Acute aerobic exercise reveals that FAHFAs distinguish the metabolomes of overweight and normal-weight runners

Background Responses of the metabolome to acute aerobic exercise may predict maximum oxygen consumption (VO2max) and longer-term outcomes, including the development of diabetes and its complications. Methods Serum samples were collected from overweight/obese trained (OWT) and normal-weight trained (NWT) runners prior to and immediately after a supervised 90-minute treadmill run at 60% VO2max (NWT = 14, OWT = 11) in a cross-sectional study. We applied a liquid chromatography high-resolution–mass spectrometry–based untargeted metabolomics platform to evaluate the effect of acute aerobic exercise on the serum metabolome. Results NWT and OWT metabolic profiles shared increased circulating acylcarnitines and free fatty acids (FFAs) with exercise, while intermediates of adenine metabolism, inosine, and hypoxanthine were strongly correlated with body fat percentage and VO2max. Untargeted metabolomics-guided follow-up quantitative lipidomic analysis revealed that baseline levels of fatty acid esters of hydroxy fatty acids (FAHFAs) were generally diminished in the OWT group. FAHFAs negatively correlated with visceral fat mass and HOMA-IR. Strikingly, a 4-fold decrease in FAHFAs was provoked by acute aerobic running in NWT participants, an effect that negatively correlated with circulating IL-6; these effects were not observed in the OWT group. Machine learning models based on a preexercise metabolite profile that included FAHFAs, FFAs, and adenine intermediates predicted VO2max. Conclusion These findings in overweight human participants and healthy controls indicate that exercise-provoked changes in FAHFAs distinguish normal-weight from overweight participants and could predict VO2max. These results support the notion that FAHFAs could modulate the inflammatory response, fuel utilization, and insulin resistance. Trial registration ClinicalTrials.gov, NCT02150889. Funding NIH DK091538, AG069781, DK098203, TR000114, UL1TR002494.


Introduction
The obesity epidemic and its associated cardiometabolic complications is an acknowledged public health crisis worldwide. Obesity, however, does not necessarily equate to obesity-related complications, as observed with metabolically healthy obese (MHO) individuals with high body mass index (BMI) (1,2). Indeed, in MHO patients (prevalence estimate of 7-13% of adults with obesity), high BMI exists in the absence of any metabolic syndrome components and presence of low HOMA-IR (1,2). Additionally, higher cardiovascular fitness, related to higher volumes of physical activity, provides multiple benefits even after adjusting for BMI, including lower mortality rates, incident of prediabetes or diabetes (3)(4)(5)(6)(7)(8)(9)(10). Studies comparing the differences between MHO and metabolically unhealthy obese (MUO) groups focus on identifying differential mortality and morbidity risks and the stability of the MHO state over time (2,11,12). Interestingly, 30-50% of MHO convert to MUO within 4-20 years (11).
Exercise continues to be a highly effective lifestyle intervention for metabolic dysfunction that decreases the risk of more than 20 chronic diseases (13)(14)(15). Exercise training improves amino acid and fatty acid profiles over a period of moderate intensity training -adaptations that resemble the metabolic differences between MHO and MUO groups (16)(17)(18). Additionally, exercise has an anti-inflammatory effect in both acute and chronic settings, contributing to the improvement of insulin sensitivity and other cardiometabolic factors in MUO groups who undergo exercise training (13,(19)(20)(21)(22). While differences between MHO and MUO are evident in the absence of lifestyle intervention, it is plausible that training may protect against transition to MUO.
The modulating effect of exercise on insulin sensitivity makes it an important context to study the differential impact of fitness versus BMI on the metabolome. However, the effect of exercise on metabolism has largely been constrained to examining normal weight trained (NWT) runners or overweight/obese subjects trained from a sedentary state, whereas the population of overweight/obese trained (OWT) runners remains understudied. This group is a relevant demographic since they embody a high fitness yet obese state whose study may reveal novel physiological adaptation to exercise (2,6,16,23). Furthermore, numerous studies in populations within the normal weight range have interrogated the metabolic responses to acute aerobic 4 exercise. However, nearly all metabolic profiling studies of overweight populations have focused on the sedentary state, or the impact of prolonged exercise interventions and weight loss.
