Abstract

Objective

   Fibroblast growth factor 21 (FGF21) was recently discovered as
   stress-induced myokine during mitochondrial disease and proposed as key
   metabolic mediator of the integrated stress response (ISR) presumably
   causing systemic metabolic improvements. Curiously, the precise
   cell-non-autonomous and cell-autonomous relevance of endogenous FGF21
   action remained poorly understood.

Methods

   We made use of the established UCP1 transgenic (TG) mouse, a model of
   metabolic perturbations made by a specific decrease in muscle
   mitochondrial efficiency through increased respiratory uncoupling and
   robust metabolic adaptation and muscle ISR-driven FGF21 induction. In a
   cross of TG with Fgf21-knockout (FGF21^−/−) mice, we determined the
   functional role of FGF21 as a muscle stress-induced myokine under low
   and high fat feeding conditions.

Results

   Here we uncovered that FGF21 signaling is dispensable for metabolic
   improvements evoked by compromised mitochondrial function in skeletal
   muscle. Strikingly, genetic ablation of FGF21 fully counteracted the
   cell-non-autonomous metabolic remodeling and browning of subcutaneous
   white adipose tissue (WAT), together with the reduction of circulating
   triglycerides and cholesterol. Brown adipose tissue activity was
   similar in all groups. Remarkably, we found that FGF21 played a
   negligible role in muscle mitochondrial stress-related improved obesity
   resistance, glycemic control and hepatic lipid homeostasis.
   Furthermore, the protective cell-autonomous muscle mitohormesis and
   metabolic stress adaptation, including an increased muscle proteostasis
   via mitochondrial unfolded protein response (UPR^mt) and amino acid
   biosynthetic pathways did not require the presence of FGF21.

Conclusions

   Here we demonstrate that although FGF21 drives WAT remodeling, the
   adaptive pseudo-starvation response under elevated muscle mitochondrial
   stress conditions operates independently of both WAT browning and FGF21
   action. Thus, our findings challenge FGF21 as key metabolic mediator of
   the mitochondrial stress adaptation and powerful therapeutic target
   during muscle mitochondrial disease.

   Keywords: Browning, FGF21, GDF15, Myokine, Mitochondrial disease,
   Muscle mitohormesis

Graphical abstract

   graphic file with name fx1.jpg
   [49]Open in a new tab

Highlights

     * •
       Muscle mitochondrial stress-induced browning of white adipose
       tissue fully requires FGF21.
     * •
       Negligible role of myokine FGF21 on whole body metabolic
       adaptations.
     * •
       Muscle mitohormesis and starvation-like response operates
       independently of FGF21 action.

1. Introduction

   During the last decade, the pleotropic hormone-like circulating protein
   fibroblast growth factor 21 (FGF21) gained momentum as a major
   metabolic regulator in metabolism and disease [50][1], [51][2],
   [52][3], [53][4]. Notably, FGF21 was found to be expressed not only in
   liver [54][5] and white adipose tissue (WAT) [55][6] but also in
   skeletal muscle [56][7]. A first study linking muscle FGF21 secretion
   to mitochondrial dysfunction and pseudo-starvation described the
   Deletor mouse, a model of late-onset mitochondrial myopathy [57][8].
   Subsequently, significant elevations of circulating FGF21 levels were
   demonstrated in muscle-specific mouse models of autophagy deficiency
   [58][9] and have been associated with a mild decrease in mitochondrial
   efficiency [59][10] as well as activation of mTOR complex1 (mTORC1)
   signaling [60][11]. In humans, muscle FGF21 expression is induced by
   hyperinsulinemia [61][12] but not by exercise [62][13]. In patients,
   the constitutive secretion of FGF21 is accepted as the common
   denominator of muscle stress conditions and mitochondrial disease
   [63][14], [64][15]. Based on these important clinical findings, a role
   of endogenous FGF21 for systemic energy metabolism has been proposed
   [65][16]. So far, the major targets and metabolic functions of
   stress-induced muscle FGF21 are poorly understood and await
   experimental establishment of causality.

   Liver and adipose tissue are considered as the main targets of the
   myokine FGF21, but not skeletal muscle, due to lack of the FGF-receptor
   cofactor β-klotho [66][10], [67][17]. However, an endocrine axis of
   FGF21 is established as it regulates adiponectin in adipocytes, thus
   coupling FGF21 action in WAT to metabolic effects in liver and muscle
   [68][18]. This strongly supports bidirectional muscle-WAT crosstalk.
   Moreover, as an adipokine, FGF21 promotes brown adipose tissue
   (BAT)-like structures within WAT depots in response to cold exposure
   [69][19]. The link between ‘browning’ and the beneficial
   pharmacological effects of FGF21 is currently in the focus of metabolic
   disease research [70][20], [71][21], [72][22]. Similarly, a role of
   myokine FGF21 action in WAT remodeling is speculated [73][9],
   suggesting that muscle FGF21 signals to peripheral tissues to enhance
   substrate utilization/mobilization, although the biological function is
   debated. On one hand, FGF21 might sequester energy substrates into
   adipose tissue, thus bypassing the muscle to avoid overload and
   metabolic stress [74][23]. Contrarily, muscle FGF21 secretion could be
   a survival signal to mobilize and supply adequate fuel to the muscle
   [75][24]. Notably, oral administration of nicotinamide riboside
   (vitamin B3) effectively delayed disease progression in Deletor mice, a
   finding which was linked to even higher levels of circulating FGF21 and
   protective mitochondrial unfolded protein response (UPR^mt) in skeletal
   muscle [76][25]. Nevertheless, the ultimate functional relevance of
   muscle FGF21 as a stress hormone and how it controls systemic,
   metabolic adaptation has not been delineated yet. Very recently,
   studying the preprogeroid polymerase gamma mutator (POLG) mouse as a
   model of mitochondrial disease and premature aging, it was proposed
   that hepatic FGF21 and high fat diet (HFD) intake cooperatively mediate
   the rescue of mitochondrial stress [77][26]. Given the (not only
   translational) potential of these findings and in order to understand
   the physiological function of endogenous FGF21, we sought to further
   explore the specific pathophysiological relevance of skeletal muscle
   FGF21 in the context of cell-non-autonomous and cell-autonomous muscle
   mitochondrial stress adaptation and systemic energy metabolism.

