Role of GCN2-Independent Signaling Through a Noncanonical PERK/NRF2 Pathway in the Physiological Responses to Dietary Methionine Restriction

Dietary methionine restriction (MR) increases life span and produces a profile of physiological responses that confer resistance to metabolic disease (1–6), but the anatomical organization and molecular identity of the sensing and translational mechanisms are unknown. General control nonderepressible 2 (GCN2) is a potential mediator because it is activated by limited essential amino acid (EAA) availability. Activated GCN2 phosphorylates the α-subunit of eukaryotic initiation factor 2 (eIF2), altering gene-specific translation and initiating a cytoprotective transcriptional program collectively described as the integrated stress response (ISR) (7,8). Activating transcription factor 4 (ATF4), a basic leucine zipper transcription factor with important roles in glucose homeostasis, lipid metabolism, and energy expenditure (EE), is among the gene products preferentially translated after eIF2 phosphorylation (9,10). During amino acid insufficiency, ATF4 coordinates induction of hepatic fibroblast growth factor 21 (FGF21) and other metabolic genes (11,12). However, the relative importance of these individual signaling components (e.g., GCN2, eIF2, ATF4) in mediating the complex physiological responses to EAA restriction is largely unknown.

Loss-of-function models have been used to assess the role of GCN2 in the transcriptional, physiological, and behavioral responses to dietary EAA deprivation. Soon after their introduction, activation of GCN2 within the anterior piriform cortex triggers an acute aversive feeding response that is absent in mice lacking Gcn2 (13,14). Leucine deprivation (LD) also produces a chronic anorexigenic response that is not compromised in Gcn2^−/− mice (15,16). However, downregulation of hepatic lipogenic genes by LD does require GCN2 (15). Together, these findings illustrate that responses to dietary LD are mediated through a combination of GCN2-dependent and -independent mechanisms. A comparison of the responses to LD and MR reveals shared features but also distinct differences, particularly, their opposite effects on food intake and lipogenic gene expression in adipose tissue (17). It remains to be established whether these differences are EAA specific or simply reflect different degrees of restriction, irrespective of the EAA being restricted. In either case, the importance of GCN2 in mediating the biological responses to dietary MR is unknown.

In response to different cellular stresses, three other eIF2 kinases participate in activation of the ISR (18–21). Several reports identify auxiliary or cooperative action between GCN2 and protein kinase R–like endoplasmic reticulum (ER) kinase (PERK) in phosphorylating eIF2 and translational control (22–24). However, the transcriptional profiles activated by each kinase are distinct (20). PERK, IRE-1, and ATF6 are all activated by accumulation of unfolded proteins in the ER, and each induces a different component of the unfolded protein response (UPR). The UPR is a multipart stress response that includes the ISR, and unresolved ER stress has been implicated in metabolic disorders associated with obesity and diabetes (25). While dietary LD does not activate PERK or the UPR, activation of the UPR does engage GCN2 in cell-cycle control and cellular fate during metabolic stress (22,24).

Our work comparing low-protein diets with dietary MR proposed the interesting possibility that hepatic detection of reduced EAAs from both diets produced integrated effects on energy balance and insulin sensitivity by inducing hepatic FGF21 (26,27). Although the acute effects of protein restriction on FGF21 and energy balance appear to be GCN2 dependent (26), we report here that GCN2 is not required for MR-dependent induction of hepatic FGF21, increased EE, or enhancement of insulin sensitivity. Instead, we propose that MR is activating the ISR and antioxidant response program in liver via a noncanonical PERK/nuclear respiratory factor 2 (NRF2) pathway that effectively senses and translates the metabolic responses to MR in the absence of GCN2.

