Hepatic ketogenic insufficiency blunts exercise-induced energy expenditure and alters mitochondrial proteins in skeletal muscle

Abstract

Ketone body (KB) utilization increases during fasting and exercise due to enhanced hepatic fatty acid oxidation and KB production via the rate-limiting mitochondrial enzyme hydroxymethylglutaryl-CoA synthase (HMGCS2). Since KB metabolism intersects with multiple metabolic pathways and skeletal muscle KB catabolism rises during exercise, we tested the hypothesis that liver-specific HMGCS2 knockouts (KO) would have reduced energy expenditure (EE) and changes in the mitochondrial proteome of skeletal muscle with chronic exercise through voluntary wheel running (VWR), time-restricted feeding (TRF), or both combined to boost hepatic KB production and utilization. Control (CON) and HMGCS2 knockout (KO) mice (n = 6–8 per group) underwent sedentary ad libitum feeding (SED + AL), SED + TRF, VWR + AL, and VWR + TRF for 16 wk, with whole body EE measured using indirect calorimetry. In CON mice, VWR increased total EE by 19.5% and nonresting EE by 50% under AL conditions, and total EE by 16% and nonresting EE by 47.9% under TRF conditions. However, the EE increases seen with VWR did not occur in KO mice. Proteomic analysis revealed that the loss of liver HMGCS2 significantly impacted proteins involved in metabolic processes within skeletal muscle, including reduced oxidative phosphorylation (OXPHOS) protein expression in SED KO mice compared with sedentary CON. Notably, VWR restored OXPHOS protein expression in the muscle of the liver HMGCS2 KO but did not alter it in the CON. Furthermore, muscle from liver HMGCS2 KO mice had elevated expression of glycolytic pathways in sedentary and VWR conditions. These results indicate that hepatic ketogenic deficiency (HMGCS2 KO) diminishes exercise-induced increases in EE and uniquely impacts baseline and exercise-related adaptations in the metabolic and mitochondrial proteome of skeletal muscle.

INTRODUCTION

Over the past several decades, interventions designed to increase serum ketone bodies (KBs) have been purported to increase fat burning and reduce hunger. Fasting, exercise, and/or ketogenic diets (high fat/low carbohydrate) are commonly used strategies to stimulate ketogenesis (1). During states of low carbohydrate availability, hepatic fatty acid oxidation is elevated to increase the rate of acetyl-CoA formation, which serves as a precursor for acetoacetate (AcAc) via the rate-limiting enzyme: 3-hydroxymethylglutary-CoA synthase (HMGCS2). AcAc is further reduced to β-hydroxybutyrate (β-OHB). Liver-derived KBs serve as additional oxidative fuel sources for several prominent metabolic tissues like the brain, heart, and skeletal muscle (2, 3). Beyond serving as metabolic substrates under fasting or carbohydrate-deficient conditions, KBs also attenuate peripheral glucose utilization, induce anti-lipolytic effects on adipose tissue, decrease proteolysis in skeletal muscle, and impact whole body energy expenditure (EE) (4, 5). Several studies have shown that increasing KB levels protects against oxidative stress and muscle atrophy while elevating oxidative metabolism (6). A recent review highlighted the mixed results of altering circulating KB levels with different diet or supplement strategies (low-carbohydrate, medium-chain triglycerides, and intake of ketone esters) on energy balance, concurring that tightly controlled mechanistic studies are needed to better understand the effects of KB on whole body energy metabolism (7). Despite various studies suggesting the importance of KBs in energy metabolism, the impact of impaired ketogenesis on whole body and skeletal muscle energy metabolism remains relatively unknown, especially during physiological stimuli like fasting and exercise.

Skeletal muscle can rapidly remodel metabolic pathways to meet metabolic demands and is the primary site for exercise-induced increases in EE and a primary regulator of resting EE. Acute exercise increases skeletal muscle KB utilization (8), and total capacity to oxidize KB in muscle is increased with chronic exercise training (9). Elevated KB levels have also been shown to induce alterations in both the skeletal muscle transcriptome (10) and proteome (11). Despite these findings, little is known regarding whether decreased systemic KB availability affects the skeletal muscle proteome in a sedentary condition or in response to chronic physiological stimuli that induce ketogenesis. Being one of the main KB consumers and a source of metabolic intermediates, muscle can significantly affect whole body energy metabolism during exercise. Thus, how impaired hepatic ketone production impacts skeletal muscle in sedentary and exercise conditions remains unknown.

Therefore, the first objective of this study was to investigate whether insufficient hepatic ketogenesis would affect whole body EE in mice undergoing long-term interventions known to induce ketogenesis. Time-restricted feeding (TRF) and exercise increase hepatic ketogenesis, and the combination of both produces a synergetic effect that further elevates ketogenesis and whole body KB turnover (8). Thus, we measured total, resting, and nonresting EE via indirect calorimetry in both control (CON) and hepatic-specific HMGCS2 knockout (KO) mice undergoing chronic TRF, exercise via voluntary wheel running (VWR), or TRF + VWR. Second, we investigated whether hepatic ketogenesis deficiency and subsequent compromised ketone utilization would alter exercise-induced proteome adaptation in gastrocnemius muscle. Our findings revealed that impaired ketogenesis significantly blunts the capacity of voluntary exercise to increase total and nonresting EE while also changing the skeletal muscle proteome under sedentary and chronic exercise conditions.

