Bivalent sequential binding model of a Bacillus thuringiensis toxin to gypsy moth aminopeptidase N receptor.

Specificity for target insects of Bacillus thuringiensis insecticidal Cry toxins is largely determined by toxin affinity for insect midgut receptors. The mode of binding for one such toxin-receptor complex was investigated by extensive toxin mutagenesis, followed by real-time receptor binding analysis using an optical biosensor (BIAcore). Wild-type Cry1Ac, a three-domain, lepidopteran-specific toxin, bound purified gypsy moth (Lymantria dispar) aminopeptidase N (APN) biphasically. Site 1 displayed fast association and dissociation kinetics, while site 2 possessed slower kinetics, yet tighter affinity. We empirically determined that two Cry1Ac surface regions are involved in in vivo toxicity and APN binding. Mutations within domain III affected binding rates to APN site 1, whereas mutations in domain II affected binding rates to APN site 2. Furthermore, domain III contact is completely inhibited in the presence of N-acetylgalactosamine, indicating loss of domain III binding eliminates all APN binding. Based upon these observations, the following model is proposed. A cavity in lectin-like domain III initiates docking through recognition of an N-acetylgalactosamine moiety on L. dispar APN. Following primary docking, a higher affinity domain II binding mechanism occurs, which is critical for insecticidal activity.

The binding kinetics of several Bt toxins with midgut receptors has been observed using optical biosensors (4 -8). Biosensors, such as the BIAcore (Biacore AB, Uppsala, Sweden), measure the affinity of a flowing molecule for another molecule immobilized on a surface as they form a real-time complex. As the molecule in solution is adsorbed by ligand, changes in mass on the surface are monitored using surface plasmon resonance (SPR) (9,10). Real-time kinetic analysis can be an important utility when macromolecular interactions deviate from simple, monophasic binding (11)(12)(13)(14)(15)(16)(17)(18). SPR studies conducted using Cry1A toxins specific for lepidopteran receptors have reported that Cry1Ac binds to two sites on purified Manduca sexta aminopeptidase N (APN) (4) and Helitothis virescens APN (6) with 2:1 toxin-receptor stoichiometries. Cry1Aa and Cry1Ab toxins (86% and 82% homologous to Cry1Ac, respectively (Ref. 19)) are also able to bind these insect APNs. However, purified Lymantria dispar APN binds Cry1Ac with a 1:1 stoichiometry (5) and does not bind Cry1Aa or Cry1Ab (20). This suggests structural differences among different insect APNs, despite sequence conservation (21).
The molecular mechanism of Cry1A toxins is best understood by examination of their three-domain structure (22,23). Domain I is involved in pore formation in the membrane, following binding (24 -30). Domain II has been shown to influence reversible binding and irreversible membrane insertion to insect brush border membrane vesicles (BBMVs) (31)(32)(33)(34)(35)(36)(37)(38)(39)(40)(41). Mutations in this domain have caused the greatest losses in toxicity. Domain III has several proposed functions. One of these roles is determining receptor specificity (8,(42)(43)(44)(45). For example, in Cry1Ac, lectin-like domain III recognizes an N-acetylgalactosamine (GalNAc) moiety on M. sexta APN (8,45). Based on homology modeling studies between Cry1Aa and Cry1Ac, it is proposed GalNAc docks in a surface cavity (46). Interestingly, this cavity is non-conserved in Cry1Aa and Cry1Ab, and the binding of these toxins to APN is not inhibited by preincubation with sugars (4,6). To the contrary, Cry1Ac binding to L. dispar APN-1 is almost completely inhibited by preincubation with GalNAc (5), indicating carbohydrate recognition is essential for this toxin-receptor interaction. This suggests Cry1Ac domain III has a sugar-dependent mechanism of binding (47) unique to Cry1A toxins. In fact, Cry1Ac domain III sequence is notably divergent from all other Cry toxins (48,49). The requirement of sugar recognition for binding, however, is common among many intestinal pathogens (50 -55).
