Mapping the ribonucleolytic active site of eosinophil-derived neurotoxin (EDN). High resolution crystal structures of EDN complexes with adenylic nucleotide inhibitors.

Eosinophil-derived neurotoxin (EDN), a basic ribonuclease found in the large specific granules of eosinophils, belongs to the pancreatic RNase A family. Although its physiological function is still unclear, it has been shown that EDN is a neurotoxin capable of inducing the Gordon phenomenon in rabbits. EDN is also a potent helminthotoxin and can mediate antiviral activity of eosinophils against isolated virions of the respiratory syncytial virus. EDN is a catalytically efficient RNase sharing similar substrate specificity with pancreatic RNase A with its ribonucleolytic activity being absolutely essential for its neurotoxic, helminthotoxic, and antiviral activities. The crystal structure of recombinant human EDN in the unliganded form has been determined previously (Mosimann, S. C., Newton, D. L., Youle, R. J., and James, M. N. G. (1996) J. Mol. Biol. 260, 540-552). We have now determined high resolution (1.8 A) crystal structures for EDN in complex with adenosine-3',5'-diphosphate (3',5'-ADP), adenosine-2',5'-di-phosphate (2',5'-ADP), adenosine-5'-diphosphate (5'-ADP) as well as for a native structure in the presence of sulfate refined at 1.6 A. The inhibition constant of these mononucleotides for EDN has been determined. The structures present the first detailed picture of differences between EDN and RNase A in substrate recognition at the ribonucleolytic active site. They also provide a starting point for the design of tight-binding inhibitors, which may be used to restrain the RNase activity of EDN.

protein stored in the matrix of the large secretory granules (1).
It is a small, basic protein (2) that belongs to the pancreatic ribonuclease A (RNase A; EC 3.1.27.5) superfamily (3) and is also known as RNase-2 or RNase U s . EDN was initially identified by its ability to induce the Gordon phenomenon (muscle stiffness, ataxia, incoordination, and spasmodic paralysis) when injected into rabbits (4,5). Its neurotoxic effect is achieved through a selective killing of cerebellar Purkinje cells (6). The protein also displays cytotoxicity against helminths, single-stranded RNA viruses, and respiratory epithelial cells; and a role as a host defense protein has been suggested (7). Damage of host tissues by EDN could contribute to the secondary effects associated with inflammatory disorders and hypereosinophilic syndromes (8).
EDN shares 36% amino acid identity with RNase A and 67% identity with a related eosinophil RNase, eosinophil cationic protein (ECP, also known as RNase-3) (7). EDNЈs enzymatic activity is essential for its neurotoxic, helminthotoxic, and antiviral activities (9 -11) and is 3-to 30-fold lower than that of RNase A, depending on the substrate used (9,12).
The crystal structure of recombinant EDN in complex with sulfate has been determined previously at 1.83-Å resolution (13). The topology of the EDN molecule includes the RNase A fold and core ribonucleolytic active site architecture (14), which is conserved among all these molecules, although both ECP (15) and EDN exhibit significant differences at the peripheral substrate-binding sites (16).
The core of the catalytic site of RNase A consists of subsites B 1 , P 1 , and B 2 . These subsites accommodate the phosphate where phosphodiester bond cleavage occurs (P 1 ) and the nucleotide bases on the 3Ј and 5Ј sides of the scissile bond (B 1 and B 2 , respectively) (17). In addition, several studies (18 -20) have identified additional sites, including P 0 and P 2 . P 0 interacts with the 5Ј-phosphate of a nucleotide base bound at B 1 and P 2 interacts with the 3Ј-phosphate of a nucleotide base bound at B 2 (for recent reviews see Refs. 20 -22). The three main catalytic residues of RNase A (His-12, Lys-41, and His-119 of the P 1 subsite) are present in EDN as His-15, Lys-38, and His-129.
The key B 1 residues, Thr-45 and Phe-120 in RNase A, are also maintained in EDN as Thr-42 and Leu-130, but the other components of this subsite differ. The B 2 subsite is partially conserved between EDN and RNase A, but subsites P 0 and P 2 are not. Although EDN and RNase A bind only pyrimidines at B 1 and prefer purines at B 2 , differences in B 1 and B 2 site structures give rise to subtle changes in substrate specificity. With polynucleotide substrates, EDN has a 20-fold preference for cytidine over uridine; with dinucleotide substrates, EDN has a 2-fold preference for cytidine at B 1 and a 100-fold preference for adenosine at B 2 (9). Two sulfate anions were found in the EDN⅐sulfate complex structure (13) occupying two distinct subsites. One of these is subsite P 1 , whereas the other is a site not identified in the structure of RNase A, suggested to correspond to a new P Ϫ1 phosphate-binding subsite.
