Insulin-regulated serine and lipid metabolism drive peripheral neuropathy

Diabetes represents a spectrum of disease in which metabolic dysfunction damages multiple organ systems including liver, kidneys and peripheral nerves1,2. Although the onset and progression of these co-morbidities are linked with insulin resistance, hyperglycaemia and dyslipidaemia3–7, aberrant non-essential amino acid (NEAA) metabolism also contributes to the pathogenesis of diabetes8–10. Serine and glycine are closely related NEAAs whose levels are consistently reduced in patients with metabolic syndrome10–14, but the mechanistic drivers and downstream consequences of this metabotype remain unclear. Low systemic serine and glycine are also emerging as a hallmark of macular and peripheral nerve disorders, correlating with impaired visual acuity and peripheral neuropathy15,16. Here we demonstrate that aberrant serine homeostasis drives serine and glycine deficiencies in diabetic mice, which can be diagnosed with a serine tolerance test that quantifies serine uptake and disposal. Mimicking these metabolic alterations in young mice by dietary serine or glycine restriction together with high fat intake markedly accelerates the onset of small fibre neuropathy while reducing adiposity. Normalization of serine by dietary supplementation and mitigation of dyslipidaemia with myriocin both alleviate neuropathy in diabetic mice, linking serine-associated peripheral neuropathy to sphingolipid metabolism. These findings identify systemic serine deficiency and dyslipidaemia as novel risk factors for peripheral neuropathy that may be exploited therapeutically.

Diabetes represents a spectrum of disease in which metabolic dysfunction damages multiple organ systems including liver, kidneys and peripheral nerves 1,2 . Although the onset and progression of these co-morbidities are linked with insulin resistance, hyperglycaemia and dyslipidaemia [3][4][5][6][7] , aberrant non-essential amino acid (NEAA) metabolism also contributes to the pathogenesis of diabetes [8][9][10] . Serine and glycine are closely related NEAAs whose levels are consistently reduced in patients with metabolic syndrome [10][11][12][13][14] , but the mechanistic drivers and downstream consequences of this metabotype remain unclear. Low systemic serine and glycine are also emerging as a hallmark of macular and peripheral nerve disorders, correlating with impaired visual acuity and peripheral neuropathy 15,16 . Here we demonstrate that aberrant serine homeostasis drives serine and glycine deficiencies in diabetic mice, which can be diagnosed with a serine tolerance test that quantifies serine uptake and disposal. Mimicking these metabolic alterations in young mice by dietary serine or glycine restriction together with high fat intake markedly accelerates the onset of small fibre neuropathy while reducing adiposity. Normalization of serine by dietary supplementation and mitigation of dyslipidaemia with myriocin both alleviate neuropathy in diabetic mice, linking serine-associated peripheral neuropathy to sphingolipid metabolism. These findings identify systemic serine deficiency and dyslipidaemia as novel risk factors for peripheral neuropathy that may be exploited therapeutically.
To explore how obesity and diabetes influence serine, glycine and one-carbon (SGOC) metabolism, we quantified serine, glycine and methionine across tissues in an established mouse model of morbid obesity, insulin resistance and hyperglycaemia (leptin receptor-deficient db/db mice on a black Kaliss background (BKS-db/db)) and compared results with age-matched wild-type C57BL/6J controls. The db/db mice showed reductions of around 30% in hepatic and renal serine levels relative to wild-type mice ( Fig. 1a and Extended Data Fig. 1a), and the more abundant glycine pools were reduced by 30-50% in liver, kidney, inguinal white adipose tissue (iWAT) and plasma ( Fig. 1a and Extended Data Fig. 1a-c). Methionine is linked to serine through one-carbon metabolism and was also reduced in liver, iWAT, and plasma ( Fig. 1a and Extended Data Fig. 1b,c), suggesting that diabetes decreases serine and glycine levels in tissues that are important for glucose and lipid homeostasis.
Mammals obtain serine from the diet, de novo synthesis from glucose and via glycine and one-carbon metabolism, with the liver and kidney serving as major sites for postprandial NEAA metabolism (Fig. 1b). To better understand the mechanistic basis of reduced hepatic serine and glycine in db/db mice, we quantified the expression of genes associated with SGOC metabolism after a 6-h fast (Fig. 1c). Genes encoding key enzymes responsible for serine and one-carbon unit catabolism or disposal were significantly upregulated in db/db mice, including serine dehydratase (Sds), serine hydroxymethyltransferase 1 (Shmt1), Shmt2 and 10-formyltetrahydrofolate dehydrogenase (Aldh1l1). The expression of two genes encoding components of the glycine cleavage system-glycine decarboxylase (Gldc) and dihydrolipoamide dehydrogenase (Dld)-was increased in db/db liver, whereas the expression of 3-phosphoglycerate dehydrogenase (Phgdh) and methylenetetrahydrofolate dehydrogenase 2 (Mthfd2) were both significantly reduced, indicating that de novo serine synthesis may also be limited in diabetic mice. These results are consistent with genome-scale metabolic modelling of transcription data from human diabetic liver 17 . The kidney is another centre for SGOC metabolism 18 and renal expression of Shmt1 https://doi.org/10.1038/s41586-022-05637-6 and several genes encoding enzymes associated with one-carbon metabolism were increased in db/db mice (Extended Data Fig. 1d).
Circadian and postprandial variations in amino acids and glucose make the diagnosis of metabolic defects challenging. We therefore hypothesized that a 'serine tolerance test' (STT) could better assess SGOC metabolism in mice and identify those with elevated serine disposal. Applying similar dosages to those used in human clinical trials (ClinicalTrials.gov: NCT03062449) (400 mg kg −1 ), we orally  13 C 3 ]serine administration via oral gavage (n = 4 per tissue) after an overnight fast. g, Combined OGTT and STT in wild-type and BKS-db/db mice (n = 6 per group) after an overnight fast. h, STT AUC in wild-type and BKS-db/db mice (n = 6 per group). i, Combined OGTT and STT in vehicle-(n = 7) and STZ-treated (n = 6) C57BL/6J mice after an overnight fast. j, STT AUC in vehicle-(n = 7) and STZ-treated (n = 6) C57BL/6J mice. Data are mean ± s.e.m., and were analysed using two-sided independent t-test (a,c,h,j) and two-way ANOVA with Fisher's least significant difference post hoc test (g,i). The schematic in Fig. 1b was prepared in BioRender.
Article administered serine to overnight-fasted wild-type mice and quantified plasma serine pharmacokinetics to gauge the dynamics of serine clearance, with levels returning to baseline in approximately 2 h (Extended Data Fig. 1e). To identify the principal pathways involved in serine disposal, we orally administered [U-13 C 3 ]serine to overnight-fasted wild-type mice and quantified the enrichment of downstream metabolites. We observed that glucose was labelled to a similar extent as glycine throughout the test (Fig. 1d), and [U-13 C 3 ]serine-derived carbon was significantly incorporated into hepatic glycine, pyruvate and citrate pools (Fig. 1e,f and Extended Data Fig. 1f,g), demonstrating that hepatic gluconeogenesis is a major pathway for serine disposal in some contexts and linking its regulation to insulin and glucagon, which are both elevated in db/db mice (Extended Data Fig. 2a-c).
To test whether insulin resistance affects serine absorption and disposal, we delivered both glucose (2 g kg −1 ) and serine (400 mg kg −1 ) to overnight-fasted wild-type and db/db mice. We observed a significant reduction in STT area under the curve (AUC) in db/db mice (Fig. 1g,h and Extended Data Fig. 2d), despite the higher dose administered. Notably, acute serine challenge had essentially no effect on circulating glycine concentrations (Extended Data Fig. 2e), despite evidence of their rapid interconversion (Fig. 1d,e). Conversely, there were no differences in STT AUC between wild-type and db/db mice when serine was administered without glucose (Extended Data Fig. 2f).
To determine whether elevated serine disposal is specific to leptin receptor-deficient db/db mice or is more generally attributable to impaired insulin signalling, we treated C57BL/6J mice with vehicle or streptozotocin (STZ) to induce insulin deficiency, hyperglycaemia and fat loss (Extended Data Fig. 2g-i). Plasma glycine and branched-chain amino acids (but not serine) were altered one week after STZ treatment (Extended Data Fig. 2j). Two weeks after injection, co-administration of glucose and serine revealed elevated serine disposal in STZ-diabetic mice relative to controls (Fig. 1i,j and Extended Data Fig. 2k), suggesting that insulin resistance or deficiency can both contribute to reduced circulating serine in diabetic mice.
Clinical studies have implicated serine deficiency in the regulation of macular disease and peripheral neuropathy 15,16 . Given the impaired serine homeostasis in db/db mice, we next confirmed that aberrant serine metabolism is coincident with peripheral neuropathy in this model. Fourteen-week-old db/db mice exhibited thermal and tactile hypoalgesia as well as decreased motor nerve conduction velocity (MNCV) (Extended Data Fig. 2l-n). These findings suggest that aberrant serine homeostasis is associated with diabetic peripheral neuropathy.

