GDF15 promotes weight loss by enhancing energy expenditure in muscle

Mice

Gfral-null mice were generated as described previously³² and breeding pairs were provided by R. Seeley. The β-less mice were generated by B. Lowell as described previously⁵¹ and breeding pairs were provided J. Wu. Studies in mice were carried out at three sites (McMaster University, Université de Sherbrooke and Novo Nordisk). All of the animals used in the study were housed and cared for in accordance with the local guidelines for animal use, and studies were approved by the Animal Ethics Research Board of McMaster University (AUP: 210104), Université de Sherbrooke (2021–3001) and Danish Animal Experiments Inspectorate (2020-15-0201-00756:C01). All of the mice were group-housed on a 12 h–12 h light–dark cycle with ad libitum access to food and water. All of the mice used were males on the C57BL/6J background. All of the mice were housed in either the Solace Zone Heated IVC 32-Cage (Alternative Design Manufacturing & Supply) with HEPA filtered ventilation and temperature-regulated cages kept at 21 or 29 °C or in specific-pathogen free microisolators in a room kept at 21 or 29 °C. Before treatment with GDF15, mice were fed a high-fat, high-fructose diet (NASH Diet: 40 kcal% fat (mostly palm oil), 20 kcal% fructose and 0.02% cholesterol) (Research Diets, D19101102; for experiments performed at McMaster and Sherbrooke University) or western diet (Research Diets, D12079B; for experiments performed at Novo Nordisk). The mice used in chronic experiments at McMaster University were placed on a NASH diet and housed under thermoneutral conditions (~29 °C, 12 h–12 h light–dark cycle) or ambient temperature (~21 °C, 40–60% relative humidity) at 8 weeks of age. Mice at Novo Nordisk were placed on a housing condition (~29–30 °C, 12 h–12 h light–dark cycle) when treatment with GDF15. Before treatment, mice were randomized and separated into different treatment groups matched on body weight and composition, and single-housed. Recombinant GDF15 (Novo Nordisk, 0247-0000-0001; dissolved in synthetic buffer with 5 mM acetate, 240 mM propylene glycol and 0.007% polysorbate 20) was delivered by subcutaneous injection at the start of the light cycle in the morning (07:00–09:00) at 0.3, 1 and/or 5 nmol per kg once daily. The individual dosage was calculated by the body weight 1 day before. Pair-fed animals received the same amount of food as ingested by the corresponding GDF15-treated groups the day before. Body weight and food weight were recorded daily during the treatment period. Food intake was calculated by subtraction of the amount of food content in food hoppers from the amount added the previous day. Spillage and grind of food in cages was carefully monitored every day. Mice were euthanized in a fed state. Terminal blood was collected by cardiac puncture and blood from live animals was collected from a tail snip. Blood samples were centrifuged at 10,000 rpm for 10 min at 4 °C after clotting at room temperature for 30 min, and the supernatant was collected. Serum was saved and stored at −80 °C until use. Mice were anaesthetized using ketamine–xylazine (McMaster) or isoflurane (Novo Nordisk).

Pharmacokinetic analysis

Mice were placed on a NASH diet and housing condition (29 °C) for 20 weeks starting from 8 weeks of age. After a single subcutaneous injection of recombinant human GDF15 (0.3, 1 and/or 5 nmol per kg) in mice (n = 3 per group), tail blood was taken at 0, 0.5, 1, 2, 4, 8 and 24 h to measure recombinant human GDF15 concentrations in the serum using the human GDF-15 DuoSet ELISA kit (R&D Systems, DY957)⁵².

Body composition measurement

Body composition (lean and fat mass) was analysed using Bruker’s Minispec Whole Body Composition Analyzer (Minispec LF 90II) based on TD-NMR or an MR-scanner from EchoMRITM at the indicated time.

Measurement of temperature of mice

The core body temperature of the mice was measured by using a digital thermometer to test the rectal temperatures. The surface temperature of the mice was measured by a standardized infrared imaging technique using an infrared camera (T650sc, emissivity of 0.98, FLiR Systems) as described previously⁵³.

