Translational Regulation of Hsp90 mRNA

Heat shock in Drosophila results in repression of most normal (non-heat shock) mRNA translation and the preferential translation of the heat shock mRNAs. The sequence elements that confer preferential translation have been localized to the 5′-untranslated region (5′-UTR) for Hsp22 and Hsp70 mRNAs (in Drosophila). Hsp90 mRNA is unique among the heat shock mRNAs in having extensive secondary structure in its 5′-UTR and being abundantly represented in the non-heat shocked cell. In this study, we show that Hsp90 mRNA translation is inefficient at normal growth temperature, and substantially activated by heat shock. Its preferential translation is not based on an IRES-mediated translation pathway, because overexpression of eIF4E-BP inhibits its translation (and the translation of Hsp70 mRNA). The ability of Hsp90 mRNA to be preferentially translated is conferred by its 5′-UTR, but, in contrast to Hsp22 and -70, is primarily influenced by nucleotides close to the AUG initiation codon. We present a model to account for Hsp90 mRNA translation, incorporating results indicating that heat shock inhibits eIF4F activity, and that Hsp90 mRNA translation is sensitive to eIF4F inactivation.

Stressful circumstances cause cellular and physiological damage, which when severe can lead to apoptosis and death (1,2). All cells and organisms have developed responses that enhance their survival following stress. At the cellular level most metabolic processes are repressed by heat stress (3), likely to prevent the accumulation of damaged molecules that could irrevocably compromise cellular function. Concurrently, large amounts of a small group of proteins are newly synthesized. These proteins are termed the heat stress, or simply stress, proteins (Hsps), 1 and function to prevent ongoing protein damage, to restore the activity of stress-damaged proteins, and to create a stress-resistant state to ameliorate future stress-based protein injury.
Because stress inhibits gene expression at transcriptional and post-transcriptional steps, the induction of stress protein biosynthesis requires unique mechanisms to evade the general metabolic inhibition. A well characterized transcriptional response activates a latent transcription factor, heat shock transcription factor, which binds to conserved heat shock element sequences in the promoter of the Hsp genes and results in rapid, highly efficient transcription (4). Unique mechanisms also allow Hsp mRNAs to exit the nucleus, whereas the normal processing and transport of mRNAs is blocked (5).
In the cytoplasm, the Hsp mRNAs are efficiently translated. For example, polysome analysis of Hsp70 mRNA translation suggests that ribosome loading is near maximal (6). Concurrently, the non-heat shock, or normal, mRNAs are virtually excluded from translation, although they are neither degraded nor physically inactivated by any nucleotide modification (7). The basis for this translational discrimination has been extensively investigated, principally in Drosophila, because the extent of mRNA discrimination and preferential translation is accentuated in this poikilothermic organism. Most of the studies have focused on Hsp70 mRNA, in part because Hsp70 is the most abundantly synthesized Hsp, by about an order of magnitude. However, it has been noted that virtually all the Drosophila Hsp mRNAs share several common features (8,9), which have been logical candidates as the regulatory feature(s) conferring their concurrent preferential translation (reviewed in Ref. 7).
First, the 5Ј-untranslated region (5Ј-UTR) of Hsp mRNAs is sufficient to confer efficient translation to a heterologous appended coding region (10,11); and conversely, the Hsp coding sequence and 3Ј-UTR are unable to be translated during heat shock when a non-heat shock 5Ј-UTR, or a mutationally disabled Hsp 5Ј-UTR, precedes it (12,13). Second, the common features found in virtually all Drosophila Hsp mRNA 5Ј-UTRs include: (i) long length (200 -250 nucleotides); (ii) two conserved sequence segments, and positionally conserved nucleotides within the initial element; (iii) a high frequency of adenosine nucleotides (ϳ50%); and (iv) a minimal extent of secondary structure (7). Investigations have been carried out to determine whether any or all of these features are necessary or sufficient for preferential translation.
Long length per se is not required, because ϳ170 nucleotides can be deleted with only a modest (30 -50%) reduction in preferential translation (12). The remaining translational activity is still Ͼ10-fold higher than non-heat shock mRNA translation. Both conserved elements can be deleted, with no significant decrement in heat shock translation (12). Neither high adenosine content nor a paucity of secondary structure is sufficient to confer preferential translation, because scrambling the order of a tract of nucleotides abolishes translation while retaining adenosine content and minimal structure (14). On the other hand, the lack of structure is required. The introduction of a modestly stable stem into the 5Ј-UTR of the Hsp70 mRNA causes no reduction in translation under normal, non-heat shock conditions, but virtually abolishes preferential translation during heat shock (13).
In this study we have initiated an investigation into the mechanism of Hsp90 mRNA translation during heat shock.
The results indicate that this mRNA possesses several unique features that suggest that its translation, and especially its preferential translation during heat shock, occurs by a mechanism that distinguishes it from the other major Drosophila Hsp mRNAs. To date, the prevailing perspective has been that all Hsp mRNAs achieve preferential translation by a common mechanism, but this conclusion has been based on studies on a group of Hsp mRNAs (e.g. Hsp70 mRNA, Hsp22 mRNA) with properties distinct from Hsp90 mRNA.
Hsp90 is an abundant protein in Drosophila cells grown at their normal temperature. It plays multiple roles in protein folding, maturation, and the regulation of protein activities (15). Its mRNA is the only Hsp mRNA present in significant quantities in the non-stressed Drosophila cell (16). Lines of evidence that suggest that its translation is uniquely regulated include: first, whereas virtually all major Hsp mRNAs lack significant 5Ј-UTR secondary structure, the 5Ј-UTR of Hsp90 mRNA contains significant structure, consistent with the extent observed in a typical non-heat shock mRNA (for illustration of these features, see Fig. 9A). Second, Hsp90 mRNA translation appears to be inhibited by eIF4F inhibition, whereas translation of the other Hsp mRNAs is not affected to any significant extent (17,18). The presence of secondary structure and eIF4F dependence are likely related. It has been hypothesized that preferential translation, embracing both the inhibition of non-heat shock mRNAs and the high activity of Hsp mRNAs, lies in an ability of Hsp mRNAs to bypass an eIF4F activity lesion. However, this proposal presents a dilemma with respect to Hsp90 mRNA, because this mRNA possesses significant secondary structure, is translationally inhibited by eIF4F inhibition, yet is efficiently translated during heat shock. In this study we have investigated the basis for the preferential translation of Hsp90 mRNA, and uncovered several unexpected features that suggest that multiple mechanisms exist that lead to preferential Hsp mRNA translation in a single organism.

