Fibroblast Growth Factor-21 Controls Dietary Protein Intake in Male Mice

Abstract

Whereas carbohydrates and lipids are stored as glycogen and fat, there is no analogous inert storage form of protein. Therefore, continuous adjustments in feeding behavior are needed to match amino acid supply to ongoing physiologic need. Neuroendocrine mechanisms facilitating this behavioral control of protein and amino acid homeostasis remain unclear. The hepatokine fibroblast growth factor-21 (FGF21) is well positioned for such a role, as it is robustly secreted in response to protein and/or amino acid deficit. In this study, we tested the hypothesis that FGF21 feeds back at its receptors in the nervous system to shift macronutrient selection toward protein. In a series of behavioral tests, we isolated the effect of FGF21 to influence consumption of protein, fat, and carbohydrate in male mice. First, we used a three-choice pure macronutrient-diet paradigm. In response to FGF21, mice increased consumption of protein while reducing carbohydrate intake, with no effect on fat intake. Next, to determine whether protein or carbohydrate was the primary-regulated nutrient, we used a sequence of two-choice experiments to isolate the effect of FGF21 on preference for each macronutrient. Sweetness was well controlled by holding sucrose constant across the diets. Under these conditions, FGF21 increased protein intake, and this was offset by reducing the consumption of either carbohydrate or fat. When protein was held constant, FGF21 had no effect on macronutrient intake. Lastly, the effect of FGF21 to increase protein intake required the presence of its co-receptor, β-klotho, in neurons. Taken together, these findings point to a novel liver→nervous system pathway underlying the regulation of dietary protein intake via FGF21.

Energy from food derives from three macronutrient forms: carbohydrates, fats, and proteins. During the absorptive state, ingested carbohydrates and fats are stored in the body as glycogen or triglycerides and can later be mobilized to provide fuel during the postabsorptive state. In contrast, there is no analogous inert storage form of protein. Thus, protein is used sparingly as fuel, and continuous adjustments in feeding behavior are needed to ensure an adequate supply of amino acids for dynamic use, maintenance, and repair (1). Considerable evidence supports a behavioral regulation of protein intake, but mechanisms that underlie such behaviors are largely undescribed (1–3).

Fibroblast growth factor-21 (FGF21) was identified as a secreted protein from mouse liver and was cloned in 2000, and an abundance of studies detailing its metabolic benefits have appeared since then (4–7). However, a clear understanding of the physiological and/or nutritional contexts regulating FGF21 secretion has been elusive. Plasma FGF21 is modestly induced by fasting (8), or by maintenance on a (low-protein) ketogenic diet (9–11) in mice and rats. As a result, it was initially characterized as a “starvation hormone” that facilitates aspects of the adaptive starvation response. However, neither fasting nor a ketogenic diet reliably increases circulating FGF21 in humans (6). New findings suggest a more nuanced role. Recent evidence supports that FGF21 is induced by a deficit of various individual essential and nonessential amino acids (12–15) or by a deficit of total dietary protein, rather than by caloric restriction per se, in both rodents and in humans (15–17). Surprisingly, FGF21 is also induced by macronutrient excess. In 2009, Sánchez et al. (18) reported that, whereas plasma FGF21 was modestly increased after a 24-hour fast, refeeding with isocaloric amounts of either carbohydrate or fat caused a substantial additional increase. They concluded that FGF21 indicates an “unbalanced nutritional situation.” In agreement with this, current work now demonstrates that FGF21 is also induced up to 10-fold in ad libitum–fed rodents and humans after consuming an excess of simple sugars relative to other dietary options (19–22).

To resolve this apparent paradox, Simpson and colleagues (2) used the multidimensional geometric framework model to distinguish the roles of macronutrients and total energy intake on FGF21 levels in 853 mice maintained on 25 different diets of varying macronutrient content and caloric density. They concluded that plasma FGF21 increases in response to low dietary protein intake and, most strongly, when low protein intake was coupled with high carbohydrate intake. Thus, FGF21 is an endocrine signal of relative dietary protein “dilution,” characterized by a low dietary protein/carbohydrate (P/C) ratio. Recent findings from Rose and colleagues (15) and Morrison and colleagues (17) agree with this interpretation. Likewise, when we sampled from mice eating isocaloric diets that contain 22% fat and either 18% protein and 60% carbohydrates (normal P/C ratio) or 4% protein and 74% carbohydrates (low P/C ratio), plasma FGF21 was increased ∼40- to 60-fold in low P/C ratio mice compared with normal P/C ratio mice (23).

