Catalysis of decarboxylation by a preorganized heterogeneous microenvironment: crystal structures of abzyme 21D81

doi:10.1006/jmbi.2000.4503 available online at http://www.idealibrary.com on J. Mol. Biol. (2000) 302, 1213±1225

Catalysis of Decarboxylation by a PreorganizedHeterogeneous Microenvironment: Crystal Structuresof Abzyme 21D8

Kinya Hotta1,3, Holger Lange2, Dean J. Tantillo2, K. N. Houk2

Donald Hilvert3 and Ian A. Wilson1

1Department of MolecularBiology and Skaggs Institutefor Chemical Biology, TheScripps Research Institute10550 North Torrey PinesRoad, La Jolla, CA 92037, USA

Abbreviations used: Ig, immunogcomplementarity-determining regiomean-square deviation(s); MPEG, pmonomethyl ether.

0022-2836/00/051213±13 $35.00/0

Antibody 21D8 catalyzes the solvent-sensitive decarboxylation of 3-car-boxybenzisoxazoles. The crystal structure of chimeric Fab 21D8 with andwithout hapten at 1.61 AÊ and 2.10 AÊ , respectively, together with compu-tational analysis, shows how a melange of polar and non-polar sites areexploited to achieve both substrate binding and acceleration of a reactionnormally facilitated by purely aprotic dipolar media. The striking simi-larity of the decarboxylase and a series of unrelated esterase antibodiesalso highlights the chemical versatility of structurally conserved anionbinding sites and the relatively subtle changes involved in ®ne-tuning theimmunoglobulin pocket for recognition of different ligands and catalysisof different reactions.

# 2000 Academic Press

Keywords: medium effects; antibody catalysis; computational docking;immune response; enzyme evolution

Introduction

Changes in solvent can dramatically alter therates of chemical reactions. For instance, simplesubstitution and elimination reactions have beenaccelerated by factors in excess of 106 simply bytransfer from water to organic solvents (Reichardt,1988). Analogous medium effects are believed tocontribute to the high ef®ciency of many enzymes(Jencks, 1975). In general, however, it is dif®cult toseparate such effects from other contributionstoward the overall rate acceleration in these highlyevolved, mechanistically complex systems. Anti-body catalysis of archetypal model reactions pro-vides an opportunity to examine this issue.

In one example, antibodies raised against the1,5-naphthalene disulfonate derivative 1a (Figure 1)have been shown to promote the highly solvent-

lobulin; CDR(s),n(s); r.m.s.d.(s), root-olyethylene glycol

sensitive decarboxylation of 3-carboxybenzisoxa-zoles (2! 4; Lewis et al., 1991). This well-studiedunimolecular reaction occurs via a charge deloca-lized transition state 3 (Kemp & Paul, 1975).Although slow in aqueous buffer (t1/2 � 12.9 hoursat 20 �C), it is accelerated by up to eight orders ofmagnitude in aprotic dipolar solvents (Kemp &Paul, 1975). Protic solvents are believed to inhibitthe reaction by stabilizing the ground state carbox-ylate group through hydrogen bonding, whileaprotic dipolar solvents may promote the reactionby directly stabilizing the polarizable transitionstate through dispersion interactions (Kemp et al.,1975).

Hapten 1 was designed to elicit a pocket ofappropriate size and charge for binding 2 and sta-bilizing transition state 3 (Lewis et al., 1991). Thesulfonate groups were included to elicit cationiccounterions for the anionic species along the reac-tion coordinate, while the naphthalene ring wasexpected to induce a proximal hydrophobic regionfor binding the aromatic portions of substrate andtransition state. In accord with this design, anti-1

# 2000 Academic Press

Figure 1. Decarboxylation of 5-nitro-3-carboxybenzi-soxazole 2 proceeds through a charge-delocalized tran-sition state 3 to give salicylonitrile 4. Naphthalenedisulfonate 1 was used to elicit the decarboxylase anti-body 21D8 through a covalent linkage to keyhole limpethemocyanin (KLH).

1214 Structures of a Decarboxylase Antibody

antibodies accelerate the decarboxylation of 2 by asmuch as 23,000-fold over the spontaneous reactionin water (Tarasow et al., 1994). Catalytic activitywas found to correlate roughly with the hydropho-bicity of the antibody active site as measured byenvironmentally sensitive ¯uorescent probes(Tarasow et al., 1994). This observation, togetherwith relatively minor variations in the carbon kin-etic isotope effect (Lewis et al., 1993), the con-strained geometry of the reacting bonds (Zipseet al., 1995), and the known insensitivity of thereaction to acid-base catalysis (Kemp & Paul,1975), suggests that antibody catalysis and solventacceleration may share common underlying mech-anisms.

Nevertheless, protein binding sites are highlyordered and decidedly inhomogeneous comparedwith the ¯uxional, but relatively uniform, environ-ment of a dipolar aprotic solvent. To elucidate theprecise combination of hydrogen-bonding, electro-static, and dispersive interactions that lead toenhanced rates of decarboxylation, detailed struc-tural information is essential. Here, we present thecrystal structure of 21D8, one of the most ef®cientantibody decarboxylases, with and without boundhapten. In conjunction with computational dockingexperiments, the structure provides valuableinsight into the use of medium effects as a generalcatalytic strategy.

Results and Discussion

Structure of 21D8

Antibody, 21D8 was cloned and expressed as achimeric murine-human Fab fragment. Fusion ofthe VL and VH variable regions of 21D8 with theCL and CH1 regions of a human k1 and IgG1 anti-body, respectively, enables relatively ef®cient peri-plasmic expression of the protein in Escherichia coli(Ulrich et al., 1995). The chimeric antibody bindshapten 1b with a Kd value of 110 nM and acceler-

ates the decarboxylation of 5-nitro-3-carboxybenzi-soxazole with a catalytic rate constant (kcat) of0.96 sÿ1 and a Michaelis constant (Km) of 5.0 mM.Thus, the net rate acceleration (kcat/kuncat � 61,000)is somewhat higher than that found for the originalmurine IgG, while af®nity for hapten and substrateis approximately an order of magnitude lower(Tarasow et al., 1994).

The crystal structures of unliganded chimeric21D8 and its complex with 1b were determined at2.10 AÊ and 1.61 AÊ resolution, respectively (Table 1).The overall structures are well resolved andresemble those of other antibodies in their generalfeatures (Figure 2(a)). Unexpectedly, the disul®debond between CysL134 and CysL194 (residues arenumbered according to the Kabat scheme (Kabatet al., 1991)) in the CL domain appears to be par-tially reduced, with the side-chain of CysL194adopting different conformations in the oxidizedand reduced forms of the antibody. This ®nding isconsistent with other reports (Frisch et al., 1996)suggesting that covalent disul®de linkage of theb-strands is not obligatory for proper folding of theimmunoglobulin domain.

