Lysine Scanning Mutagenesis Delineates Structural Model of the Nicotinic Receptor Ligand Binding Domain*

Nicotinic acetylcholine receptors (AChR) and their relatives mediate rapid chemical transmission throughout the nervous system, yet their atomic structures remain elusive. Here we use lysine scanning mutagenesis to determine the orientation of residue side chains toward core hydrophobic or surface hydrophilic environments and use this information to build a structural model of the ligand binding region of the AChR from adult human muscle. The resulting side-chain orientations allow assignment of residue equivalence between AChR subunits and an acetylcholine binding protein solved by x-ray crystallography, providing the foundation for homology modeling. The resulting structural model of the AChR provides a picture of the ACh binding site and predicts novel pairs of residues that stabilize subunit interfaces. The overall results suggest that lysine scanning can provide the basis for structural modeling of other members of the AChR superfamily as well as of other proteins with repeating structures delimiting a hydrophobic core.

Nicotinic acetylcholine receptors (AChR) and their relatives mediate rapid chemical transmission throughout the nervous system, yet their atomic structures remain elusive. Here we use lysine scanning mutagenesis to determine the orientation of residue side chains toward core hydrophobic or surface hydrophilic environments and use this information to build a structural model of the ligand binding region of the AChR from adult human muscle. The resulting side-chain orientations allow assignment of residue equivalence between AChR subunits and an acetylcholine binding protein solved by x-ray crystallography, providing the foundation for homology modeling. The resulting structural model of the AChR provides a picture of the ACh binding site and predicts novel pairs of residues that stabilize subunit interfaces. The overall results suggest that lysine scanning can provide the basis for structural modeling of other members of the AChR superfamily as well as of other proteins with repeating structures delimiting a hydrophobic core.
Traditional methods for atomic structural determination use x-ray crystallography or NMR spectroscopy. However, many proteins do not form crystals, and many are too large to solve by NMR. New methods are therefore urgently needed to determine structures of such intractable proteins. Here we develop a mutagenesis-based modeling method and apply it to the ligand binding region of the nicotinic AChR 1 from adult human muscle. The method uses lysine scanning to distinguish core hydrophobic from surface hydrophilic orientations of residue side chains and uses this information to align residues in AChR subunits with equivalent residues in the homologous AChBP, which was solved by x-ray crystallography (1). The experimentally determined alignment forms the foundation for generating an atomic structural model of the ligand binding region of the heteromeric AChR.
Knowledge of nicotinic receptor structure advanced along two independent lines of investigation over the past decade. The first line stemmed from primary sequence data deduced from cloning AChR subunits and their relatives (2). These studies included prediction of membrane spanning regions us-ing hydropathy analysis, prediction of secondary structure from the sequence data (3), identification of key residues by affinity labeling together with microsequencing, and site-directed mutagenesis combined with functional measurements (4 -6). The emerging picture of the AChR indicated a pentamer in which about half of each subunit contributes to an N-terminal extracellular domain that harbors the ACh binding sites. The remainder of each subunit consists of four transmembrane domains, the second of which contributes to the ion channel, and a large cytoplasmic domain between the third and fourth transmembrane domains. Mutagenesis and site-directed labeling localized the ACh binding sites to interfaces between ␣ and non-␣ subunits (4,6), whereas studies of residue accessibility and residues affecting ion permeability revealed the channel gate and ion selectivity filter within the second transmembrane domain (7,8).
Advances along the second line of investigation came from cryo-electron microscopy of two-dimensional arrays of AChRs from Torpedo. Data at 9-Å resolution revealed the overall shape and dimensions of the AChR and identified secondary structural elements in the vicinity of the ACh binding site and the ion channel (9). The most recent data at 4.6-Å resolution revealed several aligned ␤ strands near the putative ACh binding site, and a fenestrated basket-like structure extending into the cytoplasm (10). Images obtained following rapid application of ACh revealed that each subunit twists about the axis normal to the membrane when the ion channel opens (11). Overall, the cryo-electron microscopy data could be reconciled with the mutagenesis, labeling, and functional measurements, although the precise locations of the ACh binding sites and channel gate remained controversial.
Atomic structural insight recently emerged from the crystal structure of AChBP, a 120-kDa acetylcholine binding protein homologous to the ligand binding region of the AChR (1). The sequence of AChBP is 23.9% identical to that of the homomeric ␣ 7 AChR and harbors residues considered diagnostic of nicotinic receptor ␣ subunits, including disulfide-bonded cysteines and aromatic residues that contribute to the ACh binding site. These conserved elements suggest that AChBP may provide a model for the three-dimensional structure of the ligand binding domain of the AChR and its relatives.
Here we combine lysine scanning mutagenesis, ligand binding measurements, and homology modeling to deduce a structural model of the ligand binding region of the nicotinic AChR from adult human muscle. Lysine scanning of the ⑀ subunit reveals that it oligomerizes with complementary AChR subunits when lysine is placed at alternating positions along the protein chain, indicating the presence of ␤ strands and establishing orientation of the side chains toward core hydrophobic and surface hydrophilic environments. The side-chain orientations allow alignment of equivalent residues between the ⑀ subunit and AChBP, forming the foundation for homology mod-eling. The resulting atomic structural model provides a detailed picture of the AChR ligand binding site and discloses novel residue pairs that stabilize subunit interfaces.

EXPERIMENTAL PROCEDURES
Materials-125 I-Labeled ␣-bgt was from PerkinElmer Life Sciences, d-tubocurarine chloride from ICN Pharmaceuticals, Inc., 293 human embryonic kidney cells (293 HEK) were from the American Type Culture Collection, ␣-conotoxin GI was from Sigma Chemical Co., and the fully methylated analog of d-tubocurarine, metocurine iodide, was a gift from the Eli Lilly Co.
Plasmids and Mutagenesis-Human adult AChR subunit cDNAs were obtained as described previously (12) and subcloned into the cytomegalovirus-based expression vector, pRBG4, as described (13). Mutations were constructed using the QuikChange kit from Stratagene. All mutations were confirmed by dideoxy sequencing.