Metabolic activity

Metabolic monitoring was conducted using the Comprehensive Laboratory Animal Monitoring System (CLAMS, Columbus Instruments at McMaster) or Promethion system (Sable Systems International, Novo Nordisk). The experiment was conducted after acclimatization to the system for 12 or 24 h. Food intake, physical activity (beam breaks), oxygen consumption (VO₂), carbon dioxide production (VCO₂), RER and energy expenditure data were collected every 20 min (CLAMS) or 5 min (Promethion) for the indicated periods. Fatty acid oxidation (mg per kg per h) was calculated using the following equation (1.70 × VO₂ − 1.69 × VCO₂). Carbohydrate oxidation (mg per kg per h) was calculated using the following equation (4.58 × VCO₂ − 3.23 × VO₂).

GTT and ITT

Glucose tolerance tests (GTT) and insulin tolerance tests (ITT) were performed 3 and 2 weeks before euthanasia, respectively. Both tests were performed after a 6 h fast. For GTT, mice were injected intraperitoneally with 1.25 g per kg of d-glucose. For ITT, mice were injected intraperitoneally with 1.2 U per kg of insulin (Novorapid). For all tests, blood glucose was measured from a drop of tail blood using the ACCU-CHEK Aviva handheld glucometer (Roche) at 0, 20, 40, 60, 90 and 120 min after injection. Area under the curve analysis was performed using GraphPad Prism (v.9.3.0).

Biochemical analysis in the serum and liver

Triglycerides (Cayman Chemical, 10010303), non-esterified fatty acids (NEFA) (FUJIFILM Wako Diagnostics, 999-34691, 995-34791, 991-34891, 993-35191), insulin (Crystal Chem, 90080), glucose (FUJIFILM Wako Diagnostics, 997-03001), noradrenaline (Abbexa, abx055012) and ALT (Cohesion Biosciences, CAK1002) were measured according to the kit protocols.

Histological analysis

Liver and iWAT were collected and fixed with 10% neutral-buffered formalin for 36–48 h. After fixation, the samples were immersed in a 70% alcohol solution. The liver tissues were then processed, paraffin-embedded, serially sectioned and stained with H&E by the McMaster Immunology Research Centre histology Core Facility. Images were taken using the Nikon 90i Eclipse upright microscope. Blinded liver semiquantitative histology scores were assigned to liver sections by a pathologist. Ballooning degeneration of hepatocytes (0–2), steatosis score (0–3) and inflammation score (0–3) were evaluated according to H&E stained liver sections as described previously⁵⁴. NAFLD activity score (0–8) was defined as the sum of these three scores. Quantitative assessment of the size and number of adipocytes in the iWAT was performed using Image J as described previously⁵⁵.

Single-cell preparations and flow cytometry analysis

For the preparation of liver cells, a lobe of the liver was collected after perfusion of the liver with PBS and digested with enzyme solution buffer containing 0.5 mg ml⁻¹ pronase E, 0.088 U ml⁻¹ collagenase D and 1% (v/v) DNase I for 30 min at 37 °C. Single-cell suspension of liver non-parenchymal cells was prepared as previously described⁵⁶, with a minor modification. In brief, after digestion, the cells were filtered through a 100 μM cell strainer. After two centrifugation steps of 1 min at 50g to remove hepatocytes, the remaining cells in suspension were further filtered through a 40 μM cell strainer. The non-parenchymal single cells were centrifuged at 1,500 rpm for 5 min at 4 °C before proceeding to blocking/antibody staining for flow cytometry. For flow cytometry analysis, the cells were blocked with an antibody against Fc receptors (Fc block (1:200, BD Biosciences, 553142)) and stained for 30 min on ice with an antibody cocktail: CD45.2 BV510 (1:25, BioLegend, 109838), CD11b APC-Cy7 (1:100, Invitrogen, A15390), F4/80-APC (1:100, Invitrogen, 17-4801-82), CD3 BV605 (1:50, BD Biosciences, 563004), CD4 PerCP-Cy5.5 (1:100, BD Biosciences, 550954). 7AAD (1:100, Thermo Fisher Scientific, A1310) was used as a cell viability marker. After the staining, the cells were analysed with a CytoFlex Flow Cytometer (Beckman Coulter Life Sciences). Data analysis was performed using FlowJo (v.10.5).

