Food intake controls daily liver glycogen use and protein release

The rhythmic plasma proteins under the meal-fed pattern exhibited a bimodal phase distribution with peaks at 05:00 h and 17:00 h (Fig. 1e). Circulating proteins that lose rhythmicity under spread-out eating patterns include postprandial metabolic hormones such as INS, GCG, PYY and PPY, along with other feeding-related hormones such as LEP (Fig. 1f and Extended Data Fig. 1b). We also detected diurnal rhythmicity of proteins involved in a broad range of functions, including blood coagulation (Coagulation factor Xa, IX, IXab, X, CRP), the complement system (Complement C2, C4b, C5, factor D and CD55/DAF), molecules and lipids transport (ALB, APOB, APOD) and protease inhibitors (SERPIN A5, C1, E1; Fig. 1f and Extended Data Fig. 1b,c). The TGF-β family also exhibited rhythmicity, peaking between 16:00 h and 20:00 h under the meal-fed pattern (Extended Data Fig. 1c). Nevertheless, a few plasma proteins such as CTSH and IGFBP1 maintained their rhythmicity even under the spread-out eating pattern (Extended Data Fig. 1b). To corroborate this observation, we reanalysed data from a previously published crossover design study¹⁶ that compared healthy participants under circadian alignment (daytime food intake, nighttime sleep) with circadian misalignment (daytime sleep, nighttime food intake). Consistent with our findings, rhythmicity in the plasma proteome was markedly reduced with misaligned eating (Extended Data Fig. 1d,e).

We next identified proteins derived from specific organs to assess the contribution of individual organs to the rhythmic plasma proteome. The liver, the primary source of circulating proteins, accounted for most organ-specific rhythmic plasma proteins. Specifically, 50% of liver-specific proteins exhibited rhythmicity with acrophases around 17:00 h (Fig. 1g and Extended Data Fig. 1f). However, rhythmic secretion was not exclusive to the liver, as rhythmic organ-specific proteins were also detected from other organs contributing to the diurnal plasma proteome (Fig. 1g,h).

To further investigate the role of timed food intake, we next used a mouse TRF model (Fig. 1i). Wild-type (WT) mice were maintained under a 12 h light–dark cycle and subjected to normal chow diet under different feeding paradigms, either with unrestricted food access (ad libitum; AL) or 8 h TRF, including an activity-aligned night-restricted (NR; food access between Zeitgeber time (ZT) 14 and 22, with ZT0 as lights on and ZT12 as lights off), and an activity-misaligned day-restricted (DR; food access between ZT2 and ZT10) feeding. After 2 weeks, metabolic parameters were assessed, and serum samples were collected every 4 h across the day for the aptamer-based proteomic analysis, described above, yielding strong correlations between biological replicates (Extended Data Fig. 1g,h). Comparable with previous reports using a 6–12 h NR feeding window^17–19, animals exposed to TRF consumed less food (−11% to −16%) and exhibited a reduced body weight change (−55% to −61%) compared with AL-fed mice (Fig. 1j and Extended Data Fig. 1i). Differential rhythmicity analysis of Dbp, Bmal1 and Gys2 in mouse liver confirmed feeding rhythm-dependent entrainment of clock and clock-regulated genes (Extended Data Fig. 1j), consistent with previous reports¹⁹. Furthermore, the temporal profiles of INS and GCG were aligned with the expected timing of food intake (Fig. 1k).

Analysis of the circulating proteome with the R package dryR²⁰ uncovered that 99% (200 out of 202) of rhythmic serum proteins are differentially affected by changes in the feeding regimen (Fig. 1l,m and Supplementary Table 2). Surprisingly, almost all serum proteins that are rhythmic under AL feeding lost their rhythmicity under TRF, independent of whether it was NR or DR feeding, with an acrophase at around ZT6 (rhythmic only in AL feeding (m2); Fig. 1n,o), as exemplified with VEGFC (Fig. 1p). Additionally, 37 serum proteins lost their 24 h rhythm exclusively under DR feeding (rhythmic only in AL and NR feeding (m7); Fig. 1n,o), with fibronectin (FN1) as a notable example (Fig. 1p). Altogether, these findings underscore the pivotal role of food intake in shaping the diurnal blood proteome.

Our findings were further corroborated by ex vivo secretion assays using livers from AL-fed WT mice (Fig. 2g), which demonstrated that the secretion of total liver protein (Fig. 2h), as well as of ALB and C3 (Fig. 2i), is higher at ZT0 than at ZT12, confirming that liver protein secretion is time-of-day-dependent. Notably, treatment with brefeldin A, an inhibitor of the classical ER–Golgi dependent secretory pathway²⁴, significantly reduced the secretion of these proteins (Fig. 2j,k). By contrast, we did not detect any rhythmic variation in circulating hepatocellular leakage markers such as ALT (official protein name GPT) and AST (official protein name GOT1) in either the human or mouse blood proteomics datasets (Supplementary Tables 1 and 2 and Extended Data Fig. 2b). Furthermore, we observed no significant time-of-day differences in the degradation of circulating proteins (that is, ALB, SERPINA1; Extended Data Fig. 2c–i), thereby excluding hepatocellular leakage or protein catabolism processes as major contributors to the observed rhythmicity in blood proteins. Taken together, these findings suggest that the rhythmicity of liver-derived circulating proteins arises primarily from secretion processes, with classical secretion representing the predominant route.

