Long-term caloric restriction in ApoE-deficient mice results in neuroprotection via Fgf21-induced AMPK/mTOR pathway

Caloric restriction (CR) decelerates the aging process, extends lifespan and exerts neuroprotective effects in diverse species by so far unknown mechanisms. Based on known neuroprotective effects of fibroblastic growth factor 21 (Fgf21) we speculate that CR upregulates Fgf21, which phosphorylates neuronal AMP-activated protein kinase (AMPK), leading to a decrease of mammalian target of rapamycin (mTOR) signaling activity and an inhibition of tau-hyperphosphorylation. This in turn reduces the formation of neurofibrillary tangles, a neuropathological hallmark of Alzheimer's disease. ApoE-deficient mice (ApoE−/−), serving as a model of neurodegeneration, showed upon CR vs. ad libitum feeding increased Fgf21 levels in both, plasma and brain as well as higher phosphorylation of fibroblastic growth factor receptor 1c (Fgfr1c), extracellular signal-regulated kinases 1/2 (ERK1/2) and AMPK in brain, lower activity of mTOR and decreased Tau-phosphorylation. Finally, CR in ApoE−/− mice caused neuroprotection as indicated by a higher synaptic plasticity shown by immunohistochemical analysis with increased numbers of PSD95-positive neurons and a better cognitive performance as analyzed with Morris water maze test. These data provide substantial evidence that neuroprotection upon CR seems to be Fgf21-dependent. Further experiments are necessary to evaluate Fgf21 as a therapeutic tool to treat tauopathy for improvement of cognitive performance.


INTRODUCTION
In all species studied to date, restricted calorie intake by 20-50% while providing adequate micronutrient supply significantly extends mean and maximal lifespan [1,2]. Moreover, age-related deficits in learning and motor coordination are ameliorated by caloric restriction (CR) in rodents [3,4]. In line with this, CR attenuates amyloid deposition in monkeys and in transgenic mouse models of Alzheimer´s disease (AD) [5,6], leading to improvement of cognitive deficits [7] and to reduction of neuronal loss in neocortex, hippocampus, and striatum [8]. Further, it is described that CR promotes neurogenesis in adult rodents, probably by increasing brain-derived neurotrophic factor levels [9]. However, the underlying mechanisms in response to CR remain unclear.
Recently, the fibroblastic growth factor 21 (Fgf21) was described as starvation hormone [10]. Fgf21 is upregulated in response to CR in the liver and is secreted into plasma [11]. Fgf21 activity occurs when Fgf21 binds to fibroblast growth factors receptor (Fgfr) and β-klotho, a single transmembrane protein [12]. Fgfrs consist of seven major isoforms (1b, 1c, 2b, 2c, 3b, 3c and 4), whereby the isoform Fgfr1c is the primary receptor of Fgf21 in the mediation of its activity in in vivo studies [13]. When Fgf21 binds to its receptor, it leads to a rapid phosphorylation of downstream pathway components, including the MAPK cascade [14] and results via protein kinase A to activation of AMP-activated protein kinase (AMPK) [15]. In addition, fgf21 is also a direct target gene of the peroxisome proliferator-activated receptor-α (pparα) [16,17], a regulator for CR-induced lipolysis.

