Baculovirus protein PK2 subverts eIF2α kinase function by mimicry of its kinase domain C-lobe

Significance

Many pathogens use molecular mimicry to subvert key cellular processes of the host. RNA-dependent protein kinase (PKR), a member of the eukaryotic translation initiation factor 2α (eIF2α) kinase family, is an important component of innate immunity in vertebrates and has often been subjected to inhibition by viral mimicry. In this study we show that the paradigm of host–virus mimicry extends to invertebrates where there is no discernable PKR homologue. We characterize an eIF2α kinase-mimic protein called “PK2,” encoded by baculoviruses, that inhibits a heme-regulated inhibitor kinase (HRI)-like eIF2α kinase, possibly through a lobe-swap mechanism. The inhibition of the HRI-like kinase confers a growth advantage to the baculovirus during infection of its insect host. These experiments suggest the independent emergence of eIF2α kinase antiviral defense mechanisms in vertebrates and invertebrates.

Abstract

Phosphorylation of eukaryotic translation initiation factor 2α (eIF2α) by eIF2α family kinases is a conserved mechanism to limit protein synthesis under specific stress conditions. The baculovirus-encoded protein PK2 inhibits eIF2α family kinases in vivo, thereby increasing viral fitness. However, the precise mechanism by which PK2 inhibits eIF2α kinase function remains an enigma. Here, we probed the mechanism by which PK2 inhibits the model eIF2α kinase human RNA-dependent protein kinase (PKR) as well as native insect eIF2α kinases. Although PK2 structurally mimics the C-lobe of a protein kinase domain and possesses the required docking infrastructure to bind eIF2α, we show that PK2 directly binds the kinase domain of PKR (PKR^KD) but not eIF2α. The PKR^KD–PK2 interaction requires a 22-residue N-terminal extension preceding the globular PK2 body that we term the “eIF2α kinase C-lobe mimic” (EKCM) domain. The functional insufficiency of the N-terminal extension of PK2 implicates a role for the adjacent EKCM domain in binding and inhibiting PKR. Using a genetic screen in yeast, we isolated PK2-activating mutations that cluster to a surface of the EKCM domain that in bona fide protein kinases forms the catalytic cleft through sandwiching interactions with a kinase N-lobe. Interaction assays revealed that PK2 associates with the N- but not the C-lobe of PKR^KD. We propose an inhibitory model whereby PK2 engages the N-lobe of an eIF2α kinase domain to create a nonfunctional pseudokinase domain complex, possibly through a lobe-swapping mechanism. Finally, we show that PK2 enhances baculovirus fitness in insect hosts by targeting the endogenous insect heme-regulated inhibitor (HRI)–like eIF2α kinase.

Eukaryotic cells possess diverse mechanisms for molecular adaptation to stress conditions. Eukaryotic translation initiation factor 2α (eIF2α) kinases are evolutionarily conserved enzymes that detect specific stress signals and induce changes in translation to produce an adaptive response. The four eIF2α kinases in humans possess divergent regulatory domains that enable response to different stress signals: (i) the RNA-dependent protein kinase (PKR) senses viral infection to mediate antiviral immunity; (ii) the PKR-like endoplasmic reticulum kinase (PERK) detects unfolded proteins in the endoplasmic reticulum to regulate the unfolded protein response; (iii) the heme-regulated inhibitor kinase (HRI) senses free heme to coordinate globin synthesis in red blood cells; and (iv) the general control nonrepressible-2 kinase (GCN2) detects amino acid insufficiencies to regulate metabolite homeostasis (1). Although eIF2α kinases respond to different stress stimuli, their protein kinase domains are highly similar and allow transmission of an overlapping signal that potently inhibits cellular translation. Upon stress-induced activation, eIF2α kinases phosphorylate a common cellular substrate, eIF2α, at a conserved site corresponding to Ser51 in human eIF2α. eIF2 is a heterotrimeric (α,β,γ subunits) GTPase that delivers initiator methionyl-tRNA to the 40S ribosome during translation initiation. Following delivery, hydrolysis of GTP results in release of an inactive eIF2•GDP complex. Ser51 phosphorylation converts eIF2 from a substrate into a competitive inhibitor of its GTP-exchange factor, eIF2B, that normally recycles eIF2 back into the active GTP-bound form (2). Because eIF2B protein levels are limiting compared with eIF2 in vivo, phosphorylation of a fraction of the cellular pool of eIF2α results in the effective sequestration of eIF2B and the potent inhibition of protein synthesis (2).

The inhibition of cellular translation allows cells to reduce energy expenditure when mounting a proper signaling response. Moreover, translation of a special class of mRNAs containing upstream ORFs is enhanced following eIF2α phosphorylation, resulting in increased synthesis of the encoded stress factors (3, 4). The archetypal eIF2α kinase PKR is a key mediator of innate immunity that attenuates translation in response to viral infection. PKR has a modular domain architecture of two dsRNA-binding regulatory domains preceding a C-terminal Ser/Thr kinase domain (1). When the N-terminal dsRNA-binding domains engage dsRNA molecules, including perhaps products of viral RNA transcription or replication, PKR dimerizes and autophosphorylates on a key regulatory site, Thr446, in the kinase activation segment. eIF2α kinases share sequence similarity only within their catalytic domains, and the minimal kinase domain of PKR (PKR^KD) is sufficient to propagate a translational repression response in vivo when fused to regulated or constitutive dimerization domains (5, 6) or when mutated to promote dimerization (7). PKR^KD has a prototypical architecture consisting of a smaller N-terminal lobe (N-lobe) and a larger C-terminal lobe (C-lobe). The crystal structure of phosphorylated PKR^KD in complex with eIF2α revealed two key aspects of eIF2α kinase function (8). First, a conserved surface on the N-lobe of the PKR^KD mediated dimerization in a back-to-back configuration required for catalytic activity. Second, the conserved helix αG on the C-lobe of the PKR^KD mediated eIF2α substrate recognition by adopting a noncanonical conformation (one turn longer and tilted by 40°) unique to eIF2α kinases (8). The conservation of both the dimerization interface and the unique αG helix conformation among all eIF2α kinases reflects a conservation of kinase regulatory and substrate targeting mechanisms (7).

The importance of PKR’s ability to down-regulate translation during viral infection is underscored by the great array of viruses that successfully target PKR for inhibition (9). Elucidating the mechanism by which PKR function is subverted by viral products has enhanced our understanding of how PKR and other eIF2α family kinases naturally function. In particular, viral mimicry appeared as a recurring theme: analysis of the pseudosubstrate inhibitor K3L from vaccinia virus revealed key molecular determinants required for PKR recognition of its natural substrate eIF2α (10–12); analysis of the RNA binding protein E3L from vaccinia virus revealed how PKR itself senses dsRNA through comparable infrastructure (13–15); and analysis of the RNA inhibitor VAI from adenovirus showed how a structured RNA decoy can competitively engage the dsRNA-binding domains of PKR in a manner that circumvents kinase domain activation (16–18).

PK2, which is encoded by insect baculoviruses, is a less well-characterized modulator of eIF2α kinase function (19, 20). When PKR was overexpressed in baculovirus-infected caterpillar moth cell lines, viral PK2 reduced eIF2α phosphorylation and stimulated virus replication (20). In addition, when PKR was expressed in yeast, PK2 alleviated PKR-mediated translational arrest and toxicity by preventing phosphorylation of yeast eIF2α. Comparative analysis of available sequences showed that PK2 is similar to rabbit reticulocyte HRI and Saccharomyces cerevisiae GCN2 kinase C-lobes (19, 20), suggesting that PK2 will mimic the function of the C-lobe of eIF2α kinases. Because subsequent structural studies (7, 8, 21) revealed that dimerization-dependent activation of eIF2α kinases is the exclusive function of the kinase N-lobe, whereas substrate recognition is mediated by helix αG in the kinase C-lobe, one would reasonably predict that PK2 might function by competitively sequestering eIF2α. Surprisingly, however, PK2 coimmunoprecipitated with PKR when coexpressed in yeast and also inhibited PKR^KD autophosphorylation (20). Further suggesting that PKR, and not eIF2α, was the interaction target of PK2, a yeast slow-growth phenotype caused by PK2 overexpression was rescued by overexpression of a catalytically inactive PKR^K296R but not by eIF2α (20).