We next evaluated the rhythmicity of secreted proteins as defined by UniProt²⁵ at various gene expression stages, including transcription, translation and protein accumulation, using RNA sequencing (RNA-seq), ribosomal profiling²⁶ and total proteomics¹¹ (Extended Data Fig. 3a). This analysis showed that secreted proteins predominantly exhibited rhythmicity at the protein level in mouse liver (Extended Data Fig. 3b). By contrast, non-secreted proteins were more likely to exhibit rhythmicity at the messenger RNA (mRNA) and translational levels (Extended Data Fig. 3b), consistent with previous observations that most rhythmic mRNA accumulation and translation mainly result in non-rhythmic protein levels owing to the generally higher stability of proteins^11,26. Altogether, these results identify post-translational mechanisms as the primary drivers of hepatic diurnal protein secretion.

Our studies revealed an anti-phasic pattern in the diurnal abundance of ER and GA-specific proteins in mouse liver microsomes. Specifically, ER-resident proteins such as the chaperones CALR and HSPA5/BiP reached their peak abundance during the light phase (ZT3), while GA-resident proteins such as the Golgi matrix proteins giantin/GOLGB1 and GM130/GOLGA2 peaked during the dark phase (ZT18; Fig. 3h,i and Extended Data Fig. 3l). Although not fully understood, ER-resident and GA-resident proteins have been suggested to have a role in the generation and maintenance of organelle structure and organization, which may be crucial for organelle function^28,29. This prompted us to investigate the question of whether ER and GA abundance is subject to diurnal regulation. Using electron microscopy, we examined mouse liver hepatocytes collected at ZT4 and ZT16 to assess ER and GA volume density. Consistent with our proteomics data, we observed an anti-phasic variation in ER and GA volumes (Fig. 3j). We also measured the density of mitochondria, which was higher at ZT16, in line with previous studies³⁰.

Rhythmic hepatic microsomal proteins were particularly enriched for functions related to protein glycosylation (Fig. 3k and Extended Data Fig. 3m), suggesting that protein glycosylation may be a diurnally regulated process. Although our proteomics approach was not designed to enrich glycosylated proteins, we identified and quantified the mass of detectable glycosylated peptides (Supplementary Table 4) and found that N-glycosylation (Fig. 3l), similar to O-glycosylation³¹, exhibits a diurnal pattern, and its temporal profile aligns with the day–night variations in ER and GA protein abundance (Fig. 3h,i). Specifically, peptides with oligomannose N-glycosylation, which is initially transferred to proteins primarily processed in the ER, peaked during the light phase (ZT9; Fig. 3l–n and Extended Data Fig. 3n). By contrast, peptides with a complex type of N-linked glycosylation processed in the GA peaked some hours later (Fig. 3l–n and Extended Data Fig. 3n). In summary, these findings demonstrate that the secretory pathway in the liver is not a constitutive but a temporally dynamic process.

Given the critical role of the circadian clock in glucose metabolism³³, we next investigated its contribution to rhythmic liver glycogen metabolism using two clock-disrupted mouse models, Bmal1-knockout (KO) and Cry1/2-KO mice, which are deficient in the transcriptional activator and repressor complexes of the circadian clock, respectively²⁰. The circadian clock component CLOCK, the heterodimerization partner of BMAL1, has been shown to directly regulate the transcription of Gys2 (ref. ³⁴), the rate-limiting enzyme of glycogen synthesis (Fig. 4a). Our data showed that BMAL1 regulates the rhythmic transcription of multiple enzymes involved in glycogen metabolism, and that Bmal1-KO mice exhibit distinct expression patterns of these enzymes compared to Cry1/2-KO mice (Extended Data Fig. 4a–c). For example, mean expression levels of Gys2 were reduced by 2.6-fold in Bmal1-KO mice, whereas no significant reduction was observed in Cry1/2-KO mice (Extended Data Fig. 4a–c). By contrast, the mean expression levels of Pygl were significantly elevated and non-rhythmic in Bmal1-KO mice but were less affected in Cry1/2-KO mice (Extended Data Fig. 4a–c). Consistent with these findings, the diurnal rhythmic patterns of hepatic glycogen levels observed in AL-fed WT mice were attenuated in Bmal1-KO mice but remained intact in Cry1/2-KO mice (Extended Data Fig. 4d), probably because of a rhythmic flux of glucose into glycogen through dietary carbohydrate uptake. These results highlight that although both Bmal1-KO mice and Cry1/2-KO mice exhibit disrupted circadian clocks, only Bmal1-KO mice show impairments in glucose storage and present a unique metabolic phenotype in addition to circadian function³⁵.

To investigate these findings in the context of hepatic protein secretion, we examined microsomal liver fractions from Cry1/2-KO mice and Bmal1-KO mice across the day (Fig. 4e). The mice were maintained on a 12 h light–dark cycle with 12 h NR feeding to avoid confounding effects of altered feeding behaviour²⁰. Comparative analysis of temporal profiles of the expression of proteins involved in the different aspects of the secretory pathway between Cry1/2-KO animals and their WT counterparts revealed a strong correlation (Fig. 4f). This finding was further confirmed by a correlated rhythmic accumulation of secreted proteins such as ALB, C3 and alpha-1-antitrypsin (SERPINA1D) between Cry1/2-KO and WT mice (Fig. 4g), extending previous work¹¹. By contrast, Bmal1-KO mice showed either no correlation or a negative correlation with their WT counterparts (Fig. 4f,g), altogether providing evidence that perturbed rhythms in secretory proteins are Bmal1-KO-specific. We next analysed serum proteins of Bmal1-KO mice and found that the levels of liver-derived blood proteins, including ALB, FN1 and C3, were altered in both the serum and liver of Bmal1-KO mice (Fig. 4h,i and Supplementary Table 5). The reduced circulating protein levels of FN1 and C3 were confirmed using ELISA (Fig. 4j). Notably, the changes observed for mouse liver ALB were also evident at the gene expression level (Fig. 4k), with Alb being a target of the CLOCK/BMAL1-regulated transcription factor DBP^36,37. Taken together, these findings point to the BMAL1-regulated hepatic glycogen metabolism as a potential driver of rhythmic protein secretion.