AGING (Albany NY)
Therefore, Fgf21 plays an important role in adaptation to metabolic states, which require increased fatty acid oxidation and ketogenesis [18] as an alternative energy source [17]. Ketogenesis, which is basically triggered by Fgf21, leads to AMPK activation not only in the periphery [19] but also in the central nervous system [20], resulting in decreased mammalian target of rapamycin (mTOR) signaling [21,22]. It could additionally be shown that inhibition of the mTOR pathway with rapamycin protects the entorhinal cortex from Tau-mediated neurodegeneration [23].
Since plasma Fgf21 can cross the blood brain barrier by simple diffusion and can be detected in human either AL or CR (60% of ad libitum). These images exemplarily show that in general CR-fed mice were smaller in body size than ALfed mice. (B) Body weight (g) of AL-and CR-fed mice over a period of 68 weeks. In general, parameter of ketogenesis and lipolysis are increased in CR-fed mice when compared with the age-matched AL-fed mice indicated by a marked rise of (C) plasma βhydroxybutyrate and of neuronal (D) pparγ and (E) abca1 mRNA expressions. Values are given as mean±SEM; ANOVA, post-hoc pairwise comparison tests: * p < 0.05 vs. AL. cerebrospinal fluid and brain tissues in rodents [24,25], it is conceivable to assume that Fgf21 contributes to regulation of brain metabolism. There is growing evidence that Fgf21 has neuroprotective effects and improves cognition [26]. It could be shown that incubation of human dopaminergic neurons with Fgf21 resulted in enhanced mitochondrial function and mitochondrial respiratory capacity [27]. Based on the neuronal functional activity of Fgf21 we speculate that Fgf21 phosphorylates neuronal AMPK. Activated AMPK leads to decrease of mTOR signaling activity [21] and in consequence to inhibition of Tauhyperphosphorylation with reduction of neurofibrillary tangles (NFTs) [28] as a neuropathological hallmark of AD.
Apolipoprotein E (ApoE) binds to Tau-protein and prevents its hyperphosphorylation [29], leading to slow down of NFTs-formation. ApoE-deficient mice (ApoE-/-) represent a well-established mouse model of tauopathy [30] with memory deficits [31]. In addition, ApoE-/compared to wild-type mice revealed throughout life a 2-to 7-fold lower expression of hepatic fgf21 (own unpublished data). Since Fgf21 has neuroprotective properties, it may be assumed that low Fgf21 contributes to neurodegeneration. To pursue this issue, we fed ApoE-/-mice caloric-restricted for a long-term to raise hepatic as well as neuronal Fgf21 with the aim to prevent tauopathy via the AMPK/mTOR pathway and to improve cognitive performance.

ApoE-/mice
In general, CR-fed mice were smaller in body size than the ad libitum (AL)-fed mice (Fig. 1A). The body weight of AL-fed ApoE-/-mice continuously increased 2-fold with aging up to 28.10 ± 0.85 g. Long-term CR resulted in a very slow increase of body weight reaching values of 19.6 ± 0.63 g. In general, at all time points CR-fed mice showed significantly lower levels of body weight when compared to AL-fed mice (Fig. 1B).

CR increased ketogenesis and neuronal lipolytic gene expression in ApoE-/-mice
CR-fed mice revealed a continuous rise of ketone bodies, as given by an up to 2-fold increase of plasma βhydroxybutyrate concentrations in long-term-fed mice when compared to short-term-fed mice. On the contrary, the concentrations of β-hydroxybutyrate remained almost unchanged in AL-fed mice averaging at low values of 1 mM up to 1.6 mM (Fig. 1C). Ketogenesis was significantly higher in CR-than in AL-

AGING (Albany NY)
fed ApoE-/-mice after short-and long-term feeding. The neuronal mRNA expression of pparγ and abca1 remained unchanged with aging in AL-fed ApoE-/mice ( Fig.1 D and E) while short-and mid-term CR markedly increased the neuronal mRNA expression of pparγ and abca1 ( Fig. 1 D and E).

CR increased hepatic expression and systemic concentration of Fgf21 in ApoE-/-mice
Of note, the hepatic mRNA expression of fgf21 in ApoE-/-mice was significantly increased upon longterm CR ( Fig. 2A). Accordingly, the systemic Fgf21 Representative Western blots as well as densitometric analysis of (B) pFgfr1c and (C) AMPK expression in primary glial cells, which were treated with vehicle (DMEM/F12) and 5 µg/ml Fgf21. Representative Western blot and densitometric analysis of (D) pERK1/2 and (E) pAMPK expression in brain of ApoE-/mice. Mice were fed either AL or CR (60% of ad libitum) for a short-term (4 weeks; n=14), mid-term (20 weeks; n=14) or long-term (64 weeks; n=14). Signals were corrected to that of either ERK1/2 or AMPK. Values are given as means ± SEM; ANOVA, post-hoc pairwise comparison tests: * p < 0.05 vs. AL, # p < 0.05 vs. vehicle.

AGING (Albany NY)
concentration in ApoE-/-mice raised significantly and reached approx. 3-fold higher levels upon long-term CR when compared to AL feeding (Fig. 2B). Fgf21 was barely measureable in the brain of ApoE-/-mice ( Fig.  2C; upper panel) but was detectable at a much higher level upon a long-term CR ( Fig. 2C; arrows, lower panel) with a preferential location around glial cells in the cortex. Along with the higher neuronal Fgf21 levels upon long-term CR, the receptor for Fgf21, namely Fgfr1c, was activated, as indicated by an increased number of pFgfr1c-positive neuronal cells in the cortex ( Fig. 3A; lower panel, arrows).