Expression of Mutant Receptors and Ligand Binding Measurements-HEK cells were transfected with mutant or wild type AChR subunit cDNAs using calcium phosphate precipitation as described (13). In all experiments, AChR subunit cDNAs were co-transfected in the following quantities per 10-cm plate of HEK cells: ␣ (13.6 g) and ␤, and ⑀ and ␦ (6.8 g). Three days after transfection, intact HEK cells were harvested by gentle agitation in phosphate-buffered saline plus 5 mM EDTA. Ligand binding to intact cells was measured by competition against the initial rate of 125 I-␣-bungarotoxin (␣-bgt) binding (14). After harvesting, the cells were briefly centrifuged, resuspended in high potassium Ringer's solution, and divided into aliquots for ligand binding measurements. Potassium Ringer's solution contains 140 mM KCl, 5.4 mM NaCl, 1.8 mM CaCl 2 , 1.7 mM MgCl 2 , 25 mM HEPES, 30 mg/liter bovine serum albumin, adjusted to pH 7.4 with 10 -11 mM NaOH. Specified concentrations of competing ligand were added 30 min prior to adding 125 I-␣-bgt, which was allowed to bind for 30 min to occupy approximately half of the surface receptors. Binding was terminated by addition of 2 ml of potassium Ringer's solution containing 100 M d-tubocurarine chloride. Cells were immediately filtered through Whatman GF-B filters using a Brandel cell harvester and washed five times with 2 ml of potassium Ringer's solution. Prior to use, filters were soaked in water containing 4% dried skim milk for at least 2 h. Nonspecific binding was determined in the presence of 1 mM d-tubocurarine. The total number of ␣-bgt binding sites was determined by incubation with the toxin for 90 min. The initial rate of ␣-bgt binding was calculated as described (14) to yield fractional ligand occupancy. Competition measurements were analyzed according to one of the following equations: the Hill equation (Equation 1), the sum of two binding sites present in equal numbers (Equation 2), one binding site plus a constant (Equation 3), the sum of two binding sites present in unequal numbers (Equation 4), and weighted contributions of Equations 2 and 3 (Equation 5), where Y is fractional ligand occupancy, n is the Hill coefficient, K APP is an apparent dissociation constant, K A and K B are intrinsic dissociation constants, Fract A is the fraction of sites with dissociation constant K A , and Fract 2-site is the fraction of binding sites with equal numbers of K A and K B sites. Homology Modeling-We generated a homology model of the major extracellular domain of the adult human AChR using version 6.0 of the program MODELER (15), together with spatial restraints provided by the AChBP structure (1). We aligned AChR and AChBP sequences based on side-chain dispositions determined by lysine scanning combined with ligand binding measurements, as described under "Results" (Table I and Fig. 4). However, to model strand ␤1, spatial restraints were removed for AChR ⑀L40 and two flanking residues to account for its surface exposure (see Fig. 5), which is contrary to the corresponding aligned residue, Ile-38, in AChBP. To maintain complementarity between subunits at their interfaces, all five subunits were modeled simultaneously. AChR ␣ subunits were matched to A and C subunits of AChBP, and the ⑀, ␦, and ␤ subunits were matched with the B, D, and E subunits, respectively. We used the "patch" command in MODELER to constrain coordinates of cysteines 128 and 142, which form a disulfide bond in each subunit. Among several options in MODELER, we selected the "refine1" mode, which generates the highest level of refinement using conjugate gradients coupled with simulated annealing and molecular dynamics. Modeling included all polar hydrogens to allow for main-chain hydrogen bonding but omitted non-polar hydrogens. We used MODELER to generate 100 different structures, and evaluated each structure using the programs PROCHECK (34) and PRO-FILES-3D (InsightII, Accelrys Inc., San Diego, CA). Among the 100 structures, ϳ10 were of high quality and very similar, and we selected the best of these for further modeling.
The major insertions requiring additional modeling were the linkers between strands ␤8 and ␤9 in ␤, ⑀, and ␦ subunits, which contain eight, eight, and eleven inserted residues, respectively. The ␣ subunit contains only one insertion in this linker, which MODELER is designed to accommodate (17). To model the ␤8 -␤9 linker in the ⑀ subunit, a second round of MODELER was employed using different sequence alignments in this region and constraining coordinates of all atoms except those of the ␤8 -␤9 linker. This was an empirical process, and we selected structures that satisfied several criteria. First, we required ⑀F160 in this linker, and equivalent residues in other subunits, to project into the hydrophobic core to account for the observation that ⑀F160K produced predominantly misfolded subunits (Table II). Second, we selected structures in which endogenous positively charged residues in the linker project away from the hydrophobic interior; these included ⑀K171, ⑀K167, ␦K164, ␦K167, and ␦R170. Third, we required ⑀D175 and ␦D180 to approach within 10 -15 Å of ␣C192/193 determined in cross-linking experiments (16) and within 15-20 Å of ⑀L119 or ␦L121 determined from double-mutant cycles analysis of ␣-neurotoxin binding (18). Although these constraints are qualitative, in practice they eliminated many candidate structures. PROCHECK and PROFILES 3-D were used to select the best structure for further modeling.
Finally, we applied two rounds of energy minimization using the program CHARMM (19), version 27b4. The first round constrained the coordinates of all heavy atoms while allowing movement of hydrogens. The second round constrained the protein main chain while allowing movement of side chains. The final structure was selected based on evaluations using PROCHECK and PROFILES 3-D.

Lysine Scanning Mutagenesis to Assess AChR Secondary
Structure-We mutated individual residues to lysine along a putative ␤ strand from positions 49 through 61 of the ⑀ subunit from adult human muscle AChR. Each lysine mutation was co-transfected with complementary ␣, ␤, and ␦ subunit cDNAs in 293 HEK cells, followed by quantitation of cell surface receptors using the AChR-specific ligand 125 I-␣-bgt. The results reveal a pattern of high expression alternating with low expression as lysine is advanced along the protein chain ( Fig. 1a). High expression for odd numbered residues approaches that for wild type AChR, whereas low expression for even numbered residues approaches that for subunit-omitted ␣ 2 ␤␦ 2 receptors. High expression likely results from mutant ⑀ subunits that fold and incorporate into pentameric AChRs, whereas low expression likely results from misfolding of the mutant ⑀ subunits, leading to subunit-omitted ␣ 2 ␤␦ 2 receptors (12,20,21). These results provide strong evidence that residues 49 through 61 in the AChR ⑀ subunit form a ␤ strand.