Circulating FGF21 crosses the blood–brain barrier to activate neurons in the brainstem and hypothalamus known to be important in the regulation of energy balance and feeding behavior (21, 24–26). Therefore, it is well situated to act as a neuroendocrine signal, eliciting compensatory behavioral changes that increase the relative intake of dietary protein. Consistent with this, in a recent genome-wide association study, variants located near the Fgf21 locus associated with decreased protein (and increased carbohydrate) intake (27). The current study tested the hypothesis that FGF21 acts via its receptors in the nervous system to restore appropriate macronutrient balance, by increasing protein intake at the expense of carbohydrate and/or fat.

Materials and Methods

Animals

All animal experiments were approved by the Institutional Care and Use Committee of the University of California, Davis. Age-matched adult male C57BL/6J mice (n = 8 to 20 per group) were obtained from The Jackson Laboratory (Sacramento, CA) or were bred in-house, for up to two generations removed from the founders. Mice lacking β-klotho in neurons (Klb^ΔSynCre) were generated by crossing β-klotho-floxed mice (gift from Steven Kliewer, UT Southwestern Medical Center) with mice expressing Cre-recombinase under the Synapsin-I promoter (catalog no. 003966, The Jackson Laboratory). To avoid known germline recombination, the Cre gene was passed only through females (28). As an initial validation of this line, we extracted mRNA from whole hypothalamus of neuron-specific β-klotho-null mice (n = 3, Klb^ΔSynCre) and littermate controls (n = 5, Klb^{flox/flox/Cre−}), and using quantitative RT-PCR (see methods described in this section), we determined that β-klotho (Klb) expression was reduced by 69% in the Klb^ΔSynCre mice (100 ± 7.7 vs 31 ± 10.1, t test, P < 0.05). Male mice were used in all of the experiments and were singly housed on a 12-hour light/12-hour dark cycle in a temperature-controlled (20°C to 22°C) and humidity-controlled vivarium with ad libitum access to food and water.

Diets

Standard chow diet, used for the experiments shown in Fig. 1, was obtained from Harlan Laboratories/Envigo (Madison, WI; catalog no. 5008). Sucrose was from Sigma-Aldrich (St. Louis, MO) and was dissolved in regular tap water. Single macronutrient diets from Harlan Laboratories/Envigo, used for the experiments shown in Fig. 2, contained >99% of total calories derived from protein, carbohydrates, or fat, respectively, as in Ryan et al. (29) and Chambers et al. (30) (TD.02523, TD.02521, and TD.02522). Purified, pelleted diets used for the experiments shown in Figs. 3–6 (D11092301, D11051801, D18040303, D18040304, and D18061811) were manufactured by Research Diets (New Brunswick, NJ). The pelleted diets were based on the AIN-93 rodent diet formula (31) in which protein is derived from casein and supplemented with cystine. Full nutritional details can be found in Tables 1 and 2.

Figure 1.

FGF21 decreases sucrose intake and increases chow intake. In a two-bottle test, mice were offered 10% sucrose vs water, together with ad libitum chow. (A) FGF21 (1 mg/kg bw, IP) reduced sucrose intake and increased chow intake [P (diet × treatment) < 0.001; Tukey post hoc test, ***P < 0.001]. There was no effect of FGF21 on (B) total caloric intake or (C) body weight change. Data are shown as mean ± SEM; n = 9 to 11 mice per group.

Figure 2.

FGF21 increases protein intake and decreases carbohydrate intake. In a three-choice pure macronutrient selection paradigm, mice were offered free access to pure protein (PRO; casein), carbohydrate (CHO; 33% sucrose and 67% corn starch), and FAT (vegetable shortening) diets. We injected 1 mg/kg bw FGF21 or saline IP and measured overnight food intake. (A) FGF21 increased PRO intake and decreased CHO intake [P (diet × treatment) < 0.01; Tukey post hoc test, *P < 0.05, **P < 0.01). There was no effect of FGF21 on (B) total caloric intake or (C) body weight change. Data are shown as mean ± SEM; n = 20 mice per group.

Figure 3.

