Daily changes in glucose levels in humans and fruit flies

Glucose metabolism is tightly regulated by circadian processes and disruption results in various disorders, thus we first looked to establish the extent and robustness of time-of-day variability in a network of glucose-related metabolites using steady state measurements in humans. A curated set of 93 metabolites were measured in plasma samples collected every 4 hours over 24 hours (N = 14 subjects; S1A Fig and S1 Data) using a targeted ion-switching LC–MS method [35]. Multivariate analysis of the resulting data exhibits remarkably robust time-of-day variability (Fig 1A and S1 Data). Amino acids and related metabolites such as alanine, ornithine, cystathionine, and asparagine were noted to be elevated around zeitgeber time (ZT) 8 and ZT 12 while elevated levels of TCA cycle metabolites such as citrate/isocitrate and malate were associated with ZT 0 and 4 (Fig 1B). This result implies that the total pools of metabolites are dramatically altered throughout the day, hinting at differential pathways being available for nutrient processing. Pairwise multivariate models distinguished each single time point from a combination of the others (“one versus all”) with the exception of ZT 4. (Figs 1C and S1B–S1F). Pathway analysis of each time point model using significant metabolites demonstrates a unique set of pathways enriched at each specific time of day. For example, alanine, aspartate, and glutamate metabolism and cysteine and methionine metabolism were significantly different during the dark period; TCA cycle was significantly different throughout the day; phenylalanine, tyrosine, and tryptophan biosynthesis were significant across all of the time points, but not at a specific one (Fig 1D). Taken together the multivariate approach indicated overall differences in metabolite levels and activity of pathways at different time of the day.

Formal rhythmicity testing was performed using RAIN/JTK with Benjamini–Hochberg (BH) correction and 24-hour and 20–28-hour rhythms were observed across different q-value cut-offs (q < 0.2, 03, or 0.4, Figs 1E and S1G). For downstream analyses, we used q < 0.2 as a discovery-oriented threshold to balance sensitivity and specificity (expected FDR ≤20% among called rhythmic metabolites), while also reporting results across a range of q-value cutoffs to show robustness (Fig 1E). Metabolites with a RAIN q-value less than 0.2 were selected for further analysis (Figs 1F and S1H). Phases of cycling metabolites were found to be distributed across the day with a peak observed during the early light period (ZT 0–4) (Figs 1G and S1I). Taken together, these results imply that glucose tolerance is influenced by differential downstream pathways regulated in a time-of-day-dependent manner. Notably, the robustness of the response and similarity across the 14 subjects was remarkable and not obvious from individual metabolites levels. This suggests that the concerted network of glucose metabolism provides constraints under which observations of glucose metabolism should be contextualized.

The pathways measured related to glucose metabolism are largely conserved across species, and thus we developed a mass spectrometry-based platform in Drosophila to trace the fate of glucose metabolites through a broad network of downstream metabolism, as observed in the human study above. The use of isotope tracers such as uniformly labeled ¹³C₆-glucose enable the metabolites of interest (downstream of glucose) to be monitored at the atomic level, thus allowing further insights into how glycemic metabolism is altered as a function of time. Uniformly labeled ¹³C₆-glucose was chosen as the stable isotope tracer as it allows carbon flow to be monitored into downstream metabolites spanning the TCA cycle, pentose phosphate pathway (PPP), glycolysis, and amino acid biosynthesis. A range of glucose concentrations were tested in Drosophila to identify an optimal concentration at which Drosophila remained viable. Flies were sampled one hour after introducing the tracer to allow glucose to be metabolized into downstream pathways and to maintain sufficient time resolution to monitor diurnal shifts. The 1 M bolus of glucose was introduced intra-thoracically using a blue dye as a visual aid. The injections were performed on flies that were sedated with cold treatment to prevent metabolic changes caused by hypoxia due to carbon dioxide exposure (Fig 2A and 2B). Representative images of a needle containing the injection solution highlight the small volume (~31 nl/fly) introduced into flies (Fig 2B). Since a non-physiological glucose concentration (~25.6 µM/fly corresponding to ~0.8 pmol/fly or ~148 pg/fly) was used to allow for detection by mass spectrometry, geotaxis and locomotion assays were used to assess whether the glucose and/or blue dye were impacting physiology (S2 and S3 Data files). No significant differences were noted between the glucose and PBS injected conditions for both climbing ability and activity (Fig 2C and 2D). This indicated that the supraphysiological glucose and dye concentrations did not impair geotaxis or locomotion of WT flies and thus were chosen as a tracer to assess time-resolved metabolic nutrient challenge.

We challenged WT Drosophila with 1 M uniformly labeled ¹³C₆-glucose at 4-hour intervals across a day to determine how glucose is processed at different times in the fly body (Fig 2E and S4 Data). Each sample was analyzed using two LC–MS methods covering 22 and 15 downstream glucose metabolites in electrospray ionization positive (211 transitions) and negative (95 transitions) modes, respectively (Fig 2E and 2F). Metabolite transitions were defined by considering the number of carbons and possible labeling patterns of the precursor and fragment (Fig 2G). The resulting data were corrected for instrumental drift and normalized to the blue dye ion counts to adjust for variability in injection volumes before performing a natural abundance (NA) correction. Enrichments were calculated using the NA-corrected data and were used for further analyses (Fig 2E).

To determine data quality, enrichments for all biological and quality control (QC) samples were visualized using a PCA (Fig 2H and 2I). Overall analytical quality was robust as noted through clustering of the QC samples in the scores plot (Fig 2H) and evenly distributed compounds from both ionization modes in the loadings plot (Fig 2I). Furthermore, the platform was sensitive to detect changes in levels of detected isotopologues (chemicals with the same chemical formula but different isotopic compositions) one hour after samples were challenged with labeled glucose at 4-hour intervals. For example, opposing changes were noted in the enrichments of two hexose isotopologues, M + 0 and M + 6 (Fig 2J and 2K and S4 Data). Here, hexose M + 0 (composed primarily of glucose) peaked around ZT 12 (lights off) while the M + 6 (composed of all labeled glucose) form peaked around ZT 4. This points to potential changes in overall glucose usage during the day.