   The above described models of skeletal muscle autophagy deficiency
   [78][9], [79][27] and mitochondrial dysfunction [80][8], [81][26] show
   severe muscle disease phenotypes. Thus, we made use of UCP1 transgenic
   (TG) mice with an ectopic expression of the uncoupling protein 1 (UCP1)
   in skeletal muscle [82][28], an established model of metabolic
   perturbations by muscle-specific decrease in mitochondrial efficiency
   through increased respiratory uncoupling [83][29] linked to robust
   metabolic adaptations [84][30], [85][31], obesity resistance and
   increased longevity [86][32], [87][33]. In addition to the integrated
   stress response (ISR) induced FGF21 [88][10], TG mice share a common
   adaptive metabolic stress response with other models of mitochondrial
   dysfunction, including muscle proteostasis and amino acid biosynthetic
   pathways [89][34]. Curiously, while the idea of a systematic and
   protective role of endogenous FGF21 is widely accepted in the field,
   detailed, in-vivo, proof-of-principal experiments using genetic
   ablation of FGF21 are still lacking. Thus, in a cross of TG with
   Fgf21-knockout (FGF21^−/−) mice [90][35] we aimed to investigate the
   functional role of FGF21 as a muscle stress-induced myokine under low
   and high fat feeding conditions.

2. Material and methods

2.1. Animals

   Experiments were performed in adult male mice that were group-housed
   and random-caged with ad libitum access to food and water at 23 °C and
   a 12:12 h dark–light cycle. UCP1 transgenic (TG) mice [91][28] were
   crossed with Fgf21-knockout (FGF21^−/−) mice [92][35] to generate four
   genotypes: wild-type (WT), FGF21^−/−, TG and TG/FGF21^−/− mice. Unless
   otherwise stated, all data are from male mice at 40 wks of age fed a
   low-fat (LFD) or a high-fat diet (HFD) [93][36] for 24 weeks starting
   at wk 16 of age. Animal experiments were approved by the ethics
   committee of the Ministry of Agriculture and Environment (State
   Brandenburg, Germany, permission number GZ V3-2347-16-2013).

2.2. Phenotypical assessments

   For representative computed tomography of mice the 3rd generation
   computed tomography scanner, LaTheta LCT-200 (Hitachi-Aloka, Tokyo,
   Japan) was used [94][37]. Body composition was determined by
   quantitative magnetic resonance (QMR, EchoMRI 2012 Body Composition
   Analyzer, Houston, USA). Bone mineral density was determined by
   Dual-energy X-ray absorptiometry (DEXA, Lunar Piximus). Physical
   locomotor activity was assessed by measurement of spontaneous physical
   activity using infrared detections (TSE Systems GmbH, Germany). For the
   oral glucose tolerance test, 2 mg glucose per g body weight was applied
   2 h after food withdrawal. Blood glucose levels were measured before,
   15, 30, 60 and 120 min after glucose application using a glucose sensor
   (Bayer, Germany). Insulin levels were measured before, 15 and 30 min
   after application by an ultra-sensitive ELISA assay (DRG Instruments
   GmbH, Germany).

2.3. Microarray analysis

   Agilent 8 × 60 k whole-genome mouse microarrays (G4852A, Agilent
   Technologies Inc., Santa Clara, USA) were used. Amplification,
   labeling, and microarray hybridization of individual sWAT samples of
   40 wks old male WT, FGF21^−/−, TG and TG/FGF21^−/− mice (n = 10 per
   group) fed a LFD were performed following the Agilent's procedure for
   whole-genome microarray technology with a few modifications together
   with normalization and statistical analysis as described previously
   [95][38]. Briefly, One-way ANOVA with FDR-adjustment was used based on
   all 4 groups (n = 10) and p < 0.05 was used as threshold for
   principal component analysis (PCA). This subset of 4983 probes was
   ranked on fold change of TG over TG/FGF21^−/−, collapsed into unique
   genes and only known protein-encoding genes were selected to construct
   the heat map of the top30 genes (thus excluding Riken cDNA genes and
   lincRNAs). All microarray data follow minimal information about
   microarray experiment compliance (MIAME) and have been deposited into
   the gene expression omnibus (GEO) under accession [96]GSE71749.