Chemicals
Chemicals were purchased from Sigma unless otherwise indicated. All restriction enzymes were purchased from New England Biolabs. The Topo 2.1 vector used for ligation of PCR products was purchased from Invitrogen.

Transfection of Drosophila S2 Tissue Culture Cells
Schneider S2 cells were cultured at 22-23°C in Schneider's Drosophila Medium (Invitrogen) containing 10% fetal calf serum, 20 mM L-glutamine, 100 units/ml penicillin, 0.1 mg/ml streptomycin, 0.25 mg/ml amphotericin B (Invitrogen). 24 h prior to transfection the cells were seeded at a density of 1-1.5 ϫ 10 6 cells/ml in a T25 flask (Corning). Transfections were carried out as described (13). For the eIF4E-BP transfection, a total of 25 g of plasmid was transfected in each case, made up of equal microgram amounts of eIF4E-BP and Gal4 plasmids (19), and 5 g each of Hsp70 ⌬Cd and Hsp90 ⌬Cd ; to retain the total amount of plasmids equally transfected in all cases, a copia (promoter)/ ␤-galactosidase plasmid was added to some transfections.

Heat Shock, [ 35 S]Methionine Labeling, and Protein Extraction
Drosophila S2 cells were scraped from T25 flasks, pelleted by brief centrifugation (3 min, 3000 rpm) in a clinical centrifuge (IEC), and resuspended in Grace's media lacking methionine (Invitrogen). The cells were then transferred to 20-ml glass scintillation vials, with a stir flea, and allowed to recover Ն15 min with stirring prior to analyses (at heat shock or normal growth temperature (22-24°C)). For heat shock, a portion of the cells was incubated in a 36°C water bath with stirring for 15 min, at which time ϳ5 ϫ 10 6 cells (1 ml of suspension) were labeled with 15-20 Ci of [ 35 S]methionine/cysteine (ICN Biochemicals) for 15 min (i.e. total 30 min heat shock). For non-heat shock analysis, another equal portion of the same cell sample, maintained at normal growth temperature, was labeled concurrently as described above (for heat shock). At the end of labeling, cells were rapidly pelleted by centrifugation, resuspended in and washed twice at 4°C (50 mM KCl, 15 mM MgSO 4 , 4 mM CaCl 2 , 3 mM KH 2 PO 4 , 10 mM dextrose, 8.4 mM HEPES, pH 7.2, and 20 g/ml cycloheximide) by centrifugation. The cell pellet was lysed with (for two-dimensional gel analyses) 100 -150 l of Ampholyse buffer (ϳ9.8 M urea, 5% 3.5-10 Biolytes (Bio-Rad), 2% Nonidet P-40, 1% ␤-mercaptoethanol), microcentrifuged for 3 min at top speed (Eppendorf), and the supernatant was recovered to give a final protein concentration of ϳ2 mg/ml. Protein concentration was measured by the Bradford assay (Bio-Rad), and radioactivity/g of protein was measured by trichloroacetic acid precipitation. For samples to be analyzed by one-dimensional gel electrophoresis, washed protein pellets were lysed in hSDS (0.3% SDS, 50 mM Tris, pH 8.0, 1% ␤-mercaptoethanol, at ϳ98°C). The pellet was disrupted by pipetting. 1/20 volume of RNase/DNase solution (5 mg/ml DNase, 2.5 mg/ml RNase A, 500 mM Tris, pH 7.0, 50 mM MgCl 2 ) was added for ϳ1 min, viscosity was reduced by pipetting, then 1 ⁄4 volume of 4ϫ Laemmli formula SDS-PAGE buffer was added, and the samples were analyzed on one-dimensional slab gels as described below for the second dimension of the two-dimensional procedure.

Analysis of Proteins by Two-dimensional Isoelectric
Focusing/SDS-PAGE Two-dimensional IEF/SDS-PAGE was performed basically as described by O'Farrell (20), with modifications as described by Duncan and Hershey (21) to promote spot focusing. Gels were fixed, dried, and exposed to Kodak X-Omat film for 4 -15 days. Protein bands/spots were quantitated by densitometry (Bio-Rad VersaDoc 1000 imaging system/ Quantity 1 or PDQuest (Bio-Rad) software, or by LabWorks (UVP)). Heat shock translation was calculated as the spot IOD at heat shock (numerator) divided by the spot IOD prior to heat shock (denominator). Transgene mRNA translational efficiency was also calculated as the protein synthesis rate (spot IOD) per unit transgene mRNA (determined by Northern blot analysis, autoradiography, and densitometry). Equal RNA loading was verified by methylene blue staining or endogenous Hsp70 hybridization, as described below. RNA analyses verified that the mRNA levels of the CuSO 4 -induced transgenes neither increase nor decrease during the 30-min heat shock interval. Similarly, protein synthesis-based analyses of the transgenes are unaffected by the inclusion of actinomycin D during the heat shock interval. Thus, comparative analysis of protein synthesis rates (spot darkness) before and after heat shock accurately quantifies heat shock preferential translation.

RNA Isolation, Analysis, and Quantitation
RNA was extracted from ϳ5 ϫ 10 6 cells using TRIzol reagent (Invitrogen) as recommended by the manufacturer. The ethanol-precipitated RNA pellet was resuspended in a final volume of 25 l of diethyl pyrocarbonate-treated water at ϳ2-3 g/l, as determined in a Beckman UV spectrophotometer. Samples were analyzed by Northern blotting as described (13), and bands on the film were quantified by densitometry using the Bio-Rad VersaDoc 1000 Imaging System/Quantity 1 software.

