Antithrombin-binding Octasaccharides and Role of Extensions of the Active Pentasaccharide Sequence in the Specificity and Strength of Interaction

The antithrombotic activity of low molecular weight heparins (LMWHs) is largely associated with the antithrombin (AT)-binding pentasaccharide sequence AGA*IA (GlcNNAc/NS,6S-GlcA-GlcNNS,3,6S-IdoUA2S-GlcNNS,6S). The location of the AGA*IA sequences along the LMWH chains is also expected to influence binding to AT. This study was aimed at investigating the role of the structure and molecular conformation of different disaccharide extensions on both sides of the AGA*IA sequence in modulating the affinity for AT. Four high purity octasaccharides isolated by size exclusion chromatography, high pressure liquid chromatography, and AT-affinity chromatography from the LMWH enoxaparin were selected for the study. All the four octasaccharides terminate at their nonreducing end with 4,5-unsaturated uronic acid residues (ΔU). In two octasaccharides, AGA*IA was elongated at the reducing end by units IdoUA2S-GlcNNS,6S (OCTA-1) or IdoUA-GlcNNAc,6S (OCTA-2). In the other two octasaccharides (OCTA-3 and OCTA-4), AGA*IA was elongated at the nonreducing side by units GlcNNS,6S-IdoUA and GlcNNS,6S-GlcA, respectively. Extensions increased the affinity for AT of octasaccharides with respect to pentasaccharide AGA*IA, as also confirmed by fluorescence titration. Two-dimensional NMR and docking studies clearly indicated that, although elongation of the AGA*IA sequence does not substantially modify the bound conformation of the AGA*IA segment, extensions promote additional contacts with the protein. It should be noted that, as not previously reported, the unusual GlcA residue that precedes the AGA*IA sequence in OCTA-4 induced an unexpected 1 order of magnitude increase in the affinity to AT with respect to its IdoUA-containing homolog OCTA-3. Such a residue was found to orientate its two hydroxyl groups at close distance to residues of the protein. Besides the well established ionic interactions, nonionic interactions may thus contribute to strengthen oligosaccharide-AT complexes.

are located at different sites along the oligosaccharide chains (7,8). The increasing interest in the development of "tailored" LMWHs and very low molecular weight heparins stimulates studies aimed at a better understanding at the molecular level the mechanisms of interaction between AT and AGA*IA-containing oligosaccharides. Earlier studies on tetrasaccharides sequences adjacent to the antithrombin-binding site have demonstrated two possible variants of AT-binding sequences, suggesting a possible role of the extensions of these sequences on binding to AT (9). Longer AT-binding sequences, such as decasaccharides, were also previously isolated (8). The influence of the position of the pentasaccharide sequence along the oligosaccharide chains together with the knowledge of the role of the residues prolonging the active sequence toward both its reducing and nonreducing side are among the major goals of current heparin research (10). Although the active pentasaccharide AGA*IA is taken as paradigm for a unique heparin sequence targeting a specific protein (i.e. AT) (3), different mechanisms have been proposed for its interaction with AT in terms of position and conformation of sugar residues. The possibility of a shift along the AT D-helix for sequences longer than pentasaccharide was taken into consideration (11). Independent crystallographic and NMR studies on the structure of complexes of AT with AGA*IA and AGA*IA-containing oligosaccharides suggested that the position of the pentasaccharide in the protein binding region is unique (10,(12)(13)(14)(15). These studies provided information on both the ring conformation of the monosaccharide residues and the geometry of the glycosidic linkages of the AT-bound pentasaccharide. In particular, it was shown that the 2-O-sulfated iduronic acid residue in the pentasaccharide, which in the free state in water solution is in equilibrium between two equienergetic conformations ( 1 C 4 , and 2 S O ) (16), adopts the 2 S O conformation when AGA*IA is bound to AT. Shifting toward this conformation, facilitated by the presence of the 2-OSO 3 group, enhances the contacts between the AGA*IA and basic amino acid residues in the AT binding region (15).
In this study, four high purity octasaccharides isolated by size exclusion and AT-affinity chromatography from the LMWH enoxaparin were selected. Like all fragments generated by ␤-elimination cleavage of heparin chains (7,17), all four octasaccharides terminate at the nonreducing end with 4,5unsaturated uronic acid residues (⌬U). In two octasaccharides, AGA*IA was found to be elongated toward the reducing end by the disaccharide units IdoUA 2S -GlcN NS,6S (OCTA-1) and IdoUA-GlcN NAc,6S (OCTA-2). In the other two octasaccharides (OCTA-3 and OCTA-4), AGA*IA was found to be elongated toward the nonreducing end by GlcN NS,6S -IdoUA and GlcN NS,6S -GlcA units, respectively. Earlier NMR studies on the interaction of OCTA-1 and OCTA-3 with AT suggested a possible role of both the reducing and nonreducing end extensions in favoring binding to the protein, and supported a specific binding between the pentasaccharide and the AT-binding site (10). In this work the interaction of OCTA-1 and OCTA-3 with AT was analyzed in greater detail, and the study was extended to the two novel octasaccharides (OCTA-2 and OCTA-4) described above. The structures of the four octasaccharides are shown in Fig. 1. Affinity chromatography on immobilized AT showed the following relative binding strength: OCTA-3 Ͻ OCTA-1 Ͻ OCTA-2 Ͻ Ͻ OCTA-4. The highest affinity of OCTA-4 was confirmed also by fluorescence titration experiments. Furthermore, when this measurement is performed in 0.5 M NaCl (i.e. at the same ion strength used for the NMR studies; see "Experimental Procedures"), OCTA-4 was shown to bind AT with 1 order of magnitude higher affinity than its homolog OCTA-3. Saturation transfer difference (STD) experiments confirmed the specificity of the AGA*IA sequence for the AT binding. The conformational and AT binding properties of these octasaccharides were also investigated by NMR (transferred-NOESY) spectroscopy and docking simulations. The structural properties of the four octasaccharides have been correlated with the affinity to AT determined by affinity chromatography on a preparation that contains about 95% of active protein as judged by active site titration (18) and interpreted in terms of both ionic and nonionic interactions.