In this study, the nature of Bt Cry1Ac binding to purified L. dispar APN was examined by comparing mutant toxin affinities on an optical biosensor. APN-binding epitopes were localized to specific residues in domain II and III. Our results suggest that Cry1Ac binds L. dispar APN in sequential steps, first by domain III, then domain II. Here we present an empir-* This work was supported by National Institutes of Health Grant R01 AI29092. 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.
Construction of Mutants-Site-directed mutagenesis of cry1Ac1 gene (pOS4201) subcloned into pBluescript KSϩ (pOS11200) was performed with a Bio-Rad Muta-Gene phagemid in vitro mutagenesis kit. Mutagenic primers were purchased from Biosynthesis or Genemed. Automated DNA sequencing with a United States Biochemical Corp. kit was performed according to manufacturer's instructions. The constructs of Cry1Ac-Cry1Aa domain-switched hybrids were described previously (59). Mutant constructs were expressed in Escherichia coli MV1190.
Bt Toxin Purification and Structural Analysis-Cry1Ac crystals were purified and solubilized, and protoxin was trypsin-activated as described previously (32). Toxins were purified by size exclusion on a Superdex-200 column (Amersham Pharmacia Biotech, Uppsala, Sweden) at 1 ml/min in either HBS (Hepes-buffered saline, 10 mM Hepes, pH 7.4, 150 mM NaCl, 3.4 mM EDTA) for BIAcore, or 20 mM phosphate buffer, pH 7.4, for circular dichroism (CD) spectral analysis. The purity of toxins was checked by 10% SDS-PAGE, and concentrations were determined by Coomassie protein assay reagent (Pierce).
To ensure toxin existed as a monomer in solution, dynamic light scattering for molecular size detection was employed using a DynaPro-801 (Proteins Solutions, Inc.). Samples were syringe-filtered (0.1 m). Bovine serum albumin was used as a control sample. The average hydrodynamic radius (R H ) of molecules purified from a Superdex 200 size-exclusion column was measured and fitted by bimodal analysis. At least eight measurements of each sample were taken. An average R H of 3.9 nm was found for over 99% of the monomeric fraction sample. This overlaps with the bovine serum albumin standard, which is of similar molecular size.
To detect changes in secondary structure, CD spectra of wild-type (wt) or mutant toxins in HBS or phosphate buffer were measured with an Aviv CD spectropolarimeter (model 62ADS) in a 32-Q-10 quartz cuvette. Readings were taken from 195 nm to 280 nm, five times, with 1000 sampling number. Data were analyzed with K2D (60). Additionally, CD spectra were examined for toxins S438A, G439A, F440A, W545A, QNR-AAA, and Cry1Ac wild type dialyzed into CAPS, pH 9.7, 150 mM NaCl, 3.4 mM EDTA to test the possibility of structural changes occurring at the pH of a lepidopteran gut environment.
Toxicity Bioassays-L. dispar eggs were supplied by the United States Department of Agriculture (Otis Methods Development Center, Beltsville, MD). LC 50 (50% lethal concentration) values were measured using the surface contamination method (36). Toxins were diluted in 50 mM sodium carbonate, pH 9.5, and applied to artificial diet in 24-well tissue culture plates. Two larvae were placed in each well, and LC 50 was recorded after 5 days. Bioassays were repeated three to five times. Mortalities were scored using SoftTOX 1.1 Probit analysis (Win-dowChem, Fairfield, CA).
Purification of L. dispar Aminopeptidase N (APN) Receptor-Fifth instar L. dispar larvae were dissected as described (32). BBMVs were prepared from midguts by the magnesium precipitation method (61) and resuspended at 1 mg/ml in 20 mM Tris, pH 7.4, 3.4 mM EDTA, with Complete protease inhibitor (Roche Molecular Biochemicals). A total of 32 mg of BBMVs were solubilized with 0.5% CHAPS overnight at 4°C. Insoluble materials were removed by centrifugation at 10,000 ϫ g for 10 min. Supernatant was concentrated by Amicon YM-30 filtration and applied to a MonoQ HR 10/30 ion-exchange column (Amersham Pharmacia Biotech) at 0.5 ml/min in 20 mM Tris, pH 7.4, 3.4 mM EDTA, 0.4 mg/ml CHAPS. Proteins were eluted in a step gradient of 1 M NaCl in the same buffer. Fractions were analyzed for aminopeptidase activity by a leucine-p-nitroanilide assay described previously (5). APN fractions were probed with biotinylated Cry1Ac and anti-L. dispar APN polyclonal serum using ligand blot overlays as described (5). Appropriate fractions were reconcentrated and re-purified by Superdex 200 in HBS, 2 times. A single peak at 280 nm corresponding to 120 kDa was collected (0.3 mg/ml). The purity of APN was checked by 10% SDS-PAGE and Coomassie-stained or transferred to PVDF for ligand blot overlays using biotinylated Cry1Ac and biotinylated soybean agglutinin (Pierce). Ligand blots were incubated in Extravidin-POD and developed in diaminobenzidene/urea (Sigma).
