Referenced abstract
The interaction between high-fat diet (HFD) feeding and the gut microbiome has a strong impact on the onset of insulin resistance (IR)1-3. In particular, bacterial lipopolysaccharides (LPS) and dietary fats trigger low-grade inflammation4 through activation of Toll-like receptor 4 (TLR4), a process called metabolic endotoxemia5. However, little is known about how the microbiome can mitigate this process. Here, we investigate longitudinal physiological and metabotypical responses of C57BL/6 mice to HFD feeding. A series of in vivo experiments with choline supplementation, then blocking trimethylamine (TMA) production and administering TMA, demonstrate that this microbiome-associated metabolite decouples inflammation and IR from obesity in HFD. Through in vitro kinome screens and in silico molecular dynamics studies, we reveal TMA specifically inhibits Interleukin-1 Receptor-associated Kinase 4 (IRAK-4), a central kinase integrating signals from various TLRs and cytokine receptors. Consistent with this, genetic ablation and chemical inhibition of IRAK-4 result in similar metabolic and immune improvements in HFD. In summary, TMA appears as a key microbial effector inhibiting IRAK-4 and mediating metabolic and immune effects with benefits upon HFD. Thereby we highlight the critical contribution of the microbial signalling metabolome in homeostatic regulation of host disease and the emerging role of the kinome6 in microbial–mammalian chemical crosstalk.
Main Text
IR is key to a cluster of conditions such as impaired glucose tolerance (IGT), obesity, hyperlipidemia, hypertension and non-alcoholic fatty liver disease (NAFLD), increasing the risk of developing type 2 diabetes and cardiovascular diseases7. One of the hallmarks of these disorders is the early development of a systemic low-grade inflammation4. Increasing evidence supports that complex communication occurs between the gut microbiota and the innate immune system8, with consequences on metabolic homeostasis9. Whilst some of the functional signalling molecules mediating microbial–mammalian chemical crosstalk have been characterized, a limited number of metabolite classes and their targets have been identified (G protein-coupled10 and nuclear11 receptors). It is, however, hypothesized that microbial metabolites potentially affect other pharmacological target classes such as kinases6. In previous studies, we and others have identified a family of microbiome-associated metabolites, methylamines, associated with IR, NAFLD12 and atherosclerosis13, but their mechanisms of action on mammalian hosts remain unclear. TMA is one of the most abundant microbial metabolites produced by the gut microbiota. We first reported that TMA may be associated with IR12. TMA results from microbial metabolism of dietary choline, carnitine and TMAO14-18 before being absorbed and N- oxidized into TMAO by hepatic flavin-containing mono-oxygenase 3 (FMO3)19. After initial reports associating TMAO with adverse cardiovascularoutcomes13,16, it has since emerged that Fmo3 inactivation was beneficial for several metabolic outcomes20-23, strongly suggesting that TMA and TMAO have distinct biological roles.
Here, we present a new paradigm to decipher the microbiome–host kinome chemical crosstalk in IR: i) identification of gut-derived microbial metabolites associated with HFD- induced IGT and obesity, ii) systematic pharmacological target screening of discriminant microbial metabolites, iii) molecular modelling of the interaction, and iv) mechanistic validation of the pathophysiological relevance of pharmacological interactions with in vivo models, thereby unravelling novel mechanisms by which gut microbial TMA acts as an IRAK- 4 inhibitor and directly improves the host’s immunometabolism.
Longitudinal study leading to the choline supplementation hypothesis
We initially carried out a longitudinal pathophysiological monitoring in a cohort of C57BL/6 mice (n=472) rapidly developing obesity and IGT when fed a 65% kCal HFD (Suppl.Fig. 1a-c). Liver transcriptomics identified 1,641 significant differentially expressed genes between chow-fed (CHD-) and HFD-fed mice (Suppl. Fig. 1d,e and Suppl. Tab. 1). Gene ontology and signalling pathway impact analyses (Suppl. Fig. 2, Suppl. Tab. 2-4) demonstrated upregulation of protein processing in the endoplasmic reticulum, whereas carbohydrate metabolism, circadian rhythm and AMPK signalling were significantly inhibited, consistent with existing literature24,25. Surprisingly, inflammation-associated pathways were not differentially affected: expression of genes coding for acute phase serum amyloid A proteins Saa1 and Saa2 in response to acute inflammation and associated with type 2 diabetes was reduced whereas expression of pro-inflammatory cytokines such as Il6 and Il1β was not significant (Suppl. Fig. 1d and Suppl. Tab. 4). This is contrasting with established knowledge as low-grade inflammation is one of the key features associated with IR4,5. Metabolic profiling of the urinary metabolome of this mouse cohort (CHD and HFD, 4 time-points) by
1H-NMR spectroscopy followed by multivariate modelling showed diet is the main factor influencing the metabolic profiles, followed by age (Suppl. Fig. 1f), the model’s goodness-of-prediction parameters being highly significant upon permutation testing (Suppl. Fig. 1g). The metabolic signature of HFD-feeding compared to CHD was mainly associated with TMA excretion (Suppl. Fig. 1h), consistent with our previous reports using this diet12,26. We used variance components analysis to decompose the contribution of diet and age to the excretion of these three metabolites, showing that the HFD contribution overwhelms the contribution of age (Suppl. Fig. 1i). Since the HFD contains >15 g of choline per kg of diet, this led us to hypothesize that choline supplementation may explain some of the metabolic and immune variation in response to the HFD.
