Identification of the peptide-binding site in the heat shock chaperone/tumor rejection antigen gp96 (Grp94).

Heat shock protein (HSP)-peptide complexes from tumor cells elicit specific protective immunity when injected into inbred mice bearing the same specific type of tumor. The HSP-mediated specific immunogenicity also occurs with virus-infected cells. The immune response is solely due to endogenous peptides noncovalently bound to HSP. A vesicular stomatitis virus capsid-derived peptide ligand bearing a photoreactive azido group was specifically bound by and cross-linked to murine HSP glycoprotein (gp) 96. The peptide-binding site was mapped by specific proteolysis of the cross-links followed by analysis of the cross-linked peptides using a judicious combination of SDS-gel electrophoresis, mass spectrometry, and amino acid sequencing. The minimal peptide-binding site was mapped to amino acid residues 624-630 in a highly conserved region of gp96. A model of the peptide binding pocket of gp96 was constructed based on the known crystallographic structure of major histocompatibility complex class I molecule bound to a similar peptide. The gp96-peptide model predicts that the peptide ligand is held in a groove formed by alpha-helices and lies on a surface consisting of antiparallel beta-sheets. Interestingly, in this model, the peptide binding pocket abuts the dimerization domain of gp96, which may have implications for the extraordinary stability of peptide-gp96 complexes, and for the faithful relay of peptides to major histocompatibility complex class I molecule for antigen presentation.

Heat shock protein (HSP)-peptide complexes from tumor cells elicit specific protective immunity when injected into inbred mice bearing the same specific type of tumor. The HSP-mediated specific immunogenicity also occurs with virus-infected cells. The immune response is solely due to endogenous peptides noncovalently bound to HSP. A vesicular stomatitis virus capsid-derived peptide ligand bearing a photoreactive azido group was specifically bound by and cross-linked to murine HSP glycoprotein (gp) 96. The peptide-binding site was mapped by specific proteolysis of the cross-links followed by analysis of the cross-linked peptides using a judicious combination of SDS-gel electrophoresis, mass spectrometry, and amino acid sequencing. The minimal peptide-binding site was mapped to amino acid residues 624 -630 in a highly conserved region of gp96. A model of the peptide binding pocket of gp96 was constructed based on the known crystallographic structure of major histocompatibility complex class I molecule bound to a similar peptide. The gp96-peptide model predicts that the peptide ligand is held in a groove formed by ␣-helices and lies on a surface consisting of antiparallel ␤-sheets. Interestingly, in this model, the peptide binding pocket abuts the dimerization domain of gp96, which may have implications for the extraordinary stability of peptide-gp96 complexes, and for the faithful relay of peptides to major histocompatibility complex class I molecule for antigen presentation.
Specific protective immunity results when heat shock protein (HSP) 1 -peptide complexes purified from tumor cells are injected into inbred mice bearing the same specific type of tumor (1)(2)(3)(4)(5)(6). The HSP-mediated specific immunogenicity is also seen with virus-infected cells (7)(8)(9). This paradigm is the basis for a new therapeutic strategy against human cancers (9,10). The immune response is directed against peptides noncovalently bound to the HSP and not to the HSP (for reviews, see Refs. 11 and 12)). HSPs that form the immunogenic peptide complexes include gp96 (GRP94) (13) and calreticulin (14), which reside in the endoplasmic reticulum (ER). Cytosolic HSP70- (15,16) and HSP90- (17) peptide complexes are also immunogenic. The ER-resident chaperone gp96 (GRP94) has been the most extensively studied from an immunological standpoint (13, 18 -21). It is an abundant stress protein that displays dual functionality: it directs peptides into the immune response pathway and it assists in protein folding. The role of gp96 in the immune response is not well understood at the molecular level. gp96 binds a variety of peptides in vitro and in vivo with little or no apparent amino acid sequence specificity (13,(22)(23)(24). gp96 also binds ATP but the role of nucleotide in peptide loading/unloading is unclear (22,25). Highly purified HSP90, the cytosolic paralog of gp96, probably binds and uses ATP (26). This is in contrast to chaperones HSP70 and BiP (ER paralog of HSP70), where it is clear that ATP binding is important for the release of peptide substrates (reviewed in Ref. 27). In vivo, gp96 binds peptides in a transporter associated with antigen processing-dependent (23,28) or transporter-associated with antigen processing-independent manner (29). gp96-chaperoned peptides are represented in the context of the major histocompatibility complex class I (MHC-I) pathway in professional antigen presenting cells (30), thereby inducing a peptide-specific CD8 ϩ response (13, 19 -21, 31, 32). gp96 forms dimers and higher-order aggregates (33)(34)(35). Peptides may be bound in a gp96 dimer complex (33). To develop gp96 as a vaccine for cancer immunotherapy, an understanding of the mechanism of peptide binding and a clear definition of the peptide-binding site are essential. Toward this goal, we have, for the first time, identified the peptide-binding site in gp96.

