Activation of Protein Kinase PKR Requires Dimerization-induced cis-Phosphorylation within the Activation Loop*

Background: PKR regulates many biological processes including stress response. Results: An activation-loop-phosphorylation bypass mutant of PKR functions without a key dimer-interface residue. The PKR kinase domain together with a kinase-dead partner, but not alone, activates the catalytic function. Conclusion: Activation loop phosphorylation occurs in cis following dimerization. Significance: PKR autophosphorylates in the activation loop in cis to initiate the antiviral signaling cascade. Protein kinase R (PKR) functions in a plethora of cellular processes, including viral and cellular stress responses, by phosphorylating the translation initiation factor eIF2α. The minimum requirements for PKR function are homodimerization of its kinase and RNA-binding domains, and autophosphorylation at the residue Thr-446 in a flexible loop called the activation loop. We investigated the interdependence between dimerization and Thr-446 autophosphorylation using the yeast Saccharomyces cerevisiae model system. We showed that an engineered PKR that bypassed the need for Thr-446 autophosphorylation (PKRT446∼P-bypass mutant) could function without a key residue (Asp-266 or Tyr-323) that is essential for PKR dimerization, suggesting that dimerization precedes and stimulates activation loop autophosphorylation. We also showed that the PKRT446∼P-bypass mutant was able to phosphorylate eIF2α even without its RNA-binding domains. These two significant findings reveal that PKR dimerization and activation loop autophosphorylation are mutually exclusive yet interdependent processes. Also, we provide evidence that Thr-446 autophosphorylation during PKR activation occurs in a cis mechanism following dimerization.

The kinase domain (KD) 2 in most eukaryotic protein kinases remains in an inactive precursor state while associated with another domain in the same protein (intradomain) or with another partner protein (intersubunit) (1). Inactive-to-active transition of the KD requires binding of a specific modulator to the respective domain/or subunit or phosphorylation by an upstream kinase, or both (2)(3)(4)(5). The KD then adopts a core catalytic structure composed of a smaller N-terminal lobe (N-lobe) and the larger C-terminal lobe (C-lobe), creating an active site at the interface of these lobes (1). While adopting the core structure, an internal loop called the activation loop is often phosphorylated. The activation loop contains one or two regulatory phospho-acceptor residues that are phosphorylated by another kinase (trans-phosphorylation) or by the kinase per se (autophosphorylation). For example, the protein kinase PDK1 (phosphoinositide-dependent protein kinase 1) transphosphorylates the activation loop of PKA (cyclic AMP-dependent protein kinase A) (6), whereas DYRK1 (dual specificity tyrosine-regulated kinase 1) autophosphorylates its own activation loop (7). However, the mechanism of activation loop phosphorylation is still not clear in many kinases.
X-ray crystal structures have been resolved for the truncated PKR lacking the RBDs in apo form (i.e. K296R mutant) and in a complex with eIF2␣ (29). Several mechanistic details of PKR activation have been inferred from these structures although both are inactive under physiological conditions (20). In both structures, the KDs exist as dimers in which protomers associate with each other in a back-to-back orientation so that the catalytic sites face outward. We analyzed these structures and observed that the overall architecture of the apo-KD dimer was similar to the KD dimer complexed with eIF2␣ except at the region of the activation loop. This loop in the eIF2␣-bound form adopts an extended conformation with a phosphorylated residue Thr-446 (Fig. 1A), whereas it remains largely unstructured in the apo form (Fig. 1A). This structural difference indicates that a Thr-446 phosphorylation-dependent conformational transition must occur within the activation loop during PKR activation. Previously, we have shown that Thr-446 autophosphorylation is directly related to PKR activation and coupled with KD dimerization (20,30). Such coupled processes lead us to propose three different possible models for Thr-446 autophosphorylation (20). In the first model, the dimers phosphorylate activation loops on other dimers (trans-interdimer). Second, each protomer in a dimer phosphorylates its own activation loop (cis-intradimer). Third, one protomer in a dimer phosphorylates the activation loop of its partner (trans-intradimer). Several additional questions about the PKR activation mechanism remain unclear: what are the exact mechanisms and roles of Thr-446 autophosphorylation? How does dimerization promote Thr-446 autophosphorylation?
We showed that an engineered PKR that bypassed the need for Thr-446 autophosphorylation could function without a key residue (Asp-266 or Tyr-323) that is essentially required for the PKR dimerization; supporting the conclusion that dimerization precedes and stimulates activation loop autophosphorylation. Then, we showed that the Thr-446 phosphorylation occurs in a mechanism where each protomer within a dimer phosphorylates on its own (cis-intradimer).
