doi101006jmbi20015404 available online at httpwwwidealibrarycom on J Mol Biol (2002) 316 1061plusmn1069

Atomic (094 AEcirc ) Resolution Structure of an InvertingGlycosidase in Complex with Substrate

Diego M A GueAcircrin1 Marie-Bernard Lascombe1 Marcelo Costabel1

HeAcirc leAacutene Souchon1 Victor Lamzin2 Pierre BeAcircguin3 and Pedro M Alzari1

1UniteAcirc de BiochimieStructurale CNRS URA 2185Institut Pasteur 25 rue du DrRoux 75724 Paris ceAcircdex 15France2EMBL Hamburg OutstationNotkesstrasse 85 HamburgGermany3UniteAcirc de PhysiologieCellulaire CNRS URA 300Institut Pasteur 25 rue du DrRoux 75724 Paris ceAcircdex 15France

Present address D M A GueAcircrinDepartamento de Fisica UniversidaBahia Blanca Argentina

E-mail address of the correspondalzaripasteurfr

Coutinho P M amp Henrissat BCarbohydrate-active enzymes servecnrs-mrs frCAZYindex html

0022-283602051061plusmn9 $35000

The crystal structure of Clostridium thermocellum endoglucanase CelA incomplex with cellopentaose has been determined at 094 AEcirc resolutionThe oligosaccharide occupies six D-glucosyl-binding subsites three oneither side of the scissile glycosidic linkage The substrate and product ofthe reaction occupy different positions at the reducing end of the cleftwhere an extended array of hydrogen-bonding interactions with watermolecules fosters the departure of the leaving group Severe torsionalstrain upon the bound substrate forces a distorted boat25 B conformationfor the glucosyl residue bound at subsite yuml1 which facilitates the for-mation of an oxocarbenium ion intermediate and might favor the break-age of the sugar ring concomitant with catalysis

2002 Elsevier Science Ltd

Keywords inverting glycosidase X-ray crystallography atomic resolutionprotein-carbohydrate interactions reaction mechanism

Corresponding author

Introduction

Glycosidases (EC 321X) are ubiquitousenzymes with functions ranging from glycosidesdegradation to control and mediation of cell-cellinteractions Since the pioneering work of Phillipsand co-workers on lysozyme more than 30 yearsago1 the modes of action of glycosidases havebeen studied extensively due to their possible usein biomass energy conversion and the potential forspeciregc inhibitors to act as new therapeutic agentsin a wide range of infectious diseases Catalyticdomains of glycosidases can be grouped in at least83 distinct protein families on the basis of aminoacid sequence similarities2 with enzymes within asame family following a similar catalytic mechan-ism3 Several of these families involve enzymesthat use a single displacement mechanism leadingto inversion of the anomeric conregguration Exten-

M Costabeld Nacional del Sur

ing author

(1999)r at httpafmb

sive biochemical and crystallographic studies ofinverting glycosidases revealed a large diversity ofstructural architectures4 plusmn 13 and substrate speciregci-ties but conregrmed to a large extent that invertingglycosidases share a common general acidbasemechanism as originally proposed byKoshland1415 However the limited resolution ofmost crystallographic studies and the paucity ofstructural information on enzyme-substrate com-plexes for inverting glycosidases leave importantunanswered questions concerning the detailed con-formation of bound carbohydrate the role of sub-strate distortion on catalysis and the nature ofreaction intermediates We report here structuralevidence at atomic resolution of severe carbo-hydrate distortion for an inverting cellulaseClostridium thermocellum CelA that strongly sup-ports an oxocarbenium ion-like transition state andillustrates how substrate-binding energy can beused as a driving force in catalysis

Inverting endoglucanase CelA from the thermo-philic anaerobe C thermocellum is a component ofthe cellulosome an extracellular multi-enzymaticcomplex that is very efregcient in degrading crystal-line cellulose16 The catalytic domain of CelAwhich belongs to family 8 of inverting glycosi-dases folds into a (aa)6 barrel consisting of sixinternal mutually parallel a-helices interconnectedby six external helices17 This protein folding top-

2002 Elsevier Science Ltd

Table 1 Crystallographic data and reregnement statistics

A Data collectionNumber of measured reflections 903691Number of unique reflections 208140Resolution (AEcirc ) 30-094 (095-094)Completeness () 998 (977)R-merge () 43 (190)Percentage of refs with I gt 2s(I) 918 (729)

B RefinementResolution range (AEcirc ) 10-094 (097-094)Number of reflections 197438 10391 (19946)R-factor 0094 (0135)Free R-factor 0113rms (bond lengths) (AEcirc ) 0014rms (angle distances) (AEcirc ) 0027Protein residues 363Glucosyl residues 6Water molecules with full occupancy 401Isotropic average temperature factors (AEcirc 2)

Protein atoms 77Sugar atoms 92Water molecules 214

Values in parentheses correspond to the higher-resolutionshell

1062 Glycosidase-Substrate Complex at Atomic Resolution

ology is common among inverting glycosidaseswhere it is shared by family 9 endoglucanases4

family 15 glucoamylases5 and family 48cellobiohydrolases12 but has not been observed sofar for retaining glycosidases A glutamate residueat the centre of the active-site cleft Glu95 has beenidentireged as the proton donor in the catalytic reac-tion of CelA but the residue acting as a generalbase could not be identireged unambiguously Tofully characterize the reaction mechanism and toanalyze the role of substrate conformation on cata-lysis we have produced an inactive isostericmutant of CelA in which the proton donor Glu95was substituted by Gln (CelAE95Q) and determinedits crystal structure in complex with substrate Thecrystallographic analysis of the complex at atomicresolution (094 AEcirc ) provides a unique insight intostructural aspects of carbohydrate binding and cat-alysis

Results and Discussion

Crystallographic data and model quality

Inactive CelAE95Q was crystallized in complexwith the substrate cellopentaose a complete dif-fraction data set was collected at 094 AEcirc resolutionusing synchrotron radiation and the structure ofthe complex was reregned to a regnal crystallographicR-factor of 94 (Table 1) The regnal modelincludes 363 amino acid residues 401 fully occu-pied solvent sites and six enzyme-bound D-gluco-syl residues

The stereochemical quality of the model is com-parable to that of other crystal structures reregned atatomic resolution18 All non-glycine residues dis-play main chain dihedral angles that fall withinallowed (904 ) or additionally allowed (96 )regions of the Ramachandran plot as deregned bythe program PROCHECK19 Anisotropic reregne-ment of non-hydrogen atomic temperature factorsresulted in a signiregcant quality improvement ofthe electron density map in most cases allowingunambiguous determination of the proper orien-tation of side-chain residues by discriminatingbetween carbon nitrogen and oxygen atoms

The average error estimates for the differenttypes of chemical bonds between main-chainatoms obtained by inversion of the block matrixlie in the range 0014-0018 AEcirc Although bond dis-tance and angle restraints were relaxed during theregnal cycles of reregnement these estimates show aremarkable overall agreement with the observedrms deviations from ideal geometry20 which fallin the range 0014-0021 AEcirc However a few exper-imental parameters with low error estimates devi-ate signiregcantly from their target values (data notshown) probably indicating genuine differencesfrom standard protein dictionaries

The substrate-binding region

Carbohydrate binds to a deep acidic cleft of the(aa)6 barrel which runs across the molecular sur-face at the N-terminal end of the central helices(Figure 1(a)) The enzyme-substrate complexderegnes six D-glucosyl-binding subsites within thiscleft The sugar chain can be visualized as drawinga V with three subsites on each branch and thescissile glycosidic linkage at the vertex At the non-reducing end of the substrate subsites yuml3 to yuml1(using the nomenclature proposed by Davieset al21) the electron density map reveals a singlewell-deregned conformation for the bound D-gluco-syl residues whereas at the opposite end of thecleft there is clear indication of alternate confor-mations for the sugar residues bound at subsites1 and 2 (Figure 1(b)) With the exception of afew hydroxyl groups that face the bulk solventsugar atoms in these regve positions display a mod-erate anisotropy of their thermal displacements(Figure 2) comparable to that of neighboring pro-tein atoms

Partial binding of substrate is detected at anadditional pocket (subsite 3) within the enzymecleft However the bound glucosyl residue dis-plays higher temperature factors (Figure 2) and noglucosyl binding to this position had beenobserved in previous studies of wild-type CelAcomplexed with different cellooligosaccharides atlower resolution17 probably due to high mobilityor disorder

Bound carbohydrate was modelled as a cellopen-taose molecule (ie the substrate) bound to subsitesyuml3 to 2 and a cellotriose molecule which mimicsthe leaving group moiety bound to subsites 1 to3 as shown in Figure 1(a) This interpretation issupported by several pieces of evidence (a) a cleardisruption of the oligosaccharide chain is observed

Figure 1 (a) Top view of the CelAE95Q-cellopentaose complex Sugar residues corresponding to the substrate (yel-low) and product (violet) of the reaction are shown in stick representation within the enzyme-binding cleft (b) Elec-tron density map of the glucosyl residues bound at subsites 1 2 (c) Stacking interactions at subsite yuml2 (d) Closeview of the catalytic center The glucosyl residue bound at subsite yuml1 displays a distorted 25B boat conformationNote the split position for the water nucleophile