Serine remodels lipid metabolism
Serine and glycine restriction is widely used to modulate health outcomes in mice and is relatively well-tolerated for several months 16,[19][20][21][22] . To model the effect of systemic serine deficiency in the context of diet-induced obesity, we fed mice low-fat (10% kcal) (LFD) or high-fat (60% kcal from fat) (HFD) diets and compared their phenotypes to mice fed isonitrogenous diets lacking serine and glycine (−SG LFD or −SG HFD) (Supplementary Table 1). Both −SG diets effectively reduced circulating and hepatic serine and glycine contents during the fed state but not during the fasted state (Fig. 2a,b and Extended Data Fig. 3a,b). The HFD alone reduced liver glycine levels, but to a lesser extent than dietary serine and glycine withdrawal (Extended Data Fig. 3b). Other serine and glycine-derived hepatic metabolites, including glutathione, were not strongly affected by their restriction (Extended Data Fig. 3b). Of note, weight gain caused by HFD was attenuated by dietary serine and glycine restriction, whereas food, calorie and water intake, calorie absorption, and physical activity were all unaffected ( Fig. 2c and Extended Data Fig. 3c-h). Using echo magnetic resonance imaging (echoMRI), we quantified lean and fat masses across all groups and observed that dietary serine restriction significantly reduced fat mass but had no effect on lean mass relative to the HFD group (Fig. 2d). Consistent with these changes in adiposity, we observed that HFD feeding increased, whereas serine and glycine restriction decreased, epididymal white adipocyte size (Extended Data Fig. 3i).
To determine how feeding with −SG HFD influences glucose homeostasis, we analysed mice using standard GTTs and ITTs. Whereas −SG HFD attenuated obesity, the mice remained glucose and insulin intolerant (Fig. 2e,f and Extended Data Fig. 3j,k), suggesting that serine and glycine restriction and the consequent reduction in adiposity do not prevent HFD-induced glucose intolerance and insulin resistance. To test whether altered systemic substrate utilization of carbohydrates in preference to lipids were driving the changes in adiposity, we placed mice that had consumed each diet for 18 weeks in metabolic cages and quantified the respiratory exchange ratio (RER). Whereas the HFD altered RER as expected, no changes were observed with dietary serine/ glycine restriction (Extended Data Fig. 3l).
Next, we performed a metagenomic analysis of faecal microbiota to understand how the above diets affected microbiome composition and diversity, which correlate with and can buffer against dietary deficiencies 23 . Sustained HFD or −SG HFD feeding reduced phylogenetic alpha diversity relative to LFD-fed mice, whereas −SG LFD feeding increased alpha diversity (Fig. 2g). Conversely, PERMANOVA testing of Aitchison distances from robust principal component analysis revealed significant beta-diversity differences between −SG HFD-fed mice and other groups, highlighting the distinct effect of this low-carbohydrate, high-fat, serine and glycine-restricted diet on the faecal microbiome (Extended Data Fig. 4a). Furthermore, the log ratios of strains of microorganisms expressing complete serine biosynthesis and glycine cleavage pathways were increased and decreased, respectively, by −SG HFD (Extended Data Fig. 4b,c), and this diet reduced the log ratio of strains expressing a complete fatty acid synthesis pathway (Fig. 2h). Key strains showing the strongest alterations are listed in Extended Data Fig. 4d. Notably, −SG LFD feeding did not modulate strains in this manner, presumably owing to the higher carbohydrate content, which could facilitate serine synthesis.
Next, to directly investigate how serine deficiency affects hepatic lipogenesis, mice fed the above diets for 18 weeks were administered heavy water (D 2 O) for 18 h, and lipids were extracted for quantification of isotope enrichment, molar abundance, and synthesis 24 . Dietary serine and glycine restriction potently reduced hepatic palmitate synthesis by around 70% relative to serine-replete control diets ( Fig. 2i and Extended Data Fig. 5a). Hepatic cholesterol synthesis was increased in −SG HFD compared with HFD (Extended Data Fig. 5b). Consistent with these changes, hepatic expression of ATP-citrate lyase (ACLY), acetyl-CoA carboxylase (ACC2) and stearoyl-CoA desaturase (SCD1) were strongly reduced (around 50%) by dietary serine restriction, whereas the expression of cholesterol biosynthesis enzymes was unchanged or increased (Extended Data Fig. 5c-e). Changes in AKT phosphorylation at Ser473 and Thr308 correlated with dietary fat and carbohydrate content rather than serine and glycine levels, further suggesting that serine restriction drives changes in fatty acid metabolism that are independent from insulin signalling (Extended Data Fig. 5c).

