Cysteine Mutagenesis and Homology Modeling of the Ligand-binding Site of a Kainate-binding Protein*

Glutamate receptors comprise the most abundant group of neurotransmitter receptors in the vertebrate central nervous system. Cysteine mutagenesis in combination with homology modeling has been used to study the determinants of kainate binding in a glutamate receptor subtype, a low molecular weight goldfish kainate-binding protein, GFKARβ. A construct of GFKARβ with no cysteines in the extracellular domain was produced, and single cysteine residues were introduced at selected positions. N-Ethylmaleimide or derivatized methanethiosulfonate reagents (neutral or charged) were used to modify the introduced cysteines covalently, and the effect on [3H]kainate binding was determined. In addition, cysteine mutants of GFKARβ transiently expressed in HEK293 cells were labeled with a membrane-impermeable biotinylating reagent followed by precipitation with streptavidin beads and specific detection of GFKARβ by Western blot analysis. The results are consistent with the proposal that the energy driving kainate binding is contributed both from residues within the binding site and from interactions between two regions (i.e. two lobes) of the protein that are brought into contact upon ligand binding in a manner analogous to that seen in bacterial amino acid-binding proteins.

iGluRs have a distinct modular structure ( Fig. 1A and Ref. 8). This consists of (i) one (S1S2, also known as LAOBP-like) or two (both S1S2 and leucine, isoleucine, valine-binding proteinlike) domains that are homologous to bacterial amino acidbinding proteins (9,10), (ii) a channel domain that has some similarities to cyclic nucleotide-gated and potassium channels (8,11), and (iii) a C-terminal domain that mediates regulation by phosphorylation (12) and binds intracellular proteins (13)(14)(15). In particular, the agonist and antagonist-binding domain (S1S2 domain; Refs. 8, 10, and 16) consists of a portion of the N terminus of the protein before the first transmembrane segment (M1) and a large portion of the protein between the third and fourth membrane associated segments (M3 and M4). The S1S2 domain has been modeled previously based on the relatively low homology with bacterial amino acid-binding proteins (16 -22) and can be expressed independently in bacteria (23,24) and insect cells (25,26) with an apparently native fold. From homology modeling and site-directed mutagenesis, six regions (R1-R6; Fig. 1B) were suggested to contribute to the binding site. The three-dimensional structure of the S1S2 domain of GluR2 has been solved recently by x-ray crystallography (27). Although differences exist between the crystal structure and previous homology models, the overall structure and many of the contributions to the binding site were predicted correctly. With the availability of a three-dimensional structure of GluR2, mutagenesis studies are particularly interesting for comparing different subtypes and for testing the importance of specific interactions. In particular, the static crystal structure of the agonist-bound form suggests that the two lobes comprising the S1S2 domain may close to surround the ligand upon binding, but further evidence is required to demonstrate this experimentally.
Cysteine mutagenesis is an extremely useful tool for probing structural features of proteins (28,29) if (i) individual mutations can be made in the absence of a cysteine background (to avoid both spurious disulfide bonds and adventitious interactions with sulfhydryl reagents) and (ii) the mutation can be made with little or no change in the binding site. Postexpression modifications of the protein can then be made by treat-ment with a sulfhydryl reagent, and the accessibility of the introduced thio-group can be tested by labeling with a biotinylating reagent. We have been using the kainate-binding proteins (GFKAR␣ and GFKAR␤; Ref. 30) from goldfish brain as model systems for understanding the ligand-binding site of iGluRs (31). As described here, a construct of GFKAR␤ with high affinity kainate binding can be produced with no cysteines in the extracellular portion of the protein. Thus, a single cysteine can be introduced without the possibility of forming disulfide bonds, which could in turn lead to a misfolded protein. Cysteine mutations were designed to probe protein-ligand interactions and sites of interaction between the S1 and S2 lobes. The results are interpreted in terms of a model of GFKAR␤, which is based on homology with GluR2. Expression in Cultured HEK 293 Cells-cDNAs were subcloned from the pRBG4 vector into the expression vector, pcDNA1.1/Amp (Invitrogen). HEK 293 cells, grown to 6 ϫ 10 5 cells/10-cm dish, were transfected (20 g of plasmid DNA/dish) using the calcium phosphate precipitation method, and cell membranes were prepared 48 h after transfection as described previously (31). Membrane pellets from each dish were resuspended in 0.5 ml (ϳ1 mg protein/ml) of 1 mM phenylmethylsulfonyl fluoride and stored at Ϫ20°C. Each binding assay contained 4 -8 g of protein.