FGF21 increases protein intake and decreases carbohydrate intake. (A) In a first cohort of mice, we measured circulating hFGF21 concentrations following an IP injection of 0.1 mg/kg. A second cohort of mice was offered simultaneous access to two pelleted diets matched for dietary fat content (22%), whereas the protein/carbohydrate ratio varied (4%:74% or 18%:60%). After establishing a baseline intake for 1 wk, mice were divided into two groups that were well matched for (B) total caloric and (C) macronutrient intake. (E) Next, we delivered FGF21 (0.2 mg/kg/d, IP) or saline for 7 d. FGF21 elicited an increase in the consumption of dietary protein, offset by a decrease in carbohydrate intake. In agreement with its established role to increase energy expenditure, FGF21 also (D) increased total caloric intake and (F) induced body weight loss. Data are shown as mean ± SEM; n = 3 to 5 mice per treatment per time point for (A), and n = 15 mice per group for (B)–(F). All comparisons were made by a t test: *P < 0.05, **P < 0.01. PRO, protein.

Figure 6.

Nervous system β-klotho is necessary for the effects of FGF21 on protein intake. Mice lacking the FGF21 coreceptor, β-klotho, in neurons (Klb^ΔSynCre) and littermate controls (Klb^{flox/ flox}) were offered a choice between two pelleted diets matched for dietary fat content (22%), whereas the protein/carbohydrate ratio varied (4%:74% or 18%:60%). After establishing a baseline intake for 1 wk, mice of each genotype were divided into two groups that were well matched for (A) caloric and (C) macronutrient intake (n = 5 to 7 per group). Next, we delivered FGF21 (0.2 mg/kg/d, IP) or saline for 7 d. (D) As expected, the effect of FGF21 on macronutrient intake depended on genotype [P (treatment × genotype) < 0.05]. Only the Klb^flox/flox mice significantly altered diet selection relative to baseline by increasing protein intake in response to FGF21 (Wilcoxon signed-rank test, ^#P < 0.05). (B) FGF21 also elicited a significant increase in caloric intake, which did not depend on genotype [P (treatment) < 0.05], and (E) there was no effect of either genotype or treatment on body change. (F) Lastly, we confirmed knockdown by quantitative RT-PCR (t test, ***P < 0.001). iBAT, intrascapular brown adipose tissue; iWAT, inguinal white adipose tissue; PRO, protein.

Table 1.

Diet Composition for Three-Choice Diets

Ingredient, g/kg	Protein Only (TD.02523)	Carbohydrate Only (TD.02521)	Fat Only (TD.02522)
Casein	871.7
dl-Methionine	1.1	1.1	2.0
Cellulose	87.0	87.0	169.0
Mineral mix, AIN-76 (170915)	30.7	30.7	59.5
Mineral mix, AIN-76A (40077)	7.7	7.7	14.9
Choline chloride	1.8	1.8	3.4
Corn starch		581.1
Sucrose		290.6
Vegetable shortening, hydrogenated			751.2
kcal/g	3.2	3.3	6.9

Table 2.

Diet Composition for Pelleted Diets

Ingredient, g	D11092301	D11051801	D18040303	D18040304	D18061811
Casein	50	200	50	206	206
l-Cystine	0.75	3	0.75	3	3
Corn starch	485	375.7	176.75	180	289.2
Maltodextrin 10	150	125	60	60	125
Sucrose	107.0777	107.0777	107.0777	107.0777	107.0777
Cellulose	50	50	50	50	50
Soybean oil	25	25	69.25	53.4	34
Lard	75	75	207.75	160.1	102.1
Mineral mix S10022G	0	0	0	0	0
Mineral mix S10022C	3.5	3.5	3.5	3.5	3.5
Calcium carbonate	8.7	12.495	8.7	8.7	8.7
Calcium phosphate, dibasic	5.3	0	5.3	5.15	5.15
Potassium citrate, 1 H2O	2.4773	2.4773	2.4773	6.72	6.72
Potassium phosphate, monobasic	6.86	6.86	6.86	1.53	1.53
Sodium chloride	2.59	2.59	2.59	2.59	2.59
Vitamin mix V10037	10	10	10	10	10
Choline bitartrate	2.5	2.5	2.5	2.5	2.5
FD&C yellow dye no. 5	0	0.05	0	0.025	0.025
FD&C red dye no. 40	0.05	0	0	0.025	0
FD&C blue dye no. 1	0	0	0.05	0	0.025
kcal/g	4.15	4.082	5.35	4.75	4.27
Protein, g%	4.569	17.88	5.89	21.39	19.22
Carbohydrate, g%	81.437	66.72	52.91	47.31	60.70
Fat, g%	10.154	9.99	36.28	24.87	14.21
Protein, kcal%	4.38	17.52	4.38	18.03	18.03
Carbohydrate, kcal%	73.6	60.46	34.62	34.94	51.99
Fat, kcal%	22.021	22.021	61.00	47.03	29.97
Calcium, g	5.06	5.06	5.06	5.06	5.06
Phosphorus, g	3.17	3.17	3.17	3.17	3.17
Potassium, g	3.6	3.6	3.6	3.6	3.6
Folate, g	2	2	2	2	2
Omega-6 fatty acids	15.10	15.10	41.83	32.25	20.54
Omega-3 fatty acids	3	3	8.31	6.41	4.08
Omega-6/omega-3 ratio	5.03	5.03	5.03	5.03	5.03

Abbreviation: FD&C, food, drug, and cosmetic.