Choline supplementation corrects HFD-induced inflammation and IR
To investigate the effect of choline supplementation on glucose tolerance and IR, we initiated a series of in vivo, in vitro and in silico studies. We first fed C57BL/6 mice with a 65% kCal HFD containing 2 g/kg or 17 g/kg of choline (i.e., LC-HFD and HC-HFD, respectively; Fig. 1). Body weight (BW) was significantly increased in mice fed a LC-HFD from weaning until 5 months of age compared to mice fed a chow diet and remarkably compared to mice fed HC-HFD (Fig. 1a), which was not related to food intake (Suppl. Fig. 3a). To evaluate metabolic homeostasis, we performed glucose tolerance tests (GTT), which displayed a similar pattern with normalisation of cumulative glycemia on HC-HFD compared to LC-HFD (Fig. 1b-c). We then profiled circulating pro-inflammatory cytokines such as IL6, IL1β, TNFα (Fig. 1d-f) and phosphorylation of NF-kB regulating their transcription (Suppl. Fig. 3b,c), which was also suggestive of an improvement of HFD-induced inflammation by choline supplementation.
Targeted analysis of choline and methylamine pathways
To document the effect of choline supplementation on choline-related metabolic pathways, we further refined our isotopic UPLC-MS/MS quantification27,28 to evaluate plasma concentrations of choline- and carnitine-derived metabolites leading to TMA and TMAO (Suppl. Fig. 3d,e). An O-PLS-DA model significantly segregates the three treatment groups at 5 months and highlights a choline supplementation effect on the first predictive component (Fig. 1g,h). Quantifications all reflect an increase in TMA and TMAO in the HC-HFD in line with choline supplementation (Fig. 1i,j). In particular, the circulating TMA levels were similar in the CHD and the LC-HFD (0.38 μM in CHD vs. 0.3 μM in LC-HFD) which was about 20 times less than in the HC-HFD (5.9 μM). These results depict an increased microbial conversion of choline into TMA in our HC-HFD, an observation already made in a previous study12. These results suggest that TMA could mediate the metabolic and immune benefits of choline supplementation.
Baseline metabolic phenotyping after 2-month HFD feeding
To further assess the impact of choline supplementation on metabolic homeostasis and low- grade inflammation in HFD contexts, we performed a second experiment, feeding C57BL/6 mice a 60% kcal HFD with a time-frame comparable to the ones used in subsequent figures. Weekly BWs and the GTT showed a similar trend after 8 weeks of HFD-feeding (Fig. 2a-c). Choline supplementation not only improved glucose tolerance (Fig. 2b,c) but also the Matsuda index (Fig. 2d), a surrogate marker for IR29. The normalisation of insulin sensitivity was confirmed through insulin tolerance tests (ITT) and hepatic Akt phosphorylation assays (Fig. 2e-h). We further characterized that HFD feeding increased hepatic NF-kB phosphorylation compared to CHD, which was normalised by choline supplementation (Fig. 2i,j), a pattern that was also observed for expression of acute-phase proteins Saa1, Saa2 and Saa3 (Fig. 2k-m).
Chemical blockage of bacterial TMA biosynthesis and loss of metabolic benefits
To test whether the beneficial effects of choline supplementation are mediated by its microbial product TMA, we sought to block bacterial TMA production in HC-HFD-fed mice both non-specifically and specifically using, respectively, a wide-spectrum antibiotics cocktail and 3,3-dimethyl-1-butanol (DMB), which inhibits microbial TMA-lyase30. We first confirmed the functional blockage of TMA production in mice fed a HC-HFD, resulting in a drastic drop in circulating and excreted TMA and TMAO following 1% DMB administration (Suppl. Fig. 4a-d). In accordance with our hypothesis, both antibiotics and DMB treatments abolished the effects of choline supplementation-induced improvements in HFD, in particular for glucose tolerance, insulin sensitivity (as suggested by the Matsuda index, Fig. 3a-c, Suppl. Fig. 4e,f) and hepatic insulin sensitivity as shown by the absence of beneficial effect of choline supplementation on insulin-induced Akt phosphorylation (Fig. 3d-g). The metabolic improvements were not associated with change in BW gain (Suppl. Fig. 4e), thereby strongly suggesting that the effects on glucose metabolism induced by inhibiting bacterial TMA production in HC-HFD were decoupled from obesity.
