GDF15 Provides an Endocrine Signal of Nutritional Stress in Mice and Humans

1GDF15 (growth differentiation factor 15; also known as macrophage inhibitory cytokine-1 [MIC-1], NAG1, PLAB, and PDF) is a stress-induced cytokine and an atypical member of the transforming growth factor beta superfamily. Bootcov et al. originally characterized it as a dimeric protein secreted by activated macrophages. In healthy animals, it is predominantly expressed in the liver, lung, and kidney and, at least in humans, in large amounts in the placenta. It circulates at high levels in humans and serum levels are known to increase with age, smoking, intense exercise, cancer, and a range of other disease states (reviewed in,, ). It appears that almost any cell or tissue can express GDF15 in response to various forms of stress. The measurement of circulating concentrations of GDF15 is beginning to enter clinical practice as a diagnostic biomarker in mitochondrial disease and as a prognostic marker in conditions such as heart failure and certain cancers.

2Johnen et al. first reported that mice bearing tumors engineered to overexpress GDF15 lost weight dramatically. This could also be reproduced by injection of recombinant GDF15 and prevented by a neutralizing GDF15 antibody. Transgenic GDF15-expressing mice similarly lost weight secondary to reduced food intake. Conversely, GDF15 null mice were reported to be slightly heavier (6%–10%) than their wild-type littermates on a chow diet (CD), a difference that becomes more striking on a high-fat diet (HFD). GDF15 injection induced cFos activation in selected regions of the brainstem, particularly the nucleus tractus solitarius (NTS) and area prostrema (AP). Selective lesioning of these hindbrain regions rendered mice unresponsive to the anorexigenic actions of GDF15. Recently, it has been demonstrated that these effects of GDF15 are mediated via a receptor composed of a heterodimer of Ret and a member of the GDNF receptor alpha (GFRa) family, known as GFRa-like or GFRAL. Notably, GFRAL expression appears to be strictly confined to the AP and NTS. In addition to reporting the structure of GFRAL, these papers also showed that GFRAL knockout (KO) mice are resistant to the anorectic effects of exogenous injected GDF15 and to endogenously secreted GDF15 levels induced by cisplatin chemotherapy, clearly establishing the GDF15-GFRAL axis as critical to stress pathway-induced weight loss. Interestingly, two groups also noted that whereas body weight is similar to that of wild-type littermates in chow-fed GFRAL null mice, GFRAL null animals gain more weight on an HFD, whereas the other groups reported similar body weights in GFRAL null mice on an HFD. It is unknown whether circulating levels of GDF15 rise in response to sustained overfeeding and, if this occurs, what tisues are responsible.

3Here, we explore the relationship between GDF15 production and nutritional state and find that, in contrast to enteroendocrine hormones or leptin, GDF15 levels are not influenced by meals, by the imposition of periods of caloric deficit or caloric excess of moderate intensity and duration in mice or humans. However, GDF15 levels do increase significantly when mice are exposed to chronic high-fat feeding. We then characterize in detail the elements of the cellular integrated stress response (ISR) that are involved in the regulation of GDF15 expression and demonstrate activation of the ISR in selected tissues of high-fat-fed mice. We also show that another severe nutritional perturbation, namely provision of a lysine-deficient diet to mice, activates the ISR and increases GDF15 levels. Finally, we provide the first evidence that GDF15 generates an aversive signal through the demonstration of conditioned taste aversion (CTA) in mice.

4It is well established in humans that hormones derived from enteroendocrine cells respond to acute changes in nutritional state and play a key role in regulating energy homeostasis. To determine if GDF15 responds in a similar way, we studied the response of humans to established stimuli of the enteroendocrine system.

5In Human Study 1, fasting (overnight) healthy volunteers received a mixed macronutrient liquid drink (Figures 1A–1D) or 50 g anhydrous glucose (Figures S1A–S1D) followed by 30-min blood sampling for 180 min. In both studies, glucose peaked at 30 min (6.60 ± 0.26 mmol/L and 8.71 ± 0.16 mmol/L, respectively), with a lower blood glucose in the mixed macronutrient load reflecting the lower sugar content (22 g) in the test drink. In parallel with the glucose peak, early increases in both insulin and GLP-1 levels were observed, whereas circulating GDF15 levels briefly (at 60 min time point) fell after the mixed meal and were unchanged following glucose ingestion, as reported previously.

6A fall in the adipose-derived hormone leptin represents an important peripheral signal of nutritional deprivation, serving to induce hyperphagia and suppress selected neuroendocrine hormone axes. To address the question of whether GDF15 mirrored the behavior of leptin to changes in nutritional state, we evaluated the response of circulating GDF15 to fasting and caloric restriction in mice and humans.

7First, we examined hormone responses to a 24 h fast in mice (Mouse Study 1). Despite a 17.8% loss in body weight and a marked fall in leptin levels, circulating levels of GDF15 were unchanged (Figures 1E–1G).

8Next, circulating GDF15 levels were measured in three independent studies in humans subjected to caloric restriction of varying intensity and duration. In Human Study 2, GDF15 concentrations increased from 319.4 ± 21.27 pg/mL to 406.8 ± 31.24 pg/mL in lean healthy volunteers calorie restricted for 2 days (10% of estimated daily energy requirements) (Figure S1E).

9In Human Study 3, a cohort of obese participants consumed a low-calorie meal replacement diet (∼1,000 kcal/day) for 28 days. This resulted in a significant reduction in body weight (−5.55 ± 2.05 kg from baseline, p < 0.0001; Figure 1H) and leptin levels (Figure 1I), whereas there was a small, statistically significant increase in GDF15 levels (Figure 1J).

10In Human Study 4, a group of lean healthy participants underwent 7 days of total calorie deprivation. As expected, circulating leptin levels fell precipitously from 6.03 ± 1.18 ng/mL to 2.24 ± 0.49 ng/mL at 24 h (Figure 1K) and a marked ketogenic response (β-hydroxubutyrate) was observed in response to the fast (Figure 1L). Meanwhile, circulating GDF15 levels increased, peaking at 48 h of caloric restriction (from 371.4 ± 94.2 pg/mL at baseline to 670.2 ± 349.2 pg/mL, p = 0.003). Interestingly, despite continuation of the calorie deprivation, GDF15 levels gradually returned toward baseline levels, although they remained higher than starting values (441.2 ± 151.3 pg/mL) (Figure 1M).

11From these studies in both mice and humans, it is clear that GDF15 does not exhibit a leptin-like response to caloric restriction. Rather, a small increase in GDF15 levels is observed under conditions of severe nutritional deprivation.

12In contrast to caloric deprivation, states of nutritional excess and weight gain are associated with physiological increases in leptin. We studied the effect of short-term hyper-caloric interventions on non-obese volunteers to determine if they had a similar effect on GDF15.

