Comparative Circadian Metabolomics Reveal Differential Effects of Nutritional Challenge in the Serum and Liver^*

Mice and Diets

Age-matched, male C57BL/6J mice (JAX, 00064) were maintained on a 12-h light/12-h dark cycle. Mice had access to water and pellet food ad libitum. Animal care and use were in accordance with guidelines of the institutional animal care and use committee at the University of California (Irvine, CA). At 6 weeks of age, animals were placed on a normal chow diet (Prolab RMH 2500) or an HFD (60% of calories from fat; Research Diets, D12492) for 10 weeks. Body weight was measured weekly. Animals were separated into individual cages 1 week before sacrifice. Liver and serum were harvested across the circadian cycle every 4 h. Serum was prepared from an abdominal/thoracic blood sample. All samples were stored at −80 °C until analysis.

Mass Spectrometry for Metabolomics Analysis

Metabolomics analysis was carried out by Metabolon, Inc. (Durham, NC) using a published methodology (49). Briefly, small molecules from the liver and serum (n = 5/group and zeitgeber (ZT)) of age-matched, adult, male mice were harvested at ZT0, ZT4, ZT8, ZT12, ZT16, ZT20, and ZT24, with ZT0 corresponding to lights on and ZT12 corresponding to lights off in the animal facility. Animals were housed on a 12-h light/12-h dark schedule. Livers and serums were harvested, and livers were rinsed briefly in PBS and immediately frozen in liquid nitrogen. Frozen samples were shipped to Metabolon. The metabolic profiling platform used for this kind of analysis by Metabolon combines three independent platforms: ultra-HPLC/MS/MS optimized for basic species, ultra-HPLC/MS/MS optimized for acidic species, and gas chromatography/mass spectrometry (GC/MS). Briefly, a comparable amount of liver and serum tissue from each replicate animal for normal chow (NC) and HFD was used. Liver samples were processed essentially as described previously (26, 50). For serum samples, the LC/MS portion of the platform was based on a Waters ACQUITY UPLC system and a Thermo-Finnigan LTQ mass spectrometer, which consisted of an electrospray ionization source and linear ion trap mass analyzer. The sample extract was split into two aliquots, dried, and then reconstituted in acidic or basic LC-compatible solvents, each of which contained 11 or more injection standards at fixed concentrations. One aliquot was analyzed using acidic positive ion-optimized conditions and the other using basic negative ion-optimized conditions in two independent injections using separate dedicated columns. Extracts reconstituted in acidic conditions were gradient-eluted using water and methanol, both containing 0.1% formic acid, whereas the basic extracts, which also used water/methanol, contained 6.5 mm ammonium bicarbonate. The MS analysis alternated between MS and data-dependent MS2 scans using dynamic exclusion. The samples destined for GC/MS analysis were redried under vacuum desiccation for a minimum of 24 h before being derivatized under dried nitrogen using bistrimethyl-silyl-trifluoroacetamide. The GC column was 5% phenyl, and the temperature ramp was from 40 to 300 °C in a 16-min period. Samples were analyzed on a Thermo-Finnigan Trace DSQ fast scanning single-quadrupole mass spectrometer using electron impact ionization. The instrument was tuned and calibrated for mass resolution and mass accuracy on a daily basis. Metabolites were identified by automated comparison of the ion features in the experimental samples with a reference library of chemical standard entries that included retention time, molecular weight (m/z), preferred adducts, and in-source fragments as well as associated MS spectra and curated by visual inspection for quality control using software developed at Metabolon (51).

Adipokine Level Measurements

Analyses of IL-6, leptin, and adiponectin levels in serum and liver were carried out by Eve Technologies by multiplexing laser bead technology.

Western Blotting

Equal amounts of protein from whole liver cell lysates (30 μg) were separated on polyacrylamide gel and transferred to nitrocellulose membranes. After blocking, the membranes were incubated overnight, at 4 °C, with one of the following primary antibodies: anti-phospho-GSK3β (Ser⁹) (Cell Signaling Technology), anti-phospho-AKT (Ser⁴⁷³) (Cell Signaling Technology), anti-AKT (Cell Signaling Technology), and actin. All of the primary antibodies were used at a concentration of 1:1,000.

