Proteomic signatures of in vivo muscle oxidative capacity in healthy adults

Abstract Adequate support of energy for biological activities and during fluctuation of energetic demand is crucial for healthy aging; however, mechanisms for energy decline as well as compensatory mechanisms that counteract such decline remain unclear. We conducted a discovery proteomic study of skeletal muscle in 57 healthy adults (22 women and 35 men; aged 23–87 years) to identify proteins overrepresented and underrepresented with better muscle oxidative capacity, a robust measure of in vivo mitochondrial function, independent of age, sex, and physical activity. Muscle oxidative capacity was assessed by 31P magnetic resonance spectroscopy postexercise phosphocreatine (PCr) recovery time (τPCr) in the vastus lateralis muscle, with smaller τPCr values reflecting better oxidative capacity. Of the 4,300 proteins quantified by LC‐MS in muscle biopsies, 253 were significantly overrepresented with better muscle oxidative capacity. Enrichment analysis revealed three major protein clusters: (a) proteins involved in key energetic mitochondrial functions especially complex I of the electron transport chain, tricarboxylic acid (TCA) cycle, fatty acid oxidation, and mitochondrial ABC transporters; (b) spliceosome proteins that regulate mRNA alternative splicing machinery, and (c) proteins involved in translation within mitochondria. Our findings suggest that alternative splicing and mechanisms that modulate mitochondrial protein synthesis are central features of the molecular mechanisms aimed at maintaining mitochondrial function in the face of impairment. Whether these mechanisms are compensatory attempt to counteract the effect of aging on mitochondrial function should be further tested in longitudinal studies.


| INTRODUC TI ON
Among the biological damages and dysfunctions that occur with aging, the reduction of mitochondrial oxidative capacity and consequent effects on energy availability plays a special role (Gonzalez-Freire, Adelnia, Moaddel, & Ferrucci, 2018a;Lopez-Otin, Blasco, Partridge, Serrano, & Kroemer, 2013). In fact, decreased mitochondrial oxidative capacity is the only "hallmark" of human aging that has been associated with measures of functional capacity such as muscle strength (Zane et al., 2017) and walking speed (Choi et al., 2016), two robust functional biomarkers that convey information on global health status and risk of disability, frailty, and death in older adults (Guralnik, Ferrucci, Simonsick, Salive, & Wallace, 1995;Studenski et al., 2011).
Muscle mitochondrial function can be assessed reliably and noninvasively by phosphorous magnetic resonance spectroscopy ( 31 P MRS) (Arnold, Matthews, & Radda, 1984;Prompers, Wessels, Kemp, & Nicolay, 2014). The time constant of the mono-exponential function that describes phosphocreatine (PCr) recovery (τ PCr ) after a short intense exercise provides a well-established estimate of mitochondrial oxidative phosphorylation capacity (Arnold et al., 1984;Kemp, Ahmad, Nicolay, & Prompers, 2015;Prompers et al., 2014), with smaller values of τ PCr, (more rapid return to baseline PCr levels), reflecting better muscle oxidative capacity. Andreux et al. (2018) demonstrated that τ PCr increases with aging and is smaller in active elderly adults as compared to nonactive or prefrail adults. These authors also reported a reduction of gene expression of mitochondrial genes in prefrail adults, which was consistent with lower in vivo muscle oxidative capacity measured by 31 P MRS. In agreement with these findings, a comparison of transcriptomics analysis from skeletal muscle biopsies between prefrail and nonfrail older adults found that the 10 most down-regulated genes in prefrail adults were all mitochondria-related (Andreux et al., 2018). Finally, our group recently published a discovery proteomic analysis of skeletal muscle aging and found that mitochondrial proteins were the most important class of proteins reduced with aging (Ubaida-Mohien et al. 2019a).
The mechanisms that drive the decline of mitochondrial content and function as well as resilience mechanisms that attempt to offset this decline with aging are still unclear .
In this work, we conduct a proteomic analysis of skeletal muscle biopsies obtained from healthy individuals dispersed over a wide age range. Muscle oxidative capacity, a robust measure of mitochondrial F I G U R E 1 31 P MRS measurement of in vivo muscle oxidative capacity. (a) Schematic diagram depicting energetic fluxes associated with acute muscle contraction (left) and during postexercise recovery (right). (b) Representative 31 P spectra collected before, during, and after exercise showing phosphocreatine (PCr) depletion during exercise and concomitant increase in inorganic phosphate (Pi), corresponding to these underlying bioenergetic processes. (c) Time course of PCr changes before, during, and after exercise. Red line corresponds to the fitted mono-exponential recovery model (see Experimental Procedures section) function, was assessed in vivo by 31 P MRS. Our aim was to identify proteins that were differentially expressed according to mitochondrial function, independent of age, gender, level of physical activity, and other potential confounders. Based on the set of proteins identified, we make inferences on mechanisms that may drive changes of mitochondrial function both in terms of functional decline or resilience mechanisms aimed at optimizing energy availability and utilization. Overall, we aim to identify new targets and therapeutic strategies for prevention or reversal of the decline in muscle function and mobility with aging.  Table S1. It is worth noting that this cohort exhibited a wide range of physical activity, with the older adults being slightly more active than the younger ones.

