Solution Structure of the Major α-Amylase Inhibitor of the Crop Plant Amaranth*

α-Amylase inhibitor (AAI), a 32-residue miniprotein from the Mexican crop plant amaranth (Amaranthus hypochondriacus), is the smallest known α-amylase inhibitor and is specific for insect α-amylases (Chagolla-Lopez, A., Blanco-Labra, A., Patthy, A., Sanchez, R., and Pongor, S. (1994) J. Biol. Chem. 269, 23675–23680). Its disulfide topology was confirmed by Edman degradation, and its three-dimensional solution structure was determined by two-dimensional 1H NMR spectroscopy at 500 MHz. Structural constraints (consisting of 348 nuclear Overhauser effect interproton distances, 8 backbone dihedral constraints, and 9 disulfide distance constraints) were used as an input to the X-PLOR program for simulated annealing and energy minimization calculations. The final set of 10 structures had a mean pairwise root mean square deviation of 0.32 Å for the backbone atoms and 1.04 Å for all heavy atoms. The structure of AAI consists of a short triple-stranded β-sheet stabilized by three disulfide bonds, forming a typical knottin or inhibitor cystine knot fold found in miniproteins, which binds various macromolecular ligands. When the first intercystine segment of AAI (sequence IPKWNR) was inserted into a homologous position of the spider toxin Huwentoxin I, the resulting chimera showed a significant inhibitory activity, suggesting that this segment takes part in enzyme binding.

Plant seeds produce a large variety of enzyme inhibitors that are thought to provide protection against insects and microbial pathogens. As plant seed inhibitors are often species specific, i.e. they inhibit enzymes of a well defined group of pathogenic organisms but do not affect the mammalian counterpart, they make attractive candidates for conferring pest resistance to transgenic plants (for a review see Ref. 1).
The ␣-amylase inhibitors vary considerably in their struc-tures. Many of their structural relatives, e.g. proteinase inhibitors, osmotin, and salt-induced proteins (Table I), play roles in plant stress response. The smallest of the known ␣-amylase inhibitors, AAI, 1 is found in the seeds of Amaranthus hypochondriacus, a variety of the Mexican crop plant amaranth or Prince's feather (2). AAI is a 32-residue polypeptide with three disulfide bridges, which has no significant sequence similarity to other proteins in the data bases. It has a spurious sequence similarity to various members of the so-called knottin (3) or "inhibitor-type cystine knot" (4) family, which includes various proteinase inhibitors and toxins. AAI is species specific; it inhibits ␣-amylase of several pathogenic insect larvae (Tribolium castaneum, Prostaphanus truncatus, Periplaneta americana, and Tenebrio mollitor) but not the mammalian ␣-amylases.
Here we report the three-dimensional structure of AAI as determined by NMR spectroscopy and show via amino acid replacement and chimera construction that a short segment of the first loop of AAI is involved in enzyme inhibition. 2

EXPERIMENTAL PROCEDURES
Materials AAI was prepared as described (5). Sephadex G-75 and DEAE-Sepharose CL6B were obtained from Amersham Pharmacia Biotech. ␣-Chymotrypsin, endoproteinase Glu-C, and trypsin were obtained from Sigma; cyanogen bromide and vinyl pyridine were obtained from Aldrich. All chemicals used were of analytical or sequencing grade. HPLC grade acetonitrile and trifluoroacetic acid were obtained from Aldrich.

Peptide Mapping and N-terminal Sequencing
A 75-g sample of AAI was dissolved in 150 l of 0.2 M Tris-HCl buffer, pH 7.3, and digested with a mixture of trypsin (3 g), chymotrypsin (3 g), and endoproteinase Glu-C (3 g) at 37°C for 15 h. The mixture was separated by reverse phase-HPLC on a Vydac C 18 column (2.1 ϫ 250 mm). Selected peaks were collected, lyophilized, and covalently coupled to aminophenyl glass beads in prepacked capillary * The work at Peking University was supported by a grant from the Collaborative Research Program of the International Center of Genetic Engineering and Biotechnology (Trieste, Italy) and grants from the National Science Foundation of China and the National High Technology Program of China. 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.
The  1 The abbreviations used are: AAI, ␣-amylase inhibitor; CB, cellulosebinding domain; COSY, correlation spectroscopy; TOCSY, total correlation spectroscopy; HPLC, high pressure liquid chromatography; FMOC, 9-fluorenylmethyoxycarbonyl; DQF-COSY, double quantum filtered correlation spectroscopy; NOESY, nuclear Overhauser effect spectroscopy. 2 NMR spectroscopy was determined in the lab of Dr. X. Gu (guxc@lsc.pku.edu.cn). The disulfide structure and the chimera construction were carried out in the lab of Dr. S. Liang (liangsp@public.cs.hn.cn). AAI and its analogs were synthesized in the lab of Dr. S. Pongor (pongor@icgeb.trieste.it). columns for N-terminal sequencing carried out on a MilliGen/Biosearch model 6600 ProSequencer. The released amino acid phenylthiohydantoins were detected simultaneously at 269 and 313 nm.