TMA treatment mimics choline supplementation
To further confirm whether TMA mediates the beneficial effects of choline supplementation, we chronically treated LC-HFD-fed C57BL/6 mice with TMA for 6 weeks using subcutaneous osmotic minipumps and assessed their immunometabolism. We confirmed that chronic TMA treatment at 0.01 mM in a LC-HFD did not affect BW gain (Suppl. Fig. 4g) but effectively normalised glucose homeostasis and insulin sensitivity (Fig. 3h-l, Suppl. Fig. 4d). TMA treatment also improved the pro-inflammatory response to HFD- feeding and hepatic expression of acute phase proteins Saa1, Saa2 and Saa3 (Fig.3m-p). These results suggest that chronic TMA treatment decouples BW gain and adiposity from low-grade inflammation and glucose homeostasis, thereby mimicking the immune and metabolic benefits observed in choline supplementation.
TMA is a specific IRAK-4 inhibitor
To identify a direct mechanism linking TMA to metabolic and immune benefits in HFD-fed mice, we sought to identify its host pharmacological targets. The kinome, made of 518 kinases encoded the human genome6, represents a repertoire of critical signal transduction switches for metabolic homeostasis and inflammation. To discover physical interactions, we screened TMA against a panel of 456 clinically-relevant human kinases using a high- throughput assay31,32 and identified five preliminary hits for TMA (Fig. 4a and Supplementary Tab. 5). We then generated multiple-dose binding curves between TMA and each hit and confirmed that only TMA specifically binds IRAK-4 (dissociation constant Kd = 14 nM, Fig. 4b) whereas TMA did not bind the other four preliminary hits (flat dissociation curves with no Kd fit in Supplementary Fig.5a-d, meaning no physical binding was observed). Since Kd only addresses a physical interaction in its simplest form (i.e. binding), we functionally tested TMA as an IRAK-4 inhibitor, by quantifying IRAK-4 phosphorylation activity in presence of ATP and increasing doses of TMA to derive a half-maximal relative inhibitory constant (IC50 = 3.4 μM, Fig. 4c). We also confirmed that choline, TMAO and DMB do not inhibit IRAK-4 (Suppl. Fig. 5e-g) and that TMA does not inhibit IRAK-1 either (Suppl. Fig. 5h), as Irak-1-/- mice fed a HFD were recently shown to have metabolic improvements similar to TMA treatment33. Molecular dynamics simulations of IRAK-4 were performed in the presence of 1% TMA. In silico analysis of thousands of binding and unbinding events suggests that in solution TMA binds to several small hydrophobic cavities in IRAK-4 and to the ATP binding site (Fig. 4d and Suppl. Fig. 5i-j).
Since IRAK-4 is a central kinase in the TLR pathway sensing bacterial invasion and promoting a pyretic pro-inflammatory response in infectious contexts34,35, we sought to test its inhibition in a HFD context. The main free fatty acid (FFA) in our HFD is palmitate, a saturated FFA which triggers TLR4. We therefore developed a palmitate-activated Kupffer cell model of HFD-induced low-grade inflammation based on murine KUP5 cell line36 (Suppl. Fig. 4k). Palmitate treatment in KUP5 cells significantly increased both IL-6 and TNFα secretions, which were improved by co-treatment with 0.1 mM TMA (Fig.4e,f). Our results reveal that TMA is a specific IRAK-4 inhibitor blunting the palmitate-induced low-grade inflammation in vitro, requiring further assessment in Irak4-/- mice.
Irak4-/- mice are protected against HFD-induced immune and metabolic dysregulations
IRAK-4 is a key kinase required for defence against pyogenic infections in acute contexts34,35.To test whether this kinase plays a role in HFD-induced chronic low-grade inflammation and glucose homeostasis, we fed 5-week-old Irak4-/- mice37 and wild-type (WT) littermates in C57BL/6 background a LC-HFD, to avoid potential confounding effects from TMA for 8 weeks before assessing circulating cytokines, expression of hepatic inflammatory genes and acute-phase markers and metabolic homeostasis (Fig. 5a-h). Irak4-/- mice presented improved glycemic control compared to WT littermates (Fig.5 a,b). Likewise, the inflammatory response to LC-HFD observed in WT was obliterated in Irak4-/- littermates (Fig. 5c-g). There was a similar trend for Saa3 (Fig. 5h), whilst there was no effect on BW gains (Supplementary Fig. 6a). Hence, genetic ablation of IRAK-4 abolishes the HFD-induced pro- inflammatory response and IGT thereby decoupling obesity from impaired glucose tolerance and low-grade inflammation with a similar phenotype to Irak1 deficiency33.