Materials-[
[ 3 H]Kainate Binding Assay-[ 3 H]Kainate binding to membrane fragments of transfected cells was measured as described previously (31). The inclusion of 100 M kainate in the assays defined nonspecific binding. Equilibrium binding was analyzed by a nonlinear fit to the Hill is the concentration of bound ligand, and [L] is the concentration of free ligand. The K D values reported below correspond to the K D as given in this equation, which is the concentration of free ligand at half-maximal saturation (K D values are reported with standard errors based on three to four independent observations). Because of the variability of B max between transfections, only the K D and sensitivity to sulfhydryl reagents were compared. The data in Fig. 2 were normalized to B max in the absence of sulfhydryl reagents. To test the effect of NEM (2 mM) and the MTS reagents (200 M) on the cysteine substitution mutants of GFKAR␤, two parallel membrane samples were incubated at 25°C for 40 min in the absence and presence of a sulfhydryl reagent. The samples were then placed on ice, and the binding of [ 3 H]kainate was measured.
Biotinylation and Western Blot Analysis of Cysteine Substitution Mutants-48 h after transfection in HEK293 cells, cysteine mutants were biotinylated as described by Hall et al. (33) and Basiry et al. (34). Transfected cells were washed twice with 1ϫ PBS, and then the cells were incubated with biotin-HPDP (300 M) in 1ϫ PBS at room temperature for 1 h. After aspiration of the biotinylating solution, cells were washed twice with 1ϫ PBS to remove residual biotin-HPDP. The cells were removed from the dishes in 1ϫ PBS containing 1 mM EDTA and 1 mM phenylmethylsulfonyl fluoride, lysed by sonication, and centrifuged for 25 min at 12,000 ϫ g. The pellets containing membrane preparations were resuspended in 1 ml of solubilization buffer (1% Triton X-100 and 0.1% SDS in 1ϫ PBS) and centrifuged. The supernatant contained solubilized membrane proteins that were used for subsequent analysis. For each sample, 800 l of the solubilized membrane proteins were incubated with 120 l of streptavidin-linked beads at 4°C for 4 h. The beads were then centrifuged briefly in a microcentrifuge and washed three times with solubilization buffer. Proteins were eluted from the streptavidin-linked beads using two 60-l washes with 0.1 M glycine⅐HCl buffer (pH 2.2). Samples were mixed with 30 l of 5ϫ SDS gel loading buffer and heated for 5 min. Samples were compared using silver-stained gels before Western blotting. Each Western blot shown in Fig. 4 is representative of four to five independent replicates.
Construction of Cysteine Substitution Mutants-The cDNAs coding for three Cys to Ser single point mutations (␤C27S, ␤C305S, and Filled triangles indicate the key residues forming the binding site based on the crystal structure of the S1S2 domain of GluR2. Open triangles indicate additional residues playing an important role in binding according to the GluR2 S1S2 structure, and secondary structural elements are labeled according to Armstrong et al. (27). Those residues absent from the crystal structure are shown in lowercase letters. The figure was generated using ALSCRIPT 2.0 (44). Fig. 1B and Ref. 31) were used to create a mutant with all three extracellular cysteines mutated to serine. In addition, a 30-residue C-terminal deletion was introduced at the end of the cDNA coding sequence by PCR to remove Cys 410 and Cys 411 . An antisense primer (5Ј-CGCTCTAGATTAGGATTTCTTCTGCTC-3Ј), which contains a stop codon and a XbaI site after the Ser 409 , was used to amplify ␤C358S cDNA by PCR. The PCR DNA was cut with SacI and XbaI and used to replace the corresponding DNA sequence (310 base pairs) of ␤C305S. Next, an EcoRI/EcoRV fragment (400 base pairs) was exchanged with the corresponding one of ␤C27S. This resulted in a construct (␤⌬C5) with no extracellular cysteines and no cysteines in the C-terminal domain (i.e. ␤⌬C5 provides a background free of reactive cysteines). The construct was verified by DNA sequencing.