To further understand changes in glucose utilization as a function of time in WT Drosophila, JTK and RAIN algorithms were used to test for 24-, 20−28-, and 12-hour rhythmicity. As the FDR (q-value) cut-off was increased from 0.1 to 0.4, increasing numbers of isotopologues cycled with periods of 24-, 20−28-, and 12-hour periods (Figs 3A, S2A, and S2D and S4 Data). For further analyses, rhythmic compounds with an FDR less than 0.2 were used (Figs 3B, S2B, and S2E and S4 Data). For all periods tested, metabolite pools (sum of all isotopologues for a given metabolite) were not significantly rhythmic. For 24- and 20–28-hour rhythmic results, labeled isotopologues peaked at ZT 4 while unlabeled (M + 0) isotopologues peaked at the light:dark (LD) transition (ZT 12) (Figs 3C and S2C and S4 Data) which is to be expected based on the hexose labeling pattern. On the other hand, 12-hour rhythmic isotopologues peaked mostly around ZT 8 and ZT 20 (S2F Fig). Since 24-hour rhythmicity was not present in metabolite pools, this suggests labeled glucose carbons were incorporated into isotopologues such as serine M + 3, proline M + 2, M + 3, M + 4, and M + 5 at ZT 4. Taken together, peaks at ZT 4 for erythrose-4-phosphate M + 4 and AMP M + 5 can imply use of the PPP while M + 3 forms of serine and alanine may arise downstream of glycolysis, and glutamine and proline isotopologues may arise from the TCA cycle (Fig 3D). This points to overall increased activity of downstream glucose metabolic pathways uniquely at ZT 4, which we will refer to as a “rush hour” of biosynthesis.

To determine if time-of-day differences in glucose usage persisted at an overall metabolic profile level, a multivariate approach using orthogonal partial least squares-discriminant analysis (OPLS-DA) models was employed to determine whether any time point was unique from the rest. From all time points tested, ZT 4 emerged as a distinct time point compared to all remaining time points (Fig 3E and 3F and S4 Data). Higher enrichments of isotopologues such as alanine M + 2, glutamine M + 2 to M + 5, glutamate M + 2 to M + 4, and proline M + 3 to M + 5 and lower enrichments of M + 0 forms (alanine, glutamine, glutamate, and proline) were associated with ZT 4 (Fig 3G and S4 Data). This further supported the rhythmicity analysis in highlighting increased incorporation of labeled glucose into downstream products at ZT 4 suggesting increased biosynthesis.

Because circadian structure in small-molecule pools can reflect insect-specific redox shuttling (e.g., the proline–alanine cycle) rather than net anabolism, we sought orthogonal evidence for biosynthetic allocation into larger molecular pools. We therefore injected flies with either unlabeled glucose or [U-¹³C]-glucose and acquired ¹H–¹³C HSQC spectra (S3A and S3B Fig and S5 Data). Using ROI definitions summarized in the HSQC ROI heatmap (S3B Fig), we quantified lipid-associated regions; ROI2 (S3A Fig) corresponds to cross-peaks in the aliphatic lipid region consistent with lipid-associated methyl groups. Labeling kinetics were temporally structured. In contrast to the early “rush hour” of ¹³C in small-molecule metabolites around ZT 4, consistent with rapid glucose entry into central metabolism, the lipid-associated HSQC ROIs showed a delayed/shifted pattern. For ROI2, the labeled–unlabeled difference (net ¹³C incorporation into this lipid-associated ROI) peaked in the mid-to-late light time (~CT6–10) (Fig 3H and S5 Data) and reached a minimum later in the dark period (~CT16–20), consistent with time-of-day variation in allocation of glucose-derived carbon into lipid-associated pools. We therefore interpret the temporal program as comprising both central-metabolism flux changes (which may include proline–alanine cycling) and a measurable component of biosynthetic allocation into lipid-associated carbon.

To determine whether the observed “rush hour” of glucose flux at ZT 4 is driven by the endogenous circadian clock, we performed U-¹³C₆-glucose tracing in the arrhythmic period null mutant, per⁰¹. These mutants lack a functional molecular clock, allowing us to disentangle clock-driven regulation from passive responses to environmental cues like light. Following the same time-resolved injection protocol used for wild-type flies, we quantified the labeling fraction of glucose isotopologues (f_labeled = 1 − (M + 0/∑_iM + i) (S3C Fig and S6 Data). The analysis revealed a complete loss of rhythmic glucose processing in the absence of a functional clock. While wild-type flies showed a robust 24-hour rhythm in hexose M + 6 labeling (RAIN, p = 0.03), this rhythm was entirely abolished in per⁰¹ mutants (RAIN, p = 0.9). As illustrated in S3C Fig, the diurnal oscillation in glucose incorporation observed in wild-type flies is attenuated in per⁰¹ mutants, which exhibit an ultradian-like labeling pattern across the day. This result provides direct evidence that the temporal gating of glucose into central carbon metabolism is actively regulated by the core molecular clock.

As activity levels can act as a weak zeitgeber to influence the clock [36], we next sought to determine how glucose is used differently in a hyperactive short-sleeping mutant deficient in dopamine reuptake, fumin (fmn) [34]. The fmn mutation leads to a defective dopamine transporter resulting an increased amount of dopamine signaling in the synapse. Consistent with the critical and conserved role of dopamine system in regulating arousal and sleep, fmn flies exhibit reduced sleep and prolonged activity compared to wild-type flies. The increase in activity is accompanied by an up-regulated metabolic rate, making the fmn mutant a vital tool to provide unique metabolic fingerprints. Overall, fmn displays elevated activity profiles as compared to WT with higher levels observed during the dark period (Fig 4A and S3 Data). We challenged the mutant with glucose in a similar manner to WT. The climbing ability and activity of fmn were also not impaired in flies injected with glucose versus PBS at both ZT 0 and ZT 12 (Fig 4B and 4C; S2 and S3 Data). Therefore, the injection of supraphysiological concentrations of glucose did not dramatically alter the physiology of the flies.

