Main

The incidence of anxiety disorders has rapidly increased over the last two decades, and anxiety is now the most prevalent psychiatric disorder, affecting nearly 30% of the Western population at some point1. Acute stress-induced anxiety helps maintain the arousal and vigilance required to avoid repeated exposures to dangerous conditions2,3. Catecholamines, namely noradrenaline and adrenaline (international nonproprietary names: norepinephrine and epinephrine, respectively), are critical to coordinating the central and peripheral responses to psychological stress3,4,5,6.

Central noradrenaline, particularly from the locus coeruleus, is known to control acute anxiety-like behaviour3,4. Interestingly, acute peripheral administration of adrenaline also produces anxiety5,6,7,8,9, despite adrenaline not crossing the blood–brain barrier, and this effect is abolished by the β-blocker propranolol10,11,12,13. Central mechanisms could explain the effects of propranolol, which readily crosses the blood–brain barrier; however, treatment with practolol, a β-blocker that does not cross the blood–brain barrier, elicits similar anxiolytic effects14. These studies suggest that there may be peripheral endocrine signals linking β-adrenergic signalling with behaviour that have yet to be identified.

Growth differentiation factor 15 (GDF15) is a distant member of the transforming growth factor-β superfamily that circulates at low levels under normal physiological conditions but is elevated during cellular stress and mitochondrial diseases15,16. GDF15 signals through its receptor, glial-cell-derived neurotrophic factor receptor α-like (GFRAL), which is found exclusively in the brainstem17. GDF15–GFRAL signalling leads to behavioural changes, including reduced food intake18, nausea19 and aversive19,20 and anxiety-like behaviours20,21,22. These effects may involve activation of the sympathetic and hypothalamic–pituitary–adrenal (HPA) axes23,24,25,26, both of which are classic responses to psychological stress. In this study, we aimed to identify potential links between peripheral β-adrenergic activity and anxiety. We show that β-adrenergic stimulation activates adipocyte lipolysis, which promotes the secretion of GDF15 from adipose tissue-resident macrophages, ultimately linking changes in peripheral metabolism with behaviour.

Results

Psychological stress increases GDF15 through β-adrenergic signalling

Adipose tissue is a major endocrine organ that links metabolism with neuronal circuits and is acutely sensitive to changes in β-adrenergic signalling5,6,7,8,9,10,13,27. We treated mice with saline or adrenaline injected intraperitoneally (IP) (Fig. 1a). In agreement with previous studies28, adrenaline induced anxiety-like behaviour without affecting overall activity levels (Fig. 1b). To identify adipose-derived endocrine signals, we collected gonadal white adipose tissue (gWAT) and serum 1 h after treatment with saline or the same dose of adrenaline from another cohort of mice. As anticipated29, adrenaline produced marked and distinct separation in principal component analysis (Fig. 1c), with the most upregulated differentially expressed genes (DEGs) associated with the unfolded protein response and thermogenic and adaptive gene signatures (Fig. 1d and Extended Data Fig. 1a). From this gene set, we then isolated genes encoding secreted protein products (that is, ligands)30 and found that Gdf15 was the most highly upregulated (P = 6.59 × 10−29) secreted factor, showing a >4 log2-fold increase in adrenaline-treated mice compared with saline-treated mice (Fig. 1e and Extended Data Fig. 1b); indeed, Gdf15 was one of the most upregulated genes in gWAT. Corresponding with elevations in gWAT Gdf15 mRNA, GDF15 protein in serum was increased nearly five-fold within 1 h and remained elevated up to 4 h after injection (Fig. 1f). Reverse transcription followed by qPCR confirmed that adrenaline produced an approximately eight-fold increase in Gdf15 in gWAT without affecting Gdf15 in the liver, kidney, subcutaneous inguinal WAT (iWAT) or brown adipose tissue (Fig. 1g). Gdf15 expression was slightly increased in the lung and heart but was undetectable in skeletal muscle (Fig. 1g). Consistent with lower gWAT mass, serum GDF15 levels were lower in age-matched female mice than in male mice, but the relative response to adrenaline was similar between sexes (Extended Data Fig. 1c). Adrenaline also induced a similar increase in GDF15 whether mice were housed at room temperature or thermoneutrality (Extended Data Fig. 1d). These data indicate that adrenaline leads to marked increases in gWAT and serum GDF15 in mice.

Fig. 1: Acute psychological stress and adrenergic signalling increase GDF15 levels from gWAT.
figure 1

a, Schematic of acute vehicle and adrenaline treatment and analyses. b, Top: time in the centre and total distance travelled during an open-field test 1 h following treatment with adrenaline (n = 14) or saline (n = 13). Bottom: representative images of movements of individual mice. Data are presented as mean ± s.e.m. with P values calculated using an unpaired two-tailed t test. c, Principal component analysis of gWAT samples from saline-treated (n = 6) and adrenaline-treated (n = 6) mice using VST data from DESeq2. d, Gene-concept network diagram indicating the corresponding enriched Gene Ontology (GO) terms according to DEGs between saline-treated (n = 6) and adrenaline-treated (n = 6) gWAT. e, Volcano plot showing DEGs identified between saline- and adrenaline-treated gWAT. The P-adj was calculated using the Benjamini–Hochberg method. FC, fold change. f, Time course of circulating GDF15 before and after adrenaline (n = 5 mice) or saline (n = 3) treatment. Data are presented as mean ± s.e.m. with P values calculated using an unpaired two-tailed t test. g, Tissue screen of Gdf15 in mice 1 h after adrenaline (n = 8) or saline (n = 9) treatment. Data are expressed relative to the saline-treated liver. Data are presented as mean ± s.e.m. with P values calculated using an unpaired two-tailed t test. ND, not detected; BAT, brown adipose tissue. h, Serum GDF15 levels in mice from the following groups: maintained in their home cage (n = 5), with cage change for 30 min (n = 8), group-housed (n = 6) or single-housed for 3 days (n = 8). Data are presented as mean ± s.e.m. with P values calculated using an unpaired two-tailed t test. i, Schematic of an acute restraint test in WT and Adrb1, Adrb2 and Adrb3 triple knockout (BR−/−) mice. j, Serum GDF15 levels in WT and BR−/− mice following 4 h of tube restraint (WT n = 10, BR−/− n = 14) or the control condition (WT n = 9, BR−/− n = 12). Data are presented as mean ± s.e.m. with P values calculated using a two-way ANOVA with post hoc testing and Tukey’s correction. k, gWAT Gdf15 expression levels in WT and BR−/− mice following 4 h of tube restraint (WT n = 6, BR−/− n = 7) or the control condition (WT n = 5, BR−/− n = 6). a.u., arbitrary units. Data are presented as mean ± s.e.m. with P values calculated using a two-way ANOVA with post hoc testing and Tukey’s correction.

