GIPR agonism and antagonism decrease body weight and food intake via different mechanisms in male mice

Expression of Gipr has been demonstrated in various regions of the PNS^30,31. In light of its role in the bi-directional transfer of information between the periphery and the brain, the PNS is well positioned to control energy metabolism, not only by modulating glycaemia via regulation of sympathetic outflow to the skeletal muscle³², but also by promoting GIP-induced vasodilation in the mesenteric vasculature, including the adipose tissue^33,34. Considering these effects, we next assessed whether targeted Cre-mediated deletion of Gipr in neurons of the PNS affects energy and glucose metabolism. Mice with deletion of Gipr in neurons of the PNS were generated by crossing C57BL/6J Gipr^flx/flx mice^35,36 with C57BL/6J mice that express Cre-recombinase under control of the promoter for peripherin (MGI 3841120)³⁷. Peripherin is a neuronal intermediate filament protein with largely restricted expression in neurons of the PNS³⁸. Consistent with this, we found expression of peripherin largely absent in the hippocampus, DVC, hypothalamus, sciatic nerve, pancreas and white adipose tissue, but it had robust expression in the dorsal root ganglia (DRG) and trigeminal ganglia (Extended Data Fig. 2a). Outside the PNS, expression of peripherin was highest in the ileum, but with more than 31-fold lower expression relative to the trigeminal ganglion, and with even lower to absent expression in the cerebellum, cerebral cortex, midbrain, kidney, testis, pituitary, adrenal gland, stomach, duodenum, jejunum and colon (Extended Data Fig. 2b). Collectively, these data indicate that expression of peripherin is largely restricted to neurons of the PNS. In line with previous reports in rats showing that peripherin is expressed in only 46% of DRG neurons³⁹, we find expression of Gipr decreased by ∽43% in the DRG of Per-Cre⁺Gipr^flx/flx mice (Per-Gipr KO) relative to Per-Cre⁺Gipr^wt/wt (WT) controls, and without differences in relative expression of Gipr in either the hypothalamus, hindbrain, sciatic nerve, epididymal white adipose tissue, pancreas, cerebellum, cerebral cortex, pituitary, kidney, duodenum, jejunum, ileum or colon (Extended Data Fig. 2c–p). Consistent with the high expression of peripherin in the trigeminal ganglia and the DRG (Extended Data Fig. 2a,b), we also confirmed Per-Cre-mediated deletion of Gipr in the trigeminal ganglion and the DRG in Per-Gipr KO mice using RNAscope (Extended Data Fig. 2q,r).

We next compared the differentially expressed genes induced by either GIPR antagonism or GLP-1R agonism in the C35 Glut10, C35 GABA4 and C35 GABA3 clusters (Fig. 7e–g), that is, in the three neuronal populations most affected by GIPR antagonism (Fig. 7a). In all three neuronal clusters, we found a similar pattern of gene expression changes after treatment with the GIPR antagonist and acyl-GLP-1, with 29 genes in the C35 Glut10 cluster, 29 genes in the C35 GABA4 cluster and 45 genes in the C35 GABA3 cluster affected by both GLP-1R agonism and GIPR antagonism (Fig. 7e–g). Most of the genes downregulated by GLP-1R agonism and GIPR antagonism were associated with neural plasticity and synapse formation, including neuregulin 3 (Nrg3), neurexin 3 (Nrxn3), discs large MAGUK scaffold protein 2 (Dlg2), sodium leak channel, non-selective (Nalcn), neurotrimin (Ntm), leucine rich repeat and Ig domain containing 2 (Lingo2), leucine rich repeat containing 4C (Lrrc4c), interleukin 1 receptor accessory protein like 1 (Il1rapl1) and glutamate ionotropic receptor NMDA type subunit 2B (Grin2b) (Fig. 7e–g). Notably, Nrxn3, which encodes for a synaptic adhesion protein critical for maintaining synaptic function⁴³, was downregulated by GIPR antagonism and GLP-1R agonism in both the C35 GABA4 and the C35 GABA3 cluster (Fig. 7f,g), while Nrg3, which regulates excitatory synapse formation⁴⁴, was strongly downregulated in C35 Glut10 and C35 GABA3 neurons by GIPR antagonism but not by GLP-1R agonism (Fig. 7e,g). Furthermore, we found that Lrrc4c and Il1rapl1, both of which encode for factors that are involved in excitatory synapse formation^45–47, were downregulated by GIPR antagonism and by GLP-1R agonism in the C35 Glut10 cluster (Fig. 7e). In summary, these data not only show that GIPR antagonism and GLP-1R agonism act on the same neuronal populations in the DVC, but also that they similarly downregulate gene programmes implicated in synaptic plasticity and synapse formation.