General Procedure for Octasaccharide Isolation, Purification, and Sequencing-Octasaccharides 1-4 ( Fig. 1) were obtained by combining AT affinity chromatography and cetyltrimethylammonium-strong anion-exchange (CTA-SAX) chromatography on a semi-preparative scale, starting from octasaccharide gel permeation chromatography (GPC) fractions of enoxaparin. GPC of enoxaparin and the desalting conditions of the selected fractions were performed as described previously (19). The octasaccharide fraction was chromatographed on an AT-Sepharose column (40 ϫ 5 cm) with a stepwise gradient of NaCl. The column was prepared by coupling human AT (1 g) to CNBr-activated Sepharose 4B (Sigma) according to Höök et al. (20). The low affinity portion was eluted from the column with a 0.25 M NaCl solution buffered at pH 7.4 with 1 mM Tris-HCl at 6 ml/min. The high affinity octasaccharide fractions were eluted with a step gradient of NaCl (range between 0.25 and 3 M NaCl, 1 mM Tris-HCl, pH 7.4). The NaCl gradient was monitored by conductivity measurements, and the octasaccharides in the effluents were detected by UV at 232 nm. Octasaccharides eluted in affine fractions with conductivities between 30 and 85 mS/cm were gathered, desalted on Sephadex G-10, and used as starting material for the purification of OCTA-3. Octasaccharides eluted between 85 and 115 mS/cm were used for the purification of OCTA-1. Octasaccharides eluted between 115 and 150 mS/cm were gathered and used to purify OCTA-2. Fractions eluted for conductivities over 145 mS/cm were used after desalting to purify OCTA-4. The final purification of all the octasaccharides was achieved using CTA-SAX chromatography. CTA-SAX semi-preparative columns (25 ϫ 5 cm or 25 ϫ 2.2 cm) were prepared as described in Ref. 19 and filled with Hypersil BDS C18 (5 m particle size). Mobile phases for oligosaccharide separation were aqueous sodium methanesulfonate (Interchim) at concentrations varying between 0 and 2.5 M. The pH was adjusted to 2.5 by addition of diluted methanesulfonic acid. Separations were achieved at 40°C. Salt concentration in the mobile phase was increased linearly from 0 to 2.5 M over 60 min. Flow rate was 40 ml/min for 25 ϫ 3-cm columns, and UV detection at 234 nm was used. Collected fractions were neutralized and desalted on Sephadex G-10 after a preliminary treatment on Mega Bondelut C18 cartridges (Varian). Sequencing of the octasaccharides was performed by a combination of controlled and exhaustive cleavage with heparitinases and HPLC analysis of fragments as reported previously (19).
Separation of the Octasaccharide High Affinity Fractions and Fractionation with Ion-exchange Chromatography-An octasaccharide mixture, obtained by solubilizing a mixture of about 500 g of each crude octasaccharide, was analyzed by CTA-SAX to obtain the chromatographic T ϭ 0 prior to the affinity experiment (data not shown). The mixture of the four octasaccharides was injected on an AT-Sepharose column (40 ϫ 1.6 cm). The low affinity portion was eluted from the column with a 62.5 mM NaCl solution buffered at pH 7.4 with 1 mM Tris-HCl, and the high affinity octasaccharide fraction was eluted with a step gradient of NaCl (in a range between 0.21 and 3 M NaCl in 1 mM Tris-HCl, pH 7.4), with flow rate of 1 ml/min. Fractions (7 ml) were sampled and injected at 1 ml/min on a Carbopack AS11 column (25 ϫ 0.21 cm) (Dionex) in 2.5 mM NaH 2 PO 4 , pH 2.8 buffer. Bound oligosaccharides were eluted with a linear gradient of NaClO 4 (up to 0.6 M). Double UV detection was monitored at 232 nm and at 202-247 nm. The N-acetylated oligosaccharide selective signal, 202-247 nm, is the subtraction of UV signal 247 from 202 nm, as described previously (19).
