The Crystal Structure of Pectate Lyase PelI from Soft Rot Pathogen Erwinia chrysanthemi in Complex with Its Substrate*

The crystallographic structure of the family 3 polysaccharide lyase (PL-3) PelI from Erwinia chrysanthemi has been solved to 1.45Å resolution. It consists of an N-terminal domain harboring a fibronectin type III fold linked to a catalytic domain displaying a parallel β-helix topology. The N-terminal domain is located away from the active site and is not involved in the catalytic process. After secretion in planta, the two domains are separated by E. chrysanthemi proteases. This event turns on the hypersensitive response of the host. The structure of the single catalytic domain determined to 2.1Å resolution shows that the domain separation unveils a “Velcro”-like motif of asparagines, which might be recognized by a plant receptor. The structure of PelI in complex with its substrate, a tetragalacturonate, has been solved to 2.3Å resolution. The sugar binds from subsites -2 to +2 in one monomer of the asymmetric unit, although it lies on subsites -1 to +3 in the other. These two “Michaelis complexes” have never been observed simultaneously before and are consistent with the dual mode of bond cleavage in this substrate. The bound sugar adopts a mixed 21 and 31 helical conformation similar to that reported in inactive mutants from families PL-1 and PL-10. However, our study suggests that the catalytic base in PelI is not a conventional arginine but a lysine as proposed in family PL-9.

bacteria and fungi. In contrast to the 111 sequence-derived families of glycoside hydrolases, polysaccharide lyases have been grouped into only 18 families in the CAZy data base (1). Five polysaccharide lyase families (PL-1, 2, 3, 9, and 10) 4 contain pectate lyases (EC 4.2.2.2 and EC 4.2.2.9). These enzymes cleave polymeric ␣-1,4-linked galacturonic acid (GalA) within the pectate component of the cell wall by ␤-elimination mechanism, leaving an unsaturated C 4 -C 5 bond at the newly formed non-reducing end. The activity is maximal in the pH range 8.5-10.5 and is metal ion-dependent; Ca 2ϩ is the general cofactor except for the members of the family PL-2, which depend on Co 2ϩ , Mn 2ϩ , and Ni 2ϩ (2,3). The divalent cation mediates enzyme-substrate interaction by binding between the protein and the sugar (4). Various pectate lyases prefer either polygalacturonic acid or partially methylated pectin as substrate (5,6).
The catalytic reaction is initiated by an essential basic amino * This work was supported by grants from the Centre National de la Recherche Scientifique. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. The atomic coordinates and structure factors ( acid that abstracts the proton from the carbon C 5 . This residue appears to be an arginine in families PL-1 (7-10), PL-2 (3), and PL-10 (13,14) and a lysine in family PL-9 (12). Whether a lysine or an arginine is the catalytic base for family PL-3 remains to be determined (11).
The enterobacterium E. chrysanthemi is a causative agent of soft rot disease in a wide variety of plants. Among a set of pectin-depolymerizing enzymes produced by Erwinia, pectate lyases are the major pectinases and play a key role in plant tissue maceration. The pectate lyase PelI belongs to family PL-3 and is the only known example from the E. chrysanthemi pectinases that comprises two functional modules, termed the N-terminal domain (residues 20 -116) and the catalytic domain (residues 117-344) (6). The close homologue Pel-3 from Erwinia carotovora (19) has a similar N-terminal extension, suggesting analogous protein architecture.
PelI possesses a 19-residue signal peptide that is cleaved during export by the Sec system, following which the protein is secreted into the external medium by the type 2 secretion system (20). The two structural modules are separated in planta by E. chrysanthemi proteases (6). The resulting catalytic domain (residues 117-344) possesses similar enzymatic properties in vitro to the full-length protein. However, in contrast to the fulllength PelI, it elicits a necrotic reaction associated with an active defense by plants. This hypersensitive response is characteristic of a set of effector proteins secreted by many plant pathogenic bacteria by the type 3 secretion systems (21).
