FGF21 is not required for glucose homeostasis, ketosis or tumour suppression associated with ketogenic diets in mice
Ketogenic diets (KDs) have increasingly gained attention as effective means for weight loss and potential adjunctive treatment of cancer. The metabolic benefits of KDs are regularly ascribed to enhanced hepatic secretion of fibroblast growth factor 21 (FGF21) and its systemic effects on fatty-acid oxidation, energy expenditure (EE) and body weight. Ambiguous data from Fgf21-knockout animal strains and low FGF21 concentrations reported in humans with ketosis have nevertheless cast doubt regarding the endogenous function of FGF21. We here aimed to elucidate the causal role of FGF21 in mediating the therapeutic benefits of KDs on metabolism and cancer.
We established a dietary model of increased vs decreased FGF21 by feeding C57BL/6J mice with KDs, either depleted of protein or enriched with protein. We furthermore used wild-type and Fgf21-knockout mice that were subjected to the respective diets, and monitored energy and glucose homeostasis as well as tumour growth after transplantation of Lewis lung carcinoma cells.
Hepatic and circulating, but not adipose tissue, FGF21 levels were profoundly increased by protein starvation, independent of the state of ketosis. We demonstrate that endogenous FGF21 is not essential for the maintenance of normoglycaemia upon protein and carbohydrate starvation and is therefore not needed for the effects of KDs on EE. Furthermore, the tumour-suppressing effects of KDs were independent of FGF21 and, rather, driven by concomitant protein and carbohydrate starvation.
Our data indicate that the multiple systemic effects of KD exposure in mice, previously ascribed to increased FGF21 secretion, are rather a consequence of protein malnutrition.
KeywordsEnergy expenditure Fibroblast growth factor 21 Glucose homeostasis Ketogenic diets Protein starvation Tumour suppression
Fibroblast growth factor 21
Lewis lung carcinoma
Minimal essential medium
Non-essential amino acids
Ketogenic diets (KDs) are actively used for weight loss [1, 2, 3, 4] and anti-epilepsy therapy . More recently KDs have been studied intensively as potential adjunctive treatment for neurodegenerative diseases  and brain malignancies [7, 8, 9]. The metabolic benefits of KDs have been ascribed to an increased hepatic expression and higher circulating levels of fibroblast growth factor 21 (FGF21) [10, 11, 12]. By enhancing fatty-acid oxidation and ketogenesis in response to fasting and KD feeding in rodents, FGF21 has been suggested as a critical component of the metabolic adaptations to a state of fasting and ketosis [10, 11, 12, 13]. Profound variations in body weight, fat mass and glucose and lipid metabolism in various strains of Fgf21-knockout (Fgf21-KO) mice [12, 13, 14, 15, 16] and the wide range of serum FGF21 concentrations in humans challenged the relevance of endogenous FGF21 in metabolism [17, 18, 19]. Nevertheless, the pharmacological application of FGF21 holds promise as an effective therapeutic means for treating obesity and diabetes [20, 21, 22, 23, 24, 25] despite uncertainty remaining regarding its physiological role as endocrine mediator of ketogenic adaptations.
We recently showed that KDs need to be low in carbohydrate and protein to trigger weight loss and ketosis in rats . Accordingly, we hypothesised that the metabolic benefits of KDs on energy expenditure (EE), glucose and ketone-body metabolism result from protein starvation rather than directly from increased endogenous FGF21. We further hypothesised that protein starvation, but not FGF21, drives the therapeutic effects of KDs on tumour suppression.
To test these hypotheses, we studied male C57BL/6J mice fed a control low-fat diet (LFD) or KDs containing regular (RP-KD) or low (LP-KD) amounts of protein. We observed that circulating FGF21 levels were elevated only in LP-KD-fed mice, while increased ketone-body levels as well as enhanced expression and activity of renal gluconeogenic enzymes were found in both LP-KD- and RP-KD-fed mice. Acute effects of KDs on EE were similar in wild-type (WT) and global Fgf21-KO mice. Furthermore, FGF21 was dispensable for the maintenance of euglycaemia and for ketone-body synthesis upon KD feeding. Finally, we verified that the tumour-suppressing effects of KDs were independent of FGF21 and were mainly determined by carbohydrate and protein starvation.