Fluorescence Titration-The equilibrium dissociation constant K d for the interaction between AT and the four octasaccharides, or fondaparinux, was assessed at 25°C in 0.05 M HEPES containing 0.5 M NaCl, the salt concentration used in NMR experiments. K d was obtained by monitoring the enhancement of the intrinsic fluorescence of the serpin upon its reaction with increasing concentrations of the products, a procedure often used (21)(22)(23)(24)(25). The same procedure was also used to determine K d for the OCTA-4 or fondaparinux and OCTA-4:AT pairs in the presence of 0.1 and 0.25 M NaCl.
The fluorescence intensities measurements ( ex ϭ 280 nm, em ϭ 340 nm) were performed using an RF5000 Shimadzu fluorospectrophotometer equipped with a thermostated cell holder. A typical fluorescence equilibrium titration experiment was done as follows: an aliquot of the buffered AT stock solution was diluted to the appropriate final concentration (0.042, 0.103, or 0.276 M) into the reaction buffer contained in a 1 ϫ 1-cm path quartz cuvette (final volume ϭ 2.5 ml), and the fluorescence intensity was read at 340 nm prior to successive additions of small volumes (0.5-4.0 l) of the studied oligosaccharide solution. Before addition of AT, the apparatus was set to zero by reading against the buffer alone. The titrant concentrations were corrected for dilution when the cumulative volume of the aliquots was Ͼ35 l (about 1.4% of the total volume). The best estimate of K d was determined by fitting the fluorescence data to the quadratic equilibrium binding Equation 1, where [AT] 0 and [ols] 0 stand for the AT and oligosaccharide initial concentration, respectively. ⌬f and ⌬f max are the absolute change of fluorescence intensity (f Ϫ f 0 ) for a given oligosaccharide concentration and that of the maximum fluorescence intensity change ((f Ϫ f 0 ) max ), respectively. n is the binding stoichiometry (the ratio of the concentration of the high affinity species present in the reaction mixture versus the total concentration of the studied product). Here the value of this parameter was set to 1 because the analyzed solutions were considered to contain a single product, each molecule of which is capable of binding to one molecule of AT. Nonlinear regression analysis was done with the GraFit software (Erithacus Software).
NMR Spectra-All one-and two-dimensional NMR spectra were measured at 35°C, at 600 MHz with a Bruker Avance 600 spectrometer equipped with a high sensitivity 5-mm TCI cryoprobe. To reduce the water humping, particularly strong in cryoprobes, 3-mm NMR tubes were used instead 5-mm tubes reducing the volume from 0.6 to 0.2 ml. For proton detection, between 150 and 250 g of octasaccharide samples were dissolved in 2 H 2 O (99.9%) and freeze-dried to remove residual water. After exchanging the samples three times, samples were dissolved in 0.2 ml of 10 mM phosphate buffer (0.5 M NaCl, pH 7.4) with 3 mM EDTA in 2 H 2 O (99.996%). For the binding studies, samples were prepared by dissolving 1 mg of AT and 150 -250 g of each octasaccharide sample in the phosphate buffer so as to reach a 1:3.5 AT/octasaccharide molar ratio (10). In OCTA-1 some spectra were needed to be repeated with a smaller AT/octasaccharide ratio (1:5), to reduce the signal overlapping affecting the quantitative analysis.
Proton spectra were recorded with presaturation of the residual water signal, with a recycle delay of 12 s and 256 scans. Bidimensional double-quantum filter correlation spectroscopy (DQF-COSY) and two-dimensional TOCSY spectra were acquired using 32 scans per series of 2 K ϫ 512 W data points with zero filling in F1 (4 K ϫ 2 K), and a shifted (/3) squared cosine function was applied prior to Fourier transformation. All two-dimensional nuclear Overhauser enhancement spectroscopy (NOESY) and two-dimensional transferred NOESY experiments were performed in a similar way. A total of 48 scans was collected for each free-induction decay (matrix 2048 ϫ 512 points) and data were zero-filled to 4 K ϫ 2 K points before Fourier transformation. Mixing time values of 100, 200, and 300 ms were used.
One-dimensional STD Experiments-Samples were prepared dissolving octasaccharides in the same buffer described previously. Between 0.4 and 0.5 mg of each sample was dissolved in a 3.4 -4.2 ϫ 10 Ϫ6 mM protein solution reaching a ligand/AT ratio of 50:1. The AGA*IA/AT sample was prepared dissolving 0.5 mg of pentasaccharide in a 8.9 ϫ 10 Ϫ6 mM protein solution, maintaining the same ligand/AT ratio.
The pulse sequence used for the monodimensional STD NMR experiments includes a 30-ms spin-lock pulse to eliminate the broad resonances of the protein. A train of 40 Gaussian-shaped pulses of 50 ms each was applied to produce selective saturation. Because H5 of IdoUA residues shifts very close to water, signal solvent suppression was not included in the pulse sequence. The on-resonance irradiation was performed at the low field protein resonances (Ϸ7.2 ppm), whereas the off-resonance control irradiation was performed at 24 ppm. The STD spectrum was obtained by phase cycling subtraction of the on-resonance and off-resonance data acquired in interleaved mode. The number of scans and dummy scans were 2048 and 16, respectively.