13In Human Study 5, we assessed changes in GDF15 in healthy volunteers in response to short-term high-fat overfeeding. After 7 days of the intervention, a significant increase in body weight was observed (1.64 ± 1.07 kg, p < 0.0001) compared to baseline (Figure S1F), accompanied by a significant increase in leptin levels (Figure S1G). Despite the observed increases in fasting insulin and glucose levels (Figures S1H and S1I), there was no significant change in median (IQR) GDF15 levels (302.0 [256.0–318.0] pg/mL [baseline] versus 295.0 [258.0–343.5] pg/mL [after overfeeding]) (Figure S1J).

14Human Study 6 examined the effect of an 8-week overfeeding intervention (additional 40% of weight maintenance energy requirements) on healthy participants in an inpatient setting. This produced a significant increase in weight (5.52 ± 2.05 kg, p < 0.0001) and leptin levels (Figures S1K and S1L), but no rise in circulating GDF15. Indeed, at the end of the intervention, GDF15 levels were actually lower than at entry to the study (Figure S1M).

15Consistent with these data, GDF15 levels were unchanged in mice fed an HFD for up to 7 days (Mouse Study 2), despite the mice manifesting the anticipated increases in fat and liver weight, and rises in plasma insulin and leptin (Figure S2).

16Taken together, these data suggest that, unlike established hormonal regulators of energy homeostasis, “modest” overfeeding in humans and mice does not trigger GDF15 production.

17As GDF15 is a stress-responsive hormone, we wondered if more prolonged or severe nutritional stressors might be needed to induce its expression and secretion. To test this hypothesis, we undertook a prospective longitudinal study in mice fed either a CD or HFD from 9 weeks of age (Mouse Study 3). This resulted in progressive weight gain and fat mass in association with rising insulin and leptin levels (Figures 2A–2D). Glucose levels were initially similar in the two groups but rose significantly in the HFD-fed mice from week 4 onward (Figure 2E). GDF15 levels started to rise at the 4 week time point and were significantly higher in the HFD-fed mice from week 8 onward (Figure 2F).

18In order to clarify the source of GDF15 in this context, we analyzed GDF15 mRNA expression in a range of tissues. GDF15 expression increased in liver, white epididymal adipose tissue (WAT), and even more strikingly in brown adipose tissue (BAT), but not in subcutaneous inguinal fat, kidney, and skeletal muscle (expression was very low in the latter) (Figure 2G).

19Next we sought to establish why GDF15 expression was induced in these tissues in this context. Prior work had suggested that ATF4 and CHOP, key transcriptional regulators of the ISR, might be involved. The ISR is a cell-autonomous integrator of diverse cellular stresses, so we began by documenting changes in GDF15 mRNA in mouse embryonic fibroblasts (MEFs) treated with a range of well-characterized stressors (Figure 3). These included cobalt chloride, a chemical mimic of hypoxia, which acts by stabilizing hypoxia inducible factor 1a (HIF-1a); unfolded protein response (UPR) inducers thapsigargin, an inhibitor of sarcoplasmic/endoplasmic reticulum Ca2+-ATPase (SERCA) and tunicamycin, an inhibitor of N-linked glycosylation, both of which perturb protein folding; and histidinol, an inhibitor of histidyl tRNA synthetase, which mimics amino acid deprivation. All these agents caused a significant and robust induction of GDF15 mRNA expression, albeit to varying extents, with thapsigargin being the most potent (Figure 3A). To confirm that GDF15 can be similarly upregulated in other cell types, we documented stress-induced changes in GDF15 expression in a range of human cell lines, as well as in 3T3-L1 preadipocytes (Figures 3B and S3A–S3C). In each case, GDF15 expression was increased with the level of induction ranging from 2- to 30-fold.

20All of the stress-inducing agents used here can trigger phosphorylation of EIF2α by one of four known EIF2α kinases, so we went on to confirm that this did occur at the concentrations and over the time frames used in our study (Figure 3C). Indeed, in addition to the stress-induced phosphorylation of EIF2α at Ser51, the protein expression of both the downstream targets, ATF4 and CHOP, was increased, albeit to a different extent, with tunicamycin and thapsigargin being the most potent inducers.

21Next, we used a combination of inhibitors and KO or knockdown MEF lines to test the roles of various elements of the ISR pathway in the regulation of GDF15. First, using the PERK inhibitor GSK2606414 (abbreviated as GSK in Figure 1) or the EIF2α inhibitor ISRIB (abbreviated as ISR in Figure 1), we demonstrated that the tunicamycin-mediated induction of GDF15 was significantly reduced (Figure 3D), and that this correlated with a reduction in the activation of the ISR pathway, as judged by decreased ATF4 and CHOP expression (Figure S3D). We also found that the tunicamycin-induced expression of GDF15 was abolished in MEFs harboring a mutation at the key phosphorylation site on EIF2α (Ser51-Ala) required for ISR activation (Figure 3E). Similarly, moving downstream in the pathway, tunicamycin-induced GDF15 expression was abolished in ATF4 KO MEFs (Figures 3F and S3E). Furthermore, small interfering RNA (siRNA)-mediated knockdown of CHOP significantly reduced both basal and tunicamycin-induced GDF15 expression (Figures 3G and S3F). Coupled with the knowledge that circulating GDF15 acts via GFRAL in the hindbrain, this establishes GDF15 as a bona fide systemic endocrine signal of the ISR.

22In order to establish if the ISR is indeed responsible for the observed induction of GDF15 in long-term HF-fed mice, we next evaluated expression of ATF4 and CHOP mRNA in the same panel of tissues assessed for GDF15 expression (Mouse Study 3). These data confirmed that ATF4 and CHOP mRNA were increased in the liver and BAT (ATF4 only), but interestingly not in WAT (Figures S3G and S3H), suggesting that the induction of GDF15 in WAT may involve other signaling pathways. HF feeding in mice leads to adipocyte cell death, particularly in epididymal fat, and this has been shown to activate macrophages. Furthermore, GDF15 was originally identified in activated macrophages, so we proceeded to check mRNA expression of a macrophage marker, F4/80, in the WAT and BAT samples. F4/80 mRNA increased in parallel with the changes in GDF15 mRNA (Figure S3I), so we sought to establish if GDF15 mRNA was being induced in macrophages, or other stromovascular fraction (SVF) cells, or in adipocytes themselves. The data suggest that GDF15 mRNA is induced in both fractions (Figure S3J). However, lipid-laden macrophages may “contaminate” the adipocyte fraction as the separation is based on flotation, so we went on to check for this by analyzing mRNA expression of Plin1 (an adipocyte marker) and F4/80 in each of the fractions. These data suggest that macrophages are present in the adipocyte fraction, so it remains possible that the apparent increase in GDF15 mRNA is largely coming from macrophages, though we cannot formally exclude a contribution from adipocytes themselves.