Plasmids-A wild-type (WT) Flag-(His) 6 -tagged PKR was used (9). Mutations were introduced in the WT PKR gene by fusion PCR. WT PKR and its site-directed mutants were expressed from a galactose-inducible promoter built into the vector plasmid p1079 (pEMLyex4) or p2444 (pEMLyex4-based vector with a TRP selectable marker). The plasmids used in this report are listed in the supplemental Table S1.
Western Blot Analysis-Yeast cells were grown in a synthetic dextrose (SD) medium supplemented with appropriate nutrients until the A 600 reached 0.6 -0.7. Cells were collected and resuspended in 10% galactose (SGal) medium and allowed to grow for another 4 h at 30°C. Whole cell extracts were prepared and subjected to Western blot analysis using an appropriate antibody. Four antibodies were used in these studies: 1) a phosphospecific antibody against eIF2␣ phosphorylated on Ser-51 (gift from Dr. Thomas E. Dever, NICHD, NIH), 2) a Thr-446 phosphospecific antibody of PKR (Cell Signaling), 3) polyclonal antibody of total eIF2␣ (gift from Dr. Thomas E. Dever, NICHD, NIH), and 4) polyclonal antibody of the C-terminal end of PKR (Cell Signaling).
Protein Expression, Purification, and Kinase Assay-Flagtagged PKR and its derivatives were introduced into the S51A strain MY71. Yeast cells were then grown with appropriate nutrient in the presence of 10% galactose until the A 600 reached 4.0. Cells were harvested, and PKR protein was purified by ANTI-FLAG M2-agarose (Sigma) as described previously (9). The GST-eIF2␣ (1-180) and GST-eIF2␣-S51A (1-180) fusion constructs were introduced in Escherichia coli BL21 (DE3) plus cells (NOVAGEN). A culture of these cells (A 600 ϭ 0.5-0.8) was induced with 0.5 mM IPTG (isopropyl-␤-D-thio-galactoside) for 6 h at 37°C. Cells were collected, and the recombinant GST-eIF2␣ fusion protein was purified by glutathione-agarose resin (Roche), using the standard manufacturer's protocol. Phosphorylation of eIF2␣ by PKR was performed in a kinase buffer containing 20 mM Tris-HCl (pH 8.0), 50 mM KCl, 25 mM MgCl 2 , and 1 M PMSF (phenylmethylsulfonyl fluoride) with 10 Ci of [␥-33 P]ATP. Reactions were quenched after 20 min by addition of 2ϫ SDS gel loading buffer, and the reaction products were separated by SDS-PAGE. The gel was stained with Coomassie blue, and the dried gel was subjected to autoradiography.
Structure Analysis-We analyzed protein structure using computer software PyMol (54).

Functional Coupling between the Phosphorylated Activation
Loop and the Helix-␣C of PKR-To gain an insight into how PKR transitions from an inactive-to-active state, we analyzed two x-ray crystal structures of the PKR-KD: 1) one is complexed with eIF2␣ (PDB ID: 2A1A) (29) and 2) the other is in the inactive apo form (K296R mutant, PDB ID: 3U1U). In both structures, each kinase domain is folded into a bi-lobal structure, and two kinase domains orient in parallel to form a back-to-back dimer. We observed that these two KD dimers superimpose well on each other (root mean square deviation ϭ 1.54; supplemental Fig. S1), suggesting that the overall architecture of these proteins is similar. However, a prominent difference was observed in the region of activation loop (residues from Leu-435 to Lys-449) (Fig. 1A). In the apo-form, this loop remains largely unresolved and only three residues Val-Thr-Ser 436 -438 have been resolved as a ␤-strand. In the PKR-KD bound to eIF2␣, the activation loop adopts an extended conformation with the residue Thr-446 being phosphorylated. These structural differences suggest that, even though the activation loop is unresolved in PDB 3U1U, the inactive-to-active transition likely requires a movement of the activation loop, as observed in numerous other protein kinases (2) so that the substrate, like eIF2␣, can bind and be phosphorylated.
The structural analyses further revealed that two helix-␣C residues Lys-304 and Arg-307 co-ordinate with the phosphate moiety of Thr-446 in the activation loop and the spatial positions of the helix-␣C element remain essentially unchanged in both structures (Fig. 1A). This indicates that the inactive-toactive catalytic transition might not require re-positioning of the helix-␣C, unlike what has been observed in most active protein kinase domains (3). However, a coupling between the helix-␣C and the phosphorylated activation loop might be crucial. To determine whether these prominent structural interactions between the helix-␣C and the phosphorylated activation loop are likely to be physiologically relevant, we replaced residues Lys-304 with alanine and Arg-307 with alanine or lysine. The alanine substitution was expected to reduce the PKR function, whereas the more conservative lysine substitution might be able to retain the needed hydrogen bond.