Biosensor Analysis of Mutant Toxin Affinities-BIAcore 2000, CM5 chip, N-ethyl-NЈ-(3-diethylaminopropyl)carbodiimide, N-hydroxysuc-cinimide, and ethanolamine-HCl (Biacore AB) were used for amine coupling of APN to the dextran surface of the CM5 chip. APN (100 -200 ng) was immobilized in 20 mM ammonium acetate pH 4.2 until 300 RU (300 pg/mm 2 ) were bound and a stable base line obtained. This low capacity RU surface was chosen since we have previously determined that mass transport effects did not occur on APN surfaces of 150, 300, and 800. 2 APN-2 was immobilized at similar levels on a control flow cell. APN-2 does not bind Cry1Ac and is separable from APN-1 during ion-exchange chromatography (5,21). For all procedures, HBS buffer, pH 7.4 (without surfactant) or CAPS-buffered saline, pH 9.7 were used at a flow rate of 30 l/min. Toxins were injected at multiple concentrations (10, 50, 100, 250, 500, and 1000 nM) in randomized orders. APN was regenerated between toxin injections with 10 mM NaOH, pH 11.0, 250 M ethylene glycol, in two pulses of 10 -30 l until the RU base line returned to its pre-injection level. At least two replicate experiments were performed for each toxin using different protein preparations. Carbohydrate inhibition studies with N-acetylgalactosamine, N-acetylglucosamine, and galactose were carried out as reported elsewhere (4) using 1000 nM toxins and varying sugar concentrations (0.5, 1, 2, 5, 10, 25, 50, 100, and 200 mM) on a 300-RU APN surface.
Kinetic Analysis of Sensorgrams-Response curves were prepared for fitting by subtraction of the signal generated simultaneously on the control flow cell. BIAcore sensorgram curves were evaluated in BIAevaluation 3.0 using numerical integration algorithms. The response curves of various analyte concentrations were globally fitted to several binding models issued with BIAevaluation 3.0. These include simple bimolecular binding (A ϩ B 7 AB), heterogeneous binding (A ϩ B1 7 AB1; A ϩ B2 7 AB2), conformational change (A ϩ B 7 AB 7 ABx), and bivalent analyte (AA ϩ B 7 AAB; AAB ϩ B 7 AABB). Curves were also fitted to a receptor dimerization model (A ϩ B1 7 AB1; AB1 ϩ B2 7 AB1B2), which differs from the conformational change model since [B2] is independent of [B1]. Apparent rate constants and affinities are presented from the conformational change model. Rate constants of the first step (k a1 , k d1 ) and second step (k a2 , k d2 ) are described by the following set of equations, where T ϭ toxin and R ϭ receptor.
Total response: TR ϩ TRx ϩ RI (residual bulk shift) (Eq. 5) The dependence of TRx complex formation on TR most reasonably reflects the toxin-receptor stoichiometry and step-dependent molecular mechanism resolved by our mutant kinetics. The global fittings had a standard error value of no more than Ϯ10% of the reported value. Since mutant rate constants are compared relative to wild type on the same surface, the use of different CM5 chips does not yield higher standard errors than the use of one chip (11).