The biological properties attributed to EDN have been related to its ribonucleolytic activity (7). The analysis of its substrate specificity and the identification of the residues involved in substrate interaction would help in understanding its mechanism of action. In addition, the identification of nucleotidebased inhibitors may lead to therapeutic agents for use against the pathological conditions associated with eosinophil RNases. Here we present the first structures of recombinant EDNnucleotide complexes (at 1.8-Å resolution) and make a detailed comparison with the EDN⅐sulfate structure at higher resolution (1.6 Å) than the previous reported structure (13). The structures of the complexes have revealed a detailed picture of critical residues involved in the P 1 and B 2 substrate-binding sites and their flexibility in interaction with different adenylic nucleotides. The analysis of the sulfate-containing structure and comparison with the nucleotide complexes also confirm the presence of a previously suggested P Ϫ1 subsite for EDN (13). In addition, kinetic results suggest that these nucleotides can serve as a starting point toward the design of potent inhibitors of EDN.
Protein Purification and Crystallization-Recombinant EDN was expressed in Escherichia coli and purified as described previously (13,23). Briefly, a synthetic gene for human EDN was cloned into the pET11c expression vector, and the protein was purified from inclusion bodies. The recombinant EDN, which contains an additional methionine residue at the N terminus, has the same specific activity as the protein purified from the natural host (23) and does not have the post-translational modifications that are present in EDN isolated from human body fluids.
Crystals of recombinant EDN were grown using the hanging drop/ vapor diffusion method from drops containing 9 mg/ml protein in 0.1 M sodium cacodylate buffer (pH 6.5), 0.75 M ammonium sulfate, and 2.5% ethanol. Drops were equilibrated against reservoirs containing 0.1 M sodium cacodylate buffer (pH 6.5), 1.5 M ammonium sulfate, and 5% ethanol. Single crystals appeared after 3-4 days at 16°C.
Diffraction Measurements-The EDN crystals diffracted to a minimum Bragg spacing of 1.6 Å on a Synchrotron radiation source at 100 K using a cryoprotectant solution containing 0.1 M sodium cacodylate buffer, pH 6.5, 1.3 M sodium potassium tartrate, and 30% 2-methyl-2,4pentanediol. The systematic absences and symmetry were consistent with the space group P2 1 2 1 2 1 with the following unit cell dimensions: a ϭ 53.4 Å, b ϭ 57.2 Å, and c ϭ 42.2 Å. There is one EDN molecule per crystallographic asymmetric unit, and ϳ50% of the crystal volume is occupied by solvent. The EDN⅐2Ј,5Ј-ADP, EDN⅐3Ј,5Ј-ADP, and EDN⅐5Ј-ADP complexes were obtained by soaking native EDN crystals with 100 mM 2Ј,5Ј-ADP, 3Ј,5Ј-ADP, or 5Ј-ADP for at least 2 days prior to data collection. Because the native crystals were grown in the presence of ammonium sulfate, sulfate ions were expected to bind at the active site of EDN as observed previously (13). Hence, prior to soaking, EDN crystals were transferred to 1-2 ml of a solution containing 1.5 M sodium potassium tartrate, 5% ethanol, and 0.1 M sodium cacodylate buffer, pH 6.5, for at least 2 days to remove the bound sulfate ions.
Diffraction data for the EDN⅐sulfate complex were collected to 1.6-Å resolution at 100 K using the Synchrotron Radiation Source at Daresbury (UK) on station PX 9.5 using a MAR300 image plate and also at the Protein Diffraction Beamline at Ellettra-Trieste (Italy) using a MAR345 image plate. Data for the EDN⅐2Ј,5Ј-ADP, EDN⅐3Ј,5Ј-ADP, and EDN⅐5Ј-ADP complexes were collected (at 1.8 Å) in-house at 100 K using a MAR300 image plate mounted on an Enraf-Nonius rotating anode x-ray source with CuK ␣ radiation. Raw data images were indexed, integrated, and corrected for Lorentz and polarization effects using the program DENZO (24). All data were scaled and merged using the program SCALEPACK (24). Intensities were then truncated to amplitudes by the TRUNCATE program (25). Details of data processing statistics are presented in Table I.