Molecular Modeling-The first step in the modeling procedure (18) was to divide the amino acid sequence of GFKAR␤ into domains and to extract the amino acid sequence corresponding to the S1S2 domain (17). Scanning through the proteins of known three-dimensional structure, the kainate-bound form of the S1S2 domain of GluR2 (27) had by far the highest sequence homology to the corresponding domain in GFKAR␤ (40% identity compared with a maximum of ϳ20% among other structures). The S1S2 domain of GluR2 was therefore used as the only structural template in the homology modeling (18) of GFKAR␤. Our previous sequence alignment (17), including both GFKAR␤ and GluR2, was used in conjunction with the program MODELLER (35) to generate a set of 10 models of the kainate-bound form of GFKAR␤. To achieve this, kainate was added to the MODELLER topology library; this then allowed MODELLER to reproduce automatically in the models proteinkainate interactions that were present in the crystal structure of GluR2 and for which equivalent atoms are present in GFKAR␤. Consistent stereochemical violations across this set of 10 models were removed by manual adjustment of the sequence alignment using CAMELEON (Oxford Molecular, Oxford, UK); the adjustment was determined by visual inspection, using InsightII (MSI, San Diego, CA), of the stereochemically strained regions of the models. In adjusting the sequence alignment, care was taken wherever possible not to introduce insertions and deletions into the crystallographically determined secondary structural elements of GluR2. Once the sequence alignment had been adjusted to remove all consistent stereochemical violations, the sequence alignment was used to produce a final set of 10 models. The model with the lowest energy was selected and analyzed in detail. In particular, the ability of the model to accommodate the sulfhydryl reagents was investigated using interactive molecular graphics (InsightII).

RESULTS
Construction of ␤⌬C5-Wild type GFKAR␤ contains a total of 10 cysteines in the fully processed protein, with an additional two in the signal sequence (30). Four of these cysteines are in membrane spanning regions (two in M1 and two in M3), and one is in a cytoplasmic loop between M1 and M3. These were left intact with the assumption that cysteines in the membrane regions would not be accessible and that binding would not be affected by modification in the cytoplasmic loop (this was confirmed experimentally; see below). A 30-residue truncation after Ser 409 in the C-terminal domain of the protein was made to remove two adjacent cysteines (Cys 410 and Cys 411 ). Finally, the three cysteines (Cys 27 , Cys 305 , and Cys 357 ) in the extracellular S1S2 domain were mutated to serine. Cys 305 and Cys 357 form a disulfide bond in the native protein but elimination of the disulfide bond by reducing agents or mutation to serines results in a protein with higher affinity for kainate (31,36). The sequence of the S1S2 domain is shown in Fig. 1B Table I. Only in the cases of ␤S245C and ␤S296C were significant effects on B max observed. In all cases, data are normalized to the B max in the absence of NEM. roughly consistent with the 3-fold increase in kainate affinity when the disulfide bond alone is removed by mutation of either Cys 305 or Cys 357 to Ser (31) but with otherwise normal binding characteristics. The expression level was approximately 2-3fold lower than wild type judging from the total binding and Western blot analysis.
Design and Expression of Cysteine Substitution Mutations-The strategy was to produce mutations to cysteine in or near the glutamate-binding site or at the interlobe interface. The two considerations were (i) to make relatively conservative changes (in most cases, Ser to Cys) and (ii) to place the mutations in positions that would not interfere with folding or [ 3 H]kainate binding in the absence of sulfhydryl reagents. Mutations were made in each of the regions (R1-R6; Fig. 1B) thought to affect agonist binding. In addition, several residues (S21C and G71C) were chosen as controls. That is, the homology models and the GluR2 structure place these residues in less conserved loops and away from the binding site and the interlobe interface. The absence of other cysteines in the extracellular domain would prevent the formation of spurious disulfide bonds. The cysteine could then be modified by NEM or another sulfhydryl reagent, and the effects of this modification on binding could be determined. In addition, the accessibility of the introduced cysteine residue and the effects of ligand binding on the accessibility can be examined by labeling using the biotinylation reagent. The positions of the mutations are shown in Fig. 1B. All of the mutant proteins were transiently expressed in HEK293 cells, as indicated in the Western blot with Ab-␤1 ( Fig. 2A).