Male 8-week-old C57BL/6J mice (Jackson Laboratory, Bar Harbor, MA, USA) had ad libitum access to LFD (13% energy from protein, 10% from fat, 77% from carbohydrates), an LFD-matched RP-KD (13% energy from protein, 87% from fat, 0% from carbohydrates) or an LFD-matched LP-KD content [10, 27, 28] (5% energy from protein, 95% from fat, 0% from carbohydrates). Diet compositions (Research Diets, New Brunswick, NJ, USA) are shown in electronic supplementary material (ESM) Table 1. Experimental cohort 1 received LFD (n = 9), RP-KD (n = 15) or LP-KD (n = 15) for 8 weeks. Cohort 2 received LFD (n = 10), RP-KD (n = 10) or LP-KD (n = 10) for 2 weeks. Cohorts 3, 4 and 5, consisting of Fgf21-KO mice [15, 29] and WT littermates, received LFD (n = 8), RP-KD (n = 8) or LP-KD (n = 8), respectively, for 2 weeks. Before experiments, groups were evenly distributed according to body weight. All experiments were done non-blinded. All procedures for animal use were approved by the University of Cincinnati Institutional Animal Care and Use Committee.
Body composition was analysed using nuclear magnetic resonance (Echo Medical Systems, Houston, TX, USA).
Oral glucose tolerance test
After 6 h of fasting, mice received an oral bolus of 1.5 g glucose per kg lean mass. Blood glucose was measured at 0, 15, 30, 60 and 120 min by using a glucose analyser (Accucheck; Roche, Indianapolis, IN, USA). Plasma insulin from blood taken at 0 and 15 min was assessed by ELISA (CrystalChem, Downers Grove, IL, USA).
Pyruvate tolerance test
WT and Fgf21-KO mice that had been fed for 2 weeks with LFD, RP-KD or LP-KD were fasted for 16 h and then injected intraperitoneally with 1.5 g sodium pyruvate (Sigma Aldrich, St Louis, MO, USA) per kg body weight. Tail blood was taken at 0, 15, 30, 60 and 120 min and glucose levels were determined with an Accucheck glucose analyser.
The EE, locomotor activity, RQ and food and water intake were measured by combined indirect calorimetry (PhenoMaster; TSE Systems, Bad Homburg, Germany) as described previously . Details are given in ESM Methods.
Tumour cell injection
WT and Fgf21-KO mice received subcutaneous injections of 4 × 105 cells from the murine Lewis lung cancer cell line (LLC1; ATCC, Manassas, VA, USA), dissolved in PBS. Mice were subsequently switched from chow to LFD (n = 8), RP-KD (n = 8) or LP-KD (n = 8). Tumours were excised and weighed 14 days after tumour cell injection. One LP-KD fed WT mouse died before the end of the experiment and was excluded from the analysis.
Trunk blood from ad libitum fed mice was collected for plasma preparation. Plasma FGF21 was assessed by FGF21-ELISA (Millipore, Billerica, MA, USA). 3-Hydroxybutyrate (3-HB) was quantified using the colorimetric Autokit 3-HB (Wako, Richmond, VA, USA).
Hepatic and renal phosphoenolpyruvate carboxykinase activity assay
Cell culture experiments
HepG2 cells (ATCC HB-8065) were seeded in six-well plates and cultured in minimal essential medium (MEM, M5650; Sigma Aldrich) supplemented with 10% (vol./vol.) FBS until reaching confluence. Cells were subsequently washed with PBS and incubated for 6 h in serum-free MEM with or without non-essential amino acids (M5650 and M2279; Sigma Aldrich).
Gene expression analyses
Quantitative gene expression analyses from 7–14 random samples per group were performed as described [33, 34] using gene-specific TaqMan probes (ESM Table 2). The relative expression of each gene was normalised to that of Actb or Pgk1 .
Data are reported as means ± SEM. Statistical differences were assessed by one-way or two-way ANOVA with Tukey’s post hoc tests. EE was analysed by ANCOVA using fat and lean mass as covariates and Bonferroni post hoc tests. A p value ≤0.05 was regarded as statistically significant. All analyses were performed using Prism 6.0 (GraphPad, La Jolla, CA, USA) or SPSS 20 (IBM, Armonk, NY, USA).