Construction of Plasmid Expression Vectors
General PCR Procedures-1 ng of template DNA was amplified for 22-25 cycles, gel purified using the GeneClean II Kit (Bio 101 Inc. (per the manufacturer's instructions)), and verified by sequencing (University of Southern California Norris Cancer Center Microchemical Facility). The primers (Operon Technologies) used in the constructions are listed in Table I.
MT90-FL-The Drosophila Hsp90 gene encoded by plasmid pDm83 (gift of Dr. H. Lipschitz) was used as the PCR amplification target. The upstream 5Ј primer (DU71) contains Hsp90 5Ј-UTR nucleotides ϩ3 to ϩ23 preceded by nucleotides that introduce an EcoRI restriction site at the start of transcription to facilitate subsequent plasmid construction. The 5Ј end sequence of the transgene-expressed mRNA is GAAUUCU-UGA . . . , whereas the 5Ј end sequence of authentic Hsp90 is AGUCU-UGA . . . ; the artificial 5Ј-UTR contains two extra 5Ј-terminal nucleotides, and the 4th nucleotide in the artificial 5Ј-UTR is U, whereas the corresponding nucleotide (position ϩ2) in the authentic 5Ј-UTR is G. All other nucleotides are identical. The expressed mRNA containing the introduced EcoRI site is efficiently translated during heat shock (see Figs. 7 and 8), hence the EcoRI site does not impair heat shock translation. Similar observations were made for Hsp70 mRNA (13). The downstream 3Ј primer (DU72) hybridized to 5Ј-UTR nucleotides ϩ129 to ϩ149 (where ϩ150 is the "A" in the AUG codon). The primer altered nucleotides to create an NcoI site at the initiator AUG (new sequence is CCAUGG, which maintains the initiator AUG in an efficiently recognized context). The amplified sequence was ligated into the Topo 2.1 vector, blue colonies were identified, and accurate integrants were verified by sequencing (all other plasmids were prepared in like fashion). This plasmid was digested using EcoRI/NcoI, and inserted into an EcoRI/NcoI-digested pmthsp44-NcoI vector (13) that has been modified with the introduction of an NcoI site at the start of translation. The resulting MT90-FL vector contains the Drosophila metallothionein promoter precisely fused to the Drosophila full-length Hsp90 5Ј-UTR linked to an internally deleted Hsp70 coding region and Hsp70 3Ј-UTR. The expression construct leads to the synthesis of a unique ϳ44-kDa protein (12,13). The sequences of all plasmids used in this study were verified by DNA sequencing.

Cap-proximal Deletions of MT-FL90
MT90-⌬40 CAP -Using MT90-FL plasmid as the amplification target, a 5Ј upstream primer (DU86) was designed that hybridized to nucleotides ϩ43 to ϩ62 of the Hsp90 5Ј-UTR (numbering for this and subsequent constructions is based on the authentic Hsp90 5Ј-UTR; ϩ1 represents the first transcribed nucleotide, an A); preceding (5Ј) the hybridizing nucleotides it contained in an EcoRI site. The downstream primer (DU69) hybridized at the unique SnaBI site in the coding region FIG. 1. Translational efficiency of Hsp90 mRNA is increased proportional to temperature. Drosophila S2 cells were placed in water baths equilibrated to 29 -37°C for 15 min, then pulse-labeled with [ 35 S]methionine for 15 min. Cells were pretreated with 1 g/ml actinomycin D for 10 min (B) or not treated (A) prior to immersion in the water bath. Protein samples were prepared as described (see "Materials and Methods"). Equal amounts of protein (equal cell numbers) were loaded into each lane of the gel (based on Bradford assays, and confirmed by Coomassie Brilliant Blue staining of the gel after electrophoresis). Dried gels were exposed to film and labeled proteins were detected by autoradiography. For quantitation, films were scanned with a densitometer, and the intensity of bands determined using Labworks software (UV Products). This analysis has been repeated in part or completely Ͼ10 times, with similar results. Migration locations of the prominent heat shock proteins are labeled to the right.
FIG. 2. Translational efficiency of Hsp90 mRNA is increased proportional to temperature. Drosophila S2 cells were placed in water baths equilibrated to 29 -37°C for 15 min, then pulse-labeled with [ 35 S]methionine for 15 min. Cells were pretreated with actinomycin D for 10 min prior to immersion in the water bath (bottom rows, panel A), or not treated (top rows, panel A). Protein samples were prepared as described of the target. The PCR amplification product deletes the first 42 nucleotides of the authentic Hsp90 5Ј-UTR, but does reintroduce the EcoRI site nucleotides as ϩ1 to ϩ6 of the expressed 5Ј-UTR mRNA. The PCR product (ϳ1160 nucleotides) was digested with EcoRI/SnaBI, and inserted into EcoRI/SnaBI-digested MT90-FL to yield MT90-⌬40 CAP .
MT90-⌬75 CAP and MT90-⌬110 CAP -The procedure was identical to that described above, except the upstream primers (DU87 and DU88, respectively) were designed to hybridize to nucleotides ϩ77 to ϩ96 and nucleotides ϩ111 to ϩ130, respectively, in the Hsp90 5Ј-UTR. The resultant expressed mRNAs have nucleotides ϩ1 to ϩ76 and ϩ1 to ϩ110, respectively, deleted.
MT90-⌬75 AUG and MT90-⌬110 AUG -The procedure was identical to that described above, except the downstream primers (DU84 and DU83, respectively) were designed to hybridize to nucleotides ϩ52 to ϩ71 and ϩ17 to ϩ36, respectively, in the Hsp90 5Ј-UTR. The resultant expressed mRNAs retained Hsp90 5Ј-UTR nucleotides ϩ1 to ϩ71 and ϩ1 to ϩ36, respectively.