Structure Refinement-All crystallographic refinement was carried out using the program X-PLOR 3.851 (26) with the published structure of EDN⅐sulfate (13) used as a starting model. The behavior of the R free (27) value was monitored throughout refinement. Several rounds of refinement, model building, individual B-factor refinement, and bulk solvent correction as implemented in X-PLOR 3.851 (28) were performed until the R free value for every model could not be improved any further. During the final stages of refinement, water molecules were inserted into the model at positions corresponding to peaks in the ͉F o ͉ Ϫ ͉F c ͉ electron density maps with heights greater than 3 and at hydrogen bond forming distances from appropriate atoms. 2͉F o ͉ Ϫ ͉F c ͉ calc maps were also used to verify the presence of the peaks. Water molecules with a temperature factor higher than 65 Å 2 were excluded from subsequent refinement steps. In the case of EDN⅐ligand complexes, the ligand molecule was included during the final stages of refinement. The details of refinement are given in Table I. The program PROCHECK (29) was used to assess the quality of the final structure. Analysis of the Ramachandran ( Ϫ ) plot (30) showed that all residues lie in the allowed regions.
Determination of K i Values-RNase activity of EDN was measured by a spectrophotometric method. Assays were carried out in 0.2 M MES⅐NaOH, pH 6.5, at 25°C using 0.5-cm path length cells. UpA was used as a substrate. Substrate and inhibitor concentrations were determined spectrophotometrically using the following extinction coefficient: ⑀ 261 ϭ 23,500 M Ϫ1 cm Ϫ1 for UpA (31) and ⑀ 259 ϭ 15,400 M Ϫ1 cm Ϫ1 for 5Ј-ADP, 2Ј,5Ј-ADP, and 3Ј,5Ј-ADP (32). The activity was measured by following the initial reaction velocities using the difference molar absorbance coefficient ⌬⑀ 286 ϭ 570 M Ϫ1 cm Ϫ1 for the transphosphorylation reaction of UpA (33). Accession Numbers-The final atomic coordinates for the four complexes of EDN (sulfate, 2Ј,5Ј-ADP, 3Ј,5Ј-ADP, and 5Ј-ADP) have been deposited with the RCSB Protein Data Bank (accession codes 1HI2, 1HI3, 1HI4, and 1HI5, respectively).

RESULTS
Overall Structures-The structures of EDN detailed here are very similar to the EDN⅐sulfate structure reported previously (13). The r.m.s. difference between C ␣ atoms in the two sulfatebound structures is 0.15 Å. The corresponding values for the EDN⅐3Ј,5Ј-ADP, EDN⅐2Ј,5Ј-ADP, and EDN⅐5Ј-ADP complexes are 0.31, 0.29, and 0.28 Å, respectively. In the present EDN⅐sulfate structure, the additional methionine (Met-0) residue from the recombinant protein was observed.
The r.m.s. differences between the C ␣ atoms of the structures of the present 1.6-Å resolution EDN⅐sulfate complex and those of the 3Ј,5Ј-ADP, 2Ј,5Ј-ADP, and 5Ј-ADP complexes are 0.29, 0.27, and 0.27 Å, respectively. The corresponding values within the complexes, 3Ј,5Ј-ADP against 2Ј,5Ј-ADP and 5Ј-ADP are 0.27 and 0.31 Å, respectively, whereas the r.m.s. difference between the 2Ј,5Ј-ADP and the 5Ј-ADP complexes is 0.18 Å. The differences between the four protein structures are very small, concentrated in the loop regions, and seemingly unrelated to the presence of the different inhibitors. The EDN⅐sulfate structure, determined at 1.6-Å resolution, contains 152 water molecules. The structures of the 3Ј,5Ј-ADP, 2Ј,5Ј-ADP, and 5Ј-ADP complexes are at a slightly lower resolution (1.8 Å) and contain 123, 110, and 132 water molecules, respectively. The residues at the active site are oriented similarly in all four complexes, with the exception of His 129 (noted above), and there are no significant conformational changes due to inhibitor binding. A numbering scheme and torsion angle definitions are shown in Fig. 1 using 5Ј-diphosphoadenosine 3Ј-phosphate (ppA-3Ј-p) as a reference molecule (38) for the three adenylic mononucleotides.
