letters

584 nature structural biology • volume 8 number 7 • july 2001

Structure of the Rho-activating domain ofEscherichia coli cytotoxicnecrotizing factor 1Lori Buetow1, Gilles Flatau2, Katy Chiu1, Patrice Boquet2

and Partho Ghosh1

1Department of Chemistry and Biochemistry, University of California at SanDiego, La Jolla, California 92093-0314, USA. 2INSERM U452, Faculté deMédecine, 28 Avenue de Valombrose, 06107 Nice Cedex 2, France.

Certain uropathogenic and neonatal meningitis-causingstrains of Escherichia coli express a 114 kDa protein toxincalled cytotoxic necrotizing factor 1 (CNF1). The toxin causesalteration of the host cell actin cytoskeleton and promotesbacterial invasion of blood-brain barrier endothelial cells.CNF1 belongs to a unique group of large cytotoxins that causeconstitutive activation of Rho guanosine triphosphatases(GTPases), which are key regulators of the actin cytoskeleton.This group also includes E. coli cytotoxic necrotizing factor 2(CNF2, 114 kDa) and dermonecrotic toxins (DNT, 159 kDa) ofBordetella spp. with related sequences occuring in Yersiniaspp. Here we show that the catalytic region of CNF1 exhibits anovel protein fold as determined by its 1.83 Å resolution crys-tal structure. The structure reveals that CNF1 has a Cys-His-main chain oxygen catalytic triad reminiscent of enzymesbelonging to the catalytic triad superfamily. The position ofthe catalytic Cys residue at the base of a deep pocket restrictsaccess to potential substrates and helps explain the high speci-ficity of this and related toxins.

Cytotoxic necrotizing factor 1 (CNF1) is a 114 kDa toxinexpressed by certain extraintestinal strains of pathogenicEschericia coli, including those that cause urinary tract infectionsand neonatal meningitis1,2. A role for CNF1 in virulence is sup-

ported by epidemiological evidence2 and by experiments usingisogenic CNF1 knockout strains of E. coli in animal models ofneonatal meningitis (K.S. Kim, pers. comm.) and urinary tractinfection32. In most cell lines, constitutive activation of Rho, asmall GTPase that regulates cytoskeletal reorganization and geneexpression, by CNF1 leads to formation of stress fibers and focaladhesions, membrane ruffling and inhibition of cytokinesis withsubsequent multinucleation1. Prolonged exposure of nonphago-cytic mammalian cells to CNF1 also promotes intracellular inva-sion by E. coli3, a process that may be related to crossingblood-brain barrier endothelial cells in meningitis. Further-more, CNF1 at high concentration promotes apoptosis in blad-der epithelial cells, which may result in allowing E. coli access todeeper tissues in urinary tract infections4.

CNF1, CNF2 and the Bordetella dermonecrotic toxin (DNT)activate Rho family members constitutively through site-specificdeamidation of a Gln residue to Glu (Gln 63 in RhoA, and Gln 61in Rac and Cdc42)5,6. This glutamine is implicated in stabilizingthe transition state and activating a water molecule during GTPhydrolysis. Site-specific deamidation is a rare enzymatic activityfound in the CheB bacterial signaling pathway7 and is implicatedin the regulation of mammalian protein kinase A8. Site-specificdeamidation of Rho by toxins greatly impairs the ability of Rhoto hydrolyze GTP, with or without the assistance of GTPase acti-vating proteins, and, therefore, places Rho in a constitutivelyactivated state. These toxins, highly specific for Rho proteins, canalso constitutively activate Rho by transglutamination of thesame glutamine residue in a reaction chemically similar todeamidation9,10. In contrast to mammalian transglutaminasesthat modify a broad spectrum of targets, CNF1 acts only onmembers of the Rho family9. In addition, whereas mammaliantissue transglutaminase modifies three of the five RhoA gluta-mines, CNF1 is specific to Gln 63 (ref. 9).

