The eIF-2cr Protein Kinases, Regulators of Translation in Eukaryotes from Yeasts to Humans*

Covalent modification of translational machinery components by phosphorylation is a principal means of regulating protein synthesis in eukaryotic cells (1). Initiation of mRNA translation is a slow step in the protein synthesis pathway (2-5). A key protein factor required for translation initiation is eIF-2. Phosphorylation of the a subunit of protein synthesis initiation factor eIF-2 (eIF-Sa)' is one of the best characterized translational control mechanisms, both in lower and higher eukaryotes. The phosphorylation of eIF-Pa on serine 51 leads to an inhibition of translation (2,4). Protein kinases that catalyze the phosphorylation of eIF-2a have been identified and characterized from a variety of eukar- yotes, ranging from yeast to human cells. The eIF-2a kinases are CAMP-independent enzymes whose induction and activation are finely regulated (Fig. 1). Dependent upon the type of cell, stimuli ranging from cytokine treatment and viral infection to the avail- ability of metabolites and cofactors required for cell function may modulate eIF-2a kinase activity (2,4,6-10). This review examines what is now known about three kinds of eIF-2a protein kinases: HRI, the acid availability.

* In accord with an informal agreement among several laboratories, the term used for the RNA-dependent eIF-2a protein kinase is PKR. The PKR enzyme in earlier 1itGature is variously k n o G as: DAI; & I ; P L ; p1 kinase; Pl/eIF-2a kinase; p65, p67, or TIK (for the mouse enzyme); and p68 or p69 (for the human enzyme). In some cases the eIF-2a protein kinase may be transcriptionally induced in response to signals such as interferon, whereas in other cases the eIF-Pa kinase may be activated by signals such as amino acid or hemin availability or viral infection. Activation of eIF-Sa protein kinase catalytic activity is mediated by an autophosphorylation of the kinase protein, which may possibly affect its association with other proteins. Serine residue 51 of the a subunit of protein synthesis initiation factor eIF-2 is phosphorylated by activated eIF-2a protein kinase, leading to an inhibition of mRNA translation.  (13,18,19); HRI-Rab, the hemin-regulated eIF-Pa kinase kinases: PKR-Hum, the PKR from human cells (12,13); PKR-Mus, the PKR from rabbit reticulocytes (14); GCN2-Ysc, the eIF-Pa kinase from the yeast S. cereuisioe regulated by amino acid availability (15,16). R indicates the RNA binding domains of PKR enzymes (26,28). and HisRS indicates the histidyl-tRNA synthetase-relateddomain of GCNZ (15,17). The numbers refer to amino acid positions, numbered from the N termini. For R, the numbers refer to the approximate center of the 20-amino acid residue core (26). lysine residue is central to the phosphotransfer reaction (11). The HRI and GCN2 eIF-2a kinases possess relatively large inserts between some of the catalytic subdomains, most notably between subdomains IV and VI, relative to other protein serine/threonine kinases including PKR. Therefore, the catalytic regions of the HRI and GCNZ kinases have a larger overall size than those of the PKR-Hum and PKR-Mus enzymes (Fig. 2). In contrast to the catalytic regions of the eIF-2a kinases, the noncatalytic regions of the proteins differ significantly between PKR, HRI, and GCN2. Furthermore, the GCN2 kinase is a significantly larger protein than the PKR and HRI kinases, in part because of the larger regulatory region (HisRS), which shows significant homology with the histidyl-tRNA synthetases (15)(16)(17). Finally, the PKR, HRI, and GCNZ protein kinases are all phosphoproteins (6)(7)(8)(9)(10)(12)(13)(14)(15)(16)(17)(18)(19)(20)(21).