Genomics and epigenomics platforms have uncovered mechanistic drivers at the interface of lifestyle, environmental influences, and metabolic outcomes (24). Metabolomics technologies have also revealed prospective metabolic drivers of obesity pathogenesis and cardiometabolic complications (25). Metabolomics consists of untargeted and targeted approaches that focus on semi-quantitative or quantitative measurements, respectively. Targeted metabolomics provides absolute quantities of metabolites, however only a subset of predefined and validated metabolites are under investigation for formal quantification. On the other hand, mass spectrometry-based untargeted metabolomics surveys global metabolic shifts among samples by measuring the fluctuations of multiple chemical feature abundances detected as mass-tocharge (m/z) signals (26,27). Though untargeted pipelines also capture a large number of artifactual signals, an ever-improving toolbox supports chemical validation that allows dataset curation and ultimately, selection of signals for formal validation and quantification (28)(29)(30)(31)(32)(33).
While metabolomics tools have been applied to measure the differences between MHO and MUO, or untrained versus trained groups, here we make the novel comparison between NWT and OWT regarding metabolome differences in the resting state and in response to acute aerobic exercise (16,20,(38)(39)(40)(41). To increase the metabolome coverage that spans polar and non-polar metabolites, liquid chromatography methods can be combined. Using serum samples from NWT and OWT participants pre-and post-acute aerobic exercise, we utilized hydrophilic interaction chromatography with negative electrospray ionization (HILIC-ESI(-)-MS/MS) and reverse phase chromatography with positive electrospray ionization (RP-ESI(+)-MS/MS) to profile serum metabolites (42). NWT and OWT populations were matched for age, gender, fitness level, and the absence of clinical insulin resistance. Our findings indicate acute aerobic exercise unmasks a latent metabolome in NWT versus OWT individuals, supporting an approach that potentially increases the resolving power in the prediction of clinical outcomes, compared to the analysis of 5 samples collected at baseline. Finally, this work also provides a convergent workflow that leverages the complementary strengths of untargeted and targeted metabolomics pipelines. 6

Results
Participant characteristics and study design. We enrolled normal weight and overweight or obese individuals with a self-reported history of running 3-5 sessions per week for a minimum of 30 minutes. Trained, rather than sedentary, individuals were studied to limit the effects of a deconditioned acute stress response. The cohort consisted of twenty-five runners (OWT: n = 11, NWT: n = 14; Table 1). The two groups differ by body mass index (BMI) [kg/m 2 : 30. 9  were not significantly different between the groups as determined by t-test. Though HOMA-IR was higher in the OWT group than the NWT group [1.6 ± 0.1 vs 0.9 ± 0.2, p = 0.01], the absolute value of HOMA-IR in the OWT remains within the range for normal individuals (43). Due to absence of cardiometabolic disease, OWT meet basic criteria defining MHO (2). Correlation analysis with VO2max revealed a moderately strong negative relationship between VO2max and BMI (R = -0.64, p<0.001) (Figure 1A), and a stronger relationship between VO2max and percent body fat (R = -0.92, p<0.0001) ( Figure 1B). For this study, VO2max measures were normalized to lean body mass, which did not significantly differ between groups. These results indicate fitness is negatively associated with excess body fat, even in trained groups.