2.4. RNA isolation and gene expression analysis

   RNA isolation and quantitative real-time PCR was performed as described
   before [97][31]. Quadriceps was used for analysis of SM gene expression
   unless otherwise stated. Multi-tissue gene expression was calculated as
   dCT using Hprt or beta-2 microglobulin (B2m) for normalization. Atf4:
   forward primer 5′-GGAATGGCCGGCTATGG-3′, reverse primer
   5′-TCCCGGAAAAGGCATCCT-3′; Gdf15: forward primer 5′ GAGCTAGGGGTCGCTTC
   3′, reverse primer 5′-GGGACCCCAATCTCACCT-3′; Psat1: forward primer
   5′-AGTGGAGCGCCAGAATAGAA-3′, reverse primer 5′-CTTCGGTTGTGACAGCGTTA-3′;
   Slc3a2: forward primer 5′-CAAAGTGCCAAGAAAAAGAGC-3′, reverse primer
   5′-CTGAGCAGGGAGGAACCAC-3′; Mthfd2: forward primer
   5′-CATGGGGCGTGTGGGAGATAAT-3′, reverse primer 5′-CCGGGCCGTTCGTGAGC-3′;
   Fgf21: forward primer 5′-GCTGCTGGAGGACGGTTACA-3′, reverse primer
   5′-CACAGGTCCCCAGGATGTTG-3′; Atf5: forward primer
   5′-CTACCCCTCCATTCCACTTTCC-3′, reverse 5′-TTCTTGACTGGCTTCTCACTTGTG-3′;
   Cd36: forward primer 5′-CCAAGCTATTGCGACATGAT-3′, reverse primer
   5′-ACAGCGTAGATAGACCTGCAAA-3′.

2.5. Analysis of metabolites and circulating factors

   FGF21 values were determined by ELISA (Mouse/Rat FGF-21 Quantikine
   ELISA Kit; R&D Systems) following the manufacturer's instructions. The
   analysis of triglycerides, free fatty acids and total cholesterol in
   plasma was performed using an automated analyzer (Cobas Mira S,
   Hoffmann-La Roche, Basel, Switzerland) with the appropriate
   commercially available reagent kits (triglycerides, cholesterol CP,
   ABX, Montpellier, France; and NEFA; HR, Wako, Neuss, Germany).
   Metabolite concentrations were determined using a targeted metabolomic
   approach with the Absolute IDQ^®p180 kit (BIOCRATES Life Science AG,
   Innsbruck, Austria). Sample preparation and measurements were performed
   as described in the Biocrates user's manual UM-P180- Waters-2. Sample
   separation and analyte detection were carried out with an Acquity UPLC
   system connected to a Xevo TQ triple quadrupole mass spectrometer (all
   from Waters, Eschborn, Germany).

2.6. Enzyme activity assays

   Triglyceride concentration in liver was analyzed as described
   previously [98][31]. Cytochrome c oxidase (COX) activity in sWAT & BAT
   and citrate synthase (CS) activity in muscle & BAT [99][10], together
   with total muscle glutathione peroxidase (GPx) activity determination
   was performed as described previously [100][34].

2.7. Histology

   Liver, quadriceps skeletal muscle, sWAT and eWAT were fixed in 4%
   formaldehyde, embedded in paraffin and cut into 2 μm slices.
   Hematoxylin-eosin (H&E) staining (Roth, Fluka) was performed to
   visualize nuclear and cytoplasmic sections within the cell. Freshly
   dissected femora were collected in ice-cold PBS. Immediately after
   cleaning bones from surrounding soft tissue, bones were fixed and
   decalcified in Richard-Allan Scientific™ Cal-Rite™ solution (Thermo
   Scientific) for 48 h followed by embedding in paraffin. For phenotypic
   analysis at least three sections of each specimen were made at 6 μm
   intervals. Sections were stained with H&E (Roth, Fluka).

2.8. Western blotting

   Protein was prepared from frozen liver, skeletal muscle (quadriceps)
   and sWAT. Protein isolation, immunoblotting and detection were
   performed as previously described [101][10], [102][39]. Immunoblots
   were performed using following antibodies: oxidative phosphorylation
   complexes (OXPHOS, MitoSciences/Abcam, #MS604), PHGDH (Proteintech,
   #14719-1-AP), UCP1 (Abcam, #ab23841), PPARGC1α + β (Abcam, #ab72230),
   alpha-Tubulin (αTubulin, Sigma Aldrich, #T6074), phosphor-AMPK^Thr172
   (pAMPK, Cell Signaling, #2531), total-AMPK (tAMPK, Cell Signaling,
   #2603), CD36/FAT (R&D Systems, #MAB2519), CLPP (Proteintech,
   #15698-1-AP), MTHFD2 (Santa Cruz Biotechnology, #sc-74985), HSP25 (Enzo
   life sciences, #ADI-SPA-801), TXN2 (Cell Signaling, #13322).

2.9. FACS analysis

   Brown and inguinal fat pads, representing iBAT and sWAT, respectively,
   were cut in small pieces and incubated with 2 mg/mL collagenase II for
   60 min. The cell suspension was filtered through a 100 μm nylon mesh,
   and the stromal vascular fraction (SVF) was isolated by centrifugation
   at 1200 rpm for 10 min. The SVF was treated with ACK buffer for 3 min
   and filtered through a 40 μm nylon mesh. For FACS analysis,
   erythrocyte-free SVF cells were incubated with a mix of antibodies
   against different surface markers and sorted using an Aria flow
   cytometer (BD Biosciences). Briefly, dead cells were removed using
   propidium iodide staining. Cells negative for CD45/CD31, and positive
   for SCA1 were considered as APCs [103][40].