Computational Studies on Octasaccharide-AT Complexes-Flexible docking calculations of octasaccharides 1-4 onto AT were performed by AutoDock 3.0 program (26), following the procedure described for rigid docking (10). Each simulation was performed using 30 genetic algorithm runs and 3000 generations for each run. Eight torsions were allowed to move (see "Results"). Resulting ensembles of 30 conformations were then clustered using a root mean square deviation tolerance of 0.5 Å.
Theoretical tr-NOEs were computed on selected octasaccharide models by the CORCEMA program (27). All AT protons within 15 Å from ligands were included in the calculation, as possibly interacting with ligand protons.
For each octasaccharide, the experimental K d value measured in a 0.5 M NaCl solution was used for CORCEMA simulation to compare theoretical tr-NOEs with experimental ones at the same buffer solution. k off values were then estimated to obtain the best agreement with experimental NOEs (k off OCTA-1, OCTA-2, and OCTA-3, 8 s Ϫ1 ; k off OCTA-4, 6 s Ϫ1 ).
A ligand/protein ratio of 3.5:1 was set for OCTA-2, -3, and -4, and a ratio of 5:1 was set for OCTA-1. The correlation time of AT was considered as isotropic and estimated from published data ( AT , 46 ns) (15), whereas an average correlation time value characteristic of oligosaccarides having similar structure was used for octasaccharides ( octasaccharide , 0.9 ns) (28).
The fitting between experimental and theoretical NOEs was evaluated by computing R factors according with the Equation 2, where I ij exp indicates the experimental cross-peak intensities; I ij calc indicates the calculated cross-peak intensities, and mix indicates the mixing times.
OCTA-2 and OCTA-4 structures ( Fig. 1) were confirmed by NMR spectroscopy. Both proton (Table 1) and carbon resonances (supplemental Table S1) were assigned by one-dimensional and two-dimensional NMR homonuclear and heteronuclear spectra ( 1 H, DQF-COSY, TOCSY, and HSQC, data not shown). Proton and carbon chemical shifts are in full agreement with the proposed structures (Fig. 1). Expansions of proton spectra are shown in supplemental Fig. S1 and S2.
Relative Affinities for AT-To determine the relative affinities for AT of OCTA-1 to OCTA-4, a mixture of high affinity octasaccharides were analyzed by CTA-SAX HPLC. As illustrated in Fig. 2, OCTA-1 and -3 co-eluted in the beginning of the NaCl step gradient. However, the fraction with conductivity of 111.4 mS/cm contains only traces of OCTA-3. This fraction also contains significant amounts of OCTA-2, which was only completely eluted at conductivities between 149 and 176 mS/cm. At 149 mS/cm, OCTA-4 began to be eluted from the AT column, and its presence was still observed at 214.4 mS/cm, which corresponds to about 3 M NaCl. The final affinity order in these experiments was OCTA-3 Ͻ OCTA-1 Ͻ OCTA-2 Ͻ Ͻ OCTA-4. The affinity of OCTA-4 toward AT was unexpectedly strong.
K d Values for Oligosaccharide-AT Complexes-The interaction of the four octasaccharides and fondaparinux, the reference compound, with AT was studied by equilibrium titrations. The substantial increase of the intrinsic protein fluorescence (Ϯ33.3%) that resulted from complex formation was used as the signal. K d was determined from the experimental data sets (oligosaccharide concentration: fluorescence intensity pairs) by a least square fit based on Equation 1 (data not shown). The value of K d and its standard error for each AT-oligosaccharide complex and for the various NaCl concentrations are summarized in Table 2. The binding of fondaparinux to AT was investigated in the presence of 0.1, 0.25, and 0.5 M NaCl yielding K d values of 0.024, 0.242, and 4.75 M, respectively, a salt concentration dependence substantially in agreement with the data reported by Olson et al. (22). In the presence of 0.5 M NaCl (the salt concentration used in the NMR studies), OCTA-1, OCTA-2, and OCTA-3 bound the serpin with a moderately higher affinity than fondaparinux (Table 2), whereas OCTA-4 bound AT with a 20-fold lower K d (0.24 M). This latter interaction was also investigated in the presence of 0.1 and 0.25 M NaCl. We found K d values of 1.5 and 13.0 nM, respectively, here again 2 values lower by more than 1 order of magnitude than those measured for the binding of fondaparinux to AT ( Table 2).