Overall, wild type and fumin process glucose distinctly as indicated by a significant two-component discriminant (OPLS-DA) model (Fig 4D and S4 Data). Hexose M + 6 enrichments were relatively higher in WT while M + 0 enrichments were relatively higher in fmn suggesting that labeled glucose was more readily used in fmn. Overall, enrichments of primarily labeled isotopologues (i.e., glutamate M + 1, glutathione M + 5, and asparagine M + 2 and M + 4) contributed to the separation of fmn from WT while enrichments of several unlabeled isotopologues such as asparagine, glutamate, and glutathione were more associated with WT samples (Fig 4E). Univariate analyses of genotype differences also showed similar changes (Fig 4F). For instance, enrichments of asparagine M + 4 and leucine/isoleucine M + 5 were significantly higher (at least 2×) in fmn as compared to WT (Fig 4G). Since leucine and isoleucine are essential amino acids, the synthesis of these may be a byproduct of the fly microbiome. Additional significant differences such as higher AMP M + 5 levels in WT and elevated glutathione M + 5 levels in fmn were noted through the adjusted genotype p-value (Fig 4H). As glutathione M + 5 requires either glutamine M + 5, or glutamine M + 3 and glycine M + 2 as precursors, we also looked at the correlation of these isotopologues. Glutathione M + 5 enrichments were found to be significantly correlated with glutamine M + 3 (p < 0.05) and weakly correlated with glutamine M + 5 (p < 0.1) enrichments in fmn (S4A Fig). From both the univariate and multivariate approaches, relatively higher enrichments of AMP M + 5, a metabolite related to the PPP, and metabolites associated with glycolysis (alanine M + 2, serine M + 2 and M + 3) were observed in WT while metabolites involved in or downstream of the TCA cycle (alpha-ketoglutarate M + 5, asparagine, glutamate, and glutamine isotopologues) had relatively higher enrichments in fmn (Fig 4I). Overall, both multivariate and univariate approaches support increased TCA cycle labeling in fmn while glycolysis and PPP labeling is associated with WT.

In addition to overall genotype-level differences, we sought to determine how the two genotypes differed in glucose utilization through time. Through pairwise OPLS-DA models of the genotypes at each time point, a significant model was observed at ZT 4 (S4B and S4C Fig and S4 Data). Three types of changes were noted in labeled isotopologues: 1. Higher relative enrichments in WT (i.e., alanine M + 1 to M + 3 and proline M + 3 to M + 5); 2. Higher relative enrichments in fmn (i.e., alpha-ketoglutarate M + 2 to M + 4, asparagine M + 1 to M + 4, and glutathione M + 1, M + 2 and M + 5); and 3. Higher enrichments spread across the two genotypes (i.e., higher AMP M + 5 and lactate M + 3 in WT versus higher AMP M + 1 and lactate M + 2 in fmn) (S4D Fig). This suggests differences in downstream glucose products and/or differences in glucose utilization pathways in WT when compared to fmn. Overall, both multivariate and univariate approaches supported the idea that glucose is differentially used between the two genotypes at ZT 4 with increased TCA cycle labeling observed in fumin.

Next, to determine whether glucose is processed differently by fumin based on time-of-day, pairwise discriminant models were generated to test whether a time point(s) was distinct from all other time points. Both ZT 0 and ZT 12 emerged as being significantly distinct from other time points through one- and two-component models, respectively (Fig 5A, 5B, and 5D, and S4 Data). Increased enrichments of labeled isotopologues were noted at both time points, ZT 0 and ZT 12, pointing to increased incorporation of glucose carbons into the isotopologues (Fig 5C and 5E). Furthermore, the light time points (ZT 4 and ZT 8) were significantly different from the dark ones (ZT 16 and ZT 20) in an OPLS-DA model (S5A Fig). The light time points were associated with relatively increased enrichments of labeled isotopologues (S5B Fig).

To probe whether proline-linked redox handling contributes to the altered metabolic state of fmn, we performed additional experiments using dietary proline to stress proline-linked redox buffering and asked whether this would differentially affect wild-type versus fmn flies across time of day. Both wild-type and fmn flies were exposed to 30 and 15 mM proline (S6A–S6H Fig and S7 Data) separately during the pre-feeding phase (5 days) and on day 6 (day 0 on graphs), flies were challenged with 1 M ¹²C₆-glucose at ZT 4 and ZT 16 (injected group; Figs S6B, S6D, S6F, and S6H). First, to rule out trivial intake or baseline toxicity effects, we quantified survival of flies on 30 and 15 mM proline food during the pre-feeding phase. fmn mutants exhibited lethality (29% to 46%) during the pre-feeding phase to 30 mM but not to 15 mM proline, whereas wild-type flies tolerated proline feeding well. All the flies that survived during the pre-feeding phase were grouped on day 6 into 3 sets each (control group; Figs S6A, S6C, S6E, and S6G) and (injected group) with 20–25 flies each set for ZT 4 and ZT 16. For control group, flies were transferred to fresh proline media on Day 6 (15 or 30 mM). We observed no significant difference in lethality between wild-type and fmn flies at ZT 16 (15 or 30 mM proline control condition), indicating that both genotypes survived well under control conditions at this time point. Notably, we did not observe significant differences in lethality in flies injected with ¹²C₆-glucose at ZT 4, but fmn flies maintained on 30 mM proline exhibited clear lethality on days 2–3 at ZT 16 in the injected group, whereas wild-type flies on the same proline-supplemented medium remained largely viable over the same time frame. For proline 15 mM condition, no differences in survival were noticed in general, only wild-type flies showed some lethality at day 2 in ZT 4 injected group. Thus, proline loading selectively compromised survival in fmn in the context of repeated glucose challenges, revealing a delayed vulnerability consistent with impaired redox-buffering capacity in the mutant.