Source data

Previous studies characterizing GDF15-null mice found that these animals exhibit less anxiety-like behaviour in open-field and elevated-plus tests21,22. Psychological stress can be acutely induced in mice by changing their cages, whereas chronic single housing for 3 days is a form of long-term psychological stress31. We found that, consistent with adrenaline injection, acute cage change increased serum GDF15 levels; however, single housing did not (Fig. 1h). Consistent with our observations in mice that chronic stress and GDF15 are not linked, two-sample Mendelian randomization (2SMR) of GDF15 within the UK Biobank and serum GDF15 in children who were overweight and obese did not show associations with anxiety (Extended Data Fig. 2a–c). Although coding polymorphisms affect the measurement of circulating GDF15 in humans32, these data suggest that adrenergic-induced GDF15 may be involved in acute but not chronic psychological stress.

An acute psychological stress that reliably increases adrenaline in mice is physical restraint33. We observed increased circulating adrenaline following acute physical restraint, although the levels were lower than after treatment with IP injected adrenaline (Extended Data Fig. 1e). Adrenaline signals in adipose tissue through β-adrenergic receptors34. Therefore, to examine whether psychological stress induces GDF15 through β-adrenergic receptors, we physically restrained wild-type (WT) mice and mice lacking β1, β2 and β3 adrenergic receptors (Adrb1, Adrb2 and Adrb3 triple knockout (BR−/−)) in 50-ml conical tubes (Fig. 1i). Restraint of WT mice increased serum and gWAT GDF15; however, this response was eliminated in BR−/− mice (Fig. 1j,k). Baseline GDF15 levels were indistinguishable between genotypes and, importantly, lipopolysaccharide (LPS) increased serum GDF15 levels35 and lowered blood glucose levels similarly in WT and BR−/− mice (Extended Data Fig. 3a,b), demonstrating that these mice have an intact GDF15 response to non-β-adrenergic stimuli. These results show that psychological stress increases GDF15 and that this is dependent on β-adrenergic signalling.

ATGL is critical for promoting β-adrenergic-induced GDF15

Lipolysis is a major regulator of the transcriptional responses within adipose tissue29 and is increased by β-adrenergic-induced activation of adipose triglyceride lipase (ATGL), the rate-limiting enzyme for the release of free fatty acids from WAT36. To determine whether ATGL is responsible for regulating GDF15 secretion from adipose tissue, we treated adipocyte-specific ATGL-knockout mice (AdATGL−/−) and their floxed littermate controls (AdATGLflox/flox) with IP injected saline or CL316,243 (CL), a potent and selective β3-adrenergic agonist. ATGL expression was ~90% lower in gWAT of AdATGL−/− mice (Fig. 2a). CL increased serum nonesterified fatty acids and markers of the unfolded protein response, Atf4 and Chop, in gWAT of AdATGLflox/flox animals (Fig. 2b–d). These responses were eliminated in AdATGL−/− mice despite β-adrenergic signalling remaining intact, as evidenced by the similar increase in Ppargc1a expression and phosphorylation of cAMP-dependent protein kinase (PKA) substrates in AdATGL−/− and floxed mice (Fig. 2e,f). Consistent with our observations in BR−/− mice, basal GDF15 levels were similar between AdATGLflox/flox and AdATGL−/− mice; however, CL-induced increases in circulating and gWAT GDF15 were eliminated in AdATGL−/− mice (Fig. 2g,h). Treatment with IP injected cilostamide, a phosphodiesterase-3 inhibitor that leads to increased cellular cAMP levels, exaggerated the effects of CL on circulating GDF15 levels (Fig. 2i), supporting the involvement of the β3 adrenergic receptor–cAMP–ATGL pathway in stimulating GDF15.

Fig. 2: Adipocyte ATGL is critical for adrenergic-induced GDF15 secretion from the SVF.
figure 2

ag, gWAT Atgl expression (a), serum nonesterified fatty acids (NEFA) (b), gWAT Atf4 expression (c), gWAT Chop expression (d), gWAT Ppargc1a expression (e), gWAT phosphorylated PKA substrates with a representative Western blot image (f) and gWAT Gdf15 expression in AdATGLflox/flox (saline n = 5, CL n = 7) and AdATGL−/− (saline n = 7, CL n = 6) mice (g). Data are presented as mean ± s.e.m. with P values calculated using a two-way ANOVA with post hoc testing and Tukey’s correction. h, Circulating GDF15 levels in AdATGLflox/flox (saline n = 5, CL n = 7) and AdATGL−/− (saline n = 7, CL n = 6) mice. Data are presented as mean ± s.e.m. with P values calculated using a two-way ANOVA with post hoc testing and Tukey’s correction. i, Top: serum GDF15 after 1 h of CL and cilostamide cotreatment in mice (n = 6 per group). Data are presented as mean ± s.e.m. with P values calculated using a two-way ANOVA with post hoc testing and Tukey’s correction. Bottom: schematic of the acute restraint test in AdATGLflox/flox and AdATGL−/− mice. j, Serum glycerol levels in AdATGLflox/flox mice (control n = 6, restraint n = 6) and AdATGL−/− mice (control n = 13, restraint n = 11) following restraint. Data are presented as mean ± s.e.m. with P values calculated using a two-way ANOVA with post hoc testing and Tukey’s correction. k, Left: serum GDF15 in AdATGLflox/flox mice (control n = 6, restraint n = 6) and AdATGL−/− mice (control n = 13, restraint n = 11). Right: graph showing the fold change. Data are presented as mean ± s.e.m. with P values calculated using a two-way ANOVA (left) and an unpaired two-tailed t test (right). l, Glycerol levels in medium from white adipocytes treated with CL and/or ATGListatin (n = 9 per group; individual data points from three independent experiments). Data are presented as mean ± s.e.m. with P values calculated using a two-way ANOVA with post hoc testing and Tukey’s correction. m, GDF15 levels in medium from white adipocytes treated with CL and/or ATGListatin (n = 9 per group; individual data points from three independent experiments). Data are presented as mean ± s.e.m. with P values calculated using a two-way ANOVA. n, Gdf15 expression in adipocytes and SVF from gWAT (n = 6) and iWAT (n = 5) of mice. Data are presented as mean ± s.e.m. with P values calculated using a two-way ANOVA. o, Gdf15 expression in adipocytes and SVF from gWAT 1 h after saline (n = 7) or adrenaline (n = 9) treatment. Data are presented as mean ± s.e.m. with P values calculated using a two-way ANOVA with post hoc testing and Tukey’s correction.

Source data

Physical restraint stress promotes adipose tissue lipolysis37. Serum glycerol was increased in AdATGLflox/flox but not in AdATGL−/− mice in response to acute tube restraint (Fig. 2j). Similarly, circulating GDF15 levels were increased in AdATGLflox/flox but not in AdATGL−/− mice (Fig. 2k). We tested the behavioural response to acute restraint stress in these mice, but AdATGL−/− mice became hypoglycaemic following restraint, and this was associated with marked reductions in overall activity during open-field tests (Extended Data Fig. 4a–c), an effect that was likely secondary to the defect in adipose tissue lipolysis.