FGF21

Endotoxin-free, recombinant human FGF21 (hFGF21; ProSpecBio, Rehovot, Israel) (32) was first dissolved in H₂O, according to the manufacturer’s instructions, and was further diluted in 0.9% sterile saline for injection (1 mg/kg acutely, or 0.1 mg/kg twice daily, given just after the onset of light and again just before the onset of dark during a 12-hour light/12-hour dark cycle).

To determine plasma hFGF21 concentration in C57BL/6 mice, following a single IP injection (0.1 mg/kg), we collected blood at 0, 1, 4, and 10 hours, which was stored on ice. Plasma was separated by centrifugation for 30 minutes at 3000g at 4°C and stored at −80°C until later use. hFGF21 was measured using a commercially available ELISA kit following the manufacturer’s protocol (R&D Systems, Minneapolis, MN; catalog no. DF2100) (33). The lower limit of detection for this assay is 4.67 pg/mL.

Gene expression by quantitative RT-PCR

Total RNA was isolated from hypothalamus (see above), whole brain (Fig. 6F), and adipose depots (Fig. 6F) using the RNeasy mini kit (Qiagen, Germantown MD), according to the manufacturer’s instructions. cDNA was synthesized using a high-capacity cDNA reverse transcription kit (Life Technologies/Thermo Fisher, Carlsbad, CA), and gene expression analysis was performed, in duplicate, using TaqMan gene expression assays. Expression of Klb (Mm00473122_m1) was normalized to the housekeeping genes Rn18s (Mm04277571_s1) and Ppib (Mm00478295_m1) by the 2^−ΔΔCT method.

Statistical analysis

Data were analyzed using GraphPad Prism (GraphPad Software, La Jolla, CA) and SigmaStat (Systat Software, San Jose, CA) by a two-tailed t test, Wilcoxon signed-rank test, or ANOVA, as indicated. Planned comparisons were made using a Tukey honestly significant difference test. Data are plotted as means ± SEM unless otherwise noted.

Results

Sucrose vs water: two-choice test

Recent evidence supports that FGF21 is a negative regulator of simple sugar intake and sweet taste preference. In independent studies, Kliewer and colleagues (20) and Potthoff and colleagues (21) reported that FGF21 reduces the preference for sucrose solution vs water in a two-bottle choice test. For these experiments, however, it was not clear whether total caloric intake was also changed, or whether reduced energy intake from sucrose was offset by compensatory changes in the consumption of other dietary macronutrients. To explicitly test this, we provided adult male mice simultaneous free access to two 50-mL sippers containing 10% sucrose vs water, and with ad libitum access to regular chow. Following a 48-hour acclimation period, we administered 1 mg/kg body weight (bw) of FGF21 by IP injection, as in Kliewer and colleagues (20) and Potthoff and colleagues (21), or saline, just prior to lights out. Water, 10% sucrose, and chow intakes were measured the following morning, 2 hours into the light period.

Consistent with prior reports, we find that FGF21 reduced the consumption of sucrose vs water in the two-bottle test. FGF21-treated mice consumed fewer total kilocalories from sucrose compared with saline-treated controls. This was offset by increased consumption of normal chow [Fig. 1A, two-way ANOVA, P (diet × treatment) < 0.001; P < 0.001 by Tukey post hoc test] such that there was no difference in total caloric intake (Fig. 1B). There was no effect of FGF21 on body weight following this acute FGF21 injection (Fig. 1C). Taken together, we conclude that FGF21 decreases sucrose intake while simultaneously increasing consumption of other available dietary options.

Single macronutrient diets: three-choice test

To explicitly determine macronutrient tradeoffs in response to FGF21, we used a standard three-choice macronutrient selection paradigm, in which mice can freely distribute food intake between three single macronutrient diets. Dietary protein was from casein supplemented with methionine (TD.02523); dietary fat was from vegetable shortening (TD.02521); and dietary carbohydrate was one-third sucrose and two-thirds cornstarch (TD.02522) (see Table 1 for details). Following a 4-day acclimation period, we delivered FGF21 (1 mg/kg bw IP) or saline just before lights out and measured food intake overnight. We observed a significant increase in dietary protein intake, coupled with a significant decrease in carbohydrate intake [Fig. 2A, P (diet × treatment) < 0.01; P < 0.05, Tukey post hoc test]. There was no difference in total caloric intake (Fig. 2B), and there was no significant effect of FGF21 on body weight (Fig. 2C). We conclude that FGF21 decreases carbohydrate intake, and this is offset by increased consumption of dietary protein.

