Intermittent Leucine Deprivation Produces Long-lasting Improvement in Insulin Sensitivity by Increasing Hepatic Gcn2 Expression

Type 2 diabetes (T2D) is characterized by insulin resistance in target organs, such as the liver and other tissues, and relative insulin deficiencies caused by pancreatic β-cell dysfunction (1). The pathogenesis of T2D involves complex interactions among genetic and environmental factors (2). Among these environmental factors, nutrients play a key role in the progression of insulin resistance (3). The levels of leucine, an essential amino acid, are increased in the serum of obese humans and have been associated with the increased risk of T2D and insulin resistance (4). Consistently, complete or partial leucine deprivation improves insulin sensitivity and/or decreases fat accumulation (5,6). However, whether and how these effects can be extended remain unknown.

Metabolic memory is a phenomenon in which the effects of metabolic stimulation are retained for an extended period after stimulation (7) and is regulated by multiple nutritional status factors (8–10). For example, intermittent fasting (IF), a diet that cycles between periods of fasting and nonfasting, effectively improves the insulin sensitivity of mammals and promotes their health (11). Some of these ameliorating metabolic effects can be long-lasting via the formation of metabolic memory (12,13). That IF is normally accompanied by changes in various nutrients (14) suggests that intermittent leucine deprivation (ILD) might also induce long-lasting effects on improved insulin sensitivity through the formation of metabolic memory.

General control nonderepressible 2 (GCN2) is an amino acid sensor that is activated under amino acid starvation (15). GCN2 phosphorylates the eukaryotic initiation factor (eIF2) to regulate protein synthesis, feeding behavior, lipid metabolism, and related processes (6,16,17). Recently, GCN2 was implicated in regulating insulin sensitivity in the liver (18), suggesting that it might be involved in ILD-mediated regulation of insulin sensitivity. Regarding the mechanisms underlying metabolic memory formation, several possibilities have been proposed (19,20) and one of them is the epigenetic factors such as DNA methylation (21), which are particularly affected by leucine content (22). However, these possibilities have not been studied. In this study, we found that ILD produces a long-lasting effect on improvements in insulin sensitivity by increasing hepatic Gcn2 expression via a reduction in its DNA methylation.

Male and female C57BL/6J wild-type (WT) and leptin receptor–mutated (db/db) mice were obtained from the Model Animal Research Center of Nanjing University (Nanjing, China). To confirm the role of GCN2 in ILD, we genetically deleted Gcn2 in the liver, producing hepatic Gcn2 knockout (LGKO) mice by crossing Gcn2^loxp/loxp mice with albumin-Cre mice on a C57BL/6J background for at least four generations. Mice were maintained as previously described (23), and experiments were conducted in accordance with the guidelines of the Institutional Animal Care and Use Committee of Shanghai Institute of Nutrition and Health.

Control (nutritionally complete amino acids), leucine-deficient, threonine-deficient, valine-deficient, and isoleucine-deficient diets were obtained from Research Diets (New Brunswick, NJ). All diets were isocaloric; the missing calories owing to specific amino acid deficiency were compensated for by corn starch in each case. The food formulas of the amino acid diets and the high-fat diet (HFD) are listed in Supplementary Table 1.

Body weight and food intake were measured weekly with a precision scale. The body composition of the mice was determined with a nuclear magnetic resonance system (Bruker Corporation, Billerica, MA). Indirect calorimetry was measured with Comprehensive Lab Animal Monitoring System (CLAMS). The Vo2 was normalized to lean mass (24). Rectal temperatures were measured with a rectal probe attached to a digital thermometer (Physitemp Instruments, Clifton, NJ).

Mouse primary hepatocytes were prepared as previously described (25). For the in vitro insulin signaling assay, the primary hepatocytes were treated with 100 nmol/L insulin for 10 min. The cells were transfected with siRNA targeting Gadd45b (5′-CACTTCACCCTGATCCAGTCGTTCT-3′) to silence its expression.