At least one week prior to the acute running intervention study, the maximum oxygen consumption of each participant was assessed ( Figure 1C). The study involved a single 90minute run at 60% of individual VO2max, after overnight fast (at least 8 hours). During the run, participants wore a heart rate monitor; study staff directly supervised each participant to ensure they maintained a heart rate consistent with 60% VO2max exertion. Blood samples were collected immediately pre-and post-running bout. In parallel, all serum samples (n=50) were spiked with two internal standards (I.S. -metabolite naturally absent in the serum) whose signal was monitored to ensure method repeatability. All serum samples were extracted and analyzed using an optimized LC-MS metabolomics protocol ( Figure 1D) (44). To ensure the highest metabolome coverage, we acquired the data in positive and negative ionization modes using hydrophilic interaction (HILIC-ESI(-)-MS/MS) and reverse phase (RP-ESI(+)-MS/MS) methods 7 (42). Data analyses were performed after ensuring the I.S. areas under the curve in all analyzed samples were consistent across the batch (RSD<10%). Our untargeted workflow removes redundancies that would hinder identification of metabolites of interest in complex LC-MS signal matrices obtained for biological samples (see Methods for details). After sample processing, positive and negative mode datasets were merged, and 680 putative metabolites in total were found to be significantly altered relative to pre-exercise or NWT (adjusted p-value < 0.05, log2 fold change > 1). We then used targeted LC-MS analyses to formally identify 73 of those   Tables 3, 4). To determine the effect of exercise versus BMI, Principal Component Analysis (PCA) was performed using log10-transformed abundance of metabolites that significantly differed after correcting for multiple testing. Metabolomics shifts in LC-MS serum profile, and thus PCA group separations, were stronger in NWT or OWT groups where exercise was the studied factor (Figure 2A-B). The impact of BMI on pre-exercise serum metabolome was less easily separated by PCA, which may suggest latent differences in serum metabolome between BMI groups not evident in the resting state ( Figure 2C-D). Euclidean distance was calculated to determine the extent of distinct clustering. For the pre-exercise metabolic profile comparing NWT to OWT, two distinct clusters do indeed describe the PCA results ( Figure 2E). However, consistent with the larger number of differentially regulated metabolites between the NWT and OWT groups post-exercise, compared to pre-exercise, acute aerobic exercise caused metabolic profiles between OWT and NWT 8 groups to cluster 97.7% further apart by PCA ( Figure 2F). Additionally, distance of OWT individuals from the NWT cluster centroid is significantly greater post-exercise compared to preexercise ( Figure 2G). Together with the larger number of disrupted putative metabolites in exercise-induced profiles (349 and 241 for NWT and OWT, respectively), these data indicate stronger impact of acute aerobic exercise than BMI on the serum metabolome of trained runners, with acute aerobic exercise unveiling latent differences between NWT and OWT.
Exercise-induced metabolic profiles correlate with circulating cytokines in OWT. In addition to serum metabolomics, twelve samples (NWT = 6, OWT = 6) were used to measure cytokines pre-   Table 5). A correlation analysis between cytokine abundance and intersecting metabolic profiles (152 putative metabolites that did not vary between BMI groups, Figure 2A) revealed that OWT exhibited stronger associations with cytokine abundance compared to NWT (Figure 3C-D). Specifically, putative metabolites in the NWT profile group clustered around a Pearson Coefficient = 0, suggesting low correlation between significantly altered metabolome and cytokine production. Conversely, profile correlations in the OWT were much stronger, with most clustering higher around R = 0.7 or lower around R = -0.5. Statistical significance was determined through Fisher Z scores comparing NWT and OWT Pearson correlation coefficients for each putative metabolite. After adjusting for multiple correction, eight putative metabolites differentially correlated to IL-6 in OWT (q = 0.05) (Supplemental Table 6). Together these data suggest that changes in the metabolome are more directly related to changes in cytokine profiles in OWT than in NWT.
Untargeted metabolomics pipeline identifies canonical metabolic profile of exercise. Untargeted metabolomics is an important tool for assessing changes in metabolite pools across various groups, however due to the diversity of methods available for profiling samples, validation of an untargeted method's workflow for biologically significant features is required. The methods applied in this study revealed several putative metabolite classes known to increase with acute aerobic exercise. Among them were acylcarnitine species, whose identities were supported by comparison of MS/MS fragmentation patterns against the Metlin database ( Figure 4A); the ketone body β-hydroxybutyrate (βOHB), which was validated using an authentic internal standard ( Figure 4B); and putative free fatty acids (FFA) (Figure 4C). The abundances of these metabolites were significantly increased in both NWT and OWT groups after acute running and confirmed the expected augmentation of adipose tissue lipolysis and hepatic fat oxidation during aerobic exercise. Among these metabolites, no significant differences were observed between NWT and OWT groups post-exercise. To validate the observations revealed by the semiquantitative untargeted metabolomics pipeline, FFA species and βOHB were formally quantified using validated shotgun lipidomics and targeted UPLC-MS/MS approaches, respectively (52,53) (Supplemental Table 7). These targeted and formally quantitative approaches confirmed those generated through the untargeted metabolomics pipeline: acute aerobic exercise increased βOHB concentrations 2.1-and 3.7-fold in NWT and OWT groups, respectively (exercise effect was statistically significant in both groups, but no significant differences between NWT and OWT groups were observed); and FFA concentrations increased post-exercise between 1.3-4.6-fold in both groups, in which exercise effect was statistically significant in both groups, but with no significant differences between NWT and OWT groups (Supplemental Figure 1). Similar exercise-associated increases in acetoacetate, oxidized ketone body redox partner of βOHB, were also observed in both NWT and OWT groups (Supplemental Table 7).