Given that fmn had two peaks with strong biosynthetic activity, we then investigated whether excess glucose metabolism at ZT 0 and ZT 12 followed a similar set of pathways in fmn using a shared and unique structures (SUS) plot (Fig 5F and S4 Data). In this analysis, the loadings (or distribution of isotopologue enrichments) from both models were plotted against each other with ZT 0 versus All (Fig 5C) on the y-axis and ZT 12 versus All (Fig 5E) on the x-axis. The resulting plot highlighted a weak correlation between the two-time points through both unique (blue and orange), as well as conserved (green) metabolism (Fig 5F). For example, enrichments of ornithine M + 1 and alpha-ketoglutarate M + 2 were relatively higher at both ZT 0 and ZT 12, but were only significantly contributing to the separation of ZT 12 from other time points. On the other hand, enrichments of isotopologues such as glutamate M + 2 and M + 4 only significantly contributed to the separation of the ZT 0 time point. From both models, broad labeling of downstream glucose products with differences in metabolite and/or labeling were noted at both time points. Overall increased TCA cycle labeling was apparent at both time points through unique and common isotopologues while glucose carbons may be incorporated into glycolysis more at ZT 12 (Fig 5H). Overall, this highlights that although a few similar pathways may be activated in fmn at the two LD transitions (or dawn and dusk), generally different isotopologues are generated (i.e., unique metabolic pathways used).

Since fumin at ZT 0 and ZT 12 and wild type at beginning of the light phase (ZT 4) had all shown increased biosynthesis from labeled glucose, we next looked to determine whether fmn used glucose similarly to WT at these time points. When comparing the loading plots from the early light period from both models (e.g., WT ZT 4 versus All (Fig 3G) and the fmn ZT 0 versus All (Fig 5C), the resulting SUS model highlighted a notable degree of correlation between the two conditions (Fig 5G). Furthermore, several conserved compounds significantly contributed to the separation of both ZT 4 in WT and ZT 0 in fmn, highlighted in green. For example, proline M + 4 and glutamine M + 3, M + 4, and M + 5 enrichments were relatively higher at both ZT 4 in WT and ZT 0 in fmn relative to other time points. On the other hand, compounds significantly contributing to the separation of fmn ZT 0 (VIP compounds such as citrate M + 3, M + 5, and M + 6) were also noted and although they displayed similar changes in WT, they did not meet the significance cutoff as defined by a VIP value greater than one. In contrast, fewer concerted changes were observed between fmn at ZT 12 and WT at ZT 4 (S5C Fig). This analysis suggests that there is a similar and specific set of metabolic pathways activated at ZT 4 in WT and ZT 0 in fmn compared to other timepoints indicating a potential phase advancement in fmn of metabolic pathways primed to utilize glucose in the early morning period.

To test whether circadian variation in glycogen buffering could contribute to the bolus tracer phenotype, we performed two independent glycogen-focused assays: (1) whole-body extracts were enzymatically hydrolyzed to convert glycogen to glucose prior to isotopologue quantification of the hexose M + 6/M + 0 ratio across circadian time (S7A Fig and S8 Data), and (2) total glycogen pools were measured (S7B–S7D Fig and S9 Data). WT total glycogen/protein was comparatively stable across ZT (RAIN p = ns), whereas fmn showed markedly reduced glycogen with time-of-day structure (RAIN p = 1.73 × 10⁻⁵; amplitude ≈ 0.17), including a pronounced reduction near early day (S7A Fig). The hydrolysis-derived hexose labeling metric (M + 6/M + 0) exhibited stronger circadian structure in fmn than in the control background, indicating genotype-dependent temporal regulation of glycogen-associated labeling. Together, these data indicate that time-dependent glycogen-associated labeling is not a dominant feature of baseline glucose handling in the control background, but becomes strongly time-structured in the hyperactive, high-energy-demand fmn genotype.

Having observed unique processing of glucose at two time points in fumin, we next asked if the changes are due to underlying cycling over the course of a day. To address isotopologue rhythmicity, 20–28-, 24-, and 12-hour periods were tested. Overall, a greater number of rhythmic isotopologue enrichments and/or pools were observed in fmn than WT for all periods tested across different RAIN and JTK FDR cut-offs (Figs 6A, 6E, S8A, and S4 Data). For further analyses, compounds meeting a 0.2 RAIN FDR cut-off were used (Figs 6B, 6F, and S8B). In contrast to WT, metabolite pools cycled with 20–28-, 24- and 12-hour periods. Most of the significant 24-hour cycling labeled isotopologue enrichments were noted to peak around ZT 0 to ZT 4 while unlabeled forms peaked during the dark period between ZT 16 and ZT 20 and pools were noted to peak around ZT 4 (Fig 6C). Since the peak in overall metabolite levels (pools) and enrichments (labeled isotopologues) both occurred between ZT 0 and ZT 4, this suggested that these arose from the incorporation of labeled glucose. However, alternative sources of carbon contribution cannot be ruled out. This trend was also observed in phases of significant 20–28-hour cyclers (S8C Fig). Similar to the observed 24-hour rhythmicity, significant compounds with 12-hour periods also showed a number of labeled isotopologues with peaks between ZT 0 and ZT 4 along with pools peaking at ZT 4 and non-labeled forms, M + 0, peaking around ZT 8 (Fig 6G). In this case, because pools are peaking in between labeled and unlabeled forms, it is difficult to determine whether overall pool levels are increased due to biosynthesis from glucose. However, it is likely that glucose carbons contributed to the increased pool levels noted as alternative routes are limited. Overall, rhythmicity analyses support increased biosynthesis of isotopologues at the LD transitions in fmn with a substantial contribution to the increased biomass arising from glucose carbons. Furthermore, the peak of rhythmicity around ZT 0 in fmn also supports the idea of phase-advanced biosynthesis from WT.