Serine restriction and HFD accelerate neuropathy
Systemic serine deficiency has recently been linked to various neurodegenerative disorders 16,[25][26][27][28] . People with diabetes who have elevated serine disposal could therefore be more susceptible to neurological co-morbidities reminiscent of serine-associated peripheral neuropathy. To determine whether long-term, chronic serine deficiency is sufficient to drive peripheral neuropathy, we fed mice either control or serine-and glycine-free chow diets (19.2% of energy from fat) for up to 12 months. Temporal quantification of the thermal response to heat revealed progression to hypoalgesia after 12 months of dietary intervention (Extended Data Fig. 6a), consistent with previous observations 16 . At this time point we also detected reduced intraepidermal nerve fibre (IENF) density in paw skin (Extended Data Fig. 6b), which is also a clinical measure of small sensory fibre degeneration 29 . Of note, mice fed −SG HFD for just three months exhibited marked thermal hypoalgesia (Fig. 2j), indicating that a combination of low systemic serine and HFD accelerate the onset of peripheral neuropathy in mice.
Serine is essential for the biosynthesis of canonical sphingolipids, which are enriched in the nervous system. When serine becomes limited, serine palmitoyltransferase (SPT) incorporates other amino acids, including alanine, to form non-canonical deoxysphingolipids 30,31 . Given the importance of canonical ceramides and 1-deoxy(dihydro) ceramides in obesity and neuropathy, respectively 15,32 , we hypothesized that SPT inhibition could influence the observed obesity and neuropathy phenotypes. We therefore administered myriocin (0.3 mg kg −1 every other day), an inhibitor of SPT, or vehicle to mice fed the above diets for 6 months and quantified sphingolipid diversity and thermal sensing. Consistent with previous reports 32 , myriocin treatment attenuated HFD-induced weight gain without affecting plasma serine and glycine levels (Extended Data Fig. 6c-f) However, myriocin also mitigated thermal hypoalgesia exhibited by mice fed −SG HFD (Fig. 3a), suggesting that a reduction in SPT activity reduced serine-associated peripheral neuropathy.
To better understand the metabolic drivers of this peripheral neuropathy phenotype, we next quantified ceramides and deoxydihydroceramides (deoxyDHCer) in liver and sciatic nerve. Restriction of serine and glycine increased deoxyDHCer in LFD and HFD settings and reduced canonical ceramides in HFD background, whereas myriocin generally reduced abundances of sphingolipids ( Fig. 3b and Extended Data Fig. 6g-i). By contrast, ceramide and 1-deoxysphingolipid levels were unaltered or did not correlate with peripheral neuropathy phenotype in sciatic nerve, potentially owing to the large lipid deposits present in myelin (Extended Data Fig. 6h,i). Notably, serine restriction in a LFD background did not induce thermal hypoalgesia after six months (Fig. 3a), indicating that 1-deoxysphingolipid accumulation alone is insufficient to drive this phenotype and consistent with recent reports of an Sptlc1 C133W mouse model 33 . Next, we measured IENF density, corneal nerve density, and tissue lipids in a separate cohort of mice fed LFD, HFD, −SG HFD or −SG HFD plus myriocin for six months. Myriocin mitigated the onset of thermal hypoalgesia in mice fed −SG HFD (Fig. 3c), and this treatment also protected small fibre nerve density in the epidermis and cornea ( Fig. 3d and Extended Data Fig. 7a,b) without affecting tactile sensing or MNCV (Extended Data Fig. 7c,d), suggesting this early-onset phenotype is specific to small sensory fibres and 1-deoxysphingolipids, in contrast to other nodes in sphingolipid metabolism 34 . Mice fed −SG HFD exhibited increased hepatic 1-deoxysphingolipids and sphingomyelin, whereas myriocin strongly reduced the levels of sphingolipids as well Data are mean ± s.e.m. and minimum and maximum (g,h), and were analysed using two-way ANOVA with Fisher's least significant difference post hoc test (a-j).
Article as triglycerides and diacylglycerides (Extended Data Fig. 7e), further highlighting the pleiotropic effects of this molecule on the lipidome 32,35 . Finally, paw skin exhibited a significant increase in deoxyDHCer that was reduced with myriocin treatment (Fig. 3e), suggesting the lipid microenvironment surrounding small fibres can influence sensory function.
We next evaluated whether suppressing SPT activity could mitigate neuropathy in db/db mice, which have increased circulating deoxySA as well as increased hepatic ceramides and deoxyDHCer (Extended Data Fig. 8a,b). Consistent with results in the −SG HFD diet-induced peripheral neuropathy model, administration of myriocin at six weeks of age prevented progression to thermal hypoalgesia and restored tactile sensation in db/db mice (Extended Data Fig. 8c,d). IENF density was also increased in db/db mice dosed for eight weeks with myriocin (Extended Data Fig. 8e). Although myriocin did not affect body weight gain, hyperglycaemia or plasma serine levels (Extended Data Fig. 8f-h), it strongly reduced canonical sphingolipids in the liver but showed limited effects on paw skin 1-deoxysphingolipids and ceramides (Extended Data Fig. 8i-n). Thus, myriocin probably acts through both direct and indirect mechanisms targeting liver and other tissues, which also accounts for its toxicity 35 .