We expressed the wild-type (WT) human PKR and its mutants K304A, R307A, and R307K in yeast Saccharomyces cerevisiae from a galactose-inducible promoter. WT PKR in yeast cells expressed poorly from a galactose-inducible promoter when grown on a synthetic dextrose (SD) medium (Fig.  1B). The poor PKR expression did not affect regular cell function and physiological processes, and a normal yeast growth phenotype was observed (Fig. 1B, SD). On a synthetic galactose medium (SGal), PKR expression was induced. This induction interfered with normal cell functions, resulting in a slowgrowth phenotype (Slg Ϫ ) in an eIF2B-desensitive strain (Fig.  1B, SGal, column 1) because PKR excessively phosphorylated eIF2␣, leading to partial inhibition of the eIF2B function (20). However, expression of a catalytically inactive PKR-K296R mutant showed no growth defect (column 2). Whole cell extracts from those cells were subjected to Western blot analyses, which showed that both WT and K296R mutant proteins were expressed under inducible conditions (Fig. 1C, PKR). Phosphorylations were observed on the residues PKR-T446 (Fig. 1C, PKRϳP lane) and eIF2␣-Ser51 (eIF2␣ϳP, lane 1) in cells expressing WT PKR, but not in cells expressing K296R mutant (PKRϳP, lane 2 and eIF2␣ϳP, lane 2). These data confirmed that the Slg Ϫ phenotype was correlated with eIF2␣ phosphorylation and that the K296R substitution blocks both PKR and eIF2␣ phosphorylation, restoring rapid yeast growth (see Refs. 8 -10 for other PKR assays in yeast where PKR induction causes a lethal phenotype).

Mechanism of Activation Loop Phosphorylation of PKR
phosphorylated Thr-446 are not critical for PKR activity. Collectively, these structure-based mutational analyses reveal that a productive interaction between the Thr-446-phosphorylated activation loop and the helix-␣C is important for the PKR catalytic function.
Thr-446 Phosphorylation Occurs in a cis Mechanism following Dimerization-To determine the mechanisms of Thr-446 autophosphorylation during the KD dimerization, we used a LIM-Ldb dimerization system (31). In this system, only the kinase domain of PKR was expressed in yeast cells as a monomer or forced to form a homodimer. The KD (residues 258 to 551) was fused to the Xenopus protein LIM domain (residues 1 to 58) and to LIM-domain binding protein Ldb (residues 290 to 350) creating LIM-PKR KD and Ldb-PKR KD fusion proteins, respectively. Yeast cells expressing either LIM-PKR KD (Fig. 2, yeast growth, row 2) or Ldb-PKR KD (row 3) alone, or neither one (row 1) showed no growth defect on SD and SGal media. However, expression of LIM-and Ldb-PKR KD fusion proteins together was lethal on SGal medium (Fig. 2, yeast growth, SGal, row 4). The lethality was suppressed in an isogenic eIF2␣-S51A strain ( Fig. 2A, right panel), supporting the conclusion that the lethality was due to activation of the PKR that phosphorylated eIF2␣ (31).
The activation loop phosphorylation is a molecular marker of the active state of a protein kinase domain, including PKR (2,20). Thus, we tested the phosphorylation status of the activation loops in LIM-and Ldb-PKR KD fusion proteins. Whole cell extracts from those yeast cells were subjected to Western blot analysis using a Thr-446-phosphospecific antibody and a polyclonal antibody to total PKR (Fig. 2B). The Ldb-PKR KD fusion protein expressed better than the LIM-PKR KD fusion protein (Fig. 2B, lower panel, compare lanes 2 and 3), as a monomer as depicted in Fig. 2C. The residue Thr-446 in either fusion protein was un-phosphorylated (Fig. 2B, lanes 2 and 3; compare with total PKR at the lower panel), suggesting that Thr-446 autophosphorylation does not occur spontaneously. In contrast, Thr-446 was phosphorylated in both LIM-and Ldb-PKR KD fusion proteins was phosphorylated when two proteins FIGURE 2. Thr-446 phosphorylation occurs in a cis mechanism following dimerization. A, heterologous dimerization domains activate the PKR KD. The LIM-PKR KD fusion gene in the plasmid pC901 with a URA3 selectable marker and the Ldb-PKR KD fusion gene in the plasmid pC903 with a TRP selectable marker were introduced into WT yeast strain H2557 as well as S51A strain. The sign (ϩ) means the presence, whereas the sign (Ϫ) means the absence of pC901/pC903 plasmids but contains the respective vector plasmid. The indicated D414A mutation was introduced in to the LIM-PKR KD fusion construct. Transformants were tested for growth on SD and SGal media. The PKR was inactive when the yeast grew on SGal medium, but active if the yeast did not grow. B, analysis of Thr-446 autophosphorylation and eIF2␣ phosphorylation. Yeast strains expressing indicated LIM-PKR KD (WT or D414A) and Ldb-PKR KD fusion protein were grown in the SGal medium and whole cell extracts were subjected to Western blot analysis using either a Thr-446-phosphospecific (T446ϳP) or Ser-51 phosphospecific (eIF2␣ϳP) antibody followed by a polyclonal antibody of PKR or eIF2␣. C, models represent mono-or dimeric form of LIM-and Ldb-PKR KD fusion proteins. The bi-lobal PKR-KD is shown in gray whereas the activation loops in purple solid lines with a phosphorylated residue (red circle). The inactive form of the kinase domain is colored orange.
