Time‐restricted feeding during the inactive phase abolishes the daily rhythm in mitochondrial respiration in rat skeletal muscle

Shift‐workers show an increased incidence of type 2 diabetes mellitus (T2DM). A possible mechanism is the disruption of the circadian timing of glucose homeostasis. Skeletal muscle mitochondrial function is modulated by the molecular clock. We used time‐restricted feeding (TRF) during the inactive phase to investigate how mistimed feeding affects muscle mitochondrial metabolism. Rats on an ad libitum (AL) diet were compared to those that could eat only during the light (inactive) or dark (active) phase. Mitochondrial respiration, metabolic gene expressions, and metabolite concentrations were determined in the soleus muscle. Rats on AL feeding or dark‐fed TRF showed a clear daily rhythm in muscle mitochondrial respiration. This rhythm in mitochondrial oxidative phosphorylation capacity was abolished in light‐fed TRF animals and overall 24h respiration was lower. The expression of several genes involved in mitochondrial biogenesis and the fission/fusion machinery was altered in light‐fed animals. Metabolomics analysis indicated that light‐fed animals had lost rhythmic levels of α‐ketoglutarate and citric acid. Contrastingly, lipidomics showed that light‐fed animals abundantly gained rhythmicity in levels of triglycerides. Furthermore, while the RER shifted entirely with the food intake in the light‐fed animals, many measured metabolic parameters (e.g., activity and mitochondrial respiration) did not strictly align with the shifted timing of food intake, resulting in a mismatch between expected metabolic supply/demand (as dictated by the circadian timing system and light/dark‐cycle) and the actual metabolic supply/demand (as dictated by the timing of food intake). These data suggest that shift‐work impairs mitochondrial metabolism and causes metabolic inflexibility, which can predispose to T2DM.


| INTRODUCTION
The biological clock controls a wide variety of metabolic processes, on a systemic as well as a cellular level. This endogenous timing system enables organisms to anticipate predictable daily changes in the environment, such as food availability and prepares cellular metabolism for appropriate energy substrate selection, for example, carbohydrates during the active (awake) period and lipids during the inactive (sleep) period. Indeed, on a cellular level, many aspects of metabolism are under circadian control, including mRNA transcripts, protein levels, posttranslational modifications as well as metabolites. [1][2][3] For example, 43% of all protein-coding genes were rhythmically expressed in at least one mouse organ, with 3%-14% of all transcripts being rhythmic in the individual organs. 3 Additionally, in the mouse liver, ~50% of all metabolites, including several metabolites of the tricarboxylic acid cycle (TCA) cycle, were rhythmically fluctuating. 1 Disruption of this temporal balance between energy supply and energy demand is suspected to lead to various metabolic diseases, including type 2 diabetes mellitus (T2DM). For instance, epidemiological evidence indicates that night shift workers are at an increased risk of developing T2DM, most likely because of the frequent disturbance of their biological clock by having nocturnal light exposure, and being active and eating at the "wrong" time of day 4,5 (reviewed in Ref. [6]).
T2DM is characterized by augmented endogenous glucose production as well as lower glucose uptake as a result of insulin resistance. The full complexity underlying the pathophysiology of insulin resistance remains incompletely understood, but one possible contributing factor is a disturbance in mitochondrial substrate metabolism, 7-13 as T2DM patients have impaired skeletal muscle mitochondrial function. [14][15][16][17][18] Skeletal muscle is an important organ for glucose uptake as it comprises up to 50% of total body mass in mammals and is considered to be the primary target for insulin-stimulated glucose uptake. [19][20][21][22] Several "-omics" approaches demonstrated that metabolic gene expression in skeletal muscle shows daily rhythms, both in humans and rodents, with transcripts for mitochondrial activity clustered in the afternoon in humans and in either of two peaks in the dark and light phase in mice. 3,23,24,25,26,27 Skeletal muscle mitochondrial dynamics and respiration are also under the control of the biological clock (reviewed in Refs. [28][29][30]). However, it is currently unknown whether the mitochondrial respiration rhythm in muscle fibers can be enhanced and/or disturbed through methods other than interrupting gene expression (e.g., altered feeding behavior). Therefore, in the current study, we used high-resolution respirometry, combined with gene expression, metabolomics, and lipidomics to study skeletal muscle mitochondrial metabolism in rats with a normal ad libitum (AL) diet. Using timerestricted feeding (TRF) during the light phase as a model for shift work, we investigated how eating at the "wrong" time of the day affects the daily rhythm in mitochondrial respiration and metabolite profiles.

| Animals and housing
Ninety-six adult male Wistar WU rats (8 weeks old; 253 ± 14 g upon the start of the experiment, Charles River) were used for the current experiments. All rats were housed under 12:12=light:dark conditions in a controlled environment of ~21°C during the entire experiment. Zeitgeber Time 0 (ZT0) was defined as the time of lights on and ZT12 as the time of lights off. All animals had AL access to tap water during the entire experiment. At the start of the experiment, animals were randomly assigned to one of three citric acid. Contrastingly, lipidomics showed that light-fed animals abundantly gained rhythmicity in levels of triglycerides. Furthermore, while the RER shifted entirely with the food intake in the light-fed animals, many measured metabolic parameters (e.g., activity and mitochondrial respiration) did not strictly align with the shifted timing of food intake, resulting in a mismatch between expected metabolic supply/demand (as dictated by the circadian timing system and light/ dark-cycle) and the actual metabolic supply/demand (as dictated by the timing of food intake). These data suggest that shift-work impairs mitochondrial metabolism and causes metabolic inflexibility, which can predispose to T2DM.

