The Hsp90 Inhibitor Geldanamycin Abrogates Colocalization of eIF4E and eIF4E-Transporter into Stress Granules and Association of eIF4E with eIF4G*

The eukaryotic translation initiation factor eIF4E plays a critical role in the control of translation initiation through binding to the mRNA 5′ cap structure. eIF4E is also a component of processing bodies and stress granules, which are two types of cytoplasmic RNA granule in which translationally inactivated mRNAs accumulate. We found that treatment with the Hsp90 inhibitor geldanamycin leads to a substantial reduction in the number of HeLa cells that contain processing bodies. In contrast, stress granules are not disrupted but seem to be only partially affected by the inhibition of Hsp90. However, it is striking that eIF4E as well as its binding partner eIF4E transporter (4E-T), which mediates the import of eIF4E into the nucleus, are obviously lost from stress granules. Furthermore, the amount of eIF4G that is associated with the cap via eIF4E is reduced by geldanamycin treatment. Thus, the chaperone activity of Hsp90 probably contributes to the correct localization of eIF4E and 4E-T to stress granules and also to the interaction between eIF4E and eIF4G, both of which may be needed for eIF4E to acquire the physiological functionality that underlies the mechanism of translation initiation.

The eukaryotic translation initiation factor eIF4E plays a critical role in the control of translation initiation through binding to the mRNA 5 cap structure. eIF4E is also a component of processing bodies and stress granules, which are two types of cytoplasmic RNA granule in which translationally inactivated mRNAs accumulate. We found that treatment with the Hsp90 inhibitor geldanamycin leads to a substantial reduction in the number of HeLa cells that contain processing bodies. In contrast, stress granules are not disrupted but seem to be only partially affected by the inhibition of Hsp90. However, it is striking that eIF4E as well as its binding partner eIF4E transporter (4E-T), which mediates the import of eIF4E into the nucleus, are obviously lost from stress granules. Furthermore, the amount of eIF4G that is associated with the cap via eIF4E is reduced by geldanamycin treatment. Thus, the chaperone activity of Hsp90 probably contributes to the correct localization of eIF4E and 4E-T to stress granules and also to the interaction between eIF4E and eIF4G, both of which may be needed for eIF4E to acquire the physiological functionality that underlies the mechanism of translation initiation.
The molecular chaperone Hsp90 is required for the folding, assembly, and stability of an apparently limited subset of proteins (i.e. cellular signaling proteins that include protein kinases and transcription factors) (1)(2)(3)(4)(5)(6). Therefore, Hsp90 is not regarded as a general molecular chaperone, such as Hsp70 and chaperonin (7)(8)(9)(10)(11); however, the known substrates of Hsp90 continue to increase and to represent a broader range of protein types (see the list of Hsp90 interactors on the Didier Picard laboratory web site). In our previous study, we attempted to delineate the whole spectrum of Hsp90-interacting proteins. We disrupted the Hsp90␤ gene in chicken DT40 cells to lower the level of Hsp90 protein. Multifaceted defects, which included slow growth and the impairment of components involved in B cell antigen receptor signaling, were elicited as expected (12). In particular, the expression level of the immunoglobulin M heavy chain was reduced profoundly, and we suspected that this was due to translational repression because the transcript level and protein turnover rate were normal (12). In line with this finding, interactome analysis of an Hsp90 cochaperone Cdc37 by mass spectrometry identified many proteins that are involved in mRNA biogenesis; they included proteins involved in translation (e.g. eIF4G, eEF2, and ribosomal protein S27) (13).
Translation is an important stage in the regulation of eukaryotic gene expression and is mostly controlled at the rate-limiting step of initiation (14,15). The eukaryotic translation initiation factor complex eIF4F is composed of three proteins in mammalian systems: the cap-binding protein eIF4E, the scaffold protein eIF4G, and the ATP-dependent RNA helicase eIF4A (14,15). The assembly of eIF4F begins with the binding of eIF4E to the 5Ј cap structure (m 7 GpppN) of the mRNA, and then eIF4G is recruited. eIF4G can interact simultaneously with eIF4E, eIF4A, and the ribosome-associated initiation factor eIF3; therefore, it acts as a bridge between the mRNA and the ribosome.