were expressed together in a same cell (Fig. 2B, lane 4 and in the right panel, lane 2), suggesting that dimerization resulting from LIM⅐Ldb interaction (Fig. 2C) resulted in Thr-446 phosphorylation.
We propose three models that might explain coupling of PKR dimerization and Thr-446 phosphorylation: (I) trans-intradimer, (II) cis-intradimer, or (III) trans-interdimer. To test the first two models, we introduced a kinase-inactivating mutation in the LIM-PKR KD (D414A mutation in the conserved HRD 414 motif, data not shown) and expressed it alone or co-expressed with the Ldb-PKR KD . Expression of LIM-PKR KD -D414A alone did not affect yeast cell growth on SD and SGal media ( Fig. 2A, yeast growth, SD, and SGal, row 5), whereas co-expression with the Ldb-PKR KD fusion protein exhibited an Slg Ϫ phenotype on SGal medium ( Fig. 2A, rows 6), but not in the isogenic eIF2␣-S51A strain (phosphorylation site Ser-51 mutated to an Ala, Fig. 2A, right panel), suggesting that the reduced growth was due to eIF2␣ phosphorylation. These data suggested that the LIM-PKR KD lanes 1 and 3). Consistent with the Thr-446 autophosphorylation, eIF2␣ was phosphorylated in yeast cells expressing LIM-PKR KD together with Ldb-PKR KD or Ldb-PKR KD -D414A (Fig. 2B, Western blots,  right panel, eIF2␣ϳP, lanes 2 and 3). These results suggest that phosphorylation on residue Thr-446 in Ldb-PKR KD protein occurs in a cis mechanism following dimerization (see Fig. 2C).
Thr-446 Phosphorylation Does Not Occur in a trans Mechanism-To determine whether or not Thr-446 phosphorylation occurs in a trans-interdimer mechanism, a homodimeric GST-PKR KD fusion protein (a functional homodimer, see Ref. 31) was mixed with a kinase-dead PKR-K296R protein in a reaction buffer (Fig. 3A, Anti-PKR, 1ϫ). Then, the phosphorylation status of the activation loops of both proteins was examined by Western blot analysis using a Thr-446 phosphospecific antibody of PKR (Fig. 3A, Anti-T446ϳP). The same experiment was repeated with 5-fold higher protein concentration to examine any positive effect on Thr-446 phosphorylation (Fig. 3A,  Anti-PKR, 5ϫ). The residue Thr-446 was phosphorylated in the GST-PKR KD fusion protein in the presence or absence of PKR-K296R protein (Fig. 3A, GST-PKR KD ϳP, lanes 1 and 3; and also in 5ϫ, lanes 4 and 6); however, the residue Thr-446 was unphosphorylated in the PKR-K296R protein (lanes 2, and also in 5ϫ lane 5) even when it was mixed with GST-PKR KD fusion protein (lane 3 and in 5ϫ lane 6); arguing against the trans mechanism of Thr-446 phosphorylation.
To confirm these results under a more physiological cellular environment, we co-expressed the GST-PKR KD fusion protein and WT PKR or PKR-K296R mutant in yeast. Whole cell extracts were subjected to Western blot analysis using a Thr-446 phosphospecific antibody of PKR followed by an antibody against PKR. The residue Thr-446 in WT PKR (Fig. 3B, Anti-T446ϳP, lane 1) or GST-PKR KD fusion protein (lanes 1 and 2) was phosphorylated. However, the residue Thr-446 in the PKR-K296R protein (Fig. 3B, lane 2) was not phosphorylated. Expressions of both of these proteins (Fig. 3B, Anti-PKR) were nearly equal, indicating that the trans phosphorylation of the Thr-446 residue is not a likely mechanism.