Four-week-old male wild-type (WT) (n = 16) and Gcn2^−/− (n = 19) C57BL/6J mice were obtained from The Jackson Laboratory (Bar Harbor, ME), acclimated to a control (CON) diet for 1 week prior to phenotyping using a Comprehensive Laboratory Animal Monitoring System (Columbus Instruments, Columbus, OH). Mice were acclimated to the chambers overnight prior to measurements of O₂ consumption and CO₂ production at 16-min intervals for 18 days. All mice received the CON diet for 4 days prior to switching half the mice to the MR diet. Respiratory exchange ratio (RER) was calculated as the ratio of VCO₂ produced to VO₂ consumed. EE was calculated as follows: VO₂ * (3.815 + [1.232 * RER]) * 4.1868. Thereafter, food intake, body weights, and composition were measured weekly for 12 weeks. The mice were returned to the Oxymax for 4 days for measurement of EE, RERs, and activity. Thereafter, mice were killed and tissues were harvested.

The diets were formulated as extruded pellets. The energy content of both CON and MR diets was 15.96 kJ/g, with 18.9% of energy from fat (corn oil), 64.9% from carbohydrate, and 14.8% from a custom mixture of l-amino acids. The amino acid content of the diet on a weight basis was 14.1%. The CON diet contained 0.86% methionine, and the MR diet contained 0.17% methionine. Food and water were provided ad libitum, and lights were on 12 h/day from 7:00 a.m. to 7:00 p.m.

The temporal development of transcriptional and biochemical responses to MR was examined in 128 WT mice acclimated to CON diet in a room with a reversed light cycle (7:00 p.m. lights on and 7:00 a.m. lights off) for 1 week beginning at 6 weeks of age. The mice were then trained to eat at lights out for a period of 10 days by removal of all food at lights on (7:00 p.m.). Food was restored at lights off (7:00 a.m.). After the initial 10 days of training, half of the mice were switched to the MR diet. In groups of 16, mice were killed after 3 h, 6 h, 9 h, 12 h, 2 days, 4 days, 6 days, and 8 days on their respective diets. A cohort of Gcn2^−/− mice (n = 24) followed the same feeding paradigm, and mice were killed 6 h and 6 days after initiation of the diets.

A third cohort of WT (n = 20) and Gcn2^−/− (n = 20) mice was shipped to the Vanderbilt Phenotyping Center to undergo hyperinsulinemic-euglycemic clamps as previously described (27,28). Mice of each genotype were placed on the CON diet for 1 week before being randomized to the CON or MR diet for the following 8 weeks prior to surgery for catheter placement. After a 5-day recovery, clamps were performed in conscious mice after a 5-h fast. Insulin (2.5 mU/kg/min) was infused, along with a 50% dextrose solution at a variable rate to maintain euglycemia. Whole-body glucose turnover was assessed with a 5-μCi bolus of [³H]glucose tracer 1.5 h before insulin treatment, followed by a 0.05 μCi/min infusion during the clamp. Insulin-stimulated glucose uptake in individual organs was determined using 2-deoxy-d-[1-¹⁴C]glucose administered as a single bolus (13 μCi) 120 min after the start of clamp procedure. At t = 145 min, epididymal white adipose tissue (WAT), inguinal WAT (IWAT), brown adipose tissue (BAT), gastrocnemius, vastus lateralis, soleus, brain, and heart were harvested for analysis.

WT (n = 12) and Gcn2^−/− (n = 12) mice were acclimated to CON diet for 2 weeks. Then, half of each genotype was switched to a custom diet devoid of methionine (0% methionine). Mice were killed after 6 days on their respective diets and tissues were collected.

WT mice (n = 24) were acclimated to the CON diet for 1 week beginning at 6 weeks of age. Then, eight mice were switched to the MR diet, eight mice were switched to an MR diet containing 0.2% cysteine, and the remaining mice stayed on the CON diet. After 8 weeks on the diets, EE was measured and all mice were killed. The BIOOXYTECH GSH/GSSG-412 Assay kit (Oxis Research, Portland, OR) was used to determine hepatic glutathione (GSH/GSSG) following the manufacturer’s protocol.

WT mice (n = 24) were acclimated to CON diet for 1 week beginning at 6 weeks of age. Then, eight mice were switched to the MR diet containing 0.12% methionine, eight mice were switched to the 0.12% MR diet containing 0.2% cysteine, and the remaining mice stayed on the CON diet. The mice were killed after 8 weeks on their respective diets.