RNA isolation, cDNA synthesis and qPCR

Tissues were homogenized and lysed in TRIzol reagent. After centrifuging, the supernatant (aqueous phase) was applied to the RNeasy kit (Qiagen, 74106) for subsequent total RNA extraction and purification according to its protocols. cDNA synthesis was performed using the SuperScript IV Reverse Transcriptase kit (Invitrogen, 18090010) according to the manufacturer’s instructions. The detection of cDNA expression for specific genes was performed by quantitative PCR (qPCR) using the AmpliTaq Gold DNA Polymerase kit (Applied Biosystems, N8080241). Taqman primers were purchased from Thermo Fisher Scientific. Relative mRNA levels were quantified using the ΔC_t method, using mouse Actb (Mm02619580_g1) as an endogenous control. Gene-specific primers were as follows: Ucp1 (Mm01244861_m1), Ppara (Mm00440939_m1), Ucp3 (Mm01163394_m1), Ppargc1a (Mm01208835_m1), Atp2a1 (Mm01275320_m1), Atp2a2 (Mm01201431_m1), Sln (Mm00481536_m1), Pln (Mm04206541_m1), Ckb (Mm00834780_g1), Ryr2 (Mm00465877_m1), Gpd2 (Mm00439082_m1), Ppard (Mm00803184_m1), Cox8b (Mm00432648_m1) and Cidea (Mm00432554_m1).

RNA-seq and transcriptomic analysis

Liver and tibialis anterior muscle were collected and snap-frozen in liquid nitrogen before storage at −80 °C. Frozen liver tissues or tibialis anterior muscle (30–50 mg per sample) were homogenized and lysed in TRIzol reagent. After centrifuging, the supernatant (aqueous phase) was applied to the RNeasy kit (Qiagen, 74106) for subsequent total RNA extraction and purification according to the manufacturer’s protocols. All RNA samples passed the BioAnalyzer quality control test. RNA-seq was performed using the Illumina NextSeq 2000 (P2 Flow cell, 2 × 50 bp configuration) system. MultiQC was used for quality control of raw data from RNA-seq⁵⁷. Trim Galore was used to automate quality and adapter trimming as well as quality control. We quantified the expression of transcripts using RNA-seq data through Salmon⁵⁸. Salmon’s transcript-level quantification DESeq2 was used to detect DEGs⁵⁹ using the following threshold: for liver samples, |log₂[fold change]| > 1, adjusted P < 0.05; for tibialis anterior muscle samples: |log₂[fold change]| > 0.6, adjusted P < 0.1. PCA was performed using VST data through DESeq2. Functional enrichment analysis was performed by GO enrichment analysis⁶⁰ and Kyoto Encyclopedia of Genes and Genomes (KEGG) mapping⁶¹ using the GOstats (https://bioconductor.org) and KEGG.db (https://bioconductor.org) packages, respectively. The results were illustrated in a gene-concept network diagram using the cnetplot package (https://bioconductor.org). Transcriptomic analyses were performed using the Linux system, R and RStudio software. RNA-seq data of quadriceps samples from mice treated with β₂ agonist clenbuterol were downloaded from the NCBI Sequence Read Archive under reference number PRJNA756816 (ref. ⁴⁴). We quantified the expression of transcripts using RNA-seq data through Salmon⁵⁸. Sln expression in the quadriceps was determined using VST data.