Pharmacological inhibition of Irak4 normalises metabolism
Since the Irak4-/- mice lack the whole protein, we compared the knock-out phenotype with the phenotype of PF06650833, a recently discovered chemical inhibitor of the human IRAK4 protein38 currently being assessed in a clinical trial (NCT02996500). Treatment with PF06650833 improved BW in LC-HFD mice (Suppl. Fig. 6b). This inhibitor also yielded significant improvements in plasma glycemia at the latter time-points of the GTT and ITT and in cumulative glycemia (Fig. 5i-l), which was mirrored by an increase in Akt phosphorylation (Fig. 5m-n). These results collectively show that specific chemical inhibition of IRAK-4 leads to significant improvements in glycemic control, insulin sensitivity and insulin signalling. Further, they suggest that IRAK-4 could constitute a clinically relevant target in IR and related disorders.
DISCUSSION
The discovery that TMA is a kinase inhibitor controlling IRAK-4, a central kinase involved in innate immunity, is a major finding which provides an attractive mechanism for the metabolic and immune improvements observed with choline-supplementation in HFD contexts (Fig. 6).
IRAK-4 is the first regulatory checkpoint after MyD88, the adapter protein connecting IRAK-4 to at least six TLRs sensing bacterial compounds39. IRAK-4 deficiency is associated with bacterial infections in humans35 and in mice40. Consistent with the involvement of the TLR signalling pathway at the crossroads between gut microbiota and dietary lipids41,42, IRAK-4 deletion and its inhibition by TMA and PF0665083338 rescued HFD-induced low-grade inflammation and IR4,43,44, highlighting new unexpected roles for this microbial metabolite and its target kinase in immunometabolism. The relative IC50 (3.4 μM) we determined is two times smaller than the plasma isotopic quantifications obtained by UPLC-MS/MS in the HFD group supplemented with choline. These quantifications are comparable to circulating TMA levels previously reported in normal human plasma, ranging between 0.42 and 48 μM45,46, which makes this IC50 particularly relevant with regard to pathophysiological mechanisms: dosing mice with 0.01 mM TMA (3x IC50) was sufficient to improve inflammatory and metabolic responses. TMA can therefore be considered a “microbial signalling metabolite”47 sending a negative feedback signal to a pathway triggered by influx of saturated free fatty acids in HFD contexts, this mechanism participating in maintaining a low immunological footprint and metabolic improvements in a symbiotic relationship. TMA’s role as an IRAK-4 inhibitor could for instance explain some of the beneficial effects reported for choline supplementation in NAFLD patients48. Altogether, our results on IRAK-4 inhibition and ablation provide further insight in the phenotypic convergence between Myd88, Irak4 and Irak1 knock-out mouse models33,49, suggesting that the gut microbiota, through TMA, proceeds with a targeted hijacking of the TLR signal transduction machinery to the host’s benefit resulting in metabolic and immune improvements.
In summary, through the combination of in silico, ex vivo, in vitro and in vivo approaches, we reveal a unique mechanism in which TMA acts as a gut microbial signalling metabolite inhibiting IRAK-4, a central molecular target in the TLR pathway, thereby allowing the gut microbiota to control HFD-induced pro-inflammatory response and IR. The kinome represents a key repertoire of regulatory host targets for microbial signals and the uncharted microbiome–kinome crosstalk requires further investigation. By highlighting the physiological and therapeutic roles of TMA and IRAK-4 on HFD-induced low-grade inflammation and IR, we anticipate that its immunomodulatory properties extend beyond IR to a wider range of human pathologies involving TLR signalling and modulation of innate immunity.
Author Contributions
JC and SG performed cell-based assays, JC performed multivariate statistics and interpreted the results. JC, FB, JFF, AE, HP, LZ, DS, ARR, PDC, SC and CG contributed in vivo experiments, RHB, AM, ALN, LMG, SG, CLB and SP performed various experiments. JEF and RCG designed and performed in silico experiments. LH analysed microarray data. JS, DG, JKN, PPL, PDC, NJG and MED conceived the project, mentored and supervised its participants and interpreted its results. MED performed metabolic profiling experiments, analysed data and interpreted the results. JC, NJG, DG, PPL, PDC, and MED wrote the manuscript. All authors edited the manuscript.