Glycosidase-Substrate Complex at Atomic Resolution 1063

at the reaction center between subsites yuml1 and 1As a consequence the sugar ring assigned to thesubstrate at subsite 1 is covalently linked to bothadjacent glucosyl residues along the cleft while theequivalent sugar residue corresponding to the pro-duct of the reaction has a free OH-4 hydroxylgroup (b) transglycosylation is not possiblebecause CelA is an inverting enzyme and (c) sterichindrance due to crystal packing excludes the pre-sence of additional sugar residues beyond subsiteyuml3 whereas at the other end of the cleft (beyondsubsite 3) there is enough space to position twoadditional sugar residues without interfering withneighboring molecules in the crystal

The analysis of ligand occupation factors lendsfurther support to this model When the occu-pation factors were reregned independently for eachD-glucosyl residue the sugar rings bound at sub-sites yuml3 to yuml1 displayed similar values of occu-pancy (066-073) as did those bound to subsites1 2 corresponding to the substrate (053-054)and the product (035-037) of the reaction On thebasis of these results only regve independent occu-pation factors were regnally considered for thebound saccharide as shown in Figure 2 The twoextremities of the cellopentaose substrate (subsitesyuml3 to 2) in the reregned model display signiregcantlydifferent values of sugar ring occupancy Thiscould indicate that a fraction of these molecules

Figure 2 Drawing of the oligo-saccharidic chain showing the rela-tive anisotropic thermal dis-placement (drawn with the pro-gram ORTEP httpwwwornlgovorteportephtml) The reregnedoccupancies of the different sub-sites are indicated For clarity thethree glucosyl residues assigned tothe product of the reaction havebeen moved away from the sub-strate chain

1064 Glycosidase-Substrate Complex at Atomic Resolution

binds to the enzyme-cleft only through their non-reducing terminal residues or alternatively thatpartial substrate cleavage has taken place in thecrystal However there are some difregculties withboth models The regrst hypothesis would require anunusual torsional strain of the polysaccharidechain at the catalytic center whereas the secondhypothesis seems unlikely because the protondonor catalyst Glu95 has been mutated (no activityof CelAE95Q was detected in qualitative assays) anda signiregcant residual activity would be requiredto account for the observed differences in sugaroccupancy

The inspection of difference Fourier (Fo yuml Fc)maps reveals residual density in the neighborhoodof the reaction center that could not be modelledby a chemically feasible arrangement of solventmolecules Interestingly a sugar ring with a brokenO5-C1 bond at subsite yuml1 can regt the density andthe crystallographic reregnement of such an open-ring model produced a featureless difference Four-ier map (data not shown) It may be tempting tospeculate on the possible implications of thismodel which could account for the partial bindingof substrate to the enzyme cleft and brings to mindthe endocyclic mechanism proposed on theoreticalgrounds for retaining b-glycosidases2223 Howeverit should be kept in mind that there is strong evi-dence against an open-ring mechanism forglycosidases2425 and that the residual electrondensity could arise from an overlapping networkof partially occupied solvent sites andor multiplesugar ring conformations at subsite yuml1 Thereforefurther experiments are required to assess theactual signiregcance of the above interpretation

Protein-carbohydrate interactions

The bound oligosaccharide is stabilized by stack-ing interactions with an array of aromatic side-chain residues lined along the cleft and by an

extended network of direct and water-mediatedhydrogen bonding interactions with protein atoms(Figure 3) Protein-carbohydrate stacking inter-actions occur all along the cleft except for thesugar residue bound at subsite yuml1 (Figure 3(b))The b-faces of the glucopyranosyl rings occupyingsubsites yuml3 and yuml2 are stacked on the indole ringsof tryptophanyl residues 205 and 132 (Figure 1(c))At the other end of the cleft the sugar rings boundat subsites 1 to 3 stack on the phenol groups ofthree tyrosine residues at positions 372 277 and369 respectively In particular the phenolic groupsof Tyr372 and Tyr277 are approximately coplanarand provide a gliding surface covering subsites(1 2) that could facilitate departure of the leav-ing group In the crystal structure the bound glu-cosyl residues display two different positionswithin this plane (Figure 1(b)) which wereassigned to the substrate and product of the reac-tion respectively as discussed above Productrelease would therefore involve a regrst step duringwhich the leaving group rotates slightly and shiftsaway from the reaction center while preservingthe stacking interactions at subsites 1 and 2(Figure 3(b))

Another salient feature of the complex is theextended network of direct and water-mediatedprotein-carbohydrate hydrogen bonding inter-actions Water molecules involved in these inter-actions are well deregned in density althoughin some cases they display highly anisotropictemperature factors or split positions related toalternate sugar conformations Water-mediatedhydrogen bonds predominate at the reducing endof the enzyme cleft (Figure 3(a)) where they couldfacilitate solvatation and subsequent release of theleaving product moiety At the other end of thecleft several protein residues are involved in directhydrogen bonding interactions with sugarhydroxyl groups In particular the hydroxylgroups OH-2 and OH-3 of the glucosyl ring bound

Figure 1 (a) Top view of the CelAE95Q-cellopentaose complex Sugar residues corresponding to the substrate (yel-low) and product (violet) of the reaction are shown in stick representation within the enzyme-binding cleft (b) Elec-tron density map of the glucosyl residues bound at subsites 1 2 (c) Stacking interactions at subsite yuml2 (d) Closeview of the catalytic center The glucosyl residue bound at subsite yuml1 displays a distorted 25B boat conformationNote the split position for the water nucleophile

Glycosidase-Substrate Complex at Atomic Resolution 1063

at the reaction center between subsites yuml1 and 1As a consequence the sugar ring assigned to thesubstrate at subsite 1 is covalently linked to bothadjacent glucosyl residues along the cleft while theequivalent sugar residue corresponding to the pro-duct of the reaction has a free OH-4 hydroxylgroup (b) transglycosylation is not possiblebecause CelA is an inverting enzyme and (c) sterichindrance due to crystal packing excludes the pre-sence of additional sugar residues beyond subsiteyuml3 whereas at the other end of the cleft (beyondsubsite 3) there is enough space to position twoadditional sugar residues without interfering withneighboring molecules in the crystal

The analysis of ligand occupation factors lendsfurther support to this model When the occu-pation factors were reregned independently for eachD-glucosyl residue the sugar rings bound at sub-sites yuml3 to yuml1 displayed similar values of occu-pancy (066-073) as did those bound to subsites1 2 corresponding to the substrate (053-054)and the product (035-037) of the reaction On thebasis of these results only regve independent occu-pation factors were regnally considered for thebound saccharide as shown in Figure 2 The twoextremities of the cellopentaose substrate (subsitesyuml3 to 2) in the reregned model display signiregcantlydifferent values of sugar ring occupancy Thiscould indicate that a fraction of these molecules

Figure 2 Drawing of the oligo-saccharidic chain showing the rela-tive anisotropic thermal dis-placement (drawn with the pro-gram ORTEP httpwwwornlgovorteportephtml) The reregnedoccupancies of the different sub-sites are indicated For clarity thethree glucosyl residues assigned tothe product of the reaction havebeen moved away from the sub-strate chain

1064 Glycosidase-Substrate Complex at Atomic Resolution

binds to the enzyme-cleft only through their non-reducing terminal residues or alternatively thatpartial substrate cleavage has taken place in thecrystal However there are some difregculties withboth models The regrst hypothesis would require anunusual torsional strain of the polysaccharidechain at the catalytic center whereas the secondhypothesis seems unlikely because the protondonor catalyst Glu95 has been mutated (no activityof CelAE95Q was detected in qualitative assays) anda signiregcant residual activity would be requiredto account for the observed differences in sugaroccupancy

The inspection of difference Fourier (Fo yuml Fc)maps reveals residual density in the neighborhoodof the reaction center that could not be modelledby a chemically feasible arrangement of solventmolecules Interestingly a sugar ring with a brokenO5-C1 bond at subsite yuml1 can regt the density andthe crystallographic reregnement of such an open-ring model produced a featureless difference Four-ier map (data not shown) It may be tempting tospeculate on the possible implications of thismodel which could account for the partial bindingof substrate to the enzyme cleft and brings to mindthe endocyclic mechanism proposed on theoreticalgrounds for retaining b-glycosidases2223 Howeverit should be kept in mind that there is strong evi-dence against an open-ring mechanism forglycosidases2425 and that the residual electrondensity could arise from an overlapping networkof partially occupied solvent sites andor multiplesugar ring conformations at subsite yuml1 Thereforefurther experiments are required to assess theactual signiregcance of the above interpretation