EXPERIMENTAL PROCEDURES
gp96 Purification and Peptide Modification-His-tagged murine gp96 protein was purified as described before (33). PepK was synthesized by and purchased from Alpha Diagnostic (San Antonio, TX) and derivatized with N-hydroxysuccinimidyl-4-azidosalicylic acid (NHS-ASA, Pierce, Rockford, IL) in the dark at room temperature for 1 h. To 30 nmol of PepK was added a 10-fold excess of NHS-ASA in buffer (150 l: 20 mM sodium phosphate, 10% dimethyl sulfoxide, pH 7.6). The reaction was quenched with Tris-HCl (0.1 M, pH 6.8) for 15 min. The derivatized PepK was purified on a 3-ml Sephadex G-10 column and developed in 10 mM sodium phosphate, 15 mM NaCl, pH 7. The mass change anticipated by modification of peptide (ϩ135 Da) was verified by mass spectrometry. The derivatized peptide (1.4 nmol) was kinased with protein kinase A catalytic subunit (Roche Molecular Biochemicals) in 100 l (12.5 mM HEPES/Na ϩ , 5 mM MgCl 2 , 2 mM dithiothreitol, (pH 7) plus 100 pmol of [␥-32 P]ATP (300 Ci, NEN Life Science Products)) for 30 min at 30°C. The kinase was heat-inactivated at 60°C for 10 min and the peptide was rechromatographed on Sephadex. The specific radioactivity of the 32 P-PepK was quantified by spotting on phosphocellulose paper discs (P81, Whatman, Clifton, NJ), washing five times in 1 M glacial acetic acid for 5 min each and twice with EtOH for 5 min, followed by drying and scintillation spectrometry. Typically, ASA-modified PepK was labeled to a specific activity of 20 -40 Ci/g.
Cross-linking-His-gp96 (240 g/ml) was incubated with 32 P-labeled azidosalicylamide (ASA)-modified PepK at a 4:1 molar ratio in 300 l of * This work was supported by a research grant from Antigenics L.L.C. New York. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
¶ binding buffer (20 mM HEPES/Na ϩ , pH 7.9, 20 mM NaCl, 2 mM MgCl 2 , 100 mM KCl, 4 mM Zwittergent 3-12 (Roche Molecular Biochemicals) at 55-60°C for 15 min and cooled for 15 min to room temperature. Samples (25 l) in microtiter wells were irradiated with 320 -360 nm UV for 10 min at room temperature in a Rayonet photochemical reactor (Southern New England Ultraviolet, Branford, CT). Preparative amounts of cross-links (ϳ2 mg) were run on SDS-polyacrylamide gels and the bands representing the cross-links were identified by autoradiography and excised as gel pieces, dried in vacuo, and rehydrated in the appropriate protease buffer containing a cleaving agent. Proteolysis was done in situ at 33-37°C for 48 h with one addition of fresh cleaving agent. The ratio of cleaving agent to target was 5:1 for CNBr/mild, Ͼ100:1 for CNBr/harsh, 1:20 for enzymes except trypsin, and 3:1 for trypsin. In most cases following the digestion, samples were pooled, lyophilized, and separated by SDS-PAGE. The gel system was either standard acrylamide-SDS or Tricine-acrylamide-SDS. Radiolabeled proteolytic fragments were sized by comparison to prestained markers (Bio-Rad or Amersham Pharmacia Biotech). The apparent mass of PepK (2.6 -3.6 kDa) was subtracted from the observed mass of the cross-links. The gp96 fragment mass (cross-link mass minus PepK mass) was matched with the predicted masses (Ϯ0.4 kDa) in a gp96 fragmentation mass bank generated by computer.