CR increased Fgfr1c downstream signaling in ApoE-/-mice
Next we asked whether Fgf21 activates the neuronal downstream signaling pathway via Fgfr1c in vitro and could show that phosphorylation of Fgfr1c is significantly increased in Fgf21-treated primary glial cells compared to vehicle alone (Fig. 3B). Along with activation of Fgfr1c the AMPK protein expression was significantly increased in Fgf21-treated primary glial cells compared to vehicle alone (Fig. 3C). Further, a significant increase of ERK1/2-phosphorylation was found upon short-as well as mid-term CR vs. AL feeding in ApoE-/-mice (Fig. 3D). In addition, the phosphorylation of AMPK was consistently increased upon CR (Fig. 3E).

CR inhibited mTOR and Tau-phosphorylation in ApoE-/-mice
Since activation of AMPK inhibits mTOR, we analyzed the phosphorylation of mTOR by immunohistochemistry and could show that long-term CR reduced the activity of mTOR in the cortex of ApoE-/-mice, as indicated by a decreased number of pmTOR-positive neuronal cells ( Fig. 4A; lower panel, arrows). It could further be demonstrated that an inhibition of mTOR upon CR feeding resulted in an almost 2-fold reduction of Tau-phosphorylation at all time points, which was significant upon long-term CR (Fig. 4B) while AL-fed mice revealed an age-dependent increase of Tauphosphorylation (Fig. 4B).

Long-term CR increased working memory and neuronal plasticity in ApoE-/-mice
Mice were tested in the spatial reference memory version of the Morris water maze. AL-and CR-fed ApoE-/-mice did not show significant differences in general swimming performance, which was determined by measurement of the swimming speed during the test AGING (Albany NY) trials (data not shown). Escape latencies were displayed in five blocks. Each block consisted of four consecutive trials. It is expected, escape latencies decreased throughout the training days. Thus, all mice were able to learn the task, whereas aged mice (long-term AL/CR) vs. younger mice (short-and mid-term AL/CR) started with a per se inferior performance (Fig. 5A, C, E). While short-and mid-term-fed mice of both groups revealed almost comparable escape latency, long-term CR resulted in lower escape latency found for block 2 -4 ( Fig. 5E), suggesting that long-term CR caused a better training performance due to better cognitive activities.
All mice were exercised to the platform 4 (PF4) during training. Finally, the platform was removed and the AGING (Albany NY) number of platform crossings during 60 s was measured. Thereby, it could be shown that short-as well as mid-term-fed mice of both groups crossed PF4 nearly with the same frequency (Fig. 5B, D, F). The most striking differences between the AL-and CR-fed mice were found in the number of platform crossings of long-term-fed mice. Long-term CR-fed mice crossed PF3 up to 30% and PF4 significantly more often than AL-fed mice (Fig. 5F). Thus, long-term CR resulted in a better working memory.

CR delayed the decrease of neuronal PSD95 expression in ApoE-/-mice
AL-fed ApoE-/-mice revealed an age-dependent decrease of the number of PSD95-positive neuronal cells (Fig. 6A, C, E; arrows) while this could not be observed in CR mice, as shown by an almost stable expression of PSD95 (Fig. 6B, D, F; arrows).

