eIF2α kinases GCN2 and PERK modulate transcription and translation of distinct sets of mRNAs in mouse liver

Abstract

In eukaryotes, selective derepression of mRNA translation through altered utilization of upstream open reading frames (uORF) or internal ribosomal entry sites (IRES) regulatory motifs following exposure to stress is regulated at the initiation stage through the increased phosphorylation of eukaryotic initiation factor 2 on its α-subunit (eIF2α). While there is only one known eIF2α kinase in yeast, general control nonderepressible 2 (GCN2), mammals have evolved to express at least four: GCN2, heme-regulated inhibitor kinase (HRI), double-stranded RNA-activated protein kinase (PKR), and PKR-like endoplasmic reticulum-resident kinase (PERK). So far, the main known distinction among these four kinases is their activation in response to different acute stressors. In the present study, we used the in situ perfused mouse liver model and hybridization array analyses to assess the general translational response to stress regulated by two of these kinases, GCN2 and PERK, and to differentiate between the downstream effects of activating GCN2 versus PERK. The resulting data showed that at least 2.5% of mouse liver mRNAs are subject to derepressed translation following stress. In addition, the data demonstrated that eIF2α kinases GCN2 and PERK differentially regulate mRNA transcription and translation, which in the latter case suggests that increased eIF2α phosphorylation is not sufficient for derepression of translation. These findings open an avenue for more focused future research toward groups of mRNAs that code for the early cellular stress response proteins.

global rates of protein synthesis are repressed in response to a variety of cell stresses, including UV irradiation (14), infection by certain viruses (17, 21, 33, 56, 59, 64, 71, 77), and deprivation of essential amino acids (18, 35, 36). In part, the repression of global protein synthesis associated with such stresses is a consequence of phosphorylation of Ser51 on the α-subunit of the translation initiation factor eukaryotic initiation factor (eIF)2 (33, 56, 59, 64, 81). During one of the first steps in the initiation of mRNA translation, eIF2 binds to GTP and the initiator methionyl-tRNA_i to form the ternary complex that subsequently binds to the 40S ribosomal subunit (29). In a later step in initiation, the GTP bound to eIF2 is hydrolyzed to GDP, and the resulting eIF2·GDP complex is released from the 40S ribosomal subunit. For initiation to recur, the GDP bound to eIF2 must be exchanged for GTP, a reaction catalyzed by another initiation factor, eIF2B. Phosphorylation of eIF2α on Ser51 converts eIF2 from a substrate into a competitive inhibitor of eIF2B, leading to repression of global rates of protein synthesis (70). Paradoxically, inhibition of eIF2B by phosphorylated eIF2 leads to upregulated translation of mRNAs with particular motifs in their 5′-untranslated region (UTR), such as upstream open reading frames (uORF) and internal ribosomal entry sites (IRES) (41, 68, 73, 74). Although information concerning the number and identity of mRNAs whose translation is upregulated in response to eIF2α phosphorylation is limited, those that have been identified typically encode proteins involved in cellular adaptation to the stress (41).

The genome of the yeast Saccharomyces cerevisiae encodes a single eIF2α kinase, general control nonderepressible 2 (GCN2) (15). GCN2 is activated in response to nutrient deprivation and, in particular, to deprivation of amino acids (39). In contrast, mammals have four distinct eIF2α kinases: a mammalian ortholog of GCN2, the double-stranded RNA-activated protein kinase (PKR), the heme-controlled inhibitor (HRI), and the PKR-like endoplasmic reticulum (ER)-resident kinase (PERK) (reviewed in Ref. 76). Each of these kinases is activated in response to distinct stresses: GCN2 by deprivation of essential amino acids, PKR by infection with certain viruses, HRI during heme limitation, and PERK by ER stress [e.g., during the unfolded protein response (UPR)]. Phosphorylation of eIF2α by any of the four kinases results in inhibition of eIF2B and repressed global rates of protein synthesis. In addition, phosphorylation of eIF2α is thought to be important in effecting a recovery from the stress. For example, phosphorylation of eIF2α by PERK in response to accumulation of unfolded proteins in the lumen of the ER represses the synthesis of proteins destined for that subcellular location, thereby minimizing further accumulation of unfolded proteins (28). However, PERK-mediated phosphorylation of eIF2α also leads to increased translation of mRNAs encoding certain transcription factors such as activating transcription factor (ATF)4 (aka C/ATF/CREB2/TAXREB67) and ATF5 (aka AFTA/Atf7/Arfx/ODA-10) that promote the transcription of genes encoding proteins such as growth arrest and DNA-damage-inducible protein (GADD)34 (aka PPR1R15A) (55). GADD34 targets protein phosphatase 1 to eIF2, resulting in dephosphorylation of the protein and restoration of global protein synthesis. If this second function, i.e., increased translation of mRNAs encoding specific proteins, is as crucial as that of repressing global protein synthesis, it is logical to surmise that each of the eIF2α kinases would effect a recovery program specific to its activating stress, hence influencing the translation efficiency of different subsets of mRNAs.

In the present study we hypothesized that activation of different eIF2α kinases affects the translational regulation of both a general subset of mRNAs involved in determining cell fate as well as a specific subset of mRNAs involved in responding to the activating stress. To test the hypothesis, we performed in situ liver perfusion studies using wild-type mice and mice lacking either GCN2 (Gcn2−/−) or PERK (Perk−/−). GCN2 was activated by deprivation of the essential amino acid Met, and PERK was activated with 2,5-di-tert-butylhydroquinone (tBuHQ), a selective inhibitor of ATP-dependent microsomal Ca²⁺ sequestration and activator of ER stress (32, 51). Overall, the results support the hypothesis, i.e., activation of GCN2 resulted in altered translation of both overlapping and distinct mRNAs compared with activation of PERK.

MATERIALS AND METHODS

Animals.

All animals received water and food (Harland) ad libitum and were maintained in accordance with the protocol approved by the Pennsylvania State College of Medicine's Institutional Animal Care and Use Committee. Global Gcn2−/− and wild-type littermate mice, 3–6 mo old, were generated according to a previously published protocol (81). The liver-specific knockout of Perk was generated by crossing the AlbCre deletor strain by the floxed Perk allele (Perk f), yielding AlbCre Perkf/f and wild-type littermate mice, also 3–6 mo old, as previously described (80). The AlbCre Perkf/f mice are referred to as Perk−/− here. Both Gcn2 and Perk mice are on a mixed C57BL/6J-129SvEvTac genetic background that has been intercrossed for several generations, and thus have a similar genetic background.

Liver perfusion.