PKR-Molecular cDNA clones of PKR have been isolated from human Daudi (12) and U (13) cells, and from mouse pre-B 7021 3 (18) and FM3A (19) cells. The human PKR is a 551-amino acid protein with a molecular mass of about 62 kDa as deduced from the cDNA ORF (12,13). By contrast, the mouse PKR is a somewhat smaller protein of 518 amino acids with a molecular mass of about 59 kDa as deduced from the cDNA ORF (18). PKR is a ribosome-associated protein that can be dissociated from ribosomes by washing with high salt buffers (6,13,22). The sizes of the PKR cDNAs, about 2.4 kb, are comparable with the size 7603  of the major mRNA detected by Northern analysis, about 2.5 kb (12,13,18). The conserved catalytic subdomains characteristic of protein serinelthreonine kinases are, without exception, located in the C-terminal half of the PKR eIF-2a kinases (12,13,18). PKR kinases are activated by an RNA-dependent autophosphorylation (6,8). Substitution of the subdomain I1 invariant lysine residue 296 with arginine (K296R) eliminates autocatalytic activity and eIF-2a kinase activity associated with PKR (23)(24)(25). The Nterminal region of the PKR enzymes constitutes the regulatory domain (Fig. 2) and possesses the RNA binding activity (19,23,26,27). The multiple sites of phosphorylation on PKR-Hum (22) and PKR-Mus (18,22) proteins, predominantly serine residues but also including threonine, have not yet been identified. No PKR phosphorylation on tyrosine is detectable (18,22).
The RNA binding activity of PKR maps to a region that includes a repeated subdomain, R, the core of which is about 20 amino acid residues (26). This core sequence is conserved in several other proteins now known to be RNA binding proteins and represents a new kind of RNA binding domain (26,28). The core consensus motif is G-X-G-X-SIT-K-X-X-A-K-X-X-A-A-X-X-A-hydrophobic X-X-L. Although multiple copies of the R motif are found in many of the RNA binding proteins including PKR-Hum, PKR-Mus, TRBP-Hum, Staufen-Dm and RBPA-X1, only one copy is found in others including NS34-Rot, E3L-VV, PAC1-Ysc, and RNaseIII-Ec. Furthermore, mutational analysis of PKR has established that the N-terminal proximal copy of R is both necessary and sufficient for RNA binding activity3 (26,29). However, both copies of R appear required for optimal RNA binding activity (19,26,27,29). The significance of the fact that the two R region motifs of PKR appear to function as nonequivalent units is not yet clear.
HRI-The rabbit HRI kinase is a 626-amino acid protein with a molecular mass of about 70 kDa as deduced from the cDNA (14), although the apparent M, by gel electrophoresis is about 90 kDa ( Table I). The size of the HRI cDNA, about 2.7 kb (14), is in good agreement with the size of the single mRNA species detected by Northern analysis, about 2.8 kb (30).
The HRI kinase possesses all 11 of the conserved catalytic subdomains that are characteristic features of protein serine/ threonine kinases (14). Of particular uniqueness, however, is the insertion of about 140 amino acids between catalytic subdomains V and VI. The significance of this insert has not yet been established. HRI is present in heme-supplemented reticulocyte lysates as an inactive proinhibitor (10). The binding of hemin directly to purified heme-reversible HRI has been demonstrated. HRI is activated as an eIF-2a kinase by heme deficiency; the activation of HRI is accompanied by its autophosphorylation (31). Neither the site of hemin binding nor the site(s) of phosphorylation on HRI has yet been identified. GCN2-Two regions of the deduced 1590-amino acid sequence of the GCNZ protein display striking homology with known enzymes (Fig. 2). One region is highly homologous to the catalytic domain of protein serinelthreonine kinases; another region is homologous to histidyl-tRNA synthetases from Escherichia coli, Saccharomyces cereuisiae, and human cells (15)(16)(17). Mutation of the invariant lysine residue (Lys-559) of the conserved kinase catalytic subdomain I1 abolishes GCN2 kinase activity (21,32).
The K559R mutant GCNB protein is not able to carry out either the autophosphorylation reaction or the phosphorylation of eIF-2a in uitro (21,32). The adjacent HisRS-like domain of the GCNZ protein is dispensable for the catalytic function of the protein kinase domain in uitro, but the HisRS domain is required for positive regulatory functions of GCN2 in vivo (15,32). Finally, the GCNB protein is associated with ribosomes, possibly the 60 S subunit, and can be dissociated by washing with buffers containing 0.5 M KCl. The extreme C-terminal region of the GCN2 protein appears essential for its interaction with ribosomes (33).