23From these data, it is clear that in mice, GDF15 expression is responsive to chronic conditions of overnutrition that manifest with changes in GDF15 within adipose tissue (white and brown) and the liver. These findings are consistent with reported increases in GDF15 levels in ob/ob mice and in obese humans, though the latter will require further careful analysis as reported that in non-obese monozygotic twin pairs (n = 72 pairs), the twin with the higher GDF15 concentration had a lower BMI.

24Having demonstrated that nutritional overload can induce GDF15 expression, we wondered if other nutritional stresses might have similar consequences. Previous studies have shown that diets deficient in essential amino acids can influence food intake and increase FGF21 levels in an ATF4-dependent manner, so we wondered if such diets might have a comparable impact on GDF15 levels. This was of particular relevance as we had shown (Figure 3A) that pharmacological mimics of amino acid imbalance that activate the ISR increase GDF15 expression in cells. To test this hypothesis, mice were fasted overnight and then fed a lysine-deficient diet for 4 h (Mouse Study 4). This led to a marked increase in circulating GDF15 levels compared to chow-fed animals (Figure 4A). In keeping with activation of the ISR, ATF4, CHOP, and GDF15 mRNA were all significantly increased in the livers of these mice (Figures 4B–4D).

25Reduced food intake has been shown to mediate most of the effects of GDF15 administration or overexpression on body weight. Activation of the GDF15 receptor (GFRAL) has also been linked to subsequent cFos activation in the parabrachial nucleus (PBN), which has in turn been linked to appetite suppression in response to meal-related peptides, as well as ingestion of toxins, mimicked by lithium chloride and lipopolysaccharide. Thus, we hypothesized that GDF15 administration might result in CTA, which occurs when an animal associates the taste of a normally favored food with symptoms caused by a concomitantly administered toxic or aversive substance.

26In Mouse Study 5, we first assessed the ability of GDF15 to lower food intake in a concentration-dependent manner. A single subcutaneous injection of GDF15 in mice acutely increased plasma GDF15 concentration, with maximum exposures occurring 1 and 4 h post-treatment (Figure 4E). GDF15 treatment resulted in a corresponding dose-dependent reduction of food intake that reached statistical significance at the highest dose (0.1 mg/kg) (Figures 4F and 4G). We then addressed whether GDF15 induced CTA behavior using the two bottle saccharin preference test. Similar to the positive CTA control, lithium chloride, GDF15 treatment at 0.01 mg/kg and 0.1 mg/kg also reduced saccharin consumption and increased water consumption compared to vehicle control (Figure 4H).

27These data demonstrate that acute administration of GDF15 can elicit an aversive response in rodents.

28Although its physiological function in humans is not clearly established, FGF21 is another putative systemic signal induced by the ISR. In humans, plasma FGF21 levels did not change significantly following a mixed meal or oral glucose challenge, a week of total calorie restriction, or high-fat overfeeding in healthy volunteers (Figures S4A–S4D). In mice, short-term overfeeding induced a small increase in FGF21 levels, whereas prolonged HFD exposure was associated with a robust increase in FGF21 levels (Figures S4E and S4F). Interestingly, although FGF21 mRNA increased in similar tissues as GDF15, within WAT, FGF21 seemed to originate from adipocytes themselves rather than macrophages (Figures S4G and S4H). Lysine-deficient diet exposure in mice also elicited a significant increase in hepatic FGF21 mRNA (Figure S4I). In cells, FGF21 responses to activation of the ISR largely mirrored the GDF15 responses with one notable exception: exaggerated induction rather than amelioration of the FGF21 response to TN in CHOP knockdown cells (Figures S4J–S4N). These results suggest that while the ISR pathway regulates both hormones, the molecular mechanisms downstream of ATF4 are distinct. These data were corroborated by the finding that circulating FGF21 and hepatic FGF21 mRNA levels were significantly increased in mice following a 24 h fast, despite the lack of a significant change in ATF4 or CHOP mRNA expression (Figures S4O–S4R). Fasting-induced FGF21 expression is known to involve PPARα.

29Elevating GDF15 levels by transgenic overexpression or pharmacological administration in mice and nonhuman primates leads to a marked fall in body weight. The principal aim of our work was to understand if and how GDF15 might be involved in physiological settings of under- and over-nutrition. To this end, we combined cellular studies with in vivo work in mice and humans to establish that GDF15 expression is highly responsive to activation of the ISR in a range of cell types and that its induction in this setting is dependent upon ATF4 and CHOP. The idea that cellular stress might be translated into a systemic response initially emerged from work in C. elegans where an induction of the mitochondrial unfolded protein response (UPRmt) in neurons led to changes in mitochondria within physically distinct, non-innervated tissues, but has more recently been supported by evidence linking FGF21 to the ISR. also recently proposed that GDF15 could act as a “mitohormetic” signal of mitochondrial dysfunction. Our analysis is largely consistent with these data and provides compelling evidence of the induction of GDF15 in response to activation of the ISR.

30As GDF15 administration causes weight loss and mice lacking GDF15 are prone to gain weight on an HFD, we determined whether GDF15 shares any features in common with known hormonal regulators of post-prandial satiety (e.g., enteroendocrine hormones such as GLP-1) or longer term hormonal regulators of nutrient stores (e.g., leptin). In contrast to GLP-1, and consistent with previous reports, GDF15 did not respond acutely to a meal or a glucose load in humans. In mice fasted for 24 h, there was no change in circulating GDF15, whereas the predicted fall in leptin levels and rise in FGF21 levels was seen. In humans, 48 h of severe caloric restriction in lean healthy volunteers resulted in a significant but small increase in GDF15 concentrations. In healthy volunteers undergoing a 7 day total fast, GDF15 levels peaked at around ∼180% of baseline by day 3 and then plateaued at around 118% at day 7. This early rise in GDF15 is in the opposite direction expected of a physiological regulator of energy balance and is more consistent with GDF15 being a marker of cell/tissue stress. The mechanisms whereby GDF15 levels start to return toward baseline with more prolonged fasting are unknown, but presumably reflect some sort of adaptation to the starved state.

31In two separate studies, overfeeding of healthy humans with an ∼48% excess of ingested calories for 1 week, or 40% for 8 weeks, did not increase GDF15 concentrations. Of note, in the longer study, conducted in an inpatient setting, GDF15 levels showed a small but significant fall (Figure S1M). Among possible explanations for this fall is the fact that in this inpatient study, smoking was not permitted. GDF15 levels are known to be positively associated with smoking status and it is possible that some participants quit smoking just prior to the study.