Although PK2 can inhibit eIF2α kinase signaling in vivo, little is known about the molecular basis of its precise mechanism of action. In particular, our current understanding of eIF2α kinase structure and function does not provide a framework for understanding how a mimic of a protein kinase domain C-lobe can exert an inhibitory effect on eIF2α kinase function. In this study, we used a combination of biochemical, evolutionary genetic, biophysical, and virology approaches in vitro and in vivo to gain insight into how PK2 functions. Our results suggest that PK2 evolved from an insect HRI-like kinase and functions by engaging the N-lobe of the HRI-like insect eIF2α kinase domain to promote virus replication.

Results

PK2 Most Closely Resembles the Kinase C-Lobe of an Insect HRI but Also Possesses a Unique N-Terminal Extension.

The genome of the baculovirus Autographica californica multiple nucleopolyhedrovirus (AcMNPV) contains an ORF (ORF123) encoding a 25-kDa (215 amino acids) protein, PK2, that is conserved in other alphabaculoviruses including Bombyx mori NPV (BmNPV), Rachiplusia ou NPV (RoMNPV), and Plutella xylostella NPV (PxNPV). As assessed by BLAST analysis, the C-terminal region of PK2, encompassing residues 23–215, which we term the “eIF2α kinase C-lobe mimic” (EKCM) domain, displayed greatest sequence similarity to the C-lobe of eIF2α protein kinase domains. Top hits corresponded to the HRI orthologs from Danaus plexippus (Monarch butterfly) and B. mori (silkworm), possessing Poisson probabilities of 2e-24 and 3e-23 and percent identity of 37% and 36%, respectively. BLAST analysis against human protein kinases revealed greatest similarity to HRI (36%) followed by other eIF2α kinase family members (PKR, 25%; GCN2, 29%; PERK, 27%) and then non-eIF2α kinases (AuroraA, 23%; CDK2, 21%; PKA, 21%) (Fig. 1A). These and prior observations (19, 20) support the notion that PK2 originally evolved from an HRI ortholog of an insect species, perhaps through horizontal gene transfer from a host to a viral genome.

Fig. 1.

Sequence comparison, homology modeling, and evolutionary analysis of PK2 to eIF2α family kinases. (A, Upper) PK2 EKCM domain sequence identity with the C-lobes of different kinase domains. The EKCM domain of AcPK2 (amino acids 23–215) was aligned pairwise using EMBOSS Needle with the C-lobes of the indicated eIF2α kinase domains. (Lower) Homology model of PK2 based on a sequence alignment with PKR^KD [Protein Data Bank (PDB) ID code 2A19]. Shown on the left is the structure of the PKR^KD bound to eIF2α; on the right is a homology model of PK2. Domain structure is shown on the bottom. (B) Phylogenetic analysis of PK2. A maximum-likelihood based tree (PhyML) generated from kinase domain amino acid sequences from eIF2α kinases sampled from vertebrates and insects and PK2 sequences from several baculoviruses. The insect HRI-like eIF2α kinase and PK2 most closely associate with vertebrate HRI. The emergence of PK2 is consistent with horizontal transfer of an insect HRI-like gene fragment. Bootstrap values above 60 are indicated at branch nodes. A detailed view of the clades is shown in SI Appendix, Fig. S1B. Please refer to SI Appendix, Supplemental Methods for detailed methods.

Further sequence analysis and homology modeling using the crystal structure of PKR^KD as a template revealed the following: (i) The EKCM domain appeared to share the eIF2α substrate-recognition infrastructure consisting of an atypical helix αG that is one turn longer and tilted 40° relative to helix αG in non-eIF2α kinases (Fig. 1A and SI Appendix, Fig. S1, magenta). This observation provided additional support for the evolutionary origin of PK2 and raised the intriguing possibility that PK2 may act in part by binding to eIF2α. (ii) The EKCM domain of PK2 lacked essential residues normally required for phosphor-transfer function including the catalytic HRD motif (HHN in PK2) and the magnesium-binding DFG motif (MFG in PK2) (SI Appendix, Fig. S1). Thus, even if mated to a canonical kinase N-lobe, PK2 would not support ATP binding and phosphor-transfer functions. (iii) PK2 lacked sequences comparable to a kinase N-lobe or regulatory domains specific to any eIF2α protein kinases. Instead, PK2 possesses a 22-residue extension that is N-terminal to the EKCM domain (Fig. 1A and SI Appendix, Fig. S1, green) and well conserved across all PK2 orthologs.

Many viruses encode factors that mimic and circumvent the antiviral function of the metazoan-specific eIF2α kinase PKR. Interestingly, these mammalian viral inhibitors have undergone rapid evolution (22, 23) that in turn exerted selective pressure to accelerate the evolution of PKR itself (22, 24). That PK2 is a baculovirus-encoded inhibitor of eIF2α kinase activity and is most similar to the insect HRI-like kinase suggests that an antiviral system may have evolved in insect lineages independent of PKR. To investigate the origins of PK2 and its potential impact on kinase evolution, we performed phylogenetic analysis using amino acid sequences of various PK2 genes as well as the kinase domains of the eIF2α kinases PKR, PERK, HRI, and GCN2 from insects to vertebrates (Fig. 1B and SI Appendix, Fig. S1B and Table S1 A and B). First, PK2 robustly associated with insect HRI-like eIF2α kinases in the phylogenetic tree (Fig. 1B). This association supports a model of horizontal gene transfer between insects and viruses in which part of the adopted kinase was co-opted to subvert the host immune function. Second, whereas selective pressure from viral-encoded inhibitors induced rapid evolution of the kinase domain of PKR (Fig. 1B), no such rapid evolution was observed for the HRI-like eIF2α kinases in the lepidopteran (11 species) and mosquito (12 species) lineages tested (Fig. 1B and SI Appendix, Fig. S1B and Table S2). Thus, PK2 and other insect kinase inhibitors may not have imposed sufficient ongoing selective pressure to drive recurrent adaptation of the insect eIF2α kinase. Indeed, the restricted distribution of PK2 to a subset of alpha-baculoviruses implies a single, recent acquisition of the gene from a lepidopteran host. These observations, combined with the lack of a common origin of a single antiviral eIF2α kinase, strongly suggest that the antiviral function for eIF2α kinases emerged independently in insects and vertebrates.

PK2 Directly Binds PKR^KD but Not eIF2α in Vitro.

PKR activates in response to viral invasion in vertebrates and has been used extensively as a model eIF2α kinase to study the mechanism of viral inhibitors (25). Previous studies using yeast lysates showed that PK2 associated with the isolated kinase domain of PKR by coimmunoprecipitation (20), whereas modeling and sequence analysis suggested that PK2 should bind to eIF2α using the atypical helix αG common to all eIF2α kinases. These results raised the possibility that PK2 could inhibit eIF2α phosphorylation by directly binding to PKR kinase domain or by directly sequestering eIF2α substrate, two mechanisms of action that are not mutually exclusive. To determine whether PK2 binds to PKR and/or eIF2α directly in the absence of other cellular factors, we used an in vitro pulldown assay using bacterially expressed and purified proteins. Both wild-type and kinase-dead (K296R) GST-PKR^KD, but not GST-eIF2α, immobilized on glutathione resin bound to PK2 (Fig. 2A, lanes 6–8). In a reciprocal maltose-binding protein (MBP) pulldown experiment, MBP-PK2 immobilized on maltose resin bound to PKR^KD but not eIF2α (SI Appendix, Fig. S2A lanes 5 and 7). Consistent with PKR^KD being the direct interaction target of PK2, untagged PKR^KD and PK2 proteins comigrated together during size-exclusion chromatography with an apparent stoichiometry of 1:1 (Fig. 2B, Bottom). Together, these results suggested that PK2 acts by binding directly to PKR^KD and not to eIF2α.