In addition to exercise-engaged lipid metabolites, a small set of metabolites, the purine nucleosides or nucleobases adenosine, hypoxanthine, inosine, guanine, guanosine, and xanthine were significantly increased between 1.6-7.1-fold in the OWT group, both at baseline as well as post-exercise ( Figure 4D). Identities of these chemical features were confirmed by comparison of MS/MS fragmentation patterns against the Metlin and HMDB databases. These purine metabolites were among the top contributors to differentiating NWT from OWT both before and after exercise (Supplemental Tables 3-4, Figure 2C-D). Previous studies identified inosine and hypoxanthine as biomarkers of obesity and cardiometabolic disease (39, 54-56). Our analysis confirmed that inosine abundance shows a strong positive correlation with percent body fat (R = 0.67, p<0.0001; Figure 4E) and an inverse correlation with VO2max (R = 0.70, p<0.0001; Figure 4F). Further analysis of the relationship between putative inosine and fat mass depots revealed a positive association with subcutaneous fat (R = 0.57, p=0.003) ( Figure 4G), but not visceral fat (R = 0.34, p = 0.1). Taken together the observations generated through our untargeted metabolomics pipeline were supported by targeted approaches and were consistent with the literature, together validating our workflow to study the effect of exercise on NWT and OWT subjects. Intriguingly, abundances of known exercise-induced metabolites (FFAs, ketone bodies, acylcarnitines) did not vary between NWT and OWT runners, while the serum metabolomics profile differed primarily by purine metabolites associated with obesity.
Fatty acid esters of hydroxy fatty acids (FAHFAs) decrease in serum with acute exercise in normal weight runners. Untargeted metabolomics also revealed significant exercise-induced signals that matched the m/z of several species from the lipid class fatty acid esters of hydroxy fatty acids (FAHFAs). FAHFA lipokines have been associated with anti-inflammatory and antidiabetic effects (57,58). Palmitic acid ester of hydroxy stearic acid (PAHSA) increases in serum and adipose tissue of elderly women after a 4-month training period, though no study has revealed the responses of FAHFAs to acute aerobic exercise in trained runners (59). Given the relevance of FAHFAs to obesity, we directly quantified FAHFA lipid species with a validated shotgun lipidomics approach using NWT and OWT serum samples from pre-and post-exercise conditions (60). Consistent with prior reports, baseline concentrations among 16 FAHFA species were diminished 2-8-fold in OWT serum compared to NWT ( Figure 5A, Supplemental Table 7) (60). Baseline concentrations of PAHSA, commonly observed to be down-regulated in obesity, was decreased 1.8-fold in OWT serum relative to NWT. Other FAHFA species, such as stearic acid ester of hydroxy oleic acid (SAHOA, 5.9-fold lower in OWT relative to NWT), were even more suppressed in OWT compared to NWT ( Figure 5A). Strikingly, acute aerobic running decreased the concentrations of 22 of 25 quantified FAHFA species by 34-94% in the NWT group, while in OWT, only 5 of those were decreased, by 40-79% ( Figure 5B). In fact, in the OWT group, one FAHFA species, a linoleic acid ester of hydroxy stearic acid (LAHSA), was significantly increased (median increase of 73%) with acute running, while no FAHFA species increased with running in the NWT group. Notably, both the acute running-induced decreases in the NWT group and the dysregulated dynamic responses observed in the OWT group were in contradistinction to the conserved increases in FFA levels post-exercise in both NWT and OWT runners, which were observed using both untargeted and targeted approaches ( Figure 4C, and Supplemental Table 7).