Protein-carbohydrate interactions

The bound oligosaccharide is stabilized by stack-ing interactions with an array of aromatic side-chain residues lined along the cleft and by an

extended network of direct and water-mediatedhydrogen bonding interactions with protein atoms(Figure 3) Protein-carbohydrate stacking inter-actions occur all along the cleft except for thesugar residue bound at subsite yuml1 (Figure 3(b))The b-faces of the glucopyranosyl rings occupyingsubsites yuml3 and yuml2 are stacked on the indole ringsof tryptophanyl residues 205 and 132 (Figure 1(c))At the other end of the cleft the sugar rings boundat subsites 1 to 3 stack on the phenol groups ofthree tyrosine residues at positions 372 277 and369 respectively In particular the phenolic groupsof Tyr372 and Tyr277 are approximately coplanarand provide a gliding surface covering subsites(1 2) that could facilitate departure of the leav-ing group In the crystal structure the bound glu-cosyl residues display two different positionswithin this plane (Figure 1(b)) which wereassigned to the substrate and product of the reac-tion respectively as discussed above Productrelease would therefore involve a regrst step duringwhich the leaving group rotates slightly and shiftsaway from the reaction center while preservingthe stacking interactions at subsites 1 and 2(Figure 3(b))

Another salient feature of the complex is theextended network of direct and water-mediatedprotein-carbohydrate hydrogen bonding inter-actions Water molecules involved in these inter-actions are well deregned in density althoughin some cases they display highly anisotropictemperature factors or split positions related toalternate sugar conformations Water-mediatedhydrogen bonds predominate at the reducing endof the enzyme cleft (Figure 3(a)) where they couldfacilitate solvatation and subsequent release of theleaving product moiety At the other end of thecleft several protein residues are involved in directhydrogen bonding interactions with sugarhydroxyl groups In particular the hydroxylgroups OH-2 and OH-3 of the glucosyl ring bound

Figure 2 Drawing of the oligo-saccharidic chain showing the rela-tive anisotropic thermal dis-placement (drawn with the pro-gram ORTEP httpwwwornlgovorteportephtml) The reregnedoccupancies of the different sub-sites are indicated For clarity thethree glucosyl residues assigned tothe product of the reaction havebeen moved away from the sub-strate chain

1064 Glycosidase-Substrate Complex at Atomic Resolution

binds to the enzyme-cleft only through their non-reducing terminal residues or alternatively thatpartial substrate cleavage has taken place in thecrystal However there are some difregculties withboth models The regrst hypothesis would require anunusual torsional strain of the polysaccharidechain at the catalytic center whereas the secondhypothesis seems unlikely because the protondonor catalyst Glu95 has been mutated (no activityof CelAE95Q was detected in qualitative assays) anda signiregcant residual activity would be requiredto account for the observed differences in sugaroccupancy

The inspection of difference Fourier (Fo yuml Fc)maps reveals residual density in the neighborhoodof the reaction center that could not be modelledby a chemically feasible arrangement of solventmolecules Interestingly a sugar ring with a brokenO5-C1 bond at subsite yuml1 can regt the density andthe crystallographic reregnement of such an open-ring model produced a featureless difference Four-ier map (data not shown) It may be tempting tospeculate on the possible implications of thismodel which could account for the partial bindingof substrate to the enzyme cleft and brings to mindthe endocyclic mechanism proposed on theoreticalgrounds for retaining b-glycosidases2223 Howeverit should be kept in mind that there is strong evi-dence against an open-ring mechanism forglycosidases2425 and that the residual electrondensity could arise from an overlapping networkof partially occupied solvent sites andor multiplesugar ring conformations at subsite yuml1 Thereforefurther experiments are required to assess theactual signiregcance of the above interpretation

Protein-carbohydrate interactions

The bound oligosaccharide is stabilized by stack-ing interactions with an array of aromatic side-chain residues lined along the cleft and by an

extended network of direct and water-mediatedhydrogen bonding interactions with protein atoms(Figure 3) Protein-carbohydrate stacking inter-actions occur all along the cleft except for thesugar residue bound at subsite yuml1 (Figure 3(b))The b-faces of the glucopyranosyl rings occupyingsubsites yuml3 and yuml2 are stacked on the indole ringsof tryptophanyl residues 205 and 132 (Figure 1(c))At the other end of the cleft the sugar rings boundat subsites 1 to 3 stack on the phenol groups ofthree tyrosine residues at positions 372 277 and369 respectively In particular the phenolic groupsof Tyr372 and Tyr277 are approximately coplanarand provide a gliding surface covering subsites(1 2) that could facilitate departure of the leav-ing group In the crystal structure the bound glu-cosyl residues display two different positionswithin this plane (Figure 1(b)) which wereassigned to the substrate and product of the reac-tion respectively as discussed above Productrelease would therefore involve a regrst step duringwhich the leaving group rotates slightly and shiftsaway from the reaction center while preservingthe stacking interactions at subsites 1 and 2(Figure 3(b))

Another salient feature of the complex is theextended network of direct and water-mediatedprotein-carbohydrate hydrogen bonding inter-actions Water molecules involved in these inter-actions are well deregned in density althoughin some cases they display highly anisotropictemperature factors or split positions related toalternate sugar conformations Water-mediatedhydrogen bonds predominate at the reducing endof the enzyme cleft (Figure 3(a)) where they couldfacilitate solvatation and subsequent release of theleaving product moiety At the other end of thecleft several protein residues are involved in directhydrogen bonding interactions with sugarhydroxyl groups In particular the hydroxylgroups OH-2 and OH-3 of the glucosyl ring bound

Figure 3 Protein-carbohydrate interactions (a) A schematic detailing hydrogen bonding interactions and distancesThe catalytic residues GluGln95 and Asp278 are shown as sticks Water molecules marked with an asterisk () havepartial occupancy and are linked to either the substrate or the product at subsites 1 2 (b) Protein-sugar stackinginteractions along the substrate-binding cleft

Glycosidase-Substrate Complex at Atomic Resolution 1065

at subsite yuml1 displace two well-deregned water mol-ecules that occupy equivalent positions in the unli-ganded structure of CelA17 These two hydroxylgroups form a bidentate hydrogen bonding inter-action with the carboxylic acid group of Asp152 atthe bottom of the active center (Figure 1(d)) aninteraction that is critical to stabilize the centralsugar ring in a strained boat conformation (seebelow) A similar situation has been described fora family 13 a-amylase26 which also bind a dis-torted sugar residue at subsite yuml1 In this structureit would appear that one of the two hydrogenbonds is particularly strong (with a distance of25 AEcirc ) leading the authors to suggest that thisputative ``low barrier hydrogen bond could play

a role in stabilizing key transition-state interactionsfor a-amylase In our case the two hydrogenbonds between the sugar hydroxyl groups andAsp152 have longer distances 266 and 277 AEcirc

(Figure 3(a)) which renders less plausible thehypothesis of a low barrier hydrogen bond inCelA

The catalytic residues

Family 8 glycosidases hydrolyze the glycosidicbond via a single displacement mechanism withinversion of the anomeric conregguration The car-boxylate group of CelA Glu95 has been identiregedas the acid catalyst17 This functional assignment is

1066 Glycosidase-Substrate Complex at Atomic Resolution

now conregrmed by the strong hydrogen bondinginteraction of Gln95 with the oxygen atom of thescissile glycosidic linkage (Figure 4(a)) In our pre-

Figure 4 Role of sugar distortion in catalysis (a)Local geometry of the catalytic center showing the gen-eral acid Glu95 (Gln95 here) and the putative generalbase (Asp278) catalysts the split position assigned tothe water nucleophile and the D-glucosyl residuesbound at subsites yuml1 and 1 The local geometry of thecatalytic center is further stabilized by the guanidiniumgroup of Arg281 which makes hydrogen bonding inter-actions with the two catalysts (b) Chiral volumes of thepyranose ring carbon atoms revealing a more planararrangement for the anomeric carbon C1 at subsite yuml1that could contribute to weaken the scissile glycosidiclinkage by steric distortion (c) Proposed reaction mech-anism for CelA The boat conformation of the sugar ringfavors a transition state with substantial oxocarbeniumion-like character

vious study we have identireged two aspartic acidresidues Asp152 and Asp278 as possible candi-dates for the base catalyst but a deregnitive assign-ment could not be done in the absence of boundsubstrate The atomic-resolution structure of thereaction center now reveals the nucleophilic watermolecule which is clearly visible in the electrondensity map (Figure 1(d)) facing the anomeric car-bon atom of the glucosyl residue at subsite yuml1This water molecule occupies two alternate pos-itions 09 AEcirc apart (Figure 4(a)) consistent with apartial sugar occupancy of the enzyme cleft Oneof the two solvent sites is sterically hindered by theanomeric carbon atom of the sugar ring (distance28 AEcirc ) and superimposes well with an equivalentwater molecule observed in the crystal structure ofunliganded CelA17 Besides its interaction with theanomeric carbon at subsite yuml1 the nucleophilicwater molecule is involved in strong hydrogenbonding interactions with the hydroxyl group ofTyr215 and the carboxylate group of Asp278(Figure 4(a)) suggesting that these two residuesplay a critical role in catalysis and that Asp278 isthe putative base catalyst in CelA Interestinglythe tyrosine residue is strictly conserved in allfamily 8 endoglucanases but the aspartic acid resi-due is missing from two homologous enzymesfrom Bacillus where it is replaced by an asparagineresidue2728 It follows that at least for some family8 glycosidases the enhancement of water nucleo-philicity by a carboxylate group might not berequired for the reaction to occur Alternativelythe conserved tyrosine residue could be acting as abase catalyst in these enzymes although thiswould require a substantial modiregcation of thetyrosine pKa to account for the enzyme activity atslightly acidic pH values

