Combined genetic deletion of GDF15 and FGF21 has modest effects on body weight, hepatic steatosis and insulin resistance in high fat fed mice

Objectives Obesity in humans and mice is associated with elevated levels of two hormones responsive to cellular stress, namely GDF15 and FGF21. Over-expression of each of these is associated with weight loss and beneficial metabolic changes but where they are secreted from and what they are required for physiologically in the context of overfeeding remains unclear. Methods Here we used tissue selective knockout mouse models and human transcriptomics to determine the source of circulating GDF15 in obesity. We then generated and characterized the metabolic phenotypes of GDF15/FGF21 double knockout mice. Results Circulating GDF15 and FGF21 are both largely derived from the liver, rather than adipose tissue or skeletal muscle, in obese states. Combined whole body deletion of FGF21 and GDF15 does not result in any additional weight gain in response to high fat feeding but it does result in significantly greater hepatic steatosis and insulin resistance than that seen in GDF15 single knockout mice. Conclusions Collectively the data suggest that overfeeding activates a stress response in the liver which is the major source of systemic rises in GDF15 and FGF21. These hormones then activate pathways which reduce this metabolic stress.


INTRODUCTION
Healthy adipose tissue is essential when coping with sustained nutritional overload and manifests a remarkable capacity to increase in size [1]. Depending on factors such as the duration and macronutrient source of excess caloric intake, fat mass can double in humans [2] and quadruple in rodent models [3]. Ultimately however, adipocytes start to die triggering a chronic inflammatory response, in which macrophages play a key role [4e9]. Since excess energy cannot normally be excreted and humans, at least, manifest very small increases in energy expenditure when overfed [10], it necessarily accumulates in other ectopic sites (and in plasma lipoproteins to some extent), setting off a cascade of metabolic complications. In the liver, this ectopic lipid accumulation is strongly associated with the development of insulin resistance [11e17] and in some cases a hepatic inflammatory response. As many as a third of patients with non-alcoholic fatty liver disease (NAFLD) progress to cirrhosis and some of these go on to develop hepatocellular carcinoma [18,19]. Lipid also accumulates in skeletal muscle [15,20,21] where it has again been linked to the pathogenesis of insulin resistance and type 2 diabetes [13,22].
Although there are some reports of 'inflammation' in skeletal muscle in this context, the underlying mechanisms, and the impact on muscle itself and on systemic insulin resistance are less clear [23]. Ectopic lipid accumulation in the liver has been reported by several groups to be associated with activation of the integrated stress response (ISR) [24e27], whereas in skeletal muscle of humans or high fat fed mice there is limited evidence of induction of the ISR [24,28,29]. The ISR is triggered by a variety of stressors (such as endoplasmic reticulum stress, mitochondrial stress, hypoxia and, amino acid or glucose depletion) which activate one or more of at least four kinases leading to the phosphorylation of eukaryotic initiation factor 2 alpha (eIF2a) [30]. This in turn attenuates overall protein synthesis whilst permitting selective translation of specific proteins required for cellular adaptation, repair and alteration of metabolic homeostasis [31,32]. We [33] and others [34e36] have shown that activation of the ISR (via nutritional, genetic or pharmacological stressors) is associated with increased expression of growth differentiation factor 15 (GDF15) and fibroblast growth factor 21 (FGF21). Plasma levels of both FGF21 [37e41] and GDF15 [42e44] are known to be increased in obese humans and rodents as well as in other metabolic disease states such as insulin resistance [45,46], NAFLD [47e52] and mitochondrial disease [53,54]. Both stress-induced cytokines, GDF15 and FGF21, have attracted considerable interest as potential therapies for obesity and its associated metabolic disease [55]. FGF21 was discovered in 2000 and reported to be primarily expressed in the liver [56] though it is now known to be more widely expressed, including other key metabolic tissues such as adipose tissue, skeletal muscle and pancreas [57,58]. FGF21 was found to be a potent regulator of glucose uptake in an in vitro screen in 3T3-L1 adipocytes [41] and has subsequently been associated with a range of pleiotropic effects, including improved insulin sensitivity and b cell function, reduced hepatic lipogenesis, and increased energy expenditure via brown fat thermogenesis [59]. Using tissue-specific knockout mice, plasma FGF21 was shown to be primarily derived from the liver in response to high fat feeding [60]; it then acts centrally to regulate energy expenditure and body weight [61,62]. FGF21 is also reported to act in an