Crystal Structures of GCN2 Protein Kinase C-terminal Domains Suggest Regulatory Differences in Yeast and Mammals*

Background: GCN2 is a multidomain protein kinase that phosphorylates eIF2 to regulate translation during nutrient deprivation. Results: The CTDs of GCN2 from yeast and mammals have interdigitated dimeric structures with a conserved core, bind RNA similarly, but differ in ribosome association. Conclusion: There are key regulatory differences between yeast and mammalian CTDs. Significance: This study determines key structural features regulating GCN2 and translational control. In response to amino acid starvation, GCN2 phosphorylation of eIF2 leads to repression of general translation and initiation of gene reprogramming that facilitates adaptation to nutrient stress. GCN2 is a multidomain protein with key regulatory domains that directly monitor uncharged tRNAs which accumulate during nutrient limitation, leading to activation of this eIF2 kinase and translational control. A critical feature of regulation of this stress response kinase is its C-terminal domain (CTD). Here, we present high resolution crystal structures of murine and yeast CTDs, which guide a functional analysis of the mammalian GCN2. Despite low sequence identity, both yeast and mammalian CTDs share a core subunit structure and an unusual interdigitated dimeric form, albeit with significant differences. Disruption of the dimeric form of murine CTD led to loss of translational control by GCN2, suggesting that dimerization is critical for function as is true for yeast GCN2. However, although both CTDs bind single- and double-stranded RNA, murine GCN2 does not appear to stably associate with the ribosome, whereas yeast GCN2 does. This finding suggests that there are key regulatory differences between yeast and mammalian CTDs, which is consistent with structural differences.

In response to amino acid starvation, GCN2 phosphorylation of eIF2 leads to repression of general translation and initiation of gene reprogramming that facilitates adaptation to nutrient stress. GCN2 is a multidomain protein with key regulatory domains that directly monitor uncharged tRNAs which accumulate during nutrient limitation, leading to activation of this eIF2 kinase and translational control. A critical feature of regulation of this stress response kinase is its C-terminal domain (CTD). Here, we present high resolution crystal structures of murine and yeast CTDs, which guide a functional analysis of the mammalian GCN2. Despite low sequence identity, both yeast and mammalian CTDs share a core subunit structure and an unusual interdigitated dimeric form, albeit with significant differences. Disruption of the dimeric form of murine CTD led to loss of translational control by GCN2, suggesting that dimerization is critical for function as is true for yeast GCN2. However, although both CTDs bind single-and double-stranded RNA, murine GCN2 does not appear to stably associate with the ribosome, whereas yeast GCN2 does. This finding suggests that there are key regulatory differences between yeast and mammalian CTDs, which is consistent with structural differences.
In response to environmental stresses, eukaryotic cells rapidly reduce protein synthesis, which lowers expenditure of energy and resources and facilitates a reprogramming of gene expression designed to restore cell homeostasis. A central mechanism directing this translational control involves phosphorylation of eukaryotic initiation factor 2 (eIF2) 4 (1)(2)(3). This is illustrated during starvation for amino acids, where phosphorylation of eIF2 by the protein kinase GCN2 (EIF2AK4) reduces the exchange of eIF2-GDP to eIF2-GTP that is required for delivery of initiator Met-tRNA i Met to the translation machinery. As a consequence, there is repressed initiation of global protein synthesis, thus lowering the utilization of limiting amino acids. Accompanying this global translational control, eIF2 phosphorylation leads to preferential translation of select mRNAs such as that encoding the transcription factor ATF4 (4,5). In turn, ATF4 activates the transcriptional expression of genes involved in metabolism and nutrient uptake, anti-oxidation, and protein folding and assembly, which collectively can serve to ameliorate stress damage (6).
The mechanisms by which eIF2 kinases recognize stress signals in cells involve regulatory regions that are juxtaposed to the protein kinase domain. Mechanistic features largely derived from genetic and molecular studies in yeast Saccharomyces cerevisiae indicate that GCN2 contains two central regulatory regions, a histidyl-tRNA synthetase-like domain (HisRS) and a C-terminal domain (CTD), which function together to sense nutrient depletion through a mechanism that is not yet fully understood (18 -24). Proposed models include a role for both domains in binding uncharged tRNA, which accumulate during amino acid deprivation, followed by a conformational change that activates the kinase domain of GCN2, facilitating transautophosphorylation and induction of eIF2 phosphorylation (25). The binding of uncharged tRNA by the HisRS is supported by its sequence conservation with the histidyl tRNA synthetase. The mechanism by which the CTD would bind tRNA has yet to be established. Other proposed mechanistic features mediated by the CTD are dimerization and ribosome binding. The CTD of yeast GCN2 was shown to be sufficient to mediate binding of this eIF2 kinase to ribosomes as judged by co-migration with ribosomes separated by sucrose gradient centrifugation (19,21,24). In yeast, GCN2 association with the translational machinery has been proposed to be important for facilitating binding to accumulating uncharged tRNAs in the context of the A site of ribosomes (25). Furthermore, the CTD of GCN2 can bind double-stranded (ds) RNA through a cluster of lysine residues, which had led to the proposal that GCN2 associates with ribosomes through direct interactions with rRNA (21). The sequences of the CTD of GCN2 in vertebrates share little sequence similarity with their yeast counterpart; therefore, it is not currently known whether the functional features attributed to the yeast GCN2 CTD are functionally germane to mechanisms regulating GCN2 phosphorylation of eIF2 in mammals.
In this study we explore the structural and functional properties of the CTD of GCN2 from both yeast and mammals with the goal of answering the following key mechanistic questions. How does the CTD facilitate dimerization of yeast GCN2? Given the sequence divergence of the CTD between yeast and mammals, are there conserved structural features? Is dimerization important for GCN2 function in mammalian cells subject to amino acid deprivation? Addressing these questions will provide insight into the mechanisms controlling the activation of GCN2 eIF2 kinase and the ensuing translational control triggered by deprivation of nutrients.

