A Network of Hydrophobic Residues Impeding Helix αC Rotation Maintains Latency of Kinase Gcn2, Which Phosphorylates the α Subunit of Translation Initiation Factor 2^▿

MATERIALS AND METHODS

The phenotypes of plasmid-borne GCN2 alleles were tested in strains H1149 (MATα ura3-52 inol leu2-3 leu2-112 gcn2-Δ63-3284::LEU2 [HIS4::lacZ integrated at ura3-52]) (36) and H1817 (MATa ura3-52 leu2-3 leu2-112 gcn2Δ sui2Δ trp1-Δ63 [GCN4-lacZ TRP1 integrated at trp1-Δ63] p1098[SUI2-S51A LEU2]) (9). Strain HQY346 is a GAL2 derivative of GP3299 (MATa ura3-52 leu2-3 leu2-112 trp1-Δ63 gcn2Δ gcd2Δ::hisG pAV1033[GCD2-K627T, TRP1]) (26). The plasmids and mutagenic primers employed in this work are listed in Tables 1 and 2. All GCN2 mutations were constructed by a PCR fusion technique (39) using p722 or its derivative (p722-NruI) containing an NruI site introduced by the “silent” substitution of TCA for TCG at codon 593 in GCN2 and primers listed in Table 2. For constructing the majority of the mutations, two PCRs were carried out using p722 or p722-NruI as a template and (i) primer 189 and the appropriate mutagenic primer and (ii) primer 199 and the primer complementary to the mutagenic primer used in the first reaction. A third PCR was conducted using the products of the first two reactions and primers 189 and 199. The final PCR fragment was purified and digested with different pairs of restriction enzymes, Blp1 and BspEI, NruI and Asp718, or BspEI and Asp718, depending on the location of the mutation and whether p722 or p722-NruI was employed. The digested PCR fragment was ligated with p722 (or p722-NruI) and digested with the same pair of enzymes and, in some cases, with XhoI to eliminate the recovery of clones containing nonrecombinant, undigested p722 (or p722-NruI). (XhoI digestion was not possible when cloning BlpI/BspEI fragments, and in those cases, the digested vector was dephosphorylated with calf intestine phosphatase to reduce the background.) A similar approach was used to introduce the mutations contained in plasmids pHQ1746 and pHQ1747, except that primer 290 replaced primer 199, and the final PCR product was digested with BlpI and XhoI and used to replace the corresponding fragment in p722. DNA sequence analysis of the region encompassing the entire subcloned PCR fragment was conducted to identify recombinant plasmids containing the desired mutations and no unintended mutations. Sensitivities of yeast transformants to 3-aminotriazole (3-AT) or 5-fluorotryptophan (5-FT) were determined as previously described (30) and as indicated in the figure legends.

To measure eIF2α phosphorylation, total protein extracts were prepared by TCA extraction (31), resolved by SDS-PAGE, and subjected to Western analysis as described previously (29). Western signals were quantified by video image densitometry using Scion Image software (imported from NIH Image for the Macintosh by Scion Corporation and available on the Internet at http://www.scioncorp.com).

RESULTS

DISCUSSION

In this report, we provide strong evidence that the latency of Gcn2 in nonstarved cells depends on a network of hydrophobic residues comprised of L856 in the activation loop and hydrophobic residues in strand β3, strand β5, or helix αC. The crystal structure of inactive Gcn2 reveals a ∼90° axial rotation of helix αC, which blocks formation of the E643-K628 salt bridge crucial for positioning the triphosphate moiety of ATP. L856 appears to impede the rotation of αC to the position observed in active KD structures that is permissive for the E643-K628 salt bridge (25). The bulky hydrophobic side chain of L856 packs against the hydrophobic side chains of I640, V644, and L647 in αC, I630 in β3, and I786 in β5 (Fig. 4A), leading us to predict that the ability of L856 to block reorientation of αC depends on its insertion into a hydrophobic pocket comprised of these other residues. Supporting this prediction, our exhaustive substitution analysis of L856 revealed that replacing this residue with Ala, Gly, or any amino acid with a polar or charged side chain leads to hyperactivation of Gcn2 in vivo, bypassing the amino acid starvation signal and producing a high constitutive level of eIF2α phosphorylation that inhibits cell growth in nutrient-replete medium. Remarkably, the replacement of L856 with Ile, Val, Phe, and Trp produced little or no increase in Gcn2 function, indicating that only these four bulky hydrophobic residues can substitute for Leu-856 to maintain WT latency. Substitutions of residues comprising the hydrophobic pocket also activate Gcn2, as the double substitutions L647A I640A and I630A I640A elicited eIF2α hyperphosphorylation and Slg⁻ phenotypes approaching those of L856A. These results provide compelling in vivo evidence that the L856 hydrophobic network is crucial for Gcn2 latency in nonstarved cells.