Substrate of the eIF-2a Protein Kinases
The a subunit of protein synthesis initiation factor eIF-2 is the common substrate of PKR, HRI, and GCNB protein kinases (21,22,34). All three of the kinases phosphorylate eIF-Pa on serine residue 51; this phosphorylation mediates translational control (1,4). The amino acid sequence surrounding serine 51 is highly conserved among the eIF-2a factors from human, rat, and yeast. Between residues 41 and 59, their sequences are identical (35,36). Although PKR, HRI, and GCN2 display similar specificity with yeast and mammalian eIF-2a (21,22,34), some differences in specificity with peptide substrates are observed between the PKR and HRI enzymes (37). Two mutants of eIF-2a have been especially useful in the analysis of the functional importance of serine 51 phosphorylation in translational control (21,34,38). The S51A mutant of eIF-2a substitutes an alanine for the serine residue 51 and thus cannot be phosphorylated. The S51D mutant of eIF-La substitutes an aspartic acid for the serine residue 51; this substitution appears to functionally mimic a phosphorylated serine residue at position 51. eIF-2 forms a ternary complex, eIF-2.GTP.methionyl-tRNAi, which is an obligate intermediate in the binding of the initiator methionyl-tRNAi to the 40 S ribosomal subunit. The eIF-2 factor is subsequently released from the ribosome as an eIF-2.GDP complex. Under steady-state conditions of translation, eIF-2 must be recycled in order to participate in another round of translation initiation. Conversion of the inactive eIF-2.GDP to the active eIF-2. GTP form is catalyzed by a guanine nucleotide exchange factor, eIF-2B. Phosphorylation of the a subunit of eIF-2 inhibits translation initiation by impairing the eIF-2B-catalyzed guanine nucleotide exchange reaction (2,4).
The binding site on the kinases for the eIF-2 substrate has not yet been identified. However, more extensive homology exists around subdomains IX and X of the PKR, HRI, and GCN2 kinases than with most other protein serine/threonine kinases that do not phosphorylate eIF-2a (13,14,17). These regions may be involved in aspects of substrate binding and phosphorylation unique to the a subunit of eIF-2. A synthetic peptide, P-74, which contains residues conserved in subdomain IX, inhibits the eIF-2a kinase activity of HRI (30).

Regulation of the eIF-2a Protein Kinases
Inactivation of initiation factor eIF-2 function by serine 51 phosphorylation is an important translation control mechanism, both for total cellular protein synthesis as illustrated by HRI in reticulocytes (10) and for gene-selective translational control as illustrated by GCN2 in yeast (41, 42) and PKR in mammalian cells (8,43,44). Regulation of the level of eIF-2a protein kinase activity occurs by several different mechanisms.
PKR-PKR is regulated at the transcriptional level by interferon treatment, as measured by both Northern gel blot and nuclear run-on analyses (8,9,12,13). IFN-a is an efficient inducer of both the 2.5-and 6.0-kb mRNAs in human cells (12,13), but IFN-7 is a poor inducer (13). Tissue-specific differences in the ratios of the three PKR transcripts are observed in mice; for example, the 2.5-kb mRNA is the predominant species in testes, but in both lung and heart tissues the 4.0-kb species is predominant over the 2.5-and 6.0-kb mRNAs (18). Transcription of PKR is not rapidly down-regulated, in contrast to some other IFNinducible genes (12). However, the synthesis of PKR in transfected mammalian cells is autoregulated primarily at the level of translation by a mechanism that is likely dependent upon catalytically active PKR (25).