PET–CT Imaging

Male C57Bl/6J mice were placed on a NASH diet and thermoneutral housing conditions (29 °C) for 4 weeks starting from 7 weeks of age. Using a randomized crossover design (n = 13), mice were fasted for 7 h and then received a single subcutaneous injection of either a vehicle or GDF15 (5 nmol per kg), 4 h before a sequential dynamic PET acquisition with [¹¹C]acetate and [¹¹C]palmitate. Experimental sessions were then repeated 7 days later. PET was performed with the avalanche photodiode-based small-animal LabPET8 scanner of the Sherbrooke Molecular Imaging Center (Centre de recherche du CHUS, Université de Sherbrooke). Mice were first anaesthetized (2% isoflurane in 1.5 L min⁻¹ of oxygen) then injected intravenously with a 10 MBq bolus of [¹¹C]acetate (100 µl final volume in saline solution) through the caudal vein followed by a 15 min list-mode PET acquisition. Then, a bolus of 10 MBq of [¹¹C]palmitate (100 µl final volume in saline solution) was injected, and a 15 min list-mode PET acquisition was performed. Residual [¹¹C]acetate activity during [¹¹C]palmitate acquisition was corrected by acquiring a 60 s frame before the injection of [¹¹C]palmitate, accounting for the disintegration rate of [¹¹C]. Finally, low-dose CT images were acquired from the integrated X-O small animal CT scanner on the Triumph platform.

For [¹¹C]acetate scans, images were reconstructed into 26 dynamic frames (12 × 10, 8 × 30 and 6 × 90 s), whereas [¹¹C]palmitate scans were reconstructed into 29 dynamic frames (1 × 60, 12 × 5, 6 × 10, 6 × 30 and 4 × 150 s) using a three-dimensional maximum-likelihood estimation method with 20 iterations, span of 63, field of view of 80 mm with a final matrix resolution of 160 × 160 × 128 and a voxel size of 0.5 × 0.5 × 0.597 mm. For the [¹¹C]acetate, input curves were extracted as described previously⁶². In brief, using Amide (v.1.0.4), an image-derived input function (IDIF) was obtained by manually positioning a region of interest (ROI) in the vena cava, above the kidneys and below the myocardial blood pool. The [¹¹C]acetate IDIF was then corrected for [¹¹C]-labelled metabolites⁶³. Tissue ROIs were drawn on the liver, kidneys, myocardium, white adipose tissue, iBAT, quadricep and gastrocnemius muscles. Quantitative data was obtained from the resultant time-activity curves and used to estimate tissue blood flow index (on the basis of the uptake rate of [¹¹C]acetate, k₁ in ml g⁻¹ min⁻¹), oxidative metabolism index (the rapid fractional tissue clearance, k₂ in min⁻¹, of [¹¹C]acetate) and non-oxidative disposal (trapping of ¹¹C in tissue as free [¹¹C]acetate or other metabolites, such as lipids, k₃ in min⁻¹) using a standard two-compartment, two-tissue, kinetic model⁶³. For BAT, a four-compartment, two-tissue, kinetic model was applied, as previously described^64,65. For the [¹¹C]palmitate, IDIFs were obtained as described for [¹¹C]acetate and ROIs were drawn on the liver, myocardium, kidneys and iBAT. Fatty acid oxidation, esterification and uptake, and triglyceride release rates were calculated using a three-compartment, two-tissue, kinetic model^66,67.

Cell culture of myotubes

C2C12 cell line was purchased from ATCC and authenticated by short tandem repeat profiling at ATCC and tested negative for mycoplasma contamination. C2C12 cells were maintained in Dulbecco’s modified Eagle medium (DMEM) containing 10% fetal bovine serum at 37 °C in 5% CO₂. After reaching confluence, cells were differentiated to myotubes in DMEM supplemented with 2% horse serum for 5–7 days. Myotubes were treated with vehicle, GDF15 (10 nM) and noradrenaline (10 μM) for 15 h. Myotubes were collected, and RNA isolation, cDNA synthesis and qPCR were performed as described above.