| Association between skeletal muscle energetic capacity and proteome expression
The association of more than 4,300 skeletal muscle proteins with in vivo muscle oxidative capacity as quantified through τ PCr was examined by linear mixed regression models adjusted for age, sex, physical activity, body mass index (BMI), race, and fiber type ratio (see Experimental Procedures section). Greater muscle oxidative capacity, as indicated by smaller τ PCr , was associated with increased expression of 253 proteins and decreased expression of 98 proteins ( Figure 2a).
Overall, 30% and 23% of the proteins linked to better muscle oxidative capacity were mitochondria-and nuclear-encoded, respectively, while only 9% of underrepresented proteins were mitochondria-encoded ( Figure 2b). The proteins found to have increase expression with poorer muscle oxidative capacity were preferentially cytoplasmic. An enrichment analysis indicated that most of these proteins are linked to muscle cell development (Table S2). However, no specific pathway was found within this group of proteins. Therefore, we focused our analysis and discussion on proteins associated with higher skeletal muscle oxidative capacity. Many of these proteins are part of mitochondrial structures or connected with specific mitochondrial functions, including complex I, the electron transport chain (ETC), fatty acid metabolism, and the tricarboxylic acid (TCA) cycle ( Figure 2c). A schematic representation of the mechanism of OXPHOS ATP generation is shown in Figure 3, which shows proteins that were significantly associated with better oxidative capacity (smaller τ PCr ). Many of these proteins were components of complex I, including NADH dehydrogenase 1-α (NDUFA3, NDUFA7, NDUFA13), NADH dehydrogenase 1-β (NDUFB7, NDUFB8), NADH dehydrogenase iron-sulfur protein 4 (NDUFS4), and NADH-ubiquinone oxidoreductase (MT-ND1, MT-ND5; Table 1).
Fatty acid metabolism pathway was also enriched as a function of in vivo muscle oxidative capacity. One of the significant proteins in this pathway was the liver isoform of carnitine O-palmitoyltransferase 1 (CPT1A, β = −0.25, p = .007), located in the mitochondrial outer membrane (Figure 3), which plays an essential role in the uptake of acyl-CoA activated lipids (Eaton, 2002;Gobin et al., 2003). Of note, the muscle isoform of CPT1 (CPT1B), located in the mitochondrial inner membrane, was also negatively correlated to τ PCr (CPT1B, β = −0.09, p = .12), but the association was not statistically significant. Acyl-CoA dehydrogenase proteins (ACAD8, ACAD9, ACAD10), which catalyze the addition of trans double-bonds between C 2 (α) and C 3 (β) in the Acyl-CoA substrate during the initial step of β-oxidation (He et al., 2011) (Figure 3), were also associated with better oxidative capacity, although correlation F I G U R E 3 Schematic representation of oxidative phosphorylation pathways involved in muscle bioenergetics. Complex I (green), electron transport chain (red), fatty acid β-oxidation (gray), and tricarboxylic acid cycle (TCA; blue). Mapping of significant proteins with p < .05 are shown in both the TCA and fatty acid cycles. For clarity, complex I and electron transport chain proteins are listed in Table 1 Fa y Acid with τ PCr was statistically significant only for ACAD10 (ACAD10, β = −0.45, p = .05). Additionally, the enzyme estradiol 17-beta-hydroxysteroid 8 (HSD17B8), which catalyzes the conversion of hydroxyacyl-CoA to ketoacyl-CoA in reaction with NAD (Chen et al., 2009), was associated with τ PCr (HSD17B8, β = −0.13, p = .04). Notably, peroxisomal proteins involved in fatty acid metabolism were also significantly associated with better oxidative capacity. Examples are peroxisomal multifunctional enzyme type 2 (HSD17B4, β = −0.08, p = .03) and acyl-CoA-binding domain-containing protein 5 (ACBD5, β = −0.12, p = .02), which binds medium-and long-chain Acyl-CoA esters (Nazarko et al., 2014). Figure 3 indicates other proteins significantly associated with fatty acid metabolism.
Several enzymes in the TCA cycle also had increased expression and were significantly associated with smaller τ PCr (Figure 3 Together, these results show that better skeletal muscle oxidative capacity is significantly associated with higher representation of proteins directly involved in mitochondrial energetics (respiration, OXPHOS, TCA cycle, lipid uptake, and oxidation) and substrate transport as indicated by mitochondrial ABC transporter proteins located in the mitochondrial inner membrane.