Discussion
Here we describe how direct or indirect induction of chronic, systemic serine deficiency alters lipid homeostasis and contributes to diabetic peripheral neuropathy. Modulating dyslipidaemia with myriocin or 1-deoxysphingolipid biosynthesis with serine supplementation both mitigate thermal and tactile hypoalgesia in obese diabetic mice. These results highlight how serine deficiency can synergize with dyslipidaemia to alter neurological phenotypes both in rare disease contexts 16,25,26 and indirectly via a widespread, chronic disease such as type 2 diabetes, where it manifests as a co-morbidity experienced by a subset of patients. Reduced circulating serine and glycine in diabetic mice may be driven by increased flux through gluconeogenesis, one-carbon metabolism, renal retention 18,36 and/or disposal as acylglycines 37 , which are also influenced by dyslipidaemia 38 .
A STT, analogous to an oral glucose tolerance test (OGTT), could identify patients that exhibit elevated, postprandial serine disposal and who might be particularly susceptible to sensory neuropathy. Normalizing circulating serine levels via dietary supplementation delays the onset and progression of sensory neuropathy in db/db mice. Indeed, supplementation of serine and B vitamins improves peripheral neuropathy in some pre-clinical models and are the focus of clinical trials for various neurodegenerative disorders 39-41 (ClinicalTrials.gov: NCT03062449). Our results highlight physiologically relevant molecular links between serine and glycine homeostasis, sphingolipid metabolism, and diabetic co-morbidities. The metabolic and neuropathic phenotypes of mouse models of diabetes and obesity vary across strains and genotypes 42,43 , and C57BL6/J mice are particularly sensitive to the metabolic consequences of HFD owing to mutations in Nnt 44 and other genes. However, our data in both diabetic db/db mice and in C57BL/6J mice fed −SG HFD demonstrate that serine deficiency combined with dyslipidaemia can drive peripheral neuropathy in different genetic backgrounds.
Several key questions remain, including why serine and glycine deprivation suppresses hepatic fatty acid synthesis and gene expression in the liver. In addition, diverse sphingolipid species and/or their mis-localization contribute to neuropathy 21,45-48 , but their exact mechanisms of toxicity remain unclear. We have developed and validated a dietary model of serine-associated sensory neuropathy that develops a phenotype in three months, which could aid in understanding how neurotoxic dyslipidaemia can be managed. Diverse genetic changes may influence circulating serine and glycine, including common b S e r i n e G l y c i n e S e r i n e 10 14 Article single nucleotide polymorphisms and rare coding events 25,28,49 , or such deficiencies could be induced by diabetes-induced rewired hepatic metabolism. Thus, in both altering sphingolipid diversity and compromising the liver's ability to handle nutritional lipid overload, systemic serine deficiency emerges as a modifier of age-and diabetes-associated neuropathies.

Online content
Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-022-05637-6.

Mouse experiments
All mouse experiments were approved and conducted in accordance with the Institutional Animal Care and Use Committee (IACUC) of the University of California, San Diego and the Salk Institute for Biological Studies. Mice were housed in the same room ensuring exposure to the same temperature (21 °C), humidity (ambient humidity 65%) and a 12-h light:dark cycle (06:00-18:00). In Fig. 1, 14-to 16-week-old C57BL/6J ( JAX 000664) or BKS-db/db mice ( JAX 000642), and 10-to 12-week-old vehicle-or STZ-treated C57BL/6J ( JAX 000664) mice were fasted for 6 h prior to tissue collection. Animals were anaesthetized with isoflurane, decapitated, and tissues were rapidly collected using Wollenberger clamps pre-cooled to the temperature of liquid nitrogen and stored at −80 °C until analysis. For Fig. 2 Fig. 4, 6-week-old BKS-db/db mice ( JAX 000642) were fed with either a control or serine-supplemented diet (provided by Envigo) for a period of 8 weeks. Tissue collection took place between 07:00-10:00 h unless stated otherwise. Animals were anaesthetized with isoflurane, decapitated, and tissues rapidly collected using Wollenberger clamps pre-cooled to the temperature of liquid nitrogen and stored at −80 °C until analysis.