Irradiated cross-linked complexes that had been digested with Endo Glu-C or CNBr/harsh for the purposes of sequence determination were dialyzed extensively against H 2 O using a 1000-Da cutoff dialysis membrane, lyophilized, and run on 15% polyacrylamide analytical gels. Proteins were transferred onto a polyvinylidene difluoride membrane (Millipore, Bedford, MA) for 2 h at 250 mA at 4°C using chilled transfer buffer (10 mM CAPS, 10% methanol, pH 11). The polyvinylidene difluoride membrane was washed several times with water, dried, and exposed to x-ray film to identify the positions of the radiolabeled bands. Specific bands containing cross-links were excised with a razor blade. The peptides in the radiolabeled band were sequenced using automated Edman degradation method at the Rockefeller University Protein/DNA Technology Center.
Matrix-assisted Laser Desorption Time of Flight Mass Spectrometry-Mass spectra were acquired in a Voyager TM workstation (PerSeptive Biosystems, Framingham, MA). The spectra were calibrated using egg white lysozyme as an internal standard, and known peptide/protein molecules as external standards, which were obtained from the instrument maker. The cross-links and control samples were extensively digested with trypsin, dialyzed thoroughly against H 2 O, and lyophilized. Control samples consisted of noncross-linked UV-irradiated gp96. The lyophilized samples were dissolved in a matrix solution, which consisted of a saturating amount of ␣-cyano-4-hydroxycinnamic acid in 50% (v/v) acetonitrile, 0.1% (v/v) trifluoroacetic acid (48). The samples (1-2 l) were then spotted on a sample holder plate and allowed to air dry. Molecular weight standards were similarly prepared. The mass spectra were acquired and the mass peaks were identified by matching their masses to gp96 trypsin fragments predicted using the computer program PROWL.
Molecular Modeling-gp96 sequence comparison against most entries in the protein data bank (Brookhaven National Laboratory) using BLAST and FESTA computer programs (36) yielded poor pairwise alignment. However, comparison of the peptide-binding domain of MHC class I 2mha (37) yielded blockwise amino acid compositional similarity although there was no pairwise sequence similarity. Using 2mha crystal structure (residues 1-175) as a guide, a structural model of the gp96 peptide-binding site (residues 570 -750) was built. In this model only gp96 residues 690 -696 were excluded because they would correspond to a turn or a short loop. The physicochemical properties of the amino acids in the presumed gp96-binding pocket and those in The sequence of the peptide with photoreactive azido group is shown. PepK was synthesized and purified by HPLC. It consists of a VSV-derived core sequence (RGYVYQGLKSG) fused to a protein kinase A phosphorylation sequence (LR-RASLGRS) (47). Purified azido-modified photoreactive PepK run on a Tricine-SDS-acrylamide gel (B). The numbers at the top indicate fractions from gel permeation chromatography of the 32  the 2mha binding pocket were plotted (Fig. 6A) using GCG's peptide structure program (38,39). The comparison of the electrostatic surface potentials of 2mha and the gp96 peptide binding pockets (Fig. 6, C and D) was generated using GRASP computer program (40). In the peptide, only the vesicular stomatitis virus (VSV) nucleocapsid peptide core sequence RGYVYQGLKSG was modeled because the crystallographic coordinates for most of this sequence were available (37). The modeled peptide has the N-to C-terminal orientation that is opposite to that of the peptide in 2mha because the cross-linking results show that the unique K9 in the peptide should be close (i.e. within cross-linkable distance) to the 623 KDKALKDK 630 sequence in gp96. This would place the peptide orientation as shown in Fig. 6. The model (Fig. 6) was generated using Molecular Simulations INSIGHTII version 97 program on a Silicon Graphics workstation.