32In contrast to the studies summarized above, we found that circulating GDF15 levels rose in long-term HF feeding studies in mice. Whether or not this is also true in humans will require further studies. As recently summarized by Tsai et al., the relationship between circulating GDF15 and obesity in humans is complex. GDF15 levels rise with age and are also induced by conditions commonly associated with obesity such as diabetes and cardiovascular disease. So while positive correlations between GDF15 and measures of adiposity have been reported in several small studies, GDF15 was shown to be inversely correlated with BMI in non-obese monozygotic twin pairs. It is plausible that an inherent genetically determined increase in GDF15 levels or one induced by another cell stressor/disease might result in weight loss, and thus confound straightforward correlations between BMI and GDF15 levels.

33Ravussin et al. have drawn attention to the likely existence of leptin-independent signals of the obese state that might serve to restrain the indefinite progression of a state of positive energy balance and ever increasing obesity. The fact that mice lacking GDF15 become more obese on an HFD than wild-type mice suggests that GDF15 might at least contribute to that signal. Our studies show that while short-term overfeeding does not increase GDF15, more sustained states of caloric excess will raise circulating GDF15 levels.

34What is the source of the elevated levels of the GDF15 seen in the overnourished state? In mice, GDF15 mRNA is increased in liver and BAT, and in certain WAT depots, such as epididymal, by 18 weeks of HFD, but not in kidney or muscle. In adipose tissue it appears that most of the GDF15 mRNA is likely to be coming from infiltrating macrophages, whereas our initial analyses suggest that FGF21 mRNA is more robustly induced in adipocytes themselves. Macrophages are frequently cited as mediating many of the adverse metabolic consequences of overnutrition such as insulin resistance, through their production of a range of cytokine-like molecules. In the case of GDF15, it appears that these cells produce a circulating product that in this particular context might be beneficial to the organism.

35We also show GDF15 is significantly induced by another nutritional stress, namely a lysine-deficient diet. Interestingly, involvement of the ISR in response to perturbations affecting amino acid provision is conserved as far back as yeast, where the response is clearly cell autonomous. In metazoans, this response expanded to encompass rectifying responses to other cell-autonomous perturbations such as the UPR, hypoxia, viral infections, and iron deficiency. The link to the GDF15-GFRAL axis suggests that the ISR may now also have gained an endocrine component potentially involving an aversive response instructing the animal to change its foraging pattern. In some settings this is likely to be advantageous to the animal, though there are clearly exceptions where it is not, such as in cancer cachexia, where GDF15 levels can be as high as 40,000 pg/mL.

36What, then, are the consequences of induction of this hormone? GDF15 has been shown to reduce food intake in various species and to alter food choice. While it has been speculated that GDF15 may produce an aversive response, this had not previously been formally demonstrated. Using a CTA paradigm, we show that mice exposed to GDF15 in association with a saccharin taste will avoid saccharin-containing drinking water in future exposures. CTA is classically elicited by injection of lithium chloride and is also typical of agents that are known to produce nausea in humans. In this context, GDF15 contrasts with leptin, which reduces food intake but does not produce CTA. The lateral PBN, medial thalamus, and basolateral nucleus of the amygdala are essential for both acquisition and retention of CTA, with CGRP-expressing neurons of the PBN playing an essential role. Ascending pathways from the AP and NTS make substantial connections to the lateral PBN. It is therefore likely, though yet to be formally established, that GFRAL-expressing neurons of the caudal hindbrain project to the PBN.

37In summary, GDF15 appears to be an endocrine signal that can be produced by almost any cell type in response to activation of the ISR and presumably other signals. Our data suggest that “nutritional stress,” whether induced by sustained overnutrition as exemplified by prolonged HF feeding in mice or by provision of an amino acid imbalanced diet, leads to increased circulating GDF15 levels. We suggest that this might send an aversive endocrine signal to the brain, though we acknowledge that further work is needed to verify this hypothesis. If GDF15 is playing a role in restraining progressive weight gain, this suggests that it might have a role in the therapy of obesity. While its production of CTA might seem to suggest that it would be poorly tolerated in humans, it is premature to conclude that this should prevent its exploration as a possible obesity therapeutic. GLP-1-based drugs are licensed for the treatment of obesity, yet they too produce a CTA response in rodents and activate the lateral PBN. With careful titration, most humans can tolerate therapeutic doses of GLP-1 receptor agonists, though nausea and vomiting do lead to its cessation in a significant number. Thus, these data support the burgeoning interest in GFRAL, the GDF15 receptor, as an attractive therapeutic target in so far as it has a highly tissue-specific expression with actions of GDF15 at sites other than the caudal brain stem seemingly unlikely. Studies of the effects of GDF15 in humans are eagerly anticipated.

38Limitations of our studies include the fact that while we have shown that HF and lysine-deficient diets can induce GDF15, and that when administered at pharmacological doses, GDF15 can induce an aversive response, we have yet to formally demonstrate that this also occurs when GDF15 is induced endogenously. One might also argue that the small increases in GDF15 we detected in response to extreme fasting in humans are counter-intuitive; however, we speculate that in these circumstances, in addition to causing food aversion, GDF15 might also cause malaise, which might encourage an animal to rest and conserve energy.

39Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, David Savage (dbs23@medschl.cam.ac.uk).

40The liquid meal test included 12 healthy adult volunteers (3 male/9 female) with a mean ± SD age and BMI of 28 ± 9 years and 23.14 ± 2.74 kg/m2 respectively. Six healthy adult volunteers (4 male/2 female) were recruited to participate in the oral glucose tolerance test study with mean ± SD age of 30 ± 8 years and BMI of 25.05 ± 3.73 kg/m2. Both studies were completed at the NIHR Wellcome Trust Clinical Research Facility and the Wellcome-MRC Translational Research Facility, Cambridge Biomedical Campus, UK. Ethical approval was obtained from the East of England Cambridge South research ethics committee Ref: 16/EE/0338 and the East of England Cambridgeshire and Hertfordshire research ethics committee Ref: 013/EE/0195. Participants provided written consent prior to participation in the study.

4114 healthy male volunteers were recruited as participants in the study. The mean ± SD age and BMI of participants was 23.53 ± 2.70 years and 22.08 ± 2.0 kg/m2 respectively. Ethical approval was obtained from the Cambridge local research ethics committee (Ref: 13EE0107). All participants provided written informed consent before taking part in the study which was completed at the NIHR Wellcome Trust Clinical Research Facility, Cambridge Biomedical Campus, UK.

4250 overweight (BMI > 27 and < 30 kg/m2) or obese (BMI 30-40 kg/m2) participants were enrolled in this study. 33 (3 male and 30 female) were included in this analysis after accounting for withdrawals and insufficient biological samples. Mean ± SD age and BMI were 38.8 ± 8.8 years and 35.1 ± 3.1 kg/m2 respectively. The study protocol was approved by the Florida Hospital Institutional Review Board and was carried out in accordance with the Declaration of Helsinki. Clinical trial number: NCT01616082 (https://clinicaltrials.gov). Before taking part in the study, all participants were evaluated for eligibility. All participants provided their written consent to take part in the study.