WT mice (n = 8) and FGF21^−/− mice (n = 8) kindly provided by Dr. Steven Kliewer (University of Texas Southwestern) were acclimated to CON diet for 1 week beginning at 6 weeks of age. Then, four mice of each genotype were switched to the MR diet, while the remaining mice stayed on the CON diet. After 8 weeks of ad libitum consumption of their respective diets, EE was measured over 4 days in the respective groups as described above. Thereafter, mice were killed and IWAT was harvested.

For measurement of eIF2α and PERK phosphorylation, powdered liver tissue was homogenized in assay buffer (25 mmol/L HEPES, 2 mmol/L EDTA, 10 mmol/L dithiothreitol, 50 mmol/L sodium fluoride, 50 mmol/L β-glycerophosphate pentahydrate, 3 mmol/L benzamidine, 1 mmol/L sodium orthovanadate, 0.5% deoxycholate, 1% SDS, protease inhibitor cocktail [Sigma-Aldrich], and 1 mmol/L microcystin). Equivalent amounts of protein from each sample were separated by SDS-PAGE and transferred to polyvinylidene fluoride membranes. Blots were developed using enhanced chemiluminescence, and band densities were quantified using Alphaview SA Imaging Software (version 3.4.0). The phospho-eIF2α antibody was from Abcam (Cambridge, MA), total eIF2α antibody was from Santa Cruz (Dallas, TX), and the p-PERK and total PERK antibodies were from Cell Signaling Technology (Beverly, MA).

Total RNA was isolated using the RNeasy Mini Kit (QIAGEN, Valencia, CA). One microgram of total RNA was reverse transcribed to produce complementary DNA. Gene expression was measured by RT-PCR (Applied Biosystems, Foster City, CA), and expression of the corresponding mRNAs was normalized to cyclophilin.

Serum glucose was determined using a kit from Cayman Chemical (Ann Arbor, MI). Serum FGF21 and insulin were measured using ELISA kits from R&D Systems (Minneapolis, MN) and Millipore (Billerica, MA).

Analysis of the data was conducted with Statistica (version 8.0; StatSoft, Inc.), using two-way ANOVA with genotype (WT and Gcn2^−/−) and diet (CON and MR) as main effects. Group means of variables of interest were compared using residual variance as the error term to test for effects of diet within genotype and genotype within diet using linear contrasts and the Bonferroni correction. Protection against type I errors was set at α= 0.05.

Body weights of WT and Gcn2^−/− mice on the CON diet did not differ at the beginning or end of the study, and dietary MR reduced weight to the same extent in both genotypes (Table 1). Initial adiposity was comparable between genotypes, but Gcn2^−/− mice on the CON diet added less fat than WT mice during the study (Table 1). MR decreased adiposity to 14% of body weight in both genotypes despite producing comparable increases in food intake (Table 1). Interestingly, CON-fed Gcn2^−/− mice consumed 13% less food than CON-fed WT mice. However, on a body weight–adjusted basis, energy intake did not differ between genotypes, and MR increased energy intake by 39% in WT mice and 37% in Gcn2^−/− mice (Table 1). These findings illustrate that MR increased the energy costs of maintenance and growth similarly in both genotypes.

GCN2 plays an essential role in the acute, conditioned aversive response to EAA deprivation (13,29,30) but not the longer-term effects of LD on food intake (15,16). Therefore, we initially examined the acute responses to MR and found that EE was significantly increased 5 days after introduction of the MR diet in WT mice (Fig. 1A). The magnitude of the effect increased over the subsequent 5 days (Fig. 1A) and was also evident after 12 weeks on the respective diets. This early diet-induced increase in EE was delayed and muted in Gcn2^−/− mice, not appearing until 9 days, and increasing by a smaller amount compared with WT mice between 9 and 15 days (Fig. 1A and B). For example, the average MR-induced increase in EE after 2 weeks in the WT group was 12%, while the increase in Gcn2^−/− mice was only 7% (Fig. 1A and B).

