Internal Circadian Misalignment of the Human Metabolome Links Night Shiftwork to Metabolic Impairment

Abstract

Circadian misalignment, as experienced during shiftwork, impairs glucose metabolism and body weight regulation, yet the underlying biochemical mechanisms remain incompletely understood. Characterizing how circadian misalignment alters circulating metabolites provides a promising avenue to help identify these mechanisms. Although data from metabolomics studies have identified circulating metabolites with daily rhythms, it is not comprehensively known which rhythms shift during circadian misalignment and whether such shifts relate to metabolic impairment. We conducted 24-hour (h) metabolomic profiling every 4 h in 14 healthy adults (8 women) aged 26.4 ± 1.2 years (mean ± SD), undergoing a 6-day simulated night-shiftwork protocol. 24-h modeling analyses identified metabolite rhythms influenced by circadian versus behavioral cycles (sleep, food intake) and quantified internal circadian misalignment using acrophase shifts. Metabolic outcomes included glucose homeostasis (test meals) and energy expenditure (EE; whole-room calorimetry). Night-shiftwork produced widespread alterations in metabolite rhythms, with significant internal misalignment in multiple metabolites across pathways including pyrimidine metabolism, bile acid–microbiome signaling, and lipid metabolism. During misalignment, glucose and insulin area under the curve increased (p < 0.05) and EE decreased (p < 0.05). Internal misalignment of uridine and glycoursodeoxycholic acid was associated (p < 0.05) with impaired glucose tolerance, while their circulating concentrations were associated with decreased EE. Misalignment of uridine and glycoursodeoxycholic acid suggests dysregulated pyrimidine and bile acid–microbiome pathways as potential mechanisms linking circadian misalignment to cardiometabolic disease risk.

Introduction

Shiftworkers represent ~20% of the workforce in industrialized countries and have a higher risk of cardiometabolic disease versus non-shiftworkers (Alterman et al., 2013; Hemmer et al., 2021). Although incompletely understood, circadian misalignment is a leading mechanism hypothesized to contribute to the adverse cardiometabolic risk of shiftwork (Wittmann et al., 2006). Supporting this hypothesis, experimentally imposed circadian misalignment in humans rapidly impairs glucose metabolism and decreases energy expenditure (Scheer et al., 2009; McHill et al., 2014; Grant et al., 2017; Centofanti et al., 2025), yet the biochemical mechanisms underlying these adverse cardiometabolic health outcomes remain incompletely understood.

The circadian system times human physiology to anticipate environmental rhythms driven by Earth’s ~24-h rotation (Klerman et al., 2022). Humans have a central circadian clock in the suprachiasmatic nucleus (SCN) of the hypothalamus, and local clocks in nearly every tissue, referred to as peripheral clocks relative to the SCN. The central clock is most strongly entrained by light, whereas clocks in metabolic tissues are strongly entrained by behavioral cycles such as the timing of food intake, sleep, and physical activity (Damiola et al., 2000; Stokkan et al., 2001; Schibler et al., 2003; Davidson et al., 2009). Circadian misalignment occurs when environmental (e.g. light-dark) or behavioral (e.g. sleep-wakefulness; fasting-eating) cycles become misaligned from the central clock or when the central and peripheral clocks become misaligned with each other, termed “internal misalignment” (Klerman et al., 2022). For example, in mice, feeding at the wrong circadian time misaligns metabolic tissue clocks (e.g. liver and pancreas) from the SCN, producing internal misalignment that impairs glucose metabolism and promotes weight gain (Hampton et al., 1996; Damiola et al., 2000; Arble et al., 2009). Similarly, night shiftworkers commonly adjust their sleep and meal timing to align with work schedules—sleeping when the central clock promotes wakefulness and eating when it promotes fasting. As such, they likely experience similar internal misalignment. Yet, quantifying internal misalignment in humans remains a methodological challenge, limiting our mechanistic understanding of the adverse cardiometabolic risk of shiftwork.

One strategy to study peripheral rhythms in humans is to quantify core clock gene expression in peripheral tissues, using serial biopsy sampling (Wehrens et al., 2017). However, serial biopsies are invasive, technically challenging, and not feasible for key metabolic tissues including liver and pancreas. An emerging alternate is to quantify internal misalignment using circulating molecules (e.g. metabolites and proteins) with 24-h time-of-day patterns, as these may serve as biomarkers of peripheral rhythms (Depner et al., 2018; Skene et al., 2018; Grant et al., 2019; Kervezee et al., 2019; Kent et al., 2022; McDermott et al., 2024; McHill et al., 2024). For example, targeted mass spectrometry data, quantifying hundreds of metabolites, show 24-h time-of-day patterns in ~50% of the analyzed metabolites, with ~75%–95% of those patterns behavioral cycle-influenced (Skene et al., 2018; Kervezee et al., 2019). These findings support the hypothesis that behavioral cycle-influenced metabolites could serve as biomarkers for quantifying internal circadian misalignment in humans.

Targeted assays have the advantage of absolute quantification and enable a priori interrogation of biochemical pathways of interest. In contrast, untargeted metabolomics has the advantage of quantifying thousands of metabolites, offering broader discovery potential. Regardless of targeted or untargeted analyses, there is limited evidence linking internal misalignment of behavioral-influenced metabolites to metabolic impairments. For example, shifting food intake to the biological night shifts the timing of 24-h blood glucose patterns, resulting in internal misalignment between circulating glucose and the endogenous circadian phase marker of core body temperature minimum (Chellappa et al., 2021). Furthermore, this internal misalignment of the circulating glucose rhythm was associated with impaired glucose tolerance (Chellappa et al., 2021). Yet, such results are limited to analyses of circulating glucose rhythms. These emerging data highlight untargeted metabolomics as a powerful approach to characterize how circadian versus behavioral cycles regulate the human metabolome and to identify pathways through which internal misalignment impairs metabolic health.

Here, we tested the hypothesis that circadian misalignment alters biochemical pathways that directly relate to metabolic impairment. Using untargeted metabolomics sampled every 4 h during a simulated night-shiftwork protocol, we aimed to identify metabolites whose internal circadian misalignment, defined as acrophase shifts relative to the central circadian clock, is associated with impaired glucose and energy metabolism.