Ex vivo determination of fatty acid oxidation in soleus muscle

Soleus muscles were carefully dissected tendon to tendon for muscle incubations as described previously^68,69. Fatty acid metabolism experiments were conducted using procedures previously described^68,69. In brief, isolated soleus muscles were placed in warmed (30 °C) Krebs–Henseleit buffer pH 7.4 containing 2 mm pyruvate, 4% fatty-acid-free bovine serum albumin and 0.5 mm palmitic acid. After an initial incubation of 15–30 min in a glass vial, the incubation buffer was replaced with Krebs–Henseleit buffer supplemented with 0.5 μCi ml⁻¹ [¹⁴C]palmitate (PerkinElmer, NEC534250UC) for 60 min with vials containing 450 μl of benzethonium hydroxide. Muscles were removed at the end of the chase period and rinsed and frozen under liquid nitrogen for later use. A total of 1 ml of acetic acid was then carefully added to the glass vial, which was immediately sealed. The acetic acid liberates CO₂ produced by fatty acid oxidation through the TCA cycle. Glass vials were then placed on a shaker at 75 rpm for 1 h to allow for benzethonium hydroxide to trap the released CO₂. The inner Eppendorf tube containing benzethonium hydroxide was carefully placed into a plastic scintillation vial containing 5 ml of scintillation fluid and allowed to quench overnight in the dark. Half of the soleus muscle were homogenized in 1.5 ml of chloroform:methanol solution (2:1). Then, 2.0 ml of distilled H₂O was added to the supernatant fraction in the new tube and vortexed gently. The aqueous phase was then transferred to a plastic scintillation vial containing 5 ml of scintillation fluid. DPMs were measured by a scintillation counter (Beckman coulter, LS 6500 multi-purpose scintillation counter). Data are represented as the sum of DPMs from the CO₂ and acid soluble intermediates and normalized to tissue weight.

Chemical denervation of iBAT

Denervation of iBAT was achieved by a local injection of 6OHDA (10 mg ml⁻¹) in saline containing 1% ascorbic acid into five distinct spots across the iBAT pad (5 μl per spot)⁷⁰. Mice were allowed to recover for 48 h. For the confirmation of BAT denervation, we treated mice with the β₃ agonist CL-316,243 and measured BAT temperature using a standardized infrared imaging technique using an infrared camera (T650sc, emissivity of 0.98, FLiR Systems)⁵³ and energy expenditure in CLAMS^37,71,72.

Muscle functional testing

Muscle functional testing was performed in vitro using the horizontal bath of a whole-mouse test system (1300A, Aurora Scientific). Ringer’s solution (120 mM NaCl, 4.7 mM KCl, 2.5 mM CaCl₂, 1.2 mM KH₂PO4, 1.2 mM MgSO₄, 25 mM HEPES, 5.5 mM glucose) within the horizontal bath was bubbled with oxygen for 30 min before experimental initiation. In brief, for muscle functional testing, the EDL muscle was isolated and, using braided silk at both the proximal and distal muscle tendon junctions, secured to both a stationary lever arm hook and a force transducer (model 809c, Aurora Scientific) within the horizontal bath. In this position, the EDL muscle is aligned in parallel between two stimulation electrodes. The EDL muscle was allowed to rest for 10 min before stimulation. To determine the optimal muscle length, the EDL muscle was stimulated at different resting tensions until a maximum twitch tension was determined⁷³. The EDL muscle rested for 2 min after this optimization. A force–frequency curve was used to determine peak tetanic force. This force determination consisted of a 1 s stimulation every 30 s beginning at 10 Hz and increasing in stimulation frequency in 10 Hz increments. All data were collected and analysed using Dynamic Muscle Control and Analysis Software (v.615A, Aurora Scientific).

Immunohistochemical staining of mouse muscle

H&E staining was conducted on frozen gastrocnemius muscle sections using standard protocols. Within each gastrocnemius, the entire muscle cross-section was visualized and imaged to evaluate the whole cross-section in its entirety. Muscle fibre typing was performed as described previously⁷⁴. Immunofluorescence was visualized using the Nikon Eclipse 90i microscope (Nikon) and analysed using NIS-Elements AR software (Nikon, v.5.41.02). To determine fibre-type percentage, a total of 4,000 fibres (types I, IIa, IIb and IIx) were counted per gastrocnemius cross-section (n = 4 per group)⁷⁵.