Livers were perfused with a single-pass protocol as described previously (20, 72) with slight modifications. The perfusion medium contained a 2-to-1 ratio of buffer A [modified Ringer II buffer (mM): 110 NaCl, 4.4 KCl, 2.36 CaCl₂, 1.1 KH₂PO₄, 1.1 MgSO₄, 25 NaHCO₃, and 11 glucose, with 3% albumin] and washed bovine red blood cells. For perfusions of Gcn2+/+ and −/− mice, Met was omitted from the perfusate for the treatment condition. For perfusions of Perk+/+ and −/− mice, the final CaCl₂ concentration was lowered to 0.1 mM in buffer A. To induce ER stress, tBuHQ dissolved in DMSO was added to the perfusate immediately before perfusion to yield a final concentration of 40 μM. An equal volume of DMSO was added to the control perfusate. All perfusions were performed at 37°C for 35 min.

eIF2B guanine nucleotide exchange assay.

The guanine nucleotide exchange activity of eIF2B was measured with published protocols (13, 37). Perfused liver samples were homogenized with a Dounce homogenizer in 4 volumes of buffer B (mM: 45 HEPES, pH 7.4, 0.375 MgOAc, 0.075 EDTA, 95 KOAc, and 2 digitonin, with 10% glycerol and 3 μM microcystin), and the homogenate was centrifuged at 10,000 g for 10 min at 4°C. Supernatant (25 μl) was added to a tube containing 97 μl of H₂O and 140 μl of eIF2B assay buffer (mM: 52.1 MOPS, pH 7.4, 0.22 GDP, 104 KCl, 1.04 DTT, and 2.08 MgOAc, with 0.21 mg/ml bovine serum albumin), and the mixture was preincubated at 30°C for 1 min. The reaction was started by the addition of 35 μl of [³H]GDP·eIF2 binary complex followed by incubation at 30°C. Aliquots were removed from the assay reaction mixture into 2.5 ml of wash buffer (mM: 50 MOPS, pH 7.4, 2 MgOAc, 100 KCl, 1 DTT) at 0, 30, 60, and 90 s and vacuum filtered through 0.45-μm cellulose nitrate membrane filters (Whatman International, Maidstone, UK). The membranes were washed twice with 2.5 ml of wash buffer, dissolved in Filtron-X (National Diagnostics, Atlanta, GA) and the radioactivity contained thereon was measured in a scintillation counter (Beckman LS 6500 multipurpose scintillation counter).

SDS-polyacrylamide gel electrophoresis and Western blot analysis.

Perfused liver samples were homogenized in 7 volumes of buffer C (mM: 20 HEPES, pH 7.4, 2 EGTA, 50 NaF, 100 KCl, 0.2 EDTA, 50 β-glycerophosphate, 1 DTT, 0.1 PMSF, 1 benzamidine, and 0.5 NaVO₄, with 10 μl/ml Sigma protease inhibitor cocktail). The homogenate was centrifuged at 1,000 g for 3 min at 4°C, and an aliquot of the supernatant was added to an equal volume of SDS sample buffer [0.125 M Tris·HCl, pH 6.8, 25% (vol/vol) glycerol, 2.5% SDS, 2.5% (vol/vol) β-mercaptoethanol, 0.2% bromophenol blue] and boiled at >95°C for 4 min. Protein samples were stored at −70°C until resolved by SDS-PAGE. After electrophoresis, proteins in the gel were transferred onto 0.45-μm polyvinylidene difluoride (PVDF) membranes (Pall Life Sciences, Exton, PA) and the membranes were blocked with 5% nonfat dry milk. Membranes were then incubated with primary antibody overnight at 4°C using the following antibodies: anti-phospho-Ser51-eIF2α (Biosource International, Invitrogen, Carlsbad, CA) and monoclonal anti-eIF2α (63). The next day membranes were incubated with secondary antibody at room temperature for 1 h with horseradish peroxidase-conjugated anti-rabbit or -mouse antibodies (Bethyl, Montgomery, TX), respectively. Membranes were developed with enhanced chemiluminescence reagents (Amersham Biosciences, Piscataway, NJ), and images were captured with GeneGnome software (SynGene, Frederick, MD) and quantitated with GeneTools software (SynGene).

Sucrose density gradient centrifugation.

Discontinuous nine-step 20% [mM: 10 HEPES, pH 7.4, 250 KCl, 5 MgCl₂, 0.5 EDTA, with 20% (wt/wt) sucrose] to 47% [mM: 10 HEPES, pH 7.4, 250 KCl, 5 MgCl₂, 0.5 EDTA, with 47% (wt/wt) sucrose] sucrose density gradients were used to separate mRNAs based on the number of bound ribosomes, an indicator of translational state. Perfused liver samples (1 g) were homogenized with a Dounce homogenizer in 3 volumes of buffer D (mM: 50 HEPES, pH 7.4, 250 KCl, 5 MgCl₂, 250 sucrose, 100 μg/ml cycloheximide), and the homogenate was centrifuged at 3,000 g for 15 min at 4°C. The supernatant (1 ml) was removed to a separate tube, and 10% (wt/vol) Triton X-100 and 13% (wt/wt) sodium deoxycholate (100 μl each) were added and gently mixed. Detergent-containing supernatant (1 ml) was loaded onto the 20–47% sucrose density gradients and centrifuged at 141,371 g for 3 h 38 min at 4°C for resolution of polysomes or 91,287 g for 19 h 9 min at 4°C for resolution of ribosomal subunits and monomers. After centrifugation, gradients were fractionated with an ISCO UV gradient fractionator while the absorbance at 254 nm was continuously monitored (Teledyne Isco, Lincoln, NE). Quantitation of the area under the absorbance curves was performed with the Un-Scan-It software program (version 4.3, Silk Scientific, Orem, UT).

RNA extraction.

Total RNA was extracted from 1 ml of detergent-containing supernatant or from the sucrose density gradient fractions with acid phenol:chloroform (Ambion, Austin, TX). Per milliliter of sample, 40 μl of 0.5 M EDTA, 25 μl of 20% SDS, and 1 volume of acid phenol:chloroform were added. The mixture was centrifuged at 3,100 g for 30 min at 4°C, the top clear layer was removed, and RNA was extracted a second time by the addition of 1 volume of acid phenol:chloroform. RNA was precipitated overnight in 2.5 volumes of 100% ethanol and 0.1 volume of 5 M NH₄OAc (Ambion). RNA was pelleted, washed twice with 75% ethanol, and resuspended in RNA storage solution plus anti-RNase (Ambion). The concentration and purity of RNA samples were measured with an Agilent 2100 Bioanalyzer in the Pennsylvania State University College of Medicine Functional Genomics Core Facility before hybridization array and quantitative (q)RT-PCR analysis.

Hybridization array analysis.

Total RNA from two perfused liver samples per condition (unfractionated, nonpolysomal, subpolysomal, or polysomal) was pooled to minimize biological variation, and the experiment was performed in duplicate. Twelve micrograms of total RNA was used in the initial cDNA conversion step. All steps including double-strand cDNA synthesis, in vitro labeling, and cRNA fragmentation were done with the One-Cycle Target Labeling and Control Reagents kit and protocol (Affymetrix, Santa Clara, CA). Samples were hybridized to GeneChip Mouse Genome 430 2.0 arrays (Affymetrix) in a GeneChip Hybridization Oven 640 and washed and stained in a GeneChip Fluidics Station 400, with reagents from the above-mentioned kit. Arrays were scanned with a GCS 3000 scanner, using GCOS 1.2 software, in the Pennsylvania State University DNA Microarray Facility. Hybridization array data have been submitted to the National Center for Biotechnology Information's Gene Expression Omnibus according to the MIAME (Minimum information about a microarray experiment) protocol [Ref. 7; GSE11685, http://www.ncbi.nlm.nih.gov/projects/geo/query/acc.cgi?acc=GSE11685].