2.10. Statistics

   Statistical analyses were performed using Stat Graph Prism (6.0)
   (GraphPad, San Diego, CA). For microarray bioinformatics, see section
   above. Data are reported as mean + SEM. After testing for normal
   distribution (D'Agostino & Pearson omnibus normality test), either
   One-way ANOVA was used to evaluate differences between genotypes with
   Bonferroni's multiple comparisons test or Kruskal–Wallis and Dunn's
   multiple comparison test was used as non-parametric test. Statistical
   significance was assumed at p < 0.05.

3. Results

3.1. Phenotypic characterization of TG/FGF21^−/− mice as a model of muscle
mitochondrial stress in the absence of endogenous FGF21

   We compared male mice of the following four genotypes: wild-type (WT),
   FGF21^−/−, TG, and TG/FGF21^−/− mice. Phenotype data from mice fed LFD
   or HFD for 24 weeks (wks) from 16 wks of age confirmed that TG mice
   with mild muscle mitochondrial stress were smaller and showed reduced
   bone mass ([104]Figure S1A and B), although they did not show any
   conspicuous anatomical changes ([105]Figure 1A), compared to control
   littermates (WT and FGF21^−/− mice, [106]Figure S1A). Moreover, TG mice
   were leaner and resistant to diet-induced obesity compared to control
   mice ([107]Figure 1B,C + [108]Figure S1C and D). Irrespective of the
   diet, TG mice showed increased plasma FGF21 levels ([109]Figure 1D)
   while physical activity levels ([110]Figure 1E) and body temperature
   ([111]Figure S1E) were not affected. These results are in accordance
   with the pharmacological effects of FGF21 treatment or with
   FGF21-overexpression promoting distinct metabolic responses [112][5],
   [113][41], [114][42]. Unexpectedly, despite the loss of circulating
   FGF21 ([115]Figure 1D), the phenotype of TG/FGF21^−/− mice was almost
   identical to TG mice. Further, in contrast to prior expectations
   [116][9], [117][26], endogenous FGF21 played only a minor role in
   regulating body weight gain and fat depot mass under LFD or HFD
   conditions ([118]Figure 1A–D + [119]Figure S1). Strikingly, we obtained
   similar results from mice fed LFD or HFD for 16 wks starting from 8 wks
   of age ([120]Figure S2A–G). Furthermore, absolute energy expenditure
   was reduced in both TG groups, demonstrating that FGF21 is expendable
   ([121]Figure S1F). Additionally, energy intake and energy expenditure
   normalized to body weight ([122]Figure 1F,G + [123]Figure S1G and H) as
   well as respiratory quotient ([124]Figure 1H) revealed no effect of
   FGF21 as a myokine.

Figure 1.

   [125]Figure 1
   [126]Open in a new tab

   Muscle mitochondrial stress-driven whole body metabolic adaptations do
   not require FGF21 action. Phenotype data of 40 weeks (wks) old male WT,
   FGF21^−/−, TG, and TG/FGF21^−/− mice fed low-fat (LFD) or high-fat diet
   (HFD) for 24 wks. (A) Representative X-ray Computed Tomography
   (CT)-scan (left panel) of 10 wks old and body length (right panel) of
   40 wks old mice (n = 9–10). (B) Body weight development and (C) fat
   mass determined by quantitative magnetic resonance at wk 40 (n = 9–11).
   (D) Plasma FGF21 concentration at wk 40 (n = 5–11). (E) Physical
   activity level, (F) weight specific energy intake, (G) daily energy
   expenditure (kJ/d) as a function of body mass, and (H) mean 24 h
   respiratory quotient (VCO[2]/VO[2]) at wk 36 (n = 9–11). Except of (G)
   all data represent mean + SEM; means with different letters are
   significantly different; #, not detectable.

3.2. Muscle mitochondrial stress-induced WAT browning is fully
FGF21-dependent

   Next, we investigated adipose tissue as the major target tissues of
   FGF21 action [127][43]. Independent of FGF21, white fat depot weights
   were reduced in TG mice and TG/FGF21^−/− mice under LFD and
   particularly HFD conditions as compared to control littermates at
   24 wks and 40 wks of age ([128]Figure S1D + [129]Figure S2G). TG mice
   on LFD displayed multiple clusters of multilocular adipocytes in
   subcutaneous WAT (sWAT) depot ([130]Figure 2A) whereas the morphology
   of epididymal WAT (eWAT) was not affected ([131]Figure S3A). In mice
   and humans, a link between FGF21 and BAT activity has been shown
   [132][22], [133][44]. However, enzyme activity of cytochrome c oxidase
   (COX) and citrate synthase (CS) measured in interscapular BAT (iBAT),
   as well as adipogenic progenitor cell population analysis of iBAT
   stromal vascular fraction using FACS analysis, revealed little
   differences between the groups ([134]Figure S2B–E), excluding iBAT as a
   major target of muscle FGF21. In accordance with the morphological
   appearance of sWAT ([135]Figure 2A), biochemical analysis showed a
   robust increase of COX activity on LFD in sWAT of TG mice only
   ([136]Figure 2B) which was largely attenuated on HFD. The molecular
   extent of sWAT browning was further confirmed by protein induction of
   PPARGC1α/β, mitochondrial OxPhos complexes and UCP1. These effects were
   completely absent in TG/FGF21^−/−mice ([137]Figure 2C). Intriguingly,
   FGF21-dependent sWAT browning in TG mice was also strongly suppressed
   by HFD.