Computer-simulated GalNAc Docking-GalNAc docking into the Cry1Ac domain III cavity was simulated using DockVision (University of Alberta, Alberta, Canada). DockVision's Research program utilizes a Monte Carlo simulated annealing, with scoring based on steric fitting and a pairwise energy function. Numerous conformations of GalNAc were generated and tested. The best fit was determined by the docking conformation with the lowest binding energy score. Electrostatic calculations by the Coulomb method and hydrogen bond calculations were made in Swiss-PdbViewer version 35b4.

RESULTS
B. thuringiensis Cry1Ac "Hot Spots"-Site-directed mutagenesis was performed in Cry1Ac domains II and III by substituting alanine for selected surface amino acids. Following size-exclusion column purification of toxins, the aggregation state of toxins in solution was examined by dynamic light scattering. Essentially all of the mass within purified fractions consisted of proteins with hydrodynamic radii corresponding to monomeric toxin.
Next, mutant toxins were purified and tested for biological activity against neonate L. dispar larvae. Alanine substitutions that did not alter toxicity are shown in Fig. 1 (yellow). We found two different regions affecting toxicity to L. dispar larvae that are separated by up to 62 Å (from Asn 377 to Trp 545 ). The first region includes residues surrounding a small depression in domain III (Fig. 1, blue), including Gln 509 , Arg 511 , and Tyr 513 . We have recently shown alanine substitutions at these residues greatly affected binding to L. dispar BBMVs with only minor reductions in toxicity (62). Another mutation in domain III, W545A, which is located on the upper lip of the cavity mouth, had the largest loss in toxicity for this region. The second region affecting toxicity was in domain II. This surface, larger than the one in domain III, was more critical for toxicity ( Fig. 1, red). Domain II, loop 3 mutations from residue 438 to 443 caused the largest reductions in toxicity. Additionally, alanine substitutions of arginine residues at positions 281, 289, 368, and 369 and also at Asn 377 caused reduced toxicity. Interestingly, I375A, located at the bottom of domain II, showed a slight increase in toxicity as compared with wild type, although their 95% confidence intervals were overlapping.
CD spectra were also analyzed for each mutant and compared with wild type to ensure differences in activity were not the result of structural changes. The mutant toxins had overlapping CD spectra to that of wild-type Cry1Ac at neutral pH, with the exception of mutations ⌬F371 (deletion), I373A, G546A, L583A, I586A, V587A, which were unstable during trypsin digestion at pH 10, and therefore not used. Additionally, the CD spectra of several mutants (see "Materials and Methods") were examined at an alkaline pH to better reflect the lepidopteran gut environment. The spectra of wild-type Cry1Ac shifted at higher pH values, as observed previously (63). The mutants that were tested shifted identically, indicating a wild-type transition at basic pH (Fig. 2). It is, therefore, unlikely that differences in wild-type and mutant properties are the result of structural problems arising at the "active" pH.
Determining a Kinetic Model-Next, we explored the hypothesis that mutant toxins with reduced insecticidal activity were affected in their ability to bind L. dispar aminopeptidase N receptor. First, APN was column-purified from detergent-solubilized L. dispar BBMVs by anion-exchange and size-exclusion chromatography until a single peak was obtained (Fig. 3A). SDS-PAGE (10%) analysis revealed a single band at approximately 120 kDa (Fig. 3B). To further test for purity, the APN preparation was probed by ligand blotting with biotinylated Cry1Ac toxin, as well as biotinylated soybean agglutinin (Fig.  3C). Both the toxin and the GalNAc-specific lectin appeared to bind to a single band at 120 kDa. These results provided evidence that there are no isoforms of APN present in the sample preparation that might contribute to the toxin binding response. To study the real-time binding kinetics of Cry1Ac to purified L. dispar APN, the receptor was immobilized on a biosensor chip surface and toxin-receptor complex formation was analyzed by surface plasmon resonance (SPR). Wild-type Cry1Ac binding curves were fit to various models (see "Materials and Methods"). The best-fitting models deviated from pseudo-first order kinetics. Cry1Ac response curves gave similar fits to two different binding models. The equation describing the first model involves two independent binding sites on the receptor (T ϩ R1 7 TR1; T ϩ R2 7 TR2). The equation for the second model describes a sequential binding, or "two-step" mechanism of binding, to a single receptor molecule (T ϩ R 7 TR 7 TRx). The equation for sequential binding is identical to the conformational change equation used in BIAevaluation 3.0 software, since two binding events occur on one receptor in both scenarios. When the wild-type Cry1Ac binding curves were fit to both the "two-site" and sequential binding ("two-step") models, the "goodness-of-fit" was similar, as indicated by 2 of Ͻ1.0. Further demonstrating a similar goodness-of-fit, a statistical Ftest comparing the fits of these two models indicated no significant preference. All other models had 2 Ͼ 1, indicating higher non-random deviation from the fitted curve.