EXPERIMENTAL PROCEDURES
Preparation of Murine and Yeast CTD Proteins-Several expression vectors encoding N-terminal His 6 tagged or N-terminal His 6 -SUMO-tagged CTD proteins of yeast and murine GCN2 with different N termini were used in this study. N-terminal His 6 constructs include cDNA encoding the murine GCN2 C-terminal domain (mCTD) from residues 1493-1648, inserted between the NheI and BamHI sites of plasmid pET28a (Novagen). Previously reported yeast GCN2 CTD (yCTD) residues 1498 -1659 or 1536 -1659 were also introduced between the NdeI and BamHI sites into plasmid pET15b (Novagen) (21). N-terminal His 6 -SUMO constructs included cDNA encoding mCTD residues 1514 -1648 or yCTD encoding residues 1519 -1659 inserted into pSMT3 vector (gift from Dr. Christopher Lima, Sloan-Kettering Institute) between restriction sites BamHI and XhoI. Crystallographic studies were carried out using mCTD 1514 -1648 and yCTD 1519 -1659 proteins.
All plasmids were transformed into Rosetta (DE3) Escherichia coli (Novagen, Inc.); cultures were grown at 37°C in Luria broth containing 20 g/ml kanamycin (pET28, pSMT3) or 100 g/ml ampicillin (pET15) and 34 g/ml chloramphenicol until the optical density at 600 nm reached 0.6 and then induced overnight at 18°C by adding 1 mM isopropyl-␤-D-thiogalactoside. Selenomethionine-derivatized yeast N-terminal His 6 -SUMO-CTD (1519 -1659) was expressed in M9 medium as described (26). In this method, six amino acids, leucine, isoleucine, lysine, phenylalanine, threonine, and valine, were added to the medium to inhibit methionine synthesis, thereby forcing the bacterial cells to use the supplied selenomethionine.
Both N-terminal His 6 -SUMO fusion and N-terminal His 6 proteins were purified similarly with the exception of the cleavage step as described here. E. coli cell pellets were resuspended in 50 mM sodium phosphate, pH 7.8, 0.3 M NaCl, 10 mM imidazole, and lysed two times by using a French press (SLM-AMINCO, Spectronic Instruments, Rochester, MN) at 1000 p.s.i. and then subjected to ultracentrifugation at 35,000 rpm for 30 min. The supernatant was then incubated with nickel-nitrilotriacetic acid (Qiagen) beads for 1 h at 4°C. Protein bound beads were applied to a column and washed with 50 mM sodium phosphate buffer, pH 7.8, 0.3 M NaCl, and 20 mM imidazole until the absorbance reading at 280 nm reached background level. The His 6 -SUMO affinity tag was removed by on-column cleavage with the addition of Ulp1 protease, which was prepared using an expression plasmid kindly provided by Dr. Chris Lima (Sloan-Kettering Institute). Ulp1 protease was added at an estimated mass ratio of 1:1000 (Ulp1: His 6 -SUMO-protein) and incubated overnight at 4°C. The His 6 -tagged samples were subjected to thrombin cleavage (2 units/mg of protein) at 4°C overnight to remove the His 6 tag. The mCTD protein was then loaded on a heparin column buffered in 50 mM Tris-HCl (pH 8.5) and eluted with a linear 50 to 1000 mM NaCl gradient. Finally, mCTD-containing fractions were concentrated and then subjected to Superdex 75 gel filtration column chromatography buffered in 50 mM Tris (pH 8.0), 300 mM NaCl, and 1 mM DTT. After purification, the mCTD protein used for crystallization migrated as a single band as judged by SDS-PAGE and Coomassie staining. The native and selenomethionine yCTD samples were purified as described above for the mCTD. Purified CTD samples were concentrated, filtered to remove any particulate matter, and then stored at Ϫ80°C.
Crystallization and Data Collection-mCTD (residues 1514-1648) or yCTD (residues 1519-1659) was diluted to a final concentration of 24 and 13 mg/ml, respectively, in a solution of 50 mM Tris-Cl (pH 8.0), 150 mM NaCl, and 1 mM DTT for vapor diffusion hanging drop crystallization trials. Diffraction quality crystals of mCTD were obtained from the precipitant solution including 0.1 M Bis-Tris (pH 6.5) and 2.0 M (NH 4 ) 2 SO 4 in drops containing 1 l of protein and 1 l of precipitant solution. Crystals of yCTD were obtained from 0.1 M Bis-Tris (pH 5.5) and 21-28% (w/v) polyethylene glycol 2000 (PEG2K).
As mCTD lacks Met residues, efforts to solve the phase problem focused on obtaining a heavy atom derivative. Ethyl mercuric phosphate (EMP) was added directly into crystallization drops to a final concentration of 1 mM overnight. Crystals of mCTD were removed from the mother liquor with a fiber loop, soaked in the Cryo solution (0.1 M Bis-Tris (pH 6.5), 2.1 M (NH 4 ) 2 SO 4 , 10% ethylene glycol, and 1 mM EMP) for about 1 min, and then immediately frozen in liquid nitrogen. Selenomethionine derivative crystals of yCTD were obtained from identical conditions as used for the native crystals. A heavy atom derivative of yCTD was prepared by soaking crystals overnight in a solution of 0.1 M Bis-Tris (pH 5.5), 25% PEG2K, and 1 mM EMP. Crystals were cryocooled in 0.1 M Bis-Tris (pH 5.5), 21-28% (w/v) PEG2K, 10% ethylene glycol, and 1 mM EMP. All data were collected at beam line GM/CA CAT 23-ID at the Advanced Photon Source, Argonne National laboratory, and processed with HKL-3000 (27). mCTD crystals belong to P3 2 21 space group; yCTD crystals belong to space group P2 1 2 1 2 (see Table 1). Both mCTD and yCTD crystals contain two polypeptide chains per asymmetric unit.
Structure Determination and Refinement-The crystal structure of mCTD was solved by single wavelength anomalous dispersion phasing from the EMP derivative. Two heavy atom mercury positions in the asymmetric unit were identified in an anomalous difference Patterson analysis by using SHELXD as implemented in HKL-3000 (27). Starting phases were then calculated to 2.2 Å using SHELXE after refinement of the sites in MLPHARE, and RESOLVE was used to perform solvent flattening and obtain a partial structural model using HKL-3000. The initial figure of merit for the single wavelength anomalous dispersion phased map was 0.33. After density modification, the figure of merit was 0.87, and the electron density maps were easily interpreted. Subsequent model building was performed using COOT (28). Mercury atoms were bound to the unique cysteine 1619 of each polypeptide chain in the asymmetric unit. Crystallographic refinement was performed using restrained maximum likelihood refinement with individual B-factors in REFMAC (29) and PHENIX (30) with iterative model building in COOT. The native structure was solved by molecular replacement using the mercury derivative structure as the search model. Model building and refinement were done as described for the mercury structure. The final refinement statistics for the native structure are shown in Table 1.
