Heme-Regulated eIF2α Kinase Modulates Hepatic FGF21 and Is Activated by PPARβ/δ Deficiency

Fibroblast growth factor 21 (FGF21), a member of the FGF family, is a hormone with a wide range of endocrine and autocrine actions on carbohydrate and lipid metabolism (1) and is considered a novel therapeutic target for the treatment of nonalcoholic fatty liver disease (NAFLD), insulin resistance, and type 2 diabetes. Several studies in genetic and diet-induced models of obesity have demonstrated that FGF21 administration ameliorates a large number of metabolic parameters (2–5), including plasma glucose and triglyceride levels and hepatic steatosis. However, despite these beneficial effects of administered FGF21, serum FGF21 levels are paradoxically increased in animal models of obesity (6). Likewise, circulating FGF21 levels are elevated in patients with obesity, hypertriglyceridemia, type 2 diabetes, or NAFLD (7–9).

The pleiotropic effects of FGF21 on target tissues, including adipose tissue, β-cells, and liver (1,2,10), are mediated by its binding to FGF receptors in a β-klotho–dependent manner (11–13). β-Klotho is almost exclusively expressed in liver, adipose tissue, and pancreas (13), which might explain why these specific tissues are the predominant sites of action of FGF21. Circulating FGF21 is liver derived (14), and hepatic FGF21 expression is upregulated, following extended periods of fasting, by peroxisome proliferator–activated receptor (PPAR) α, a nuclear receptor that induces the expression of numerous genes involved in mitochondrial fatty acid (FA) oxidation (15). In recent years, FGF21 has been reported to be regulated by additional transcription factors, including CREBPH (16) and retinoic acid receptor–related orphan receptor (ROR) α (17). Moreover, hepatic FGF21 is negatively regulated by the PPARγ coactivator 1α (PGC-1α) by modulating the heme-sensing nuclear receptor REV-ERBα, which acts as a transcriptional repressor through its binding to ROR response elements (18). Impaired mitochondrial oxidative phosphorylation (mtOXPHOS) is also responsible for Fgf21 induction by activating transcription factor (ATF) 4. ATF4 is a transcriptional effector of the protein kinase R–like endoplasmic reticulum (ER) kinase branch of the ER stress/unfolded protein response pathway, which is essential for Fgf21-induced expression (19). However, although circulating FGF21 has been reported to be increased following treatment with PPARβ/δ agonists (20), little is known about the effects of PPARβ/δ on FGF21 regulation in liver. PPARβ/δ is a ligand-activated transcription factor involved in the regulation of glucose and lipid homeostasis (21) and has been proposed as a therapeutic target for the treatment of the metabolic syndrome (22). Thus, genetic manipulations of PPARβ/δ as well as it activation by agonists attenuates dyslipidemia and hyperglycemia, improves whole-body insulin sensitivity, and prevents diet-induced obesity (23).

In this study, we examined the effects of PPARβ/δ deficiency on hepatic Fgf21 expression. Pparβ/δ-null mice showed enhanced hepatic Fgf21 expression, which appears to depend on a reduction in PGC-1α levels and the subsequent reduction in hemin levels that finally activate the heme-regulated eukaryotic translation initiation factor 2α (eIF2α) kinase (HRI). This kinase phosphorylates eIF2α, which in turn increases ATF4 levels, resulting in enhanced Fgf21 expression. Overall, the findings point to HRI as a new therapeutic target for regulating Fgf21 expression and metabolic dysregulation.

N,N′-diarylureas 1-(benzo[d][1,2,3]thiadiazol-6-yl)-3-(3,4-dichlorophenyl)urea (BTdCPU) or 1-(benzo[d][1,2,3]thiadiazol-6-yl)-3-(4-chloro-3-[trifluoromethyl]phenyl)urea (BTCtFPU), were synthesized as previously described (Supplementary Data) (24).

Male Pparβ/δ knockout (Pparβ/δ^−/−) mice and their wild-type littermates (Pparβ/δ^+/+) with the same genetic background (C57BL/6 × 129/SV) (25) and initial weight of 20–25 g were fed a standard diet. Lipid-containing media were prepared by conjugation of palmitic acid with FA-free BSA as previously described (26). Primary mouse hepatocytes were isolated from nonfasting male C57BL/6 mice (10–12 weeks old) by perfusion with collagenase as described elsewhere (27).