Browning of WAT contributes to increased EE in rodents (31), and chronic MR (11 weeks) induces significant browning of WAT in mice (32). To explore the relationship between the acute increase in EE and the browning of WAT during MR, we measured thermogenic markers in IWAT after short- and long-term dietary MR. The MR diet increased markers of browning and thermogenesis (Ucp1, Ppargc1a, Cox7a, Cox8b, Cidea, and Dio2) within 6 days in WT mice (Fig. 1C) but not Gcn2^−/− mice (Fig. 1D). However, after 14 weeks, MR increased mRNA expression of Ucp1 in WAT to the same extent in WT and Gcn2^−/− mice (Fig. 1E). These data show that the delayed increase in EE in Gcn2^−/− mice after initiation of MR (Fig. 1B) is temporally related to a delayed induction of thermogenic genes in IWAT, whereas after chronic consumption of the diet, the increase in browning of WAT was comparable. In addition, measurement of EE after 12 weeks showed that the MR-induced increase in EE was comparable in WT and Gcn2^−/− mice (Fig. 1F). Overall, the effects of MR on energy balance were not compromised in Gcn2^−/− mice (Table 1 and Fig. 1F).

For examination of the relationship between eIF2 phosphorylation and FGF21 induction, WT mice were studied at different times after introduction of MR. A diet-induced increase in hepatic eIF2 phosphorylation was not detected before 8 days (Fig. 2A). In contrast, within 6 h of initiation of MR and at all subsequent time points, hepatic Fgf21 mRNA (data not shown) and serum FGF21 (Fig. 2B) were significantly increased. Notwithstanding previous in vitro and in vivo findings linking eIF2 signaling to FGF21 expression (11,33), we find that eIF2 phosphorylation is not required for the acute MR-induced increase in hepatic FGF21.

To examine potential causes for the delayed increase in EE in Gcn2^−/− mice after initiation of MR, we measured diet-induced induction of serum FGF21, a hormone that increases sympathetic nerve activity and EE (34–36). Dietary MR produced a 7.7-fold, 4.2-fold, and 6.8-fold increase in serum FGF21 after 6 h, 6 days, and 14 weeks, respectively, in WT mice, and at all three time points the MR diet produced comparable increases in serum FGF21 in Gcn2^−/− mice (Fig. 2C and D). The fold increase in FGF21 in Gcn2^−/− mice after 6 h was somewhat larger than the increase in WT mice, but by 6 days, both the net and fold increases were larger in Gcn2^−/− mice (Fig. 2C). After 14 weeks, both basal and MR-induced levels of FGF21 were comparable between genotypes (Fig. 2D).

To examine the role of FGF21 in the MR-induced increase in adipocyte browning and increased EE, we measured MR-dependent induction of IWAT Ucp1 mRNA and total EE in WT and Fgf21^−/− mice. MR produced a four- to fivefold increase in Ucp1 mRNA in WT mice but produced no detectable increase in Ucp1 mRNA in IWAT from Fgf21^−/− mice (Fig. 2E). Total EE was comparable in WT and Fgf21^−/− mice on the CON diet, while MR increased EE by 22% in the WT mice (Fig. 2F). In contrast, MR failed to increase EE in the Fgf21^−/− mice (Fig. 2F), illustrating that the MR-dependent increase in FGF21 is functionally linked to the diet-induced increase in IWAT Ucp1 mRNA and total EE.

Dietary MR for 8–10 weeks produces a two- to threefold increase in overall insulin sensitivity (27,32). Short-term LD also increases insulin sensitivity (37), ostensibly through a mechanism requiring GCN2 (38). To test whether GCN2 was also required for MR to increase insulin sensitivity, hyperinsulinemic-euglycemic clamps were used to compare the response between genotypes. The glucose infusion rates (GIRs) needed to establish euglycemia after initiating insulin infusion were achieved within 40 min in all groups, and steady-state glucose was maintained thereafter across the groups (Fig. 3A). The GIRs were indistinguishable in WT and Gcn2^−/− mice on the CON diet (Fig. 3A) but were 2.2-fold higher in mice on the MR diet, regardless of genotype (Fig. 3A). Thus, the absence of GCN2 affected neither overall insulin sensitivity nor the responses of the mice to dietary MR. Liver 3-[³H]glucose infusion was used to calculate basal endogenous glucose production (endo R_a) and compare insulin-dependent suppression of endo R_a during the clamp. Figure 3B shows that the efficacy of insulin to suppress hepatic glucose production was equivalently enhanced by MR in both genotypes. The glucose metabolic index (R_g) showed that MR produced significant two- to threefold increases in uptake of [¹⁴C]2-deoxyglucose in gastrocnemius muscle, vastus lateralis muscle, soleus muscle, IWAT, BAT, and heart in both genotypes (Fig. 3C and D). Together, these data provide compelling evidence that GCN2 is not essential to the insulin-sensitizing effects of MR.