43In total 13 healthy adult participants (7 male/6 female) were recruited to the study. The inclusion criteria for participation were age 18-45 years, percent body fat from DEXA > 12% for males and > 15% for females. Mean ± SD age of participants was 29.7 ± 6.1 years and BMI was 25.04 ± 3.32 kg/m2. The study protocol was initially reviewed by the Regional Ethics Committee of Norway (2017/1052; REK sør-øst B) with the decision that the research project was outside the Act on Medical Health Research, confirmed in a letter of exemption (2017/1052b). The study was then approved by the Ethics Committee at the Norwegian School of Sport Sciences (15-220817). The study was undertaken at the Norwegian School of Sports Sciences, Oslo, Norway. Written consent was obtained from volunteers prior to participation in the study.

44A total of 28 adult participants (25 males/3 females) were included in the study. Mean ± SD age and BMI of the study cohort were 22.6 ± 3.7 years and 24.2 ± 2.5 kg/m2 respectively. Subjects were physically active (taking part in at least 3 × 30 min of moderate-intensity physical activity each week), non-smokers, with no diagnosis of cardiovascular or metabolic disease, not taking any medication, and body mass was stable for at least 3 months. Both studies were approved by the Loughborough University Ethical Subcommittee for Human Participants (R13-P171 and R16-P132). All participants gave written consent to take part after the experimental procedures.

4520 healthy adult volunteers (11 male /9 female) with a mean ± SD age and BMI of 24.3 ± 4.3 years and 25.2 ± 3.0 kg/m2 respectively who led a sedentary lifestyle (less than 2 h of moderate to vigorous exercise per week) were recruited to an 8 week overfeeding study. Written consent approved by the Pennington Biomedical Research Center Institutional Review Board was provided by all participants. This trial was registered at clinicaltrials.gov (number NCT00565149). The study was undertaken at the Pennington Biomedical Research Center (Baton Rouge, Louisiana, USA).

46Ten C57BL/6J mice were purchased from Charles River (Charles River Ltd, Manston Rd, Margate, Kent, CT9 4LT) at 7-8 weeks of age. Mice were maintained in open vented cages with group housing (2 or 3 per cage) in a 12 h light/12 h dark cycle (lights on 07:00–19:00), temperature-controlled (22°C) facility, with ad libitum access to food and water. This research was regulated under the Animals (Scientific Procedures) Act 1986 Amendment Regulations 2012 following ethical review by the University of Cambridge Animal Welfare and Ethical Review Body (AWERB).

47C57BL/6J mice were purchased from Charles River (Charles River Ltd, Manston Rd, Margate, Kent, CT9 4LT) or bred in-house for some long-term HFD experiments. Mice were maintained in ventilated cages with group housing (2 or 3 per cage) in a 12 h light/12 h dark cycle (lights on 06:00–18:00), temperature-controlled (20-24°C) facility, with ad libitum access to food and water. All mice were fed either ad libitum or as stated otherwise prior to harvesting tissue and serum analysis. This research was regulated under the Animals (Scientific Procedures) Act 1986 Amendment Regulations 2012 following ethical review by the University of Cambridge Animal Welfare and Ethical Review Body (AWERB).

48Mice were originally purchased at Janvier Labs (Route du Genest, 53940 Le Genest-Saint-Isle, France) and bred in-house. All mice were maintained in standard housing conditions (22°C) on a 12 h light-dark cycle (lights on 08:00-20:00). Animal experiments were carried out in accordance with INRA guidelines in compliance with European animal welfare regulation. Mouse maintenance and all experiments have been approved by the institutional animal care and use committee, in conformance with French and European Union laws (permission to experiment on mice #5558, animal facilities agreement D6334515, GMO agreement #4713).

49For the lysine nutritional deficiency experiment, an experimental diet was manufactured in the INRA diets core facility (Unité de Préparation des Aliments Expérimentaux, INRA) and nutritional experiments were performed as previously described. Briefly, the nutritional deficiency in an essential amino acid is carried out by means of experimental diets in which the protein fraction is replaced by a mixture of free amino acids.

50Male C57BL/6N mice were obtained from Taconic Farms (25-30 g). All mice were maintained in standard housing conditions (21-24°C; 45% humidity) on a 12 h light-dark cycle (lights on 06:00, lights off 18:00). Mice were singly housed in the BioDAQ caging system (Research Diets, New Brunswick, NJ, USA) and allowed ad libitum access to tap water and standard rodent chow (Purina 5053) unless otherwise noted. All procedures were approved by the Pfizer-Massachusetts Animal Care and Use Committee.

51Mouse embryonic fibroblast (MEF) cells lines were obtained from David Ron (CIMR/IMS, Cambridge) and maintained in Dulbecco’s Modified Eagle’s Medium supplemented with 10% (vol/vol) fetal bovine serum (FBS), 2 mM L-glutamine, penicillin/streptomycin, 1% Sodium Pyruvate, 1% Non-Essential Amino Acids and 2-Mercaptoethanol. HeLa (human cervical carcinoma obtained from ATCC), HuH7 (human hepatocarcinoma obtained from Albert Pol, IDIBAPS, Barcelona), A549 (human lung epithelial carcinoma obtained from ATCC) were cultured in the same media as MEFs but without 2-Mercaptoethanol. 3T3-L1 preadipocytes (obtained from Zenbio) cells were cultured in complete DMEM containing 10% newborn calf serum (NCS), 2 mM L-Glutamine, Penicillin-Streptomycin, 1% Non-Essential Amino Acids and 1% Sodium Pyruvate. All cells were maintained at 37°C in a humidified atmosphere of 5% CO2

52On the day before the assessment all participants received a standardized evening meal at 18:00, before commencing an overnight fast. The energy content of the meal was one third of a participant’s daily requirements estimated from predicted resting metabolic rate and multiplied by an activity factor of 1.35. Meal composition consisted of 30%–35% fat, 12%–15% protein and 50%–55% carbohydrate by energy. Anthropometric measurements were acquired for all participants on arrival to the clinical research facility. Participants were cannulated prior to administration of an oral liquid meal consisting of a 200 mL Ensure Plus (Total energy 330 kcal; Protein 16.7%, Carbohydrate 53.8%, Fat 29.5%) or a glucose drink (50 g anhydrous glucose in 200 mL water) with these particular participants described in. Blood samples were taken at 30 min intervals over the 180 min duration of the study. EDTA and Lithium heparin samples were placed immediately on ice while serum samples remained at room temperature for 30 min prior to centrifugation at 4°C at 3500 rpm for 10 min, plasma was frozen on dry ice and stored at −70°C until the time of biochemical analysis. Assays were completed by the Cambridge Biochemical Assay Laboratory, University of Cambridge. Serum GDF15 measurements were undertaken with antibodies & standards from R&D Systems (R&D Systems Europe, Abingdon UK) using a microtiter plate-based two-site electrochemiluminescence immunoassay using the MesoScale Discovery assay platform (MSD, Rockville, Maryland, USA). Plasma glucose was determined using a hexokinase assay on a Siemens Dimension ExL Analyzer. Plasma insulin measurements using the Diasorin Liaison XL automated onestep chemiluminesence immunoassay (Diasorin S.p.A, 13040 Saluggia (VC), Italy). Plasma total GLP-1 was measured by microtiter plate-based two-site electrochemiluminescence immunoassay using a Meso Scale Discovery kit (Gaithersburg, MD, USA). FGF21 levels were measured in duplicate on serum samples using the human FGF21 Quantikine ELISA kit (R&D Systems).