Substrate distortion

Comparison with the ligand-free structure ofwild-type CelA indicates that carbohydrate bind-ing induces small structural changes in the proteinmostly slight reorientations of aromatic and polarside-chains in contact with substrate In contraststacking and hydrogen bonding interactions withprotein residues impose a continuous torsionalstrain upon the substrate that modireges the confor-mation characteristic of b(14)-linked glucosyl resi-dues in cellulose (Figure 3(b)) signiregcantly A kinkis observed for the oligosaccharidic chain at thescissile glycosidic linkage which in turn imposes adistorted 25B conformation to the sugar ring in theyuml1 subsite (Table 2) Presumably full substrateoccupancy of available subsites (at least yuml3 to 2)is energetically required to stabilize the observedoligosaccharide conformation thus explaining whyCelA and other family 8 glycosidases cleave onlyoligosaccharides with a high degree of polymeriz-ation (DP 55)29 Assuming that the enzyme under-goes no major conformational change theobserved torsional stress on the substrate impliesthat the initial formation of the protein-sugar com-

Table 2 Oligosaccharide conformation

A Sugar ring2 (deg) (deg) QT (AEcirc )

Substrateyuml3 yuml35 9 058yuml2 yuml6 4 059yuml1 127 103 0541 yuml116 11 0622 40 6 058Product1 61 3 0592 19 4 0593 yuml87 10 066

B Glycosidic bond

(deg) (deg)Dist O5i-O3i 1 (AEcirc )

Substrateyuml3 yuml 2 yuml78 126 315yuml2 yuml 1 yuml89 147 453yuml1 1 yuml44 145 3911 2 yuml75 120 308Product1 2 yuml74 113 2762 3 yuml84 138 331

The puckering parameters (QT 2) were calculated asdescribed37 dihedral angles are deregned as (O5i-C1i-O4i 1-C4i 1) and (C1i-O4i 1-C4i 1-C3i 1)

Glycosidase-Substrate Complex at Atomic Resolution 1067

plex should be a sequential process in which oneend of the oligosaccharide chain is free to rotatewhile the other binds to the cleft

Analysis of the oligosaccharide torsion anglesand ring puckering parameters reveals a signiregcantdistortion for the sugar ring bound at subsite yuml1This ring displays a strained boat 25B confor-mation whereas all other glucosyl residues (fromboth substrate and product) adopt the more favor-able 4C1 chair conformation (Table 2) The unfavor-able 25B conformation is stabilized by the tightanchoring of adjacent glucosyl residues within theenzyme cleft and by a bidentate hydrogen bondinginteraction of the carboxylate group of Asp152with the sugar OH-2 and OH-3 atoms as discussedabove (Figure 1(d)) Previous crystallographic stu-dies of other inverting5913 and retaining3031 glyco-sidases in complex with sugar ligands haddemonstrated that the pyranose ring at the point ofcleavage deviates from the standard 4C1 confor-mation although in some cases the limited resol-ution of those studies was not sufregcient todistinguish among closely related sugar ring con-formations or indeed from a mixture of them Toour knowledge the present example marks theregrst crystallographic observation at atomic resol-ution of a distorted boat conformation in a glycosi-dase-substrate complex

The observed substrate distortion has clearimplications for the hydrolytic mechanism Theplanarity of the atoms C5 O5 C1 and C2 for thesugar ring in the 25B conformation facilitates theformation of a partial double bond between O5and C1 during the hydrolytic reaction (Figure 4(c))

which is a requirement of inverting catalytic mech-anisms proceeding through a transition state withsubstantial oxocarbenium ion character1524 Inter-estingly a similar 25B conformation has beenobserved for the 2-macruoro-xylose residue bound inthe yuml1 subsite of the glycosyl-enzyme intermediateof a retaining family 11 xylanase31 and similararguments were forwarded concerning the planar-ity of C5 O5 C1 and C2 Therefore a parallel canbe established between the single step of theinverting CelA reaction and the second step (degly-cosylation) of the retaining xylanase reactionAlthough the two reactions have opposite stereo-chemical outcomes the formation and hydrolysisof the transition state is facilitated considerably inboth cases by the achievement of a 25B sugar con-formation at the initial state (enzyme-substratecomplex or glycosyl-enzyme intermediate)

Materials and Methods

Mutagenesis crystallisation data collectionand processing

Overlap extension of mutated PCR fragments wasused to generate the 488 bp PstI-BamHI segment ofpCT12817 bearing the E95Q mutation which wasexchanged against the corresponding wild-type segmentOverexpression puriregcation and crystallization ofCelAE95Q were performed as described32 A complete dif-fraction data set at 094 AEcirc resolution was collected at theBW7 beamline of EMBLDESY at Hamburg using a30 cm MarResearch Imaging Plate scanner and mono-chromatic wavelength (l 08815 AEcirc ) from a singlefrozen crystal at T 100 K Data reduction was carriedout with the computer programs DENZO andSCALEPACK33 Data collection statistics are summarizedin Table 1

Structure refinement

Crystallographic reregnement at atomic resolution wascarried out using the restrained conjugate gradient least-squares algorithm as implemented in the programSHELXL9734 Throughout the whole process 197438remacrections (representing 95 of the total data between10 and 094 AEcirc resolution) were used for reregnement and10391 remacrections were reserved for cross-validation anal-ysis using the free R-factor35 Manual modelling was car-ried out using the program O36 Stereochemical restraintswere applied during reregnement to bond lengths andangles chiral volumes and planar groups Additionalrestraints were applied to the temperature factors ofbonded atoms and to preclude large anisotropy of sol-vent molecules Non-structured solvent was modelledaccording to Babinets principle as implemented inSHELXL97

The unliganded structure of CelA reregned at 165 AEcirc

resolution17 was used as the starting model for reregne-ment The initial model included regve glucosyl residuesregtted manually to a preliminary (Fo yuml Fc) difference elec-tron density map A few cycles of reregnement were car-ried out assuming atomic isotropic temperature factorsfollowed by a long stage of several reregnement rounds(220 cycles) with anisotropic temperature factors for allnon-hydrogen atoms After each round (2Fo yuml Fc) and(Fo yuml Fc) electron density maps were inspected visually

1068 Glycosidase-Substrate Complex at Atomic Resolution

to model protein residues with alternate conformationsand to add or remove water molecules as requiredHydrogen atoms were included during the regrst reregne-ment rounds following the introduction of anisotropictemperature factors but were removed temporarilywhile modelling alternate side-chain conformations toavoid numerical instabilities Towards the end of reregne-ment stereochemical restraints were softened graduallyby duplicating their standard deviations from ideal tar-get values The regnal protein model includes 363 aminoacid residues 26 of which were modelled in alternateconformations Residues in alternate conformationsoccur mostly at the surface of the protein and in somecases split solvent sites could be attached at properhydrogen-bonding distances to side-chains displayingdouble conformations

Addition of solvent molecules was carried out usingthe automatic water divining procedure as implementedin SHELXL97 selecting from density peaks higher than04 eAEcirc 3 In parallel water molecules with equivalentisotropic U-factor values greater than 08 AEcirc 2 (B gt 63 AEcirc 2)were removed systematically from the atomic modeland the remaining solvent structure was checked manu-ally using stereochemical and electron density criteria Atotal of 401 solvent sites with full occupancy and 98additional solvent sites with partial occupancy (corre-sponding to split positions or related to disorderedamino acid side-chains) could be positioned unambigu-ously in the electron density map However the structur-al comparison of the solvent shell with that modelled forthe structure of wild-type CelA at 165 AEcirc suggests thatonly about half of the observed water molecules occupysimilar chemically equivalent positions in both struc-tures Solvent occupancies were not reregned except forsplit sites and water molecules coordinated to a proteinregion with multiple conformations The regnal parametersof the reregnement are shown in Table 1

The electron density is well deregned for the whole pro-tein model except for two short protein regions exposedto the solvent (residues 49-51 at the N-terminal end ofthe regrst a helix and residues 394-395 at the C terminus)Furthermore in well-structured regions of the proteindifference Fourier (Fo yuml Fc) maps (calculated from a pro-tein model including only heavy atoms) clearly indicatethe positions of missing hydrogen atoms as illustratedin Figure 5 for the hydrogen bonding interactionbetween Asp91 and His133

Figure 5 View of the 3Fo yuml 2Fc (blue) and Fo yuml Fc

(orange) electron density maps around His133

Protein Data Bank accession code

Atomic coordinates have been deposited with theRCSB Protein Data Bank with accession code 1KWF

Acknowledgements

This work has been supported by grants from InstitutPasteur and CNRS We thank the European Unionfor support under the TMRLSF program to theEMBL Hamburg Outstation reference number ERBFM-GECT980134 DMAG and MC are UNS-FOMECfellows