Figure 2.

   [138]Figure 2
   [139]Open in a new tab

   Muscle mitochondrial stress-induced WAT browning is fully
   FGF21-dependent. (A–C) Subcutaneous WAT (sWAT) of 40 weeks (wks) old
   male WT, FGF21^−/−, TG, and TG/FGF21^−/− mice fed low-fat (LFD) or
   high-fat diet (HFD) for 24 wks (A) sWAT morphology (H&E); bars
   represent 50 μm. (B) Cytochrome c oxidase (COX) activity per mg sWAT
   tissue (n = 9–11). (C) Representative immunoblots of UCP1, PPARGC1α/β
   and mitochondrial OxPhos complexes (CIII, Complex 3 core protein 2;
   CIV, Complex 4 subunit I; ATP-S, ATP-Synthase alpha subunit); α-Tubulin
   was used as loading control. (D) FACS analysis and (E) quantification
   of frequency of SCA1^+/Lin^−(CD45/CD31) progenitor cells from sWAT
   stromal vascular fraction (SVF) of young male mice at 8 wks of age
   (n = 4). (F–H) Transcriptomic profiling of FGF21-dependent metabolic
   remodeling in sWAT from 40 wk old males mice fed LFD (n = 10 per
   group). (F) Heat map of gene expression levels of the top 30 most
   significant regulated genes using whole genome microarray analysis
   (ANOVA FDR p < 0.0005 followed by FC TG over TG/FGF21^−/−). (G)
   Principal component analysis (PCA) showing the distinct profile of TG
   mice compared to the all other groups using One-way ANOVA-FDR
   (p < 0.05; 4983 probes). (H) Pathway enrichment analysis using MetaCore
   (FC > 1.15, FDR-adjusted p < 0.05); FAO, fatty acid beta-oxidation;
   T2D, Type 2 diabetes. All data represent mean + SEM; means with
   different letters are significantly different.

   To examine whether the FGF21-induced browning under LFD conditions was
   due to changes in precursor cell populations, we performed FACS
   analysis in sWAT stromal vascular fraction of young male mice.
   Quantification of adipogenic SCA1^+/Lin^−(CD45/CD31) progenitor cells
   [140][40] showed no differences between the groups ([141]Figure 2D,E).
   To identify key mechanisms of myokine-induced adipose tissue
   remodeling, we next performed whole genome gene expression analysis in
   sWAT of male mice on LFD. Gene clustering and principal component
   analysis (PCA) showed that gene expression profile ([142]Figure 2F,G)
   of TG sWAT was distinct from the other three groups. A heat map of the
   30 most upregulated genes in sWAT of TG mice includes BAT-markers Ucp1,
   Acot11, Cidea, Cox7a1, Cpt1a, Elovl3, Fabp3, Ppargc1b and Ppara
   ([143]Figure 2F, full gene names are listed in [144]Table S1). In
   addition, pathway enrichment analysis using MetaCore revealed a robust
   and unique browning transcriptome pattern in sWAT of TG mice, including
   oxidative phosphorylation, mitochondrial ubiquinone metabolism
   signaling and BAT-differentiation ([145]Figure 2H, for details see also
   [146]Table S2). Strikingly, this gene expression profile was fully
   absent in TG/FGF21^−/− mice although both TG genotypes displayed a
   similar fat depot mass.

3.3. FGF21 action is dispensable for muscle mitochondrial stress-improved
glycemic control

   Apart from the muscle mitochondrial stress-induced anti-obesogenic
   effects, the improved whole body insulin sensitivity also was recently
   related to FGF21 induction and WAT metabolic activation [147][9],
   [148][11]. In line with that, the induction of recruitable brown
   adipocytes, also called beige cells, has been linked not only to
   obesity resistance but also to improved glucose homeostasis [149][45].
   As mentioned above, we here performed two long-term diet intervention
   studies ([150]Figure 1 and [151]Figure S2). Surprisingly, despite the
   absence of FGF21 action and sWAT browning, we observed that glucose
   tolerance did not differ between TG and TG/FGF21^−/− mice in both
   studies ([152]Figure 3A–C + [153]Figure S2H and I). Intriguingly,
   comparing all groups, FGF21^−/− mice were most susceptible to
   HFD-induced impairment of glycemic control. Additionally, plasma
   insulin levels during oral glucose tolerance test were strongly reduced
   in both TG and TG/FGF21^−/− mice ([154]Figure 3D–F + [155]Figure S2J
   and K), in particular under HFD conditions. Remarkably, this effect was
   again independent of WAT browning and diet intervention. Thus, improved
   glycemic control and insulin sensitivity induced by muscle
   mitochondrial stress occurred independent of FGF21, suggesting that
   FGF21 as myokine is of little significance for the systemic adaptation
   of glucose metabolism.