Affymetrix Chip files containing qualitative and quantitative values of each probe set on the arrays were imported into GeneSpring (Agilent Technologies, Palo Alto, CA) and sorted for raw expression values >80 in at least 1 of 16 conditions, as an initial filter to eliminate background noise. The resulting lists were imported into Microsoft Excel and analyzed with the Z-score method (10, 48). First, the data were log₁₀ transformed. The transformed values were used in the following calculations: Z_score = (intensity_G − mean intensity_G1…Gn)/SD_G1…Gn, where G is any gene on the hybridization array and G1…Gn is the aggregate measure of all genes, and Z_ratios = [(Zscore_G1ave)_Exp − (Zscore_G1ave)_Con]/SD of Z score differences_G1…Gn, where G1ave = average Z score for any particular gene being tested under multiple experimental conditions [experimental (Exp) vs. control (Con)].

Lists of different criteria were generated by sorting the appropriate Z scores in Microsoft Excel. The resulting probe lists were categorized based on their biological process with the GeneSpring Gene Ontology.

Quantitative RT-PCR.

Total RNA (1 μg) was converted into cDNA with the SuperScript First Strand Synthesis Kit (Invitrogen). A wild-type, untreated, unfractionated liver cDNA sample was used to create a standard curve. The remaining cDNA samples were diluted 1:16 and an amount equal to 1/10th of the final reaction volume was used per PCR reaction. For validation of changes in Stch (aka HSPA13) expression observed with the ABI gene expression assay, the plates were run with the TaqMan reporter on an ABI 7000 sequence detection system following the manufacturer-recommended settings (Applied Biosystems, Foster City, CA). For all other qRT-PCR validation reactions using SYBR Green, the final primer concentration was 100 μM. The plates were run on either an MJ Research Opticon 2 or an ABI 7000 thermal cycler with the following settings: 95°C for 15 min, 50 cycles of 94°C for 15 s, 55°C for 20 s, 72°C for 15 s. Primer sequences are listed in Supplemental Table S1.1 Expression level was calculated with the absolute quantitation method (8). The amount of the cDNA of interest was normalized to the average expression amount of β-actin, GAPDH, and Tbp (aka GTF2D1/Gtf2d/SCA17/TFIID) cDNA in the same fraction. The amount of mRNA encoding individual proteins present in fractions from sucrose density gradients was expressed relative to the total amount of that mRNA present in the combined non-, sub-, and polysome fractions.

Statistical analysis.

Statistical analyses were done with the InStat software program (GraphPad Software, La Jolla, CA) as detailed in Figs. 1–9.

RESULTS

Validation of experimental model systems.

GCN2 is activated in response to individual deprivation of any of the essential amino acids (39). However, in liver, certain amino acids, e.g., the branched-chain amino acids and tryptophan (2), also modulate signaling through the mammalian target of rapamycin (mTOR) complex 1 (mTORC1) pathway. Therefore, in pilot studies, mouse livers were perfused with medium lacking individual amino acids, and the effect on phosphorylation of eIF2α, an index of GCN2 activation, and 4E-BP1 and S6K1, indexes of mTORC1 activation, was assessed. In contrast to Leu or Trp, deprivation of Met resulted in a significant increase in eIF2α phosphorylation, with no effect on mTORC1 signaling (data not shown). Consequently, Met deprivation was used as a means of specifically activating GCN2 in subsequent experiments.

To functionally confirm the absence of amino acid-regulated eIF2α kinase activity in livers of Gcn2−/− mice, livers from Gcn2+/+ or Gcn2−/− mice were perfused with medium containing or lacking Met and changes in eIF2α phosphorylation were assessed by Western blot analysis. As shown in Fig. 1A, Met deficiency caused eIF2α phosphorylation to increase ∼2.2-fold in livers from Gcn2+/+ mice. In contrast, Met deficiency had no effect on eIF2α phosphorylation in livers from Gcn2−/− mice. Moreover, eIF2B activity was reduced in livers from Gcn2+/+, but not Gcn2−/−, mice in response to Met deprivation (Fig. 1C). Inhibition of translation initiation relative to elongation/termination leads to disaggregation of polysomes, with a resulting accumulation of 40S and 60S ribosomal subunits as individual subunits or as 80S monosomes (referred to here as the subpolysomal fraction). The inhibition of eIF2B activity observed in the present study would therefore be expected to result in polysome disaggregation. Indeed, as shown in Fig. 1, E and G, in livers from Gcn2+/+ mice, Met deficiency caused a decrease in the number of ribosomes present in the polysome fraction and a corresponding increase in the number of 80S monomers present in the subpolysomal fraction.

Fig. 1.

Regulation of eukaryotic initiation factor (eIF)2α, eIF2B, and polysome aggregation in perfused general control nonderepressible 2 (GCN2)- and double-stranded RNA-activated protein kinase (PKR)-like endoplasmic reticulum-resident kinase (PERK)-perfused mouse livers after methionine (Met) deprivation or 2,5-di-tert-butylhydroquinone (tBuHQ) treatment, respectively. A, C, E, G: Gcn2+/+ or −/− mouse livers were perfused for 35 min with perfusate containing complete amino acids (open bars, A and C) or deficient in Met (gray bars, A and C). B, D, F, H: Perk+/+ or Alb/Cre Perk−/− mouse livers perfused for 35 min with perfusate containing DMSO (open bars, B and D) or 40 μM tBuHQ (gray bars, B and D). A and B: insets show representative Western blots for Ser51 phosphorylated (2αP, top) and total (2αT, bottom) eIF2α. Bars represent quantitation of Western blot band density, calculated as ratio of phosphorylated over total eIF2α. Values are means ± SE. C and D: guanine nucleotide exchange activity of eIF2B. n = 9 or 10 livers/group. Statistical significance was determined by unpaired t-test: *P < 0.001, **P < 0.025. E–H: polysome profiles of GCN2 (E, G)- and PERK (F, H)-perfused mouse livers. Supernatant from liver homogenate was resolved on a 20–47% sucrose density gradient. Absorbance at 254 nm was continuously measured and recorded to generate profiles. Assignment of the 40S, 60/80S, subpolysome, and polysome peaks was as demonstrated in E. E and F: representative gradient profile for conditions of wild-type control condition. G and H: representative gradient profile for wild-type treated condition. Profiles are representative of n = 9 or 10 livers/group.