The Binding of 3Ј,5Ј-ADP to EDN-The 3Ј,5Ј-ADP molecule is reasonably well defined in the electron density map ( Fig. 2A). The conformation of 3Ј,5Ј-ADP when bound to EDN is very similar to that observed previously for B 2 -bound adenosine in the complexes of RNase A with d(pA) 4 (43), d(ApTpApApG) (44), d(CpA) (37,45), and 2Ј,5Ј-CpA (45), as well as those frequently observed in free adenylic nucleotides (46). The glycosyl torsion angle adopts the anti-conformation, whereas the ribose is at the C2Ј-endo conformation. The ␥ torsion angle is in the unusual sp range (Table II), but its value (Ϫ29°) is very close to the highly favorable Ϫsc region (Ϫ30°to Ϫ90°) (46) and the 5Ј-phosphate group is oriented toward the adenosine. The ␦ torsion angle, dictated by the orientation of the 3Ј-phosphate group, is in the ϩac region as commonly found in both free and protein-bound nucleotides (46).
The inhibitor binds to the P 1 -B 2 region of the catalytic site of EDN in a manner similar to the binding of the analogous parts of d(ApTpApApG) (44,45), d(CpA) (37,45), and 2Ј,5Ј-CpA (45) to RNase A (Fig. 2B). The molecular surface of EDN calculated using the GRASP program (47) shows how the active site cleft accommodates the 3Ј,5Ј-ADP nucleotide (Fig. 3A). EDN and 3Ј,5Ј-ADP engage in seven hydrogen-bond interactions, and three water molecules form hydrogen bonds with the inhibitor (Table III). The 5Ј-phosphate binds to subsite P 1 (the distance between the phosphorous to sulfur of the SO 4 2Ϫ (A) of EDN⅐sulfate complex is 0.3 Å) and is involved in hydrogen bond interactions with the imidazole rings of His-15 and His-129. It also forms hydrogen bonds with the side chain of Gln-14, and the main-chain amide nitrogen of Leu-130 and makes a watermediated interaction with the side chain of Lys-38 (Fig. 2B). The adenosine binds to the B 2 subsite with its five-membered ring being almost parallel to the imidazole ring of His-129 (active conformation A) and thereby involved in ring-stacking interactions. In addition, atoms N1 and N6 of the adenine are located at hydrogen bonding distance from the side-chain atoms of Asn-70. The 3Ј-phosphate group forms a hydrogen bond with the ␣-amino group of Met-0 at the N terminus and participates in a water-mediated interaction with the side chain of Trp-10. There are also numerous van der Waals contacts between the inhibitor and Trp-7, Cys-62, Arg-68, Asn-70, Ala-110, Val-128, and His-129. With the exception of residue His-129, a shift in Asp-112 side-chain orientation and a slight movement of Arg-68 (ϳ1.0 Å) from its position in the EDN⅐sulfate complex toward the inhibitor, there are no other significant conformational changes in the catalytic site of EDN upon 3Ј,5Ј-ADP binding.
Comparison of the EDN-3Ј,5Ј-ADP binding mode with the RNase A⅐ppA-3Ј-p complex (38) (PDB code 1AFK) shows equivalent interactions at the main phosphate site P 1 and at the   secondary base site B 2 (Fig. 3B). Both of the 5Ј-phosphate of 3Ј,5Ј-ADP and the 5Ј-␤-phosphate of ppA-3Ј-p are located at P 1 . His-119 in RNase A and His-129 in EDN are both in the same plane and stack against the five-membered ring of adenine.
Although the binding of adenine in both EDN-3Ј,5Ј-ADP and RNase A-ppA-3Ј-p is almost coplanar (as observed in RNase A complexes with substrate analogs d(CpA) (37) and d(ApTpApA) (44)), the six-and five-membered rings are reversed. Finally, the 3Ј-phosphates of ppA-3Ј-p and 3Ј,5Ј-ADP are located at the N-terminal ends of RNase A and EDN, respectively. The Binding of 2Ј,5Ј-ADP to EDN-The structure of 2Ј,5Ј-ADP is very well defined within the electron density map (Fig.  2C). The conformation of 2Ј,5Ј-ADP is typical for protein-bound nucleotides and the deoxyribose takes up the energetically favored C2Ј-endo anti conformation (Table II) (46). The ␥ torsion angle around the C4Ј-C5Ј bond of the adenosine is in the ϩsc range, and the ␦ torsion angle has a value of 158°. Both these angles are in the range frequently observed in proteinbound nucleotides (46).