Statistics

Data from each experimental group were assessed through replicates (n = 5) for representative mean and image. Data represent mean ± S.E. of experiments. Experiments with two variables were analyzed by two-way analysis of variance followed by Tukey's post hoc test for multiple comparisons (GraphPad Prism version 5.0). For analysis of rhythmic metabolites, the nonparametric test JTK_CYCLE was used, incorporating a window of 20–28 h for the determination of circadian periodicity, as described previously (26, 50). Metabolites with p < 0.05 were considered significant (supplemental Files 1 and 2).

Comparative Circadian Metabolomics

We used LC-MS metabolomics to analyze the relative abundance of metabolites in the mouse serum and liver throughout the circadian cycle under NC and HFD conditions. Wild-type mice were divided in two experimental groups, the first fed an NC diet and the second fed an HFD (60% of calories from fat (26)) for 10 weeks. Liver and serum were harvested across the circadian cycle every 4 h. Although the hepatic circadian metabolome and transcriptome have been previously reported to undergo extensive reprogramming following nutritional challenge (26), the degree and specificity to which circulating metabolites oscillate have not been determined.

Our metabolome analysis identified 362 known metabolites in the serum (Fig. 1A), belonging to major metabolic pathways. Although a large fraction of metabolites are present in both tissues (222 metabolites), 140 metabolites, corresponding to 38.6% of the total, were detected only in serum and not in the liver (supplemental Files 1 and 2). This indicates that either 1) their abundance is too low in the liver to be detected; 2) these metabolites are not typically made in or transported to the liver; or 3) due to technological variability, they were unable to be detected.

In both tissues, a large number of metabolites (62% in serum, 77% in the liver) were affected by HFD. When analyzed by analysis of variance, 40% of serum metabolites and 45% of liver metabolites deviated in abundance throughout the circadian cycle. However, when analyzed for circadian oscillation specifically (p < 0.05, JTK_cycle (26)), the fraction of oscillating metabolites was the same. Specifically, 46% oscillated in the serum, and 46% cycled in the liver under a distinct feeding condition (i.e. NC, HFD, or both (BOTH)) (Fig. 1B). Strikingly, in serum, HFD induced an extensive disruption in the oscillation of metabolites that cycle in NC (55%). A smaller fraction of metabolites (27%) oscillated in both feeding conditions, and only 18% oscillated in HFD. This profile is in stark contrast to the situation in the liver, where 43% of metabolites oscillate in both feeding conditions, 30% oscillate only in NC (i.e. oscillation is lost under HFD), and 27% oscillate only in HFD (26) (Fig. 1C). Thus, although a similar number of metabolites oscillated in both feeding conditions in the liver, there was a 3-fold decrease in the number of metabolites oscillating in HFD in the serum relative to NC. Thereby, it appears that the serum metabolome is much more sensitive to nutritional challenge than the liver. Moreover, an analysis of the phase of oscillation in both tissues under NC and HFD revealed some important differences (see Figs. 4F, 5 (E and J), 6F, 7D, and 8 (E and J)). In serum, apart from the loss of oscillation in carbohydrates and cofactors in HFD, the remaining oscillatory metabolites were phase-delayed in HFD. Moreover, some variance was noted in the phase of oscillation for metabolites that oscillated in both diets. In particular, serum metabolites oscillating in both feeding conditions were phase-advanced in HFD (Fig. 1D), whereas, considering the phase of metabolites that oscillated only in NC or only in HFD, liver metabolites in HFD were somewhat phase-advanced compared with metabolites oscillating only in NC (26) (Fig. 1E). Also, we analyzed the relative abundance of major classes of metabolites at ZT0 and ZT12 for serum and liver (37). Significant decreases in serum were detected at ZT0 under HFD compared with NC diet for peptides and xenobiotics. Moreover, lower levels of peptides were also found at ZT12 compared with ZT0 under NC. Interestingly, some major classes of metabolites showed a diet effect in the liver. This is the case for amino acids, xenobiotics, and nucleotides, with a marked decrease in their content under HFD for both of the time points analyzed.