To evaluate whether β-adrenergic signalling increases GDF15 in a cell-autonomous manner, we cultured and differentiated mouse white adipocytes before treating them with CL and the ATGL inhibitor ATGListatin38. As expected, CL increased glycerol release in the medium, and this was blocked by ATGListatin; however, neither CL nor ATGListatin altered GDF15 levels in the medium (Fig. 2l,m). These data suggest that GDF15 is unlikely to be derived from adipocytes in response to adrenergic stimulation.

As adipocyte-specific ablation of ATGL inhibited GDF15 secretion but isolated adipocytes did not secrete GDF15 in response to CL, we hypothesized that another cell type within adipose tissue might be responsible for GDF15 secretion. To test this, we fractioned gWAT and iWAT into adipocytes and the stromal vascular fraction (SVF). Consistent with our hypothesis and recent studies39, Gdf15 expression was higher in the SVF of both WAT depots but particularly in gWAT (Fig. 2n). Moreover, in response to IP injected adrenaline, Gdf15 was increased in only the SVF (Fig. 2o), whereas the related cytokine, Fgf21, was increased in only the adipocyte fraction (Extended Data Fig. 5a). Appropriate separation was confirmed as leptin was found only in the adipocyte fraction (Extended Data Fig. 5b). These results point towards ATGL-dependent intercellular communication mediating β-adrenergic-induced adipose tissue GDF15 secretion.

Macrophages are the primary source of serum GDF15

To determine the probable cell type secreting GDF15 in adipose tissue, we analysed publicly available single-cell RNA-sequencing (scRNA-seq) data of cells from gWAT of CL-treated mice40. Cells were divided into either lineage marker positive (Lin+) (including monocytes, macrophages and B cells) or Lin (including fibroblasts and adipocytes). Gdf15 was primarily expressed in Lin+ cell types, especially in macrophages (Fig. 3a,b), with negligible expression in any Lin cell type, including adipocytes (Extended Data Fig. 5c). In response to CL treatment, there was increased Gdf15 expression only in the macrophage population (Fig. 3a). Because macrophage populations in adipose tissue are highly heterogeneous41, we used established gene signatures to identify three major clusters of macrophage populations: classically activated M1-like (Adgre1/F480 and iNos/Cd86/Cd80), alternatively activated M2-like (Adgre1/F480 and Cd163/Arg1) and macrophages that express Adgre1/F480 but no other markers for M1- or M2-like activation (Fig. 3c). We observed that CL led to the accumulation of both M1- and M2-like macrophages in gWAT, as noted in the original report40. CL increased Gdf15 in M2-like Adgre1/F480- and Cd163/Arg1-positive macrophages to a greater degree than observed in M1-like macrophages (Fig. 3c,d). We subsequently examined whether increases in Gdf15 also occurred after acute treatment with adrenaline by isolating F4/80+ cells from the SVF of gWAT (Extended Data Fig. 5d). Consistent with the scRNA-seq results, we found that Gdf15 expression was higher in F4/80+ cells than in other cells within the SVF and that adrenaline induced increases in Gdf15 in F4/80+ cells without any change in F4/80 cells of the SVF (Fig. 3e). Taken together, these data demonstrate that macrophages are the primary source of GDF15 in response to β-adrenergic-induced lipolysis.

Fig. 3: Adipocyte lipolysis mediates GDF15 secretion from alternatively activated M2-like macrophages through fatty acids.
figure 3

a, Top: t-distributed stochastic neighbour embedding plot of stromal vascular cells from gWAT of mice treated with CL for 3 days. Clustering identified ten major cell types or states. cDCs, conventional dendritic cells; NKT, natural killer T; NK, natural killer; VECs, vascular endothelial cells. Bottom: data were queried for cell clusters expressing Gdf15. exp., expression. b, Heat map showing the expression of Gdf15 and other cell-identifying factors in the various cell populations identified in scRNA-seq data. c, Data were queried to determine Gdf15 expression in the identified macrophage populations. d, Median normalized average Gdf15 expression in the M2-like, M1-like and macrophage populations from CL-treated mouse scRNA-seq data. The box-and-whisker plot is defined by the median (centre line) with the first quartile (Q1, lower line), third quartile (Q3, upper line), maximum (Q1 − 1.5 × interquartile range) and minimum (Q3 + 1.5 × interquartile range). Whiskers: 1.5 × (Q3 − Q1). P values were calculated using a one-way ANOVA with Benjamini–Hochberg correction for multiple tests. e, Gdf15 expression in mouse F4/80+ and F4/80 fractions (n = 3 per group). Data are presented as mean ± s.e.m. with P values calculated using a two-way ANOVA with post hoc testing and Tukey’s correction. f, Experimental schematic of the isolation, activation and treatment of BMDMs. g, GDF15 levels in medium from BMDMs treated with fatty acids (n = 3) or tunicamycin (n = 2). Individual data points represent triplicates from three independent experiments. Data are presented as mean ± s.e.m. with P values calculated using a two-way ANOVA with post hoc testing and Tukey’s correction. h, Gdf15 expression in BMDMs treated with fatty acids (n = 3) or tunicamycin (n = 2). Data points represent triplicates from three independent experiments. Data are presented as mean ± s.e.m. with P values calculated using a two-way ANOVA with post hoc testing and Tukey’s correction. i, GO enrichment analysis of M1- and M2-like macrophage populations from scRNA-seq data. The P-adj was calculated using the Benjamini–Hochberg method. j, GDF15 levels in medium from M2-like BMDMs treated with rosiglitazone (Rosi) or T0070907 (T007) (n = 9 per group). Individual data points represent triplicates from three independent experiments. Data are presented as mean ± s.e.m. with P values calculated using a one-way ANOVA with post hoc testing and Tukey’s correction. k, GDF15 levels in medium from M2-like BMDMs treated with palmitate (Palm) or T0070907 (T007) (n = 9 per group). Individual data points represent triplicates from three independent experiments. Data are presented as mean ± s.e.m. with P values calculated using a one-way ANOVA with post hoc testing and Tukey’s correction. f created using BioRender.com.