Excellent density was found for the hapten atthe combining site formed by the complementar-ity-determining regions (CDRs) of the VL and VH

domains (Figure 2). With the exception of CDR H3,the hypervariable loops of 21D8 adopt canonicalconformations (Al-Lazikani et al., 1997). The H3loop is unusually short, consisting of only threeresidues (Figure 2(a)). In fact, this segment is theshortest structurally characterized CDR H3, andadopts a conformationally restricted type II b-turn(Wilmot & Thornton, 1988). This short hairpinappears to be stabilized by a hydrogen bondbetween the guanidinium group of residueArgH94 and the carbonyl group of IleH95 at thetip of the turn. The side-chains of ArgH94 andTyrH102 face away from the binding pocket andform a tightly packed cluster with ValH2, TyrH27,and TyrH32, which may further bolster the hairpinstructure. The indole side-chain of conservedTrpH103 is often hydrogen bonded to other CDRH3 residues, providing additional conformationalconstraints to longer loops (Shirai et al., 1996), butthis is not possible in 21D8 because of the shortlength of this segment. Instead, a well-de®nedwater molecule (B-value � 37 AÊ 2) hydrogen bondsto TrpH103 in the unliganded antibody, while the5-sulfonate oxygen atom of the hapten assumesthis role in the complex (Figure 3(b) and (c)).

Although antibodies are subject to a variety ofligand-induced conformational changes, compari-son of the hapten complex with the unligandedantibody shows that only minor structuralrearrangements accompany hapten binding to21D8. Essentially no change (0.4 �) occurs in theelbow angle where the pseudo 2-fold axes of thevariable and the constant domains intersect(Figure 4). In addition, the root-mean-square devi-ation (r.m.s.d.) for all atoms involved in haptenrecognition is only 0.63 AÊ for the two structures

Table 1. Diffraction data and re®nement statistics for Fab 21D8

Diffraction data Hapten complex Unliganded

Space group P212121 P1Unit cell a � 39.2, b � 43.9, c � 221.2 a � 39.3, b � 44.3, c � 59.5

(AÊ , � for a, b, and g) a � 88.0, b � 79.1, g � 89.9Data collection temperature ÿ176 �C Room temperatureResolution range (AÊ ) 50.0-1.61 50.0-2.10Highest resolution shell (AÊ ) 1.67-1.61 2.18-2.10Number of observations 176,796 46,419Number of unique reflections 49,213 21,083Completeness (%) 96.6 (98.8) 91.8 (90.8)Mean I/sI 20.5 (3.4) 11.4 (2.8)Rsym (%) 5.4 (42.1) 9.4 (43.4)

RefinementRcryst (%) 18.8 19.6Rfree (%) 25.1 23.8Number of protein atoms 3270 3270Number of hapten molecules 1 0Number of glycerol molecules 2 0Number of water molecules 297 179Average B-values (AÊ 2):

Protein 22.9 29.8Hapten 16.7 N.A.Glycerol 20.3 N.A.Water 34.1 39.6

R.m.s. deviations from:Ideal bond length (AÊ ) 0.008 0.015Ideal bond angle (�) 2.1 2.0Ideal dihedral angles (�) 29.3 27.9Ideal improper angles (�) 1.3 1.0

Ramachandran plot (%)Favored 90.7 94.2Allowed 8.7 4.9Generously allowed 0.3 0.5Disallowed 0.3 0.3

Values in parentheses indicate the statistics for the highest resolution shell. Rcryst and Rfree values were calculated for all data. TheRfree value was calculated with approximately 10 % of all re¯ections that were not included in the re®nement. Ramachandran plotswere generated with PROCHECK (Laskowski et al., 1993). As observed for many Fabs, residues 127 to 136 of CH1, as well as threeC-terminal residues in the light chain and residues 228 to 230 of the heavy chain have poor electron density and high B-values(Arevalo et al., 1993). Electron density for residues beyond 230 of the heavy chain was also undiscernible. The single residue in a dis-allowed region of the Ramachandran plot is L51, which forms a strained g-turn, as in most other Vk structures (Arevalo et al., 1993).Two glycerol molecules were found in the hapten complex structure; one is very well de®ned and is deeply buried at the CLÐCH1interface, where it forms extended hydrogen-bonding interactions with surrounding side-chain and backbone atoms, as well as abound water molecule. The second, less well de®ned glycerol molecule is located at the VHÐCH1 interface, where it forms onehydrogen bond to a side-chain with poor geometry, and two other hydrogen bonds to proximal, well-de®ned water molecules.

Structures of a Decarboxylase Antibody 1215

(Figure 3(c)). The relative immobility of the 21D8pocket is unexpected in light of the relatively smallburied surface area of the VH/VL-interface (ca1100 AÊ 2), which has been found to correlate withconformational ¯exibility in other antibodies(Wilson & Stan®eld, 1994). However, the small sizeof CDR H3 may help rigidify the structure. Thisloop generally undergoes the greatest rearrange-ment upon antigen binding (Wilson & Stan®eld,1994), particularly towards its tip, but its shortlength and well-packed hairpin structure in 21D8could limit the conformational possibilities.

Hapten recognition

The 21D8 binding pocket is a relatively narrowslot in the surface of the protein, approximately4 AÊ wide, 10 AÊ long, and 11 AÊ deep (Figure 3(a)).The naphthalene disulfonate portion of 1b is essen-tially completely buried (97.2 %) in the complex,with the light and heavy chains each contributing

roughly half of the productive interactions (50.6and 49.4 %, respectively). CDRs L3, H1, and H3provide most of the contacts with the ligand (40.3,16.9, and 29.9 %, respectively); L1 plays but aminor role (2.6 %), while CDRs H2 and L2 makeno contacts with the ligand at all (Figure 3(b)). Thedominance of H3 and L3, as well as the exclusionof L2 contacts, is typical for small molecule recog-nition by antibodies, but the contribution of theheavy chain is proportionally less than in manyantibody-hapten complexes, largely because of theshort H3 (Wilson & Stan®eld, 1994).

The ¯at hydrophobic naphthalene portion of thehapten is sandwiched between two hydrophobicsurfaces formed by the H3 (ValH93 and IleH95)and L3 (LeuL89 and TyrL91) CDR loops(Figure 3(b)). A total of 60 van der Waals contactsare observed between the hapten and the ``hydro-phobic slot'' of the antibody. Notably, IleH95 andTyrL91 each contribute around 25 % of all contacts.A number of other hydrophobic residues from

Figure 2. Fab 21D8-hapten complex structure. (a) Structure of chimeric 21D8 looking into the binding cleft. Electrondensity for the hapten in the binding pocket is shown. Ca-traces of the light and the heavy chains are colored pinkand blue, respectively; the segments corresponding to the CDR loops are slightly darker and labeled. Carbon atomsare shown in yellow, oxygen atoms in red, nitrogen atoms in blue, and sulfur atoms in green. (b) A sA-weighted2Fo ÿ Fc map calculated to 1.61 AÊ resolution and contoured at 1.5 s in the region of the bound hapten. Alternativeconformations of SerL34 are shown.

