Feeding-induced resistance to acute lethal sepsis is dependent on hepatic BMAL1 and FXR signalling

In mice, time of day strongly influences lethality in response to LPS, with survival greatest at the beginning compared to the end of the light cycle. Here we show that feeding, rather than light, controls time-of-day dependent LPS sensitivity. Mortality following LPS administration is independent of cytokine production and the clock regulator BMAL1 expressed in myeloid cells. In contrast, deletion of BMAL1 in hepatocytes globally disrupts the transcriptional response to the feeding cycle in the liver and results in constitutively high LPS sensitivity. Using RNAseq and functional validation studies we identify hepatic farnesoid X receptor (FXR) signalling as a BMAL1 and feeding-dependent regulator of LPS susceptibility. These results show that hepatocyte-intrinsic BMAL1 and FXR signalling integrate nutritional cues to regulate survival in response to innate immune stimuli. Understanding hepatic molecular programmes operational in response to these cues could identify novel pathways for targeting to enhance endotoxemia resistance.

C ircadian rhythmicity in the innate immune response was first reported in 1960 1 in the high-dose LPS model of sepsis, with mice exposed to LPS at zeitgeber time 0 (ZT0, beginning of the light cycle) having a survival advantage over those exposed at ZT12 (end of the light cycle). Since then, daily changes in lethality have been described upon stimulation with a variety of infammogenic materials 2,3 , and recent years have seen advances in understanding the molecular and cellular basis of this phenomenon, particularly building on the discovery of the cellintrinsic circadian clock [4][5][6][7] . The myeloid clock has thereby received much attention and was proposed to be decisive in regulating mortality due to circadian variability in the production of inflammatory cytokines regulated by components of the cellintrinsic clock, such as BMAL1 4,5 .
The last decade has also seen an increased understanding of metabolic cues in inflammation. While reversal of the feeding cycle can reprogramme circadian clock-controlled genes in the periphery 8,9 , the influence of the feeding compared to the light cycle on daily immune responses is not known. However, it should be noted that the circadian machinery integrates multiple environmental factors, such as the feeding and sleep-wake cycles that align under natural conditions, making it impossible to decipher the contributions of a single zeitgeber (an environmental signal that can regulate the internal clock) without specifically adjusted experimental settings. It is therefore likely that other factors will influence the host response to acute inflammatory stimuli beyond the light cycle. Indeed, metabolic cues such as prolonged fasting and the Farnesoid X-receptor, a component of the feeding-regulated bile acid cycle have been recently shown to influence inflammatory responses and survival in sepsis models 10,11 .
Importantly, sepsis is a leading cause of death in intensive care units and patient care is currently solely supportive due to a lack of effective therapies. Although inflammation is the foundation of sepsis, the exact cause of mortality is not clear 12 and previous efforts using anti-cytokine antibodies, such as anti-TNF, anti-IL-1 and anti-IL-6, have remained unsuccessful in humans [13][14][15] . Understanding novel aspects and pathways is therefore urgently needed to find new targets to enhance sepsis survival. In this study, we reveal a critical role for the feeding cycle in regulating survival in the LPS model of sepsis. This feeding-cycle regulated susceptibility depends on BMAL-1 and FXR signalling in hepatocytes, identifying potential new targets for therapeutic intervention.

Results
Lethality in the LPS model of sepsis follows the feeding, not the light cycle. To dissociate the role of the feeding from the light cycle in LPS susceptibility, we studied mice under time-restricted feeding (TRF) conditions. Food access was restricted to either the dark phase (night-time fed, NF), the predominant feeding time for nocturnal species, or the light phase (day-time fed, DF), where food and light cycles become dissociated (Fig. 1a). As expected, activity paralleled the light cycle ( Supplementary Fig. 1a), and night-time feeding closely approximated feeding patterns of ad libitum food availability ( Supplementary Fig. 1b). Food consumption and activity levels were comparable between groups ( Supplementary Fig. 1a, b) and the respiratory exchange ratio (RER) was >1 in the fed state, reflecting a switch to glucose utilization, decreasing to~0.8, indicative of fatty acid oxidation, with food-deprivation (Fig. 1b). The RER thus closely paralleled the feeding rather than the light cycle. TRF mice displayed similar maximal and minimal RER, while differences remained less pronounced in ad libitum-fed animals (Fig. 1b).