A partial structure of yCTD was solved by using single wavelength anomalous dispersion phasing implemented in PHENIX from the selenomethionine derivative. Two selenium sites were identified out of four expected methionine residues in the asymmetric unit. The partial model was built based on a 3 Å experimentally phased electron density map and included about 50% of the two polypeptides in the asymmetric unit. This model was then used in combination with single wavelength anomalous dispersion phasing from the two-site mercury derivative to obtain the whole structural model by using MRSAD phasing in PHENIX. The structure was initially refined by using restrained maximum likelihood refinement with individual B-factors in REFMAC and followed by TLS refinement in PHENIX with iterative cycles of model building in COOT (see Table 1). There remains some disorder associ-ated with this structure that is not easily modeled and contributes to higher R values than obtained for the mCTD structure. Consistent with this view is the small number of water molecules associated with this structure, 109 as compared with 169 for the murine structure, and the disordered loop between ␤1 and ␣1 in each subunit of the structure. Coordinates and structure factors have been deposited with the Research Collaboratory for Structural Bioinformatics.
Preparation of Functional Mutants-By using site-directed mutagenesis, alanine substitutions were introduced in the mCTD and full-length murine GCN2 for residues predicted to play important roles in either stabilizing the dimeric structure or binding RNA. The QuikChange site-directed mutagenesis kit (Stratagene Inc.) was used for creating mCTD mutants, whereas the QuikChange XL site-directed mutagenesis kit (Stratagene Inc.) was used for GCN2 full-length mutants. The mutations were confirmed by DNA sequencing.
Substituted mCTD proteins were purified as His 6 -SUMO fusion proteins using the same protocol as described for mCTD, excluding the proteolytic removal of the N-terminal His 6 SUMO tag. Use of the His 6 -SUMO fusion proteins facilitated characterization as several of the substituted proteins (Y1614A, L1564A, L1561A, Y1639A, I1646A, L1587A) were found to degrade readily after removal of the N-terminal affinity tag, and dimerization of the wild-type protein was not significantly affected by inclusion of the tag. Size exclusion chromatography was used to characterize the estimated molecular weight of the wild-type and substituted mCTD samples by using a Superdex 200 10/300 GL (GE Healthcare) column buffered in 50 mM Tris (pH 8.0) and 0.30 M NaCl. Although several different size exclusion chromatography experiments were performed on tagged and untagged versions of mCTD mutants, ultimately the mCTD mutants were analyzed for dimerization as fusion proteins with results provided in Table 2 for a single representative experiment. For characterization of RNA binding activity, substituted mCTD proteins were prepared as described for mCTD that was used in the crystallographic studies in which the N-terminal affinity tag was removed.
Cell Culture and Luciferase and Immunoblot Assays-Wildtype and GCN2 Ϫ/Ϫ mouse embryonic fibroblast (MEF) cells were described previously (31). MEF cells were cultured in Dulbecco's modified eagle media (DMEM) supplemented with 1 mM nonessential amino acids, 10% (v/v) fetal bovine serum, 100 units/ml penicillin, and 100 mg/ml streptomycin at 37°C. The P TK -ATF4-Luc activity measurements were similar to those previously described (5). Briefly, GCN2 Ϫ/Ϫ MEF cells were plated at a density of 10 5 cells per well in 6-well plates and grown to ϳ40% confluency. An N-terminal FLAG-tagged cDNA encoding the wild-type mouse GCN2 (p767) or the indicated CTD mutation was inserted into a derivative of the pcDNA3 expression plasmid that was transfected along with the P TK -ATF4-Luc reporter using the FuGENE 6 transfection reagent (Roche Applied Science). The Renilla luciferase plasmid was co-transfected and served as an internal control for transfection efficiencies (Promega). 24 h after transfection cells were treated with 1 M thapsigargin, 5 mM histidinol, or no stress and cultured for 8 h. Lysates were prepared by 1ϫ passive lysis buffer (Promega), and dual-luciferase assays were per-formed as described by the Promega instruction manual. Values reported are a measure of the ratio of firefly versus Renilla luciferase units (relative light units) and represent the mean values determined for three independent transfections for each plasmid. Renilla luciferase values did not change significantly in the dual reporter assays. The results represent the means, and the S.D. is shown as an error bar. Statistical significance was calculated by using the two-tailed Student's t test.
In parallel GCN2 Ϫ/Ϫ MEF cells were transfected with the wild-type or mutant GCN2 expression plasmids or parent pcDNA3 vector alone using FuGENE transfection reagents (Promega). Transfected cells were cultured in DMEM as described above. Lysates were prepared, and equal amounts of protein were separated by SDS-PAGE followed by transfer to nitrocellulose filters as described (32). GCN2 levels were measured using polyclonal antibody-specific to this eIF2 kinase (Cell Signaling catalog #3302) followed by incubation with horseradish peroxidase-tagged secondary antibody. FLAGtagged GCN2 was measured by immunoblot analysis using monoclonal antibody that specifically recognizes the FLAG epitope (Sigma catalog #F1804). As a control for equal protein loading, actin levels were also measured by immunoblot using specific monoclonal antibody (Sigma catalog #A5441).
RNA Binding Assays-Filter binding assays were performed as described previously (33). ssRNA labeled at the 5Ј-end with either 32 P or a fluorescent tag were used to analyze mGCN2 for associated RNA binding activity by a filter binding assay. Binding of other nucleic acid substrates, including dsRNA and dsDNA, were analyzed using fluorescently labeled probes. For radioactive ssRNA binding assays, polycytidylic acid potassium salt (poly(C), Sigma catalog #P4903) with an average length of about 450 nucleotides was radioactively labeled at its 5Ј-end using [␥-32 P]ATP. The 50-l reaction mixture contained 8.3 nmol of poly(C), 25 units of T4 polynucleotide kinase (New England Biolabs), and 125 Ci of [␥-32 P]ATP incubated at 37°C for 30 min followed by the addition of 1 mM ATP for 5 min. The labeling reaction was stopped by adding 5 mM EDTA and heating the reaction mixture at 65°C for 10 min. To remove unincorporated nucleotide, the reaction mixture was passed through a G-25 Sephadex column (Roche Applied Science), and the flow-through was collected. The 32 P-labeled poly(C) had a specific activity of ϳ243 cpm/pmol, determined by using a liquid scintillation counter.
