Domain Structures and Immunogenic Regions of the 90-kDa Heat-shock Protein (HSP90)

Domain structures of the 90-kDa heat-shock protein (HSP90) have been investigated with a library of anti-HSP90 monoclonal antibodies (mAbs) and by limited proteolysis with trypsin and chymotrypsin. Thirty-three mAbs were obtained by immunization with bacterially expressed human HSP90α and HSP90β isoforms. Among them, ten and three mAbs reacted specifically with HSP90α and HSP90β, respectively. Immunoblotting and enzyme-linked immunosorbent analyses revealed that major immunogenic domains were located at two restricted regions of HSP90α, i.e. amino acids 227–310 (designated Region I) and 702–716 (Region II), corresponding to a highly charged region and a region near the C terminus, respectively. Taken together with the characteristics of the amino acid sequences, these two immunogenic regions appeared to be exposed at the outer surface of HSP90. We further investigated the domain structures of HSP90 by limited proteolysis in combination with N-terminal sequencing and immunoblotting analyses. Tryptic cleavages of HSP90α at low concentrations revealed the existence of major susceptible sites at Arg400-Glu401, Lys615-Ala616, and Arg620-Asp621. Proteolysis at higher trypsin concentrations caused successive cleavages only toward the N-terminal direction from these sites, and Region I was included in the region selectively deleted during this process, thereby further suggesting its surface location. From these results, we propose three domain structures of HSP90 consisting of amino acids 1–400, 401–615, and 621–732. Differences in the protease sensitivity and immunogenicity further suggest that every domain is composed of two subdomains. This is the first study describing the domain structures and the immunogenic regions of HSP90.

The 90-kDa heat-shock protein (HSP90) 1 is one of the major stress proteins in eukaryotic cells. There are at least two HSP90 genes, and two HSP90 isoform proteins, ␣ and ␤, are expressed in the cytosolic compartment (1). The amino acid sequence of HSP90␣ is 85 (human; see Refs. 1 and 2) to 90% (yeast; see Ref. 3) homologous to HSP90␤. Either one of the isoforms is indispensable for the growth of yeast cells at higher temperatures (3). Biochemical characterization of purified HSP90 indicates that HSP90␣ predominantly exists as a homodimer and HSP90␤ exists mainly as a monomer (4). Dimer formation is mediated by the interaction at the C-terminal 191 amino acids in which the C-terminal region (Met 628 -Asp 732 ) of one subunit associates with the adjacent region (Val 542 -Tyr 627 ) of the other subunit (5). The amino acid substitutions at 561-685 between ␣ and ␤ isoforms are responsible for the impeded dimerization of HSP90␤ (5).
HSP90 is believed to have a chaperone-like activity for particular molecules that are involved in signal transduction, such as steroid receptors (6), casein kinase II (7), pp60 v-src (8), elF2␣ kinase (9), and aryl hydrocarbon receptor (dioxin receptor) (10). HSP90 specifically binds to these proteins; and, in most cases, this interaction is essential for the function of the proteins (9,11,12). However, a variety of evidence, i.e. the abundance of HSP90 in cells even under nonstressed conditions, the conserved amino acid sequences from prokaryotic to eukaryotic cells (13), and the indispensability in yeast (3), strongly suggests that HSP90 is involved in more fundamental functions of cells. In fact, several studies have recently shown that HSP90 functions as a general chaperone. That is, it interacts with various proteins less specifically and modulates their conformation. For instance, the refolding of citrate synthase is significantly enhanced by the co-presence of HSP90 (14). The spontaneous refolding of denatured dihydrofolate reductase and irreversible denaturation of firefly luciferase are prevented by association with HSP90 (15). The latter study further demonstrated that the chaperone-like activity of HSP90 is closely related to the oligomerization of HSP90 at temperatures higher than 46°C. Although HSP90 possesses an ATPase activity (16), this activity does not appear to be needed for the oligomerization of HSP90.
Epitope mapping is a useful approach for investigating the structures of proteins of interest. Several anti-HSP90 monoclonal antibodies (mAbs) have been produced to date (6,10,17). Although the epitopes of several anti-HSP90 mAbs have been described (18), systematic assignment of the immunogenicity of HSP90 has not been conducted. In this study, we developed a library of the mAbs that specifically recognize human HSP90 and with them demonstrated that the major immunogenic domains are located in two restricted regions. The domain structures of HSP90 were further analyzed by limited proteolysis. We finally proposed three domain structures of HSP90. In addition, this is the first report of developing isoform-specific mAbs against HSP90.