Fig. 2.

PK2 directly binds PKR^KD but not eIF2α in vitro in a manner dependent on the N-terminal extension of PK2. (A) GST pulldown assay. Purified wild-type PK2 was incubated with wild-type GST-PKR, GST-PKR K296R (catalytically dead GST-PKR), GST-eIF2α, or GST alone, bound to Glutathione Sepharose resin as indicated, washed, and then subjected to SDS/PAGE analysis and Coomassie staining. (B) Size-exclusion chromatography coelution assay. Sixty-micromolar amounts of PKR (Top), PK2 (Middle), or PK2–PKR complex (Bottom) were applied to a Superdex S75 column. Fractions corresponding to elution volumes were subjected to SDS/PAGE analysis. (C) GST pulldown assay. Equimolar mixtures of wild-type PK2 and the N-terminal deletion mutant (PK2^Δ1–21) were incubated with wild-type GST-PKR, GST-PKR^K296R (catalytically dead GST-PKR), GST-eIF2α, or GST as indicated. Pulldown assays were performed as in A. (D) PK2 limited proteolysis. Two micrograms of wild-type PK2 were incubated for 30 min with 2 ng of trypsin, chymotrypsin, or thermolysin, and samples were analyzed by SDS/PAGE and Coomassie staining. Edman sequencing of the protease-resistant fragment following trypsin digestion identified the PK2 peptide IYFAS and the protease cleavage site as being Arg15. (E) MBP pulldown assay. PKR^KD was incubated with C-terminal MBP-tagged full-length PK2, PK2^1–22, PK2^1–44, or MBP alone as indicated. Samples were bound to amylose resin, washed, and subjected to SDS/PAGE followed by Coomassie staining.

The N-Terminal Extension of PK2 Is Required but Not Sufficient for Binding to PKR^KD.

In the crystal structure of human PKR^KD in complex with yeast eIF2α, the N-lobe of PKR^KD mediated kinase domain dimerization, and the C-lobe of PKR^KD mediated eIF2α recognition through helix αG. The observation that PK2 can interact directly with PKR^KD raised the question of how this interaction occurs, because PK2 lacks a recognizable kinase N-lobe. Because PK2 possesses a 22-residue N-terminal extension in place of a kinase N-lobe, we reasoned that the N-terminal extension might serve an analogous dimerization function. In support of this hypothesis, both wild-type and kinase-dead (K296R) GST-PKR^KD immobilized on glutathione resin interacted with wild-type PK2 but not with N-terminally truncated PK2 in a competitive GST-pulldown experiment (Fig. 2C, lanes 5 and 6).

To probe the structural nature of the N-terminal extension of PK2, in particular whether it is extraneous to or an integral part of the EKCM domain, we performed limited proteolysis. In the presence of dilute concentrations of three different proteases, wild-type PK2 was cleaved from a 25-kDa full-length protein into a smaller 22-kDa protease-resistant fragment (Fig. 2D, lanes 2–4). Edman sequencing identified Arg15 as the N-terminal boundary of the protease-resistant fragment of PK2 by trypsin digestion. Further supporting a role for its involvement in binding PKR^KD, the N-terminal extension of PK2 was protected from proteolysis in the presence of increasing PKR^KD. (SI Appendix, Fig. S2B; compare lane 2 with lanes 5 and 11).

To investigate whether the N-terminal extension of PK2 was sufficient to bind PKR^KD, we expressed and purified MBP fusions of wild-type PK2 or PK2 lacking the EKCM domain and evaluated binding to PKR^KD. Although full-length PK2 bound robustly to PKR^KD (Fig. 2E, lane 6), the N-terminal extension of PK2, corresponding to residues 1–22 or residues 1–44, did not bind PKR^KD (Fig. 2E, lanes 7 and 8). Together, these results showed that the N-terminal extension of PK2 is a protease-labile element peripheral to the EKCM domain that is required, but not sufficient, for high-affinity binding to PKR^KD in vitro. That the minimal N-terminal extension of PK2 was not sufficient for binding PKR^KD suggested that the EKCM domain has a supporting role in PKR binding.

Lastly, we finely probed the determinants in PK2 for binding PKR^KD. Removal of the first eight residues of PK2 (PK2^Δ1–8) did not affect binding to GST-PKR^KD, but removal of the first 21 residues of PK2 (PK2^Δ1–21) abolished binding (Fig. 3A, lanes 3 and 4). This result localized an essential binding determinant to a region encompassing PK2 residues 9–21. Alanine-scan mutational analysis of PK2 residues 9–21 revealed that a single amino acid substitution, F18A, abolished PK2 binding to GST-PKR^KD (Fig. 3B, lane 11). Consistent with a loss of PKR binding, proteolytic protection assays showed that PKR protects wild-type PK2, but not PK2^F18A, from proteolysis (Fig. 3C, lanes 7 and 8).

Fig. 3.

Systematic mutational analysis of the N-terminal extension of PK2. (A) GST pulldown assay with wild-type PK2 and N-terminal truncation mutants using the protocol described in Fig. 2A. (B) GST pulldown with wild-type PK2 and the indicated alanine mutants using the protocol described in Fig. 2A. (C) Protease protection assay. Two micrograms of wild-type PK2 or PK2^F18A were subjected to chymotrypsin proteolysis in the presence or absence of a fourfold molar excess of PKR^KD.

The N-Terminal Extension of PK2 Is Required but Not Sufficient for Inhibiting PKR in Vitro.

Our in vitro studies showed that PK2 binds PKR^KD in a manner dependent on the PK2 N-terminal extension centered on the critical residue Phe18. We next asked whether the N-terminal extension of PK2 also was required for PK2 to inhibit PKR kinase function in vitro. In an in vitro kinase assay using purified proteins, we observed that wild-type PK2 inhibited PKR autophosphorylation on Thr446 and eIF2α phosphorylation on Ser51 (Fig. 4 A and B). In contrast, mutants PK2^F18A and PK2^Δ1–21, which were deficient for binding to PKR^KD, were also fully compromised in inhibiting both PKR kinase activities in vitro. Interestingly, PK2^F18I [corresponding to a known AcMNPV PK2 variant (accession number NP_054153)], PK2^T148D, and PK2^T148A (bearing mutations to the eIF2α recognition helix αG of PK2) were fully competent in PKR inhibitory function. Lastly, in contrast to full-length PK2 fused to MBP, the N-terminal extension alone fused to MBP was not sufficient to inhibit PKR kinase activity. Taken together, these results demonstrate a direct correlation between the ability of PK2 to bind and to inhibit PKR.

Fig. 4.