Although eIF4E is localized predominantly in the cytoplasm, a certain fraction is found in the nucleus, where it colocalizes with splicing factors in speckles (16). The nuclear import of eIF4E, which occurs via the importin ␣␤ pathway, is mediated by a nucleocytoplasmic shuttling protein eIF4E transporter (4E-T) 5 (17). 4E-T binds to eIF4E through a conserved binding motif (YXXXXL⌽, where ⌽ is Leu, Met, or Phe, and X is any amino acid) that is also found in eIF4G and the family of translational suppressors known as eIF4E-binding proteins (4E-BPs) (17,18). Translation of mRNAs that require the eIF4A-driven unwinding machinery is inhibited by sequestration of eIF4E by reversible binding to the 4E-BPs (15, 19 -22).
The control of mRNA turnover also plays a part in the regulation of eukaryotic gene expression. mRNAs are monitored by a quality control system in the cytoplasm, and they are triaged between translation and repression/degradation. Many mRNAs that are not being actively translated accumulate in discrete cytoplasmic domains referred to as processing bodies (P-bodies) (23)(24)(25)(26)(27). These translationally inactive mRNAs colocalize with the translational repression and mRNA decay machinery in P-bodies; thus, the mRNAs are either stored for return to translation or degraded. P-bodies contain proteins that are required for mRNA decay in the 5Ј to 3Ј direction: the decapping enzyme Dcp1/Dcp2; the activators of decapping Dhh1p/ RCK/p54, Pat1p, and the Lsm1-7 complex; and the 5Ј to 3Ј exonuclease Xrn1 (23)(24)(25)(26)(27). In addition, the components of P-bodies include several proteins that are involved in mRNA surveillance, RNA interference, and translational repression, such as Argonaute, GW182, Scd6p/RAP55, and 4E-T (23)(24)(25)(26)(27). RCK and Pat1p, mentioned above, reportedly have a role in general translational repression (28 -31). Therefore, they act as both repressors of translation and activators of mRNA decapping.
When cells are subjected to multiple stresses, such as heat shock, oxidative stress, or viral infection, the expression of stress proteins is selectively induced (7)(8)(9)(10)(11). In contrast, the translation of mRNAs that encode constitutively expressed proteins is abrogated, and the translationally inactive mRNAs are redirected to another class of RNA granule, the stress granule, that is assembled in the cytoplasm (23,32,33). Stalled 48 S preinitiation complexes are the core components of stress granules, which contain small but not large ribosomal subunits in addition to the early translation initiation factors eIF2, eIF3, eIF4E, and eIF4G. In addition, stress granules contain the poly(A)-binding protein, RCK/p54, RAP55, and many RNAbinding proteins (e.g. TIA-1, TIA-R, and G3BP) (23,(31)(32)(33).
P-bodies and stress granules are closely related in that both are assembled on translationally inactive mRNAs, and they contain a subset of shared proteins (25,32,33). Furthermore, P-bodies and stress granules are linked spatially and functionally. Although the directionality of mRNA flux between P-bodies and stress granules has not been established (34,35), it is speculated that mRNAs and associated proteins are exchanged between the two RNA granules according to decisions on the fates of specific mRNAs (25,32,33). In the dynamic process of messenger ribonucleoprotein remodeling, P-bodies serve as the machinery for mRNA decay, whereas stress granules act as sites en route for reentry into translation (25,32,33). However, as described above, some protein components of P-bodies are also implicated in translational repression (23)(24)(25)(26)(27), and mRNAs that are associated with P-bodies may reenter translation either directly or through stress granules (25,(32)(33)(34)(35).
Given that our previous data suggest that depletion of Hsp90␤ causes translational repression of the immunoglobulin M heavy chain mRNA (12), we became interested in the possible role of Hsp90 in translational control. In the study described herein, we examined whether abrogation of Hsp90 activity would affect the formation of P-bodies and stress granules. We show that geldanamycin (GA) treatment resulted in a substantial reduction in the number of HeLa cells that contained P-bodies. Stress granules were also affected by this treatment, although to a lesser degree, and eIF4E and 4E-T, which are components of stress granules, specifically dissociated from these granules. Moreover, treatment with GA weakened the physical interaction of eIF4E with eIF4G. We conclude that the chaperone activity of Hsp90 is likely to be required for the activities of eIF4E that are necessary for its physiological function.