auto/paracrine fashion when secreted by adipocytes and pancreatic exocrine cells [63e66]. Molecularly, FGF21 signals via the FGF receptor-1c and its cognate co-receptor bklotho which is predominantly expressed in target tissue such as the CNS and adipocytes [67e69]. Both transgenic overexpression and exogenous administration of supraphysiologic levels of FGF21 in genetic-or diet-induced obese (DIO) rodents, substantially reduces body weight, hypertriglyceridaemia and hyperglycaemia [41,70,71]. These responses were shown to be associated with reductions in hepatic and intramyocellular lipid content (TAG and DAG) [72]. Beneficial metabolic effects of exogenous FGF21 administration were also observed in obese or diabetic primates [73,74] and in humans [74,75], although the improvements in hyperglycaemia were disappointingly modest in humans. FGF21 deficiency in mice has been linked with reduced insulin sensitivity in a high fat diet setting [76], but the impact on body weight remains unclear as studies have reported both lower and higher body weight in FGF21 null mice [76,77]. Some studies suggested that the weight of FGF21 knockout mice was increased early after transition to a high fat diet but that this difference was subsequently lost [66,78]. GDF15, originally identified as a gene upregulated in activated macrophages [79], can be expressed in almost all cell types, and is relatively highly expressed in several tissues including the liver, kidneys, intestines and especially the placenta. Plasma GDF15 is elevated in a range of human diseases, in addition to obesity and the metabolic syndrome, where it is widely considered to be a useful biomarker [80e 82]. GDF15 mRNA expression is elevated within the liver and adipose tissue of high fat fed mice [33]. GDF15 null mice weigh more than WT littermates on a high fat diet and are glucose intolerant [83]. Meanwhile, GDF15 over-expressing transgenic mice are protected from dietinduced obesity and display improved insulin sensitivity [84]. Concordantly, pharmacological treatment with recombinant GDF15 reduces food intake and body weight in obese rodents and primates [43,85,86]. These metabolic impacts of GDF15 are mediated via the GFRAL receptor along with its tyrosine kinase coreceptor Ret in the hindbrain [87,88]. Here we sought firstly to clarify the principal source of GDF15 in HFD fed mice. Having shown that the liver is a major source of GDF15 in this context, similarly to what has previously been reported for FGF21, we proceeded to evaluate the phenotypic impact of deleting both genes in mice. As both are present at least as far back as zebrafish [89,90] and both have been implicated in weight loss, we hypothesized that they might act synergistically to alleviate the stress imposed by nutritional overload so we generated and characterised GDF15:FGF21 double knockout mice.

Animal husbandry
Mice were maintained in ventilated cages with group housing (2e5 per cage), unless specified otherwise for indirect calorimetry and food intake experiments, on a 12 h light/12 h dark cycle (lights on 06:00e 18:00), in a temperature-controlled (20e24 C) facility, with ad libitum access to food and water. During the experimental protocol, all mice were fed either ad libitum or fasted as stated prior to some tests. All animal studies were performed on male mice and carried out at two facilities at the University of Cambridge, UK. 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).

Original Article
Myeloid-specific deletions of GDF15 were generated using two separate strategies:  mm cell strainer and centrifuged at 400 Â g for 5 min and resuspended in macrophage differentiation medium [RPMI-1640 with 20% L929 conditioned medium, 10% heat-inactivated FBS and 100 U/ml penicillin-streptomycin (Thermo Fisher Scientific)]. Total bone-marrow cells were counted using a Countess II automated cell counter (Thermo Fisher) and seeded onto 10 cm non-culture treated plates (Falcon) at a density of 5 Â 10 6 cells per plate and maintained for 7 days at 37 C in a humidified atmosphere of 5% CO 2 . On day 5 of differentiation, medium was removed, and 10 ml of fresh macrophage differentiation medium was added to each plate. On day 7 of differentiation, macrophages were detached using ice-cold PBS containing 1 mM EDTA and centrifuged at 400 g for 5 min. The cells were resuspended in macrophage differentiation medium and seeded onto 24 well plate for 24 h prior to experiments. To make L929 conditioned medium, L929 cells (CCL-1, ATCC) were seeded in DMEM supplemented with 10% heat-inactivated FBS, 100 U/ml penicillin-streptomycin and 2 mM L-glutamine (Sigma) at a density of 250,000 cells per 50 ml of medium per T175 tissue culture flask. Medium was harvested after 1 week of culture, and then 50 ml of fresh DMEM supplemented with 10% heat-inactivated FBS, 100 U/ ml penicillin-streptomycin and 2 mM L-glutamine was added onto cells and harvested 1 week later. Batches obtained after the first and second weeks of culture were mixed at a 1:1 ratio, aliquoted and stored at À20 C.