Alternatively, lysine mutations at even numbered positions may produce a subunit that can assemble with complementary subunits, but the mutations reduce the efficiency of receptor assembly. To distinguish between omission of the ⑀ subunit and inefficient assembly, we took advantage of the observation that ligand binding properties of the receptor depend on subunit composition (12,20,22). Thus for each mutation, we measured binding of ACh and ␣-conotoxin GI (CTx GI) by competition against the initial rate of 125 I-␣-bgt binding. Receptors containing the ⑀ subunit (␣ 2 ␤⑀␦) show a monophasic competition curve for the agonist ACh, whereas receptors lacking the ⑀ subunit (␣ 2 ␤␦ 2 ) show a distinctive biphasic curve with a plateau at half occupancy extending over three orders of magnitude in ACh concentration (12,21) (Fig. 1b). Analogous results are obtained with the competitive antagonist CTx GI, where ␣ 2 ␤⑀␦ receptors show a competition curve with two closely spaced components, whereas subunit-omitted receptors show a high affinity component accompanied by a plateau extending over nearly three orders of magnitude of CTx GI concentration. Thus competition profiles for ACh and CTx GI clearly distinguish receptors containing the ⑀ subunit from those with the composition ␣ 2 ␤␦ 2 .
For the odd numbered lysine mutations, both ACh and CTx GI fully compete against binding of the reporter ligand 125 I-␣bgt (Fig. 1b). Quantitative differences in the competition pro-  Table III. For the CTx GI competition measurements, the smooth curves are fits using the two-site equation (Equation 2) or one site plus a constant (Equation 3), with the fitted parameters given in Table I; the fitted curve for I56K was obtained using Equation 4, with the fitted parameters given in Table II files are observed among the various mutations, because residues in this region contribute to the ligand binding site (23,24). By contrast, lysine mutations at even numbered positions lead to biphasic profiles characteristic of subunit-omitted ␣ 2 ␤␦ 2 receptors ( Fig. 1b). Together with the alternating pattern of high and low expression, these results indicate that lysine mutations at odd numbered positions permit incorporation of ⑀ subunits into pentameric AChRs, whereas lysine mutations at even numbered positions lead to ⑀ subunits unable to incorporate into pentamers.
For even numbered lysine mutations in strand ␤2, all but one show a plateau in the competition curve corresponding to ϳ50% of the binding sites, as observed for subunit-omitted ␣ 2 ␤␦ 2 receptors. However, for the ⑀I56K mutation, the plateau corresponded to 28% of the binding sites (Fig. 1b). Because a plateau of 50% corresponds to 100% ␣ 2 ␤␦ 2 receptors, a plateau of 28% indicates 56% ␣ 2 ␤␦ 2 and 44% containing the mutant ⑀ subunit (i.e. ␣ 2 ␤␦⑀I56K). Co-expression of the ⑀I56K mutant yields total expression of ␣-bgt binding sites of 25% of control, of which only 11% contain the mutant ⑀ subunit. The reduced expression, together with the reduced plateau in the competition measurements, mirrors observations for a disease-causing mutation in the human ⑀ subunit (12). Thus the ⑀I56K mutation markedly impairs the ability of the subunit to oligomerize into assembled receptors.
Residues 49 through 61 are analogous to those in strand ␤2 of AChBP, which is the center of three antiparallel ␤ strands that form a sheet delimiting a hydrophobic core on one face and the protein surface on the other (1). The inability of subunits with even numbered lysine mutations to assemble with complementary subunits likely owes to a large energetic penalty for placing a positively charged side chain in a hydrophobic core, together with the hydrogen-bonded ␤ strands that attempt to hold it in place. We therefore conclude that residues 49 through 61 in the ⑀ subunit form a ␤ strand. Fig. 2 exhibits structures of the antiparallel ␤ sheet containing strand ␤2 from our model of the ⑀ subunit described herein and the corresponding ␤ sheet in AChBP. In strand ␤2 of the ⑀ subunit, residues intolerant to lysine mutation are highlighted in blue, whereas those tolerant to mutation are highlighted in yellow. In strand ␤2 of AChBP, residues projecting into the hydrophobic core are highlighted in blue, whereas those on the protein surface are highlighted in yellow. Our results demonstrate that residues intolerant to lysine mutagenesis project their side chains into the hydrophobic core, whereas those tolerant to lysine mutagenesis project their side chains on the protein surface. Moreover, the results establish alignment of equivalent residues between the ⑀ subunit and AChBP where key residues contributing to the ligand binding site in the ⑀ subunit, ⑀W55 and ⑀D59, align with their counterparts in AChBP (Trp-53 and Thr-57).
Lysine Scanning Mutagenesis of Putative ␤ Strands 1 through 10 of the ⑀ Subunit-We next applied lysine scanning and ligand competition measurements to the nine remaining regions in the ⑀ subunit that potentially form ␤ strands. For these measurements, we used a single ligand, CTx GI, to distinguish receptors that incorporate the mutant ⑀ subunit from those that do not (i.e. subunit-omitted ␣ 2 ␤␦ 2 receptors). Lysine mutations that permit incorporation into pentameric AChRs are identified as those yielding ␣-bgt binding sites in the range observed for wild type AChR, together with complete competition by CTx GI against the initial rate of ␣-bgt binding. Lysine mutations that prevent incorporation into pentameric AChRs are identified as those yielding low expression of ␣-bgt sites, together with the signature biphasic competition profile for CTx GI indicating subunit-omitted receptors.
