Low plasma lysophosphatidylcholines are associated with impaired mitochondrial oxidative capacity in adults in the Baltimore Longitudinal Study of Aging

Abstract The decrease in skeletal muscle mitochondrial oxidative capacity with age adversely affects muscle strength and physical performance. Factors that are associated with this decrease have not been well characterized. Low plasma lysophosphatidylcholines (LPC), a major class of systemic bioactive lipids, are predictive of aging phenotypes such as cognitive impairment and decline of gait speed in older adults. Therefore, we tested the hypothesis that low plasma LPC are associated with impaired skeletal muscle mitochondrial oxidative capacity. Skeletal muscle mitochondrial oxidative capacity was measured using in vivo phosphorus magnetic resonance spectroscopy (31P‐MRS) in 385 participants (256 women, 129 men), aged 24–97 years (mean 72.5) in the Baltimore Longitudinal Study of Aging. Postexercise recovery rate of phosphocreatine (PCr), k PCr, was used as a biomarker of mitochondrial oxidative capacity. Plasma LPC were measured using liquid chromatography–tandem mass spectrometry. Adults in the highest quartile of k PCr had higher plasma LPC 16:0 (p = 0.04), 16:1 (p = 0.004), 17:0 (p = 0.01), 18:1 (p = 0.0002), 18:2 (p = 0.002), and 20:3 (p = 0.0007), but not 18:0 (p = 0.07), 20:4 (p = 0.09) compared with those in the lower three quartiles in multivariable linear regression models adjusting for age, sex, and height. Multiple machine‐learning algorithms showed an area under the receiver operating characteristic curve of 0.638 (95% confidence interval, 0.554, 0.723) comparing six LPC in adults in the lower three quartiles of k PCr with the highest quartile. Low plasma LPC are associated with impaired mitochondrial oxidative capacity in adults.


| INTRODUCTION
Progressive mitochondrial dysfunction is an important hallmark of aging (López-Otín, Blasco, Partridge, Serrano, & Kroemer, 2013). A variety of changes in mitochondria have been described in aging skeletal muscle, including a reduction of number, morphological changes, reduced oxidative phosphorylation efficiency that impairs ATP production, and excess release of reactive oxygen species that cause oxidative damage and possibly accumulation of mitochondrial DNA mutations (Gonzalez-Freire et al., 2015;Kent & Fitzgerald, 2016). The decline of skeletal muscle mitochondrial oxidative capacity has been associated with lower gait speed, especially in task that require endurance (Choi et al., 2016). Higher physical activity has been associated with higher mitochondrial mass and function (Kent & Fitzgerald, 2016). Insulin resistance is associated with lower mitochondrial function, although the direction of this association is still a matter of discussion (Fabbri et al., 2017).
Recent epidemiological studies show that low concentrations of circulating species in the lipid class of lysophosphatidylcholines (LPC) are independent predictors of several aging phenotypes. Low plasma LPC 18:2 predicted memory impairment and/or Alzheimer's disease in a case-control study (Mapstone et al., 2014). Low serum LPC 17:0 and 18:2 predicted incident myocardial infarction in populationbased studies (Ward-Caviness et al., 2017). In community-dwelling adults ≥50 years, low plasma 18:2 predicted greater decline of gait speed (Gonzalez-Freire et al., 2015). LPC are biologically active lipids that comprise a major class of lipids in human plasma. LPC can serve as ligands for specific G protein-coupled signaling receptors and as a precursor to lysophosphatidic acid (LPA). LPA is an intermediate in the biosynthetic pathway of cardiolipin, an important dimeric phospholipid that is found almost exclusively in the inner mitochondrial membrane (Schlame & Greenberg, 2017). LPA also has specific receptors involved in growth and differentiation (Aikawa, Hashimoto, Kano, & Aoki, 2015).
We hypothesized that lower plasma LPC were associated with lower mitochondrial oxidative capacity. To address this hypothesis, we measured plasma LPC in a sample of adults who had muscle bioenergetics assessed using phosphorus magnetic resonance spectroscopy ( 31 P-MRS) for quantifying postexercise recovery rate of phosphocreatine (PCr), k PCr , a measure of mitochondrial oxidative capacity.