It could be argued that the activating substitutions we identified in residues comprising the hydrophobic pocket stimulate Gcn2 function by a mechanism unrelated to their proposed role in buttressing the inhibitory orientation of L856. We consider this unlikely for several reasons. First, it should be recalled that we targeted these residues for mutagenesis precisely because they make direct interactions with L856 in the crystal structure of the inactive Gcn2 KD (25). Second, we found that substitutions at L647 conform to the rule we established for activating substitutions at L856, in that replacing L647 with Ala activates Gcn2, whereas its replacement with another bulky hydrophobic residue, Ile or Val, maintains Gcn2 latency. Similarly, we found that replacing I640 with Leu or Val maintains nearly WT or partial latency, respectively, whereas the Ala substitution strongly activates Gcn2. Thus, residues I640 and L647 resemble L856 in requiring a bulky hydrophobic side chain to block Gcn2 activation under nonstarvation conditions. Third, we found that Ala substitutions at L647 and I640 individually conferred 5-FT^R and no Slg⁻ phenotype, but they produced a synthetic Slg⁻ phenotype when combined in the same mutant. I640A and I630A also produced a Slg⁻ phenotype only in the double mutant. Such synthetic interactions frequently indicate that both single mutations partially impair the same pathway or function (17). Finally, we showed that the I640A L647A double mutation resembles L856A by strongly suppressing the activation defect conferred by the m2 mutation. Such suppression is not a general feature of activating mutations in GCN2, as most GCN2^c mutations remain dependent on the m2 motif for efficient tRNA binding for strong kinase activation (29). Hence, the weight of the evidence supports the idea that the inhibitory conformation of L856 is buttressed by its hydrophobic interactions with (at a minimum) I630, I640, and L647.

The crystal structure of inactive Gcn2 predicts that the inactive orientation of αC is also stabilized by the incorrect salt bridge formed between E643 (in αC) and R834 in the catalytic loop (25) (Fig. 1B). As proposed previously (20), autophosphorylation of T882 in the activation loop should destabilize the incorrect rotation of αC by neutralizing the positive charge of R834 and disrupting the inhibitory E643-R834 salt bridge (Fig. 1B). Supporting this hypothesis, the L856A mutation suppresses the activation defect conferred by the autophosphorylation site mutation T882A. Thus, it appears that decreasing the impediment to αC reorientation by disrupting the L856 hydrophobic network helps to overcome the second impediment posed by the inhibitory E643-R834 salt bridge in the T882A mutant, where the latter cannot be disrupted by autophosphorylation. T882 phosphorylation still contributes to Gcn2 activation when the L856 hydrophobic network is impaired, which is fully consistent with the idea that the hydrophobic network and E643-R834 salt bridge both contribute to impeding the reorientation of αC to the active conformation.

It is highly interesting that L856 corresponds to a key residue in the epidermal growth factor receptor (EGFR), L858, which appears to help buttress an outward displacement of αC in the inactive form of this kinase (38). Moreover, mutating L858 to Arg activates EGFR in vitro, and this mutation is found frequently in a subset of lung cancers (40, 41). In the case of inactive EGFR, L858 resides within a helical turn in the activation loop that seems to force αC into the incorrect position, and L858 makes hydrophobic contacts with other residues in the KD interior. Presumably, the charged side chain introduced by the activating substitution L858R is incompatible with these hydrophobic interactions and destabilizes the helical structure needed to displace αC. However, most of the hydrophobic residues interacting with L858 in inactive EGFR, located in β2, β3, β5, and the activation loop (22), are distinct from those in the hydrophobic pocket surrounding L856 in inactive Gcn2. Unlike in Gcn2, residues in αC itself do not participate in the EGFR hydrophobic pocket, and two Leu residues in the activation loop helical turn in EGFR (L861 and L862) are not conserved, and do not participate, in the hydrophobic pocket of Gcn2. Hence, while the equivalent Leu residue plays a key role in both Gcn2 and EGFR latency, the hydrophobic pocket that supports its inhibitory function is constituted quite differently in the two kinases.