Respiration in mitochondria isolated from red skeletal muscle

Skeletal muscle mitochondria were isolated using temperature controlled (4 °C) differential centrifugation as previously described⁷⁶. In brief, hindlimb skeletal muscles (red gastrocnemius, plantaris, red tibialis anterior, soleus and red portion of the quadriceps) were excised, trimmed of visible fat and connective tissue, weighed and minced in isolation buffer (100 mM sucrose, 100 mM KCl, 50 mM Tris-HCl, 1 mM KH₂PO₄, 0.1 mM EGTA, 0.2% BSA and 1 mM ATP, pH 7.4). Minced tissue was homogenized and centrifuged at 800g for 10 min to separate the subsarcolemmal and intermyofibrillar mitochondrial fractions. The pellet containing intermyofibrillar mitochondria was resuspended and treated with a protease subtilisin A (0.025 g per mg wet tissue) for 5 min and ice-cold isolation buffer was subsequently added to stop the protease. The samples were immediately centrifuged at 5,000g for 5 min and the pellet was resuspended and centrifuged at 800g for 10 min to liberate the intermyofibrillar mitochondria in the supernatant. Subsarcolemmal and intermyofibrillar mitochondria were further centrifuged at 10,000 g for 10 min, resuspended, and combined before centrifugation twice at 10,000g for 10 min to recover the final mitochondrial pellet. These pellets were resuspended in Mg²⁺ absent MiR05 (0.5 mM EGTA, 60 mM potassium lactobionate, 10 mM KH₂PO₄, 20 mM HEPES, 110 mM sucrose, 1 g l⁻¹ fatty acid free BSA, pH 7.2) and kept on ice until respiration experiments were conducted.

Respiration experiments were performed in the Oroboros Oxygraph-2k system at 37 °C with constant stirring. A total of 20 µg of mitochondrial protein was loaded per 2 ml chamber (quantified by Bradford protein assay). ADP/O ratios were calculated using the change in oxygen content (nmol) after the addition of ADP (separately following a 50 µM (100 nmol) bolus and, when depleted, a 100 µM (200 nmol) bolus) in the presence of 5 mM pyruvate and 2 mM malate. Maximal respiration was assessed with subsequent additions of 5 mM ADP, 10 mM glutamate (maximal complex I supported respiration) and 10 mM succinate (maximal complex II supported respiration). RCRs were quantified as the ratio of state 3 (saturating ADP) to state 4 (presence of pyruvate and malate, absence of ADP) respiration. In a separate experiment, 0.5 µM oligomycin was initially added to the chamber and respiration determined in the presence of saturating mixed substrates (pyruvate, malate, ADP, glutamate and succinate).

Mitochondrial purity was checked by western blotting as previously described⁷⁷. In brief, mitochondrial samples were added on top of 1 ml 60% Percoll (1.336 ml 5× SMEA, 4 ml Percoll, 1.334 ml distilled H₂O; density 1.08–1.12 g ml⁻¹). The samples were centrifuged at 20,000g for 5 h at 4 °C. The purified mitochondrial layer was removed and suspended in 1 ml isolation buffer and centrifuged at 12,000g for 10 min at 4 °C to remove the residual Percoll. The final pellet was resuspended in isolation buffer at stored at −80 °C until the samples were prepared for western blotting. The primary antibodies included COXI (1:500, OXPHOS cocktail, Abcam, Ab110413), COXIV (1:30,000, Invitrogen, A21347), GLUT4 (1:2,500, Abcam, Ab654), calnexin (1:2,000, Sigma-Aldrich, C4731) and SERCA2 (1:1,000, Abcam ab2861).