Subsequent analysis of N-glycoproteins with terminal α-D-mannosyl and α-D-glucosyl groups in mouse liver revealed a transient decrease in protein N-glycosylation upon CP-91149 treatment (Fig. 5d). In line with these changes, circulating C3 levels were altered (Fig. 5e), altogether confirming a role of PYGL-mediated glycogenolysis for protein N-glycosylation and secretion. To further study the downstream effects of CP-91149, we performed RNA-seq analysis on treated mouse livers and identified 130 differentially expressed genes at 6 h post treatment, which decreased to only nine differentially expressed genes after 24 h (Supplementary Table 6). A gene ontology enrichment analysis revealed that drug detoxification pathways were upregulated in the liver as early as 6 h following CP-91149 treatment (Extended Data Fig. 5e), suggesting an induction of drug metabolizing enzymes and, consequently, a rapid breakdown of CP-91449 and reduction of its effects.

To circumvent potential compensatory effects in vivo, we extended our investigations to AML12 mouse hepatocytes (Fig. 5f). CP-91149 treatment did not cause adverse effects (Extended Data Fig. 5f,g), but effectively reduced UDP-glucose + UDP-galactose levels within 3 h (Fig. 5g) and altered other glycosylation substrates (Fig. 5h) and downstream metabolites of glucose or glycogen metabolism (Extended Data Fig. 5h). Additionally, CP-91149 treatment resulted in a rapid decreased phosphorylation of ribosomal protein S6 (RPS6-P; Extended Data Fig. 5i), a target of the glucose-sensing mTOR pathway⁴¹, which further supports effective PYGL inhibition. Similar to in vivo, CP-91149 treatment significantly decreased the abundance of N-glycoproteins in AML12 cells (Fig. 5i). It also impaired the secretion of ALB (a non-glycosylated protein), as well as FN1 and C3 (both glycosylated proteins), into the culture medium (Fig. 5j). Altogether, these results confirm that PYGL-mediated hepatic glycogenolysis is required for the synthesis of glycosylation substrates and, subsequently, protein N-glycosylation and secretion. To confirm the role of UDP-glucose in the mechanism, we added exogenous UDP-glucose in addition to CP-91149 (Fig. 5k). Our observation that UDP-glucose supplementation only restored C3 secretion (Fig. 5l) and partially rescued intracellular C3 levels (Fig. 5m), while ALB secretion remained suppressed by CP-91149 treatment (Fig. 5l,m), underscores the differential dependence of glycosylated versus non-glycosylated proteins on UDP-glucose for liver protein secretion and suggests that additional mechanisms are involved in the regulation of the secretion of non-glycosylated proteins.

Therefore, we aimed to test whether alleviating ER stress would restore protein secretion. To this end, AML12 cells were treated with salubrinal (SAL; Fig. 6i), a selective inhibitor of eIF2α phosphatases that stabilizes phosphorylated eIF2α and modulates the PERK–ATF4 arm of the UPR, thereby mitigating ER stress^44,45. SAL-treatment markedly increased the secretion of non‑glycosylated ALB (Fig. 6j) and, to a lesser extent, intracellular ALB content (Fig. 6k). CP-91149 treatment in the presence of SAL did not differ from CP-91140 treatment alone, indicating that glycosylation stress overrides eIF2α-mediated ER stress mitigation. For glycosylated C3, both SAL and CP-91149 reduced secretion (Fig. 6j) and intracellular levels (Fig. 6k), with combined treatment resulting in additive inhibition (Fig. 6j,k). Analysis of gene expression further corroborated these observations. As expected, CP‑91149 induced robust upregulation of canonical UPR and ERAD markers (Fig. 6l and Extended Data Fig. 6k), whereas SAL alone selectively increased only Herpud1 (Fig. 6l), an ERAD adaptor that recruits misfolded proteins for degradation, without affecting expression of protein glycosylation-associated genes (Fig. 6l). Notably, Calr and Mgat1 transcripts were downregulated in CP‑91149-treated AML12 cells, both in the absence and presence of SAL. Together, these data support a model in which glycosylation defects elicit ER stress and activate UPR and ERAD pathways that differentially impact the secretion and degradation of glycosylated and non-glycosylated proteins. Although glycosylated proteins such as C3 appear to be directly targeted by ER quality control and degradation mechanisms, non-glycosylated proteins such as ALB are indirectly affected through a global UPR-mediated increase in protein-folding factors, which are known to enhance protein secretion^46,47.