The catalytic domain of PelI shows a sequence identity of 26% with the unique structural representative of the family PL-3, the alkaline pectate lyase Pel-15 of Bacillus sp. strain KSM-P15 (11). This protein exhibits the classical superhelical fold, and its putative active site is located in a conserved cleft that stretches along the external core of the superhelix.
Here, we present the crystallographic structures of fulllength PelI and that of the single catalytic domain termed PelI cata in order to visualize the rearrangements that likely occur in planta after maturation. Furthermore, the associated structure of PelI in complex with a polygalacturonate fragment (a tetragalacturonate GalA 4 ) provides new insight into the catalytical mechanism.

MATERIALS AND METHODS
Protein Purification-PelI was expressed in Escherichia coli BL21 (DE3) (Stratagene) cells carrying a high expression plasmid with the pelI gene and purified to homogeneity as described before (22). The module PelI cata was obtained by treatment of pure full-length PelI with the E. chrysanthemi protease PrtC (6). PelI cata was then separated from the N-terminal domain by chromatography on Protein-Pak CM 8HR (Waters) column equilibrated with 20 mM sodium acetate, 0.1 mM EDTA, pH 5.0 buffer. The protein was eluted with a 0 -0.5 M NaCl linear gradient in the same buffer and then concentrated to 15 mg/ml in a Centricon 10 (Amicon) in 10 mM Tris-HCl with 0.1 mM CaCl 2 .
Construction of Mutant Proteins and Enzyme Assay-Single amino acid mutations were introduced in the PelI sequence by site-directed mutagenesis using the QuikChange kit (Stratagene). The primers used are PelIK224R (5Ј-ggcggcaaaccggac-agagtgctgcagcacaattcc-3Ј), PelIK249R (5Ј-ctgaccggtgaacacgggagattgtggcgttcctgcgg-3Ј), PelIR252K (5Ј-ggtgaacacgggaaattgtggaagtcctgcggcgactgctcc-3Ј), and the corresponding reverse complementary primers (mutated bases are in bold). The nucleotide sequences of mutant genes were checked (Genome express). The mutant proteins were produced and extracted from E. coli BL21 cells as described above for the wild type PelI. Their amounts were estimated by immunoblotting with PelI antibodies, and then equivalent quantities of each mutant protein and the wild type PelI were used in enzyme assay. Pectate lyase activity was measured spectrophotometrically by monitoring the formation of unsaturated products from pectin. The assay mixture contained 50 mM Tris-HCl, pH 8.5, 1 or 3 mM CaCl 2 , and 0.5 g⅐liter Ϫ1 pectin with a degree of methylation of 45%. The appearance of unsaturated products was monitored at 37°C over a period of 30 s with 6-s intervals and used to calculate enzyme activity as described (6).
Crystallization-Crystals of PelI were grown as reported previously (22). They belong to space group P2 1 with unit-cell parameters a ϭ 61.6 Å, b ϭ 70.7 Å, c ϭ 73.4 Å, ␤ ϭ 112.8°and contain two molecules in the asymmetric unit. Crystals of PelI cata were obtained in the same conditions, i.e. by the vapor diffusion method from a solution containing 10 mM zinc sulfate, 100 mM MES, pH 6.5, 25% polyethylene glycol 550 (w/v) at a temperature of 292 K. A 1:1 ratio of protein to reservoir solution was used. Crystals belong to space group P2 1 2 1 2 1 and have cell dimensions a ϭ 45.4 Å, b ϭ 62.6 Å, c ϭ 128.1 Å. The calculated V m is 1.9 Å 3 /Da for two molecules in the asymmetric unit.
The structure of the Michaelis complex was obtained by soaking a crystal of PelI for 24 h in the crystallization buffer depleted in zinc sulfate and with increasing concentrations of GalA 4 (1, 2, 3 mM). These successive soakings were performed in calcium-free solutions at pH 6.5 to avoid enzymatic reaction. The soaking experiment induces a significant modification of unit cell parameters up to a ϭ 61.2 Å, b ϭ 68.7 Å, c ϭ 73.4 Å, ␤ ϭ 113.9°.