METHODS

Ethical Approval and Experimental Protocol

Liver-specific HMGCS2 knockout (KO) mice were generated via a cross of the Albumin-Cre strain (Jackson Laboratories, Strain No. 035593) to the floxed Hmgcs2 allele on a C57BL/6NJ background as described previously (12). Male C57BL/6NJ control (CON) and liver-specific HMGCS2 KO mice (n = 6–8 per treatment group) were individually housed at 22°C on a reverse light cycle (dark: 1000–2200) with ad libitum access to water and were allocated to one of four treatments at an average 34–35 wk of age. Treatments included AL, TRF, VWR, or VWR + TRF. The following sample size was included in each group (SED + AL; n = 6 for controls, n = 8 for KO), (TRF + SED; n = 7 for controls, n = 8 for KO), (VWR + AL; n = 6 for controls, n = 8 for KO), and (VWR + TRF; n = 7 for controls and n = 8 for KO). The AL group had ad libitum access to low-fat diet [LFD; D12110704: 10% kcal fat, 70% carbohydrate (3.5% kcal sucrose), and 20% kcal protein with an energy density of 3.85 kcal/g, Research Diets, New Brunswick, NJ], whereas the TRF group were fasted from 0830 to 1630 Monday through Friday, with ad libitum access to food on Saturday and Sunday. Food was pulled 90 min before the dark cycle (0830) to preclude food consumption occurring before the start of the dark cycle. In addition, mice regularly start running at the onset of the dark cycle; thus, VWR + TRF induced a treatment that is on par with “fasted exercise,” resulting in a synergetic ketogenesis response before food was replaced at 1630. The mice had ad libitum access to food for the remaining 5.5 h of the dark period (from 1630 to 2200). The VWR group had ad libitum access to vertical running wheels (ENV-047V, Med Associates Inc), and running distance was continuously recorded throughout the intervention. The running wheel data over the 3 wk before indirect calorimetry were used for assessing running distance. The interventions occurred for 16 wk before the mice were placed in indirect calorimetry cages for 4 days. All animal use was in accordance with the National Institutes of Health (NIH) Guide for the Care and Use of Laboratory Animal and conformed to the principles specified in protocols approved by the University of Kansas Medical Center and University of Minnesota Institutional Animal Care and Use Committees (IACUCs).

Anthropometric

Body weight and body composition were collected weekly and before and after indirect calorimetry. Food intake was measured weekly during the 16-wk intervention and during indirect calorimetry. Body composition was determined via MRI (EchoMRI-1100, EchoMRI, Houston, TX), with the fat-free mass calculated as the difference between body mass and fat mass.

Indirect Calorimetry

Energy metabolism was measured using the Promethion indirect calorimetry metabolic monitoring system, as previously described (Sable Systems International, Las Vegas, NV) (13, 14). Food restriction for TRF groups was continued within the indirect calorimetry system, and all VWR groups maintained access to wheels while in the system. Mice were acclimated in the indirect system cages for 3 days, with data collection occurring on the 4th day. Respiratory exchange rate (RER) was calculated as Vco₂/Vo₂. Total EE was calculated as the mean EE per h using a modified Weir equation [EE (kcal/h) = 60 × (0.003941 × Vo₂ + 0.001106 × Vco₂) and extrapolated to 24 h. Resting EE was calculated from the 30-min period with the lowest EE and extrapolated for a 24-h period. Nonresting EE was calculated as total EE minus resting EE. Ambulatory cage activity was quantified by total X, Y, and Z beam breaks, as previously described (13). For VWR groups, ambulatory cage activity was added to wheel meters during data collection to determine total meters. “Fasting” and “fed” RER, EE, and total meters were calculated by averaging the data during the 8-h fasting period (0830 to 1630) and the 16-h fed period separately. Efficiency of movement was determined from the following equation: nonresting EE/cage activity [Fletcher and MacIntosh (15)]. Net cost of transport was calculated as the slope of the regression of joules_area_hrs versus wheel meters for each wheel running bout during the time period presented. Joules_area_hrs represents the EE during a wheel running bout above the interpolated REE. Linear mixed-effects modeling using a three-way analysis of covariance (ANCOVA) was performed as previously described (see Statistical Analysis) (13).

Tissue Collection

To prevent the acute effects of exercise on tissue outcomes, the axles were removed from the running wheels and placed in the cage bottom 48 h before tissue collection. Food was pulled 2 h before tissue collection at 1000. Mice were anesthetized with an intraperitoneal injection of 25∼30 µL of 50 mg/mL phenobarbital, followed by exsanguination. Blood was collected via cardiac puncture. Gastrocnemius muscle was dissected, frozen in liquid nitrogen, and stored at −80°C and processed for proteomics analysis.

Serum Ketone Measurements

Serum D-beta-hydroxybutyrate (D-βOHB) and acetoacetate (AcAc) concentrations were quantified via UHPLC-MS/MS as reported previously (16). In brief, AcAc and D-βOHB were extracted from 10 µL of serum into 40 µL of ice-cold (−20°C) methanol: acetonitrile (1:1) spiked with [U-¹³C₄]AcAc and D-[3,4,4,4-²H₂]βOHB internal standards at a final concentration of 50 µM. Samples were spun at 18,000 g for 10 min at 4°C, then 1 µL of the supernatant was injected onto a C₁₈ Cortecs UPLC T3 column (150 × 2.1, 1.6 µm) (Waters 186008500), separated via reverse-phase chromatography, and detected via parallel reaction monitoring (PRM) on a QExactive Plus hybrid quadrupole-orbitrap mass spectrometer, equipped with a heated electrospray ionization (ESI) source operated in negative ionization mode.

Gastrocnemius Muscle Preparation for LC-MS/MS Analysis

Proteins were extracted by adding 30 µL per mg of tissue RIPA buffer containing protease inhibitors and incubating on ice for 30 min followed by sonicating in a water bath for 15 min. A bicinchoninic acid assay (BCA) was performed to determine protein content, and based on the concentrations, 50 µg of protein from each homogenized and clarified mouse gastric sample was transferred to a new tube, and volumes were increased to 50 µL using 50 mM TEAB. Samples were reduced with 5 mM TCEP followed by incubation at 55°C for 30 min. Cysteines were alkylated by the addition of 375 mM iodoacetic acid (IAA) to a final concentration of 10 mM, followed by incubation at room temperature in the dark for 30 min. Ice-cold acetone was added at a 1:5 ratio, followed by incubation at −20°C overnight. After precipitation, samples were centrifuged at 14,000 g at 4°C for 10 min to pellet the proteins. The supernatant was removed, and proteins were allowed to air dry on the bench top for 15 min. The proteins were resuspended in 100 µL of 50 mM TEAB (pH 8) with 2 mM CaCl2, and proteins were digested by adding 500 ng of trypsin and incubating overnight at 37°C with shaking at 500 RPM (Thermomixer, Eppendorf). Digestion reactions were quenched by the addition of 10% formic acid to a final concentration of 1%. Digested samples were centrifuged at 10,000 g for 10 min to remove particulates, and the supernatant was transferred to a fresh tube. Peptide concentrations were measured using a Nanodrop spectrophotometer (Thermo Scientific) at 205 nm.