GCN2 is responsible for eIF2 phosphorylation in response to LD (39) and asparaginase treatment (40), as both responses are compromised in Gcn2^−/− mice. In contrast, dietary MR increased phosphorylation of hepatic eIF2 by four- to fivefold in both genotypes (Fig. 4A), showing that GCN2 is not required for eIF2 phosphorylation by MR. Low-protein diets also increased eIF2 phosphorylation in Gcn2^−/− mice (26), suggesting the presence of an alternative eIF2 kinase that is activated by MR and low-protein diets. One possibility is PERK, which is activated by ER stress and reductions in GSH (18,41). PERK phosphorylation was similarly increased by MR in WT and Gcn2^−/− mice (Fig. 4B). To explore the involvement of PERK activation by MR, we tested the possibility that MR was disruptive to ER homeostasis and protein folding, eliciting an ER stress or UPR. Measurement of genetic markers that correspond with activation of the UPR showed that MR did not provoke ER stress in the livers of WT or Gcn2^−/− mice (Table 2). To further explore the transcriptional consequences of PERK activation, we measured MR-dependent changes in known targets of ATF4 and the transcription factor NRF2, which is activated after phosphorylation by PERK (42). MR strongly induced the expression of eIF2/ATF4 genes, Asns, Trib3, Gadd34, and Vldlr, in livers of WT and Gcn2^−/− mice (Table 2). Known targets of NRF2 were upregulated (Aox1, Cbr1, Ephx1, Gsta1, Mgst3, Nqo1, Psat1, and Cyp4a14) and downregulated (Thrsp), providing strong evidence that NRF2 was also activated by MR (Table 2). Together, these findings show that MR activates the ISR and NRF2 pathways regardless of GCN2 status and without induction of the UPR.

Given the role of GCN2 in responding to LD (39) but not MR, we tested the possibility that GCN2 could distinguish between EAA deprivation and EAA restriction by measuring eIF2 phosphorylation in WT and Gcn2^−/− mice after methionine deprivation. Methionine deprivation for 6 days increased eIF2 phosphorylation in livers of WT and Gcn2^−/− mice, indicating that unlike LD, GCN2 is not required for the methionine deprivation–induced phosphorylation of eIF2 (Fig. 4C). Together, these studies illustrate that GCN2 does not direct either the short-term signaling responses to dietary MR or its long-term effects on energy balance or insulin sensitivity. The findings also show that PERK is sensitive to MR and activates the ISR but not the UPR in response to the diet.

To explore whether PERK activation was related to hepatic GSH, 0.2% cysteine was added back to the MR diet. This dietary addition reversed the reduction in hepatic GSH and the increase in nighttime EE by MR (Fig. 5A). The addition of cysteine also reversed the MR-induced increase in serum FGF21 and the reduction in fasting insulin (Fig. 5B). In addition, cysteine blocked the MR-induced increases in NRF2-sensitive and eIF2/ATF4-sensitive genes in the liver (Fig. 5C). To test whether adding back cysteine also reversed the activation of eIF2α and PERK, we measured the phosphorylation of hepatic eIF2α and PERK after 8 weeks of 0.12% MR or 0.12% MR plus 0.2% cysteine. Figure 6A illustrates that adding cysteine back reversed the MR-dependent activation of both eIF2α and PERK (Fig. 6A). These findings link the MR-induced reduction of GSH to activation of the ISR and significant components of the metabolic profile produced by the diet. A proposed model showing the GCN2-independent, GSH-dependent signaling through PERK to eIF2/ATF4 and NRF2 is shown in Fig. 6B.