We report here the X-ray crystal structure of the catalyticregion of CNF1 (residues 720–1,014, called CNF1-C), whichwhen introduced into target cells is sufficient to cause morpho-logical changes characteristic of intact CNF1 (ref. 9). The struc-ture has a novel protein fold and an active site reminiscent of the

Fig. 1 Structure of CNF1-C. a, Cα trace in stereo view. b, Ribbons repre-sentation (α-helices are blue, β-strands red and coils green). c, Topologyrepresentation (α-helices as rectangles and β-strands as arrows) with thecatalytic groups (Cys 866, His 881 and main chain carbonyl of Val 833)indicated. (a,b) was created with Molscript29.

a b

c

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://s

tru

ctb

io.n

atu

re.c

om

© 2001 Nature Publishing Group http://structbio.nature.com

letters

nature structural biology • volume 8 number 7 • july 2001 585

catalytic triad enzyme superfamily. The location of the active sitein a deep pocket explains how the toxin excludes random sub-strates and indicates that a conformational change is likely to berequired for Rho modification.

Catalytic region exhibits a novel foldThe catalytic region of CNF1 (32 kDa) was expressed in E. coliand purified. Crystals grown from the purified protein diffractedto at least 1.76 Å resolution, and the structure was determined bymultiwavelength anomalous dispersion (MAD) using a quadru-ple methionine substitution mutant of CNF1-C (see Methods)and refined to 1.83 Å resolution (Table 1). DNA sequencingidentified an inadvertent Pro to Leu substitution at position 794introduced during cloning by the polymerase chain reaction.The mutation is functionally and structurally inconsequential;the residue in DNT corresponding to Leu 794 is Pro, and L794Pis as active as wild type CNF1-C (see below). Additionally, themutation occurs on a surface-exposed, flexible loop regionrather than within the structural core. Herein, CNF1-C refers tothe L794P isoform unless wild type is explicitly denoted.

CNF1-C forms a single compact domain composed of a cen-tral β-sandwich surrounded by helices and extensive loopregions (Fig. 1). The β-sandwich is formed by two mixedβ-sheets (one containing six and the other seven strands) thatform a previously undiscovered pattern of complex interconnec-tions. Curiously, a buried lysine (Lys 879) and two waters arefound in the hydrophobic core of the β-sandwich. There is nocompensating negative charge for Lys 879, which instead formshydrogen bonds with main and side chain oxygens. Lys 879 is notconserved among CNF2 and DNT (Fig. 2), and functional rele-vance for the buried residue is uncertain. Searches of databasesreveal no structural homology between CNF1-C and other pro-tein structures, and, likewise, the sequence of CNF1-C is unrelat-ed to any except those of CNF2, DNT and the putative YersiniaCNF-like toxins. Given the sequence and functional similaritiesof these toxins, it is likely that their catalytic domains constitute aunique family that shares the CNF1-C fold.

Catalytic triad at the active siteResidues Cys 866 and His 881 are conserved among the toxinsand are required for activity9. The positions of these two residues

resemble the active site of enzymesbelonging to the catalytic triad superfami-ly (Fig. 3a,b). The superfamily includes alarge number of structurally unrelated butmechanistically similar enzymes, includ-ing blood coagulation factor XIII (ref. 11)and GMP synthetase12, which both deami-date glutamine as part of their activity.Factor XIII (a transglutaminase) possessesa Cys-His-Asp triad and GMP synthetase(an amidotransferase) a Cys-His-Glu

triad. In these enzymes, the Cys-His pair is thought to exist as athiolate–imidazolium ion pair, with the thiolate acting as theattacking nucleophile11,12.

