Role of activating transcription factor 4 in the hepatic response to amino acid depletion by asparaginase

The anti-leukemic agent asparaginase activates the integrated stress response (ISR) kinase GCN2 and inhibits signaling via mechanistic target of rapamycin complex 1 (mTORC1). The study objective was to investigate the protective role of activating transcription factor 4 (ATF4) in controlling the hepatic transcriptome and mediating GCN2-mTORC1 signaling during asparaginase. We compared global gene expression patterns in livers from wildtype, Gcn2 −/−, and Atf4 −/− mice treated with asparaginase or excipient and further explored selected responses in livers from Atf4 +/− mice. Here, we show that ATF4 controls a hepatic gene expression profile that overlaps with GCN2 but is not required for downregulation of mTORC1 during asparaginase. Ingenuity pathway analysis indicates GCN2 independently influences inflammation-mediated hepatic processes whereas ATF4 uniquely associates with cholesterol metabolism and endoplasmic reticulum (ER) stress. Livers from Atf4 −/− or Atf4 +/− mice displayed an amplification of the amino acid response and ER stress response transcriptional signatures. In contrast, reduction in hepatic mTORC1 signaling was retained in Atf4 −/− mice treated with asparaginase. Conclusions: GCN2 and ATF4 serve complementary roles in the hepatic response to asparaginase. GCN2 functions to limit inflammation and mTORC1 signaling whereas ATF4 serves to limit the amino acid response and prevent ER stress during amino acid depletion by asparaginase.


Results
Whole body responses to asparaginase in Atf4 −/− mice differ from Gcn2 −/− mice. We sought to understand the role of ATF4 relative to GCN2 in hepatoprotective responses to asparaginase treatment. To accomplish this and to capture both the acute as well as the auxiliary stress responses to asparaginase, we injected asparaginase (3 IU/g) into WT, Gcn2 −/− , and Atf4 −/− mice once daily for 8 d using phosphate buffered saline (PBS) as a control. We chose this time point based on our previous studies which show that a single injection of asparaginase alters hepatic signaling and transcriptional responses in WT and Gcn2 −/− mice similar to multiple injections 5,10,11 . Before treatment commenced, Atf4 −/− mice had significantly less fat mass than WT and Gcn2 −/− mice ( Supplementary Fig. S1), in agreement with the observation that Atf4 −/− mice are leaner than Atf4 +/+ mice 27 . Following asparaginase treatment, WT mice experienced minimal change in body weight and body composition, but Atf4 −/− and Gcn2 −/− mice both lost body weight and body fat ( Supplementary Fig. S1). Compared to WT mice, asparaginase increased liver and pancreas mass and reduced spleen mass in Gcn2 −/− mice ( Fig. 1) in accordance with our previous findings 6, 10,12,27 . None of these asparaginase-associated effects in Gcn2 −/− mice were observed in Atf4 −/− mice; instead, Atf4 −/− organ/tissue masses did not significantly change following asparaginase treatment. ATF4 and GCN2 control overlapping patterns of hepatic gene expression in response to asparaginase. To further define the role of ATF4 and GCN2 in liver in response to asparaginase, we performed transcriptional profiling of poly-A mRNA via massively parallel sequencing of cDNA (RNA-seq). To accomplish this, total RNA was extracted from liver of WT, Gcn2 −/− and Atf4 −/− mice treated with asparaginase or PBS excipient. After RNA-Seq, we examined data quality by examining the percentages of mapped reads and evaluated fragments per kilobase of transcript per million mapped reads (FPKM) values for isoforms, transcripts and annotated genes (Supplementary Table S1). Next, we confirmed the Gcn2 −/− and Atf4 −/− deletions at the genome level ( Supplementary Fig. S1). We also assessed the degree of global changes in gene expression among the samples visualized as cross-comparison Volcano plots ( Supplementary Fig. S3). Furthermore, Principle Component Analysis indicated that transcriptional profiles from the livers of asparaginase-treated Atf4 −/− and Gcn2 −/− mice were quite different from the other treatment groups ( Supplementary Fig. S3).