Online Content
Methods, along with any additional Supplementary display items and Source Data, are available in the online version of the paper; references unique to these sections appear only in the online paper
Online Methods
Protocols
All experimental procedures involving mice were carried out in accordance with U.K. Home Office, Canadian Council on Animal Care, the ethics committee of the French Research Ministry (authorization number 00486.01), Belgian Law of May 29, 2013 regarding the protection of laboratory animals (agreement number LA1230314) and local guidelines on animal welfare and license conditions and the University of Oxford, University of Ottawa, Université Pierre et Marie Curie and Université Catholique de Louvain guidelines on animal welfare.
Cell-based assays
C-myc-immortalised KUP5 cells, originally harvested from a C57BL/6 mouse, were purchased from RIKEN BioResource Center, which certified the cells were not contaminated by Mycoplasma. They were cultured in splitting medium at 37°C and 5% CO2 atmosphere under sterile conditions. Cells were split every 48-72 h using splitting medium (DMEM GlutaMAX-I supplemented with 10% FBS, 1% penicillin and streptomycin, 0.1% insulin and 0.044% thioglycerol), maintained at a density between 5 x 105 – 1 x 106 cells/mL and used between passages 2 and 20. For palmitate (PA) treatment, 125,000 cells/cm2 were seeded in treatment medium (DMEM GlutaMAX-I supplemented with 5% FBS, 1% penicillin and streptomycin, 0.1% insulin and 0.044% thioglycerol) on a 48-well plate with a working volume of 250 μL and incubated at 37°C and 5% CO2 atmosphere overnight (16-18 h). 0.5 M PA stock solution was prepared in 100% ethanol and 0.5 mM PA treatment solution was prepared in treatment medium containing a final concentration of 0.5% low-endotoxin-fatty acid free BSA. TMA (0.1 mM) treatment was performed in a simultaneous fashion with PA. All solutions containing PA were prepared in borosilicate glass tubes. Upon overnight incubation of seeded cells, the medium was replaced with the same volume of treatment solutions and incubated at 37°C and 5% CO2 atmosphere for 24 h before measurement of cytokines in the medium.
Mouse Models
Longitudinal HFD-feeding in mice
All experiments were approved by the ethical committee of the University of Oxford. Male mice from C57BL/6J inbred strain were bred in our animal facility by using a stock originating from The Jackson Laboratory. At 5 weeks of age, groups of n=8-10 mice were transferred to a 40 %w/w high fat diet (65% kcal) (Special Diets Services), containing 32 % lard and 8 % corn oil, whereas control groups remained on a 5 % Low Fat Diet (B & K Universal) for up to 6 months. Detailed diet formulations were published previously12 and summarised in Supplementary Table 6.
Mice were housed under a 12 h–12 h light–dark cycle. For physiological profiling, several mouse groups fed CHD or HFD were tested to assess consistency of results and discard any impact of potential batch effects. Intra peritoneal glucose tolerance tests (ipGTT) were performed in mice after 2-, 3-, 5-and 7-month-old mice after an overnight fast, as previously published50 (see also metabolic phenotyping below). Four days after the GTT, 24 h urinary samples (9 a.m. to 9 a.m.) were collected from mice maintained in individual metabolic cages. Urinary samples collected in a solution of 1 % (wt/vol) sodium azide were centrifuged to remove solid particles and kept at -80°C until assayed. After an overnight fast, mice were killed by exsanguination. Plasma was separated by centrifugation and stored at -80 °C until 1H-NMR analysis.
Choline supplementation on HFD
At 5 weeks of age, mice were fed either a chow diet containing 2g of choline/kg of diet (Research Diets, D12450J), a low-choline HFD containing 2g of choline/kg of diet (Research diets, D12492), a high-choline HFD containing 17 g of choline/kg of diet (Research diets, D16100401), a HC-HFD containing 1% of DMB or a HC- HFD combined with a cocktail of antibiotics (0.5 g/L vancomycin hydrochloride, 1 g/L neomycin trisulfate, 1 g/L metronidazole, 1 g/L ampicillin sodium) in drinking bottles (n=6- 10 per group) for 8 weeks (see diet formulations in Supplementary Table 6). Mice then were killed by decapitation and organ were dissected and weighed.
Irak4-/- mice. Irak4-/- mice on C57BL/6J background as already described37 were bred with C57BL/6J female mice (The Jackson Laboratory), and the F1 offspring were subsequently bred to produce Irak4-/- mice and WT littermate used for this study. After eight weeks of HFD-feeding, mice were killed by decapitation and organ were dissected and weighed.
Chronic TMA and PF06650833 treatment in LC-HFD-fed mice
Five-week-old C57BL/6J mice (Charles River) were housed a week before experiment in a controlled environment. Mice were housed under a 12 h–12 h light–dark cycle. At day 0, the 10-week-old mice were anaesthetised with isoflurane (ForeneH, Abbott). Mini-osmotic pumps were implanted subcutaneously (Model 2006, Alzet) (flow rate: 0.15 mL/h, total filling volume: 200 mL, delivery duration: 42 days) as previously described51. The osmotic mini-pump contains either vehicle or TMA (0.1 mM in circulation) or PF06650833 (50 nM in circulation). After six weeks of metabolite treatment, mice were killed by decapitation and organ were dissected and weighed.