2.10. RNA isolation, cDNA synthesis and qPCR for mouse studies Following 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 À80 C until further analysis. For RNA isolation, approximately 30e50 mg of tissue was placed in Lysing Matrix D tubes and homogenized in 800 ml TRI reagent (Sigma-Aldrich) using the previously described [92]. Briefly, punch biopsies were collected from a sun-protected area of the abdomen from each participant, from which adipose tissue was separated and RNA extracted. Gene expression was measured by RNA-Seq, with RNA-Seq reads aligned to the hg19 reference genome using STAR [97] version 2.4.0.1, as fully described elsewhere [98]. Gene level counts were generated using the quan function of QTLtools [99] and Gencode version 19 [100], and rank-based inverse normal transformation then applied to gene counts per million (CPMs) prior to all downstream analyses. Gene expression levels were then adjusted for RNA-Seq technical covariates using linear mixed effects models, with expression levels of each gene in turn treated as a dependent variable, with median insert size and mean GC content included as fixed effects, and date of RNA sequencing and RNA-Seq primer index as random effects.
In the obese cohort, paired samples of abdominal omental AT (visceral, VAT) and subcutaneous AT (SAT) were obtained from 525 Caucasian men (n ¼ 143) and women (n ¼ 382), who underwent open abdominal surgery as described previously [94,95]. The age ranged from 18 to 80 years and body mass index (BMI) from 19 to 60 kg/m 2 . AT was immediately frozen in liquid nitrogen and stored at À80 C. RNA was extracted from AT by using the RNeasy Lipid Tissue Mini Kit (Qiagen, Hilden, Germany), and qPCR was performed as described elsewhere [101,102]. Real-time quantitative PCR was performed with the TaqMan Assay predesigned by Applied Biosystems (Foster City, CA, USA) for the detection of human GDF15 (Hs00171132_m1), CD68 (Hs02836816_g1) and GAPDH (Hs 02786624_g1) mRNA expression in AT. All reactions were carried out in 96-well plates using the Quant-Studio (TM) 6 Flex System Fast Real-Time PCR system. GDF15 mRNA expression was calculated relative to GAPDH mRNA expression 2.13. Estimation of adipose tissue cell type proportions Adipose tissue cell type proportions were estimated from RNA-Seq data in TwinsUK using CIBERSORT [103], as reported previously elsewhere [98]. Estimated cell types included in our analysis were adipocytes, microvascular endothelial cells (MVEC) and macrophages.
2.14. Quantification and statistical analysis Cell and mouse quantitative data are reported as mean AE standard deviation (SD). As indicated in the figure legends, differences between means were assessed by two-tailed Student's t tests or One-way ANOVA or Original Article random effects that permit identification of the family a twin belongs to, and their clonality (MZ/DZ status).
To assess association of adipose tissue GDF15 expression levels with gene expression levels of macrophage markers CD68 and EMR1, and adipose tissue estimated macrophage proportion, GDF15 gene expression residuals were treated as a dependent variable, and each of adipose tissue expression residuals of CD68, and EMR1, and adipose tissue estimated macrophage proportion, were used in turn as predictor variables. Models included the same fixed and random effect covariates as described for the GDF15 serum level model. For each analysis, the full model was then compared to a null model where the trait of interest was omitted, using a 1 degree of freedom ANOVA. For obese human subject data, prior to statistical analysis, non-normally distributed parameters were logarithmically (ln) transformed to approximate a normal distribution. Results are expressed as mean AE SD. Linear regression analysis were used to assess the relationships between GDF15 mRNA expression and phenotypic traits. Pearsons correlation analyses were conducted using two-way bivariate correlations. Differences in GDF15 mRNA expression between visceral and subcutaneous AT were assessed using the paired Student's t-test or one-way ANOVA. Statistical analyses were performed using SPSS/ PC þ for Windows statistical package (version 27.0; SPSS, Chicago, IL, USA).