Next, we examined hepatic glycogen levels and found that their diurnal rhythms were maintained across all conditions (Fig. 7g), with reverse-phase (DR) feeding causing a 12 h shift in the glycogen peak in both WT and Ob/Ob mice. Nevertheless, Ob/Ob mice exhibited a markedly lower amplitude (Fig. 7g,h and Extended Data Fig. 7d), probably resulting from the strong attenuation of rhythmic Gys2 expression and the increased expression of Pygl observed in these animals (Fig. 7i and Extended Data Fig. 7e). Accordingly, glycogen levels were significantly elevated in Ob/Ob mice, with higher trough levels compared to AL-fed WT mice (Fig. 7g,h), consistent with other reports in genetically obese animals^48,50. Interestingly, mean glycogen levels were also elevated in WT mice under TRF (Fig. 7h), probably because of the associated caloric restriction (Figs. 1j and 7b), which promotes glucose storage via glycogen synthesis rather than glycogenolysis⁵¹. The calorie-restriction effect of TRF on mean or trough levels of glycogen was less pronounced in Ob/Ob mice (Fig. 7h), which are metabolically skewed to favour glucose storage. Strikingly, elevated glycogen levels in Ob/Ob mice and WT mice under TRF negatively correlated with the number of rhythmic blood proteins, thus linking reduced glycogenolysis to a loss of rhythmicity in secretory proteins (Fig. 7d,e).

We then investigated ER stress/UPR gene expression, which typically exhibit 12 h ultradian rhythms in AL-fed WT mice^52,53, a pattern we likewise observed in our study (Fig. 7j and Extended Data Fig. 7f). However, Ob/Ob mice showed overall lower expression levels of several ER stress/UPR markers, and the characteristic expression peaks at ZT0–4 and ZT16 observed under AL feeding were lost (Fig. 7j and Extended Data Fig. 7f). WT mice also lost the ER stress-activation peak at ZT0–4 and ZT16 when exposed to NR and DR feeding, respectively. Notably, the detected loss of ER stress-activation peaks in these mice coincides with troughs in secretory liver protein levels and their ubiquitination (Extended Data Fig. 7g), an observation that, together with our PYGL inhibition studies, supports the involvement of protein degradation processes. Altogether, these findings confirm that impaired glycogen metabolism, whether caused by obesity or calorie restriction, is associated with altered protein secretion and impaired initiation of rhythmic physiological ER stress. Consistently, Bmal1-KO mice, which exhibit defects in both glycogen metabolism and protein secretion (Fig. 4e–k), lack the ER stress-activation peak at ZT0–4, resulting in a single remaining peak at ZT14 (Extended Data Fig. 7h). This provides a mechanistic explanation for the reported loss of 12 h ER stress-activation rhythms and their shift to 24 h rhythms in Bmal1-KO mice^26,52,53.

All experimental studies were conducted in accordance with the regional committee for ethics in the regulations of the veterinary office of the Canton of Vaud, Switzerland (VD2459, VD2801, VD2720), and the University of Queensland Animal Ethics Committee (2021/AE000004↗, 2022/AE000274↗). The studies involving human participants were conducted in compliance with the Declaration of Helsinki and received ethical approval from both the Health Research Authority (Cornwall & Plymouth National Health Service Research Ethics Committee; reference 14/SW/0123) and the Commission Cantonale d’Éthique de la Recherche of Canton de Vaud (CER-VD; Protocol 337/12). Participants provided written informed consent.

For investigating the human diurnal secretome, blood samples were collected from two studies, each study containing nine healthy male volunteers (Fig. 1a,b and Extended Data Fig. 1a). The volunteers had no history of sleep disorder, as confirmed by questionnaires MCTQ⁷⁹ and Horne-Ostberg⁸⁰. For the meal-fed pattern study, nine healthy men aged between 22 and 27 years were recruited. Participants could follow their normal breakfast routine before participating in the study. At 08:00 h, each participant had a catheter installed for blood sampling. Blood samples were collected every 4 h throughout the day. Standardized meals were provided at 12:00 h and 18:00 h, with participants allowed one snack during the day. Throughout the study period, participants were instructed to maintain normal daily activities but were restricted from engaging in sports and consuming alcohol. Lights were turned off at 23:00 h to standardize sleep conditions. The design and metadata on the spread-out eating pattern study were previously published in detail⁸¹. In brief, nine healthy men, aged 22 to 54 years, arrived at the laboratory at 19:00 h the day before the testing. They remained semi-recumbent throughout the study with a standardized dinner provided the evening before testing. Lights were on from 07:00 h to 22:00 h. Participants were woken at 07:00 h, and a catheter was placed at 08:00 h to collect blood samples. On the day of testing, participants ingested Fortisip (Nutricia) hourly during waking hours (08:00 h to 22:00 h) to maintain energy balance. The amount of Fortisip matched each participant’s daily energy expenditure, which had been assessed upon waking that morning by indirect calorimetry⁸².

Experiments with WT mice were conducted using C57BL/6J (RRID:IMSR_JAX:000664) purchased from Charles River (L’Arbresle, France) or Ozgene ARC (Australia). The generation of SILAC C57BL/6J mice was previously described¹¹. Studies involving clock-depleted animals were carried out using Bmal1-KO and Cry1/2-KO mice, along with their respective WT controls, generated as previously described^83,84. C57BL/6J Ob/Ob mice were obtained from Charles River (RRID:IMSR_JAX:000632). Unless otherwise noted, all animals were maintained under a standard 12 h light–dark cycle (ZT0, lights on; ZT12, lights off) in regular housing conditions (temperature of 23 ± 1 °C; humidity ranging between 30% and 70%), with free access (AL) to food (sterile chow diet; SAFE, U8213G10R; Specialty Feeds, SF00-100) and water. All mice used in the experiments were adult males, aged 9–15 weeks.