Internal Deletion Mutant of MT90-FL
MT90-⌬40 -110 I -The procedure was identical to that described above for MT90-⌬40 CAP , except the upstream primer (DU89) was designed to hybridize to nucleotides ϩ112 to ϩ128 in the Hsp90 5Ј-UTR. Preceding this hybridization segment the primer contained 5Ј-UTR nucleotides ϩ1 to ϩ37 (including the EcoRI site at the start of transcription). The resultant expressed mRNA has nucleotides ϩ38 to ϩ111 deleted.

Coding Sequence Deletion Mutant of Hsp90
MT90 ⌬Cd -The upstream primer (DUC3) was designed to hybridize to 5Ј-UTR nucleotides ϩ138 to ϩ149, and extend 18 nucleotides into the coding sequence. The downstream primer (DUC4) was designed to hybridize to nucleotides 1160 to 1185 in the Hsp90 coding sequence (based on the numbering in NM 079175 (Entrez nucleotide)). The primer appends an AflII site preceding the coding nucleotides (or, in the orientation of the mRNA, downstream of Hsp90 coding nucleotides) for plasmid construction purposes. The primers were used to amplify the 5Ј-UTR and coding sequence nucleotides using pDm83 as the amplification target. The PCR amplified DNA was digested with NcoI and AflII, and inserted into MT90-FL digested with the same enzyme pair. The resultant plasmid replaces the Hsp70 ⌬Cd coding sequence with Hsp90 ⌬Cd . In MT90 ⌬Cd the "UAA" sequence within the CTTAAG AflII restriction site is in-frame to constitute the stop codon. The resulting plasmid expresses ϳ840 nucleotides of the Hsp90 coding sequence, resulting in a ϳ30-kDa protein product (see Fig. 5).

Heat Stress Increases the Translation of Hsp90 mRNA-
Hsp90 mRNA is unique among the Drosophila Hsp mRNAs in being present in non-heat stressed cells in significant amounts. To investigate the translational characteristics of Hsp90 mRNA (throughout this report), S2 cells were pulse-labeled with [ 35 S]methionine for 10 -15 min, and the production of newly synthesized Hsp90 protein was assessed by gel electrophoresis, autoradiography, and densitometry.
The rate of synthesis of Hsp90 is rapidly increased by heat shock (Fig. 1A). There is a detectable increase when temperature is raised to 29°C, a progressively larger induction as temperature is raised from 29 to 35°C, and then its protein synthesis begins to decrease as temperature is further increased (to 37°C in this experiment).
To assess whether this increase in protein synthesis represented increased translational efficiency (i.e. protein synthesis per mRNA), or simply occurred because there were more Hsp90 mRNAs, transcription was blocked by treatment with actinomycin D. In this case, there was still a significant increase in the synthesis of Hsp90 (Fig. 1B), indicating that the mRNA is relatively inefficiently translated under normal circumstances, and that translation is specifically increased by heating. To more precisely and accurately quantify the amount of Hsp90 synthesized during heat shock, without and with actinomycin D, labeled proteins were analyzed by two-dimensional IEF/ SDS-PAGE ( Fig. 2A) and the Hsp90 mRNA translational efficiency was quantified (Fig. 2B). Hsp90 mRNA increases its translational efficiency proportional to temperature, reaching a FIG. 3. Hsp90 and Hsp70 mRNA expression as temperature is increased. Drosophila S2 cells were placed in water baths equilibrated to 29 -37°C for 30 min. Cells were pretreated with 1 g/ml actinomycin D for 10 min prior to immersion in the water bath (ϩAct D), or not treated (No ActD). RNA samples were prepared as described (see "Materials and Methods"). Equal amounts of RNA (equal cell numbers) were loaded into each lane of the gel (based on A 254 , and confirmed by methylene blue staining of the nylon membrane after electrophoresis and transfer). Nylon membranes were probed using a 32 P-labeled plasmid fragment for Hsp90 (A) or Hsp70 (B). Temperatures of heat shock are shown above the lanes. Dried membranes were exposed to film and labeled bands were detected by autoradiography. Panels shown for Hsp70, ϩ actinomycin D, were prepared from equal amounts of RNA (verified by methylene blue staining) and exposed for the same interval. This analysis has been repeated in part or completely Ͼ5 times, with similar results. The apparent lower expression of Hsp90 mRNA seen at 34°C in the top portion of panel A was not detected in other analyses.
(see "Materials and Methods"). Equal amounts of protein (equal cell numbers) were loaded into each first dimension gel (based on Bradford assays, and confirmed by Coomassie Brilliant Blue staining of the gels after electrophoresis). Dried gels were exposed to film and labeled proteins were detected by autoradiography. For quantitation, films were scanned with a densitometer, and the intensity of spots determined using Labworks software (UV Products). The pH gradient runs from more acidic to the left to more basic to the right. The coordinates of Hsp70 and Hsp90 are shown in the 37°C, No Act D panel. The translational efficiency of Hsp90 and -70 is shown in panel B, based on the translation in actinomycin D-treated cells depicted in panel A for Hsp90, and non-treated cells for Hsp70. The translational efficiency was calculated by quantifying the spot darkness shown in panel A, divided by the relative mRNA expression, as measured by Northern analyses (Fig. 3). This analysis has been repeated in part or completely Ͼ5 times, with similar results. The significantly enhanced translation was seen for Hsp70 at 29°C, and at 30°C to a lesser extent, is a consequence of the very low levels of mRNA detected at these temperatures (see Fig. 3). The exposures shown are darker than the ones used to quantify expression to more fully reveal the spot patterns of proteins with a lower rate of synthesis. The darker exposures underestimate the expression differences because some of the spots shown, including the Hsps, are saturated in some panels. maximum activity at 35°C that is ϳ3-4 times that observed at normal growth temperature (22-24°C).