Impact of GDF15 deletion on body weight regulation in mice
In order to confirm prior reports suggesting that GDF15 null mice are heavier when fed a high fat diet (HFD) [83], wild-type (WT) and GDF15 KO mice were fed a 60% HFD for w20 weeks. The GDF15 KO mice displayed a small but significant increase in absolute body weight compared to their WT littermates ( Figure 1A), the difference becoming evident from about 10 weeks into the HFD, particularly when the data was displayed as percent body weight gain ( Figure 1B). Daily food intake was similarly very modestly increased in the GDF15 null mice whereas energy expenditure was similar to that of WT mice (Supplemental Figure 1A and B). Subcutaneous, epididymal and brown adipose tissue weights were similar between the groups, but GDF15 KO mice manifested increased liver tissue mass ( Figure 1C). Biochemical ( Figure 1D) and histological analyses revealed elevated hepatic lipid content in GDF15 KO mice ( Figure 1E and Supplemental Figure 1C) and plasma triglycerides; cholesterol, leptin and liver enzyme concentrations were also higher than those of wildtype littermates ( Figure 1F). Blood glucose concentrations were similar in 6 h fasted mice whereas plasma insulin and HOMA-IR were elevated suggesting mild insulin resistance ( Figure 1G). Glucose tolerance was also mildly impaired in the GDF15 KO group whereas insulin tolerance was similar to that of WT mice suggesting that the insulin resistance might primarily relate to the liver rather than peripheral tissues ( Figure 1H,I). High fat feeding is known to increase plasma FGF21 in mice [33] and was even higher in GDF15 KO mice after 16 weeks of HFD feeding ( Figure 1J). In keeping with this observation, FGF21 mRNA was increased in the liver of the GDF15 KOs but not in WAT, BAT or skeletal muscle ( Figure 1K).

GDF15 is upregulated in adipose tissue macrophages and hepatocytes
Having previously shown that high fat feeding in mice is associated with elevated plasma GDF15 levels and increased GDF15 mRNA in white and brown adipose tissue, and the liver, we used RNA scope to identify the cellular source. In white adipose tissue, the vast majority of GDF15 expression was detected within macrophages contributing to the formation of 'crown-like structures', along with occasional spots which appeared to be in adipocytes (Figure 2AeC). In chow fed mice, RT-PCR quantification confirmed that GDF15 mRNA was mainly expressed in myeloid cells (CD11b þ SVF) with very little expression in adipocytes or CD11b-SVF (stromal vascular fraction) ( Figure 2D). In HFD fed mice, GDF15 was readily detectable within all three eWAT fractions. However, GDF15 expression within the CD11b-SVF and adipocyte fractions is likely due to macrophage contamination, as they express the macrophage marker EMR1 which correlates with GDF15 expression (Figure 2E,F).
In keeping with these murine data, GDF15 mRNA expression in human subcutaneous adipose tissue samples in 733 individuals from the TwinsUK cohort was strongly associated with the estimated adipose tissue macrophage proportion (b ¼ 32.81; P < 2.2 Â 10 À16 ) ( Figure 2G), as well as with expression of macrophage markers CD68 (b ¼ 0.68; P < 2.2 Â 10 À16 ) and EMR1 (b ¼ 0.21; P ¼ 3.5 Â 10 À9 ) ( Figure 2H,I). We confirmed these findings in a second large obese cohort (n ¼ 525) in which we also had access to paired visceral adipose tissue samples. In this sample set, GDF15 mRNA expression was again strongly correlated with CD68 mRNA in subcutaneous fat. The correlation was also present in visceral fat, though interestingly was slightly weaker in that depot (Subcutaneous fat: r ¼ 0.562, P < 0.001; Visceral fat: r ¼ 0.453, P < 0.001). Both associations withstand adjustment for age, sex and BMI (Padj < 0.001) ( Figure 2J,K). In liver samples from HFD fed mice, the RNA scope analysis suggested that GDF15 mRNA was primarily detectable in hepatocytes with no visually apparent contribution from non-hepatocyte cell populations ( Figure 2L). Previously published work has suggested that GDF15 expression is also increased in human liver tissue samples in the context of NAFLD [106]. In brown adipose tissue, the GDF15 mRNA signal appeared to originate predominantly from selected brown adipocytes ( Figure 2M). In contrast, HFD-induced FGF21 is reported to be predominantly expressed within adipocytes themselves whereas within the liver it is also expressed in hepatocytes [60]. We confirmed the former using RT-PCR in epididymal WAT fractions ( Figure 2N).