References

1 Blake C C Koenig D F Mair G A North A CPhillips D C amp Sarma V R (1965) Structure ofhen egg-white lysozyme A three-dimensional Four-ier synthesis at 2 AEcirc resolution Nature 206 757-761

2 Henrissat B amp Bairoch A (1996) Updating thesequence based classiregcation of glycosyl hydrolasesBiochem J 316 695-696

3 Claeyssens M amp Henrissat B (1992) Speciregcitymapping of cellulolytic enzymes classiregcation intofamilies of structurally related proteins conregrmed bybiochemical analysis Protein Sci 1 1293-1297

4 Juy M Souchon H Alzari P M Poljak R JClaeyssens M BeAcircguin P amp Aubert J P (1992)Three-dimensional structure of a thermostable bac-terial cellulase Nature 357 89-91

5 Aleshin A Golubev A Firsov L M amp HonzatkoR B (1992) Crystal structure of glucoamylase fromAspergillus awamori var X100 to 22 AEcirc resolutionJ Biol Chem 267 19291-19298

6 Mikami B Hehre E J Sato M Katsube YHirose M Morita Y amp Sacchettini J C (1993) The20 AEcirc resolution structure of soybean b-amylasecomplexed with a-cyclodextrin Biochemistry 326836-6845

7 Spezio M Wilson D B amp Karplus P A (1993)Crystal structure of the catalytic domain of a ther-mophilic endoglucanase Biochemistry 32 9906-9916

8 Davies G J Dodson G Moore M H Tolley S PDauter Z Wilson K S et al (1996) Structuredetermination and reregnement of the Humicola inso-lens endoglucanase V at 15 AEcirc resolution ActaCrystallog sect D 52 7-17

9 Tews I Perrakis A Oppenheim A Dauter ZWilson K S amp Vorgias C E (1996) Bacterial chito-biase structure provides insight into catalytic mech-anism and the basis of Tay-Sachs disease NatureStruct Biol 3 638-648

10 Sakon J Irwin D Wilson D B amp Karplus P A(1997) Structure and mechanism of endoexocellu-lase E4 from Thermomonospora fusca Nature StructBiol 4 810-818

11 Petersen T N Kauppinen S amp Larsen S (1997)The crystal structure of rhamnogalacturonase Afrom Aspergillus aculeatus a right-handed parallelb-helix Structure 5 533-544

12 Parsiegla G Juy M Reverbel-Leroy C Tardif CBelaich J P Driguez H amp Haser R (1998) Thecrystal structure of the processive endocellulaseCelF of Clostridium cellulolyticum in complex with a

1066 Glycosidase-Substrate Complex at Atomic Resolution

now conregrmed by the strong hydrogen bondinginteraction of Gln95 with the oxygen atom of thescissile glycosidic linkage (Figure 4(a)) In our pre-

Figure 4 Role of sugar distortion in catalysis (a)Local geometry of the catalytic center showing the gen-eral acid Glu95 (Gln95 here) and the putative generalbase (Asp278) catalysts the split position assigned tothe water nucleophile and the D-glucosyl residuesbound at subsites yuml1 and 1 The local geometry of thecatalytic center is further stabilized by the guanidiniumgroup of Arg281 which makes hydrogen bonding inter-actions with the two catalysts (b) Chiral volumes of thepyranose ring carbon atoms revealing a more planararrangement for the anomeric carbon C1 at subsite yuml1that could contribute to weaken the scissile glycosidiclinkage by steric distortion (c) Proposed reaction mech-anism for CelA The boat conformation of the sugar ringfavors a transition state with substantial oxocarbeniumion-like character

vious study we have identireged two aspartic acidresidues Asp152 and Asp278 as possible candi-dates for the base catalyst but a deregnitive assign-ment could not be done in the absence of boundsubstrate The atomic-resolution structure of thereaction center now reveals the nucleophilic watermolecule which is clearly visible in the electrondensity map (Figure 1(d)) facing the anomeric car-bon atom of the glucosyl residue at subsite yuml1This water molecule occupies two alternate pos-itions 09 AEcirc apart (Figure 4(a)) consistent with apartial sugar occupancy of the enzyme cleft Oneof the two solvent sites is sterically hindered by theanomeric carbon atom of the sugar ring (distance28 AEcirc ) and superimposes well with an equivalentwater molecule observed in the crystal structure ofunliganded CelA17 Besides its interaction with theanomeric carbon at subsite yuml1 the nucleophilicwater molecule is involved in strong hydrogenbonding interactions with the hydroxyl group ofTyr215 and the carboxylate group of Asp278(Figure 4(a)) suggesting that these two residuesplay a critical role in catalysis and that Asp278 isthe putative base catalyst in CelA Interestinglythe tyrosine residue is strictly conserved in allfamily 8 endoglucanases but the aspartic acid resi-due is missing from two homologous enzymesfrom Bacillus where it is replaced by an asparagineresidue2728 It follows that at least for some family8 glycosidases the enhancement of water nucleo-philicity by a carboxylate group might not berequired for the reaction to occur Alternativelythe conserved tyrosine residue could be acting as abase catalyst in these enzymes although thiswould require a substantial modiregcation of thetyrosine pKa to account for the enzyme activity atslightly acidic pH values

Substrate distortion

Comparison with the ligand-free structure ofwild-type CelA indicates that carbohydrate bind-ing induces small structural changes in the proteinmostly slight reorientations of aromatic and polarside-chains in contact with substrate In contraststacking and hydrogen bonding interactions withprotein residues impose a continuous torsionalstrain upon the substrate that modireges the confor-mation characteristic of b(14)-linked glucosyl resi-dues in cellulose (Figure 3(b)) signiregcantly A kinkis observed for the oligosaccharidic chain at thescissile glycosidic linkage which in turn imposes adistorted 25B conformation to the sugar ring in theyuml1 subsite (Table 2) Presumably full substrateoccupancy of available subsites (at least yuml3 to 2)is energetically required to stabilize the observedoligosaccharide conformation thus explaining whyCelA and other family 8 glycosidases cleave onlyoligosaccharides with a high degree of polymeriz-ation (DP 55)29 Assuming that the enzyme under-goes no major conformational change theobserved torsional stress on the substrate impliesthat the initial formation of the protein-sugar com-

Table 2 Oligosaccharide conformation

A Sugar ring2 (deg) (deg) QT (AEcirc )

Substrateyuml3 yuml35 9 058yuml2 yuml6 4 059yuml1 127 103 0541 yuml116 11 0622 40 6 058Product1 61 3 0592 19 4 0593 yuml87 10 066

B Glycosidic bond

(deg) (deg)Dist O5i-O3i 1 (AEcirc )

Substrateyuml3 yuml 2 yuml78 126 315yuml2 yuml 1 yuml89 147 453yuml1 1 yuml44 145 3911 2 yuml75 120 308Product1 2 yuml74 113 2762 3 yuml84 138 331

The puckering parameters (QT 2) were calculated asdescribed37 dihedral angles are deregned as (O5i-C1i-O4i 1-C4i 1) and (C1i-O4i 1-C4i 1-C3i 1)

Glycosidase-Substrate Complex at Atomic Resolution 1067

plex should be a sequential process in which oneend of the oligosaccharide chain is free to rotatewhile the other binds to the cleft

Analysis of the oligosaccharide torsion anglesand ring puckering parameters reveals a signiregcantdistortion for the sugar ring bound at subsite yuml1This ring displays a strained boat 25B confor-mation whereas all other glucosyl residues (fromboth substrate and product) adopt the more favor-able 4C1 chair conformation (Table 2) The unfavor-able 25B conformation is stabilized by the tightanchoring of adjacent glucosyl residues within theenzyme cleft and by a bidentate hydrogen bondinginteraction of the carboxylate group of Asp152with the sugar OH-2 and OH-3 atoms as discussedabove (Figure 1(d)) Previous crystallographic stu-dies of other inverting5913 and retaining3031 glyco-sidases in complex with sugar ligands haddemonstrated that the pyranose ring at the point ofcleavage deviates from the standard 4C1 confor-mation although in some cases the limited resol-ution of those studies was not sufregcient todistinguish among closely related sugar ring con-formations or indeed from a mixture of them Toour knowledge the present example marks theregrst crystallographic observation at atomic resol-ution of a distorted boat conformation in a glycosi-dase-substrate complex

The observed substrate distortion has clearimplications for the hydrolytic mechanism Theplanarity of the atoms C5 O5 C1 and C2 for thesugar ring in the 25B conformation facilitates theformation of a partial double bond between O5and C1 during the hydrolytic reaction (Figure 4(c))

which is a requirement of inverting catalytic mech-anisms proceeding through a transition state withsubstantial oxocarbenium ion character1524 Inter-estingly a similar 25B conformation has beenobserved for the 2-macruoro-xylose residue bound inthe yuml1 subsite of the glycosyl-enzyme intermediateof a retaining family 11 xylanase31 and similararguments were forwarded concerning the planar-ity of C5 O5 C1 and C2 Therefore a parallel canbe established between the single step of theinverting CelA reaction and the second step (degly-cosylation) of the retaining xylanase reactionAlthough the two reactions have opposite stereo-chemical outcomes the formation and hydrolysisof the transition state is facilitated considerably inboth cases by the achievement of a 25B sugar con-formation at the initial state (enzyme-substratecomplex or glycosyl-enzyme intermediate)