Plasmid Construction and Purification of Recombinant Proteins-Amino acid numbers are presented as those of human HSP90␣ (732 amino acids) throughout this study, in which the amino acid sequence of HSP90␤ is aligned with that of HSP90␣ in consideration of the eight amino acid deletions corresponding to amino acids 1-5 and 238 -240 of HSP90␣. Full-length forms of human HSP90␣ and HSP90␤ were expressed in Escherichia coli Y1090 as glutathione S-transferase (GST)fusion proteins, and the recombinant proteins were purified as described (5). GST-HSP90␣ or ␤ encoding amino acids 535-732 and their chimeric proteins, GST-HSP90␣458 -732 and GST-HSP90␣1-43/604 -732, were expressed as described previously (5). Other expression vectors with deleted forms of HSP90 isoforms were constructed by the cleavage of pGST-HSP90␣ or ␤ with appropriate restriction enzymes. H 6 HSP90␣, a recombinant HSP90␣ tagged with a dodecapeptide (MRGSH 6 GS), was expressed as described (19). pQE-9 series of expres-sion plasmids, i.e. pH 6 HSP90␣542-732, ␣542-728, ␣542-720, and ␣542-697, were prepared by the polymerase chain reaction technique with appropriate primers as described previously (5). The details on the constructions are available on request. H 6 HSP90␣s were purified with Talon TM metal affinity resin according to the manufacturer protocol.
Preparation of Anti-HSP90 mAbs-Following thrombin cleavage of full-length forms of GST-HSP90␣ and ␤, the HSP90 moieties were purified by DEAE ion-exchange HPLC as described (20). The hybridoma cells producing anti-HSP90 mAbs were prepared according to Oli and Herzenberg (21). The fusions were performed separately with the mice injected with HSP90␣ and HSP90␤. Anti-HSP90 mAb-producing hybridomas were selected by use of an enzyme-linked immunosorbent assay (ELISA) with purified HSP90␣ or ␤ (0.5 g) coated on each well of a 96-well titer plate (Falcon). Peroxidase-conjugated anti-mouse (IgG ϩ IgM) antibodies were used as second antibodies. Absorbance at 492 nm was measured following incubation with O-phenylenediamine for 1 h at 30°C. Positive hybridomas were cloned by limiting dilution. Large scale production of mAbs was carried out by growing the hybridoma cells in ascites of mice. The immunoglobulin fraction was precipitated with 50% (w/v) ammonium sulfate, and the precipitate was dialyzed against saline containing 0.1% sodium azide.
Nomenclature of the mAbs-K3700 and K41000 number series indicate the mAbs produced from the hybridoma cells of mice immunized with HSP90␤ and HSP90␣, respectively. When more than one hybridoma colony grew up in a culture well, they were separated at the following screening step. Those mAbs were distinguished by addition of alphabets A-D. Their independence was confirmed by characterization of the biochemical properties.
Antibody Class-Class and subclass of the heavy chain of the mAbs were determined by the micro-Ouchterlony method.
Polyacrylamide Gel Electrophoresis and Immunoblotting-Proteins were subjected to SDS-PAGE at a polyacrylamide concentration of 12.5 or 15%. Separated proteins were stained with Coomassie Brilliant Blue or transferred to a polyvinylidene difluoride membrane (Millipore Corp.). The membrane was incubated with anti-HSP90 mAbs (0.1-0.5 g/ml) at 25°C for 2 h. Alkaline phosphatase-conjugated goat antimouse IgG or IgM was used as the second antibody at a 1:5000 dilution in 50 mM Tris-HCl (pH 7.6), 0.15 M NaCl (TBS) containing 0.05% Tween 20 and 0.25% bovine serum albumin. After having been washed with TBS containing 0.1% Tween 20, blots were visualized by incubation with BCIP and NBT. Low molecular weight markers and rainbow markers were used as standards for Coomassie staining and immunoblotting, respectively. Epitope Mapping by ELISA-The regions recognized by the mAbs were investigated by ELISA and immunoblotting with various forms of HSP90s fused to GST. The purified protein (0.5 g) coated on each well of a 96-well titer plate was incubated with 0.3 g of purified mAbs in phosphate-buffered saline (pH 7.3). Alkaline phosphatase-conjugated goat anti-mouse IgG or anti-mouse IgM was used as the second antibody at a 1:5000 dilution. BCIP and NBT in a soluble buffer system (Kirkegaard & Perry Lab.) were used as the substrates, and absorbance at 600 nm was measured following a 1-h incubation at 25°C.