We then challenged mice conditioned for 4 days to NF or DF schedules with LPS at ZT0 vs. ZT12 (Fig. 1a). Mice were monitored frequently for behavioral changes due to sepsis and euthanized at pre-specified morbidity (Supplemental methods). Congruent with the previous reports 1 , lethality was greater in NF animals following the LPS challenge at ZT12 (Fig. 1c). Strikingly, in DF mice, susceptibility was reversed, with 87% mortality following ZT0 LPS administration versus 82% survival following ZT12 administration (Fig. 1d). While LPS susceptibility, in concordance with previous reports 16 , was generally less pronounced in male than female mice, feeding timedependent sensitivity was evident in both sexes ( Supplementary  Fig. 1e-h). The majority of subsequent experiments were therefore performed on mixed populations (female data points in black and males in red), as has been done in comparable studies 17 . The regulation of susceptibility by feeding cycle persisted whether the food was provided ad libitum after LPS or remained restricted according to the prior feeding schedule ( Supplementary Fig. 1c, d), indicating that the outcome was not determined by the availability of food post-stimulation. It should be noted, that in concordance with previous findings 18 , food intake was significantly blunted following LPS stimulation ( Supplementary Fig. 1i). As food intake remained voluntary, this phenotype may not be applicable to studies investigating anti-anorexic forced feeding effects 19 . Systemic cytokine production, a measure of the inflammatory state, was not correlated with food-regulated LPS susceptibility, with similar levels of proinflammatory cytokines IL1β, IL6, TNF, the chemokine KC, and the counter-regulatory cytokine IL10, regardless of the feeding schedule at baseline and following LPS ( Fig. 1e-j, Supplementary Fig. 1j-o). Soluble CD14, a biomarker for systemic LPS exposure, was also similar between feeding groups at baseline ( Supplementary  Fig. 1p) and after LPS challenge ( Supplementary Fig. 1q). Thus, the feeding cycle influence on acute mortality after LPS appears not to be dependent on inflammatory cytokine response or LPS bioavailability.
Profound hypoglycaemia is associated with increased LPS sensitivity. LPS additionally induces metabolic reprogramming, including a shift to glycolysis 20 . As systemic hypoglycaemia can be lethal in the absence of compensatory gluconeogenesis 21 , we studied glucose metabolism in LPS-exposed mice after different feeding schedules. Mice susceptible to LPS (NF-ZT12, DF-ZT0) developed profound hypoglycaemia of <50 mg/dl, significantly less than mice in the resistant groups, which averaged 100 mg/dl (Fig. 2a). Kinetic analysis showed that blood glucose began falling as early as 2 h after LPS challenge, and in animals resistant to LPS, began to rise after 6 h ( Supplementary Fig. 2a). TRF itself was not sufficient to affect baseline glucose levels ( Supplementary Fig. 2b), but fasted groups (NF12 and DF0) had significantly depleted liver glycogen stores (Fig. 2b), which contribute to ineffective maintenance of glucose levels after LPS challenge. Fasting also induced ketogenesis, resulting in the production of β-hydroxybutyrate (BHB), which was reported to inhibit NLRP3 inflammasome activation 22 . In TRF mice, BHB increased during the foodrestricted phase in DF and NF groups (Fig. 2c). LPS, independent of feeding status, further induced BHB, with higher levels at LPSsusceptible time points (Fig. 2d). This supports the notion of metabolic distress after LPS challenge in susceptible groups and shows that elevated BHB is not sufficient to reduce mortality or inflammatory cytokine levels in this model. Endogenous corticosteroids influence diurnal cycles of cytokine production 23 and might also protect mice from LPS challenge. However, peak serum levels of corticosterone, the principal active circulating corticosteroid in mice, was found at ZT12 in nocturnally fed mice, and ZT0 in DF mice, when susceptibility to LPS is the greatest (Fig. 2e). As expected, glucocorticoids rose after the LPS challenge but were not different between feeding groups and time of stimulation nor the post-stimulation feeding regimen (Fig. 2f), showing that endogenous corticosteroids are not likely a key protective factor in food-cycle controlled LPS-induced mortality.