Body weight changes in response to KDs depend on the relative amounts of dietary fat and protein
Exposure to protein-starved KD results in muscle wasting in mice
Liver but not adipose tissue FGF21 expression upon ketosis depends on the amount of dietary protein
Short-term dietary effects of KDs on EE are independent of endogenous FGF21
EE was similar in animals fed chow-diet (ESM Table 3). Switching the diet to LFD, RP-KD or LP-KD revealed a significant effect on dark phase EE in WT (Fig. 4k) and Fgf21-KO (Fig. 4l) mice after ANCOVA with lean and fat mass as covariates (ESM Fig. 1). The covariate lean mass, but not fat mass, was significantly related to EE (ESM Table 4). Bonferroni post hoc tests showed a significant decrease in dark phase EE in WT mice fed with LP-KD compared with LFD (p = 0.031), and a trend for decreased total EE in WT mice fed with LP-KD compared with mice fed with LFD (p = 0.053). Similarly, Fgf21-KO mice fed with LP-KD displayed significantly decreased dark phase EE (p = 0.008) when compared with LFD-fed mice. We did not observe any significant interaction effects between genotypes and dietary exposure (Genotype × Diet for EE total: F = 0.004, p = 0.996; EE dark phase: F = 0.113, p = 0.894; EE light phase: F = 0.078, p = 0.925), demonstrating that WT and Fgf21-KO mice respond similarly to the exposure to LFD, RP-KD or LP-KD, respectively. Interestingly, after 7 weeks of feeding the dietary impact on EE was lost (Fig. 4m); EE continued to be strongly related to lean mass, but LP-KD or RP-KD exposure no longer had a significant effect (ESM Table 4).
FGF21 is not required for maintaining glucose levels upon ketosis
Both RP-KD and LP-KD-fed mice were able to maintain normoglycaemia despite the absence of carbohydrates (Fig. 1g). Based on this finding and our findings of muscle catabolism and higher liver Fgf21 expression in LP-KD-fed mice but not RP-KD-fed mice, we hypothesised that FGF21 could be an important factor for muscle protein catabolism to ultimately provide amino acids as substrates for gluconeogenesis.
Exposure to protein-starved KD suppresses tumour growth independently of FGF21
KDs have garnered attention as an effective treatment for obesity by causing weight loss and improving glucose tolerance and plasma lipid profiles [1, 2, 3, 4], potentially via release of FGF21 [10, 12]. Through a series of experiments, we reveal that FGF21 is not an essential mediator of the metabolic alterations seen with KD feeding in mice. Our data suggest that the overall protein content in a KD is an important contributor to its effects on body weight, energy and glucose homeostasis.
We reveal that body weight and lean mass are potently decreased by LP-KD feeding. In contrast, a gain in fat mass in RP-KD-fed mice led to significantly higher body weights compared with LFD- and LP-KD-fed mice. Both LP-KD and RP-KD feeding induced ketosis, as revealed by increased 3-HB levels in RP-KD- and LP-KD-fed mice, compared with LFD controls. However, circulating FGF21 levels were only increased in LP-KD-fed mice compared with RP-KD- or LFD-fed mice. Accordingly, our data corroborate previous findings that hepatic and plasma FGF21 levels are indicative of protein starvation [26, 37, 38, 39] but not the state of ketosis . In contrast to the protein-dependent regulation of hepatic Fgf21 expression and FGF21 release, RP-KD- and LP-KD-fed mice displayed a similar upregulation of Fgf21 in the gonadal white adipose tissue, suggesting alternative regulation of FGF21 in adipose tissue compared with liver.
The mechanisms underlying the weight-lowering effects of KDs remain a matter of debate. In our model, we observed a similar energy intake in all dietary groups but severely diminished food efficiency in LP-KD-fed mice. The inability of LP-KD-fed mice to translate ingested energy into weight gain may either be due to malabsorption, urinary loss of ketones or increased EE. Indeed, recent evidence suggests that KD feeding increases EE [27, 28]. FGF21 may play a role in this thermogenic effect of KDs, as also suggested by increased EE in obese rats or mice after central or intraperitoneal infusion, respectively, of pharmacological doses of FGF21 [40, 41]. Further, hypothalamic β-klotho, the obligate co-receptor for the FGF21 receptor fibroblast growth factor receptor 1, is required for the central effects of FGF21 on EE in transgenic Fgf21 mice .
To investigate the impact of endogenous FGF21 on EE, we performed indirect calorimetry measurements in WT and Fgf21-KO mice following an acute challenge with LFD, RP-KD or LP-KD. The acute feeding paradigm assured that lean mass as predictive covariate for EE was similar for all groups. LP-KD- but not RP-KD-fed mice showed a significant decrease in EE during the dark phase compared with LFD-fed mice. This stands in contrast to previous findings by Leager et al  showing increased EE in mice fed a protein-restricted regular diet. Further, in contrast to Laeger et al, who showed that dietary effects on EE were no longer present in an alternative strain of Fgf21-KO mice, we observed a similar decrease in EE in LP-KD-fed KO and WT mice, suggesting that this effect was FGF21-independent . To assess whether these differences were due to a shorter exposure of our mice, we compared C57BL/6 WT mice that were fed chronically with RP-KD and LP-KD and displayed low and high circulating FGF21 levels, respectively. ANCOVA analyses revealed that effects on EE were strongly related to changes in lean mass, but not to LP-KD or RP-KD exposure. Together, our findings do not support a functional role for endogenous FGF21 in EE, at least not in a model of KD exposure.