Step 1 displays fast association and fast dissociation kinetics, while step 2 binding is slower, but enables adherence to APN.

Substitutions in the Binding Sites Alter
Step-specific Rate Constants-The response curves for mutant toxins were analyzed, and individual rate constants for each mutant were compared with wild type (Table I). For each mutant, rate constants for step 1 and for step 2 were determined. Changes in rate constants with respect to wild type are presented from curve fittings using the sequential binding model. In general, all mutants with decreased toxicity had a lower overall affinity for APN. The mutations can be categorized into two groups, those made in domain II and those made in domain III. This is illustrated in Table I, where boldface type denotes changes in binding greater than 2-fold. Domain II mutations affected step 2 rate constants of the sequential binding model. In contrast, domain III mutations affected step 1 rate constants.
Within domain II, we observed that mutations mostly caused decreases in k a2 (association) or increases in k d2 (dissociation). Replacing Arg at 281, 289, 368, and 369 by Ala or Glu tended to slow on-rates. Changes made in loop 3 were detrimental to k d2 , causing increased off-rates. These loop 3 mutants displayed the most dramatic losses in biological activity against L. dispar. However, mutating Ile at 375 to Ala decreased its off-rate. The delayed adherence time reflects its slightly augmented toxicity (Table I), emphasizing the biological importance of this kinetic step.
Domain III mutations more severely affected association and dissociation rates of step 1 when compared with domain II mutants. Analysis of W545A indicates a dependence of step 2 binding on step 1 binding. This is suggested since this domain III mutation completely disrupts toxin binding to the APN receptor. Thus, step 1 binding of domain III may be a prerequisite to step 2 binding.
For visual comparison, relative changes in apparent rate constants were plotted for each mutation (Fig. 5). Mutations   5, gray bars). For mutant toxins that altered binding more than 2-fold, the domain in which the mutation was made appears to correlate with which set of rate constants are affected.
Step-specific alteration of rate constants can also be observed by overlaying response curves from each mutant type (Fig. 6), as obtained from SPR experiments. Domain II mutations that alter step 2 binding only moderately affect total RUs bound during the association phase. After 240 s of association time, the "k d2 mutants" were the only type to have a similar amount bound as wild type. Alanine substitutions in loop 3, such as F440A (curve b, Fig. 6), fall off APN faster after toxin injection is replaced by buffer flow. Mutant R281A decreases k a2 (curve c, Fig. 6). Mutant R368A/R369A nearly eliminates step 2 binding (Table I), and approximately half of the binding signal (curve d, Fig. 6), but has no affect on step 1 binding. Conversely, when step 1 binding is decreased by Y513A (curve e, Fig. 6), total binding was diminished. W545A in the step 1 binding epitope eliminated binding to APN (curve f, Fig. 6). Other domain III mutants affecting step 1 also showed great reductions in APN binding.
The mutant toxins that most affected binding and toxicity were also compared with wt Cry1Ac using SPR at pH 9.7. We found that the affinities obtained for wild-type and mutant toxins were the same as those obtained at pH 7.4 (Fig. 7). Interestingly, the total binding response to APN of every toxin tested at this pH was lower than the total response observed at neutral pH, despite having the same affinity at both pH values. The observation of a reduced R max without reduced K D is similar to what has been found in binding studies using brush border membrane vesicles, where increasing pH from 7.4 to 10 reduced the binding site concentration for toxin (R t ) by more than half, while insignificantly affecting affinity (64). In this study, the reduction of total binding at higher pH values was correlated with decreased nonspecific binding. An alternative explanation for the reduced signal response is that increasing pH may induce repulsion from residual free carboxyl groups on the surface of the sensor chip. However, the behavior of wild type and these mutant toxins is conserved at varying pH values, and thus the reductions in toxicity and binding of these mutant toxins is not a reflection of unique damage inflicted in a basic environment.