Pelleted diets: series of two-choice tests

In three separate cohorts of mice, we completed a series of experiments in which individuals were allowed to choose between two pelleted diets that were matched for one of the three macronutrients, whereas the other two macronutrients varied. To isolate the contribution of carbohydrates per se vs simple sugars and/or sweet taste, sucrose content was held constant for all of the pairwise diet choices discussed below. Detailed descriptions of the diets can be found in Table 2. For these chronic experiments, FGF21 was delivered at 0.1 mg/kg, twice daily. This dose was chosen according to a previous study (34) demonstrating that it is effective to induce weight loss and improve glucose tolerance in diet-induced obese mice. To determine the plasma concentration of hFGF21 resulting from this dose, we collected blood from mice at 0, 1, 2, 4, and 10 hours after its IP injection. Circulating concentrations reached 65.7 ng/mL hFGF21 at 1 hour after the injection. By 4 hours, plasma hFGF21 was comparable to the levels we and others observe during dietary protein dilution (23) (7.86 ng/mL), and by 10 hours no hFGF21 was detectible. As expected, we did not detect hFGF21 from saline-injected mice at any time point (Fig. 3A).

In the first of the two-choice experiments mice were offered simultaneous access to two diets that were matched at 22% fat (iso-FAT), together with 18% protein and 60% carbohydrates or 4% protein and 74% carbohydrates [diets as in Maida et al. (15), Laeger et al. (17), and Larson et al. (23); D11051801 and D11092301, Table 2]. First, mice were maintained on the two diets for 1 week to establish a baseline ratio of macronutrient intake. Next, mice were divided into two treatment groups matched for baseline caloric and macronutrient intake, expressed here as percentage of total kilocalories from protein (Fig. 3B and 3C). Lastly, we delivered FGF21 (0.1 mg/kg/d, twice daily, IP) and measured intake of the two diets for 7 days. Chronic administration of FGF21 induces weight loss via increased energy expenditure, and this is often accompanied by increased caloric intake (35). In agreement with this, mice treated for 7 days with FGF21 lost weight (Fig. 3F, t test, P < 0.05) and ate more calories (Fig. 3D, t test, P < 0.01) compared with saline-treated controls. Moreover, as we observed during our three-diet choice test (Fig. 2), mice treated for 7 days with FGF21 significantly increased the percentage of kilocalories consumed from dietary protein (and decreased the percentage of kilocalories consumed from carbohydrates), whereas saline-treated mice did not change their macronutrient consumption (Fig. 3E, t test, P < 0.01).

In the second such experiment, mice were offered simultaneous access to two diets that were matched at 35% carbohydrate (iso-CHO), together with 18% protein and 47% fat or 4% protein and 61% fat (D18040304 and D18040303, Table 2). Mice were maintained on the two diets for 1 week to establish a baseline, as above. Next, mice were divided into two treatment groups matched for baseline caloric (Fig. 4A) and macronutrient intake, expressed here as percentage of total kilocalories from protein (Fig. 4C). Finally, we delivered FGF21 (0.1 mg/kg, twice daily, IP) and measured intake of the two diets for 7 days. We observed a decrease in total caloric intake during the 7 days of treatment (Fig. 4B), relative to baseline (Fig. 4A), for both treatment groups. We attribute this to mild stress associated with increased handling and the twice-daily injections, together with decreasing novelty of the higher fat diets. Mice treated for 7 days with FGF21 significantly increased the percentage of kilocalories consumed from dietary protein (and decreased the percentage of kilocalories consumed from fat). In contrast, saline-treated mice did not change their macronutrient consumption (Fig. 4D, t test, P < 0.001). FGF21-treated mice lost weight relative to saline-treated controls (Fig. 4E, t test, P < 0.001) despite no effect on caloric intake (Fig. 4B).

Figure 4.