K E Y W O R D S
lipidomics, metabolomics, mitochondrial respiration, soleus muscle, time-restricted feeding feeding groups: AL fed, dark-fed TRF, or light-fed TRF. Ad libitum fed animals had AL access to food 24 h/day, TRF animals had also AL access to food, but only for 10 h during either the light or dark phase. Dark-fed animals had access to food from ZT13-23, while light-fed animals had access to food from ZT1-11 ( Figure 1A). Bodyweight of the animals was measured weekly. Food intake was measured after the animals were habituated to the TRF protocol either in the third or fourth experimental week (7 days average). All experiments were approved by the Dutch government and performed in accordance with the guidelines on animal experimentation of the Netherlands Institute for Neuroscience (License number: AVD801002016693/IVD 576).

| Metabolic cages
Metabolic PhenoCages (TSE systems) were used to assess metabolic parameters such as locomotor activity, food intake, and the respiratory exchange ratio (RER) in a separate group of experimental animals. Eight rats were placed in these metabolic cages and remained inside these cages for 4 weeks (4 days AL feeding and 3.5 weeks of TRF in either the light (n = 4) or dark (n = 4) phase). Cages were cleaned twice a week. Data shown are from the last 2 days of AL feeding and the last 3 days of the TRF feeding phase.

| Body temperature loggers
To assess body temperature, loggers (DST nano-T, STAR ODDI) were placed subcutaneously in another subset of animals (n = 5), while the animals were anesthetized. Temperature loggers were placed at the lumbar back region to prevent interference with brown adipose tissue activity. After recovery from the surgery, the temperature loggers were measured continuously until the end of the experiment with one sampling point every 15 minutes. Each animal was first measured for 3 days during AL feeding and subsequently during TRF to either the light (n = 3) or dark (n = 2) phase for 3.5 weeks.

| Muscle tissue preparation
After 4 weeks of TRF (between 3.5 and 4.5 weeks) animals were sacrificed at ZT3, 7, 15, or 19. As only one animal could be sacrificed per measurement session, the timing of sacrifice was highly standardized (1-minute deviation from the actual ZT). Furthermore, all further steps involving tissue preparation up until the start of the oxygen consumption measurement were highly standardized in their timing to minimize variation between the measurements.

| Mitochondrial respiration
After sacrifice, the left soleus muscle was dissected, snapfrozen in liquid nitrogen, and stored at -80°C for later analysis. The middle part of the right soleus muscle was excised, and stored in respiration solution B (2.8 mM CaK 2 EGTA, 7.2 mM K 2 EGTA, 5.8 mM ATP, 6.6 mM magnesium chloride, 20 mM taurine, 15 mM phosphocreatine, 20 mM imidazole, 0.5 mM DTT, and 50 mM MES (pH 7.1)) on ice, and further separated into fiber bundles (3-10 fibers) under a light microscope. Next, saponin (50 μg/ml) was added for 30 minutes to permeabilize the outer cellular membrane. Subsequently, the muscle fibers were washed twice with Mir05 (0.5 mM EGTA, 3 mM magnesium chloride, 60 mM K-lactobionate, 20 mM taurine, 10 mM potassium dihydrogen phosphate, 20 mM HEPES, 110 mM sucrose, and 1 g/L fatty acid-free BSA (pH 7.1)), quickly blotted dry and weighed before placing them in a respirometer (O2k, Oroboros Instruments, Innsbruck, Austria). The following substrate-uncoupler-inhibition-titration (SUIT) protocol was used for all measurements, similar to the previous studies. 31,32 To start, background respiration was measured for 10-15 minutes. Next, leak respiration was measured by the addition of sodium glutamate (10 mM), sodium malate (0.5 mM), and sodium pyruvate (5 mM). Maximal complex I-stimulated respiration with NADH producing substrates was measured after the addition of 2.5 mM ADP. The integrity of the mitochondrial membrane was tested by a lack of stimulation by 10 μM cytochrome c. Maximal oxidative phosphorylation (OXPHOS) was measured by subsequent convergent input of electrons via complex I and II after adding 10mM succinate. Carbonylcyanide-4-(trifluoromethoxy)-phenylhydrazone (FCCP) was stepwise (0.01 mM steps) titrated until maximal uncoupled respiration was reached. Subsequently, 0.5 μM rotenone was added to block mitochondrial complex I. Lastly, 2.5 μM antimycin A was used to block all electron transport-related respiration and this oxygen consumption level was used to subtract from all values. Chamber oxygen levels were maintained above 300 nM to prevent oxygen being limited. Respiration was measured in duplo for each animal at the same time.