Serine tolerance test
Age-matched 14-to 16-week-old wild-type and BKS-db/db, and 10-to 12-week-old vehicle-and STZ-treated C57BL/6J ( JAX 000664) mice were fasted overnight with water access provided ad libitum. For a STT, animals were weighed, and serine and/or glucose were administered via oral gavage at a dose of 400 mg kg −1 and 2 g kg −1 , respectively, with tail tip blood samples collected into EDTA-coated microvette tubes (Sarstedt) before, and 15, 30, 60, 120 and 180 min after an oral gavage. EDTA microvettes were spun at 2,000g at 4 °C for 5 min to obtain plasma, and samples stored at −80 °C until analysis. Blood glucose and serine concentrations were quantified using Contour Next glucometer (Bayer) and gas chromatography-mass spectrometry as described below, respectively. Plasma serine pharmacokinetics were determined for a 400 mg kg −1 dose using PK solver 50 .
To qualify downstream fate of serine, wild-type mice were fasted overnight, weighed in the morning, and [U-13 C 3 ]serine administered via oral gavage at a dose of 400 mg kg −1 , with tissues collected, using Wollenberger clamps pre-cooled to the temperature of liquid nitrogen, before, and 15, 30, 45, 60, and 120 min after oral gavage, and samples stored at −80 °C until analysis.

Serum insulin and glucagon measurements
Commercially available kits were used to determine serum insulin (Mouse Insulin ELISA 10-1247-01, Mercodia) and glucagon (Glucagon ELISA 10-1271-01, Mercodia) following a 6-h fast in mice according to the manufacturer's instructions.

Lipogenesis D 2 O experiments
C57BL/6J mice fed diets for 18 weeks were injected intraperitoneally with D 2 O (in 0.9% NaCl) at a dose of 0.027 ml per g of body weight with drinking water replaced with 6% D 2 O-enriched solution for a period of ~18 h. In the morning (07:00-10:00 h) tissues were rapidly collected using Wollenberger clamps pre-cooled to the temperature of liquid nitrogen and stored at −80 °C until analysis.
Plasma D 2 O enrichment was determined using deuterium-acetone exchange protocol as previously described 24 . In brief, 5 µl of plasma were incubated with 4 µl of 5% acetone in acetonitrile solution and 4 µl of 10 M NaOH for 24 h. Next, 500 mg of Na 2 SO 4 and 600 µl of chloroform were added, and samples vortex-mixed. After 2 min centrifugation at 3,000g, 80 µl was transferred in triplicate into gas chromatographymass spectrometry (GC-MS) vials, and plasma D 2 O enrichment was quantified from an external standard curve on an Agilent DB-35MS column (30 m by 0.25 mm internal diameter × 0.25 µm, Agilent J&W Scientific) installed in an Agilent 7890 A gas chromatograph (GC) interfaced with an Agilent 5975 C mass spectrometer with the following temperature program: 60 °C initial, increase by 20 °C min −1 to 100 °C, increase by 50 °C min −1 to 220 °C, and hold for 1 min.
To quantify tissue D 2 O labelling, ~20 mg of frozen tissue was homogenized with 250 µl −20 °C methanol, 250 µl ice-cold saline and 500 µl −20 °C chloroform spiked with internal standards palmitate-d 31 (Cambridge Isotope Laboratories, DLM-215-PK) and coprostanol (Sigma, 7578). After a 5 min spin at 4 °C at 21,000g, the chloroform fraction was collected, dried, and resuspended with 500 µl of 2% H 2 SO 4 in methanol for 2 h at 50 °C. Next, 100 µl of saturated NaCl and 500 µl of hexane were added, samples vortex-mixed, and upper hexane phase collected and transferred into a GC-MS vial. Fatty acid methyl esters were analysed using a Select FAME column (100 m × 0.25 mm internal diameter) installed in an Agilent 7890 A GC interfaced with an Agilent 5975 C MS using the following temperature program: 80 °C initial, increase by 20 °C min −1 to 170 °C, increase by 1 °C min −1 to 204 °C, then 20 °C min −1 to 250 °C and hold for 10 min. The percent isotopologue distribution of each fatty acid and polar metabolite was determined and corrected for natural abundance using in-house algorithms adapted from a previous report 51 .

GTT and ITT
For GTT and ITT, C57BL/6J mice fed the diets for 18 weeks were fasted overnight with water provided ad libitum. In the morning animals were weighed and fasting blood glucose was determined from a tail bleed using a Contour Next glucometer (Bayer). For GTT, the animals were injected intraperitoneally with a bolus of glucose at a dose of 2 g kg −1 of body weight, and blood glucose determined at 15, 30, 60, 120 and 180 min post-injection. For ITT, the animals were injected intraperitoneally with a bolus of insulin (100 IU ml −1 Humulin Insulin, Eli Lilly) at a dose of 0.5 IU kg −1 , and blood glucose was quantified at 15, 30, 60 and 90 min post-injection as previously described 52 .

