M. Hricovíni, Institute of Chemistry, Slovak Academy of Sciences, Dubravska cesta 9, SK-842 38 Bratislava, Slovakia. Fax: +421 75941 0222, Tel.: +421 75941 0323, E-mail: email@example.com
A complex of the synthetic tetrasaccharide AGA*IM[GlcN,6-SO3-α(1–4)-GlcA-β(1–4)-GlcN,3,6-SO3-α(1–4)-IdoA-αOMe] and the plasma protein antithrombin has been studied by NMR spectroscopy. 1H and 13C chemical shifts, three-bond proton–proton (3JH-H) and one-bond proton–carbon coupling constants (1JC-H) as well as transferred NOEs and rotating frame Overhauser effects (ROEs) were monitored as a function of the protein : ligand molar ratio and temperature. Considerable changes were observed at both 20 : 1 and 10 : 1 ratios (AGA*IM : antithrombin) in 1H as well as 13C chemical shifts. The largest changes in 1H chemical shifts, and the linewidths, were found for proton resonances (A1, A2, A6, A6′, A1*, A2*, A3*, A4*) in GlcN,6-SO3 and GlcN,3,6-SO3 units, indicating that both glucosamine residues are strongly involved in the binding process. The changes in the linewidths in the IdoA residue were considerably smaller than those in other residues, suggesting that the IdoA unit experienced different internal dynamics during the binding process. This observation was supported by measurements of 3JH-H and 1JC-H. The magnitude of the three-bond proton–proton couplings (3JH1-H2 = 2.51 Hz and 3JH4-H5 = 2.23 Hz) indicate that in the free state an equilibrium exists between 1C4 and 2S0 conformers in the ratio of ≈ 75 : 25. The chair form appears the more favourable in the presence of antithrombin, as inferred from the magnitude of the coupling constants. In addition, two-dimensional NOESY and ROESY experiments in the free ligand, as well as transferred NOESY and ROESY spectra of the complex, were measured and interpreted using full relaxation and conformational exchange matrix analysis. The theoretical NOEs were computed using the geometry of the tetrasaccharide found in a Monte Carlo conformational search, and the three-dimensional structures of AGA*IM in both free and bound forms were derived. All monitored NMR variables, 1H and 13C chemical shifts, 1JC-H couplings and transferred NOEs, indicated that the changes in conformation at the glycosidic linkage GlcN,6-SO3-α(1–4)-GlcA were induced by the presence of antithrombin and suggested that the receptor selected a conformer different from that in the free state. Such changes are compatible with the two-step model [Desai, U.R., Petitou, M., Bjork, I. & Olson, S. (1998) J. Biol. Chem.273, 7478–7487] for the interaction of heparin-derived oligosaccharides with antithrombin, but with a minor extension: in the first step a low-affinity recognition complex between ligand and receptor is formed, accompanied by a conformational change in the tetrasaccharide, possibly creating a complementary three-dimensional structure to fit the protein-binding site. During the second step, as observed in a structurally similar pentasaccharide [Skinner, R., Abrahams, J.-P., Whisstock, J.C., Lesk, A.M., Carrell, R.W. & Wardell, M.R. (1997) J. Mol. Biol.266, 601–609; Jin, L., Abrahams, J.-P., Skinner, R., Petitou, M., Pike, R.N. & Carrell, R.W. (1997) Proc. Natl Acad. Sci. USA94, 14683–14688], conformational changes in the binding site of the protein result in a latent conformation.