Peptide Synthesis and Refolding Studies
AAI mutants (Table II) were synthesized manually by solid phase peptide synthesis methods based on FMOC chemistry and oxidative refolding as described (5). The Huwentoxin I chimera (Table II) was synthesized by FMOC chemistry on a PIONEER peptide synthesizer (PE Biosystems) using FMOC-L-Leu-polyethylene glycolpolystyrene resin as described (8). The oxidative refolding of the chimera was carried out in the presence of 0.1 mM oxidized glutathione and 1 mM reduced glutathione in 0.1 M acetate buffer, pH 9.5, containing 0.2 mM EDTA. The final product was purified by reverse phase-HPLC.
The qualitative ability of the peptides to fold was tested as follows. A stock solution of 1.0 mg of totally reduced and HPLC-purified peptide sample dissolved in 1 ml of freshly prepared argon-saturated refolding buffer (0.1 M NH 4 Ac, 2 mM EDTA, 1 M guanidine hydrochloride, pH 7.8) was added to 49 ml of the same buffer to which 1 mM cysteine and 0.05 mM cystine were added immediately before use (the final concentration of the peptide was 20 g/ml). After 16 h of stirring at 25°C the reaction was stopped by adjusting the pH to 4.0 with acetic acid. The reaction mixture was analyzed by reverse phase-HPLC as described (5).The folding ability was evaluated from the chromatogram. Typical chro-matograms of "well folding" (ϩ) and "poorly folding" (Ϫ) samples are shown in Fig. 1,A and B, respectively. The peak corresponding to the folded product, indicated by the arrow, was collected and lyophilized for amylase inhibition assay.

NMR Studies
Sample Preparation-Samples were prepared by dissolving AAI powder in 0.5 ml of 20 mM phosphate or acetate buffer (90% H 2 O, 10% D 2 O) containing 0.02% NaN 3 and 0.1 mM EDTA with the final concentration of AAI being 2-4 mM. The pH of the solution was adjusted to 6.5 with 1N HCl and NaOH. Sodium 3-(trimethylsilyl)-2,2,3,3,-tetradeuteropropionate was added as an internal reference for chemical shifts to a final concentration of 200 M. For the experiments in D 2 O, AAI samples dissolved in H 2 O were lyophilized and redissolved in 99.8% D 2 O (Cambridge Isotope Laboratories). The solution was then allowed to stand at room temperature for 24 h before re-lyophilization and a second reconstitution with 99.996% D 2 O.
NMR Spectroscopy-All of the two-dimensional spectra, including COSY, DQF-COSY, TOCSY, and NOESY, were collected at 500 MHz using a Bruker 500 AMX spectrometer at pH 6.5, 300 K. The data points were 512 in the F1 dimension and 2048 in the F2 dimension. Most two-dimensional spectra were recorded by the time proportional phase incrementation method (9). Solvent suppression was carried out by the presaturation method.  Table III  ϩ a Peptides exhibiting Ͼ1% inhibition at 15 M concentration and 100:1 substrate/enzyme ratio were scored as "ϩ." b The folded products ( Fig. 1) were purified by reverse phase-HPLC, and their quantity was estimated based on HPLC peak height (taking AAI as 100%). Cases where the quantity of the folded product reached 5% were scored as "ϩ." c U, norleucine.  Richardson (33) and completed with recent data (2). ND, no data. Data Processing-SYBYL software (Tripos Inc.) was used on an Indigo 2 Silicon Graphics workstation. All the data were zero-filled to 1000 in the F1 dimension, resulting in a 2048 ϫ 1024 (F2 ϫ F1) real matrix. The sequential assignment method of Wü thrich (10) allowed identification of all backbone and side chain protons with the exception of a few side chain protons of Ile 2 , Lys 4 , Arg 7 , Lys 11 , and Glu 19 . A total of 348 distance constraints were derived from the NOESY spectra (100 and 400 ms in H 2 O and D 2 O). 87 long distance constraints, 39 medium distance constraints, 101 sequential constraints, and 121 intraresidue constraints were found. Three distance constraints (S(i) Ϫ S(j) (2.02 Ϯ 0.02 Å), S(i) Ϫ C ␤ (j) (2.99 Ϯ 0.5 Å), and S(j) Ϫ C ␤ (i) (2.99 Ϯ 0.5 Å)) were used for each disulfide, and 8 dihedral constraints were calculated from one-dimensional spectrum and DQF-COSY. These constraints were used for calculating the structure with Brü nger's X-PLOR, version 3.851 (11). The nuclear Overhauser effect spectroscopy constraints were divided into three classes (strong, medium, and weak), corresponding to the ranges of 1.8 -2.8 Å, 1.8 -3.5 Å, and 1.8 -5.0 Å, respectively.