Physiological phenotyping
After four weeks of treatment an intra-peritoneal glucose tolerance test (2g/kg) was performed in conscious mice following an overnight fast. Blood was collected from the tail vein before glucose injection and 30, 60, 90 and 120 minutes afterwards. Blood glucose levels were determined using an Accu-Check® Performa (Roche Diagnostics, Meylan, France). Additional blood samples were collected at baseline and 30 minutes after glucose injection in Microvette® CB 300 Lithium Heparin (Sarstedt, Marnay, France). Plasma was separated by centrifugation and stored at -80°C until insulin radioimmunoassay. Circulating insulin levels were determined using Insulin ELISA kits (Mercodia, Uppsala, Sweden). The Matsuda insulin sensitivity index was calculated as previously published29
After five weeks, we performed an insulin tolerance test. 5 hour-fasted mice were injected intraperitoneally with insulin (0.75 mU/g, Actrapid, Novo Nordisk). Blood glucose levels were measured immediately before and 15, 30, 45, 60, 90 and 120 min after insulin injection with a standard glucose meter (Accu Check, Roche, Basel, Switzerland) on the tip of the tail vein.
Gene expression
Groups of 6 mice showing consistent pathophysiological profile in response to CHD or HFD treatment were selected for microarray analysis performed as previously described52 and data were deposited in array express with accession number: E-MEXP-1755. The Bioconductor53 package Limma54 was used to generate the list of differentially expressed genes. Gene ontology was implemented using Enrichr55 and signalling pathway impact analysis using SPIA56.
For qPCR analysis, total RNA was prepared from tissues using TriPure reagent (Roche). Quantification and integrity analysis of total RNA were performed by analysing 1 μL of each sample in an Agilent 2100 Bioanalyzer (Agilent RNA 6000 Nano Kit, Agilent). cDNA was prepared by reverse transcription of 1 mg total RNA using a Reverse Transcription System kit (Promega). Real-time PCR was performed with the StepOnePlus real-time PCR system and software (Applied Biosystems) using Mesa Fast qPCR (Eurogentec) for detection according to the manufacturer’s instructions. RPL19 RNA was chosen as the housekeeping gene. All samples were performed in duplicate in a single 96-well reaction plate and data were analysed according to the 2_DDCT method. The identity and purity of the amplified product were assessed by melting curve analysis at the end of amplification. The primer sequences for the targeted mouse genes are: SAA1 forward CAT-TTG-TTC-ACG-AGG-CTT- TCC, SAA1 reverse GTT-TTT-CCA-GTT-AGC-TTC-CTT-CAT-GT, SAA2 forward GGG-GTC-TGG- GCT-TCC-CAT-CT, SAA2 reverse CCA-TTC-TGA-AAC-CCT-TGT-GG, SAA3 forward CGC-AGC- ACG-AGC-AGG-AT, SAA3 reverse CCA-GGA-TCA-AGA-TGC-AAA-GAA-TG.Q-RT-PCR assays were performed in a single batch with the personnel blind to treatment groups.
Circulating cytokine quantification
Circulating cytokines were quantified using MSD V-PLEX Plus Proinflammatory Panel 1 kit. Plasma samples were diluted 2 times in diluent 41 provided and experiment was processed as request by manufacturer and read on a SECTOR imager 2400. Cytokine assays were performed in a single batch with the personnel blind to treatment groups.
Western-blotting
Western blot analyses were performed as described previously 28. To analyze the insulin signaling pathway, mice were allocated to either a saline-injected subgroup or an insulin- injected subgroup so that both subgroups were matched in terms of BW and fat mass. They then received 1 mU insulin/g BW (Actrapid; Novo Nordisk A/S, Denmark) or an equal volume of saline solution. Three minutes after injection, mice were killed and liver was harvested. 30mg of liver were homogenized in 680 μL of RIPA buffer containing a cocktail of protease and phosphatase inhibitors. The homogenate was then centrifuged at 12,000G for 20 min at 4 °C. Equal amounts of proteins were separated by SDS–PAGE and transferred to nitrocellulose membranes. Membranes were incubated at 4 °C with antibodies diluted in Tris-buffered saline Tween-20 containing 1% bovine serum albumin: p-Akt Ser473 (1:1,000; #4060, Cell Signaling), Total Akt (1:1000, #9272S, Cell Signalling), p-NF-kB (1:3000, #ab86299, AbCam), Total NF-kB p65 (1:3000, #8242, Cell Signalling).