Respiration in permeabilized muscle fibres from red skeletal muscle

Permeabilized muscle fibres were prepared from red gastrocnemius muscle as previously described^78,79. In brief, muscle was placed in ice-cold BIOPS (50 mM MES, 7.23 mM K₂EGTA, 2.77 mM CaK₂EGTA, 20 mM imidazole, 0.5 mM dithiothreitol, 20 mM taurine, 5.77 mM ATP, 15 mM PCr and 6.56 mM MgCl₂·H₂O, pH 7.1) and fibre bundles were separated with fine-tipped forceps underneath a microscope (MX6 Stereoscope, Zeiss Microsystems). Fibres were incubated in 40 µg ml⁻¹ saponin for 30 min and washed in MiR05 respiration buffer (respiration experiments) for 15 min. Mitochondrial respiration experiments were performed in MiR05 respiration buffer in an Oxygraph high-resolution respirometer at 37 °C (Oroboros Instruments) with constant spinning at 750 rpm. Experiments were conducted at room air saturation with reoxygenation after the addition of each substrate (~180–195 µM O₂). All experiments were performed in the presence of 5 µM blebbistatin, 5 mM pyruvate and 1 mM malate. For submaximal ADP experiments, ADP was titrated at various concentrations (25, 100, 250, 500, 1,000, 2,000, 4,000, 6,000, 8,000, 10,000 µM ADP) followed by the addition of 10 mM glutamate, 10 mM succinate and 10 µM cytochrome c. RCRs were calculated by dividing maximal state 3 respiration (presence of ADP) by state 4 respiration (pyruvate + malate, absence of ADP). Ca²⁺ experiments were performed with the addition of 5 mM ATP before titrations of CaCl₂ (25, 50, 100, 200, 250, 300, 350 µM CaCl₂) as recently reported⁸⁰. After reaching a plateau, 40 µM CPA was added to inhibit SERCA activity. During the Ca²⁺ titration, SERCA hydrolyses ATP, generating ADP to stimulate respiration. We therefore used the one-phase association curve from the ADP titration to estimate the ADP generated during the Ca²⁺ titration as an index of SERCA efficiency. The regression equations from the ADP titrations were as follows: Vehicle: JO₂ = 127 + (598 − 127) × (1 − exp(−0.0005292 × [ADP])), GDF15: JO₂ = 128 + (463 − 128) × (1 − exp(−0.0006502 × [ADP])), pair-fed: JO₂ = 119 + (590 − 119) × (1 − exp(−0.0005858 × [ADP])). Fibre bundles were recovered from all of the experiments, dried and weighed to normalize respiration to tissue weight (pmol s⁻¹ mg⁻¹ dry weight).

Respiration in isolated soleus muscles

Respiration experiments in isolated soleus muscles from mice were performed as previously described⁸¹ with minor modifications. Soleus muscles were isolated from obese mice fed the NASH diet for more than 13 weeks that were calorically restricted and injected with vehicle (pair-fed) or GDF15 for 21 days. In brief, soleus muscles were excised and placed in a sealed vial containing 2 ml of pre-gassed (95% O₂, 5% CO₂) modified Kreb’s Ringer (MKR) buffer at 30 °C (115 mM NaCl, 2.6 mM KCl, 1.2 mM KH₂PO₄, 10 mM NaHCO₃, 10 mM HEPES) supplemented with 4% BSA, 0.5 mM palmitate and 10 mM d-glucose for 1 h. After the 1 h preincubation, soleus muscles were transferred into respirometry systems (Oroboros O2k) containing hyper-oxygenated (starting [O₂], ~500 μM) supplemented MKR buffer with constant stirring at 30 °C. The rate of oxygen consumption (JO₂) was determined at the baseline for 20 min before the addition of 3 mM caffeine (Sigma-Aldrich, C0750) to stimulate calcium leak and increase ATP hydrolysis. After 10 min of caffeine-mediated respiration, dantrolene (Sigma-Aldrich, D9175, 10 μM in DMSO) was added to the chamber to assess the effects of inhibiting Ca²⁺ release. All soleus muscles were recovered, trimmed of any remaining tendon/connective tissue, blotted dry and weighed for the normalization of respiration to wet muscle weight. All experiments were determined in duplicate (paired soleus muscles) and technical replicates were averaged for each JO₂ determination.

Bioinformatics analysis of the GTEx dataset

We have access to the GTEx Analysis V8. The data used for the analyses described in this Article were obtained from dbGaP accession number phs000424.v8.p2 on 11 May 2022. To study the effect of physiological levels of GDF15 on skeletal muscle in humans, we assessed the raw RNA-seq gene count data from the muscle of 803 individuals. We compared the muscle gene expression by establishing two groups on the basis of the GDF15 expression level in the skeletal muscle using the R and RStudio software.