ELISA with an Octapeptide Library-A noncleavable octapeptide library covering amino acids 212-312 of HSP90␣ was prepared according to the manufacturer method (Chiron Mimotopes Pty. Ltd. Clayton, Victoria, Australia). Octapeptides with their N terminus shifted sequentially to each amino acid residue from 212 to 305. The reactivity of the mAbs to the peptides was determined by ELISA as described above.
Limited Proteolysis and N-terminal Amino Acid Sequencing-Proteins in 50 mM Tris-HCl (pH 8.0) containing 100 mM NaCl and 10% glycerol were incubated at various concentrations of trypsin (N-tosyl-Lphenylalanine chloromethyl ketone treated) or chymotrypsin (N-␣-tosyl-L-lysine chloromethyl ketone treated) at 30°C for 6 h. Proteolytic peptides (13 g/lane) were denatured, separated by SDS-PAGE, and transferred to a polyvinylidene difluoride membrane (Bio-Rad). The protein bands visualized with Coomassie Brilliant Blue were subjected to N-terminal sequencing with a model 477A protein sequencer (Applied Biosystems) equipped with an on-line model 120A analyzer.
Phage Display System-Amino acid sequence with an affinity to an anti-HSP90 mAb (K41007) was determined by use of the phage display system with a heptapeptide library according to the manufacturer protocol. At each step of panning, 2 ϫ 10 11 phages were incubated with 15 g of the mAb precoated on a well of a 96-well Maxisorp plate (Nunc). Following four pannings, the phages were amplified and purified, and then the amino acid sequence of the heptapeptide was deduced from the nucleotide sequence.

RESULTS
Isoform Specificity and Immunoglobulin Class of the mAbs-We obtained 33 independent mAbs against human HSP90 isoforms. For determination of the isoform specificity of the mAbs, ELISA was performed with various amounts of HSP90␣ and HSP90␤ (0.07-150 ng). According to the results, we tentatively classified the mAbs into the following three classes: mAbs with cross-reactivity to the other isoform of less than 1% were isoform-specific; mAbs with the cross-reactivity to the other isoform from 1 to 66.7% were isoform-preferential; and mAbs reacting with the other isoform with more than 66.7% efficiency were equivalently recognizing ones (Table I). Unexpectedly, K3725A, obtained from a mouse immunized with HSP90␤, was specific for HSP90␣ on the basis of ELISA. The isoform specificity was further investigated by immunoblotting with bacterially expressed HSP90␣ and HSP90␤. The immunoblotting data were in accord with the isoform specificities of the mAbs estimated by ELISA with the exception of K3725A (Table I). K3725A bound to both HSP90␣ and HSP90␤ on immunoblotting. As a result, isoform specificities of the mAbs were assigned as shown in Table I. We obtained ten (numbers 1-10) and three (numbers 31-33) mAbs that specifically interacted with HSP90␣ and ␤, respectively; five mAbs (numbers [11][12][13][14][15]) that preferentially interacted with HSP90␣; five mAbs (numbers 23 and 27-30) that preferentially interacted with HSP90␤; and nine mAbs (numbers 16 -22, 25, and 26) that were equivalently reactive to both isoforms. The isoform specificity of K3725A (No. 24) remained to be established. Immunoglobulin class and IgG subclass of the mAbs were also determined ( Table I).
Epitope Mapping of HSP90 -Next we determined the regions of HSP90 recognized by the mAbs by using various deletion mutants of HSP90␣ (Fig. 1a) and HSP90␤. Most forms were purified to homogeneity by affinity chromatography on SDS-PAGE (data not shown). However, several forms revealed additional bands on SDS-PAGE. Even in such cases, the positions of the expressed proteins were defined by comparison with calculated molecular masses. Bands with molecular masses lower than expected seemed to be proteolytic products of the recombinant proteins. The purified samples were used for ELISA; whereas, the immunoblotting analysis was done with the bacterial lysates containing the expressed proteins to avoid proteolysis during the purification step.