By analyzing a variety of parameters, we have previously shown that, as expected, HFD-fed mice develop an obese phenotype (26). Adipose tissue in mammals not only acts as storage for excess of nutrients; it also acts as an endocrine organ secreting adipokines that are involved in a wide range of functions (52–55). Specifically, obesity is associated with oxidative stress and inflammatory responses in adipose tissue (due to adipocytes hypertrophy and hyperplasia) with consequent increased levels of local and systemic pro-inflammatory cytokines. We analyzed the levels of IL-6, leptin, and adiponectin (56–59) in serum and liver at two different time points (ZT4 and ZT16) to monitor whether the obese status elicited by the high fat diet could affect the circadian secretion of these adipokines (Fig. 2, A–C). No changes or little change was observed in IL-6 and adiponectin levels between NC and HFD in both tissues. This could be due to various factors, such as the diet composition and/or the circadian changes occurring at other ZTs. Importantly, the levels of leptin at ZT16 were significantly higher in HFD as compared with NC in both serum and liver. This result not only indicates a temporal disruption of this adipokine; it also suggests a misreading of the signal for the brain of the body's energy stores (Fig. 2C). Moreover, because leptin is implicated in the etiology of insulin resistance (60–62), we extended our analysis by monitoring AKT and GSK3β. As expected, we found that HFD induced an increase in basal (non-insulin-stimulated) AKT phosphorylation at Ser⁴⁷³ and GSK3β inactivation (as measured by phosphorylation at Ser⁹) (Fig. 2D). Importantly, we also observed a complete loss of rhythmicity in AKT phosphorylation in the animals fed an HFD, whereas robust rhythmicity in phosphorylation of AKT is seen in NC-fed animals.

Next, in the pool of shared metabolites between serum and liver (Fig. 3A), we analyzed in detail the metabolites present in both tissues and oscillating only in NC (Fig. 3, B–E). Of the eight metabolites oscillating in both tissues under NC (Fig. 3, B and C), most were synchronous by showing a similar phase peak at ZT16 (Fig. 3D). Importantly, there was a different composition of the metabolites shared between serum and liver in NC. Specifically, 75.5% of the serum lipids oscillated in NC versus 36.4% in the liver. Another striking difference relates to nucleotide metabolites; 18% of the shared nucleotide oscillated in the liver, whereas none oscillated in the serum (Fig. 3E). A parallel analysis for oscillating metabolites only in HFD revealed that only two are shared between liver and serum (allantoin and 2-oleoylglycerophosphoethanolamine) (Fig. 3F).

Another revealing difference relates to metabolites oscillating in HFD, where 52% of liver lipid metabolites cycled, whereas only 33% did so in serum (Fig. 3G). Interestingly, 14 metabolites are shared and oscillate under both feeding conditions in both tissues, and most of them belong to the amino acid pathway (Fig. 3H). Analysis of the phase of oscillation for these shared metabolites revealed that, unlike the liver, the serum metabolome is not phase-advanced under HFD (Fig. 3I).

Amino Acids and Peptides

A significant fraction of amino acid metabolites are common to serum and liver, although some unique profiles are distinctive of the two tissues. Specifically, a number of amino acids were found exclusively in the serum (40% of all serum amino acid metabolites), including sarcosine, N⁶-acetyllysine, phenylpyruvate, and creatinine (Fig. 4 (A and B) and supplemental Files 1 and 2). A smaller fraction of amino acid metabolites are unique to the liver (17% of all liver amino acid metabolites), including glutarate, hypotaurine, and S-adenosylmethionine. Comparison of the profiles reveals that fewer amino acids oscillate in the serum than in the liver (48% versus 61.3%) (Fig. 4C). In particular, we observed a 2-fold reduction in circadian amino acids affected by HFD in the serum as compared with the liver (Fig. 4D).

During the analysis of the amino acid pathway, we found that several amino acids, including glycine, serine, and threonine, display rhythmicity only under HFD in the serum. Conversely, metabolites belonging to the subpathway of the tryptophan metabolism, including tryptophan, indolelactate, indolepropionate, and kinurenine, completely lost oscillation in HFD (Fig. 4E and supplemental File 1). These metabolites, however, showed a significant decrease in levels, indicating that one of the effects of HFD is to influence the homeostasis of tryptophan metabolism. Moreover, HFD results in an increase of the overall levels of tyrosine and induces temporal regulation of tyrosine-related metabolites, such as cresol sulfates and 3-(4-hydroxyphenyl) lactate. Although HFD did not appear to regulate the circadian levels of other metabolites belonging to this pathway (phenol sulfates, 4-hydroxyphenylpyruvate), it decreased temporal aspects of their degradation. Interestingly, serotonin levels were significantly decreased in the serum under HFD compared with NC (supplemental File 1), in contrast to observations in human plasma following sleep deprivation (17).