Mobilization of calcium from the ER promotes misfolding of proteins within the ER lumen, leading to activation of PERK and subsequently to phosphorylation of eIF2α. The absence of ER stress-induced eIF2α kinase activity in livers from Perk−/− mice was confirmed by perfusing livers from Perk+/+ and Perk−/− mice with tBuHQ to promote ER calcium release (51) and then measuring eIF2α phosphorylation and eIF2B activity. As shown in Fig. 1, B and D, perfusion of livers from Perk+/+, but not Perk−/−, mice with tBuHQ resulted in both increased eIF2α phosphorylation and decreased eIF2B activity, respectively. Moreover, polysome disaggregation was induced by perfusion of livers from Perk+/+ mice with tBuHQ (Fig. 1, F and H). Interestingly, both eIF2B activity and polysome aggregation were affected to a larger extent in response to perfusion with tBuHQ compared with Met deprivation.

Global alterations in mRNA expression in livers lacking GCN2 or PERK.

Initially, the expression of individual mRNAs in livers of Gcn2+/+ mice perfused with all amino acids, or without Met, was compared with those expressed in livers of Gcn2−/− mice perfused under the same condition. The time point chosen for these studies, i.e., 35 min, was selected based on previous studies (45a) showing that in cells in culture mRNAs shift from the RNP fraction to polysomes within 30 min. This duration was likely too short to allow for detection of some transcriptional changes. However, even at this relatively early time point, the expression of 574 (all amino acids) and 628 (without Met) mRNAs was greater in Gcn2−/− compared with Gcn2+/+ mice (Supplemental Table S2, tabs 1 and 2, respectively), and 141 mRNAs were common to both lists (Fig. 2 and Supplemental Table S2, tab 3). In parallel, the expression of 616 (all amino acids) and 675 (without Met) mRNAs was lower in livers of Gcn2−/− compared with Gcn2+/+ mice (Supplemental Table S2, tabs 4 and 5, respectively), and 163 mRNAs were common to both lists (Fig. 2 and Supplemental Table S2, tab 6). Those mRNAs exhibiting altered expression in livers lacking GCN2, and whose expression was changed in response to perfusion both with and without Met, were functionally grouped with GeneSpring. As shown in Fig. 3A, mRNAs exhibiting increased expression in the absence of GCN2 grouped into categories such as transport, with some mRNAs exhibiting Z score of 10 or greater (Supplemental Table S2, tab 3). In contrast, those mRNAs whose expression was reduced in Gcn2−/− compared with control livers include those in categories such as regulation of physiological processes, immune response, and apoptosis (Fig. 3A). Of particular interest is the finding that the expression of mRNAs encoding two proteins involved in amino acid synthesis, asparagine synthetase (Asns) and dopachrome tautomerase (Dct/TRP2/Tyrp-2), was dramatically less in Gcn2−/− compared with wild-type liver (Supplemental Table S2, tab 6). To verify the change in Asns mRNA expression identified by microarray analysis, Asns mRNA abundance was quantitated by qRT-PCR. As shown in Fig. 4, Asns mRNA expression was significantly lower in livers from Gcn2−/− compared with Gcn2+/+ mice (P < 0.001) as well as in Perk−/− compared with Perk+/+ mice (P < 0.05). However, the magnitude of the PERK-dependent change was smaller compared with the GCN2-dependent change. Similarly, the GCN2-dependent change in Cyp2c55 mRNA expression (P < 0.01) was also larger than the PERK-dependent change (not significant). Recent studies suggest that Asns mRNA expression is regulated in part by the transcription factor ATF5 (1). Therefore, the expression of ATF5 mRNA was measured by qRT-PCR. In agreement with the observed changes in Asns expression, the expression of ATF5 mRNA was significantly lower in livers from Gcn2−/− compared with control mice (Fig. 4; P < 0.05), suggesting that GCN2 may act through ATF5 to modulate Asns expression.

Fig. 2.

Number of mRNAs whose expression changed in eIF2α kinase−/− vs. +/+ perfused livers. GCN2 or PERK mouse livers were perfused for 35 min with perfusate ± Met or ± tBuHQ, respectively. Total RNA was extracted from the unfractionated supernatant and used in hybridization arrays experiments. A: analyses of the effect of eIF2α kinase on mRNA abundance were done by sorting of the Z score of genes in the arrays. mRNAs with Z score >1.5 when comparing expression in −/− vs. +/+ samples were considered to have increased expression in the absence of the eIF2α kinase. mRNAs with Z score less than −1.5 when comparing expression in −/− vs. +/+ samples were considered to have decreased expression in the absence of the eIF2α kinase. B: Venn diagram of mRNAs whose expression increased in the absence of GCN2 or PERK. C: Venn diagram of mRNAs whose expression decreased in the absence of GCN2 or PERK.

Fig. 3.

Functional categories of genes whose mRNA expression changed in the absence of GCN2 or PERK. Mouse livers were perfused for 35 min with or without treatment as described in Fig. 2. Total RNA was extracted from liver homogenates for use in hybridization array experiments. Analyses of eIF2α kinase-dependent increase or decrease in mRNA expression were done by sorting of genes with Z score greater than 1.5 or less than −1.5. Sublists of sorted genes were used for biological process classification in GeneSpring. Bars represent number of genes within the functional category whose mRNA expression increased (positive values) or decreased (negative values) in the absence of GCN2 (A) or PERK (B). A functional category is included only if it contains ≥3 genes and has a GeneSpring P value of <0.05.

Fig. 4.

Validation of hybridization array results of mRNAs whose expression decreased in the absence of GCN2 (left) or PERK (right). Experimental conditions and analyses used for the identification of genes whose mRNA expression decreased in the absence of GCN2 or PERK were as described in Figs. 2 and 3. Total RNA extracted from unfractionated liver samples was used in quantitative (q)RT-PCR to validate mRNAs with Z score less than −1.5 (Asns, Dct, Calr, Cyp2c55) or those reported in the literature to affect Asns transcription (ATF5) or to be transcriptionally regulated by PERK (Herpud1).

The studies described in the previous paragraph were repeated with liver-specific Perk-knockout mice, referred to here as Perk−/− mice, and Perk+/+ mice. Thus mRNA expression in livers of Perk−/− mice perfused in the absence or presence of tBuHQ was compared with livers of Perk+/+ mice perfused under the same conditions. In livers perfused without tBuHQ, 690 mRNAs exhibited increased abundance and 1,320 decreased abundance in livers of Perk−/− compared with Perk+/+ mice (Supplemental Table S3, tabs 1 and 4, respectively). Moreover, in livers perfused with tBuHQ, 886 mRNAs were present in greater amounts and 802 in lesser amounts, in livers of Perk−/− compared with Perk+/+ mice (Supplemental Table S3, tabs 2 and 5, respectively). In contrast to the GCN2 mice, only 59 mRNAs were commonly upregulated in control and treated Perk−/− compared with wild-type mouse liver (Fig. 2B), and the highest Z score was only 4.95 (Supplemental Table S3, tab 3). The mRNAs functionally grouped into categories including response to stimulus, positive regulation of biological process, and enzyme-linked receptor protein signaling pathway (Fig. 3B). The absence of PERK also led to decreased expression of a smaller common set of mRNAs compared with GCN2 (Fig. 2C; Supplemental Table S3, tab 6), and those that changed grouped into categories such as cell motility and cell adhesion (Fig. 3B). Only two mRNAs were upregulated in livers from both Gcn2−/− and Perk−/− mice compared with wild-type control mice (Fig. 2B), C-type lectin-related f (GenBank AK017207) and an mRNA encoding a hypothetical protein with similarity to zinc finger protein 691 (GenBank AV344364). No mRNAs were commonly downregulated in the absence of GCN2 and PERK (Fig. 2C).