To infer whether the observed changes in transcriptional gene programmes indicative of reduced synaptic plasticity by GIPR antagonism and GLP-1R agonism in the DVC translate to altered signalling in the hypothalamus, we performed a cell–cell communication analysis using the LIANA+⁴⁹ implementation of the CellPhoneDB⁵⁰ algorithm, and the receptor–ligand database from NeuronChat⁵¹ (Fig. 8h). Cell–cell communication analysis in dissociated single-cell data infers likely communication events from the expression of known ligand–receptor pairs across different cell types to predict interactions based on transcriptomic profiles. We found similar alterations in the probability of cell–cell communication events between C35 GABA4 and C35 Glut10 sender neurons and hypothalamic receiver neurons after treatment with acyl-GLP-1 or the GIPR antagonist compared with vehicle DIO controls, which clearly diverge from that of acyl-GIP (Fig. 8h). Treatment with the GIPR antagonist and acyl-GLP-1 both led to a reduction in the likelihood of Nrxn3-Nlgn1 signalling between C35 GABA4 neurons and Pomc.GLU-5 and Agrp.GABA-4 neurons, a change that was specific to these feeding-related neurons and absent in other hypothalamic neurons (Fig. 8h). Similarly, we observed a decrease in the likelihood of Nrxn1-Nlgn1 signalling from C35 Glut10 neurons to Pomc.GLU-5 and Agrp.GABA-4 neurons (Fig. 8h). We did not observe a difference between Pomc.GLU-5 and Agrp.GABA-4 neurons and all other hypothalamic neurons in these signalling pathways in mice treated with acyl-GIP or MAR709 (Fig. 8h and Extended Data Fig. 5i). Together, these data suggest that GLP-1R agonism and GIPR antagonism may exert their effects on energy balance by downregulating signalling from DVC C35 GABA4 and C35 Glut10 neurons to hypothalamic feeding circuits.

In non-neuronal cells, we found in the DVC the highest expression of Gipr in oligodendrocytes (Extended Data Fig. 6a–d), which further constituted the most affected cell type in this area after treatment with either acyl-GIP or MAR709 (Extended Data Fig. 6e–h). In contrast to this, while non-neuronal Gipr expression was also in the hypothalamus highest in oligodendrocytes (Extended Data Fig. 7a–e), this cell type was, in this area, among the least affected after treatment with either acyl-GIP or MAR709 (Extended Data Fig. 7f–i). We also found tanycytes and ependymal cells among the most affected cell types in all treatment groups in the hypothalamus, that is, cell types with privileged access to the third ventricle and which have previously been implicated in the food intake inhibitory effects of the GLP-1R agonist liraglutide⁵².