Activation of PKR is RNA-dependent but CAMP-independent (6,8,9,45,46,48). A number of effectors of the RNA-dependent activation of PKR have been identified (Fig. 3). Activators include both synthetic and natural double-stranded RNA, for example (rI),,-(rC),, and reovirus genome dsRNA, respectively. Certain natural single-stranded RNA species are also activators, for example human immunodeficiency virus (HIV) TAR RNA and reovirus s l mRNA. Inhibitors of the autophosphorylation activation of PKR include dsRNA at high concentration and three highly structured ssRNA species: adenovirus VAI RNA; Epstein-Barr virus EBER RNA; and HIV TAR RNA. RNA binding proteins that sequester activator RNAs, for example the reovirus a3 protein and the vaccinia virus E3L protein, also antagonize the PKR activation process (48). In addition, 2-aminopurine is a potent inhibitor of PKR, although cellular protein kinases in general are not inhibited by the purine analog (48). Some of the most direct evidence for the involvement of PKR in translational repression in mammalian cells emerges from the analysis of various RNA, protein, and purine inhibitors of kinase activity (8,39,46,48). The basis of the RNA selectivity of PKR activation is an important question. Kinase activation is associated with the formation of a stable PKR. dsRNA complex that requires at least 30-50 bp of duplex and is optimal with about 80 bp, although the PKR protein appears to interact with as little as 11 bp of dsRNA (47). Reovirus s l mRNA, an activator of P K R adenovirus VAI RNA, an inhibitor of PKR activation; and HIV TAR RNA and reovirus genome dsRNA all bind to the same R region of the PKR protein (23,26). Mutations in PKR that impair RNA binding have similar effects on the binding of both activator and inhibitor RNAs3 (29). This suggests that the discrimination between activator and inhibitor RNAs presumably takes place subsequent to RNA binding. Little is known concerning the potential association of PKR with itself or with other proteins that may regulate PKR localization to the ribosome and activity. Cell fractionation studies reveal that PKR is ribosome-associated (6,13,22); in situ immunofluorescent staining studies reveal that PKR largely localizes to the rough endoplasmic reticulum (50). A soluble, RNAindependent phosphatase constitutively present in animal cells catalyzes the dephosphorylation of PKR and eIF-2a (51,52). The phosphatase has been identified as a type I, manganese-dependent protein phosphatase (52). The ribosome-associated form of PKR appears to be a monomer (22,53), but the soluble form of PKR-Mus may be a partially phosphorylated homodimer (53). A 90-kDa protein of unknown function has also been described that complexes with VAI RNA and with PKR, and is phosphorylated by PKR (54).
PKR is an important component in the antiviral action of interferon (8,49). Viruses that have deleted from their genome genes that antagonize the action of PKR, for example adenovirus VAI RNA and vaccinia virus K3L, are reported to have an increased sensitivity to IFN relative to the wild-type virus (40,55). Constitutive expression of wild-type but not K296R mutant PKR-Hum in mouse cells mediates phosphorylation of eIF-2a and partial resistance to encephalomyocarditis virus growth (56).
It has been proposed that PKR may also function as a tumor suppressor. Expression of functionally defective PKR-Hum in mouse NIH 3T3 cells causes malignant transformation, and 3T3 cells overexpressing inactive PKR are highly tumorigenic when injected into nude mice (57,58). In contrast, 3T3 cells expressing the wild-type PKR-Hum or the endogenous PKR-Mus are not tumorigenic (57). The mechanism by which the mutant PKR proteins transform cells and wild-type PKR apparently suppresses the transformation and tumor formation is not yet clear.
HRZ-HRI is not ubiquitous. It may be erythroid-specific, because among several tissues examined from anemic rabbits, HRI was detected only in the reticulocytes and bone marrow (31). In hemin-deficient reticulocytes, inhibition of protein synthesis occurs as a result of activation of the HRI eIF-2a kinase (10). However, protein synthesis is maintained even under conditions of heme deficiency in HRI-depleted lysates (31). The regulation of protein synthesis by HRI is not restricted to globin but is a general mechanism controlling the synthesis of soluble and membrane-bound polypeptides in reticulocytes (10).
Activation of HRI in reticulocytes is mediated by various stimuli in addition to hemin deficiency, including heat shock, sulfhydryl reagents such as N-ethylmaleimide, oxidized glutathione, and heavy metal ions (10). Native HRI appears to be a dimer composed of two 90-kDa polypeptides that may be in part disulfide-linked (59). Binding of hemin to HRI promotes an intersubunit disulfide bond formation that may be involved in the negative regulation of HRI (60). Heat shock proteins hsp90 and hsp7O also interact with HRI (61). The interaction may be of regulatory significance. For example, hemin may regulate eIF-201 HRI kinase activity by promoting the formation of an inactive HRI. hsp90 (p87) dimer (62).