1216 Structures of a Decarboxylase Antibody

Figure 3. The Fab 21D8 combining site with bound hapten. (a) Molecular surface representation of the 21D8 bind-ing pocket region. Hapten sulfonate groups bind in regions that are positively charged (blue), while the naphthalenemoiety is entirely buried within an apolar pocket (gray). (b) View of the 21D8 hapten binding site showing all thehapten-contacting residues and their molecular interactions with the hapten. The coloring scheme is the same as forFigure 2. (c) View of the hapten binding pocket of the unliganded 21D8 structure, showing the side-chain confor-mations of the hapten-contacting residues and the bound water molecules (red spheres). The hapten complex struc-ture (darker color for Fab) is overlaid to emphasize the position of the bound hapten (semitransparent structure) andthe slight differences in the binding site conformations.

Structures of a Decarboxylase Antibody 1217

Figure 4. Comparison of free and bound Fab 21D8. Overlay of the Ca-traces of the entire Fab fragment of the unli-ganded 21D8 and its hapten complex, indicating that the entire structure undergoes only a small conformationalchange upon hapten binding. The coloring scheme follows that of Figure 2, with the liganded Fab in light pink andlight blue.

1218 Structures of a Decarboxylase Antibody

both the light and heavy chains (LeuL36, PheL98,and ValH37) further increase the hydrophobicity ofthe pocket, although they make few explicit con-tacts with bound hapten.

The 5-sulfonate group of the hapten docks in ananion binding site at the bottom of the pocket,while the 1-sulfonate interacts with a second anionbinding site closer to the surface. This orientationallows the acetamide linker to emerge from thecavity (Figure 3(a)). The more deeply buried anionsite is constructed from the side-chains ofTrpH103, SerL34, ArgL46 and the backbone amidegroup of AlaH101. The side-chains of HisH35 andArgL96 (through a bidentate interaction) constitutethe second site. The hydrogen-bonding geometryof both sulfonate binding sites is well suited forrecognition of tetrahedral oxyanions (Baker &Hubbard, 1984). No water molecules are observedwithin the pocket of the complex, but in the un-liganded structure several well-de®ned water mol-ecules occupy positions adopted by the sulfonateoxygen atoms in the hapten complex (Figure 3(c)).

Catalytic mechanism

Antibody 21D8 exhibits excellent shape andchemical complementarity to its hapten. Like 1,substrate 2 and transition state 3 should easilydock into the hydrophobic slot between the H3and L3 CDR loops, placing its carboxylate groupin or near one of the anion binding sites. Becausethe availability of two anion binding pockets cre-ates ambiguity about the preferred orientation ofthe substrate, attempts were made to obtain astructure of the antibody complexed with 5-nitro-3-

carboxyindole, a stable substrate analog. Unfortu-nately, crystals grown in the presence of this com-pound had no interpretable density within thebinding pocket, presumably re¯ecting either poorsolubility of the analog, its relatively low af®nityfor 21D8, or multiple bound conformations. Dock-ing experiments with substrate 2 and the transitionstate 3 were, therefore, performed computationally.

The geometries of 2 and 3 were fully optimizedat the B3LYP/6-31 � G(d) level of theory usingGaussian98 (Frisch et al., 1998). Selected bondlengths and partial charges are shown in Figure 5.The optimization revealed that substrate and tran-sition state possess three distinct regions of nega-tive charge: the carboxylate group, the nitro group,and the NÐO moiety in the isoxazole ring, whilehapten 1 has two anionic sulfonate groups. Opti-mized 2 and 3 were docked into 21D8 with bothAutoDock V.2.4 (Morris et al., 1996, which uses aMonte Carlo simulated annealing technique tosample binding modes and grid-based af®nitypotentials to compute gas phase interaction ener-gies and AutoDock V.3.0 (Morris et al., 1998, whichuses a genetic algorithm to sample binding modesand is parameterized such that predicted bindingenergies re¯ect free energies of binding from aqu-eous solution). In all calculations, the carboxylategroup was allowed to rotate freely, since it hasbeen shown that this rotation has a shallow poten-tial (Zipse et al., 1995).

Both versions of AutoDock predict several bind-ing modes for 2 and 3 within 1 kcal/mol of eachother. In all complexes, the benzisoxazole ring sys-tem docks in the hydrophobic slot in an orientationsimilar to that of the aromatic portion of the hap-

Figure 5. Structures on the decarboxylation reactioncoordinate optimized at the B3LYP/6-31 � G(d) level.Selected bond lengths (AÊ ) and CHELPG atomic charges(italic type) are shown for substrate 2 and transitionstate 3. Carbon atoms are shown in gray, oxygen atomsin red, and nitrogen atoms in blue.

Figure 6. Computational docking of the transitionstate into Fab 21D8. (a) One of the transition state bind-ing modes predicted by AutoDock V.2.4 (Morris et al.,1996) where the carboxylate group interacts with theanion binding site that is conserved in certain esterolyticantibodies (see Figure 8). (b) An alternative predicteddocking orientation is shown where the carboxylategroup of the transition state is now bound in the deeperanion binding site and the nitro group in the conservedanion binding site. In this docking mode, an additionalspeci®c hydrogen-bonding interaction at the developingoxyanion of the breaking NÐO bond, which has beenproposed to introduce transition state stabilization (Gao,1995; Na et al., 1996; Zipse et al., 1995), is observed.

Structures of a Decarboxylase Antibody 1219

ten. This allows the carboxylate group to occupyeither of the anion binding sites, where it formsone or two hydrogen bonds with antibody combin-ing site residues (Figure 6). In the lowest-energycomplexes, the nitro substituent docks simul-taneously at the second anion binding site.

More sophisticated quantum mechanical calcu-lations will be needed to pin down the precise ori-gins of catalysis in this system, but the AutoDockprediction that 21D8 stabilizes the substrate andtransition state to a similar extent as in the gasphase or aprotic solvent (where the reaction isexpected to be extremely rapid) is intuitivelyreasonable. The partially polar antibody bindingsite can exploit extensive van der Waals contactsand electrostatic interactions as the driving forcefor binding throughout the decarboxylation reac-tion. By contrast, water is expected to stabilize thelocalized charge of the carboxylate group muchmore than the delocalized nitrophenoxide anion inthe transition state, leading to a signi®cantly higheractivation barrier for reaction in aqueous solution(Gao, 1995; Zipse et al., 1995).