A comparative examination of PBS-injected mice by Venn analysis showed many hepatic genes required Gcn2 and/or Atf4 for basal expression ( Fig. 2a; Supplementary Table S2). We found 128 genes were expressed differently upon Gcn2 deletion whereas deletion of Atf4 changed basal expression of 145 genes compared to WT. Among these, 65 genes were altered in both Gcn2 −/− and Atf4 −/− mice (category B). Ingenuity pathway analyses of these gene categories ( Fig. 2b; Supplementary Table S3) indicated that loss of either Gcn2 or Atf4 resulted in basal changes in the expression of genes involved in calcium and protein kinase A signaling, chemokine receptor 4 (CXCR4) signaling, glycolysis and gluconeogenesis, creatine phosphate biosynthesis, retinoid acid receptor (RXR) signaling and function, bile acid biosynthesis and acute phase response signaling. Loss of GCN2 in basal state uniquely affected p38 MAPK signaling, steroid hormone biosynthesis, and growth arrest and DNA damage 45 (GADD45) signaling in liver. Loss of ATF4 in basal state uniquely affected cytochrome P450-mediated processes and oxidative phosphorylation. Thus, while the global influence of GCN2 in the non-stressed liver was largely mediated by ATF4, a few key biological processes were unique to each protein. These findings indicate that both GCN2 and ATF4 function to regulate gene expression in the absence of apparent stress and these key components of the ISR serve complementary roles in regulating liver metabolism and signaling.
We then utilized Venn analysis to identify shared versus unique hepatic gene patterns altered in Gcn2 −/− and Atf4 −/− mice treated with asparaginase (ASNase versus PBS in each genetic strain) (Fig. 2c, Supplementary  Table S2). In WT mice, asparaginase altered expression of 195 genes compared to PBS. Consistent with our Principal Component Analysis, deficiency in either Gcn2 or Atf4 promoted much larger asparaginase-induced shifts in gene expression, reflected by increased numbers of genes in Gcn2 −/− (915 genes) and Atf4 −/− (1,155 genes) categories. Roughly one-half of these gene expression changes were common to loss of Gcn2 and Atf4 but not shared with WT (category F). Among the genes altered by asparaginase, 114 genes in Gcn2 −/− were not shared with either Atf4 −/− or WT (category E) whereas 337 genes altered in Atf4 −/− mice were not shared with Gcn2 −/− or WT (category G). Ingenuity pathway analysis of these specified gene categories ( Fig. 2d; Supplementary  Table S3) identified hepatic inflammation, endocytosis and phagosome maturation as altered by Gcn2 status (category E) whereas loss of Atf4 influenced signaling via eIF2, eIF4, p70S6K/mTOR, ER stress and the UPR as well as cholesterol biosynthesis (category G). We also found that a variety of metabolic processes were misaligned in both Gcn2 −/− and Atf4 −/− mice (category F). For example, genes implicated in cytoplasmic nuclear receptor activation, glutathione-mediated reactions, nuclear factor 2 (NRF2)-mediated oxidative stress, triglyceride, ethanol and amino acid degradation, fatty acid oxidation, acute phase response, cytochrome P450-mediated reactions, bile acid synthesis, and hepatic fibrosis were altered in both Gcn2 −/− and Atf4 −/− mice. Additional examination of shared gene categories by heat map showed gene expression changes in Gcn2 −/− and Atf4 −/− in category F were largely in the same direction whereas many gene expression changes in category I were not shared in direction ( Supplementary Fig. S4).
Among the biological processes identified, cytochrome P450-mediated reactions and activation of cytoplasmic nuclear receptors (e.g., FXR, RXR, PXR, LXR, VDR) stood out as shared among categories E, F, G. Interestingly, biomolecular data mining using EVEX (evexdb.org) indicated that ATF4 binds LXR, PXR and RXR and coregulates multiple gene families; many of these encode transcriptional and translational responses to ER and oxidative stress in addition to the metabolism of xenobiotics, lipids, cholesterol and bile acids 28,29 . The directional change in the expression of genes involved in these two biological processes was for the most part shared, reflecting downregulation during asparaginase exposure when the ISR is disrupted.
Loss of Atf4 versus Gcn2 differentially regulates the amino acid response to asparaginase. GCN2 is critical for activation of the ISR by asparaginase in liver but to what extent its downstream effector ATF4 in the implementation of the resulting AAR transcriptional signature is undefined. To address this question, we first measured phosphorylation of eIF2 alpha (p-eIF2) and found that in contrast to loss of Gcn2, loss of Atf4 amplified basal p-eIF2 similar to levels induced by asparaginase (Fig. 3a). To visualize how these differences in phosphorylation impacted the amino acid response we crafted a z-score heat map of selected hepatic genes, with the selection list based on a previously published mRNA expression profile in liver cells subjected to amino acid deprivation 30 (Fig. 3b). Cluster analysis of these data again revealed shared and unique response patterns upon asparaginase treatment: reduced AAR gene levels in both Gcn2 −/− and Atf4 −/− (I), increased AAR gene levels in both Gcn2 −/− and Atf4 −/− (II) and increased AAR gene levels in Atf4 −/− but not Gcn2 −/− (III). We used RT-qPCR to further assess specific gene transcripts and confirmed that loss of either Gcn2 −/− or Atf4 −/− altered Fgf21 mRNA expression to asparaginase in a similar direction (Fig. 3c) whereas Asns, Atf3, Eif4ebp1, and Sestrin2 were differentially altered (Fig. 3d,e; Supplementary Fig. S5). These analyses indicate that overall, loss of Gcn2 dampens the AAR whereas loss of Atf4 amplifies the AAR following asparaginase exposure.