[ 3 H]Kainate Binding-Each of the mutations, with the exception of ␤T247C, resulted in a protein with high affinity for kainate. As shown in Fig. 2B and Table I, of the cysteine mutants only ␤S271C differed by more than a factor of 3 in affinity from wild type; others were within a factor of 2 of the affinity of kainate for wild type GFKAR␤. The only mutant that did not exhibit measurable [ 3 H]kainate binding was T247C. As shown in Fig. 2A, this protein was well expressed. This observation is consistent with the mutational results on the homologous Ser 267 of chick kainate-binding protein (20). Based on the three-dimensional structure of GluR2 (27) and the homology model shown in Fig. 5A, the reason for the lack of binding is the direct role of this conserved threonine in coordinating the carboxymethyl group of kainate. The assertion that ␤⌬C5 is free of reactive cysteines is demonstrated by the negligible change in kainate binding when thiol reagents are added and the lack of labeling with biotin-HPDP for this construct (see Fig. 4A).
In lobe 1, two cysteine mutations were sensitive to NEM treatment (␤D13C and ␤A51C). ␤D13C is located next to the consensus VTTILE sequence ( 6 VTTIKQ 12 in GFKAR␤); mutations of the position corresponding to Gln 12 have pronounced effects on agonist affinity (20,31,37). The crystal structure of GluR2 places Glu 402 (which is equivalent to Gln 12 of GFKAR␤) at the interface between the two lobes, thus suggesting an involvement in interlobe interactions. ␤D13C is also near the domain interface. The ␤D13C mutation itself decreases affinity by 3-fold (Table I), and NEM decreases the binding by approximately 30% (Figs. 2B and 3B). The MTS series of sulfhydryl reagents (Fig. 3A) show results similar to that of NEM (Fig.  3B), suggesting that modification of this position most likely produces only a modest steric inhibition at the interlobe interface. The conserved Tyr 52 is critical for ligand binding (31,37), and two residues flanking Tyr 52 were tested here. In a previous study (31), ␤A51K decreased kainate affinity (31), although in these studies the mutational analysis does not distinguish changes in folding from changes in direct interactions with ligand. The modest effects of NEM, MMTS, and MTSES on ␤A51C (Fig. 3B) may be consistent with a minor allosteric contribution of this residue to binding. The GluR2 structure and the homology model described below (see Fig. 5A) place this residue near the interlobe interface. Thus, the contributions of this residue to the binding energy would be due to allosteric interactions rather than a direct contact with the ligand. Notably, however, the positively charged MTSEA and MTSET had no effect on [ 3 H]kainate binding to ␤A51C. In contrast, [ 3 H]kainate binding to ␤S21C, ␤R54C, ␤G71C, and ␤A84C is unaffected by NEM (Fig. 2B). The residues corresponding to ␤S21C, ␤R54C, ␤G71C, and ␤A84C in GluR2 are solvent exposed in the crystal structure (27) and the GFKAR␤ homology model based on GluR2 and therefore predicted not to be involved in either agonist binding or the interlobe interface. In the absence of agonist and antagonist, all cysteine mutations (in both lobe 1 and lobe 2) could be labeled in situ with biotin-HPDP (Fig. 4, B and C; see below).
In lobe 2, binding to ␤S245C is strongly affected by NEM modification. ␤S245C is adjacent to two serines that are involved directly in agonist binding in GluR2. The ␤S245C mutation alone decreases the binding affinity relative to wild type by 2-fold ( Fig. 2B and Table I), and modification by NEM reduces B max relative to unmodified ␤S245C by 2-to 5-fold. The decrease in B max (with no evidence for a second binding site) likely indicates that binding in all modified receptors is undetectable and that the binding that was detected may be to unmodified receptors. This is not unexpected given the predicted close proximity of Ser 245 to kainate in the binding site (as judged from the homology model presented below). Interestingly, NEM, MMTS, MTSEA, and MTSET all significantly decrease binding to ␤S245C. In contrast, negatively charged MTSES has little or no effect.