In summary, the results of our mutant toxin analysis by SPR provided evidence that domain III appears necessary for recognition of APN since mutations in this domain can eliminate all binding. This agrees well with the findings of Lee et al. (42), who showed domain III determines receptor specificity in L. dispar for Cry1A toxins. Domain II binding is secondary, but also necessary. In combination with previous findings of a 1:1 stoichiometry of Cry1Ac binding to L. dispar APN (5), our results favor a two-step binding process of toxin to a single receptor molecule. Sugar-binding Domain III Is Initially Required for Any Binding-Since our Cry1Ac binding steps to L. dispar APN are domain modulated, we analyzed the binding response of domain-switch hybrids to APN after preincubation with N-acetylgalactosamine (GalNAc). One hybrid used, hybrid 4109, consisted of domains I and II of Cry1Aa and domain III of Cry1Ac (1Aa/1Aa/1Ac). Hybrid 4209 had a complementary construction (1Ac/1Ac/1Aa) (31). After a 30-min preincubation in increasing concentrations of GalNAc, toxins (500 nM) were injected over immobilized APN (Fig. 8). The IC 50 of GalNAc for Cry1Ac was 4.5 mM, which agrees with GalNAc inhibition constants previously observed (4,5,7). Hybrid 4109 possessed a slightly lower inhibition constant (IC 50 ϭ 1 mM). It was expected that hybrid 4209 would not bind APN since the domain III of Cry1Aa does not recognize APN (42). As expected, hybrid 4209 was unable to bind L. dispar APN at all. Mutant Q509A, which exhibited a FIG. 6. Component curves of Cry1Ac are altered by mutagenesis of the binding epitopes. Sensorgram overlay of wild-type Cry1Ac with mutant toxins affecting different binding steps: a, Cry1Ac wild type; b, F440A, a k d2 mutant; c, R281A, a k a2 mutant; d, R368A/ R369A, a k a2 /k d2 mutant; e, Y513A, a k a1 mutant; f, W545A, a step 1 "knock out" mutant.

FIG. 5. Relative changes in individual rate constants of mutant toxins.
Relative changes in association rates are k a(mut) /k a(wt) , and relative changes in dissociation rates are k d(mut) /k d(wt) . Mutations are identified by residue number of Cry1Ac and are assumed to be alanine substitutions except 368/369 where noted. Mutations occurring in domain II or domain III are divided by vertical dashed line. Relative rates Ͻ1.0 indicate slower association or faster dissociation compared with wild type APN-binding. Mutants with greater than 2-fold changes in relative rates are colored gray.
reduced k a1 , was also tested for GalNAc inhibition. Although total binding of Q509A to APN after 4 min of association is only 12% of the total binding of wild type, it is still completely inhibited by increasing GalNAc concentrations, affirming step 2 binding relies on step 1 (Fig. 8, open squares).
Finally, we tested galactose and epimeric GlcNAc for inhibition of binding. Talose and gulose (epimers of galactose at the C2 and C3 positions, respectively) were also tested (data not shown). None of these sugars was able to inhibit Cry1Ac binding to APN. Our results confirm previous reports, which found sugar binding requires a galactoside orientation at the C-4 hydroxyl (4, 7), and also suggest that Cry1Ac prefers an acetamido group at the C-2 position.