The liver-intrinsic circadian clock rapidly adapts to changes in the feeding cycle. To better understand the influence of food and light cycle on the circadian clock and downstream metabolic and inflammatory outputs, we measured expression of key circadian clock components in the liver and spleen in NF or DF mice ( Fig. 3a, b). In the liver, time-of-day-dependent hepatic expression of many clock genes and selected metabolic and inflammatory regulators were clearly reversed at ZT0 and ZT12 in DF animals. In contrast, DF merely dampened the patterns of gene expression in spleen, in macrophage-enriched peritoneal exudate cells, and adipocytes at the selected time-points on day 5 of TRF, indicating greater liver transcriptional sensitivity to feeding cues ( Supplementary Fig. 2c-f).  Myeloid BMAL1 is not required for feeding cycle-dependent LPS mortality. Macrophages play an important role in sensing innate immune stimuli and previous reports linked diurnal changes in the inflammatory response to the myeloid compartment 4,6,24 . However, in concordance with no clear transcriptional adaptation to reversed feeding cycles in peritoneal macrophages, deletion of BMAL1 in the myeloid lineage did not alter the food-regulated sensitivity to LPS, as reversal of the feeding cycle in LysM Cre × BMAL1 loxP/loxP (Bmal1 ΔMye ) mice reversed the peak sensitivity to LPS from ZT12 to ZT0 (Fig. 3c, d), The myeloid clock is dispensable for feeding time-dependent mortality. a, b Liver (a) and spleen (b) gene expression profile of selected clock genes measured by NanoString on day 5 of TRF at indicated times in female mice. Unsupervised gene clustering. c, d Morbidity in night-time (c) or daytime (d) fed myeloid cell specific Bmal1 ΔMye (solid) (NF ZT0 vs. 12 p = 0.0032, DF ZT 0 vs. 12 ns, ZT0 NF vs. DF p = 0.0005, ZT12 NF vs. DF ns), or control Bmal1 fl/fl (dashed line) mice (NF 0 vs. 12 p = 0.0008, DF 0 vs. 12 p < 0.0001, ZT0 NF vs. DF p < 0.0001, ZT12 NF vs. DF p < 0.0001) following 20 mg/kg LPS i.p. on day 5 of TRF at ZT0 or ZT12 as indicated (Bmal1 ΔMye vs. respective Bmal1 fl/fl all ns). Statistical analysis according to Mantel-Cox. Experimental numbers:   (Fig. 3e-h). Although this result contrasts with previous reports of enhanced sensitivity to LPS in myeloid cellspecific BMAL1 deficiency 4,5 , our findings are in a model with earlier onset of symptoms, and are supported by a recent study 25 , which also challenges the notion of enhanced mortality in these conditional knockout mice.
Daily changes in LPS sensitivity depends on BMAL1 in hepatocytes. Given the importance of the liver in metabolism and dependence of clock gene expression on the feeding cycle, we deleted hepatocyte BMAL1 in Albumin Cre × BMAL1 loxP/loxP (Bmal1 ΔHep ) mice, which effectively disrupted time-of-daydependent gene expression of BMAL1 and core hepatic clock genes (Fig. 4a). Remarkably, these mice entirely lost food and time dependency in their susceptibility to LPS, with >90% of mice in all groups succumbing to LPS (Fig. 4b, c). Deletion of BMAL1 in hepatocytes did not alter baseline glucose levels ( Supplementary  Fig. 3a), the hypoglycaemia in NF mice challenged with LPS at ZT12 and DF mice at ZT0 (Fig. 5a), or glycogen stores in the fed vs. fasted state (Fig. 5b). Gluconeogenesis, as measured by pyruvate tolerance test (PTT) (Supplementary Fig. 3b), as well as liver expression of Pepck and Pygl ( Supplementary Fig. 3c, d), and a selection of metabolic markers in the serum of Bmal1 ΔHep mice also mirrored Bmal1 fl/fl values under similar feeding conditions (Supplementary Fig. 4a-g). Bmal1 ΔHep mice further had no spontaneous phenotype as reflected by appearance, lean and fat mass (as defined by EchoMRI), RER or food intake ( Supplementary  Fig. 5a-c). Measurement of serum cytokines in Bmal1 ΔHep mice revealed that levels of IL-6, which is predominantly synthesized by hepatocytes after LPS 26 , were significantly higher in Bmal1 ΔHep mice after LPS challenge compared to Bmal1 fl/fl (Fig. 5c) or Bmal1 ΔMye (Fig. 3e) mice, while other cytokines as well as soluble CD14 at baseline and post LPS mirrored Bmal1 fl/fl levels ( Fig. 5d-f, Supplementary Fig. 5d, e). These data suggest that BMAL1 in hepatocytes may contribute to an attenuation of the inflammatory response, at least in part by dampening IL-6 production. The mechanism and biological consequences of this elevated, yet feeding-time-dependent IL-6 response in the Bmal1 ΔHep mice warrants further investigation.