Recombinant adenoviral vector expressing mouse GCN2 (Ad-GCN2), adenoviral vector expressing green fluorescent protein (Ad-GFP), adeno-associated virus expressing shRNAs specific for Gadd45b (AAV-shGadd45b), and adeno-associated virus expressing scrambled RNA (AAV-scramble) were purchased from Vigene Biosciences (Shandong, China). The scrambled sequence was 5′-TTCTCCGAACGTGTCACGT-3′, and the shRNA sequence for mouse Gadd45b was 5′-TGAAGAGAGCAGAGGCAATAATTCAAGAGATTATTGCCTCTGCTCTCTTCATTTTTT-3′. Adenoviral viruses were diluted in PBS and administered via a tail vein injection using 5 × 10⁸ plaque-forming units per mouse. Adeno-associated virus were diluted in PBS and administered via a tail vein injection using 3 × 10¹¹ plaque-forming units per mouse.

The Gcn2 promoter (from −1187 to −98) and its mutant versions were constructed in pGL3-Basic (Promega, Madison, WI). The pCpGL-CMV-firefly luciferase plasmid (without CpG loci, for methyltransferases treatment) and pCpGL-CMV-Renilla plasmid were kindly provided by Professor Guoliang Xu from the Shanghai Institute of Biochemistry and Cell Biology, Chinese Academy of Sciences. The Gcn2 promoter was constructed in the pCpGL-CMV-firefly luciferase plasmid to avoid the disturbance of the CpG loci in carrier vector, and it was methylated in vitro with M.SssI, HpaII, and HhaI (New England Biolabs, Ipswich, MA). The firefly and Renilla luciferase activities were assayed with the Dual-Glo Luciferase Assay System (Promega) after 24 h.

DNA extraction (26) and bisulfite sequencing PCR (27) were performed as previously described. The converted DNA was collected with the Promega Wizard DNA clean up kit following the instructions of the manufacturer. The regions of interest of the converted DNA were amplified with TaKaRa Ex Taq DNA Polymerase (Takara Bio, Shiga, Japan). The sequences of the primer were as follows: F, 5′- GTAAATAAGATTGATAGAGATAT-3′, and R, 5′-AAAAAATATAACTCTACTCTAAAAAAAC-3′. The PCR was performed with a DNA Engine Thermal Cycler (Bio-Rad Laboratories). The PCR products were gel purified with a Gel Purification Kit (Sangon, Shanghai, China). The purified fragments were used to ligate the pMD19-T Vector (TaKaRa, Dalian, China). The positive recombinants were selected by PCR and then sequenced.

Levels of blood glucose were measured with a Glucometer Elite monitor (Abbott Diabetes Care, Oxon, U.K.). Serum insulin was measured with the Mercodia Ultrasensitive Rat Insulin ELISA kit (ALPCO Diagnostics, Salem, NH). The glucose tolerance tests (GTTs), insulin tolerance tests (ITTs), and HOMA of insulin resistance (HOMA-IR) index assessment were conducted as previously described (25).

Mice maintained on different diets were fasted for 6 h before insulin injection, and the insulin signaling pathway was detected as previously described (25).

The serum amino acids were analyzed with high-performance liquid chromatography (UltiMate 3000)–tandem mass spectrometry (API 3200 QTRAP) methods of Beijing MS Medical Research Co. Ltd (Beijing, China).

RNA was isolated and RT-PCR was performed as previously described (18). The sequences of the primers used in this study are available in Supplementary Table 2.

Western blotting was performed as previously described (18), with the following primary antibodies: anti-GCN2 (1:1,000, 65981S; Cell Signaling Technology [CST], Danvers, MA), anti–phosphorylated (p)-GCN2 (1:500, bs-3155R; Bioss, Beijing, China), anti-GADD45b (1:500, sc377311; Santa Cruz, Dallas, TX), anti–p-eIF2α (1:1,000, 3398S; CST), anti–total (t)-eIF2α (1:1,000, 9722S; CST), anti–p-IR (1:1,000, 3024S; CST), anti–t-IR (1:1,000, 3025S; CST), anti–p-AKT (1:1,000, 9271S; CST), anti–t-AKT (1:1,000, 9272S; CST), anti–p-GSK3β (1:1,000, 9336S; CST), anti–t-GSK3β (1:1,000, 9315S; CST), and anti-TUBULIN (1:3,000, ab0037; Shanghai Abways Biotechnology, Shanghai, China).