Source data

Fatty acids are the primary mediators of the transcriptional response to adipocyte lipolysis29. To determine whether fatty acids are responsible for GDF15 secretion from macrophages, we turned to bone marrow-derived macrophages (BMDMs) polarized with LPS + interferon-γ (IFNγ) (M1-like activation) or interleukin-4 (IL-4) (M2-like) or left untreated (M0) (Fig. 3f)42. We then treated BMDMs with two of the most abundant fatty acid species that are increased in the circulation by β-adrenergic agonists43, namely oleate, an unsaturated fatty acid, and palmitate, a saturated fatty acid. We also included tunicamycin, a compound that activates the unfolded protein response and is known to increase GDF15 (refs. 44,45). In line with our scRNA-seq results, palmitate and tunicamycin stimulated the secretion of GDF15 only in M0 and M2-like macrophages but not in M1-like macrophages (Fig. 3g). Similar results were seen at the transcriptional level, with Gdf15 expression being increased by palmitate only in M0 and M2-like macrophages (Fig. 3h). These results align with earlier work showing that adipocytes preferentially secrete palmitate, which is a critical regulator of macrophage metabolism46. Macrophage polarization was confirmed by the expression levels of Arg1 and iNos, which were elevated in M2- and M1-like macrophages, respectively (Extended Data Fig. 5e). Under basal conditions, GDF15 secretion into the medium was similar between M1-like and M2-like polarized cells, both of which showed greater GDF15 levels than unpolarized M0 cells (Extended Data Fig. 5f). Importantly, treatment of macrophages with adrenaline did not affect GDF15 secretion (Extended Data Fig. 5g), indicating that adrenergic signalling does not directly stimulate macrophage GDF15 secretion.

We then used these scRNA-seq data to compare the M1- and M2-like macrophage populations to identify possible mechanisms behind the M2-like-specific palmitate-induced GDF15 secretion. M2-like macrophages are known to use fatty acid metabolism to a greater degree than M1-like macrophages; consistent with this, our analysis of the scRNA-seq data found that fatty acid metabolic processes (Fig. 3i) and fatty acid transporters (Extended Data Fig. 5h) were upregulated in M2- compared with M1-like macrophages. Many fatty acid transporters are upregulated by the transcription factor peroxisome proliferator-activated receptor-γ (PPARγ), which is enriched in M2-like macrophages41 and was recently shown to increase the secretion of GDF15 from hepatocytes when activated47. To determine whether PPARγ is involved in controlling the secretion of GDF15, we treated M2-like BMDMs with rosiglitazone, a PPARγ agonist, and T0070907, a PPARγ antagonist. We observed that rosiglitazone increased GDF15 levels in the medium and that this effect was blocked by T0070907 (Fig. 3j). T0070907 also blocked palmitate-induced GDF15 secretion from M2-like macrophages (Fig. 3k). Together, these data show that palmitate drives the secretion of GDF15 specifically from macrophages polarized to an alternative M2-like phenotype through a mechanism that probably involves PPARγ.

Feeding mice a diet high in saturated fatty acids can promote anxiety-like behaviour (reviewed in ref. 48). Exogenous delivery of fatty acids such as palm oil by acute oral gavage increases the expression of GDF15 in the kidney and other tissues, and this leads to elevations in serum GDF15, an effect that peaks after 4 h (ref. 49). We tested whether an oral gavage of palm oil can produce anxiety-like behaviour after 4 h and, if so, whether this effect is blunted in mice lacking the receptor for GDF15, GFRAL (Gfral−/− mice). In contrast to adrenaline injection or tube restraint, treatment with palm oil did not increase anxiety-like behaviour in either genotype (Extended Data Fig. 6a). We next tested the effects of a high-fat diet predominantly composed of palm oil for 4 weeks and found similarly increased anxiety-like behaviour in both WT and Gfral−/− mice (Extended Data Fig. 6b,c). These results suggest that GDF15–GFRAL signalling is not involved in mediating anxiety-like behaviour in response to acute or chronic exogenous fatty acids.

GDF15 promotes anxiety-like behaviour through GFRAL

GDF15 regulates food intake and energy expenditure through its receptor, GFRAL15,18,26, the expression of which is isolated to the area postrema and nucleus tractus solitarius of the hindbrain17. To date, no other ligands besides GDF15 have been shown to signal through GFRAL. To establish whether GFRAL signalling is important for mediating anxiety-like behaviour, we treated WT or GFRAL-knockout (Gfral−/−) mice with IP injected adrenaline (as in Fig. 1a) 1 h before open-field tests. Adrenaline again induced anxiety-like behaviour in WT mice, but this response was lost in Gfral−/− mice (Fig. 4a) without affecting total activity (Fig. 4b). Anxiety is also reflected by nonambulatory movements, such as grooming50. Further supporting that GFRAL is necessary for the behavioural responses to acute stress, nonambulatory movements were increased in WT but not in Gfral−/− mice in response to IP injected adrenaline (Extended Data Fig. 7a); again, total physical activity was not altered by adrenaline (Extended Data Fig. 7b). Other responses to adrenaline, including food intake, energy expenditure and substrate use, were similar between genotypes (Extended Data Fig. 7c–e).

Fig. 4: GDF15 regulates behavioural and physiological responses to acute psychological stress through multiple anxiogenic brain regions.
figure 4

a, Time in the centre during the open-field test following treatment with saline (WT n = 5 mice, Gfral−/− n = 6) or adrenaline (WT n = 6, Gfral−/− n = 6). Data are presented as mean ± s.e.m. with P values calculated using a two-way ANOVA with post hoc testing and Tukey’s correction. b, Top: total movement during the open-field test following treatment with saline (WT n = 5, Gfral−/− n = 6) or adrenaline (WT n = 6, Gfral−/− n = 6). Bottom: representative images of movements of individual mice. Data are presented as mean ± s.e.m. with P values calculated using a two-way ANOVA with post hoc testing and Tukey’s correction. c, Left: schematic of the acute restraint test. Right: time in the centre during the open-field test following 4 h of tube restraint (WT n = 11 mice, Gfral−/− n = 8) or the control condition (WT n = 10, Gfral−/− n = 9). Data are presented as mean ± s.e.m. with P values calculated using a two-way ANOVA with post hoc testing and Tukey’s correction. d, Total distance travelled during the open-field test following restraint (WT n = 11 mice, Gfral−/− n = 9) or the control condition (WT n = 10, Gfral−/− n = 9). Data are presented as mean ± s.e.m. with P values calculated using a two-way ANOVA. e, Top: quantification of c-Fos+ cells in the bed nucleus of the stria terminalis (BNST), central amygdala (CeA) and paraventricular nucleus of the hypothalamus (PVN) of mice 90 min following treatment with IP injected GDF15. Bottom: representative pictures of the staining (n = 4 per group). Data are presented as mean ± s.e.m. with P values calculated using an unpaired two-tailed t test. Scale bar, 200 µm. 3V, third ventricle; ac, anterior commissure; BLA, basolateral amygdala; LV, lateral ventricle. f, Serum corticosterone following restraint (WT n = 9 mice, Gfral−/− n = 8) or the control condition (WT n = 9, Gfral−/− n = 8). Data are presented as mean ± s.e.m. with P values calculated using a two-way ANOVA with post hoc testing and Tukey’s correction. g, Serum GDF15 after treatment with CRH (WT n = 5 mice, Gfral−/− n = 6) or vehicle (WT n = 3, Gfral−/− n = 4). Data are presented as mean ± s.e.m. with P values calculated using a two-way ANOVA with post hoc testing and Tukey’s correction. h, Schematic summary of the mechanism of GDF15-mediated anxiety-like behaviour. h created using BioRender.com.