Deletion of Atf4 promotes ER stress to asparaginase. Amplification of the hepatic AAR in asparaginase-treated Atf4 −/− mice and the pathway analysis revealed that components of the ER stress pathway were also highly enriched. To test this premise further, we assessed the phosphorylation status of PERK by immunoblot, examining electrophoretic migration of PERK protein and measuring signal intensity following incubation with an antibody that specifically recognizes PERK phosphorylated at threonine 980. Both immunoblots showed that PERK phosphorylation was not detectable in any samples, only the positive control (Fig. 4a). However, CHOP protein was detectible in the liver of asparaginase-treated Atf4 −/− mice (Fig. 4b). Also, an important measure of ER stress, Xbp1 mRNA splicing, was greatly elevated in livers from asparaginase-treated Atf4 −/− mice (Fig. 4c) as was Atf6 mRNA expression (Fig. 4d). These results indicate that loss of ATF4 function during sustained amino acid depletion results in selective engagement of the UPR upon asparaginase treatment by a mechanism that does not require PERK phosphorylation.
GCN2 regulates mTORC1 independent of ATF4. Mice lacking Gcn2 show hyperactivation of mTORC1 in liver during asparaginase 5,11 and during dietary leucine deprivation 31,32 . To explore the connection among GCN2, ATF4, and mTORC1 in response to asparaginase, we first examined the phosphorylation state of S6K1 and 4E-BP1, two readouts of mTORC1 activity. Consistent with our previously published findings, phosphorylation levels of S6K1 and 4E-BP1 in liver were stable or even reduced by asparaginase in WT but amplified in Gcn2 −/− mice (Fig. 5a-c). One idea to explain this disruption involves GCN2 regulation of mTORC1 via ATF4 23 . In contrast with this idea, mTORC1 signaling was repressed in Atf4 −/− similar to WT. Further, Ei4ebp1 hepatic gene expression was increased by asparaginase in WT and Atf4 −/− mice but not Gcn2 −/− mice ( Supplementary Fig. S5). To address the underlying mechanism, we measured phosphorylation of protein kinase B/Akt (p-Thr308) to determine if hyperactivated mTORC1 was due to enhanced insulin/growth factor signaling in the Gcn2 −/− condition (Fig. 5a,d). However, Akt phosphorylation levels were unchanged in WT and reduced in both Gcn2 −/− and Atf4 −/− , indicating that differences in mTORC1 activity were not due to increased growth factor signaling via Akt. We also examined Sestrin2 because increases in expression and phosphorylation inversely correlate with mTORC1 activity during leucine deprivation 25,33 . In contrast with this idea, asparaginase increased Sestrin2 mRNA expression and protein phosphorylation in all mouse strains, with Atf4 −/− mice demonstrating the largest increases (Supplementary Fig. S5; Fig. 5a,e). These results indicate that GCN2 but not Sestrin2 or ATF4 are required for hepatic mTORC1 inhibition during sustained amino acid depletion by asparaginase. Heterozygous loss of Atf4 promotes hepatic ER stress and apoptosis to asparaginase. Atf4 −/− mice exhibit pleiotropic phenotypes of reduced growth, lack of vision, and defects in hematopoiesis and glucose homeostasis 27,[34][35][36] . These health challenges may serve as confounders in the assessment of Atf4 deletion during asparaginase exposure. To help assess this, we decided to examine the impact of asparaginase exposure in Atf4 heterozygotes (Atf4 +/− ) which appear normal and healthy. To understand if ATF4 haploinsufficiency altered the hepatic ISR and health status to asparaginase, we examined the effects of asparaginase on Atf4 +/− mice compared to WT and Atf4 −/− . Prior to treatment, Atf4 −/− but not Atf4 +/− mice were leaner than Atf4 +/+ mice  ( Supplementary Fig. S6). Asparaginase reduced body weight in all strains, with Atf4 +/− mice displaying body weights that were intermediate between Atf4 +/+ and Atf4 −/− mice (Supplementary Fig. S6). Asparaginase did not significantly alter liver or pancreas mass but reduced spleen weight in all strains ( Supplementary Fig. S6). Asparaginase also elevated liver triglycerides in all strains, although to a lesser extent in Atf4 −/− mice (Supporting Fig. S6).