Although suboptimal binding of the planar car-boxylate anion in a pocket evolved for recognitionof a tetrahedral oxyanion is likely to favour decar-boxylation, the mechanism of the antibody-catalyzed reaction appears to be distinct from atypical medium effect in that the carboxylategroup is not simply desolvated in an apolar pock-et. Instead, as the reaction proceeds within theantibody combining site, interactions with the car-boxylate in the ground state are replaced by effec-tive interactions with the charge-delocalizedtransition state. Notable in this regard are speci®chydrogen-bonding interactions between hydrogenbond donors in the combining site and the heteroa-toms of the breaking NÐO bond (e.g. Figure 6(b)).Previous theoretical studies on similar reactionshave shown that hydrogen bonding to the incipientphenoxide can reduce the activation energy signi®-cantly (Gao, 1995; Na et al., 1996; Zipse et al., 1995),and such interactions may be essential to catalysisof the decarboxylation reaction by 21D8. Detailed

quantum mechanical investigations to assess thispossibility are in progress.

Comparison with serum albumins andother decarboxylases

Like 21D8, serum albumins have well-de®nedhydrophobic pockets functionalized with cationicresidues (Carter & Ho, 1994). They also bind small,heterocyclic carboxylic acids and catalyze a varietyof reactions (Carter & Ho, 1994), including asolvent-sensitive deprotonation (Hollfelder et al.,1996; Kikuchi et al., 1996). Not surprisingly,bovine serum albumin has been found to promotethe decarboxylation of 2 (kcat � 1.67 � 10ÿ3 sÿ1,Km � 125 mM), but its catalytic ef®ciency

1220 Structures of a Decarboxylase Antibody

(kcat/kuncat � 110) is substantially lower than that of21D8. The carboxylate group of the bound sub-strate may be more extensively hydrogen bondedthan in 21D8, inhibiting the reaction. Alternatively,stabilizing interactions with the incipient phenox-ide may be lacking.

Several natural decarboxylase enzymes, includ-ing histidine (Gallagher et al., 1989) and ornithine(Grishin et al., 1999) decarboxylases, also possesspredominantly hydrophobic pockets for bindingtheir substrate carboxylate group. However, theytypically contain at least one ionizable group in thevicinity of the substrate carboxylate that may facili-tate substrate binding in a relatively low dielectricenvironment and/or serve as a proton source fol-lowing decarboxylation (Hurley & Remington,1992; Jencks, 1975). Extensive non-covalent inter-actions with non-reacting portions of the substratecan be used to position the carboxylate group forreaction, as seen recently in the case of orotidinemonophosphate decarboxylase (Appleby et al.,2000; Harris et al., 2000; Miller et al., 2000; Wu et al.,2000). Alternatively, in many of these systems,covalent interactions between substrate and anenzyme-bound cofactor (pyruvamide (Gallagheret al., 1989), pyridoxal phosphate (Grishin et al.,1999), or thiamine pyrophosphate (KoÈnig, 1998))are exploited to hold the substrate in place. Thecofactors further augment catalysis by providingan electron sink for stabilizing the anion formedupon decarboxylation. In contrast, 21D8 promotesthe direct decarboxylation of its substrate withoutrecourse to covalent intermediates and must stabil-ize the incipient phenoxide anion at the transition

Figure 7. Amino acid sequences of 21D8 and its putativemature 21D8 sequence are designated for the germline sequand square brackets, respectively. CDR residues are shownagainst the international ImMunoGeneTics (IMGT) databaseJ-segment of the 21D8 light chain are predicted to be Ig kgene differs from its precursor at 12 of 285 bases, while therespond to seven amino acid changes in the V-segment,V-segment of the heavy chain appears to be Ig H-VI, whichchanges. The D-minigene seems to provide the ®rst of the thidenti®cation of the germline gene is dif®cult because of itsdues of CDR H3, appears to derive from the germline geneintroduced by the primers used to clone the genes. Enginechanges at positions L1, L2, L107, H5, and H6, but the cataly

state through non-covalent hydrogen bondingand/or electrostatic interactions.

Similarities to other antibodies

The wealth of sequence information available forantibodies (Kabat et al., 1991) permits tracing of theimmunological origins of 21D8. In Figure 7, thesequences of the heavy and light-chain variableregions of 21D8 are compared with their putativegermline precursors. During the process of af®nitymaturation, 14 somatic mutations appeared in theheavy chain and seven in the light chain. Of thealtered amino acid residues, only two interactdirectly with the hapten, the Pro! Arg mutationat position L96 provides one of the polar contactswith the 1-sulfonate of 1, and the Ala! Valmutation at position H93 extends the hydrophobicsurface in contact with the naphthalene ring. Theother mutated residues are located distant from thebinding site. As suggested previously (Patten et al.,1996; Wedemayer et al., 1997), accumulation ofsuch changes may ®ne-tune the shape of the bind-ing site and/or contribute to reduced side-chainand backbone ¯exibility of the binding-site CDRloops.

In addition, structural analysis has revealedstriking similarities between 21D8 and a series ofunrelated esterolytic antibodies (Figure 8(a)). Theesterases were raised against aryl phosphonatehaptens and share a number of structural featuresdespite low overall sequence identity (MacBeath &Hilvert, 1996). First, they all have a relatively deephydrophobic pocket that extends into the frame-

germline precursor. Only residues that differ from theences. The J and D-regions are enclosed by parenthesesin red. Based on a sequence-similarity search performed(Lefranc et al., 1999), the germline genes for the V and

-V9S3 and Ig k-J1*02, respectively. The 21D8 V-segmentJ-segment differs by only one base. These mutations cor-and one in the J-segment. The germline gene for thecontains 26 mismatches corresponding to 15 amino acidree residues that constitute CDR H3, although a positiveshort length. The J-segment, encoding the last two resi-Ig H3. Mutations at positions L4 and H9 were probablyering of the chimeric Fab fragment required additionaltic ef®ciency of 21D8 was not greatly affected.

Figure 8. Structural convergence of decarboxylase and esterase antibodies to create an anion binding site. (a) Com-parison of the aryl tetrahedral anion recognition by 21D8 (upper left), 48G7 (Patten et al., 1996; upper right), 17E8(Zhou et al., 1994; lower left), and CNJ206 (Charbonnier et al., 1995; lower right), shows remarkable structural conver-gence among these independently derived antibodies. The coloring scheme is the same as for Figure 2. Only the resi-dues that signi®cantly contribute to the aryl tetrahedral anion recognition are displayed. Linkers of the haptens for21D8 and 17E8 are omitted for clarity. (b) Overlay of active site residues in 21D8 (blue) and the esterolytic antibody48G7 (Patten et al., 1996; red). The extraordinary similarity of the positions of the bound sulfonate sulfur for 21D8and phosphonate phosphorus for the hydrolytic antibodies are illustrated by green and purple spheres, respectively.Only TyrH100 of 48G7 lacks a counterpart in the corresponding ion binding site of 21D8. The residue in 48G7 thatcorresponds to SerL34 in 21D8 is a glycine, denoted by a blue sphere that indicates its Ca position.