To compare cell types with the neuronal populations we identified, we integrated our DVC data with two publicly available murine DVC datasets from Hes et al.⁵³ and from Ludwig et al.⁵⁴. This presented a particular challenge as, in addition to the expected variation from different laboratories, each dataset had been produced using different experimental groups. As each dataset has multiple experimental groups, to correct for the variance between studies while preserving variance between cell types and experimental groups, we trained the single-cell variational inference (scVI) model⁵⁵ on the control groups from each study with the study as the batch key, and then integrated the experimental groups (Extended Data Fig. 8a–d). Most experimental groups from all three datasets integrated well, however, neurons from mice treated once daily with semaglutide in the Ludwig⁵⁴ dataset did not integrate well, suggesting that longer-term administration of GLP-1R agonists continue to have a large impact on DVC neuron cell state after the initial dose, although this difference may be confounded by the reduction in body weight (Extended Data Fig. 8e).

To validate our cell typology framework, we predicted the cell-type labels from our framework using progressive learning through scHPL⁵⁶ and compared them to the author-provided cell types for both the Hes⁵³ and Ludwig⁵⁴ datasets (Extended Data Figs. 9a and 10a). We observed good concordance of major cell types at the C12 annotation level between our data and both the Hes⁵³ and Ludwig⁵⁴ datasets; however, at the C35 level, most neuron subclusters mapped to the largest glutamatergic or GABAergic neuron cluster, probably owing to the persistent differences between cells from the different experimental conditions.

Experiments were performed in accordance with the Animal Protection Law of the European Union after permission by the Government of Upper Bavaria, or the Eli Lilly and Company Institutional Animal Care and Use Committee. Mice were double or single housed and, unless otherwise indicated, fed ad libitum with either a regular chow (1314, Altromin) or a HFD (58% fat, D12331, Research Diets) diet under constant ambient conditions of 22 ± 2 °C with constant humidity (45–65%) and a 12 h/12 h light/dark cycle. C57BL/6J Vgat-ires-cre knock-in mice were purchased from The Jackson Laboratory (028862) and crossed with C57BL6/J Gipr^flx/flx mice^35,36 to generate Vgat-cre^+/−Gipr^flx/flx (Vgat-Gipr KO) mice and Vgat-cre^+/−Gipr^wt/wt (WT) controls. Per-Cre mice³⁷ (MGI ID:3841120) were crossed with C57BL/6J mice for >10 generations before pairing with C57BL6/J Gipr^flx/flx mice^35,36 to receive Per-cre^+/−Gipr^flx/flx (Per-Gipr KO) mice and Per-cre^+/−Gipr^wt/wt (WT) controls. Body composition was analysed using a magnetic resonance whole-body composition analyser (EchoMRI).

For assessment of drug effects under room temperature (22 ± 2 °C), male age-matched mice were double housed and fed with a 58% HFD (D12331, Research Diets) for approximately 20 weeks, followed by random assignment into groups of matched genotype, body weight and body composition. Mice were treated at the indicated doses with either long-acting acyl-GIP (IUB0271)^13–15, acyl-GLP-1 (IUB1746)^13–15, the GIPR:GLP-1R co-agonist MAR709 (refs. ^13–15) or an acylated peptide GIPR antagonist ([N^α-Ac,L¹⁴,R¹⁸,E²¹]hGIP_(5–31)-K¹¹(γE-C16))²². All peptides were provided by the Novo Nordisk Research Center Indianapolis, and have been previously validated in vitro and in vivo for receptor specificity and their ability to decrease body weight in DIO mice^{13–15,22,41}. All sequences of the used peptides are published elsewhere^13–15,22. For assessment of drug effects under thermoneutrality (28 °C), 12–14-week-old male age-matched mice were acclimatized to the housing temperature 2 weeks before start of the studies. At study start, male C57BL6J WT, as well as global Glp-1r^−/− and Gipr^−/− deficient mice were continued to be housed at thermoneutrality (28 °C) and given ad libitum access to a HFD (60% fat, D12492; Research Diets) and treated with a single dose (30 mg kg⁻¹) of either a control mAb or a GIPR antagonist mAb²³ (synthesized and provided by Eli Lilly and Company).