complete relaxation and conformational exchange matrix
heteronuclear single quantum coherence
intensive nuclei enhanced by polarization transfer
rotating frame Overhauser effect
The sulfated glycosaminoglycan heparin and some heparin-derived oligosaccharides are known for their anticoagulant properties mediated by a serine protease inhibitor, the plasma protein antithrombin. The need to understand these biological properties at the molecular level has stimulated both theoretical and experimental studies involving analysis of the structure of both saccharides and the plasma protein during the binding process [1–7]. It has been shown [8,9] that the most important features of the antithrombin-binding site lie in the specific pentasaccharide sequence [–GlcN,6-SO3-α(1–4)-GlcA-β(1–4)-GlcN,3,6-SO3-α(1–4)-IdoA2-SO3-α(1–4)-GlcN,6-SO3-α- (AGA*IA)]. Systematic chemical modifications of this pentasaccharide, as well as of other heparin-derived oligosaccharides, have been used to analyse the effect of individual monosaccharide units and functional groups on the biological activity of these compounds . It was found that most sulfate and carboxylate groups are important in the activation of antithrombin. For example, the loss of a single carboxylate can lead to a decrease (more than 90%) in the antifactor Xa activity. Comparable changes in activity can occur when one of the N-sulfate groups is lacking in the glucosamine residue structure. Computer modelling of heparin-derived oligosaccharides, complexed with antithrombin, suggests that the presence of the pentasaccharide can induce elongation of helix D in antithrombin . Recently, an analysis of the crystal structure of the antithrombin–pentasaccharide complex provided direct evidence that the formation of the complex is accompanied by a change in the conformation of the active site of antithrombin . In the latent conformation, helix D is unkinked and the side chains of three of the binding residues are hydrogen-bonded to other regions of the molecule. In the active conformation, helix D was observed as slightly kinked and the side chains of the most important binding residues are not stabilized by hydrogen-bonding and are thus readily available to form ionic interactions with the heparin saccharides.
Despite the progress in understanding the mechanism of heparin-induced potentiation of antithrombin and the role of amino acids in the receptor site during the complexation, many phenomena are still not clear and require further analysis. In order to understand the binding processes at the molecular level and the thermodynamics of complexation, a knowledge of possible changes in the three-dimensional structure of the ligand in solution during the binding process and the role of the charged groups involved in the weak interaction with the antithrombin-binding residues appears to be very important.
NMR spectroscopy, together with molecular modelling, is often a major tool used to study protein–ligand binding processes; transferred NOE experiments can provide information on ligand structure in both the free and the bound state when a ligand exchanges rapidly, compared with the time scale of spin-lattice relaxation rates [12–16]. Such experiments have proved useful for analysing the conformation in several protein–carbohydrate systems [17–25]. Nevertheless, in addition to data based on NOE analysis, important information on the ligand structure can be obtained from J-couplings and chemical shifts. In fact proton chemical shifts can be affected by the presence of protein, thus the cited variables can provide quite a detailed picture of the influence of the receptor on the individual atoms in the ligand molecule. Furthermore, it is possible to monitor the torsion angle changes on glycosidic linkages through variations of both proton and carbon shieldings as well as one-bond proton–carbon coupling constants.
The present study deals with the analysis of the synthetic tetrasaccharide GlcN,6-SO3-α(1–4)-GlcA-β(1–4)-GlcN,3,6-SO3-α(1–4)-IdoA-αOMe (AGA*IM, Fig. 1) in the free state and in the presence of antithrombin. AGA*IM represents a good model for the interaction of other heparin-derived oligosaccharides with high affinity for antithrombin, and the results are discussed and compared with data obtained in kinetic studies on the same system . The lower binding affinity of the compound allowed the use of NOE and rotating frame Overhauser effect (ROE) transfer experiments at various temperatures. Thus, the NMR analysis was based on the measurement of 1H and 13C chemical shifts, proton–proton and proton–carbon coupling constants as well as transferred NOEs and ROEs. The data provide information on the behaviour of heparin-derived oligosaccahride during the binding process, and describe the three-dimensional structure of this molecule in the presence of antithrombin.
Materials and methods
The synthetic tetrasaccharide AGA*IM was prepared using well-established methods . For the study of free ligand, 1.42 mg of the sample was dissolved in 0.4 mL of D2O (99.996% D). A preparation of partially purified antithrombin (≈53 kDa) isolated from human plasma and stabilized with 50% glucose was a gift from Professor J. Fareed (Loyola University, Chicago, IL, USA). SDS/PAGE-pure antithrombin was obtained as a high-affinity fraction eluted from heparin–Sepharose and desalted by centrifugation on a 10KD Microsep device (Filtron Technology Corp., Northborough, MA, USA) The protein was freeze-dried at 4.125 mg·mL−1 in 10 mm phosphate buffer, pH 7.2, containing 0.1 m NaCl. For the first part of the binding studies, 83 µm antithrombin and 1.66 mm AGA*IM were used, at a molar protein : ligand ratio of 1 : 20; later the same concentration (83 µm) of antithrombin was added to achieve a molar ratio of 1 : 10.