EXPERIMENTAL PROCEDURES
Cell Culture and Immunocytochemistry-HeLa S3 cells were maintained as described previously (36). Cells that had been grown on coverslips in 24-well plates were treated with 2 M GA (Sigma) or dimethyl sulfoxide for 24 h and fixed for immunocytochemistry. For heat stress, the 24-well plates were floated in a water bath in a CO 2 incubator at 44°C for 30 min just before fixation (37). Immunocytochemistry was performed as described previously (31,38). Scoring of P-bodies was performed in a blind manner with a minimum of ϳ100 cells scored.
Plasmids-An expression plasmid for FLAG/hemagglutinintagged Argonaute 2 (Ago2; Addgene plasmid 10822) was supplied by Dr. Thomas Tuschl (The Rockefeller University, New York). To produce an expression construct for Myc-tagged eIF4E, the coding region of eIF4E was amplified by PCR from a human cDNA plasmid (Open Biosystems; clone ID 5295521); a BamHI site was added at the 5Ј-end, and an XhoI site was added after the stop codon (TAA) at the 3Ј-end. The cDNA fragment was inserted into pcDNA3Myc1 plasmid DNA (13) that had been cut with both BamHI and XhoI. The construct obtained was verified by DNA sequencing.
Phosphatase Treatment-Cells were lysed in protein phosphatase buffer (New England Biolabs) supplemented with 0.1% Nonidet P-40 and Complete protease inhibitor mixture and then incubated with phosphatase as described (13).
Transfection and Immunoprecipitation-Cells that had been grown in a 100-mm dish were transfected with 8 g of plasmid DNA using Lipofectamine 2000 (Invitrogen), following the manufacturer's recommendations. Two days later, the cells were lysed in Tween 20 lysis buffer (50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1 mM EDTA, 0.2% Tween 20, 10 mM NaF, and Complete protease inhibitor mixture). Cell lysates were precleared with Protein G-Sepharose (GE Healthcare) and then incubated for 2 h with antibodies that had been prebound to Protein G-Sepharose. The beads were collected by centrifugation and washed three times with the lysis buffer.
Binding to m 7 GTP-Sepharose 4B-Cell lysis and binding to m 7 GTP-resin were performed according to the method described by Pyronnet et al. (40) with slight modifications. Briefly, the cells were lysed in buffer A (50 mM Tris-HCl, pH 7.5, 50 mM KCl, 1 mM EDTA, 0.5% Nonidet P-40, and Complete protease inhibitor mixture), and the protein in the resulting cell extracts was quantified using the BCA protein assay kit (Thermo Scientific). Aliquots of the extracts that contained 650 g of protein were incubated with 25 l of m 7 GTP Sepharose 4B resin (GE Healthcare) for 3 h at 4°C. The resin was washed three times with 1 ml of buffer A.

P-Bodies
Are Decreased by GA Treatment-To investigate further our previous finding that depletion of Hsp90␤ causes translational repression of immunoglobulin M heavy chain mRNA (12), we attempted to reproduce this effect, which initially was induced by gene disruption, by treating HeLa cells with the Hsp90 inhibitor GA. GA blocks the essential nucleotide binding site of Hsp90 and thus inhibits its ATPase activity, which is required for chaperone function (1)(2)(3)(4)(5)(6)41). It is known that Hsp90 co-immunoprecipitates with Ago2 and that treatment with GA reduces the amount of Ago2 protein in cells (42,43). These data suggest that Ago2 is a substrate of Hsp90; GA often induces the degradation of Hsp90 substrates (12,44). Initially, to establish appropriate conditions for the GA treatment, the concentration of GA and the duration of treatment were varied, and the efficacy of the different conditions was evaluated in terms of the reduction of the amount of Ago2 protein.
Ago2 was decreased profoundly in HeLa cells that had been treated with GA at 2 M for 24 h (Fig. 1A, left). This effect was reproduced with exogenously expressed FLAG/hemagglutinintagged Ago2 in HeLa cells (Fig. 1A, right). On the basis of these observations, we treated cells with 2 M GA for 24 h throughout this study, unless otherwise mentioned.