Plasma levels of glucose and insulin were measured after 6 h fasting. For assessment of glucose tolerance, glucose was administered i.p. at a dose of 1.5–2 g kg⁻¹. For assessment of insulin tolerance, insulin (Humalog; Eli Lilly) was injected i.p. at a dose of 0.75–1.5 U kg⁻¹. HbA1c was assessed from fresh blood using the DCA Vantage Analyzer (Siemens). For assessment of glucose-induced insulin secretion, glucose was given orally at a dose of 4 g kg⁻¹ in 6 h fasted mice, followed by blood sampling at timepoints 0, 2, 5, 15 and 30 min after glucose administration. Commercially available enzyme-linked immunosorbent assays (ELISAs) were used according to the manufacturer's instruction to measure insulin (Crystal Chem Zaandam, 90080), total GIP (Sigma-Aldrich, EZRMGIP-55K), triglycerides (Wako Chemicals, 290-63701 or Abcam, ab65336) or total cholesterol (Thermo Fisher Scientific, 10178058).

Energy expenditure, food intake, respiratory exchange ratio (RER) and locomotor activity were assessed for 96–132 h, and after 24 h of acclimatization, in single-housed mice using the Promethion climate-controlled indirect calorimetric system (Sabel Systems). For assessment of acute food intake, mice were treated with either vehicle or acyl-GIP (IUB0271)^13–15 at the indicated doses, followed by measurement of food intake for 16 h. Data for energy expenditure were analysed using analysis of covariance (ANCOVA) with body weight as a covariate^70,71.

Total RNA was isolated using the RNeasy Kit (Qiagen) according to the manufacturer's instructions. cDNA synthesis was performed using the QuantiTect Reverse Transcription kit (Qiagen) or the High-Capacity cDNA Reverse Transcription kit (Thermo Fisher Scientific), according to the manufacturer's instructions. Gene expression was profiled using SYBR green (Thermo Fisher Scientific) and the Quantstudio 7 flex cycler (Applied Biosystems). The relative expression levels of each gene were normalized to the housekeeping gene peptidylprolyl isomerase A (Ppia), hypoxanthin-phosphoribosyl-transferase 1 (Hprt) or the TATA-binding protein (Tbp). The decision to use either Ppia, Hprt or Tbp was made based on the lowest variability of the housekeeper across the samples in the given tissue. Primer sequences were Ppia-F: 5′-GAG CTG TTT GCA GAC AAA GTT C-3′; Ppia-R: 5′-CCC TGG CAC ATG AAT CCT GG-3′; Hprt-F: 5′-AAG CTT GCT GGT GAA AAG GA-3′; Hprt-R: 5′-TTG CGC TCA TCT TAG GCT TT-3′; Gipr-F: 5′-GGC CCA GAT CAT GAC CCA AT-3′; Gipr-R: 5′-AGC CAA GAA GCA GGT AGC AG-3′; Prph-F: 5′-AAG TTT AAA GAC GAC TGT GCC TG-3′; Prph-R: 5′-TGC TGT TCC TTC TGG GAC TCT-3′; Tbp-F: 5′-GAA GCT GCG GTA CAA TTC CAG-3′; Tbp-R: 5′-CCC CTT GTA CCC TTC ACC AAT-3′. All raw CT values are stated in the Source data.