53Study volunteers attended the clinical research facility after an overnight fast. During the study, participants were required to eat all of the meals provided by the research team. For the first 24 h of the study (day 1), all of the meals provided to participants contained 100% of their estimated daily energy requirements based on the Scholfield equation and were composed of 50% carbohydrate, 30% fat and 20% protein. Baseline blood tests were acquired upon waking on day 2 of the study. For the following 48 h (day 2 and 3) participants were calorie restricted to meals containing 10% of their daily estimated energy requirements. Blood sampling was repeated upon waking on day 4 of the study protocol, prior to refeeding. Serum samples remained at room temperature for 30 min prior to centrifugation at 4°C at 3500 rpm for 10 min, plasma was frozen on dry ice and stored at −80°C until the time of biochemical analysis. Assays were completed by the Cambridge Biochemical Assay Laboratory, University of Cambridge. Serum GDF15 measurement were undertaken with antibodies & standards from R&D Systems (R&D Systems Europe, Abingdon UK) using a microtiter plate-based two-site electrochemiluminescence immunoassay using the MesoScale Discovery assay platform (MSD, Rockville, Maryland, USA).

54At day 0, prior to initiation of the low-calorie dietary (LCD) intervention, baseline fasting blood sampling and anthropometry were measured. Participants were provided with dietary counselling and meal-replacement shakes at this and subsequent visits, and the LCD was initiated. Participants received a low-calorie diet for 8 weeks at approximately 1000 kcal per day, replacing 2 meals with approximately 600 kcal of meal replacement shakes followed by a dinner of approximately 400 kcal. Dinners were chosen from an approved list of Lean Cuisine and Healthy Choice brand meals. Participants were free living for the duration of the study and attended at days 14 and 28 of the intervention for assessment where blood sampling and anthropometric measures were repeated.

55EDTA plasma samples underwent centrifugation at 4°C, 4000 rpm for 15 min and stored at −80°C until analysis. Biochemical assays were undertaken at the Translational Research Institute for Metabolism and Diabetes (Florida Hospital). Plasma GDF15 was measured using antibodies & standards from R&D Systems (R&D Systems Europe, Abingdon UK) using a microtiter plate-based two-site electrochemiluminescence immunoassay using the MesoScale Discovery assay platform (MSD, Rockville, Maryland, USA). Plasma leptin was assayed using the MesoScale Discovery platform, (Human Leptin).

56Participants were free living for the duration of the caloric restriction. Anthropometric measurements were acquired at baseline and at the end of the study. On day 0 all participants had a breakfast meal ad libitum prior to commencing the caloric restriction to 0 kcal per day for a total of 7 days. Water was permitted throughout the study. During the study, weight (mean ± SD) fell from 79.6 ± 5.0 kg at baseline to 73.8 ± 4.8 kg after 1 week of fasting. The measurements of mean ± SD body fat were acquired by dual-energy X-ray absorptiometry (DEXA) and are reported as an average of two measurements at baseline on day −1 and 0 (18.6 ± 1.8 kg) or following the fast on days 6 and 7 (17.3 ± 1.9 kg). Participants attended the research facility each morning for phlebotomy where both serum and plasma (EDTA and Lithium Heparin) samples were acquired. Plasma samples were immediately placed on ice while serum samples remained at room temperature for 30 min to coagulate prior to centrifugation at 4°C at 3500 rpm for 10 min. Samples were then immediately frozen on dry ice and stored at −80°C until the time of biochemical analysis. Plasma Leptin and GDF15 measurements were undertaken at the Cambridge Biochemical Assay Laboratory, University of Cambridge using antibodies & standards from R&D Systems (R&D Systems Europe, Abingdon UK). A two-site microtiter plate-based Delfia assay measured Leptin. GDF15 was measured using a microtiter plate-based two-site electrochemiluminescence immunoassay using the MesoScale Discovery assay platform (MSD, Rockville, Maryland, USA). The analytic processes of β-Hydroxybutyrate were conducted according to the manufacturer’s instructions. Plasma concentration of β-Hydroxybutyrate (mmol/l) were undertaken at the Department of Clinical Medicine, Diabetes and Hormone Diseases - Medical Research Laboratory, Aarhus University, Denmark using a kinetic enzymatic method, based on the oxidation of D-3-hydroxybytyrate to acetoacetate by the enzyme 3-Hydroxybutyrate dehydrogenase (Randox Laboratories Ltd., Crumlin, UK) and measured on the Cobas c111 system (Roche Diagnostics International Ltd, Rotkreuz, Switzerland). FGF21 levels were measured in duplicate on serum samples using the human FGF21 Quantikine ELISA kit (R&D Systems).

57Prior to the start of the study, subjects attended the research facility at Loughborough University for an initial assessment of their baseline anthropometric characteristics. This information was then used to estimate resting energy expenditure (REE) using validated equations. A standard correction for physical activity (1.6 and 1.7 times REE for females and males, respectively) was applied in order to estimate total daily energy requirements. This information was then used to determine individual energy intakes for the overfeeding period. Experimental trials were conducted immediately before and after 7 days of high-fat overfeeding. Briefly, subjects arrived at the laboratory in the morning (07:00-09:00) after an overnight fast (≥10 h), having refrained from strenuous exercise for 48 h and having avoiding alcohol or caffeine intake for 24 h. Body mass was recorded after subjects had voided. A venous blood sample was then obtained after 30 min of seated rest. Blood samples were collected for plasma (EDTA) or serum. Blood samples were then centrifuged, and the resulting plasma / serum stored at −20°C until analysis. Upon completion of the first experimental trial, subjects were provided with all food to be consumed for the following 7 days. The high-fat diet provided 19974 ± 474 kJ per day (48 ± 1% greater than estimated daily requirement), with 178 ± 5 g [15%] protein, 245 ± 5 g [21%] carbohydrate, and 342 ± 9 g [64%] fat intake. Diet compliance was assessed by daily interview. Plasma glucose concentration was determined using a spectrophotometric assay (Glucose PAP; Horiba Medical, Northampton, UK) and semi-automatic analyzer (Pentra 400; Horiba Medical, Northampton, UK). Serum insulin concentration was determined by ELISA (EIA-2935; DRG Instruments GmBH, Marburg, Germany). Serum Leptin and GDF15 measurements were undertaken at the Cambridge Biochemical Assay Laboratory, University of Cambridge using antibodies & standards from R&D Systems (R&D Systems Europe, Abingdon UK). A two-site microtiter plate-based Delfia assay was used to measure Leptin. GDF15 was measured using a microtiter plate-based two-site electrochemiluminescence immunoassay using the MesoScale Discovery assay platform (MSD, Rockville, Maryland, USA). FGF21 levels were measured in duplicate on serum samples using the human FGF21 Quantikine ELISA kit (R&D Systems).