Current studies investigating the effect of FGF21 on carbohydrate metabolism are controversial. FGF21 has been implicated in the regulation of gluconeogenesis during the progression from fasting to starvation . Administration of recombinant FGF21 induced hepatic expression of key regulators of gluconeogenesis such as Pgc1α and Pck1 but not G6pc . In contrast, therapeutic administration of FGF21 had a potent glucose-lowering effect in ob/ob and db/db mice . Glucose output was suppressed in H4IIE hepatoma cells exposed to FGF21 .
In our experimental set-up neither the expression of hepatic gluconeogenic enzymes G6pc, Pck1 and Pgc1α nor PEPCK activity were altered by 8 weeks of KD feeding. This is consistent with findings from rats fed with KD for 4 weeks  and WT and Fgf21-KO mice fed for 6 days with KD . In contrast, G6pc and Pck1 expression, as well as PEPCK activity, was elevated in the kidney cortex of our KD-fed mice. Overall, our data suggest that enhanced renal gluconeogenesis is an important contributor to systemic glucose control in a state of ketosis. However, similar changes in blood glucose levels during 14 days of KD feeding and similar glucose excursions following an acute pyruvate challenge in KD-fed WT and Fgf21-KO mice do not support an essential role for endogenous FGF21 in the overall maintenance of normoglycaemia during KD feeding.
We finally aimed to investigate whether FGF21 might be involved in the anti-tumorigenic effects of KD feeding, and implanted LLC1 cells into WT and Fgf21-KO mice fed with LP-KD, RP-KD and LFD. We observed significantly diminished tumour growth in both WT and KO mice after LP-KD feeding, suggesting that concomitant dietary protein and carbohydrate deprivation but not FGF21 are necessary to reduce tumour growth. This finding may have important clinical implications for cancer patients that are subjected to adjuvant KD therapy. The prospect of anti-tumorigenic efficacy being increased by adjuvant LP-KD may be promising for cancer patients and warrants future clinical studies, but adverse clinical symptoms of protein malnutrition should be monitored closely.
In conclusion, our studies demonstrate that endogenous FGF21 is a biomarker for protein starvation but is dispensable for the maintenance of normoglycaemia, ketone-body synthesis and tumour suppression in mice fed a KD. The metabolic alterations in EE and glucose and ketone-body metabolism caused by KD in mice were rather a consequence of concomitant carbohydrate and protein starvation. Future studies will need to address remaining questions such as the correct protein content and composition of KD to induce similar states of ketosis in mice and humans, or the often-discrepant findings obtained by endogenous FGF21 induction vs pharmacological doses of FGF21. Nevertheless, we believe that our physiological model of increased or decreased FGF21 in response to protein-starved or regular-protein KD provides a useful tool with which to investigate the impact of endogenous FGF21 on the therapeutic effects of KD on metabolism and cancer.
The authors thank N. Itho (Kyoto University Graduate School of Pharmaceutical Sciences, Sakyo, Kyoto, Japan) for providing the Fgf21-KO mice.
This work was supported in part by NIH grant U01 CA141464-02. KS was supported by a fellowship from the German Research Foundation (STE 1466/4-1). KMH was supported by ADA grant 1-13-JF-21.
Duality of interest
RJS has received research support from Novo Nordisk, Ethicon Surgical Care, Eisai and Boehringer Ingelheim and has served as a paid consultant for Novo Nordisk, Ethicon Surgical Care, Boehringer Ingelheim, Sanofi, Novartis and Takeda. RJS has equity in Zafgen and Endobetix. All other authors declare that there is no duality of interest associated with their contribution to this manuscript.
KS, FZ, KMH, CN, PK, MB, SY, AA, ML, PJFM, TDM, PTP and RJS made substantial contributions to conception and design, acquisition of data, or analysis and interpretation of data. KS and PTP drafted the manuscript. FZ, KMH CN, PK, MB, SY, AA, ML, PJFM, TDM, and RJS critically revised the article for intellectual content. All authors approved the final manuscript. KS is responsible for the integrity of the work as whole.
- 9.Seyfried TN, Kiebish MA, Marsh J, Shelton LM, Huysentruyt LC, Mukherjee P (1807) Metabolic management of brain cancer. Biochim Biophys Acta 2011:577–594Google Scholar