GCN2 and PERK activation leads to recruitment of different functional groups of mRNAs for translation.

The effect of GCN2 or PERK activation on the distribution of individual mRNAs present in polysomes was assessed by perfusing livers in the presence or absence of Met or tBuHQ, respectively, followed by sucrose density gradient centrifugation to separate mRNAs present in polysomes from those not present (i.e., those being translated from those that are untranslated). Three criteria had to be met in order for an mRNA to be considered as translationally upregulated in a GCN2-dependent manner. First, the amount of a particular mRNA had to be the same or lower in the total fraction of wild-type mouse liver perfused in the absence compared with the presence of Met (i.e., no increase in mRNA transcription). Second, the amount of a particular mRNA had to be greater in the polysome fraction from livers of wild-type mice perfused in the absence compared with the presence of Met (i.e., increased translation by shifting into polysomes after stress). Third, the amount of that mRNA had to be the same or lower in the polysome fraction from livers of Gcn2−/− mice perfused in the absence compared with the presence of Met (i.e., increased translation following stress only in the presence of GCN2). Thus, if an mRNA exhibited increased abundance in polysomes in response to Met deprivation in livers of both Gcn2+/+ and Gcn2−/−, it was not considered to be translationally upregulated in a GCN2-dependent manner. A total of 490 mRNAs, or ∼3% of the 15,700 mRNAs identified by GeneSpring, exhibited a GCN2-dependent shift into polysomes in response to Met deprivation (Fig. 5). Grouping of the 490 mRNAs into functional classifications revealed that the proteins encoded by the mRNAs participate in a variety of cellular processes related to metabolism and energy production, with the three processes with the lowest P values being generation of precursor metabolites and energy, electron transport, and lipid metabolism (Table 1 and Supplemental Table S4, tab 1). Of mRNAs that shifted into polysomes in a GCN2-dependent manner, the greatest number encode proteins involved in generation of precursor metabolites and energy (Fig. 6).

Fig. 5.

Number of mRNAs whose abundance in the polysome fraction changed after treatment in wild-type livers. GCN2 or PERK mouse livers were perfused for 35 min with perfusate ± Met or ± tBuHQ, respectively. Total RNA was extracted from the unfractionated supernatants and polysome fractions for use in hybridization array experiments. A: analyses of the effect of treatment on mRNA abundance in the polysome fraction were done by sorting of the Z scores of genes in the arrays. mRNAs with Z score > 1.5 in polysome fraction of wild-type treated livers were considered to have shifted into polysomes after activation of the eIF2α kinase. mRNAs with Z score less than −1.5 in polysome fraction of wild-type treated livers were considered to have shifted out of polysomes after activation of the eIF2α kinase. B: Venn diagram of mRNAs that shifted into polysomes after activation of GCN2 or PERK. C: Venn diagram of mRNAs that shifted out of polysomes after activation of GCN2 or PERK.

Table 1.

Group classification based on biological function of mRNAs that exhibit GCN2-dependent or PERK-dependent shift into polysomes in response to Met deprivation or tBuHQ treatment, respectively

Biological Process	P Value	No. of mRNAs Affected
GCN2 dependent
Generation of precursor metabolites and energy	1.50E-06	26
Electron transport	9.39E-06	21
Lipid metabolism	1.41E-04	20
Blood coagulation	3.92E-04	6
Carbohydrate metabolism	9.94E-03	11
Death	1.21E-02	13
Proteolysis	1.52E-02	18
Endocytosis	2.32E-02	7
Steroid metabolism	3.94E-02	5
Alcohol metabolism	4.34E-02	5
PERK dependent
Response to biotic stimulus	2.15E-06	26
Electron transport	1.17E-03	12
Response to wound healing	2.81E-03	9
Response to stress	5.71E-03	20
Humoral immune response	8.45E-03	5
Protein folding	1.25E-02	5
Anion transport	2.08E-02	6
Translation	2.44E-02	7
Programmed cell death	4.37E-02	12

GCN2, general control nonderepressible 2; PERK, double-stranded RNA-activated protein kinase (PKR)-like endoplasmic reticulum-resident kinase; tBuHQ, 2,5-di-tert-butylhydroquinone. The mRNAs in each category are listed in Supplemental Table S9.

Fig. 6.

Functional categories of genes whose mRNAs shift into or out of polysomes after Met deprivation, in the presence of GCN2. Gcn2+/+ or −/− mouse livers were perfused for 35 min ± Met, and supernatant was fractioned on a 20–47% sucrose density gradient. Total RNA was extracted from the unfractionated supernatant and the polysomal gradient fraction for use in hybridization array experiments. Analyses of GCN2-dependent shift of mRNAs into polysomes were done by sorting of Z scores of genes in arrays containing unfractionated supernatant and polysomal gradient fraction. Sublists of sorted genes were used for biological process classification in GeneSpring. Bars represent number of genes within the functional category whose mRNA shifted into (positive values) or out of (negative values) polysome fraction. A functional category is included only if it contains ≥3 genes and has a GeneSpring P value of <0.05.

The criteria for designating an mRNA as translationally downregulated in a GCN2-dependent manner were as follows. First, the amount of a particular mRNA had to be the same or more in the total fraction from livers of wild-type mice perfused in the absence compared with the presence of Met (i.e., no decrease in mRNA transcription). Second, the amount of a particular mRNA had to be less in the polysome fraction from livers of Gcn2+/+ mice perfused in the absence compared with the presence of Met (i.e., decreased translation by shifting out of polysomes after stress). Third, the amount of that mRNA had to be the same or greater in the polysome fraction from livers of Gcn2−/− mice perfused in the absence compared with the presence of Met (i.e., decreased translation is dependent on Gcn2). Thus, if an mRNA exhibited decreased polysome abundance in response to Met deprivation in livers of both Gcn2+/+ and Gcn2−/−, it was not considered to be translationally downregulated in a GCN2-dependent manner. As shown in Fig. 5C, 527 mRNAs were found to fit this pattern (also see Supplemental Table S4, tab 2). When grouped by biological function, >100 mRNAs encoding proteins involved in cellular metabolism and >60 mRNAs encoding proteins involved in either cellular protein or macromolecule metabolism shifted out of the polysome fraction.