The N-terminal extension of PK2 is required for PKR inhibitory function in vitro and in vivo. (A) In vitro PKR autophosphorylation assay. Dephosphorylated GST-PKR (8 µM) was incubated with or without ATP in the presence of 30 µM wild-type or mutant PK2 proteins as indicated. Samples were subjected to SDS/PAGE, followed by immunoblot analysis using an anti-pThr446 PKR antibody. (B) In vitro eIF2α phosphorylation assay. PKR^KD (10 nM) was incubated with 10 µM eIF2α, with or without ATP, in the presence of 30 µM wild-type or mutant PK2 proteins as indicated. Samples were subjected to SDS/PAGE, followed by immunoblot analysis using an anti-pSer51 eIF2α antibody. (C) Fluorescent polarization (FP)-based ATP displacement assay. PKR (30 µM) was incubated with 20 nM BODIPY-γ-S-ATP with increasing concentrations of untagged wild-type or mutant PK2 proteins, as indicated. (D) Yeast growth assay. Empty control vector (pEMBLyex4) or plasmids expressing the indicated forms of AcMNPV or BmNPV PK2 under the control of a yeast GAL-CYC1 hybrid promoter were introduced into isogenic yeast strains J673 (no kinase) or J865 (GAL-CYC1-HsPKR). Transformants were grown to saturation in synthetic complete (SC) medium, and 4-µL serial dilutions (OD₆₀₀ = 1.0 or 0.1) were spotted on SD medium, in which PKR and pk2 expression are repressed, or on SGal medium, in which PKR and pk2 expression are induced. Growth on SGal medium indicates inhibition of HsPKR kinase activity by a functional PK2.

Next, we asked how binding of PK2 to PKR^KD might influence the latter’s ability to act as an enzyme. Toward this end, we first probed whether PK2 perturbed the ability of PKR to bind nucleotide. Using a fluorescence polarization assay with BODIPY-γ-S ATP at 20 nM as a probe and increasing concentrations of PKR^KD, we measured a dissociation constant (K_d) of 23 μM (SI Appendix, Fig. S3A). In subsequent competition assays holding the PKR^KD concentration fixed at 30 μM, we observed that titration of wild-type PK2 caused displacement of BODIPY-γ-S ATP from PKR^KD with an IC₅₀ of 30 μM, which translates into a K_i of 3 μM using the Munson–Rodbard adjustment equation (Fig. 4C, squares). Notably, the PK2^F18A and PK2^Δ1–21 mutants, which were compromised for PKR binding and inhibitory function, showed 20-fold and 107-fold impairments, respectively, in their ability to displace BODIPY-γ-S ATP (as reflected by K_i values compared with wild-type PK2) (Fig. 4C, circles and triangles). In contrast, both the PK2^T148D and PK2^F18I mutants were similar to wild-type PK2 in inhibitory potency (Fig. 4C, filled and open downward-pointing triangles). Finally, in contrast to wild-type PK2 fused to MBP, the PK2 N-terminal extension alone fused to MBP was not able to displace nucleotide from PKR (SI Appendix, Fig. S3B, upward- and downward-pointing triangles). These results extend the correlation of PK2 binding and inhibitory function on PKR and reveal that PK2 may act in part by disrupting the ability of PKR^KD to bind nucleotide.

The N-Terminal Extension of PK2 Is Important for PKR Inhibition in Yeast.

We next investigated whether the relationships discerned for PK2 function in vitro extended to a cellular context. Expression of full-length human PKR in yeast under the control of a galactose-inducible promoter was toxic on galactose (SGal) medium because of sustained phosphorylation of eIF2α on Ser51 (Fig. 4D, row 1). Coexpression of wild-type AcMNPV PK2 partially suppressed PKR-mediated toxicity as evidenced by rescued growth (Fig. 4D, row 4), although not to the level observed in cells lacking PKR. In contrast to the ability of the wild-type AcMNPV PK2 to suppress PKR toxicity in yeast, expression of the PK2^F18A mutant failed to suppress toxicity, whereas the PK2^F18I allele functioned comparably to the wild-type protein (Fig. 4D, rows 5 and 6). Finally, mutation of the projected eIF2α docking site on helix αG (T147A, E150A) of AcMNPV PK2 had no effect on PKR inhibition in vivo (Fig. 4D, rows 7 and 8). These results were fully consistent with our in vitro findings that the N terminus of PK2, but not the helix αG determinants, is critical for PKR inhibition and thus provide further support for the notion that PK2 functions by directly targeting PKR^KD and not eIF2α. To extend these studies and to correlate with the virology studies of BmNPV that are discussed next, we switched to using PK2 from BmNPV. As shown in SI Appendix, Fig. S1, the PK2 from BmNPV has a shorter, 10-residue C-terminal tail compared with the 29-residue tail on the AcMNPV protein. Moreover, within the conserved body of the proteins BmNPV PK2 differs from AcMNPV PK2 at eight of 215 residues (96% identity) within the overlapping regions of the two proteins. Under the stringent conditions used for the test, the wild-type PK2 expressed by BmNPV weakly alleviated the growth inhibition caused by PKR overexpression in yeast (Fig. 4D, row 2). As observed with AcMNPV PK2, this activity was lost by the F18A mutation (Fig. 4D, row 3). Wild-type and mutant PK2 protein expression levels were similar in all yeast strains tested, and inhibition of yeast toxicity by PK2 correlated with reduced eIF2α phosphorylation levels (SI Appendix, Fig. S7 A and B).

The N-Terminal Extension of PK2 Is Important for eIF2α Kinase Inhibition in Baculovirus-Infected Insect Cells.

We next asked whether the mechanistic insights gained in our in vitro and yeast studies using PKR as a model target of PK2 had physiological relevance in lepidopteran cells, the natural host of the BmNPV baculovirus. We characterized a pk2-deleted BmNPV (KO) in B. mori cultured cells (BmN cells) and host larval insects (B. mori). We observed that budded virus (BV) production was reduced, but eIF2α phosphorylation increased in KO-infected BmN cells compared with cells infected with wild-type virus (Fig. 5A). The larval bioassays also revealed that the KO virus took 12 h longer than the wild-type virus to kill B. mori larvae (Fig. 5B). These results indicate that pk2 is involved in BmNPV growth in cultured cells and virulence in insect larvae.

Fig. 5.

The N-terminal extension is required for PK2 function during baculovirus infection. (A, Upper) Viral propagation of wild-type and KO viruses in BmN cells. *P < 0.05; Student’s t-test. (Lower) eIF2α phosphorylation in BmN cells infected with wild-type or KO viruses. BmN cells were infected with wild-type or KO BmNPV at an MOI of 5. At 48 hpi, cells were harvested and subjected to Western blot analysis using phospho-specific or total eIF2α antibodies. (B) Survival curves for B. mori larvae infected with wild-type and KO viruses. The LT₅₀s of wild-type and KO viruses were 116 h and 132 h, respectively. A log-rank (Mantel–Cox) test with Bonferroni correction was used, comparing each of the viruses with KO. *P < 0.05 vs. KO. (C) BV production in BmN cells infected with KO viruses expressing BmNPV pk2 variants. BV production was assessed at 48 hpi in BmNPV-infected BmN cells. One-way ANOVA was performed with a post hoc Tukey’s test. *P < 0.05 vs. KO. Data show means ± SD of triplicates, and similar results were obtained in two independent experiments. (D) Survival curves for B. mori larvae infected with wild-type, KO, KO+BmPK2, or KO+BmPK2^F18A virus. The LT₅₀s were 120 h, 132 h, 124 h, and 136 h, respectively. A log-rank (Mantel–Cox) test with Bonferroni correction was used, comparing each of the viruses with KO. *P < 0.05 vs. KO.