A PKR phk1 Chimera Bypasses the Requirement for Activation Loop Phosphorylation-PKR activation requires both dimerization and Thr-446 autophosphorylation. To further understand whether these processes are dependent, independent, and interdependent, we engineered the PKR gene so that it could bypass the requirement for activation loop phosphorylation. Initially, two PKR mutants (PKR-T446D and PKR-T446E) were created with the expectation that aspartate/glutamate could function as a phospho-mimetic substitution. As shown in Fig.  4A, yeast cells expressing the PKR-T446D mutant (column 4), like PKR-K296R (column 2), and PKR-T446A mutants (column 3), grew well on SGal medium. Western blot analysis showed that phosphorylation of eIF2␣ was substantially reduced in those cells (Fig. 4B, eIF2␣ϳP, lanes 2, 3, and 4), although mutant PKR proteins expressed comparatively better than WT (Fig. 4B, Western blots, compare lanes 1-4, PKR). These results indicated that the ability of PKR-T446D or PKR-T446E (data not shown) mutant to phosphorylate eIF2␣ is substantially lower than WT PKR. Collectively, we conclude that phosphomimetic aspartate or glutamate does not bypass the requirement for activation loop phosphorylation. The reaction products were then separated by SDS-PAGE and subjected to Western blot analyses using phosphospecific antibody of Thr-446 followed by a polyclonal antibody of PKR. B, analysis of Thr-446 phosphorylation. GST-PKR KD co-expressed with WT PKR or PKR-K296R mutant in yeast. Whole cell extracts were prepared from these cells and subjected to Western blot analysis using a Thr-446 phosphospecific antibody of PKR (T446ϳP). The membrane was stripped and re-probed with a polyclonal antibody of PKR. FEBRUARY 28, 2014 • VOLUME 289 • NUMBER 9

Mechanism of Activation Loop Phosphorylation of PKR
Then we attempted to replace the activation loop by a loop that would mimic the phosphorylation state. Thus, we performed a global search of kinase domain sequences and structures and identified kinases Phk1 (phosphorylase kinase 1) (32) and Chk1 (checkpoint kinase 1) (33) in which the activation loop elements are devoid of any regulatory phospho-acceptor residues (Fig. 5A). Moreover, we observed that the structure of PKR kinase domain (PDB ID: 2A1A) superimposed well onto the structure of Phk1 kinase domain (PDB ID: 2PHK, Fig. 5B) as well as onto the structure of Chk1 kinase domain (PDB ID: 1IA8), and that the phosphorylated residue Thr-446 in PKR aligned with the phospho-mimetic residue E182 in Phk1, whereas the helix-␣C residue R307 in PKR aligned with K72 (Lys-72) in Phk1 (Fig. 5B) or K54 (Lys-54) in Chk1 (not shown). These structural analyses indicated that activation loop residues of Phk1 or Chk1 might functionally substitute the corresponding residues of PKR.
We replaced the activation loop residues of PKR by the respective residues from Phk1 or Chk1, generating PKR phk1 and PKR chk1 chimera, and expressed these chimeras in yeast. Yeast cell expressing PKR phk1 , but not PKR chk1 , grew slowly like WT PKR on SGal medium (Fig. 4A, compare columns 1, 5, and 6). Western blot analysis showed that the slow-growth correlated with eIF2␣ phosphorylation (Fig. 4B, Western blots, eIF2␣ϳP, lanes 5 and 6). These results indicated that activation loop residues of Phk1, but not of Chk1, functionally substituted the activation loop residues of PKR. To confirm the in vivo results, we performed an in vitro kinase assay to test the ability of PKR phk1 to phosphorylate eIF2␣. Purified PKR phk1 protein was mixed with the recombinant eIF2␣ and [␥-33 P]ATP in a kinase reaction buffer. In parallel kinase assays that were performed under the same conditions, we used WT PKR and K296R protein (Fig. 4C, stain). As shown in Fig. 4C, WT PKR (lane 1) and the PKR phk1 chimera (lane 4) both autophosphorylated (PKRϳP) and phosphorylated the substrate eIF2␣ on the residue Ser-51 (eIF2␣ϳP), whereas the PKR-K296R protein neither autophosphorylated nor phosphorylated eIF2␣ (lane 3), confirming that the PKR Phk1 chimera is a functional eIF2␣ kinase.