The criteria used for designating mRNAs as being translationally up- or downregulated in a PERK-dependent manner were similar to those described above for GCN2, except that the experimental conditions utilized tBuHQ treatment rather than Met deprivation. Activation of PERK resulted in an increase in the abundance of 447 mRNAs in the polysome fraction, or ∼2.4% of the 18,700 mRNAs identified by GeneSpring (Fig. 5). Grouping of the mRNAs into functional categories (Fig. 7, Table 1, and Supplemental Table S5) revealed that a number of the biological processes associated with PERK activation were distinct from those affected by GCN2 activation. PERK activation also resulted in 967 mRNAs shifting out of polysomes (Fig. 5C). Of these mRNAs, >60 encode proteins involved in development and >40 are involved in either biosynthesis or cellular biosynthesis (Fig. 7). Notably, of those mRNAs that shifted into the polysome fraction in response to activation of GCN2 or PERK, 11 were common to both groups (Fig. 5B). In addition, 44 mRNAs shifted out of the polysome fraction in response to activation of either kinase (Fig. 5C).

Fig. 7.

Functional categories of genes whose mRNAs shift in or out of polysomes after tBuHQ treatment in the presence of PERK. Perk+/+ or −/− mouse livers were perfused for 35 min ± tBuHQ, and supernatant was fractioned on a 20–47% sucrose density gradient. Total RNA was extracted from the unfractionated supernatant and the polysomal gradient fraction for use in hybridization array experiments. Analyses of PERK-dependent shift of mRNAs into polysomes were done by sorting of Z scores of genes in arrays containing unfractionated supernatant and polysomal gradient fraction. Sublists of sorted genes were used for biological process classification in GeneSpring. Bars represent number of genes within the functional category whose mRNA shifted into (positive values) or out of (negative values) polysome fraction. A functional category is included only if it contains ≥3 genes and has a GeneSpring P value of <0.05.

Validation of microarray data.

qRT-PCR was used to validate the changes in mRNA abundance in the polysome fraction for selected mRNAs. The mRNAs selected for validation include mRNAs encoding three putative “housekeeping” proteins (β-actin, GAPDH, and Tbp), six mRNAs whose abundance in polysomes was altered in either a GCN2 [IGFBP-2 (aka BP2/IBP2), Slc3a2 (aka 4F2/4F2HC/CD98/Ly-10/Ly-m10/Mdu1/Mgp-2Hhc), and Trfr2 (aka Tfr2)]- or a PERK [Ldlr (aka Hlb301), Slc2a2 (aka Glut-2), and Stch]-dependent manner, and mRNAs previously reported to be translationally regulated [ATF3, ATF4, Bag-3 (aka Bis) (16, 60), Ddit3 (aka CHOP10/chop/gadd153) (5), and Hspa5 (aka Bip/Hsce70/Sez7/Grp78)]. The Z scores obtained by microarray analysis for the selected mRNAs are shown in Supplemental Table S6. In addition to analysis of mRNA abundance in the polysome fraction, the amount of the individual mRNAs present in the nonpolysomal and subpolysomal fractions was also quantitated. As shown in Fig. 8, A and B, in livers from Gcn2+/+ mice Met deprivation had no significant effect on the abundance of any of the housekeeping mRNAs present in the polysome fraction, except that the amount of Tbp mRNA was decreased in livers of wild-type mice. With the exception of Ldlr, all of the mRNAs identified by microarray analysis as being translationally regulated in response to activation of GCN2 and/or PERK exhibited a shift in distribution from the non- and/or subpolysomal fractions into polysomes (Fig. 8C). In contrast, in livers from Gcn2−/− mice, none of these mRNAs shifted into polysomes (Fig. 8D). Of the mRNAs shown in previous studies to be subject to translational regulation, only Hspa5 shifted into polysomes in response to Met deprivation (Fig. 8E), and the effect was dependent on GCN2 (Fig. 8F).

Fig. 8.

qRT-PCR validation of polysome distribution of selected mRNAs in Gcn2 wild-type vs. knockout perfused livers. Mouse livers were perfused for 35 min ± Met. Total RNA was extracted from the unfractionated supernatant and the nonpolysomal, subpolysomal, and polysomal gradient fractions for use in hybridization array and qRT-PCR experiments. Expression of mRNA was determined by qRT-PCR and normalized to the average expression of the housekeeping genes β-actin, Gapdh, and Tbp (see materials and methods for details on calculations). Values are means ± SD; n = 6 or 7. A and B: putative housekeeping genes. C and D: genes identified in this experiment to have shifted into polysomes after activation of GCN2. E and F: genes reported in the literature whose mRNA shifted into polysomes after activation of GCN2. Statistical analysis of the data is presented in Supplemental Table S7.

PERK activation had little effect on the polysomal distribution of the mRNAs encoding the three housekeeping proteins other than GAPDH exhibiting a PERK-dependent shift into polysomes (Fig. 9, A and B). The mRNAs encoding Slc3a2, Trfr2, Ldlr, and Slc2a2 also shifted from the non- and/or subpolysomal fractions into polysomes, and Stch exhibited a similar trend, although the change was not large enough to reach statistical significance (Fig. 9C). Interestingly, all of these mRNAs also shifted into polysomes in livers from Perk−/− mice perfused with tBuHQ (Fig. 9D). Of the five mRNAs previously shown to be subject to translational regulation, all demonstrated a dramatic shift from the non- and/or subpolysome fraction into polysomes in wild-type liver (Fig. 9E). Although there was a trend for these mRNAs to shift into polysomes in tBuHQ-treated livers from Perk−/− mice, only Hspa5 reached statistical significance, and even for that mRNA the magnitude of the shift was blunted in Perk−/− compared with Perk+/+ livers (Fig. 9F). Although the basis for the observed change is unknown, it must occur through a mechanism distinct from phosphorylation of eIF2α and inhibition of eIF2B, because neither was changed in PERK−/− livers treated with tBuHQ.

Fig. 9.

qRT-PCR validation of polysome distribution of selected mRNAs in Perk wild-type vs. knockout perfused livers. Mouse livers were perfused for 35 min ± tBuHQ. Total RNA was extracted from the unfractionated supernatant and the nonpolysomal, subpolysomal, and polysomal gradient fractions for use in hybridization array and qRT-PCR experiments. Expression of mRNA was determined by qRT-PCR and normalized to the average expression of the housekeeping genes β-actin, Gapdh, and Tbp (see materials and methods for details on calculations). Values are means ± SD; n = 6 or 7. A and B: putative housekeeping genes. C and D: genes identified in this experiment to have shifted into polysomes after activation of PERK. E and F: genes reported in the literature whose mRNA shifted into polysomes after activation of PERK. Statistical analysis of the data is presented in Supplemental Table S8.