We performed the epitope mappings with the rest of the mAbs by the same procedures. The full-length form and four deleted forms of HSP90␤ were used for the investigation of the mAbs specifically or preferentially bound to HSP90␤. In consequence, their recognition sites were defined as shown in Table II. These observations demonstrated that the epitopes defined by most mAbs were located in particular regions: the epitopes recognized by 25 out of the 33 mAbs (76%) were found within amino acids 185-335, most probably in 216 -312. We tentatively designated this region as Region I. The second one, designated Region II, seemed to be found at the C-terminal region at 533-732. As expected, the isoform specificity of mAbs seemed to be dependent on the degree of amino acid substitutions in the recognized region: mAbs recognizing conserved sites, such as amino acids 290 -312, had no or little isoform preference; and amino acids 216 -285 containing the least conserved region (amino acids 238 -273, 47% homology between HSP90␣ and HSP90␤) were recognized by isoform-specific or highly isoform-preferential mAbs (Table II).
Epitope Mapping at Region I-Immunoblotting analysis and ELISA indicated specific localization of the immunogenic region (Region I) most probably within amino acids 216 -312. Because the region consisted of ϳ100 amino acid residues, we prepared 94 octapeptides of which N termini were shifted in every amino acid from 212 to 305 of HSP90␣. ELISA with the octapeptide library was performed with all mAbs recognizing Region I except K3701 and K3716, which did not react with the ␣ isoform (Table I). As shown in Fig. 2, two-thirds of the mAbs (15/23) revealed single binding peaks at their respective positions; and accordingly, their epitopes were clearly defined. On the other hand, two binding peaks were observed with K3725D, K3738, and K41331; three peaks with K41110 and K41116C; and four major peaks with K3729 (Fig. 2, c and d). No bindings were observed with K41002 and K41028 (Fig. 2d). With four mAbs (K3725D, K41110, K41116C, and K41331) among those that revealed multiple peaks, the peaks were close together (Fig. 2, c and d). This suggested that more than eight amino acids were involved in recognition of the mAbs and that several combinations of amino acids less than eight residues were sufficient for the bindings.
K3738, an HSP90␤-preferential mAb, revealed two peaks relatively distant from each other (Fig. 2d, open squares). However, the two binding regions (DDEAEEKEDKEEE, 232-244 and EEEKEKEEKESEDK, 242-255) could not be separated when all peptides interacting with the mAb were aligned. That is, the last peptide (amino acids 237-244) forming the first peak overlapped the first peptide (242-249) forming the second peak. In addition, it should be remembered that the hybridoma producing K3738 was derived from a mouse immunized with the ␤ isoform. HSP90␤ has a three-amino acid deletion corresponding to 238 -240 of HSP90␣. Moreover, the two binding regions are considerably similar to each other; and thus, we postulate that either the deletion, the similarity between the two regions, or both explain the two binding peaks of K3738. Accordingly, we prudently propose that the epitope of K3738 is localized within amino acids 232-255. Unexpectedly, another HSP90␤-preferential mAb, K3729, bound to various peptides at completely distinct positions, i.e. DKEVSDDEAEEK (227-238), IEDVGSDEEEEKK (258 -270), EKYIDQEELN (282-291), and PDDITNEEYG (301-310) (Fig. 2d). Again, there are sequence similarities between amino acids 227-238 and 258 -270 and between 282-291 and 301-310. If not all, this may explain the multiple binding peaks of K3729.
The results of ELISA with the octapeptide library clearly showed that the boundaries of Region I were amino acids 227  a Isoform specificities of mAbs are presented as described in Table I , and the fourth one with Site Ic occurred. These results indicate that the mAbs recognizing Region I bound to one of Sites Ia-Ic or to the regions that overlapped with them (Fig. 3).