To address the next obvious question of whether or not the PKR phk1 chimera required an active interaction between the activation loop and the helix-␣C, in particular the interaction between the residues Glu-182 and Arg-307 (Fig. 5), we substituted each residue in PKR phk1 by alanine and expressed mutants PKR phk1 E182A and PKR phk1 R307A in yeast. Yeast cells expressing PKR phk1 E182A (Fig. 4D, column 2) or

. Functional substitution of activation loop residues of PKR by residues from the activation loop of phosphorylase kinase 1 (Phk1). A, in vivo
analysis of PKR mutants by growth in yeast. The yeast strain H17 (eIF2B desensitive) or MY71 (eIF2␣-S51A) harboring indicated WT PKR and its derivatives were serially diluted and spotted on SD and SGal media. B, in vivo analysis of eIF2␣ phosphorylation by PKR mutants. Whole cell extracts were prepared from yeast cells indicated in A and subjected to Western blot analysis using phosphospecific antibodies against Ser-51 (eIF2␣ϳP). The membrane was stripped and re-probed with a polyclonal antibody of eIF2␣ and PKR. C, in vitro analysis of eIF2␣ phosphorylation by PKR mutants. Purified PKR protein (WT, K296R, or PKR phk1 ) was mixed with the recombinant eIF2␣ and [␥-33 P]ATP in a reaction buffer for 10 min. The reaction products were then separated using SDS-PAGE. The gel was stained, dried, and subjected to autoradiography to monitor the incorporation of 33 P in eIF2␣ proteins. D, catalytic function of PKR phk1 chimera requires an active interaction between the activation loop and the helix-␣C. Left panels, eIF2B desensitive strain H17 expressing indicated PKR mutants were tested for growth on SD and SGal media. Right panels, yeast cells were grown in the presence of galactose (10%) and harvested after 2 h. Whole cell extracts were then prepared and subjected to Western blot analyses using antibodies of phosphorylated eIF2␣ (eIF2␣ϳP), eIF2␣, and PKR. PKR phk1 R307A (column 4) grew rapidly like the PKR-R307A mutant (column 3) on SGal medium. Western blot analyses showed that rapid growth correlated with reduced eIF2␣ phosphorylation (Fig. 4D, Western blots, compare lanes 1-4) though expression of these PKR derivatives was better than WT (Fig. 4D,  Western blot, PKR, lanes 1-4). These results suggest that, in the PKR Phk1 chimera, the positively charged residue Arg-307 in helix-␣C and the phospho-mimetic glutamate in the activation loop (Glu-182 corresponding to Thr-446) play important roles in the PKR catalytic function. The residue Phk1-E182 aligns with the phosphorylated Thr-446 of PKR (Fig. 5); thus it appears that the residue Glu-182 in the PKR phk1 activation loop maintains the salt bridge interaction with the helix-␣C.
Dimerization Precedes and Stimulates Activation Loop Phosphorylation-Though PKR has been reported to exist as a monomer (17,34), numerous studies have challenged whether PKR functions as a monomeric protein. These studies include: 1) PKR forms a dimer at higher concentration (34) and 2) PKR exists as a monomer-dimer equilibrium that is enhanced by autophosphorylation (35). Furthermore, the crystal structure reveals that the PKR-KD forms a dimer by a back-to-back interaction between two N-lobes where several intra-and inter-molecular salt-bridges and hydrogen bonds stabilize the interaction. The prominent interactions include a salt-bridge between residues Arg-262 and Asp-266 and a hydrogen bond network between residues Asp-289, Tyr-293, and Tyr-323 (Fig. 6A). We  shown. Several intra-and inter-molecular hydrophobic and salt-bridge interactions stabilize the dimer interface between protomer 1 (gray) and protomer 2 (green). Only two prominent salt-bridge interactions (R262⅐D266 and D289⅐Y293⅐Y323) are zoomed in a separate box. B, in vivo analysis of PKR mutants by growth in yeast and by eIF2␣ phosphorylation. The gcn2⌬ yeast strains harboring indicated PKR mutants were tested for growth on SD and SGal media (upper panels). Whole cell extracts from these cells were subjected to Western blot analyses using antibodies of phosphorylated eIF2␣ (eIF2␣ϳP) and eIF2␣.
have shown previously that mutations of residues Arg-262 and Asp-266 reduce the ability of PKR both to autophosphorylate Thr-446 and to phosphorylate eIF2␣, whereas reciprocal exchange of these two residues restores both functions (30). We conclude that the KD dimerization is coupled with Thr-446 phosphorylation in the activation loop. But, it is still unclear how these processes are inter-related.
We used the PKR phk1 chimera to understand the importance of dimerization on Thr-446 autophosphorylation. We hypothesized that PKR phk1 should be non-functional or partially functional if the RBD-mediated KD dimerization was required for Thr-446 autophosphorylation. To test this, residues Asp-266 and Tyr-323 in the PKR phk1 chimera were individually substituted to Arg and Ala, generating PKR phk1 D266R and PKR phk1 Y323A mutants, respectively. We then expressed these PKR derivatives in yeast, tested for growth on SGal medium, and examined eIF2␣ phosphorylation. The results were then compared with cells expressing WT PKR and PKR mutant K296R, D266R, or Y323A (Fig. 6B).