Out of 110 lysine mutations examined, the majority yielded clear-cut distinction between those that incorporate into pentameric AChRs and those that do not (Table I). However, nine mutations yielded intermediate results suggesting both assembly-competent and assembly-incompetent ⑀ subunits. For these mutations, competition profiles for CTx GI were similar to but quantitatively distinct from that for subunit-omitted ␣ 2 ␤␦ 2 receptors ( Fig. 3a). In particular, the plateau in the competition profiles ranged from 5 to 35% of the maximal rate of ␣-bgt binding, rather than the expected 50%, as seen for the mutation ⑀I56K in strand ␤2 (Fig. 1b). Furthermore, close inspection of the competition profiles revealed a component with intermediate affinity for CTx GI, suggesting the presence of receptors containing the mutant ⑀ subunit along with subunit-omitted ␣ 2 ␤␦ 2 receptors. In support of this interpretation, the CTx GI competition data are well fitted by the weighted sum of contributions of ␣ 2 ␤⑀␦ and ␣ 2 ␤␦ 2 receptors ( Fig. 3a and Table II).
To confirm that mutant ⑀ subunits in this intermediate assembly class are present in cell surface AChRs, we carried out competition measurements using metocurine, the fully methylated analog of d-tubocurarine, which binds to the ␣-⑀ site with high affinity. The metocurine competition measurements reveal a high affinity component characteristic of the ␣-⑀ site, along with a low affinity component characteristic of the ␣-␦ site ( Fig. 3b and Table II). Relative contributions of high and low affinity components, determined from fitting the metocurine competition data, differ among the different mutations, but are similar to the relative contributions estimated from fitting the CTx GI competition measurements (Table II). Thus competition measurements using metocurine confirm intermediate assembly characteristics of several mutations.
We classified mutations with intermediate assembly charac- Mapping side-chain dispositions from lysine scanning on our model of the AChR ⑀ subunit and the known dispositions in AChBP (1). The antiparallel strands ␤1, ␤2, and ␤6 are shown, with space-filling rendering of residues in strand ␤2; residues colored yellow project on the protein surface, whereas those colored blue project into the hydrophobic core.

TABLE I Expression and CTx GI competition parameters for lysine mutations
Expression is the total number of cell surface ␣-bgt binding sites relative to that for wild type ␣ 2 ␤⑀␦ AChRs (see Fig. 1a and "Experimental Procedures"). For receptors in which CTx GI competed against all ␣-bgt binding (Fig. 1b), K A and K B are dissociation constants determined by fitting Equation 2 to the data. For receptors in which Ctx GI did not compete against all ␣-bgt binding (highlighted in boldface; see Fig. 1b), K A is the dissociation for the high affinity component, and Fraction A is the fraction of sites with dissociation constant K A determined by fitting Equation 3 to the data.   . 1b). For AChBP, residues highlighted in boldface orient toward the hydrophobic core (see Fig. 4, legend). b NM, residue was not mutated.
teristics into assembly-competent and assembly-incompetent categories. The assembly-incompetent category comprises ⑀I56K, ⑀L92K, ⑀W118K, ⑀S148K, ⑀F160K, and ⑀I215K. These mutations yield similar numbers of ␣ 2 ␤⑀␦ and ␣ 2 ␤␦ 2 receptors while reducing expression of receptors containing the mutant ⑀ subunit to 10 -15% compared with the wild type ⑀ subunit (Table II). These hallmarks of reduced expression parallel those described for a disease-causing mutation in the human ⑀ subunit (12), supporting the classification as assembly-incompetent. The assembly-competent category comprises ⑀P191K, ⑀I194K, and ⑀T208K; these incorporate into ␣ 2 ␤⑀␦ receptors, where expression is 30 -50% compared with the wild type ⑀ subunit, and subunit-omitted ␣ 2 ␤␦ 2 receptors form. Four other mutations, ⑀L109K, ⑀Y111K, ⑀T117K, and ⑀P121K, express in high amounts similar to wild type AChR yet show biphasic competition curves for CTx GI. Because each of these residues contributes to the ligand binding site (4,6), the biphasic competition profiles likely owe to impaired binding of CTx GI to the mutant binding sites. To confirm that these mutant ⑀ subunits incorporate into pentamers, rather than yield subunit-omitted receptors, we examined ACh competition against 125 I-␣-bgt binding. The mutations ⑀Y111K and ⑀T117K exhibit monophasic competition profiles similar to wild type AChR, confirming that these mutant subunits incorporate into cell surface AChRs (Fig. 3c and Table III). The mutations ⑀L109K and ⑀P121K show broad competition profiles shifted to high concentrations of ACh, but the profiles clearly differ from that for subunit-omitted ␣ 2 ␤␦ 2 receptors (compare Figs. 1b and  3c). Thus, although these four mutations impair ligand binding, they incorporate into cell surface AChRs. In summary, out of 110 mutations examined, 68 permit incorporation into cell surface receptors, whereas 42 produce subunits with limited or no ability to incorporate into surface receptors.
In addition to experimental results for 110 mutations, we infer surface or core side chain dispositions for 12 more residues. Residues with positively charged side chains are assumed to project on the protein surface; these include ⑀K34, ⑀R79, ⑀R125, ⑀R147, ⑀R195, ⑀R196, ⑀R202, and ⑀R203. A glycosylation site, ⑀N141, and its consensus residue, ⑀S143, are also assumed to project on the protein surface. On the other hand, dispositions of the disulfide-bonded cysteines 128 and 142 could not be assessed by mutagenesis, owing to their absolute requirement for subunit folding (25,26). These cysteines are assumed to project into the hydrophobic core of the subunit, as observed for the corresponding cysteines in AChBP (1).
Homology Modeling of the Major Extracellular Domain of the Human Muscle AChR-The most important requirement in homology modeling is correct alignment of the sequence to be modeled with that of the template structure. Our determination of side-chain disposition in the ⑀ subunit, together with known side-chain dispositions in AChBP, allows correct alignment of sequences in the ten ␤ strands and several linkers between the strands. Furthermore, the alignment of ⑀ subunit and AChBP sequences provides a basis for aligning ␣, ␤, and ␦ sequences with that of AChBP. Additionally, for five of the ten loops between ␤ strands, alignment is straightforward, because there are no insertions between AChR subunits and AChBP. The remaining five loops contain insertions and therefore constitute the greatest potential sources of error in modeling; however, four of these loops contain one to three insertions, which the homology modeling program MODELER is designed to accommodate (15,17). The remaining loop is located between strands ␤8 and ␤9 and contains an eight-to eleven-residue insertion in ␤, ⑀, and ␦ subunits. Our determinations of sidechain disposition are overlaid upon the overall sequence alignment used in modeling (Fig. 4).