Although 2Ј,5Ј-ADP binds to the same P 1 -B 2 region of the EDN active site as 3Ј,5Ј-ADP (Fig. 2D), there are striking differences in their interaction with EDN. Interestingly, in the 2Ј,5Ј-ADP⅐EDN complex, a 2Ј-rather than the 5Ј-phosphate occupies the P 1 site. Indeed, the 2Ј-phosphate group forms a similar extensive set of hydrogen bonds with the side chains of Gln-14, His-15, and His-129 and with the amide nitrogen of Leu-130 (Table IV) as observed for the 5Ј-phosphate group in the 3Ј,5Ј-ADP⅐EDN complex. However, although the adenine is almost parallel to the imidazole of His-129, the five-membered ring does not stack against the imidazole ring as in the 3Ј,5Ј-ADP complex. Instead, the imidazole ring of His-129 adopts conformation B (inactive) and packs against the five-membered ring of adenine in a different orientation (Fig. 2D). There are three water molecules making hydrogen bond interactions with the inhibitor, and one of them mediates interactions between the adenine moiety and the side chain of Asn-70. A shift in orientation of both Arg-68 and Asp-112 is observed when compared with the sulfate bound EDN or the 3Ј,5Ј-ADP complex, although these residues are not directly interacting with the adenine. The 5Ј-phosphate is not involved in any direct interactions with EDN residues and is only involved in a watermediated interaction with the ␣-amino group of Met-0 at the N terminus (Fig. 2D). The binding of 2Ј,5Ј-ADP does not trigger any conformational changes at the active site of the EDN molecule and, apart from residues Gln-14, His-15, His-129, and Leu-130, 2Ј,5Ј-ADP does not interact directly with any other parts of EDN.
The Binding of 5Ј-ADP to EDN-The nucleotide is very well defined in the electron density map (Fig. 2E). The deoxyribose adopts the C2Ј-endo anti energetically favored conformation FIG. 2. A, C, and E, diagrams of the 1.8-Å sigmaA 2͉F o ͉ Ϫ ͉F c ͉ electron density map of 3Ј,5Ј-ADP, 2Ј,5Ј-ADP, and 5Ј-ADP, respectively. Electron density maps were calculated using the standard protocol as implemented in X-PLOR 3.851 (60) from the EDN model before incorporating the coordinates of each inhibitor, are contoured at the 1.0 level, and the refined structure of the inhibitor is shown. B, D, and F, diagrams showing the interactions of 3Ј,5Ј-ADP, 2Ј,5Ј-ADP, and 5Ј-ADP with EDN, respectively. EDN residues are drawn as ball-and-stick models, water molecules appear as gray spheres, and the nucleotide molecules are shown in dark gray. Hydrogen bonds are indicated by dashed lines. (46), and the ␥ torsion angle is in the ϩsc range. The ␣ and pp torsion angles around the phosphoester bond are in the ϩsc and ϩac range, respectively. These values are in the accepted range for both free and protein-bound nucleotides (Table II) (46). 5Ј-ADP is bound at the active site of EDN in an extended conformation with the ␤rather than the ␣-phosphate group at subsite P 1 and the adenosine located in a region away from the B 2 subsite close to but not exactly at B 1 (Fig. 2F). The ␤-phosphate group engages in a hydrogen-bonding network with the peptide nitrogen of Leu-130 and the side-chain atoms of Gln-14, His-15, and His-129 (found in conformation B). The ␣-phosphate forms a hydrogen bond with the ⑀-amino group of Lys-38, whereas the 2Ј-hydroxyl group of the ribose makes two hydrogen bonds with the main-chain atoms of Gln-40 (Table V). Furthermore, EDN and 5Ј-ADP participate in an extended water-mediated hydrogen bond network involving seven water molecules, and the residues Trp-7, Gln-14, Arg-36, Asn-39, Gln-40, Val-128, His-129, Leu-130, and Asp-131. In addition, 5Ј-ADP also has van der Waals interactions with His-82.