58Details of this study were previously described. Briefly, this was a randomized, parallel-arm, in-patient study. Participants remained in-patients at the Biomedical Research Center for approximately 12 weeks without leaving. The first 13–25 days of the in-patient stay (Baseline) were used to establish energy requirements for weight maintenance. The baseline diet consisted of 361 g of carbohydrates, 67 g of fat, 90 g protein for a total of 2412 kcal. Once weight stability was achieved, baseline assessments were performed, including blood draws and measurements of body composition by dual-energy X-ray absorptiometry (DEXA). Overfeeding was planned at approximately 40% above energy requirements for weight maintenance or approximately 1000 kcal/d (4180 kJ/d). During the final 24 h period, the diet was returned to the baseline components and the same baseline assessments were performed at the end of the 8 week overfeeding. Participants ate all food provided during the study period. Plasma glucose was measured using a glucose oxidase electrode (DXC 600 Pro; Beckman Coulter), and insulin was measured by an immunoassay (Immulite 2000; Siemens). Plasma free fatty acids (FFAs) were measured with a high-sensitivity Wako kit. Plasma GDF15 measurements were undertaken with antibodies & standards from R&D Systems (R&D Systems Europe, Abingdon UK) using a microtiter plate-based two-site electrochemiluminescence immunoassay using the MesoScale Discovery assay platform (MSD, Rockville, Maryland, USA).

59When aged 11-12 weeks old, on day 1 of the study the mice were divided into two groups of five mice, either “fed” or “fasted.” The body weight of the two groups at study start were matched (fed versus fast, mean ± SEM; 27.68 ± 0.45 g versus 27.70 ± 0.59 g). At 09:00, mice were transferred into clean cages in the same grouped arrangement as during the maintenance period. Home cage bedding was also transferred into clean cages to reduce stress. Animals in the “fasted” group had all food removed, animals in the “fed” had ad libitum access to food. All animals had free access to water. 24 h later (09:00 on day 2 of study) mice were weighed then received a terminal dose of anesthetic (Dolethal 200 mg/mL solution, Vetoquinol UK Ltd.) given via the intraperitoneal route. Once unresponsive, blood was collected by cardiac puncture, transferred into a Microtube 1.1 mL Z-Gel (Sarstedt AG & Co) and spun at 10 000 x g for 5 min at room temperature. Serum was collected, frozen on dry ice and stored at −80°C until analyzed. After cardiac puncture, body composition was measured using DEXA with a Lunar PIXImus Mouse Densitometer (GE Healthcare Systems) and tissue was harvested, frozen on dry ice and stored at −80°C until being processed.

60For the short-term high fat diet study (Mouse Study 2), 17-18 week old mice were fed a 45% high-fat diet (D12451i, 45% kcal as fat, 4.7 kcal/g, Research Diets) for 1, 3 or 7 days. A separate control group was kept on a chow diet (Safe Diets, DS-105). On the morning (10:00) of the specified days, mice were weighed and blood collected by cardiac puncture into microtubes containing serum gel with clotting activator and centrifuged at 13 000 rpm for 10 min at 4°C and stored at −80°C for serum GDF15 and insulin measurements. Mouse glucose levels were measured from approximately 2 μl blood drops using a glucometer (AlphaTrak2; Abbot Laboratories) and glucose strips (AlphaTrak2 test 2 strips, Abbot Laboratories, Zoetis). Tissues were harvested and weighed.

61For the long-term chronic high fat diet study (Mouse Study 3), 9 week-old male mice were subjected to either a chow or high fat diet (as in short-term HFD) over a period of 18 weeks. All mice were weighed weekly and body composition was determined every 4 weeks by Time-Domain Nuclear Magnetic Resonance (TD-NMR) using a Minispec Live Mice Analyzer (LF50, Bruker). Tail blood samples were collected into heparinized micro blood tubes (01605-00, Hawksley), centrifuged at 13,000 x g for 4 min for plasma GDF15, leptin and insulin measurement. Mouse glucose levels were measured as described above in the short-term HFD study. At the end of the experiment, mice were fasted for 4 h and tissue was harvested and stored at −80°C. For isolating stromo-vascular and adipocyte fractions, epididymal adipose tissue was minced into small pieces and resuspended in 5 mL Hanks’ Balanced Salt Solution (Sigma) containing collagenase Type I (Sigma). The tissue was completely disaggregated by incubation in a 37°C shaker for approximately 10 min. The digested material was filtered through a 100 μm nylon mesh cell strainer, and 10 mL of 10% FBS DMEM added. After a 5-10 min incubation at room temperature, the upper phase containing the adipocytes was transferred into a new tube. The remaining supernatant was centrifuged at 700 x g for 10 min and the pellet containing the stromo- vascular fraction was collected. Both fractions were frozen at −80°C until further analysis.

62Eighteen 10 week old C57BL/6J female mice were habituated to the control experimental diet (containing 20 free amino acids) for one week. The night before the experiment, mice were fasted for 16 h before offering them a control meal or a meal lacking lysine. Mice were divided into three groups of six mice: the first group was fasted overnight; the second group was fasted overnight, then fed the control diet; the third group was fasted overnight, then fed the lysine devoid diet. A blood sample was collected from fed mice at 1 h after the beginning of the meal by sub-mandibular sampling. At the time of sacrifice (4 h after the beginning of the meal for fed mice), the blood of all mice was withdrawn by cardiac puncture and treated with EDTA (500 mM) along with the tissues being harvested and stored at −80°C for analysis.