AML12 (alpha mouse liver 12) hepatocyte cells from a 3-month-old mouse donor (CD1 strain, line MT42) were purchased from ATCC (CRL-2254). Cells were maintained in an incubator (37 °C, 5% CO₂) in AML12 base medium (DMEM/F-12 medium with L-glutamine and 15 mM HEPES (Gibco, 11330032), supplemented with 10% FBS, 10 µg ml⁻¹ insulin, 5.5 µg ml⁻¹ transferrin, 5 ng ml⁻¹ selenium and 40 ng ml⁻¹ dexamethasone).

For TRF experiments, mice were subjected to either 8 h or 12 h NR or DR feeding schedules, whereby food access was limited to the respective day or night periods; food was provided in bulk at the start and removed at the end of each feeding window. In the 12 h NR feeding regimen, food was provided from ZT12 to ZT24, starting 4 days before tissue collection. For the 8 h TRF, food access was available between ZT2 and ZT10 for DR feeding, or between ZT14 and ZT22 for NR feeding, beginning 2 weeks before tissue collection. Organs were snap-frozen in liquid nitrogen and stored at −80 °C until further processing.

C57BL/6J mice were exposed to 12 h NR feeding for 4 days before receiving a subcutaneous injection (40 mg kg⁻¹; volume of 10 µl per g body weight) of CP-91149 (Sigma-Aldrich, PZ0104) or vehicle (5% PEG400/5% Tween80/5% DMSO) at ZT0. Body weight and food intake of mice were measured after 24 h of CP-91149 or vehicle treatment. Blood glucose was measured with the Breeze2 system (Bayer).

Cell cytotoxicity was assessed using the LDH-Glo Cytotoxicity Assay according to the manufacturer’s protocol (Promega Corporation, J2380), and cell viability was measured using resazurin (Sigma-Aldrich, R7017) as previously described⁸⁵. Untreated cells and Triton X-100 (TX100) were used as controls.

For metabolite analysis, AML12 cells (6,500,000 cells per dish) were seeded into sterile 150 mm dishes and incubated overnight at 37 °C. The following day, cells were washed twice with 1× PBS, and the medium was replaced with AML12 base medium containing either 67 μM CP-91149 or vehicle (0.03% DMSO). At 3 h, 6 h, 14 h and 24 h post treatment, cells were washed twice with ice-cold 0.9% NaCl and collected using a cell scraper in 800 µl of the same solution. Cell pellets were collected by centrifugation at 400g for 5 min at 4 °C, weighed to determine wet mass and snap-frozen in liquid nitrogen. For all other cell experiments, AML12 cells (300,000 cells per well) were seeded into sterile six-well plates. Treatments were carried out in AML12 base medium and included the following conditions: 67 μM CP-91149 or 0.03% DMSO as vehicle for 3–24 h; 2 mM UDP-glucose (Merck Life Science, U4625) or 2% water as vehicle for 14 h in the absence or presence of 67 μM CP-91149. For the SAL (Merck Life Science, SML0951) experiments, AML12 cells were first pre-incubated for 14 h with 15 μM SAL and then, following a wash in 1× PBS, were treated for 6 h with 15 μM SAL in the absence or presence of 134 μM CP-91149.

For total protein extraction from mouse livers, 40–100 mg of snap-frozen liver tissue was homogenized in a buffer containing 20 mM HEPES (pH 7.6), 100 mM KCl, 0.1 mM EDTA, 1 mM NaF, 1 mM NaVO₄, 1% Triton X-100, 0.5% NP-40, 0.15 mM spermine, 0.5 mM spermidine, 1 mM dithiothreitol, 1.6 µM aprotinin, 1 µM pepstatin A, 1.7 µM leupeptin and 0.5 mM PMSF. The homogenates were incubated on ice for 30 min, followed by centrifugation at 16,000g for 10 min to collect the supernatant.

To measure protein in cell culture medium, the media were centrifuged at 10,000g for 10 min at 4 °C to remove debris and collect the supernatant. For cell lysates, cells were first incubated on ice in 200 µl of RIPA lysis buffer supplemented with a protease and phosphatase inhibitor mixture (0.2 tablet cOmplete and one tablet PhosSTOP per 10 ml; Roche) for 15 min. The cells were then scraped from the dish and incubated for an additional 30 min on ice. The lysates were centrifuged at 10,000g for 10 min at 4 °C to collect the supernatant. Protein concentrations in the supernatants were determined using a Bradford assay.

For microsomal protein extraction, liver tissue was homogenized in 0.2 M sodium phosphate buffer supplemented with a protease and phosphatase inhibitor mixture (0.2 tablet cOmplete and one tablet PhosSTOP per 10 ml; Roche). The homogenate was clarified by centrifugation at 9,000g for 20 min at 4 °C. Microsomal fractions were recovered from the supernatant by centrifugation at 105,000g for 1 h. The resulting pellet was resuspended in 8 M urea buffer with inhibitors. After a 20 min incubation on ice, the extracts were centrifuged at 20,000g for 10 min to collect the supernatant. Protein concentrations were determined using a BCA protein assay kit.

Tandem mass spectrometry (MS/MS)-based SILAC analysis was conducted as previously described^11,27, with the modification that microsomal proteins were analysed. A common microsomal fraction reference SILAC protein mix was prepared from 16 SILAC protein samples (six male and ten female SILAC livers) collected at ZT0 and ZT12 and added in a ratio of 1:1 to each sample. In brief, protein samples were precipitated with trichloroacetic acid and deoxycholate, redissolved in 8 M urea and digested with trypsin. Peptides were fractionated by off-gel isoelectric focusing, and all 12 fractions of each mix were analysed on a Fusion tribrid mass spectrometer (Thermo Fisher Scientific) on 60 min gradients. MS data were analysed with MaxQuant (v.1.5.1.2)⁸⁶ against the UniProt database (release 2014_10) restricted to mouse (Mus musculus) taxonomy. The same procedure was applied to protein samples from Cry1/2-KO and Bmal1-KO mice using the same SILAC mix as the reference.