We used MODELER in the Insight II software environment to generate a homology model of the major extracellular domain of the adult human AChR using spatial constraints provided by AChBP (see "Experimental Procedures"). To maintain complementarity between subunits at their interfaces, all five subunits were modeled simultaneously. The output structure from MODELER was corrected for misfolded linkers between strands ␤8 and ␤9 in ␤, ⑀, and ␦ subunits, followed by energy minimization (see "Experimental Procedures").
Mapping the Lysine Scanning Data on the Model of the AChR ⑀ Subunit-The overall mutagenesis results are mapped on our structural model of the ⑀ subunit, using yellow to indicate mutations producing folded and blue to indicate mutations producing misfolded subunits (Fig. 4). The map reveals alternating patterns of folded and misfolded subunits in nine of the ten ␤ strands, whereas strand ␤5, which contains two residues, shows no alternating pattern ( Fig. 4 and Table I). Nine of the ␤ strands align as pickets of a fence along the vertical axis of the subunit, forming a ␤ sheet that wraps in a cylinder to delimit a hydrophobic core on one face and the protein surface on the other. In the upper half of the sheet, the mutagenesis map reveals precisely aligned stripes of yellow alternating with blue that run normal to the long axis of the subunit. Each row of the For reference, the thin lines are mean results for ␣ 2 ␤⑀␦ and ␣ 2 ␤␦ 2 receptors obtained by fitting Equations 2 and 3, respectively. Smooth curves through the data are fits to the weighted sum of these two equations (Equation 5), assuming 50% high affinity sites for ␣ 2 ␤␦ 2 receptors, with parameters given in Table II. b, metocurine competition measurements for the indicated mutations and control receptors. The smooth curves are fits obtained using the two-site equation in which the fraction of each site is variable (Equation 4); fitted parameters are given in Table II. c, ACh competition measurements for the indicated receptors. The smooth curves are fits obtained using either Equation 1 or 2, with fitted parameters given in Table III. yellow or blue encoded residues forms a ring perpendicular to the long axis of the subunit, and these rings stack one upon another.
In the lower portion of the ␤ sheet, the map reveals alternating core and surface side chains, as observed in the upper portion of the sheet, but in several strands consecutive residues project their side chains on the protein surface. These consecutive surface residues correspond to local regions where the strand twists away from the hydrophobic core, exposing consecutive side chains on the protein surface. These twisted regions of strands ␤1, ␤6, and ␤9 correspond to regions in AChBP in which the ␤ strands also twist away from the core (1). Thus lysine scanning not only establishes alignment of side chains between the ⑀ subunit and AChBP but also identifies local regions where the ␤ strand twists away from the hydrophobic core.
In a departure from the AChBP structure, we find that the side chain of ⑀L40 projects on the protein surface, contrary to the equivalent residue I38 in AChBP, which projects into the hydrophobic core (Fig. 4). Incorporated into our structural model, the surface disposition of ⑀L40 means that five consecutive residues of strand ␤1 project their side chains on the protein surface. Moreover, surface disposition of ⑀L40 eliminates an inward kink in strand ␤1 present in AChBP; instead the main chain of strand ␤1 aligns co-linearly with the main chain of strand ␤6 as the pair wrap around the ⑀ subunit (Fig.  5). Thus surface exposure of these five consecutive side chains owes to twisting of strand ␤1 away from the hydrophobic core together with removal of the kink at position ⑀40. This differ-ence between the AChR ⑀ subunit and AChBP may be structurally important, because leucine is conserved at this position in all heteromeric AChR subunits, whereas isoleucine is present in homomeric AChR subunits.
Projection of Side Chains into the Hydrophobic Core and on the Protein Surface-Space-filling rendering shows that residues tolerant to lysine mutagenesis project on the protein surface, whereas residues intolerant to mutagenesis project into the protein core (Fig. 6). To illustrate side-chain disposition, we divide the ⑀ subunit into three regions bounding the hydrophobic core: upper, lower, and outer regions.
The upper region contributes to the subunit interface formed with the ␣ subunit to create one of two ACh binding sites and comprises strands ␤3, ␤5Ј, ␤8, and the upper halves of strands ␤6, ␤2, and ␤1 (Fig. 6a). All residues tolerant to lysine mutagenesis project on the protein surface. Out of 18 side chains that project on the surface, nine are hydrophobic, including Thr, Leu, Pro, and Trp, whereas nine are hydrophilic or charged, including Asp, Asn, Glu, Gln, Lys, and Arg. Residues previously shown to contribute to the ligand binding site, ⑀W55 (24), ⑀L109 (36), ⑀Y111 (27), ⑀T117 (13), and ⑀L119 (28), project on the protein surface. By contrast, all residues intolerant to lysine mutagenesis project into the hydrophobic core. These core residues are exclusively hydrophobic, and include Val, Leu, Ile, Phe, and Trp.
The lower region comprises the lower portions of strands ␤1, ␤2, ␤6, and the short ␤5. These collectively form the inner wall of the subunit facing the central vestibule, through which ions pass in route to the transmembrane channel (Fig. 6b). A relatively large proportion of residues in this region tolerates mutation to lysine, owing to twisting of strands ␤1, ␤6, and ␤5 away from the hydrophobic core. As a result, one edge of the protein backbone projects toward the central vestibule, allowing exposure of polar groups in the backbone and consecutive side chains to the aqueous exterior. Surface exposure of the polar backbone may compensate for the relatively small proportion of hydrophilic side chains on the protein surface. Only six out of 18 surface exposed residues in this region are hydrophilic, including Ser, Gly, Asn, and Arg, whereas the remaining residues are hydrophobic, including Ala, Thr, Val, Leu, Ile, and Tyr.