The orientation of Cys 866 in CNF1 supports the view that theresidue existing in its thiolate form is responsible for nucleophilicattack on the δ-carboxamide of RhoA Gln 63. Two conformersare observed for Cys 866 (denoted ‘1’ and ‘2’ in Fig. 3a,c), withconformer 1 being dominant (Fig. 3, legend). As expected for ionpair formation13, Sγ of conformer 1 is coplanar to the imidazolering of His 881 and forms a close hydrogen bond (3.0 Å) to the Nεatom. Conformer 2 is more surface-accessible than conformer 1;Cys 886 in this conformer is too distant (4.2 Å) to form a hydro-gen bond to His 881. Conformer 2 is closer to the expected posi-tion of incoming RhoA, but its role in catalysis is not yet clear.The third member of the catalytic triad usually plays a role in ori-enting the His imidazole ring through a hydrogen bond. UnlikeFactor XIII and GMP synthetase, CNF1 does not have a Glu orAsp residue to orient the His imidazole ring. Instead, the mainchain carbonyl of Val 833 provides a hydrogen bond (2.8 Å) to Nδof His 881. The involvement of a main chain atom is unusual for acatalytic triad but has been observed14–16.

CNF1 also has an unusual structural element orienting thenucleophilic Cys. In Factor XIII and GMP synthetase, the loca-tion of the Cys at the N-terminus of an α-helix is thought to aidin the stabilization of the thiolate through the helix dipole11,12. Incontrast, Cys 866 of CNF1 is positioned on a short loop betweenβ-strands 6 and 7 (Fig. 3a). This highly structured loop appearsto be stabilized by the conserved residue Tyr 962, which formshydrogen bonds to the main chain of the loop. Despite these dif-ferences, key atoms of the triads (Cys Sγ, His imidazole nitro-gens, and main or side chain oxygen) superimpose among theseenzymes remarkably well (Fig. 3b) and provide evidence for con-vergent evolution in catalytic mechanism.

Substitution of Cys 866 with Ser renders CNF1 inactive9. Thecatalytically inactive C866S mutant of CNF1-C (Table 1) isessentially indistinguishable in structure from catalytically activeCNF1-C (root mean square (r.m.s.) deviation 0.16 Å Cα, datanot shown). The only slight difference is that Ser 866, whichforms a hydrogen bond (2.7 Å) to His 881, occupies conformer 1but not conformer 2. Stereoelectronic effects of the active siteappear to promote the catalytic ability of Cys over Ser.

Fig. 2 Sequence alignment of CNF1, CNF2 andDNT catalytic regions. Conserved and conserva-tively substituted residues (with common chem-ical properties) are in red. Secondary structuralelements and catalytic residues are displayedabove the sequences. Lines above the sequencedelineate loop regions that surround the cat-alytic pocket. Sequences were aligned usingCLUSTALW30 and adjusted based on inspectionof the CNF1-C structure. CNF1-C is 84% and22% identical to equivalent regions of CNF2and DNT, respectively.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://s

tru

ctb

io.n

atu

re.c

om

© 2001 Nature Publishing Group http://structbio.nature.com

letters

586 nature structural biology • volume 8 number 7 • july 2001

Structure-based mutagenesisResidues in the active site pocket that are conserved among thetoxins were mutated to determine their contribution to activity(Figs 2, 4). Asn 835 and Ser 864, which bracket Cys 866 (Figs 3and 4), were individually mutated to alanines. The activity ofthese mutants was assessed by taking advantage of the observa-tion that deamidation of RhoA Gln 63 leads to decreased elec-trophoretic mobility on denaturing polyacrylamide gels. A timecourse of deamidation shows that wild type and L794P (the iso-form used for structural determination) are equally active(Fig. 4b). As reported, substitution of Cys 866 by Ser completelyabolishes activity9. By comparison, Ala substitution of Asn 835greatly reduces the catalytic rate but does not abolish activity,and Ala substitution of Ser 864 results in a small but distinguish-able decrease in catalytic rate. Mutations of conserved residueson the face of CNF1 surrounding the active site pocket weredesigned to identify potential interactions with Rho. However,Ala substitution of Glu 943 or Asn 966, or Met substitution ofLeu 769 fail to show any effect on activity (Fig. 4b).