To examine the role of the PK2 N-terminal extension in baculovirus infection, we generated derivatives of the pk2 KO virus in which BmNPV pk2 variants are expressed under the ie1 promoter (SI Appendix, Fig. S4A). As shown in Fig. 5C, the amounts of infectious BVs produced in cultured cells by KO viruses expressing BmNPV PK2 (KO+BmPK2) were similar to those produced by wild-type BmNPV. However, like the pk2 KO virus, the virus expressing PK2^F18A (KO+BmPK2^F18A) showed lower BV production than wild-type BmNPV in cultured cells. These results indicate that residue Phe18 of PK2 is important for virus propagation in cultured cells. In addition, we examined whether Phe18 also is required for PK2 function in larval insects. Fifth-instar B. mori larvae were intrahemocoelically inoculated with BV, and survival curves were determined. The assay clearly showed that viruses lacking PK2 (KO) or expressing the PK2^F18A mutant (KO+BmPK2^F18A) took 12–18 h longer to kill larvae than did wild-type BmNPV (Fig. 5D), indicating that Phe18 is critical for BmNPV virulence in insect hosts. Similar results were observed when the pk2-deleted BmNPV expressing AcMNPV PK2^F18A (KO+AcPK2^F18A) was tested in cultured cells and larval insects (SI Appendix, Fig. S4 B and C). Finally, eIF2α phosphorylation levels were reduced in cultured cells infected with BmNPV containing functional wild-type AcPK2 compared with virus expressing the defective PK2^F18A mutant, consistent with the inhibition of a host eIF2α kinase in vivo (SI Appendix, Fig. S4D).

Taken together, these results showed that PK2 inhibits eIF2α phosphorylation in vivo and is required for efficient baculovirus infection in both cultured cells and larval insects. Moreover, these results indicated that the functional relationships discerned for the PK2 mechanism of action on PKR had relevance in a cellular context with the full complement of insect host eIF2α kinases.

PK2 Inhibits PKR^KD by Engaging the Kinase N-Lobe.

The insufficiency of the N-terminal extension of PK2 for binding and inhibition of PKR^KD provided indirect evidence of an essential role for the adjacent EKCM domain. To gain insight into the function this domain might serve, we performed a gain-of-function screen in yeast to identify PK2 mutations that enhance the ability to alleviate toxicity caused by eIF2α kinase overexpression. For these studies we used PK2 from BmNPV. As observed previously (Fig. 4D), the wild-type PK2 from BmNPV had negligible ability to alleviate the growth inhibition caused by PKR overexpression in yeast, whereas the hyperactive K3L-H47R mutant of the vaccinia virus pseudosubstrate inhibitor of PKR displayed a potent ability to alleviate toxicity (Fig. 6A, Left, row 2 vs. row 6). To test the activity of PK2 against an authentic host eIF2α kinase, we fused the kinase domain of the B. mori HRI-like kinase (BmHRI) to GST, to promote dimerization and activation of the kinase, and then expressed the fusion protein (GST-BmHRI^KD) in yeast under the control of the galactose-regulated GAL1 promoter. Like the expression of human PKR (HsPKR), induction of GST-BmHRI^KD expression on galactose medium impaired yeast cell growth (Fig. 6A, Center, row 1). However, in contrast to its weak ability to suppress growth inhibition caused by expression of HsPKR in yeast, the PK2 from BmNPV robustly restored the growth of yeast expressing GST-BmHRI^KD (Fig. 6A, Center, row 2). Using the yeast strains expressing HsPKR and GST-BmHRI^KD, we screened for mutants of BmNPV PK2 that were more effective than wild-type PK2 in suppressing the growth inhibition caused by eIF2α kinase overexpression in yeast. In total three PK2 enhancer mutations were isolated, namely PK2^I82S, PK2^G95S, and PK2^K102E, that when coexpressed with HsPKR restored yeast cell growth much more effectively than wild-type PK2 (Fig. 6A, Left, rows 3–5 vs. row 2). These PK2 mutants also appeared to be moderately more effective than the wild-type protein in suppressing GST-BmHRI^KD toxicity in yeast [although this effect was difficult to assess, because the wild-type protein already was a potent inhibitor of GST-BmHRI^KD toxicity (Fig. 6A, Center, rows 3–5 vs. row 2)].

Fig. 6.

PK2-activating mutations support an inhibitory model whereby PK2 engages the eIF2α kinase N-lobe, possibly through a lobe-swapping mechanism. (A) Yeast growth assay. Empty control vector (pEMBLyex4) or plasmids expressing the indicated forms of BmNPV PK2 or the hyperactive vaccinia virus (vv) K3L-H47R protein under the control of a yeast GAL-CYC1 hybrid promoter were introduced into isogenic yeast strains J673 (no kinase), J865 (GAL-CYC1-HsPKR), or J1221 (GAL-CYC1-GST-BmHRI^KD). Transformants were grown to saturation in SC medium, and 4-µL serial dilutions (OD₆₀₀ = 1.0 or 0.1) were spotted on SD medium, in which kinase and PK2 expression are repressed, or on SGal medium, in which kinase and PK2 expression are induced. Growth on SGal medium indicates inhibition of kinase activity by PK2 or K3L. (B) Projection of PK2-activating mutations on the inferred structure of PK2. A homology model of PK2 is shown, superimposed on the structure of PKR^KD (PDB ID code 2A1A). (C) PK2 lobe-swapping inhibition model. The N-terminal extension of PK2 centered on Phe18 mediates high-affinity binding to the N-lobe of PKR^KD. The EKCM domain of PK2 may displace the C-lobe of PKR^KD, resulting in a pseudo protein kinase domain that lacks the ability to bind ATP and phosphorylate substrate. PKR contains the essential, invariant kinase motifs DFG and HRD, whereas PK2 contains the nonproductive motifs MFG and HHN. (D) GST pulldown assay for PKR^KD, PKR N-lobe, or PKR C-lobe binding to wild-type or mutant PK2, as indicated. (E) Two-hybrid assays. Vector pGAD424 expressing the GAL4 AD or derivatives expressing the indicated AD-HsPKR or AD-BmHRI fusion proteins were introduced into yeast strain Y190 along with the vector pGBT9 expressing the GAL4 DNA BD or a derivative expressing the BD-BmPK2 fusion protein. Transformants were grown to saturation in SC medium. Four-microliter serial dilutions (OD₆₀₀ = 1.0, 0.1, 0.01, 0.001, or 0.0001) were spotted on medium containing 10 mM 3-AT and were incubated for 5 d at 30 °C. (F) Inhibition of PKR in vitro kinase activity by wild-type PK2 and PK2 K102E.

Mapping of the identified mutations onto a homology model of the PK2 structure revealed clustering to a region that forms the bottom surface of the catalytic cleft in bona fide protein kinases (Fig. 6B, red residues). The G95S mutation localized to the magnesium-binding loop, the I82S mutation localized to the catalytic loop, and the K102E mutation localized to the activation segment. From the predicted positions of these mutations in the PK2 homology model, and a working knowledge of protein kinase structure and function, we hypothesized that PK2 may inhibit eIF2α kinase function through a lobe-swapping mechanism. Accordingly, upon binding to PKR, the EKCM domain of PK2 replaces the kinase domain C-lobe of PKR (Fig. 6C). In support of this possible model, we found, using a GST pulldown assay, that AcMNPV PK2 selectively bound to the isolated N-lobe and not the isolated C-lobe of PKR^KD (Fig. 6D, lanes 2–4). Importantly, the interaction between AcMNPV PK2 and the N-lobe of PKR^KD also was dependent on the N-terminal extension of PK2 (Fig. 6D, lanes 6–8 and 10–12).