The disparate effect of acute aerobic exercise between BMI groups on static FAHFA concentrations suggests FAHFA turnover during exercise may be modulated by body fat and/or relative insulin resistance. To determine if baseline FAHFA concentrations were related to specific adipose depots, we performed Pearson correlation analysis, which revealed that many species were negatively correlated with visceral fat mass, total body weight, and BMI ( Figure   5C), but not subcutaneous or total body fat mass. Several of the species with the strongest negative associations [e.g., palmitic acid hydroxy ester of oleic acid (PAHOA), oleic acid hydroxy ester of oleic acid (OAHOA), and SAHOA] were those that decreased with exercise in both NWT and OWT groups, despite OWT showing lower circulating concentrations pre-exercise. As expected, FAHFAs also negatively correlated with HOMA-IR ( Figure 5C). Because specific adipose depots have differential impact on circulating cytokines, we next determined if the observed variations in exercise-induced changes in FAHFA concentrations unveiled latent relationships between BMI and circulating inflammatory cytokine levels. Linear correlation analysis of covariance revealed that numerous FAHFA species showed a strong, inverse relationship with circulating IL-6 abundance in the NWT group ( Figure 5D, Supplemental Table   8), while the OWT group showed nominally positive relationships to IL-6. The species exhibiting these relationships are those that decrease with exercise in NWT, but not in OWT. Therefore, while IL-6 was not significantly different between BMI groups ( Figure 3B and Supplemental Table 5), this correlation suggests there are pathways linked to IL-6 and acute aerobic exercise involving FAHFAs that are dynamically regulated by adipose tissue mass. Taken together, integrated, adipose depot-specific, and FAHFA species-specific mechanisms may modulate both FAHFA turnover and cytokine production. Furthermore, exercise-responsive regulation of FAHFAs could depend on selective adipose depots and cytokine action, as several FAHFAs are selectively regulated in OWT despite depleted abundance at baseline.
Exercise reveals BMI impact on FAHFA and FFA fold changes in trained runners. As described above, we observed an increased number of significantly altered putative metabolites between NWT and OWT groups post-exercise and surprising differences in metabolite associations to circulating cytokines ( Figure  and other lipid mediators that are derived from elongation steps of 18:2 and are also significantly impacted by BMI. Together these results underscore that BMI and specific fat depots are differentially associated with FAHFA and FFA turnover. Targeted shotgun lipidomics showed modest differences in FFA fold change with exercise between NWT and OWT that were not statistically significant, suggesting that variation in FFA is related to underlying differences in metabolic activity of subcutaneous adipose depots (Supplemental Figure 1). 13 To evaluate the ability of combined pre-exercise FAHFA, FFA, and purine nucleoside abundances to predict BMI and cardiovascular fitness (VO2max), we applied a machine learning approach to these data. To prevent model overfitting, we applied regularized regression models (ridge regression), which includes a penalty during model fitting to yield sparse models that depend only on the most influential metabolites (see Methods for details). Resulting models applying FAHFA + FFA + purine profiles predicted BMI with an average R 2 = 0.33, (explaining 33% of variance in BMI, p = 0.03), while prediction for VO2max reached R 2 = 0.53 (explaining 53% of the variance in VO2max, p = 0.002) (Figure 6C-D), which are magnitudes consistent with other metabolomics correlation studies (39). The most significant contributors for BMI and VO2max are shown in Figure 6E-F (all model coefficients can be found in Supplemental Tables 9-10). A positive model coefficient indicates a direct relationship to the predicted variable (BMI or VO2max), while negative coefficients indicate an inverse relationship with the predicted variable of interest. Interestingly, purines grouped with several FAHFA species, including PAHSA, with a negative relationship to VO2max while the remaining highly ranked FAHFAs had the opposing sign. Among these, palmitic acid ester of hydroxy palmitoleic acid (PAHPO), palmitoleic acid ester of hydroxy linoleic acid (POHLA), and palmitoleic acid ester of hydroxy stearic acid (POHSA) were some of the few FAHFAs that did not differ significantly between NWT and OWT at baseline and may be important species for training adaptation independent of BMI (Supplemental Table 7). Notably, PAHSA was not highly ranked for predicting BMI. Oleic acid ester of hydroxy oleic acid (OAHOA), the FAHFA whose exercise-induced fold change was most influenced by visceral fat mass, also contributed to the prediction of both BMI and VO2max ( Figure 5D, 6E-F). SAHOA also contributed to both BMI and VO2max, while some FAHFA species selectively contributed to only BMI or VO2max prediction. While the significance of fatty acid chain versus hydroxy fatty acid chain is not well understood in the synthesis of FAHFAs, our results show they tend to change to similar degrees with exercise. More study is required to determine how specific FAHFA composition, including regioisomerization is to physiological impact (61). Together these data suggest FAHFAs play a role in the training effect of chronic running and may be in opposition of adenine metabolism-related effects. FAHFAs have been 14 primarily studied in relation to their influence on insulin sensitivity but this study reveals a novel role in acute aerobic exercise adaptation. 15

Discussion
This study recruited an understudied population of overweight trained individuals (OWT) to assess differences in the serum metabolome compared to NWT counterparts before and immediately following non-exhaustive aerobic exercise. The OWT group shares similarities with MHO, an important demographic, whose fitness may preempt transition to MUO. NWT and OWT groups did not differ in their fitness levels as determined by VO2max, allowing evaluation of BMI and body fat effect on the response to exercise. Notably, BMI was not associated with substantial differences in serum metabolome either before or after acute aerobic exercise in trained OWT or NWT runners. Lipidomics measurements showed exercise-induced depletion of circulating FAHFA lipokines in NWT, including several species that have yet to be studied in depth ( Figure   5B). Finally, a machine learning approach also revealed FAHFAs strongly predict VO2max ( Figure 6D, F).