The original report of dietary MR showed that ad libitum consumption of a diet containing 0.17% methionine and no cysteine increased longevity in rats by 25–30% (43,44). The paradoxical finding that MR also reduced adiposity despite increasing food intake focused growing interest on the short-term metabolic effects of this diet. The responses include increased food intake, increased EE, reduced fat deposition, increased insulin sensitivity, reduced tissue and circulating lipids, and tissue-specific transcriptional effects (1–6,45,46). In the present work, a loss-of-function approach was used to test whether GCN2 is an essential mediator of these metabolic responses to MR. Using a combination of metabolic phenotyping and examination of tissue-specific responses, we provide compelling evidence that GCN2 is not required for the metabolic and adaptive responses to dietary MR. In contrast, we provide evidence that MR signals to eIF2/ATF4 and NRF2 through the eIF2 kinase, PERK, which is activated by MR in the absence of ER stress. The result is activation of both the ISR, through eIF2/ATF4, and the NRF2-sensitive antioxidant response program (Table 2). Based on the significance of reduced formation of hepatic GSH to enhancement of insulin signaling in MR mice and HepG2 cells (27), we hypothesize that the reduced formation of hepatic GSH also represents a key metabolite through which dietary MR is affecting signaling through a novel PERK-eIF2-ATF4-NRF2 pathway.

Significant progress has been made in identifying molecular and cellular mechanisms of EAA sensing (30,39,47–53), but the more difficult challenge has been to understand how these sensing systems are anatomically and temporally organized to produce the complex physiological responses that occur when an EAA reduction is detected. Restricting the availability of EAAs activates GCN2, and previous studies have shown that GCN2 is an important mediator of some but not all of the effects of a diet lacking leucine (15,16). Stated another way, these findings illustrate that mechanisms additional to GCN2 sense and respond to the absence of leucine. The role of GCN2 as a mediator of EAA restriction, particularly restriction of methionine, is unknown. There is an assumption that any degree of EAA restriction sufficient to activate GCN2 would produce a common set of responses, but clear exceptions are evident with leucine. For example, LD (e.g., 100% restriction) causes a significant decrease in food intake and lipogenic gene expression in the liver (15), whereas leucine restriction by 85% increases food consumption and has no effect on hepatic lipogenic genes (54). It is unclear whether GCN2 is comparably activated by these two diets, but the responses to them are fundamentally different. While the present work widens the apparent conundrum by showing that MR is fully effective without GCN2, these results are consistent with an array of recent reports defining unique molecular events after the removal of different amino acids from cells in culture (55–57). Together, these findings emphasize that both the EAA being restricted and the degree of restriction play important roles in determining the responses to EAA-modified diets.

The uniqueness of methionine was illustrated in previous work showing that the decrease in hepatic GSH produced by MR extends the activation cycle of insulin signaling to protein kinase B by slowing the dephosphorylation of phosphatidylinositol (3,4,5)-triphosphate (27). This occurs because reducing dietary methionine limits the formation of the GSH precursor, cysteine, and the resulting reduction of GSH limits GSH-dependent degradation of phosphatidylinositol (3,4,5)-triphosphate by phosphatase and tensin homolog (27). The present work extends these findings by proposing that MR increases ATF4-sensitive genes by activating PERK. Surprisingly, markers of ER stress, the normal activator of PERK, were unchanged, indicating that MR was not activating PERK as part of the UPR. Given the essential role of GSH in disulfide bond formation during protein folding in the ER (18,58), it seems plausible that the MR-dependent decrease in hepatic GSH, rather than an accumulation of misfolded proteins, is an alternate signal for PERK activation (19). For example, although PERK is activated by glucose restriction in ESCs, its activation depends on depletion of cellular GSH (41). PERK is not typically activated by EAA depletion (41), so it seems likely that the ability of MR to activate PERK and initiate components of the ISR derives from its unique effects on hepatic GSH. In addition to eIF2, our findings indicate that the NRF2-dependent antioxidant response program is activated by MR. NRF2 is activated by PERK (42), although NRF2 transcriptional activity also responds to low GSH (59). It is interesting that the ISR and the antioxidant stress response share genes whose primary function is to increase production of GSH (19). Our findings suggest the interesting possibility that PERK activation by low GSH may be the common signal.