X-ray Diffraction Experiments and Structure Determinations-Before data collection, all crystals were rapidly transferred into a cryoprotectant solution consisting of the crystallization solution plus 15% (v/v) ethylene glycol. Synchrotron data were collected at 100 K at European Synchrotron Radiation Facility beamlines, Grenoble, France. The programs XDS and XSCALE were used for data reduction and scaling (statistics in Table 1). 5% of randomly selected reflections were kept apart for cross-validation, and the crystallographic refinement was performed with CNS using the method of slow cooling simulated annealing (23). After each cycle, the model was manually improved using the graphic program TURBO FRODO (24). Final statistics are presented in Table 1. The quality of the refined structures was assessed with PROCHECK (25).
PelI-The first model of PelI was obtained at 1.6 Å resolution by single wavelength anomalous dispersion phasing of a gold derivative (22). A native structure was subsequently determined to 1.45 Å resolution from synchrotron data collected at 100 K at a wavelength of 0.94 Å at beamline BM30A. The crystallographic refinement yields an R-factor of 16.8% (R free 19.5%). The model contains two monomers, termed A and B. Residues 108 -118 were not visible in electron density maps.
According to geometric criteria and to the highest peaks in the 2F o Ϫ F c maps, five Zn 2ϩ and two Ca 2ϩ ions substitute the seven gold ions observed in the derivative structure. Zn 2ϩ ions were further identified by collecting synchrotron data around the zinc K edge and calculating anomalous difference Fourier maps. An extra Zn 2ϩ ion was positioned in the final structure as well as one SO 4 2Ϫ , seven ethylene glycol, and 752 water molecules.
PelI cata -Data were collected to 2.1 Å resolution at beamline ID14 -3. The structure was solved by molecular replacement using the program AMoRe (26) and a two-body search taking as a model the catalytic domain from the structure of PelI (residues 119 -344). The first two residues of PelI cata (residues 117-118) are not visible in electron density maps as in the PelI structure. The final R-factor is 20.4% (R free 27.1%). Two Ca 2ϩ and four Zn 2ϩ are observed at positions similar to those of PelI. Two SO 4 2Ϫ and 325 water molecules have been placed in the asymmetric unit.
PelI-GalA 4 -Data were collected to 2.3 Å resolution at a wavelength of 0.93 Å at beamline ID14 -3. Attempts to solve the structure of the complex by a crystallographic rigid body refinement were unsuccessful due to non-isomorphous crystals (shift in ␤ angle). The structure was solved by molecular replacement using the program AMoRe (26) and a two-body search based on the substrate-free structure. F o Ϫ F c maps were inspected after crystallographic refinement using CNS. The tetrasaccharide bound to molecule A is unambiguously defined in electron density, whereas the one bound to molecule B is weakly defined at its non-reducing end. The four sugar rings of the two ligands were built and oriented in the electron density. No electron density was found for segments 148 -150 and 216 -222 from monomer A and segments 77-81, 145-150, and 215-222 from monomer B. These amino acids were deleted from the refined structure as well as one Ca 2ϩ and five Zn 2ϩ ions. Further cycles of crystallographic refinement yield an R-factor of 20.0% (R free 27.9%). The final model contains two GalA 4 , one Ca 2ϩ ion, one Zn 2ϩ ion, and 171 water molecules.
Structure Analysis-Secondary structure elements were assigned according to main-chain hydrogen bonds and / angles calculated with STRIDE (27). Pair-wise structure comparison was accomplished with DaliLite (28). Images of sequence alignments were prepared using ESPript/ENDscript (29). Interactions between tetragalacturonate and PelI were examined with LIGPLOT (30). Other images were generated with Molscript/Bobscript (31,32). Atomic coordinates and structure factors of PelI, PelI cata , and PelI-GalA 4 have been deposited in the Protein Data Bank (PDB) under the codes 3B4N, 3B90, and 3B8Y, respectively.

RESULTS AND DISCUSSION
Overall Fold of PelI-The structure of full-length PelI has been determined to 1.45 Å resolution. The protein crystallizes with two monomers, A and B, in the asymmetric unit but is monomeric in solution according to gel filtration experiments (not shown). The two monomers A and B are very similar, and their C ␣ traces can be superimposed with a root mean square deviation (r.m.s.d.) of 0.4 Å.