Materials and Methods

Human Participants

The scientific and advisory review committee of the Colorado Clinical and Translational Sciences Institute and the Colorado Multiple Institutional Review Board approved the study. All participants provided written informed consent. Data were collected in a previously described study (Figure 1) (McHill et al., 2014). Fourteen participants (8 women) aged 26.4 ± 1.2 years (mean ± SD), body mass index (BMI) 22.7 ± 0.5 kg/m², and percent body fat 27.4% ± 2.2% (dual-energy X-ray absorptiometry; Prodigy Advance; GE Lunar) completed the study. Exclusion criteria were medical, psychiatric, or sleep disorders, current smoking, working night shiftwork in the preceding year, pregnancy, and self-reported habitual sleep duration of <7 h or >9.25 h per night. All participants completed medical, psychological, sleep history, and overnight polysomnographic sleep disorders screening (McHill et al., 2014). Participants were not allowed to cross more than one time zone in the 3 weeks prior to the study.

Figure 1.

Modified constant routine protocol. White bar represents room light at <40 lux. Black bars represent scheduled sleep during circadian alignment. Red bars represent scheduled sleep during circadian misalignment. Gray bars represent scheduled wakefulness in dim light <10 lux. Red dots indicate blood draws for analyses of melatonin and untargeted metabolomics. “B” breakfast, “L” lunch, “D” dinner, “S” snack. The x-axis represents relative clock hour with scheduled waketime arbitrarily assigned a value of 0800 h. All other times are referenced to this value.

Study Design

Pre-Laboratory Ambulatory Monitoring.

Participants maintained a self-selected habitual sleep schedule of ~8 h per night for 1 week before admission to the University of Colorado Hospital Clinical Translational Research Center. Wrist actigraphy (Actiwatch-L; Mini-Mitter/Respironics), sleep-wake diaries, and daily call-in sleep and waking times via time-stamped voice recorder verified sleep. Caffeine, alcohol, nicotine, and over-the-counter medications were proscribed 1 week before laboratory admission. Urine toxicology and breath alcohol testing (Lifeloc Technologies; model FC10) confirmed drug-free status. Three days before admission, exercise was proscribed and participants consumed outpatient diets designed to meet daily caloric needs (resting metabolic rate times a 1.5 activity factor) (McHill et al., 2014).

Laboratory Protocol.

The simulated night shiftwork protocol (Figure 1) had 2 days of habitual nighttime sleep (circadian alignment, sleep opportunity based on participant’s habitual sleep schedule), 1 day of transitioning to shiftwork, and 1 day of simulated night shiftwork with daytime sleep (circadian misalignment. The parent study (McHill et al., 2014) consisted of 2 days of simulated night shiftwork followed by recovery sleep before laboratory discharge. Timing of all protocol events was determined by individual habitual sleep-wake schedules during ambulatory monitoring, and timing of protocol events was shifted by 8 h for the simulated night shiftwork protocol. Participants followed a modified constant routine on days 2–4 as previously described (McHill et al., 2014). Continuous monitoring including daytime and nighttime electroencephalography (EEG) verified scheduled wakefulness. Participants received scheduled meals (percent daily caloric intake: 30% breakfast, 30% lunch, 30% dinner, 10% snack) at ~1.5, 5.5, 10.5, and 14.5 h after waketime. Thus, the timing of food intake shifted in parallel with the timing of sleep. During study day 3, percent daily caloric intake and timing were adjusted to accommodate the nap opportunity (30% breakfast, 25% lunch, 10% snack, 25% dinner, 10% snack) with meals provided at 1.5, 5.5, 10.5, 15.5, and 20.5 h after scheduled waketime. Total calories and macronutrient composition were similar across study conditions.

Breakfast Test Meals.

Participants consumed identical breakfast test meals on study days 2 (circadian alignment) and 4 (circadian misalignment) at 1.5 h after the scheduled waketime, following a >10-h fast. Meals were ~55% carbohydrate, 30% fat, and 15% protein. Participants had 30 min to consume meals. Glucose (Beckman Coulter) and insulin (Beckman Coulter) were analyzed in blood samples collected at −30, 30, 60, 90, and 210 min from the start of each meal (Depner et al., 2018). Trapezoidal area under the curve (AUC) was analyzed using OriginPro (v 92E; OriginLab Corporation).

Melatonin.

Blood was collected hourly during days 2–5 for melatonin analyses. A three-harmonic fit to 24-h melatonin profiles determined peak-to-trough amplitude for each participant (Wright et al., 2001). Dim-light melatonin onset (DLMO) was calculated for each study condition for each participant as the linear interpolated point in time when melatonin increased above 25% of the calculated peak-to-trough amplitude.

Energy Expenditure.

Daily energy expenditure was assessed continuously by whole-room indirect calorimetry across study days 2–5 as described previously (McHill et al., 2014).

Metabolomics.

Blood samples for untargeted metabolomics were collected at 1, 5, 9, 13, 17, and 21 h after habitual waketime on study day 2 for circadian alignment and study day 4 for circadian misalignment (Figure 1). Samples were prepared using previously described methods, including liquid chromatography, mass spectrometry, and data extraction (Cruickshank-Quinn et al., 2014; Depner et al., 2020). Detailed methods are in Supplementary Materials. Briefly, blood samples were collected into ethylenediaminetetraacetic (EDTA) tubes, centrifuged, aliquoted, and stored at −80C. Pooled aliquots from each participant served as sample prep and instrument quality control (QC) samples, prepared and analyzed together with experimental samples. Liquid-liquid extraction using methyl tert-butyl ether (MTBE) separated the samples into hydrophobic (lipid-rich) and hydrophilic (aqueous) fractions. Sample preparation order was randomized with samples run on LC/MS (Agilent 6545 Q-TOF) in re-randomized order. MS/MS generated additional annotation information for compounds with significant 24-h time-of-day patterns during circadian alignment and misalignment. The LC/MS chromatographic method was replicated for LC–MS/MS with 10, 20, and 40 eV collision energies on an Agilent 6545 Q-TOF, a scan rate of 2 spectra/s, 1.3 m/z isolation width, and a 30-s retention time window. Fragmentation data were matched to standards from the NIST 14, METLIN, and in-house MS/MS spectral libraries. Based on MS and MS/MS data acquisition, 21 potential duplicate compounds were removed from analyses.

Statistical Analyses

Metabolomics Data Processing.

Metabolomics data were filtered by removing compounds with missing data, so only compounds detected in all samples were analyzed. Assessment for possible batch effects was conducted by principal components analysis (PCA) and by comparing relative standard deviation of experimental samples versus QCs (Han and Li, 2022) (Supplemental Figures S4 and S5). Major batch effects were not detected. Potential outliers were identified by PCA and metabolite z-scores (Supplemental Figure S4). Six samples from one participant during circadian alignment were identified as outliers and removed (Supplemental Figure S4). Metabolomics data were Log2 transformed and converted into difference from the mean (DFM) for analyses of 24-h time-of-day patterns, similar to our prior research (Depner et al., 2018; Depner et al., 2020; Cogswell et al., 2021; Gombert et al., 2023; McHill et al., 2024).