1H-NMR spectra were recorded at 500 MHz on a Bruker AMX500 spectrometer, equipped with a 5-mm inverse broadband probe with a shielded z-axis gradient, at 313 K and 298 K, respectively. 1H and 13C chemical shifts are expressed relative to external sodium 3-(trimethylsilyl)propionic acid. One-dimensional 1H-NMR experiments, used for determination of 3JH1-H2 in IdoA, were performed with the pulse sequence in which the dephasing delay was inserted before the acquisition period. A series of experiments were carried out in which the delays varied systematically from 5 µs up to 600 ms (35 experiments altogether); the carrier frequency was positioned in the middle of the resonance of interest. The true 3JH1-H2 was then determined by computer fitting of the amplitude of I1 signals to the function A[B + cos(πJτ)][exp(–τ/T2)], where J is the coupling, τ is the dephasing delay, T2 is the spin–spin relaxation time and A and B are constants. The recycle time was 14.5 s, and the digitization of spectra 0.022 Hz per point after Fourier transformation. 1H spin–spin relaxation times were measured by Carr–Purcell–Meiboom–Gill spin-echo experiments using 16 points in the variable delay list. Bruker software was used to process the data running on a Silicon Graphics indy workstation. Two-dimensional COSY and heteronuclear single quantum coherence (HSQC) experiments were performed using z-gradients for coherence selection. HSQC spectra were collected with and without decoupling during the acquisition period in phase sensitivity-enhanced pure-absorbtion mode ; the matrix size was 1024 × 512 points (ω2 × ω1). The experiment performed without decoupling was zero-filled to 4K × 2K and multiplied with shifted sine-bell-squared prior Fourier transformation giving the resolution of 0.72 Hz per point (for the spectra of free ligand) and 0.97 Hz per point (spectra of complex) in ω2 domain; 32 (free AGA*IM) or 512 (AGA*IM–antithrombin complex) transients were acquired for each free-induction decay and the corresponding experimental time was 6 h and 3 days, respectively. Two-dimensional phase-sensitive NOESY experiments were recorded using standard pulse sequence with 16 transients with three different mixing times; the relaxation delay was 4 s. Presaturation of HDO resonance was achieved by low-power irradiation during part of the relaxation delay and during the mixing time. The spectral width was 3000 Hz, and data (matrix size 1024 × 512 points) were zero-filled to 2K × 2K before Fourier transformation and multiplied with a shifted (π/3) squared cosine functions; fourth-order polynomial baseline correction was applied in both dimensions. Two-dimensional ROESY spectra were measured with the same spectral characteristics; during the mixing time a single spin-lock pulse was applied (carrier frequency positioned at 6 p.p.m.) with a strength of 4 kHz. The experimental ROE intensities were then recalculated according to the resonance frequencies in the spectrum . Two-dimensional transferred NOESY spectra were performed in a similar way to two-dimensional NOESY in the case of free ligand: the spectral width was 4000 Hz, and a short (20 ms) spin-lock pulse was performed after the π/2 excitation pulse . A total of 32 transients were collected for each free-induction decay; the total experimental time was about 12 h depending on the mixing time. Seven relaxation delays were used to measure spin-lattice relaxation times at 7 T from two-dimensional double intensive nuclei enhanced by polarization transfer (INEPT) spectra  collected with suppression of the effects of cross-correlation between dipolar and chemical shift anisotropy relaxation mechanisms. T1 values were calculated from a two-parameter fit to the experimental cross-peak volumes using the non-linear Levenburg–Marquart algorithm.
A Monte Carlo conformational search was carried out using the MacroModel V5.0 program . The modified force field MM2 was used, in which the main difference between the authentic field is in Coulomb’s law treatment of electrostatics (instead of bond dipoles/Jeans equation) and out of plane bending equation. The continuum model for evaluation of tetrasaccharide solvation (GB/SA model) where the GB (generalized Born equation) solvent electrostatics treat solvent polarization explicitly, the solvent-accessible (SA) surface is treated empirically . A dielectric constant of 1.0 was used for Coulomb’s law electrostatics within this model. As tetrasaccharide could adopt multiple conformations in aqueous solution, a search of conformational space was carried out using the Monte Carlo conformational search routine . Conformational searching was conducted in two steps using the Pollack–Ribiere minimization procedure. In the first step, 500 trial structures were generated starting from the geometry (at the glycosidic linkages) computed for a structurally similar pentasaccharide AGA*IAM. In the second step, further minimization (500 steps each) was carried out starting from the geometry of the six lowest minima found in the previous procedure. The conformation at the glycosidic linkages is characterized in terms of torsion angles which are defined as φ = (H-1)–(C-1)–(O-1)–(C-4) and ψ = (C-1)–(O-1)–(C-4)–(H-4).