Glycopeptides in mouse liver microsomal fractions were identified in the Byonic (Protein Metrics v.4.3.4) node of Proteome Discoverer (Thermo Fisher Scientific, v.2.5.0.400) searching against a database of 16,964 mouse proteins (UniProt reviewed, downloaded 20 April 2018) and a database provided by Byonic containing 83 mouse plasma N-glycans. The cleavage enzyme was specified as trypsin with no missed cleavages. Precursor mass tolerance was set to 6 ppm, and fragment mass tolerance was set to 0.5 Da. Modifications were set as follows: carbamidomethyl at cysteine (fixed), SILAC heavy label at lysine (common 1) and all N-glycans (rare 1).

Human plasma and mouse serum proteins were quantified using the SomaScan assay (v.3.2). SomaScan uses dilution groups in its assays to ensure that the measurement for each individual aptamer remains within the linear range, accommodating the wide range of endogenous protein concentrations present in the blood. The most abundant proteins are found in the highest dilution groups⁸⁷. Samples were processed according to the manufacturer’s protocol at Somalogic, and the resulting relative fluorescence units were normalized and log₂-transformed for downstream statistical analysis. The SomaScan assay (v.3.2) is optimized for human proteins (1,305 protein targets); thus, we only considered 896 SOMAmers identified as having medium to high quality in mouse serum samples, based on performance in dilution linearity and assay variability in technical replicates. We identified organ-specific human plasma proteins by analysing gene expression patterns using the GTEx project RNA-seq dataset⁸⁸. A plasma protein was classified as organ-specific if its encoding gene was four times more highly expressed in one organ compared to others. Serum AST and ALT activities were measured by an IFCC kinetic UV assay, using Cobas 8000 (Roche).

Mouse serum protein samples were digested, and peptides were purified using previously established methods⁷⁰. Liquid chromatography–MS/MS analysis of these samples used the scanning SWATH methodology on a Sciex 6600 mass spectrometer with a 5 min gradient as described previously⁸⁹. Raw MS data were analysed using quantitative DIA proteomics software DIA-NN (v.1.8). The complete mouse proteome database from Uniprot was used for neural network generation, with deep spectral prediction enabled. Protease digestion was set to trypsin, allowing for one missed cleavage and one variable modification. Oxidation of Met and acetylation of the protein N-terminus were set as variable modifications. Carbamidomethyl on Cys was set as a fixed modification. Match between runs and remove likely interferences were enabled. The neural network classifier was set to double-pass mode. Protein interferences were based on genes. Quantification strategy was set to any liquid chromatography (high accuracy). Library profiling was set to smart profiling.

For each time point (ZT0 and ZT12), livers were dissected from C57BL/6J mice under AL feeding. Each liver was cut into two ~100 mg pieces, which were individually washed three times with 1× PBS, transferred into 25 cm² tissue flasks (one tissue piece per flask) containing 5 ml CHO medium (Gibco, 10743029) and incubated at 37 °C for 30 min. Medium was then collected, centrifuged at 200g for 5 min at 4 °C and supernatants were stored at −80 °C until analysis. For each mouse liver, the secretion measurements from two tissue pieces were averaged to generate a single biological replicate. For the brefeldin A experiments, the same procedure was followed as described above, with the modification that the two liver pieces per mouse were incubated in parallel with CHO medium containing either 2 mM brefeldin A or vehicle (2% DMSO).

Protein samples were immersed in Bolt LDS Sample Buffer (1×) substituted with 50 mM dithiothreitol and subsequently heat-denatured at 70 °C for 10 min. Denatured protein samples (10–30 µg) were fractionated using an Invitrogen 4–12% Bis-Tris plus gel. Proteins were then transferred to an Immobilon-FL low-fluorescence PVDF membrane, and western blot analysis was performed using the Odyssey CLx system according to the manufacturer’s protocol. Intercept blocking buffer (LI-COR, LCR-927-60001) was the standard blocking buffer, but for biotinylated lectin detection, Intercept Protein-Free Blocking buffer (LI-COR, 927-80001) was used. Primary antibodies were used at the following dilutions: 1:1,000 for ATF4 (Cell Signaling Technologies, 11815), ARFGAP1 (Cell Signaling Technologies, 14608), Phospho-RPS6 (Cell Signaling Technologies, 2211), Total-RPS6 (Cell Signaling Technologies, 2217), GABARAPL1 (Genetex, GTX132664) and ConA Lectin (Vector Laboratories, B-1005) and 1:2,000 for STX4 (ProteinTech, 14988-1-AP). Secondary anti-Mouse (Thermo Fisher Scientific, SA5-35521) and anti-Rabbit polyclonal (Thermo Fisher Scientific, SA5-35571) antibodies were used at a 1:10,000 dilution. For biotinylated SERPINA1 or lectin detection, LI-COR IRDye 800CW Streptavidin (LI-CORbio, 926-32230) was used at a 1:5,000 dilution.

ELISAs targeting ALB (ab108792), FN1 (ab210967) and C3 (ab157711) were performed according to the manufacturer’s protocols (Abcam). Dilutions of ex vivo supernatants were 1:20 for ALB and undiluted for C3. For the cell media, dilutions were 1:10 for ALB, 1:20 for FN1 and 1:10 for C3. For cell lysates, dilutions were 1:20 for ALB, 1:32 for FN1 and 1:16 for C3. For mouse serum, a 1:40,000 dilution was used for C3.