RNAseq correlates hepatic FXR signalling with BMAL1dependent sepsis mortality. To better understand the pathways underlying feeding-regulated LPS susceptibility in the presence and absence of BMAL1 and a functional molecular clock in hepatocytes, we performed RNAseq, which revealed tight clustering of gene expression in the 12-h starved (NF-ZT12, DF-ZT0) and fed (NF-ZT0, DF-ZT12) groups in Bmal1 fl/fl and Bmal1 ΔHep , with the groups protected in the LPS sepsis model (Bmal1 fl/fl NF-ZT0, DF-ZT12) separated from susceptible groups by the strongest principal component ( Supplementary Fig. 6a). More than 80% of 549 genes regulated in the liver by the feeding cycle in Bmal1 fl/fl mice lost differential expression in Bmal1 ΔHep animals (Supplementary Table 2, Supplementary Fig. 6b). Pathway analysis of differentially expressed (DE) genes comparing LPS resistant and susceptible Bmal1 fl/fl groups identified upregulation of farnesoid X receptor (FXR) and retinoid X receptor (RXR) signalling pathways with LPS resistance, whereas these pathways were not feeding-regulated in the Bmal1 ΔHep liver (Supplementary Fig. 6c, d). Strikingly, when we compared gene expression among all groups of Bmal1 fl/fl and Bmal1 ΔHep mice, all BMAL1deficient groups clustered with LPS sensitive Bmal1 fl/fl mice (Fig. 6a), with 232 DE genes between Bmal1 fl/fl LPS-resistant mice and the other LPS susceptible groups (Supplementary Table 3). Pathway analysis of these DE genes revealed similar pathways correlating with LPS resistance, including FXR and RXR signalling (Fig. 6b).
Hepatic FXR expression regulates feeding-cycle-dependent LPS sensitivity. We tested the function of the FXR pathway in feeding-cycle regulation of LPS response by hepatocyte-specific disruption of the Fxr gene (Fxr ΔHep ). At the same dose of LPS which induced high mortality in C57BL/6J mice, Fxr ΔHep and control Cre-negative littermates did not reach disease scores requiring euthanasia. However, using disease severity scores, hepatocyte-specific FXR disruption rendered both night-and DF animals as sensitive to LPS as the susceptible (DF) Fxr fl/fl group following ZT0 stimulation (Fig. 6c). Mice in the DF group were equally hypoglycaemic irrespective of FXR signalling status, but in this non-lethal setting, IL-1β and IL-6 levels were higher in the DF groups in both Fxr fl/fl and Fxr ΔHep mice (Fig. 7a-c), recapitulating previous reports in WT mice 27 and arguing for a ceiling effect which dampened differential cytokine production at lethal doses of LPS, as seen in Fig. 1e-j. With a higher dose of 35 mg/kg LPS, Fxr ΔHep animals in the fed state (DF-ZT12) were still protected from overall mortality, but had increased mortality at timepoints before 20 h compared to Fxr fl/fl DF mice following ZT12 stimulation (Fig. 7d). These results demonstrate that at early time-timepoints, FXR deficiency in the liver (FXR ΔHep , WT 12h-food deprived), partially overcomes the protective effect of feeding on LPS sepsis. This is supported by Gehan-Breslow-Wilcoxon statistical analysis, which is more sensitive to early events. Using this method, FXR ΔHep mice no longer show a statistical difference in survival between DF and NF groups. In agreement with the constitutively high serum cytokine levels noted after LPS administration regardless of mortality rates in C57BL/6J mice (Figs. 1e-j, 2k-n), serum cytokines were comparably high in Fxr ΔHep and control mice (Fig. 7e, f). These results support a role for FXR in protecting mice from LPSinduced lethal sepsis as they transition out of the feeding period. Given the importance of FXR signalling in promoting glucose utilization 28 , hepatic FXR deficiency may allow euglycemic mice in the fed state to phenocopy hypoglycaemic counterparts. At later time points, where the mortality of DF Fxr ΔHep converged towards DF Fxr fl/fl levels (Fig. 7d), compensatory pathways may become more critical. RNAseq analysis also identified the LXR pathway as feeding and BMAL1 dependent in the mouse liver (Fig. 6b). Given the interplay between the FXR and LXR receptors, and enhanced glucose utilization upon LXR activation 29,30 , both pathways may contribute to protective glucose metabolism and survival in mice with available glycogen stores. Susceptibility of FXR/LXR double-deficient animals in the time-restrictedfeeding model, remains to be determined and will aid in unravelling the exact molecular mechanism.