Change in 24-h Average Abundance.

Linear mixed-effects regression identified compounds with increased or decreased abundance during circadian misalignment versus alignment, using the R package lme4 (v1.1–33), with condition (circadian alignment versus misalignment) as a fixed effect and participant as a random effect. p-values were adjusted for multiple testing via the Benjamini-Hochberg FDR method (Benjamini, 1995) and statistical significance set at the 5% (<0.05) FDR level.

24-h Time-of-Day Patterns.

Because not all biological rhythms conform to a single rhythmic model, we used multiple complementary approaches to detect 24-h time-of-day patterns, consistent with recommendations in the circadian field (Deckard et al., 2013). First, we applied MetaCycle (Wu et al., 2016) (v1.2.0) in R to identify 24-h time-of-day patterns separately for circadian alignment and misalignment conditions. Specifically, we used the “meta2d” function with “JTK” and “LS” methods, as previously described (Depner et al., 2020; McHill et al., 2024). Second, we also fit a cosine model (equation (1)) using lme4 to detect 24-h time-of-day patterns separately for circadian alignment and misalignment conditions

$$Y=M+\beta 1\cos \left(\frac{2\pi t}{24}\right)+\beta 2sin\left(\frac{2\pi t}{24}\right)$$ (1)

In equation (1), participant is a random factor, $M$ is the intercept, $\beta 1$ and $\beta 2$ are cosinor coefficients, and t is the time of sample collection in h from habitual waketime. For each condition, model fit was compared against a null model ($\beta 1$ and $\beta 2=0$) via the likelihood ratio test. p-values from MetaCycle and cosine analyses were combined using the minP method (Tseng et al., 2012). Statistical significance to identify 24-h time-of-day patterns was set at the 5% FDR level.

Modeling Approach #1: Dichotomous Circadian Versus Behavioral Cycle Regulation.

The slam-shift feature of our protocol, conducted in dim-light conditions, allowed us to distinguish whether 24 h time-of-day patterns of compounds were influenced by the circadian or behavioral cycle. Because DLMO did not change between circadian alignment and misalignment conditions, compounds with a stable acrophase across conditions were likely circadian cycle-influenced. In contrast, compounds with significant shifts in acrophase between conditions were likely behavioral cycle-influenced. To determine these classifications, we focused on compounds with significant 24-h time-of-day patterns during both circadian alignment and misalignment, applying a dichotomous model selection approach (Kervezee et al., 2019; McHill et al., 2024). First, two independent linear mixed-effects models, a circadian and a behavioral model, were separately fit and analyzed. For the circadian model, equation (1) was fit to all data with t representing sampling time as relative clock hour. As noted, because DLMO did not change between study conditions, the circadian model analysis was aligned with endogenous circadian phase, independent of the shifted sleep and food intake timing during circadian misalignment. Model significance was determined by comparing the fit to a null model via the likelihood ratio test. Bayesian Information Criterion (BIC), a measure of model fit, was computed. For the behavioral model, equation (1) was also fit to all data with t representing sampling time in hours after experimentally imposed lights on as a measure of behavioral time. Thus, the behavioral model analysis accounted for the shifted sleep and food intake timing during circadian misalignment. Significance was assessed via the likelihood ratio test, and BIC was calculated. p-values were corrected for multiple testing with the Benjamini-Hochberg FDR method. Next, compounds were classified as circadian-influenced if the circadian model FDR was < 0.05 and BIC was less than the behavioral model BIC. Alternatively, compounds were classified as behavioral-influenced if the behavioral model FDR was <0.05 and BIC was less than the circadian model BIC. If neither criteria were met, the 24-h time-of-day pattern was classified as neither strongly circadian nor behavioral-influenced. At the group level, compound acrophase was calculated using cosinor coefficients (Cornelissen, 2014). t tests were used to analyze shifts in acrophase between circadian misalignment and alignment for the behavioral- versus circadian-influenced compounds.

Modeling Approach #2: Estimated Simultaneous Contribution of Circadian and Behavioral Cycles.

In our modified constant routine protocol, behavioral cycles were always present throughout the study. As a result, compound abundance at any given time point could reflect the combined influence of both circadian and behavioral cycles, rather than exclusively one or the other as in modeling approach #1. Some compounds may gain or lose a significant 24-h time-of-day pattern—or may not exhibit one at all—due to the joint influence of both cycles. To quantify these joint effects and the range of influence from circadian to behavioral, we modeled compound time series as a linear combination of independently derived circadian and behavioral components, similar to prior analyses in this field (Archer et al., 2014). Specifically, we fit a linear model using the “lm” function in R (equation (2))

$$X_{mdt}=\text{β}1C_{mdt}+\text{β}2B_{mdt}$$ (2)

In equation (2), $X_{mdt}$ is the median abundance of compound $m$ on study day $d$ at time-point $t$, $C_{mdt}$ is the fitted median abundance of compound $m$ on study day $d$ at time-point $t$ from the circadian model, and $B_{mdt}$ is the fitted median abundance of compound $m$ on study $d$ at time-point $t$ from the behavioral model. Median abundance was calculated across all participants, per study day, per sampling time point, using z-scored time-series data to produce standardized β estimates. Model fit was assessed using $R^2$. This approach yields a continuous estimate of the relative contribution of circadian $\left({\text{β}}_1\right)$ and behavioral $\left({\text{β}}_2\right)$ influences for each metabolite, as well as the overall variance explained by their combined effects. Similar to prior analyses (Archer et al., 2014), we focused on compounds with substantial model fit (overall R² ≥ 0.60).

To aid interpretation, we additionally classified these compounds into five groups based on (1) whether the circadian and/or behavioral cycles significantly contributed to the model (FDR-corrected p-values); and (2) the directionality (positive or negative) of the standardized $\left({\text{β}}_1\right)$ and $\left({\text{β}}_2\right)$ coefficients. Grouping criteria were used to describe whether compound concentrations increased or decreased during wakefulness, whether concentrations were high or low during the circadian biological night, and whether model fit was significant for the circadian cycle, behavioral cycle, or both. Importantly, this categorical grouping is intended as a simplified summary of the underlying continuous distribution of circadian–behavioral cycle contributions, which we also present (Figure 4b).