Proteomic LC-MS/MS Detection and Data Analysis

Samples were injected using the Vanquish Neo (Thermo) nano-UPLC onto a C18 trap column (0.3 mm × 5 mm, 5 µm C18) using pressure loading. Peptides were eluted onto the separation column (PepMap Neo, 75 µm × 150 mm, 2 µm C18 particle size, Thermo) before elution directly to the MS. In briefy, peptides were loaded and washed for 5 min at a flow rate of 0.350 µL/min at 2% B (mobile phase A: 0.1% formic acid in water, mobile phase B: 80% ACN, 0.1% formic acid in water). Peptides were eluted over 100 min from 2% to 25% mobile phase B before ramping to 40% B in 20 min. The column was washed for 15 min at 100% B before reequilibrating at 2% B for the next injection. The nano-LC was directly interfaced with the Orbitrap Ascend Tribrid MS (Thermo) using a silica emitter (20 µm id, 10 cm) equipped with a high field asymmetric ion mobility spectrometry (FAIMS) source. The data were collected by data-dependent acquisition with the intact peptide detected in the Orbitrap at 120,000 resolving power from 375 to 1,500 m/z. Peptides from samples with a charge of +2–7 were selected for fragmentation by higher-energy collision dissociation (HCD) at 28% NCE and were detected in the ion trap using a rapid scan rate. Dynamic exclusion was set to 60 s after one instance. The mass list was shared between the FAIMS compensation voltages. FAIMS voltages were set at −45 (1.4 s), −60 (1 s), −75 (0.6 s) CV for a total duty cycle time of 3 s. Source ionization was set at 1,700 V with the ion transfer tube temperature of 305°C. Raw files were searched against the mouse protein database downloaded from Uniprot on September 26, 2023, using SEQUEST in Proteome Discoverer 3.0 (17).

Gastrocnemius Muscle Proteomics Data Analysis

All proteins identified from proteomics analysis were included for pathway analysis using Ingenuity Pathway Analysis (IPA, Qiagen). log2FC and P value were used to identify up- and downregulated cellular processes, as previously performed (18). As such, the heat map in Fig. 3D represents differences in the weighted log 2-fold abundance ratio and P value rather than absolute protein intensities, and is limited by the proteins assigned to “oxidative phosphorylation.” Mitocarta3.0 (19) was used to examine mitochondrial-specific proteins identified as enriched within the proteomics data set, as done previously (18, 20). In brief, MS2 raw intensity of all mitochondrial proteins was summed and corrected for the total raw intensity of the entire proteome to compare protein abundance differences of total mitochondrial proteins and OXPHOS across all groups.

Figure 3.

Energy cost of VWR. Net cost of transport (A and B), average distance ran per bout (C and D), average wheel speed (E and F), and number of running bouts (G and H) of VWR groups during the fasting period and across 24 h, respectively. Data presented as means ± SE (n = 6–8 per group). ^tP < 0.05 main effect of TRF; ^ggP < 0.05 genotype within feeding group and time frame; ^ttP < 0.05 TRF within genotype and time frame. TRF, time-restricted feeding; VWR, voluntary wheel running.

Statistical Analysis

For each group, data are presented as means ± standard error (SE). Outliers were identified and removed using the two-standard-deviations test. A three-way ANOVA was used to test the effects of genotype (CON vs. KO), exercise (VWR vs. SED), and feeding status (TRF vs. AL) for anthropometric and indirect calorimetry data (Prism 10, GraphPad). Two-way ANOVA was explicitly used when assessing wheel meters in the VWR groups. In addition, three-way ANCOVA was used to test whether differences in fat mass and fat-free mass were significant covariates of total EE, nonresting EE, and resting EE. Adjusted marginal means and effect size [partial eta (2)] based on differences in body composition were also calculated (SPSS Statistics, IBM Corp., Armonk, NY). For analysis of protein content within the proteomics data, two-way ANOVA was conducted [genotype (control vs. KO) vs. interventions (SED + AL, VWR + AL, and VWR + TRF)]. When significant interactions and/or main effects were detected, Fisher’s least-significant difference (LSD) post hoc analysis was performed. Main effects were only described when all within-group post hoc comparisons were significant.

RESULTS

Anthropometrics and Ketosis

There was no significant difference in final body weight or fat mass between CON and KO mice, or with VWR or TRF (Table 1). Fat-free mass was reduced by ∼8% in KO compared with CON within SED + AD LIB. Fat-free mass also increased by ∼9% in KO VWR group compared with KO in SED fed AD LIB (Table 1, P < 0.05). VWR increased average weekly food intake and energy intake (∼28%) (P < 0.05) during the 16 wk intervention before indirect calorimetry compared with SED in both CON and KO. Daily running distance before the indirect calorimetry tended to be greater in CON AD LIB compared with both KO AD LIB (P = 0.1), whereas KO TRF running distance tended to be greater than CON TRF (P = 0.06) and was greater than KO AD LIB (P < 0.05) (Table 1). Following 48 h of no VWR and a 2-h food withdrawal at the end of the light cycle (0800–1000), application of our sensitive UPLC-MS/MS method showed that KO mice showed consistently lower KB concentrations than controls across all conditions (Supplemental Fig. S1), with no statistically significant effects of the TRF or VWR interventions.

Table 1.

Anthopometrics

	SED				VWR
	AD LIB		TRF		AD LIB		TRF
	CON	KO	CON	KO	CON	KO	CON	KO
BW, g	38.5 ± 1.4	35.4 ± 1.9	37.3 ± 1.7	36.5 ± 1.1	35.4 ± 2.1	36.7 ± 1.5	36.7 ± 2.5	34.1 ± 1.3
FM, g	11.2 ± 0.9	10.6 ± 1.5	10.9 ± 1.1	10.9 ± 0.8	8.5 ± 1.6	9.0 ± 1.6	9.1 ± 1.8	8.6 ± 1.2
FFM, g	25.6 ± 0.4	23.5 ± 0.3++	24.6 ± 0.4	24.4 ± 0.4	26.0 ± 0.8	25.6 ± 0.6**	25.9 ± 0.8	25.1 ± 0.5
Weekly energy intake, kcal	83.6 ± 1.30	78.7 ± 2.1	80.2 ± 3.2	79.3 ± 1.3	102.0 ± 2.8*	103.9 ± 1.9*	101.9 ± 2.5*	104.1 ±.2.6*
Wheel meter per day, km					3.2 ± 0.5	2.0 ± 0.6	2.2 ± 0.5	3.7 ± 0.5++

Data are reported as the stimated means ± SE, n = 6–8/group. AD LIB, ad libitum; KO, knockout; TRF, time-restricted feeding; VWR, voluntary wheel running.