Materials and Methods

Mutagenesis crystallisation data collectionand processing

Overlap extension of mutated PCR fragments wasused to generate the 488 bp PstI-BamHI segment ofpCT12817 bearing the E95Q mutation which wasexchanged against the corresponding wild-type segmentOverexpression puriregcation and crystallization ofCelAE95Q were performed as described32 A complete dif-fraction data set at 094 AEcirc resolution was collected at theBW7 beamline of EMBLDESY at Hamburg using a30 cm MarResearch Imaging Plate scanner and mono-chromatic wavelength (l 08815 AEcirc ) from a singlefrozen crystal at T 100 K Data reduction was carriedout with the computer programs DENZO andSCALEPACK33 Data collection statistics are summarizedin Table 1

Structure refinement

Crystallographic reregnement at atomic resolution wascarried out using the restrained conjugate gradient least-squares algorithm as implemented in the programSHELXL9734 Throughout the whole process 197438remacrections (representing 95 of the total data between10 and 094 AEcirc resolution) were used for reregnement and10391 remacrections were reserved for cross-validation anal-ysis using the free R-factor35 Manual modelling was car-ried out using the program O36 Stereochemical restraintswere applied during reregnement to bond lengths andangles chiral volumes and planar groups Additionalrestraints were applied to the temperature factors ofbonded atoms and to preclude large anisotropy of sol-vent molecules Non-structured solvent was modelledaccording to Babinets principle as implemented inSHELXL97

The unliganded structure of CelA reregned at 165 AEcirc

resolution17 was used as the starting model for reregne-ment The initial model included regve glucosyl residuesregtted manually to a preliminary (Fo yuml Fc) difference elec-tron density map A few cycles of reregnement were car-ried out assuming atomic isotropic temperature factorsfollowed by a long stage of several reregnement rounds(220 cycles) with anisotropic temperature factors for allnon-hydrogen atoms After each round (2Fo yuml Fc) and(Fo yuml Fc) electron density maps were inspected visually

1068 Glycosidase-Substrate Complex at Atomic Resolution

to model protein residues with alternate conformationsand to add or remove water molecules as requiredHydrogen atoms were included during the regrst reregne-ment rounds following the introduction of anisotropictemperature factors but were removed temporarilywhile modelling alternate side-chain conformations toavoid numerical instabilities Towards the end of reregne-ment stereochemical restraints were softened graduallyby duplicating their standard deviations from ideal tar-get values The regnal protein model includes 363 aminoacid residues 26 of which were modelled in alternateconformations Residues in alternate conformationsoccur mostly at the surface of the protein and in somecases split solvent sites could be attached at properhydrogen-bonding distances to side-chains displayingdouble conformations

Addition of solvent molecules was carried out usingthe automatic water divining procedure as implementedin SHELXL97 selecting from density peaks higher than04 eAEcirc 3 In parallel water molecules with equivalentisotropic U-factor values greater than 08 AEcirc 2 (B gt 63 AEcirc 2)were removed systematically from the atomic modeland the remaining solvent structure was checked manu-ally using stereochemical and electron density criteria Atotal of 401 solvent sites with full occupancy and 98additional solvent sites with partial occupancy (corre-sponding to split positions or related to disorderedamino acid side-chains) could be positioned unambigu-ously in the electron density map However the structur-al comparison of the solvent shell with that modelled forthe structure of wild-type CelA at 165 AEcirc suggests thatonly about half of the observed water molecules occupysimilar chemically equivalent positions in both struc-tures Solvent occupancies were not reregned except forsplit sites and water molecules coordinated to a proteinregion with multiple conformations The regnal parametersof the reregnement are shown in Table 1

The electron density is well deregned for the whole pro-tein model except for two short protein regions exposedto the solvent (residues 49-51 at the N-terminal end ofthe regrst a helix and residues 394-395 at the C terminus)Furthermore in well-structured regions of the proteindifference Fourier (Fo yuml Fc) maps (calculated from a pro-tein model including only heavy atoms) clearly indicatethe positions of missing hydrogen atoms as illustratedin Figure 5 for the hydrogen bonding interactionbetween Asp91 and His133

Figure 5 View of the 3Fo yuml 2Fc (blue) and Fo yuml Fc

(orange) electron density maps around His133

Protein Data Bank accession code

Atomic coordinates have been deposited with theRCSB Protein Data Bank with accession code 1KWF

Acknowledgements

This work has been supported by grants from InstitutPasteur and CNRS We thank the European Unionfor support under the TMRLSF program to theEMBL Hamburg Outstation reference number ERBFM-GECT980134 DMAG and MC are UNS-FOMECfellows

References

1 Blake C C Koenig D F Mair G A North A CPhillips D C amp Sarma V R (1965) Structure ofhen egg-white lysozyme A three-dimensional Four-ier synthesis at 2 AEcirc resolution Nature 206 757-761

2 Henrissat B amp Bairoch A (1996) Updating thesequence based classiregcation of glycosyl hydrolasesBiochem J 316 695-696

3 Claeyssens M amp Henrissat B (1992) Speciregcitymapping of cellulolytic enzymes classiregcation intofamilies of structurally related proteins conregrmed bybiochemical analysis Protein Sci 1 1293-1297

4 Juy M Souchon H Alzari P M Poljak R JClaeyssens M BeAcircguin P amp Aubert J P (1992)Three-dimensional structure of a thermostable bac-terial cellulase Nature 357 89-91

5 Aleshin A Golubev A Firsov L M amp HonzatkoR B (1992) Crystal structure of glucoamylase fromAspergillus awamori var X100 to 22 AEcirc resolutionJ Biol Chem 267 19291-19298

6 Mikami B Hehre E J Sato M Katsube YHirose M Morita Y amp Sacchettini J C (1993) The20 AEcirc resolution structure of soybean b-amylasecomplexed with a-cyclodextrin Biochemistry 326836-6845

7 Spezio M Wilson D B amp Karplus P A (1993)Crystal structure of the catalytic domain of a ther-mophilic endoglucanase Biochemistry 32 9906-9916

8 Davies G J Dodson G Moore M H Tolley S PDauter Z Wilson K S et al (1996) Structuredetermination and reregnement of the Humicola inso-lens endoglucanase V at 15 AEcirc resolution ActaCrystallog sect D 52 7-17

9 Tews I Perrakis A Oppenheim A Dauter ZWilson K S amp Vorgias C E (1996) Bacterial chito-biase structure provides insight into catalytic mech-anism and the basis of Tay-Sachs disease NatureStruct Biol 3 638-648

10 Sakon J Irwin D Wilson D B amp Karplus P A(1997) Structure and mechanism of endoexocellu-lase E4 from Thermomonospora fusca Nature StructBiol 4 810-818

11 Petersen T N Kauppinen S amp Larsen S (1997)The crystal structure of rhamnogalacturonase Afrom Aspergillus aculeatus a right-handed parallelb-helix Structure 5 533-544

12 Parsiegla G Juy M Reverbel-Leroy C Tardif CBelaich J P Driguez H amp Haser R (1998) Thecrystal structure of the processive endocellulaseCelF of Clostridium cellulolyticum in complex with a

Table 2 Oligosaccharide conformation

A Sugar ring2 (deg) (deg) QT (AEcirc )

Substrateyuml3 yuml35 9 058yuml2 yuml6 4 059yuml1 127 103 0541 yuml116 11 0622 40 6 058Product1 61 3 0592 19 4 0593 yuml87 10 066

B Glycosidic bond

(deg) (deg)Dist O5i-O3i 1 (AEcirc )

Substrateyuml3 yuml 2 yuml78 126 315yuml2 yuml 1 yuml89 147 453yuml1 1 yuml44 145 3911 2 yuml75 120 308Product1 2 yuml74 113 2762 3 yuml84 138 331

The puckering parameters (QT 2) were calculated asdescribed37 dihedral angles are deregned as (O5i-C1i-O4i 1-C4i 1) and (C1i-O4i 1-C4i 1-C3i 1)

Glycosidase-Substrate Complex at Atomic Resolution 1067

plex should be a sequential process in which oneend of the oligosaccharide chain is free to rotatewhile the other binds to the cleft

Analysis of the oligosaccharide torsion anglesand ring puckering parameters reveals a signiregcantdistortion for the sugar ring bound at subsite yuml1This ring displays a strained boat 25B confor-mation whereas all other glucosyl residues (fromboth substrate and product) adopt the more favor-able 4C1 chair conformation (Table 2) The unfavor-able 25B conformation is stabilized by the tightanchoring of adjacent glucosyl residues within theenzyme cleft and by a bidentate hydrogen bondinginteraction of the carboxylate group of Asp152with the sugar OH-2 and OH-3 atoms as discussedabove (Figure 1(d)) Previous crystallographic stu-dies of other inverting5913 and retaining3031 glyco-sidases in complex with sugar ligands haddemonstrated that the pyranose ring at the point ofcleavage deviates from the standard 4C1 confor-mation although in some cases the limited resol-ution of those studies was not sufregcient todistinguish among closely related sugar ring con-formations or indeed from a mixture of them Toour knowledge the present example marks theregrst crystallographic observation at atomic resol-ution of a distorted boat conformation in a glycosi-dase-substrate complex