The plots of log K d versus log[Na ϩ ] shown in Fig. 3 illustrate the Na ϩ concentration dependence of OCTA-4 and fondaparinux-AT interactions. Assuming that fondaparinux and OCTA-4 behave as polyelectrolytes in solution, we analyzed the effect of Na ϩ concentration on their binding to AT according to the theory of macromolecule-polyelectrolyte interactions. This approach, used by Record et al. (29) to investigate ligand nucleic acid interaction in the presence of monovalent ions, was also used by others to investigate the effect of the salt concentration on the binding properties of heparin or pentasaccharide to antithrombin (22,30,31), thrombin (32), fibroblast growth factor (33), or peptide (34). Complex formation between a polyelectrolyte with bound counterions (Na ϩ ) such as an oligosaccharide (fondaparinux or OCTA-4) and a protein (AT) is accompanied by an entropically favorable release of Na ϩ from the oligosaccharide chain by cationic residues located within the binding site of the protein. Although this process (the polyelectrolyte effect (29)) accounts for the ionic component of the global interaction of AT with the oligosaccharide, its nonionic component results from hydrogen bonding and/or hydrophobic interactions. For these systems the whole interaction is described by Equation 3, which relates the observed equilibrium dissociation constant K d to the Na ϩ concentration (29), where K dn-i is the equilibrium dissociation constant characterizing the nonionic component at 1 M NaCl; Z is the number of ionic interactions (or ion pairs) formed between oligosaccharide and AT, and is the fraction of Na ϩ counterion bound to oligosaccharide per unit of charge. , a parameter related to the axial charge density of the polyelectrolyte, was determined to be 0.8 for heparin (30,32). Equation 3 predicts a linear depend- ence of log K d on log [Na ϩ ], and Z can thus be derived from the slope of the plot, whereas log K dn-i represents its intercept with the y axis. From the theoretical straight lines generated by linear regression using the data of Fig. 3, we found that between 4 and 5 ionic interactions are involved in the binding to AT of OCTA-4 (Z ϭ 4.7 Ϯ 0.6). A similar result was found for fondaparinux, the reference product (Z ϭ 4.5 Ϯ 0.6), in agreement with the literature (22). From the y intercepts of the plots, we found that the nonionic contribution to the binding of fondapariux to AT is characterized by K dn-i ϭ 42.6 M, in reasonable agreement with published data (31). In contrast, K dn-i ϭ 1.8 M for the interaction of OCTA-4 with AT, a 22-fold lower value than for fondaparinux, indicates that the nonionic contributions to the protein binding are significantly enhanced in the case of OCTA-4. Due to K d values for the complex of OCTA-4 with AT ϭ 1.5 nM at 0.1 M NaCl and pH 7.4, we calculated from the ratio log K dn-i /log K d that the nonionic interactions account for about 65% of the total free energy of binding in these conditions. Molecular Conformation of the Oligosaccharides-As reported previously for OCTA-1 and OCTA-3 (10), the conformation of OCTA-2 and OCTA-4 in buffer solution was deter-mined by analysis of 3 J H-H (three-bond proton-proton coupling constants) and NOEs. 3 J H-H couplings measured by one-dimensional 1 H spectra (Table 1) indicated that all glucosamine and glucuronic acid residues were present in aqueous solution in the 4 C 1 conformation. In contrast, the conformations of the unsaturated terminal uronic acid residues (⌬U and ⌬U 2S ) were influenced by 2-O-sulfation. Their measured 3 J H-H values are consistent with a preferred 2 H 1 half-chair conformation for ⌬U residue of OCTA-2 and a preferred 1 H 2 half-chair conformation for ⌬U 2S residue of OCTA-4 (35), similarly to what was observed for OCTA-1 and OCTA-3, respectively (10).  be related to the relative percentage of the two conformers. An H5-H2 NOE having smaller intensity than its corresponding H5-H4 NOE was measured for the I moiety in both OCTA-2 and OCTA-4 ( Table 3), indicating that this residue is present in solution in equilibrium between the 1 C 4 and 2 S O conformations. On the contrary, no H5-H2 correlation was detected for IЉ in OCTA-2, confirming that this moiety is present in solution in a pure 1 C 4 form. Iduronic acid residue regions of the two-dimensional NOESY spectra of OCTA-2 and OCTA-4 are shown in supplemental Fig. S3.
STD Experiments-To identify the ligand epitope binding, one-dimensional STD experiments (36,37) were performed on both the pentasaccharide fondaparinux and the four octasaccharides in the presence of AT (about 50-fold excess of the ligand). The STD spectra in comparison with their corresponding reference spectra are shown in Figs. 4 and 5. The STD signals of AGA*IA residues are the most intense ones, whereas signals of residues belonging to the reducing and nonreducing end extensions are weaker or almost disappear from the spectra. Because H2 of the trisulfated glucosamine (GlcN NS,3,6S ) residue is not affected by signal overlapping in all the spectra, it was chosen as reference peak. The STD intensity of anomeric and H2 signals of AGA*IA ranges from 80 to 120% in all octasaccharides. The relative STD intensity of each AGA*IA signal remains constant independently from the structure of the octasaccharides (e.g. it is 90 -100% for both H1 of I and H2 of G) (Fig. 6), as expected when their pentasaccharide sequence is located in the same position within the AT-binding site.
Relatively intense signals (40 -60%) were observed also for H1 and H4 of the unsaturated uronate residue in the STD spectrum of the OCTA-1, indicating that this residue is close to the binding region. On the contrary, both H1 and H2 of AЉ show weak STD signals, indicating a larger distance of this residue

600-MHz proton chemical shifts of OCTA-2 (top) and OCTA-4 (bottom) residues
Proton spectra were measured at T ϭ 35°C in 10 mM phosphate buffer, pH 7.4, and 0.5 M NaCl. Standard errors, 1    from AT. Moreover, the H1 signal of IЉ is almost absent in the STD spectrum, whereas the H5 signal of the same residue is relatively intense, suggesting that this proton is oriented toward the protein surface. Because signals of the ␣and ␤-forms of the reducing N-acetylglucosamine of OCTA-2 overlap with the anomeric signal of I and the H5 of IЉ, respectively, it is not possible to define the position of this residue relative to the AT-binding site. As observed for OCTA-1, the presence of signals of H4 and H1 of the ⌬U residue indicates that this residue receives a part of the saturation because of its proximity to the binding region. An opposite situation was observed for OCTA-3 and OCTA-4, where very weak signals belonging to the nonreducing end disaccharides (H1/H4 of ⌬U 2S and H1/H2 of AЈ) indicate that this part of the oligosaccharide chains is not located in the proximity of the binding region. On the contrary, the STD intensities of H2 of GЈ of the OCTA-4 and H1 of IЈ of the OCTA-3 suggest a contribution of these residues to the binding. The STD signal of H2 of GЈ suggests an orientation of the corresponding residue having hydroxyl groups oriented toward AT.