Figure 3.

   [156]Figure 3
   [157]Open in a new tab

   FGF21 action is dispensable for muscle mitochondrial stress-improved
   glycemic control. Phenotype data of 40 weeks (wks) old male WT,
   FGF21^−/−, TG, and TG/FGF21^−/− mice fed low-fat (LFD) or high-fat diet
   (HFD) for 24 wks. (A) Basal blood glucose in postabsorptive state and
   (B) during an oral glucose tolerance test (oGTT) together with (C) the
   corresponding total area under curve of blood glucose during oGTT at wk
   38 (n = 9–11). (D) Basal plasma insulin levels in postabsorptive state
   and (E) during the oGTT together with (F) the corresponding total area
   under curve of insulin during oGTT. All data represent mean + SEM;
   means with different letters are significantly different.

3.4. Differential role of muscle FGF21 on plasma and hepatic lipid
homeostasis

   Browning of WAT has been associated with improved metabolic health,
   including diminished hepatic lipid content and attenuated dyslipidemia
   [158][46]. Thus, we further evaluated parameters of plasma and hepatic
   lipid homeostasis. Plasma triglycerides and cholesterol were
   significantly lower in TG mice than in the other three genotypes,
   independent of diet and sWAT browning, whereas plasma free fatty acids
   were not affected ([159]Figure 4A–C). The induction of FGF21 as
   stress-induced myokine was recently related to improved hepatic
   metabolic profile and whole-body metabolic homeostasis [160][9]. In
   contrast, we here clearly demonstrate a negligible role of muscle
   mitochondrial stress-induced FGF21 in hepatic lipid homeostasis.
   Histological analyses showed that although HFD induced hepatic
   steatosis in WT and FGF21^−/− mice, very little intracellular hepatic
   lipid accumulation was observed in TG and TG/FGF21^−/− mice on HFD
   ([161]Figure 4D) consistent with reduced hepatic triglyceride content
   ([162]Figure 4E). HFD did not lead to an induction of hepatic Fgf21
   gene expression whereas it highly increased hepatic gene expression of
   the fatty acid translocase Cd36 in control mice, which was completely
   abolished in both TG and TG/FGF21^−/− mice ([163]Figure 4F). Thus,
   despite the significant effects on sWAT remodeling, muscle secreted
   FGF21 had only minor effects on plasma and hepatic lipid homeostasis.

Figure 4.

   [164]Figure 4
   [165]Open in a new tab

   Differential role of muscle FGF21 on plasma and hepatic lipid
   homeostasis. Plasma and liver analyses of 40 weeks (wks) old male WT,
   FGF21^−/−, TG, and TG/FGF21^−/− mice fed low-fat (LFD) or high-fat diet
   (HFD) for 24 wks. (A) Plasma triglycerides, (B) free fatty acids, and
   (C) cholesterol levels (mmol/l) (n = 9). (D) Liver morphology (H&E);
   bars represent 50 μm. (E) Liver triglyceride content (mg) per mg
   protein (n = 9–11 per group). (F) qPCR of hepatic Cd36 (fatty acid
   translocase FAT/CD36) and Fgf21 gene expression (n = 8); qPCR cycle
   time values are displayed for the group with highest expression level.
   All data represent mean + SEM; means with different letters are
   significantly different.

3.5. Metabolic rescue of mitochondrial stress in muscle operates independent
of FGF21 action

   The precise pathophysiological relevance of mitochondrial
   stress-induced FGF21 especially for skeletal muscle itself is still
   poorly understood. Recent in-vitro studies proposed muscle FGF21 as a
   compensatory cell-autonomous stress response to improve mitochondrial
   function [166][47]. However, independent of FGF21 action and diet,
   skeletal muscle morphology was preserved ([167]Figure 5A), despite a
   greatly reduced muscle mass in both TG and TG/FGF21^−/− mice
   ([168]Figure 5B). Strikingly, the induction of phospho-AMPK^Thr172 and
   the fatty acid translocase CD36 in muscle of TG mice were not affected
   by the loss of FGF21 action ([169]Figure 5C). Furthermore, in line with
   a preserved mitochondrial OxPhos complex protein expression
   ([170]Figure 5C), muscle mitochondrial content evaluated by citrate
   synthase (CS) activity was also not affected by FGF21 ablation
   ([171]Figure 5D). Further, muscle glutathione peroxidase (GPx) activity
   as marker of cellular oxidative stress response remained increased in
   TG/FGF21^−/− mice ([172]Figure 5E). In addition, the robust gene
   expression induction of the myokines Fgf21 and Gdf15, as shown
   previously in TG mice [173][10], [174][34], could be confirmed, the
   latter similar in both TG and TG/FGF21^−/− ([175]Figure 5F). Recently,
   we described a detailed profile of muscle metabolic remodeling
   [176][34] adding to the already established mitochondrial
   stress-related pseudo-starvation response [177][8]. In agreement with
   these studies, key markers related the serine, one-carbon, glycine
   (SOG) pathway (Psat1, Mthfd2) were induced in skeletal muscle of TG
   mice ([178]Figure 5G). This metabolic profile was still preserved in
   the absence of FGF21, excluding FGF21 as control element of the
   pseudo-starvation response. Remarkably, immunoblots of UPR^mt marker
   proteins HSP25, TNX2 and ATP-dependent CLPP as well as selected SOG
   pathway proteins PHGDH and MTHFD2 in muscle showed no differences
   between TG and TG/FGF21^−/− mice ([179]Figure 5H), demonstrating that
   muscle metabolic remodeling and mitohormetic response does not require
   FGF21 action.