The magnitude of three-bond 1H–1H coupling constants, based on intensities of the selected signal, was obtained by fitting the amplitude to the function A[B + cos(πJτ)][exp(–τ/T2)], where A, B are constants, τ is the dephasing delay, T2 is proton transversal relaxation time and J is unknown coupling . The fit to the intensities, using 35 delays varying from 5 µs up to 600 ms, was performed with the routine written within sigma plot software running on a PC. Theoretical transferred NOEs were obtained using a complete relaxation and conformational exchange matrix (corcema) program  running on a Power Challenge Silicon Graphics under irix6.2. The program computes the NOESY intensities for a given structural model using Eqns (2) and (3)
where M and M0 are the magnetization vector and its equilibrium value. I(τ) is the intensity matrix at mixing time τ, D is the dynamic matrix which is a sum of the relaxation matrix R and the exchange matrix K. In the present case, a two-state model consisting of free and bound states was used. Thus, the NOEs were based on the tetrasaccharide geometry found in the Monte Carlo conformational search using full relaxation and conformational exchange matrix analysis. As the structure of the complex of tetrasaccharide AGA*IM and antithrombin is unknown, the protein protons were not evaluated. The intensities of cross-peaks were referenced to the diagonal peaks at each mixing time. The R factors were computed according to the formula:
The 500-MHz 1H-NMR spectrum of tetrasaccharide AGA*IM in aqueous solution at 40 °C is shown in Fig. 2. The assignment of 1H and 13C resonances was straightforward using standard one-dimensional and two-dimensional methods. The values of 1H chemical shifts, collected at 40 °C and 25 °C, are listed in the Tables 1 and 2, and 13C chemical shifts at 40 °C are in Table 3. Three-bond proton–proton (3JH-H) coupling constants (not shown) confirmed the 4C1 chair conformation of glucosamine and glucuronic acid residues. The conformation of the IdoA residue was subjected to a more detailed analysis, as this residue exists in other glycosaminoglycans in an equilibrium of three conformers, namely 1C4, 4C1 and 2S0[37–39]. The conformer population depends on substitution at C-2 and the structure of neighbouring units. The conformer abundance can be estimated from the magnitude of the proton–proton coupling constants. The splitting of both I1 (1.82 Hz) and I5 (2.23 Hz) (Table 4) indicated the prevailing abundance of the 1C4 conformer with a smaller fraction of 2S0 conformer. For a more precise evaluation of the conformer populations, the coupling constant of the anomeric proton had to be determined; direct measurement from splitting would lead to a biased result, because of an insufficiently resolved I1 signal . The computer fitting of the amplitude of the I1 signal, collected in a series of one-dimensional experiments with various dephasing delays, yielded 3JH1-H2 = 2.51 ± 0.08 Hz using the value of 1H T2 (I1) = 210 ms. The value of the computed coupling (2.51 Hz), higher than the experimental splitting (1.82 Hz), is due to the shift of the doublet maxima in an unresolved multiplet . Based on this value, as well as other 3JH-H values (Table 4), the estimated ratio of conformers 1C4 to 2S0 was found to be approximately 75 : 25 in the tetrasaccharide AGA*IM in aqueous solution at 40 °C.
Table 1. 1H chemical shifts (referenced to external sodium 3-(trimethylsilyl)propionic acid) in tetrasaccharide AGA*IM in aqueous solution at 40 °C in the absence (δfree) and presence (δ20 : 1, δ10 : 1) of antithrombin (20 : 1 and 10 : 1, AGA*IM : antithrombin).
δ20 : 1
δ10 : 1
Table 2. 1H chemical shifts (referenced to external sodium 3-(trimethylsilyl)propionic acid) in tetrasaccharide AGA*IM in aqueous solution at 25 °C in the absence (δfree) and presence (δ20 : 1, δ10 : 1) of antithrombin (20 : 1 and 10 : 1, AGA*IM : antithrombin).