Characterisation of macrophage specific GDF15 KO mouse models
As macrophages appear to be the predominant site of GDF15 expression in adipose tissue, we proceeded to generate myeloidspecific GDF15 knockout mice (KO) using Cre recombinase under the control of the Lysozyme M promoter (LysM Cre). LysM-GDF15 KO were viable, fertile and born with no obvious abnormalities. Deletion of GDF15 was confirmed within isolated bone marrow derived macrophages (BMDMs), where we observed a >90% reduction in GDF15 mRNA (Supplemental Fig. 3). In keeping with these data, secretion of GDF15 into the media from isolated GDF15-null BMDMs was undetectable (Supplemental Figure 3). To verify that GDF15 was effectively deleted within the myeloid population in vivo, adipose tissue macrophages (ATMs) were purified from WT and LysM-GDF15 KO epididymal adipose tissue. Like in BMDMs, mRNA analysis revealed >90% reduction of GDF15 expression in the LysM-GDF15 KO ATMs ( Figure 3A). Furthermore, analysis of gene expression from WAT displayed significant loss of GDF15 expression in LysM-GDF15 KO mice but not in other tissues ( Figure 3B). As expected, plasma concentrations of GDF15 increased during HFD feeding in all mice, but GDF15 concentrations were w30% lower in the LysM-GDF15 KO cohort ( Figure 3C).     concentrations were similar between genotypes ( Figure 3D,E). Weekly weight analysis revealed no differences in body weight or percent body weight gain in the LysM-GDF15 KO mice on a HFD compared to their WT counterparts ( Figure 3F,G). Tissue weights, hepatic lipid content and plasma leptin were similar to WT mice (Figure 3HeJ) as were fasting glucose, insulin, HOMA IR and both glucose and insulin tolerance (Figure 3KeM). Plasma lipids (TG and cholesterol) were similar, as were plasma liver enzyme concentrations ( Figure 3N). Since the LysM-Cre driven deletion is present from birth and specifically targets the myeloid population, it is possible that compensatory mechanisms might account for the very modest impact on plasma GDF15 concentrations. To test this possibility, we transplanted bone marrow from GDF15 KO mice to regenerate the hematopoietic lineage in irradiated WT mice. GDF15 mRNA expression was effectively deleted within the BMT-GDF15 KO WAT depots but not in other tissues (liver and BAT) ( Figure 4A). Within WAT, GDF15 mRNA was substantially reduced in the adipocyte fraction and in the CD11b -and CD11b þ SVFs (Supplemental Figure 4). The fact that EMR1 mRNA expression was unchanged in these fractions in HF fed mice is consistent with macrophages being present in all 3 fractions as expected i.e. the macrophages are present but do not express GDF15 mRNA (Supplemental Figure 4A). In contrast, Plin1 mRNA was only present in the adipocyte fraction. We interpret these observations as being consistent with GDF15 mRNA in WAT largely being present in macrophages. Plasma GDF15 concentrations were similar in both lines ( Figure 4B) and were substantially higher than in the LysM-Cre cohorts ( Figure 3C). We presume that this relates to the irradiation therapy which is known to increase GDF15 [107,108]. However, similarly to what was observed in the whole body GDF15-null mice, plasma FGF21 concentrations and hepatic FGF21 mRNA expression were increased in the BMT-GDF15 KO mice ( Figure 4C,D). When both groups were fed a 60% HFD, the BMT-GDF15 KO cohort displayed similar weight gain to the WT cohort ( Figure 4E,F). Tissue weight analyses revealed a slight reduction in epididymal fat mass and a trend for increased liver weight ( Figure 4G) but plasma TG, cholesterol, leptin concentrations as well as liver enzymes were similar ( Figure 4H). Furthermore, there was no difference in hepatic fat content ( Figure 4I). Fasting glucose and insulin levels, and HOMA IR were similar to WT levels as was glucose tolerance ( Figure 4J,K), whereas insulin tolerance appeared to be modestly impaired ( Figure 4L). Collectively, these data suggest that myeloid cells are the main producer of adipose tissue GDF15 but contribute very little to the rise in circulating GDF15 concentrations seen in HFD fed mice. An implication of this finding is that white adipose tissue is likely to be a minor contributor to the rise in circulating GDF15 concentrations associated with weight gain. In keeping with this observation, plasma GDF15 concentrations were only weakly associated with white adipose tissue GDF15 mRNA in the human TwinsUK cohort referred to previously (b ¼ 0.08; P ¼ 0.02) (Supplemental Figure 4B).