Another mutation is strongly affected by NEM modification (␤S296C). ␤S296C is near Glu 293 , which is predicted to interact with the amide of kainate and mutation of which severely affects ligand binding (38). NEM modification of ␤S296C could easily disrupt this interaction. However, it is more likely to interact directly with kainate; kainate comes within 6 Å of the side chain of S296, and the NEM could therefore partially block the binding site. The modification of ␤S296C by MTSEA, MTSET, and MTSES decreases binding to the same or a greater extent as NEM. MMTS, on the other hand, has little effect. This result suggests that the effects of the modification are likely to be steric rather than affected by charge.
Previous mutagenesis results and modeling studies (17, 18, 20, 21, 38) differed significantly in their interpretations of the TABLE I Binding parameters of cysteine substitution mutants K D values for the 12 mutants whose binding is displayed in Fig. 2B. Because of the lack of significant cooperativity, the Hill coefficient was restrained to 1, and nonlinear least squares fits were used to calculate K D and B max .
a B max was decreased in NEM-treated membranes relative to untreated membranes. role of the sequence between R4 and R6 (Fig. 1B). The cysteine substitution mutations ␤S258C, ␤S271C, ␤S275C, and ␤K282C were introduced to explore the role of this region. ␤S258C was chosen as a site in lobe 2 that was not expected to be associated with the binding site. ␤S271C is homologous to Ser 684 of GluR6, which was originally thought to be a site for cAMP-dependent protein kinase phosphorylation (39,40). Ser 275 is the position equivalent to Asn 721 in GluR6, the residue that seems to be involved in controlling affinity for AMPA (21). Finally, the conserved Lys 282 was originally thought to be in the binding site based on modeling studies (17). All four mutants expressed normally ( Fig. 2A) but are insensitive to NEM modification (Fig. 2B), suggesting that all four are outside of the binding site. This interpretation is entirely consistent with the GluR2 structure (and our GluR2-based model), which places these residues away from the binding cleft (27). However, although the mutation of Ser 271 to Cys did not affect expression, it decreased noticeably affinity for [ 3 H]kainate (104 Ϯ 23 nM versus 13 Ϯ 2 nM for ␤⌬C5). As will be discussed below, this site is likely to be involved in a conformational change associated with ligand binding.
Biotinylation of cysteine Substitution Mutants-Of the total of 12 cysteine mutants of GFKAR␤, four (D13C, A51C, S245C, and S296C) showed sensitivity of [ 3 H]kainate binding to chemical modifications by NEM or MTS derivatives. Labeling with the membrane-impermeable biotinylation reagent (biotin-HPDP) was used to address the question of whether introduced cysteine residues are solvent exposed and, more interestingly, whether ligand-binding would change the accessibility.
HEK293 cells expressing single cysteine mutants were reacted with biotin-HPDP. After solubilization of the membrane proteins, proteins with covalently bound biotin were recovered on and subsequently eluted from streptavidin-linked beads. Western blots were then used to determine whether the mutant GFKAR␤ was among the proteins eluted from the streptavidin-linked beads, indicating labeling by biotin-HPDP. Wild type GFKAR␤ was not labeled by biotin-HPDP (Fig. 4A, lanes  1 and 2), consistent with the fact that of the three extracellularly located cysteines, Cys 305 and Cys 358 form a disulfide bond (31), and Cys 27 is partially buried. The C305S mutant was readily labeled (Fig. 4B, lanes 3 and 4) because the disulfide can no longer form and Cys 358 becomes a free and solventexposed cysteine. ␤⌬C5, which has the three extracellularly located cysteines removed, showed no biotin-HPDP labeling, as expected (Fig. 4C, lanes 5 and 6).