RESULTS AND DISCUSSION
Peptide Design and Photocross-linking of the Peptide Ligand-In a previous study we showed that a VSV nucleocapsidderived peptide (SLSDLRGYVYQGLKSGNVS) bound to gp96 under a variety of conditions (33). Earlier, others (3,37,42) had shown that this VSV peptide sequence or subsequences thereof, were chaperoned by gp96 and were represented by MHC class I molecule for cytotoxic T-lymphocyte-specific immune response. Therefore we chose the immunogenic core sequence of the VSV peptide for mapping the peptide-binding site in gp96. We designed a chimeric 20-amino acid peptide (PepK; Fig. 1A) containing the VSV core sequence RGYVYQGLKSG followed by the sequence LRRASLGRS. The latter has a phosphorylation motif (RRXS), which permitted the introduction of a 32 P radiolabel on the Ser at position 16 of PepK using protein kinase A and [␥-32 P]ATP (see "Experimental Procedures"). The 32 P label allowed the gp96-PepK cross-links to be visualized by autoradiography following gel electrophoresis. This facilitated the specific identification and subsequent purification of crosslinks for the mapping of the peptide-binding site in gp96 (see below). To achieve covalent photocross-linking to gp96, PepK was modified by the attachment of a photoreactive azido group to the unique Lys (Fig. 1A) and the gp96-peptide complexes were exposed to near UV light. In a typical cross-linking experiment, PepK was first modified with the azido cross-linker (ASA), purified from the unincorporated cross-linker and then radiolabeled with 32 P. The 32 P-labeled ASA-modified PepK was further purified by gel filtration chromatography (Fig. 1B). The purified PepK (Fig. 1B) was then incubated with and crosslinked to gp96 to form SDS-resistant complexes (Fig. 1C), in agreement with previous observations (16,33). Covalent crosslinking of PepK to gp96 could be achieved by shining either room light or near UV (320 -360 nm) from a UV light source. Cross-linking with near UV gave better cross-link yields. The cross-linking resulted in a single sharp band in SDS-polyacrylamide gels suggesting a homogeneous specimen (gp96XL in Fig. 1C). Control lanes indicated that there were no cross-links to gp96 in the absence of the azido modification on PepK either with or without UV.