1HNMR spectroscopy and multivariate statistics
Mouse urine and plasma samples were prepared and measured on a spectrometer (Bruker) operating at 600.22 MHz 1H frequency; full resolution 1H NMR spectra were then processed and analysed using orthogonal partial least square discriminant analysis as described previously12. Variance component analysis was performed as described previously57. 1H NMR profiling was performed in a single batch with the personnel blind to treatment groups. The O-PLS-DA model was validated using 10,000 random permutations of the original class membership variable to explain (i.e. diets, treatments or genotypes), as described previously57.
Plasma methylamine quantification by UPLC-MS/MS
Methylamines were quantified as previously described. Plasma samples (20 μL) were spiked with 10 μL Internal Standard (IS) solution (13C3 /15N-TMA, d9 -TMAO, d4 -choline, d3 -carnitine and d9 -betaine in water; 1 mg/L) and 45 μL of ethyl 2-bromoacetate solution (15 g/L ethyl 2-bromoacetate, 1% NH4OH in acetonitrile) were added to derivatize methylamines (TMA and 13C3/15N-TMA) to their ethoxy- analogues, completed after 30 min at room temperature. 935 μL of protein/lipid precipitation solution (94% acetonitrile/5% water/1% formic acid) was added; samples were centrifuged for 20 min (4°C, 20,000g) and were transferred to UPLC-autosampler vials. Sample injections (10 μL loop) were performed on a Waters Acquity UPLC-Xevo TQ-S UPLC- MS/MS system equipped with an Acquity BEH HILIC (2.1 × 100 mm, 1.7 μm) chromatographic column. An isocratic elution was applied with 10 mM ammonium formate in 95:5 (v/v) acetronitrile:water for 14min at 500 μL/min and 50 °C. Positive electrospray (ESI+) was used as ionization source and mass spectrometer parameters were set as follows: capillary, cone and source offset voltages at 500, 93 and 50 V, respectively, desolvation temperature at 600°C, desolvation/cone/nebulizer gases were high-purity nitrogen at 1000 L/hr, 150 L/hr and 7 bar respectively. Collision gas was high-purity argon. Mass spectrometer was operated in multiple reactions monitoring (MRM) mode. The monitored transitions were the following: for derivatized-TMA, +146 -> +118/59 m/z (23/27 V); for derivatised-13C3 /15N-TMA, +150 -> +63 (27V); for TMAO, +76 -> +59/58 m/z (12/13 V); for d 9-TMAO, +85 -> +68 m/z (18 V); for choline, +104-> +60/45 m/z (14/16 V); for d4-choline, +108 -> +60 m/z (15 V); for γ-butyrobetaine, +146 -> +60/87 m/z (12/12 V); for carnitine, +162 -> +103/60 m/z (16/14 V); for d3-carnitine, +165 -> +103 m/z (16 V); for betaine, +118 -> +59/58 m/z (16/16 V); for d9-betaine, +127 -> +68 m/z (16 V).
Kinome screen, Kd and IC50 determination
TMA was assessed using KdELECT screening service (DiscoveRx) as described previously31,32.This technique is based on a competition-binding assay that quantitatively measures the ability of a compound to compete with an immobilized, active site-directed ligand. The assay consists of DNA-tagged kinase, immobilized ligand, and the potent inhibitor. The ability of TMA to compete with the immobilized ligand was measured by quantitative PCR of the DNA tag. The binding constant (Kd) was then calculated from duplicate 11-point dose– response curve. Kinase interaction tree plots were generated using TREEspot™ Software Tool and reprinted with permission from KINOMEscan®, a division of DiscoveRx Corporation, © Discoverx corporation 2015. The TMA IC50 on IRAK-4 was determined using Kinexus kinase-inhibitor activity profiling service (Kinexus, Vancouver, Canada). The technique is based on the direct quantification of radiolabelled phosphate from ATP onto a protein substrate of a target protein kinase.
Molecular dynamics simulations
Molecular dynamics simulations were performed within the Amber package58 based on an X-ray structure of human IRAK-459. After removal of the bound ligand and sulfate ions, the system was prepared using protonate3d60. Protein atoms were described using the Amber fore field 99SB-ILDN61, ligand TMA was described within the Generalized Amber Force Field62 after fitting partial charges using RESP at the Hartree-Fock/6-31G* level (see Supplementary 2e for resulting parameters). A periodic truncated octahedral shell of TIP3P water molecules was added63 10Å to our systems. 1% of approximately 10,000 water molecules with a minimum distance of 4Å to the protein were randomly replaced by TMA to study binding. After applying an elaborate equilibration protocol64, we sampled each system without restraints for 100 ns using the GPU implementation of pmemd58. Analyses of simulation data were performed using ptraj and cpptraj from AmberTools65. After performing standard stability checks for energy and overall structure, hydrogen bonding between protein and ligands was counted using cpptraj’s default criteria. All protein–ligand contacts within a distance cutoff of 5Å were considered for contact analysis. Grid-based analyses of TMA positions were performed on a grid capturing the entire simulation box with a spacing of 0.5Å. The resulting grids were overlaid with an ATP-bound structure of IRAK-4 (PDB: 2OID)66.