All cysteine substitution mutants on lobe 1 could be labeled with biotin-HPDP (Fig. 4B), indicating that they are all solvent exposed. Of these, the labeling of ␤D13C was inhibited by kainate. The modest effect of NEM on binding and the potent inhibition of biotin labeling by kainate supports the notion that this residue is at the interface between the two lobes. Likewise, in lobe 2, all cysteine substitution mutants could be labeled with biotin-HPDP (Fig. 4C). Of those tested, the labeling by biotin-HPDP of ␤S271C, ␤S275C, and ␤S296C was inhibited by kainate (Fig. 4C). In addition to kainate (lane 2), CNQX (lane 4), and glutamate (lane 3) inhibited labeling of ␤S275C and ␤S296C by biotin-HPDP (Fig. 4D). In the case of ␤S271C, both kainate and glutamate inhibited biotin labeling, but CNQX was less potent (Fig. 4D).
Homology Model of GFKAR␤-Previous homology models of the binding domain of glutamate receptors were based on bacterial amino acid-binding proteins (16 -18, 20 -22). Considering the low resolution inherent in homology models, the overall structure of the S1S2 domain was reasonably well represented. Nevertheless, the relatively low homology between the bacterial proteins and glutamate receptors made modeling somewhat problematic because of uncertainties in the sequence alignment. With the availability of the three-dimensional structure of the S1S2 domain of GluR2 (27), more accurate models of other subunits can be constructed because of the much higher sequence homology (sequence identity of ϳ40% versus ϳ20%) that likely corresponds to a more similar struc-ture. The differences between the previous homology models and the crystal structure arise for two reasons: (i) difficulties in aligning the sequences of the iGluRs with the sequences of the LAOBP-like structures, leading to errors in mapping structural features from the templates to the models and (ii) the kainatebound form of GluR2, unlike complexes of the LAOBP-like structures, is thought not to have undergone full domain closure (26). Solvent accessibility calculations on the model of GFKAR␤ (using NACCESS) reveal that, unlike in the previous models, the reactive thiol is exposed to solvent in all 13 introduced cysteine residues. The accuracy of the new GluR2-based models is supported further by the finding that all cysteine substitution mutants could be labeled with biotin-HPDP (Fig.  4), as predicted by the model. DISCUSSION The model of GFKAR␤ built based on the GluR2 S1S2 domain template can guide the interpretation of mutagenesis studies, help to demonstrate differences in subtypes, and provide information on conformational changes upon binding. Both mutagenesis studies and the GluR2 structure suggest that agonist affinity is controlled by two classes of residues: (i) those that are found in the binding site interacting with ligand and (ii) those that are involved in interlobe contacts, which in turn can affect the energetics of lobe closure and therefore indirectly affect ligand binding. Homology modeling of glutamate receptors based on bacterial amino acid-binding proteins identified six regions (R1-R6) thought to form the binding site (18,19). Of those regions, R2, R3, R4, and R6 are found in the binding site. Amino acids in R1, R2, R4, and R5 are involved in interlobe contacts.
Binding Site-Considering first the residues that line the binding pocket (Fig. 5B), most determinants of binding seem to be well conserved across iGluR subtypes. In R2, Tyr 52 interacts with the pyrrolidine ring and isopropenyl groups of kainate (27), and mutation to serine (but not phenylalanine) completely eliminates binding activity (31). The amino acid at this position is tyrosine or phenylalanine in all iGluRs. In R3, a highly conserved arginine residue (Arg 86 ) forms a salt bridge to the ␣-carboxyl group of kainate. In an AMPA receptor, the modification of the homologous Arg (Arg 507 ) to Lys was sufficient to abolish binding (37,38). The homologous mutation in the chick kainate-binding protein also completely abolished binding, although this residue was excluded from the binding site in the model of that protein (20).