Since rhythmicity was observed in both genotypes, we next explored whether similar compounds were cycling in WT and fmn. For 24-hour rhythmic compounds, about 50% of the isotopologues cycling in WT were also noted to be 24-hour rhythmic in fmn (Fig 6D). A similar observation was also noted with 20–28 hour rhythmic compounds (S8D Fig). In contrast, fmn primarily contained unique isotopologues and/or pools that were rhythmic, but it is important to note that a few are trending towards significance in WT (glutathione M + 1 and M + 3 and glutamine M + 0 and M + 2). A similar trend was also noted for 12-hour rhythmic compounds (Fig 6H). However, glutamine M + 0 and M + 3 were significant in WT for 12-hour periods but trended towards significance in fmn. This highlights that a majority of the cycling observed in fmn was uniquely gained rather than conserved or lost from the cycling observed in WT. Overall, rhythmicity analyses also supported TCA cycle labeling in fumin with contributions from both diurnal and ultradian rhythmicity while cycling of glycolysis and PPP related isotopologues was primarily diurnal.

In order to understand how carbon flow was occurring through the TCA cycle, we employed a product-to-precursor relationship analysis [37]. This analysis generates ratios (phi values) using different mathematical relationships between TCA cycle precursors and products to provide further insight into how glucose is being used. For example, the SM4 ratio looks at the formation of malate M + 4 from succinate M + 4. Since malate M + 4 is only derived from succinate M + 4, the ratio provides insight into stepwise flow of oxidative carbons at this TCA cycle step relative to other pathways like anaplerotic pyruvate carboxylase. Neither of the inputs for this ratio displayed significant rhythmicity (24- or 12-hour), however, a 12-hour rhythm trending towards significance was observed in the WT ratio (Fig 7A and S4 Data). This highlights the additional value gained from the ratio analysis which was not captured at the enrichment-level alone and suggests rhythmicity between oxidative and synthetic mitochondrial metabolism.

Overall, the Malate_PEP phi value demonstrated significant 24-hour rhythmicity and two others, POM and SM2, trended towards significance in fmn (Fig 7B and 7D). In contrast to 24-hour rhythms, 12-hour cycling trended towards significance in three WT phi values (SM4, MO3, and Malate_PEP) (Fig 7C and 7D). For SM4, a ratio of one (i.e., ZT 4) indicated the formation of malate M + 4 from succinate M + 4 was not diluted by non-oxidative carbon flows. However, a ratio greater than one (i.e., ZT 20) could indicate either reversal of the pathway or the entry of alternately labeled carbons in the formation of malate (Fig 7A middle panel). Similarly, ratios greater than one in the 12-hour rhythmic MO3 phi value suggested that either the malate was being formed from oxaloacetate or alternative routes with labeled carbons were being used to form oxaloacetate (Fig 7E). Given that decapitated fly bodies (and not single tissues) were examined and that some compartmentation is to be expected, then phis become more semi quantitative or qualitative indices of flux since true single-pool steady state assumptions may not apply. Interestingly, the Malate_PEP ratio switched from 12-hour rhythmicity in WT to 24-hour rhythmicity in fmn (Fig 7F) and indicates a relative shift in glycolysis versus gluconeogenesis. However, the ratio was less than one in both genotypes indicating that unlabeled carbons from alternate metabolites, such as glucose, were diluting the formation of PEP from malate. A different ratio looking at the formation of malate from succinate using the M + 2 forms, SM2, displayed 24-hour rhythmicity in fmn with values less than one indicating the entrance of alternate sources of unlabeled carbons (Fig 7G). fmn also displayed 24-hour rhythmicity in the POM ratio which looked at the formation of malate from pyruvate (Fig 7H). Here, in instances where the ratio decreased less than 1, it implied alternative sources of unlabeled carbons diluting the pathway. Overall, the product to precursor analysis highlighted a shift in key oxidative and synthetic metabolic flows from ultradian rhythmicity in WT to 24-hour rhythmicity in fmn pointing not only to differences in glucose utilization but also potential differences in demand.

Since the timing of food availability can be a zeitgeber for molecular clock rhythmicity, we asked how nutrient factors impact the time shift in biosynthesis observed between WT and fmn. To determine whether altered feeding rhythms between the genotypes were impacting downstream glucose biosynthesis, feeding profiles of WT and fmn were monitored through the Activity Recording CAFE (ARC) assay revealing similar levels of food consumption in both genotypes overall and during the light and dark periods. However, the feeding peak in fmn (ZT 4.5) was phase delayed in comparison to WT (ZT 3.25) (Fig 8A and S10 Data). Since the biosynthesis peak in fmn was observed to be phase-advanced from WT while the feeding rhythms were phase-delayed, we concluded that overall feeding was not an underlying cause of the biosynthesis shift in fmn.

To further establish whether the availability of food was driving biosynthesis in WT, a shortened time course from ZT 0 to ZT 12 was analyzed under ad libitum and a 4-hour fasted condition, where food was withheld for 4 hours prior to injection with the glucose tracer. Distinct metabolic profiles were noted for the two conditions, with different labeled isotopologue enrichments associated with each (Figs 8B and S9A). We next determined whether ZT 4 was still metabolically distinct from remaining time points under the short-term fasted condition. The ZT 4 model did emerge as significant in a pairwise discriminant model to all remaining time points (Fig 8C and 8D). Similar to the earlier observation, a majority of the labeled isotopologues were associated with ZT 4 (S9B Fig and S10 Data) pointing to increased biosynthesis from glucose. To further determine whether the presence of increased biosynthesis was arising from similar compounds and/or pathways observed previously (Fig 3G), we performed a SUS analysis using the multivariate OPLS-DA loadings from both ZT 4 versus All (WT—Fig 3G and short-term fasted—S9B Fig) models. The resulting plot showed conserved and similar changes occurring at both ZT 4 time points (highlighted in green) (Fig 8E). It is important to note that several compounds uniquely contributed to the separation of ZT 4 in the short-term fasted model, but this is to be expected as food restriction would impact metabolism and excess glucose processing. We also performed a complementary analysis using the enrichments identified from the WT ZT 4 versus All model (Fig 3G) as significantly contributing to the separation of ZT 4 from all other time points (VIP greater than 1). Using these VIP metabolites in a clustering analysis with calculated enrichments from ZT 0–12 from the WT, short-term fasted WT and fmn time courses, two main clusters emerged for both time and compounds (Figs 8F and S9C). On the time axis, ZT 4 from both the WT and short-term fasted condition clustered with the fmn ZT 0 time point while all remaining time points formed a second cluster. On the metabolite axis, labeled and unlabeled isotopologues formed the two main clusters. This clustering further highlighted the peak in labeled isotopologue enrichments at ZT 4 and ZT 0 in WT and fmn, respectively. Since rhythmicity analyses had also previously shown peaks in labeled isotopologues and troughs for unlabeled forms at ZT 4, we assessed the overall enrichment profiles of the short-term fasted condition using the 24-hour rhythmic metabolites (identified in Fig 3B). This further showed similar patterns across the two data sets (Fig 8G). Overall, this highlighted sustained increased downstream biosynthesis from glucose carbons and peaks in rhythmicity at ZT 4 in WT irrespective of food pointing to a potential underlying circadian driver rather than the availability of food.