Statistical analyses were performed in GraphPad Prism, version 8.0 (GraphPad Software, San Diego, CA). All values are presented as the mean ± SEM. The two-tailed unpaired Student t test was used for comparisons between two groups, and ANOVA was used for multiple comparisons, followed by the Student-Newman-Keuls test. Statistical significance was defined as P < 0.05.

The data sets generated during and/or analyzed during the current study are available from the corresponding author upon reasonable request. No applicable resources were generated or analyzed during the current study.

We started by investigating the effect of 1-day leucine deprivation and found that it decreased blood glucose levels in both the fed and fasting periods (Supplementary Fig. 1A). This effect lasted for 3 days but disappeared on the fourth day after a control diet was resumed (Fig. 1A). Although the fed and fasting serum insulin levels remained unchanged, the HOMA-IR index decreased significantly after 1 day of leucine deprivation and lasted for 3 days (Fig. 1B and C and Supplementary Fig. 1B and C). Glucose tolerance and insulin tolerance were examined with GTTs and ITTs, respectively. The GTT showed that the blood glucose levels were significantly lower after 1 day of leucine deprivation, in line with a previous report (5), and the effect lasted for 3 days but disappeared at the fourth day (Fig. 1D and E and Supplementary Fig. 1D). Similar results were obtained in the ITT (Figs. 1F and G and Supplementary Fig. 1E).

Then we explored whether other leucine deprivation–induced changes could also be extended. Although food intake, body weight, and fat mass decreased on the day of leucine deprivation, they all returned to normal levels within the next 4 days (Supplementary Fig. 2A–C). The lean mass remained unchanged throughout the experiment (Supplementary Fig. 2D). Body temperature, which reflects energy expenditure (26), also increased after leucine deprivation and returned to normal levels after a control diet was resumed (Supplementary Fig. 2E).

Because longer periods of leucine deprivation (7 days) have previously been shown to effectively improve glucose and lipid metabolism in WT mice (18), we speculated that longer periods of leucine deprivation would extend its ameliorating effects on glucose metabolism. However, the ITT results were similar to those of mice maintained on a leucine-deficient diet for 1 day (Supplementary Fig. 3A and B).

To clarify whether these effects were leucine specific, we examined the prolonged effects of deficiencies in other essential amino acids, including valine and isoleucine (another two branched-chain amino acids) (5) and threonine, on insulin sensitivity. The prolonged improvement of insulin sensitivity was observed after valine and isoleucine deprivation but not threonine deprivation (Supplementary Fig. 4A–F). Together, 1 or 7 days of leucine deprivation could improve the insulin sensitivity for 3 days and the lasting effects also occurred with valine and isoleucine deprivation.

IF treatment can ameliorate insulin resistance and obesity after a normal diet is resumed (12), suggesting that ILD might induce long-term effects. We then conducted a diet-switch experiment that comprised various cycles of ILD treatment (1-day leucine-deficient diet and 3-day control diet) in WT mice (Fig. 2A). On the seventh day after seven cycles of ILD, glucose metabolism was improved (Supplementary Fig. 5A–E), and these effects persisted as examined at day 49 (Fig. 2B–F) or day 84 (Supplementary Fig. 5F) after ILD treatment. ILD treatment also improved insulin sensitivity in female mice (Supplementary Fig. 6A–E)

Lipid and energy metabolism is affected by leucine deprivation in mice (6), which might also contribute to insulin sensitivity regulation (28). Most related parameters examined in ILD-treated mice 49 days after a control diet was resumed in this study remained unchanged (Supplementary Fig. 7B–K). We found that food intake changed on the day of leucine deprivation (Supplementary Fig. 7A); however, food intake was unlikely to contribute to the improved insulin sensitivity, as shown by our pair-fed experiments (Supplementary Fig. 8A–H). Furthermore, fewer cycles of ILD might also result in similar positive outcomes. We therefore examined the effects of four cycles of ILD treatment. Despite glucose metabolism showing improvements on the fourth day after four cycles of ILD, the effects disappeared quickly (by the seventh day after a control diet was resumed) (Supplementary Fig. 9A–E). Thus, in the subsequent experiments, we conducted ILD for seven cycles. Taken together, seven cycles of ILD improved glucose metabolism in both male and female mice and had no effect on lipid and energy metabolism.