In R4, T247 interacts with the carboxymethyl group of kainate, and mutation of this residue to cysteine, as described above, completely abolishes binding but not expression of the protein. In the model, Ser 245 is also in a position to form a hydrogen bond with the carboxymethyl group of kainate. The equivalent position in GluR2 is Gly 653 . Mutation of Ser 245 to cysteine decreases binding affinity by 4 -5-fold relative to ␤⌬C5, suggesting that this hydrogen bond could form but that its contribution to the binding energy is less than Thr 247 . Modification of ␤S245C with NEM and the MTS reagents reveals an interesting pattern (Fig. 3B). MMTS, MTSEA, and MTSET significantly reduce binding, but negatively charged MTSES has no effect. This is a bit surprising because the MTSESmodified S245C would be in a position to interact with Arg 86 (the residue in R3 that forms the salt bridge with the 2-carboxyl group of kainate) and the amide of Gly 53 . The negative charge could interfere with binding both because of its proximity to the kainate carboxyl and its putative interaction with R86. However, as shown in Fig. 5C, MTSES can be positioned so as to allow kainate to bind. In contrast, the positively charged MTSET can lie in the same position as the amide of kainate (Fig. 5D), thereby blocking the binding site. Interest- ingly, ␤S245C can be labeled with biotin-HPDP both in the presence and absence of kainate. This would suggest that at least in the cysteine mutant, position 245 is solvent exposed in the bound complex. Thus, Ser 245 is likely to contribute only a small amount of binding energy and, because the equivalent position is glycine in AMPA receptors as well as GluR5, GluR6, and GluR7, its role may be unique to kainate-binding proteins.
The highly conserved Glu 293 in R6 coordinates the amino group of kainate. ␤S296C has no effect on kainate affinity, but modification by sulfhydryl reagents significantly affects binding. As shown in Fig. 3B, NEM and the MTS reagents (with the exception of MMTS) dramatically decrease kainate binding. Presumably, this is a steric effect because MMTS is the smallest reagent used, and both positively and negatively charged reagents have the same effect. The side chain is predicted to point directly into the binding site and could easily block the binding of kainate (Fig. 5E). As indicated by inhibition of biotin-HPDP labeling by kainate, glutamate, and CNQX, this site is no longer solvent exposed in the presence of agonist or antagonist.
Interlobe Interface-The interlobe interface consists of several points of contact that can affect the ligand specificity allosterically (21,27). Two points of contact have been identified in the crystal structure of GluR2 (27) and consist of contacts between R1 and R5 and between R2 and R4. In GluR2, the corresponding contacts are Glu 402 -Thr 686 and Lys 449 -Asp 651 -Ser 652 . The fact that GFKAR␤ has a Gln (Gln 12 ) corresponding to Glu 402 and an Ala (Ala 51 ) corresponding to Lys 449 could provide additional insight into this interface. Q12E results in an increase in kainate affinity (31), suggesting that, as with the equivalent residue Glu 402 in GluR2, this residue could be involved in interlobe interactions. Inspection of the model suggests that a hydrogen bond with the backbone amide of Met 276 (the position equivalent to Thr 686 of GluR2, Fig. 1B) is the most likely interaction and that interaction with Ser 275 is unlikely. The latter is consistent with other experiments; with Gln in position 12, ␤S275C does not affect kainate affinity, nor does modification of ␤S275C by NEM. Both ␤D13C and ␤S275C are occluded upon agonist binding in that they are no longer available for biotin-HPDP labeling in the presence of either agonist or antagonist. This is consistent with the positioning of these residues outside of the binding site but at the domain interface. The fact that the sites are exposed in the absence but not the presence of agonist and antagonist is suggestive of a domain closure similar to that seen in bacterial amino acid-binding proteins. Because Ser 275 is analogous to the position in GluR6 that controls AMPA affinity (21), the differences in agonist affinity between AMPA and kainate receptors maybe be at least partially because of allosteric interactions at the interlobe interface rather than to direct contact with ligand.
␤A51K decreases the binding affinity for kainate by 2.5-fold (31). NEM treatment of ␤A51C decreases kainate affinity by approximately 3-fold (Fig. 2B). Perhaps surprisingly, however, MTSEA and MTSET have no effect on ␤A51C (Fig. 5B). This suggests that the aliphatic chain of these two reagents could produce a hydrophobic interaction that cannot be produced by the shorter aliphatic chain of lysine. Linked to this hypothesis, the model also suggests the possibility that the charged groups on the longer MTSEA and MTSET could interact with Asp 56 , whereas a lysine could not reach this far. Thus, at least in the case of kainate binding, the putative interaction between the lobes at this site does not extensively affect ligand affinity. This is supported by the finding that ␤A51C was accessible to biotin-HPDP in the presence and absence of kainate, suggesting that, at least in its unmodified form, conformational changes resulting from ligand binding do not affect the solvent accessibility of this residue.