Figure 4.

Change in compound acrophase timing between circadian alignment and misalignment. The shift in acrophase between conditions is represented as the absolute value of study day 4 minus study 2 in hours. Compounds on y-axis are ordered sequentially from the smallest to largest shift in acrophase. Error bars represent 95% CIs. (a) Colored by groups derived from linear models described in the section “Modeling Approach #1: Dichotomous Circadian Versus Behavioral Cycle Regulation.” (b) Colored by groups derived from linear models described in the section “Modeling Approach #2: Estimated Simultaneous Contribution of Circadian and Behavioral Cycles.”

Association Between Internal Circadian Misalignment and Dysregulated Glucose Metabolism.

Differences in DLMO between study conditions were analyzed using paired t tests. To quantify the magnitude of internal circadian misalignment of all circadian and behavioral cycle-influenced compounds during both study conditions, we calculated the phase relationship between DLMO and compound acrophase for circadian alignment and misalignment days separately (defined as time of DLMO minus time of acrophase). Five participants had missing DLMO values on study day 4 (circadian misalignment) due to missing samples. Day 3 DLMO values were used for three of these five participants. Linear mixed-model analyses using lme4 tested associations between compound phase relationships and glucose and insulin AUC from the breakfast test meal. Insulin AUC data for one participant was missing due to missed sample collection. p-values were adjusted for multiple testing via the Benjamini-Hochberg FDR method and statistical significance was set at the 5% FDR level.

Association Between Energy Expenditure and Metabolite Concentrations.

As an exploratory analysis, we examined compounds potentially linked to both glucose and energy expenditure. Because energy expenditure was quantified as 24 h profiles in 1-h bins, and our prior analyses showed the most prominent changes specifically during sleep opportunities (McHill et al., 2014), we applied linear mixed-model analyses using lme4 to test associations between hourly energy expenditure and abundance of the seven compounds associated with glucose AUC. Hourly energy expenditure was derived from the corresponding hour for each metabolomics blood-collection time point. Study condition and time were entered as fixed factors, and participant was entered as a random factor. To further characterize the uridine, uracil, and glycoursodeoxycholic acid (GUDCA) associations with energy expenditure, a one-tailed t test assessed differences in compound abundance between timepoint 21 (nighttime sleep) during circadian alignment and timepoint 5 (daytime sleep) during circadian misalignment.

Exploratory Pathway Analyses.

Human Metabolome Database (Wishart et al., 2018) (HMDB, v4.0) IDs for compounds of interest were loaded into the Network Explorer module in Metaboanalyst (Chong et al., 2018). This matched compounds to a gene-metabolite interaction network in the Reactome (Fabregat et al., 2017) pathway database.

Results

Participants

Fourteen adults (eight female, six male) with no history of working shiftwork for 1 year prior to study enrollment, aged 26.4 ± 1.2 years (mean ± SD), body mass index (BMI) 22.7 ± 0.5 kg/m², and percent body fat 27.4 ± 2.2% completed the study.

Sleep and Circadian Timing

Sleep data were reported previously (McHill et al., 2014). Briefly, total sleep time decreased (p < 0.05) from 7.23 ± 0.16 h (mean ± SEM) during circadian alignment (study day 2) to 6.32 ± 0.20 h during circadian misalignment (study day 4). DLMO was similar across study days 2, 3, and 4 (all p ≥ 0.16; Table I). Thus, during circadian misalignment, participants slept during the biological day when melatonin levels were low, and remained awake during the biological night when melatonin levels were high.

Table I.

Dim-light Melatonin Onset (DLMO).

Study condition	DLMO (Relative clock time)
Study day 2: Circadian Alignment	22.46 (0.28)
Study day 3: Transition Day	22.39 (0.49)
Study day 4: Circadian Misalignment	23.13 (0.38)

Data are mean ± SEM, n = 13.

Change in 24-h Average Abundance of Compounds

Our final data set had 2903 compounds after filtering. Circadian misalignment altered overall abundance (false discovery rate [FDR] < 0.05) for 157 compounds with 134 decreased and 23 increased (Supplemental Figure S1A). Exploratory pathway analyses using all compounds with altered abundance revealed the top three pathway hits (ranked by FDR) were related to respiratory electron transport, with additional notable hits linked to amino acid and pyrimidine metabolism. Supplemental Table S1 represents the top 10 pathway hits using the compounds with change in 24 h abundance.

24-h Time-of-Day Patterns

We identified 320 compounds (11%) with significant 24-h time-of-day patterns (FDR < 0.05) during at least one study condition. Among these, 41 were only significant during circadian alignment, 161 only during circadian misalignment, and 118 during both circadian alignment and misalignment (Supplemental Figure S1B and Table S2), illustrating substantial temporal reorganization of the circulating metabolome during circadian misalignment.

Circadian Versus Behavioral Cycle-Influenced Compounds (Modeling Approach #1)

Initial analyses focused on the 118 compounds with 24-h time-of-day patterns during both conditions. Among these, 44 were classified as circadian-influenced, 70 as behavioral-influenced, and 4 did not meet criteria for either (Supplemental Table S3). Representative 24-h profiles are shown in Figure 2 and Supplemental Figure S2. The 44 circadian-influenced compounds displayed a single tight cluster of acrophases near habitual waketime, with minimal shifts between study conditions. Specifically, during circadian alignment, 31 of 44 circadian-influenced compounds peaked between 0700 and 1200 (Figure 3a), with a similar distribution during circadian misalignment, where 32 of 44 compounds peaked between 0700 and 1300 (Figure 3b). In contrast, behavioral-influenced compounds showed marked reorganization. During circadian alignment, most (37 of 70) peaked between 0400 and 0900 (Figure 3c). However, during circadian misalignment, their acrophases split into two distinct clusters, one cluster of 17 compounds peaked between 0100 and 0500, whereas a second cluster of 37 compounds peaked between 1200 and 1600 (Figure 3d). Consistent with these observations, acrophase shifts between conditions were significantly larger (p < 0.001) for behavioral-influenced (7.13 ± 2.08 h; mean ± SD) versus circadian-influenced (2.52 ± 1.67 h) compounds (Figure 3e and f).

Figure 2.

Example compounds influenced by the circadian (a-c) and behavioral (d and e) cycles (n = 14). (a) Unknown, highlighted in Supplemental Table S3; (b) Decanoylcarnitine; (c) Creatine; (d) Isobutyryl-L-Carnitine; and (e) Uridine. Bold lines represent fitted cosinor model data. Fine lines represent individual participant data. DFM, difference from the mean. Black and red boxes represent scheduled sleep opportunities for circadian alignment and misalignment, respectively.