The outer region contributes to the outer wall of the subunit and the subunit interface not involved in ligand binding and includes strands ␤9, ␤10, ␤7, and ␤4 (Fig. 6c). Residues that tolerate lysine mutagenesis project their side chains on the protein surface, with one exception: ⑀I215. ⑀I215 is expected to project on the protein surface yet is classified as assemblyincompetent because mutation to lysine leads to similar num-TABLE II Ctx GI and metocurine competition parameters for lysine mutations yielding subunit-containing (␣ 2 ␤⑀␦) and subunit-omitted receptors (␣ 2 ␤␦ 2 ) For CTx GI competition results, parameters are fits of Equation 5 to the data where K A and K B are dissociation constants and Fraction two-site is the fraction of sites corresponding to ␣ 2 ␤⑀␦ receptors, which is used to compute the ratio ␣ 2 ␤␦ 2 /␣ 2 ␤⑀␦. For metocurine competition results, parameters are fits of Equation 4 to the data where K A and K B are dissociation constants and Fraction K A is the fraction of sites with dissociation constant K A . The fraction of ␣ 2 ␤⑀␦ receptors is computed as twice Fraction A and used to compute the ratio ␣ 2 ␤␦ 2 /␣ 2 ␤⑀␦.

Mutant
CTx GI parameters Metocurine parameters bers of ␣ 2 ␤⑀␦ and ␣ 2 ␤␦ 2 pentamers and expression is markedly reduced (Table II). Reduced expression of ⑀I215K may owe to instability between the engineered lysine side chain and the endogenous ⑀R217, which projects on the same side of strand ␤10. Of the remaining 15 surface exposed side chains in the outer region, eight are charged or hydrophilic, similar to the upper hydrophobic region, and include Ser, Asp, Asn, Glu, and Arg. The remaining surface exposed side chains are hydrophobic and include A, V, I, P and F. All residues intolerant to lysine mutagenesis project their side chains into the hydrophobic core; these are predominantly hydrophobic, including Val, Leu, Ile, Phe, Tyr, and Trp, with Ser and Gly the hydrophilic exceptions.
AChR Subunit Interfaces-Our model of the AChR ligand binding domain shows complementarity between juxtaposed subunits, with the characteristic twisting of each subunit forming protrusions and grooves that interlock the pentameric array (Fig. 7). Each pair of subunits shows three regions of close approach at the top, middle, and bottom of the interface. Particular residues in each region span the subunit interface, with the partners approaching close enough to interact through a variety of non-covalent forces, including salt bridges, hydrogen bonds, and hydrophobicity.
In the top region of inter-subunit contact, our model reveals a salt bridge between ␣R20 and ⑀E4 and between equivalent residues at each subunit interface (Fig. 7b). These salt bridges are distinct from the bifurcated salt bridge in AChBP (1), which FIG. 4. Side-chain dispositions and resulting sequence alignments of AChR subunits with AChBP. Lower panel, side-chain dispositions and resulting sequence alignments of AChR subunits with AChBP. Surface and core residues are highlighted in yellow or blue, respectively, for the experimentally determined orientations from lysine scanning the AChR ⑀ subunit and AChBP (1). For AChBP, surface and core side-chain orientations were assessed from side-chain contacts using the program 3D-PSSM (35) and confirmed by visual inspection of the structure. ␤-Strands are shown above the sequences. The upper panel shows secondary structure rendering of the AChR ⑀ subunit, with results from lysine scanning overlaid. Regions colored yellow correspond to positions where lysine mutation produced a folded subunit, whereas regions colored blue correspond to positions where lysine mutation produced a misfolded subunit. The left view is from the subunit interface that forms the ligand binding site with the ␣ subunit. The right view is rotated 180°and shows the opposite interface and the region facing the inner vestibule. Results from lysine scanning are shown, with residues in yellow corresponding to folded and residues in blue corresponding to misfolded following mutation to lysine. a, ␤-sheet delimiting the upper hydrophobic core, including strands ␤3, ␤5Ј, ␤8, and the upper halves of ␤6 and ␤1. The left view is from the subunit interface that forms the ligand binding site with the ␣ subunit, and the right view is rotated 180°. b, ␤-sheet facing the inner hydrophobic core, including strands ␤5 and the lower portions of ␤1, ␤2, and ␤6. The left view is from the inner vestibule, and the right view is from the top and rotated 180°. c, ␤-sheet delimiting the outer hydrophobic core, including strands ␤9, ␤10, ␤7, and ␤4. The left view is from the outer surface of the subunit, and the right view is from the top of the subunit.
FIG. 7. Homology model of the overall structure of the ligand binding domain of the adult human AChR. a, view of the structure from the top with the ␣ subunit highlighted in magenta, the ⑀ subunit in yellow, and the remaining three subunits in gray. b, the structure viewed from the side with residue side chains highlighted at three key regions of close approach of the subunits (see text). c, a stereo view of the ACh binding site at the ␣-⑀ subunit interface. Main chains are magenta for the ␣ subunit and yellow for the ⑀ subunit. Side chains of key conserved aromatic and hydrophobic residues are displayed along with three non-conserved residues that align in a row above and to the right of the binding site center (see text). muscle and heteromeric neuronal AChR subunits. These salt bridges may stabilize the pentamer in either open or closed states and are candidates for breaking and reforming during channel gating, as established for salt bridges in prototypical allosteric proteins (29).
The middle region of inter-subunit contact gives rise to the ACh binding sites, which are accessible from the periphery of the pentamer (Fig. 7, b and c). These are formed at ␣-⑀ and ␣-␦ subunit interfaces, as established by mutagenesis, site-directed labeling, and the AChBP structure. The center of the ACh binding pocket contains aromatic and hydrophobic side chains from each interface partner. The aromatic residues, ␣W149, ␣Y93, and ⑀W55, merge at the center of the binding site, and these are flanked by the hydrophobic residues ␣V91, ⑀L119, ⑀P121, and ⑀T36 (Fig. 7c). Additional aromatic residues, ␣Y190 and ␣Y198, line the outer wall of the binding site cavity. Each of these nine residues, or their aligned equivalents, is highly conserved and present at both ␣-⑀ and ␣-␦ binding sites; they may constitute minimal structures necessary for ACh binding. Residues ⑀L109, ⑀T117, and ⑀D59 form a row above and to the right of the binding site. ⑀D59, which is not conserved, contributes to selectivity of competitive antagonists for the ␣-⑀ over the ␣-␦ site, because it differs from its counterpart in the ␦ subunit, ␦G61 (23). Similarly ␥Y117 and its equivalent residue ␦T119 in the fetal mouse AChR contribute to selectivity of competitive antagonists for the ␣-␥ over the ␣-␦ site (13). These and other non-conserved residues at the binding site may constitute structures that fine tune ACh binding to meet the demands of the motor synapse in a given species.