Conformations of AT-bound Octasaccharides-Similarly
to what was observed for OCTA-1 and OCTA-3 (10), the 1 H NMR spectra of OCTA-2 and OCTA-4 in their complexes with AT, compared with the corresponding ones in the free state (supplemental Figs. S1 and S2), show small shifts of the proton res-   Tables S2-S5). Notably, only a weaker increment is detectable in OCTA-1 because the quantitative analysis of its transferred-NOESY spectra needed to be performed using a smaller protein/ ligand ratio (see "Experimental Procedures"). Iduronic acid conformations were investigated by analyzing intra-residue tr-NOE effects (supplemental Fig. S3) because the increased line width of bound spectra does not allow measurement of 3 J H-H . A significant enhancement of H5-H2/H5-H4 NOE ratio of the I residue was observed for both OCTA-2 and OCTA-4. This indicates that the conformation of such moiety is driven toward the 2 S O form by the presence of AT, as observed in all AGA*IAcontaining oligosaccharides so far described (10,12,15). In contrast, the IЉ residue of OCTA-2 does not show an H5-H2 tr-NOE signal, indicating that IЉ maintains its 1 C 4 conformation in the presence of AT. In OCTA-4, GЈ maintains its 4 C 1 conformation as confirmed by its 3 J H1-H2 that assumes the same value measured in the free state (8 Hz).

Computational Studies on Octasaccharide-AT Complexes-
The earlier conformational studies on OCTA-1 and OCTA-3 in their complex with AT performed by rigid docking simulations were refined in the present work by performing flexible runs and extended to OCTA-2 and OCTA-4. Ring conformations were set as indicated previously (10). Eight selected glycosidic torsions in each octasaccharide were allowed to move freely. On the basis of STD results, glycosidic linkage geometries of residues inside the AGA*IA sequence were taken as invariable in all the analyzed ligands and they were kept fixed during docking runs. Conversely, glycosidic torsions and (defined as: ϭ H1-C1-O1-C4; ϭ C1-O1-C4-H4) between the moieties external to AGA*IA (i.e. ⌬U-A, A-IЉ, IЉ-AЉ in OCTA-1 and OCTA-2; ⌬U 2S -AЈ, AЈ-IЈ/GЈ, IЈ/GЈ-A in OCTA-3 and OCTA-4) were allowed to move. The glycosidic linkage of the AGA*IA disaccharide moieties nearest to the extension (I-A in OCTA-1 and OCTA-2 and A-G in OCTA-3 and OCTA-4) was also allowed to move, because its geometry is expected to affect the conformation of the extension.
At least 8 of 30 AT-binding structures with AGA*IA in the same position found in the pentasaccharide-AT x-ray complex (12) were calculated by the docking simulation of each octasaccharide. From all the simulated docking oligosaccharide/protein ensembles, the structure that was able to better fit the experimental STD data was selected. In all analyzed ligands, the models maintaining AGA*IA in its original position show essentially the same distances between protons of the AGA*IA sequence and AT residues. In contrast, ligand-protein distances involving proton residues of both reducing and nonreducing end extensions are, on average, larger and different among the four octasaccharides (supplemental Table S6). These findings are in good agreement with experimental data, indicating that STD effects involving AGA*IA residues have similar magnitudes in all the analyzed octasaccharides, whereas smaller STD magnitudes were detected on proton signals of residues corresponding to both reducing and nonreducing end extensions.
In both OCTA-3 and OCTA-4, from all the simulated docking oligosaccharide/protein ensembles, two main clusters of structures can be identified. Models belonging to the first cluster show IЈ/GЈ residues with their 2-OH and 3-OH groups oriented toward the AT surface (Fig. 7). In models of the second cluster, the orientation of these groups is completely reversed by a drastic change of IЈ/GЈ-A glycosidic linkage, driving the carboxylic group toward AT surface (supplemental Fig. S4). Consequently, in the latter cluster, the orientation of ⌬U 2S and AЈ residues is also reversed. Comparison with STD data indicates that models of the first cluster are the only ones able to interpret experimental data. Indeed, such models account for interactions between AT and H2 and H1 of both GЈ/IЈ and the AЈ residues, whereas in models of the second cluster these protons are far from the AT surface.