Analysis of the Recognition Sites at the C-terminal Region (Region II)-
We also characterized the recognition sites at the C-terminal region by immunoblotting analyses of both chimeric proteins and deletion mutants. First, because the mAbs recognizing this region were specific or highly preferential to either one HSP90 isoform or the other (Table II), we investigated the recognition sites with chimeric proteins of the two isoforms (Fig. 4a). Although all chimeric proteins migrated at the same position on SDS-PAGE, they were clearly distinguished by the immunoblotting. K41009, an HSP90␣-specific mAb, bound to HSP90␤535-700/␣701-732 but did not do so to the counter chimera (Fig. 4b, lanes 3 and 4). On the other hand, K3705, an HSP90␤-specific mAb, bound to HSP90␣535-700/␤701-732 (Fig. 4c, lane 3). Thus, K3705 and K41009 recognized amino acids 701-732 of their respective isoforms. K3714, K3725B, and K41007 also recognized the same region of their respective isoforms (data not shown). Furthermore, when the bindings of K3705, K3714, and K3725B to amino acids 535-719 of HSP90␤ are taken into account (Table II), it seems reasonable to further narrow the recognition site of the HSP90␤-specific mAbs to amino acids 701-719.
The C-terminal recognition sites of the HSP90␣-specific mAbs were further investigated with the deletion mutants (Fig. 5a). Immunoblotting analysis with K41009 demonstrated its binding to HSP90␣542-720, but not to HSP90␣542-697 (Fig. 5b, lanes 4 and 5). An identical result was obtained with K41007 (data not shown). Taken together with the results of immunoblotting of the chimeric proteins (Fig. 4b), we concluded that K41009 and K41007 recognized amino acids 701-720. Therefore, two HSP90␣-specific mAbs (K41007 and K41009), two HSP90␤-specific mAbs (K3705 and K3725B), and a mAb highly preferential to HSP90␤ (K3714) should recognize identical or very close sites within amino acids 701-720. Isoform-specific bindings of the mAbs that bound to Region II further suggest that the highly substituted amino acids (702-716) correspond to their epitope (Fig. 4a). We tested this possibility by use of the phage display system. We obtained a phage that specifically accumulated after four pannings with an HSP90␣-specific mAb, K41007. The heptapeptide sequence expressed on the phage surface was WVADTSY, of which ADTS is similar to the ADDTS (705-709) of HSP90␣. Therefore, it is most likely that the highly substituted site at amino acids 702-716 contains the epitope recognized by K41007.
According to the epitope mapping described above, it was obvious that the recognition site of K3725A (533-603) was distinct from Region II. Hence, we obtained three mAbs (K3725A, K41016, and K41218) that recognized respective regions distinct from Regions I and II. We also determined the epitope for AC88, a mAb bound to HSP90 of various species (17), because its epitope is not precisely defined, but is reported to reside in the C-terminal region (18). We found that the epitope for AC88 was located within 604 -697, where no mAbs developed herein bound (data not shown).
Limited  Table I. FIG. 3. Summary of the epitopes determined by ELISA with the peptide library. The epitopes of the mAbs determined by ELISA with the peptide library are summarized. The region of the peptide library used in this study is represented at the top. A hatched area represents the KED triplet lacking in HSP90␤. Asterisks indicate amino acid replacements between the isoforms. The boundary of Region I and those of the mAbs are indicated by an arrow and bars, respectively. Isoform specificities the mAbs are indicated by characteristic letters as described in Table I. dodecapeptide, was used in this study because the peptide was small enough not to affect the electrophoretic mobility of the protein, and the recombinant protein exists as a dimer as does the native protein (19). H 6 HSP90␣ was purified to near homogeneity by affinity chromatography (Fig. 6, lane 1). The N-terminal sequencing showed that the major contaminant with a 20-kDa molecular mass was not an endogenous bacterial protein, but a C-terminal fragment of HSP90␣ produced by the cleavage at Lys 573 -Ile 574 (Table III). Since the C-terminal region of 191 amino acids forms a dimer (5), we supposed that the 20-kDa fragment was associated and copurified with intact H 6 HSP90␣. Cleavage at the lowest trypsin concentration generated 15-, 50-, and 80-kDa fragments (lane 2). Subsequently, 27-and 40-kDa fragments appeared at intermediate enzyme concentrations (lanes  3 and 4). At higher trypsin concentrations, 22-, 32-, and 33-kDa peptides appeared (lanes 5-10). We found that the degradation profile of chymotrypsin treatment was similar to the tryptic cleavage profile (data not shown).