SERPINA1D was synthesized by in vitro translation using the TNT Quick Coupled Transcription/Translation Kit (Promega, L1170) with 1 µg of a T7 promoter-driven plasmid encoding mouse Serpina1d and containing an SV40 polyadenylation signal (VectorBuilder, T7::Serpina1d_SV40polyA) and biotinylated tRNA (Promega, L5061) at 30 °C for 90 min. The translation mix was combined 1:1 (v/v) with serum and incubated at 37 °C for the indicated times; biotinylated SERPINA1D degradation was quantified by densitometric analysis of western blots. Serum-protein stability (including ALB) was assessed by incubating serum 1:1 with PBS at 37 °C, resolving by SDS–PAGE, staining with Coomassie Blue and quantifying band intensities by densitometry.

Glycogen extractions from the liver were performed using a phenol-sulfuric acid quantification as previously described⁹⁰. In brief, 40–100 mg of snap-frozen liver tissue was immersed in a 30% KOH solution and incubated for 20–30 min at 100 °C, with intermittent vortexing, until fully homogenized. A 95% ethanol solution was then added, and the samples were vortexed thoroughly before being incubated on ice for 30 min. This was followed by centrifugation at 2,000g for 10 min at 4 °C. The pelleted glycogen was resuspended in water, and debris was sedimented at 10,000g for 30 s before glycogen was measured in a phenol-sulfuric acid solution (67.8–70% (v/v) sulfuric acid 0.71% (w/v) phenol) at an absorbance of 490 nm.

Frozen cell pellets (7–14 mg wet mass) were processed using a modified method for targeting energy carriers and nucleotide sugars⁹¹, including polarity switching and additional metabolites of interest. In brief, cell pellets were extracted on ice using 250 µl of cooled extraction buffer (acetonitrile:MeOH:15 mM ammonium acetate in H₂O (3:1:1), pH 10). The samples were then sonicated in a sonication bath (Transsonic 460, Elma) for 5 min at the highest frequency on ice to ensure complete cell disruption. Afterwards, the samples were centrifuged at 13,000g for 15 min at 4 °C, and the supernatant was transferred to a new liquid chromatography–MS grade autosampler vial and immediately frozen at −80 °C.

Metabolite separation and detection were carried out using an ACQUITY I-class PLUS UPLC system (Waters) coupled with a QTRAP 6500+ (AB SCIEX) mass spectrometer equipped with an electrospray ionization source. Metabolites were separated on an ACQUITY Premier BEH Amide Vanguard Fit column (100 mm × 2.1 mm, 1.7 µm; Waters) with a constant column temperature of 35 °C. Separation of nucleotide sugars was achieved using the following liquid chromatography gradient scheme with mobile phase A (5 mM ammonium acetate in H₂O + 0.05% (v/v) ammonium hydroxide, pH 10) and mobile phase B (acetonitrile + 0.05% (v/v) ammonium hydroxide, pH 10). Data acquisition was performed using Analyst (v.1.7.2) (AB SCIEX) and processed with the OS software suite (v.2.0.0) (AB SCIEX).

C57BL/6J mice were exposed to 12 h NR feeding for 3 weeks and then sacrificed for tissue collection at ZT4 and ZT16. Small pieces of fresh liver tissue were transferred to ice-cold PBS and then into 2.5% glutaraldehyde in PBS. Fixed tissue was processed as described previously⁹² (method A), and sections were viewed without further on-grid staining or with a modified protocol involving incubation in 1% osmium tetroxide, followed by 1% aqueous uranyl acetate, before dehydration (method B). For the latter method, sections were stained with aqueous uranyl acetate, then with Reynolds lead citrate stain. The two different methods were used to ensure that different compartments were readily distinguishable at low magnification (method B providing optimal staining of rough ER-associated ribosomes). Random images of hepatocytes were captured across the section on a JEOL1011 transmission electron microscope at a primary magnification of ×10,000 for each biological replicate. Images were then analysed (46–105 per condition) by two independent investigators who were blind to the identity of the samples using the stereology quantification method in ImageJ to assess the volume density of the ER, GA and mitochondria relative to the cell volume.

Cells were lysed on ice in QIAzol Lysis Reagent, collected using a cell scraper and homogenized by passing several times through a syringe. Snap-frozen liver tissue was homogenized in 2 ml tubes with QIAzol Lysis Reagent and 1.4 mm ceramic beads (Cappella Science, 19-627) using a Precellys tissue homogenizer. RNA from cells and liver tissue was extracted using the RNeasy Plus Universal Mini Kit (Qiagen, 73404), following the manufacturer’s protocol. The purity, integrity and quality of the total RNA were assessed using an agarose gel and a NanoDrop spectrometer (Thermo Scientific). Only RNA with an A260/280 ratio of ≥2 and an A260/230 ratio of ≥1 were used for subsequent analysis.

cDNA synthesis was performed using 1 μg of RNA with the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems, 4368814). Gene expression levels were quantified using ABI SYBR GREEN PCR MASTERMIX (Applied Biosystems, 4312704) according to the manufacturer’s protocol. Primer sequences for all analysed genes are listed in Supplementary Table. 8

Transcriptomic analysis of RNA extracted from AML12 cells and mouse livers was performed and analysed using bulk RNA barcoding and sequencing (BRB-seq) performed at Alithea Genomics (Epalinges, Switzerland)⁹³. Raw reads were mapped using STAR on the GRCm38/mm10 genome assembly (Ensembl release 102). The BRB-seqTools suite (http://github.com/DeplanckeLab/BRB-seqTools↗) was subsequently applied to retrieve raw counts per gene.