δ20 : 1
δ10 : 1
Table 3. 13C chemical shifts (± 0.03 p.p.m., referenced to external sodium 3-(trimethylsilyl)propionic acid) and one-bond 1H-13C coupling constants (values in Hz) in tetrasaccharide AGA*IM in aqueous solution at 40 °C in the absence (δfree) and presence (δ10 : 1) of antithrombin (10 : 1, AGA*IM : antithrombin).
δ10 : 1
1JC-H (10 : 1)
Table 4. Measured splittings [3JH-H (exp), values in (Hz)] and linewidths (Hz) in iduronic acid residue in tetrasaccharide AGA*IM in the free state and in complex with antithrombin (20 : 1, AGA*IM : antithrombin) at 40 °C. Computed coupling 3JH1-H2 (calc) was determined by fitting the amplitude of the I1 signal to the function A[B + cos(πJτ)][exp(–τ/T2)] using T2(I1) = 210 ms. 3JH2-H3 (exp) and 3JH3-H4 (exp) were determined experimentally, with 0.2 Hz precision.
20 : 1
13C chemical shifts and one-bond proton–carbon coupling constants (1JC-H) were determined from two-dimensional HSQC spectra (Table 3) collected, respectively, with and without decoupling during the acquisition period. In fact, both 13C shifts and 1JC-H values can be used as additional parameters to monitor changes in the ligand molecule during protein binding. 1JC-H values, because of their great magnitude, are expected to change more than three-bond proton–proton couplings, which have values more than one order smaller. Furthermore, the 1JC-H values in carbohydrates depend, among other parameters, on the stereochemistry of atoms in the vicinity [41–43] and can thus be used to monitor the conformational changes at the glycosidic linkage (in AGA*IM based on the values of 1JC1-H1 and 1JC4-H4) and the conformational equilibria in the IdoA residue. The measured magnitudes of 1JC1-H1 varied (Table 3) depending on the type of glycosidic linkage (α or β) at the anomeric centre, and were larger in the α-type of the linkage (≈174–175 Hz). Time-averaged 1JC1-H1 (170.6 Hz) in the IdoA residue was slightly smaller, and its reduced magnitude was due to the contribution of the skew 2S0 conformer. For ring atoms, 1JC-H differed because of the effect of substitution and ψ torsion angles.
Two-dimensional NOESY and ROESY spectra were measured at 25 °C and 40 °C for three different mixing times. Cross-peaks observed in two-dimensional ROESY at 40 °C (Fig. 3) indicated numerous interproton interactions, both within the monosaccharide units and across the glycosidic linkages. Relative magnitudes of ROE cross-peaks depended on temperature, the ROEs being larger at 25 °C (Tables 5 and 6). Similarly, NOE cross-peaks varied with temperature; however, as expected, their variations were considerably larger. NOEs were small (≈1%) and positive at 40 °C and negative at 25 °C (Table 7). Cross-sections along the ω2 axis through the A1* signal at different temperatures are shown in Fig. 4. NOE between A1* and I3 was 0.2% at 40 °C (300 ms mixing time), and NOE (A1*)–(I4) was ≈1%. As mentioned, negative NOEs were observed at 25 °C: NOE (A1*)–(I3) was −2.4% and (A1*) –(I4) was −0.1%. In the additional NOE experiment performed at 30 °C, NOE (A1*)–(I3) was found to be positive, and at the same time, NOE (A1*)–(I4) was negative, indicating rather complex internal dynamics in the reducing part of the molecule. Pseudo-rotation of the IdoA residue [37,38,44], with possible contributions of flexibility of this residue on the A*–I glycosidic linkage, caused different correlation times of the (A1*) – (I3) and (A1*)–(I4) relaxation vectors, respectively . In addition, non-isotropic overall molecular motion of AGA*IM is another factor that cannot be fully ruled out, and might also cause the differences in the correlation times of both relaxation vectors [46,47].
Table 5. Experimental and computed ROEs (in percentage with respect to A1 diagonal peak intensity) in tetrasaccharide AGA*IM at 40 °C for three different mixing times (150, 250 and 350 ms). The geometry, φ1, ψ1 (−45°, −30°), φ2, ψ2 (42°, 18°) and φ3, ψ3 (−27°, −48°), was used to calculate theoretical ROEs. The R factor is given in the last column.