To carry out the filter binding assay with 32 P-labeled poly(C) as the RNA substrate, 30 pmol of the radiolabeled poly(C) was incubated with purified wild-type (WT) mCTD in concentrations ranging from 0.625 to 30 M in binding solution (50 mM Tris, pH 8.0, 2 mM DTT, 0.1 mM EDTA, and 100 mM NaCl). After incubating the mixture for 30 min at room temperature, 35 l of the solution was applied to a nitrocellulose filter followed by washing with 35 l of binding solution. Duplicate measurements were carried out for each concentration of protein, and the experiments were done twice. A DEAE membrane was included directly below the nitrocellulose membrane to trap any radiolabeled RNA not retained by the nitrocellulose membrane. Both nitrocellulose and DEAE membranes were dried and quantitated by FLA-5100 ␤-emission imaging system (Fujifilm) after exposure on a Fuji image plate BAS E2 2325 for 1 h. Thus, the total poly(C) in each binding assay was the sum of quantitated radiolabeled poly(C) from nitrocellulose and DEAE membranes. A ligand binding curve was plotted as the fraction of RNA molecules bound ([RNA] nitrocellulose /[RNA] nitrocellulose ϩ [RNA] DEAE ), representing bound divided by total as a function of mCTD concentration. The background obtained on the filter with RNA in the absence of protein was subtracted from all values of bound RNA. The data with the log of protein concentration versus the fraction of RNA bound was fit with a sigmoidal dose-response curve (variable slope) using SigmaPlot Version 11.0 (Systat Software Inc.). Use of standard one-site or two-site ligand binding models produced unsatisfactory fits to the data. We report the concentration for which half-maximal binding (the inflection point of the sigmoidal curve) was obtained from this analysis as an approximation of the relative binding affinity of WT mGCN2 to nucleic acid in this assay.
Filter binding assays with WT mCTD were carried out using 30 nM 5Ј-rhodamine-labeled ssRNA, dsRNA, or dsDNA in a similar manner as that described above for the binding assay using radiolabeled poly(C). The mCTD concentrations used to address binding of these oligonucleotides varied from 0.05 to 7.5 M. Binding of yCTD to ssRNA and dsRNA was also tested by using the filter binding assay with yGCN2 concentrations ranging from 0.05 to 7.5 M. Both nitrocellulose and DEAE membranes were dried and quantitated by FLA-5100 ␤-emission imaging system (Fujifilm) using a 532-nm laser. CTD binding curves for all nucleic acids tested were plotted as described above. Moreover, mCTD mutants, R1547A, K1540A, and K1603A, were also analyzed for their association with ssRNA using the filter binding assay. 30 nM 5Ј-rhodamine-labeled ssRNA was used with the same protein concentrations ranging from 0.05 to 7.5 M. The data analysis utilized a sigmoidal doseresponse curve with variable slope as described above.
Ribosome Association Assays-GCN2 Ϫ/Ϫ MEF cells were transfected with plasmids expressing wild-type versions of fulllength GCN2 or mCTD, each tagged with a FLAG epitope at the N terminus. Cells were cultured as described above and treated with 50 mg/ml cycloheximide 10 min before harvesting. Collected cells were washed with a cold solution of phosphatebuffered saline (pH 7.4) containing 50 g/ml cycloheximide, and cell lysates were prepared in a solution of 20 mM Tris-HCl (pH 7.5), 5 mM MgCl 2 , 100 mM NaCl, and 0.4% Nonidet P-40 supplemented with 50 g/ml cycloheximide. Cell lysates were processed by passage through a 23-gauge needle and precleared by microcentrifugation (10,000 rpm for 10 min at 4°C). The supernatant was then layered onto a 10 -50% sucrose gradient solution containing 20 mM Tris-HCl (pH 7.5), 5 mM MgCl 2 , 100 mM NaCl, and 50 g/ml cycloheximide and then subjected to centrifugation in a Beckman SW-41Ti rotor for 2 h at 40,000 rpm at 4°C. Gradients were fractionated using a Biocomp Gradient Station, and absorbance of RNA at 254 nm was recorded using an in-line UV monitor. Proteins from equivalent volumes of the gradient fractions were precipitated with TCA and separated by SDS-PAGE, and FLAG-tagged GCN2 was measured by immunoblot analysis using antibody that recognizes the FLAG epitope.
To measure ribosome association of murine GCN2 expressed in yeast S. cerevisiae, full-length murine GCN2, which was tagged with a FLAG epitope at the N terminus, was expressed in yeast strain J82 (MATa ura3-52 leu2-3 leu2-112  gcn2⌬ trp1⌬-63 p1098[SUI2-S51A LEU2]). The murine GCN2 cDNA was encoded in a URA3-marked high copy number plasmid that was transformed into the yeast strain and expressed using the galactose-inducible-CYC1 hybrid promoter, as described (34). Strain J82 expresses eIF2␣-S51A, rendering the strain insensitive to growth inhibition by expressed murine GCN2 phosphorylation of the yeast eIF2␣. This would allow for sufficient levels of the GCN2 proteins to be detected by immunoblot analyses. Lysates were prepared from the yeast strain expressing murine GCN2, and lysates were subjected to sucrose gradient centrifugation and fractionated using Biocomp Gradient Station as described (21). The FLAG-tagged murine GCN2 was visualized in the fractions by immunoblot analysis using monoclonal antibody specific for the FLAG epitope. As a control we also expressed full-length yeast GCN2 tagged at the N terminus with FLAG expressed using a galactose-inducible promoter in yeast (23). Lysates were prepared, and yeast GCN2 was similarly analyzed for association with ribosomes.