Comparison to DNTAlthough CNF1 and DNT share common specificities, theydiffer in that DNT is preferentially a transglutaminase, where-as CNF1 is preferentially a deamidase5,6,9,10. In transglutamina-tion, a primary amine rather than a hydroxyl replaces the sidechain amide of glutamine. Modeling of DNT suggests a possi-ble explanation for this preference. In contrast to the electro-statically nondescript catalytic pocket of CNF1, the DNTpocket is predicted to carry a substantial negative charge thatmay attract positively charged primary amines. DNT also dif-fers from CNF1 in preferentially acting on GDP-bound RhoA,whereas CNF1 acts equally on GDP- and GTP-bound RhoA10.No obvious explanation for this preference is evident from thestructure. DNT possesses a sequence, 1,304-AFYHTGK-1,310,that has been speculated to form a nucleotide binding P-loop17. The equivalent sequence in CNF1, 878-YKVHTGT-884, forms a β-strand rather than a canonical P-loop. Theseregions of CNF1, CNF2 and DNT contain the catalytic His andtherefore are unlikely to differ in structure, suggesting that the

a b

c

a b

Fig. 3 Catalytic triad. a, CNF1-C catalytic triad pocket.Cys 866 has two conformations, indicated by an arrowbetween ‘1’ and ‘2’. The likely direction of the approachof RhoA to Gln 63 is indicated. Tyr 962 forms hydrogenbonds to main chain amide of Gly 865, and main chainamide and carbonyl of Cys 866. The loop on whichCys 866 is located is well ordered (average Cα B-factor16.8 Å2). Side chains and part of the main chain (residues865–866) are shown as bonds (carbon is yellow; nitro-gen, blue; oxygen, red; and sulfur, green). The gray rib-bon represents a Cα trace. b, Catalytic triads of CNF1-C(main chain carbonyl of Val 833, His 881 and Cys 866),Factor XIII (Asp 396, His 373 and Cys 314) and GMP syn-thetase (Glu 183, His 181 and Cys 86). c, Stereo view of a2Fo – Fc simulated anneal omit electron density map(atoms within 5 Å of Cys 866, inclusive, were omitted) ofthe active site contoured at 1.3 σ with the final model ofCNF1-C superimposed. Cys 866 conformers are indicatedas ‘1’ and ‘2’. Conformers 1 and 2 were assigned occu-pancies of 0.5 and were found to refine to B-factors of11.7 Å2 and 42.8 Å2, respectively. Figs 3a,b and 4a werecreated with MidasPlus31 and Fig. 3c with O26.

Fig. 4 Surface of CNF1-C pocket and activity of mutants. a, Molecular surface representation of CNF1-C. The view is of the ‘top’ of the molecule(related to the view in Fig. 1b by a rotation 90° towards the viewer along a horizontal axis). Surface-accessible residues conserved among DNT, CNF2and CNF1 are labeled. b, Quantitative analysis of catalytic activity. Percent product (RhoA Glu 63) formed as a function of time by wild type andmutants of CNF1-C is plotted. At 3 h, deamidation of RhoA is complete for all CNF1-C constructs except C866S. Error bars reflect standard deviations.All CNF1-C proteins except wild type encode L794P in addition to the indicated mutation.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://s

tru

ctb

io.n

atu

re.c

om

© 2001 Nature Publishing Group http://structbio.nature.com

letters

nature structural biology • volume 8 number 7 • july 2001 587

sequence in DNT is unlikely to provide a P-loop nucleotidebinding site.

Target specificityCys 866 is located at the base of a narrow (15 × 30 Å2) and deep(10 Å) pocket whose dimensions probably exclude randomsubstrates from access to the active site (Fig. 4a). However, thepocket is also too small to accommodate RhoA unless confor-mational changes occur during binding. A conformationalchange in one or both proteins is plausible. The switch IIregion of RhoA that contains Gln 63 and is involved in speci-ficity18,19 is highly flexible, as are the CNF1-C loops that sur-round the catalytic pocket. These CNF1 loops have the highestB-factors in the structure. A conformational change may notonly be involved in catalysis but also in specificity. This possi-bility is raised by the observation that the face of CNF1 inter-acting with RhoA has a surprisingly small number of residuesthat are conserved among the related toxins (Fig. 4a). TheCNF1-C structure provides little evidence for the existence ofa conserved Rho-specific binding site, but instead suggests

that specificity is derived from an induced fit between Rho andCNF1.