| RNA isolation
Soleus muscle tissue was kept on dry ice to preserve mRNA integrity during mechanical homogenization. RNA isolation was performed using the ISOLATE II RNA Mini Kit (Bioline). RNA was eluted from the spin column using 40 μl of H 2 O after which RNA concentration and quality of the RNA were determined using a DS-11 (DeNovix) spectrophotometer and using the Agilent 2100 Bioanalyzer (Agilent Technologies), respectively. Although RNA integrity number (RIN) values above 5 were considered acceptable, all samples had a RIN above 8.

| cDNA synthesis
Three hundred and fifty ng of RNA was used as input for the cDNA synthesis. The Transcriptor First Strand cDNA synthesis kit (Roche) was used with oligo-dT primers and several additional samples without reverse transcriptase (-RT) were used as negative controls to check for genomic DNA contamination during the RT-qPCRs. RT-PCRs were run using an UNO-Thermoblock (Biometra) (30 min at 55°C, 5 min at 85°C).

| qPCR
One to 19 (1:19) diluted cDNA was used for all qPCRs to detect daily gene expression profiles. Expression levels of all genes were standardized by dividing over the geometric mean of three reference genes: RSP S18, βactin, and Cyclophilin. This geometric mean showed no significant effects of time of day (ZT) or feeding condition (TRF) nor for the ZT × TRF interaction. RT-qPCR was performed using a LightCycler 480 (Roche). Expression levels were calculated using dedicated software for linear regression of qPCR data (LinRegPCR). All primers used are listed in Table S1. Melting curves of the RT-qPCR and fragment length of the DNA amplicons were inspected as a means of quality control.

| mtDNA/nDNA isolation and quantification
DNA isolation was performed using the DNA mini isolation kit (Qiagen) according to the manufacturer's guidelines. DNA was eluted from the spin column using 200 μl of H 2 O after which DNA concentration was determined using a DS-11 (DeNovix). For the RT-qPCR of the mtDNA/nDNA ratio and the subsequent analysis and quality control, similar procedures were used for the gene expression described above (Section 2.5.3). However, for the mtDNA/nDNA ratio qPCRs, an input of 6ng per reaction well was used. Primers used for the nuclear and mitochondrial DNA fragments are listed in Table S1.

| Citrate synthase measurements
Tissue samples (1-2 mg wet weight) were disrupted in 100 mM Tris/HCl-KCl buffer, pH 7.6 and 1 g/L Triton X-100 using a TissueLyser II (Qiagen) for 1 min at a frequency of 30 times/sec, followed by sonication (twice at 8-watt output, 40 J, on ice). The homogenates were centrifuged for 5 minutes at 500 g. Protein concentration was determined in the supernatant using the BCA method. The homogenates were diluted to 0.3 mg/ml. Citrate synthase (CS) was measured spectrophotometrically at 412nm by the formation of free CoA using 5,5'-Dithio-Bis (2-Nitrobenzoic Acid) (DTNB). 33 Table 1 summarizes the main statistical findings for the metabolic parameters displayed in E-H. Note that for E-H the figures to the left show data from one full week, that is, 4 days of ad libitum feeding and the first 3 days of the TRF phase, the ad libitum-and TRF-phase being separated by a dashed line. This means the TRF data in the left figures are different from those in the right figures, as the former displays the first 3 days of the TRF period, whereas the latter are summaries of the 4th week of the TRF period. *p < .05, **p < .01, ****p < .0001 using Tukey's post hoc tests period of 5 min. The reduction of DTNB was followed in time on a CobasFara centrifugal analyzer (Roche) at 412 nm, with the use of a molar extinction coefficient of 13 600 L/mol/cm.

| Polar metabolomics and lipidomics
Metabolomics and lipidomics were performed as previously described, with minor adjustments. 34 Briefly: for each sample, approximately 5 mg of freeze-dried soleus muscle was added to a 2 ml Eppendorf Safe-Lock tube. Internal standards for both metabolomics and lipidomics were added, as well as solvents for a total of 500 µl water, 500 µl methanol, and 1 ml chloroform. Samples were homogenized using a stainless steel bead before centrifugation, creating a twophase system. Polar metabolomics was performed on the top layer using a ZIC-cHILIC LC-MS platform, while lipidomics was performed on the bottom layer, using both a normal and reversed-phase LC-MS platform.