RESULTS
Crystallization of C-terminal Domains of GCN2-Crystallization of the CTD from both murine and yeast GCN2 proteins, designated mCTD and yCTD, respectively, required identification of an optimal N terminus for each. We initially purified an N-terminally His 6 -tagged mCTD (1493-1648) and then proteolyzed the sample with thrombin to remove the N-terminal tag. However, after cleavage, we observed two distinct electrospray ionization masses consistent with two different N termini, the expected N terminus for thrombin cleavage and a second N terminus beginning at residue 1514. We then expressed and purified mCTD (1514 -1648) and obtained diffraction quality crystals of this protein. Similarly, initial crystals of yCTD (1498 -1659 or 1536 -1659) were very small, but diffraction quality crystals were subsequently obtained for yCTD including residues 1519 -1659, analogous to the mCTD (1514 -1648) (see "Experimental Procedures"). Crystals of both murine and yeast CTD contain two polypeptides in the asymmetric unit.
The CTDs from Yeast and Murine GCN2 Form Interdigitated Dimeric Structures-The crystal structures of murine and yeast CTDs reveal novel interdigitated dimeric structures with no obvious matches to known three-dimensional structures (Fig.  1). Both CTD structures were determined by using experimental phasing methods (see "Experimental Procedures" and Table  1). The mCTD structural model includes residues 1530 -1648 for the A chain and 1526 -1648 for the B chain; the yCTD includes residues 1537-1549 and 1559 -1659 for the A chain and residues 1538 -1555 and 1560 -1659 for the B chain. Missing residues in yCTD are disordered and correspond to the connecting loop region between the N-terminal ␤-strand and ␣-helix. Formation of the mCTD and yCTD dimers buries an unusually large accessible surface area for each molecule, 4200 and 4300 Å 2 , respectively, potentially contributing to the stability of each molecule.
In comparing a single subunit of the murine and yeast CTD structures, the r.m.s.d. is 2.1 Å for superpositioning of 74 C␣ atoms, consistent with the same overall fold ( Fig. 2A). However, the N-terminal regions have different relative dispositions despite sharing a conserved core structure. As shown in the structure-based sequence alignment in Fig. 2B, the lengths of the comparable secondary structural elements and intervening loop regions differ significantly in the two structures with only 12 residues that are identical in the two sequences. There are also important structural differences in the overall shape of the dimeric CTD structures as well as the relative positioning of secondary structural elements within these dimeric structures despite similarities in overall organization.
The dimeric structures of the murine and yeast CTDs have three interfaces that are each situated between the two polypeptide chains. We refer to these interfaces as I, II, and III, with III being the central interface between the two subunits (Fig. 3). Interface I in both the yeast and murine CTDs arises principally from hydrogen-bonding interactions between the N-terminal ␤-strand 1 and ␤-strand 2 from the second polypeptide chain. Interface II results from the interaction of the N-terminal ␣-helical region in the mCTD structure with the second polypeptide chain within each subunit. This interface features the following residues, His-1549, Gln-1552, Val-1553, Leu-1557, Thr-1560, Leu-1561, and Leu-1564 from the A chain and residues Ile-1571, Ile-1573, Phe-1635, Tyr-1637, Tyr-1639, and Tyr-1644 from the B chain, with A and B chains referring to the two different polypeptides that make up the dimer (Fig. 3A). By comparison in interface II of yCTD, the N-terminal ␣-helix from one polypeptide interacts with the other N-terminal ␣-helix involving residues Leu-1574 and Leu-1577 from each poly-peptide. This interaction is unique to yCTD; the long ␣-helices in mCTD do not interact directly with one another. Additional interactions involve the long ␣-helix include residues Trp-1641, Tyr-1562, Ile-1573, Ser-1569, and Ala-1566 from the A chain and Tyr-1645, His-1647, and Ser-1652 from the B chain (Fig. 3B). Interface II in mCTD and yCTD is only approximately equivalent due to structural differences. Residues involved in interface II that are located in structurally similar locations include the following (murine/yeast): His-1549/Tyr-1562, Gln-1552/Ala-1565, Leu-1557/Ser-1569, and Thr-1560/Ile-1573 in the A chain and Tyr-1639/His-1647, Tyr-1637/Tyr-1645, and Tyr-1644/S1652 in the B chain.
Interface III arises from the interactions of the C-terminal strands and ␣3 in the mCTD polypeptide or ␣2 in the yCTD. Interactions in interface III of mCTD include residues Ler-1584, Leu-1587, Arg-1645, Ile-1646, and Phe-1648 from the A and B chains (Fig. 3C), whereas those in yCTD include residues Glu-1594, S1597, Ile-1598, and Arg-1659 in the A and B chains (Fig. 3D). Due to structural differences, the only residues within interface III that are structurally equivalent in the two structures are (murine/yeast) Leu-1584/Glu-1594 and Leu-1587/ Ser-1597. The C-terminal strand in the yCTD is longer by three residues, and the C-terminal Arg-1659 is buried in the interface through both charge-charge and hydrogen-bonding interactions with Glu-1594 and Ser-1597 of the other chain. The C-terminal residue of the mCTD Phe-1648 packs within the hydrophobic core. As a result of the larger interface III in the yCTD, the relative disposition of the two subunits is different than that of the mCTD structure. Dimerization is known to be important in the function of GCN2, and the structures pro- Ribbon renderings are shown for the dimeric CTD structures murine, one polypeptide chain in red, the second in green (A) and yeast, one polypeptide in magenta and the second in cyan (B) . The murine and yeast CTDs are oriented such that the ␤-sheets in the murine red chain and yeast cyan chain are in approximately the same view. C, a portion of the experimental electron density map for the mCTD structure is shown contoured at 1.5 with the ball-and-stick final model (yellow, C; yellow; red, O; blue, N) for two ␤-strands, left strand including residues 1643-1648, the C-terminal strand, and the right strand residues 1632-1637. Experimental phasing was obtained for a mercury derivative as described under "Experimental Procedures." The final model was superimposed on the initial model built for the mercury phased map. vide a basis for evaluating the role of specific residues in dimeric interactions that may regulate function.