*Main effect of VWR P < 0.05;

++genotype within activity and treatment P < 0.05.

**Activity within genotype and treatment P < 0.05.

Male HMGCS2 KO Mice Display Lower VWR-Induced Energy Expenditure

Indirect calorimetry data were collected over 24 h, and were performed while mice were maintained through their respective daily interventions of VWR or SED and the 8-h food restriction (TRF) or AD LIB feeding. VWR increased total EE in CON (11%) mice compared with the SED (Fig. 1A, P < 0.05). Interestingly, this increase did not occur in KO VWR mice, suggesting that hepatic ketogenic deficiency blunts exercise-induced total EE. Quantification of the two main components of total EE, resting and nonresting EE, revealed no differences in resting EE between genotypes, activity, or treatment groups (Fig. 1B). In contrast, both CON VWR groups had ∼40% greater nonresting EE compared with CON SED (Fig. 1D, P < 0.05), whereas VWR did not increase nonresting EE in KO mice compared with SED (Fig. 1D).

Figure 1.

HMGCS2 KO displays lower energy expenditure (EE) phenotype during voluntary wheel running (VWR). Total energy expenditure (EE) (A), resting EE (B), nonresting EE (C), average respiratory exchange ratio (RER) (D), total meters (E), efficiency of movement in CON and HMCGCS2 KO mice after 16 wk of TRF with and without VWR (F). Wheel meters (G) and cage activity (H) for VWR groups during indirect calorimetry. Data presented as means ± SE (n = 6–8 per group). g*t, genotype*TRF interaction. ^vP < 0.05 main effect of VWR; ^ggP < 0.05 genotype within feeding and activity; ^ttP < 0.05 feeding within genotype and activity; ^vvP < 0.05 activity within genotype and feeding. HMGCS2, hydroxymethylglutaryl-CoA synthase; KO, knockout.

Exercise and TRF impact substrate utilization by increasing systemic fat metabolism and enhancing ketogenesis (8). We hypothesized that TRF would lead to greater fat utilization and, consequently, a lower RER in the CON TRF groups. Contrary to our expectations, TRF in CON SED mice did not change RER compared with AD LIB, whereas CON VWR TRF mice had a higher 24-h average RER compared with AD LIB (Fig. 1C, P < 0.05). RER was not different between KO mice and CON mice; moreover, TRF and VWR did not influence RER in the KO mice quantified over an entire 24-h period (Fig. 1C). EE is highly correlated with body mass and fat-free mass (21) and is impacted by adiposity (22, 23). To assess whether differences in EE were due to the differences in body composition, we performed an analysis of covariance using fat mass and fat-free mass as covariates to generate estimated marginal means and assess covariate effect size (Table 2). Although both fat and fat-free mass were significant predictors of variability in total and resting EE (P < 0.05), the effect sizes for fat and fat-free mass were small, and the comparisons of the estimated marginal means were not different from those in Fig. 1.

Table 2.

Energy expenditure: analysis of covariance

	SED				VWR
	AD LIB		TRF		AD LIB		TRF
	CON	KO	CON	KO	CON	KO	CON	KO
Total EE, kcals, covariates: FM + FFM	11.450 ± 0.491	11.333 ± 0.469	11.269 ± 0.458	11.290 ± 0.443	13.374 ± 0.528**	12.001 ± 0.425++	12.779 ± 0.475**	12.097 ± 0.434

Covariate Effect Size	Covariate	P Value	Partial eta2
	FFM	0.005	0.153
	FM	0.015	0.117
Resting EE, kcals, covariates: FM + FFM	8.404 ± 0.412	8.328 ± 0.393	8.220 ± 0.384	8.118 ± 0.372	8.930 ± 0.443	8.691 ± 0.356	8.394 ± 0.398	8.611 ± 0.364

Covariate Effect Size	Covariate	P Value	Partial eta2
	FFM	0.014	0.120
	FM	0.012	0.124
Nonresting EE, kcals, covariates: FM + FFM	3.046 ± 0.214	3.006 ± 0.204	3.049 ± 0.199	3.172 ± 0.193	4.443 ± 0.230**	3.310 ± 0.185++	4.385 ± 0.206**	3.486 ± 0.189++

Covariate Effect Size	Covariate	P Value	Partial eta2
	FFM	0.071	0.066
	FM	0.442	0.012

Data are reported as the estimated mean ± SE, n = 6–8/group. AD LIB, ad libitum; EE, energy expenditure; KO, knockout; TRF, time-restricted feeding; VWR, voluntary wheel running.

++Genotype within activity and treatment P < 0.05;

**activity within genotype and treatment P < 0.05.

We also examined whether genotype, VWR, or TRF influenced total movement in the indirect calorimetry cages. Total meters encompasses both ambulatory cage movement and VWR distance over a 24-h period in the calorimetry cages. Total Meters were not affected by genotype or treatment in SED mice (Fig. 1E), whereas VWR increased activity across all groups as expected (P < 0.05). Within the VWR groups, an interaction between genotype and TRF was observed (Fig. 1E, P < 0.05). Total Meters was lower in CON TRF mice compared with CON AD LIB (P < 0.05). However, although KO AD LIB mice showed decreased Total Meters than CON KO AD LIB mice (P < 0.05), KO TRF mice exhibited greater Total Meters than KO AD LIB (P < 0.05), and Total Meters of KO TRF trended higher than CON TRF (P = 0.054). To explore whether this interaction was due to differences in the ambulatory cage movement versus VWR components of cage activity, we compared wheel meters (Fig. 1G) and ambulatory cage activity in VWR groups (Fig. 1H). The interaction between genotype and TRF was again evident in wheel running distance, but not in ambulatory cage activity. Given the differences in nonresting EE and cage activity between genotypes and treatments, we calculated movement efficiency, defined as nonresting EE divided by cage activity (cal/m). A higher value indicates less efficiency, and a lower value indicates greater efficiency (15). As expected, VWR significantly increased movement efficiency in both genotypes (Fig. 1F, P < 0.05). An interaction between genotype and treatment was also observed in VWR mice (P < 0.05). CON TRF mice tended to be less efficient than AD LIB (P = 0.052). Although KO AD LIB mice demonstrated lower efficiency than CON (P < 0.05), KO TRF mice showed greater efficiency than both KO AD LIB and CON TRF (P < 0.05). To assess the impact of voluntary wheel running on the total meters of the VWR group, wheel meters (Fig. 1G) and cage activity (Fig. 1H) were determined. Similar findings to total meters were observed for wheel meters, and no difference was observed in cage activity between VWR groups. Overall, these findings indicate that during increased activity (wheel running), hepatic ketogenic insufficiency reduces EE and interacts with food restriction to influence movement efficiency.