| Overrepresentation of proteins of the mRNA splicing machinery is significantly associated with better muscle oxidative capacity
Several proteins involved in mRNA splicing were also significantly associated with better in vivo mitochondrial oxidative capacity ( Figure S1). The greatest degree of upregulation was seen with TA B L E 1 List of proteins that are significantly (p < .05) associated with smaller τ PCr and identified in complex I (CI) and the electron transport chain (ETC). These proteins localize in mitochondria (M) and mitochondrial inner membrane (MIM) Enrichment analysis also unveiled a third significant protein cluster comprising several constituents of the mitochondrial protein translation machinery, including initiation, elongation, and termination. Of the proteins present in this cluster which were significantly associated with τ PCr , 39S ribosomal protein L18 (MRPL18, β = −0.43, p = .01), which facilitates the import of the 5S rRNA ribosome to mitochondria (Smirnov, Entelis, Martin, & Tarassov, 2011), exhibited the strongest association (see also Table   S3).
To further ascertain the robustness of our findings, reactome and KEGG pathway analyses were performed on proteins that were significantly overrepresented with better mitochondrial oxidative capacity. Pathways profiled according to molecular action (e.g., binding and catalysis) were in agreement with the enrichment findings; that is, the same protein clusters were found corresponding to mRNA splicing, mitochondrial protein translation and muscle bioenergetics as shown in Figure 5 (see also Figure S2).
Taken together, these findings suggest that overrepresentation of proteins that regulate alternative mRNA splicing and translation in mitochondrial ribosomes, in addition to proteins related to mitochondrial oxidative phosphorylation, may contribute to better oxidative capacity.

| D ISCUSS I ON
In this study, we combined a quantitative discovery proteomics  The current study shows that the TCA and β-oxidation cycles are also significantly associated with better in vivo oxidative capacity.
Importantly, our findings suggest that fatty acid oxidation disorder (Vishwanath, 2016) is one of the mechanisms involved in the decline of in vivo muscle oxidative capacity. Consistent with this notion, Bian and colleagues proposed that ACAD10 variation in Pima Indians may increase susceptibility to type 2 diabetes and that this effect may be mediated by impairment of insulin sensitivity via abnormal lipid oxidation (Bian et al., 2010). In addition, in a previous study, we found that in older adults without diabetes, impaired in vivo mitochondrial function (i.e., bigger τ PCr values) is associated with greater insulin resistance and a higher likelihood of prediabetes (Fabbri et al., 2017). Thus, our results confirm those suggestions in the literature that alteration of lipid metabolism and lipid oxidation may affect mitochondrial function, thereby contributing to a wide range of metabolic disorders.
Interestingly, we found that ATP-binding cassette sub-family B proteins were significantly associated with better oxidative capacity (smaller τ PCr ), while a recent study reported no association between of the expression of these proteins and chronological age or the level hand, it has been demonstrated that overexpression of ABCB10 protein increases heme synthesis and plays a role in cellular protection F I G U R E 5 Network representation of the 253 proteins associated with better muscle oxidative capacity as organized by molecular action. Nodes correspond to proteins while edges connecting nodes represent molecular action (e.g., binding, catalysis). STRING network construction was performed using high confidence values (0.7), and the robustness of major protein cluster detection was ascertained by reproducing the results even after allowing for higher number of cluster formation. Color nodes highlight enriched pathways according to reactome analysis, that is, mRNA splicing, respiratory electron transport chain, and mitochondrial protein translation. Nonconnected proteins are not represented mRNA Splicing