To address the origins of the proteolytic fragments, we determined their N-terminal amino acid sequences. Despite the appearance of many proteolytic peptides, only four N termini were detected: Glu 401 , Ala 616 , Asp 621 , and the N terminus of the recombinant protein (Table III). The 15-kDa species was a mixture of two peptides starting at Ala 616 and Asp 621 , the same as that found in our previous study of thrombin cleavage (5). We also found that liberation of the N-terminal dipeptide (MR, Ϫ12-11) caused alteration of the electrophoretic mobility from 32 to 33 kDa (Table III).
Taken together, we concluded that there were two major sites susceptible to limited proteolysis: one at Arg 400 -Glu 401 and the other at Lys 615 -Ala 616 /Arg 620 -Asp 621 . Furthermore, it should be noted that the 32/33-and 40-kDa fragments were derived from the N-terminal 50-kDa fragment by successive cleavages at its C-terminal part. These observations suggest that the N-terminal 32/33-kDa part is the most protease-resistant region of HSP90␣. In contrast, the adjacent C-terminal region encompassing the region up to Arg 400 was susceptible to proteolytic cleavage. The 22-kDa fragment starting at Glu 401 seemed to be generated from the larger 27-kDa one through a similar mechanism.
Immunoblotting Analysis of Tryptic Peptides-The apparent molecular mass of the region susceptible to the proteolysis was estimated to be 18 kDa from the difference between the Nterminal 32/33-and 50-kDa fragments. Since Region I (amino acids 227-310) should be localized around this region, we tried to determine its location among the tryptic peptides by immunoblotting with two mAbs (K41102 and K41122B) that recognized Sites Ia and Ic, respectively. K41102 as well as K41122B reacted with the 40-, 50-, 80-, and 90-kDa fragments, but did not bind the 32/33-kDa fragments (Fig. 7), thereby indicating that the N-terminal 40-kDa fragment still retained Region I and that the 32/33-kDa fragments had lost it. Moreover, because of the sensitivity of the immunoblotting analysis, several additional bands, e.g. 42-and 47-kDa ones (Fig. 7), were apparently detected, while they were faint by Coomassie staining (Fig. 6). These results indicate that Region I is truely localized in the C-terminal, 18-kDa portion of the N-terminal 50-kDa fragment. Taken together with these findings, the origins of tryptic peptides and the 20-kDa one, and their relationship to Region I, were deduced as shown in Fig. 8.

DISCUSSION
In this study, we developed 33 anti-HSP90 mAbs, including ␣ and ␤ isoform-specific ones (Table I), and determined the epitopes recognized by them (Table II and Figs. 3 and 9). We confirmed that all mAbs specifically recognized HSP90 endogenously expressed in human salivary gland adenocarcinoma (22) and T47D cells, and we found that most of the mAbs cross-reacted with rat, mouse, and rabbit HSP90s. 2 This is not surprising in view of the evolutional conservation of HSP90. To our knowledge, this is the first report of the development of isoform-specific mAbs against HSP90. Since the epitopes for the mAbs have been characterized, these antibodies should be valuable to investigate the structure and function of HSP90 isoforms.
We found that two regions (Regions I and II) confer most immunogenicity of HSP90. Only three mAbs (K3725A, K41016, and K41218) developed in this study and AC88 recognized respective regions different from Regions I and II (Table II). Region I was defined to amino acids 227-310 by ELISA with the octapeptide library. Notably, there is a charged region (223-289) containing six amino acid substitutions of opposite net charges at 241, 242, 247, 257, 271, and 273 (Fig. 9). These characteristics strongly suggest that amino acid substitutions are permissive only when charged properties are maintained. Among anti-HSP90 mAbs reported previously, BF4 (6) also recognizes the charged region (18). This charged region is proposed to form an ␣-helical structure that the polyglutamic acids of which are stabilized by ionic interactions with lysines (23). In HSP90, there is another charged region at amino acids 534 -585 (23), suggesting its exposure to the outer surface. However, only K3725A recognized this region among all mAbs developed (Table II)   acid sequences of the former two are highly charged, and substituted between HSP90␣ and HSP90␤; and hence, the mAbs recognizing these two sites are specific or highly preferential to HSP90␣. We could not define the epitopes of K3701 and K3716, mAbs recognizing Region I and specific or highly preferential to HSP90␤. However, it seems reasonable to postulate that their recognition sites are also located at Site Ia or Ib. In contrast, the amino acid sequence of Site Ic is less charged and is completely conserved between the isoforms. Hence, the mAbs bound to Site Ic equivalently recognize the two isoforms (Fig.  3). Prediction of the secondary structures indicates that the charged region (223-289) carrying Sites Ia and Ib, and the region of amino acids 321-345 form ␣-helical structures and that their intermediate region (290 -320) carrying Site Ic (291-304) may form a ␤-turn or coiled structure (23). The strong immunogenicity of the ␣-helical structure composed of the charged amino acids and the adjacent ␤-turn or coiled structure strongly suggests the protruding of these conformations into the aqueous environment.