The bottom region of inter-subunit contact contains residue partners that may form hydrogen bonds across the interface (Fig. 7b). At the ␣-⑀ site, these residues include ␣Y127, ⑀N39, and ⑀N182, whereas equivalent residues are present at the ␣-␦ site. Residue ␣Y127 flanks ␣C128 that forms the signature Cys loop found in all AChR subunits, ⑀N39 projects on the protein surface from strand ␤1, and ⑀N182 is located in the linker just N-terminal to strand ␤9. Residues ␣Y127 and ⑀N39 are situated close enough to form an amide-aromatic hydrogen bond. Toward the periphery from this pair, the amide group of ⑀N182 closely approaches the main chain carbonyl of ␣C128 to which it may form a hydrogen bond. ␣Y127, ⑀N39, and ⑀N182 are conserved in all species of the corresponding AChR subunits, suggesting critical contributions at both ligand binding interfaces. DISCUSSION We introduce lysine scanning mutagenesis as a method to delineate repeating structures that separate hydrophobic from hydrophilic regions in proteins. The method takes advantage of instability of the positively charged side chain in a hydrophobic environment and of its stability in the polar environment of the protein surface. We show that the AChR ⑀ subunit contains ␤-strands that delimit a hydrophobic core and compare predicted side-chain dispositions with known dispositions in the homologous AChBP (1). Our results establish register of side chains between ␤ strands in the ⑀ subunit and those in AChBP, identify local twisted regions in the ␤-strands and identify structural differences between AChBP and the AChR. Our experimentally determined sequence alignment provides the basis for building a homology model using spatial constraints from the homomeric AChBP. The resulting structural model provides a detailed picture of the ACh binding site of the adult human AChR, which is heteromeric and contains multiple conserved residues that differ from residues in equivalent positions in AChBP. The model also discloses previously unrecognized residue pairs at subunit interfaces, which emerge as candidates for mediating allosteric transitions triggered by agonist. The structural model provides clear predictions that can be tested by mutagenesis to further refine the structure and provide insight into function. The overall findings suggest that lysine scanning may be used to generate reliable structural models of other members of the AChR superfamily, as well of other proteins not amenable to x-ray crystallography or NMR.
Lysine is rarely found in protein interiors unless stabilized by other charged or polar groups. An engineered lysine is not likely to encounter such stabilizing groups in the protein interior and, therefore, should be highly unstable. When the engineered lysine is located in a repeating structure such as a ␤-strand, main-chain hydrogen bonds attempt to hold it in place, causing a structural perturbation that prevents oligomerization. Lysine scanning mutagenesis has been applied to peripheral loops in human thyrotropin to produce superactive analogs (38) and to voltage-gated sodium channels to identify sites for local anesthetic block (39). However, the present study is the first to apply lysine scanning to distinguish core hydrophobic from surface hydrophilic environments of residue side chains. Using ligands specific for subunit composition, our experiments assess the presence or absence of the structural perturbation by determining whether the mutant subunit oligomerizes with complementary wild type subunits to form cell surface AChRs. Null mutants, which do not oligomerize, give rise to subunit-omitted ␣ 2 ␤␦ 2 pentamers, which produce a unique biphasic ligand binding signature (12,21). The two ligand binding sites in the ␣ 2 ␤␦ 2 pentamer bind ␣-bgt at the same rate, but only one of these binds ACh and CTx GI with high affinity. Why only one binding site in ␣ 2 ␤␦ 2 receptors achieves high affinity for these ligands is not known but owes to the absence of the ⑀ subunit, which allows ready identification of null mutants.
The null mutants likely fold partially and associate with ␣ and ␤ subunits, but these do not assemble into cell surface AChRs. Evidence for partial folding and association comes from our observation that the number of ␣-bgt binding sites was reduced in the presence of the null mutants compared with ␣, ␤, and ␦ subunits alone (Table I and Fig. 1a); the null mutant ⑀ subunits likely compete for ␣ and ␤ subunits, reducing expression of ⑀-omitted receptors. Out of 110 mutants examined, nine exhibit detectable but incomplete incorporation into cell surface AChRs, because both ⑀-containing and ⑀-omitted receptors are detected. All but one of these mutations localizes to the C terminus of a ␤-strand ( Fig. 4 and Table II), a position that may more readily allow the strand to twist away from the hydrophobic core.
Our structural model of the AChR extracellular domain (Fig.  7) shows differences as well as similarities compared with the AChBP structure (1). Because our experimental results allowed side-chain alignment of the ␤ strand regions, these regions are the most similar between the two proteins. However, strand ␤1 shows altered side-chain disposition in a stretch of consecutive surface-exposed residues, which we incorporate into our structural model (Fig. 5). Additional differences are evident in the peripheral strands ␤9 and ␤10; these are greatest in non-␣ subunits, which contain one or more prolines in strand ␤9, unlike AChBP, and smallest in the ␣ subunit, which lacks proline in this region (Fig. 4). Differences are also present in the ligand binding sites where residues conserved in all heteromeric AChRs, ⑀T36, ␣V91, ⑀L109, ⑀L119, and ⑀P121, differ from the corresponding aligned residues in the homomeric AChBP, Lys-34, Ala-87, Arg-104, Met-114, and Ser-116 (Figs. 4 and 7). Larger differences are evident in linkers between ␤-strands, particularly the linkers between strands ␤8 and ␤9 and between strands ␤9 and ␤10. For the ␤8 -␤9 linker, the differences are small in the ␣ subunit, which contains only one residue inserted, but largest in the non-␣ subunits, which contain eight to eleven inserted residues. For the ␤9 -␤10 linker, differences are again smallest in the ␣ subunit, which contains a single inserted residue and the vicinal cysteines 192 and 193, also present in AChBP, and greatest in non-␣ subunits, which contain one to three inserted residues and lack vicinal cysteines.