Table 6. Experimental and computed ROEs (in percentage with respect to A1 diagonal peak intensity) in tetrasaccharide AGA*IM at 25 °C for three different mixing times (150, 250 and 350 ms). The geometry, φ1, ψ1 (−45°, −30°), φ2, ψ2 (42°, 18°) and φ3, ψ3 (−27°, −48°) was used for the calculation of theoretical ROEs. The R factor is given in the last column.
Table 7. Experimental NOEs (in percentage with respect to A1 diagonal peak intensity) in tetrasaccharide AGA*IM at 40 °C and 25 °C.
The theoretical interpretation of experimental ROEs in AGA*IM in aqueous solution was based on the geometry of minima found during Monte Carlo conformational search  using the MM2 force field. Four energy minima were found with relative energies within 10 kJ·mol−1 (Table 8). The dihedral angles φ1, ψ1 were −20° and −38°, respectively, for the lowest energy minimum. Minima 2 and 3 had slightly larger φ1, ψ1 values (comparable with each other), with a relative energy 1.4 kJ·mol−1 and 5.0 kJ·mol−1. Minimum 4 (ΔE = 8.4 kJ·mol−1) had a different conformation at the A-G glycosidic linkage (φ1 = 39°, ψ1 = 12°). φ2, ψ2 dihedral angles were nearly identical for conformers 1, 2 and 4. The dihedral angles φ3, ψ3 could be considered comparable in all four minima, ranging from −27° up to −56° for φ3 and from −41° up to −63° for ψ3. Because of the very small magnitudes of the NOEs (both positive and negative) in water at both temperatures, and thus relatively large experimental error, NOEs were not used for interpretation of theoretical calculations. The overall molecular correlation time of tetrasaccharide in the free state was calculated from 13C spin-lattice relaxation times measured at 7 T at 40 °C using two-dimensional double INEPT spectra with suppression of cross-correlation between dipolar and chemical shift anisotropy relaxation mechanisms. 13C T1 values were 0.20–0.22 s for all carbons except 13C in the IdoA residue, where slightly longer relaxation rates were found (13C T1 = 0.23–0.24 s), suggesting different motional properties of this residue. In the present case, however, a single correlation time (τlig = 0.32 ns), computed from 13C T1, was used for characterization of the overall motion of tetrasaccharide and was used in the calculation of ROE. As mentioned, although NOEs were not used for quantitative interpretation, the magnitude of the τlig obtained corresponded to experimental observations (ωτlig≈1). The independent way of estimating τlig was based on fitting experimental ROEs across fixed internuclear distances [(A1)–(A2) and (A1*)–(A2*)], and the value of τlig was found to be the same (τlig = 0.32 ns) as computed from 13C T1 data. The same approach was used to calculate τlig of tetrasaccharide at 25 °C (τlig = 0.47 ns).
Table 8. Relative energy (in kJ·mol−1) and torsion angles (in degrees) at the glycosidic linkages for four lowest enegy minima found during a Monte Carlo conformational search.
Based on the geometry of computed minima, theoretical ROEs were obtained using the full relaxation matrix approach (Tables 5 and 6). The computed ROEs, based on the geometry of the lowest energy minimum (Table 8), had larger (A1)–(G4) interglycosidic ROE than that observed in experiments; at the same time discrepancies were observed also for (A1*)–(I4) and (A1*)–(I3) ROEs. Better agreement was obtained using the geometry of minimum 2 (ΔE = 1.4 kJ·mol−1) with φ1, ψ1 (−59°, −64°), φ2, ψ2 (42°, 18°) and φ3, ψ3 (−27°, −48°), although the computed (A1)–(G4) ROE was found to be too small. The latter discrepancy was due to the large (A1)–(G4) distance (3.236 Å) computed for the conformation with φ1 = −59°, ψ1 = –64°, thus the theoretical ROEs were considerably smaller than the experimental values. ROEs computed using the geometry of minima 3 and 4 did not allow interpretation of the experiments, and large differences were observed in practically all interglycosidic ROEs. In order to interpret (A1)–(G4) ROEs, the geometry of the local minimum with φ1 = –45°, ψ1 = –30° (the conformer with ΔE > 10 kJ· mol−1) had to be used. In the latter conformation the internuclear distance (A1)–(G4) was 2.548 Å and the computed ROE data matched well with the experiment (R = 0.07 at both temperatures).