Hourly EE, RER, and Movement

After identifying differences in total meters, RER, and EE, we investigated whether these factors exhibit temporal variations (hour by hour) between genotypes and treatments within either sedentary or VWR conditions. Although no significant differences in 24-h EE were found among SED groups (Fig. 1A), both CON and KO SED + TRF mice exhibited reduced EE during the TRF period (Fig. 2B, P < 0.05) and increased EE when they had access to food (P < 0.05) compared with their within-genotype AD LIB counterparts. In addition, KO SED AD LIB mice had lower EE than CON during the restricted feeding period (Fig. 2B, P < 0.05), and SED KO TRF mice showed higher EE compared with KO AD LIB during ad lib food access (P < 0.05). In VWR groups, CON AD LIB mice displayed higher EE than both CON TRF and KO AD LIB during the food-restricted period (Fig. 2D, P < 0.05). Unlike the SED groups, no differences in EE were identified during periods when mice had access to food between VWR groups. However, CON AD LIB mice had decreased EE during ad lib food access compared with the food-restriction period (P < 0.05). Similar to the EE data in SED mice, RER measurements in SED groups were influenced by food availability when analyzed on an hourly basis. SED RER was lower in both TRF groups during restricted food access (Fig. 2F, P < 0.05) and higher during food-access periods (P < 0.05) compared with AD LIB groups. Furthermore, SED CON TRF mice exhibited higher RER during ad lib food access than CON AD LIB (P < 0.05). In VWR groups (Fig. 2H), TRF tended to lower RER in both CON (P = 0.09) and KO (P = 0.06) compared with AD LIB during food restriction (Fig. 2H). During ad lib feeding, RER was elevated in both KO groups and CON TRF compared with the respective food restriction periods (P < 0.05). KO AD LIB mice and CON TRF mice showed higher RER than CON AD LIB during the ad lib feeding period (P < 0.05). Interestingly, RER was increased in VWR KO AL mice over VWR CON AL mice during the ad lib feeding period, an effect that not found in SED mice (Fig. 2H). Similar to the 24-h cage activity data, SED total meters was lower in KO TRF mice compared with AD LIB during food restriction (Fig. 2J, P < 0.05) and tended to be higher during ad lib feeding in this group compared with restricted feeding (P = 0.07). The distinctive interaction observed in VWR mice regarding total meters in the averaged 24-h cage activity analysis was further emphasized in the hourly analysis (Fig. 2, K and L). Specifically, KO AD LIB and CON TRF mice had lower total meters than CON AD LIB during restricted food access (P < 0.05), whereas KO TRF total meters was higher than both KO AD LIB and CON TRF in the same period (P < 0.05). Although all groups except KO AD LIB exhibited lower total meters during ad lib feeding compared with restricted feeding (P < 0.05), the same pattern was observed across groups during ad lib feeding (P < 0.05). To confirm that differences in hourly movement and EE were not artifacts of single-day analysis, we examined the movement data for the VWR groups on the final day of habituation (day 3). All VWR groups exhibited similar outcomes as they did on day 4 (Supplemental Fig. S2). Together, these findings demonstrate that hepatic ketogenic insufficiency and TRF interact to create temporal differences in EE (Fig. 2, B and D) and Total Meters (Fig. 2L) in mice with access to running wheels during the normal dark cycle feeding period.

Figure 2.

Hourly energy expenditure (EE), respiratory exchange rate (RER), and movement (left, sedentary; right, VWR). Hourly EE for SED group (A); average EE for SED group in TRF Time frame vs. Ad Lib time frame (B); hourly EE for VWR group (C); average EE for VWR group in TRF time frame vs. ad lib time frame (D); hourly RER for SED group (E); average RER for SED group in TRF time frame vs. ad lib time frame (F); hourly RER for VWR group (G); average RER for VWR group in TRF time frame vs. ad lib time frame (H); hourly cage activity for SED group (I); average movement for SED group in TRF Time frame vs. Ad Lib time frame (J); hourly cage activity for VWR group (K); average movement for VWR group in TRF time frame vs. ad lib time frame (L). Data presented as means ± SE (n = 6–8 per group). ^ggP < 0.05 genotype within feeding group and time frame; ^ttP < 0.05 TRF within genotype and time frame; **P < 0.05 time frame within genotype and feeding group. SED, sedentary; TRF, time-restricted feeding; VWR, voluntary wheel running.

Running Wheel Patterns

The increased wheel running and decreased EE of KO TRF during food restriction suggest greater efficiency of movement while running. Therefore, we calculated net cost of transport for wheel running during the food restriction period and over 24 h (Fig. 3, A and B, respectively). Net cost of transport was reduced in both TRF groups compared with AD LIB (P < 0.05), and tended to be lower in KO TRF compared with CON TRF (P = 0.1). Differences in wheel running bout intensity (i.e., distance and speed per bout) could be involved in differences in net cost of transport (24, 25). During the fasting period, KO AD LIB and CON TRF mice ran ∼40% less per running bout compared with CON AD LIB (Fig. 3C, P < 0.05), whereas no differences in running bout length were observed between KO or TRF groups. Also, average wheel speed was consistent across groups during the fasting period (Fig. 3E). The data for running bout distance and wheel speed across the 24-h period (Fig. 3, D and F, respectively) were similar to the fasting period. Finally, the number of running bouts during the fasting period was ∼40% less in KO compared with CON in the AD LIB fed mice, whereas KO TRF had increased fasting period running bouts compared with KO AD LIB (Fig. 3G, P < 0.05). The 24 h number of running bouts was not different by genotype or feeding paradigm (Fig. 3H). These findings demonstrate that chronic TRF results in adaptations resulting in reduced energy cost during wheel running, with a further improvement in efficiency in mice that lack hepatic ketone body production.