To understand tissues involved in ILD-improved insulin sensitivity, we examined the phosphorylation of three key components in the insulin signaling pathway, insulin receptor (IR) on Tyr^1150/1151 (p-IR), protein kinase B on Ser⁴⁷³ (p-AKT), and glycogen synthase kinase 3β on Ser⁹ (p-GSK3β) (23). The insulin-stimulated phosphorylations of IR, AKT, and GSK3β were higher in the liver, but not white adipose tissue (WAT) or muscle, of ILD-treated mice 49 days after resuming a control diet (Fig. 2G and Supplementary Fig. 10A and B).

For confirmation of the liver specificity, primary hepatocytes were cultured and subjected to an ILD mimic treatment. We exposed the cultured primary hepatocytes to seven cycles of 10 min of leucine-deprived medium and 30 min control medium, before 8 h of control medium, mimicking the effects of ILD in vivo, in manner similar to that shown in another study (29). Similar to the results observed in vivo, the insulin signaling pathway was promoted in the ILD-treated primary hepatocytes (Fig. 2H). These results indicated that ILD improved the hepatic insulin sensitivity in vivo and in vitro.

GCN2 is an amino acid sensor and regulates insulin sensitivity under leucine deprivation (18). Consistent with increased insulin sensitivity in the liver, the phosphorylation of GCN2 and eIF2α in the liver was increased in the ILD mice, but the phosphorylation of eIF2α in WAT and muscle was unchanged (Fig. 3A and Supplementary Fig. 11A and B). To confirm the role of GCN2 in ILD, we genetically deleted GCN2 in the liver, producing liver GCN2-knockout (GCN2^loxp/loxp; Alb^Cre: LGKO) mice, by crossing GCN2^loxp/loxp mice with albumin-Cre mice. The GCN2-knockout efficiency was validated based on significantly decreased GCN2 mRNA and protein levels in the liver, with unchanged Gcn2 mRNA levels in other tissues (Fig. 3B and C and Supplementary Fig. 11C). Then we treated LGKO and control mice with control diet or ILD and found that there was no difference in total food intake and body weight (Supplementary Fig. 11D and F). After a control diet was resumed for 49 days after ILD, the improved blood glucose, HOMA-IR, GTT, and ITT were blocked when hepatic GCN2 was deleted but the serum insulin levels were not changed (Fig. 3D–H). GCN2 knockout also significantly blocked the beneficial effects of ILD on the insulin signaling pathway and increased p-GCN2 and p-eIF2α in liver but not in other tissues examined (Fig. 3I and Supplementary Fig. 11G–I). Similar effects were observed in hepatocytes with GCN2 knockdown (Supplementary Fig. 12A–C).

We further investigated the role of GCN2 by injecting WT mice with Ad-GCN2 and Ad-GFP. The efficiency of GCN2 overexpression was confirmed with Western blotting (Supplementary Fig. 15A). Compared with that in control mice, glucose metabolism was indeed improved in GCN2-overexpressing mice (Supplementary Fig. 15B–F). In summary, the improvement in glucose metabolism induced by ILD depended on hepatic GCN2.

GCN2 is normally activated by amino acid starvation (30). However, serum leucine was not affected by ILD treatment after the control diet was resumed for 7 or 49 days (Supplementary Figs. 13A and 14A). Surprisingly, hepatic GCN2 protein and Gcn2 mRNA were higher than those in control mice after ILD treatment (Fig. 4A and B and Supplementary Fig. 14B and C). Moreover, this effect was liver specific, as the Gcn2 mRNA levels in the muscle and WAT were unchanged by ILD treatment (Supplementary Figs. 13B and 14C).