Determination of body composition and systemic metabolic rates
Lean and fat masses were determined using a EchoMRI 3-in-1 instrument (quantitative nuclear magnetic resonance (qNMR) imaging system). Comprehensive Laboratory Animal Monitoring System (CLAMS) (Oxymax, Columbus Instruments) was used to quantify systemic metabolic rates in individually housed mice during a period of 6 days. Water, food, and calorie intake were calculated from individually housed animals over a period of 6 days when subjected to CLAMS. Whole-body oxygen consumption (VO 2 ) and carbon dioxide (VCO 2 ) rates were normalized to corresponding total body weights, and RER was calculated as ratio of VCO 2 to VO 2 .

Faecal bomb calorimetry and calorie absorption
Approximately 1 g faeces was desiccated overnight and ground using a mortar and pestle. Powdered sample was reconstituted into a pellet with 300 µl ddH 2 O and weighed. Pellet was placed in bomb cylinder surrounded by 2,000 ml ddH 2 O (Parr 6100 Compensated Jacket Calorimeter). Heat produced by combustion was measured by change in water temperature. The calorimeter energy equivalent, W (Cal °C −1 ), was calculated with standardized benzoic acid. The final energy content of each pellet was calculated as follows: Calorie absorption was calculated by subtracting gross energy (faecal calorie extraction) from calorie intake.

Microbiome analysis
DNA was extracted from 10-30 mg of stool using the MoBio PowerFecal DNA isolation kit (12830-50). Extracted DNA was quantified using a Nanodrop (ThermoFisher Scientific). The whole-genome sequencing raw data was uploaded to Qiita 53 , where we followed their default processing workflow. In summary, the raw reads were adapter filtered using the auto-detect parameters in fastp version 20 54 and host (mouse) filtered using minimap2 version 2.17 55 . The resulting sequences were aligned using Bowtie 2 version 2.4.2 56 to the Web of Life (WoL) reference database 57 via the Web of Life Toolkit App (https://github.com/ qiyunzhu/woltka); this step generated tables at genus, species, per genome, and per gene tables. For all analyses we used the per genome table; then for alpha diversity we removed any samples below 1,273,062 sequences per sample and for beta-diversity analysis we rarefied at the same value. Downstream analyses were performed in QIIME 2 version 2020.11 58 . To asses global microbiota alterations, alpha diversity analysis was performed through Faith's PD 59 and beta diversity through robust principal component analysis (RPCA) 60 and resulting Aitchison distances were evaluated through permutational multivariate analysis of variance (PERMANOVA) 61 .
We then designed a Bayesian hierarchical model for differential abundance incorporating diet type as a fixed effect and cage as a random effect. We model the count generating process as a negative binomial distribution to account for overdispersion. Due to the sparsity of microbiome data, we also accounted for zero-inflation by assigning each microbe a probability of being unobserved separately from the count generating process: We wrote this model using the Stan probabilistic programming language 62 and fit the model using BIRDMAn (https://github.com/gibsramen/BIRDMAn). To account for compositionality, we fit this model using the first microbe in the table as an additive log ratio reference and converted log fold changes into centred log ratio coordinates after fitting. We used the following as prior distributions for the target parameters: in which i is the sample, j is the feature, y is the microbial count, θ is the indicator for non-biological zero, η is the mean feature count, x is the covariate, β is the regression coefficients to be estimated (log-fold changes), π is the probability of non-biological zero, z is the cage identifier variable, u is the random effect of cage, and ϕ is the overdispersion parameter. In order to compare functional changes associated with strain level differential abundances a comparative genomics pathway completeness approach was taken. First, each genome was assessed via MetaCyc 63 pathway completeness, a proportion ranging from zero to one, by mapping characterized genes to reactions and finally to pathways. Each pathway was then correlated by Spearman's rank correlation to the beta differential abundance determined from the above model. Serine biosynthesis, glycine cleavage, and fatty acid synthesis pathways were significantly correlated to betas. To validate these correlations, the log ratio of the sum of the abundance of genomes with complete pathways (completeness = 1) vs. those without (completeness < 1) were evaluated between treatment groups.
Behavioural assays Thermal sensation. Small sensory C fibre function was quantified by behavioural responses to heat using a thermal nociception test device (UARD) as previously described 64 . In brief, the apparatus surface was warmed up to 30 °C, and animals were placed in individual testing chambers for 20-30 min prior to testing. Four separate response latency measurements were performed, and the mean of the last triplicate taken to represent response latency for each animal. All measurements were made on coded animals by an observer unaware of the treatment groups.  ). Testing began with the 3.84 filament and the pressure applied was repeated five times. If a positive response was observed, the next lower weighted filament was used in the sequence. In the case of a negative response, the next higher weighted filament was applied. All measurements were made on coded animals by an observer unaware of the treatment groups.
Nerve conduction velocity. Conduction of a motor nerve was quantified in anaesthetized mice using EZ Anesthesia Versaflex system (Braintree Scientific, Z-7150). In brief, lightly anaesthetized mice were transferred onto a water-heated pad with anaesthesia maintained via a face mask. Two recording platinum electrodes were inserted between the animal's second, third, and fourth toes, and a grounding electrode into the skin at the neck. PowerLab stimulator delivered a 200-mV, 50-µs square-wave stimulus every 2 s. The stimulating electrode was inserted in the ankle near the Achilles tendon and subsequently into the sciatic notch at the hip, and M waves were recorded. The latency between Achilles tendon and sciatic notch was used to calculate nerve conduction velocity as described 64 . All measurements were made on coded animals by an observer unaware of the treatment groups.