DISCUSSION
The main finding of the present study is that long-term CR in ApoE-/-mice increased lipolysis and ketogenesis in the animals. This was reflected by higher systemic ketone body concentrations and increased neuronal mRNA expressions of pparγ and abca1. CR further caused both, a systemic and neuronal increase of Fgf21, leading to higher activation of Fgfr1c in the brain and, in consequence, to activation of the AMPK/mTOR pathway. AMPK-induced inhibition of mTOR in CRfed ApoE-/-mice reduced Tau-phosphorylation and improved their cognition performance.
Most recently, we could demonstrate that long-term CR i.e. starting at the age of weaning until pre-senescence reduced the normal age-related decline in cognitive performance [4]. This underlines the importance of lifestyle factors such as nutrition in the context of metabolic syndrome-associated neurodegeneration [32]. However, the mechanism of CR-induced neuroprotection is not yet completely understood. To address this issue, we used ApoE-/-mice an accepted model of metabolics-driven neurodegeneration. The animals were maintained under CR for a long-term to evaluate neuroprotective molecular mechanisms. ApoE-/-mice are known for their significant loss of synapses with increasing age [33] and further are also described as model of tauopathy [30].
In response to fasting, lipolysis and β-oxidation are enhanced to provide alternative energy resources for the organism [17,34]. In addition, marked elevations in Fgf21 expression were observed in mice upon fasting [18] and food restriction [11]. Elevated fgf21 gene expression and serum Fgf21 concentrations in foodrestricted animals were accompanied by an upregulation of pparα mRNA expression [11]. PPARα regulates the hepatic lipid metabolism by induction of ketogenesis and lipolysis via increased β-oxidation [11,35].
Accordingly, the present study shows an accumulation of ketone bodies upon CR. Ketone bodies may act neuroprotective [36] and thereby, enhance memory performance [37]. In addition, we found that lipolysis and β-oxidation are also enhanced in brain as shown by increased neuronal mRNA expression of pparγ and of its target gene abca1. Since activation of PPARγ reduces AD-pathology and improves cognitive function in mouse models of AD [38] this current finding might also contribute to neuroprotection.
Fgf21 is present in the brain [39] including midbrain and in vitro is expressed by glial cells [27]. Mäkelä and coworkers found additionally that Fgf21 plays a regulatory role in brain energy metabolism as shown by enhanced mitochondrial capacity in Fgf21-treated dopaminergic neurons [27]. Further, it is known that Fgf21 can cross the blood brain barrier [24]. Thus, it is conceivable to state that CR-induced hepatic Fgf21 might be responsible for the positive immunostaining of glial cells in vivo, as observed in the present study (Fig.  2C). Fgfr1c, the receptor for Fgf21, is found to be expressed in hypothalamic region of the brain [40] and in cortical neurons (current study; Fig. 3A). Activation of Fgfr1c -most occasionally via Fgf21-positive glial cells-phosphorylates specific tyrosine residues that mediate interaction with cytosolic adaptor proteins and intracellular signaling pathways [14]. This view is supported by our in vitro data showing the activation of Fgfr1c and the increase of downstream AMPK protein upon Fgf21 exposure of glial cells and by our in vivo data showing increased Fgfr1c-as well as increased ERK1/2-and AMPK-phosphorylation in CR-fed mice. Since Fgf21 is capable to elevate phosphorylation of AMPK [19], we assume that CR-induced rise of Fgf21 triggers the AMPK pathway. However, it has to be mentioned that conditions of perceived energy deprivation, such as fasting, food restriction, and exercise, increase the AMP/ATP ratio, eventually causing an activation of AMPK [41 and current study].
AMPK is a suppressor of mTOR signaling [42]. Abrogation of TOR signaling causes rejuvenation in the C. elegans [43]. This new function for TOR signaling in ageing control may represent a link between nutrition, metabolism and longevity [43]. In support of this function, inhibition of mTOR is described to be protective in neurodegenerative diseases such as AD by enhancement of autophagy, a biological process that not only facilitates the clearance of mutant proteins but also significantly reduces the build-up and accumulation of toxic protein aggregates such as NTFs, being caused by Tau-hyperphosphorylation [44,45]. Therefore, reduced mTOR and Tau-phosphorylation in the long-term CRfed ApoE -/-mice can be most supposedly attributed to the activation of AMPK pathway by increased Fgf21. In parallel, Ma and coworkers showed that food restriction could lead to activation of SIRT1 and suppression of mTOR activation in mice and speculated that the SIRT1/mTOR signaling pathway may be involved in the neuroprotective effect of food restriction [46]. This view is supported by the study of Guo et al. [47], which demonstrated that food restriction-induced SIRT/mTOR activation contributed to brain integrity. In addition, positive properties of SIRT signaling in ApoE-/-mice was reported by the groups of Stein et al. [48] and Palacios et al. [49]. These studies showed that of SIRToverexpression diminishes atherogenesis and that dietand exercise-induced SIRT/AMPK pathway regulates muscle energy homeostasis.
Moreover limited Tau-phosphorylation by Fgf21dependent AMPK/mTOR signaling, improved cognitive performance upon long-term CR restriction might also be due to maintenance of PSD95-positive cells and thus, neuronal plasticity.
In summary, these data provide substantial evidence that neuroprotection upon CR seems to be Fgf21/AMPK/mTOR-dependent. Further experiments are necessary to evaluate Fgf21 as a therapeutic tool to treat tauopathy for improvement of cognitive performance.