The Kinase Domain of PKR phk1 Functions without RNA Binding Domains-Next, to test the hypothesis that the kinase domain of PKR phk1 should be functional without its N-terminal RNA binding domains, we expressed only the kinase domain of the PKR phk1 in yeast, tested cell growth on SGal medium and examined eIF2␣ phosphorylation in those cells. Expression of LIM-PKR phk1 -KD fusion (Fig. 7, eIF2␣-WT, lane 4), like PKR phk1 (lane 2), substantially reduced yeast growth on SGal medium. However, no growth defect was observed in an isogenic S51A strain (Fig. 7, eIF2␣-S51A), suggesting that growth reduction was due to eIF2␣ phosphorylation. As expected, expression of the PKR-KD only like a vector control (lane 1) showed no growth defect in both WT and S51A yeast strains (Fig.  7, lanes 1 and 4, see also Fig. 2A, column 2). As expected, Western blot analysis showed that PKR phk1 phosphorylated eIF2␣ (Fig. 7B, lane 2) and no phosphorylation of eIF2␣ was observed in yeast expressing PKR-KD (lane 1). Interestingly, we observed that the PKR phk1 KD phosphorylated eIF2␣ (Fig. 7B, lane 3), suggesting that PKR phk1 KD bypassed the requirement of absolute requirement for KD dimerization for its catalytic activity.
To test further that the PKR phk1 KD actively phosphorylated eIF2␣, yeast cell expressing PKR-KD or PKR phk1 KD was transformed with a GCN4 LacZ reporter plasmid p180 (36). Transformants were grown till A 600 reached ϳ0.6. Then, cells were harvested, resuspended on a medium containing 10% galactose, FIGURE 7. The kinase domain of PKR phk1 chimera is partially active. A, in vivo analysis of PKR mutants by growth in yeast. The WT (eIF2␣-WT) and S51A (eIF2␣-S51A) yeast strains harboring indicated PKR mutants were serially diluted and tested for growth on SGal medium. B, PKR KD -Phk1 phosphorylates eIF2␣. Whole cell extracts from cells expressing indicated PKR derivatives were subjected to Western blot analyses using antibodies of phosphorylated eIF2␣ (eIF2␣ϳP) and eIF2␣. C, PKR KD -Phk1 can activate the GCN4 translational control. Whole cell extracts of yeast cells carrying a GCN4 LacZ reporter plasmid p180 were prepared and ␤-gal activities were monitored as described previously (36). and grown for another 6 h. Whole cell extracts were prepared, and ␤-gal activity was monitored (Fig. 7C). The LacZ expression was not induced in cells expressing PKR-KD, but was significantly induced in PKR phk1 KD cell (5-fold), confirming that the PKR phk1 KD is an active kinase and can activate the GCN4 translational control in yeast. Collectively, our data suggest that dimerization and KD activation are mutually exclusive yet interdependent events.

DISCUSSION
The mechanism of Thr-446 phosphorylation in the activation loop has been determined by expressing the PKR KD in yeast cells using a heterologous dimerization system. We showed that the PKR KD fused to LIM or Ldb is inactive when either one is present alone. When these fusion proteins are present together in the same cell, the PKR KD becomes functional because the LIM and Ldb domains form a heterodimer that holds two kinase domains together. Separation of LIMand Ldb-PKR KD fusion proteins showed that each one was phosphorylated on the residue Thr-446, supporting that dimerization is required for Thr-446 autophosphorylation. Then, we showed that a kinase-dead isoform of LIM-PKR KD was able to activate the catalytic function of the partner Ldb-PKR KD fusion protein. Consistently, we showed that the residue Thr-446 in the Ldb-PKR KD , but not the kinase-dead isoform of LIM-PKR KD in the heterodimer, was phosphorylated. These results suggest that Thr-446 autophosphorylation does not occur via the trans-intradimer model, and must have occurred in mechanisms where 1) each protomer within a dimer phosphorylates on its own (cis-intradimer) and/or 2) an active partner in a dimer phosphorylates a monomeric protein or a protomer of the dimeric protein (trans-interdimer). To determine the trans-interdimer model of Thr-446 autophosphorylation, a homodimeric form of PKR (GST-PKR KD ) was mixed with a kinase-dead PKR-K296R mutant. Absence of any detectable Thr-446 phosphorylation in the PKR-K296R protein (Fig. 3) suggests that Thr-446 phosphorylation in trans is not a likely mechanism.