The efficacy of actinomycin D treatment can be observed in the inhibition of Hsp70 synthesis. Hsp70 mRNA is virtually absent in non-heat shocked cells. Hence, all its synthesis requires new, heat-induced transcription. The efficacy of treatment was also directly assessed by analysis of Hsp90 and Hsp70 mRNAs by Northern blotting. The heat shock-induced increase in Hsp90 mRNA was largely suppressed by actinomycin D treatment, and induction of Hsp70 mRNA was reduced by Ͼ95% (Fig. 3). This is similar to the extent of Hsp70 protein synthesis inhibition seen in Fig. 2.
The increase in Hsp90 synthesis as temperature is raised could theoretically be because of temperature generally activating the translational machinery (a "Q 10 -like" effect), or could reflect a shared characteristic of all the Hsp mRNAs. However, it is neither because of a general nor class-specific activation; first, there is no significant increase in the translation rate of numerous non-heat shock mRNAs at very mild heat shock temperatures (i.e. 30 -32°C) that do increase the synthesis of Hsp90 (see, for example, bands/spots in Figs. 1 or 2 representing synthesis of non-heat shock proteins (e.g. actin)). Second, the temperature-dependent translational activation is specific to Hsp90 mRNA because when Hsp70 mRNA was expressed at normal temperature (see Fig. 4, legend, for details), its translation did not increase with temperature (Fig. 4). Thus, there is no general increase in translation of Hsp mRNAs as the temperature is increased; Hsp90 mRNA possesses unusual characteristics that may extend to a unique pathway to preferential translation, as detailed below.
Hsp90 mRNA Translation Is Cap-dependent, as Is Hsp70 mRNA Translation-A potential unique pathway for Hsp90 mRNA translation would be IRES-mediated cap-(and eIF4F-) independent translation. Considering the sensitivity of Hsp90 mRNA translation in vitro to antibody-mediated eIF4F inhibition (17,18), we wished to verify that the same sensitivity to eIF4F inhibition applied to in vivo translation, to more rigorously address the possibility that Hsp90 mRNA is translated via an IRES-mediated pathway under natural circumstances. To specifically inhibit cap-dependent translation in intact cells, Drosophila eIF4E-BP was overexpressed (Fig. 5A) by transfection. Reporter genes for Hsp90 and Hsp70 mRNA translation were cotransfected. These mRNAs contain their respective Hsp fulllength 5Ј-UTRs followed by their respective coding sequence, each containing an internal deletion to allow unique identification of the protein expression product (see McGarry and Lindquist (12) for Hsp70 mRNA, and see "Materials and Methods" for Hsp90 mRNA). Translation of the Hsp reporter mRNAs was measured by pulse labeling with [ 35 S]methionine and twodimensional gel electrophoresis, autoradiography, and densitometry (Fig. 5, B-E). Initially, experiments were carried out at normal growth temperature to investigate the basic mRNA properties. Overexpression of eIF4E-BP caused significant inhibition of both Hsp90 and Hsp70 mRNA translation. At high-level overexpression of eIF4E-BP (corresponding to Fig. 5A, lane 4) Hsp90 mRNA translation was reduced by Ͼ95%, and Hsp70 mRNA translation by ϳ65% (densitometric quantitation of panels B and C). At lower levels of eIF4E-BP overexpression (corresponding to Fig. 5A, lanes 2 and 3) there was undetectable to minor (Ͻ50%) inhibition of Hsp mRNA translation for both 70 and 90 (data not shown). Hsp90 mRNA translation appears to be slightly more sensitive to eIF4E-BP overexpression at normal growth temperature. Most significantly, neither Hsp90 nor Hsp70 mRNA translation is unaffected by (or increased by) eIF4E-BP overexpression, as would be expected to occur if these mRNAs are translated via a cap-independent pathway. At higher levels of eIF4E-BP expression, translation of both transgenes was undetectable (data not shown).
To investigate whether an independence from eIF4E-BPmediated inhibition was induced by heat shock, an aliquot from each transfected cell culture was heat shocked and labeled as above to determine protein expression of the Hsp90 and Hsp70 Protein samples were prepared as described (see "Materials and Methods"). Equal amounts of protein (equal cell numbers) were loaded into each first dimension gel (based on Bradford assays, and confirmed by Coomassie Brilliant Blue staining of the gels after electrophoresis). Dried gels were exposed to film and labeled proteins were detected by autoradiography. This analysis has been repeated Ͼ5 times, with similar results. Locations of Hsps and Grp78 (dmHsc72) are indicated with arrows. The reporter transgene, like Hsp70 itself, splits into two isoforms upon twodimensional IEF/SDS-PAGE. For quantitation, the spot densities of both forms were summed.
transgenes. The influence of overexpressed eIF4E-BP on Hsp70 mRNA translation at heat shock was very similar to that observed at normal growth temperature (Fig. 5, D and E), whereas Hsp90 was significantly less affected; the extents of inhibition for Hsp70 and Hsp90 mRNA were ϳ65 and ϳ60%, respectively. The reduced sensitivity of Hsp90 mRNA to cap-dependent translation inhibition under heat shock conditions parallels results obtained using rapamycin in heat-shocked cells. 2 Two distinct conclusions may be drawn from this analysis. First, the translation characteristics of Hsp90 mRNA are altered by heat shock to reduce its dependence on eIF4F. Second, and equally important, Hsp90 mRNA translation is cap-dependent, because its elevated resistance to moderate eIF4E-BP overexpression only results in partial translation, and its translation is completely abrogated by high level eIF4E-BP overexpression. These results showing Hsp mRNA preferential translation is cap-dependent corroborate previous investigations by ourselves and others using different approaches (12,13). The observation that translation of Hsp90 mRNA is cap-dependent at both normal and heat shock temperatures influences the model we propose for its translation (see "Discussion").