MethodsPurification. CNF1-C was cloned into pET28a (Novagen), andexpression was induced in E. coli BL21(DE3) with 0.5 mM IPTG(25 °C). Bacteria were sonicated in 150 mM NaCl, 50 mM phosphatebuffer, pH 8, 0.5 mM EDTA and 10 mM β-mercaptoethanol. CNF1-Cwas purified by anion exchange (Q Sepharose Fast Flow, Poros20HQ/M), hydrophobic interaction (Poros 20PE) and size-exclusionchromatography (Superdex 200). Purified CNF1-C was dialyzed in150 mM NaCl, 20 mM Tris, pH 8, 2 mM DTT and concentrated to ∼ 15 mg ml–1. The yield of 98% pure CNF1-C is 8 mg l–1.

Crystals. Crystals were grown by vapor diffusion at 37 °C using 0.65 MNa2HPO4, 0.65 M KH2PO4, 100 mM HEPES, pH 7.5, 5% (v/v) 1,4-butane-diol and 2 mM DTT. Crystals belong to space group P65 (a = b = 55.0 Å,c = 176.2 Å), and the asymmetric unit contains one CNF1-C monomer.Crystals were cryoprotected in Paratone-N, mounted in fiber loops andflash-cooled at ∼ 100 K (Oxford Cryostream). Oscillation data (1°) werecollected on a Rigaku FR5 generator equipped with focusing mirrorsand Mar345 detector, and processed as described20.

Table 1 Structural statistics

Data collectionDataset Native C866S Se λ1 Se λ2 Se λ3 Se λ4

Wavelength (Å) 1.54180 1.54180 0.97943 0.980035 0.925256 1.06883Resolution (Å) 20.0–1.76 15.0–1.86 20.0–1.95 20.0–1.95 20.0–1.95 20.0–1.95Completeness (%) 99.9 99.8 98.2 97.7 96.1 91.2Redundancy 8.7 10.0 4.2 3.9 3.7 3.5I / σi

1 36.7 (2.6) 33.8 (3.5) 13.3 (2.8) 14.7 (2.6) 12.8 (2.7) 15.6 (2.5)Rmerge

2 5.6 7.9 9.1 8.2 9.0 6.8

PhasingDispersive and anomalous differences (%)3

Se λ1 7.5Se λ2 4.1 6.4Se λ3 4.9 5.7 7.3Se λ4 4.6 5.4 4.2 5.7

Phasing power (isomorphous / anomalous)4

Centric 2.1 / – 2.1 / – 0.05 / 0 – / –Acentric 2.5 / 1.3 2.6 / 0.8 0.04 / 1.0 – / 0.04Number of sites 4Figure of merit (20.0–1.95Å)Centric 0.42Acentric 0.38

Refinement Native C866S Native C866SRcryst (%)5 19.1 19.8 R.m.s. deviationRfree (%)5 21.4 22.0 Bonded B-factor (Å2) 3.1 1.8Number of reflections From ideal geometry

Working set 25,141 23,901 Lengths (Å) 0.0053 0.0051Test set 1,306 1,241 Angles (°) 1.25 1.25

Number of Atoms Average B-factors (Å2)Protein 2,262 2,260 Protein 28.1 29.5Solvent 204 157 Solvent 34.7 37.2Ion 15 10

1Highest resolution shell in parentheses.2Rmerge = 100 × ΣhΣi| Ih,i – Ih| / ΣhΣi Ih,i, where Ih is the mean intensity of symmetry-related reflections, Ih,i.3Anomalous and dispersive differences = r.m.s. ∆F / r.m.s. F, where ∆F for anomalous differences is (F+h – F-h) / 2 (diagonal element) and for dispersivedifferences is Fλi – Fλi (off diagonal).4Isomorphous and anomalous phasing power are FH / E and FH'' / E, respectively. FH and FH'' are the mean amplitude and the anomalous component,respectively, of the heavy atom structure factor. E is the r.m.s. lack-of-closure error.5R-factor = 100 × Σhkl |Fobs − Fcalc| / ΣhklFobs, where Rfree is calculated for a randomly chosen 5% of reflections (F > 0) omitted from refinement, and Rcryst iscalculated for the remaining 95% of reflections (F > 0) included in refinement.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://s