Discussion
In summary, these findings point to BMAL1 in hepatocytes as a key transducer of nutritional cues into outputs which modulate susceptibility to the lethal effects of LPS. Enhanced susceptibility was further associated with more severe hypoglycaemia rather than enhanced serum cytokines. We have identified the FXR/RXR pathway as a feeding-cycle and BMAL1-dependent pathway and a possible regulator of LPS susceptibility, and confirmed this with functional studies in hepatocyte-specific FXR knockout animals. While LPS stimulation is commonly used to induce TLR4mediated immune responses and is one model for bacterial sepsis, the translational aspects of these findings in more complex, clinically relevant models, such as CLP, will be of particular interest in future studies, and will allow to dissociate the role for bacterial resistance and tolerance. Given the close relation of food intake and time-of-day under homoeostatic conditions (Supplementary Fig. 1b AL group), our findings are of direct pertinence to the previously described time-of-day-dependent changes in CLP-induced sepsis lethality 17 . The relevance of timed feeding for activation of various immune pathways and the role of food composition in humans will require further investigation and may identify targets to enhance endotoxemia resistance and improve the currently poor survival in severe sepsis.
One of the characteristics of a circadian response is sustained cycling after entrainment. We cannot rule out the possibility that LPS susceptibility depends only on the antecedent feeding state or liver Bmal1, rather than being entrained in a circadian manner. However, previous studies with a single 48-hour food deprivation showed reduced hepatic injury and cytokine release in response to LPS 10 , arguing that reversal of a 12-hour feeding cycle has divergent effects on inflammation compared to more prolonged starvation. Our data rather demonstrates, that while feeding dependent availability of glucose and glycogen is a prerequisite, optimal protection from sepsis mortality requires the presence of BMAL1 regulated FXR. In support of this, ad libitum fed animals generally do not phenocopy food deprived animals at any time of the day, but still show differences in mortality at ZT0 vs. ZT12 1,4,31 . Together, these findings change the paradigm of our understanding of diurnal sensitivity to innate immune stimuli and provide new pathways connecting nutritional intake to host defense and adaptive responsiveness to LPS. Fxr Alb-Cre colony. Conditional Fxr knockout and control animals were generated and provided by F. J. Gonzalez 32 . Animals were maintained on a C57BL/6J;129S1 background.

Methods
General animal holding. Animals at 9-13 weeks of age were housed in 12 h:12 h light:dark conditions (~70 Lx at times of illumination) at least one week prior to use, in rooms accessible under red lights during the dark period. Experiments were mainly conducted using male and female mice. No difference was apparent between sexes (red vs. black data points). If not indicated otherwise, only female mice were used.
Time restricted feeding (TRF). Mice were housed in a 12:12 light:dark environment on ad libitum feeding for at least one week, before food was restricted to either the light or the dark phase (Damiola et al., 2000) for 4 days if not specified otherwise. On day 5, mice were either stimulated or killed for tissue collection at ZT0 or ZT12.