The ACh binding sites are formed at ␣-⑀ and ␣-␦ subunit interfaces, as established by mutagenesis, site-directed labeling, and the AChBP structure. Mutagenesis and site-directed labeling established that seven loops, far apart in the linear sequence, contribute to the ACh binding site: loops A, B, and C in the ␣ subunit and loops D, E, F, and G in ⑀ or ␦ subunits (4,6). The AChBP structure confirmed that residues in each of these loops are present at the ACh binding site, and our model provides a detailed picture of both common and unique residues in the human muscle AChR. In the AChR, residues in each of the seven loops localize to the ligand binding site: ␣Y93 in loop A, ␣W149 in loop B, ␣Y190 and ␣Y198 in loop C, ⑀T36 in loop D, ⑀W55 in loop E, ⑀L119 and ⑀P121 in loop F, and ⑀D175 in loop G. Each of these residues is highly conserved and may constitute minimal structures necessary for ACh binding. Additional residues, predominantly non-conserved, have been identified by mutagenesis and are located at the periphery of the binding site. These include ␣G153 in loop B (37); ␣S187, ␣V188, and ␣T189 in loop C (18,30); ⑀K34 in loop D (27, 31); ⑀D59 in loop E (23); ⑀L109, ⑀Y111, ⑀S115, and ⑀T117 in loop F (13,27,32,36); and ⑀E177 in loop G (18,33). These peripheral residues may constitute structures specialized for binding ACh at concentrations found at the motor synapse or for releasing bound ACh with sufficient speed to terminate the response.
Beyond their importance in forming the ACh binding sites, subunit interfaces are critical because the fundamental twisting movement in channel activation displaces one subunit relative to its neighbors (11). Inter-subunit interactions can be broadly classified into those stabilizing resting, active, or desensitized states and further classified into strong or weak interactions. The strong interactions are candidates for breaking and reforming during allosteric transitions between functional states, whereas the weak interactions may provide complementary surfaces between juxtaposed subunits in one state or the other. Our model contains a salt bridge between ␣R20 and ⑀E4, as well as between equivalent residues at each subunit interface (Fig. 7b). The salt bridges in the AChR are distinct from the bifurcated salt bridges in AChBP (1), which form between side chains from different regions of the juxtaposed subunits. Salt bridges in the AChR may provide strong inter-subunit stabilization in one functional state or another and therefore emerge as candidates for breaking and reforming during channel gating or desensitization, analogous to salt bridges that break and reform during activation of classical allosteric proteins (29). Thus our results provide tangible starting points for future experiments aimed at correlating AChR structure with function.
A general theme regarding subunit interface structure emerges from our model of the AChR, namely that each subunit interface contains contact regions that are essentially the same at all interfaces, as well as contact regions that are unique and highly specialized. Common interface structures include the salt bridges at the top region of contact in all subunits. Specialized structures include the ligand binding sites, which are present only at ␣-⑀ and ␣-␦ interfaces. At the remaining three subunit interfaces analogous to the ACh binding site, specialized contact residues are present and likely make critical contributions to function. For example at the ⑀-␣ interface, ⑀E93 (equivalent to ␣Y93) closely approaches ␣R55 (equivalent to ⑀W55), suggesting a salt bridge. ␣R55 is conserved in all muscle ␣ subunits, whereas ⑀E93 is conserved in all ⑀ and ␥ subunits, supporting a critical functional role for a salt bridge between these residues.
Our model of the AChR extracellular domain naturally has limitations. Regions containing insertions relative to AChBP are the greatest sources of uncertainty in modeling; these comprise five of the ten linkers between ␤-strands and the peripheral ␤-strands ␤9 and ␤10. The linker between strands ␤8 and ␤9 requires further modeling, perhaps using distance constraints between bound ligand and residue side chains, combined with measurements of ligand binding following mutation of candidate residues. The three ␣-helices predicted from AChBP have not been examined in the AChR, but as they lodge against the hydrophobic core, they could be examined by lysine scanning. Our results from mutating ⑀T215 are not readily explained by our structural model, because the ⑀T215K mutation incorporates only partially into cell surface receptors, yet its side chain is predicted to project onto the protein surface. A likely explanation is that the ⑀T215K mutation places its positive charge proximal to the endogenous residue ⑀R217 on the same side of the ␤ strand, which is unstable due to electrostatic repulsion.
Lysine scanning may also prove useful when applied to other structurally intractable proteins. It may allow side-chain disposition to be determined in other regions of the AChR ⑀ subunit, including helices and perhaps short linkers. Longer linkers are expected to be flexible and allow reorientation of the substituted lysine side chain, unlike ␤-sheets where hydrogen bonds hold the main chains in place and thus constrain the side chains. Lysine scanning also holds promise for examining the AChR ␣, ␤, and ␦ subunits as well as subunits in other members of the AChR superfamily, such as ␥-aminobutyric acid (type A), glycine, and 5-hydroxytryptamine-3 receptors. Perhaps the greatest potential lies in deducing reliable structural models of the many intractable proteins expected to emerge from genomics and proteomics studies. When an unknown protein is homologous to a known protein with repeating structures delimiting a hydrophobic core, secondary structure and side chain register can be determined by lysine scanning coupled with a measure of protein folding. The results should allow correct alignment of model and template sequences and generation of reliable homology models.
Future studies will be aimed at refining our structural model of the AChR ligand binding domain. These refinements will likely require distance constraints generated by docking ligands whose three-dimensional structures are known, coupled with the effects of mutations on their binding. Molecular dynamics simulations are possible with the structural model presented here and will yield additional refinement by including explicit solvent molecules to provide more realistic energy minimization of the structure. Once a refined structural model is obtained, simulation of ligand association with the ACh binding site should reveal the nature of molecular recognition in the AChR.