We then explored the mechanisms underlying the increase in Gcn2 mRNA after ILD treatment. As DNA methylation is an epigenetic modification that regulates gene expression and mediates IF-regulated glucose and lipid metabolism (31), we speculated that ILD would increase Gcn2 expression by reducing DNA methylation. To test this, we screened a genomic region of 990 base pairs (from −1109 to −120) in the Gcn2 promoter for CpG sites (the targets of DNA methylation). We identified the methylation patterns of 24 CpG sites using bisulfite sequencing PCR (Fig. 4C). Although large parts of the chosen genomic region were not differentially methylated (Supplementary Fig. 13C) or completely unmethylated (data not shown), two CpG sites, CpG⁻⁷⁵⁸ and CpG⁻⁵⁷¹, were significantly less methylated in the livers of ILD mice that had resumed a control diet for 49 days relative to levels in control mice (Fig. 4D). Similar results were observed in the liver after a control diet was resumed 7 days after ILD (Supplementary Fig. 14D and E).

The methylation pattern of a promoter is often related to its activity (32). To verify whether the activity of the Gcn2 promoter could be affected by ILD, we generated a luciferase reporter construct (pGL3-Gcn2), carrying 1,090 base pairs (from −1187 to −98), upstream of the transcription start site in a luciferase reporter vector (pGL3-Basic). The activity of the Gcn2 promoter treated with ILD was markedly increased in primary hepatocytes (Fig. 4E). To confirm the function of the two CpG sites (CpG⁻⁷⁵⁸ and CpG⁻⁵⁷¹) in regulating the activity of the Gcn2 promoter, we further generated a point mutation luciferase reporter (CG→AT) in which the two sites of interest could not be methylated (27). As expected, the single and double mutants all increased the activity of the Gcn2 promoter, and the double mutants displayed higher activity than the single mutants in primary hepatocytes (Fig. 4F). We then wondered whether increasing the levels of methylation in the Gcn2 promoter could reverse the effect of ILD on the activity of the promoter. We in vitro methylated pCpGL-Gcn2 using HpaII (CCGG motif), HhaI (GCGC motif), and M.SssI (CG motif) methyltransferases and quantified luciferase activity in primary hepatocytes. Partial methylation of pCpGL-Gcn2 with HpaII and HhaI and full methylation with M.SssI significantly reduced luciferase activity (Fig. 4G). Under ILD treatment, the increased luciferase activity was also blocked by HpaII, HhaI, and M.SssI (Fig. 4G). Taken together, ILD could increase the expression of hepatic GCN2 by decreasing the methylation of CpG⁻⁷⁵⁸ and CpG⁻⁵⁷¹ in the promoter of Gcn2.

We next examined the mRNA expression of several epigenetic modifiers, including demethylases such as growth arrest and DNA damage inducible (GADD45)a, GADD45b, and GADD45g and ten-eleven translocation (TET)1, TET2, and TET3 and methylases such as DNA methyltransferase (DNMT)1, DNMT3a, and DNMT3b (33). Although these methylation modifiers showed substantial changes in the liver 7 days after ILD, only demethylase GADD45b remained at high levels 49 days after ILD (Fig. 5A and B and Supplementary Fig. 16A and B).