| RESULTS
The characteristics of the study population by quartile of mitochondrial oxidative capacity (k Pcr ) are shown in Table 1. The mean (standard deviation) for kPCr was 0.020 (0.005). Adults with higher k Pcr were significantly younger and taller and more likely to be male.
There were no significant differences across quartiles of k PCr by weight or BMI. Aerobic capacity (VO 2 max) was significantly higher in adults with higher k PCr . Eight LPC species were measured using liquid chromatography-tandem mass spectrometry (LC-MS/MS). The chemical structures and fatty acid chains of the LPC species are shown in Figure 1. Mean plasma LPC concentrations by quartile of mitochondrial oxidative capacity are shown in Table 2. LPC 16:1, 17:0, 18:0, 18:1, 18:2, and 20:3 concentrations were significantly higher across quartiles of k PCr , adjusting for age, sex, and height.
Adults in the highest quartile of mitochondrial oxidative capacity were also compared with the lower three quartiles combined (Table 3). Higher plasma LPC 16:0, 16:1, 17:0, 18:1, 18:2, and 20:3 concentrations were significantly associated with higher k PCr in multivariable linear regression model adjusting for age, sex, and height (p < 0.05). LPC 18:0 and 20:4 were higher in the highest quartile of mitochondrial oxidative capacity compared with the lower three quartiles in the models above, although differences were not statistically significant (p = 0.07, p = 0.09, respectively). In an alternative multivariate linear regression model adjusting for age and sex but not height, higher plasma LPC 16:1 (p = 0.005), 18:1 (p = 0.001), 18:2 (p = 0.008), and 20:3 (p = 0.001) were associated with higher k PCr, but plasma LPC 16:0, 17:0, 18:0, and 20:4 were not significantly associated with k PCr. Additional analyses were conducted using an interaction term between k PCr and sex added to the multivariable linear regression model described above to determine whether the relationship detected was between men and women.
The p-values for the interaction term (k PCr and sex) for the six different LPC species ranged between 0.19 and 0.98.
The major species of 20:3 and 20:4 are in the n-6 configuration in plasma. To our knowledge, this is the first study to identify circulating biomarkers associated with mitochondrial oxidative capacity in skeletal muscle in humans.
F I G U R E 1 Chemical structure of lysophosphatidylcholine molecular species with fatty acid group noted on the right column. LPCs consist of a glycerol backbone, a phosphate head group at the sn-3 position, a choline (C 5 H 14 NO), and a single fatty acid chain at the sn-1 or sn-2 position. A potential biological mechanism by which LPC could influence mitochondrial oxidative capacity is through its role in the synthesis pathway of cardiolipin. Cardiolipin is a unique dimeric phospholipid containing four fatty acid chains that is specific to mitochondria.
Cardiolipin is an essential constituent of mitochondrial membranes (Schlame & Greenberg, 2017). LPC in human plasma can be generated by the activity of: (a) phospholipase A 2 (PLA 2 ) on phosphatidylcholine; (b) by the activity of endothelial lipase (EL), including phospholipase A 1 (PLA 1 ), on high-density lipoprotein (Gauster et al., 2005); (c) from phosphatidylcholine during the for-

| Mitochondrial oxidative capacity measured by magnetic resonance spectroscopy in vivo
Measurements of 31 P-MRS of phosphorus-containing metabolites were obtained from the vastus lateralis muscle of the left thigh using a 3T Philips Achieva MRI scanner (Philips, Best, The Netherlands) and a 10-cm 31 P-tuned flat surface coil (PulseTeq, Surrey, UK). Participants were placed in a supine position on the bed of the scanner, with a foam wedge placed under the knee to maintain slight flexion.
| 5 of 8 baseline period of one minute immediately prior to the initiation of exercise, during exercise, and after exercise, using a pulse-and-collect sequence with adiabatic radiofrequency excitation pulses with a repetition time of 1.5 s. Four signal averages were performed for a time resolution of 6 s and total duration of MR data acquisition of 7.5 min (Choi et al., 2016). Spectra were processed using jMRUI (version 5.0), with metabolites quantified using a nonlinear least squares algorithm implemented through (AMARES) (Naressi, Couturier, Castang, Beer, & Graveron-Demilly, 2001;Naressi, Couturier, Devos et al., 2001).

| Collection of plasma
Blood was collected from participants who stayed overnight at the NIA Clinical Research Unit, Medstar Harbor Hospital in Baltimore, Maryland following a standardized protocol. Blood samples were drawn from the antecubital vein between 07:00 and 08:00 hr after an overnight fast. Participants were not allowed to smoke, engage in physical activity, or take medications before the blood sample was collected. Blood samples were immediately stored at 4°C, centrifuged within 4 hr, then immediately aliquoted and frozen at −80°C. where x denotes the number of carbons and y denotes the number of double bonds. The kit potentially measures 14 LPC, but LPC 14:0, 24:0, 26:0, 26:1, 28:0, and 28:1 were excluded from the analysis because they were below the limit of detection in nearly all subjects.

| Measurement of plasma metabolites
The MS spectra were evaluated using Analyst/MetIDQ (Biocrates) software. Human plasma samples spiked with standard metabolites were used to monitor the reproducibility of the assay. The inter-assay and intra-assay coefficients of variation ranged from 5% to 15% for nearly all analytes.

| Statistical analysis
Demographic and other characteristics were compared across quartiles of k PCr using Wilcoxon rank-sum test for continuous variables and Pearson chi-square test for categorical variables. Multivariable linear regression models were used to examine the relationship between plasma lysophosphatidylcholines and the highest quartile versus the bottom three quartiles of k PCr . Classification modeling with eight machine-learning algorithms was used to examine the ability of the six LPC to classify the highest quartile of mitochondrial oxidative capacity from the lower three quartiles. SuperLearner is an ensembling approach to machine learning is based upon the idea that no single machine-learning algorithm is optimal for all data sets (van der Laan et al., 2007). The SuperLearner algorithm was used to fit, internally cross-validate, and combine the results into a single classifier, as described in detail elsewhere; Semba et al., 2016). First, multiple individual machine-learning classification algorithms (e.g., logistic regression) were implemented, using only a portion of the data to train each algorithm. Second, we combined estimates using the data left out of the training data, producing a cross-validated estimate. The cross-validated results from the individual machinelearning algorithms were then combined via weighted average to minimize a cross-validated mean squared error. The weighted average potentially fits the data better than each individual algorithm.
Cross-validated SuperLearner additionally leaves out a portion of data in the second step to produce cross-validated estimates of the weights. SuperLearner tests multiple machine-learning algorithms and optimally combines the results to find the lowest mean squared error (Polley & van der Laan, 2015;Rose, 2013). SuperLearner was applied by splitting the data into ten mutually exclusive and collectively exhaustive groups.
We selected eight individual machine-learning algorithms to pro-

CONF LICT OF I NTEREST
The authors declare no conflicts of interest.