For brain isolation, mice were perfused with PBS, followed by 4% paraformaldehyde (PFA). Brains were then fixed for 24 h at 4 °C in 4% PFA and then transferred to 15% sucrose for 24 h, followed by 24 h in 30% sucrose at 4 °C. For DRG, trigeminal ganglion and nodose ganglion, tissues were extracted and fixed for 1–2 h in 4% PFA and transferred to 30% sucrose overnight at 4 °C. All tissues were frozen in Tissue-Tek O.C.T (Sakura Finetek, 4583), cut in 12–14 µm sections and placed on microscopic slides (Thermo Fisher Scientific, 10149870). The slides were heated for 30 min at 60 °C followed by antigen retrieval using a steamer, then processed by the RNAscope Multiplex Fluorescent Reagent kit v2 (Advanced Cell Diagnostics, 323270) according to the manufacturer's instructions. In brief, a custom-made probe was designed to bind to the deleted exons of mouse Gipr (Advanced Cell Diagnostics, 1138821-C1) and Vgat (Advanced Cell Diagnostics, 319191-C2) hybridized to the RNA, before preamplifiers, amplifiers and dyes were added for visualization of GIPR and Vgat. The slides were incubated with rabbit anti-peripherin antibody (Thermo Fisher Scientific, PA316723; 1:200) for 1 h at room temperature, followed by 30 min incubation with goat anti-rabbit-HRP (Thermo Fisher Scientific, A16096, 1:1,000) at room temperature. TSA vivid dyes 650 and 520 (Advanced Cell Diagnostics, 323271 and 323273, both 1:500 dilution) were added to detect GIPR, peripherin or vesicular GABA transporter (VGAT), respectively. Slides were counterstained with 4,6-diamidino-2-phenylindole (DAPI) (Advanced Cell Diagnostics, 320858) and imaged using Leica SP8 Laser Confocal Microscope using LAS X (version 3.5.7.23225).

Mice were killed by cervical dislocation, followed immediately by clamping of the bile duct and perfusion with collagenase P (Roche Diagnostics, 11249002001). Tissues were incubated in a 15 ml Falcon tube with 1 ml of collagenase P solution for 15 min at 37 °C, followed by the addition of 12 ml of cold G-solution (Sigma-Aldrich) and centrifugation at 586g at room temperature. The pellet was subsequently washed with 10 ml of G-solution (500 ml HBSS (Life Technologies, BE10-508F) with 10% BSA (Sigma-Aldrich, 126615-25 ml) and 1% penicillin–streptomycin (Life Technologies, 15140122)) and resuspended in 5.5 ml of gradient solution comprising 15% Optiprep (5 ml 10% RPMI (Life Technologies, 11875093) + 3 ml of 40% Optiprep that was diluted from 60% Optiprep with G-solution (Sigma-Aldrich, D1556)) per sample, and placed on top of 2.5 ml of the gradient solution. To form a three-layer gradient, 6 ml of the G-solution was added on the top. Samples were then incubated for 10 min at room temperature and centrifuged at 630g. The interphase was then collected and filtered through a 70 μm nylon filter (BD Falcon, 352350), before washing with G-solution. Islets were handpicked by a micropipette under the microscope and cultured in RPMI 1640 medium (Life Technologies, 11875093) overnight.

Culture medium was removed and islet microtissues were equilibrated for 1 h with Krebs Ringer HEPES buffer (131 mM NaCl, 4.8 mM KCl, 1.3 mM CaCl₂, 25 mM HEPES, 1.2 mM KH₂PO₄, 1.2 mM MgSO₄ and 2% BSA) containing 2.8 mM glucose. The supernatant was collected as a sample under the low glucose condition for 45 min incubation, and islets were incubated for another 45 min at 37 °C with Krebs Ringer HEPES buffer containing 16.7 mM glucose and supplements as above. The supernatant was collected as a sample under the high glucose condition and stored at −20 °C. For drug-induced insulin secretion, native mouse GIP or GLP-1 (provided by Novo Nordisk) were diluted in 1× KRK buffer with 20 mM glucose to reach a concentration of 50 nM. Cells were subsequently treated with either mouse GIP or GLP-1 for 45 min. Insulin concentrations were determined using a Mouse Insulin ELISA (Crystal Chem, 90082).