Computer-simulated Docking-The results of our GalNAc competition studies prompted us to test docking of GalNAc to domain III by computational methods as well. Electrostatic calculations of Cry1Ac indicate a strong, positive field around domain II loops, and near the domain III GalNAc-binding epitope, while the domain I ␣-helical bundle is negatively charged (Fig. 9). Given the alkaline environment of the insect gut (65), the orientation of positive surface charges on this and other Bt toxins may serve to direct toxin receptor-binding epitopes toward the negatively charged surface of the insect brush border membrane. Using a computer-simulated docking method, we observed that GalNAc binding in the putative cavity of domain III was favorable in steric and energy-pairing analyses. Multiple docking positions were possible for GalNAc. The least potential energy obtained for binding was Ϫ24 kcal/ mol (Fig. 9). Potential hydrogen bonds between the GalNAc-Cry1Ac complex were calculated for various rotamers of Gal-NAc at the lowest energy positioning. Potential hydrogen bonding occurred between the GalNAc acetamido group and Cry1Ac residues Gln 509 , Asn 544 , and the backbone of Asn 547 . These interactions may account for the specificity of Cry1Ac for GalNAc, but not galactose. Additionally, the backbones of  Asn 544 and Gly 546 interact with C-3 and C-4 hydroxyls, indicating a preference for a galactoside orientation. Thus, the geometry around the C-4 hydroxyl may account for Cry1Ac's preference for GalNAc, but not GluNAc (a C-4 epimer). Our results also indicated the C6 hydroxyl might serve as a hydrogen bond donor to protein atoms at the Gly 512 backbone and Arg 511 . Finally, Trp 545 or Tyr 513 may contribute by stacking against the sugar ring, as observed of aromatic residues in other GalNAc-binding pockets (66). Conversely, in docking simulations with Cry1Aa and GalNAc, no favorable annealings were found. This agrees with previous reports that Cry1Aa binding to M. sexta APN is not inhibited by GalNAc (4).

DISCUSSION
Toxicity and receptor binding studies with our Cry1Ac mutants provided evidence of functional epitopes on Cry1Ac domains II and III. Each domain contributed its own set of rate constants during binding. Biosensor analysis of domain II and III mutant toxins indicated domain III binds and releases quickly (k a1 and k d1 ), while domain II binds slower and tighter (k a2 and k d2 ). Loss of domain III binding by mutagenesis or domain-exchanges eliminates APN binding, demonstrating the dependence of secondary domain II binding on initial domain III binding. Domain III binding is specifically inhibited by GalNAc. Given these results and the 1:1 binding stoichiometry of Cry1Ac to APN, we propose a two-step, or sequential binding, model rather than a two-site model. In this model, APN recognition is determined by Cry1Ac domain III binding through a GalNAc moiety, followed by contact of domain II loop residues (Fig. 10).
Previously, we have observed that, under our experimental conditions, the effects of mass transport and analyte rebinding do not contribute to Cry1Ac's deviation from simple bimolecular binding. 2 We note that there are some possible alternative explanations of our findings. First, a heterogeneous surface of APN molecules may result from amine coupling. However, it is unlikely that our two sets of rate constants are induced by constrained receptor positions since mutations in separate domains of Cry1Ac affect both toxicity and APN binding. Two separate structural and functional regions make monophasic binding improbable, and thus, amine coupling is an unlikely source of heterogeneity. Second, the two Cry1Ac domains could bind independently to APN. Such bivalency is also seen in antibody-antigen interactions, but separate antibody fragments are monovalent (12). Cry1Ac is different from an antibody since the second binding site depends on the first one. In addition, the bivalent analyte model did not fit our biosensor curves as well as the sequential binding model (data not shown). A third alternative is that APN receptors dimerize on the biosensor surface in a sequential fashion, like the human growth hormone interaction with its receptor hGHbp (67). This could occur by domain III binding to one APN and domain II binding to a second APN. Curve fittings with a receptor dimerization model (see "Materials and Methods") did not fit better than the sequential binding model (data not shown), and our predicted 1:1 stoichiometry does not suggest dimerization occurs. Additionally, our apparent kinetic rate constants do not vary with different densities of receptor immobilized, indicating ligand interactions do not affect complex formation. 2 One final possibility is that a conformational change actually occurs upon APN binding. It has been suggested that Cry1Ac does undergo a conformational change upon binding to M. sexta APN in a lipid monolayer (7). This would initiate membrane insertion of domain I ␣-helices involved in pore formation. It remains to be determined if binding to a functional toxin receptor, alone, can trigger this event.