The mechanisms of FAHFA turnover, including acyl chain specificity in FAHFA formation, have not been completely elucidated. Thus, the differential influence of individual FAHFA families is of interest. FAHFAs were first identified by Kahn and colleagues as a class of Glut4-regulated lipids that positively correlated with insulin sensitivity (58). Chronic PAHSA treatment in mice fed a high fat diet improved glucose uptake and insulin sensitivity in heart, skeletal muscle, and liver and protects against colitis in the gut (62,63). Efforts to identify regulation of FAHFA synthesis have linked these lipids to the Nrf2-regulated anti-oxidation pathway (64). While studies to uncover regulation of FAHFA synthesis and hydrolysis are still ongoing, ChREBP-regulated de novo lipogenesis and the Nrf2 antioxidant system in adipose tissue may play important roles (58,64). Nrf2 is activated during exercise, increasing activity of antioxidant defense pathways (65, Exercise produced similar yet distinct metabolic profiles between NWT and OWT groups that differentially associated with circulating IL-6, further suggesting an underlying influence of BMI on the physiological effect of acute running. During exercise, IL-6 mediates crosstalk between skeletal muscle and target organs, including adipose tissue, possibly stimulating lipolysis and mobilization of non-esterified fatty acids (14,(67)(68)(69)(70). IL-6 can also be induced by Nrf2 under conditions of oxidative stress. Previous study of the association between IL-6 production and lipid metabolite abundance in lean, male runners saw only a minor relationship after endurance running (71). Thus, our study is the first to show BMI-related shifts in metabolites that present a stronger relationship to IL-6 and MCP-1 production and may represent an important difference in exercise physiology due to excess body weight. Circulating metabolite and cytokine abundance is limited in its ability to reveal directionality. Correlations suggest a possible relationship between IL-6 and FAHFA clearance from circulation in NWT. Dynamic turnover of circulating FAHFAs could be attributable to numerous mechanisms that are not mutually exclusive, including synthesis, hydrolytic catabolism, or uptake/release within tissues. Our lipidomics platform measures FAHFAs composed of 16:0, 16:1, 18:0, 18:1, and 18:2, with complement hydroxy fatty acids, which decreased during exercise. Depletion of FAHFAs during exercise may be due to hydrolysis to provide fatty acids for oxidation during a long running bout (60). However, FAHFAs may also regulate glucose transport through activation of GPR120 thus their downregulation during exercise may be due to metabolic switch from glucose to fatty acid oxidation (57,58). However, FAHFA roles in skeletal muscle remain largely unknown. Our results are consistent with the notion of metabolic crosstalk between specific adipose tissue depots and skeletal muscle. Further study of FAHFAs in trained contexts is needed to further elucidate this relationship.
Together with FAHFAs, purine nucleosides contributed to VO 2max prediction by machine learning models. Purines, including putative inosine, and hypoxanthine were among the top metabolites contributing to variation between NWT and OWT before and after exercise and were highly associated with percent body fat. Inosine showed fold changes of 2.4 pre-exercise and 3.7 post-exercise in OWT over NWT. Elevation in adenine catabolism products has been reported in metabolomics analyses of pathologies associated with obesity and its levels improve with exercise training (55,(72)(73)(74)(75). A recent -omics analysis identified a downstream metabolite of purine nucleotides, uric acid, as highly associated to BMI in individuals with an abnormal metabolome (39). Exercise training of db/db mice showed restoration of uric acid and its intermediates in skeletal muscle (74). Increasing levels of hypoxanthine have also been observed in human adipose tissue under hypoxic conditions (56). These results may reflect ongoing risk for obesity-related pathologies in OWT (14,(67)(68)(69)(70)(71)(76)(77)(78).