Mapping the Peptide-binding Site-Using the UV induced cross-linking technique shown in Fig. 1C, we prepared a large amount (ϳ2 mg of gp96) of cross-linked gp96 for the purpose of mapping the peptide-binding site. The gp96 cross-links were proteolyzed and the fragments were identified using a judicious combination of SDS-PAGE, peptide sequence determination using Edman degradation, and mass spectrometry. For each proteolytic cleavage, a computer-generated peptide mass bank of gp96 was used to order the proteolytic fragments on the gp96 primary sequence. The numbering system of gp96 peptides and amino acid sequences given in this work corresponds to the murine gp96 primary sequence including the signal sequence (Swiss-Prot accession number P08113; Refs. 41 and 44). A self-consistent pattern of fragmentation emerged, yielding the location of the peptide-binding site. Cleavage of cross-links with formic acid (Fig. 2A), which normally cleaves Asp-Pro bonds and X-Pro (X is any other amino acid) under more severe conditions (43), yielded two fragments by cleavage at Asp-Pro, 66.2 kDa (residues 236 to 802) and 50.5 kDa (369 to 802), and two more by cleavage at X-Pro, 41 kDa (residues 443 to 802) and 33.5 (residues 443 to 729) ( Fig. 2A). Mild digestion with cyanogen bromide (CNBr), which cleaves on the carboxyl side of Met, generated one prominent fragment of 33.5 kDa (lane 4, Fig. 2B), which could be residues 337 to 622, 371 to 658, or 427 to 711. Cleavage with endoproteinase Asn-C, which cuts after Asn, yielded fragments of 41.1 and 22.8 kDa, which were unambiguously assigned to residues 338 to 688 and 492 to 688, respectively (Fig. 2C). Harsh digestion with CNBr generated a 13.3-kDa fragment corresponding to residues 542 to 658, and a 3.9-kDa fragment that was assigned to residues 623 to 658 (Fig. 2D). The 3.9-kDa fragment arose from the 13.3-kDa frag- ment by CNBr cleavage at the single internal Met (Fig. 2D). The 4.5-kDa fragment extended from amino acids 623 to 662. This further narrowed the peptide-binding site to between residues 623 and 658. Digestion with endoproteinases Lys-C or Glu-C each resulted in cross-linked fragments containing the 623 to 658 region. However, precise assignment was difficult because of closely spaced potential cleavage sites. Therefore, NH 2 -terminal sequencing of a cross-link fragment was done. First, a cross-linked fragment containing a gp96 peptide of 11 kDa (Fig. 2, B, lane 3, and E, lane 2) that was generated by digestion with Endo Glu-C was selected for automated sequencing by Edman degradation. This fragment was selected because the probable NH 2 terminus of this fragment (residue 610) was closest to the NH 2 -terminal amino acid (residue 623) in the putative binding site (CNBr/harsh, Fig. 2D). The determined partial sequence from the NH 2 terminus of this 11-kDa Endo Glu-C fragment was 610-ATEKEFEPLLNW, which matched perfectly (12 of 12 matches) with the expected amino acid sequence based on the known DNA sequence of the mouse gp96 gene (41,44). A second cross-linked Endo Glu-C fragment was also partially sequenced by Edman degradation. It began with residue 615, FEPLLNWMKD-A (11 of 12 matches), and also fit the known gp96 sequence. The line (-) signifies that a Lys (Lys 625 ) was missing from the eleventh sequencing cycle, presumably because it photoreacted with the azido group on PepK (see below). These results from automated peptide sequencing confirmed that the peptide-binding site must be located between residues 615 and 658. To further refine the peptide-binding site, we used matrix-assisted laser desorption time of flight mass spectrometry of cross-links cut with trypsin. This robust protease is highly suitable for this purpose because of its high specificity and the large number of potential cleavage sites in gp96 (118 cleavage sites). In addition, trypsin cuts within PepK itself (5 cleavage sites), which could reveal the smallest contact region protected by gp96 in the cross-link. Two prominent mass peaks were present only in the cross-link spectrum (XL in Fig. 3) but not in the control (NC, Fig. 3). The observed mass peak ‫,ء(‬ m/z ϭ 3086 in Fig. 3) matched the sum of the mass of a cross-link between gp96 trypsin fragment DKALKDK (residue 624 to residue 630; mass 816 Da) and a trypsinized cross-linked PepK (mass 2272 Da after removal of the NH 2 -terminal Arg by trypsin). The expected mass of the conjugate (3086 Da) exactly matched the observed mass (3086 Da). No trypsin fragment of uncross-linked gp96 could have the observed mass, the closest being either 12 or 27 atomic mass units away. The mass determinations had an accuracy of 0.01% in the linear mass range of the analysis. Therefore we ruled out those fragments. Note that the sequence of the second crosslinked Endo Glu-C fragment (see above; 615 FEPLLNWMK-DKA 626 ) partially overlapped with that of the trypsin fragment 624 DKALKDK 630 identified by mass spectrometry, thus confirming that the smallest known region in gp96 contacting the peptide was from residue 624 to 630. Fig. 4 summarizes the mapping results from gel electrophoresis, mass spectrometry, and automated peptide sequencing within the context of the organization of the gp96 primary sequence. The important fragments are ordered according to the positions of their NH 2 and COOH termini. The peptide fragments nicely fall into an overlapping pattern in decreasing order of size. The smallest fragment comprising residues 624 to 630, which was generated by trypsin, defines the minimal peptide-binding site (P). There is only one peptide-binding site in gp96. Otherwise, parallel lineages of one or more large proteolytic fragment(s) giving rise to non-overlapping daughter fragments would have been observed. This was clearly not the case (Fig. 4). The gp96 protein has a signal sequence located at the NH 2 terminus and an endoplasmic reticulum retention/ retrieval signal (ER) at the COOH terminus. There is extensive homology with the cytosolic counterpart of gp96, HSP90 (41). Toward the NH 2 -terminal region there are two Walker boxes (A and B) which are suggestive of ATP-binding site(s). The dimerization domain is located in the COOH-terminal region (46), as is the peptide-binding site P (this work). As discussed below, the close juxtaposition of the peptide-binding site and the dimerization domain may have functional implications.