Structures of a Decarboxylase Antibody 1221

work region where the aryl group of the haptenbinds. Second, they exploit multiple hydrogenbonds and salt bridges near the surface of thecavity for recognition of the anionic, tetrahedralphosphonate. These features correspond to thenaphthalene and 1-sulfonate binding regions in21D8 respectively. In fact, direct comparison of the

structures of 21D8 and esterase 48G7 (Patten et al.,1996; Wedemayer et al., 1997) shows that ten of the14 hapten-contacting residues are conserved withan r.m.s.d. of 0.78 AÊ for all atoms in those residues(Figure 8(b)).

These observations suggest that the immune sys-tem is strongly biased in its response toward tetra-

1222 Structures of a Decarboxylase Antibody

hedral aryl oxyanions. Rather than observing anessentially limitless variety of structural solutions,a small set of similarly con®gured combining sitesappears to be the dominant response for the bind-ing of such compounds. The use of a commongermline framework for recognizing chemicallydistinct ligands has been seen previously with var-ious hydrophobic haptens (Xu et al., 1999). How-ever, structural convergence in the present instanceis all the more remarkable in that different germ-line sequences are utilized. Further ®ne-tuning ofthis basic structural motif is then achieved throughsomatic mutation and the diversity inherent in theH3 loop. For example, residues SerL34 andTrpH103 contribute important interactions withthe 5-sulfonate of 1 in the 21D8 complex. On theother hand, in the mature 48G7 esterase, residueL34 is mutated to a Gly, providing room for bind-ing the aromatic portion of its aryl phosphonatehapten (Wedemayer et al., 1997), while TrpH103 isengaged in a canonical hydrogen-bonding inter-action with the backbone carbonyl oxygen ofGlyH100K in the H3 loop and is unavailable forligand recognition. The longer CDR H3 in 48G7(®ve rather than three residues) also provides anextended binding surface for the long aliphaticlinker of its hapten, which has no counterpart in21D8.

Given the observed similarities between anti-bodies raised to different haptens, the range ofpossible structural variation elicited in response toan individual hapten is of considerable interest.The large majority of anti-1 antibodies from thefusion that yielded 21D8 are catalytically inactive,while some of the active variants appear to bemechanistically distinct from 21D8, as judged bytheir activation parameters and other properties(Tarasow et al., 1994). It will be interesting to deter-mine how such antibodies recognize 1, to whatextent they differ from 21D8 genetically and struc-turally, and whether they are related to the anti-phosphonate antibodies.

In this context, the chemical versatility of the tet-rahedral oxyanion binding sites in 21D8 and 48G7is reminiscent of the active sites of functionallydiverse enzymes that have evolved from proteinscatalyzing similar types of reaction (Babbitt &Gerlt, 1997). In the case of the esterases, the arylphosphonate hapten induces an anion binding sitethat binds hydroxide, a localized anion, and pro-motes its attack on an ester by stabilizing theresulting tetrahedral oxyanionic intermediate andits ¯anking transition states (MacBeath & Hilvert,1996). The anion binding site of 21D8 elicited bythe aryl sulfonate analogously binds a localizedcarboxylate anion in the ground state and facili-tates transfer of charge to the developing phenox-ide in the transition state; here, the phenoxideoxygen atom roughly corresponds to the oxyanionin the tetrahedral intermediate formed duringesterolysis. This insight may prove useful indesigning more effective haptens for other chemi-cal transformations.

Conclusions

The activity of 21D8 approximates the accelera-tion achieved upon transfer of the substrate fromwater to acetonitrile (Kemp & Paul, 1975). Optim-ization of the catalytic ef®ciency of the antibody isnow an obvious challenge, particularly as decar-boxylation is up to four orders of magnitude fasterin other solvents, such as dimethylsulfoxide or hex-amethylphosphoramide, than in acetonitrile (Kemp& Paul, 1975). The structure of 21D8 suggests thatresidual hydrogen-bonding interactions with thesubstrate carboxylate group, though less optimalthan those in aqueous solution, are probably theprimary limitation on catalysis at the active site.Mutations that reduce or remove these interactionswithout sacri®cing the af®nity for the transitionstate are likely to lead to substantial increases inthe catalytic rate. Furthermore, modi®cation of thetwo anion binding sites and improvement of thepacking interactions with the transition state mayallow differentiation of the several predicted bind-ing mode; and enhance catalytic ef®ciency. Theresults of such studies will be relevant to under-standing and exploiting medium effects more effec-tively for promoting this and other solvent-sensitive processes.

Materials and Methods

Cloning of the antibody genes

Poly(A)� mRNA was isolated from the hybridomaproducing antibody 21D8 (Lewis et al., 1991). A cDNAlibrary was constructed with the Great Lengths cDNAsynthesis kit (Clontech). The VL and VH geneswere cloned using the polymerase chain reaction withthe following primers: VL sense, GCGGTTCA-GAGCTCCAGCTGACTCAGTCT (the SacI restrictionsite is underlined); VL anti-sense, TTTATTT-CAAGCTTGGTGCCTCCACCGAA (HindIII); VH sense,TCCAACTGCTCGAGCCTGGGACTGAACTG (XhoI);VH anti-sense, GCAGAGACGGTGACCAGAGTCCCTTG(BstEII). The ampli®ed fragments were puri®ed, clonedinto appropriate sites in the vector p4xH-M13 (Ulrichet al., 1995) and sequenced. The resulting expressionplasmid, p4xH-21D8, allows production of 21D8 as a chi-meric murine-human Fab fragment in which the VL andVH segments of the catalytic antibody are fused tohuman Ck and gamma CH1 regions, respectively (Ulrichet al., 1995).

Production, purification and characterization ofchimeric 21D8 Fab

The TOPP2 E. coli strain (Strategene) was transformedwith the p4xH-21D8, and high-density fermentation withBIOFLO 3000 (New Brunswick) was performed asdescribed (Ulrich et al., 1995). The chimeric Fab fragmentwas puri®ed from crude periplasmic lysates by proteinG af®nity chromatography, followed by monoS cation-exchange chromatography. The purity of the sample wascon®rmed by SDS-PAGE and isoelectric focusing gelelectrophoresis. The ®nal yield of Fab protein wasapproximately 10 mg/l. Hapten af®nity was determinedby enzyme-linked immunosorbent assay (ELISA). Cata-

Structures of a Decarboxylase Antibody 1223

lytic activity was assessed by monitoring the decarboxy-lation of 5-nitro-3-carboxybenzisoxazole at 380 nm asdescribed (Lewis et al., 1991).