We then examined the subcellular localization of Ago2. Immunofluorescence showed that, in untreated cells, Ago2 was localized in the cytoplasm and concentrated in discrete foci, as reported previously (45)(46)(47) (Fig. 1B, GA Ϫ). To confirm that these cytoplasmic foci were P-bodies, we performed co-immunostaining with an antibody against the decapping enzyme Dcp1a, which is one of the conserved signature components of the P-body (23)(24)(25)(26)(27). Dcp1a clearly colocalized to the Ago2-containing foci in untreated cells, which indicated that these foci were P-bodies. Interestingly, in the majority of GA-treated cells, P-bodies were largely undetectable with the antibody against Dcp1a, and Dcp1a was distributed diffusely in the cytoplasm (Fig. 1B, GA ϩ). In addition, Ago2 disappeared from P-bodies (Fig. 1B, GA ϩ), which appeared to reflect the reduction of Ago2 in GA-treated cells shown above (Fig. 1A).
Given that neither Dcp1a nor Ago2 is required for the formation of P-bodies (48,49), it was unclear whether these results represented the disappearance of microscopically visible Pbodies per se or the loss of the two components from P-bodies that still existed. To distinguish between these two possibilities, we performed immunofluorescence analysis with antibodies against 4E-T (Fig. 2, A and B), RAP55 (Fig. 2, A and C), and RCK (data not shown), which are all essential for P-body formation (29,31,48,50). As shown in Fig. 2A by the staining of 4E-T and RAP55, GA-treated cells contained fewer P-bodies than the control cells, which supported the possibility that P-bodies per se had disappeared. Furthermore, co-immunostaining of 4E-T and its partner eIF4E, which is also present in P-bodies (23)(24)(25)(26)(27), revealed that the cytoplasmic foci in which 4E-T and eIF4E colocalized were reduced substantially by GA treatment (Fig.  2B). This observation is consistent with the previous finding that 4E-T is a prerequisite for the association of eIF4E with P-bodies (50). Thus, we conclude that the number of P-bodies per se was decreased by treatment with GA.
Next, we investigated the changes of P-bodies that occurred during the time course of GA treatment. Treatment of cells with GA for only 6 h resulted in a significant reduction in the number of P-bodies visualized by staining with anti-RAP55 and anti-Dcp1a antibodies compared with that in control cells, which had not been exposed to GA (Fig. 2C, GA 6 h). However, the reduction in P-bodies was more prominent, and the remaining P-bodies appeared very faint, at 24 h after the initiation of GA treatment (Fig. 2C, GA 24 h). For quantitative evaluation, cells with foci that were revealed by 4E-T staining ( Fig.  2A) or colocalization of eIF4E/Dcp1a (data not shown) were scored. The percentage of cells that contained foci was decreased by GA treatment from 93% (113 positive cells per 121 total cells) to 49% (117 of 237 cells) and from 90% (155 of 172 cells) to 51% (92 of 181 cells), respectively; thus, the number of cells that contained P-bodies decreased to ϳ50% upon GA treatment for 24 h. In addition, similar observations were made using a different Hsp90 inhibitor, radicicol, 6 which is distinct from GA in structure (41). Therefore, the reduction in P-bodies could be induced by two distinct inhibitors of Hsp90. Taken together, it is highly plausible that the impairment of Hsp90 activity is responsible for the reduction in P-bodies in cells treated with these drugs. 7 It has been reported that GA induces mitotic arrest in HeLa cells (52), and in addition, P-bodies are known to disassemble prior to mitosis (53). These two facts suggest that the decrease in P-bodies after GA treatment may be due merely to blocking of the cell cycle of HeLa cells at the G 2 /M boundary. To address this issue, cells were synchronized at G 1 /S using a double thymidine block and exposed to 10 M GA (this concentration meets the dual requirements of minimizing the duration of treatment and preventing damage to cells) for 6 h (this duration is insufficient to allow entry into the next M phase even if the cells are released from the block) with or without release from the block. Under these conditions, we also observed a reduction in the number of cells that contained P-bodies (data not shown). Therefore, we have demonstrated conclusively that the decrease in P-bodies after GA treatment is not caused by mitotic arrest of the cells.