HMGCS2 KO Mice and Skeletal Muscle Proteome Adaptation to Exercise

Skeletal muscle is a crucial site of energy metabolism during exercise. Since KO showed decreased nonresting EE during activity, we performed untargeted proteomics on skeletal muscle (whole gastrocnemius) to explore pathways that might influence EE. Our rationale was that skeletal muscle is a key site of energy metabolism during exercise and also a major site for ketone utilization. Driven by the calorimetry outcomes, conditions focused on SED + AL, VWR + AL, and VWR + TRF, all in CON and KO cohorts. The top upregulated and downregulated pathways are shown in Fig. 4, using IPA analysis (IPA, cutoff Z-score = ±2.0, −log P value > 1.3). Within genotype, condition-dependent comparisons are shown in Supplemental Fig. S3. In the SED + AL condition, KO muscle shows a general reduction in proteins involved in oxidative phosphorylation (OXPHOS) (Fig. 4A, Z-score = −4.92) compared with controls. VWR slightly lessens this difference between KO and controls (Fig. 4B), changing the Z-score from −5 to −3.25 for OXPHOS and from −4.92 to −2.25 for ETS, although they remain significantly suppressed in KO mice. When VWR is combined with TRF, the OXPHOS pathway is no longer significantly diminished in KO skeletal muscle (Fig. 4C) compared with CON, and indeed the VWR + TRF condition in KO increases OXPHOS-related protein expression compared with SED AD LIB (Supplemental Fig. S4). Pathway analysis of KO compared with CON was combined to illustrate how VWR and VWR + TRF influence the gastrocnemius proteome (Fig. 4D). In the VWR + TRF groups, KO shows higher protein levels linked to OXPHOS (Fig. 4D, Z-score = 2.77) than controls. Due to differences in pathways involving OXPHOS proteins across genotypes and TRF conditions, we cross-referenced these proteins with MitoCarta3.0 to identify known mitochondrial proteins within our dataset. The comparison between CON and KO mice revealed no differences in total mitochondrial proteins between KO and CON (Fig. 4E), indicating that overall mitochondrial content differences do not cause variations in the OXPHOS pathway. We then normalized all OXPHOS proteins to total mitochondrial protein content. In the SED + AL condition, KO mice have significantly fewer OXPHOS proteins in skeletal muscle than controls, but VWR in KO mice mitigates this difference regardless of whether it occurs under AL or TRF conditions (Fig. 4F). Interestingly, VWR + TRF has contrasting effects across genotypes. In CON, VWR + TRF reduces OXPHOS protein levels in skeletal muscle, whereas in KO mice, it significantly increases them (Fig. 4F, genotype × VWR + TRF < 0.05). Aside from mitochondrial proteins, IPA analysis shows that KO skeletal muscle has increased expression of various other pathways in response to VWR + TRF compared with CON. For example, the EIF2AK4 [general control nonderepressible 2 (GCN2)] pathway responds to VWR + TRF, with a Z-score shifting from −3.24 in SED + AL to 2.60 in VWR + TRF in KO mice (Fig. 4D). Other pathways affected by VWR + TRF in KO include neutrophil extracellular trap signaling, eukaryotic translation elongation, and ribosomal quality control signaling (Fig. 4D). We also performed separate analyses within the CON and KO groups to visualize how different interventions affected skeletal muscle protein expression differently (Supplemental Fig. S4).

Figure 4.

Comparison of skeletal muscle proteome between CON and KO. Pathway analysis for SED+AD LIB groups—KO relative to CON (A), pathway analysis for VWR + AD LIB groups—KO relative to CON (B), pathway analysis for VWR + TRF—KO relative to CON (C), heat map visualization for IPA across three different groups KO relative to Con (D), mitochondrial protein over total protein expression across all groups (E). F: OXPHOS protein normalized by mitochondrial protein. For E and F, data are presented as means ± SE (n = 4 per group). ^ggP < 0.05 genotype within feeding and activity; ^vvP < 0.05 activity within genotype and feeding; ^v+tP < 0.05 feeding and activity within genotype. KO, knockout; OXPHOS, oxidative phosphorylation; TRF, time-restricted feeding.

Significant Genotype Difference in Protein Expression Associated with Glycolysis

Although ketone metabolism pathways in muscle were not significantly different between genotypes (absolute Z-score < 2.0), we specifically examined proteins associated with KB metabolism due to its relevance to our study (Supplemental Fig. S3). Z-scores were shown under all three conditions (Supplemental Fig. S3A). Under SED + AL condition, CON has higher 3-hydroxy-3-methylglutaryl-CoA lyase like 1 (HMGCLL1) protein expression, a protein involved in leucine degradation and KBs synthesis (26). VWR + AL and VWR + TRF both decreased HMGCLL1 protein expression in the CON but did not affect the KO (Supplemental Fig. S3B, interaction genotype × VWR < 0.05; interaction genotype × VWR + TRF < 0.05). To our surprise, the genotype difference in protein expressions related to ketone utilization was minimal. Different interventions have minimal impacts on protein expression associated with ketolysis (Supplemental Fig. S3, C–E). Since we observed a consistent increase in glycolysis-associated protein expression (Fig. 3, A and B), we examined the abundances of specific glycolytic enzymatic mediators. Gastrocnemius muscle from HMGCS2 KO showed lower hexokinase 2 (HK2) protein expression across all three conditions (Fig. 5), but higher levels of glucose-6-phosphate isomerase (GPI) and triosephosphate isomerase 1 (TPI1) (Fig. 5; main effects of genotype, P < 0.05.) Under both AD LIB and TRF conditions, VWR increased GPI expression compared with SED + AL (Fig. 5). When VWR was combined with TRF, KO had higher GPI expression than when VWR was combined with AL, but this increase was not seen in CON mice (Fig. 5, genotype × TRF < 0.05). This genotype × TRF interaction was also captured in phosphofructokinase (PFKM) (catalyzes the rate-limiting glycolytic reaction) and TPI1 proteins (Fig. 5). VWR plus TRF increased expression for PFKM, TPI1, and GAPDH in the KO compared with SED + AL, but this increase did not occur in CON muscle (genotype × VWR + TRF < 0.05). VWR + TRF increased phosphoglycerate kinase 1 (PGK1) and phosphoglucomutase 1 (PGM1) protein levels regardless of genotype compared with SED + AL (main effect, P < 0.05). Across all conditions, KO generally showed higher PGM1 protein levels than CON. These results further demonstrate that liver HMGCS2 KO leads to substantially different regulation of multiple pathways in skeletal muscle in response to various interventions, highlighting the pronounced influence of hepatic KB production on skeletal muscle metabolism and mitochondrial profiles. However, future work is necessary to assess whether these protein expression differences result in functional changes in glycolysis and glucose oxidation.