PelI is composed of a fibronectin type III domain (Fn3D) at its N terminus (residues 20 -107) fused to a catalytic domain that displays a parallel ␤-helix fold (residues 119 -331). The connecting segment where cleavage by E. chrysanthemi proteases occurs (residues 108 -118) (6) was not observed in the density and is not modeled. This linker is also predicted as natively disordered by the Regional Order Neural Network server (33), and the crystallographic packing was carefully examined to identify the biological monomer.
The more plausible arrangement of entire protein is presented in Fig. 1A. It possesses the largest buried interface between the two domains, 750 Å 2 , as well as the shortest dis-  (16). The Fn3 domain is bound to the opposite of the active site in both proteins but is oriented perpendicularly to the ␤-helix axis in PelI and parallel in YeGH28. Fn3 Domain-The N-terminal domain of PelI exhibits a conventional seven-stranded Fn3 fold with a Greek key motif composed of four ␤-strands termed ␤C, ␤CЈ, ␤F, and ␤G (Fig. 1). As reported in some other animal and bacterial Fn3Ds, the last strand ␤G is split into two short strands (34). The Fn3D does not contain any disulfide bonds and interacts with the catalytic domain via its flat ␤A, ␤B, ␤E, façade (Fig. 1A). Two hydrophobic residues, Met 25 and Tyr 34 , give extensive contacts at this interface, whereas polar residues mediate hydrogen bonds (Ser 27 , Ser 38 , and Thr 72 , Fig. 1B).
A structural search performed with the DALI server shows that this N-terminal domain superimposes with the best match on an Fn3 module of a neural cell adhesion molecule (r.m.s.d. of 2 Å for 85 aligned C ␣ with an amino acid identity of 13%) (Protein Data Bank code 1CFB) (35). It superimposes also extremely well on the Fn3D of the modular YeGH28 polygalacturonase (16), which includes an additional helix-loop-helix motif that further stabilizes the interdomain interface (r.m.s.d. of 2 Å for 79 aligned C ␣ and a sequence identity of 11%).
Fn3Ds are found in many eukaryotic proteins, where they are frequently implicated in cell adhesion and signaling processes (36). They display a variety of binding modes with other proteins, especially via their BC, CЈE, and FG loops (37,38). Fn3Ds have also been reported in bacterial carbohydrate active enzymes, where they may act as spacers between sugar-binding and catalytic domains (34). It has also been suggested that they can help binding of sugar groups or promote hydrolysis (39,40). However, the distinctive features of carbohydrate binding modules such as a large aromatic platform or a bound Ca 2ϩ (41) are not detected on the Fn3D surface of PelI. Moreover, PelI displays the same enzyme activity on pectin regardless of the presence of its N-terminal domain (6). In addition, following crystal-soaking experiments, a pectin fragment, tetragalacturonate GalA 4 , has been detected only within the catalytic domain, but not in the Fn3D of PelI (see below).
Thus, there is no evidence of direct involvement of this Fn3D (or any other modular carbohydrate active enzymes) in polysaccharide recognition and degradation. On the other hand, eukaryotic Fn3Ds are known for their ability to interact with cell receptors, and this bacterial module may have been a potential pathogen elicitor. However, the hypersensitive response of the plant following the proteolytic cleavage of PelI seems to be triggered by the sole catalytic domain freed from the Fn3D (6).
Catalytic Domain-The catalytic domain is made of eight coils stacked on top of one another (Fig. 1). As in pectate lyases of families 1, 3, and 9, each coil contains three consecutives strand-turn motifs termed (PBn-Tn) nϭ1,3 (7, 8, 12). The char-acteristic ladders of inward-facing asparagines observed in family PL-1 and PL-9 within regular stacked turns (7,12) are absent in PelI as in Pel-15 of the same family PL-3 (11). The stability of the cylindrical core is ensured by inner stacks of aliphatic residues and by five disulfide bonds (highlighted in Fig. 1B).