Male mice (Harlan Ibérica SA, Barcelona, Spain) were randomly distributed into three experimental groups (n = 8 each): standard diet, Western-type high-fat diet (HFD) (35% fat by weight, 58% kcal from fat; Harlan Ibérica SA) plus one daily oral gavage of vehicle (0.5% weight for volume carboxymethylcellulose), and HFD plus one daily oral dose of 3 mg ⋅ kg⁻¹ ⋅ day⁻¹ of the PPARβ/δ agonist GW501516 dissolved in the vehicle (volume administered 1 mL ⋅ kg⁻¹). In a second study, male mice received one daily intraperitoneal administration of DMSO (vehicle) or BTdCPU (70 mg ⋅ kg⁻¹ ⋅ day⁻¹) for 7 days. In a third study, male mice were randomly distributed into two experimental groups: standard chow (n = 8) or HFD (n = 16) for 3 weeks. Mice fed standard chow and one-half of the mice fed the HFD received one daily intraperitoneal administration of DMSO (vehicle) for the last week. The rest of the mice fed the HFD received one daily intraperitoneal administration of BTdCPU (70 mg ⋅ kg⁻¹ ⋅ day⁻¹) for the last week. In a fourth study, male knockout (Fgf21^−/−) mice (B6N;129S5-Fgf21^tm1Lex/Mmcd obtained from the Mutant Mouse Regional Resource Centre) and their wild-type littermates (Fgf21^+/+) were treated as described in the third study.

The relative levels of specific mRNAs were assessed by real-time RT-PCR as previously described (26). Primer sequences used for real-time RT-PCR are displayed in Supplementary Table 1.

Western blot analyses were performed as previously indicated (26).

Heme content in liver was quantified by measuring the oxidized version of the protein hemin by using an enzymatic assay kit (Hemin Assay Kit; Sigma-Aldrich).

Lipid accumulation in hepatocytes was assessed by Oil Red O (ORO) staining as previously reported (28).

Results are expressed as mean ± SD. Significant differences were established by one-way ANOVA by using the GraphPad Instat program (GraphPad version 5.01; GraphPad Software, La Jolla, CA). When significant variations were found by one-way ANOVA, the Tukey-Kramer multiple comparison posttest was performed. Differences were considered significant at P < 0.05.

PPARβ/δ^−/− mice displayed higher hepatic Fgf21 mRNA levels than wild-type littermates (threefold induction, P < 0.001) (Fig. 1A). Consistent with circulating FGF21 being liver derived (14), we found increased plasma FGF21 levels in Pparβ/δ-deficient mice (2.6-fold increase, P < 0.001) (Fig. 1B) and reduced expression of one of the receptor and coreceptor pairs used by FGF21, Fgfr1c (49% reduction, P < 0.05) and β-klotho (57% reduction, P < 0.001) (Fig. 1C). Carnitine palmitoyltransferase 1A (Cpt-1a), 3-hydroxy-3-methylglutaryl-CoA synthase 2 (Hmgcs2) (29), 3-β-hydroxysteroid dehydrogenase type 5 (Hsd3b5) (30), and major urinary protein 1 (Mup1) (30) are FGF21-responsive genes. In accordance with increased FGF21 plasma levels, hepatic expression of Cpt-1a and Hmgcs2 was increased, whereas the expression of Hsd3b5 and the mRNA and protein levels of MUP1 were decreased in Pparβ/δ^−/− mice (Fig. 1D and Supplementary Fig. 1A). In agreement with the reported observation of increased glucose uptake in adipocytes through enhanced Glut1 expression by FGF21 secreted by the liver (31), the expression of this glucose transporter was higher in white adipose tissue of Pparβ/δ^−/− mice than in the wild-type animals (Fig. 1E). In contrast to the liver, the expression of Fgf21 in white adipose tissue was not significantly increased in Pparβ/δ-deficient mice (Fig. 1E), pointing to a tissue-specific effect. Because increased serum free fatty acids (32) and glucose levels (33) are two important stimuli that upregulate Fgf21 expression in liver, we measured their levels in serum of Pparβ/δ^−/− and wild-type mice. No differences were observed in serum free fatty acids and glucose levels (Supplementary Fig. 1B and C), suggesting that they were not involved in the reported increase in Fgf21 expression. In addition, the expression levels of Pparα and its target genes Acox and Mcad were not significantly increased, rendering it unlikely that PPARα is involved in the increase in Fgf21 expression in these mice (Supplementary Fig. 1D). Pparβ/δ-null mice did not present upregulation of Chop, Orp150, or Atf3 expression (Supplementary Fig. 1E) or phosphorylated (phospho)-IRE1α and BiP levels (Supplementary Fig. 1F), suggesting that ER stress was not the stimulus responsible for the increase in Fgf21 expression. Primary hepatocytes transfected with either control small interfering RNA (siRNA) or Pparβ/δ siRNA showed a significant reduction (71%, P < 0.001) in this transcription factor (Supplementary Fig. 1G) and a significant increase in Fgf21 gene expression (Fig. 1F), confirming that downregulation of Pparβ/δ in hepatocytes raises Fgf21 expression.