Characterisation of liver specific GDF15 null mice
In order to evaluate the contribution of hepatocyte derived GDF15, we used an Albumin driven Cre line (Alb-Cre) to generate hepatocytespecific GDF15 null mice. Effective deletion of GDF15 was confirmed within Alb-GDF15 KO liver tissue where there was a 90% reduction in GDF15 mRNA expression without any changes in GDF15 mRNA in the other tissues examined ( Figure 5A). Furthermore, in this instance plasma GDF15 was significantly reduced (w50%) in HFD fed mice ( Figure 5B), establishing the liver as a major source of circulating GDF15 in the HFD fed state. Alb-GDF15 KO mice also displayed elevated circulating FGF21 levels upon HFD feeding, but FGF21 mRNA expression in the liver was unchanged ( Figure 5C,D). Alb-GDF15 KO mice were slightly heavier when fed a HFD compared to the WT control groups ( Figure 5E,F) though leptin concentrations were similar to those of WT littermates ( Figure 5H) and we were not able to detect significant differences in food intake or energy expenditure (Supplemental Figure. 5A and B). However, the Alb-GDF15 KO mice did also manifest larger livers ( Figure 5G), higher plasma lipids (TG, cholesterol) and liver transaminases (ALT, AST) ( Figure 5H). Histologically, Alb-GDF15 KO livers appeared to be similar to WT control mice (Supplemental Figure 5C), though total hepatic lipid content was significantly higher as the livers were heavier ( Figure 5I). mRNA analysis revealed no difference in hepatic expression of genes involved in lipid synthesis or fat oxidation (Supplemental Figure 5D). Metabolically, Alb-GDF15 KO mice showed minor impairments in glucose and insulin tolerance as well as higher fasting insulin concentrations and thus higher HOMA IR scores (Figure 5JeL).

Characterisation of GDF15/FGF21 double KO mice
Similarly to GDF15, circulating FGF21 has also been shown to be largely derived from the liver in obese HFD fed mice [60] raising the question of why two stress responsive hormones are secreted in this context of overnutrition. Furthermore, we had noted increased FGF21 concentrations in the plasma of GDF15 null mice suggesting that this might be attenuating the impact of GDF15 deficiency on HFD associated weight gain. In order to test this hypothesis we next studied mice lacking both GDF15 and FGF21 i.e. 'double knockouts'. The GDF15/ FGF21 double knockout (dKO) strain displayed normal viability and fertility and had no gross abnormalities. Loss of both hormones in dKO mice was confirmed by checking mRNA expression in the liver ( Figure 6A,B) and circulating concentrations ( Figure 6C,D). Interestingly, as seen for circulating FGF21, plasma GDF15 in FGF21KO mice was significantly higher than in wild-type littermates within 8 weeks of commencing a HFD diet and then further elevated after 16 weeks of HFD feeding ( Figure 6C). dKO mice were slightly heavier than their WT littermates, but their weight was not significantly different from GDF15 KO mice ( Figure 6E,F) and we did not detect significant differences in food intake or energy expenditure (Supplemental Figure 6A and B).