The nonreducing end extension residues of OCTA-3 and OCTA-4 models belonging to the first cluster show essentially the same pattern of distances between their protons and AT residues (supplemental Table S6). The short distances between H2 of their A1Ј and Arg-132 and Lys-133 account for the occurrence of ionic interactions between the N-sulfate group of AЈ residue and such amino acids. Short distances between hydroxyl groups of GЈ/IЈ moieties and Lys-125, Lys-129, and Arg-132 can also be detected, suggesting that nonionic interactions involving these groups could occur in both octasaccharides (Table 4). Notably, the two octasaccharides differ only in the epimerization and conformation of the uronic acid preceding the AGA*IA sequence, which is that of GЈ glucuronic acid in OCTA-4 and IЈ iduronic acid in OCTA-3. The selected model of OCTA-3, having IЈ in skew-boat conformation, shows a slight increment of the distance from the protein surface of OH-2 and OH-3 of IЈ, as well as of H2 of A1Ј (Table 4 and supplemental Table S6).
The docking output models both OCTA-1 and OCTA-2, maintaining AGA*IA in its original position (Fig.  8) shows distances between H1 and H5 atom of IЉ moieties and the Glu-113 residue (about 7.0 Å) (supplemental Table S6) accounting for the small STD effects experimentally detected on signals of these protons. Moreover, shorter distances between H1 and H4 atoms of the ⌬U moiety and the AT surface were measured in these models and found to be in agreement with the observed STD effects. Both octasaccharides show OH-2 and OH-3 of the ⌬U residue close to Lys-125 and Arg-132, indicating possible nonionic interactions between these hydroxyl groups and AT (Table 4).
Theoretical tr-NOEs were computed on octasaccharide models by the CORCEMA program, evaluating full relaxation and exchange matrix. Following the procedure previously adopted for the AGA*IA pentasaccharide (15), we evaluated the agreement between theoretical and experimental tr-NOEs by calculating R factors. To evaluate the ability of our models to interpret the position of the ligand in the heparin-binding site of AT regardless of ligand internal conformation, tr-NOEs of H1-H2 protons cross-relaxing within glucosamine residues were first analyzed. Models in Figs. 7 and 8 showed a good fitting between theoretical and experimental intra-residue tr-NOEs (R factors ranged from 0.05 to 0.30; see Table 5). H1-H2 tr-NOEs were then also computed on a model of OCTA-1 moved up by a disaccharide unit from the normal AT-binding site (supplemental Fig. S5). This type of shifted structures was also found using rigid docking simulation of OCTA-1, and it corresponds to one possible docking to AT described for an heptasaccharide containing AGA*IA at its nonreducing end (11). However, tr-NOEs com-

TABLE 4 Selected ligand-protein distances (in Å) measured on models of OCTA-1, -2, -3 and -4/AT complexes obtained by flexible docking simulations (models are shown in Figs. 7 and 8)
Values are referred to distances between the oxygen atom of the indicated hydroxyl group and the heteroatom of the indicated AT residue nearest to such oxygen. Only distances smaller than 7 Å are shown. puted on this shifted model gave significantly higher R factors (from 0.23 to 0.40; see Table 5) with respect to those found for models preserving the AGA*IA original position. This confirms that the CORCEMA program can discriminate among differently shifted ligands and that, in all the analyzed octasaccharides, the models maintaining the AGA*IA sequence in its original placement best fit the experimental data.
Inter-residue tr-NOEs between protons cross-relaxing across the glycosidic bonds were also analyzed. Models in Figs. 7 and 8 give relatively good R factors ranging from 0.10 to 0.62 (supplemental Tables S2-S5). Such models thus need to be refined by further calculation, to reach a complete satisfactory agreement with the experimental data. However, because small variations of and glycosidic torsions give rise to strong variations of inter-glycosidic proton distances (and consequently in theoretical tr-NOE magnitudes), no dramatic change in octasaccharide geometry is expected to occur during the optimization. Therefore, the models so far obtained provide useful indications on the "true" octasaccharide geometry (Table 6).

DISCUSSION
In this work, the structure of two novel AT-binding octasaccharides (OCTA-2 and OCTA-4) isolated from enoxaparin have been characterized by sequence analysis and NMR spectroscopy. Their AT binding properties and molecular conformations in the absence and presence of AT have been compared with those of previously described OCTA-1 and OCTA-3 (10). In OCTA-4, the AGA*IA sequence is preceded by a glucuronic acid (GЈ) residue instead of iduronic acid that normally occurs in most common heparins and low molecular weight heparins (5).
Fluorescence titrations performed in 0.5 M NaCl indicated that at this salt concentration the affinity to AT of octasaccha-rides 1-4 is characterized by equilibrium dissociation constants 2-20-fold lower than those measured for the pentasaccharide-AT complex, OCTA-4 forming the tightest complex with the serpin-K d ϭ 0.24 M. Affinity chromatography on immobilized AT of an octasaccharide mixture indicated the following order of elution: OCTA-3 Ͻ OCTA-1 Ͻ OCTA-2 Ͻ Ͻ OCTA-4. These results confirmed that both reducing and nonreducing end AGA*IA extensions contribute to binding to AT. Moreover, they confirmed that the different affinity of AGA*IA-containing oligomers depends on the structure of these extensions.