In addition to the mass effects of eIF4E-BP overexpression on its association with eIF4E and consequent protein synthesis inhibition, eIF4E-BP dephosphorylation can further increase its inhibitory effect by stabilizing its interaction with eIF4E. Our previous results had shown that mammalian eIF4E-BP transfected into Drosophila cells was dephosphorylated by heat shock (37°C) (24), as well as showing that eIF4E-BP was dephosphorylated in mammalian cells by heat shock at temperatures Ն43°C (24). To determine whether the effects of DmeIF4E-BP overexpression in heat shocked Drosophila cells included enhanced repression because of heat-induced dephosphorylation, the lower molecular weight region of the two-dimensional gels was examined. At 36°C heat shock causes dephosphorylation of Drosophila eIF4E-BP, as seen by the reduction in the higher M r , more acidic, phosphorylated variants (Fig. 6, arrows). The extent of phosphorylation at normal temperature is less than typically observed in mammalian cells (corroborated in numerous experi-2 R. Duncan, unpublished results.  (19). The amount of eIF4E-BP plasmid transfected was varied to yield different extents of expression (panel A, immunoblot analysis). Transfected cells (72 h post-transfection) were incubated with 500 M CuSO 4 for 3 h at normal growth temperature (22-24°C) to induce mRNA expression. The culture was split, and aliquots of cells were incubated in a water bath equilibrated to 36°C for 15 min (D and E), or left at 22-24°C (B and C), then pulse-labeled with [ 35 S]methionine for 15 min. Protein samples were prepared as described (see "Materials and Methods"). Equal amounts of protein (equal cell numbers) were loaded into each first dimension gel (based on Bradford assays, and confirmed by Coomassie Brilliant Blue staining of the gels after electrophoresis). Dried gels were exposed to film and labeled proteins were detected by autoradiography. The locations of the reporter-expressed Hsp70 and Hsp90 proteins are indicated with arrows. This analysis has been repeated 3 times. ments), 2 but the more highly phosphorylated forms virtually disappear, and one lowest molecular weight variant (Fig. 6B, bolder arrow) significantly increases. This two-dimensional analysis of Dm eIF4E-BP resembles one recently described by Miron et al. (25), with the significant difference that the highly abundant, most basic variant (see legend for details) does not appear to be detected in their analysis; this variant may correspond to wholly dephosphorylated eIF4E-BP. These results suggest that heat-induced eIF4E-BP dephosphorylation contributes to its inhibitory activity during heat shock, yet Hsp90 mRNA translation is significantly less inhibited under heat shock conditions compared with normal growth temperature where eIF4E-BP phosphorylation is significantly greater. These results also confirm that eIF4E-BP dephosphorylation is a common response to heat shock, although contrary results have been reported (26).
AUG-proximal Nucleotides Are Critical for Preferential Translation of Hsp90 mRNA-Two Drosophila heat shock mRNAs, encoding Hsp70 and Hsp22, have been dissected to identify where sequence elements critical to heat shock translation are located. In both instances, the first ϳ60 nucleotides of the transcript (the cap-proximal region of the 5Ј-UTR) were shown to have the greatest effect on preferential translation during heat shock. For 5Ј-UTR Hsp70, the terminal ϳ180 nucleotides could be replaced with little diminution of preferential translation (10). For 5Ј-UTR Hsp22, the first ϳ25 nucleotides have been suggested to be sufficient (27).
We have carried out a similar analysis to determine the location within the Hsp90 mRNA sequence of signals required for its preferential translation. First, the entire 5Ј-UTR was appended to a reporter coding sequence/3Ј-UTR to create expression plasmid MT90-FL. This coding body/3Ј-UTR cannot be translated during heat shock unless it has a preferential translation-promoting 5Ј-UTR (12,13). mRNAs were expressed under the control of a metallothionein promoter. Expression was induced at normal growth temperature for 3 h using 500 M CuSO 4 . Translation was assessed as above, using pulse labeling with [ 35 S]methionine and quantification of reporter protein synthesis by two-dimensional IEF/SDS PAGE, autoradiography, and densitometry, at normal growth temperature and under heat shock conditions.
The 5Ј-UTR of Hsp90 mRNA is sufficient to confer translation during heat shock. The translation of MT90-FL mRNA remains high during heat shock, as evidenced by the robust production of transgene protein (Fig. 7, indicated with arrows). There was little to no decrease in translation rate relative to normal growth temperature (Figs. 7 and 8), mirroring results obtained when the Hsp70 5Ј-UTR is appended to this transgene (Refs. 13; Fig. 4). Thus, the full-length Hsp90 5Ј-UTR contains all the sequence information required for preferential translation. This observation parallels results regarding Hsp70 and Hsp22.
To identify which regions of the Hsp90 5Ј-UTR were necessary for preferential translation, two series of truncation mutants were constructed. In the first series 3 progressively longer blocks of nucleotides were removed from the cap end of the 5Ј-UTR, to create plasmids MT90-⌬40cap, MT90-⌬75cap, and MT90-⌬110cap. In the second series 3 progressively longer blocks of nucleotides were removed from the AUG-proximal end of the 5Ј-UTR, to create plasmids MT90-⌬40AUG, MT90-⌬75AUG, and MT90-⌬110AUG. In addition, a 5Ј-UTR comprised of the first 35 nucleotides linked to the last 35 nucleotides was created, MT90-⌬40 -110I. All of these 5Ј-UTRs are diagrammed in Fig. 8. Translation ability during heat shock was determined as described above for MT90-FL. mRNAs in which either 40 or 75 nucleotides have been deleted from the cap proximal region are translated relatively well during heat shock (Fig. 8). Translation rate decreases about 50%, which is similar to the decrement seen when similar lengths are truncated from the cap-proximal region of Hsp70 mRNA (14). The retained translation potency remains ϳ5-10-fold greater than the typical non-heat shock mRNA, which are inhibited by Ͼ90% on average (many examples can be seen in Fig. 7, comparing the spot intensities of non-shock protein synthesis at the two temperatures; spot quantitation of Ͼ10 randomly selected spots showed an average reduction in translation rate to 10% that observed in non-heat shocked cells, with greater than half (8/13) reduced to Ͻ5% the non-heat shock rate). Deletion of 110 nucleotides from the cap results in severely compromised translation, typical of a non-heat shock mRNA. In summary, cap proximal nucleotides in Hsp90 mRNA influence preferential translation, but they can be deleted and significant preferential translation during heat shock is retained as long as a minimum amount of Hsp90 5Ј-UTR is present. Additionally, there are no required elements in internal nucleotides 38 -110, because the internal deletion is translated relatively well during heat shock.