Exercise is a 'formidable regulator of insulin sensitivity and overall systemic metabolism' (14).
Acute and chronic effects of exercise force adaptation in several systems including adipose tissue, skeletal muscle, and the liver. For this reason, exercise continues to be the most effective intervention for metabolic diseases, such as type 2 diabetes and cardiovascular disease, and could be an important strategy in preventing MHO to MUO conversion. This study showed FAHFAs and purine nucleosides significantly contributed to variation in VO2max after normalizing for lean body mass. Intriguingly, FAHFAs were negatively associated with visceral fat mass while inosine was positively associated with subcutaneous fat. These relationships may indicate competing metabolic impacts from specific adipose depots that influence overall metabolic health. Future studies to uncover the role of FAHFAs in both acute and chronic exercise may provide insight into adipose tissue remodeling in exercise and offer a node for therapeutic intervention.

Limitations
This study has several important considerations. We report the metabolomics shift in serum of well-trained individuals with normal and high BMI. Previous studies have demonstrated BMI incompletely characterizes metabolic health (39). Some individuals within the OWT group had very low body fat and their exercise-induced changes were minimal for the identified metabolic profile (Figure 2B). This cross-sectional study sought individuals with an established exercise habit, and did not acquire further details on training history, diet, or body composition prior to training. These factors need to be considered in future human studies. Due to the small sample size, additional studies of FAHFAs in both untrained and trained individuals are required to demonstrate reproducibility of the relationships among FAHFAs, cardiovascular fitness and longterm health outcomes.

Methods
Study participant details. We recruited OWT (n=11) and NWT (n=14) participants who selfreported aerobic exercise (3-5 sessions/week) from the Twin Cities metro area between July 2014 and April 2017. We preferentially recruited participants from recent running events, to ensure that they are capable to complete a prolonged (90 minute) run. Inclusion criteria were: 1) Age 18-40 years, and 2) Regular aerobic exercise, preferably running, at least 3-5 sessions/week. Individuals with 1) Self-reported clinically significant medical issues (for example diabetes, cardiovascular disease, uncontrolled pulmonary disease), 2) abnormal EKG indicating cardiac disease (study EKG performed) and 3) current pregnancy (screening pregnancy test performed) were excluded. Participants were recruited with the goal to achieve similarity in age and sex between the two groups. Acute Aerobic Exercise Intervention. The acute aerobic exercise intervention was scheduled at least 1 week after the VO2max testing to minimize influence of the strenuous exercise from the VO2 max test. Subjects were instructed to avoid intentional exercise for two days before the second visit and arrive after an overnight fast (at least 8 hours). The exercise bout was a prolonged run designed to promote fat oxidation. The heart rate reserve was calculated from the subject's resting heart rate and maximum heart rate from the VO2 max testing. For the supervised exercise bout, all subjects ran for 90 minutes on a treadmill. Each subject's run pace was initially selected by adjusting the speed and incline that achieved 60% heart rate reserve (HRR) during the VO2max, to keep all participants running at 0% grade. Heart rate was monitored during the entire run by study staff to maintain proper running intensity with Polar heart rate monitor and the Polar Beat Multi-Sport Fitness Tracker smartphone app (Polar Electro Inc., Bethpage, NY, USA). If heart rate fluctuated more than 5% HRR, study staff adjusted the treadmill speed in 0.2 mph increments until the target heart rate was maintained. Subjects were offered free access to water during their exercise bout. Plasma samples were collected pre-and post the acute aerobic exercise intervention. Data processing. Untargeted metabolomics RAW files were treated using Compound Discoverer 2.1 software. Positive and negative mode data processing used the following parameters: mass tolerance, 5 ppm; maximum retention time drift, 4 minutes using the adaptive curve algorithm, minimum scans per peak, 5; maximum peak width (full-width-half-height), 0.5 minutes. We excluded background signals using parameter setting Sample/Blank signal ratio > 3 and merged chemical features corresponding to isotopes and adducts into one putative metabolite (29).