Because AMPK (34) and SIRT1 (35) stimulate Fgf21 expression, we explored the levels of these proteins in Pparβ/δ^−/− mice. No changes were observed either in phosphorylated or total AMPK protein levels in Pparβ/δ-deficient mice compared with wild-type animals, whereas in accordance with the reported regulation of SIRT1 by PPARβ/δ (36), the protein levels of SIRT1 were reduced in Pparβ/δ^−/− mice (Fig. 2A). Next, we focused on PGC-1α because this transcriptional coactivator negatively regulates hepatic levels of Fgf21 (18). Pparβ/δ-null mice showed reduced nuclear PGC-1α protein levels compared with wild-type mice (Fig. 2B). Because the reduction in PGC-1α upregulates Fgf21 mRNA levels by decreasing the expression of the transcriptional repressor Rev-Erbα (18), we measured its mRNA and protein levels. Pparβ/δ^−/− mice exhibited lower Rev-Erbα expression and protein levels than wild-type mice (Fig. 2B and C), which is consistent with the upregulation of Rev-Erbα–repressed Bmal1 in Pparβ/δ-null mice (Fig. 2C). Because PGC-1α stimulates the expression of the Nrf-1 gene (37), its reduction was in accordance with the reduction in hepatic PGC-1α (Fig. 2B). mtOXPHOS is regulated by both PGC-1α (38) and REV-ERBα (39), and their decrease is consistent with the reduction observed in the protein levels of several members of mtOXPHOS in the livers of Pparβ/δ^−/− mice (Fig. 2D). Impaired mtOXPHOS is reportedly responsible for FGF21 induction through activation of the eIF2α-ATF4 pathway (19). Although this pathway is activated by ER stress, the increase in phosphorylated eIF2α and activation of its downstream ATF4 signaling pathway can occur independently of ER stress because eIF2α can also be phosphorylated by other kinases, including HRI (40). Of note, HRI is activated by heme deprivation (41), and PGC-1α is an important regulator of heme in liver cells because it coactivates NRF-1 and other transcription factors that increase the expression of Alas1, the rate-limiting enzyme in heme biosynthesis. In addition, the repressive activity of REV-ERBα on Fgf21 expression is potentiated by the binding of its ligand heme (18). Given that Pparβ/δ^−/− mice exhibited reduced PGC-1α and NRF-1 protein levels, we assessed heme content levels by measuring the oxidized form of this protein hemin. Pparβ/δ^−/− mice showed reduced levels of hemin compared with wild-type mice (Fig. 2E) and increased HRI levels in Pparβ/δ^−/− mice compared with wild-type littermates (Fig. 2F). In agreement with the increase in HRI, levels of the downstream proteins of this pathway, phospho-eIF2α and ATF4, were also upregulated (Fig. 2F).

In accordance with the findings observed in the livers of Pparβ/δ-deficient mice, siRNA knockdown of Pparβ/δ in primary hepatocytes led to enhanced protein levels of HRI, phospho-eIF2α, and ATF4 and reduced levels of MUP1 (Fig. 3A). Transfection of primary hepatocytes with siRNA against Hri caused a significant reduction in hemin levels and in the expression of Fgf21, Atf4, and Chop, the latter being a direct ATF4 transcriptional target (42) used as a marker of the activation of the eIF2α-ATF4 pathway following activation of HRI (Fig. 3B–E) (24). Likewise, the protein levels of phosphorylated eIF2α, ATF4, and CHOP were decreased, whereas the protein levels of MUP1 were increased (Fig. 3F), confirming through a genetic approach that HRI controls Fgf21 expression in hepatocytes.