The Cross-linking Site-In an attempt to identify the exact amino acid that potentially reacted with the azido group in the modified PepK (Fig. 1), we turned to automated sequencing by Edman degradation of a cross-linked peptide generated in a harsh digest with CNBr (CNBr/harsh; peptide 623-658; Fig. 4). This peptide should be ideal for the resolution of the cross-linking amino acid site by automated sequencing because its NH 2 terminus contains the tryptic peptide identified in the mass spectrum (Fig. 4). Unfortunately, this cross-linked CNBr fragment (apparent combined mass ϳ6.9 kDa, labeled 3.9 in Fig. 2D) co-migrates on SDS gels with a noncross-linked CNBr fragment (7.1 kDa; residues 472 to 531). On an autoradiogram only the cross-linked fragment is seen because it carried the radiolabeled PepK (Fig.  2D). Both the radiolabeled cross-linked fragment and the nonradiolabeled noncross-linked fragment were transferred onto polyvinylidene difluoride membrane, thus yielding a mixture of the two sequences upon sequencing using Edman degradation. As expected, sequencing revealed the cross-linked CNBr fragment 623 KDKALKDK 630 , which overlaps with the tryptic fragment ( 624 DKALKDK 630 ) that was identified in the mass spectrum (see above). This result provides additional confirmation of the minimal peptide-binding site. However, the contaminating noncross-linked CNBr fragment had the sequence 472 IKKIA-DEK 479 . Note that both the cross-linked and noncross-linked fragments have Lys at positions 3 and 8. This introduced an unexpected complexity for the identification of the cross-linking site. The azido group in PepK undergoes photoreaction preferentially with the ⑀-amino group of lysines (45). Ideally, during automated sequencing using Edman degradation, the crosslinked Lys should be represented in very low picomole yield relative to other Lys in the sequenced region. This is because the cross-linked Lys is not a standard amino acid. Normally, crosslinked amino acids should be missing with the appearance of a blank in the sequencing cycle at that particular position. However, the presence of Lys from the noncross-linked peptide would obscure any blank cycle originating from the cross-linked peptide. It was impossible to apportion the picomole yields of Lys at positions 3 and 8 in the amino acid sequences for the mixture of CNBr fragments. However, when the picomole yields of the position-unique lysines for cross-linked (Lys 628 , KDKALKDK) and uncross-linked (IKKIADEK) peptides were taken as references, the Lys at position 3 in the cross-linked peptide (Lys 625 , KD-KALKDK) had an estimated 10-fold lower picomole yield than did Lys 628 or Lys 630 . Putting together this result and the fact that the Endo Glu-C fragment that was sequenced earlier had Lys 625 missing from its sequence (see above), we tentatively conclude that Lys 625 might have been the site of azido cross-linking. Although we believe that these results are compelling, we hasten to add that further analysis using other techniques may be necessary to confirm the cross-linked amino acid. From our point of view (see model in Fig. 6), the peptide binding pocket is a threedimensional entity and may indeed consist of ϳ200 amino acids within a domain. Therefore we believe that more definitive information regarding the exact cross-linked Lys in the sequenced contact region is not critical for our present understanding of the architecture of the peptide-binding site in gp96.