LPS mortality. Mice were acclimatized to the TRF schedule for 4 days. 24 h before the subjects received an intra peritoneal injection of 20 mg/kg LPS (Sigma, Cat# L2630 Lot# 014M4018V) (sonicated for 5 min prior to preparation in sterile PBS), serum was collected via the submandibular route for baseline level determination. Following stimulation, mice received ad libitum feeding if not indicated otherwise. 2 or 6 h post injection a second submandibular bleed was performed for serum CLAMS, comprehensive lab animal monitoring system. The CLAMS (Columbus Instruments) is an enclosed animal holding system that allows measurement of various metabolic parameters such as oxygen consumption, carbon dioxide production, locomotor activity as measured by beam breaks in all three dimensions, and food consumption. Food access as well as timing and duration of the light phase was regulated and adapted to experimental requirements. 12 mice were single housed and data was acquired from each individual separately. Peak RER values continuously increased, but as activity recordings were not indicative of intense activity, an activator bicarbonate buffering which elevates the RER > 1 (Supplementary Fig. 1a), and given that this increase was present in time-restricted and ad libitum-fed animal, suggests a technical rather than biological cause.
RNA analysis RNA preparation. Tissue samples were temporarily stored at −20°C in RNAlater stabilization solution (Ambion, Cat# AM7021) to preserve tissue RNA integrity. The sample was homogenized in a Micro-Bead tube 2.38 mm (MoBio, Cat# 13117-50) containing 1 ml Trizol Reagent using a Precellys 24 and subsequently centrifuged at 12,000 × g for 10 min at 4°C. The clear supernatant (without the fatty layer if present) was transferred to a new vial. RNA was isolated using TRIzol Reagent (Life Technologies, Cat# 15596-018) according to the manufacturer's instructions. Homogenized samples containing 1 ml TRIzol were incubated for 5 min at RT to allow denaturation of proteins before 0.2 ml chloroform was added. During centrifugation at 12,000 × g for 15 min at 4°C, the RNA containing aqueous phase (top phase) was separated from the DNA at the interphase and protein in the (bottom) organic phase. RNA was precipitated from the aqueous phase during incubation with 0.5 ml isopropanol. Following centrifugation at 12,000 × g for 10 min at 4°C, the supernatant was discarded and the pellet washed with 1 ml 75% ethanol, before the pellet (at 7500 × g for 5 min at 4°C) was air-dried for~10 min at RT. The purified RNA was then resuspended in an appropriate amount of ultra pure water (10-100 µl) and incubated for 12 min at 57°C to facilitate complete solubilization of the RNA. In order to remove DNA contaminations, the samples were digested with DNase using Turbo DNA-free Kit (Life Technologies, Cat# AM1907) according to the manufacturer's instructions. 0.1 volume of DNase buffer was added to the sample together with 1 µl of DNase and incubated for 30 min at 37°C, before 0.1 volume of inactivation reagent was added. Following a 5 min incubation at RT, the sample was centrifuged at 10,000 × g for 1.5 min to pellet the DNase inactivation reagent. The DNase digested RNA was then transferred into a fresh vial and stored at −80°C until further use.
NanoString. NanoString technology was used according to the manufacturer's instructions. Genes involved in the circadian machinery, metabolism and immune response were measured amongst five housekeeping gene mRNAs (Decr1, Fpgs or Eif4a2, Hmbs, Hprt1, Ppib). RNA (800 ng) was mixed with 8 µl Reporter CodeSet in a total volume of 13 µl, before 2 µl Capture ProbeSet was added and the reaction immediately incubated for~18 h at 65°C in a thermal cycler. The hybridized samples were then loaded to the NanoString PrepStation for further processing and data collection in the NanoString Digital Analyzer. Target sequences of the CodeSet can be found in Supplementary Method Table 2.