RESULTS
Structural Studies-Previous to the current study, the disulfide topology of AAI was only partially known, i.e. the connectivities of two vicinal Cys residues (17 and 18) were inferred from homology modeling rather than from experimental data. Because this is crucial to the NMR studies, we chose the Edman degradation method combined with phenylthiohydantoin analysis to confirm this disulfide topology at 313 nm. This combined method makes it possible to identify dehydroalanine, the ␤-elimination product of phenylthiohydantoin-cystine, which forms when the N-terminal sequencing process reaches a Cys residue disulfide bonded to a sequentially upstream Cys  (Table II) under the same conditions. Column, Supersphere 100 C 18 4 mm 4 ϫ 125 column (Merck GmbH); buffer A, 0.1% trifluoroacetic acid in water; buffer B, 0.1% trifluoroacetic acid in acetonitrile; linear gradient from 15 to 40% buffer B in 25 min.

FIG. 2. Identification of dehydroalanine (indicated as ⌬Ser)
during Edman sequencing by monitoring the HPLC effluent at 313 nm. Appearance of dehydroalanine in position 18 indicates only that this residue is disulfide-bonded to a sequentially upstream cystein, which, by way of elimination (see "Results"), is Cys 1 .   FIG. 3. The C ␣ trace for the 10 best AAI structures obtained by NMR spectroscopy. The C ␣ traces of the 10 best structures (blue) were superimposed, and the root mean square deviations were calculated using the molecular simulation program InsightII. The average root mean square distance was 0.32 Å for the backbone heavy atoms (N, C ␣ , C, and O) and 1.07 Å for all heavy atoms. N and C termini are indicated. The C ␣ trace of the final structure is shown in red.

FIG. 4.
Ribbon diagram of the three-dimensional structure of AAI. The best structure chosen from the results of PROCHECK (15) calculation is presented as a ribbon diagram produced with the program SETOR (44). The three ␤-strands, which form an antiparallel ␤-sheet are highlighted by blue arrows, and the three disulfide bridges are indicated in yellow. N and C termini are labeled. residue (12). Fig. 2 clearly shows that the degradation product appears in position 18, confirming that AAI in fact has the disulfide topology Cys 1 -Cys 18 , Cys 8 -Cys 23 , Cys 17 -Cys 31 , a topological pattern characteristic of the knottin family. 50 structure calculations were carried out using the X-PLOR program of Brü nger (11). The 10 best structures, based on their stereochemical energies (excluding the electrostatic term) and PROCHECK (15) results, were chosen for structure analysis. These structures (Fig. 3) had no nuclear Overhauser enhancement spectroscopy violations larger than 0.3 Å and no dihedral violations larger than 2°with 84.3% of all residues (excluding Gly and Pro) in the most favored regions and 15.7% in the additional allowed regions. A root mean square distance of 0.32 Ϯ 0.05 Å was found for backbone atoms, whereas that for all heavy atoms was 1.04 Ϯ 0.12 Å. Both values are satisfactorily low for a peptide of this size. These ten structures have been deposited in the Protein Structure Data Bank under the identification code "1qfd".
Using the protein structural element criteria described by Wü thrich (10), Richardson (16), and Wilmot and Thornton (17), we identified three antiparallel ␤-strands and five ␤-turns in the three-dimensional structure of AAI (Fig. 4). The three short ␤-strands include residues Arg 7 -Cys 8 , Thr 22 -Thr 24 , and Gly 29 -Ser 32 . The first strand is very short, a feature often prevalent in the knottin fold (4,18). Of the five ␤-turns, Pro 3 -Asn 6 belongs to type II, and Gly 9 -Met 12 and Pro 10 -Asp 13 belong to type I, whereas Glu 19 -Thr 22 and Asp 26 -Gly 29 belong to type VIII. Of the three disulfide bridges in this structure, Cys 1 -Cys 18 is a right-handed spiral, whereas Cys 8 -Cys 23 and Cys 17 -Cys 31 are left-handed.
Structure Comparison and Residue Replacement-The motif consists of three antiparallel ␤-strands accompanied by special disulfide bonding patterns and corresponds exactly to the socalled knottin (3) or inhibitor cystine knot fold (4). We therefore made a structural alignment for AAI with a set of known three-dimensional structures from this family (Fig. 5). In this figure we divided the sequences into segments between cysteine residues. To identify functionally and structurally important residues, we prepared a series of peptides carrying replacements or deletions in segment 1 or segment 2 and monitored, in a qualitative way, both their in vitro refolding ability (see "Experimental Procedures") as well as their ␣-amylase inhibitory activity (Table II). We found that although replacements in segment 2 impaired the folding ability of AAI, they did not abolish ␣-amylase activity. In contrast, replacements in segment 1 abolished ␣-amylase inhibition without drastic consequences for folding. To test whether segment 1 of AAI is directly involved in enzyme binding, we grafted it into Huwentoxin I, a protein, which although structurally similar to AAI, is not an ␣-amylase inhibitor. The resulting chimera (sequence shown in Table II) folded well (data not shown) and had a low but clearly significant ␣-amylase inhibitory activity (Table III). The inhibitory activity was 10 Ϫ4 times that of AAI but very much higher than that of Huwentoxin I (inhibition was not detectable at the concentrations achievable with this peptide); because a successful scorpion toxin/charybdotoxin chimera had a bioactivity 10 Ϫ5 that of the original charybdotoxin molecule (19), we infer that the inhibitory activity of our Huwentoxin I/AAI chimera construct is satisfactory proof that the segment 1 sequence is a part of the enzyme binding region of AAI.