FGF21 increases protein intake and decreases fat intake. Mice were offered simultaneous access to two pelleted diets matched for dietary carbohydrate content (35%), whereas the protein/fat ratio varied (4%:61% or 18%:47%). After establishing a baseline intake for 1 wk, mice were divided into two groups that were well matched for (A) caloric and (C) macronutrient intake. (D) Next, we delivered FGF21 (0.2 mg/kg/d, IP) or saline for 7 d. FGF21 elicited an increase in the consumption of dietary protein, offset by a decrease in fat intake. In agreement with its established role to increase energy expenditure, FGF21 also (E) induced body weight loss, despite (B) no significant effect on caloric intake. Data are shown as mean ± SEM; n = 8 to 9 mice per group. All comparisons were made by a t test: ***P < 0.001, ****P < 0.0001. PRO, protein.

In the third such experiment, mice were offered simultaneous access to two diets that were matched at 18% protein (iso-PRO), together with 60% carbohydrate and 22% fat or 52% carbohydrate and 30% fat (D11051801 and D18061811). Mice were maintained on the two diets for 1 week to establish a baseline, as above. Next, mice were divided into two treatment groups matched for baseline caloric (Fig. 5A) and macronutrient intake, expressed here as percentage of total kilocalories from carbohydrate (Fig. 5C). Finally, we delivered FGF21 (0.1 mg/kg, twice daily, IP) and measured intake of the two diets for 7 days. FGF21 increased caloric intake (Fig. 5B, t test, P = 0.0505), but there was no significant effect of FGF21 on macronutrient selection in this experiment (Fig. 5D, t test, P = 0.3588). Again, consistent with the literature, FGF21-treated mice lost weight relative to saline-treated controls (Fig. 5E, t test, P < 0.01).

Figure 5.

FGF21 does not change macronutrient intake when dietary protein is held constant. In this experiment, mice were offered simultaneous access to two pelleted diets matched for dietary protein content (18%), whereas the carbohydrate/fat ratio varied (60%:22% or 52%:30%). After establishing a baseline intake for 1 wk, mice were divided into two groups that were well matched for (A) caloric and (C) macronutrient intake. Next, we delivered FGF21 (0.2 mg/kg/d, IP) or saline for 7 d. (D) There was no effect of FGF21 on macronutrient selection. In agreement with its established role to increase energy expenditure, FGF21 (E) induced weight loss, (B) despite a tendency to increase caloric intake. Data are shown as mean ± SEM; n = 12 mice per group. All comparisons were made by a t test: **P < 0.01. CHO, carbohydrate.

Comparing the outcomes from this series of two-choice experiments identifies protein as the primary regulated variable in our three-choice study. The results from the iso-FAT and iso-CHO experiments support that FGF21 increases protein intake at the expense of either carbohydrate or fat. The results of the iso-PRO experiment support that FGF21 does not change macronutrient preference when protein (and sucrose) content is held constant. Therefore, we conclude that, with regard to dietary macronutrient preference, FGF21 specifically acts to increase the relative intake of dietary protein.

Macronutrient intake in neuronal Klb-null mice

FGF21 preferentially signals to tissues using a heteromeric cell-surface receptor comprised of the FGF receptor-1 (Fgfr1) in complex with its obligate co-receptor β-klotho (Klb) (36, 37) to signal via RAS/MAPK, phosphatidylinositol 3-kinase/AKT, and protein kinase C (6). Whereas Fgfr1 is broadly distributed, Klb is discretely localized and provides tissue and cell-type specificity for FGF21 action (38). In the brain, Klb is highly expressed in the suprachiasmatic nucleus of the hypothalamus, in the nucleus of the solitary tract, and in the area postrema (39). It is also modestly expressed in the paraventricular nucleus of the hypothalamus, where it directs feeding responses to sweet taste (26). To determine the contribution of neuronal β-klotho to mediate increased protein intake after peripheral FGF21 administration, we performed a two-diet choice experiment using the iso-FAT diets described above. First, we maintained neuron-specific β-klotho-null mice (Klb^ΔSynCre) and littermate controls (Klb^{flox/flox/Cre−}) on diets that were matched at 22% fat, together with 18% protein and 60% carbohydrates or 4% protein and 74% carbohydrates to establish a baseline, as above (D11051801 and D11092301 as in Fig. 3; see Table 2). Next, within each genotype, mice were divided into two treatment groups matched for baseline caloric (Fig. 6A) and macronutrient intake (Fig. 6C). Lastly, we delivered FGF21 (0.1 mg/kg, twice daily, IP) and measured intake of the two diets for 7 days. FGF21 increased total caloric intake [Fig. 6B, two-way ANOVA, P (drug) < 0.05] and there was no interaction with genotype. As expected, however, the effect of FGF21 treatment on macronutrient intake depended on genotype [Fig. 6D, two-way ANOVA, P (treatment × genotype) < 0.05]. Put another way, only the Klb^{flox/flox/Cre−} mice significantly altered diet selection relative to baseline by increasing protein intake in response to FGF21 (Wilcoxon signed-rank test, P < 0.05). There was no significant effect of FGF21 on body weight in this experiment (Fig. 6E), likely because of the smaller group size and the small effect size among mice consuming low-fat diets (34). Lastly, we confirmed the efficacy of our cre-lox recombination by measuring Klb mRNA expression from the mice involved in this experiment. We observed a 59% decrease in Klb expression from whole brain of Klb^ΔSynCre mice compared with the Klb^{flox/flox/Cre−} littermate controls, whereas there was no difference in Klb expression in intrascapular brown adipose or inguinal white adipose tissue (Fig. 6F, t test, P < 0.001). We conclude that FGF21 acts via its receptors in neurons to elicit changes in dietary protein intake. Further studies are needed to identify the critical neuronal populations.