Figure 3.

Acrophase timing during circadian alignment and misalignment (n = 14). (a–d) Radial plots showing phase distribution of circadian compounds during (a) circadian alignment and (b) misalignment, and behavioral compounds during (c) circadian alignment and (d) misalignment. Data represent number of compounds with an acrophase at a given relative clock hour. Gray and red shading represent sleep opportunities. Radial axes represent number of compounds and range from 0 to 24 with 6, 12, and 18 indicated by black concentric circles. (e) Acrophase timing of 24-h time-of-day patterns during circadian alignment (day 2) and circadian misalignment (day 4) for compounds classified as behavioral or circadian-influenced. Pink Square: timing of DLMO. The dashed diagonal lines indicate where compounds would fall if the timing of their acrophase was unchanged between circadian misalignment and alignment. (f) Absolute magnitude of change in acrophase timing (phase shift) between circadian misalignment and alignment. *p < 0.001 for behavioral cycle-influenced compounds versus circadian cycle-influenced compounds.

Although circadian- and behavioral-influenced compounds showed different acrophase shifts, sequential ordering of absolute acrophase shifts revealed a continuous distribution with an inflection point near ~4 h separating most circadian- from behavioral-influenced compounds (Figure 4a). Notably, three compounds classified as circadian-influenced showed acrophase shifts greater than 4 h, suggesting this dichotomous modeling approach likely resulted in a misclassification of these compounds. Just four circadian-influenced compounds (Gamma-Glutamyl Glutamine, L-alpha-Amino-1 H-pyrrole-1-hexanoic acid, and two compounds with unknown annotations, one of which is the compound in Figure 2a) showed 95% confidence intervals overlapping with 0 h (indicating no difference between conditions), and nine compounds showed 95% confidence intervals overlapping with the 9 h experimentally imposed shift in the behavioral cycle. In addition, 10 compounds showed 95% confidence intervals with shifts greater than 9 h. These observations suggest a complex regulatory pattern, beyond a simple dichotomy between circadian versus behavioral cycle-influence. Instead, many compounds were likely influenced by the simultaneous combination of the circadian and behavioral cycles to varying degrees, especially because the behavioral cycle was present in both conditions during sample collection in our protocol.

Simultaneous Influence of Circadian and Behavioral Cycles (Modeling Approach #2)

To estimate the simultaneous contribution of the circadian and behavioral cycles, from predominantly circadian- to predominantly behavioral-influenced, we applied a second modeling approach similar to prior work (Archer et al., 2014). As compounds lacking 24 h time-of-day patterns in one or both conditions could be jointly influenced by circadian and behavioral cycles, we included all 2903 compounds in this analysis—rather than limiting analyses to only the 118 compounds with 24 h time-of-day patterns in both conditions. Briefly, for each compound, this second approach modeled the linear combination of the fitted data derived from the individual circadian and behavioral models described in the prior section. This allowed us to estimate the total variance (R²) explained by the combined influence of the circadian and behavioral cycles, along with estimates of the individual contribution from each cycle (Figure 5a and b). Based on an overall R² ≥ 0.60, along with statistical significance and direction of the β estimates of the circadian and behavioral cycle models, we identified five distinct groups, consisting of 216 total compounds (Figure 5c and Supplemental Table S4). Groups 1–3 were significant (FDR < 0.05) for both the circadian and behavioral cycle components, suggesting joint influence by both cycles. Specifically, group 1 (one compound) showed decreasing concentrations over hours awake with higher concentrations during the circadian biological night (Figure 6a). Group 2 (two compounds) showed increasing concentrations over hours awake with lower concentrations during the circadian biological night (Figure 6b). Group 3 (48 compounds) showed increasing concentrations over hours awake with higher concentrations during the circadian biological night (Figure 6c). In contrast, group 4 (99 compounds) was only significant (FDR < 0.05) for the behavioral cycle with concentrations increasing over hours awake (Figure 6d). Group 5 (66 compounds) was only significant (FDR < 0.05) for the circadian cycle with the highest concentrations during the circadian biological night (Figure 6e). Consistent with the hypothesized joint influence of the circadian and behavioral cycles, the acrophase shifts between conditions were generally lowest in group 5, the only group exclusively influenced by the circadian cycle, with a mixed distribution of acrophase shifts across the other four groups (Figure 4b). These findings represent a gradual transition from predominantly circadian-influenced to predominantly behavioral-influenced compounds.

Figure 5.

Modeling the simultaneous influence of behavioral and circadian cycles for all 2903 compounds. Each compound was analyzed by fitting the linear combination of the behavioral and circadian models for each timepoint, using z-scored data to produce standardized β estimates. (a and b) Two rotational views of a 3D scatter plot representing the overall model fit (R², y-axis), the standardized β estimate for behavioral cycle-influence (x-axis), and the standardized β estimate for circadian cycle-influence (z-axis). Each data point represents a single compound. The large pink data point represents uridine (strongly behavioral influenced) for reference. Shading of data points reflects R² values. (c) Representation of the five groups, based on overall R² ≥ 0.60, statistical significance for the behavioral and circadian cycles (FDR < 0.05), and the direction of the standardized β estimate from each cycle (x- and y-axes represent the standardized β estimates for behavioral and circadian cycles, respectively). Groups 1–3 were significant for behavioral and circadian cycles. Group 4 was only significant for the behavioral cycle. Group 5 was only significant for the circadian cycle. NS, not significant.

Figure 6.

Modeled contributions of the simultaneous behavioral and circadian cycle-influence for the 5 groups (a-e, respectively) presented in Figure 5 C. Each panel represents the median of the modeled z-score concentrations of all compounds in the group, at each combination of circadian phase and hours since waketime, as derived from linear models described in the section “Modeling Approach #2: Estimated Simultaneous Contribution of Circadian and Behavioral Cycles.” Circadian clock time is double plotted for graphical purposes only to help illustrate the magnitude of circadian cycle-influence. Hours since waketime is plotted with 0 representing waketime and the black shading representing scheduled 8-h sleep opportunities from hours 16 to 24 after waketime. Data are plotted in 2 h bins.