We found a second immunogenic site (Region II) at the Cterminal region. The C-terminal region (amino acids 699 -732) has hydrophilic characteristics although the degree is less than that of Region I (Fig. 9). This difference is mainly due to the absence of basic amino acids except Arg 727 and the presence of several hydrophobic amino acids. Nevertheless, it is important to note that acidity of the five amino acid residues at 702, 706, 707, 714, and 723 is conserved in the replacements between the isoforms. Considering the isoform-specific properties of the mAbs recognizing Region II, the highly substituted amino acids at 702-716 are the most probable recognition site of the mAbs. In fact, we selected a heptapeptide with an affinity for an HSP90␣-specific mAb (K41007). The sequence (WVADTSY) was similar to amino acids 705-709 of HSP90␣ (Fig. 9). Thus, Region II is ascribed to amino acids 702-716 (Fig. 9). HtpG, an E. coli homolog of mammalian HSP90, has 37% amino acid homology with human HSP90␣ (13). Interestingly, Region I, except Site Ic, and Region II are highly substituted between HSP90 isoforms and are even deleted in HtpG. These facts strongly suggest that these regions are dispensable for the function of HSP90.
We previously reported that bacterially expressed HSP90␣1-47/290 -732, i.e. HSP90␣ with Sites Ia and Ib deleted tends to form a self-oligomer and is able to form a heteromeric complex with the estrogen receptor (24). In contrast, HSP90␣ possessing Sites Ia and Ib (HSP90␣1-312) neither forms an oligomer nor interacts with the receptor (24). These results indicate that HSP90␣, with the major part of Region I deleted, acquired enhanced activity for associating with the estrogen receptor as well as its self-oligomerizing activity. A recent study indicates that the oligomerization of HSP90 is closely related to its chaperone activity (15). Thus, we suspect that one possible role of Region I is to suppress the oligomerization activity of HSP90 under nonstressed conditions.
We further analyzed the domain structures of HSP90 by means of limited proteolysis. The N-terminal sequencing analysis demonstrated that cleavages occurred only at a few susceptible sites at Arg 400 -Glu 401 and Lys 615 -Ala 616 /Arg 620 -Asp 621 at lower protease concentrations. The former site had not been recognized before because our previous study focused on the C-terminal dimer-forming region (5). We found that, even at higher enzyme concentrations, trypsin attacted several particular sites. Immunoblotting analysis confirmed that Region I was localized in the protease-sensitive region.
Based on the present results, we finally propose a model for the domain structures of HSP90 (Fig. 8). HSP90 is composed of three domains, Domains A-C, corresponding to amino acids 1-400, 401-615, and 621-732, respectively. Further, the division of each domain into two subdomains is suspected from the differences in proteolytic susceptibility and immunogenicity. The subdomain structures of Domains B and C are also suggested from our previous study, demonstrating that chymotrypsin cleaves at Leu 541 -Val 542 , Tyr 627 -Met 628 , and Met 628 -Ala 629 bonds (5). The deletion in HtpG corresponding to amino acids 696 -732 further suggests the structural and functional differences between amino acids 604 -700 and 701-732. The characteristics of each region are as follows: subdomain A-I is protease-resistant; subdomain A-II is protease-sensitive, is abundant in charged amino acids, and carries the most immunogenic region of HSP90 (Region I), which may be exposed to the outer surface of the molecule; subdomain B-I is resistant to proteolysis; subdomain B-II interacts with subdomain C-I of another HSP90 subunit, and this interaction mediates the dimerization of HSP90; and subdomain C-II carries the second immunogenic region (Region II) and is not involved in dimer formation. Although the validity of this model, especially as to the subdomain structures, should be tested by functional and chemical studies, we believe that this provides an appropriate working model for investigating the structure and function of HSP90.