The comparison between serum and liver amino acid metabolites revealed that some metabolites, including betaine, glutamate, glutamine, and 3-methylcrotonylglycine, are not oscillatory in serum, whereas they are robustly cyclic in the liver only in HFD. Similarly, amino acids of glutathione metabolism were devoid of oscillation in the serum (except for the ophthalmate and 5-oxoproline), in contrast to the liver, where they are highly oscillatory in both feeding conditions. In contrast, isobutyrylcarnitine, tyrosine, tryptophan, p-cresol sulfate, 3-indoxyl sulfate, and arginine were not cyclic in the liver but were cyclic in serum in NC or in HFD (Fig. 4E and supplemental Files 1 and 2). Comparison of the oscillation phase revealed that metabolites of both tissues tended to peak later under HFD than what was typically observed in NC. Interestingly, under NC, most of serum and hepatic amino acids peaked at ZT16 (Fig. 4F). Serum also showed an absence of fibrinogen cleavage peptides and dipeptide derivatives and a higher number of metabolites belonging to the dipeptide subpathway compared with the liver (Fig. 5 (A and B) and supplemental Files 1 and 2). Also, there was an increase in oscillating peptides in the serum compared with liver in HFD and a complete loss of rhythmicity for serum peptides that cycle only in NC as compared with liver (Fig. 5D and supplemental Files 1 and 2). No temporal similarity for the oscillation phase was found between the two tissues analyzed (Fig. 5E).

Nucleotides

Most nucleotide metabolites are shared between liver and serum. The liver is the primary organ of de novo nucleotide synthesis, although many tissues use salvage pathways to generate nucleotide levels sufficient for cellular functions (63). Several differences between liver and serum were, however, found, specifically in the content of the purine metabolites adenine and guanine (Fig. 5, F and G). Interestingly, there was a loss in nucleotides oscillating in the serum compared with the liver (37% versus 54%) (Fig. 5H). In particular, no oscillating nucleotide metabolites were found under NC conditions in the serum (Fig. 5I). Moreover, in the serum, nucleotide metabolites tended to peak mostly at ZT4 under both of the feeding conditions, whereas in the liver, they were phase-advanced under HFD (Fig. 5J).

Carbohydrates

Analysis of carbohydrate metabolites showed that 15 are shared between serum and liver. Almost one-third (29%) of the total serum carbohydrate metabolites were found exclusively in the serum, whereas 50% of total liver carbohydrates were found only in the liver (Fig. 6 (A and B) and supplemental Files 1 and 2). There was a striking difference in the number of carbohydrate metabolites whose levels changed in a circadian manner in the liver versus serum. Indeed, whereas only four metabolites (mannose, mannitol, sucrose, and xylose) oscillated in the serum (19%), 20 did so in the liver (61%) (Fig. 6C). Moreover, all circadian serum carbohydrates in NC lost their cycling profile in HFD, whereas in the liver, the oscillations were conserved also under nutritional challenge (Fig. 6, D and F). The cycling of carbohydrates in general appears to be more prominent in the liver, the primary site of both glucose uptake and glucose production. The loss of cycling under HFD of most carbohydrate metabolites in the serum reinforces the notion that the general effect of nutritional challenge is the disruption of homeostasis.

In addition to carbohydrates, another pathway that in the serum undergoes circadian disruption by HFD is glycolysis. Under nutritional challenge, all metabolites involved in glycolysis lost oscillation in the serum, including glucose 6-phosphate, lactate, glucose, and 3-phosphoglycerate, all of which remained oscillatory in the liver (Fig. 6E and supplemental Files 1 and 2). An intriguing example is sucrose (present in serum but not in the liver). Sucrose, which cycles in NC conditions, completely lost oscillation in HFD (supplemental File 1). A likely explanation could be that mice under HFD have a delay in glucose clearance compared with those in NC, mostly because of peripheral insulin resistance that results from higher levels of circulating free fatty acids.