Discussion

In this study, we identify a novel role for the polypeptide hormone FGF21 in the neuroendocrine control of dietary protein intake. First, we find that the previously reported effect of FGF21 to reduce simple sugar intake (20, 21) involves caloric tradeoffs with other available dietary options (Fig. 1). To more clearly define the nature of these tradeoffs, we used a standard three-choice pure macronutrient diet paradigm. In response to FGF21, mice increased consumption of protein while reducing consumption of carbohydrate. There was no effect on fat intake (Fig. 2). Next, to determine whether protein or carbohydrate was the primary regulated nutrient, we employed a series of two-choice experiments to isolate the effect of FGF21 on preference for each of the three macronutrients. Sweetness was well controlled by holding sucrose constant across all of the diets. Under these conditions, we found that FGF21 increased consumption of dietary protein while decreasing the consumption of either carbohydrate or fat (Figs. 3 and 4). However, when protein was held constant between the two-diet options, there was no effect of FGF21 on the relative consumption of carbohydrate vs fat (Fig. 5). Finally, we report that the effect of FGF21 to alter dietary protein intake was dependent on the presence of its co-receptor, β-klotho, in neurons (Fig. 6). Thus, FGF21 is a hepatokine that is secreted into circulation during dietary protein or amino acid restriction, in both rodents and humans (2, 12–15, 17, 23), and it feeds back via the nervous system to increase the relative intake of dietary protein.

The homeostatic regulation of energy balance is essential to survival. Consequently, well-described and robust neuroendocrine pathways control caloric intake to facilitate the defense of total energy stores (40–43). However, food intake is critical not only to supply energy substrates, but also to provide organic building blocks supporting growth (e.g., during development) and/or for dynamic use, maintenance, and repair (e.g., in adulthood). Moreover, although carbohydrates and lipids can be stored as glycogen and fat, there is no easily accessible, inert storage form of protein. Continuous adjustments in feeding behavior are therefore needed to ensure an adequate supply of amino acids for ongoing needs.

Perhaps the strongest evidence for a behavioral control of dietary protein and amino acid homeostasis comes from the multidimensional geometric framework model, which provides a quantitative assessment of macronutrient intake in a free-feeding situation (3, 44–48). The model has been applied across diverse species, including rodents (3) and humans (49), and consistently finds a strong behavioral regulation of dietary protein intake. Moreover, the data support that when the ratio of dietary protein/total energy is low, a drive to consume sufficient protein supersedes the homeostatic control of energy balance. It has been suggested, therefore, that this “protein leverage” may be a driving force contributing to the global rise in obesity [(50, 51), but see Simpson et al. (49)]. If so, defining the underlying mechanisms may provide new opportunities for manipulating food intake to improve metabolic health. Yet there has been little progress defining critical neuroendocrine mediators (1–3, 44, 52).

Amino acids can be directly sensed by neurons that control feeding behavior. For example, diets devoid of one or more essential amino acids are aversive and are rejected by both rodents and humans (53–55). This learned taste aversion, and subsequent hypophagia, is thought to rely on direct neuronal sensing of amino acid depletion or imbalance by the GCN2 kinase in neurons of the anterior piriform cortex (APC) (52), because it is prevented in GCN2-null mice [but see Leib and Knight (48)] by lesioning the APC, or by directly injecting the missing dietary amino acid locally within the APC (56–60). Careful recent studies by Leib and Knight (48) indicate the rapid aversive response to amino acid–depleted diets requires a prior physiologic need; that is, the mice must be previously deprived of one or more individual amino acids, or of total dietary protein, for 2 days prior to behavioral testing. They concluded that this “need-based mechanism” is the primary innate mechanism animals possess to select among diets based on their amino acid and/or protein content. The present data position FGF21 as a likely neuroendocrine signal of this “prior physiologic need” and suggest future research to determine how/whether the FGF21 signal interacts with direct amino acid sensing in the APC.