Crystallization and data collection

Chimeric 21D8 Fab, concentrated to 26 mg/ml in50 mM sodium acetate buffer (pH 5.5), was crystallizedfrom 33 % (v/v) polyethylene glycol monomethyl ether(MPEG) 550 in 100 mM imidazole malate (pH 6.0) in thepresence of a tenfold molar excess of 2-acetamido-1,5-naphthalene disulfonate 1b. Crystals of the unligandedantibody were obtained from 21.5 % (w/v) MPEG 5000in 100 mM sodium acetate buffer (pH 5.5). Crystals weregrown by the sitting-drop vapor-diffusion method at22.5 �C.

A high-resolution (1.6 AÊ ) data set was collected atÿ176 �C from a single crystal of the 21D8-hapten com-plex with glycerol (25 % (v/v)) as cryoprotectant onbeamline 9-1 at the Stanford Synchrotron Radiation Lab-oratory. Addition of cryoprotectants to crystals of theunliganded antibody dramatically increased mosaicity.Thus these diffraction data were collected at room tem-perature from two crystals on an in-house facilityequipped with a 30 cm MAR-Research image platedetector mounted on a Siemens generator operated at50 mA and 50 kV. Each data set was integrated, scaled,and reduced with the HKL (DENZO and SCALEPACK)program suite (Otwinowski & Minor, 1997). Crystallo-graphic parameters and data statistics are summarizedin Table 1.

Structure determination and analysis

The structure of the 21D8-hapten complex was deter-mined by molecular replacement, performed with EPMR(Kissinger et al., 1999), over the resolution range 10.0 to4.00 AÊ . The variable domain of the murine anti-in¯uenzahemagglutinin HA1 peptide antibody 26/9 (Churchillet al., 1994) and the constant domain of the chimeric Fab48G7 (Patten et al., 1996) were employed as initial searchmodels. The best solution had a correlation coef®cient of54.3 and was subjected to X-PLOR (BruÈ nger, 1992) rigid-body re®nement to give an R-value of 39.6 %. Initialcycles of re®nement were performed with data between10.0 and 1.90 AÊ using bulk-solvent correction, conju-gated gradient minimization, and individual positionaland B-factor re®nement protocols from X-PLOR. Manualrebuilding of the model against 3Fo ÿ 2Fc omit mapswas performed during each re®nement cycle with O(Jones et al., 1991). The Rfree was calculated at each re®ne-ment step using 4790 (8.9 %) of the 42,517 observedre¯ections as the reference data set; the quality of theelectron density maps was also evaluated by visualinspection. The hapten was constructed and energy-minimized at the semi-empirical SH2 level in Spartanv5.0 (Wavefunction, Inc.), and incorporated into themodel at the ®nal stage of X-PLOR re®nement. Furtherre®nement, using data from 10.0 to 1.61 AÊ , was per-formed with SHELXL-97 (Sheldrick & Schneider, 1997).Water molecules were modeled by SHELXWAT(Sheldrick & Schneider, 1997). Any water moleculeslacking a plausible hydrogen-bonding partner, havingB-values higher than 60 AÊ 2, or with electron densityweaker than the 1.0 s level in the 3Fo ÿ 2Fc maps, wererejected in subsequent re®nement cycles. Two glycerolmolecules were identi®ed during model rebuilding andwere included in the model. Similarly, nine residues

which have side-chains with apparent multiple confor-mations were re®ned to plausible occupancies and B-values.

The structure of unliganded 21D8 was determined bymolecular replacement using the hapten complex as asearch model. The program AMoRe (Navaza, 1994) wasutilized for the resolution range 10.0 to 3.00 AÊ . Anunambiguous solution with an R-value of 30.6 % and acorrelation coef®cient of 72.8 was obtained. Subsequentre®nement, model rebuilding and water modeling wereperformed as described above for the resolution range10.0 to 2.20 AÊ , except that the Crystallography and NMRSystem (CNS) program (BruÈ nger et al., 1998), which uti-lizes the maximum-likelihood re®nement protocol, wasused. Rfree was calculated with 2002 (11.0 %) re¯ectionsof the 18,180 observed re¯ections as the reference dataset.

The quality of the structures was examined by theprogram PROCHECK v3.3 (Laskowski et al., 1993). Devi-ations from ideal geometry were further evaluated usinga protocol in CNS. Ligand-antibody and VH/VL inter-actions were analyzed with the program MS (Connolly,1983) using a probe atom radius of 1.7 AÊ . All structuresuperimpositions and r.m.s.d. calculations were donewith ProFit V1.7 (A.C.R. Martin, SciTech Software). AllFigures except 1, 5, and 7 were created with the pro-grams Bobscript (Esnouf, 1997), Raster3D (Merritt &Murphy, 1994), and Adobe Photoshop 5.0 (Adobe Sys-tems Incorporated). The molecular surface representationin Figure 3(a) was created with GRASP (Nicholls et al.,1991), using a probe atom radius of 1.6 AÊ , and sub-sequently rendered in Raster3D.

Theoretical calculations and analysis

The geometries of substrate 2 and transition state 3 fordecarboxylation were fully optimized at the B3LYP/6-31 � G(d) level of theory using Gaussian98 (Frisch et al.,1998). Partial charges were calculated using the electro-static potential-based CHELPG scheme (Breneman &Wiberg, 1990). Substrate and transition state weredocked into 21D8 using both AutoDock V.2.4 (Morriset al., 1996) and AutoDock V.3.0 (Morris et al., 1998).AutoDock V.2.4 utilizes a Monte Carlo simulated anneal-ing technique to sample potential binding modes, whileAutoDock V.3.0 utilizes a genetic algorithm. The inter-action energy (consisting of van der Waals plus electro-static interactions) of each ligand and antibody in eachbinding orientation was calculated internally by Auto-Dock, utilizing grid-based af®nity potentials to describethe protein. Additionally, interaction energies producedby AutoDock V.3.0 have been parameterized to repro-duce experimental free energies of binding from aqueoussolution.

In each docking simulation, the substrate or transitionstate was allowed to move in a 22.5 AÊ � 22.5 AÊ � 22.5 AÊ

box (i.e. a cubic grid with 61 points per edge, each separ-ated by 0.375 AÊ ) centered roughly on the geometric cen-ter of the hypervariable loops in 21D8. Only the CÐCbond connecting the carboxylate group to the aromaticring system was allowed to rotate during docking. Auto-Dock V.2.4 calculations used an initial annealing tem-perature of 800 K and a temperature-reduction factor of0.92 K per cycle. A schedule of 120 runs, 100 cycles, 3000steps accepted, and 3000 steps rejected was used fordocking each ligand. A population size of 50 and a totalof 120 runs were used for calculations with AutoDockV.3.0. The binding orientations resulting from each calcu-

1224 Structures of a Decarboxylase Antibody

lation were clustered into groups based on the r.m.s.d. ofthe ligand atom positions.

Protein Data Bank accession codes

The coordinates for antibody 21D8 and its haptencomplex have been deposited in the Protein Data Bankwith accession codes 1C5B and 1C5C, respectively.