DISCUSSION
The structure of AAI corresponds to the so-called knottin (3) or inhibitor cystine knot fold (4), which is characterized by three antiparallel ␤-strands and a disulfide topology of the "abcabc" type, as schematically shown in Fig. 5. The first of the  three ␤-strands is usually distorted, and it is missing in some members of this family so the core of this motif is often referred to as a short distorted triple-stranded ␤-sheet. In the solution structure of AAI the first ␤-strand is in fact less clearly defined than the other two. Members of this family do not show high sequence similarity, and, partly because of their small size, they cannot be easily identified by sequence similarity searches. We previously built a model of AAI based on its homology with the Trichoderma reesei CB and squash proteinase inhibitor (2). Although the global features of AAI were correctly captured by this model, the atomic coordinates were only in moderate agreement with the experimental data. The knottin family contains several distinct subtypes, according to structure-based sequence alignment (Fig. 5). AAI, along with many of the fungal type CBs and some spider toxins, lacks segment 3, so these two central cysteines are vicinal. The CB of T. reesei cellobiohydrolase lacks the disulfide bridge a, whereas -conotoxin GVIA has an additional d. Huwentoxin I and other spider toxins form a distinct subgroup characterized by a very short third ␤-strand. AAI most closely resembles the fungal CB group.
The knottin (3) or inhibitor cystine knot fold (4) was found in various proteins from fungi, plants, spiders, and cone shells. The fact that these proteins fulfill a large variety of biological functions leads us to suggest that this fold may have emerged by convergent evolution. Thus the weak sequence similarities among various knottin proteins may be because of common structural determinants rather than a common evolutionary origin. In fact we have found that certain residues can be replaced without impairing the folding of the molecule, whereas other replacements seem to interfere with the folding. For example we found that segment 1 of AAI can accommodate residue replacements, which is in very good agreement with the finding that the T. reesei CB readily accepts mutations in the homologous region (20). It is interesting, in this respect, that formation of this simple fold is apparently not a one-step process. A two-step mechanism that includes the reshuffling of the first disulfide intermediates was suggested for AAI (5) and also for the related potato proteinase inhibitor (21). It is tempting to speculate that this suggested two-step mechanism may be in fact a condition for the formation of the knottin-structure; quantitative folding studies are now underway.
Members of the knottin family bind to various macromolecular ligands as diverse as cellular receptors, enzymes, or cellulose (for a recent overview see Ref. 20). In the structural alignment in Fig. 5 green displays the residues that are thought to be involved in binding. According to the distribution of these binding residues, knottins can be divided into three broad categories. The proteinase inhibitors from Cucurbita maxima (22) and Ecballium elaterium (18) bind through segment 1. The group, including the CB (23-25), charybdotoxin (26), and Huwentoxin I, 3 contains additional binding residues in segment 5. The third group, -conotoxin GVIA and -agatoxin interact via residues in segment 2 as well as segment 1. Our residue replacement results indicate that segment 1 of AAI seems to be involved in amylase binding but presumably is not solely responsible for it; however, the inhibition level of the chimera is low. Because our residue replacement studies failed to reveal binding residues in segment 2, which lead us to conclude that the binding mode of AAI may be different from that of -conotoxin GVIA and -agatoxin. So by elimination we are left with the conclusion that the binding mode of AAI may be similar to the T. reesei CB, i.e. the molecule may bind via segments 1 and 5. Conclusive data on the binding mode of AAI await analysis of the three-dimensional structure of an AAI/␣-amylase complex.
Our AAI mutagenesis studies confirm that the knottin structure can accept a wide range of mutations. This is important for the engineering of small binding proteins with altered biological activities (20,27). Finally, we note that the species-specific biological activity of AAI makes it an attractive target both for the development of insect-resistant transgenic plants (1), as well as in engineering studies (27).