GCN2 Dimerization Occurs in Vitro and Is Required for Translational Control-To assess the role of residues within interfaces II and III of the mCTD in dimerization, we introduced alanine substitutions for the following residues: Leu-1561, Leu-1564, Leu-1587, Tyr-1614, Tyr-1639, and Ile-1646 (Fig. 3E). Residues Leu-1561, Leu-1564, and Tyr-1639 are involved in forming hydrophobic dimer interface II; Leu-1587 and Ile-1646 form hydrophobic interface III; Tyr-1614 stacks with Pro-1538 and hydrogen bonds to Leu-1536 of another polypeptide chain. As interface I is formed primarily by backbone hydrogen-bonding interactions, no residues were selected from this interface for further analysis. We also selected residues that may have a role in ribosome/RNA binding activities, functions previously attributed to the yeast CTD (21), and characterized their dimerization properties through alanine substitution of Lys-1540, Arg-1547, and Lys-1603 (Fig. 3E). These mCTD residues are located on the surface of mGCN2 and roughly define possible RNA binding surfaces. Lys-1540 is located in a loop between ␤1 and ␣1; Lys-1552 and Lys-1553 are located in this same region of the structure in yCTD. Arg-1547 is located within ␣1, and the structurally equivalent residue in yCTD is the solvent-exposed residue Trp-W1560. Lys-1603, located in ␣4, is structurally equivalent to Arg-1609 in yCTD. The substituted mCTDs were expressed and purified as N-terminal His 6 -SUMO fusion proteins from E. coli and then characterized by size exclusion chromatography. The percentage of dimeric protein observed for each of the substituted fusion proteins is shown in Table 2. Substitution of residues within the interfaces II or III, with the exception of Leu-1587, significantly reduced the amount of dimeric protein observed by size exclusion chromatography. The CTD featuring L1587A exhibited intermediate behavior, with approximately half of the protein observed in a dimeric state, whereas all of the other CTDs were ϳ25% or less dimer. Given the nature of the dimeric mCTD structure, substitutions of buried residues within these interfaces may also affect the overall stability or folding of the protein. In contrast, substitution of residues predicted to play a role in the RNA binding activity of murine CTD had no effect on the percentage of dimer, with each displaying wild-type levels of dimerization.
To determine the contribution of the CTD dimerization to GCN2 function, WT and mutant versions of full-length murine GCN2, which were tagged with the FLAG epitope at the N terminus, were expressed in GCN2 Ϫ/Ϫ MEF cells. Although there were some variations in the levels of the mutant versions of GCN2 compared with WT, each was expressed at substantially higher levels than that measured for endogenous GCN2 protein in GCN2 ϩ/ϩ MEF cells (Fig. 4A). Next, we measured the translational expression of ATF4 using a well-characterized reporter containing sequences encoding the 5Ј-leader of the ATF4 mRNA inserted between the constitutive TK promoter and the firefly luciferase reporter. There was a 2-fold increase in ATF4 expression in the GCN2 Ϫ/Ϫ MEF cells upon treatment with histidinol, a drug that induces a starvation response and thereby elicits GCN2 phosphorylation of eIF2 (Fig. 4B). There was no increase in luciferase activity in GCN2-deficient cells upon histidinol treatment. Further illustrating the selectivity for activation of GCN2 in response to nutrient depletion, ATF4 FIGURE 2. Comparison of murine and yeast CTD subunits reveals a conserved core structure. A, the red and cyan subunits from murine and yeast, respectively, are shown superimposed (74 C␣ positions, r.m.s.d. 2.1 Å) revealing a conserved core structure but different relative positions for the N-terminal portions of each subunit. B, a structure-based sequence alignment is shown for murine and yeast CTD, with secondary structural elements indicated above for murine and below the sequence for yeast. The sequence identity is Ͻ10%. Overall, the secondary structural elements are largely conserved with differences in the length of the N-terminal ␣-helices and the ␤-strands.
translational expression was induced independent of this eIF2 kinase in response to treatment with thapsigargin, a potent inducer of endoplasmic reticulum stress and an alternative eIF2 kinase PERK. Transfection of a cDNA expressing the WT GCN2 restored ATF4 translational expression in the GCN2 Ϫ/Ϫ cells, with more than a 4-fold increase in luciferase activity upon nutrient stress (Fig. 4B). By comparison, for each of the GCN2 proteins with substitutions of important residues in interfaces II or III, significantly reduced ATF4 expression was observed, consistent with the idea that CTD dimerization is required for GCN2 activity. GCN2 containing the L1587A expressed at levels higher than wild-type GCN2 but exhibited intermediate levels of CTD dimerization in the in vitro assay and also displayed partial induction of ATF4 expression during nutrient deprivation, suggesting that its luciferase activity is correlated with dimerization. Finally, the MEF cells expressing GCN2 with Ala substitutions of Lys-1540, Arg-1547, and Lys-1603 displayed sharply lowered levels of ATF4 translational expression, suggesting that these surface-accessible basic residues in the CTD are required for induced GCN2 activity (Fig.  4B).
Murine CTD Binds to Both Single-stranded and Doublestranded Nucleic Acid-To determine whether the WT mCTD binds RNA, we investigated the interactions between RNA and mCTD by using a filter binding assay. The binding of WT mCTD to 32 P-poly(C) followed a sigmoidal dose-response curve with a half-maximal binding value of 5.6 M (Fig. 5A). To . The murine interface involves primarily hydrophobic interactions which bury its C-terminal residue Phe-1648, whereas that in yeast involves a hydrogen-bonding network with the C-terminal Arg-1659. E, functional studies were conducted on mCTD proteins with individual substitutions for the residues shown in space-filling van der Waal sphere models; yellow green, residues involved in interfaces II or III as specified under "Results"; blue, potential nucleic acid binding residues.

TABLE 2 Analysis of dimerization for substitutions in dimer interfaces or potential RNA binding sites
Proteins with substitutions expected to affect either dimerization or RNA binding were analyzed by size exclusion chromatography. Residue changes for each mCTD mutant are listed along with the site affected by the residue substitution, and the percentage of dimerization is compared to wild-type mCTD.