For recruitment of human participants, Institutional Review Board of the University of Pennsylvania and the Clinical and Translational Research Center of the University of Pennsylvania granted the study protocol for subject recruitment (approval no. 807347). Written informed consent was obtained from all research subjects.

Drosophila melanogaster, wild type (isogenic w¹¹¹⁸ stock), fumin [34] and per⁰¹ mutants [25] as characterized previously, were maintained on standard cornmeal/molasses medium at 25°C under 12:12 LD conditions.

After three days under LD entrainment, 5-to-7-day-old wild type or fumin flies were anesthetized on ice for 15 min at ZT 0 (lights on) or 12 (lights off). Flies were either not injected (control) or injected with PBS or 1 M Glucose. For geotaxis assays [61,62], flies were placed in new vials without food for one hour after which climbing activity was measured by the time taken to climb 4 cm per fly, measured in triplicate using 5–8 flies per group. For locomotion assays [63,64], flies were placed in 5 × 65 mm glass tubes with 5% sucrose post-injections and monitored using Activity Monitoring System devices (DAMS) from Trikinetics (Waltham, MA) for a minimum of 24 hours in 25 °C LD incubators. Locomotion data was analyzed using a custom MATLAB (MathWorks, Natick, MA Versions 2013B and 2020A) script [65] with additional analysis and plotting performed in R (version 4.0.3). Comparisons for both assays were made using a two-sided t test using p < 0.05 as a threshold for significance.

An ARC assay [66] was used to record feeding rhythms of wild type and fumin. Male flies were entrained under 12 hours Light:Dark conditions within one day of eclosion at 21 °C. On day 6, entrained flies were used to set up the feeding assay with one fly per well containing 300 µl of 2% agar. Genotypes were loaded in alternating wells. A capillary containing a (infrared) dye (for detection of food level) and 2.5% yeast and 2.5% sucrose was placed into each well as a food source. After allowing for habituation to the setup, data was recorded for 48 hours with capillaries replaced each day before ZT 12. This assay was performed at 21 °C to minimize evaporation of the capillaries. Data was visualized in R (version 4.0.3) and any wells containing flies that had not survived, evaporation only of food and/or inconsistent food level recordings were discarded. Data was processed using a python program (Noah15.2) with additional analyses in R (version 4.0.3).

Within one day after eclosion, 15–18 males were sorted into a new vial for entrainment to LD rhythms for a minimum for 3 days in appropriate incubators at 25°C. All flies were between 5–7 days old at the time of injection. One vial was placed on ice to immobilize flies 15 min prior to each injection time point. The injection solution consisted of 1 M uniformly labeled (¹³C₆) glucose (CIL D-Glucose, 99%) and a blue dye (50× dilution of McCormick, FD&C Blue Dye No. 1) to allow for visualization during injections and normalization for variations in injection volumes during data processing. The solution was injected into the thorax using glass capillaries (9 cm × 1.14 mm diameter, Drummond Scientific, Broomall, PA) on ice. After injection, flies were moved to empty vials containing a kimwipe moistened with 1 ml of water to prevent flies from becoming dehydrated and returned to the appropriate light or dark 25 °C incubator to metabolize the glucose tracer. After one hour, flies were collected on dry ice and stored at −80 °C until extractions. Five separate days of injections were performed as biological replicates for both WT and fumin mutants. Any flies which died after the injections were discarded. No difference in survival rate (>95%) was noted between the genotypes.

For the shortened ad libitum versus 4-hour short term fasted time course, injections were modified as follows: 20–24 male wild-type flies were collected and entrained within one day of eclosion and were injected on day 7. Four hours prior to each time point, flies were transferred either to fresh food vials (ad libitum condition) or agar vials (4-hour fasted condition). Each group was injected with either 1 M ¹³C₆-glucose or 1 M ¹²C₆-glucose (Acros Organics alpha-D(+) Glucose, 99%) with blue dye (504 µM Sigma Aldrich Brilliant Blue FCF, analytical standard) dissolved in PBS.

Fly heads and bodies were separated prior to metabolite extraction, using an adaptation of the Bligh-dyer extraction [35,67]. Briefly, to each fly body sample, a total of 600 µL of cold 2:1 methanol:chloroform was added and homogenized in a bead-based tissue homogenizer at 25 Hz for 4 min (TissuLyser II, Qiagen, Hilden, Germany). 200 µL each of water and chloroform were then added, followed by centrifugation at 18787g for 7–10 min at 4 °C. 350–400 µL of the upper (aqueous) layer was collected and dried under vacuum until dry or overnight. Samples were resuspended at 4 µl/fly body for negative mode and diluted to 5.33 µl/fly body for positive mode using 50:50 water:acetonitrile. Each sample was analyzed separately in ESI positive and negative modes. Transitions included in each method are shown in Table A in S4 and S10 Data files. Chromatographic separations utilized an ACQUITY UPLC BEH Amide column (2.1 × 150 mm, 1.7 µm) on a Waters ACQUITY H-Class UPLC coupled to a triple quadrupole Waters Xevo TQ-S Micro MS (Milford, MA). LC conditions for the positive-ionizing method were performed as described previously [68] with a modified gradient from 100% to 20.6% B over 15 min at 0.35 mL/min, followed by a wash of 100% A for 5 min. Mobile phase B was changed from 0% to 100% from 20 to 22 min and held for column equilibration until 30 min. For the negative-ionizing method, solvents consisted of 95:5 water:acetonitrile with 20 mM ammonium bicarbonate, pH 9 (mobile phase A) and 90:10 water:acetonitrile with 20 mM ammonium bicarbonate, pH 9 (mobile phase A). The gradient was changed from 100% to 20.6% B over 15 min at 0.4 mL/min, followed by a wash of 100% A for 5 min. Mobile phase B was changed from 0% to 100% from 20 to 22 min and held for column equilibration until 30 min. Ion counts were acquired through multiple reaction monitoring (MRM).