Similarly to the GDF15 KOs, the dKO mice displayed mildly impaired glucose and insulin tolerance (as reflected by AUC) (Figure 6G,H). However, higher fasting glucose, insulin and thus HOMA IR scores suggested that they were significantly more insulin resistant than the GDF15 null mice, at least in the fasting state ( Figure 6I). In contrast, FGF21 KO mice displayed relatively normal glucose tolerance and circulating insulin concentrations ( Figure 6I). Despite their body weight being similar to that of the single GDF15 KOs, the most prominent phenotype of the dKO mice was their increased liver size ( Figure 6J). This was associated with increased hepatic lipid content and elevated circulating triglycerides (Figure 6KeM and Supplemental Figure 6C). Furthermore, plasma cholesterol, NEFA, AST and ALT levels were also significantly higher in the dKOs compared to the other genotypes (Supplemental Figure 6D). Gene expression analysis revealed increases in genes encoding proteins involved in de novo lipogenesis (DNL) whereas genes involved in fat oxidation were expressed at similar levels to the other genotypes studied ( Figure 6N). This is likely to relate to the higher insulin concentrations in the dKO mice.

DISCUSSION
GDF15 was identified more than 20 years ago as a gene that was upregulated in 'activated macrophages' but has since been shown to be elevated in a range of human disease states [55,80,109]. Our original interest in understanding GDF15 biology in the context of weight gain Original Article  and metabolic disease related to three key observations: (i) GDF15 is elevated in human and rodent models of obesity [42e44]; (ii) transgenic overexpression or pharmacological treatment with GDF15, suppresses food intake and reduces body weight in mice [43,84e86]; and (iii) this effect is mediated exclusively via the hindbrain restricted expression of the GFRAL receptor [87,88]. Whilst significant progress has been made in understanding the mechanism by which GDF15 leads to suppression of food intake and modulation of body weight [84,87,88], the cellular source of GDF15 in this context remains unclear.
Here we first sought to determine (i) the source of circulating GDF15 in HFD fed mice and (ii) if selective deletion of this source replicates the more obese phenotype observed in global GDF15 KO mice on a HFD. We previously showed that expression of GDF15 mRNA is increased in white and brown adipose tissue and in the liver in mice fed a HFD for up to 18 weeks, whereas its expression in skeletal muscle in this context is very low and does not differ from that of chow fed mice [33]. We used RNA scope to reveal the cellular source of GDF15 in WAT to be almost exclusively macrophages and in the liver, to be the hepatocytes themselves. In brown fat, we see a weak signal in selected adipocytes as well as in stromal macrophages. In keeping with these murine observations, we observed a strong correlation between GDF15 mRNA and macrophage markers (CD68 and EMR1) in human white fat samples from two large independent cohorts. It is well established that the number of macrophages in white adipose tissue increases with weight gain so these findings could either simply relate to the higher number of macrophages present in the tissue and/or increased expression within individual macrophages. Interestingly, despite this evidence suggesting that macrophages account for the increase in GDF15 mRNA in WAT associated with weight gain in humans and mice, our data does not suggest that macrophages contribute a great deal to circulating GDF15 concentrations in this context. This is borne out by the very modest impact of both LysM Cre driven GDF15 deletion in macrophages and the reconstitution of hematopoietic lineage cells from GDF15-null mice in irradiated wild type mice, on serum GDF15 levels in HFD fed mice. Furthermore, this and the hepatocyte specific deletion of GDF15, indicate that Kuppfer cells (liver resident macrophages) are not a major source of GDF15. In humans, we also noted that plasma GDF15 levels correlate weakly with WAT GDF15 expression. When GDF15 was deleted in hepatocytes using the Alb-Cre promoter, we observed a more substantial impact on GDF15 concentrations in mice gaining weight on a HFD and an associated small increase in body weight, suggesting that the liver is likely to be the major source of GDF15 in this context. However, the data suggest that additional tissues probably also contribute small amounts towards circulating GDF15 concentrations in this paradigm. These data are also consistent with a recent human study which showed that plasma GDF15 levels correlated strongly with NAFLD progression [106]. We have previously shown that activation of the integrated stress response is likely to be mediating the increase in GDF15 mRNA expression in the liver, highlighting the 'stress' imposed on hepatocytes by overfeeding. Hepatocytes appear to be a major site of ectopic lipid accumulation when the capacity of white adipocytes to store surplus energy is overloaded, and these data suggest that this results in a stress response in the hepatocytes, one