Figure 5.

Significant genotype differences in glycolytic-associated proteins. Raw intensity protein abundance for glycolysis proteins: HK2 (A), GPI (B), PFKM (C), TPI1 (D), GAPDH (E), PGK1 (F), PGM1 (G). Data are presented as means ± SE (n = 4 per group). ^vP < 0.05 main effect of VWR within feeding, ^tP < 0.05 main effect of feeding within activity, ^v+tP < 0.05 main effect of VWR+TRF vs. Sed+Ad Lib, ^ggP < 0.05 genotype within feeding and activity; ^vvP < 0.05 activity within genotype and feeding; ^ttP < 0.05 TRF within genotype and activity; ^vv+ttP < 0.05 feeding and activity within genotype. TRF, time-restricted feeding.

DISCUSSION

Our main finding is that mice with hepatic ketogenic insufficiency show impairment in both 24 h total and nonresting EE in response to exercise compared with controls and a dissociation between EE and physical activity. This phenotype was latent in the sedentary state but was provoked by chronic exposure to daily VWR. In addition, the lower EE phenotype in KO mice occurs in both AD LIB and TRF VWR groups. Interestingly, although reduced nonresting EE was associated with reduced total activity in the KO VWR AD LIB mice, KO VWR TRF mice showed greater total activity than KO AD LIB and CON TRF. These differences in the relationship of EE to total activity resulted in KO VWR TRF mice having better efficiency of movement. Perhaps most striking was the observation of increased activity of KO VWR TRF during the food-restricted period without a consummate increase in EE, as observed in CON VWR TRF mice. We also investigated muscle proteomics to assess whether reduced ketone body availability impacted ketone and mitochondrial metabolic pathways. Our results show that the liver ketogenic insufficiency of KO mice markedly reduces skeletal muscle OXPHOS proteins in SED AD LIB mice, which is reversed by the increased activity of VWR and time-restricted feeding. Overall, our findings suggest that hepatic ketogenesis plays a crucial role in regulating both whole body energy metabolism and skeletal muscle proteome adaptations to exercise and TRF.

As an alternative fuel source, ketone metabolism interacts with various energy pathways such as β-oxidation of fatty acids, de novo lipogenesis, sterol biosynthesis, and glucose metabolism (27). The capacity of exercise to increase ketone metabolism is well described (5); however, the impact of physical activity/exercise on systemic EE is an ongoing topic of debate (28, 29). Specifically, whether is a linear relationship between total EE, constrained at high levels of physical activity, or whether reductions in components of total EE or increased sedentary behaviors compensate for the increased activity EE. Often constraint/compensation in energy budget is supported by observed reductions in resting EE (30); however, these are often weight loss studies (31). Recently, data from overweight, sedentary men and women showed increased total EE following a 12-wk physical activity intervention, however, a reduction in resting EE was also observed (30). Alternatively, in a cross-sectional study, total EE was observed to be directly associated with physical activity level in trained, weight-stable, men and women (32), with no relationship between physical activity and resting EE. In rodent experiments measuring EE in weight-stable animals, total EE typically is observed to increase with increases in physical activity (24, 33–37), with no change in resting EE (33, 34, 37, 38). However, a compensatory decrease in resting EE has been suggested (36). Nevertheless, our findings of no increase in 24 h total or nonresting EE in mice with liver-specific ketogenesis insufficiency following over 16 wk of wheel access compared with wild-type littermates are notable (Fig. 1), in particular, with no observed differences in resting EE. This finding of no apparent adaptive increase in EE is more novel when considered against the physical activity data. The more than 50% reduction in total meters of ad lib fed KO mice compared with CON is associated with decreased nonresting and no increase in total EE. However, the similar total EE and nonresting EE of TRF mice is dissociated with total meters. Where WT mice had an ∼45% reduction, and KO mice had ∼2-fold increase in total meters compared with ad lib. To our knowledge, our data represent the first observations of a dissociation of EE and physical activity through an interaction of loss of hepatic ketogenesis and TRF.

Wheel running represents a measure of increased physical activity in rodents, as it represents a rewarding voluntary behavior (39). However, it has been observed to model the physiological effects of aerobic exercise in humans (40, 41). The majority of running bouts occur in the dark cycle, with most occurring in the first half of the dark cycle (42). In general, all for VWR groups modeled this behavior (Fig. 2K). The dissociation between EE and physical activity is visually apparent during the food-restricted period of the dark cycle (Fig. 2, C and K). Averaging across the food-restricted period highlights the reduced EE of KO and TRF mice (Fig. 2D) and the very similar pattern of total meters (Fig. 2L). To date, a small number of studies have assessed energy expenditure with some type of TRF protocol with/without exercise; however, the comparison of these findings to ours is complicated by all being high-fat diet studies, and none use VWR (43–46). Importantly, all these studies show significant decreases in RER during fasting, with associated increases upon food availability. TRF in both SED and VWR results in lower RER during the fasting period and increased RER during ad lib feeding (Fig. 2, E–H), with VWR having no apparent impact. As the majority of food intake occurs in the dark cycle, future work will aim to precisely assess food intake during TRF in VWR mice to examine how this variable interacts with VWR and ketogenic insufficiency.

Typically, rodents spontaneously run when presented with a wheel and gradually increase wheel running over the first few weeks until a steady-state is reached (39, 42). Some of the previously cited studies have examined energy expenditure and energetic cost of movement in mice during the initial exposure to wheel running (24, 36), before many of the observed tissue-specific and systemic physiological adaptations often associated with exercise can occur (39, 41). As such, our chronic TRF/VWR experimental paradigm allows for assessment of the interaction of these experimental stimuli with loss of hepatic ketogenesis on the energy cost of movement. Our summary data of the efficiency of movement (Fig. 1F) demonstrate that a significant interaction between these factors does exist. However, this measure does not consider EE or movement specific to wheel running. Calculation of the net cost of transport (Fig. 3, A and B) shows that chronic TRF results in reduced energetic cost of wheel running, which tends to be further reduced in KO mice. Interestingly, these differences in the net cost of transport for wheel running do not exist when assessed across a 24-h period. Previously, the intensity of voluntary wheel running and off-wheel activity has been shown to impact the relationship of EE to physical activity and the energy cost of movement (24, 25, 37). Future studies will aim to examine this relationship in our model of reduced liver ketogenesis and the molecular and functional phenotypes of this observed difference in the efficiency of movement. Overall, these novel indirect calorimetry findings demonstrate that hepatic ketone insufficiency during VWR and TRF results in reduced energy cost of movement, resulting from a dissociation between EE and physical activity.