We conducted exploratory pathway analyses to help identify potential biochemical functions associated with each group. All compounds in groups 1 and 2 had unknown annotations (Supplemental Table S4) and therefore we were unable to conduct further biochemical functional analyses of these groups. For group 3 (circadian and behavioral cycle-influence), top pathway hits were related to G protein-coupled receptor signaling (the largest class of drug targets), serotonin and melatonin biosynthesis, and catecholamine biosynthesis. For group 4 (behavioral cycle-influence), top pathway hits included pyrimidine metabolism, heme degradation, and pyrimidine and purine catabolism. For group 5 (circadian cycle-influence), top pathway hits included 4 pathways linked to phospholipid metabolism and 4 pathways linked to lipoprotein metabolism. The top 10 pathway hits for groups 3, 4, and 5 are presented in Supplemental Table S5.

Finally, we compared results from modeling approach #1—which modeled the circadian versus behavioral cycles as separate, dichotomous influences—against modeling approach #2 that modeled the simultaneous circadian and behavioral cycle-influence and used the overall model fit threshold of R² ≥ 0.60 (Supplemental Table S4). Among the 44 circadian-influenced compounds from modeling approach #1, eight were in group 5, consistently showing strong influence from the circadian cycle, and two were in groups with behavioral-influence (groups 3 and 4). Thus, the majority (34 out of 44; ~77%) of circadian-influenced compounds from modeling approach #1 did not meet the R² ≥ 0.60 threshold for overall model fit used in modeling approach #2. Alternately, among the 70 behavioral-influenced compounds in the initial analysis, 35 compounds were in group 4 and 6 compounds were in group 3 consistently showing strong influence from the behavioral cycle. Thus, relative to the circadian-influenced compounds, a larger proportion of behavioral-influenced compounds (41 out of 70; ~59%) showed concordance between modeling approaches #1 and #2. Combined, both modeling approaches identified a total of 279 compounds with some influence by the circadian or behavioral cycles, or a combination of both. In aggregate, Supplemental Tables S3 and S4 list each of these 279 compounds, with the noted overlapping compounds listed in both tables.

Compounds Associated with Glucose and Insulin Responses to Breakfast Test Meals

Fasting glucose and insulin were similar during circadian alignment and misalignment, similar to our previously reported data in a male-only subset of the current participants (Depner et al., 2018). However, glucose and insulin AUC increased (p < 0.05) during circadian misalignment versus circadian alignment (Figure 7a and b). We next conducted association analyses between compound phase relationships (duration in hours between acrophase and DLMO) and glucose and insulin AUC, using all 279 circadian and behaviorinfluenced compounds. Seven compounds had phase relationships associated (FDR < 0.05) with glucose AUC (Figure 7c-f and Figure S3A-D)—uracil, uridine, linoleyl carnitine, GUDCA, pipecolinic acid, 1-methylinosine, and one unannotated compound. All of these compounds consistently showed behavioral cycle-influence using both modeling approaches, except pipecolinic acid that was classified as behavioral-influenced using modeling approach #1 and had R² = 0.59 in modeling approach #2, just under the predefined R² ≥ 0.60 threshold. A single unannotated compound had a phase relationship associated (FDR < 0.05) with insulin AUC (Supplemental Figure S3E).

Figure 7.

Effects of circadian misalignment on (a) plasma glucose and (b) insulin AUC in response to a breakfast test meal, and (c-e) compound phase angles associated with glucose AUC (n = 14). (c) uridine, (d) uracil, and (e) glycoursodeoxycholic acid (n = 11). Data in panels a and b are mean ± SEM. White rectangles represent the 30 minute time allocated to ingest the breakfast test meal. AUC, area under the curve. DLMO, dim-light melatonin onset. *p < 0.05 for circadian misalignment versus alignment.

Compounds Associated With Energy Expenditure

Among compounds with phase relationships associated with glucose AUC, uridine, a pyrimidine nucleoside composed of uracil and ribose, has roles in DNA and RNA synthesis and glycogen and lipid metabolism, and is also linked to regulating energy expenditure and food intake in rodents and humans. This is hypothesized to occur through concentrations rising during fasting and decreasing following food intake (Steculorum et al., 2015; Deng et al., 2017; Steculorum et al., 2017; Hanssen et al., 2023; Felix et al., 2025). Because of these properties and because our participants lived in a whole-room calorimeter, we explored associations between uridine and energy expenditure. As previously published (McHill et al., 2014), hourly energy expenditure decreased (p = 0.01) during circadian misalignment (5.59 ± 0.21 kJ; mean ± SEM) versus circadian alignment (6.13 ± 0.18 kJ), primarily due to lower energy expenditure during the daytime sleep of circadian misalignment. Uridine concentrations were associated (p < 0.001) with hourly energy expenditure across both conditions (Figure 8a). We explored this association for the other six compounds associated with glucose AUC and found uracil and GUDCA were also associated (p < 0.001) with hourly energy expenditure (Figure 8b and c). Moreover, during daytime sleep in the misalignment condition versus nighttime sleep in the aligned condition, uridine and uracil concentrations were higher (p < 0.05) and GUDCA was lower (p < 0.01), paralleling changes in energy expenditure and reinforcing their categorization as behavioral-cycle regulated.

Figure 8.

Associations between compound abundance and average hourly energy expenditure across all metabolomics blood collection time-points in both conditions (n = 14). Significant (p < 0.001) associations between abundance of (a) uridine, (b) uracil, and (c) glycoursodeoxycholic acid and average hourly energy expenditure are shown. EE, energy expenditure.

Discussion

We characterized the impact of circadian misalignment on the human plasma metabolome during a simulated night shiftwork protocol. Across 279 compounds influenced by the circadian cycle, behavioral cycle, or both, the majority (179; ~64%) were predominantly behavioral-influenced. Notably, only three behavioral-influenced compounds, uridine, uracil, and GUDCA, were associated with changes in both glucose metabolism and energy expenditure. Uridine is a pyrimidine nucleoside composed of uracil and ribose, and thus our findings raise the hypothesis that dysregulated pyrimidine metabolism is a potential contributor to the adverse cardiometabolic risk of night shiftwork. GUDCA is a glycine conjugated form of ursodeoxycholic acid, a secondary bile acid produced by gut bacteria, and is lower in people with elevated cardiometabolic risk (Cheng et al., 2021; Huang et al., 2021; Fleishman and Kumar, 2024). Thus, our findings also raise the hypothesis that potential changes in bile acid metabolism and the gut microbiome may contribute to the adverse cardiometabolic risk of night shiftwork. Collectively, our findings illustrate that circadian misalignment exerts broad effects on the human blood metabolome and support the possibility of leveraging circulating metabolites as potential biomarkers of peripheral rhythms in humans.