After transfection with siGadd45b, the decreased methylation of CpG⁻⁷⁵⁸ and CpG⁻⁵⁷¹ induced by ILD were reversed in the primary hepatocytes (Fig. 5C). As expected, the ILD-induced increase in Gcn2 promoter activity was reversed upon transfection with siGadd45b (Fig. 5D). The ILD-induced increase in Gcn2 mRNA was also reversed when Gadd45b was knocked down in the primary hepatocytes (Fig. 5E). The ILD-induced increases in the levels of t- and p-GCN2 also returned to normal when GADD45b was knocked down, as did downstream p-eIF2α, with consequent recovery of the insulin signaling pathway (Fig. 5F and G). To investigate the role of GADD45b in ILD in vivo, we constructed AAV-shGadd45b and AAV-scramble and injected them into WT mice via the tail vein. The knockdown efficiency of GADD45b was confirmed by Western blotting. The increased p-GCN2, t-GCN2, and p-eIF2α levels induced by ILD were reversed when GADD45b was knocked down (Fig. 5H). After the control diets were resumed for 7 days after ILD, the improved blood glucose, HOMA-IR, GTT, and ITT were reversed when hepatic GADD45b was knocked down, but the serum insulin levels were not changed (Fig. 5I–M). Together, the demethylation of hepatic Gcn2 in ILD mice was induced by GADD45b, and the in vivo experiments also confirmed the crucial role of GADD45b in the improved glucose metabolism mediated by ILD.

The effect of ILD was then tested in insulin-resistant mice. After maintenance on an HFD for 8 weeks, the mice were subjected to ILD treatment with a leucine-deficient HFD for seven cycles and then returned to the HFD for a further 49 days. The HFD-mediated impaired blood glucose, HOMA-IR, GTT, and ITT in the mice were improved by ILD treatment, but the serum insulin levels were not changed (Fig. 6A–E). Unlike the unchanged body weight and fat mass in ILD-treated WT mice, the body weight and fat mass in the ILD-treated HFD mice decreased, except for the lean mass (Supplementary Fig. 17A–C). Similar results showing glucose metabolism improvement were obtained in the db/db mice on the 49th days after ILD treatment as in the WT mice, but their body composition remained unchanged (Fig. 7A–E and Supplementary Fig. 18A–C). In conclusion, the ILD treatment could also improve glucose metabolism in insulin-resistant conditions.

Leucine deprivation significantly ameliorates insulin resistance (5,18); however, whether the effects of leucine deprivation can be extended was unknown. We found that seven cycles of ILD was sufficient to form a metabolic memory, producing long-term improvements in insulin sensitivity in both normal and insulin-resistant mice. Moreover, ILD had a direct effect on insulin sensitivity, independent of the change in body weight. Therefore, we demonstrated a new ILD regimen that produced long-lasting improvements in insulin sensitivity, without other side effects. IF has various biological benefits, including improved hepatic insulin sensitivity in animal models and humans (34). Our results suggest that the single nutrient leucine might play a fundamental role in IF-induced improvements in insulin sensitivity; however, this possibility requires future investigation. A recent study showed that leucine restriction has no effect on glucose metabolism (35), which seems to be different from our results. However, the difference in the treatment and leucine content between our and their study (18) might explain the difference in some of the results obtained.

Metabolic memories can be formed in various tissues (10,29,36). However, metabolic memory in the liver has not yet been reported. A striking finding from this study is that the improved insulin sensitivity induced by ILD existed only in the liver, though the reasons for this liver specificity remain unknown. We suspect that shorter durations of ILD might only affect the liver, whereas longer durations might affect the other tissues, as shown in another study reporting the difference between shorter and longer periods of IF treatment (37,38). This could be an advantage in avoiding side effects such as muscle loss and anorexia during long-term continuous leucine deprivation, thereby expanding the potential usage of leucine-deprived diets.

GCN2 is a well-known amino acid sensor that regulates lipid and glucose metabolism (18,26). Here, we demonstrate the crucial role of hepatic GCN2 in ILD-regulated hepatic insulin sensitivity, whereas the downstream signal is unknown. The S6K1 pathway mediates GCN2 regulation of insulin sensitivity (18), suggesting the possible involvement of it in ILD, which needs to be investigated in the future. Leucine deprivation normally activates GCN2 by stimulating its phosphorylation rather than by changing its expression (15). Unlike these results observed with a simple leucine deprivation treatment, both GCN2 phosphorylation and its expression levels were increased by ILD treatment. Because the serum leucine levels were not changed in ILD mice, we speculate that the increased phosphorylation of GCN2 was likely caused by the increase in GCN2 expression. Consistent with this possibility, we found that the mRNA levels of Gcn2 were increased by ILD, suggesting transcriptional control of its expression.