Inhibition of EDN by Mononucleotides 3Ј,5Ј-ADP, 2Ј,5Ј-ADP, and 5Ј-ADP-The inhibition constants for these adenylic mononucleotides have been determined spectrophotometrically. The K i values for 3Ј,5Ј-ADP, 2Ј,5Ј-ADP, and 5Ј-ADP are 32 Ϯ 2, 64 Ϯ 4, and 92 Ϯ 7 M, respectively. These results are consistent with the crystallographic analysis of EDN complexes described above, i.e. among the three complexes, 3Ј,5Ј-ADP binds most avidly to EDN with a maximum number of contacts with the protein atoms. However, when comparing the determined K i values with the reported ones in the literature, differences in pH and ionic strength of the assay mixture have to be taken into account. Kinetic assays were performed at pH 6.5, the pH used for the crystallization buffer. Previous kinetic characterization of eosinophil RNases indicated that they have a lower optimum pH for catalytic efficiency in comparison with RNase A. Sorrentino and Libonati (48) reported that of the optimum pH for catalytic efficiency is 7.5-8.0 for RNase A and 6.5-7.0 for EDN. Thus a lower pH optimum for the nucleotide interaction should be expected, because RNase A has an optimum pH of 7.5 for the catalytic constant and of 5.5 for the substrate affinity constant (49). Therefore, lower K i values would be expected for the assayed mononucleotides if they are analyzed at a pH lower than 6.5. Moreover, when comparing the reported K i with previous kinetic analyses of EDN, it should be considered that the present structural and kinetic characterization has been performed with the recombinant protein, whereas the reported kinetic assays used the natural enzyme (50). Native EDN has several N-glycosylation sites (8) and is also C-mannosylated at Trp-7 (51). The reported inhibition constants of the same adenylic mononucleotides for RNase A are in the range of 1 to 8 M (52, 53), indicating that these nucleotides have a much higher affinity for RNase A. A 10-fold difference in the K i values has also been observed while studying the affinity of dinucleotides for either RNase A or EDN (50), in agreement with kinetic data that show a lower catalytic efficiency for the RNA transphosphorylation reaction with EDN (9).
EDN⅐Sulfate Complex-As reported by Mosimann et al. (13), two sulfate ions were found in the EDN structure occupying two distinct subsites (Fig. 4). The first sulfate, SO 4 2Ϫ (A), is bound at P 1 and interacts in both structures with the two catalytic histidines (His-15 and His-129), Gln-14, the amide nitrogen atom of Leu-130, and two conserved water molecules (Table VI). In addition, the sulfate ion in the present structure is stabilized by an additional hydrogen bond interaction with a water molecule (Fig. 4B). The second sulfate, SO 4 2Ϫ (B), was located in both structures at an additional binding site, not previously observed in RNase A. This site has been designated as P Ϫ1 on the basis of its proximity to the P Ϫ1 phosphate of (Tp) 4 upon superposition of the EDN⅐sulfate structure onto the RNase A-(Tp) 4 complex (54). In both structures the SO 4 2Ϫ (B) interacts with both the N1 and the N2 atoms of the side chain of Arg-36, and with the main-chain atoms of Asn-39 and Gln-40; there is also an interaction with a conserved water molecule ( Fig. 4D and Table VI). The only interaction described in the previous structure (13), which is not observed in the present structure, is the involvement of the side chain of Gln-40 with SO 4 2Ϫ (B). Indeed, the conformation of the Gln-40 side chain is different in each structure. Although Gln-40 interacts with the side chain of Asn-39, the SO 4 2Ϫ (B), and a water molecule in the 1.8-Å structure, the side chain of Gln-40 in the 1.6-Å structure is exposed to the solvent.
The ␣ torsion angle for 3Ј,5Ј-ADP and 2Ј,5Ј-ADP cannot be defined, because the position of atom O3A is ambiguous (the phosphate group is free).