| Statistics
Rhythmicity of physiological parameters (food intake, RER, locomotion, and body temperature) was determined using SigmaPlot 14.0 (Systat Software) as this software package can take into account within-individual measures. Rhythmicity of gene expression, metabolomics/lipidomics, and mitochondrial respiration measures was determined using the JTK_Cycle 3.0 script for nonparametric cosine regression, as this software package is best suited for independent sampling with lower temporal resolution (less than 24 time points per 24 h cycle) and can correct for the multiple comparisons problems, which is necessary when, for example, testing a multitude of gene expression profiles from the same mRNA/ cDNA pool during qPCR. The JTK_Cycle script was run in RStudio (R version 3.3.3). All graphs were plotted using Graphpad Prism 7. All other statistics, including the one-way and two-way ANOVA's, were executed by Graphpad Prism 7.

| Physiological parameters display altered rhythms upon TRF
To confirm that during the TRF phase animals showed similar metabolic phenotypes as in our previous studies, we monitored food intake and body weight and placed a subset of animals in metabolic cages to assess rhythms in metabolic parameters such as RER. [35][36][37][38] In the final week of the TRF protocol, the daily food consumption ranged between 19 and 24 g per day and was lowest for the light-fed animals, but did not significantly differ between AL fed and dark-fed animals ( Figure 1B). Animals continuously gained body weight throughout the experiment ( Figure 1C,D), but light-fed animals gained the least amount, likely due to the necessary adaptation to the TRF in the first week, as food intake was ~25% less in this first week compared to the AL group (data not shown). In the remainder of the experiment light-fed animals gained similar amounts of body weight as the other groups ( Figure 1D), despite eating significantly less ( Figure 1B). Thus, contrary to most mice studies, rats eating at the wrong time of day did not show an increased body weight, this could be due to the fact that in the current experiment the feeding period was only 10 h, instead of 12 h, or a stronger light-induced inhibition of activity and feeding behavior during the light period in rats as compared to mice. In contrast, feeding efficiency was increased in the light-fed animals and in previous experiments, we found increased adiposity in the light-fed animals. [37][38][39] A subset of animals was housed in metabolic cages, first under AL (baseline) feeding conditions and subsequently under TRF conditions (4 weeks) to assess physical activity and substrate metabolism. As expected, light-fed and dark-fed animals did not eat during the dark and light phases, respectively ( Figure 1E). Similar to our previous experiments, under AL conditions animals consumed most of their food in the dark phase (85%), with 61% of total food intake in the dark phase being consumed in the first half of the dark phase. Such a bimodal feeding pattern was also found during the TRF phase, with the obvious difference that the light-fed animals had their two peaks in food consumption during the light phase ( Figure 1E). Light-fed animals ate slightly more during the first half of their feeding period (63% of total food intake) as compared to the dark-fed animals (56% of total food intake).
The amplitude of the daily rhythm in RER increased in animals on TRF, both for the light and dark-fed groups, especially due to a greater reliance on lipid substrate utilization during the non-feeding period ( Figure 1F, Table 1). Dark-fed animals were relatively more active during the dark phase and light-fed animals were more active during the light phase ( Figure 1G, Table 1), as compared to AL feeding. However, the shift in locomotor activity in the light-fed animals was clearly less pronounced than the shift in food intake, as the light/dark (L/D) ratio for locomotor activity went from 20/80 during AL feeding to 50/50 during light phase TRF. Importantly, the TRF-induced alterations in RER and locomotor activity were highly similar to those reported previously. 35,37,38 The switch from AL feeding to light phase feeding also drastically changed the daily pattern in body temperature ( Figure 1H right panel, Table 1). Such a change was not observed when animals switched from AL feeding to dark phase feeding.

| Mitochondrial respiration rhythm is abolished upon light phase TRF feeding
After confirming the metabolic effects of TRF we assessed if the soleus muscle of rats displays rhythms in mitochondrial respiration and whether such rhythms could be altered through TRF. Mitochondrial respiration of the soleus muscle was determined at four timepoints along the 24-h cycle. The light-fed animals had an overall lower OXPHOS capacity, maximal uncoupled respiration, and succinate/ rotenone-linked respiration compared to AL animals, with the largest differences (~30% lower) at ZT7 (Figure 2A-C). Mitochondrial respiration showed significant differences over the 24 h cycle, but no interactions of ZT × TRF were found (Figure 2A-C, Table 2). The strongest rhythmicity was found for maximal OXPHOS capacity with combined NADH-substrates and succinate in the dark-fed animals. The AL fed group showed a trend toward a time-of-day effect, but no significant time-of-day effect was found for the light-fed group. The difference between the highest and lowest maximal OXPHOS was 17% in AL and dark-fed animals, but 11% in light-fed. Similar results were obtained for the maximal uncoupled respiration. For uncoupled respiration rates, the difference between the highest and lowest rates was about 15% for the AL group, 19% for dark-fed, and only 8% for the light-fed animals. None of the groups showed any significant daily rhythmicity after inhibition of mitochondrial complex I by rotenone, although the dark-fed animals still showed a trend towards rhythmicity, suggestive of a contributing role for mitochondrial complex I in the regulation of daily rhythmicity. Together, these data indicate that enhancing the daily feeding rhythm enhances mitochondrial respiration rhythms, while disturbing the daily feeding rhythm abolishes the mitochondrial respiration rhythm and impairs the 24-h mean respiration. Additional to the potential role of complex 1 described above, likely the mitochondrial number per muscle fiber plays a role as uncoupling respiration from OXPHOS did not lead to similar respiration between the three TRF groups.
To further assess the mechanisms that drive these alterations in respiration upon manipulation of the daily feeding rhythm, we measured the number of mitochondria per muscle fiber. The relative number of mitochondria per fiber was quantified using the ratio of DNA copy numbers of a mitochondrial DNA (mtDNA) fragment divided by the copy numbers of a nuclear DNA (nDNA) fragment ( Figure 2D). No significant effects were found for the time of day (ZT), feeding group (TRF), or for the interaction ZT × TRF. However, when we performed rhythmicity analysis with JTK_Cycle, we found that only in the light-fed group the mtDNA/nDNA ratio was rhythmic (p = .0198), which likely can be attributed to the diurnal fluctuations in mtDNA abundance and not the nDNA abundance (p = .0248 and p > .9999, respectively, JTK_ Cycle, data not shown). For the AL and dark-fed groups no rhythms in mtDNA/nDNA ratio were found (p > .9999 for both groups). Strikingly, similar rhythmicity patterns T A B L E 1 Analysis of rhythmicity of the physiological parameters from animals that went from ad libitum feeding (AL phase) to either dark or light phase feeding (TRF phase)