Source data

We next tested whether GFRAL is important for the behavioural responses to psychological stress by applying acute physical restraint before the open-field test. As expected3, restraint stress reduced the time spent in the centre, but, remarkably, this response was lost in Gfral−/− mice (Fig. 4c). Importantly, this effect was independent of genotypic differences in the total activity levels during the open-field test, which were similarly reduced regardless of genotype (Fig. 4d). Anxiety-like behaviour was distinct from food intake or nausea, both of which are altered by stress and can be controlled by GDF15 (ref. 19), given that the same restraint protocol did not affect chow or kaolin clay consumption in WT or Gfral−/− mice (Extended Data Fig. 7f,g); stress-induced hypophagia and nausea have both been attributed to glucagon-like peptide 1 (refs. 51,52,53). In support of the anxiogenic effects of GDF15, a single intraperitoneal injection of a pharmacological dose of GDF15 (ref. 26) produced anxiety-like behaviour in the open-field and light–dark box tests without affecting overall physical activity (Extended Data Fig. 7h,i). Critically, the anxiogenic effects of GDF15 persisted following repeated daily exposure, suggesting that tachyphylaxis did not occur (Extended Data Fig. 7j–l), similar to what we and others have shown for food intake and body weight loss with repeated GDF15 treatment26,54.

Several brain regions are involved in the acute stress response2,3,55,56. To determine the neurobiological substrates of GDF15-induced anxiogenesis, we measured Fos, an immediate early gene associated with neuronal activation, in several structures known to be involved in stress-related behaviours. The locus coeruleus is the main source of noradrenaline in the mammalian brain and is one of the first structures recruited following stressful stimuli3,4,57. Given that GDF15’s metabolic effects were recently shown to be dependent on β-adrenergic activity26, the locus coeruleus represented a particularly relevant target. However, high-dose GDF15 did not increase c-Fos expression in locus coeruleus noradrenergic neurons (Extended Data Fig. 8a) nor in the medial prefrontal cortex and basolateral amygdala, both downstream projections of the locus coeruleus known to drive anxiogenesis (Extended Data Fig. 8b)4,58. In contrast, GDF15 showed increased activity in the central amygdala and the bed nucleus of the stria terminalis (Fig. 4e), each of which has well-established roles in anxiogenesis59,60,61,62,63. Consistent with other studies, we also observed activation within the paraventricular nucleus of the hypothalamus (Fig. 4e)23,64.

High levels of circulating GDF15 can activate corticotropin-releasing hormone (CRH) neurons within the paraventricular nucleus of the hypothalamus and activate the HPA axis, leading to the secretion of corticosterone in mice23,24,64. We explored whether GDF15–GFRAL signalling is important for activating the HPA axis under more physiological conditions by applying physical restraint stress to WT and Gfral−/− mice. Expectedly, restraint stress increased circulating corticosterone in both genotypes, reflecting activation of the HPA axis, but significantly less so in Gfral−/− mice (Fig. 4f), despite circulating GDF15 levels being modestly elevated by restraint stress in both genotypes (Extended Data Fig. 8c). CRH is critical in driving stress-induced behavioural responses, such as anxiety and aversion3,50. Intriguingly, exogenous CRH led to a marked increase in circulating GDF15 levels (Fig. 4g), further demonstrating the involvement of GDF15 and GFRAL in the stress response. Importantly, exogenous CRH increased adrenocorticotropic hormone (ACTH) and corticosterone levels, whereas dexamethasone decreased corticosterone levels similarly in WT and Gfral−/− mice, suggesting that there was no impairment in HPA axis signalling (Extended Data Fig. 8d–f).

Together, these data support the role of GDF15 in regulating acute stress-induced anxiety-like behaviour (Fig. 4h), possibly through the recruitment of multiple anxiogenic brain regions56, and emphasize that GDF15–GFRAL signalling is transmitted broadly throughout the brain despite the brainstem-localized receptor expression.

Discussion

Mammals have evolved many mechanisms for detecting and avoiding noxious stimuli such as foodborne toxins, bacterial infections and allergens16,65,66,67,68. GDF15 has been linked to all of these by inducing nausea, emesis (or pica) and aversive behaviour19,64,69. However, in addition to physical stimuli, psychological stress produces its own set of physiological and behavioural responses intended to protect the organism. The fight-or-flight response is a classic example of psychological stress leading to metabolic responses that mobilize endogenous energy stores from adipose tissue13,70 and elicit behavioural responses such as anxiety5,6,7,8,9. For decades, there has been evidence that adipose tissue lipolysis and peripheral adrenergic activity are associated with feelings of anxiety10,11,12,13, but the link between these responses has remained elusive.

Here, we identified a mechanism whereby adrenergic activation of adipocyte lipolysis promotes the secretion of GDF15 from M2-like alternatively activated macrophages within adipose tissue, which, through its receptor, GFRAL, is critical for appropriate behavioural and physiological responses to acute psychological stress. These results are consistent with previous studies in which chronically supraphysiological levels of GDF15 promoted anxiety-like behaviour whereas lifelong deletion of GDF15 reduced anxiety21,22. We also identified various anxiogenic brain regions, including the central amygdala, bed nucleus of the stria terminalis and paraventricular nucleus, that could be involved in mediating these responses. GDF15, at least at high circulating levels, has been shown to reduce physical activity71,72, produce anxiety-like behaviour20,21,22 and activate the β-adrenergic25,26,68 and HPA axes23,24. Indeed, optogenetic activation of GFRAL-expressing neurons is sufficient to activate the HPA axis and increase the circulating cortisol levels24.

Clinical trials with GDF15 analogues did not report increases in anxiety73,74. However, while our paper was in review, a study was published in which plasma proteomics conducted in individuals from the UK Biobank demonstrated that GDF15 is the circulating factor with the strongest association with anxiety in humans compared with more than 3,000 other measured proteins75. Herein, we show that GDF15–GFRAL signalling is a critical component of responses to acute psychological stress, including anxiety-like behaviour and HPA axis activation.

Some important limitations to the current data should be considered. First, the M1- and M2-like macrophage distinction is an oversimplification of the highly heterogeneous macrophage populations. Other publications have identified distinct macrophage subpopulations that secrete GDF15, at least in skeletal muscle76, so it will be interesting to determine whether a particular subpopulation is responsible for the fatty acid-dependent secretion of GDF15 demonstrated here. Moreover, it will need to be determined how palmitate drives GDF15 secretion from macrophages and whether endogenous adipocyte-derived palmitate is sufficient to elicit this response. It is appealing to speculate that fatty acid-sensing PPARs, which have been shown to regulate GDF15 (ref. 47), could be involved, but this will need to be confirmed in vivo.