The observed substrate distortion has clearimplications for the hydrolytic mechanism Theplanarity of the atoms C5 O5 C1 and C2 for thesugar ring in the 25B conformation facilitates theformation of a partial double bond between O5and C1 during the hydrolytic reaction (Figure 4(c))

which is a requirement of inverting catalytic mech-anisms proceeding through a transition state withsubstantial oxocarbenium ion character1524 Inter-estingly a similar 25B conformation has beenobserved for the 2-macruoro-xylose residue bound inthe yuml1 subsite of the glycosyl-enzyme intermediateof a retaining family 11 xylanase31 and similararguments were forwarded concerning the planar-ity of C5 O5 C1 and C2 Therefore a parallel canbe established between the single step of theinverting CelA reaction and the second step (degly-cosylation) of the retaining xylanase reactionAlthough the two reactions have opposite stereo-chemical outcomes the formation and hydrolysisof the transition state is facilitated considerably inboth cases by the achievement of a 25B sugar con-formation at the initial state (enzyme-substratecomplex or glycosyl-enzyme intermediate)

Materials and Methods

Mutagenesis crystallisation data collectionand processing

Overlap extension of mutated PCR fragments wasused to generate the 488 bp PstI-BamHI segment ofpCT12817 bearing the E95Q mutation which wasexchanged against the corresponding wild-type segmentOverexpression puriregcation and crystallization ofCelAE95Q were performed as described32 A complete dif-fraction data set at 094 AEcirc resolution was collected at theBW7 beamline of EMBLDESY at Hamburg using a30 cm MarResearch Imaging Plate scanner and mono-chromatic wavelength (l 08815 AEcirc ) from a singlefrozen crystal at T 100 K Data reduction was carriedout with the computer programs DENZO andSCALEPACK33 Data collection statistics are summarizedin Table 1

Structure refinement

Crystallographic reregnement at atomic resolution wascarried out using the restrained conjugate gradient least-squares algorithm as implemented in the programSHELXL9734 Throughout the whole process 197438remacrections (representing 95 of the total data between10 and 094 AEcirc resolution) were used for reregnement and10391 remacrections were reserved for cross-validation anal-ysis using the free R-factor35 Manual modelling was car-ried out using the program O36 Stereochemical restraintswere applied during reregnement to bond lengths andangles chiral volumes and planar groups Additionalrestraints were applied to the temperature factors ofbonded atoms and to preclude large anisotropy of sol-vent molecules Non-structured solvent was modelledaccording to Babinets principle as implemented inSHELXL97

The unliganded structure of CelA reregned at 165 AEcirc

resolution17 was used as the starting model for reregne-ment The initial model included regve glucosyl residuesregtted manually to a preliminary (Fo yuml Fc) difference elec-tron density map A few cycles of reregnement were car-ried out assuming atomic isotropic temperature factorsfollowed by a long stage of several reregnement rounds(220 cycles) with anisotropic temperature factors for allnon-hydrogen atoms After each round (2Fo yuml Fc) and(Fo yuml Fc) electron density maps were inspected visually

1068 Glycosidase-Substrate Complex at Atomic Resolution

to model protein residues with alternate conformationsand to add or remove water molecules as requiredHydrogen atoms were included during the regrst reregne-ment rounds following the introduction of anisotropictemperature factors but were removed temporarilywhile modelling alternate side-chain conformations toavoid numerical instabilities Towards the end of reregne-ment stereochemical restraints were softened graduallyby duplicating their standard deviations from ideal tar-get values The regnal protein model includes 363 aminoacid residues 26 of which were modelled in alternateconformations Residues in alternate conformationsoccur mostly at the surface of the protein and in somecases split solvent sites could be attached at properhydrogen-bonding distances to side-chains displayingdouble conformations

Addition of solvent molecules was carried out usingthe automatic water divining procedure as implementedin SHELXL97 selecting from density peaks higher than04 eAEcirc 3 In parallel water molecules with equivalentisotropic U-factor values greater than 08 AEcirc 2 (B gt 63 AEcirc 2)were removed systematically from the atomic modeland the remaining solvent structure was checked manu-ally using stereochemical and electron density criteria Atotal of 401 solvent sites with full occupancy and 98additional solvent sites with partial occupancy (corre-sponding to split positions or related to disorderedamino acid side-chains) could be positioned unambigu-ously in the electron density map However the structur-al comparison of the solvent shell with that modelled forthe structure of wild-type CelA at 165 AEcirc suggests thatonly about half of the observed water molecules occupysimilar chemically equivalent positions in both struc-tures Solvent occupancies were not reregned except forsplit sites and water molecules coordinated to a proteinregion with multiple conformations The regnal parametersof the reregnement are shown in Table 1

The electron density is well deregned for the whole pro-tein model except for two short protein regions exposedto the solvent (residues 49-51 at the N-terminal end ofthe regrst a helix and residues 394-395 at the C terminus)Furthermore in well-structured regions of the proteindifference Fourier (Fo yuml Fc) maps (calculated from a pro-tein model including only heavy atoms) clearly indicatethe positions of missing hydrogen atoms as illustratedin Figure 5 for the hydrogen bonding interactionbetween Asp91 and His133

Figure 5 View of the 3Fo yuml 2Fc (blue) and Fo yuml Fc

(orange) electron density maps around His133

Protein Data Bank accession code

Atomic coordinates have been deposited with theRCSB Protein Data Bank with accession code 1KWF

Acknowledgements

This work has been supported by grants from InstitutPasteur and CNRS We thank the European Unionfor support under the TMRLSF program to theEMBL Hamburg Outstation reference number ERBFM-GECT980134 DMAG and MC are UNS-FOMECfellows

References

1 Blake C C Koenig D F Mair G A North A CPhillips D C amp Sarma V R (1965) Structure ofhen egg-white lysozyme A three-dimensional Four-ier synthesis at 2 AEcirc resolution Nature 206 757-761

2 Henrissat B amp Bairoch A (1996) Updating thesequence based classiregcation of glycosyl hydrolasesBiochem J 316 695-696

3 Claeyssens M amp Henrissat B (1992) Speciregcitymapping of cellulolytic enzymes classiregcation intofamilies of structurally related proteins conregrmed bybiochemical analysis Protein Sci 1 1293-1297

4 Juy M Souchon H Alzari P M Poljak R JClaeyssens M BeAcircguin P amp Aubert J P (1992)Three-dimensional structure of a thermostable bac-terial cellulase Nature 357 89-91

5 Aleshin A Golubev A Firsov L M amp HonzatkoR B (1992) Crystal structure of glucoamylase fromAspergillus awamori var X100 to 22 AEcirc resolutionJ Biol Chem 267 19291-19298

6 Mikami B Hehre E J Sato M Katsube YHirose M Morita Y amp Sacchettini J C (1993) The20 AEcirc resolution structure of soybean b-amylasecomplexed with a-cyclodextrin Biochemistry 326836-6845

7 Spezio M Wilson D B amp Karplus P A (1993)Crystal structure of the catalytic domain of a ther-mophilic endoglucanase Biochemistry 32 9906-9916

8 Davies G J Dodson G Moore M H Tolley S PDauter Z Wilson K S et al (1996) Structuredetermination and reregnement of the Humicola inso-lens endoglucanase V at 15 AEcirc resolution ActaCrystallog sect D 52 7-17

9 Tews I Perrakis A Oppenheim A Dauter ZWilson K S amp Vorgias C E (1996) Bacterial chito-biase structure provides insight into catalytic mech-anism and the basis of Tay-Sachs disease NatureStruct Biol 3 638-648

10 Sakon J Irwin D Wilson D B amp Karplus P A(1997) Structure and mechanism of endoexocellu-lase E4 from Thermomonospora fusca Nature StructBiol 4 810-818

11 Petersen T N Kauppinen S amp Larsen S (1997)The crystal structure of rhamnogalacturonase Afrom Aspergillus aculeatus a right-handed parallelb-helix Structure 5 533-544

12 Parsiegla G Juy M Reverbel-Leroy C Tardif CBelaich J P Driguez H amp Haser R (1998) Thecrystal structure of the processive endocellulaseCelF of Clostridium cellulolyticum in complex with a

1068 Glycosidase-Substrate Complex at Atomic Resolution

to model protein residues with alternate conformationsand to add or remove water molecules as requiredHydrogen atoms were included during the regrst reregne-ment rounds following the introduction of anisotropictemperature factors but were removed temporarilywhile modelling alternate side-chain conformations toavoid numerical instabilities Towards the end of reregne-ment stereochemical restraints were softened graduallyby duplicating their standard deviations from ideal tar-get values The regnal protein model includes 363 aminoacid residues 26 of which were modelled in alternateconformations Residues in alternate conformationsoccur mostly at the surface of the protein and in somecases split solvent sites could be attached at properhydrogen-bonding distances to side-chains displayingdouble conformations