For the feeding time course samples, the LC–MS analyses was performed as described above with the addition of 5 µM ammonium phosphate to both mobile phases in both ionization modes.

Chromatograms were processed using TargetLynx under MassLynx version 4.1 or using El-MAVEN (version 0.12.1 beta) and exported as ion counts for further processing in R (version 4.0.3). Column pressure deviations were monitored to identify poor sample injections which were excluded from further analysis. MRMs which were either not detected or overlapped with other chromatographic peaks and thus prevented unequivocal peak integration were set to zero. For each sample, analytical duplicates were injected in a randomized order, along with QC samples every 4–10 injections, which was comprised of a pool of all samples.

In order to perform an instrumental drift correction, consecutive QCs were averaged. Next, ion counts were used to calculate metabolite pools and were also used to calculate a ratio of each transition acquired to the calculated pool. A LOESS correction was performed on the metabolite pools to minimize the impact of differences in detection of isotopologues across samples [69]. The corrected pools were multiplied by the ratios calculated earlier to regenerate transition-level data corrected for instrumental drift. A validated FCF transition from the negative mode data was used to normalize for variation in glucose injection volumes for the wild type and fumin time courses. Analytical replicates were then averaged and used to perform a NA correction using the Polly Phi LC–MS/MS application [70,71]. Transitions were summed to obtain isotopologue-level data which was then used to calculate enrichments of each isotopologue on a metabolite level. Enrichments were used for multivariate and rhythmicity analyses. The NA-corrected data was also used to calculate phi values (product/precursor relationships) using the Polly Phi LC–MS/MS application. The resulting ratios were used for rhythmicity analyses.

The ad libitum versus short-term fasted time course data acquired validated FCF transitions in both ESI negative and positive modes and were used to normalize the data within each mode. After averaging the analytical replicates, a background correction was performed using the Polly Phi LC–MS/MS application using the ¹²C injected samples as background samples for the ¹³C injected samples. The background-corrected data was then corrected for NA.

Principal component analyses through SIMCA (Umetrics, version 17) were periodically checked throughout the data processing steps to ensure data quality was being maintained. Enrichments were used for further OPLS-DA models in SIMCA. OPLS-DA models were generated to test for time-dependent within each genotype and/or group by defining each time point as a class with remaining time points as a second class. Genotype/Group level differences were also tested overall as well as for each time point. For each model, the number of components was trimmed to obtain a model with the lowest CV-ANOVA p-value to avoid overfitting. A CV-ANOVA p-value cutoff of 0.1 was used to determine significant models for further analysis. SUS analyses [72] were also performed in SIMCA using significant OPLS-DA models. Each metabolite enrichment was treated as an independent variable in multivariate models for computational simplicity.

In addition to multivariate analyses, univariate analyses were also performed. A two-way ANOVA using genotype and time was performed followed by a post-hoc Tukey HSD test for any significant interaction compounds. P-values were adjusted using a BH correction. The adjusted ANOVA genotype p-values along with calculated fold changes of fumin to wild type were used in a volcano plot.

To identify metabolites with 24-h rhythmic patterns, we applied two complementary rhythm-detection methods (JTK_CYCLE and RAIN), which test whether a time series follows a repeating daily pattern while accounting for the sampling interval. Using both approaches provides a robustness check, as each method has different sensitivities to waveform shape and noise. Using the enrichment-level data as 5 biological replicates from ZT 0–20, RAIN [73] and JTK [74,75] algorithms were used to assess diurnal (24- and 20–28-hour periods) and ultradian (12-hour) rhythmicity through the R Rain and MetaCycle packages, respectively. Results from RAIN using a q-value cutoff of 0.2 were used for further analyses. For the product to precursor relationships (phi values), RAIN p-values less than 0.1 were used to determine significance. For feeding data, rhythms were tested using food consumption combined into 10-minute bins through JTK.

Representative sample of wild-type flies were injected with a glucose and blue dye solution, collected, and extracted as described earlier. Uninjected flies were collected and extracted to use as a background matrix for the calibration curve. To ensure a similar background, the upper fractions from uninjected samples were pooled and then aliquoted before drying. Injected samples were resuspended in diluent while uninjected samples were resuspended in appropriate calibration solutions containing varying FCF concentrations in the diluent. Integration was performed in El-Maven as described with data processing in R (version 4.0.3) and Prism (version 9).

Plasma samples from 14 patients (8 males and 6 females; 12 White; average standard deviation age 29.14 9.37) collected as described in [76] were used. These samples were collected every four hours from an in-patient protocol and the first 24 h of samples analyzed here. The consent of the Institutional Review Board of the University of Pennsylvania and the Clinical and Translational Research Center of the University of Pennsylvania was granted for this study and informed consent obtained from all research subjects. 150 µL of plasma was used as a starting material and extracted using the method described in metabolite extraction and LC–MS measurements section with the following differences: To each sample 900 µL of 2:1 methanol:chloroform was added followed by 300 µL of chloroform and water. Samples were vortexed and centrifuged and 200 µL of the upper fraction was collected for LC–MS analyses and dried for 6 hours under vacuum. Paired samples were acquired using an ion-switching method using transitions, solvents, instrumentation, LC and MS parameters described in [35]. For sample resuspension, dried upper fractions were resuspending using 200 µL of 15:85 milliQ water:acetonitrile. Acquired data was integrated using El-MAVEN (version 0.12.1 beta) as described in the data processing and normalization section. Exported ion counts were corrected for instrument drift using a linear correction followed by median fold change normalization. The data was combined for all subjects into a single dataset. To normalize for interindividual differences in concentration, the ratio of each metabolite to the average for that metabolite was taken on an individual level. The resulting ratios were used for multivariate and rhythmicity analyses as described in the multivariate and rhythmicity analysis section.