Despite the rapid increase in the rates of anxiety, there is a dearth of new targets for its treatment. The interactions between the brain and peripheral immune cells in the development of stress have been a growing area of interest77,78,79, but previous studies have focused largely on classic inflammatory cytokines. Here, we provide additional insights into these interactions by identifying a new pathway for immune–brain crosstalk mediated by adrenergic-induced ATGL-dependent lipolysis and M2-like macrophage-secreted GDF15. These results raise the possibility that blocking GDF15–GFRAL signalling could mitigate acute stress-induced anxiety.

Methods

Mice

Germline GFRAL-knockout mice (Gfral−/−) were generated as previously described, with breeding pairs provided by R. Seeley26. Mice lacking β1, β2 and β3 adrenergic receptors (BR−/−) were generated as previously described26,80. Adipocyte-specific ATGL-knockout animals were generated by crossing Adipoq-Cre mice with Atgl/Pnpla2-floxed mice81; Atgl/Pnpla2flox/flox littermates were used as controls. Animal studies were carried out at McMaster University (210104), the University of Toronto (21-467 and 24-0362H) or Washington University in St. Louis (20-0139). All animals used in the study were housed and cared for in accordance with the Guidelines for Animal Use at McMaster University and were approved by the McMaster University Animal Ethics Research Board or the ethics boards of their respective facilities. All mice were group-housed on a 12-h light–dark cycle with ad libitum access to food and water. The animals were housed either in a HEPA-filtered unit at room temperature or in specific pathogen-free microisolators at 29 °C and 40–60% relative humidity. Experiments were performed on mice at ages between 16 and 24 weeks. All mice were male unless stated otherwise, including for cell isolations.

Experiments were performed, and mice were killed in a fed state between 09:00 and 12:00. Terminal blood was collected by cardiac puncture, and blood from live animals (for example, time-course GDF15 levels) was collected from a tail vein. Blood samples were centrifuged at 8,000g for 10 min at 4 °C after clotting at room temperature for 30 min, and the supernatant was collected. Tissues were collected following anaesthetization with ketamine (75 mg kg−1) and xylazine (10 mg kg−1). Tissues were frozen at −80 °C until future analyses.

Male WT and Gfral−/− mice were placed on a high-fat, high-fructose diet (40 kcal% fat (mostly palm oil), 20 kcal% fructose with 0.02% cholesterol; Research Diets, D19101102) starting at 12–20 weeks of age for 4 weeks. Palm oil (Sigma-Aldrich, 70905) was gavaged at ~09:00 as described previously49.

Mouse drug treatments

Adrenaline hydrochloride (Sigma-Aldrich, E4642) was diluted in saline and injected IP (~09:00) at a dose of 0.5 mg per kg body weight based on previous studies82,83. CL (Sigma-Aldrich, C5976) was diluted in saline and injected IP (~09:00) at a dose of 1 mg per kg body weight84. Cilostamide (Sigma-Aldrich, C7971; 10 mg per kg body weight, injected IP) was prepared in 10% DMSO and 5% Kolliphor85. All control mice received saline IP injected at the same volume. Recombinant human GDF15 (5 nM kg−1; Novo Nordisk) or vehicle was injected IP at ~09:00 (ref. 26). To test for tachyphylaxis, we injected mice with either vehicle or GDF15 (5 nM kg−1) for 9 days at ~09:00. Chow diet was measured daily. On the 10th day, mice were injected with either vehicle or GDF15 (5 nM kg−1), and open-field tests were performed 1 h later (described below). LPS (Sigma-Aldrich, 0111:B4, L2630) was injected IP at a dose of 2 mg kg−1 starting at 09:00, and control mice were injected with saline. CRH (Sigma-Aldrich, C3042) was injected IP at ~09:00 at a dose of 90 μg kg−1, and blood was collected from a tail vein at 0.5 and 3 h after injection86. Dexamethasone (Sigma-Aldrich, D4902) was injected IP at ~09:00 at a dose of 100 μg kg−1, and blood was collected from a tail vein 6 h following injection87.

Human paediatric anxiety samples

Children with obesity who were enrolled in the Canadian Pediatric Weight Management Registry at the McMaster site were included in this study. Study participants with an available fasting blood sample at their initial visit and a clinical diagnosis of anxiety but without antipsychotic or antidepressant medications (n = 23) were compared with those without any diagnosis of anxiety and also free from the use of medications (n = 24). Control participants and those with anxiety were matched for age (12.50 ± 2.97 and 12.44 ± 2.86 years), sex (12/11 and 12/12 male-to-female ratio), body weight (87 ± 31.88 and 88 ± 35.32 kg) and body mass index (33.18 ± 6.41 and 34.08 ± 8.07 kg m−2). Serum GDF15 was assessed from previously frozen samples in duplicate, as described below.

Separation of adipose tissue fractions

Each gWAT or iWAT sample was combined from two mice. WAT was collected, quickly rinsed in warm PBS, minced and incubated in collagenase (10 mM HEPES–Krebs–Ringer buffer, 4% BSA, 1.5 mg ml−1 collagenase type I (Gibco, 17100017)) at 37 °C with gentle agitation for 30 min (gWAT) or 45 min (iWAT). Tissues were filtered (100 μm), and fractions were separated by repeated centrifugation at 500g for 5 min. F4/80+ cells were isolated using a commercially available kit as per the manufacturer’s instructions (EasySep Mouse F4/80 Positive Selection Kit; STEMCELL Technologies, cat. no. 100-0659) along with an EasySep magnet (cat. no. 18000).

Adipocyte differentiation

Immortalized mouse white adipocytes were generated as previously described88. Preadipocytes were grown to confluence in DMEM supplemented with 1% GlutaMAX, 1% penicillin–streptomycin, 10% FBS, insulin (20 nM) and T3 (1 nM; differentiation medium). Confluent cells were incubated for 48 h in differentiation medium further supplemented with isobutylmethylxanthine (0.5 mM), dexamethasone (0.5 μM) and indomethacin (0.125 mM; induction medium). All chemicals were from Sigma, and experiments were performed within 20 passages following immortalization.

Adipocytes were then treated with either CL (10 μM; Sigma-Aldrich, C5976), ATGListatin (40 μM; Sigma-Aldrich, SML1075), neither of the compounds (control) or a combination of the two compounds for 24 h in serum-free DMEM supplemented with 1% BSA (Sigma-Aldrich, cat. no. A8806). Medium was collected, and excess lipids and cellular debris were removed by centrifugation. Cells were washed with PBS and snap-frozen either in TRIzol for RNA extraction or in cell lysis buffer for protein content quantification using BCA. Lipolysis was assessed by measuring glycerol levels in the medium using a commercially available kit (Cayman Chemical, cat. no. 10010755).