Local excess of certain amino acids can also be sensed by neurons to alter feeding behavior. Rodents maintained on high-protein diets are hypophagic, and this is recapitulated by dietary supplementation with l-leucine (3, 52, 61). Moreover, injecting a leucine-containing mixture of amino acids, or l-leucine alone, into the brain of fasted rodents reduces total energy intake during the subsequent refeeding period (62–66). The leucine-induced hypophagia is thought to depend in part on direct activation of mTORC1 and AMP-activated protein kinase (62, 64, 66, 67) and in part on local branched chain amino acid catabolism (67) in key neuronal populations of the medial basal hypothalamus and brainstem. The relevance of brain leucine signaling to a specific regulation of dietary protein intake vs calories per se, however, is unknown. Notably, brain leucine sensing is most effective to reduce total caloric intake in fasted rodents [reviewed in Heeley and Blouet (52)] or in rodents fed ad libitum low-protein diet (68). Thus, any potential effect of hypothalamic or brainstem leucine-sensing on macronutrient selection may also require a signal of “physiologic need” to shift diet preferences. Future studies will investigate whether direct central nervous system infusion of amino acids and/or l-leucine is sufficient to blunt the FGF21-induced increase of dietary protein intake.

The current study has several limitations. First, we used casein as the sole source of dietary protein in all of the experiments. This logistical decision allows for continuity with the original literature describing FGF21 as a signal of dietary protein restriction (15–17), and it maintains internal consistency to facilitate comparisons across all of the choice paradigms. Because deficits in a number of essential (13, 14) and nonessential (12, 15) amino acids are known to increase FGF21 secretion, we expect that this neuroendocrine system is responding to “dietary protein” in the generic sense. Thus, we anticipate that the homeostatic model supported by the current studies will generalize to other protein sources—but certainly this is an important empirical question that will be addressed in future work. Likewise, it remains uncertain to what extent FGF21 contributes to dietary preferences for specific amino acids, but we hypothesize that it acts as a “signal of physiologic need” to modulate the behavioral response to direct individual amino acid sensing, as discussed above. Second, the experiment with neuron-specific knockout mice did not include Klb^Cre+/flox− controls. Although we consider it unlikely, it is possible the expression of Cre recombinase alone interferes with FGF21 action, independent of β-klotho. This potential confound will likely be resolved by future use of nucleus-specific viral approaches to delineate the intracellular signaling mechanisms and specific neuroanatomic circuits mediating the observed behaviors. Third, the current experiments all involved exogenous administration of recombinant FGF21. Further research using FGF21-null mice will be necessary to clarify the physiological relevance of these findings. Lastly, the current findings include only male mice. In light of the divergent energetic and nutritional costs of reproduction in male and female mammals, it will be especially important to compare the role of FGF21 to control dietary protein intake in females and males.

Conclusions

Because it is the primary anatomical site of amino acid catabolism and biosynthesis, the liver is well situated to function as a “canary in the coal mine,” signaling physiologic need to the brain during times of protein deficit, and eliciting behaviors that replenish an adequate supply for cells throughout the body. Together with previous reports that define FGF21 as an endocrine signal of dietary amino acid and/or protein restriction (2, 12–15, 17, 23), the present findings support a critical liver → nervous system mechanism facilitating the control of dietary protein intake via FGF21. Amino acids are one of the main building blocks of life and are used not only as fuels, but as precursors for hormones, neurotransmitters, enzymes, and structural proteins throughout the body (69). Therefore the quantity and quality of dietary protein intake has a critical influence on diverse aspects of organismal health (70), including development (71, 72), growth (73), metabolism (1), performance (71, 72), immune function (74, 75), cognition (71, 72), and aging (76). Identifying the critical role of FGF21 signaling in the nervous system to control dietary protein intake represents a first step toward new opportunities for manipulating macronutrient intake to benefit various aspects of health and longevity.

Glossary

Abbreviations:

APC

anterior piriform cortex

bw

body weight

FGF21

fibroblast growth factor-21

Fgfr1

fibroblast growth factor receptor-1

hFGF21

human fibroblast growth factor-21

P/C

protein/carbohydrate