Lipids

Many lipid metabolites are shared between serum and the liver. However, 33% of the lipid species found in the serum were not present in the liver (supplemental Files 1 and 2). Most of these belong to the subpathways of medium-chain fatty acids, monohydroxy fatty acids, branched-chain fatty acids, lysolipids, and metabolites in the carnitine metabolism pathway. Conversely, 18% of liver lipids are not found in the serum. Importantly, whereas more than half of the serum lipids oscillated across the circadian cycle, fewer did so in the liver (55% versus 33%) (Fig. 7A). Diets also differentially affect lipid metabolites in the liver and serum. In particular, under NC, a higher number of lipids oscillated in the serum compared with the liver (Fig. 7B). Conversely, there was a massive loss of lipid metabolites under HFD in the serum as compared with the liver. Strikingly, most metabolites of the lysolipid pathway (55%) cycled in NC but lost oscillation in HFD, in striking contrast to the situation in the liver, in which only 28% of the lysolipids were shown to have a diet effect (Fig. 7C and supplemental Files 1 and 2). Also, metabolites in the essential fatty acid and long chain fatty acid pathways oscillated in the serum only under NC, although they showed a non-cyclic trend in the liver under any diet condition. On the other hand, few lipids of the carnitine metabolism subpathway (e.g. myristate, carnitine, acetylcarnitine, and stearoylcarnitine) oscillated in the liver but not in the serum under HFD (supplemental Files 1 and 2). For a number of metabolites, there were also changes in the phase. Some serum lipid metabolites were phase-delayed in HFD; also, whereas most of the liver metabolites oscillated in NC between ZT0 and ZT12, most serum lipids peaked at ZT8 (Fig. 7D). Thus, circadian lipid profiles are profoundly affected by HFD and show loss of circadian oscillation in the serum.

Cofactors and Xenobiotics

The comparison between cofactors revealed the absence of vitamin B₆, folate, and thiamine metabolites in the serum, whereas these metabolites were highly present in the liver (Fig. 8 (A and B) and supplemental Files 1 and 2). Interestingly, 91% of serum cofactor and vitamin-related metabolites were not oscillating throughout the circadian cycle compared with the liver (57%), whereas the remaining 9% oscillated only in NC conditions (Fig. 8, C and D). This difference was inverted for xenobiotic metabolites (Fig. 8, F–J). Indeed, 41% of serum xenobiotic metabolites were circadian versus only 27% in the liver (Fig. 8, H and I), where there was a complete loss of oscillation under HFD. Moreover, serum xenobiotic-related metabolites were phase-advanced in NC compared with those in the liver (Fig. 8J).

Energy Metabolism

Metabolites related to the Krebs cycle and oxidative phosphorylation showed no major changes along the circadian cycle (Fig. 8, K–N). Importantly, only two metabolites (22%) oscillated across the circadian cycle in the serum and only one in the liver (14%) (Fig. 8, M and N). These data suggest that these critical metabolites must maintain relatively stable levels throughout the circadian cycle and under nutritional challenge.

Serum Metabolites as Markers for Metabolic Disease

The comparison between serum and liver metabolomes showed that almost 39% of the total metabolites identified, across most of the metabolic pathways, were exclusively present in serum (supplemental File 1). Most of the metabolites found in the serum are known to be present in all tissues and organs of the body, including intestine, muscle, brain, epidermis, and spleen, indicating that a given metabolite may be not specific to serum but rather present in low abundance or is not found in the liver. Considering serum metabolites only, 60% of them undergo changes in response to the diet, independent of whether it is a decrease or an increase upon HFD (Figs. 1A and 9 (A–D) and supplemental File 1). Interestingly, many of these metabolites can be considered as markers for various diseases, including cancer, cardiovascular and renal diseases, and metabolic disorders. For example, in our study, we found altered levels of arachidonate, cholesterol, stearate, betaine, glycerol, sucrose, 2-aminoadipate, eicosapentaenoate, 3-methyl-2-oxovalerate, and oleate (Fig. 9, A and B). These metabolites have been shown to be associated with obesity, metabolic syndrome, or type II diabetes (64–71). Moreover, the levels of several metabolites associated with cardiovascular diseases and/or renal failure/dysfunctions were significantly changed, such as, for example, fatty acid, p-cresol sulfate, inosine, genistein, and daidzain (72–79) (Fig. 9, A–C and E). Serotonin, which has a key role in appetite control, was severely affected by HFD, possibly underscoring the tight link between this neurotransmitter and obesity (Fig. 9A). Peripheral serotonin could be involved in the obesity-induced adipose tissue inflammation in our mice (80–84).

Other examples of metabolites that we have found altered and are known to be associated with human diseases are β-sitosterol, hydroxyproline, N¹-methyguanosine, cytidine, indoxyl sulfate, pyruvate, and creatinine, all of which have been reported to be involved in cancer (85–93) (Fig. 9, A, B, D, and E). Interestingly, our analysis showed higher levels of tyrosine, pyruvate, and glutamine, paralleling the abnormal concentration of these metabolites in patients affected by schizophrenia (94, 95) (Fig. 9, A and D). Finally, our analysis also revealed some metabolites used as markers for muscle damage, such as hydroxyproline and 3-methyl histidine (96, 97), which are more abundant in HFD (Fig. 9A).