Deletions from the AUG-proximal region of Hsp90 5Ј-UTR suggest these nucleotides are required for significant heat shock translation. Deletion of 40 nucleotides reduced reporter gene translation to the minimal level characteristic of a non-heat shock mRNA (Fig. 8). Deletions of larger amounts from the AUGproximal regions were consistent with this result, also showing very low synthesis of reporter protein during heat shock (Fig. 8). In all the AUG-proximal truncations, the nucleotides preceding and following the AUG are part of the NcoI site, which retains an adequate context for efficient translation (e.g. MT90-FL). These  Fig. 5, panel A). The coordinates of three non-eIF4E-BP protein spots are indicated by asterisks in both panels, for orientation purposes. The positions of Hsp22 and Hsp23, which migrate at similar M r and pI to certain eIF4E-BP variants, are shown in panel B (labeled H22 and H23). The isoelectric point of Hsp22 is virtually identical to actin, and the isoelectric point of the most basic eIF4E-BP variant is more basic than all Hsp70 variants, and is detected on the right (basic) edge of the sector shown in panels in Fig.  5. The pH gradient runs from more acidic to the left to more basic to the right. The most acidic eIF4E-BP variants migrate to the left border of the gel sectors shown in Fig. 5, which corresponds to the acidic terminus of the isoelectric focusing gel. results suggest that unique requirements and considerations apply to the mechanism of Hsp90 mRNA translation, because the strong dependence on AUG-proximal nucleotides has been unambiguously refuted for Hsp70 and Hsp22 mRNA translation during heat shock (10,27).

DISCUSSION
Hsp90 mRNA and protein are abundant in Drosophila cells at normal growth temperature. Whereas virtually all mRNAs expressed under non-heat shock conditions are translationally repressed, synthesis of Hsp90 remains high, and the translational efficiency of Hsp90 mRNA even increases. Hsp90 mRNA translation is relatively inefficient at normal growth temperature, and this inefficiency is relieved by heat shock. This heatdependent activation of translation distinguishes Hsp90 from other Hsp mRNAs, such as Hsp70 mRNA, whose translation is very efficient at normal growth temperature and achieves preferential translation during heat shock by evading the global Protein samples were prepared as described (see "Materials and Methods"). Equal amounts of protein (equal cell numbers) were loaded into each first dimension gel (based on Bradford assays, and confirmed by Coomassie Brilliant Blue staining of the gels after electrophoresis). Dried gels were exposed to film and labeled proteins were detected by autoradiography. Eight plasmid-expressed mRNAs are depicted to the left. The 5Ј-UTR nucleotides in these eight were: 1-149 (full-length; see "Materials and Methods" for precise description of the 5Ј-UTR structures for each construct); the cap-proximal deletions that retained 43-149, 77-149, 111-149; the AUG-proximal deletions that retained 1-105, 1-71, 1-36; and a 5Ј-UTR containing 1-37/112-149 (internal deletion). Segments deleted are indicated by thin dashed line. Heat shock mRNA translation was measured as: cpm spot heat shock/cpm spot normal temperature. Northern analysis indicates that the mRNA content does not change over the ϳ30-min analysis interval, so translation rate equals translational efficiency. This analysis has been repeated Ͼ5 times for each plasmid-expressed mRNA.
inhibitory mechanism(s) induced by heat shock. This provides novel evidence that there exist two fundamentally different patterns, and likely pathways, for achieving preferential heat shock translation.
Several lines of evidence have conclusively excluded an IRES-mediated pathway for preferential translation of Hsp mRNA, including abrogation of its translation by appending nucleotides to its 5Ј terminus (12) or by the introduction of a stem-forming region proximal to the cap site (13). IRES-mediated translation would represent an obvious distinct pathway for Hsp90 mRNA heat shock translation, and no previous experiments have addressed this possibility. To investigate this possibility, the sensitivity of Hsp90 mRNA translation to eIF4E-BP overexpression was determined, because it has been consistently documented that IRES element-mediated translation is resistant to eIF4E-BP inhibition. The results clearly show that Hsp90 mRNA translation is suppressed by high-level overexpression of eIF4E-BP, indicating that this mechanism does not account for Hsp90 mRNA preferential translation. Other pathways must be entertained, and other molecular interactions determined.
Studies to identify the nucleotides that allow continued translation of Hsp90 mRNA during heat shock identified the 5Ј-UTR as sufficient to promote preferential translation, paralleling studies by others and ourselves investigating which portions of Hsp70 and Hsp22 mRNA confer preferential translation (10,12,13,27). However, in distinction to those mRNAs, we find that the AUG-proximal nucleotides of Hsp90 mRNA are of critical importance, because their removal reduces reporter mRNA translation during heat shock to levels characteristic of a non-heat shock mRNA. These characteristics are featured in a model described in the following paragraphs.
Studies by Sierra and colleagues (17,18) determined that Hsp mRNA translation in general is significantly resistant to FIG. 9. Structural features of the Hsp90 5-UTR. A, the extent of 5Ј-UTR secondary structure for a panel of Drosophila mRNAs is depicted. Overall stability was determined by folding the entire 5Ј-UTR secondary structure of each mRNA using the Mfold algorithm (33), then dividing the total energy (Ϫ⌬G) by the 5Ј-UTR length (because Ϫ⌬G increases with length, at a rate proportional to the extent of secondary structure along the length) to achieve the Ϫ⌬G/nucleotide scale on the y axis. B, the predicted secondary structure of Drosophila Hsp90 5Ј-UTR based on the Mfold algorithm (33). The location of the initiator AUG codon at ϩ150 is shown. the inhibition of eIF4F activity. Along with other results directly measuring eIF4F activity and showing that it is reduced by heat shock (29,30), this has led to the proposal that preferential translation of Hsp mRNA occurs via their ability to evade a heat-induced lesion in eIF4F activity. Our earlier results demonstrated that an Hsp70 mRNA variant in which the 5Ј-UTR is modified to contain a modest extent of secondary structure is efficiently translated at normal growth temperature, but completely loses its capacity to be translated during heat shock (13). This is wholly consistent with a model in which reduced eIF4F activity leads to inadequate unwinding activity for efficient translation of the stem-containing variant, whose secondary structure is similar to the extent found in non-heat shock mRNAs.