The structures of PelI and Pel-15 superimpose well on each other with a r.m.s.d. of 1.7 Å on 177 C ␣ atoms of 226 (Fig. 1A). The main differences lie in the two extremities of the ␤-helix. The catalytic domain of PelI begins with a portion of random coil whereas Pel-15 harbors a ␤-strand. This feature might be related to the proximity of the cleavage site in PelI and increases accessibility to proteases. The unique disulfide bond of Pel-15 is conserved in PelI in positions Cys 177 and Cys 182 . Finally, PelI displays in its C-terminal a distinctive ␤-hairpin made of strands PB1.8 and PB1Ј.8 (Fig. 1). The structure of PelI is more distantly related to that of PelC of family PL-1. 176 C ␣ atoms of 226 overlap with a r.m.s.d. of 2.9 Å, and the sequence identity is only 16%. Two disulfide bonds of PelC maintaining extended loops have no structural equivalents in PelI.
Turns (Tn) nϭ1,3 of the ␤-superhelix generally start by an amino acid in the conformation of a left-handed ␣ helix (␣L). The secondary structure assignment of Fig. 1B shows that several turns T of PelI and Pel-15 are made of a single amino acid in ␣L conformation, which sharply bends the solenoid fold. Of particular interest, the T2 amino acids from coils 2 to 5 line up at the surface of the superhelix (Asn 163 , Asn 186 , Gly 218 , Asn 240 ) and form a highly accessible "Velcro"-like motif of asparagines that anchors the Fn3D. The 2.1 Å structure of the catalytic domain alone, PelI cata , has been determined. This structure is quasi-equivalent to that of PelI and demonstrates that this region retains its rigidity when the Fn3D is removed. Hence, this T2 pattern of outward-facing asparagines (NNGN) located in the interdomain interface of full-length PelI becomes accessible after proteolytic maturation. It then might be recognized by an unknown receptor of the infected plant, leading to the defense response.
Ca 2ϩ and Zn 2ϩ Ions-Calcium ions are essential for the catalytic mechanism of all pectate lyases except the family PL-2, which employs other divalent metal cations (2,3). They may neutralize the charges of the acidic substrate as observed in the inactive mutant R218K of PelC and pentagalacturonate GalA 5 that possesses four Ca 2ϩ ions in the binding site (4). PelI was purified in the presence of calcium and was crystallized at pH 6.5 with zinc sulfate (22). One Ca 2ϩ ion is present in an equivalent position in each molecule A and B of PelI and PelI cata . It binds to an external region of the ␤-superhelix and coordinates the main-chain carbonyl O of Ile 192 and Tyr 218 and the sidechain O␦ of Asp 191 and Asn 213 .
In molecule A of PelI, the Ca 2ϩ ion also coordinates two water molecules and displays a classical octahedral geometry. In molecule B, the Ca 2ϩ ion binds three water molecules, giving rise to pentagonal bipyramidal coordination geometry. In both sites, coordination distances are in the range 2.2-2.5 Å. The bound Ca 2ϩ displays a high affinity for the site. Indeed, it has been observed in crystals of PelI grown from protein purified in the presence of the chelating agent EDTA (data not shown). This Ca 2ϩ ion is not located in the active site like those identified in PelC but stabilizes a long turn T3 comprising Tyr 218 .
This loop is rich in glycine and forms a lid over the bound Ca 2ϩ (Fig. 1A). A structural Ca 2ϩ is found at an equivalent position in Pel-15, crystallized at pH 6.5 (11).
The increase of the pH to 9.5 in Pel-15 involves the binding of a catalytic Ca 2ϩ at a neighboring cluster of acidic residues (Asp 63 , Glu 83 , and Asp 84 ) in the putative sugar binding site (11). Asp 63 and Glu 83 are essential for lyase activity (42), and this acidic cluster is highly conserved among pectate lyases (8,12,14). However, in both PelI and PelI cata , the two equivalent aspartic residues Asp 173 and Asp 195 already coordinate two Zn 2ϩ ions provided by the crystallization buffer. These interactions may interfere with the binding of catalytic calcium ions at this site even at alkaline pH. Interestingly, Zn 2ϩ ions have a strong inhibitory effect on PelI activity (data not shown). The other Zn 2ϩ ions mediate contacts between domains in the crystal. All Zn 2ϩ ions display tetrahedral coordination geometry with an average distance of 2.0 Å.