Hepatic Fgf21 expression increases in response to ER stressors in liver, where it seems to display an adaptive response to these stimuli (43). In fact, exogenous administration of FGF21 alleviates the tunicamycin-induced eIF2α-ATF4-CHOP pathway, whereas it shows an insignificant effect on the IRE1α-XBP1s pathway (43). We hypothesized that the increase in FGF21 levels in Pparβ/δ-deficient mice may protect the liver from ER stress, which is consistent with the fact that in contrast to the liver, skeletal muscle of these mice showed increased expression of ER stress markers (44). To test this, wild-type and Pparβ/δ^−/− mice were treated with the ER stressor tunicamycin for 24 h. Wild-type mice treated with tunicamycin exhibited a ninefold increase (P < 0.001) in Fgf21 expression, and this increase was higher (∼21-fold induction, P < 0.001 vs. tunicamycin-treated wild-type mice) in livers of Pparβ/δ^−/− mice (Fig. 4A). A similar effect was observed in plasma FGF21 levels (Fig. 4B). The mRNA and protein levels of the ER stress marker BiP were higher in the livers of tunicamycin-treated Pparβ/δ-deficient mice than in tunicamycin-treated wild-type mice, suggesting that ER stress is exacerbated in the former (Supplementary Figs. 2A and 4C) probably as a result of the increase in ER stress pathways, such as the IRE1α-XBP1s, which are not inhibited by FGF21. The higher levels of FGF21 in vehicle-treated Pparβ/δ^−/− mice compared with vehicle-treated wild-type mice were accompanied by a reduction in phospho-eIF2α protein levels (Fig. 4C). This was especially marked in tunicamycin-treated Pparβ/δ^−/− mice compared with tunicamycin-treated wild-type mice (Fig. 4C), suggesting that the higher levels of FGF21 in Pparβ/δ^−/− mice inhibited the phosphorylation of eIF2α as previously described (43). To demonstrate this more clearly, we used an FGF21-neutralizing antibody. In wild-type mice, treatment with the FGF21-neutralizing antibody for 14 h did not significantly affect eIF2α phosphorylation compared with IgG-treated mice (Supplementary Fig. 2B). In contrast, when Pparβ/δ^−/− mice were treated with the FGF21-neutralizing antibody for the same amount of time, a significant increase was observed in phospho-eIF2α levels (Fig. 4D). Pparβ/δ^−/− mice treated for 14 h with tunicamycin and IgG showed a reduction in the levels of phospho-eIF2α, suggesting that the additional increase in FGF21 levels caused by tunicamycin treatment was responsible for this effect. In line with this, injection of the FGF21-neutralizing antibody raised phospho-eIF2α levels (Fig. 4D). The protein levels of ATF4 also showed an increase following treatment with the FGF21-neutralizing antibody (Fig. 4D). Through its negative action on eIF2α and ATF4, FGF21 can activate a negative feedback loop that reduces Fgf21 expression (43). Consistently, administration of the FGF21-neutralizing antibody raised the mRNA levels of Fgf21 (Fig. 4E). The increase in BiP and Atf3 expression and BiP protein levels confirmed that tunicamycin treatment for 14 h results in ER stress (Supplementary Fig. 2C–E).

Next, we explored whether the described mechanisms also operated in human HepG2 cells. Treatment with Pparβ/δ siRNA led to a significant increase in phosphorylated eIF2α and ATF4 levels (Fig. 4F), confirming that Pparβ/δ deficiency activates the eIF2α-ATF4 pathway. In addition, in the presence of the FGF21-neutralizing antibody, the increase in phospho-eIF2α and especially in ATF4 was exacerbated compared with IgG-treated cells. We then examined the effects of the FGF21-neutralizing antibody on tunicamycin-treated cells. In these cells, the FGF21-neutralizing antibody increased the levels of phosho-eIF2α and ATF4 protein compared with IgG-treated cells (Fig. 4F). Overall, these findings confirm that an increase in FGF21 levels in Pparβ/δ^−/− mice prevents an increase in the eIF2α-ATF4 pathway, alleviating part of the ER stress process in liver.