Figure 5.

   [180]Figure 5
   [181]Open in a new tab

   Metabolic rescue of mitochondrial stress in muscle operates independent
   of FGF21. Muscle analyses of 40 weeks (wks) old male WT, FGF21^−/−, TG,
   and TG/FGF21^−/− mice fed low-fat (LFD) or high-fat diet (HFD) for
   24 wks. (A) Muscle (tibialis anterior) morphology; bars represent
   50 μm. (B) Quadriceps muscle mass (n = 9–11). (C) Representative
   immunoblots of pAMPK^Thr172/tAMPK, mitochondrial OxPhos complexes (CI,
   Complex 1 subunit NDUFB8; CIII, Complex 3 core protein 2; CIV, Complex
   4 subunit I; ATP-S, ATP-Synthase alpha subunit) and fatty acid
   translocase FAT/CD36 (CD36). (D) Citrate synthase (CS) activity in m.
   gastrocnemius (n = 5). (E) Total glutathione peroxidase (GPx) activity
   in m. quadriceps (n = 9–11). (F–G) qPCR of myokines (Fgf21; Gdf15,
   growth differentiation factor 15) and key markers of SOG pathway
   (Psat1, phosphoserine aminotransferase 1; Mthfd2,
   methylentetrahydrofolate dehydrogenase 2) (n = 8); qPCR cycle time
   values are displayed for the group with highest expression level. (H)
   Representative immunoblots of UPR^mt marker proteins HSP25 (heat shock
   protein 25), TNX2 (thioredoxin-2), CLPP (Clp protease) and key SOG
   pathway proteins (MTHFD2; PHGDH, phosphoglycerate dehydrogenase). qPCR
   and immunoblots were performed in m. quadriceps; α-Tubulin was used as
   loading control. All data represent mean + SEM; means with different
   letters are significantly different. #, not detectable.

4. Discussion

   FGF21 currently receives general interest as one of the most promising
   clinical anti-diabetic drug candidate [182][4], but, paradoxically,
   circulating endogenous FGF21 is elevated during obesity [183][48],
   [184][49], type 2 diabetes [185][50], and coronary artery disease
   [186][51]. In contrast, its induction and secretion from skeletal
   muscle in response to mitochondrial perturbations and ISR activation
   [187][8], [188][9], [189][10], [190][11] raised the idea of an adaptive
   mitochondrial stress-related pseudo-starvation response via endogenous
   FGF21. However, previous studies so far only speculated on the
   beneficial metabolic role of FGF21, whereas detailed long-term
   proof-of-principle experiments using genetic ablation of FGF21 were
   lacking. Here, we conclusively tested this hypothesis using the
   established UCP1 transgenic mouse [191][28], [192][31], [193][33], a
   model of muscle mitochondrial perturbations and a recently published
   common metabolic remodeling profile [194][34], together with a
   comprehensive array of FGF21-ablated mouse models and dietary
   challenges. Surprisingly, we found new evidence supporting a
   dispensable role of FGF21 for muscle cell-autonomous as well as
   systemic metabolic adaptations.

   One effect that is clearly dependent on endogenous FGF21 action is the
   browning of sWAT induced by muscle mitochondrial stress in a
   cell-non-autonomous manner. The first to suggest an induction of
   browning by endogenous FGF21 was the group of Bruce Spiegelman who
   showed that FGF21-deficient mice displayed diminished cold induced
   browning of WAT. During cold exposure, FGF21 expression was found to be
   increased in different adipose depots suggesting an autocrine/paracrine
   manner of action [195][19]. Here we enlarge this concept to endocrine
   acting muscle FGF21. Surprisingly, since the study of Fisher et al.
   [196][19] our study is the only one to investigate browning in
   FGF21^−/− mice. Several studies have shown that FGF21 treatment or
   overexpression is associated with a browning of WAT [197][52],
   [198][53], [199][54] but without demonstrating that the observed
   browning was actually dependent on FGF21. Notably, browning of WAT has
   been associated with improved metabolic profiles making this an
   interesting target to combat obesity associated disorders [200][46]. On
   the other hand, recent studies found that WAT browning can also
   contribute to body weight loss and cachexia in cancer patients
   [201][55], [202][56]. Moreover, WAT browning can also be induced by
   circulating metabolites from muscle in response to exercise [203][57].
   Noteworthy, Fndc5/irisin, the only other myokine reported to induce WAT
   browning [204][58], is not induced by respiratory uncoupling in muscle
   [205][10], [206][39]. Overall, although the browning of WAT has been
   well established in different rodent models and also in humans, its
   precise physiological function and relevance remain hotly debated
   [207][59]. Here we show that FGF21-dependent browning is attenuated by
   HFD. This is not attributable to obesity, i.e. WAT expansion, because
   it is also observed in TG mice, which did not increase WAT mass on HFD.
   The attenuation of WAT browning by HFD thus seems to be related to
   dietary fat intake itself and suggests that it might play a role in
   lipid/fatty acid metabolism and homeostasis. In line with this,
   previous studies on Deletor and POLG mice showed that mitochondrial
   dysfunction is associated with lipid metabolic alterations, which can
   be compensated by a ketogenic or high-lipid diet [208][8], [209][26],
   [210][60]. Additionally, the muscle mitochondrial-stress related
   reduction of plasma triglycerides and cholesterol was abolished in the
   absence of FGF21, similarly under low and high fat feeding conditions.
   Browning of sWAT in general was largely attenuated by HFD questioning
   its role in these metabolic improvements. Further studies are warranted
   to examine the interaction of dietary fat and the mitochondrial
   stress-induced metabolic fine-tuning including WAT remodeling.