Rhythmicity
Acrophase Amplitude a were found when we measured citrate synthase enzyme activity, which is another standard assay to reflect the mitochondrial abundance and activity ( Figure 2E).

| Gene expression of some mitochondria-related genes is altered during TRF
To pinpoint the mechanisms underlying the alterations in mitochondrial respiration described above, we also measured the expression profiles of several genes that are associated with (mitochondrial) metabolism ( Figure 3D-I). TRF during the light phase dampened the daily rhythmicity of the muscle clock (the core clock gene Bmal1) and altered the expression of several metabolic genes, including uncoupling protein 3 (Ucp3) and fatty acid synthase (Fasn). Ucp3 expression profile was inverted with an overall lower expression in the light-fed group compared to the dark and AL-fed animals. TRF induced rhythmicity in Fasn expression in both the light and dark-fed groups, albeit in exact antiphase (acrophases at ZT8 and ZT20, respectively) ( Figure 3A-C), which is likely explained by the food intake per se. Also, several genes important for mitochondrial biogenesis and dynamics showed different expression profiles in the AL and the dark-fed group as compared to the light-fed group: Ppargc1α showed a loss of rhythm, while Mfn1 showed a gain of rhythm. Additional genes important for mitochondrial functioning that showed altered expression in the light-fed group, included Nocturnin (Noct, loss of rhythm), Sirtuin 3 (Sirt3, gain of rhythm), mechanistic target of rapamycin kinase (mTOR; gain of rhythm), and Sarcolipin (Sln, loss of rhythm).

| TRF alters rhythms in major metabolic pathway intermediates
To further characterize metabolic changes in soleus muscle resulting from TRF we performed polar metabolomics  Table 2 summarizes the main statistical findings for the mitochondrial respiration states and mtDNA/nDNA ratio. Colorcoded "~" indicates that the (metabolic) parameter is significantly rhythmic for that TRF group as determined by JTK_ Cycle (p < .05)

(E)~
and lipidomics. Out of 97 polar metabolites that we identified in the AL and dark-fed groups, 18% and 16% were rhythmic in the soleus (17 and 16 metabolites, respectively; Figure 4, Figure S1, Table S4), versus only 8% in light-fed animals (8 metabolites). Only malate showed significant daily rhythmicity in all three groups. Coenzyme A, citric acid, and succinyl AMP (which is cleaved into both AMP and fumarate) showed significant rhythmicity in AL and dark-fed animals, whereas cis-aconitate, fumarate, and oxaloacetate trended to have a 24 h rhythm in AL and dark-fed groups (p < .1).
Intermediates from the pentose phosphate pathway (erythrose-4-phosphate, gluconate-6P, and ribose) were significantly rhythmic in dark-fed animals, while no intermediates were rhythmic in muscle tissue of AL or lightfed animals. In AL animals one glycolysis intermediate was found to be rhythmic (glucose), while in dark-fed animals four rhythmic glycolysis metabolites were found (fructose-6P, glucose-6P, uridine, and its precursor cytidine). Only in light-fed animals the breakdown products of both anaerobic glycolysis (lactate) and creatine phosphate metabolism (creatinine) were found to be rhythmic. Surprisingly, many of the metabolites that were significantly rhythmic in at least one of the TRF groups were amino acids and nucleic acids (Table S4).
For most metabolites overall daily (i.e., 24-h average) values did not differ between the three TRF groups, except for aspartate, βalanine, and cis-aconitate (Figure 4C-E; Figure S1, p < .05 and VIP score >1.8 for all three metabolites; ANOVA). This indicates that the alterations in metabolite rhythms mostly result from shifts in the peak phase and not from strongly increased or decreased levels of the metabolites at a certain timepoint (which can also induce/abolish rhythmicity).
We used MetaboAnalyst to compare differences in rhythmic metabolite sets between the light-and dark-fed animals (i.e., the metabolites that were rhythmic in only one of these two groups). The top 6 highest metabolite enrichment set results were (in order of significance): Warburg effect, Purine metabolism, pentose phosphate pathway (PPP), pyrimidine metabolism, TCA-cycle, and gluconeogenesis (all p-values <.005; all false-discoveryrates (FDR) <0.1; Table S5).