Significant pathways as identified by VIP values were used in MetaboAnalyst 5.0 for a pathway analysis [77]. Metabolites were uploaded using HMDB identifiers and processed using a hypergeometric enrichment method, relative-betweenness centrality for topology analysis using the Homo sapiens (KEGG) pathway library. The resulting data was imported into R (Version 4.0.3) and further analyzed. Significant pathways were defined using an FDR less than 0.05. To be included in further analyses, an impact value greater than zero was required for at least one time comparison.

Glycogen measurements were done as previously [78] documented. Fly bodies of 5- to 7-day-old adult male flies were collected on dry ice prior to homogenization. The decapitated fly bodies (N = 3 groups/ five flies each group/genotype/time point) homogenized in 200 µl of 0.1 M NaOH. The homogenate was then centrifuged at 13,300 rpm for 10 min at 4 °C. The supernatant was extracted from each replicate separately and 20 µl of the lysate was treated with 5 mg/ml Amyloglucosidase (Sigma-Aldrich, CAS-No-9032-08-0) in 0.2 M acetate, pH 4.8. Amyloglucosidase enzyme catabolizes glycogen to yield free glucose molecules. Simultaneously, another 20 ul aliquot of the lysate was treated with 0.2 M acetate, pH 4.8 alone (untreated lysate). Both reactions incubated for 2 h at 37 °C on the heating blocks. Subsequently, Amplex Red Glucose/Glucose Oxidase Assay kit (Invitrogen-A22189) was used to measure the free glucose content in each reaction in triplicate. The protein concentration of these reactions was then measured using the Pierce^TM BCA Protein Assay Kit (Thermo Scientific, Rockford, IL) for normalization. Finally, the glycogen content of each sample was calculated by subtracting the free glucose concentration of the untreated lysate from the free glucose concentration of the lysate that was treated with amyloglucosidase.

Lysates post Amyloglucosidase reaction, as mentioned in previous section, was subjected to Bligh-Dyer extraction. Polar fraction of the extract was evaporated, reconstituted in 98:2 methanol/water, and 2.5 μl spotted on PTFE-coated glass slides for further analysis in DESI-MS.

Glucose isotopomer intensities were obtained from DESI-MS analysis of fly body extracts and exported as peak areas for the [M + ³⁵Cl]⁻ adduct at m/z 215.0328–221.0529 (M + 0 to M + 6). For each sample, the total isotopomer pool was calculated as the sum of the M + 0–M + 6 signals, and fractional enrichment was defined as the M + 6 fraction of this pool (M6/pool = M + 6 ÷ ΣM + 0–M + 6). Flies were sampled every 4 h from ZT 0 to ZT 20 with three biological replicates per group (fmn and wild type) at each time point (40 male fly bodies each replicate/genotype/time point). To visualize multi-day dynamics, these replicates were ordered by sample identifier and mapped onto three consecutive “virtual” days (replicate 1: ZT 0–20 hours, replicate 2: ZT 24–44 hours, replicate 3: ZT 48–68 hours), yielding a continuous 72 hours pseudo-time series for each group.

Flies were reared under standard conditions (25 °C, 12 hours:12 hours light–dark cycle) on conventional cornmeal–molasses food. 2–4-day-old adult males were used for the experiments. For proline supplementation, L-proline (Sigma) was dissolved in distilled water to prepare a 1 M stock solution. The stock was added to melted, cooled (~50 °C) fly food to a final concentration of 30 mM and 15 mM, mixed thoroughly, and poured into standard vials.

For each genotype, flies were transferred to vials containing 30 or 15 mM proline-supplemented food (“proline media”) and allowed to feed ad libitum for 5 consecutive days. Multiple vials contained 20 flies each, and flies were transferred to fresh 30 and 15 mM proline food every 2–3 days as needed to maintain food quality. Only flies that survived on 30 and 15 mM proline through the pre-feeding period were used for subsequent glucose injection experiments and controlled condition. On day 6 (Day 0 on graphs) of proline pre-feeding, surviving flies from each genotype were randomly allocated to experimental and control groups: ¹²C Glucose-injected groups (Proline + Glucose): For each genotype, three independent biological replicates were set up, each consisting of 20–25 flies (n = 3 vials/genotype). Flies were kept briefly on ice and injected with a ¹²C-glucose solution as described in the “Glucose injection and metabolic labeling” section. Following injection, flies were immediately returned to fresh 30 and 15 mM proline food and maintained under standard conditions. Non-injected control groups (Proline only): In parallel, for each genotype, three independent biological replicates of 20–25 flies each (n = 3 vials/genotype) were handled identically (including ice exposure, if matched to injected groups), but were not injected. These flies were also maintained on 30 and 15 mM proline food for the duration of the assay. After the day 6 intervention (glucose injection or control handling), survival was monitored daily for at least 3 days. Dead flies were counted and removed at the same circadian time each day, and the number of surviving flies per vial was recorded. Flies were maintained on 30 and 15 mM proline food throughout this period, with food changed if necessary to prevent desiccation or mold. For each genotype and condition, survival was summarized as the mean ± SEM across the three biological replicates. Genotype- and treatment-dependent effects on survival at each time point (e.g., days 2 and 3 post-injection) were assessed using appropriate statistical tests (ANOVA Sidak’s multiple comparison test).

All relevant data are within the paper and its Supporting information files. Raw data are uploaded in metabolomics workbench (https://www.metabolomicsworkbench.org/↗) under data track IDs ST004644 (human LC–MS/MS data), ST004674 (glycogen DESI-MS assay data), ST004676, ST004677 (WT and fmn time course LC–MS data, negative and positive mode data, respectively), ST004685, ST004686 (WT Fasting time course LC–MS negative and positive mode data respectively), ST004687 (Time course with per01 mutant, both modes). All data is publicly available at metabolomics workbench.