63Human recombinant GDF15 (Peprotech) was prepared in saline, which was used as the vehicle control. GDF15 was administered via subcutaneous (SC) injection as a single dose in all mouse studies. LiCl (Sigma) was also prepared in saline and administered via SC injection as a single dose in all mouse studies. Mice were acclimated (up to 3 days) to drinking from two water bottles to confirm lack of side preference prior to habituation. Mice were then habituated to overnight water restriction (days 1-3) followed by 1 h water bottle presentation (two bottles) and saline SC injection. On day 4 to begin conditioning, mice were instead given a novel 0.15% saccharin solution in both bottles instead of water for 1 h, followed by a SC injection of either saline, GDF15 (within the dose range that induces anorexia) or the positive control LiCl. Access to saccharin water was allowed for an additional 30 min and was then changed back to water until the next restriction. Day 5 was used as a GDF15 washout period using the days 1-3 bottle protocol. A second conditioning period was performed on day 6 followed by a washout period on day 7. On day 8, a standard two bottle preference test (saccharin versus water) was used to assess CTA development to the saccharin solution (1 h presentation after overnight water restriction). The CTA test was performed 48 hours after the last GDF15 injection, and volume measurements were for 1 hour. Fluid intake volume was calculated for both saccharin and water. The total volume drunk in saline group was 1.4 mL and there was no statistical differences in the treatment groups. Food weight was measured manually using a digital scale. Food weight was measured at 1 h and 4 h following a single injection of GDF15 given immediately prior to the onset of the dark cycle. A separate group of mice was used for plasma GDF15 pharmacokinetic analysis. Blood was collected at 0.25 h, 1 h, 4 h, 8 h, and 24 h after GDF15 injection in EDTA tubes containing AEBSF and aprotinin, and then centrifuged (10 000 rpm; 10 min) for plasma separation all at 4°C and then stored at −80°C. Plasma human GDF15 was measured using the human GDF15 Quantikine ELISA kit per manufacturer instructions.

64Cells were seeded onto 6- or 12-well plates prior to stress treatments the following day for the times and concentrations indicated. Vehicle treatments (e.g., DMSO or ethanol) were used for control cells when appropriate.

65Wild-type MEFs were seeded onto 12-well plates and transfected with 30 nM control siRNA or siRNA for mouse CHOP using Lipofectamine RNAi MAX (Invitrogen) according to the manufacturer’s instruction. 48 h post siRNA transfection, cells were treated with ISR stressors for 6 h and subsequently processed for RNA and protein expression analysis.

66Following treatments, cells were lysed with Buffer RLT (QIAGEN) containing 1% 2-Mercaptoethanol and processed through a Qiashredder with total RNA extracted using the RNeasy isolation kit according to manufacturer’s instructions (QIAGEN). Meanwhile for mice, tissues were harvested and immediately snap frozen in liquid nitrogen and stored at −80°C until further analysis. For RNA isolation, approximately 30-50 mg of tissue was placed in Lysing Matrix D tubes and homogenized in 800 μl Triazol (Qiazol,QIAGEN) using the Fastprep-24 Homogenizer for 30 s at 4-6 m/s (MP Biomedical), or a rotor-stator homogenizer. The resultant supernatant was transferred to an RNase free tube and 200 μl chloroform (Sigma) added. The samples were vortexed and centrifuged at 13 000 rpm for 15 min at 4°C. The upper phase was then transferred to a RNase free tube and mixed with equal volume of 70% ethanol before loading onto RNA isolation spin columns (QIAGEN). RNA was then extracted (and in some instances treated with DNase1 on-column) using the RNeasy isolation kit following the manufacturer's instructions.

67RNA concentration and quality was determined by Nanodrop. 400 ng - 500 ng of total RNA was treated with DNAase1 (Thermofisher Scientific) and then converted to cDNA using MMLV Reverse Transcriptase with random primers (Promega). Quantitative RT-PCR was carried out with either TaqMan Universal PCR Master Mix or SYBR Green PCR master mix on the QuantStudio 7 Flex Real time PCR system (Applied Biosystems). All reactions were carried out in either duplicate or triplicate and Ct values were obtained. Relative differences in the gene expression were normalized to expression levels of housekeeping genes, HPRT or GAPDH for cell analysis and to B2M and 36b4 geometrical mean for mouse data, using the standard curve method. Primer sequences are shown in the Key Resources Table.

68Tail blood samples from random fed or 4 h fasted (for the long term HFD study) mice were collected for serum analysis. Mouse leptin and insulin were measured simultaneously using a 2-plex Mouse Metabolic immunoassay kit from Meso Scale Discovery Kit (Rockville, MD, USA). The assay was performed according to the manufacturer’s instructions and using calibrators provided by MSD. Mouse GDF15 was measured using a Mouse GDF15 DuoSet ELISA (R&D Systems) which had been modified to run as an electrochemiluminescence assay on the Meso Scale Discovery assay platform. Mouse FGF21 was analyzed by FGF21 Quantakine ELISA (R&D Systems) following the manufacturer’s instructions. Mouse sample measurements were performed by the MRC MDU Mouse Biochemistry Laboratory [MRC_MC_UU_12012/5]. For the human studies, the details of the serum/plasma analysis performed are described separately for each study (see Method Details section of each study above). All FGF21 measurements were completed by the Cambridge Biochemical Assay Laboratory, University of Cambridge using the human FGF21 Quantikine ELISA kit (R&D Systems).

69Following treatments, cells were washed twice with ice cold D-PBS and proteins harvested using RIPA buffer supplemented with cOmplete protease and PhosStop inhibitors (Sigma). The lysates were cleared by centrifugation at 13 000 rpm for 15 min at 4°C, and protein concentration determined by a Bio-Rad DC protein assay. Typically, 20-30 μg of protein lysates were denatured in NuPAGE 4 × LDS sample buffer and resolved on NuPage 4%–12% Bis-Tris gels (Invitrogen) and the proteins transferred by iBlot (Invitrogen) onto nitrocellulose membranes. The membranes were blocked with 5% nonfat dry milk or 5% BSA for 1 h at room temperature and incubated with the antibodies described in the STAR Methods table. Following a 16 h incubation at 4°C, all membranes were washed five times in Tris-buffered saline-0.1% Tween-20 prior to incubation with horseradish peroxidase (HRP)-conjugated anti-rabbit immunoglobulin G (IgG), HRP-conjugated anti-mouse IgG. The bands were visualized using Immobilon Western Chemiluminescent HRP Substrate (Millipore). All images were acquired on the ImageQuant LAS 4000 (GE Healthcare).

70Quantitative data are reported as mean ± SD for cells and mean ± SEM for mouse data. As indicated in the figure legends, differences between means were assessed by two-tailed Student’s t tests or One-way ANOVA or Two-way ANOVA with Bonferroni multiple comparisons test using either GraphPad Prism software (GraphPad, San Diego) or with SAS version 9.4, Cary, N. Carolina. Statistical significance was defined as p < 0.05.

71Data from human studies was analyzed using GraphPad Prism software (GraphPad, San Diego). Parametric quantitative data is expressed as mean ± SEM and the difference in the mean was assessed using a Two tailed Student’s t test. In the case of non-parametric data, it is reported as median (interquartile range) and compared using a Wilcoxon signed rank, Kruskal-Wallis or Mann Whitney test. Details of specific analyses are reported in the respective figure legends. Statistical significance was defined as p < 0.05.