Both eIF4E and 4E-T Dissociate from Stress Granules in GAtreated Cells-Given that P-bodies and stress granules are linked functionally, as described above (25,32,33), it is rational to speculate that stress granules could also be affected by GA treatment. Consequently, to gain more insight into the above results, we attempted to investigate stress granules in GA-treated cells. We chose to induce stress granules by heat shock treatment.
First, we analyzed stress granules by staining with anti-eIF4E and anti-YB-1 antibodies. YB-1 is detected in P-bodies at 37°C (data not shown), but it localizes predominantly to stress granules upon heat shock (Fig. 3A, GA Ϫ) (54). We observed that eIF4E was detected into P-bodies, in addition to stress granules, in heat-shocked cells, whereas the P-bodies were only weakly stained with anti-YB-1 antibodies ( Fig. 3B; a representative P-body is shown in the bottom panel (arrowheads)). This observation agrees with a previous report (54) and may reflect the role of YB-1 in translational repression (55). In many of the cells that were not treated with GA, stress granules formed in the perinuclear region and eIF4E localized to these granules (Fig. 3, A and B, GA Ϫ). In the cells that were exposed to GA, stress granules still assembled upon heat shock, but they were slightly smaller and more dispersed in the cytoplasm as compared with the untreated cells (Fig.  3A); this difference is more evident in zoomed images (Fig. 3B). Notably, eIF4E did not localize significantly to stress granules that were stained by anti-YB-1 antibodies in GA-treated cells (Fig. 3, A and B). The zoomed images clearly demonstrate that eIF4E did not colocalize with the stress granules that contained YB-1 (Fig. 3B, GA ϩ, Merge). We also analyzed the localization of RAP55, which is a common component of both P-bodies and stress granules (31), and found that RAP55 was present in stress granules in GA-treated cells (data not shown). These results imply that stress granules do not contain eIF4E after treatment with GA, although stress granules are formed in these cells.
Second, the cells were stained with anti-eIF4E and anti-4E-T antibodies. We found that not only eIF4E but also 4E-T was recruited to stress granules (Fig. 3C, GA Ϫ); therefore, 4E-T is a genuine component of stress granules, which is a fact that has not been described previously. In addition, in cells exposed to GA, both eIF4E and 4E-T were mostly undetectable in stress granules (Fig. 3C, GA ϩ), which indicated that 4E-T as well as eIF4E is lost from stress granules after treatment with GA.
These observations suggest that the deterioration of Hsp90 function partially affects the formation and/or stability of stress granules per se and that some components, such as eIF4E and its binding partner 4E-T, can no longer accumulate in the stress granules that remain in GA-treated cells. This finding presumably indicates that stress granules can be assembled even in the absence of P-bodies, which was consistent with a previous report (35). Furthermore, it should be emphasized that the Hsp90 inhibitor GA serves as a useful tool to analyze the differences between the mechanisms of assembly of stress granules and P-bodies. 7 During the preparation of this paper, Pare et al. (51) reported that GA treatment of HeLa cells resulted in impairment of the association of Ago2 with stress granules (but did not affect their formation) and the biogenesis and/or stability of P-bodies. The Physical Interaction of eIF4E with eIF4G Is Diminished by GA Treatment-We suspected that either eIF4E or 4E-T or both may be a direct target(s) of GA treatment (namely an Hsp90 client protein(s)). First, we analyzed the binding of Hsp90 to eIF4E and 4E-T by immunoprecipitation. Given that some Hsp90 client proteins, such as steroid hormone receptors, associate tightly with Hsp90 during their biogenesis, immunoprecipitation can reveal the ability of these proteins to bind Hsp90 (1-6). However, this was not the case for eIF4E or 4E-T (data not shown).