Mitochondrial translaƟon
Respiratory electron transport against oxidative stress during cardiac recovery after ischemic injury (Chloupkova, LeBard, & Koeller, 2003;Liesa et al., 2011). The strong association between τ PCr and ABCB10, which remained significant even after adjusting for possible confounders, suggests that adequate heme synthesis is important for oxygen transport and muscle oxygenation, especially under the high energy demand state of acute muscle contraction.
A major and novel finding of our work is the enrichment of proteins components of the mRNA splicing machinery with better muscle oxidative capacity. Alternative splicing was first described 40 years ago (Alt et al., 1980), revealing that multiple splicing variants can be generated from a single gene, thereby multiplying the protein repertoire that can be generated from a limited number of genes (Nilsen & Graveley, 2010). Previous work showed that the modulation of alternative splicing correlates with ATP depletion in neuronal cells (Maracchioni et al., 2007) and that higher physical activity is

| Limitation of this work
Overall, this work identifies the skeletal muscle proteins that are associated with in vivo oxidative capacity of skeletal muscle independent of several parameters such as age and physical activity. However, changes in protein composition that occur with aging can also be reflected by systematic changes that occur with aging in the representation of different cell types. Of primary concern is the welldescribed preferential loss of type II fibers as oppose to type I fibers, because the latter contains more mitochondria and rely on oxidative metabolism. To mitigate, at least in part, this potential source of bias, we adjusted our analyses for the ratio between MYH7 (the myosin of type I fibers) and the sum of MYH1, MYH2 and MYH4 (myosin in type II fibers). In addition, we acknowledge that interferences from other cell types and noncellular matrix that are differentially represented in older adults compared with younger adults may also affect our findings. These include fibroblast, adipocytes, endothelial cells, neurons, intercellular matrix and others. The mass of these alternative cells compared with myofibers is relatively small, their interference is probably limited, and there is currently no established method to offset their contribution in skeletal muscle proteomic analyses.
Finally, our findings were obtained in a cross-sectional analysis, and therefore, any causal inference between mitochondrial function and proteomic composition remains hypothetical and should consid-

| CON CLUDING REMARK S
To our knowledge, this is the first study that identifies skeletal mus-

| Study design and participates
Muscle biopsy and 31 P MRS data were collected from 57 healthy adults aged 23-87 years (

| In vivo muscle oxidative ATP synthesis determined by 31 P MRS
Phosphorus-31 nuclear magnetic resonance spectra were collected Exercise duration was defined to approximately achieve 50% depletion of PCr peak height (at least 30% but not more than 70%) compared with initial baseline values, while avoiding profound intracellular muscle acidosis (defined as pH < 6.8) by monitoring the chemical shift of inorganic phosphate (Pi) peak relative to PCr as a measure of pH (Paganini, Foley, & Meyer, 1997). The average value of PCr depletion overall sample size was 58.3% (±14.5) as indicated in Table S1, and a pH of 6.87 (±0.06). The series of acquired spectra were processed using Java-based jMRUI software (version 5.2) and quantified using a nonlinear least-squares algorithm, AMARES (Naressi, Couturier, Castang, Beer, & Graveron-Demilly, 2001).
The approach utilized here for determining muscle oxidative ATP synthesis rate relies on the consumption of the PCr trough exercise and then allowing a recharging of PCr through mitochondria oxidative phosphorylation after cessation of exercise.
Since the creatine kinase reaction is constantly near equilibrium, the rate of postexercise replenishment of PCr reflects the rate of the muscle oxidative ATP synthesis (Arnold et al., 1984;Kemp et al., 2015;Prompers et al., 2014). A mono-exponential function, PCr(t) = PCr ₀ + ∆PCr(1−exp(−t/τ PCr )), was used to fit the time-dependent postexercise PCr recovery, where PCr ₀ is the PCr spectral line area of phosphocreatine at the commencement of recovery (t = 0), ∆PCr is the decrease in PCr line area during exercise, and τ PCr is the recovery time constant of the phosphocreatine. τ PCr is inversely proportional to in vivo oxidative capacity of skeletal muscle, that is, oxidative ATP synthesis rate, with negligible contribution from anaerobic metabolism. In vivo muscle oxidative capacity measured using 31 P MRS recovery rate has been validated against direct mitochondrial respirometry (McCully, Fielding, Evans, Leigh, & Posner, 1993) and demonstrated to be independent of exercise intensity when intracellular pH changes are small (Arnold et al., 1984;Kemp et al., 2015;Prompers et al., 2014).