| TRF leads to distinct lipid signatures
In addition to the polar metabolomics, nonpolar lipidomics was performed on the soleus muscle tissue during which >1000 lipid species could be identified and assessed for their rhythmicity. Light-fed animals had almost double T A B L E 2 Summary of the two-way ANOVA of oxidative phosphorylation (OXPHOS) capacity, maximal uncoupled respiration, succinate/rotenone-linked respiration, mtDNA/nDNA ratio, and CS activity the number of intracellular skeletal muscle lipid species that showed significant rhythmicity as compared to AL and dark-fed animals (20%, 11%, and 12% of 1088 annotated lipid species, respectively) ( Figure 5A and When assessing the individual lipid species within each specific class more closely, we found striking differences in the proportion of significantly rhythmic lipid species between the TRF groups ( Figure 5A, Table S6). For example, in dark-fed animals, 48 out of 69 ether lipid (TG(O)) species were significantly rhythmic, while three species were rhythmic in light-fed animals, and none in the AL group ( Figure 5A,C). In stark contrast, from the regular TG species, only one out of 259 species was significantly rhythmic in dark-fed animals, while in AL and light-fed animals 45 and 115 of the regular TG species were rhythmic, respectively.
We found no clear differences in the timing of the lipid rhythmicity that could be attributed to altered TRF feeding behavior ( Figure 5 and Figure S2), except for triglycerides for which the average acrophase was ZT13.3 in AL animals, versus ZT10.5 for light-fed animals ( Figure 5C), and for PI (AL average acrophase at ZT18.8; dark-fed average acrophase at ZT7.3; Figure S2).
Overall, daily (i.e., 24h average) values of lipids also differed between the TRF groups, with for example 14 different TG ether lipid species being higher in LF animals (20% of all TG(O) species measured) and three BMP species being lower in LF animals (10%) as compared to dark-fed animals (Table S7). Compared to the AL group, DF animals had higher overall levels of, for example, Cer F I G U R E 3 (A-L) Daily profiles of (mitochondrial) mRNA expression in the soleus muscle of the 3 experimental groups (a selection of 9 out of 18 tested genes). Table S1 lists the full name of all genes. Table S2 and Table S3 summarize the main statistical findings from respectively the 2 way ANOVA and JTK_Cycle analysis of all tested genes. Color-coded "~" indicates that the gene expression is significantly rhythmic for that TRF group as determined by JTK_Cycle (p < .05) To zoom in on the muscle lipid species that are under the control of the circadian timing system we eliminated the influence of feeding status by comparing only the time points at which the light-fed and dark-fed animals just had obtained access to their food ( Figure S3; i.e., Just-Ate, which is ZT3 for the light-fed and ZT15 for the dark-fed animals) or when they both had been fasting for 14 h (i.e., Hungry, which is ZT19 for the light-fed animals and ZT7 for the Dark-fed animals). With this analysis, we found some remarkable differences between LF and DF animals, which are thus independent of feeding status. For instance, in the Just-Ate condition, LF animals had higher levels of two ether lipid LPE species (22% of total LPE(O) species), but lower levels of ten BMP species (34%) and 96 TG species (37%) (Table S8). Conversely, in the Hungry condition LF animals had lower levels of one ether lipid LPE species (11%), but higher levels of nine BMP species (31%).

| DISCUSSION
The current data provide clear evidence for significant daily rhythmicity in rat muscle maximal ex vivo mitochondrial respiration. TRF to the dark/active phase (i.e., aligned with the circadian timing system) strengthened this daily rhythm in mitochondrial metabolism, whereas feeding restricted to the inactive phase strongly dampened the amplitude of this rhythm and reduced the overall 24-h respiration rate. Analysis of mitochondrial abundance, activity, and gene expression indicated a possible role for the mitochondrial biogenesis and fission/fusion machinery. Metabolomics demonstrated a multitude of rhythmic metabolites of especially intermediates of the pentose phosphate pathway and the TCA cycle in AL and dark-fed Depicted are malate (B; the only metabolite rhythmic in all TRF groups) and the three metabolites that showed a significant difference between groups in their 24 h-mean levels (C-E), that is, lower levels in the ad libitum group. Color-coded "~" indicates that a metabolite has significantly rhythmic abundance levels for that TRF group as determined by JTK_Cycle (p < .05) animals, but no daily rhythms in light-fed animals were found for these pathways. Contrastingly, lipidomics showed that light-fed animals abundantly gained rhythmicity in levels of triglycerides. Bis(monoacylglycero) phosphate and both diglycerides and ether triglycerides were the most affected lipid species.