For snRNA-seq, 35-week-old DIO mice were treated 2 h before the end of the light phase with a single s.c. injection of either vehicle (PBS), acyl-GIP (150 nmol kg⁻¹)^13–15, acyl-GLP-1 (50 nmol kg⁻¹)^13–15, the GIPR:GLP-1R co-agonist MAR709 (50 nmol kg⁻¹)^13–15 or an acylated peptide GIPR antagonist (1,500 nmol kg⁻¹)²². The hypothalamus and DVC were collected 8 h after drug administration and stored in liquid nitrogen. Mice were euthanized followed by immediate decapitation and then the skull was removed. An earlier alignment of the brain was determined using a brain matrix and the entire hypothalamus (includes all the nuclei) was collected by microdissection. The hindbrain DVC was microdissected in an area postrema-centric manner after removal of cerebellar cortex. Tissue samples were flash frozen into liquid nitrogen and frozen tissues were stored in liquid nitrogen vapour phase for further processing to single-nuclei isolation. Nuclei were isolated using the 10X Genomics Chromium Nuclei Isolation kit including RNase Inhibitor (10X Genomics, PN-1000494), and using the 10X Genomics protocol for Single Cell Multiome ATAC + Gene Expression (10X Genomics, CG000505 Rev A). Nuclei concentration was determined using a Luna-II Automated Cell Counter (Logos biosystems, L40002) and adjusted to 6,250 nuclei per microlitre after pooling of n = 3 mice per sample. Nuclei were then processed using the 10X Genomics Chromium Next GEM Single Cell Multiome ATAC + Gene Expression (Rev. E) according to the manufacturer's instructions. Pooled samples were loaded into two lanes per group for a total of 24 lanes across three 10X Chromium chips. Equal numbers of cells per sample were loaded on a 10X Genomics Chromium controller instrument to generate single-cell gel beads in emulsion at the Helmholtz Munich Genomics Core Facility. Single-nucleus multimodal libraries were sequenced using the Illumina NovaSeq 6000. FASTQ files were generated from base calls with bcl2fastq software v2.20 (Illumina). Reads were mapped to the pre-built mm10 mouse reference (University of California Santa Cruz mm10 reference genome) using Cell Ranger ARC (v2.0.2, 10X Genomics) with default parameters. The resulting cell-by-peak and cell-by-gene matrices (ATAC and gene expression assays, respectively) from the 24 samples were integrated using Cellranger aggr (10X Genomics).

The raw gene expression matrix was filtered after removal of cells with either more than five mean absolute deviations more mitochondrial gene expression unique molecular identified counts, fewer than 500 detected genes or with more than 5 mean absolute deviations in total unique molecular identified counts. Scrublet⁷² was used to identify likely doublets, Leiden clustering was performed and clusters containing majority likely doublets were removed. After filtering, 211,537 nuclei from the hypothalamus and 57,798 nuclei from the DVC were used for further analysis. The processed gene expression matrix was imported into Scanpy (v1.9.8)⁷³ and normalized using Scran⁷⁴. The 4,000 most-variable genes were used for principal component analysis and the top 50 principal components were used for the Uniform Manifold Approximation and Projection (UMAP) visualization. We built a k-nearest neighbour graph for clustering using k = 50 nearest neighbours. Then, the Leiden clustering algorithm was used to group the cells into different clusters. To annotate hypothalamic cells, we used scArches⁷⁵ to transfer labels from the HypoMap⁴⁸ at the C66 cell annotation level. Then the expression of marker genes from the HypoMap was evaluated in each Leiden cluster, and then was manually annotated informed by marker gene expression and the majority cell type from scArches label transfer. The HypoMap hierarchical cell-type annotation framework was then used to map C25, C7 and C2 cell-type annotations. To annotate DVC cell types, as there is no comparable atlas and annotation framework to the hypomap for the DVC, each Leiden cluster was manually annotated into 35 fine-grained cell types (C35 cell type), and were then mapped to the coarser-grained C12 and C2 levels of cell type based on the expression of marker genes. DVC neurons were further subdivided into individual clusters labelled by major neurotransmitter expression.