BMDM isolation, differentiation and polarization

Femurs and tibias were collected from mice and cleaned of remaining muscle and soft tissues, and both ends of each bone were removed89. The bone marrow was extracted by centrifugation at 1,900g for 5 min. The marrow was resuspended in 1 ml DMEM (Wisent, cat. no. 319-005-CL, 25 mM glucose) and strained through a 40-μm cell strainer into a 50-ml Falcon tube. Cells were cultured for 4 h at 37 °C and 5% CO2. After incubation, nonadherent cells were collected and combined with M-CSF (20 ng ml−1; Peprotech, 315-02) to stimulate macrophage differentiation and plated for 7 days in DMEM (Wisent, cat. no. 319-005-CL) supplemented with FBS (10%, Wisent, cat. no. 098150) and penicillin–streptomycin (1%, Invitrogen). After differentiation, cells were washed, collected and plated onto 12-well plates (1 × 106 cells per ml) and left to adhere overnight.

Fully differentiated BMDMs were then polarized either with 100 ng ml−1 LPS (Sigma-Aldrich, L2630) + 20 ng ml−1 IFNγ (Peprotech, 315-05) to an M1-like phenotype or with 20 ng ml−1 IL-4 (Peprotech, 214-14) to an M2-like phenotype for 24 h (ref. 42). Following polarization, cells were treated with either sodium palmitate (500 μM, Sigma-Aldrich, P9767), sodium oleate (500 μM, Sigma-Aldrich, O7501), tunicamycin (5 ng ml−1, Cayman Chemical, 11089-65-9), rosiglitazone (1 μM; Sigma-Aldrich, R2408) or T0070907 (1 μM; Sigma-Aldrich, T8703) for another 24 h in DMEM (Wisent, 319-005-CL) supplemented with 1% fatty acid and endotoxin-free BSA (Sigma-Aldrich, cat. no. A8806). Medium was collected after 24 h and frozen at −80 °C until analysis. Cells were rinsed in PBS and collected either in TRIzol for RNA extraction (see below) or in cell lysis buffer for protein quantification using BCA.

RNA isolation, cDNA synthesis and qPCR

Tissues were homogenized in TRIzol reagent. An RNeasy kit (Qiagen, cat. no. 74106) was used for subsequent total RNA extraction and purification according to the manufacturer’s instructions. cDNA synthesis was performed using the SuperScript IV Reverse Transcriptase kit (Invitrogen, cat. no. 18090010) following the manufacturer’s instructions. cDNA expression for specific genes was detected by qPCR using the AmpliTaq Gold DNA Polymerase kit (Applied Biosystems, cat. no. N8080241). Relative mRNA levels were quantified with the ΔCT method using mouse Ppia (Mm02342430_g1) as an endogenous control, except for adipocyte and SVF comparisons for which Polr2a was used (Mm00839502_m1). The gene-specific primers used were as follows: Gdf15 (Mm00442228_m1), Atf4 (Mm00515324_m1), Chop/Ddit3 (Mm01135937_g1), iNos (Mm00440502_m1), Arg1 (Mm00475988_m1), Ppargc1a (Mm01208835_m1), Emr1/F480/Adgre1 (Mm00802529_m1), Fgf21 (Mm07297622_g1), Atgl/Pnpla2 (Mm00503040_m1) and Lep (Mm00434759_m1). All probes were purchased from Thermo Fisher.

RNA-seq and analyses

For RNA-seq, mice were injected with adrenaline (0.5 mg kg−1 IP at 09:00) and gWAT was collected after 1 h. RNA was extracted as described above. Raw RNA-seq FASTQ data were imported into Galaxy for quality control and processing steps. The FastQC tool was used to check read quality, and the Cutadapt tool was used to trim adaptor sequences and remove low-quality reads. The remaining reads were aligned to the mm10 Mus musculus reference genome with the HISAT2 tool and quantified with the featureCounts tool. These counts data were imported into R for differential expression analysis with the DESeq2 package to detect DEGs (adjusted P value (P-adj) < 0.05). Principal component analysis was performed in R using variance stabilizing transformation (VST)-normalized data from the DESeq2 analysis. Pathway analysis was performed with the GSEA software in combination with MSigDB gene sets. A list of human genes encoding secreted proteins (ligands) was obtained from the HUGO Gene Nomenclature Committee website. Mouse genes orthologous to these human ligand-encoding genes were obtained using the g:Profiler tool. A heat map was created with DEGs belonging to this group of ligand-encoding genes in R with the pheatmap package, using z-scored VST-normalized data. These genes were also highlighted in a volcano plot, which was created with the Enhanced Volcano package in R, using fold-change and P-adj values from the DESeq2 analysis.

Single-cell RNA-seq

The scRNA-seq data used for the analyses described in this article were obtained from the National Center for Biotechnology Information Sequence Read Archive under reference number SRP145475 (ref. 40). Cell Ranger was used to perform sample qualification, alignment, filtering, counting and aggregation using the Linux system or R and RStudio software. Clustering and gene expression were visualized with the 10X Genomics Loupe Browser (v.6.5.0).

2SMR using GWAS summary data

2SMR was performed using the exposure and outcome from two nonoverlapping and independent datasets to conduct the summary-level instrument–exposure analysis and the instrument–outcome association analysis. 2SMR was performed using the R package TwoSampleMR (v0.5.6)90. To verify the causal effect of GDF15 on nerves, anxiety, tension or depression in humans, we performed 2SMR using the exposure dataset (GDF15, GWAS ID: ebi-a-GCST90011998, sample size: 21,758)91,92 and outcome dataset (nerves, anxiety, tension or depression, GWAS ID: ebi-a-GCST90013910, sample size: 407,746)93. We identified genetic variants (single-nucleotide polymorphisms) associated with blood GDF15 protein levels in the GWAS catalogue dataset based on cis-protein quantitative trait loci (within 500 kb of the Gdf15 gene) and further selected proxy single-nucleotide polymorphisms by linkage disequilibrium clumping (P1 = 5 × 10−8, clump = TRUE, P2 = 1 × 10−7, r2 = 0.01, kb = 10,000). 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 Mendelian randomization analysis and visualized the results using the scatter plot and forest plot functions in the R package TwoSampleMR. 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.