Azam et al. (2) recently proposed an alternative mechanism for the role of L858 in maintaining the latency of EGFR through destabilizing the “hydrophobic spine,” which links the N and C lobes in the active conformation of EGFR and numerous other protein kinases (21). The equivalent Leu residues in ABL and SRC also seem to participate in dismantling the hydrophobic spines in the inactive forms of these kinases (2). In particular, L403 in inactive ABL interacts with the Phe residue of the DFG motif (F401) in the “DFG-out” conformation, where F401 is flipped out of the active site. Furthermore, the DFG Asp is not hydrogen bonded to the DFG Gly, so that inactive ABL lacks an interaction thought to be critical for proper positioning of the DFG Asp for catalysis (21). It is unclear whether a similar mechanism applies to the role of L856 in Gcn2 latency. Although the DFG Asp and Gly are too far apart (∼9 Å) to hydrogen bond in the inactive Gcn2 KD (D853 and G855; Fig. 8A), we found that the hydrophobic spine appears to be intact in inactive Gcn2 (25).

In the PKR KD, all but one of the residues corresponding to the L856 hydrophobic network in Gcn2 are perfectly conserved or (in the case of Gcn2 I630) replaced by a different bulky hydrophobic amino acid. Thus, a hydrophobic network involving the equivalent of L856 (L435) might contribute to the latency of PKR in its monomeric state in the manner described here for Gcn2. In the crystal structure of active PKR, L435 is oriented away from the putative hydrophobic pocket (6) (Fig. 8B), adopting essentially the same position as the corresponding “DFG + 1” residue observed in the active conformations of many different kinases (21). Moreover, the DFG Asp and Gly are close enough (3.8 Å) to hydrogen bond in active PKR (D432 and G434; Fig. 8B). In activated Gcn2, it seems likely that L856 will isomerize to the outward position observed for L435 in active PKR, eliminating the impediment to αC reorientation and achieving the stereotypical active conformation of the DFG + 1 motif (21). As shown in Fig. 8C, the inactive form of kinase CDK2 (33) shows the unfavorable positions of the Asp (D145) and + 1 (L148) residues of the DFG + 1 motif, similar to those present in (inactive) Gcn2 (cf. Fig. 8A and C), whereas the active form of CDK2 (3) displays the favorable conformation of the DFG + 1 motif (cf. Fig. 8B and D) (21). Interestingly, we realized that inactive CDK2 also contains a hydrophobic network involving L148 that is highly similar to that identified here in Gcn2, with residues I35, I52, and L55 corresponding to I630, V644, and L647 in Gcn2 (cf. Fig. 8A and C), which might contribute to the latency of this critical kinase as well.

Activation of Gcn2 in amino acid-starved cells requires the binding of uncharged tRNA to the HisRS-like domain, dependent on motif 2. If the postulated reorientation of αC is a crucial step in the activation pathway, then decreasing the impediment to αC reorientation by disrupting the L856 hydrophobic network should reduce the requirement for motif 2. Our results confirm this prediction by showing that the L856A and I640A L647A activating mutations restore kinase function to the m2 mutant. As would be expected if αC reorientation is not the only conformational change during kinase activation, the m2 substitutions, in turn, reduce the functions of the L856A and I640A L647A proteins. We propose that stimulatory input from uncharged tRNA and the HisRS domain is still required to remodel the hinge and expose the ATP-binding site even after a key impediment to αC reorientation has been eliminated by disrupting the L856 hydrophobic network. The lethality of the Hyper mutations is likewise suppressed by the m2 mutation (29), which can now be explained by proposing that tRNA-dependent reorientation of αC is still necessary for full kinase activation even when hinge rigidity and obstruction of the ATP-binding pocket has been diminished by the Hyper substitutions. This inference is supported by the crystal structure of the R794G KD, in which the hinge is remodeled but αC remains in the inactive orientation, presumably because T882 is unphosphorylated (25).

The results presented here also provide strong evidence that adopting the PKR mode of parallel KD dimerization (6) is a regulated step in the Gcn2 activation pathway. Whereas inactive Gcn2 displays a distinct, antiparallel mode of dimerization (25), the functional importance of Gcn2 residues (R594-D598) corresponding to the crucial salt bridge at the PKR parallel interface argues that Gcn2 assumes the PKR mode of dimerization in the active state (11). Supporting this conclusion, Gcn2 is constitutively activated by substitutions (L650F and K850R) corresponding to the same substitutions in PKR that stimulate dimerization and kinase function in the absence of the extrinsic (double-stranded RNA binding) dimerization domain (10). The fact that these mutations activate Gcn2 in nonstarved cells suggests that stabilizing the parallel dimer interface normally occurs in WT Gcn2 upon amino acid starvation. Supporting this interpretation, the L650F K850R double mutation restores significant kinase activity to the m2 mutant, indicating that stabilization of the parallel dimer interface decreases the requirement for strong tRNA binding in Gcn2 activation.