Next, we examined whether the levels of eIF4E and/or 4E-T were decreased by GA treatment. As mentioned above, Hsp90 client proteins are often susceptible to degradation induced by GA (12,44), because client proteins that are folded imperfectly due to a lack of Hsp90 chaperone activity are thought to be targeted for degradation by the ubiquitin-proteasome system (10,56,57). In the case of eIF4E, its level was not altered by treatment with GA (Fig. 4A). In contrast, 4E-T in the GAtreated cells appeared as a doublet band, although the total amount of protein in the doublet appeared to be comparable with the amount of 4E-T in the control cells (Fig. 4A), which was quantitatively ascertained by densitometry using ImageJ software (data not shown). The appearance of this doublet may be ascribed to post-translational modification of 4E-T, and phosphorylation is the most probable modification (58). 4E-T migrated as a doublet despite the inclusion of NaF, a broad spectrum serine/threonine phosphatase inhibitor, in the lysis buffer. Therefore, the doublet was unlikely to be caused by dephosphorylation during cell lysis. We treated the cell lysates prepared from GA-treated cells with phosphatase and found that the lower band of the doublet corresponded to the unphosphorylated form of 4E-T (Fig. 4B). Therefore, we demonstrated   DECEMBER 18, 2009 • VOLUME 284 • NUMBER 51 that the levels of eIF4E and 4E-T were largely unaffected by treatment with GA, but the phosphorylation of 4E-T was diminished.

An Hsp90 Inhibitor Abrogates eIF4E Activities
Given that the largest group of Hsp90 client proteins corresponds to protein kinases (1-6) (see the list of Hsp90 interactors on the Didier Picard laboratory web site), it is conceivable that abolishing Hsp90 activity directly or indirectly inactivates an as yet undiscovered protein kinase that is responsible for the phosphorylation of 4E-T. Prolonged hypoxia results in the subcellular redistribution of eIF4E and 4E-T, and this redistribution is coincident with the gradual dephosphorylation of 4E-T (58). These results suggest that dephosphorylated 4E-T binds eIF4E more strongly than the phosphorylated form (58). Therefore, a possibility arises that 4E-T that has been unphosphorylated by treatment with GA interacts more strongly with eIF4E, and this results in the redistribution of eIF4E, as shown by the dissociation from stress granules (Fig. 3C).
To test this possibility (i.e. whether the interaction between eIF4E and 4E-T is strengthened by GA treatment), cells were transfected with an expression plasmid for Myc-tagged human eIF4E, and the endogenous 4E-T that co-immunoprecipitated with Myc-tagged eIF4E in GA-and mock-treated cells was compared by Western blot analysis. As shown in Fig. 4C, the total amount of co-immunoprecipitated 4E-T was comparable between mock-and GA-treated cells. Therefore, treatment with GA did not enhance the interaction between eIF4E and 4E-T, although the relative amount of unphosphorylated 4E-T was increased by GA treatment (Fig. 4C, Input). Furthermore, it appears unlikely that the unphosphorylated form of 4E-T is bound more strongly to eIF4E than the phosphorylated form. Thus, the increase in unphosphorylated 4E-T on its own does not influence the interaction of 4E-T with eIF4E. The possible association between 4E-T unphosphorylation upon GA treatment and the decrease in P-bodies and/or delocalization of eIF4E and 4E-T from stress granules remains to be elucidated.
Subsequently, we focused upon the function of eIF4E, in particular its cap-binding activity, because this activity is considered to be fundamental for eIF4E to accomplish its physiological roles in both translation initiation and P-body formation (14,15,(23)(24)(25)(26)(27). As shown in Fig. 5A, the ability of eIF4E to bind the 5Ј cap structure was unchanged by treatment with GA. Next, we investigated whether the localization of eIF4G in stress granules was altered. eIF4G is another known component of stress granules (23,32,33), and the binding of eIF4G to eIF4E is involved in the formation of an eIF4F complex in mammalian cells (14,15). We found that the accumulation of eIF4G in stress granules was somewhat reduced in GA-treated cells; however, the decrease was much less obvious than that of eIF4E (Fig. 5B). Thus, although eIF4E was lost to a significant degree from stress granules in GA-treated cells, the majority of eIF4G remained in the stress granules, which suggested that the interaction between eIF4E and eIF4G might be reduced by treatment with GA.