   In contrast to previous studies on muscle autophagy-deficient mice
   [211][9], our study on TG/FGF21^−/− mice clearly shows that
   mitochondrial stress-promoted obesity resistance and glycemic control
   occurs independently of endogenous FGF21, suggesting that it is of
   little significance for a systemic adaptive response. Noteworthy,
   autophagy plays a crucial role in the regulation of cellular processes
   to maintain skeletal muscle integrity and function [212][27],
   [213][61]. Thus, the metabolic response through disruption of autophagy
   studied by Kim et al. [214][9] may not arise from
   mitochondrial-specific stress but rather from a general impaired
   cellular homeostasis which requires FGF21 action. However, the
   metabolic profile of muscle autophagy-deficient mice under HFD
   conditions was still improved when crossed with FGF21^−/− mice, whereas
   standard/LFD conditions and the issues of FGF21-dependent WAT/BAT
   activation as well as cell-autonomous muscle ISR were not investigated
   at all [215][9]. Moreover, to our knowledge, there are presently no
   data available on Deletor and POLG mice with genetic ablation of FGF21.
   Nevertheless, although the different muscle-specific metabolic stress
   mouse models are based on distinct genetic manipulations [216][8],
   [217][9], [218][10], they all show a common adaptive response including
   the induction of muscle ER stress-mediated FGF21 and amino acid
   biosynthetic pathways [219][34]. In the present study, we show that the
   common mitochondrial stress-related pseudo-starvation response
   [220][8], [221][34], including key markers related to the serine,
   one-carbon, glycine (SOG) pathway, was still highly induced in
   TG/FGF21^−/− mice. This indicates that the muscle cell-autonomous
   metabolic remodeling occurs independently of endogenous FGF21, which
   needs to be confirmed in other described mitochondrial stress mouse
   models.

   If not FGF21, what might be the key mediators for muscle mitochondrial
   stress response? It is generally accepted that multiple organ crosstalk
   involves both cell-autonomous and cell-non-autonomous mechanisms
   [222][62]. Importantly, AMPK as a key regulator of cellular energy
   homeostasis and cell-autonomous stress adaption [223][39], [224][63],
   [225][64] was still activated in the absence of FGF21. Moreover, we
   here confirm that muscle mitochondrial stress promotes the expression
   of Gdf15, encoding a cytokine of the TGF-β superfamily (also known as
   MIC-1/NAG-1). This is in line with a recent transcriptomic study on
   mitochondrial myopathy patients [226][65]. Notably, we here show that
   the induction of Gdf15 was not affected by the loss of FGF21 action. It
   was previously shown that overexpression of Gdf15 in mice prevents
   diet-induced obesity and increases lifespan by activating adipose
   tissue as well as systemic energy metabolism [227][66], [228][67].
   Conversely, circulating levels of GDF15 are also highly elevated in
   patients with muscle atrophy [229][68] and in cancer patients with
   severe anorexia and weight loss [230][69] and chronic inflammation
   [231][70]. Hence, future studies are required to delineate the
   protective or detrimental role of GDF15 as mitochondrial stress-induced
   cytokine. Moreover, further efforts are warranted to identify
   additional critical cell-non-autonomous mediators of muscle
   mitochondrial stress adaptation, independent of endogenous FGF21.

5. Conclusion

   The endocrine acting pleiotropic protein FGF21 was proposed as key
   metabolic mediator of the mitochondrial stress adaptation. Notably, we
   here confirmed that WAT is a major target of circulating myokine FGF21.
   However, in contrast to prior expectations [232][9], [233][10],
   [234][11], [235][26], the present study on TG/FGF21^−/− mice as a model
   of muscle mitochondrial stress in the absence of endogenous FGF21
   clearly shows that the adaptive metabolic stress response operates
   independently of both WAT browning and FGF21 action. Thus, our data are
   pivotal for the progressively emerging concept of cell-non-autonomous
   and cell-autonomous muscle mitochondrial pseudo-starvation response.
   Taken together, these findings challenge FGF21 as a powerful
   therapeutic target during muscle mitochondrial disease and future
   studies are required to identify other, yet unknown adaptive processes
   apart from FGF21.

Author contributions

   MO and SK designed and conducted the study; VC, AV and MO performed
   animal experiments, collected and analyzed most of the data except
   from: EvS and IvdS performed microarray analysis and bioinformatics;
   SKei performed COX- and CS-activity assay in BAT; SR performed liver
   sample analysis; AG performed FACS analysis in adipose tissue, TA
   performed bone histology, MO, VC and SK wrote the paper, which was
   proof-read and corrected by all authors, who all contributed to data
   evaluation.

Acknowledgments