Behavioral and Circadian Cycle Influence

Circadian misalignment caused widespread restructuring of rhythmicity: nearly four times as many compounds gained (161) as lost (41) 24-h time-of-day patterns. Because the behavioral cycles were present throughout our protocol, these 24-h time-of-day patterns likely represent composite effects of endogenous circadian rhythms and evoked rhythms in response to behaviors, mirroring real-world shiftworkers. Among compounds gaining or losing 24-h time-of-day patterns, lipids were the largest category, including 9 phospholipids. This is consistent with prior work (Skene et al., 2018) and broadly suggests circadian misalignment uncouples lipid metabolism from the central circadian clock.

Consistent with targeted analyses (Skene et al., 2018; Kervezee et al., 2019), more compounds were behavioral-influenced than circadian-influenced. Several behavioral-influenced compounds overlapped between our data and prior studies (Skene et al., 2018; Kervezee et al., 2019), including multiple acylcarnitines and the amino acid proline, highlighting reproducibility. Notably, we identified 12 behavioral-influenced acylcarnitines, representing the largest class of behavioral-influenced compounds, reinforcing their potential use as biomarkers of peripheral rhythms in humans.

Despite this agreement, some compounds diverged from prior studies. For example, we classified decanoylcarnitine and creatine as circadian-influenced, whereas prior analyses classified decanoylcarnitine as behavioral-influenced and reported no rhythmicity in creatine following circadian misalignment (Kervezee et al., 2019). One potential explanation for these differences is the duration of circadian misalignment. Our data reflect the second night of misalignment, whereas prior studies assessed responses after four days (Skene et al., 2018; Kervezee et al., 2019). It is therefore possible that our findings underestimate behavioral cycle effects that emerge more fully with longer exposure. Supporting this interpretation, decanoylcarnitine is thought to originate mainly from cardiac tissue (Makrecka-Kuka et al., 2017), which shifts phase more slowly than the liver in response to altered timing of food intake in mice (Damiola et al., 2000). Together, these observations raise the possibility that differences in rates of compound phase-shifting may reflect different rates of phase-shifting between tissues. Under this framework, compounds derived from slower-shifting tissues (e.g. cardiac) may appear circadian-influenced early during misalignment but then show stronger behavioral cycle influence over longer timeframes. Defining the tissue(s) of origin for such compounds will be critical to better understand this possibility and any potential health implications.

Behavioral-influenced compounds exhibited significant internal misalignment. Specifically, during circadian alignment the behavioral-influence compounds clustered around a single acrophase, whereas during misalignment, two distinct acrophase clusters emerged. This produced internal misalignment on multiple levels, within the behavioral- influenced compounds, between behavioral- and circadian-influenced compounds, and relative to the central circadian clock (DLMO). Potential physiological consequences of such multi-level internal misalignment between compounds remains unknown but our exploratory pathway analyses suggest potential links to metabolic regulation and pyrimidine metabolism. Implementing two modeling approaches yielded further insights. Among the 44 circadian-influenced compounds identified in the dichotomous analysis, only 8 mapped to the exclusively circadian-influenced group in the second modeling approach. In contrast, 34 of the 44 circadian-influenced compounds identified in the dichotomous analysis did not meet the R² ≥ 0.60 threshold in the second modeling approach, including the three circadian-influenced compounds with >4 h acrophase shifts. This suggests the circadian cycle likely contributes a relatively smaller proportion of the temporal variation of these compounds when analyzed with the behavioral cycles present. In contrast, behavioral-influenced compounds showed stronger concordance across modeling approaches. Of the 70 behavioral-influenced compounds identified in the dichotomous analysis, none mapped to the exclusively circadian-influenced group and 35 mapped to the exclusively behavioral group in the second modeling approach. Another 6 compounds were in group 3 of the second modeling approach, indicating some behavioral cycle-influence. This greater consistency among behavioral-influenced compounds may reflect the larger number of compounds showing behavioral-versus circadian-influence, and that the behavioral cycles were present in both conditions, potentially masking circadian-influence. These results highlight the robustness of our protocol to identify behavioral-influenced compounds and strengthens confidence in compounds with consistent outcomes across modeling approaches, including uridine and uracil. The differences between approaches highlights an important need to understand both the statistical significance and physiological relevance of the identified behavioral and circadian cycle-influences.

Glucose Metabolism

Circadian misalignment impaired glucose metabolism, as indicated by elevated glucose and insulin AUC, consistent with prior work (Morris et al., 2015; Chellappa et al., 2021). Building on this, we identified seven compounds (uridine, uracil, GUDCA, linoleyl carnitine, pipecolinic acid, 1-methylinosine, and one unknown compound) showing significant associations between their phase relationships with DLMO and glucose AUC. These results suggest internal misalignment between these compounds and the central circadian clock may contribute to impaired glucose metabolism.

Uridine, a ribonucleoside composed of the pyrimidine base uracil, has roles in DNA/RNA synthesis, glycogen metabolism, and lipid glycosylation (Yamamoto et al., 2011), and shows 24-h time-of-day patterns that follow the light-dark cycle in mice (el Kouni et al., 1990). In addition, uridine rises during fasting and decreases following food intake in rodents and humans (Deng et al., 2017; Hanssen et al., 2023). In mice, the fasting-associated rise in uridine leads to a decreased metabolic rate and core body temperature, whereas the post-food intake decrease in uridine promotes enhanced insulin sensitivity (Deng et al., 2017). Human data similarly suggest uridine correlates with hunger, with experimentally increased uridine leading to increased hunger and food intake (Hanssen et al., 2023). Intriguingly, data implicate the uridine-diphosphate (UDP) receptor P2Y6 on agouti related protein (AgRP) neurons in the arcuate nucleus of the hypothalamus as key mediators of these effects (Steculorum et al., 2015, 2017). Moreover, rodent data indicate AgRP neuronal regulation of feeding and glucose homeostasis is influenced by AgRP clocks (Cedernaes et al., 2019; Sayar-Atasoy et al., 2024; Douglass et al., 2025). These prior findings identify a potential link between uridine metabolism and the circadian clock in AgRP neurons, with potential implications for regulating whole-body glucose and energy metabolism.