KBs also exert many signaling functions that regulate exercise-induced adaptation in skeletal muscle. We investigated whether there were specific skeletal muscle adaptations that could contribute to the lower TEE/nonresting EE phenotype seen in the liver HMGCS KO. Utilizing pathway analysis, we identified that the SED KO group possess lower mitochondrial respiratory proteins than the equivalent CON mice, but this genotype difference was attenuated with VWR. Oxidative phosphorylation is significantly more efficient than glycolysis, and this higher OXPHOS protein expression is observed in KO, which may facilitate a more efficient energy production process and result in lower TEE and NREE. Although these proteomics findings are suggestive, without more in-depth assessment of skeletal muscle substrate flux, mitochondrial respiratory capacity, and OXPHOS enzyme activities, we are unable to make conclusions about the efficiency of OXPHOS capacity between the CON and KO. Interestingly, we did not observe any significant differences in the protein associated with ketolysis across the genotype. However, we did notice a significant difference in protein expression associated with the glycolysis pathway: KO has higher glycolysis protein expression than CON. It suggested that in response to the lack of ketone as a fuel substrate, KO switches to glycolysis as one of its major energy source. Previously, studies have shown that ketogenic insufficiencies increase glycogen storage and glycogenolysis within the liver. It is plausible that ketogenic insufficiencies and the subsequent reduction in ketone availability alter muscle glycogen storage. This change in muscle glycogen storage could, in turn, induce distinct glycolysis protein expression patterns when comparing control (CON) and knockout (KO) groups. As mentioned earlier, without data demonstrating changes in skeletal muscle glucose storage and utilization, we are unable to draw any specific conclusions regarding the efficiency of ketone metabolism or glycolysis between the CON and KO.

Skeletal muscle is also an energy consumer that detects nutrient availability, which could possibly affect energy metabolism. Proteomic analysis revealed that KO had more significant changes in proteins associated with general control nonderepressible 2 (GCN2) response to amino acid deficiency. GCN2 is a serine/threonine-protein kinase that senses amino acid deficiencies and plays a vital role in modulating amino acid metabolism in response to nutrient deprivation (47). GCN2 also promotes the phosphorylation of eIF2a, which facilitates adaptation to nutrient stress (47) and lowers protein synthesis via conserving energy and nutrients (48). Although many studies have shown that amino acid deprivation activates GCN2 (49, 50), little is known about how ketone deficiency alters the GCN2 pathway. Proteomics analysis showed that skeletal muscle in the HMGCS2KO upregulates protein expression associated with GCN2 pathway, which, as a result, could potentially trigger a series of changes to lower energy expenditure and conserve nutrients/energy. However, whether or not the GCN2 pathway plays a primary regulatory role is unknown.

Our ability to interpret the findings is limited by several factors. The relatively small sample size is a major limitation, particularly in regard to the comparisons across three experimental levels and the inability to perform regression analysis of EE components and total meters. Also, the lack of consideration of sex as a biological variable is a limitation. Previous studies suggest that sex differences in lipid metabolism exist, with females demonstrating a higher fat utilization during exercise (51). We would anticipate that female mice would present a different response to the 16-wk intervention under the various conditions. In addition, food intake directly changes RER and impacts systemic energy metabolism during exercise. The small sample size and observed hoarding within these experiments limited the collection of rigorous food intake data. Future experiments will focus on the impact of food intake on the observed HMGCS2 RER and EE phenotype. Furthermore, the serum ketone analysis was not performed when VWR animals had access to wheels and were actively running. This limits the observation of the potential additive impact of food restriction and exercise on systemic ketone levels (5, 8) and how this may be involved in our observations. Future work will dedicate experimental groups to the collection of this important data. Finally, our assessment of skeletal muscle adaptation to increased activity and TRF is limited to proteomic analysis of the gastrocnemius muscle and did not include functional measures of mitochondrial respiratory capacity or glucose/fat oxidation. Our recent publication describes experimental conditions to assess mitochondrial ketone body respiration (52), allowing for more thorough interrogation of tissue ketone body metabolism.

In summary, this study uses liver-specific HMGCS2 knockout mice to demonstrate that a significant reduction in hepatic ketogenic capacity interacts with food restriction to blunt exercise-induced energy expenditure and alters mitochondrial and metabolic proteomic signatures in skeletal muscle. These baseline differences in skeletal muscle proteome between CON and KO also likely affected responsiveness for various pathways to adapt to VWR, TRF, and TRF + VWR. Future investigation should focus on whether it is the metabolic pathways that synthesize KB within the liver or reduced utilization of KB in peripheral tissues like skeletal muscle that alters the capacity for exercise to modulate energy expenditure at the whole body level.

DATA AVAILABILITY

Data will be made available upon reasonable request.

GRANTS

This study was funded by S10 OD028598 (to J.P.T.), K01DK112967 (to E.M.M.), T32DK128770 (to E.F. and S.F.S.), P20GM144269 (to E.M.M and J.P.T), and R01AG069781 (to J.P.T. and P.A.C.).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

X.C.D.: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Writing – original draft, Writing – review & editing; C.S.M.: Data curation, Formal analysis, Methodology, Writing – review & editing; S.F.S.: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Writing – review & editing; E.F.: Data curation, Formal analysis, Investigation, Methodology, Writing – review & editing; J.A.A.: Data curation, Investigation, Methodology, Supervision, Writing – review & editing; E.D.Q.: Conceptualization, Formal analysis, Methodology, Writing – review & editing; K.L.F.: Data curation, Formal analysis, Methodology; P.P.: Data curation, Formal analysis, Investigation, Methodology, Writing – review & editing; P.A.C.: Conceptualization, Formal analysis, Funding acquisition, Methodology, Project administration, Writing – review & editing; J.P.T.: Conceptualization, Funding acquisition, Investigation, Project administration, Supervision, Writing – review & editing; E.M.M.: Conceptualization, Formal analysis, Investigation, Methodology, Supervision, Writing - review & editing.