Sumo-mGCN2
Site % Dimer further verify if ssRNA binding was length-or sequence-specific, we carried out filter binding assays with a 5Ј rhodaminelabeled 35-mer ssRNA as the substrate for WT mCTD. Quantification of the results indicated that binding of rhodamine-labeled ssRNA and wt mCTD followed a sigmoidal doseresponse curve that was similar to its binding curve with 32 Plabeled poly(C). However, mCTD bound the shorter ssRNA with a half-maximal binding value of 0.52 M (Fig. 5B) as compared with 5.6 M for poly(C). This difference in apparent binding affinity may be due to the length of the ssRNA molecules used for each experiment. For each ssRNA substrate, the sigmoidal dose-response curve is consistent with non-sequence specific binding of mCTD. The steeper slope associated with the binding to poly(C) likely reflects binding of more mCTD molecules per poly(C) due to its length. To determine whether WT mCTD also binds other nucleic acids, we carried out filter binding experiments for dsRNA and dsDNA. WT mCTD bound dsRNA and dsDNA with relative binding affinities similar to that exhibited for ssRNA. In previous studies yCTD was reported to bind dsRNA. To address whether yCTD also binds ssRNA, filter binding experiments were carried out using rhodamine-labeled ssRNA and dsRNA, and the binding was compared with that of wt mCTD. Yeast CTD exhibited a half-maximal binding value of 1.8 M for ssRNA and 1.2 M for dsRNA, respectively, values quite comparable to those obtained for mCTD. Thus, both yeast and murine GCN2 can bind ss-and dsRNA.
To assess the contribution of conserved basic residues in the mCTD binding of ssRNA, filter binding experiments were performed with mutants, K1540A, R1547A, and K1603A, identified as defining potential nucleic acid binding surfaces in mCTD using the rhodamine-labeled 35-mer ssRNA. K1540A, R1547A, and K1603A mCTD mutants exhibited a maximal binding of 15.5, 57.6, and 71.3%, respectively, as compared with FIGURE 4. Amino acid substitutions in the CTD of murine GCN2 reduce ATF4 translational expression during nutrient stress. A, lysates were prepared from GCN2 ϩ/ϩ and GCN2 Ϫ/Ϫ MEF cells or GCN2 Ϫ/Ϫ cells transfected (tsf) with plasmids expressing WT or the indicated mutant versions of mGCN2, each tagged with an N-terminal FLAG epitope. Equal amounts of proteins were separated by SDS-PAGE, and the levels GCN2 levels were measured immunoblots by using antibody specific to this eIF2 kinase (top panel) or the FLAG tag (bottom panel). In parallel actin levels were measured by immunoblot to ensure equal protein loading. B, ATF4 translational expression was measured in GCN2 ϩ/ϩ MEF cells and its GCN2 Ϫ/Ϫ counterpart expressing WT GCN2, the indicated mutant versions of GCN2, or vector alone. These cells were also transfected with the P TK -ATF4-Luc reporter and treated with histidinol (His-OH), thapsigargin (TG), or no treatment (NT). Relative luciferase activities are shown, with the error bars indicating the S.D., and asterisks indicating significant differences from the non-stressed samples (p Ͻ 0.05). WT mCTD. This suggests that substitution of charged lysine and arginine residues with alanine negatively impacts the ability of the CTD of mGCN2 to bind ssRNA.
Murine CTD Does Not Stably Associate with Ribosomes-We next addressed whether the CTD mediates interaction of mGCN2 with ribosomes through fractionation of lysates by sucrose gradient centrifugation as previously described for yGCN2 (19,21,32). We followed this procedure for ribosome association using lysates prepared from GCN2 Ϫ/Ϫ MEF expressing full-length or the CTD portion of murine GCN2, each tagged with the FLAG epitope at the N terminus. Cell lysates were applied to sucrose gradients and subjected to centrifugation, and after fractionation, ribosomes were visualized (Fig. 6). Tagged mGCN2 proteins in the fractions were visualized by immunoblot. Yeast GCN2 was reported to stably associate with free 60 S ribosomes and translating ribosomes independent of stress (19,21). We did not observe significant association of the full-length mGCN2 or the mCTD proteins with ribosomes in the MEF cells (Figs. 6A). In each case the GCN2 proteins were localized to the top portion of the gradient fractions with no GCN2 observable in the ribosomal fraction. As expected ribosomal S6 protein was found in the ribosome fractions. Tagged mGCN2 was also found largely free of ribosomes when the MEF cells were subjected to histidinol treatment before harvesting (Fig. 6A).
The N-FLAG murine GCN2 was previously expressed in yeast gcn2⌬ mutant cells and shown to complement for the function of this eIF2 kinase (34). We wished to determine whether expression of the full-length murine GCN2 tagged at the N terminus with FLAG in yeast displayed association with ribosomes that were separated by sucrose gradient centrifugation. Although the majority of mGCN2 protein was localized to the top portion of the sucrose gradient, free of ribosomes, we could reproducibly detect a minor portion of the expressed murine GCN2 that co-migrated with free ribosomes and polysome upon overexposure (Fig. 6B). By contrast, yGCN2 that was also N-FLAG-tagged co-migrated with the 60 S ribosomal subunit and monosomes. These results suggest that mGCN2 does not stably bind to ribosomes under the same experimental conditions that were successfully used to demonstrate ribosomal association with yeast GCN2.

DISCUSSION
GCN2 protein kinase is expressed among virtually all eukaryotes, suggesting that phosphorylation of eIF2 and translational control is central to eukaryotic stress responses to starvation for nutrients. Most of our understanding of the regulation of GCN2 activity comes from studies in yeast S. cerevisiae in which the activity of the eIF2 kinase domain of GCN2 is controlled by the flanking HisRS-related domain and CTD in response to nutrient availability. Whereas the GCN2 protein kinase and HisRS-related domains show sequence similarities that are broadly shared among eukaryotes, the CTDs show considerably more variation, as illustrated by the sequence alignments from yeast and mouse GCN2 (Fig. 2B).