Consistent with these prior observations, we show uridine and uracil are behaviorally influenced and that their degree of internal misalignment relates to impaired glucose tolerance. Furthermore, in our exploratory pathway analyses pyrimidine metabolism and pyrimidine and purine catabolism were among the top pathway hits from group 4 (behavioral cycle-influence), providing additional evidence that uridine and uracil are behaviorally-influenced. Together, our findings support the hypothesis that disrupted temporal coordination of metabolic pathways—now including uridine and uracil—may contribute to adverse metabolic risk in shift-workers. Critical next steps include disentangling the potential influence of the fasting-eating versus sleep-wakefulness cycles, determining tissue(s) of origin, full characterization of the pyrimidine pathway using targeted metabolomics, and understanding key signaling targets/pathways in the brain. For example, rodent data indicate liver, adipose tissue, and gut have roles in regulating uridine, with potentially critical signaling in the hypothalamus (Steculorum et al., 2015; Deng et al., 2017; Steculorum et al., 2017). Our findings lay the groundwork for testing if behavioral interventions—such as aligning the timing of food intake with the circadian biological day—can preserve uridine rhythmicity and mitigate impaired glucose homeostasis in shift-workers. These insights also raise the possibility that uridine could serve as a biomarker for internal metabolic misalignment, with potential applications in circadian medicine (Klerman et al., 2022).

GUDCA also showed behavioral influence and internal misalignment associated with impaired glucose metabolism. Recent findings show GUDCA exhibits a 24 h time-of-day pattern during circadian alignment (Bello et al., 2024), which is maintained with total sleep deprivation but lost when food intake is evenly distributed across the 24 h day in a constant routine protocol, suggesting strong regulation by the timing of food intake (Bello et al., 2024). Our findings are consistent with this feeding-driven regulation and extend prior work by linking internal misalignment of GUDCA to impaired glucose tolerance. As with uridine, a critical next step includes full characterization of the bile acid-gut microbiome pathways using targeted metabolomics. As GUDCA and related bile acids are being explored as therapeutic agents in preclinical diabetes and cardiovascular disease research (Huang et al., 2021; Chen et al., 2023), our results emphasize the need to carefully track and control the timing of intake in clinical interventions targeting bile acid-gut microbiome pathways.

Linoleyl carnitine, pipecolinic acid, and 1-methylinosine also had phase relationships associated with glucose AUC. Linoleyl carnitine is an 18-carbon acylcarnitine involved in transporting fatty acids into the mitochondrial matrix for fatty acid oxidation. Pipecolinic acid is a lysine derivative and higher levels of pipecolinic acid have been associated with increased cardiometabolic disease risk (Ouyang et al., 2021; Liu et al., 2024). 1-methylinosine is a tRNA nucleoside produced by tRNA adenosine deaminases. The potential clinical significance of the association between glucose AUC and 1-methylinosine is unclear at this time. In general, the shifted 24-h time-of-patterns of these compounds suggests our healthy participants experienced a rapid shift in both fatty acid metabolism and lysine metabolism during circadian misalignment. These rapid shifts induced a state of internal misalignment with the central SCN clock that may contribute to adverse cardiometabolic risk, especially dysregulated glucose metabolism.

Energy Expenditure

As previously published (McHill et al., 2014), total daily energy expenditure decreased during circadian misalignment, primarily due to reduced energy expenditure during the daytime sleep opportunity. Given the continuous structure of energy expenditure data, we conducted exploratory association analyses between hourly compound abundance and hourly energy expenditure. Uridine and uracil were negatively associated, and GUDCA was positively associated with hourly energy expenditure. These associations were directionally consistent with preclinical data showing higher uridine decreases metabolic rate (Deng et al., 2017), and bile acids, including GUDCA, can promote thermogenesis and increased energy expenditure (Watanabe et al., 2006; Broeders et al., 2015; Chen et al., 2023). Higher uridine and uracil, and lower GUDCA, during daytime sleep of circadian misalignment may therefore contribute to reduced energy expenditure. Together, our findings potentially implicate altered uridine and bile acid metabolism as candidate pathways linking circadian misalignment to impaired energy balance.

Limitations

Our findings should be interpreted in the context of some limitations. First, the intensive 24 h blood sampling may have affected sleep and induced physiological stress. Although sampling was performed uniformly across all study conditions—minimizing differential effects—we could not directly quantify stress-related responses. Replication in both controlled laboratory settings and larger free-living cohorts is therefore important. Second, our sample size (n = 14), although similar to prior controlled circadian metabolomics studies (Skene et al., 2018; Kervezee et al., 2019), limits statistical power and increases risk of both false positives and false negatives. Accordingly, our findings should be considered hypothesis-generating. At the same time, our tightly controlled protocol helps minimize variability and the consistency of several findings with prior work provides additional confidence. Differences across studies likely reflect multiple factors, including protocol design, duration of misalignment, and analytical approaches. Ultimately, validation in larger and more diverse populations, including real-world shiftworkers, is needed. Third, we did not report circadian MESOR or amplitude, as our modified constant routine included behavioral cycles (e.g. sleep and feeding), which introduce masking and limit mechanistic interpretability of these metrics (Klerman et al., 2022). Future studies using full constant routine or forced desynchrony protocols will be necessary to isolate these parameters. Fourth, our pathway analyses are limited by incomplete coverage across all pathways of interest. As a result, findings are exploratory and should be interpreted cautiously. Targeted metabolomics approaches with full pathway coverage will be important to validate and extend our observations to the pathway level, particularly for pathways such as bile acid–microbiome interactions and pyrimidine metabolism. Despite these limitations, our untargeted approach enabled identification of novel compound signals and pathways associated with circadian misalignment, providing a foundation for future mechanistic and translational studies.

Conclusions

Our findings show circadian misalignment produces significant disruptions of the temporal patterns and composition of the human plasma metabolome. This includes changes in compound abundance, gain or loss of 24-h time-of-day patterns, and phase shifts of compounds influenced by behavioral cycles. Specifically, we linked internal misalignment of several compounds, particularly uridine and GUDCA, to impaired glucose tolerance and lower energy expenditure. These results provide evidence that desynchronization of behavioral-influenced compounds is not merely a byproduct of shiftwork but may represent a mechanistic link between circadian disruption and cardiometabolic dysfunction. By identifying circulating metabolites that both reflect internal misalignment and associate with metabolic impairment, this work establishes a framework for developing biomarker-based approaches to more comprehensively quantify circadian misalignment and for targeting entrainment of peripheral rhythms as a strategy to mitigate disease risk in shift-working populations.

Supplementary Material

Supplementary material is available for this article online.

Data Availability Statement

Raw mass spectrometry data files used in this publication will be available at the Metabolomics Workbench database upon publication. Processed peak height mass spectrometry data (height counts) is available as a Supplementary Data File.