Yeast two-hybrid assays were used to examine further the binding of PK2 to eIF2α kinase domains. The GAL4 DNA-binding domain (BD) alone or fused to full-length PK2 from BmNPV (BD-BmPK2) was tested for interaction with GAL4 activation domain (AD) fusions containing various portions of human PKR or BmHRI. Positive interactions between the BD and AD fusion proteins will induce HIS3 expression and confer growth on medium containing 3-aminotriazole (3-AT). As shown in Fig. 6E, full-length BmHRI interacted readily with PK2 (row 2). In contrast, no interaction above background was observed between PK2 and the full-length HsPKR^KD (row 5). The more pronounced interaction between PK2 and BmHRI^KD vs. HsPKR is consistent with the greater ability of PK2 to suppress the growth inhibition in yeast caused by expression of BmHRI vs. HsPKR (Fig. 6A, row 2). Next, the two-hybrid assays were used to test the binding of PK2 to the N- and C-lobes of the BmHRI and HsPKR kinase domains. The N-lobe of BmHRI^KD (residues 119–368) interacted with PK2, yielding weak growth on the 3-AT medium (Fig. 6E, row 3), whereas the C-lobe of BmHRI^KD (residues 369–579) failed to interact with PK2 to a greater extent than the empty AD control (row 4 vs. row 1). This interaction of PK2 with the N-lobe, but not the C-lobe, of BmHRI is consistent with a lobe-swapping mechanism of kinase inhibition; however, the weaker interaction of PK2 with the BmHRI N-lobe vs. full-length BmHRI was unexpected. It is noteworthy that the AD-BmHRI^119–579 fusion protein was expressed to higher levels than the AD-BmHRI^119-368 fusion, possibly accounting for the weaker-than-expected two-hybrid interaction with the isolated N-lobe (SI Appendix, Fig. S6A). In contrast to the robust interaction between BD-PK2 and the full-length AD-BmHRI fusion, no interaction was detected between BD-PK2 and the full-length AD-HsPKR^KD fusion (Fig. 6E, row 2 vs. row 5). Interestingly, a two-hybrid interaction was detected between BD-PK2 and the HsPKR^KD N-lobe (residues 258–369) but not the HsPKR^KD C-lobe (residues 370–551) (Fig. 6E, rows 6 and 7). Thus, PK2 readily interacted with full-length or the N-lobe of BmHRI^KD but interacted only with the free N-lobe of the HsPKR^KD. Perhaps PK2 can readily displace the C-lobe of the insect BmHRI kinase and thus interact with the full-length BmHRI^KD; PK2 may not displace the C-lobe of HsPKR as effectively and thus requires its removal to interact strongly with the N-lobe of HsPKR.

These results are consistent with the lobe-swapping model in which the EKCM domain of PK2 interacts with the N-lobe of HsPKR or BmHRI to displace its C-lobe. Therefore we reasoned that the increased inhibitory function of PK2 activating mutations arose from an enhanced interaction across the catalytic cleft between the EKCM domain of PK2 and the kinase N-lobe of PKR. Consistent with this hypothesis, PK2^K102E was sixfold better than wild-type PK2 as an inhibitor of PKR in vitro kinase activity (Fig. 6F). We note that in purifying the three PK2 activating mutants, only PK2^K102E, but not PK2^I82S or PK2^G95S, was soluble in isolation when expressed from bacteria (SI Appendix, Fig. S5A). Interestingly, PK2^I82S and PK2^G95S were soluble when coexpressed with the kinase N-lobe of PKR (SI Appendix, Fig. S5B). In addition, wild-type PK2 comigrated with the minimal PKR N-lobe, forming a tight complex and coeluting during size-exclusion chromatography (SI Appendix, Fig. S5C). Taken together, our results support a mode of interaction whereby PK2-activating mutations enhance binding across the catalytic cleft to the PKR N-lobe. In this model, PK2 competitively displaces the C-lobe of PKR in a lobe-swapping fashion and forms nonenzymatically productive interactions with the PKR N-lobe (Fig. 6C).

HRI Is the Endogenous Target of PK2 During Baculovirus Infection.

Having observed that PK2 can inhibit both HsPKR and BmHRI, we next set out to address which of the three B. mori eIF2α kinases, namely BmHRI, BmGCN2, or BmPERK, is the likely in vivo target of PK2. As observed for the overexpression of human PKR in yeast, overexpression of fusion proteins consisting of the constitutive dimer GST fused to the kinase domains of BmHRI or BmPERK inhibited yeast cell growth (Fig. 7A, row 1). Overexpression of GST-BmGCN2^KD in yeast gave no observable phenotype and hence was not investigated further. Overexpression of nonphosphorylatable eIF2α-S51A partially suppressed the toxicity associated with expression of the heterologous eIF2α kinases in yeast (Fig. 7A, row 4), indicating that the growth inhibition was caused by impaired translation. Moreover, the comparable levels of growth restoration in the cells coexpressing eIF2α-S51A with the various kinases indicates that the three kinases impair yeast cell growth to similar extents. Expression of the hyperactive K3L-H47R mutant of the vaccinia virus pseudosubstrate restored growth in the strains expressing HsPKR or GST-BmHRI^KD but not in the strain expressing GST-BmPERK^KD (Fig. 7A, row 3). Finally, expression of PK2 from BmNPV was able to restore growth substantially only in the strain expressing GST-BmHRI^KD (Fig. 7A, row 2). Importantly, in all these experiments, yeast cell growth correlated with a reduction in eIF2α phosphorylation (SI Appendix, Fig. S6B), demonstrating that the inhibitors functioned by impairing eIF2α kinase activity. Consistent with this marked ability of PK2 to inhibit BmHRI function in yeast, we found that PK2 was a superior inhibitor of eIF2α phosphorylation by recombinant, purified BmHRI with an IC₅₀ of 1.1 μM, compared with HsPKR (IC₅₀ = 2.5 μM) and BmPERK (IC₅₀ = 20.2 μM) (Fig. 7B). We were unable to test BmGCN2 in this analysis because the protein could not be produced recombinantly. Taken together, these results show that PK2 is most efficient at inhibiting the HRI-like kinase in B. mori and, in agreement with our phylogenetic analysis (Fig. 1B), raise the possibility that the HRI-like kinase performs an antiviral role during baculovirus infection.

Fig. 7.

BmHRI is the in vivo target of PK2. (A) BmNPV PK2 is a potent inhibitor of BmHRI in yeast. Empty control vector (pEMBLyex4), plasmids expressing the indicated forms of BmNPV PK2 or the hyperactive vaccinia virus (vv) K3L-H47R protein under the control of a yeast GAL-CYC1 hybrid promoter, or a plasmid expressing nonphosphorylatable yeast eIF2α-S51A under the control of the native promoter were introduced into isogenic yeast strains J673 (no kinase), J865 (GAL-CYC1-HsPKR), J1221 (GAL-CYC1-GST-BmHRI^KD), or J1222 (GAL-CYC1-GST-BmPERK^KD). Transformants were grown to saturation in SC medium, and 4-µL serial dilutions (OD₆₀₀ = 1.0, 0.1, 0.01, 0.001, or 0.0001) were spotted on SGal medium, in which kinase, PK2, and K3L expression is induced. (B) In vitro eIF2α phosphorylation assay using the protocol described in Fig. 5B. eIF2α substrate (10 μM) was incubated with kinase reaction buffer, the indicated protein kinase (10 nM HsPKR or 25 nM BmHRI or 25 nM BmPERK), and increasing concentrations of PK2 (0.09, 0.18, 0.35, 0.71, 1.4, 2.8, 5.6, 11.2, 22.5, and 45 μM). Samples were subjected to SDS/PAGE and immunoblot analysis using an anti-pSer51 eIF2α antibody. (C) Effect of knockdown of three host eIF2α kinases on viral growth in cultured cells. BmN cells were transfected with siRNAs against BmHRI, BmPERK, and BmGCN2 and then were infected with KO and KO-WT-bm. BV production at 48 hpi was assessed by plaque assay. *P < 0.05 vs. siGFP with Student’s t-test.

If PK2 inhibits BmHRI to increase viral fitness during baculovirus infection, we reasoned that knockdown of BmHRI, but not the other eIF2α kinase genes, in the host would rescue the impaired replication of the BmNPV mutant lacking pk2. To test this hypothesis, we first established RNAi knockdown conditions for BmHRI, BmPERK, and BmGCN2 and confirmed the knockdown efficiency by quantitative RT-PCR (SI Appendix, Fig. S6C). We next examined whether knockdown of these kinase genes affects virus propagation in cultured cells. Compared with a nonspecific control siRNA against GFP, knockdown of BmHRI, BmPERK, or BmGCN2 did not affect BV production by the reconstituted wild-type virus (KO+BmPK2) (Fig. 7C, compare bars 3 and 4 in each panel). Strikingly however, knockdown of BmHRI, but not BmPERK or BmGCN2, caused a significant rescue of low BV production by the pk2-disrupted BmNPV (KO) (Fig. 7C, compare bars 1 and 2 in each panel). These observations strongly support the notion that BmHRI is the bona fide target of PK2 in vivo.