Corneal confocal imaging
Quantification of corneal nerves was performed in anaesthetized mice (using EZ Anesthesia Versaflex system, Braintree Scientific, Z-7150) using Retina Tomograph 3 with Rostock Cornea Module (Heidelberg Engineering) equipped with Tomocap (Heidelberg Engineering, 0220-001) as previously described 64 . In brief, lightly anaesthetized mice were transferred onto a small animal platform with anaesthesia maintained via a face mask. Forty sequential images of uniform magnification and size were collected and those containing nerves of the sub-basal plexus identified. ImageJ software (ImageJ 1.53e Java 1.8.0_172) was used to quantify corneal nerve area within each image, with data presented as pixels/image. All measurements were made on coded animals and images by an observer unaware of the treatment groups.

Epidermal innervation
Quantification of epidermal innervation was performed in paw skin samples by immunostaining for the pan-neuronal protein PGP9.5, as described previously in detail 64 . In brief, paw skin samples were collected into 4% buffered paraformaldehyde (Thermo scientific, J19943-K2). Staining of epidermal nerves was performed using anti-PGP9.5 antibody (ProteinTech, 14730-1-AP; 1:500 dilution). Using 40× magnification of a light microscope, the number of PGP9.5-positive profiles present in the epidermis was calculated, length of skin section calculated, and IENF profile density expressed as profiles mm −1 . All measurements were made on coded slides by an observer unaware of the treatment groups.

Metabolite extraction and quantification
Plasma polar metabolites were extracted from 3 µl of plasma spiked with a known amount of 13 C-and 15 N-labelled standards (Cambridge Isotope Laboratories, MSK-A2-1.2). Tissue metabolite extraction was performed as described before 21 . In brief, ~20 mg of tissue was homogenized for 2 min using Precellys beads with 500 µl −20 °C methanol, 400 µl ice-cold saline and 100 µl ice-cold water and spiked with 13  ) standards, and normalization for 1-deoxydihydroceramides was done with the 13C-dihydroceramide-d7 standard. Homogenate aliquot of 50 µl was taken to determine tissue protein content using BCA protein assay (Lambda Biotech, G1002). The remaining homogenate was transferred to a 2 ml Eppendorf tube and 1 ml of −20 °C chloroform was added. Samples were vortex-mixed for 5 min and spun down for 5 min at 4 °C at 15,000g. The organic phase was collected and 2 µl of formic acid was added to the remaining polar phase which was re-extracted with 1 ml of −20 °C chloroform. Combined organic phases were dried and the pellet was resuspended in 100 µl of buffer containing 100% methanol, 1 mM ammonium formate and 0.2% formic acid. Data represents ion counts normalized by class-specific internal standards and tissue protein content, with stacked plots to represent acyl-chain distribution.
Targeted sphingolipid quantification. Quantification of sphingolipid metabolites was determined using triple quadrupole liquid chromatography-mass spectrometry platform (Agilent 6460). Sphingolipid species were separated on a C8 column (Spectra 3 µm C8SR 150 × 3 mm inner diameter, Peeke Scientific) as previously described 67 . Mobile phase A was composed of 100% HPLC-grade water containing 2 mM ammonium formate and 0.2% formic acid and mobile phase B consisted of 100% methanol containing 0.2% formic acid and 1 mM ammonium formate. The flow rate was 0.5 ml min −1 . The gradient elution programme consisted of the following profile: 0 min, 82% B; 3 min, 82% B; 4 min, 90% B, 18 min, 99% B; 25 min, 99%, 27 min, 82% B, 30 min, 82% B. Column re-equilibration followed each sample and lasted 10 min. The capillary voltage was set to 3.5 kV, the drying gas temperature was 350 °C, the drying gas flow rate was 10 l min −1 , and the nebulizer pressure was 60 psi. Sphingolipid species were analysed by selective reaction monitoring (SRM) of the transition from precursor to product ions at associated optimized collision energies and fragmentor voltages (Supplementary Table 2). Quantification of sphingolipid species was performed using spiked-in deuterated standards.

High-resolution LC-MS/MS of polar metabolites.
Around 10-20 mg of frozen tissue was extracted with 800 µl of −20 °C 5:3:2 acetonitrile: methanol:water solution spike with a known concentration of norvaline as an internal standard using the Precellys Evolution Homogenizer (Bertin Technologies) 68 . After extraction, a 50-µl aliquot was taken for protein quantification using BCA protein assay (Lambda Biotech, G1002), and the remaining extract was spun for 10 min at 21,000g at 4 °C. The supernatant was then transferred into a glass vial, and chromatographic separation and compound identification performed using Q Exactive Orbitrap MS with a Vanquish Flex Binary UHPLC system (ThermoFisher Scientific) on an iHILIC-(P) Classic, 150 mm by 2.1 mm, 5-mm particle, 200-Å (Hilicon) column at 45 °C. Chromatography was performed using a gradient of 20 mM ammonium carbonate, adjusted to pH 9.4 with 0.1% ammonium hydroxide (25%) solution (mobile phase A) and 100% acetonitrile (mobile phase B), both at a flow rate of 0.2 ml min −1 . The liquid chromatography gradient ran linearly from 80 to 20% B from 2 to 17 min and then from 20 to 80% B from 17 to 18 min and then held at 80% B from 18 to 25 min.