Correlation analysis of GDF15 and TSH in human

Blood samples were collected after an overnight fast from women with obesity (n = 22)³¹. TSH measurements were conducted by the Ottawa Hospital Laboratory Services. GDF15 levels were analysed using a Human GDF-15 Quantikine ELISA Kit according to the manufacturer’s instructions (R&D Systems, DGD150).

Correlation study on GDF15 and RMR in humans

The RMR of 154 participants was measured using a ventilated hood⁴⁵ (JAEGER Oxycon Pro, Viasys Healthcare). The measurement was performed after an overnight fast between 08:00 and 10:00. The hood was placed over the head of recumbent subjects. The measurement lasted for 40 min, during when the participants were required to keep still yet remain awake. The mean values of every 10 min were then calculated and the minimum values were used as the RMR of the participants. RMR was adjusted for body composition on the basis of TANITA data using our published equation Natural logarithm (Ln)_{BEE (basal expenditure)} = −0.954 + 0.707 Ln_{FFM (fat-free mass)} + 0.019 Ln_{FM (fat mass)} (ref. ⁴⁶). GDF15 levels in human plasma were tested by using human GDF-15 DuoSet ELISA kit (R&D Systems, DY957)⁵². GDF15 levels were corrected for weight and age by multiple linear regression using R.

2SMR using GWAS summary data

2SMR was performed using the exposure and the outcome from two non-overlapping and independent datasets to conduct the summary-level instrument exposure analysis and the instrument–outcome association analysis. 2SMR was performed using the TwoSampleMR R package (v0.5.6)⁸². To verify the causal effect of GDF15 on liver fat in humans, we performed 2SMR using the exposure dataset (GDF15, GWAS ID: ebi-a-GCST90011998, sample size: 21,758)^83,84 and outcome dataset (liver fat percentage, GWAS ID: ebi-a-GCST90016673, sample size: 32,858)^85,86. To examine the effect of GDF15 on liver volume in humans, we performed 2SMR using the exposure dataset (GDF15, GWAS ID: ebi-a-GCST90011998, sample size: 21,758) and outcome dataset (liver volume, GWAS ID: ebi-a-GCST90016666, sample size: 32,858)^85,86. We identified genetic variants (SNPs) associated with blood GDF15 protein levels in the GWAS catalogue dataset based on cis-pQTL (within 500 kb of the Gdf15 gene), and further selected proxy SNPs by linkage disequilibrium (LD)-clumping (p1=5e-08, clump=TRUE, p2 = 1e-07, r2 = 0.001, kb = 10000). After dropping duplicate exposure–outcome summary sets, we further performed sensitivity analyses, including heterogeneity statistics, horizontal pleiotropy and leave-one-out analysis. After confirming that there was no heterogeneity or horizontal pleiotropy, we next performed MR analysis and visualized the results using the scatter plot and forest plot functions in the TwoSampleMR R package. We used MR Steiger directionality test⁸⁷ to evaluate causal direction between GDF15 and liver fat in humans. The inverse variance weighted method was used to assess the significance of the causal effect of the exposure on the outcome. 2SMR was performed using R and RStudio.

Statistics

Statistical analyses were performed using GraphPad Prism (v.8.4.1, v.9.3.0) or R (v.4.2.3), RStudio software (v.1.3.1056). All values are reported as mean ± s.e.m. unless stated otherwise. Data were analysed using one-way or two-way ANOVA with Tukey’s, Dunnett’s or Šidák’s post-hoc tests where appropriate. Differences were considered to be significant when P < 0.05. Statistical significance of histological scores was evaluated using unpaired Mann–Whitney non-parametric tests. ANCOVA was used to correct for the influence of variability of covariates (for example, body mass) on main variates (for example, treatment). ANCOVA was performed and visualized using the HH package (v.3.1-47)⁸⁸ in R and RStudio software after checking the homogeneity of regression slopes. The correlation analysis was performed using Pearson’s product-moment correlation.

Reporting summary

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