Results from negative and positive modes data were analyzed separately. Differential abundance analysis was performed to analyze the effect of 1) BMI or 2) Acute exercise. Peaks with log2 fold change > 1 and p value < 0.05 were selected and evaluated using Compound Discoverer 2.1 visualization tools for quality of spectra, peak picking, and area integration. Poorly integrated peaks of interest were manually integrated in Thermo Quan Browser. Further negative and positive results were manually merged. Metabolites of interest were targeted using tandem mass spectrometry, and the identification was based on the MS/MS spectra available in publicly accessible HMDB, Metlin, or mzCloud databases (45)(46)(47)(48)(49)(50)(51).
Quantitation of FAHFA and FFA. Quantification of FAHFA and FFA was carried out using the validated internal standard addition methods. FAHFA and FFA species were identified and quantified through multidimensional MS-shotgun lipidomics as previously described (53,60). Derivatized extract were infused in TSQ triple quadrupole mass spectrometer (Thermo Fisher Scientific, San Jose CA) equipped with an automated nanospray device (Trivers Nanomate, Advion Biosciences, Ithaca, NY). Identification and quantification of derivatized FFA was performed using selective precursor ion mode scans using in-house automated software. FAHFA identification and quantitation was performed in product-ion analysis. Optimized collisioninduced dissociation was also used for neutral-loss scanning (60). Data processing was performed as previously described (80).  Table 1). Comparisons between the NWT and OWT groups were performed using paired t-test. Statistical significance was defined as p ≤0.05 due to the sample size and exploratory nature of this study. All statistical analyses of metabolite profile used Benjamini-Hochberg to adjust for multiple testing across metabolites (81). Cytokine analyses were performed using SAS 9.3 (SAS Institute, Cary NC). Principal Component Analysis used log10-transformed raw intensities and R packages FactoMineR and Factoextra.
Correlation between metabolic profiles and cytokines IL-6 and MCP-1 was computed for each metabolite individually using the Pearson correlation in the R base package, based on logtransformed raw intensities and cytokine abundance. The significance of differential correlations were calculated using the Fisher z-score, corrected for multiple testing, and graphed to show 95% confidence intervals.
The relationships between exercise-related change in metabolite levels (difference in log2 concentrations between post-and pre-exercise) and pre-exercise BMI, total fat mass, visceral fat mass, and subcutaneous fat mass were examined using linear models of metabolite changes, in univariate models and in models including sex and age covariates. Effects are reported as model coefficients with 95% confidence intervals after correction for multiple testing.
For predictive models for BMI and VO2max, ridge regression was used to build models using glmnet library in R. LASSO was also tested but was outperformed by ridge in terms of model generalizability, and thus, results from ridge are reported here. Due to small sample size, five unique models were generated and are summarized. Samples were randomly sampled without replacement for 5 unique training and testing sets of combined NWT and OWT (80/20 split), and the sampling was constrained to maintain the participant sex distribution in each split to remove sex differences as a confounding factor. For each training/test set split, the lambda parameter was tuned based on the training set only using 10-fold cross-validation. The optimal lambda was chosen on the basis of the training set cross-validation performance, and the resulting model was then used to predict BMI or VO2max on the test set. All correlation analyses with model predictions are based on test set predicted values. Model variables included scaled 24 concentrations of FAHFA and FFA species, scaled intensities for putative purine species, scaled age, and sex. Performance was evaluated based on the Pearson correlation coefficient between the test set predicted values and actual values. Additional models were trained with the inclusion of sex and age; however, these did not significantly improve model performance. Optimal lambda values and variable coefficients for each training set can be found in Supplemental Tables 9-10               Post-exercise abundance of βOHB in NWT and OWT groups relative to their respective pre-exercise abundance, identified by internal standard. (C) Abundances of putative free fatty acid (FFA) species in NWT and OWT groups relative to their respective pre-exercise abundance. Significance symbols (A)-(C) denote pre-to post-exercise comparison, symbols with brackets denote NWT to OWT comparison. (D) Putative purine nucleoside abundance in OWT before and after exercise relative to NWT level; significance symbols indicate OWT to NWT comparison. Scatter plot baseline abundance of inosine for NWT (black circles) and OWT (blue squares) groups against (E) percent body fat, (F) VO2max, and (G) subcutaneous fat. R value of correlation pre-and post-exercise (NWT+OWT) abundance to study measurements; significance symbols denote comparison to Pearson correlation R = 0. Putative species identified by m/z and MS 2 fragmentation. *: p≤0.05; ¥: p≤0.01; #: p≤0.001; ‡: p≤0.0001 by Student's t test with Benjamini-Hochberg correction for multiple testing. Data represent mean ± SEM.