Discussion
Liver dysfunction as a result of oxidative stress and hepatic steatosis is an asparaginase-associated adverse event that complicates remission induction and reduces treatment success [38][39][40] . A maladaptive GCN2/eIF2-directed ISR is proposed by our laboratory as a molecular basis for asparaginase-associated metabolic complications and hepatotoxicity 5, 6, 10-12 but the role of ATF4 in this process is unknown. Herein we present a global view of how genetic loss of Gcn2 versus Atf4 influences hepatoprotection during asparaginase treatment by altering the liver transcriptome and signal transduction responses. First, we identify shared versus unique hepatic processes altered by GCN2 versus ATF4 status in the basal state and during chronic amino acid depletion by asparaginase. Second, we find that coordination of mTORC1 signaling by GCN2 does not require ATF4 or Sestrin2. Finally, we determine that both total and partial loss of Atf4 expression is unable to self-limit the adaptive AAR to asparaginase, resulting in ER stress and greater cell death. These findings both confirm and challenge prevailing ideas on ATF4 function and add new mechanistic insight into how the liver adapts to amino acid deprivation.
This study provides the first view of the hepatic transcriptome on the background of global Atf4 deletion and a first comparison to global deletion of Gcn2. This study strength must be viewed in balance with limitations that are inherent when studying mice with whole body gene deletion. Specifically, because ATF4 is a master regulator of metabolism and thermogenesis, global deletion of Atf4 results in a lean phenotype with defects in eye, bone and blood cell development 27,[34][35][36] . Potential confounding influences of these health challenges on the liver transcriptome are acknowledged, particularly in light of the chronic treatment imposed. Indeed, these concerns formed the basis for pursuing experiments in Atf4 +/− mice, which are healthy. In this regard, we note that the hepatic AAR response patterns following asparaginase exposure in Atf4 +/− and Atf4 −/− mice are directionally similar. Also, studies by our group and others using mice harboring a liver-specific deletion of Atf4 show that its expression in liver plays important roles in regulating processes that were also identified in the current study such as cholesterol and lipid metabolism and oxidative stress responses 18,41 . Furthermore, injection of asparaginase into obese mice shows that genetic ablation of Atf4 in liver does not amplify mTORC1 signaling but instead amplifies a transcriptional signature similar to the current results 42 . Thus, the likelihood these outcomes are nonspecific or not relevant to ATF4 is low.
Mice lacking Gcn2 do not show evidence of hepatic ER stress to asparaginase 11 . In contrast, Atf4 −/− mice display an ER stress transcriptional signature in liver upon asparaginase treatment. These results suggest that ATF4 is necessary to self-limit the AAR in order to prevent inappropriate accumulation of an ER stress response transcriptional signature. Recently, we reported that in the absence of Atf4, ATF6 and CHOP assume auxiliary functions during pharmacological ER stress, driving the cell toward a maladaptive cell fate 18 . This report supports our current findings showing that in response to asparaginase treatment, expression of CHOP, Atf6 and IRE1-directed splicing of Xbp1 mRNA are potently induced in the livers of Atf4-deficient mice. These signals may allow other transcription factors to gain auxiliary function, directing the organism toward cellular death pathways rather than adaptive health outcomes. Additional studies are needed to more fully define how the binding partners versus targets of ATF4 direct cellular consequences to asparaginase. This is especially important because it will help explain why loss of one Atf4 allele amplifies the ER stress transcriptional signature following asparaginase exposure.