To address this issue, we examined the interaction of eIF4E and eIF4G using m 7 GTP-Sepharose and cell lysates that had been prepared from cells exposed to heat shock at 44°C in the presence or absence of GA. As shown in Fig. 5C, eIF4E prepared from the heat-shocked cells retained the capability to bind to the mRNA 5Ј cap regardless of GA treatment. This may imply that the binding of eIF4E to the 5Ј cap is not solely responsible for the association of eIF4E with stress granules. However, substantially less eIF4G was bound to the resin in lysates prepared from GA-treated cells than in lysates prepared from control cells. We observed an upward shift of the protein bands only for eIF4G bound to the resin (Fig. 5C, Bound), which was consist-FIGURE 5. The interaction between eIF4E and eIF4G is reduced by GA treatment. A, Lysates prepared from untreated (-) or GA-treated (ϩ) cells were incubated with m 7 GTP resin. Input (Input) and equivalent amounts (relative to the input material) of each bound (Bound) and flow-through (FT) fraction were subjected to Western blotting with the indicated antibodies. Actin serves as a loading control. B and C, untreated (Ϫ) or GA-treated (ϩ) cells were exposed to heat shock. The cells were then stained with anti-eIF4G (green) and anti-eIF4E (red) antibodies. Merged images (Merge) with TO-PRO-3 staining (blue) are also shown (B). The corresponding cell lysates were subjected to m 7 GTP resin chromatography as in A, and input (Input) and equivalent amounts of each bound (Bound) and flow-through (FT) fraction were subjected to Western blotting with the indicated antibodies except for 4E-BP1, where a 5 times greater amount than the input fraction was applied for the bound fraction (C). D, lysates prepared from the heat-shocked cells were either mock-treated (Phosphatase Ϫ) or treated with protein phosphatase (Phosphatase ϩ) and subjected to Western blotting with antibodies against 4E-BP1. ent with a previous report (58). These data imply that eIF4E could no longer localize to (or remain in) stress granules upon treatment with GA despite the retention of cap-binding ability, at least in vitro. GA treatment reduced the physical interaction between eIF4E and eIF4G (Fig. 5C), which suggested that the molecular integrity of eIF4G may be affected by the deterioration of Hsp90 chaperone activity that occurs upon treatment with GA. In this context, it is noteworthy that eIF4G is included in the list of proteins that interact with the Hsp90 co-chaperone Cdc37 (13). As can be seen in Fig. 5C, Input, the amount of eIF4G was not necessarily decreased by GA treatment. We performed immunoprecipitation with anti-eIF4G antibodies to examine its binding to Hsp90 in the presence and absence of GA and observed no difference in Hsp90 binding between the two conditions (data not shown).
When cells are exposed to stresses, such as amino acid starvation or hypoxia, hypophosphorylated 4E-BPs compete with eIF4G for the binding to eIF4E and prevent interaction between eIF4E and eIF4G (40,58), because 4E-BPs and eIF4G share a conserved eIF4E recognition motif (15,21). Therefore, it may be that GA treatment abrogated a particular protein kinase that is responsible for phosphorylation of 4E-BPs and thereby increased the hypophosphorylated form of 4E-BPs, which led to a reduction in the eIF4E-associated eIF4G. However, as shown in Fig. 5C, the amount of 4E-BP1 bound to the cap resin was not changed between control and GA-treated cells. Furthermore, the more slowly migrating bands of 4E-BP1 increased in intensity in GA-treated cells (Fig. 5C, Input, GA ϩ), which could be attributed to hyperphosphorylation of 4E-BP1. To examine this presumption, we treated the cell lysates with phosphatase and found that the upward shift of 4E-BP1 was indeed a result of hyperphosphorylation (Fig. 5D). This finding is contrary to the aforementioned anticipation, although the reason why GA treatment enhanced the phosphorylation of 4E-BP1 is uncertain. Our data clearly show that treatment with GA impairs both the colocalization of eIF4E with 4E-T to stress granules and its association with eIF4G, both of which are presumably important for the physiological functionality of eIF4E.