However, Sierra and colleagues (17,18) also report the perplexing finding that Hsp90 mRNA translation is strongly inhibited by eIF4F inhibition. This is consistent with our analysis of the extent of secondary structure in Hsp90 mRNA relative to the other Hsp mRNAs, and relative to non-heat shock mRNAs, which show that the Hsp90 5Ј-UTR is characteristic of an eIF4F-dependent non-heat shock mRNA, and quite distinct from the other Hsp mRNAs (Fig. 9A). Yet, Hsp90 mRNA translation is demonstrably efficient during heat shock. Two alternative scenarios can account for this discrepancy. First, it is possible that eIF4F activity is not significantly compromised by heat shock. We believe this to be unlikely, insofar as the direct activity measurements and the stem-containing Hsp70 mRNA translation analysis, which shows that a modest stem structure in Hsp70 mRNA inhibits translation at normal temperature but not at heat shock (13), strongly suggests that a significant impairment of eIF4F activity occurs. Furthermore, it is unlikely that this scheme can be quantitatively modified to postulate that eIF4F inhibition occurs but its extent is insufficient to affect Hsp90 mRNA, because (i) the extent of secondary structure in Hsp90 mRNA is significantly greater than that in the inhibited Hsp70 mRNA stem-containing variant (pSL17.11) (13), and (ii) Hsp90 mRNA translation is not only resistant to heat shock inhibition but in fact enhanced by heat shock.
The alternative hypothesis is that the mechanism of Hsp90 mRNA translation is altered by heat shock such that it becomes less eIF4F-dependent. This heat-dependent transition leading to decreased eIF4F dependence would represent a novel pathway to achieve preferential heat shock translation, and would account for the inability to observe eIF4F independence in the in vitro analyses cited above (17,18). A mechanism can be proposed that is wholly supported by our 5Ј-UTR deletion analyses, presents a strong parallel to mechanisms of Hsp mRNA preferential translation in prokaryotes (e.g. Ref. 31), and has parallels in the translation of mammalian Hsp70 mRNA (32). Folding analysis of the complete Hsp90 5Ј-UTR using Mfold (33) presents the theoretical structure shown in Fig. 9B. Notable features are the extensive regions of secondary structure, and specifically a long stem in the AUG proximal half of the UTR, as well as a short stem including the AUG initiation codon. We suggest that one or both of these regions of secondary structure comprises a heat-sensitive inhibitory element that impedes access to the initiation codon at normal growth temperature. Furthermore, the ability of ribosomal subunits to recognize this region could be heat enhanced, presumably through thermal destabilization of the stem. This model draws by analogy on studies that have elucidated a mechanism of bacterial heat shock preferential translation (28,31). In that instance, a series of studies have shown that thermal melting of a stem-containing region including the Shine-Dalgarno region, and perhaps also a downstream box segment, allows rRNA base-pairing and ribosome recruitment only at elevated (heat shock) temperatures (28,31). Whereas we do not yet have any direct evidence that a similar mechanism applies to Hsp90 mRNA translation in Drosophila, the concept that a prokaryotic mechanism of preferential translation might be retained as the foundation for a lower eukaryote is intriguing. Supporting evidence comes from studies of Hsp70 mRNA heat shock translation in human cells, where it has been shown that AUGproximal sequences may recruit ribosomal subunits for shunting-mediated translation based on mRNA-rRNA base-pairing (32), analogous to the bacterial situation.
An aspect of this model must be its relative independence from eIF4F. We suggest that, in parallel to documented mechanisms in prokaryotes and human cells, the segment of nucleotides preceding the Hsp90 mRNA of AUG can recruit ribosomes during heat shock providing reduced eIF4F dependence. This may occur through a base-pairing mechanism, perhaps including a direct transfer of ribosomal subunits to the AUG from the cap-proximal region where they initially associate (i.e. shunting, as described for human Hsp70 mRNA). Assessing potential base pairing regions between the AUG-proximal Hsp90 5Ј-UTR and 18 S rRNA is equivocal, insofar as potential regions can be identified (e.g. segments with 6 out 7 nucleotides paired), but in no case do these examples involve the terminal nucleotides of 18 S rRNA with an mRNA segment close to the AUG, as occurs in the Shine-Dalgarno interaction. Our hypothesis predicts that there are two modes of Hsp90 mRNA translation. Under normal temperature conditions, translation occurs by a typical scanning mechanism, is eIF4F-dependent, and relatively inefficient because of the secondary structure elements. During heat shock, translation shifts to a mode in which ribosomal subunits are more directly recruited to the AUG, promoted by direct mRNA-rRNA base pairing. This model predicts that deletion of the AUG proximal nucleotides would severely compromise heat-dependent translation, but have little effect on non-heat shock Hsp90 mRNA translation (which uses active eIF4F to unwind the AUG-proximal region, albeit inefficiently). This is exactly what we observe based on the experiments in this study. The AUG-proximal nucleotides could be a heat-activated IRES, promoting cap-independent translation, but we do not believe this is likely as discussed above. Thus, a model positing eIF4F-mediated cap-dependent ribosome subunit recruitment seems most consistent with our data; the initial eIF4F-mediated binding step may be tolerant of reduced eIF4F activity (allowing cap engagement during heat shock), whereas the subsequent eIF4F-dependent unwinding steps are bypassed. Further experiments are in progress to provide direct evidence for this hypothetical model of Hsp90 mRNA translation in Drosophila. In conclusion, we suggest that certain types of preferential heat shock translation may reflect the adaptation of prokaryotic mechanisms to eukaryotic cells.