Serum and medium analyses

Serum and cell-based GDF15 levels were measured by ELISA (R&D Systems, DY6385) as per the manufacturer’s instructions18,45. Serum corticosterone (Thermo Fisher, EIACORT), adrenocorticotropic hormone (Abcam, ab263880) and adrenaline (Abnova, KA1882) were measured following the manufacturer’s instructions.

c-Fos immunofluorescence and quantification

C57BL/6J mice were injected with GDF15 (5 nM kg−1) or vehicle; 90 min later, the mice were anaesthetized with a mixture of ketamine (69.57 mg ml−1), xylazine (4.35 mg ml−1) and acepromazine (0.87 mg ml−1) IP and perfused with 10 ml of 0.1 M phosphate buffer (PB) followed by 25 ml of 4% paraformaldehyde (PFA) in 0.1 M PB. The brains were removed, postfixed overnight (4% PFA in 0.1 M PB) and kept at 4 °C in 30% sucrose solution until cutting. Coronal sections (40 µm) were obtained using a microtome (Leica SM 2000R) and serially collected in PBS. Sections were washed in PBS (3 × 10 min) and preincubated in PBS containing Triton X-100 (0.3%) and normal goat serum (5%) for 1 h. Sections were then incubated overnight at room temperature in PBS containing Triton X-100 (0.3%), normal goat serum (1%) and rabbit anti-c-Fos (1:1,000, Cell Signaling Technology). Sections were then washed in PBS (3 × 10 min), incubated with Alexa Fluor 488 goat anti-rabbit secondary antibody (1:400, Invitrogen) in PBS containing Triton X-100 (0.3%) and normal goat serum (1%) for 2 h, and washed in PBS (3 × 10 min). Sections were finally serially mounted in Vectashield medium with DAPI (Vector Laboratories). Image acquisition was performed using Zeiss Axio Scan 7.

Western blotting

Proteins were extracted using lysis buffer containing 50 mM HEPES (pH 7.4), 150 mM NaCl, 100 mM NaF, 10 sodium pyrophosphate, 5 mM EDTA, 250 mM sucrose, 1 mM DTT, 1 mM sodium orthovanadate, 1% Triton X-100 and cOmplete Protease Inhibitor Cocktail (Roche). The total protein concentration in the samples was measured by the BCA protein assay. Protein concentrations were adjusted and diluted in 4× SDS sample buffer. Proteins were separated using SDS–PAGE gels and transferred to poly(vinylidene difluoride) membranes. After blocking for 1 h with 5% BSA in TBST at room temperature, membranes were incubated with the primary antibodies overnight at 4 °C (phosphorylated PKA substrates (1:1,000, Cell Signalling #9624) or tubulin (1:1,000, Abcam #ab4074)). After washing, membranes were incubated with a secondary antibody at room temperature for 1 h. Protein bands were visualized with the Fusion FX7 system (MBI) and quantified using ImageJ software.

Tube restraint and stress tests

Mice were put into 50-ml Falcon tubes (beginning at ~09:00) with holes at both ends so the tail was exposed and there was sufficient air flow3. Control mice remained in their home cage with food and water removed at the initiation of tube restraint. Immediately following tube restraint, mice were either anaesthetized for tissue collection or submitted to open-field test protocols (see below).

Alternatively, social isolation was induced by individually housing previously group-housed mice for 3 days and comparing them with their continually group-housed littermates31. In the cage-switch model, mice were removed from their home cage and placed in an identical cage with the dirty bedding of nonlittermate males31. Nondisturbed socially housed littermates were applied as the controls.

To assess stress-induced food intake and nausea, we provided mice with kaolin clay (Research Diets, K50001) in addition to their regular diet for 3 days before testing. Mice were restrained (starting at ~09:00) for 4 h and returned to their cages (group-housed) with access to standard chow and kaolin clay pellets, which were weighed before and after 18 h (~09:00 the following day).

Open-field and light–dark box tests

Mice were moved to an isolated room the evening before open-field testing to minimize stress on the day of testing. Immediately following 4-h tube restraint or 1 h following exogenous GDF15 injection (5 nM kg−1 IP) or 1 h following exogenous adrenaline treatment (0.5 mg kg−1 IP), mice were placed in the centre of a box (40 × 40 × 40 cm, black walls and white plastic bottom). Mice remained in the box in the same isolated room, and videos were recorded using a camera (GoPro, 1080p resolution and a sample rate of 3 frames per second) mounted to the ceiling. Videos were recorded for 20 min. The two-dimensional mouse pose was analysed using DeepLabCut94 to extract the locations of arena boundaries and anatomical landmarks (nose, left ear, right ear, tail base). The DeepLabCut model (ResNet50 architecture) was trained using 20 distinct frames (k-means clustering) from 20 input videos for a total of 200,000 epochs, resulting in a mean testing error of 3.74 pixels across all tracked positional landmarks. Weighted spatial means of all four anatomical points taken to represent the location of the centre of the head were then analysed using custom MATLAB (v2022b) scripts for the time spent in the centre (>8 cm from the arena boundaries), time spent in the periphery (<8 cm from one arena boundary), time spent in the corners (<8 cm from two arena boundaries) and total distance moved during the test. Afterwards, the mice were returned to their home cages, and the boxes were thoroughly cleaned and left to dry before subsequent rounds. The boxes in which groups (that is, WT/GFRAL, control/restraint, control/GDF15, etc.) were placed were systematically rotated to control for any possible differences in box positioning during recording. Alternatively, mice were placed into an automated open-field test system (Opto-Varimex-5 Auto-Track, Columbus Instruments) for 20 min, and the time spent in the centre and periphery and the total movement were analysed using equipment software.

For light–dark box testing, mice were placed into the ‘light’ area and videos were recorded using a camera (GoPro, 1080p resolution and a sample rate of 3 frames per second) for 10 min. The videos were analysed manually by a blinded assessor for the time spent by the mice in the light area.

Behavioural and metabolic activity

Metabolic and behavioural monitoring was conducted using the Promethion system (Sable Systems International). Data were collected following acclimation to the system for 24 h. After acclimation, mice were injected with adrenaline (0.5 mg kg−1 IP) at 09:00, after which the mice remained in the system for another 24 h. Data on food intake, physical activity and nonambulatory movements (beam breaks), oxygen consumption (VO2), carbon dioxide production (VCO2), respiratory exchange ratio (VCO2/VO2) and energy expenditure (kcal h−1) were collected at 15-min intervals.

Statistics

Statistical analyses were performed using GraphPad Prism (version 10) or R software (RNA-seq). Data were analysed using an unpaired t test, a one-way analysis of variance (ANOVA) or a two-way ANOVA. When significant interactions were detected through ANOVA, subsequent post hoc analyses were reported with appropriate corrections; where no post hoc correction is stated, there was no interaction detected. Correlational analyses were performed with Pearson’s correlation. No statistical methods were used to predetermine sample sizes, but our sample sizes are similar to those reported in previous publications18,26. Normality was tested using the Shapiro–Wilk test. Outliers were defined as points outside 2 s.d. away from the group mean. All data are presented as mean ± s.e.m. with individual data points. Differences were considered significant when P < 0.05.

Reporting summary

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