| Rhythmicity of skeletal muscle mitochondrial respiration
Our results confirm that clear daily rhythms in mitochondrial respiration exist in permeabilized muscle fibers, which is in agreement with other small animal studies [40][41][42][43][44] and in vitro approaches. [45][46][47][48][49] A recent study in healthy humans observed a similar difference between the highest and lowest daily respiration rate (20% vs. 18% in this study). 27 In that study, no rhythmicity in maximal uncoupled respiration was found, but this is likely due to the large inter-person variation. 27 In both studies, the daily peak in OXPHOS capacity was not in the active phase (daytime in humans, nighttime in rats), but rather in the inactive phase (11:00 p.m. in humans, ZT6 in our study). The underlying cause of the higher maximal OXPHOS capacity during the inactive phase is currently unknown. Furthermore, we observed that the overall maximal OXPHOS capacity was 11% lower in the light-fed group (i.e., our circadian misalignment group), which is quite similar to the 14% difference observed when comparing humans in an acute circadian aligned and misaligned shift-work protocol. 50 Therefore, disturbing the circadian system, such as occurs during shift-work, could be causal or at least contribute significantly to the pathology of T2DM. An important observation was that the total mitochondrial abundance and activity (assessed by mtDNA/nDNA and citrate synthase activity) was not different across the day in AL and dark-fed animals, indicating that the cause of this daily rhythmicity likely resides in mitochondrial proteostasis or other factors of mitochondrial quality such as mitochondrial dynamics. Surprisingly, in the light-fed group mitochondrial abundance appeared to be rhythmic. We think that feeding during the inactive phase disturbs the balance in mitochondrial biogenesis, mitophagy, and fission/fusion, resulting in abnormal mitochondrial abundance patterns throughout the day. Several studies have indeed provided evidence for daily rhythms in mitochondrial dynamics. 27,46,50,51,52 To further provide mechanistic insight into the reason for the lower respiration and dampened rhythmicity in the light-fed animals, we analyzed gene expression profiles of several important mitochondrial genes.
We found several mitochondria-related genes that lost their rhythmic expression in the light-fed group, such as Ppargc1a, Sln, and Noct, while others gained rhythmic expression (Mfn1, mTOR, and Sirt3). The dysregulation that we found in the expression of these genes thus fits well with the idea of circadian disruption of mitochondrial functioning through daytime TRF.
However, most mitochondria-related genes tested showed no rhythmic expression and thus the effect of TRF on rhythmicity could not be assessed. This is not surprising as only an estimated 4% of all transcripts are rhythmic in mice skeletal muscle. 3