The dimeric structures of the murine and yeast CTD are unusual in lacking similarity to any other known structures and in their organization including secondary structural elements from two different polypeptide chains within each subunit. In contrast to three-dimensional domain-swapped dimers, the CTD dimer is more accurately described as interdigitated with the core ␤-sheet formed through interactions of strands from two different polypeptide chains. A potential benefit of this dimeric arrangement is that it may provide additional stability to the dimeric structure. This idea is supported by the unusually large surface area of ϳ4000 Å 2 that is buried upon dimer formation in each CTD. Perhaps the most surprising result of our FIGURE 6. Murine GCN2 is not stably associated with ribosomes. A, lysates were prepared from GCN2 Ϫ/Ϫ MEF cells expressing either full-length mGCN2 (mGCN2-FL) or the mCTD (mGCN2-CTD) that contained a FLAG epitope and the N terminus. Cells were grown in non-stressed media (No stress), and lysates were prepared, subjected to sucrose gradient centrifugation, and fractionated, and profiles were monitored by absorbance at 254 nm. The 40 S and 60 S ribosomal subunits, 80 S monosomes, and polysomes are highlighted. Equal volumes from each fraction sample were separated by SDS-PAGE, and the GCN2 proteins were measured by immunoblot using monoclonal antibody that specifically recognizes the FLAG tag. Ribosomal S6 protein in the fractions was also measured by immunoblot. Alternatively, MEF cells expressing mGCN2-FL were grown in the presence of histidinol for 6 h (Histidinol treatment), and after sucrose gradient centrifugation the GCN2 protein was measured by immunoblot in each fraction. B, lysates were prepared from yeast expressing either mGCN2 or yGCN2 and analyzed by sucrose gradient centrifugation. The levels of FLAG tagged mGCN2 and yGCN2 in the gradient fractions were measured by immunoblot analysis. structural analysis was that both yeast and murine GCN2s share a similar dimeric organization and a common core structure. At the same time, there are significant structural differences including the overall shape and size of the dimers dictated largely by interface III in each of the structures. There are also significant functional differences between the yeast and murine CTDs.
Functional activities associated with yCTD include dimerization, nucleic acid binding, and ribosome association. The dimeric mCTD structure provides a basis for assessing the role of specific subunit interactions in dimerization and the functional consequences of disrupting these interactions in the mammalian system. Collectively, size exclusion chromatographic characterization of substituted mCTD proteins and analysis of the ability of these same substitutions to disrupt induced ATF4 translational expression in cells supports the hypothesis that dimerization of the CTD is important for function in mGCN2. In this structure-based approach, we found that substitution of any of the residues involved in interactions within dimer interfaces negatively impacted the ability of GCN2 to activate ATF4 expression after treatment with histidinol. Thus, as was reported for yGCN2 (22,24,25), dimerization appears to be important for the function of mGCN2, and in this regard the proteins are similar.
One of the reported regulatory features associated with yCTD is ribosomal association. Yeast GCN2 was found to stably bind to free and translating ribosomes that were separated by sucrose gradient centrifugation (19,21), and it has been proposed that association with the translational machinery provides GCN2 with access to bind uncharged tRNA that enters the A site of ribosomes during periods of amino acid starvation (25). We carried out a sucrose gradient analysis using lysates from MEF cells expressing FLAG-tagged full-length GCN2 or mCTD and found minimal binding to ribosomes (Fig. 6A). Furthermore, when we expressed the full-length mGCN2 in yeast, we observed minimal ribosome association, with the majority of the protein situated at the top of the gradients, free of ribosomes (Fig. 6B). Thus, in contrast to its yeast counterpart, mGCN2 does not appear to exhibit stable ribosomal association, suggesting a regulatory difference between the mGCN2 and its yeast counterpart.
Binding to dsRNA by the CTD had been proposed as the basis for the association of yGCN2 with ribosomes. Because mGCN2 does not appear to stably associate with ribosomes, one possibility was that mGCN2 does not bind dsRNA. However, this proved not to be the case, as both yCTD and mCTD bind ssRNA and dsRNA as analyzed by filter binding assay. These results are consistent with a general affinity for nucleic acid, and in accord with this idea, mCTD also binds dsDNA. Binding of mCTD to ssRNA is consistent with its role in sensing infection by ssRNA viruses as previously reported (7,8). Interestingly, GCN2 was also shown to provide for resistance to DNA viruses mouse cytomegalovirus (MCMV) and human adenovirus, and loss of GCN2 (EIF2AK4) was shown to block eIF2 phosphorylation upon MCMV infection (35). This would suggest that GCN2 can be activated in response to a broader range of nucleic acids than previously suggested. Given the lack of selectivity between ss-and dsRNA and dsDNA, it is likely that the interaction of the CTD with nucleic acid is governed in part by electrostatic interactions. This idea is supported by the moderate binding affinities that we infer from the filter binding experiments.
Both mCTD and yCTD have similar nucleic acid binding properties, and yet mGCN2 does not appear to be stably associated with ribosomes. This suggests that the nucleic acid binding properties of mCTD may be central for mGCN2 activation in response to diverse stress signaling involving nucleic acids, including those involving virus infection. Furthermore, the GCN2 CTD may serve in conjunction with the adjacent HisRSrelated domain to bind different uncharged tRNAs that accumulate during nutrient deprivation, facilitating activation of the eIF2 kinase (23). The functional relevance of nucleic acid binding by mCTD was further investigated by identifying structurally conserved basic residues found in both yCTD and mCTD. Three conserved basic residues, Lys-1540, Arg-1547, and Lys-1603, were substituted with Ala in mGCN2 and then tested for dimerization, ATF4 translational expression, and nucleic acid binding. None of these conserved basic residues was found to impact dimerization, but all were found to decrease ATF4 expression after treatment with histidinol, suggesting that these residues are important for GCN2 function. All three substituted proteins showed reduced binding to ssRNA as compared with the WT mCTD. Of these, K1540A resulted in the most significant reduction in binding activity, retaining only 15% of the activity observed for WT mCTD. In yCTD, the structural equivalent of Lys-1540 is Lys-1553; a mutant including K1552L/K1553I/K1556I was shown to selectively block the interaction between yeast GCN2 and 60 S ribosomal subunits and reduce translational control induced by yGCN2 in response to amino acid limitation (21). This same yCTD mutant was also reported to block the interaction of yGCN2 with uncharged tRNA (23), supporting a common role in both yeast and mammalian systems for interaction of the CTD with uncharged tRNAs.
Taken together, our results suggest that although some aspects of GCN2 regulation involving the CTD are conserved between yeast and mammals, others such as ribosomal association are not. Furthermore, our studies are consistent with a broader functional role for the CTD through its recognition of both ss and ds nucleic acid than previously suggested. Not only does the CTD play a role in recognition of uncharged tRNA, but it may also play a role in recognizing viral nucleic acid through a bipartite binding site including the HisRS-related domain.