Acknowledgments

This work was supported, in part, by the NIHCA27489 (I.A.W.), GM38273 (D.H. and I.A.W.), NSFCHE-9986344 (K.N.H.), and Novartis Pharma AG (D.H.).We thank H.D. Ulrich and P.G. Schultz for kindly pro-viding expression plasmid p4xH-M13, R.A. Lerner, R.C.Stevens, and P.G. Schultz for helpful discussions, andthe staff of beamline 9-1 at SSRL. This is manuscript no.13011-MB of The Scripps Research Institute.

References

Al-Lazikani, B., Lesk, A. M. & Chothia, C. (1997). Stan-dard conformations for the canonical structures ofimmunoglobulins. J. Mol. Biol. 273, 927-948.

Appleby, T. C., Kinsland, C., Begley, T. P. & Ealick, S. E.(2000). The crystal structure and mechanism of oro-tidine 50-monophosphate decarboxylase. Proc. NatlAcad. Sci. USA, 97, 2005-2010.

Arevalo, J. H., Stura, E. A., Taussig, M. J. & Wilson, I. A.(1993). Three-dimensional structure of an anti-ster-oid Fab0 and progesterone-Fab0 complex. J. Mol.Biol. 231, 103-118.

Babbitt, P. C. & Gerlt, J. A. (1997). Understandingenzyme superfamilies. Chemistry as the fundamen-tal determinant in the evolution of new catalyticactivities. J. Biol. Chem. 272, 30591-30594.

Baker, E. N. & Hubbard, R. E. (1984). Hydrogen bond-ing in globular proteins. Prog. Biophys. Mol. Biol. 44,97-179.

Breneman, C. M. & Wiberg, K. B. (1990). Determiningatom-centered monopoles from molecular electro-static potentials. The need for high sampling den-sity in formamide conformational analysis.J. Comput. Chem. 11, 361-373.

BruÈ nger, A. T. (1992). X-PLOR, Version 3.851: A Systemfor X-ray Crystallography and NMR, Yale University,New Haven, CT.

BruÈ nger, A. T., Adams, P. D., Clore, G. M., Delano,W. L., Gros, P., Grosse-Kunstleve, R. W., Jiang, J.-S.,Kuszewski, J., Nilges, M., Pannu, N. S., Read, R. J.,Rice, L. M., Simonson, T. & Warren, G. L. (1998).Crystallography & NMR System: a new softwaresuite for macromolecular structure determination.Acta Crystallog. sect. D, 54, 905-921.

Carter, D. C. & Ho, J. X. (1994). Structure of serum albu-min. Advan. Protein Chem. 45, 153-203.

Charbonnier, J.-B., Carpenter, E., Gigant, B., Golinelli-Pimpaneau, B., Eshhar, Z., Green, B. S. & Knossow,M. (1995). Crystal structure of the complex of a cat-alytic antibody Fab fragment with a transition stateanalog: structural similarities in esterase-like cataly-tic antibodies. Proc. Natl Acad. Sci. USA, 92, 11721-11725.

Churchill, M. E., Stura, E. A., Pinilla, C., Appel, J. R.,Houghten, R. A., Kono, D. H., Balderas, R. S.,

Fieser, G. G., Schulze-Gahmen, U. & Wilson, I. A.(1994). Crystal structure of a peptide complex ofanti-in¯uenza peptide antibody Fab 26/9. Compari-son of two different antibodies bound to the samepeptide antigen. J. Mol. Biol. 241, 534-556.

Connolly, M. L. (1983). Analytical molecular surface cal-culation. J. Appl. Crystallog. 16, 548-558.

Esnouf, R. M. (1999). Further additions to MOLSCRIPTversion 1.4, including reading and contouring ofelectron density maps. Acta Crystallog. sect D, 55,938-940.

Frisch, C., Kolmar, H., Schmidt, A., Kleemann, G.,Reinhardt, A., Pohl, E., UsoÂn, I., Schneider, T. R. &Fritz, H.-J. (1996). Contribution of the intramolecu-lar disul®de bridge to the folding stability of REIV,the variable domain of a human immunoglobulin klight chain. Fold. Des. 1, 431-440.

Frisch, M. J., Trucks, G. W., Schlegel, H. B., Scuseria,G. E., Robb, M. A., Cheeseman, J. R., Zakrzewski,V. G., Montgomery, J. A. J., Stratmann, R. E.,Burant, J. C., Dapprich, S., Millam, J. M., Daniels,A. D., Kudin, K. N. & Strain, M. C. et al. (1998).Gaussian 98, Gaussian, Inc., Pittsburgh, PA.

Gallagher, T., Snell, E. E. & Hackert, M. L. (1989). Pyru-voyl-dependent histidine decarboxylase. Active sitestructure and mechanistic analysis. J. Biol. Chem.264, 12737-12743.

Gao, J. (1995). An automated procedure for simulatingchemical reactions in solution. Application to thedecarboxylation of 3-carboxybenzisoxazole in water.J. Am. Chem. Soc. 117, 8600-8607.

Grishin, N. V., Osterman, A. L., Brooks, H. B., Phillips,M. A. & Goldsmith, E. J. (1999). X-ray structure ofornithine decarboxylase from Trypanosoma brucei:the native structure and the structure in complexwith a-di¯uoromethylornithine. Biochemistry, 38,15174-15184.

Harris, P., Navarro Poulsen, J. C., Jensen, K. F. &Larsen, S. (2000). Structural basis for the catalyticmechanism of a pro®cient enzyme: orotidine 50-monophosphate decarboxylase. Biochemistry, 39,4217-4224.

Hollfelder, F., Kirby, A. J. & Taw®k, D. S. (1996). Off-the-shelf proteins that rival tailor-made antibodiesas catalysts. Nature, 383, 60-63.

Hurley, J. H. & Remington, S. J. (1992). Contribution ofcharged side chains, Mg2�, and solvent exclusion toenyzmatic b-decarboxylation of a-keto acids. J. Am.Chem. Soc. 114, 4769-4773.

Jencks, W. P. (1975). Binding energy, speci®city, andenzymic catalysis: the Circe effect. Advan. Enzymol.43, 219-410.

Jones, T. A., Zou, J. Y., Cowan, S. W. & Kjeldgaard, M.(1991). Improved methods for building proteinmodels in electron density maps and the location oferrors in these models. Acta Crystallog. sect. A, 47,110-119.

Kabat, E. A., Wu, T. T., Perry, H. M., Gottesman, K. S.& Foeller, C. (1991). Sequences of Proteins of Immuno-logical Interest, 5th edit., National Institutes ofHealth, Bethesda.

Kemp, D. S. & Paul, K. G. (1975). The physical organicchemistry of benzisoxazoles. III. The mechanismand the effects of solvents on rates of decarboxyla-tion of benzisoxazole-3-carboxylic acids. J. Am.Chem. Soc. 97, 7305-7312.