For comparison of gene expression differences in DVC and hypothalamic neurons, the mean log fold difference in normalized expression of each gene was compared between the treatment groups and the DIO control group. Linear regression and Spearman correlation coefficients were calculated between treatment groups. Cell type prioritization was done using the Pertpy⁷⁶ implementation of Augur⁴², which uses a random forest classifier to assess how accurately the experimental condition of cells within a given cell type can be predicted based on their gene expression profiles. The Augur score is given by the performance of the classifier, measured as the area under the receiver operating characteristic curve. Cell type prioritization comparisons are always made between the experimental group and the DIO vehicle control. Differential gene expression analysis was performed using Scanpy's tl.rank_genes_groups function to identify genes that were differentially expressed between two experimental groups within a given cell type, genes with fewer than 30 counts were filtered for each comparison. The Wilcoxon rank-sum test was applied to assess differences in gene expression between groups. Default parameters were used, and multiple testing was accounted for by adjusted P values using the Benjamini–Hochberg method. Only genes with an adjusted P value <0.05 were considered statistically significant. Cell–cell communication between DVC and hypothalamic cells was performed using LIANA+⁴⁹ implementation of the CellPhoneDB⁵⁰ algorithm combined with the receptor–ligand database from NeuronChat⁵¹. DVC neurons were specified as sender cell types and hypothalamic neurons as receiver cell types.

We selected the 2,500 most variable genes in our DVC snRNA-seq data and used scVI⁵⁵ to integrate snRNA-seq data from Ludwig et al.⁵⁴ and Hes et al.⁵³, subset to the same 2,500 genes. We used treeArches to create a manual tree with three layers of granularity in cell type in our data. We trained the scVI model on the control groups from each dataset using the study as the batch variable, then updated the model with the experimental groups. We mapped our own data with the Hes⁵³ and Ludwig⁵⁴ datasets into a joint latent space using scArches⁷⁵, and then mapped the parameters for hierarchical progressive learning from scHPL v1.0.5 (ref. ⁵⁶) to predict the cell type from our annotation framework each cell type from the Hes⁵³ and Ludwig⁵⁴ datasets correspond to.

In vivo studies were performed in male or female age-matched mice that were randomly distributed to achieve groups of equal body weight and body composition. The number of independent biological samples per group is indicated in the figure legends and. No animals were excluded from the studies unless health issues demanded exclusion of single mice (for example, due to fighting injuries) as indicated in the. For in vivo studies, drugs were aliquoted by a lead scientist in number-coded vials and most, but not all, handling investigators were blinded to the treatment condition. Analyses of glucose and insulin tolerance were performed by experienced research assistants who did not know prior treatment conditions. Source data Source data

For animal studies, sample sizes were calculated based on a power analysis assuming that a body weight difference of ≥5 g between the treatment groups can be captured with a power of ≥75% when using a two-sided, two-tailed statistical test under the assumption of a s.d. of 3.5 and an α level of 0.05. Statistical analyses were performed using the statistical tools implemented in GraphPad Prism10 (version 10.0.3) and after testing of data for normal distribution using the Kolmogorov–Smirnov test, D'Agostino and Person test, Anderson–Darling test or Shapiro–Wilk test implemented in GraphPad Prism (version 10.0.3). Nonparametric tests such as the Mann–Whitney U test or the Kruskal–Wallis test were used to analyse data that were not normally distributed. Normally distributed data were analysed with the following parametric tests: two-tailed Student's t-test, one-way analysis of variance (ANOVA) or two-way ANOVA with time and genotype as co-variants followed by Bonferroni's post hoc multiple comparison test for individual timepoints. All data met the assumption of the statistical tests used. All results are given as mean ± s.e.m. P < 0.05 was considered statistically significant, with asterisks indicating significance at *P < 0.05, **P < 0.01 and ***P < 0.001. Differences in energy expenditure were calculated using ANCOVA with body weight as co-variate using SPSS (version 24). No data were excluded from the analysis unless for animal welfare reasons (for example, injury due to fighting) or identification of singular outlier using Grubbs test. Individual P values and outliers are shown in the Source data, unless P < 0.0001.

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