RNA sequencing. RNA-seq libraries of purified liver RNA were prepared using NEBNext Ulta II RNA Library Prep Kit for Illumina (New England BioLabs, Cat# E7770S) and sequenced on Illumina Hiseq 3000 with 50 base single end reads. Illumina BCL sequencing results were demultiplexed and converted to FastQ using b2fastq version 2.17.1.14. Reads were mapped to mm10 using Tophat 2.1.1 and RPKM calculations made using Partek Genomic Suite 7.18.0723. ANOVA comparisons were performed on Log2-transformed RPKM (with a 0.1 offset). Cutoffs were based on Fold Change < 2 or <−2; FDR < 0.2 and an RPKM value in at least one animal of >1. Pearson correlations, heat maps and PCA plots were generated using Partek GS. Ingenuity pathway analysis was used to analyse filtered ANOVA gene lists for potentially relevant pathways. For Canonical Pathway results, the height of the bar represents the −log p-value of the enrichment of genes within the pathway. The connected line shows the ratio, which is the number of overlapped genes in the pathway vs. the total number of genes in the pathway. The colour of the bar represents the z-score, IPA's prediction of whether the pathway is induced or suppressed based on fold changes.
For RNA-Seq analysis ( Fig. 6 and Supplementary Fig. 5), raw FastQ files and processed RPKM table and ANOVA results have been submitted to NCBI GEO under accession number GSE123909.
Magnetic bead separation for F4/80 positive cell enrichment. Macrophages were isolated from the peritoneal lavage by negative magnetic isolation, using the EasySep system. To ensure the shortest possible purification times, Mouse Streptavidin Rapidspheres (StemCell Technologies, Catalogue # 19860) were used in conjunction with a custom selection cocktail. The peritoneal cells were pelleted at 805 g for 5 min at 4°C and subsequently resuspended in 100 µl PBS in a polystyrene FACS tube. Unspecific binding was prevented using 2.5 µl Fc-block ((Bioxcell Cat# CUS-HB-197-A02) 1.2 mg/ml) for 5 min prior to incubation with biotinylated antibodies directed against B-cell (CD19 (BioLegend, Cat# 115504, RRID:AB_313639) & B220 (BioLegend, Cat# 103204, RRID:AB_312989), 0.5 µg), T-cell (Thy1.2, (BioLegend, Cat# 140313), 1 µg) and erythrocyte (TER119 (Bio-Legend, Cat# 116204, RRID:AB_313705), 1 µg) surface marker for 10 min. The magnetic Rapidspheres were vortexed 1 min before 1.5 µl were added to the isolation mix for 2.5 min. The reaction was resuspended in 2.5 ml PBS and inserted into a purple EasySep magnet for an additional 2.5 min before the purified macrophage suspension was carefully decanted into a new tube. To ensure homogenous distribution, the reaction was vigorously mixed after every addition of reagents.
Blood works Serum preparation. Serum was separated from whole blood in Z-gel micro tubes (Sarstedt Cat# 41.1500.005) after a 30 min incubation and centrifugation at 10,0000 × g for 5 min. Serum was stored at −20°C until use.
Glycogen measurement. Liver glycogen content was measured using Glycogen Assay Kit (Sigma, Cat#MAK016) according to the manufacturer's instructions for the colorimetric detection. In short, 10 mg of liver was homogenized in 100 µl water on ice and boiled for 5 min. Following 5 min at 13,000 × g, the supernatant was collected and brought to a final volume of 50 µl with Hydrolysis Buffer. 2 µl of hydrolysis enzyme mix was added, the reaction mixed well and incubated for 30 min at RT. Blanks were set up for each reaction by omitting the hydrolysis enzyme mix. After addition of 50 µl master reaction mix, wells were mixed and incubated for 30 min at RT in the dark. The absorbance was measured at 570 and glycogen content was calculated by linear regression from the standard curve.
Glucose measurements. Serum glucose was measured using a commercially available kit according to the manufacturer's instructions (Abcam, Cat# ab65333). In short, 0.5-1.5 µl serum was diluted with Glucose Assay Buffer to a final volume of 50 µl and mixed in a 96-well flat bottom plate with glucose reaction mix. Following an incubation for 30 min at 37°C in the dark the OD was measured at 570 nm in a microplate reader. Glucose concentrations were then calculated by linear regression from the standard curve.
