reported, showing variability in turnover within well-defined complexes (17, 56). In our data, we detected a wide range of turnover rates between and within RC complexes. This variability in part correlated with protein localization in the mitochondria. For instance, proteins located in the matrix generally had faster turnover than their respective membrane or IMS proteins (Fig. 3A). One explanation for this could be increased oxidative damage and faster turnover in the high-ROS matrix, owing mostly to CI and -III, which produce the most of the ROS created by the RC (1, 57). Of particular interest, assembly factors in all complexes displayed faster turnover rates than did other localization groups (e.g., in CI 20% faster vs. matrix subunits, P < 0.0001; CIII 54–58% faster vs. matrix and membrane subunits, P < 0.0001), concordant with recent reports that showed faster turnover kinetics for mt assembly factors or chaperones (17, 18). Overall, these data suggest that proteins within RC protein complexes turn over in a heterogeneous fashion and that turnover correlates with subunit localization. In contrast, there was no significant correlation in HLs with stages of CI assembly, inconsistent with the view that exposed subunits that are incorporated last would turn over faster than those that are incorporated at early stages, becoming “protected” by their internalization.

Our data also show that the location of RC protein synthesis (i.e., the cytoplasm for the nDNA-encoded subunits vs. mt matrix for the mtDNA-encoded proteins) had a small, but significant effect on protein turnover: RC proteins synthesized in the cytoplasm turned over 14–16% faster, but this effect was observed equally in both total and purified mt fractions, seemingly at odds with the concept that additional proteolysis in the cytoplasm explains the HL difference in coding origins. The current literature on mt protein import suggests, however, that most nDNA-encoded proteins are translated on the OMM, tethered by the nascent polypeptide or the polysome-bound mRNA (42). If these are tethered in a way that they are copurified with the mitochondria, it would explain our observation. As an alternative, RC proteins synthesized in the cytoplasm may be subject to additional proteostatic processes during their transit through the mt IMS.

If RC proteins are recycled before they are incorporated into complexes, the 7.3-fold range of mean normalized HLs that we observed would require at least 86% of some proteins to be degraded before entry into complexes. It would also require that this process be specific to the mitochondrion, seeing that there is a very modest HL difference between proteins encoded by nDNA vs. mtDNA. Understandings of the mechanisms of mt protein turnover and the contribution from various proteolytic systems is still evolving. Mitochondria contain proteases in each mt compartment that sustain a stoichiometric balance between nDNA- and mtDNA-encoded subunits, removing excess subunits and dysfunctional, oxidatively modified, or otherwise damaged subunits (58). These proteases are found within mitochondria and play important roles in clearing proteins. Lon and ClpXp are 2 ATPase associated with various activities (AAA)+ proteases residing in the matrix, whereas 2 other AAA metalloproteases, matrix (m)AAA and intermembrane (i)AAA, embedded in the inner membrane, perform quality control in their respective compartments (5, 6). As noted above, this contrasts with mitophagy, a process that degrades whole mitochondria or mt fragments (47), although autophagy has occasionally been suggested to play a role in the selective turnover of some mt proteins (38, 59), including RC proteins, possibly involving the UPS (50).

The alternative model of dynamic RC protein exchange in which intermediate subunits and proteins can be readily exchanged or recycled when damaged or unfolded seems equally implausible, as the observed heterogeneity in turnover would require that damaged or misfolded proteins be released from complexes, followed by reassembly with new protein. In this model, factors such as the extent of damage and diffusivity of the proteins (60) may affect the rates of exchange of the different subunits, thus leading to heterogeneity in turnover rates. This outcome would be inconsistent with the current understanding of a highly structured order of complex assembly (61) and would require a view that allows for reassembly of complexes to continue from “checkpoints” in assembly rather than starting over completely, or perhaps in an order that is not well fixed at all. Such a theory would also have to account for the large variation seen in turnover rates for RC CI, -III, and -IV, which are supercomplexes (61, 62). It has been assumed that the formation of these supercomplexes prevents destabilization and degradation while enhancing electron transport efficiency. Is it possible that, even within supercomplexes, protein components are in a state of fluidity where they can be ejected and replaced as needed? A recent report supports a “plasticity model” of RC organization where supercomplexes are dynamic aggregates, a state that allows adaptability to various substrates and specific cellular requirements (62). This notion challenges the “solid-state” model of rigid high-order assembly (61).

In conclusion, this study demonstrated for the first time that the relative turnover of RC proteins is highly conserved across tissues and conditions that differ substantially in overall homeostasis. The high interprotein heterogeneity in turnover between and within RC complexes, as well as in non-RC proteins of the mt, supports a dynamic, nonunitary mechanism of protein quality control. The high degree of conservation of heterogeneity between tissues and treatments suggests that the mechanisms and rules of protein turnover are uniform, regardless of tissue type and relative long or short HLs of tissues or treatments. The similarity to previously reported Drosophila turnover rates suggests that these rules are evolutionarily conserved. The contrasting hypotheses of dynamic RC protein exchange vs. highly heterogeneous degradation of RC proteins in mt before RC complex assembly should be studied further. Regardless of the mechanism, the very high relative turnover rates of some proteins indicate that mt impose very stringent quality control, despite the high energetic cost of protein synthesis, and that understanding the rules that govern heterogeneity in turnover should provide new insights into mt dynamics.

Glossary

AAA

ATPase associated with various activities; CA, carbonic anhydrase

CI–V

complex I–V

CR

calorie restriction

EDL

extensor digitorum longus

ETC

electron transport chain

HEPES

4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid

HL

half-life

IMS

intermembrane space

LC-MS/MS

liquid chromatography tandem mass spectrometry

LECA

last eukaryotic common ancestor

MS

mass spectometry

mt

mitochondrial

nDNA

nuclear DNA

OMM

outer mitochondrial membrane

RC

respiratory chain

ROS

reactive oxygen species

RP

rapamycin

RSE

relative standard error

UPS

ubiquitin proteasome system