Kemp, D. S., Cox, D. D. & Paul, K. G. (1975). The physi-cal organic chemistry of benzisoxazoles. IV. Theorigins and catalytic nature of the solvent rate accel-

Structures of a Decarboxylase Antibody 1225

eration for the decarboxylation of 3-carboxybenzi-soxazoles. J. Am. Chem. Soc. 97, 7312-7318.

Kikuchi, K., Thorn, S. N. & Hilvert, D. (1996). Albumin-catalyzed proton transfer. J. Am. Chem. Soc. 118,8184-8185.

Kissinger, C. R., Gehlhaar, D. K. & Fogel, D. B. (1999).Rapid automated molecular replacement by evol-utionary search. Acta Crystallog. sect. D, 55, 484-491.

KoÈnig, S. (1998). Subunit structure, function and organ-isation of pyruvate decarboxylases from variousorganisms. Biochim. Biophys. Acta, 1385, 271-286.

Laskowski, R. A., MacArthur, M. W., Moss, D. S. &Thornton, J. M. (1993). PROCHECK: a program tocheck the stereochemical quality of protein struc-tures. J. Appl. Crystallog. 26, 283-291.

Lefranc, M. P., Giudicelli, V., Ginestoux, C., Bodmer, J.,Muller, W., Bontrop, R., Lemaitre, M., Malik, A.,Barbie, V. & Chaume, D. (1999). IMGT, the inter-national ImMunoGeneTics database. Nucl. AcidsRes. 27, 209-212.

Lewis, C., KraÈmer, T., Robinson, S. & Hilvert, D. (1991).Medium effects in antibody-catalyzed reactions.Science, 253, 1019-1022.

Lewis, C., Paneth, P., O'Leary, M. H. & Hilvert, D.(1993). Carbon kinetic isotope effects on the spon-taneous and antibody-catalyzed decarboxylation of5-nitro-3-carboxybenzisoxazole. J. Am. Chem. Soc.115, 1410-1413.

MacBeath, G. & Hilvert, D. (1996). Hydrolytic anti-bodies: variations on a theme. Chem. Biol. 3, 433-445.

Merritt, E. A. & Murphy, M. E. P. (1994). Raster3D Ver-sion 2.0. A program for photorealistic moleculargraphics. Acta Crystallog. sect. D, 50, 869-873.

Miller, B. G., Hassell, A. M., Wolfenden, R., Milburn,M. V. & Short, S. A. (2000). Anatomy of a pro®cientenzyme: the structure of orotidine 50-monophos-phate decarboxylase in the presence and absence ofa potential transition state analog. Proc. Natl Acad.Sci. USA, 97, 2011-2016.

Morris, G. M., Goodsell, D. S., Huey, R. & Olson, A. J.(1996). Distributed automated docking of ¯exibleligands to proteins: parallel applications of Auto-Dock 2.4. J. Comput. Aided Mol. Des. 10, 293-304.

Morris, G. M., Goodsell, D. S., Halliday, S., Huey, R.,Hart, W. E., Belew, R. & Olson, A. J. (1998).Automated docking using a Lamarckian geneticalgorithm and an empirical binding free energyfunction. J. Comput. Chem. 19, 1639-1662.

Na, J., Houk, K. N. & Hilvert, D. (1996). Transition stateof the base-promoted ring-opening of isoxazoles.Theoretical prediction of catalytic functionalitiesand design of haptens for antibody production.J. Am. Chem. Soc. 118, 6462-6471.

Navaza, J. (1994). AMoRe: an automated package formolecular replacement. Acta Crystallog sect. A, 50,157-163.

Nicholls, A., Sharp, K. A. & Honig, B. (1991). Proteinfolding and association: insights from the interfacialand thermodynamic properties of hydrocarbons.Proteins: Struct. Funct. Genet. 11, 281-296.

Otwinowski, Z. & Minor, W. (1997). Processing of X-raydiffraction data collected in oscillation mode.Methods Enzymol. 276, 307-326.

Patten, P. A., Gray, N. S., Yang, P. L., Marks, C. B.,Wedemayer, G. J., Boniface, J. J., Stevens, R. C. &Schultz, P. G. (1996). The immunological evolutionof catalysis. Science, 271, 1086-1091.

Reichardt, C. (1988). Solvent Effects on the Rate ofHomogeneous Chemical Reactions, 2nd edit., VCH,Weinheim.

Sheldrick, G. M. & Schneider, T. R. (1997). SHELXL:high-resolution re®nement. Methods Enzymol. 277B,319-343.

Shirai, H., Kidera, A. & Nakamura, H. (1996). Structuralclassi®cation of CDR-H3 in antibodies. FEBS Letters,399, 1-8.

Tarasow, T. M., Lewis, C. & Hilvert, D. (1994). Investi-gation of medium effects in a family of decarboxy-lase antibodies. J. Am. Chem. Soc. 116, 7959-7963.

Ulrich, H. D., Patten, P. A., Yang, P. L., Romesberg, F. E.& Schultz, P. G. (1995). Expression studies of cataly-tic antibodies. Proc. Natl Acad. Sci. USA, 92, 11907-11911.

Wedemayer, G. J., Patten, P. A., Wang, L. H., Schultz,P. G. & Stevens, R. C. (1997). Structural insightsinto the evolution of an antibody combining site.Science, 276, 1665-1669.

Wilmot, C. M. & Thornton, J. M. (1988). Analysis andprediction of the different types of b-turn in pro-teins. J. Mol. Biol. 203, 221-232.

Wilson, I. A. & Stan®eld, R. L. (1994). Antibody-antigeninteractions: new structures and new conformation-al changes. Curr. Opin. Struct. Biol. 4, 857-867.

Wu, N., Mo, Y., Gao, J. & Pai, E. F. (2000). Electrostaticstress in catalysis: structure and mechanism of theenzyme orotidine monophosphate decarboxylase.Proc. Natl Acad. Sci. USA, 97, 2017-2022.

Xu, J., Deng, Q., Chen, J., Houk, K. N., Bartek, J.,Hilvert, D. & Wilson, I. A. (1999). Evolution ofshape complementarity and catalytic ef®ciency froma primordial antibody template. Science, 286, 2345-2348.

Zhou, G. W., Guo, J., Huang, W., Fletterick, R. J. &Scanlan, T. S. (1994). Crystal structure of a catalyticantibody with a serine protease active site. Science,265, 1059-1064.

Zipse, H., Apaydin, G. & Houk, K. N. (1995). A quan-tum mechanical and statistical mechanical explora-tion of the thermal decarboxylation of Kemp's otheracid (benzisoxazole-3-carboxylic acid). The in¯uenceof solvation on the transition state geometries andkinetic isotope effects of a reaction with an awe-some solvent effect. J. Am. Chem. Soc. 117, 8608-8617.

Edited by D. Rees

(Received 6 July 2000; received in revised form 3 August 2000; accepted 4 August 2000)