tru

ctb

io.n

atu

re.c

om

© 2001 Nature Publishing Group http://structbio.nature.com

letters

588 nature structural biology • volume 8 number 7 • july 2001

Mutagenesis. Mutants were constructed using divergent primers21

or QuikChange (Stratagene) and checked by DNA sequencing.Mutant proteins were purified as above, except size exclusion chro-matography was eliminated for wild type, S864A, N835A, E943A,N966A and L769M/F771M, which were judged from sodium dodecylsulfate polyacrylamide gel electrophoresis (SDS-PAGE) to be suitablypure (95%) for enzymatic assays.

MAD phase determination. Selenomethionine was incorporat-ed into a quadruple methionine mutant (L769M, F771M, L906Mand L907M)22, and purification and crystallization protocols wereas above. Inverse beam, four wavelength (Table 1) diffractiondata were collected from cryocooled crystals at beamline 1-5(Stanford Synchrotron Radiation Laboratory) using an ADSCQuantum 4 CCD detector, and processed as above. Positions forfour selenomethionines were determined using SOLVE23, andrefinement, phase calculation and solvent flattening were carriedout using SHARP24.

Refinement. WarpNTrace25 was used with 1.95 Å resolution datafor automated model building (83% complete), and was followedby manual building using O26. Refinement protocols using CNS27

were as described20, except that an automated water building rou-tine27 was used. Waters were required to be within 2.6–3.4 Å of ahydrogen bond donor or acceptor and located ≥3 σ in Fo − Fc density.The CNF1-C model includes residues 720–1,014 without any breaksin the main chain. In later refinement stages, occupancy of disor-dered atoms was set to zero. All residues are within allowed regionsof the Ramachandran plot.

Structure of C866S mutant. CNF1-C C866S was expressed, puri-fied and crystallized (P65, a = b = 54.9 Å, c = 178.8 Å) similarly toCNF1-C, and diffraction data were treated as above (Table I).CNF1-C with residue 866 modeled as Ala served as an initial phas-ing model. Simulated anneal omit maps (atoms within 5 Å ofresidue 866, inclusive, omitted) were used in rebuilding the activesite.

Activity assays. RhoA was expressed and purified as described28,except that thrombin-cleaved RhoA was further purified by glu-tathione and size exclusion chromatography (Superdex 200).RhoA at 0.1 mg ml–1 was incubated for varying times at 37 °C in150 mM NaCl, 20 mM Tris, pH 8, 1 mM MgCl2, 10 mM DTT, 100 µMGDP with wild type or mutant CNF1-C at 0.01 mg ml–1. Sampleswere separated on 12% SDS-PAGE-urea (1 M) gels and stainedwith Coomassie blue. Bands were integrated using NIH imagefrom scanned gels (AlphaImager). Densitometric readings werecalibrated, and a standard curve was constructed from knownmixtures of RhoA Gln 63 and RhoA Glu 63. Assays were repeatedthree times.

Coordinates. Coordinates and structure factors have been deposit-ed in the Protein Data Bank (PDB accession codes 1HQ0 and 1HZGfor CNF1-C and CNF1-C C866S, respectively).

AcknowledgmentsWe thank L. Knight and N. Nguyen for excellent technical assistance, H. Bellamyof SSRL for help in MAD data collection and members of the laboratory fortechnical assistance and discussion. We thank A. Hall for his gift of the RhoAexpressing vector. This work is based upon research conducted at the SSRL, whichis funded by the Department of Energy (BES, BER) and the National Institutes ofHealth (NCRR, NIGMS). L.B. was supported by a GANN fellowship. P.G. is a W.M.Keck Distinguished Young Scholar, and research was supported in part by a grantfrom the Universitywide AIDS Research Program.

Correspondence should be addressed to P.G. email: pghosh@ucsd.edu

Received 27 December, 2000; accepted 29 March, 2001.