GCNZ"GCN2 is an eIF-2a protein kinase that plays a central roIe in the reguIation of amino acid biosynthesis in S. cereuisiae (7,42). Amino acid starvation of bakers' yeast results in the increased synthesis of a transcriptional activator (GCN4) required for the expression of a number of unlinked genes encoding amino acid biosynthetic enzymes. The GCNB eIF-201 protein kinase stimulates, at the level of translation initiation, the synthesis of the GCN4 activator (15,16). The increased synthesis of GCN4 is accomplished by overcoming the inhibitory effects of multiple short upstream ORFs present in the GCN4 mRNA leader (21,41).
The expression of the GCN2 protein itself does not appear to be increased under conditions of amino acid starvation (32). Likewise, although GCNB is a ribosome-associated protein, the binding of GCNB to ribosomes is not regulated by the availability of amino acids (33). Rather, the catalytic activity of the GCN2 kinase appears to be regulated in response to amino acid levels. The HisRS domain present within the C-terminal half of the GCNB protein regulates its catalytic activity (15,17).
Conceivably the HisRS-like domain of GCNZ monitors amino acid abundance by interacting with uncharged tRNA. Binding of uncharged tRNA to the HisRS domain of the GCNZ protein might activate the catalytic activity of the adjacent eIF-Za protein kinase domain via an allosteric change in the GCNZ protein.
Indeed, dominant GCNZ mutations that lead to derepression of GCN4 expression in the absence of amino acid starvation have been mapped to both the kinase domain and the HisRS-like domain of the GCNZ protein (15,17). Because starvation of S. cereuisiae for a single amino acid leads to the GCN2-dependent increased synthesis of GCN4, the HisRS-like domain of GCNZ presumably has diverged sufficiently in order to bind other uncharged tRNA species in addition to tRNAHi". Indeed, mutation of yeast lysyl-tRNA synthetase in a manner that causes reduced charging of tRNALys also results in the increased expression of GCN4 at the translational level; the uncharged tRNALy8 is postulated to stimulate the GCNB e 1 F -h protein kinase (63). Among the least conserved regions of the GCNZ HisRS-like domain are the extreme N-and C-terminal portions of the HisRS sequences (17), regions that are thought to have accessory functions specific for histidine (64).
When the human PKR kinase is expressed in S. cereuisiae, a growth suppression phenotype is produced resulting from an inhibition of translation initiation (65). When yeast GCNZ is overexpressed on a multiple-copy plasmid, a similar slow growth phenotype is observed (66). The slow growth phenotype mediated by PKR-Hum is reverted in yeast by two mechanisms involving perturbation of kinase function: expression of the N-terminal region of PKR-Hum that binds the activator dsRNA; and expression of the S51A mutant of the yeast eIF-2a that cannot be phosphorylated (19,65).

Perspective
The eIF-Za protein kinases are fundamentally important regulators of translation in eukaryotic cells. These kinases mediate gene-selective as well as general protein synthesis regulation. The nature of the translational repression is dependent upon the type of cell, the environmental stimulus, and the type of eIF-201 kinase mediating the regulation. Full-length cDNAs have recently been obtained and expressed for the three presently known types of eIF-Pa protein kinase, PKR, HRI, and GCN2. The availability of these clones provides an approach to generate mutant forms of the kinases. This, together with the diversity of organisms available for their combined genetic and biochemical analyses should help to unravel the structure-function relationships for the eIF-Za kinases. A number of laboratories are actively pursuing the identification and characterization of proteins and nucleic acids that may modulate kinase function; the growth of crystals as a step toward the elucidation of a 3-dimensional structure for an eIF-fa protein kinase; and the effects of knocking out as opposed to overexpression of the kinase in cells and animals. Such studies of the eIF-2a kinases will likely continue to provide us with many surprises.   . .