The Conformation of Bound Substrate-The structure of PelI in complex with pectin fragment GalA 4 has been determined at a resolution of 2.3 Å after crystal soaking in the absence of calcium. One substrate molecule binds in each molecule A and B to an elongated groove conserved in families PL-1, PL-3, and PL-9. The binding site of PelI stretches from coils 3 to 6, and its base is formed from amino acids of strands PB1. The reducing end of the substrate, termed Ada1, is oriented toward coil 6. The sugar subsites have been identified after superimposition between PelI-GalA 4 and inactive PelC-GalA 5 (4). One galacturonate unit was not visible in PelC, and the substrate occupies subsites Ϫ1 to ϩ3.
Of particular interest in PelI, the substrate binds from subsites Ϫ2 to ϩ2 in molecule A, whereas it lies on subsites Ϫ1 to ϩ3 in molecule B (Figs. 2 and 3). These two sugar binding modes are consistent with the dual mode of bond cleavage in this substrate. PelI cleaves GalA 4 with a frequency of 62% between moieties Ada3 and Ada4 and with a frequency of 38% between moieties Ada2 and Ada3 (43). Thus, the two observed Michaelis complexes perfectly confirm the positions of the two scissile glycosidic bonds in GalA 4 acted   end (A1 or B1), and the moiety Ada4 is at the non-reducing end (A4 or B4). A yellow arrow symbolizes the proton abstraction at the C 5 atom.

Structure of Pectate Lyase PelI with and without Substrate
upon by PelI. In both molecules, the sugar folds so as to form a 2 1 helix (2 units/helical turn) between subsites Ϫ1 and ϩ1 and a 3 1 helix between subsites ϩ1 and ϩ2 (Fig. 4). Such distorted helical topology between the essential subsites Ϫ1 and ϩ2 has also been reported in PelC and in Pel10Acm complexed with substrate and Ca 2ϩ (4,14). The latter protein exhibits an ␣-barrel fold, (␣/␣) 3 , instead of the classical ␤-superhelix fold. Therefore, the conserved conformation of the substrate at this central zone Ϫ1 to ϩ2 appears to be a requisite for the lyase activity and can be induced even in the absence of Ca 2ϩ .
Rearrangements in the Active Site-The structure of PelI-GalA 4 indicates that the active site does not overlap with the Fn3D interface on the surface of the catalytic domain. However, in PelI crystals, positive subsites of monomer A and negative subsites of monomer B are partially buried in non-crystallographic interfaces. Monomers A and B shift by ϳ2 Å with respect to each other during the soaking of PelI crystals with tetrasaccharide. As a consequence, the substrate has gained access to the active site in PelI-GalA 4 and loops 144 -150 and 215-222 are disordered at these interfaces. These structural constraints at the A/B interfaces are very likely to be responsible for the two sugar binding modes observed in the crystal. They also have an impact on adjacent Asp 223 and Lys 224 located at subsite ϩ1 whose side chains are disordered. Beside these structural rearrangements due to the crystal packing, it can be noted that strand PB1Ј.8, which is poorly conserved in the PL-3 family, has moved by 2 Å in PelI-GalA 4 to accommodate the substrate. Hence, Lys 317 and Thr 322 come close to the tetrasaccharide to bind it at subsites Ϫ1 and ϩ2, respectively (Fig. 3).
Interestingly, the two Zn 2ϩ ions coordinated to the acidic cluster in PelI (Asp 173 , Asp 195 ) disappear in PelI-GalA 4 (the soaking experiment was performed in zinc-depleted solutions) and the side chain of Asp 195 has rotated by 106°around 1 . It now has a suitable orientation to bind the sugar at subsite Ϫ1 (Fig. 2). Finally, a sulfate molecule has been substituted by the substrate at subsite ϩ2 of monomer B (Fig. 2). This sulfate molecule was stabilized by the amine N of Lys 224 and by the guanidinium of Arg 252 . The latter residue is the unique arginine of the active site and raises the possibility of it catalyzing the ␤-elimination of the proton at subsite ϩ1.