| Rhythmicity of skeletal muscle metabolite and lipid concentrations
Using the same soleus muscle tissue for both polar metabolomics as well as lipidomics, we were able to get a better understanding of the individual metabolites affected by TRF. The polar metabolomics showed profound changes in some of the major muscle metabolic pathways as a result of TRF, such as the TCA cycle, glycolysis, and the pentose phosphate pathway, both in light-fed and darkfed animals. More specifically, although the total number of rhythmic metabolites remained unaltered, the muscle tissue of dark-fed animals showed an enrichment of PPP metabolites as compared to AL feeding. Conversely, TRF during the inactive phase showed a quite drastic change in the pool of rhythmic metabolites, with a roughly halving and doubling of rhythmic polar metabolites and lipids, respectively. Additionally, various alterations in the types of metabolites that were rhythmic were observed in these light-fed animals, such as a strong reduction in the number of rhythmic TCA cycle and glycolysis intermediates. Furthermore, a gain of rhythm in the muscle waste products creatinine and lactate was observed. One explanation for the latter findings is that maximal OXPHOS capacity is reduced in light-fed animals, especially during the light phase, while during this period the activity of the animals is higher. Therefore, during the light phase, the ATP demand is increased while OXPHOS is decreased, likely resulting in an increase in alternative metabolic pathways such as (anaerobic) glycolysis and creatine metabolism and thus altered levels of creatinine and lactate. Adding to this notion is the increased RER of light-fed animals during the light phase, as also observed previously, 35,37,38 which supports the notion of increased lactate production through glycolysis. Taken together these findings suggest that eating at the wrong time disrupts metabolic flexibility in skeletal muscle which can contribute to pathological conditions such as T2DM (reviewed in Ref. [53]). When looking at peak abundance of the different lipid species it becomes clear that the peak time of most rhythmic lipid species clusters in the dark phase for all three TRF groups ( Figure 5B-E and Figure S2). When averaging the acrophases of all significantly rhythmic lipids only small shifts are found for the dark-fed compared to the light-fed animals. This indicates that these rhythms are not driven by the timing of food intake, but are mostly determined by the circadian timing system. Especially in the light-fed group, much larger shifts would be expected when the timing of food intake would determine the phasing of these rhythms. The fact that these rhythms are not adjusted upon TRF implies a consequent imbalance between energy supply and demand for light-fed animals. Regardless of its timing, TRF clearly induces an unique muscle lipid signature for both regular and ether TGs. A similar conclusion was recently drawn by Mehus et al. for the effects of TRF on the liver of mice fed a high-fat diet. 54 Since increased TG levels are known to interfere with the functioning of the main insulin-dependent glucose transporter GLUT4, the disturbances in TG metabolism in the light-fed group provide a possible mechanism for the disturbed glucose metabolism found in our shift-work model. 55 For ether lipid DGs something similar was observed regarding rhythmicity, with dark-fed animals having 52% of the DG(O) lipid species showing significant fluctuations throughout the day, but in light-fed and AL animals these figures were only 5% and 0%, respectively. The specific functions of most of these lipid species remain to be investigated, as well as the mechanisms responsible for these large changes in abundance.
BMPs are negatively charged glycerophospholipids that are considered important for nutrient cycling through lysosomal activity. 56 The role of BMPs in skeletal muscle is not well studied, but BMPs are known to be affected by feeding status with lowered levels in skeletal muscle within 2 h of refeeding fasted mice, 56 which fits with the changes in 24h mean levels induced by TRF ( Figure 5, Tables S6-S8). In addition, our data indicate that BMPs in skeletal muscle are likely also under the control of the biological clock. Strikingly, whereas only five out of the 1088 annotated lipids showed a significant rhythm in all three experimental groups, three of those were BMPs. Intriguingly, the percentage of annotated BMPs that showed a significant rhythm more than doubled in the light-fed group (45% of all 29 identified BMPs) as compared to both AL and dark-fed animals, while light-fed animals overall had lower BMP levels as compared to dark-fed animals. Future studies should elucidate the exact roles of lipids such as DG(O)s, TG(O)s, and BMPs in the regulation of daily metabolism in skeletal muscle. Finally, our finding that ~11% of all measured lipids are rhythmic matches closely the finding that 13% of measured lipids in the human vastus lateralis display rhythmicity and thus adds to the validity of our animal model. 57 Concluding, we report the presence of a daily rhythm in mitochondrial respiration in skeletal muscle in rats. The amplitude of this daily rhythm in respiration was enhanced by restricting food intake to the active phase. Contrasting, eating at the wrong time of day, as is often the case during shift work, abolished the rhythm in mitochondrial respiration. Moreover, 24-h respiration levels were significantly decreased in the misaligned animals, an observation that parallels the reduced oxidative capacity observed previously in T2DM patients. Metabolomics and lipidomics indicated that these misaligned animals had lost a rhythmic abundance of αketoglutarate and citric acid levels, while triglyceride levels gained rhythmicity. Furthermore, due to the significant influence of the circadian timing system many measured metabolic parameters (e.g., activity and mitochondrial respiration) did not strictly align with the shifted timing of food intake, resulting in a mismatch between expected metabolic supply/ demand (as dictated by the circadian timing system and L/D cycle) and the actual metabolic supply/demand (as dictated by the timing of food intake and locomotor activity). Adding to this are the observations that the RER did shift entirely while peak timing of most lipid classes assessed did not shift at all with the altered feeding behavior.
In this regard, it is unfortunate that with the current experimental set-up it is not possible to completely separate the effects of the L/D cycle and timing of food intake. Ideally, these experiments should also be performed in dark/dark conditions, with food restricted to either the subjective day or subjective night. However, the freerunning rhythms that will be initiated in these constant conditions make this a challenging experimental design, i.e., the timing of the access to food will have to be adapted every day for every individual animal. Interestingly, recently, Heyde & Oster 58 used another experimental design to approach this question. They exposed their animals to a combined 28-h L/D cycle and a 24-h feeding/fasting cycle, or the other way around, to separate the effects of the L/D cycle and feeding time. Liver and adipose tissue exhibited the strongest correlation with the feeding cycle, whereas the SCN and adrenal gland showed a stronger influence by light.
Taken together, these findings indicate that shiftwork impairs muscle mitochondrial functioning on multiple levels, which eventually could lead to metabolic disorders such as T2DM. Thus, our results suggest that mitochondria are an interesting therapeutic target for the prevention or even treatment of T2DM as they can be targeted not only by pharmaceutical drugs, but can also be influenced by lifestyle interventions. Examples of such lifestyle interventions are dietary interventions (both caloric restriction as well as nutritional supplements such as resveratrol, carnitine, and NAD + precursors) and (timed) exercise and other measures to strengthen circadian rhythms. Therefore, these interventions urgently need further confirmation in human studies in the context of treatment and/or prevention of T2DM, 11 especially whether an optimized combination and timing of exercise and diet can further enhance their separate effects. ACKNOWLEDGEMENT P. de Goede was funded by a ZonMW TOP grant (#91214047).

DISCLOSURES
The authors have stated explicitly that there are no conflicts of interest in connection with this article.