Cytokine measurement. In vivo cytokines were measured at the indicated time points by terminal blood collection following decapitation or via the submandibular route for survival bleeds. Serum was separated in Z-gel microtubes (Sarstedt Cat# 41.1500.005) after a 30 min incubation and centrifugation at 10,000 × g for 5 min. Serum was stored at −20°C until use. Cytokines were measured using the Bio-Plex system (BioRad) in a multiplex assay according to the manufacturer's instructions. In short, appropriate magnetic capture beads were washed twice with wash buffer in a magnetic plate washer, before the samples and standards were added to the plate and incubated for 30 min while shaking. Following three washes, the respective biotinylated detection antibodies were added for 30 min while shaking, before the plate was washed three times. Lastly, PE-conjugated Streptavidin was added for 10 min before the magnetic beads were washed three times and resuspended in 125 µl assay buffer. The conjugates were then analysed using a Bio-Plex MagPix Multiplex Reader (BioRad). Serum metabolic markers (Supplementary Fig. S4) were similarly measured using this protocol.
CD14 measurement. Soluble CD14 levels were quantified using a commercially available ELISA kit (R&D Cat# MC140) according to manufacturer's instructions. In short, samples were diluted 1:100 and 1:1000 by serial dilution in calibrator diluent, standards were prepared as instructed in calibrator diluent. Assay diluent was added to the microplate pre-coated in a monoclonal antibody specific to mouse CD14. Standards, diluted samples, and sCD14 controls were then added to the plate and incubated for 2 h while shaking. Plate was washed four times before a polyclonal antibody specific for mouse CD14 conjugated to horseradish peroxidase was added then incubated for 2 h while shaking. Wash steps were repeated, and a substrate solution consisting of equal parts stabilized hydrogen peroxide and stabilized chromogen (tetramethylbenzidine) were added and incubated for 30 min without shaking. A stop solution of hydrochloric acid was added and the plate was read at 450 nm. sCD14 measurements were calculated by plotting against the standard curve and multiplying by the dilution factor.
Corticosterone measurement. Corticosterone levels were quantified using a commercially available ELISA kit (Enzo Cat# ADI-900-097). In short, serum samples were incubated with steroid displacement reagent to free bound Corticosterone from steroid binding proteins. 1:40 diluted samples and freshly prepared Corticosterone standards were then loaded to a Donkey anti-Sheep IgG Microtiter Plate and incubated on a plate shaker for 2 h at 453 × g RT. Following three washes, pNpp Stop solution was added and absorption was read at 405 nm, while absorption at 580 nm was used for correction. Corticosterone concentrations were extrapolated from a percent bound versus corticosterone standard plot.
β-Hydroxybutyrate measurement. β-Hydroxybutyrate was measured using a commercially available colorimetric assay (Cayman, Cat# 700190). Oxidation of βhydroxybutyrate to acetoacetate results in simultaneous reduction of NAD + to NADH + H + . The colorimetric detector WST-1 then reacts with NADH to form a formazan dye, whose absorbance at 450 nm is proportional to the βhydroxybutyrate concentration. To initiate the reaction, serum samples and βhydroxybutyrate standards were incubated for 30 min at RT in the dark with Developer solution. Absorbance of β-hydroxybutyrate standards was plotted against the concentrations followed by linear regression, from which sample concentrations were extrapolated.
Statistics. Error bars represent the standard error of the mean. Statistical analysis was performed using appropriate tests in Prism7. Based on the relatively small sample sizes, non-parametric analysis (Kruskal-Wallis or Mann-Whitney) were used for one-factorial testing. Due to lack of readily available two-factorial nonparametric tests, parametric two-way ANOVA was performed for two-factorial testing. The authors refrained from using higher order statistical analysis (e.g. three-way ANOVA) as the complexity of the results was not proportional to the addition of useful information they would provide. Statistical testing of survival data was performed using Mantel-Cox and where appropriate (due to early/late time-point differences) Gehan-Breslow-Wilcoxon analysis. CLAMS data was analysed using CalR 1.2 (https://calrapp.org) 34 . Data points represent distinct samples in all figures.

Data availability
Source data are provided with this paper, Raw FastQ files and processed RPKM table and ANOVA results have been submitted to NCBI GEO under accession number GSE123909. Source data are provided with this paper.