Our lab was the first to identify hyperactivation of mTORC1 signaling in the Gcn2 −/− condition during leucine deprivation and asparaginase 5,31 , leading to the hypothesis that GCN2 or another ISR target coordinates amino acid sensing with mTORC1 signaling. One candidate is Sestrin2, an mTORC1 inhibitor regulated by ATF4 under conditions of amino acid deprivation or ER stress [23][24][25]33 . Under conditions of amino acid deficiency, Sestrin2 becomes highly phosphorylated, promoting its interaction with a complex that subsequently blocks lysosomal localization of mTORC1 25,43 . Using cells in culture, these studies identify a reciprocal relationship between the degree of Sesn2 phosphorylation and the degree of mTORC1 inhibition, specific to leucine availability 25 . However, others report that deprivation of other amino acids, such as glutamine, can also activate Sestrin2 inhibitory effect toward mTORC1 23 . Our results show Sestrin2 mRNA expression levels and phosphorylation of the protein are increased following asparaginase exposure even in the absence of Gcn2, Atf4 or PERK activation and are not reciprocally related to mTORC1 signaling. We speculate that in the absence of GCN2-mediated signaling, Sestrin2 may be induced by a different profile of transcription factors such as CCAAT-enhancer-binding protein beta 44 which in the current study was significantly elevated only in the livers of Gcn2 −/− mice exposed to asparaginase (FPKM: Gcn2 −/− PBS = 38; Gcn2 −/− ASNase = 103, q = 0.0008). In the absence of Atf4, we suggest that auxiliary action by other ER stress-responsive transcription factors may promote Sestrin2 expression, based on experiments showing that both the PERK and the IRE1/XBP1 arms of the UPR contribute to regulating Sestrin2 45 . Experimental confirmation of these ideas still leaves open the question why is hepatic mTORC1 activity unleashed in Gcn2 −/− but not Atf4 −/− mice? Fafournoux's group shows that GCN2-regulation of mTORC1 in cultured cells is ATF4-independent but involves phosphorylation of eIF2 46 . Our findings partially align with this idea, leaving open the need to resolve the contribution of eIF2 phosphorylation by other kinases to mTORC1 signaling versus independent control of mTORC1 by other transcription factors such as TRIB3 42 versus regulation of mTORC1 secondary to the impact of Gcn2 deletion on autophagy 12 .
In conclusion, this is the first comprehensive dataset identifying the role of Gcn2 versus Atf4 in control of hepatic transcriptome levels both basally and during chronic amino acid depletion. This information emphasizes how the function of the AAR contributes to the molecular basis of hepatoprotection in response to asparaginase. Application of these findings to dietary amino acid insufficiency in general is evident based on a recent study 47 showing that multiple dietary restriction models elicit a gene expression signature sharing many of the same overrepresented canonical pathways as the current study and another report 48 showing that removal of asparagine and/or glutamine from cells activates the ISR similar to dietary protein dilution. Future areas of investigation will use this information to develop personalized screening tools and/or discover targeted therapies to prevent adverse events to asparaginase or other therapies that utilize amino acid depletion as a treatment strategy. Methods Animals and Care. All animal protocols were reviewed and approved by the Institutional Animal Care and Use Committee at Rutgers, The State University of New Jersey and all animals received humane care according to the criteria outlined in the "Guide for the Care and use of Laboratory Animals" prepared by the National Academy of Sciences and published by the National Institute of Health (NIH publication 86-23 revised 1985) and ARRIVE (articulated at www.nc3rs.org.uk/ARRIVE). In these experiments, male and female C57BL/6J mice (8-12 wk old) carrying a homozygous deletion of Gcn2 (Gcn2 −/− ) or Atf4 (Atf4 −/− ) were used alongside wildtype controls (WT, Atf4 +/+ ) and Atf4 heterozygote nulls (Atf4 +/− ) (Jackson Laboratories, Bar Harbor, ME).
RNA-Sequencing. RNA-Seq was performed at the JP Sulzberger Columbia Genome Center (Columbia University, NY, NY). Poly-A purified RNA was used with the TruSeq RNA Sample Prep Kit v2 from total RNA having a RIN ≥ 8.0. Libraries were sequenced on an Illumina HiSeq 2500 with a read depth of 30 million, 100 bp single-end reads. The resulting fastq files were subject to the Tuxedo bioinformatics pipeline 49,50 . The reads were mapped to the mouse genome (mm10) with Tophat v2.0 then differences in gene expression were determined using Cuffdiff and evaluated according to drug treatment and genetic strain.
Statistical Analysis. Global transcriptome analyses (n = 3 per group) were conducted in R using cummeRbund. Differentially expressed genes were considered to be statistically significant when the q value (or False Discovery Rate, FDR) was <0.1 (unadjusted p < 0.017). All other measurements utilized 3-6 animals per group and are reported as the average value ± standard error of the mean (SEM). Homoscedasticity was evaluated using Levene's Test and data normality evaluated using Shapiro-Wilk-Test. Log transformation of data was performed when assumptions of distribution normality and homoscedasticity were violated. Differences between multiple treatment groups were analyzed by one and two factor ANOVA. When a statistically significant overall difference was detected, differences among individual means were evaluated using Tukey's post-hoc test with an alpha level of P < 0.05 (STATISTICA; StatSoft, Tulsa, OK).