Discussion

PK2 Inhibits eIF2α Kinases by Engaging the N-Lobe of the Kinase Domain.

In vitro and in vivo analyses reveal two distinct regions of PK2 domain architecture as important for function. The first region, which we refer to as “the N-terminal extension of PK2,” has been delineated by sequence homology and limited proteolysis analyses as a labile 22-aa peptide sequence at the very N terminus of PK2. Because mutation (PK2^F18A) or deletion (PK2^Δ1–21) of this epitope abolished the ability of PK2 to bind and inhibit eIF2α kinases in vitro and in vivo, and because the isolated epitope (residues 1–22) is insufficient for binding or inhibition, we believe that a chief function of the N-terminal extension of PK2 is to promote high-affinity interaction with the eIF2α kinase domain. A second region of PK2, with functionally distinct properties, is the EKCM domain immediately adjacent to the N-terminal extension of PK2. Sequence analyses revealed that the EKCM domain is homologous to the C-lobe of an eIF2α kinase domain and contains the eIF2α substrate-docking infrastructure centered on helix αG that is unique to eIF2α family kinases. However, our study indicates that PK2 does not function by competitive sequestration of eIF2α via its EKCM domain, because specific mutation of helix αG (PK2^T148D) did not affect PK2 function in vitro or in vivo. Instead, we propose that the EKCM domain acts at the level of the eIF2α kinase domain and serves to displace the C-lobe competitively by binding the N-lobe in a reconstituted kinase-like manner. Because the EKCM domain is deficient in crucial motifs normally required in bona fide protein kinase domains to coordinate magnesium/ATP or catalyze phosphor-transfer, the resulting PK2–PKR interaction would yield a nonfunctional pseudokinase–kinase complex. We refer to this previously unidentified mechanism of kinase inhibition as the “lobe-swapping” inhibitory model. To our knowledge, this is the first example of kinase regulation in which an entire C-lobe of a kinase domain is displaced by a pseudokinase C-lobe mimic. The structural implication would be that the hinge region connecting the bilobal architecture of the eIF2α kinase domain is much more flexible and dynamic than previously envisioned. Whether PK2 does in fact engage the N-lobe of eIF2α kinases in a manner that structurally mimics the bilobal architecture of the eukaryotic protein kinase domain or through an altogether novel binding mechanism awaits final validation by a high-resolution structure of PK2 in complex with its target.

PK2 Reveals the Convergent Evolution of Antiviral eIF2α Functions.

PK2 is a baculovirus-encoded protein that has no discernable orthologs in mammalian viruses. Our analysis revealed that it shares greatest sequence identity with HRI and HRI-like kinases over other eIF2α and non-eIF2α kinases. Using in vitro kinase assays and in vivo yeast toxicity assays, we showed that PK2 inhibits BmHRI more efficiently than other eIF2α kinases. Moreover, siRNA-mediated knockdown of BmHRI, but not BmGCN2 or BmPERK, in host silkworm cells was able to rescue virus production by a pk2-disrupted virus. These observations strongly suggest that the HRI-like kinase is the natural target of PK2 and that PK2 originated from an HRI-like gene from an insect host via horizontal gene transfer. Interestingly, because invertebrates lack red blood cells, HRI-like homologs from these species are unlikely to respond to iron/heme deficiency as do their vertebrate counterparts. It is clear, however, that when HRI becomes activated in vivo, its function creates unfavorable replicative conditions for the baculovirus that can be reversed by viral expression of PK2 protein. Based on the restricted phylogenetic distribution of key N-terminal histidine residues and cysteine/proline motifs implicated in the recognition of heme by HRI (26, 27), we speculate that the ancestral function of HRI-like genes was antiviral. This model is consistent with evolutionary analysis suggesting that the heme responsiveness of HRI appeared only after the split of ancestral insects and metazoans. Under this scenario, the eIF2α kinase PKR took up an antiviral function through convergent evolution of this crucial host defense mechanism in vertebrates around the time HRI-like proteins acquired heme recognition to become HRIs. Furthermore, although long-term antagonistic evolution is observed between the pseudosubstrate mimic K3L protein from vaccinia virus and PKR (22, 24), no such evolutionary signal was observed between PK2 and the insect HRI-like kinase or any of the other insect eIF2α kinases. This absence of signal may indicate that the recent emergence of PK2 and other potential kinase inhibitors has not imposed sufficient selective pressure to drive rapid evolution of insect eIF2α kinases.

The Kinase Noncatalytic Domain, a Metazoan C-Lobe Domain.

The kinase noncatalytic domain (KIND) has been identified as a metazoan-specific protein–protein interaction domain consisting of a minimal kinase C-lobe similar to PK2. Unlike PK2, in which an atypical helix αG clearly delineates its origin from an eIF2α kinase C-lobe, the evolutionary origin of the KIND domain is unclear. The protein interaction function of KIND domains has been proposed to derive from a vestigial remnant of substrate recognition infrastructure. However, the crystal structure of the KIND domain of Spir in complex with interaction peptides did not resemble the substrate-targeting mode for any kinase solved to date and instead may have evolved independently via deletion of a conserved structural element of the kinase C-lobe (28, 29). Our study raises the intriguing possibility that, in addition to binding peptides, the KIND domain also may bind to protein kinase domains to regulate their catalytic function.

PK2 Is Important for Baculovirus Fitness During Infection of Insect Host.

PK2 exists specifically in a subset of alphabaculoviruses that are closely related to BmNPV and AcMNPV. Thus, the ancestral gene might have been transferred from the specific lepidopteran host. Previous studies showed that pk2 KO virus infection of Sf9 cells generated higher eIF2α phosphorylation than wild-type virus infection, concomitant with increased apoptosis (30). Moreover, UV-induced eIF2α phosphorylation was mitigated more efficiently by wild-type virus than by pk2 KO virus. In this study, we have shown, for the first time to our knowledge, that PK2 is required in vivo for efficient baculovirus infection in both cultured cells and larval insects (in the silkworm B. mori). Thus, the paradigm of inhibiting host eIF2α phosphorylation by viruses extends to insects.

Materials and Methods

Protein Expression, Mutagenesis, and Purification.

Expression and purification of human PKR (amino acids 258–551, H412N, β4-β5 kinase insert delete) and yeast eIF2α (amino acids 3–175) was performed as described previously (21). PK2 was expressed in Escherichia coli BL21 cells as an N-terminal 6×His-tagged fusion protein and purified similarly to yeast eIF2α, as described previously (21). N-terminally tagged B. mori GST-HRI^KD (amino acids 119–579) and B. mori GST-PERK^KD (amino acids 367–764) were purified similarly to human GST-PKR^KD, as described previously (21). Please see SI Appendix, Supplemental Methods for detailed methods.

GST Pulldown Assay.

In a 50-μL GST pulldown assay reaction using wash buffer [20 mM Hepes (pH 7.0), 100 mM NaCl, 1 mM DTT, 1% glycerol], a 10-μL bed volume of Glutathione-Sepharose 4B resin (GE Healthcare), 10 μg of GST-fusion bait protein, and 50 μg of untagged prey protein (PK2) were incubated at 25 °C for 20 min with constant nutation. The reactions then were washed with 3× 500 mL wash buffer, mixed with 10 μL 6× SDS loading buffer, boiled, and electrophoresed on 15% SDS polyacrylamide gels (Novex; Invitrogen).