| Muscle biopsy and tissue processing
A region above the vastus lateralis muscle, identified as the midpoint of a line drawn between the great trochanter and the midpatella upper margin, was identified as the site of muscle biopsy.
A trained physician performed the muscle biopsy following a standardized protocol as described previously (Gonzalez-Freire, Scalzo, et al., 2018b;Ubaida-Mohien, Gonzalez-Freire, et al., 2019a;Ubaida-Mohien, Lyashkov, et al., 2019b;). Briefly, using conventional sterile technique and after provision of local anesthetic, a 6-mm Bergstrom biopsy needle was inserted transcutaneously into the muscle following the introduction of a gauge needle. Muscle samples were obtained under suction after incising with a coaxial blade. The tissue then was cut into small sections, which were snap frozen in liquid nitrogen and subsequently stored at − 80°C until used for analyses.
Protein concentration was determined using a 2-D Quant Kit (GE Healthcare Life Sciences) following pulverization and sonication of the muscle sample. Detergents and lipids were removed using a methanol/chloroform extraction protocol (Bligh & Dyer, 1959

| LC-MS analysis and proteomic informatics
High pH reversed-phase chromatography with a fraction concatenation approach (Wang et al., 2011)

| Other covariates
Covariates include sex, self-reported black or non-black race, and body mass index (BMI). The fiber type ratio was calculated for each participant and was expressed as the ratio between type I (MYH7) and type II myosin fibers (MYH1, MYH2, MYH4). The level of moderate-to-vigorous physical activity (MVPA) was determined using a standard questionnaire as described previously (Ubaida-Mohien, Gonzalez-Freire, et al., 2019a;Ubaida-Mohien, Lyashkov, et al., 2019b). Briefly, participants were asked whether they performed physical activity in the past two weeks. Total participation in each level of physical activity was then calculated by multiplying frequency by amount of time performed in each activity. Finally, the total activity was divided by two, to obtain minutes of MVPA per week, with four categories defied as follows: MVPA < 30 min per week coded as 0, MVPA ≥ 30 and less than 70 min per week coded as 1, MVPA ≥ 75 and less than 150 min per week coded as 2, and MVPA ≥ 150 coded as 3. The resulting ordinal variables from 0 to 3 were used in the analysis as a continuous variable.

| Statistical and enrichment analysis
The association between muscle oxidative capacity (τ PCr ) and the relative abundance of different proteins was calculated using a linear mixed regression model (R statistical software lme4, v1.1. library) that adjusted for age, physical activity, sex, race, BMI, and fiber type ratio. The TMT mass spectrometry batches were added as a random effect. The protein abundances were analyzed and reported as log2 transformed to account for the prevalence of low versus high abundance, thereby compensating for the limited dynamic range of the muscle protein distribution. Fiber type ratio and τ PCr were also log2 transformed, while those covariates that had a small range of variation (e.g., categorical), or exhibited a normal distribution (i.e., physical activity, BMI) were not log transformed. The beta coefficiant of τ PCr was calculated from the above mixed linear regression model for each protein. Statistical significance was defined as p < .05, as indicated using the lmerTest package in R Statistical Software. Proteins with a negative beta coefficient were associated with better muscle oxidative capacity (shorter τ PCr ) and proteins with a positive beta coefficient were associated with poorer muscle oxidative capacity (longer τ PCr ). Analysis was performed using R packages (3.3.4 version, Team, 2016).
To detect enrichment of certain biological processes and molecular functions involving the 253 proteins that were significantly over represented with better in vivo oxidative phosphorylation (shorter τ PCr ), an enrichment analysis was run using the Search Tool for Recurring Instances of Neighbouring Genes (STRING) tool (Szklarczyk et al., 2015). The Reactome and KEGG pathway enrichment, functional annotation clustering, and biological processes (GO function) were assessed in STRING. Annotation and localization of quantified proteins was performed based on the UniProt (UniProt 2019) and STRING database (Szklarczyk et al., 2015). A pathway was considered significantly enriched if FDR was < 0.05.

ACK N OWLED G M ENTS
This work was funded by the Intramural Research Program of the National Institute on Aging, National Institutes of Health, Baltimore, MD. We are grateful to the GESTALT and BLSA participants and the clinical team at Harbor Hospital and NIA, Linda Zukley, and Dr. Chee W. Chia for sample collection and project coordination.

CO N FLI C T O F I NTE R E S T
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
The mass spectroscopy proteomic-related data have been deposited in the ProteomeXchange consortium via the PRIDE partner repository with the dataset identifier PXD011967. The 31P MRS data will also be included in the repository and it will be publicly released after the publication of the manuscript.