DISCUSSION

In the present study, perfused mouse liver preparations were used to delineate the similarities and differences between activation of GCN2 compared with PERK in the regulation of mRNA translation. Both kinases phosphorylate the same residue, Ser51, on the α-subunit of eIF2 (28, 65). However, the two kinases do not appear to contribute equally to basal eIF2α phosphorylation. Thus, in livers perfused with a complete amino acid mixture in the absence of tBuHQ, eIF2α phosphorylation was similar in wild-type and Gcn2−/− livers, whereas phosphorylation of the protein was lower in livers from Perk−/− compared with Perk+/+ mice. This result suggests that, under the conditions used here, PERK contributes more to basal eIF2α phosphorylation compared with GCN2. Interestingly, eIF2B activity was not significantly different in livers from Perk−/− compared with Perk+/+ mice perfused without tBuHQ, despite the difference in eIF2α phosphorylation. The basis for this finding may lie in the absolute amount of eIF2α present in the phosphorylated form in the control condition. Although not determined for mouse liver, we have previously shown (38) that in rat livers treated with perfusate of the same composition used the present study, <10% of eIF2α is present in the phosphorylated form. Consequently, the difference in eIF2α phosphorylation between livers of Perk−/− and Perk+/+ mice may have caused an increase in eIF2B activity that was too small to detect with the GDP exchange assay. A related observation is that, although eIF2α phosphorylation was increased to a similar extent upon activation of either PERK or GCN2, eIF2B activity was decreased to a greater extent in response to activation of PERK compared with GCN2. This finding suggests that activated PERK regulates additional proteins involved in controlling eIF2B activity. Of the kinases known to phosphorylate eIF2B, GSK-3 is the only one whose action is inhibitory (61). Recent studies (4, 34) demonstrate that activation of either PERK or PKR promotes GSK-3β phosphorylation on Ser9, an event known to inhibit its kinase activity. However, if activated PERK inhibited GSK-3 activity in perfused liver, it would be expected to increase rather than inhibit eIF2B activity. Therefore, it is unlikely that PERK-specific inhibition of GSK-3 explains the differential decrease in eIF2B activity.

Most studies examining alterations in gene expression in response to cell stress have focused on transcriptional rather than translational mechanisms. In this regard, several studies have used hybridization array analysis to document changes in mRNA expression following activation of PERK or GCN2. For example, induction of ER stress in the rat mast cell line RBL-2H3 following a 3-h treatment with tBuHQ increased the expression of mRNAs encoding proteins involved in the immune response and cell cycle regulation (including Gadd153) (54). In another study, induction of ER stress by tunicamycin led to increased expression of Asns, Ddit3, Slc3a2, and Stch mRNA expression in primary fibroblasts (45). Of these, only Asns was found to be transcriptionally regulated in the present study in Gcn2−/− compared with Gcn2+/+ mice. However, all of the mRNAs encoding these proteins, except Asns, were translationally regulated in response to activation of GCN2 and/or PERK.

Changes in mRNA expression in response to activation of either GCN2 or PERK were also observed in the present study. For example, Met deprivation led to a significantly greater increase in expression of the Cyp2c39, Clrf, Isg20, Abhd1, and Hgf mRNAs in the liver of Gcn2−/− compared with Gcn2+/+ mice. In addition to GCN2, the other major intracellular nutrient sensor is mTORC1. However, the observed changes are not likely due to alterations in signaling through mTORC1 because, as noted above, no change in phosphorylation of the mTORC1 substrates 4E-BP1 and S6K1 was observed in livers deprived of methionine. Whether or not another, as yet unidentified, amino acid sensor might exist in mammalian cells (e.g., a protein similar to the yeast Ssy1 protein) is unknown. Less surprising, but still interesting, was the relatively large number of genes that were induced by tBuHQ in the liver of Perk−/− compared with Perk+/+ mice. For example, perfusion with tBuHQ led to a greater increase in expression of the BECN1, EPPK1, CTSB, HIPK2, IGF-6, and ABCC9 mRNAs in the liver of Perk−/− compared with Perk+/+ mice. Although we did not investigate the possibility, the results could be due to induction of other ER stress signaling pathways such as ATF6 or XBP-1.

Translational derepression of ATF4 has been repeatedly documented under different stress conditions in cell models. For example, ATF4 protein expression has been shown to increase after leucine deprivation (31), treatment with thapsigargin (27), or hypoxia (6). In part, the upregulation of ATF4 protein expression is due to a shift in the distribution of the mRNA from small to larger polysomes. A similar phenomenon has been observed for GCN4 in S. cerevisiae in response to nutrient deprivation and treatment with butanol. The proposed model for the derepression event in yeast is that reduction of the met-tRNA_i·eIF2·GTP ternary complex allows the ribosomes to bypass uORF and reinitiate at the GCN4 AUG. A similar model has been proposed for the ATF4 mRNA in mammalian cells by Harding et al. (27) and Wek and colleagues (31). In the present study, we observed a shift in the ATF4 mRNA from the subpolysomal to the polysomal fraction in the liver of PERK+/+, but not PERK−/−, mice treated with tBuHQ. At most, only one ribosome can be associated with a given mRNA in the subpolysomal fraction, suggesting that a fraction of the ATF4 mRNA is poorly translated under control conditions. Moreover, upon activation of the UPR, the poorly translated fraction shifts into polysomes, a result consistent with previous studies showing increased expression of the protein. However, even though the magnitude of the increase in eIF2α phosphorylation in the liver of GCN2+/+ mice deprived of Met was similar to that in Perk+/+ mice treated with tBuHQ, there was no significant change in the polysomal distribution of the ATF4 mRNA in response to Met deprivation. The basis for this observation is unknown. However, the finding may relate to the greater reduction in eIF2B activity in response to tBuHQ treatment compared with Met deprivation. The difference in eIF2B activity was mirrored by alterations in polysome aggregation, such that polysome aggregation was reduced to a greater extent in livers perfused with tBuHQ compared with livers deprived of Met. Because the proposed mechanism for upregulation of GCN4 and ATF4 mRNA translation depends on the activity of eIF2B, it is possible that the magnitude of the change induced by Met deprivation for 35 min is not sufficient to cause a significant shift in the ATF4 mRNA from the subpolysomal to the polysomal fraction. It is possible that longer periods of deprivation might promote ATF4 mRNA redistribution. However, such studies cannot be performed with the present model, because global rates of protein synthesis recover over longer periods of amino acid deprivation (e.g., 1–2 h) because of an increase in hepatic proteolysis (23).