DISCUSSION
All three ligands, 3Ј,5Ј-ADP, 2Ј,5Ј-ADP, and 5Ј-ADP bind to the catalytic site of EDN, each one in a significantly different manner. In the EDN⅐3Ј,5Ј-ADP complex, the 5Ј-phosphate binds at the P 1 site engaging in similar interactions with EDN to those observed for phosphate groups bound to RNase A subsite P 1 . This was anticipated, because subsite P 1 is conserved between RNase A and EDN. The adenosine binds in an identical manner to that observed for adenosines bound to the subsite B 2 of RNase A. Subsite B 2 is only partially conserved between EDN and RNase A; residues Asn-67, Gln-69, Asn-71, and Glu-111 in RNase A are replaced by residues Asn-65, Arg-68, Asn-70, and Asp-112 in EDN. Atoms N6 and N1 of the adenosine of 3Ј,5Ј-ADP form hydrogen bonds with Asn-70, and the adenine ring is engaged in stacking interactions with the imidazole of His-129. All these interactions are very similar to those observed for the adenosine bound to B 2 in the RNase A complex with d(ApTpApApG) (44), d(CpA) (37), and 2Ј,5Ј-CpA and 3Ј,5Ј-d(CpA) (45). However, the replacement of RNase A residue Gln-69 by Arg-68 in EDN restricts EDN from forming an additional hydrogen bond with adenine as seen in the RNase A⅐d(ApTpApApG) complex (44). In addition, the substitution of a Gln residue by Arg seems to force the adenine ring to occupy a slightly different position. The hydrogen bond between Arg-68 and Asn-65 observed in the sulfate-bound structure is not present in the 3Ј,5Ј-ADP and 2Ј,5Ј-ADP complexes, where the Arg-68 side chain can shift toward the base. The position of the 3Ј-phosphate of 3Ј,5Ј-ADP indicates a possible location of subsite P 2 in EDN. In RNase A, Lys-7 is the main   The 3Ј-phosphate group forms a hydrogen bond with the mainchain nitrogen of the N-terminal residue Met-0 (an interaction that should not exist in the natural EDN because Met-0 is an addition due to the expression system) and a water-mediated interaction with the side chain of Trp-10. In addition, this phosphate group forms several van der Waals contacts with Trp-10. Therefore, it appears that Trp-10 is the sole component of EDN subsite P 2 and contributes only through nonpolar interactions to substrate binding. The observed binding mode of 2Ј,5Ј-ADP to EDN was not anticipated on the basis of the 3Ј,5Ј-ADP binding to EDN. In the EDN⅐2Ј,5Ј-ADP complex, it is the 2Ј-phosphate that binds to the P 1 subsite, whereas the 5Ј-phosphate points toward the N terminus and is involved only in water-mediated interactions with the N-terminal residue Met-0. In addition, there are no interactions between this phosphate group and Trp-10 (the sole component of subsite P 2 ), which is 6.9 Å away. The adenosine binds to the purine binding subsite in a different mode to that observed in previous RNase A⅐nucleotide complexes (37-39, 44, 45). The adenine ring is in a different orientation and does not pack against the imidazole ring of His-129 due to steric hindrance. In doing so, His-129 adopts conformation B (as described for RNase A His-119, the inactive conformation). On the other hand, the 2Ј-phosphate group of 2Ј,5Ј-ADP forms hydrogen bond interactions with EDN at the main phosphate active site P 1 similar to those observed for the 5Ј-phosphate of 3Ј,5Ј-ADP. From the present study, we are unable to explain why it is the 2Ј-phosphate instead of the 5Ј that binds to subsite P 1 of EDN. Modeling of the 2Ј,5Ј-ADP molecule onto the 3Ј,5Ј-ADP structure revealed that the EDN active site could easily accommodate 2Ј,5Ј-ADP with the 5Ј-phosphate group in P 1 and the 2Ј-phosphate pointing toward the solvent. This mode of binding might also bring the adenine moiety to form stacking interactions with His-129. However, the binding mode of 2Ј,5Ј-ADP to EDN shows the flexibility of the B 2 site in EDN, the ability of the P 1 site to accommodate either a 5Ј-or a 2Јphosphate and the low affinity for phosphate anions of subsite P 2 noted above.
In the EDN⅐5Ј-ADP complex it is the ␤-phosphate rather than the ␣-phosphate that occupies the P 1 subsite while the adenosine binds close to the EDN pyrimidine binding subsite B 1 . The ␤-phosphate engages in interactions similar to those made by the phosphate groups and sulfate ion located at P 1 as seen in the other two EDN complexes described above. In addition, the ␣-phosphate group forms a hydrogen bond with Lys-38. This mode of binding for the pyrophosphate group has been observed previously in RNase A complexes. In previous structural studies of RNase A with three potent nucleotide inhibitors that had a 5Ј-pyrophosphate group (38,39), the ␤instead of the ␣-phosphate was found to bind to subsite P 1 . That mode of binding drove the adenosine to adopt the syn instead of the anti conformation with the six-instead of the five-membered ring of the adenine stacking against the imidazole ring of His-119. However, in EDN this does not occur. Instead, the binding of the ␤-phosphate group in the P 1 position forces the adenosine out of B 2 to a new location. Thus in this complex the binding of the adenosine to EDN is stabilized through two hydrogen bonds made by the ribose with Gln-40 and through van der Waals interactions of the adenine ring with Lys-38, Gln-40, and His-82.