Exposure to an HFD increases Fgf21 expression (10), and we have previously reported that exposure to an HFD reduces hepatic mRNA levels of Pgc-1α (45). This suggests that exposure to an HFD might activate the HRI-eIF2α-ATF4 pathway and contribute to an increase in hepatic Fgf21 expression. In mice exposed to an HFD for 3 weeks in the presence or absence of the PPARβ/δ activator GW501516, we observed that the HFD increased hepatic Fgf21 expression by eightfold (P < 0.05), whereas this increase was prevented by GW501516 (Fig. 5A). Consistent with the changes in Fgf21 expression, feeding with an HFD reduced MUP1 protein levels, whereas the PPARβ/δ agonist prevented the decrease caused by the HFD (Fig. 5B). No change was observed in the protein levels of the ER stress marker BiP (Supplementary Fig. 3A), rendering it unlikely that ER stress might be responsible for the increase in Fgf21 expression caused by an HFD. Of note, the HFD strongly reduced the protein levels of PPARβ/δ, PGC-1α, and REV-ERBα, whereas ATF4 protein levels were increased (Fig. 5C). In contrast, these changes were abolished in mice fed the HFD and treated with GW501516. Furthermore, the reduction in PPARβ/δ, PGC-1α, and REV-ERBα was accompanied by a reduction in hemin levels (Supplementary Fig. 3B) and an increase in HRI levels in mice fed the HFD, whereas this increase was blunted following drug treatment (Supplementary Fig. 3B and Fig. 5D). To confirm whether FAs were the HFD component responsible for the changes observed in the in vivo study, we exposed human Huh-7 hepatocytes to the saturated fatty acid (SFA) palmitate. Cells incubated with palmitate showed a huge increase in FGF21 and ATF4 expression (Fig. 5E). Palmitate also elicited a reduction in PPARβ/δ, PGC-1α, and REV-ERBα and a subsequent increase in HRI, phospho-eIF2α, and ATF4 protein levels (Fig. 5F). These findings suggest that by reducing PPARβ/δ and PGC-1α, SFAs lead to activation of the HRI-eIF2α-ATF4 pathway and a subsequent increase in FGF21 expression.

Given that HRI regulates FGF21, we next explored its suitability as a pharmacological target to modulate hepatic FGF21 expression. For this purpose, we used two N,N′-diarylureas, BTdCPU and BTCtFPU, which are HRI activators, thereby causing the phosphorylation of eIF2α and the increased expression of the transcription factor ATF4 (24). Exposure of human Huh-7 hepatocytes to 10 μmol/L of either BTdCPU or BTCtFPU for 24 h strongly increased Fgf21 and Atf4 mRNA levels (Fig. 6A and B). The increase in the expression of CHOP (Fig. 6C), a direct ATF4 transcriptional target (42), confirmed that this pathway was activated by the N,N′-diarylureas. These compounds did not increase PPARα, its target genes, BiP mRNA (Supplementary Fig. 4A–D), or BiP protein levels (Fig. 6D), suggesting that these mechanisms were not involved in Fgf21 upregulation. In line with the reported activation of HRI by these N,N′-diarylurea compounds, we observed increased protein levels of phospho-eIF2α and ATF4, especially following treatment with BTdCPU (Fig. 6D). Treatment of mice with BTdCPU for 1 week also increased hepatic Fgf21 expression (Fig. 6E) without changes in BiP protein levels, whereas phospho-eIF2α and ATF4 were increased (Fig. 6F). Moreover, consistent with the increased expression of Fgf21, Glut1 mRNA levels were higher in the white adipose tissue of mice treated with BTdCPU (Fig. 6G).