Animals
Female Apolipoprotein (Apo) E deficient (ApoE-/-) mice at the age of 4 weeks were fed either ad libitum (AL) or caloric-restricted (CR, 60% of ad libitum) for 4

AGING (Albany NY)
additional weeks (short-term, n=10 for each group), 20 weeks (mid-term, n=10 for each group) or 64 weeks (long-term, n=10 for each group). All mice were housed in standard cages in a temperature-controlled room (22°C±2°C) on a 12 h light/dark cycle with free access to food (4.2% fat) and water under specific pathogen free (SPF) conditions. The mice were weighed weekly and directly before they were sacrificed. The

Sampling and assays
For the analysis of Fgf21-kinetics blood samples were taken from all mice at weeks 8, 16, 24, 32, 40, 48 and 64. At the end of the experiment, all mice were exsanguinated by puncture of the vena cava inferior for immediate separation of plasma, followed by harvest of brain and liver tissues. Measurements of Fgf21 and ketone bodies in plasma were performed using the Fgf21 immunoassay and β-hydroxybutyrate assay kit methods according to the manufacturer's instructions (Fgf21: R&D System; β-hydroxybutyrate: Cayman Chemical Company).

Immunohistology
For immunohistochemical analysis, brain tissue was fixed in 4% phosphate-buffered formalin for 5-6 weeks and subsequently embedded in paraffin. From the paraffin-embedded tissue blocks, 4 µm thin sections were put on X-tra Adhesive Precleaned Micro Slides (Leica) and exposed to a rabbit monoclonal anti-Fgf21 antibody (1:500, Abcam), rabbit polyclonal anti-pFgfr1c antibody (1:100, Cell Signaling), rabbit monoclonal anti-pmTOR (IHC specific, 1:100, Cell Signaling) and rabbit polyclonal anti-PSD95 (1:1000, Abcam). For the development of the primary antibodies with DAB chromogen Universal LSAB  kits (System-HRP; DakoCytomation, Dako) were used according to the manufacturer's instructions. The sections were counterstained with hemalaun and analyzed with a light microscope (Olympus BX51). Images were acquired with a Color View II FW camera (Color View).

Morris water maze test
The Morris water maze (MWM) was performed as a task to measure spatial reference memory. Mice under dim light conditions (indirect illumination, 3.0 Lux) at days -4 to -1 d were trained to locate a submerged hidden platform (11 cm diameter) in a fixed quadrant of a circular pool filled with opaque water (diameter of pool 107 cm, filled to a depth of 20 cm, water temperature 17º C, platform submerged 1 cm beneath the surface). For each trial (1 x at -4 d to familiarize the animals with test apparatus, 8 x at -3 d, 8 x at -2 d, 4 x -1 d = 5 training blocks each consisting of 4 consecutive trials) mice are gently placed in the water, hind legs first. On each trial, the starting position was randomized between four possible quadrant positions. The location of the platform remained constant throughout the whole training period. Each trial lasted until the animal located the platform. However, after max. 60 s the mouse was guided to the platform. The latter was given a latency score of 60 s. Mice were allowed to rest 10 s on the escape platform, were then removed, dried with a towel and warmed under a heating lamp. After further 30 s the next trial started. The time to reach the platform (latency to escape) was recorded for each trial. On day -1, mice finally were trained for four trials. All training trials were monitored online by a video camera [4]. After training, the platform was removed from the pool and the mouse performed a 60 s test trial. The start position of the test trial was located opposite to the quadrant that had contained the platform. The swimming route and the number of times the mouse crossed each of the four possible platform positions (the previous platform position and 3 imaginary platform position in the other quadrants and platform crossings) was monitored online by a video camera [4].

Statistical analysis
All data are expressed as means ± SEM. Statistical differences were determined using one-or two-way ANOVA followed by post-hoc pairwise comparison AGING (Albany NY) tests for analysis between either feeding groups or duration of feeding. Data were considered significant if p<0.05. Statistical analysis was performed using the SigmaStat software package (Jandel Scientific).