Differential rhythmicity analysis was performed using the R package dryR²⁰. We fitted nested cosinor models using the functions dryseq (negative-binomial generalized linear model for count data) or drylm (ordinary linear regression) and selected the optimal model per gene by minimizing the Bayesian information criterion. Schwarz weights, calculated from the change in Bayesian information criterion values, were interpreted as approximate posterior probabilities for each model (for more details, see previous work²⁰). If only one condition was tested for rhythmicity, we used the f_24 function of the dryR package. Human somalogic data were fitted with a mixed linear model as previously described⁹⁴. Specifically, we used the lmer function from the lme4 R package on log₂-normalized relative fluorescence unit values. A 24 h harmonic regression model with a subject-specific random effect on the baseline was applied as follows: log₂(y_i,t) ~ cos(2π / 24t) + sin(2π / 24t) + (1|i), where y_i,t represents the log₂-normalized relative fluorescence units for subject i at time t. Model comparison between this full model and a null model (excluding harmonic terms) was performed using the likelihood ratio test from the lmtest R package. P values were corrected using the Benjamini–Hochberg method.

To analyse BRB-seq data, we used DESeq2 (ref. ⁹⁵) to statistically evaluate differences between the two treatment groups, with time included as a covariate in the model. For cell experiments, we also accounted for batch effects by including batch as a covariate. StageR⁹⁶ post hoc analysis was performed to further evaluate the statistical significance of treatment within timepoints. Wet-mass-normalized metabolite abundances from metabolic profiling of AML12 cells were analysed using a mixed linear model (~time × treatment + (1|batch)), where time represents the duration after treatment and treatment denotes CP-91149-treated or vehicle-treated samples. The model accounted for batch effects by including batch as a random effect. Post hoc analysis was performed using emmeans and stageR. Total paired-end RNA-seq data from previous work²⁶ was mapped using kallisto⁹⁷ to GRCm39/mm10 (Ensembl release 110) quantify transcript levels.

For non-depleted plasma proteomic analysis of Bmal1 WT and Bmal1-KO mice, we used maxLFQ to assess mean differences by comparing a full linear model (~time + genotype) with a reduced model (~time), where time is time of tissue collection.

For functional enrichment analysis, we performed hypergeometric testing or a mean-rank gene set enrichment analysis for standard gene sets, such as Gene Ontology Cellular Component and Biological Process, and custom gene sets based on a previous publication⁹⁸.

Unless otherwise stated, comparisons between two independent groups were made by unpaired two-tailed Student’s t-test for normally distributed data and by Wilcoxon rank-sum test for non-normal data. Rhythmicity within a single condition was assessed by cosine-fit linear regression using the f_24 function from the dryR package²⁰. To assess rhythmicity in temporally resolved measurements from human subjects, we used a linear mixed-effects model with harmonic regression, as previously described⁹⁴. Experiments with two factors (for example, time × genotype) were analysed by two-way ANOVA with Šidák/Šidák–Holm’s post hoc test, or with P value adjustment by the Benjamini–Hochberg method, as indicated. Phase distributions were compared by a two-sample Kolmogorov–Smirnov test. Correlations were evaluated by Pearson’s correlation coefficient. One-sample t-tests were used to test whether group means differed from zero. Overlap between protein, pQTLs or gene sets was assessed by hypergeometric testing, and when sample sizes were small, contingency tables were analysed by Fisher’s exact test. In cell-based and ex vivo assays, batch effects were modelled using mixed-effects linear models. We used StageR correction for post hoc comparisons if more than two timepoints were assessed. Unless otherwise specified, P < 0.05 was considered statistically significant. Details of the statistical analysis are available in Source Data figures and Source Data Extended Data Figs. 1–8.

No statistical methods were used to pre-determine sample sizes, but our sample sizes for animal studies were designed to capture 24 h rhythms follow previously published guidelines⁹⁹. The number of human participants was comparable to those reported in a previous blood proteomics study with similar sampling intervals and objectives for 24 h rhythm analysis¹⁶. For in vitro experiments, a minimum of n = 4 biological replicates was used. Mice were randomized to experimental conditions based on body weight, except for experiments involving Bmal1-KO and Cry1/2-KO animals, in which mice were randomly assigned. Experimental conditions were interleaved to avoid clustered treatment or sampling and order effects. Data collection and analysis were not performed blind to experimental conditions, except for electron microscopy image analysis, which was conducted by analysts blinded to experimental conditions. In addition, investigators were blinded to the injection treatments, so they were unaware whether a mouse received the vehicle or CP-91149 injection. All data were included in the analyses unless exclusion was required owing to documented technical errors during sample processing or non-concordant technical replicates; additionally, limited material precluded inclusion in certain analyses. Stereological analyses of electron microscopy images were assumed to be non-normally distributed and were analysed using non-parametric tests. Proteomics and SomaLogic data distributions were assessed using diagnostic visualizations, including Q–Q plots. BRB-seq count data were modelled using a negative binomial distribution as implemented in DESeq2, with standard quality control procedures applied. For all other data, including RT–qPCR, western and lectin blotting, ELISA, cytotoxicity assays, metabolite measurements and blood glucose analyses, data distribution was assumed to be normal but was not formally tested.

Further information on research design is available in thelinked to this article. Nature Portfolio Reporting Summary