Epigenetic changes (e.g., DNA methylation), influenced by nutrient status (39), play an important role in gene expression regulation. DNA methylation often occurs on the cytosine residues of CpG dinucleotides to silence genes (40). However, the methylation pattern of Gcn2 has not previously been studied. We found that two CpG sites in the Gcn2 promoter (CpG⁻⁷⁵⁸ and CpG⁻⁵⁷¹) in the liver were demethylated by ILD treatment. Beyond its role in cell cycle arrest, apoptosis, and cell survival, increased GADD45b is reported to optimize lipid and glucose metabolism, which is in line with our results (41). In this study, we identified that the demethylation and expression of GCN2 are regulated by GADD45b. These results demonstrate a novel epigenetic mechanism underlying Gcn2 expression control via GADD45b. Increased GCN2 expression has been observed not only under conditions of essential amino acid deprivation but also in cancer and inflamed intestines (42,43). However, the underlying mechanisms remain unknown. Our results might provide important insights into the altered expression of GCN2 in other conditions.

Nevertheless, several questions remain unanswered. For example, the reasons for the specific effect of ILD on insulin sensitivity but not body weight in WT or db/db mice are unknown. Consistently, similar distinct effects have been reported with other IF treatments (44). We suspected that the effect on adipose tissue might have quickly vanished after the stimuli was stopped because of the obesogenic inflammation induced earlier (45). In contrast to the changes in normal mice, fat weight was decreased in ILD-treated HFD mice. We assumed that this reduced fat mass could be the result of the increased insulin sensitivity caused by ILD, as shown previously (46). Another interesting finding is that the effect of ILD might be branched-chain amino acid specific, based on the results of 1-day deficiencies of different amino acids. However, these possibilities must be tested in the future by investigating the effect of intermittent valine or isoleucine deprivation with more cycles.

Furthermore, the mechanisms by which ILD induces the long-lasting increase in GADD45b are still unclear. The expression of GADD45b is inhibited by the methylation of its promoter, and PPARa is a regulator of Gadd45b demethylation (47). Furthermore, how GADD45b influences the demethylation of Gcn2 is unknown. GADD45a, which also belongs to the GADD45 family, promotes TET1 recruitment and DNA demethylation at CpG island promoters (48). Similar mechanisms might exist for GADD45b. Finally, apart from DNA demethylation, other epigenetic modifications such as histone acetylation (49) could also participate in the regulation of GCN2 expression, which will be studied in the future.

In conclusion, our study suggests that ILD induces long-lasting, positive effects on insulin sensitivity through the formation of metabolic memory via a reduction in Gcn2 DNA methylation (Figs. 7F). This work provides novel insights into the mechanisms underlying long-lasting, IF-induced effects on metabolism and uncovers a potential nutritional intervention strategy for treating insulin resistance and related diseases. Our study also demonstrates novel epigenetic mechanisms involved in the regulation of Gcn2 gene expression.

Funding. This work was partly supported by grants from the National Key R&D Program of China (2018YFA0800600), the National Natural Science Foundation (91957207, 31830044, 81870592, 81770852, 81970742, 81970731, and 82000764), Chinese Academy of Sciences Interdisciplinary Innovation Team, and China Postdoctoral Science Foundation (2020M681433 and 2021T140690).

Duality of Interest. This work was also supported by Novo Nordisk–Chinese Academy of Sciences Research Fund (NNCAS-2008-10). No other potential conflicts of interest relevant to this article were reported.

Author Contributions. H.Y. and F.G. planned and supervised the experimental work and data analysis. H.Y. performed the experiments. H.Y. and F.Y. wrote the manuscript. F.Y., F.J., Y.N., X.J., J.D., and Y.G. researched data and provided technical support. S.C., Q.Z., C.H., and Y.L. helped with planning for the experiments, provided technical support, and contributed to research design, data analysis, and discussion, F.G. directed the project, contributed to discussion, and wrote, reviewed, and edited the manuscript. The manuscript was revised and approved by all authors. F.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.