DISCUSSION
Hsp90 is known to act as a molecular chaperone for a plethora of signaling molecules, in particular protein kinases, and thereby influences multiple aspects of cellular function (1-6). In fact, multiple defects occurred in avian DT40 cells in which Hsp90␤ was depleted (12). At the start of the present study, we envisaged that GA would abolish Hsp90 activity in a manner similar to the gene disruption and therefore would cause the translational repression that we observed in our previous study (12) (see Introduction). Hsp90 is associated with Ago2 and is involved in the stability of this protein (42,43). In addition, Ago2 is essential to microRNA-dependent translational repression and therefore is an important component of P-bodies (59 -61). Thus, these earlier findings also prompted us to investigate the possible role of Hsp90 in P-body formation. We report here that treatment with GA leads to a decrease in the number of P-bodies ( Figs. 1 and 2).
Given that Hsp90 acts as a chaperone for a variety of protein kinases, the results obtained may be accounted for by the dete-rioration of kinase(s) that are responsible for P-body formation and/or maintenance. However, this notion is apparently incompatible with the current prevailing model for the mechanism by which the rate of translation initiation is controlled in mammalian cells, which is dependent on recognition of the mRNA 5Ј cap by eIF4E (14,15,21,22). If one assumes that GA treatment abrogates the activity of the most critical kinase in this signaling pathway, mammalian target of rapamycin (14,15), the eIF4E-eIF4G interaction would be prevented, and this would culminate in the repression of cap-dependent translation. Thus, it may be concluded that P-bodies are dispensable for the repression of translation. 4E-T is an essential component of P-bodies (48,50), although the molecular mechanism by which it is involved in P-body assembly has not yet been elucidated. Therefore, loss of the ability of 4E-T to localize to P-bodies may be sufficient to reduce the number of P-bodies. In this context, it is of interest that 4E-T exhibits a doublet band in cells treated with GA (Fig. 4), where the lower band represents unphosphorylated 4E-T. This altered phosphorylation state of 4E-T may be attributable to the reduced activity of an as yet unidentified protein kinase, which would presumably be induced by GA treatment. It remains to be determined whether the unphosphorylated form of 4E-T is responsible for the observed decrease in P-bodies.
Alternatively, Hsp90 may function directly as a molecular chaperone for a component(s) that facilitates P-body assembly. Our current understanding of the molecular mechanisms that underlie P-body formation suggests that deadenylated mRNAs are assembled into inactive messenger ribonucleoproteins with various mRNA decay factors and/or translational repressors, and these proteins often contain domains that mediate proteinprotein interactions and thereby enable multiple messenger ribonucleoproteins to aggregate into microscopically visible P-bodies (27,62,63). According to this model, our finding may imply that the molecular chaperone activity of Hsp90 is involved in the aggregation process.
Treatment with GA also affected the integrity of another type of cytoplasmic RNA granule that contains translationally inactive mRNAs, the stress granule. Stress granules per se were not disrupted to the same extent as P-bodies; thus, despite the absence of P-bodies, stress granules were able to assemble. However, more detailed observations showed that the size and subcellular localization of the stress granules were altered by treatment with GA (Fig. 3). Moreover, the amounts of eIF4E and 4E-T (which was identified as a component of stress granules in this study) in stress granules were reduced markedly in the GA-treated cells (Fig. 3). In this regard, it is noteworthy that mutated 4E-T that is defective in eIF4E binding still localizes to P-bodies, but mutated eIF4E that lacks the 4E-T binding site does not (50). Therefore, the loss of 4E-T from stress granules may trigger the corresponding loss of eIF4E by a mechanism that is as yet unknown but similar to that which results in the reduction in P-bodies. It would be interesting to determine whether the loss of eIF4E and 4E-T is related to the observed alterations in the size and localization of stress granules upon treatment with GA.
We found that GA treatment diminished the ability of eIF4E to interact physically with eIF4G (Fig. 5C). Furthermore, we found that eIF4E retained its cap-binding activity after GA treatment (Fig. 5, A and C), although eIF4E was largely lost from stress granules (Figs. 3C and 5B). One possible explanation could be that although the cap-binding of eIF4E is a prerequisite for its accumulation in stress granules, other factors are involved in its stable localization to stress granules. In fact, stress granules are highly dynamic structures, and many components shuttle rapidly in and out of them (34); therefore, for instance, association with eIF4G may be necessary to anchor eIF4E in stress granules. Our previous finding suggested that eIF4G is a possible Hsp90 client protein (13); thus, deterioration of eIF4G after GA treatment may be responsible for the diminished interaction between eIF4E and eIF4G.