The ability of MR to produce its full range of metabolic effects is not compromised in Gcn2^−/− mice, and we have proposed that hepatic PERK, activated by the decrease in hepatic GSH, is a critical sensing event linking the MR diet to its effects. We have supported this hypothesis by showing that within hours, dietary MR activates hepatic Fgf21 and increases its circulating levels. Based on the known physiological effects of FGF21 (34,36,60,61), the MR-induced browning of WAT, activation of thermogenesis in BAT, increase in EE, and overall increase in insulin sensitivity are fully consistent with the MR-induced increase in hepatic FGF21 (27,62). In addition, we show that the MR-dependent increase in EE and UCP1 expression in IWAT is dependent on FGF21. We have also shown that FGF21 acts directly in adipose tissue to enhance basal and insulin-dependent glucose uptake (27). Third, we have shown that the MR-dependent decrease in hepatic GSH amplifies insulin-dependent signaling in the liver (27). The central importance of reduced GSH to the metabolic phenotype is illustrated through a gain-of-function approach produced by adding back a small amount of the GSH precursor, cysteine (0.2%), to the MR diet. If cysteine repletion restored hepatic GSH levels and hepatic GSH was critical to Fgf21 transactivation, we predicted that FGF21 and other components of the phenotype would be reversed. This is precisely what was observed. The addition of cysteine to the MR diet restored hepatic GSH to control levels, along with plasma FGF21, EE, NRF2- and ATF4-sensitive genes, fasting insulin, and eIF2α phosphorylation. Collectively, our findings provide the first evidence of a nutrient-sensing system for MR that bypasses GCN2 and activates hepatic PERK through a noncanonical GSH-dependent mechanism.

Acknowledgments. The authors thank Cindi Tramonte for administrative support and Manda Orgeron, Alicia Pierse, Alexandra N. Daniel, and Phillip H. Behrens for excellent technical support. Cindi Tramonte, Manda Orgeron, and Alicia Pierce are at the Pennington Biomedical Research Center (Baton Rouge, LA), and Alexandra N. Daniel and Phillip H. Behrens are in Biochemistry and Molecular Biology, Indiana University School of Medicine-Evansville, Evansville, IN. The authors thank Carrie Elks at Pennington Biomedical Research Center (Baton Rouge, LA) for her valuable input on Western blotting of PERK.

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

Author Contributions. D.W., T.G.A., and T.W.G. contributed to the writing and editing of the manuscript. D.W., K.P.S., L.A.F., C.C.C., K.N.D., and J.S. conducted the animal experiments. D.W., K.P.S., L.A.F., K.N.D., J.S., M.X., E.C.H., I.A.N., and A.P.P. conducted the in vitro analysis and associated Western blots and mRNA and metabolite measurements. T.G.A. and T.W.G. analyzed data used to produce illustrations. T.W.G. 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.

Funding. This work was supported in part by the American Diabetes Association (1-12-BS-58 to T.W.G.), National Institutes of Health (NIH) (RO1-HD-070487 to T.G.A. and RO1-DK-096311 to T.W.G.), U.S. Department of Agriculture Cooperative State Research, Education, and Extension Service (Multistate NC1184 to T.G.A.), and the Vanderbilt University Mouse Metabolic Phenotyping Center (NIH U24-DK-059637). This work also made use of the Genomics core facilities supported by NIH P20-GM-103528 to T.W.G. and NIH 2P30-DK-072476. D.W. was supported by NIH National Research Service Award 1 F32-DK-098918 and NIH P20-GM-103528. L.A.F. is supported by an ADA mentor-based postdoctoral fellowship (7-13-MI-05). E.C.H. was supported by summer student fellowship (NIH 5T35OD011151). A.P.P. is supported by an NIH Institutional Research and Academic Career Development Award (New Jersey/New York for Science Partnerships in Research & Education at Rutgers-Robert Woods Johnson Medical School).