The L650F and K850R substitutions at the predicted dimer interface also restore function to the T882A autophosphorylation mutant. One way to interpret this result is to propose that autophosphorylation promotes parallel dimerization, so that the requirement for T882 is diminished by mutations that strengthen the parallel dimer interface. Indeed, there is evidence that autophosphorylation is a prerequisite for PKR KD dimerization (10). A second, not mutually exclusive, explanation is that parallel dimerization and autophosphorylation both promote the same activating conformational transition in the KD, so that enhancing dimerization lessens the importance of autophosphorylation in achieving the active conformation. If T882 autophosphorylation promotes reorientation of αC, as argued above, then this second interpretation would imply that αC reorientation is also stimulated by parallel dimerization. In fact, this conclusion fits perfectly with our finding that L856A and I640A L647A, which facilitate αC reorientation, restore activity to the dimerization-defective mutant R594D lacking the critical salt bridge at the parallel dimer interface. Therefore, it seems very likely that autophosphorylation stabilizes reorientation of αC indirectly by enhancing parallel dimerization in addition to dissolving the inhibitory salt bridge E643-R834 by the charge neutralization mechanism mentioned above (Fig. 1B and C).

Our proposal that parallel dimerization facilitates αC reorientation is consonant with the fact that αC in its proper location constitutes a portion of the parallel dimer interface in active PKR (6). Thus, it is easy to imagine that parallel dimerization helps to push αC into the correct orientation. Indeed, a related mechanism is supported for EGFR, in which asymmetric KD dimerization achieves αC reorientation, wherein the C lobe of one protomer interacts with the N lobe of the second protomer at an interface containing αC in its inwardly rotated, active conformation (41). This is also analogous to a role played by cyclin in cyclin-dependent kinase, wherein inwardly rotated αC is present at the cyclin-KD interface (19).

The role of the hinge region in Gcn2 latency represents another layer of control not described for other kinases. The Hyper activating mutations, which alter the hinge conformation and expose the ATP-binding site (25), elicit a strong recovery of activity in the face of defects in tRNA binding (m2 substitutions), parallel dimerization (R594D), or autophosphorylation (T882A). These findings imply that, in WT Gcn2, tRNA binding, autophosphorylation, and parallel dimerization all combine to stabilize the favorable hinge conformation mimicked by the Hyper substitutions. We argued above that these events also promote the reorientation of αC. Although the R794G KD crystal structure displays an altered hinge conformation in the presence of the incorrect orientation of αC (25), it is possible that αC reorientation actually facilitates hinge remodeling. In fact, we envision that these various conformational transitions are mutually reinforcing to provoke a concerted switch from inactive to active conformation on tRNA binding and that the ensuing autophosphorylation of T882 locks in the fully activated state.

Integrating our current findings with previous results allows us to propose the hypothetical model for Gcn2 activation shown in Fig. 9. In nonstarved cells with low levels of uncharged tRNA, we propose that the Gcn2 KD exists in equilibrium between two inactive states. In state A, the KDs of the full-length Gcn2 dimer are monomeric and contain an improperly rotated αC and rigid hinge, incompatible with ATP binding or catalysis. (The KDs might undergo antiparallel dimerization in this state; however, this possibility is disfavored by our finding that destabilizing substitutions at the antiparallel dimer interface do not activate Gcn2.) State B contains a parallel KD dimer with properly oriented αC that still binds ATP weakly and is defective for catalysis, owing to hinge rigidity. Binding of uncharged tRNA reduces hinge rigidity via interaction with the HisRS domain, establishing a new equilibrium between the monomeric (state C) and parallel (state D) dimer forms, which both bind ATP more tightly and possess the requisite interlobe flexibility needed for catalysis. Stochastic transition from the monomeric (state C) to the parallel (state D) mode of dimerization elicits rapid autophosphorylation of T882 in state D to produce a highly stable, fully activated conformation (state E). According to this model, the Hyper substitutions shift the equilibrium from states A and B to C and D, even at the low-level tRNA occupancies of the HisRS domain in nonstarved cells. Given that the Hyper substitutions suppress the deleterious effects of alterations at the parallel dimer interface, they also appear to shift the equilibrium toward the parallel dimer in the conversion of state C to state D to enhance the formation of fully activated Gcn2 (state E). Disrupting the hydrophobic network by L856A strongly shifts the equilibrium from state A to state B in nonstarved cells, so that even the low occupancy of bound tRNA under nonstarvation conditions provokes substantial state B to D conversion and, with ensuing autophosphorylation, fully activated Gcn2 (state E). States A and C in our model correspond to the crystal structures of the inactive forms of the WT and R794G KDs determined previously (25). In the future, it would be interesting to determine whether the KDs of the L856A mutant and L856A R794G double mutant crystallize in states B and E, respectively, with parallel dimer interfaces and αC oriented in the active conformation, and (for state E) autophosphorylation of T882.