1. Caprioli, A., Falbo, V., Roda, L.G., Ruggeri, F.M. & Zona, C. Infect. Immun. 39,1300–1306 (1983).

2. Andreu, A. et al. J. Infect. Dis. 176, 464–469 (1997).3. Falzano, L. et al. Mol. Microbiol. 9, 1247–1254 (1993).4. Mills, M., Meysick, K.C. & O’Brien, A.D. Infect. Immun. 68, 5869–8580 (2000).5. Flatau, G. et al. Nature 387, 729–733 (1997).6. Schmidt, G. et al. Nature 387, 725–729 (1997).7. West, A.H., Martinez-Hackert, E. & Stock, A.M. J. Mol. Biol. 250, 276–290 (1995).8. Pepperkok, R. et al. J. Cell Biol. 148, 715–726 (2000).9. Schmidt, G., Selzer, J., Lerm, M. & Aktories, K. J. Biol. Chem. 273, 13669–13674

(1998).10. Schmidt, G., Goehring, U.M., Schirmer, J., Lerm, M. & Aktories, K. J. Biol. Chem.

274, 31875–31881 (1999).11. Yee, V.C. et al. Proc. Natl. Acad. Sci. USA 91, 7296–7300 (1994).12. Tesmer, J.J., Klem, T.J., Deras, M.L., Davisson, V.J. & Smith, J.L. Nature Struct. Biol.

3, 74–86 (1996).13. Rullmann, J.A., Bellido, M.N. & van Duijnen, P.T. J. Mol. Biol. 206, 101–118 (1989).14. Wilson, K.P. et al. Nature 370, 270–275 (1994).15. Wei, Y. et al. Nature Struct. Biol. 2, 218–223 (1995).16. Serre, L., Verbree, E.C., Dauter, Z., Stuitje, A.R. & Derewenda, Z.S. J. Biol. Chem.

270, 12961–12964 (1995).17. Walker, K.E. & Weiss, A.A. Infect. Immun. 62, 3817–3828 (1994).18. Flatau, G., Landraud, L., Boquet, P., Bruzzone, M. & Munro, P. Biochem. Biophys.

Res. Commun. 267, 588–592 (2000).19. Lerm, M., Schmidt, G., Goehring, U.M., Schirmer, J. & Aktories, K. J. Biol. Chem.

274, 28999–29004 (1999).20. Marino, M., Braun, L., Cossart, P. & Ghosh, P. Mol. Cell 4, 1063–1072 (1999).21. Higuchi, R., Krummel, B. & Saiki, R.K. Nucleic Acids Res. 16, 7351–7367 (1988).22. Budisa, N. et al. Eur. J. Biochem. 230, 788–796 (1995).23. Terwilliger, T.C. Methods Enzymol. 276, 530–537 (1997).24. de La Fortelle, E. & Bricogne, G. Methods Enzymol. 276, 472–494 (1997).25. Perrakis, A., Morris, R. & Lamzin, V.S. Nature Struct. Biol. 6, 458–463 (1999).26. Jones, T.A., Zou, J.-Y., Cowan, S.W. & Kjeldgaard, M. Acta Cystallogr. A 47,

110–119 (1991).27. Brünger, A.T. et al. Acta Cystallogr. D 54, 905–921 (1998).28. Self, A.J. & Hall, A. Methods Enzymol. 256, 3–10 (1995).29. Kraulis, P. J. Appl. Crystallogr. 24, 946–950 (1991).30. Thompson, J.D., Higgins, D.G., & Gibson, T.J. Nucleic Acids Res. 22, 4673–4680 (1994).31. Huang, C.C., Pettersen, E.F., Klein, T.E., Ferrin, T.E. & Langridge, R. J. Mol. Graphics

9, 230–236 (1991).32. Rippere-Lampe, K.E., O’Brien, A.D., Conran, R. & Lockman, H.A. Infect. Immun.

69, 3954–3964 (2001).

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://s

tru

ctb

io.n

atu

re.c

om

© 2001 Nature Publishing Group http://structbio.nature.com