Translational upregulation of ATF4 protein has been commonly reported to precede transcriptional upregulation of Ddit3 mRNA (47); however, there is evidence to suggest that the latter can also be translationally regulated, although little is yet known about the mechanism. As shown by the results of the present study, translation of the Ddit3 mRNA is also derepressed in response to activation of PERK, but not GCN2. Derepressed translation of the Ddit3 mRNA subsequently leads to decreased expression of Asns (69). Results from the present study show that the basal mRNA expression of two proteins involved in amino acid metabolism, Asns and Dct, depends on the presence of GCN2. Asns, whose expression was previously shown to depend on the presence of GCN2 (3), encodes asparagine synthetase, an enzyme that catalyzes the hydrolysis of the amido-N group of glutamine, which acts as a donor to convert aspartate to asparagine. Dct, shown to be regulated in response to GCN2 activation for the first time in this study, is involved in tyrosine metabolism and melanin synthesis. Recent reports have placed ATF5 upstream of Asns, proposing that in opposition to Ddit3 (aka CHOP), ATF5 binds to the C/EBP-ATF regulatory sequence in the Asns promoter, thus upregulating Asns gene transcription (1). The decrease in ATF5 and Asns mRNA expression in Gcn2−/− mice observed in this study provides further support for this idea. Although only a small number of transcripts, and none of those identified through the array analysis as being transcriptionally dependent on PERK, have been analyzed by qRT-PCR, their Z score value typically showed smaller changes than those identified in the GCN2 study.

Surprisingly, Hspa5 (GRP78/BiP), an mRNA whose expression (46) and translational state are most closely linked to the UPR, was found in this study also to be regulated after amino acid deprivation. Hspa5 mRNA was shown by Stephens and colleagues (67) to shift into polysomes after UPR, in both the cytosolic and ER-bound mRNA populations. In the present study, the Hspa5 mRNA also shifted into polysomes after activation of either GCN2 or PERK. Interestingly, the mRNA encoding Hspa5 contains an IRES in its 5′-UTR. Even more surprising is the finding that the Stch mRNA shifted into polysomes after Met deprivation but not tBuHQ treatment. Stch mRNA, whose expression is induced by ER stress associated with treatment with the calcium ionophore A-23187, codes for a constitutively expressed, microsome-associated chaperone protein that has a cellular distribution similar to BiP (57). Together, these data would suggest that translation of the Stch mRNA would be expected to be regulated by PERK, but not GCN2.

Regulation of cellular transporter expression appears to be a common response to stress at both the transcriptional (53) and translational levels. A well-studied example of the latter is the amino acid transporter cationic amino acid transporter 1 (Cat-1) mRNA, which contains both an uORF and an IRES (78). Its translation after glucose deprivation is proposed to be PERK dependent and to occur through the cooperative action of the two 5′-UTR structures (22, 78). In the present study, the finding that mRNAs encoding three transporters (Slc2a2, Slc3a2, and Trfr2) shifted into polysomes in response to activation of both PERK and GCN2 suggests a common pool of cellular mRNAs whose translation is derepressed under any type of stress that increases eIF2α phosphorylation. Under either amino acid deprivation or ER stress, increased expression of the protein encoded by these mRNAs would allow cells to increase nutrient uptake (glucose, amino acids, and iron, respectively) that provides building blocks for cellular molecules. Of the three transporters, most is known about the regulation of the Trfr2 mRNA via its iron-regulatory element.

The observation that the mRNAs encoding Igfbp-2, Ldlr, and Bag3 shift into polysomes in response only to activation of either PERK or GCN2 provides an example of unique regulation by each eIF2α kinase. The IGFBP protein family has been strongly implicated in metabolic processes, and transcription of the genes encoding the proteins is modulated after various stresses (3, 40). For example, Igfbp-1 mRNA expression is increased in Leu-deprived HepG2 human hepatoma cells after small interfering RNA (siRNA) knockdown of ATF4 as well as in hepatocytes from Gcn2−/− mice, suggesting that Igfbp-1 expression is regulated by amino acids independently of the GCN2/ATF4 pathway (3). Igfbp-2 expression has been reported to specifically increase after protein or caloric restriction and is involved in regulating liver growth (50). Although Ldlr mRNA expression has been linked to UPR caused by treatment with protease inhibitors (83), and recent work has shown that retention of mutated forms of Ldlr in the ER can lead to UPR and PERK activation (66), little is known about the role of PERK in regulating the translation of mRNAs involved in lipid metabolism. Similarly, besides studies showing that the Bag1 mRNA is translationally regulated through an IRES in the 5′-UTR of its mRNA (11), evidence for regulation of other family members is sparse.

An important question is how activation of one eIF2α kinase might result in differential translation of specific mRNA subsets. Although the answer to this question is unknown, it is tempting to speculate that, in part, the effect may be a result of differences in the subcellular localization of the kinases, leading to a gradient of eIF2α phosphorylation within the cell. Evidence in support of this idea is provided by a recent study (42) showing that activation of PERK in response to ER stress results in the accumulation of eIF2 phosphorylated on Ser51 of the α-subunit in the cytosol adjacent to the ER. The concentration of phosphorylated eIF2 near the ER could preferentially repress the translation of mRNAs destined for that subcellular compartment.

Sequence analyses and experiments initiated in the 1980s by Kozak (43, 44) have popularized the concept of upstream, 5′ noncoding region sequences positively influencing translation initiation. Studies done in yeast and mammalian model systems since then have provided evidence for this theory, and two working models involving leaky scanning or reinitiation in mRNAs containing uORF or cap-independent translation initiation in those containing IRES have evolved (9, 24–26, 75, 78, 82). Despite the continuing prolific amount of new data and new mRNA candidates whose translation is being regulated in this manner, the effort continues to be piecemeal. A comprehensive survey of the mammalian genome for genes coding for mRNAs that contain uORF/s or IRES has only been done in silico (in the case of uORF) or not at all (in the case of IRES). Computer analyses of coding sequences have shown that 10–50% of the mammalian mRNAs contain uAUGs (12, 30, 49, 62, 79). Of these, only one-quarter have complete ORF (12, 30, 49, 62, 79). These uORF along with out-of-frame uAUGs are more conserved across species than in-frame uAUGs (12, 30, 49, 62, 79). The uORF code for proteins ranging from 20 to 100 amino acids in length, with some being synthesized in vivo (58). In addition to reconfirming Kozak's rules of conserved bases surrounding the coding sequence AUG and translation initiation at the first AUG, they also presented a population of potential mRNA candidates with uORF that can be regulated by leaky scanning, reinitiation, or ribosome shunting during translation. The percentage of mRNAs identified in the present microarray study that were potentially regulated through these alternative translation initiation pathways is on the low end of the in silico data. Two caveats must be considered: 1) the applied experimental techniques may not favor detection of many mRNAs whose expression is low, and 2) this experimental percentage is a sum of mRNA groups containing uORF or IRES, whereas that from the in silico study may include mainly just those mRNAs containing uORF. A future application for this microarray analysis is to identify biological groups of mRNAs whose translation initiation is similarly regulated after a stressful stimulus.

In summary, the results obtained in the present study show that, despite sharing a common downstream target, GCN2 and PERK regulate the transcription and translation of distinct, as well as overlapping, mRNA subsets. In addition, the percentage of mRNAs translationally regulated after activation of either GCN2 or PERK is in the same range as that identified through sequence analyses to contain uORF. The hybridization array data presented here provide important selective targets for further comparative study concerning GCN2, PERK, and their activating stressors.

GRANTS

This study was supported by National Institute of Diabetes and Digestive and Kidney Diseases Grant DK-13499.