Despite the strong evidence accumulated on the specificity of heparin-AT binding, (3,10,15), recent studies suggested other possible assemblies between the negatively charged heparin chains and AT (38,39). In these latter studies, the possibility has been considered that sulfated residues in heparin sequences, different from that of the AT-binding site, may activate AT through nonspecific interactions. The STD analysis carried out in this study, and for the first time applied to glycosaminoglycan-protein complexes, indicates that all the octasaccharide-AT complexes show the whole pattern of contacts identified in the pentasaccharide-AT complex, further supporting the specificity of binding. This study also suggests that, to a lesser extent, additional contacts involving reducing and nonreducing extensions of the AGA*IA sequence contribute to the binding. In fact, STD experiments indicate that the sequence AGA*IA lies closer to the AT-binding site than its reducing or nonreducing end extensions. All analyzed compounds showed either comparable STD effects on AGA*IA signals.
tr-NOEs experiments indicated also that the bound conformation of the AGA*IA sequences in the four octasaccharides  Fig. S5). The R factor is also given. (A NAc ϭ GlcN NAc,6S of AGA*IA sequence). NI, signal not integrated because of strong signal overlapping; ND, not determined R factor. was essentially the same as that assumed in the pentasaccharide, including the 2 S O skew-boat conformation for the I residue. tr-NOE analysis also indicated that in the presence of AT both the 2-O-sulfated residue IЉ in OCTA-1 (10) and the nonsulfated IЉ residue of OCTA-2 adopt the 1 C 4 conformation. These findings suggest that the 1 C 4 conformation of the iduronate residue located immediately after the AGA*IA sequence could enhance the AT affinity of AGA*IA containing oligosaccharides, either by optimizing contacts between AGA*IA and AT or promoting additional contacts involving the AGA*IA reducing extension. On the other hand, the IЈ residue of OCTA-3, despite lacking in 2-O-sulfation, is known to be driven to the 2 S O form by the presence of AT (10), whereas GЈ of OCTA-4 maintains its 4 C 1 conformation. We can thus speculate that 4 C 1 and 2 S O conformations of the GlcA and IdoUA residues, respectively, preceding the AGA*IA sequence could enhance the AT affinity of AGA*IA-containing oligosaccharides, whereas the 1 C 4 chair form does not optimize contacts with the protein.
Model outputs from flexible docking simulations, with the AGA*IA sequence maintaining the same position adopted in pentasaccharide/AT structures, can interpret both STD and tr-NOE data. In such models the distances between protons of the AGA*IA sequence and AT residues are quite similar, whereas ligand-protein distances involving residues at the extensions are, on the average, larger and different among the four octasaccharides (supplemental Table S6). Moreover, in all octasaccharide-AT complexes, glycosidic linkages within the AGA*IA sequence show essentially the same geometry, supporting the concept that this sequence interacts with AT with high specificity (Table 5). In all octasaccharide-AT complex models, short distances were found between uronic acid OH groups of both reducing and nonreducing end AGA*IA extensions and AT, accounting for the occurrence of nonionic interaction that could contribute to enhance the affinity of the octasaccharide with respect to the pentasaccharide ( Table 3).
The higher nonionic contribution to AT affinity of OCTA-4 (K dn-i ϭ 1.8 M), with respect to the pentasaccharide (K dn-i ϭ 42.6 M) (Fig. 3), can be attributed to the stronger nonionic interactions between AT and OH-2 and OH-3 groups of GЈ residue. Besides nonionic contacts with Lys-125 and Arg-129, the proposed model (Fig. 7) shows particularly short distances between these hydroxyl groups and Arg-132 (Table 4).
These observations support the idea put forward in a recent study (31), indicating that nonionic interactions can play important roles in the binding of charged saccharides to proteins and further emphasize the importance of characterizing these binding components. The presence of iduronic acid instead of glucuronic acid at the nonreducing extension of OCTA-3 reduces the affinity by 1 order of magnitude. The conformational flexibility of iduronic acid may affect the binding efficacy of the octasaccharide both in terms of entropic and enthalpic contributions and by increasing intermolecular distances between IЈ hydroxyl groups and AT as proposed in our model. Such hypothesis needs to be confirmed by further studies using isothermal titration calorimetry and designed AT mutants.
The occurrence of a GlcA residue near AGA*IA in OCTA-4 deserves a special comment. In fact, IdoUA residues are the prevalent uronic acids in this position near the active site of heparin for AT (3). Because relatively higher amounts of 2-Osulfated GlcA were found in enoxaparin as compared with TABLE 6 , dihedral angles at the glycosidic linkages measured on models of OCTA-1, -2, -3 and -4/AT complexes obtained by flexible docking simulations (models are shown in Figs. 7

and 8)
Glycosidic linkages within AGA*IA sequence are highlighted in bold boxes. unfractionated heparin (40), the present finding suggests that GlcA could be generated by C5-epimerization of IdoUA residues under the basic conditions used for the preparation of enoxaparin. Studies are underway to validate this hypothesis. The extension role of the active pentasaccharide sequence on AT binding properties of heparin oligosaccharides has to be considered in the design of a new generation of tailored low and very low molecular weight heparins. The type of depolymerization process applied to complex heparin chains may generate many structural variants that influence the AT binding properties and regulate the interaction with several other proteins as well.