In our earlier report (20), we ruled out that the cis-intradimer Thr-446 autophosphorylation is not a possible mechanism because the in vitro PKR autophosphorylation reaction shows second order kinetics with respect to PKR concentration (35,37). The autophosphorylation reaction occurs at multiple residues other than Thr-446, including residues Ser-340, Ser-343, Ser-344, Ser-351, Ser-354, Ser-355, Ser-357, Thr-359, Ser-242, Thr-255, and Thr-258. Alanine mutation of these residues has no or minimum effect on the catalytic function of PKR (38), suggesting that those sites are not physiologically relevant phosphorylation sites. Therefore, we conclude that the in vitro PKR autophosphorylation reaction may not explain adequately the mechanism of Thr-446 phosphorylation.
Dar et al. reported that the PKR kinase domains form a backto-back dimer (29). If this geometry needs to be maintained in an active PKR protein, the mechanism of trans-intradimer Thr-446 autophosphorylation is not a likely mechanism because a protomer in a dimer would not access the active site of the partner. Supporting this hypothesis, it was shown that the residue Thr-446 of LIM-PKR KD -D414A remained un-phosphorylated in an Ldb⅐LIM-PKR KD heterodimer (Fig. 2). However, the heterodimer was able to activate the catalytic function of its partner Ldb-PKR KD fusion protein while inducing Thr-446 phosphorylation (Fig. 2), fitting perfectly with the possibility  (40). Consequently, the kinase domains of PKR adopt a favorable dimeric configuration (even with an inactive KD partner shown in gray) that facilitates autophosphorylation on the residue Thr-446 in cis. As shown by dotted lines, the phosphorylated Thr-446 couples with the helix-␣C (orange cylinder) and an active closed conformation is achieved. Then PKR phosphorylates other residues in trans (shown in red circles) and becomes fully active.
that Thr-446 phosphorylation occurs in a cis mechanism following dimerization.
Although the structure of the PKR holoenzyme is still lacking, it has been shown that the isolated RBDs forms a dumbbell shaped monomer in solution (39). It has also been shown that the RDB2 interacts with the C-lobe of PKR-KD (40). Together with these results, we propose that RBD2 remains bound to the inactive kinase domain until an effector molecule (dsRNA, PACT, or RAX) binds and stimulates its release (Fig. 8). The allosteric interaction between the RBDs and the dsRNA, PACT, or RAX favors a conformation that triggers Thr-446 autophosphorylation in a cis mechanism. Then, PKR autophosphorylates in trans at multiple residues as reported earlier (38) and becomes fully active. The active PKR induces antiviral defense responses and functions in regulatory pathways of a plethora of biological processes, including myogenesis (41), cell cycle progression (42), and inflammation (43).
PKR belongs to a family of eIF2␣ kinases that also contain GCN2, PERK, and HRI (44). Each of these kinases carries distinct regulatory domains that sense a specific cellular stress. HRI contains heme-binding domains that sense the intracellular heme deprivation (45), PERK contains a lumenal domain that senses the accumulation of unfolded protein in the endoplasmic reticulum (46), and GCN2 contains a HisRS-like (histidyl-tRNA synthetase) domain that senses the metabolite stress (47). Upon sensing stress through the sensor domain, each one activates its kinase domain and phosphorylates eIF2␣ to down-regulate protein synthesis. Each of these kinases functions as a dimer, and is regulated by autophosphorylation at a conserved threonine (corresponding to Thr-446), located in the activation loop. The kinase domains of PERK, similar to PKR, have been shown to form a dimer in a back-to-back orientation (48). Also, it has been shown that the phosphorylated activation loop in PERK kinase domain salt-bridges with the helix-␣C. Thus, we propose that the activation of PERK-KD might resemble PKR-KD activation.
Approximately 10% of protein kinase domains encoded by the human genome lack either an N-lobe, a C-lobe, or one/two canonical kinase motifs (e.g. DFG and RD). They are unable to phosphorylate a substrate and are called pseudokinase domains (49). These pseudokinases positively or negatively regulate the functions of the partner domain (50). For example, a pseudokinase domain of human RNase L positively regulates its cognate RNase domain function (51), whereas a pseudokinase domain JH2 negatively regulates the tyrosine kinase activity of JAK2 (52). We showed that the kinase-dead PKR-D414A isoform (Fig. 2) activated the catalytic function of its partner, reminiscent of the functional interaction between the pseudokinase KSR1 (kinase suppressor of Ras) with Raf, MEK, or ERK (53). Thus, the PKR pseudokinase provides a molecular explanation of a pseudokinase domain function.