␤S271C-Position 271 is predicted to be neither in the binding site nor at the interdomain interface. Nevertheless, the ␤S271C mutation decreased binding affinity by more than 4-fold. Likewise, the labeling by biotin-HPDP is decreased in the presence of kainate and glutamate and consistently less so with CNQX. Similar results have been observed in GluR6 (34). Although this position is at the end of a stretch of relatively low homology between GFKAR␤ and GluR2, it is almost certainly located in the loop between the G and H helices and is outside of the cleft that defines both the binding site and the interface between the two lobes. The postulated conformational change, which accompanies binding and which is based on homology with bacterial amino acid-binding proteins, is assumed to be a closure of the two lobes. Because the conformational change decreases the solvent accessibility of Ser 271 , which is neither at the binding site nor the interlobe interface, a modification of the structure of at least lobe 2 is likely to take place in addition to the lobe closure. Thus, conformational changes accompanying binding likely involve more than a rigid body movement of the two lobes relative to each other and may differ for agonists and antagonists. This position is of particular interest because in GluR6, a mutation to alanine abolishes the effects of cAMPdependent protein kinase phosphorylation (39,40).
Site-directed mutagenesis has been used to probe the ligandbinding site of various iGluRs. These included, among others, the AMPA receptor subunits (37,38,41), the kainate receptor subunits (21), chick (20, 42) and goldfish (31) kainate-binding proteins, and the N-methyl-D-aspartate receptor subunits (22,43). The results of these studies have provided valuable infor-mation that was an indispensable component in the development of homology models (17)(18)(19). However, the interpretation of mutagenesis results are complicated by two factors: (i) In some cases, differential effects of mutations at homologous sites of different subtypes occurred. For example, E424Q and E424A of GluRD (GluR4; Ref. 38) did not affect kainate binding, but the homologous E33V mutation of chick kainate-binding protein decreased kainate binding by greater than 10-fold (20) and the homologous Q12E mutant of GFKAR␤ increased affinity for kainate and glutamate (31). (ii) The loss of binding activity cannot be unambiguously interpreted as a loss of a specific interaction in the binding site because it could equally be because of a large structural disturbance. In the case of a model of the chick kainate-binding protein (20), some key residues with dramatic effects on ligand binding (e.g. Arg 107 and Glu 316 ) were placed outside of the binding site, resulting in a model of the binding site with no positively charged residues and based on hydrogen bonding alone. Cysteine substitution mutagenesis can partially overcome some of the problems of standard mutagenesis. Assuming that the mutation to cysteine does not greatly affect binding and that the cysteine is accessible for modification, the effects of cysteine modification are less likely to result in misfolding of the protein. Furthermore, a single mutation can be modified by a variety of sulfhydryl reagents to test the effects of positive and negative charges. Thus, cysteine mutagenesis is a useful complement to standard mutagenesis, homology modeling, and direct structural analysis.
Summary-With the availability of an atomic resolution structure of GluR2, which has considerable sequence identity to GFKAR␤, and the subsequent production of a model based on the GluR2 structure, the interpretation of the cysteine mutants and their modification by sulfhydryl reagents as well as site-directed mutagenesis can be made with a much higher level of confidence than has been previously possible. The results support the notion that the energy for binding kainate is contributed both by essential residues in the binding site as well as by allosteric interactions between two distinct regions of the protein. The use of biotin-HPDP to assess the solvent accessibility of each cysteine substitution provided additional information on the conformational changes associated with binding. The findings support previous suggestions that the closure of the two lobes accompanies binding (8,16); however, the changes in the accessibility of position 271 suggest that the closure of the lobe likely involves more than a rigid body movement of the lobes. Furthermore, based on the inhibition of biotin labeling for residues 13 and 275, the lobe closure can occur both upon binding of agonist and antagonist, suggesting that a more subtle conformational change may be associated with gating.