EDN is one of a relatively small array of proteins that are C-mannosylated which involves the attachment of an ␣-mannosyl residue via a C-C link to the indole moiety of the first tryptophan in the recognition sequence of Trp-X-X-Trp (55,56). The site of mannosylation is Trp-7, and the modification is absent in the recombinant protein (enzymatically active) used in this study (23). There is much interest in this novel postbiosynthetic modification but its role in structure and/or activity is not yet known. In the recombinant EDN⅐nucleotide complexes presented here, Trp-7 makes van der Waals interactions with the nucleotide in both 3Ј,5Ј-ADP and 5Ј-ADP complexes at the active site of the protein. Based on these structural observations, we predict that these interactions will also be present in the natural protein. It is quite likely that the mannosyl residue(s) is(are) positioned on the opposite face of the active site and may not have a direct role on inhibitor binding to the protein. This hypothesis can only be tested through structural study of one or more of these inhibitors with the C-mannosylated protein.
The EDN sulfate structure has confirmed the position of the two sulfate molecules at P 1 and P Ϫ1 sites. The presence of an additional anchoring site at P Ϫ1 was also proposed for ECP (a close homolog of EDN) both by structural analysis (15) and kinetic studies (16). Superpositions of the structures of the EDN⅐3Ј,5Ј-ADP, EDN⅐2Ј,5Ј-ADP, EDN⅐5Ј-ADP, and EDN⅐sulfate complexes (Fig. 5) reveal that the positions and the orientations of all three phosphate groups and sulfate A at the P 1 subsite are almost identical. All these groups participate in a similar hydrogen bond pattern with EDN and seem to be the anchoring point for each inhibitor. The binding of the adenosine is optimal only in the case of 3Ј,5Ј-ADP, because this is the only inhibitor where the adenosine ring forms hydrogen bonds with EDN and binds in the purine binding subsite. The positions of the adenosine moiety in the other two complexes seem to be dictated by the binding of the phosphate in P 1 . To optimize their interactions, each ligand adopts a different conformation, indicating that diversity in ligand binding can be achieved with subtle modifications to the parent ligand molecule mainly through the flexible side chain of His-129. Examples where analogous inhibitors adopt different binding mode/s are well documented (57). For example, 2-deoxy-D-glucose 6-phosphate and D-glucose 6-phosphate, which differ only in one hydroxyl group, bind in a totally different manner to glycogen phosphorylase (58).
The analysis of EDN⅐ligand complexes and the comparison with the new sulfate bound EDN at higher resolution (1.6 Å) allowed the identification of some of the key residues implicated in EDN substrate binding. The residues implicated in the main phosphate active site P 1 have been confirmed, and we have been able to analyze the B 2 site environment and its flexibility for different adenylic mononucleotides. Furthermore, we have confirmed that Trp-10 is the sole component of subsite P 2 . Although the mononucleotides interact with the enzyme in quite different orientations, some common features are observed. In all three complexes, one of the phosphates is invariably located at P 1 , even when the adenine does not directly interact with the protein (e.g. 5Ј-ADP complex) and adopts a completely different orientation. We can conclude that the interactions at P 1 are the main driving force for all the observed nucleotide binding analyzed. Interactions at P 1 are conserved for the three adenylic nucleotide complexes and the sulfate complex, as is the case in the well documented RNase A⅐nucleotide complexes. On the other hand, the position of the adenine base is considerably different for each complex. A position of the adenine analogous to the substrate interaction is only feasible in the 3Ј,5Ј-ADP complex. The comparative analysis of EDN complexes and the sulfate-bound structure have allowed the identification of the residues directly involved in the ligand interaction. We can therefore conclude that adenine binding in EDN mainly involves ribonucleolytic active site residues Asn-70 and His-129.
The EDN⅐inhibitor complexes presented here suggest ways for further rational design of tight binding inhibitors of this enzyme against pathological conditions associated with eosinophil RNases. It also highlights that subtle alterations in the chemical structure of an inhibitor can generate significant changes upon binding to the protein. Thus the process of rational design may not follow a predictable course. However, the observations presented here emphasize the importance of crystal structure analysis intertwined with modeling studies toward achievement of significant enhancement in potency in the inhibitor design process. Finally, the binding mode of nucleotides with EDN should also prove useful for the design of inhibitors of other biologically active RNase superfamily enzymes.