The Brønsted Base-In the sugar binding site of PelI, the possible catalytic bases are Lys 224 , Lys 249 , and Arg 252 . It is noteworthy that these three amino acids are invariant in Pel-15 where they display a similar orientation of their side chains.
In PelI-GalA 4 , the guanidinium group of Arg 252 firmly binds the carboxylic group of the sugar at subsite ϩ2 (Fig. 3), which prevents any displacement toward subsite ϩ1. The side chain of the neighboring Lys 249 is also involved in a hydrogen bond with the sugar at subsite ϩ2. This orientation is surprising because Lys 249 is equivalent in sequence to the catalytic base Arg 218 of PelC. Thus, a similar extended conformation of its side chain toward subsite ϩ1 (leaving group) was expected. However, Arg 218 belongs to a turn T3 in PelC and adopts an ␣L conformation, whereas the isostructural Lys 249 has a ␤ conformation in PelI. Such divergence in the hydrogen bond network of the two proteins explains the opposite direction taken by the side chains of these two equivalent residues.
The last candidate, Lys 224 , is the only invariant basic amino acid of the active site in family PL-3 and is well conserved in family PL-1 (Fig. 1B). It is close to the flexible segment 215-222, and its side chain is disordered in PelI-GalA 4 . Nevertheless, its amine function N can point next to the C 5 atom of the sugar at subsite ϩ1 if the orientation of the side chain in substrate-free monomer B is taken as a model (its side chain has another orientation in substrate-free monomer A because it is hydrogen-bonded to O␦ of Asp 195 ). Hence, Lys 224 seems the best candidate for a catalytic base within PelI. To better address this question, conservative point mutations K224R and K249R, as well as R252K, have been introduced and analyzed ( Table 2).
The mutant K224R is completely defective in lyase activity, unlike K249R and R252K, keeping ϳ40 -60% and 6 -9% of wild type activity, respectively (Table 2). Interestingly, at a higher Ca 2ϩ concentration in the substrate medium, enzymatic activities of K249R and R252K mutants are less affected. Hence, it seems likely that this cation can partially compensate for these mutations either by changing the pK a of the residues or by altering the substrate conformation. These three mutations maintain the electrostatic character of the binding site, and Lys 224 appears again as the best candidate catalytic base. A lysine has also been proposed as the catalytic base for Pel9A from family PL-9, although without the support of a liganded structure (12). The putative catalytic bases of PelI and Pel9A are not equivalent in sequence, but their amine functions N are located in similar positions in the active site.

CONCLUSION
Our work demonstrates that modular PelI, despite its 11amino acid-long loop sequence, is a compact protein with two interacting domains. Notably, the interdomain interface comprises a remarkable pattern of stacked asparagines from the catalytic domain. The hypersensitive response of the plant might be elicited by exposure of this motif (NNGN) following elimination of the Fn3 domain. Hence, the response would be independent of lyase activity as observed in HrpW from Erwinia amylovora (44).
Our work also answers the question raised by the structure of Pel-15 (11) by showing that the catalytic base in family PL-3 is not an arginine as in family PL-1, PL-2, and PL-10 but a lysine. The use of a lysine, whose pK a is 10.5, as a base in enzymes active at pH ϳ9.0 seems more appropriate than the use of an arginine, whose pK a is 12.5. Ca 2ϩ ions close to the arginine in the Michaelis complex could, however, lower its pK a to a value that would make the enzyme active (4). The putative catalytic base Lys 224 of PelI is located one coil below the catalytic base Arg 218 found in the paradigm PelC from family PL-1 (Fig. 1B). The specific activity of PelI on polygalacturonate is lower than that of PelC, and K249R substitution in PelI may have resulted in a PL-1-like efficient enzyme (Lys 249 in PelI is equivalent to Arg 218 in PelC), yet this mutant completely lacks activity. Structural comparisons suggest that the key feature is the mainchain conformation of this amino acid, ␣L in family PL-1 against ␤ in family PL-3.