Addition of solvent molecules was carried out usingthe automatic water divining procedure as implementedin SHELXL97 selecting from density peaks higher than04 eAEcirc 3 In parallel water molecules with equivalentisotropic U-factor values greater than 08 AEcirc 2 (B gt 63 AEcirc 2)were removed systematically from the atomic modeland the remaining solvent structure was checked manu-ally using stereochemical and electron density criteria Atotal of 401 solvent sites with full occupancy and 98additional solvent sites with partial occupancy (corre-sponding to split positions or related to disorderedamino acid side-chains) could be positioned unambigu-ously in the electron density map However the structur-al comparison of the solvent shell with that modelled forthe structure of wild-type CelA at 165 AEcirc suggests thatonly about half of the observed water molecules occupysimilar chemically equivalent positions in both struc-tures Solvent occupancies were not reregned except forsplit sites and water molecules coordinated to a proteinregion with multiple conformations The regnal parametersof the reregnement are shown in Table 1

The electron density is well deregned for the whole pro-tein model except for two short protein regions exposedto the solvent (residues 49-51 at the N-terminal end ofthe regrst a helix and residues 394-395 at the C terminus)Furthermore in well-structured regions of the proteindifference Fourier (Fo yuml Fc) maps (calculated from a pro-tein model including only heavy atoms) clearly indicatethe positions of missing hydrogen atoms as illustratedin Figure 5 for the hydrogen bonding interactionbetween Asp91 and His133

Figure 5 View of the 3Fo yuml 2Fc (blue) and Fo yuml Fc

(orange) electron density maps around His133

Protein Data Bank accession code

Atomic coordinates have been deposited with theRCSB Protein Data Bank with accession code 1KWF

Acknowledgements

This work has been supported by grants from InstitutPasteur and CNRS We thank the European Unionfor support under the TMRLSF program to theEMBL Hamburg Outstation reference number ERBFM-GECT980134 DMAG and MC are UNS-FOMECfellows

References

1 Blake C C Koenig D F Mair G A North A CPhillips D C amp Sarma V R (1965) Structure ofhen egg-white lysozyme A three-dimensional Four-ier synthesis at 2 AEcirc resolution Nature 206 757-761

2 Henrissat B amp Bairoch A (1996) Updating thesequence based classiregcation of glycosyl hydrolasesBiochem J 316 695-696

3 Claeyssens M amp Henrissat B (1992) Speciregcitymapping of cellulolytic enzymes classiregcation intofamilies of structurally related proteins conregrmed bybiochemical analysis Protein Sci 1 1293-1297

4 Juy M Souchon H Alzari P M Poljak R JClaeyssens M BeAcircguin P amp Aubert J P (1992)Three-dimensional structure of a thermostable bac-terial cellulase Nature 357 89-91

5 Aleshin A Golubev A Firsov L M amp HonzatkoR B (1992) Crystal structure of glucoamylase fromAspergillus awamori var X100 to 22 AEcirc resolutionJ Biol Chem 267 19291-19298

6 Mikami B Hehre E J Sato M Katsube YHirose M Morita Y amp Sacchettini J C (1993) The20 AEcirc resolution structure of soybean b-amylasecomplexed with a-cyclodextrin Biochemistry 326836-6845

7 Spezio M Wilson D B amp Karplus P A (1993)Crystal structure of the catalytic domain of a ther-mophilic endoglucanase Biochemistry 32 9906-9916

8 Davies G J Dodson G Moore M H Tolley S PDauter Z Wilson K S et al (1996) Structuredetermination and reregnement of the Humicola inso-lens endoglucanase V at 15 AEcirc resolution ActaCrystallog sect D 52 7-17

9 Tews I Perrakis A Oppenheim A Dauter ZWilson K S amp Vorgias C E (1996) Bacterial chito-biase structure provides insight into catalytic mech-anism and the basis of Tay-Sachs disease NatureStruct Biol 3 638-648

10 Sakon J Irwin D Wilson D B amp Karplus P A(1997) Structure and mechanism of endoexocellu-lase E4 from Thermomonospora fusca Nature StructBiol 4 810-818

11 Petersen T N Kauppinen S amp Larsen S (1997)The crystal structure of rhamnogalacturonase Afrom Aspergillus aculeatus a right-handed parallelb-helix Structure 5 533-544

12 Parsiegla G Juy M Reverbel-Leroy C Tardif CBelaich J P Driguez H amp Haser R (1998) Thecrystal structure of the processive endocellulaseCelF of Clostridium cellulolyticum in complex with a

Glycosidase-Substrate Complex at Atomic Resolution 1069

thiooligosaccharide inhibitor at 20 AEcirc resolutionEMBO J 17 5551-5562

13 Zou J Kleywegt G J Stahlberg J Driguez HNerinckx W Claeyssens M et al (1999) Crystallo-graphic evidence for substrate ring distortion andprotein conformational changes during catalysis incellobiohydrolase Cel6A from Trichoderma reeseiStruct Fold Des 7 1035-1045

14 Koshland D E Jr (1953) Stereochemistry and themechanism of enzymatic reactions Biol Rev 28416-436

15 Ly H D amp Withers S G (1999) Mutagenesis ofglycosidases Annu Rev Biochem 68 487-522

16 BeAcircguin P amp Alzari P M (1998) The cellulosome ofClostridium thermocellum Biochem Soc Trans 26 178-185

17 Alzari P M Souchon H amp Dominguez R (1996)The crystal structure of endoglucanase CelA afamily 8 glycosyl hydrolase from Clostridium thermo-cellum Structure 4 265-275

18 Longhi S Czjzek M amp Cambillau C (1998) Mess-ages from ultrahigh resolution crystal structuresCurr Opin Struct Biol 8 730-737

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

20 Engh R A amp Huber R (1991) Accurate bond andangle parameters for X-ray protein structure reregne-ment Acta Crystallog sect A 47 392-400

21 Davies G J Wilson K S amp Henrissat B (1997)Nomenclature for sugar-binding subsites in glycosylhydrolases Biochem J 321 557-559

22 Franck R W (1992) The mechanism of b-glycosi-dases a reassessment of some seminal papersBioorg Chem 20 77-88

23 Karplus M amp Post C B (1996) Simulations oflysozyme internal motions and the reaction mech-anism Experientia 75(Suppl) 111-141

24 Sinnott M L (1990) Catalytic mechanisms of enzy-mic glycosyl transfer Chem Rev 90 1171-1202

25 Sinnott M L (1993) There is no experimental evi-dence for endocyclic cleavage in the action of e e(retaining b) glycopyranosidases and much againstit Bioorg Chem 21 34-40

26 Brzozowski A M amp Davies G J (1997) Structureof the Aspergillus oryzae a-amylase complexed withthe inhibitor acarbose at 20 AEcirc resolution Biochemis-try 36 10837-10845

27 Bueno A Vazquez de Aldana C R Correa J ampdel Rey F (1990) Nucleotide sequence of a 13-14-beta-glucanase-encoding gene in Bacillus circulansWL-12 Nucl Acids Res 18 4248

28 Ozaki K Sumitomo N amp Ito S (1991) Molecularcloning and nucleotide sequence of the gene encod-ing an endo-14-beta-glucanase from Bacillus spKSM-330 J Gen Microbiol 137 2299-2305

29 PeAcirctreAcirc J Longin R amp Millet J (1981) Puriregcationand properties of an endo-b-14-glucanase fromClostridium thermocellum Biochimie 63 629-639

30 Sulzenbacher G Driguez H Henrissat BSchuEgrave lein M amp Davies G J (1996) Structure of theFusarium oxysporum endoglucanase I with a nonhy-drolyzable substrate analogue substrate distortiongives rise to the preferred axial orientation for theleaving group Biochemistry 35 15280-15287

31 Sidhu G Withers S G Nguyen N T McIntoshL P Ziser L amp Brayer G D (1999) Sugar ringdistortion in the glycosyl-enzyme intermediate of afamily G11 xylanase Biochemistry 38 5346-5354

32 Souchon H BeAcircguin P amp Alzari P M (1996) Crys-tallization of a family 8 cellulase from Clostridiumthermocellum Proteins Struct Funct Genet 25 134-136

33 Otwinowski Z amp Minor W (1997) Processing ofX-ray diffraction data collected in oscillation modeMethods Enzymol 276 307-326

34 Sheldrick G M amp Schneider T R (1997) SHELXLhigh resolution reregnement Methods Enzymol 277319-343

35 BruEgrave nger A T (1997) Free R value cross-validationin crystallography Methods Enzymol 277 366-396

36 Jones T A Zou J Y Cowan S W amp KjeldgaardM (1991) Improved methods for binding proteinmodels in electron density maps and the location oferrors in these models Acta Crystallog sect A 47110-119

37 Cremer D amp Pople J A (1975) A general dereg-nition of ring puckering coordinates J Am ChemSoc 97 1354-1358

Edited by R Huber

(Received 28 September 2001 received in revised form 21 December 2001 accepted 27 December 2001)