MBP Pulldown Assay.

In a 60-μL MBP pulldown assay reaction, a 10-μL bed volume of amylose resin (New England Biolabs), 10 μg of MBP-fusion bait protein, and 50 μg of untagged prey protein were incubated at 25 °C for 20 min with constant nutation. Pulldown assays were performed similarly to the GST pulldown assay protocol, except that amylose resin was used.

PKR–PK2 Gel Filtration Coelution Assay.

A 500-mL volume of 60-μM PKR^KD alone, 60-μM PK2 alone, or a 1:1 stoichiometric 60-μM PKR^KD–PK2 complex was applied to a Superdex 75 gel filtration column at 4 °C in running buffer [0.1 M Hepes (pH 7), 100 mM NaCl, 1 mM Tris(2-carboxyethyl)phosphine (TCEP)].

Limited Proteolysis of PK2 and Protease Protection Assays.

In a 30-μL reaction, 2 μg of wild-type PK2 protein was incubated at 25 °C for 1 h in the absence or presence of 0.5 ng thermolysin, chymotrypsin, or trypsin. The trypsin-proteolyzed PK2 fragment was subjected to Edman sequencing. For the proteolytic protection assay, proteolysis by chymotrypsin was performed as described using 2 μg of PK2^WT or PK2^F18A in the presence or absence of fourfold molar excess PKR^KD.

eIF2α Phosphorylation in Vitro Kinase Assay.

In a 60-μL reaction, 20 nM PKR^KD was incubated with 3 μM wild-type or Ser51Ala yeast eIF2α in the absence or presence of 10 μM wild-type or mutant PK2 for 30 min at 25 °C. Reaction buffer for the kinase reaction consisted of 0.5 mM PMSF, 10 mM MgCl₂, 2 mM ATP, 60 mM Hepes (pH 7.5), 50 mM NaCl, and 0.2 mM Tris(2-carboxyethyl)phosphine (TCEP). The reactions then were mixed with 12 μL 6× SDS loading buffer, boiled, and electrophoresed on 15% SDS polyacrylamide gels (Novex; Invitrogen). Phosphorylation of Se51 on eIF2α was detected using primary anti–phospho-Ser51-eIF2α antibody from Cell Signaling (product 9721) followed by secondary HRP-conjugated goat anti-rabbit antibody from Pierce (product 1858415).

PKR Autophosphorylation in Vitro Kinase Assay.

To produce dephosphorylated kinase protein, human GST-PKR^KD was dephosphorylated using recombinant 6xHis-tagged Lambda phosphatase in dephosphorylation buffer [50 mM Hepes (pH 7.0), 100 mM NaCl, 2 mM MnCl₂, 1 mM DTT], followed by Ni-affinity subtraction to remove the phosphatase (SI Appendix, Fig. S6). In a 60-μL reaction, dephosphorylated GST-PKR^KD was incubated in reaction buffer [0.5 mM PMSF, 10 mM MgCl₂, 2 mM ATP, 60 mM Hepes (pH 7.5), 50 mM NaCl, 0.2 mM TCEP, 2 mM sodium orthovanadate] for 50 min at 25 °C, in the absence or presence of wild-type or mutant PK2 constructs. The reactions then were mixed with 12 μL 6× SDS loading buffer, boiled, and electrophoresed on 15% SDS polyacrylamide gels (Novex; Invitrogen). Phosphorylation of Thr446 on PKR was detected using primary anti–phospho-Thr446 antibody followed by secondary HRP-conjugated goat anti-rabbit antibody from Pierce (product 1858415).

Fluorescence Polarization Assay.

For competition assays, in reaction buffer [30 mM Hepes (pH 7.5), 0.5 mM TCEP, 5 mM Mg²⁺], 30 μM of PKR^K296R kinase domain was incubated with 20 nM BODIPY-γ-S-ATP, in the absence or presence of increasing concentrations of wild-type or mutant PK2 proteins. Reactions carried out in the absence of PKR^K296R kinase domain were used as background and were subtracted from the competition curves to yield the final data.

Yeast Assays.

For information on the construction of yeast plasmids and strains, please refer to SI Appendix, Supplemental Methods. Yeast two-hybrid assays were performed using strain Y190 (MATa leu2-3 leu2-112 ura3-52 trp1-Δ901 his3-200 lys2-801 ade2-101 Gal4-Δ gal80-Δ URA3::GAL1-lacZ LYS2::GAL1-HIS3) (31). Plasmids encoding fusion proteins between the GAL4 DNA BD and BmNPV PK2 or the GAL4 AD and the intact N-lobe or C-lobe of HsPKR or BmHRI were constructed by PCR using oligonucleotide primers that introduced BamHI and PstI (PK2 and HsPKR) or EcoRI and SalI (BmHRI) sites at the 5′ and 3′ ends, respectively. The PCR products were digested with BamHI and PstI or EcoRI and SalI and then were subcloned between the same sites of pGBT9 (BD fusion) or pGAD424 (AD fusion). To assess interactions between PK2 and HsPKR or BmHRI, pGBT9 or the pGBT9 derivative expressing BD-BmPK2 was cotransformed into strain Y190 with pGAD424 or the pGAD424 derivatives expressing AD-HsPKR or AD-BmHRI fusion proteins. The strength of the PK2–kinase interaction was measured by stimulation of the HIS3 reporter in Y190 as assayed by growth on medium containing 10 mM 3-AT. The expression of the AD fusion proteins was assessed by immunoblot analysis. Transformants were grown overnight in synthetic dextrose (SD) medium containing adenine and then were transferred to SD + adenine medium containing 10 mM 3-AT. After 5 h growth at 30 °C, 0.1 volumes of 100% trichloroacetic acid were added, and cells were broken by vigorous mixing with glass beads as described previously (32). Whole-cell extracts were subjected to SDS/PAGE and then immunoblotted using rabbit polyclonal anti–GAL4-TA antibodies (Santa Cruz Biotechnology).

B. mori Cell Lines, Insects, and Viruses.

The B. mori ovary-derived cell line, BmN-4 (BmN), was cultured at 27 °C in TC-100 medium (AppliChem) supplemented with 10% (vol/vol) FBS (33). B. mori larvae were reared as described previously (33). The BmNPV T3 bacmid (34) was used as the wild-type virus in this study.

SDS/PAGE and Western Blotting Analysis of eIF2α Phosphorylation in BmN Cells.

BmN cells were infected with BmNPV at a multiplicity of infection (MOI) of 5. At 48 h postinfection (hpi), cells were harvested and subjected to SDS/PAGE and Western blotting as described previously (33). Western blot analysis of B. mori eIF2α was carried out using polyclonal antibodies specific for phospho-Ser51 on eIF2α or with monoclonal eIF2α antibodies that cross-react with insect eIF2α (20).

Assays for BV Production.

For virus growth curves, BmN cells were infected with recombinant BmNPVs at an MOI of 5. After 1 h of incubation (0 hpi), virus-containing culture medium was removed, and fresh serum-free medium was added. A small amount of culture medium was harvested at 48 hpi, and BV production was determined by plaque assay (33).

Larval Insect Bioassays.

The median lethal dose (LT₅₀) of BV was determined in fifth-instar larvae by intrahemocoelic injection with TC-100 medium containing 100 pfu of BV (33). B. mori larvae were intrahemocoelically inoculated with BV within 12 h after molting to the fifth instar. Survival curves were made using Prism 5 (GraphPad) as previously described (35). A log-rank (Mantel–Cox) test with Bonferroni correction was used, comparing each of the viruses with KO. Fifteen larvae per treatment were used in each experiment.

Statistic Analysis.

Statistic analysis was performed by Prism 5 (GraphPad). P < 0.05 was considered statistically significant.