Relation to Previous Work-Earlier, we showed by fluores- FIG. 6. A, comparison of physicochemical properties (38,39) of gp96 peptidebinding site (top) and that of MHC class I molecule (40 -50% similarity). The gp96binding site residues (561-750) including the minimal contact region (KD-KALKDK) are compared with MHC class I molecule peptide-binding site 2mha (residues  cence that two peptides, each bound to monomer gp96 in a gp96 dimer or higher order complex, were held extremely close to each other (3.5 to 10 Å) in the complex (33). This observation is further supported by the direct observation by mass spectrometry of a cross-linked PepK dimer from complexes (**, m/z ϭ 4259 Ϯ 2; Fig. 3, XL). The observed mass of cross-linked PepK dimer fit the predicted mass (4257 Da). No peptide of this mass was seen in the control (Fig. 3, NC). The PepK dimer was also observed in gels (Fig. 2F, PD, lane 3) but only in the peptide-gp96 complexes (absent in lane 2). The covalent homodimerization of PepK independently confirmed the intermolecular distance (3.5 to 10 Å) that was ascertained by fluorescence (33) because the azido cross-linking group was on an 8.0-Å arm. In the PepK dimer cross-link (Fig. 3, XL), trypsin cleaved off Arg 1 , as it did in the monomer PepK-gp96 cross-link, and also removed Ser 20 . But the middle portion of the peptide is protected suggesting that the peptide-binding pocket is probably an openended partially solvent-exposed channel or a groove, which is in agreement with our fluorescence studies (33). Combining our fluorescence (33) and cross-linking results, it appears that residues 624 to 630 in gp96 must interface with PepK in a proposed peptide binding pocket.
A Molecular Model and Implications for Peptide Loading and Immunogenicity-The region around the peptide-binding site is highly conserved among gp96 paralogs (Fig. 5) implicating these residues in the formation of the peptide binding pocket. The gp96 peptide-binding domain was modeled based on amino acid compositional homology with the crystal structure of the peptide binding pocket of MHC class I molecule (37). Our model offers a framework for engineering by design new gp96 mutant proteins with altered peptide-binding properties that may result in better gp96-peptide vaccines, and it suggests that the dimerization domain may play a role in stable peptide binding.
Comparison of hydrophobicity, flexibility, and surface probability plots and of the electrostatic surface potential distribution of peptide binding pockets of gp96 and MHC class I molecule revealed partial similarity (Fig. 6, A, C, and D). This is interesting because peptide ligand is thought to be relayed by gp96 to MHC class I molecule for immunogenicity (12). Our core peptide RGYVYQGLKSG and that of the octapeptide RGYVYQGL in the MHC class I molecule co-crystal structure are largely identical. The peptide lies in a pocket surrounded by amphipathic ␣-helices (Fig. 6B). The floor of the pocket consists of antiparallel ␤-sheets as in the co-crystal of the MHC class I molecule with peptide. As per our cross-linking result, K9 (blue CPK) of the peptide is close to Lys 625 (blue sticks, bottom of Fig.  6B). The 8.0-Å linker attached to Lys 9 of the peptide should permit cross-linking to Lys 625 . An open conformation of the pocket, which is suggested based on Trp fluorescence (33), may allow peptide binding in a clamp-like fashion. gp96 residues 697 to 741, which form a dimerization domain (46) (green sticks, Fig. 6B), are close to the bound peptide. It is envisioned that gp96 dimerization may stabilize peptide binding such as to facilitate intercellular peptide transportation and faithful relay to MHC class I molecule (12).