Reagents
Glutamine (Glutamax, 35050061, Life Technologies), Fetal bovine serum (Life Technologies), crystal violet (C6158) and Trimethylamine (W324108) were from Sigma-Aldrich. Mouse IL-6 Quantikine ELISA kits (M6000B, R&D system), RNeasy Micro Kit (Qiagen). SuperScript II Reverse Transcriptase, IL6 Taqman probe Hs00174131_m1 and FAST master mix (Invitrogen). High Fat Diet (Special Diets Services) and Low Fat Diet (B & K Universal) were specifically formulated in12. Control Diet (D12450K; Research diet), LC-HFD (60% kcal fat and 20% kcal carbohydrates, D12492, Research diet), HC-HFD (60% kcal fat and 20% kcal carbohydrates with 17g of choline/kg, D16100401i, Research diet). Mouse Insulin ELISA (10- 1249-01, Mercodia). Isoflurane (10014451, Forene, Abbott). TriPure reagent (1667165, Roche). Reverse Transcription System kit (A3500, Promega). Mesa Fast qPCR (CS-CKIT-PROD, Eurogentec). MSD V-PLEX Plus Proinflammatory Panel 1 kit (K15048G Meso Scale Diagnostics).
Statistics
Potential outliers were identified by a Grubbs test. For statistical comparisons between study groups, normality was tested using D’agostino-Pearson omnibus normality test, then one-way ANOVA was used, followed by Tukey’s post hoc testing when data were normally distributed, otherwise groups were compared using the two-tailed Mann-Whitney test (P < 0.05 considered to be statistically significant). Data are displayed as mean ± s.e.m in all figures. All cell culture experiments included at least three biological replicates (as indicated in figure legends). All animal cohorts included at least five animals in each study group (as indicated in figure legends) and animals were randomized to treatment groups.
Supplementary Figure Legends
Supplementary Tab. 1: Hepatic genes differentially expressed by HFD-feeding. Hepatic transcriptomes were analysed using Limma54 and genes are considered significant if Fdr<0.1.
Supplementary Tab. 2: Down-regulated HFD-responsive hepatic pathways. Gene ontology analysis was performed using Enrichr55 and pathways are considered significant if Fdr<0.1.
Supplementary Tab. 3: Up-regulated HFD-responsive hepatic genes. Gene ontology analysis was performed using Enrichr55 and pathways are considered significant if Fdr<0.1.
Supplementary Tab. 4: Hepatic signalling pathways impacted by HFD. Signalling pathway impact analysis was performed using SPIA56 and pathways are considered significant if pGFWER<0.1.
Supplementary Tab. 5. TMA kinome screen. Screening results for single-dose TMA binding to 456 kinases. “Percent control” corresponds to the percentage of DNA amplified by qPCR compared to the control condition. Values below 35 % (in green) are considered as positive hits.
Supplementary Tab. 6. Diet formulations. Sepcific fat, protein, carbohydrate and choline content in regimens formulated by Research Diet are summarized based on %w/w and %kcal.
Acknowledgments
This work was supported by grants from the Wellcome Trust (Functional Genomics Initiative grant Biological Atlas of Insulin Resistance 06678 to JKN, DG and JS), the European Commission (METACARDIS HEALTH-F4-2012-305312 to DG and MED, Neuron II under agreement 291840 and the MRC (MR/M501797/1 to MED), Institut Mérieux through a grant awarded to MED, and Canadian Institutes of Health Research (CIHR) grants (MOP-49413 & MOP-142471) awarded to PPL. DG held a Wellcome Trust Senior Fellowship in Basic Biomedical Science (057733). JEF was supported by the Medical Research Council (grant number MR/K020919/1). PDC is a senior research associate and AE a research associate at FRS-FNRS (Fonds de la Recherche Scientifique). PDC is the recipient of grants from FNRS. LH is in receipt of an MRC Intermediate Research Fellowship in Data Science (grant number MR/L01632X/1, UK Med-Bio). This work was supported by FRFS-WELBIO under grant: WELBIO-CGR-2017, by the Funds InBev-Baillet Latour (Grant for Medical Research 2015) and ERC Starting Grant 2013 (European Research Council, Starting grant 336452-ENIGMO).