Pathogenic bacterial toxins that target cell membranes appear to possess a similar functional construction. It has been observed in well characterized toxins, such as cholera and shigella, that a "B" domain functions in binding to cell surface receptors, while an "A," or activity, domain exerts the toxin's specific biological activity (68,52). A and B domains may be synthesized together or separately. It is further postulated that regions of hydrophobicity on one of the domains or on a separate domain called "E" (entry domain), plays a role in facilitating insertion of the toxin after receptor binding (68). The ABE model of toxin structure may be analogous to the domains of Cry1Ac. Our data suggest that L. dispar APN specificity is determined by sugar binding of domain III. Mutations around the domain III cavity affect initial binding rates. This lectinlike, jellyroll structure acts like a B domain. Domain II mutations affect rate constants of the subsequent step. It is tempting to speculate secondary domain II binding is critical for facilitating entry into the membrane, acting as an E domain. It is not known if this would involve binding to a second receptor site or initiating a conformational change. However, the loss of exposed, hydrophobic Phe 440 caused the most dramatic effects on k d2 , fitting with its potential role as an E domain. Interestingly, changing charged arginine residues to alanine resulted in slower on-rates during the second step. Positively charged arginines might serve to orient hydrophobic loops to an APN binding site or toward the membrane surface. Finally, the ability of domain I ␣-helices to form membrane pores suggests it may be an A domain.
A comparison of mutant toxicity and APN binding affinity in this study yields a strikingly contradictory result. Complete loss of APN binding caused by domain III mutation W545A only results in 50-fold decreased activity. Cry1Ac appears to retain slight toxicity to L. dispar without APN or GalNAc binding capability. On the other hand, domain II loop 3 mutations cause greater decreases in toxicity (Ͼ600-fold) than can be accounted for by loss of APN binding. One explanation for this contradiction is that during toxicity bioassays (5-day time period), the domain III mutations that weakened APN recognition in SPR studies (4-min association time) are less critical than domain II mutations, which affect adherence to APN. If insertion into the membrane depends on step 2 adherence to the APN, domain II mutations may affect toxicity more than domain III mutations over time. A second possibility is that other receptors in vivo may compensate for loss of domain III binding. Like APN, these receptor sites may be affected by domain II mutations, resulting in greater toxicity losses for mutations such as F440A in domain II than for W545A in domain III. The possibility of other Cry1Ac receptors agrees with a previous study that showed APN competes for Cry1Ac binding to L. dispar BBMV, but does not eliminate all binding (69). Additionally, Cry1Ac inhibition of short circuit current in L. dispar midgut was reduced by APN cleavage, but some inhibition still remained (69), suggesting other receptors were involved. In another insect, M. sexta, Carroll et al. (70) found evidence that a GalNAc-independent mechanism of Cry1Ac BBMV permeabilizing activity occurs. Complete loss of APN binding in this insect caused only slightly decreased Cry1Ac toxicity (45,46,62). Another Cry1A-binding receptor, a cadherin-like glycoprotein called BT-R 1 , has previously been identified in M. sexta (71), which may compensate for loss of APN binding (46). Such a receptor in L. dispar could contribute to the residual toxicity of our domain III mutants that lost APN binding. For example, a glycosylated Cry1Aa/b toxin receptor was recently identified in L. dispar, with relatively low affinity for Cry1Ac (72). Taken together, these results implicate APN is a major receptor for Cry1Ac in L. dispar, but not necessarily the only receptor. A final possibility for the toxicity contradiction is that a domain II mechanism is critical for all binding events at the brush border membrane. It is conceivable domain II is involved in insertion or initiates a conformational change necessary for insertion. This is consistent with the idea of domain II as an E domain, facilitating entry. Liang et al. (34) observed that irreversible binding of Cry1A toxins to L. dispar BBMVs was more directly related to toxicity than initial binding. Domain II has previously been implicated in both initial and irreversible binding of Cry1A toxins (35)(36)(37). For this reason, mutating amino acids in domain II is potentially more detrimental than altering receptor-specifying residues in domain III.
Understanding the division of Cry1Ac domain III as a receptor-recognition domain and the role of domain II in facilitating binding and toxicity helps direct continuing efforts in engineering insect specificity or improved toxicity of Bt toxins.