High-resolution LC-MS/MS of lipids.
Liver samples were extracted in 400 µl of −20 °C methanol using the Precellys Evolution Homogenizer (Bertin Technologies) spiked with EquiSPLASH labelled standard (Avanti Polar Lipids, 330731) and norvaline. After extraction, 50 µl aliquot was taken to quantify protein content using BCA protein assay (Lambda Biotech, G1002), and to the remaining extract were added 400 µl of −20 °C chloroform and 400 µl of ice-cold water. After vortexing for 5 min, samples were spun for 5 min at 4 °C at 15,000g, and the organic phase was collected. Two microlitres of formic acid were added to the remaining polar phase which was re-extracted with 400 µl of −20 °C chloroform, samples were vortex-mixed, and spun as described above. Combined organic phases were dried and the pellet was resuspended in 100 µl of isopropanol. Chromatographic separation and lipid species identification was performed using Q Exactive orbitrap mass spectrometer with a Vanquish Flex Binary UHPLC system (Thermo Scientific) equipped with an Accucore C30, 150 × 2.1 mm, 2.6 µm particle (Thermo) column at 40 °C. Five microlitres of sample was injected. Chromatography was performed using a gradient of 40:60 v/v water: acetonitrile with 10 mM ammonium formate and 0.1% formic acid (mobile phase A) and 10:90 v/v acetonitrile: propan-2-ol with 10 mM ammonium formate and 0.1% formic acid (mobile phase B), both at a flow rate of 0.2 ml min −1 . The liquid chromatography gradient ran from 30% to 43% B from 3-8 min, then from 43% to 50% B from 8-9 min, then 50-90% B from 9-18 min, then 90-99% B from 18-26 min, then held at 99% B from 26-30 min, before returning to 30% B in 6 min and held for a further 4 min.
Lipids were analysed in positive mode using spray voltage 3.2 kV. Sweep gas flow was 1 arbitrary units, auxiliary gas flow 2 arbitrary units and sheath gas flow 40 arbitrary units, with a capillary temperature of 325 °C. Full mass spectrometry (scan range 200-2,000 m/z) was used at 70,000 resolution with 10 6 automatic gain control and a maximum injection time of 100 ms. Data dependent MS2 (Top 6) mode at 17,500 resolution with automatic gain control set at 10 5 with a maximum injection time of 50 ms was used. Data were analysed using EI-Maven software, and peaks normalized to Avanti EquiSPLASH internal standard. Lipid species specific fragments used for identification and quantification are presented in the Supplementary Table 3.

Plasma sphingoid base extraction, hydrolysis and LC-MS analysis.
Plasma sphingolipids were processed as previously described with minor modifications 69 . In brief, 50 µl of plasma was mixed with 0.5 ml of methanol and spiked with internal standards, sphinganine-d7, sphingosine-d7 and deoxysphinganine-d3 (Avanti lipids). The samples were placed on a mixer for 1 h at 37 °C, centrifuged at 2,800g and the supernatant collected and acid hydrolysed overnight at 65 °C with 75 µl of methanolic HCl (1N HCl,10M H 2 O in methanol). Next, 100 µl of 10 M KOH was added to neutralize. 625 µl of chloroform, 100 µl of 2N NH 4 OH and 500 µl of alkaline water were added, samples vortex-mixed and centrifuged for 5 min at 16,000g. The lower organic phase was washed three times with alkaline water and dried under air. LC-MS analysis was performed on an Agilent 6460 QQQ LC-MS/MS. Metabolite separation was achieved with a C18 column (Luna 100 × 2.1 mm, 3 µm, Phenomenex). Mobile phase A was composed of a 60:40 ratio of methanol:water and mobile phase B consisted of 100% methanol with 0.1% formic acid with 5 mM ammonium formate added to both mobile phases. The gradient elution programme consisted of holding at 40% B for 0.5 min, linearly increasing to 100% B over 15 min, and maintaining it for 9 min, followed by re-equilibration to the initial condition for 10 min. The capillary voltage was set to 3.5 kV, the drying gas temperature was 350 °C, the drying gas flow rate was 10 l min −1 , and the nebulizer pressure was 60 psi. Sphingoid bases were analysed by SRM of the transition from precursor to product ions at associated optimized collision energies and fragmentor voltages 16 . Sphingoid bases were then quantified from spiked internal standards of known concentration.

Serine dehydratase activity assay
Frozen liver and kidney samples were extracted in an ice-cold buffer containing 50 mM KH 2 PO 4 , 1 mM Na 2 EDTA, and 1mM DTT, pH 8.0 using glass homogenizer. Maximal enzyme activity was determined using coupled-enzyme reaction with lactate dehydrogenase (Sigma 10127230001) in the presence of 200 mM serine, 0.25 mM NADH, 0.17 mM pyridoxal phosphate, and 1 mM DTT for 3 min. Tissue homogenate protein quantification was subsequently determined using BCA protein assay (Lambda Biotech, G1002), and maximal enzyme activity expressed in international units (U) per mg of protein.

Statistical analysis
Data are expressed as mean ± s.e.m. unless stated otherwise. Statistical analysis was performed with Prism software (GraphPad Prism 9.3.1) using two-sided independent t-test to compare two groups, one-way ANOVA with Fisher's least significant difference post hoc test to compare more than two groups, two-way ANOVA with Fisher's least significant difference post hoc test to compare two-factor study design, and PERMANOVA analysis to explore RPCA plots. For all tests, P < 0.05 was considered significant. All data points in the manuscript represent individual biological replicates. No statistical methods were used to predetermine sample size.

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

Data availability
Source data for microbiome algorithms and immunoblots are provided as Supplementary Information. The whole-microbiome genome sequencing raw data was uploaded to Qiita 53 , where we followed the default processing workflow. High-resolution and targeted mass spectrometry data data is available at the NIH Common Fund's National Metabolomics Data Repository (NMDR) website, the Metabolomics Workbench, https://www.metabolomicsworkbench.org where it has been assigned Project ID M8JD81 (https://doi.org/10.21228/M8JD81). The data can be accessed directly via the Project ID M8JD81. Additional data that support the findings are available from the corresponding author upon reasonable request.