Next, we explored the effects of BTdCPU on human Huh-7 hepatocytes exposed to the SFA palmitate. Cells exposed to this FA showed increased FGF21, ATF4, and CHOP mRNA levels, and when cells were coincubated with the FA and BTdCPU, a significantly higher increase was observed in the expression of these three genes (Fig. 7A). Of note, when Huh-7 cells were exposed to palmitate, a high accumulation of triglycerides was observed in the cells, as demonstrated by ORO staining, but this accumulation was prevented in the presence of the BTdCPU compound, and this effect was attenuated in the presence of the FGF21-neutralizing antibody (Fig. 7B and Supplementary Fig. 5A). Similar effects were observed in mouse Hepa-1c1c7 hepatocytes (Supplementary Fig. 5B). Likewise, this compound partially restored the reduction in insulin-stimulated Akt phosphorylation caused by palmitate (Fig. 7C and Supplementary Fig. 5C), showing that this drug treatment prevents SFA-induced attenuation of the insulin signaling pathway. Next, we examined the effects of the BTdCPU compound in mice fed an HFD. BTdCPU administration prevented the glucose intolerance caused by HFD feeding (Fig. 7D), increased Glut1 expression in white adipose tissue (Fig. 7E), and prevented hepatic steatosis as demonstrated by ORO and hematoxylin-eosin staining and quantification of hepatic triglyceride levels (Fig. 7F and G). Drug treatment did not affect the expression of PPARα-target genes (Supplementary Fig. 5D and E), whereas the expression of the lipogenic gene fatty acid synthase (Fas) and stearoyl-CoA desaturase 1 (Scd1) was reduced (Supplementary Fig. 5F and G), which is consistent with the reported reduction of these genes by FGF21 (46).

To demonstrate more clearly that the improvement in glucose tolerance and hepatic steatosis caused by the administration of the HRI activator in mice fed an HFD depends on FGF21, we used wild-type and Fgf21-null mice. The reduction in glucose intolerance observed in wild-type mice fed an HFD caused by administration of the HRI activator was completely abolished in Fgf21-null mice (Fig. 8A), suggesting that in the absence of FGF21, the effect of the HRI activator on glucose tolerance is lost. Similarly, the reduction in hepatic triglyceride accumulation in wild-type mice fed an HFD caused by BTdCPU administration was suppressed in Fgf21-null mice, as demonstrated by the analysis of the levels of triglycerides and the ORO and hematoxylin-eosin staining (Fig. 8B and C).

FGF21 has emerged as an important regulator of glucose and lipid metabolism and, hence, is a promising agent for the treatment of obesity, NAFLD, insulin resistance, and type 2 diabetes . Unraveling the mechanisms that regulate Fgf21 expression in liver may provide pharmacological targets for modulating its expression to prevent metabolic diseases. In this study, we demonstrate that Pparβ/δ-null mice show enhanced hepatic Fgf21 expression and circulating levels of this hormone. Pparβ/δ deficiency caused a reduction in transcriptional coactivator PGC-1α levels that resulted in a reduction in hemin levels and the subsequent activation of HRI and the eIF2α-ATF4 pathway, which is essential for Fgf21-induced expression (Fig. 8D). Likewise, activation of this pathway could be an additional mechanism that contributes to the increase in Fgf21 expression under lipid overload conditions. The findings reveal HRI as a regulator of Fgf21 expression and, consistent with this, knockdown of Hri or its pharmacological activation regulate Fgf21 expression in hepatocytes.

The finding that Pparβ/δ deficiency upregulates Fgf21 expression in liver was unexpected because it has previously been reported that pharmacological activation of this nuclear receptor increases FGF21 levels (20). The latter is consistent with the discovery of two putative peroxisome proliferator response elements in the mouse and human FGF21 promoters (47) and that both PPARα and PPARγ activators increase FGF21 levels in hepatocytes and adipocytes, respectively. However, genetically reduced PGC-1α, a transcriptional coactivator that controls the expression and activity of PPARs and is regulated by these transcription factors, likewise results in increased FGF21 levels (18). In fact, reduced hepatic PGC-1α upregulates Fgf21 expression by reducing REV-ERBα and the levels of its ligand heme. This alleviated the repressive activity of REV-ERBα on the Fgf21 promoter, leading to enhanced Fgf21 expression. In the current study, we show that Pparβ/δ deficiency results in a reduction in PGC-1α levels that leads to a reduction in REV-ERBα and hemin, the oxidized form of heme, suggesting that this mechanism can also contribute to an increase in Fgf21 expression in the liver of Pparβ/δ-null mice. In addition, the current findings reveal a new biochemical pathway that can contribute to the enhanced expression of Fgf21 in the liver of these mice. Indeed, the reduction in heme can increase HRI levels (42), which in turn can activate the eIF2α-ATF4 pathway where the increase in ATF4 is one of the most important transcription factors that upregulate Fgf21 expression. In fact, overexpression of ATF4 has been reported to induce Fgf21 expression (19), whereas knockdown of Atf4 reduces basal and ER stress–induced Fgf21 expression (48). Therefore, the current findings indicate that Pparβ/δ deficiency in liver results in increased levels of ATF4 and a subsequent increase in Fgf21 expression. The increase in ATF4 levels seems to be a crucial step in Fgf21 upregulation because this transcription factor is involved in the increase of this hormone in impaired mtOXPHOS induced by autophagy deficiency (19), ER stress (48), metformin-induced inhibition of the mitochondrial respiratory chain (49), and now PPARβ/δ-PGC-1α deficiency as shown in the current study.