of the consequences of which is increased GDF15 secretion. In contrast to this stress response in hepatocytes, skeletal muscle does not appear to manifest a similar response in high fat fed mice. We are also not aware of activation of the ISR having been demonstrated in skeletal muscle in obese humans. In terms of the impact of the changes in circulating GDF15 concentrations on body weight, we saw very modest changes which broadly correlated with the changes or lack thereof in GDF15 concentrations. This is consistent with the rather subtle impact of global GDF15 deficiency on body weight, and, in a fashion reminiscent of what has been reported for FGF21, contrasts with the more striking impact of 'pharmacological' administration of GDF15 [66,76e78,88,110e112]. Interestingly we did observe higher FGF21 expression in the GDF15 null mice and in some of the tissue specific GDF15 knockout lines. In order to determine if this response might be mitigating the impact of GDF15 deficiency on body weight we proceeded to study mice deficient for both hormones. In this context, we also showed that circulating concentrations of GDF15 are higher in FGF21 deficient mice. Whilst body weight gain of the double KO (dKO) mice was slightly greater than that of wildtype littermates in response to HFD feeding, the weight of the dKO mice was no higher than that of GDF15 null mice ( Figure 6C). We attribute our inability to detect significant changes in food intake or energy expenditure in the dKO mice to the very modest changes in body weight i.e. we would only anticipate very small changes in food intake or energy expenditure in this context. Recent work by Katsumura et al. [113] clearly showed that more substantial increases in hepatic and serum FGF21 and GDF15 induced in response to inhibition of CNOT6L deadenylase reduced food intake and increased energy expenditure, in keeping with prior work on their respective mechanisms of action [61,112]. Intriguingly, despite the dKO mice having very modestly increased body weight compared to WT controls and similar body weight to GDF15 KOs, they did manifest significantly higher fasting glucose, insulin and HOMA IR levels (5Â fold) than wild-type mice and, both GDF15 and FGF21 'single knockouts'. However, the changes in glucose and insulin tolerance in these mice were very modest. We interpret these data as suggesting that loss of GDF15 and FGF21 primarily impacts the liver and thus hepatic rather than peripheral insulin sensitivity. In keeping with this suggestion, liver triglyceride concentrations were significantly higher in the dKO mice and this was accompanied by higher plasma triglycerides and liver transaminases. Some have speculated that GDF15 may directly, or more likely indirectly, impact hepatic fat metabolism [50,114e116] though exactly how it might do so remains to be elucidated and will require careful investigation. In this context, Kang et al. studied a model in which both GDF15 and FGF21 expression were induced in the liver in response to activation of the integrated stress response, though in this case deletion of a mitochondrial ribosome protein CRIF1 directly impaired mitochondrial function which has its own impacts on metabolism whilst also increasing serum concentrations of both GDF15 and FGF21. Their mechanistic studies led them to suggest that GDF15 was involved in regulating fat mass, hepatic lipid metabolism and energy expenditure whereas FGF21 was involved in regulating insulin sensitivity and glucose homeostasis (and in some instances EE and hepatic fat accumulation) [117,118].
Collectively the data indicate that similarly to the situation with FGF21, the liver is a major source of circulating GDF15 in the context of high fat feeding. The fact that Alb-Cre driven deletion of GDF15 does not entirely alleviate the HFD induced rise in GDF15 suggests that other tissues are also likely to be contributing smaller amounts. Importantly however, we have never seen an increase in GDF15 or FGF21 in skeletal muscle in the context of HF feeding despite other data suggesting that muscle can secrete both hormones in the face of mitochondrial stress for example [34,119,120]. GDF15 deficiency does appear to result in an increase in high fat feeding induced weight gain though the size of this effect is modest. We interpret this as reflecting the primary role of GDF15 as a stress responsive hormone which has evolved to reduce eating and promote rest in the context of more acute insults such as might occur after ingesting a toxin. FGF21 is also highly expressed in the liver in high fat fed mice and does appear to have an additive impact on alleviating fatty liver and preserving insulin sensitivity. Further work will be required to elucidate exactly how these effects are mediated. Despite the very modest impact of GDF15 and FGF21 'loss-of-function' on body weight, they remain of considerable interest as pharmacological 'gain-of-function' targets in relation to their impact on body weight and on hepatic steatosis.

AVAILABILITY OF DATA AND MATERIALS
Further 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).