Our structure of the complex PelI-GalA 4 shows that the sugar should bind in pectate lyases with a characteristic distorted helical conformation. It indicates that this topology is not calcium-dependent, although this cation is required for pectate lyase activity. In PelI, the presence of a structural Ca 2ϩ ion near the active site may influence the efficiency of the catalytic mechanism. This cation stabilizes the flexible loop 215-222 adjacent to Asp 223 and the putative catalytic base Lys 224 . Without this structural Ca 2ϩ , the essential Lys 224 is disordered in PelI-Gal4. Catalytic Ca 2ϩ ions certainly bind to the active site of PelI at alkaline pH. In any case, pectate lyases are secreted by bacteria in the plant tissue, where they act on cell wall pectin. Ca 2ϩ ions are frequently coordinated to free carboxyl groups of pectin, so catalytic cations can be naturally found complexed with the substrate rather than bound to the enzyme.
The structures of inactive PelC and GalA 5 have previously revealed how these catalytic Ca 2ϩ ions can mediate interactions between the protein and the sugar at alkaline pH (4). The three Ca 2ϩ ions that are unambiguously defined in the electron density maps of PelC (the fourth Ca 2ϩ has half occupancy) can be modeled in our structure of PelI in complex with GalA 4 with limited structural rearrangements. These ions are of particular interest because, in this configuration, they may connect together deprotonated carboxyl groups of PelI and of the sugar: one Ca 2ϩ between Asp 173 , Asp 195 , and subsite sugar Ϫ1, a second between Asp 195 and subsite sugar ϩ1, and a third between Asp 223 , Asp 256 , and subsite sugar ϩ1. Asp 223 is invariant in family PL-3 like the adjacent essential Lys 224 , and a sequence signature -DK-of the active site can be proposed for this family.
A detailed catalytic mechanism has been proposed for Pel9A, with a lysine as a Brønsted base and the support of a More O'Ferrall diagram (12). The interaction of the carboxyl group from the ϩ1 sugar with positive charges can be an important factor for proton abstraction and stabilization of an anionic intermediate. In our high pH model of PelI, the two putative Ca 2ϩ ions that are coordinated to the conserved Asp 195 and Asp 223 may play this role by connecting the sugar carboxyl. A reasonable assumption is that the transiently protonated base Lys 224 may then act as the catalytic acid to eliminate the substituent at O 4 and give a C 4 -C 5 unsaturated product. Interaction of polar residues with the Ϫ1 and ϩ1 sugar oxygen atoms can facilitate the whole reaction (Gln 227 , Asn 229 , and Lys 317 according to Fig. 3).
Finally, the comparison between PelI and PelI-GalA 4 sheds new light on the importance of the conserved acidic cluster Asp 173 , Glu 194 , and Asp 195 at the negative subsites. This acidic cluster can favorably complex with Zn 2ϩ ions, which results in the obstruction of several Ca 2ϩ binding sites and prevents the formation of an active enzyme-substrate complex. Moreover, the side chain of the putative catalytic base Lys 224 needs to be released from its interaction with Asp 195 to reach an appropriate orientation at subsite ϩ1 to perform ␤-elimination. In either PelC or in the complex PelC-GalA 5 from family PL-1, the equivalent lysine Lys 190 is not released from its contact with a similar acidic cluster and the catalytic base is Arg 218 .
This head-to-head interaction between an acidic and a basic amino acid at the essential Ϫ1/ϩ1 subsites evokes the acid-base mechanism observed in family GH28 at acidic pH between the catalytic aspartate and histidine residues (isostructural to Glu 194 /Asp 195 and the putative catalytic base Lys 224 in PelI). Thus, not only do the overall architectures of the modular GH28 and PL-3 resemble one another but the positions of their catalytic residues are also similar. The divergent evolution of carbohydrate-degrading enzymes bearing a ␤-superhelix fold may have given rise first to the families GH28 and PL-3 and then to family PL-1 with a catalytic base at a distinct position.