As has been demonstrated by injection of recombinant FGF21 (43), the eIF2α-ATF4 pathway is activated by ER stress, which leads to enhanced Fgf21 expression that in turn alleviates ER stress by suppressing this eIF2α-ATF4 pathway. By using a neutralizing antibody against FGF21, we confirm that this negative feedback is also activated by FGF21.

Of note, the reduction in PPARβ/δ and PGC-1α following exposure to an HFD might also contribute to increased Fgf21 expression. In fact, livers of mice exposed to an HFD or hepatocytes exposed to the SFA palmitate showed reduced PPARβ/δ and PGC-1α levels, resulting in an increase in the levels of HRI and subsequent activation of the eIF2α-ATF4 pathway. These findings suggest that activation of this pathway could also contribute to the increase in FGF21 in those conditions associated with fat overfeeding, such as obesity, type 2 diabetes, and NAFLD, which show increased levels of this hormone (7–9). In contrast, activation of PPARβ/δ under conditions of lipid overload recovers PGC-1α levels and attenuates the eIF2α-ATF4 pathway, restoring Fgf21 expression.

The regulation of FGF21 by HRI provides a target for regulation of the levels of this hormone to prevent or treat metabolic diseases, including NAFLD, insulin resistance, and diabetes. N,N′-diarylureas are activators of HRI that cause phosphorylation of eIF2α, which were developed as anticancer drugs (24). Here, we show that these drugs can increase Fgf21 expression in hepatocytes and that their administration to mice fed an HFD prevents the accumulation of triglycerides in liver and improves glucose intolerance. This is consistent with liver-specific Fgf21 knockout mice fed an HFD displaying glucose intolerance and increased hepatic lipid accumulation (14). Moreover, we demonstrate that the improvement in glucose tolerance and hepatic steatosis caused by the HRI activator BTdCPU in mice fed an HFD depends on FGF21 because it was not observed in Fgf21-null mice.

In summary, we propose that Pparβ/δ deficiency results in a reduction in hepatic PGC-1α and hemin levels that in turn increase HRI levels, leading to activation of the eIF2α-ATF4 pathway and a subsequent increase in Fgf21 expression. Moreover, pharmacological activation of HRI increases Fgf21 expression, and treatment with a drug that activates HRI prevents glucose intolerance and hepatic triglyceride accumulation in mice fed an HFD. These findings point to HRI as a target for the treatment of metabolic diseases by modulating FGF21 levels.

Acknowledgments. M.Z. thanks Bijan Almassian, Massachusetts College of Pharmacy and Health Sciences. R.L. thanks the Spanish Ministry of Education for a PhD grant (FPU program). M.B.-X. thanks the Institute of Biomedicine of the University de Barcelona for a PhD grant.

Funding. This study was partly supported by funds from the Spanish Ministry of the Economy and Competitiveness (SAF2015-65267-R to Á.M.V., SAF2014-55725 to F.V., and SAF2012-30708 and SAF2015-64146-R to M.V.-C.) and the European Union European Regional Development Fund. CIBERDEM and CIBEROBN are Instituto de Salud Carlos III Health Institute projects. T.Q.-L. is supported by a CONACyT (National Council for Science and Technology in Mexico) PhD scholarship. W.W. is supported by start-up grants from the Lee Kong Chian School of Medicine, Nanyang Technological University, and by the Région Midi-Pyrénées, France.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. M.Z. contributed to the study design, experiments, review of the results, and editing and final approval of the manuscript. E.B., X.P., V.P., Á.G.-R., Á.M.V., and T.Q.-L. contributed to the experiments and editing and final approval of the manuscript. R.L., M.B.-X., E.P., C.E., and S.V. synthesized the compounds and contributed to the editing and final approval of the manuscript. F.V. and W.W. contributed to the data analysis, review of the results, and editing and final approval of the manuscript. M.V.-C. contributed to the study design, experiments, review of the results, and writing, editing, and final approval of the manuscript. M.V.-C. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.