Tetrasaccharide AGA*IM in the presence of antithrombin
Chemical shifts and coupling constants. The dissociation constant (Kd) for AGA*IM–antithrombin was recently published for various pH values and ionic strengths and is in the millimolar to micromolar range (≈10−5m) . This corresponds to the fast exchange regime with respect to chemical shift, thus one observes a single set of ligand resonances where the values of chemical shifts and coupling constants are the weighted average, determined by free and bound populations. 1H-NMR spectra of the AGA*IM–antithrombin complex (20 : 1 and 10 : 1) at 40 °C are shown in Fig. 5, and four selected regions are shown in Fig. 6. As the protein–ligand interaction can induce chemical shift changes in both ligand and antithrombin, and the magnitude of the variation depends on the fraction of ligand molecules in the binding site and Kd, 1H and 13C chemical shifts were monitored at two temperatures and two molar ratios (Tables 1 and 2). Anomeric proton resonances (A1 and I1) shifted upfield from the free to the bound state (10 : 1 ratio, 40 °C, Table 1) (−0.024 p.p.m. and −0.009 p.p.m., respectively), and A1* shifted downfield (0.017 p.p.m.) (Fig. 6a). Strong interaction between A1 and A1* and antithrombin was also manifested by changes in linewidth. The linewidth of both A1 and A1* increased considerably, from 5.9 Hz (both protons) to 10.4 Hz (A1) and 11.7 Hz (A1*). Very strong effects were observed, mainly in the GlcN,3,6-SO3(A*) residue for A2* (−0.024 p.p.m.) and A4* (−0.025 p.p.m.) as well as for A2 of the non-reducing glucosamine residue (−0.031 p.p.m.) (Fig. 6b-d). Induced shifts were also detected for G2; however, the values could not be precisely determined because of resonance overlap with the OMe signal. The changes in linewidth of the latter signals were not measurable because of signal overlap with other ligand and protein resonances, except for A2 where the linewidth changed quite noticeably (from 15.6 Hz to 26.1 Hz). Selective shielding/deshielding effects, induced by protein, were visible mainly in the A* unit, A2*, A3* and A4* resonances shifting downfield, and A1*, A5* and A6* shifting upfield; A6*′ shift remained nearly constant. The observed changes in A1*δ values could also be partially caused by the conformational changes at the glycosidic linkages (see Discussion). Hydroxymethyl protons in GlcN,6-SO3 experienced somewhat larger variations than that for A6* and A6*′, suggesting the stronger influence of antithrombin. For the same reason, the observed shift changes in A6 and A6′ could have been partially caused by the change in the ω dihedral angle (O-5 – C-5 – C-6 – O-6). The smallest variations in proton chemical shifts were detected for IdoA resonances where all the signals varied within 0.009 p.p.m. In addition, the observed narrower linewidth of these signals suggested different internal dynamics of this residue in the free state and during the binding process in comparison with the other three residues.
Similar trends have been observed for proton chemical shifts at 25 °C at both molar ratios where the largest changes were mainly detected in A and A* residues (Table 2). For example, chemical shifts of A1 and A2 changed significantly (−0.012 p.p.m. and −0.018 p.p.m.) in the non-reducing glucosamine unit A; in the A* residue the most pronounced change was for A2* (−0.018 p.p.m.) and A4* (−0.016 p.p.m.). The larger linewidth observed for A1 (Δν1/2 = 14.3 Hz) and A1* (Δν1/2 = 15.1 Hz) were caused partially by the increased non-homogeneity of the solution and the increased correlation times of the molecules.
13C chemical shieldings are generally less sensitive to the presence of protein than 1H chemical shifts, and the observed differences in the δ values in the free state and in the presence of antithrombin (ratio 10 : 1) were marginal in most cases. However, the variations seen for the anomeric (0.13 p.p.m.) and the ring carbons (0.10–0.15 p.p.m.) in the A* residue, were among the largest observed in the tetrasaccharide (Table 3) and corresponded well to the 1H-NMR spectra trends. Furthermore, large variation was also detected for the G4 chemical shift (0.17 p.p.m.). Such a profound difference in the δ value of the carbon, not linked directly to SO3− or COO− groups, is rather surprising. It seems that, in addition to the electron-density change at the C-4 nucleus (caused by the interaction of the nearest COO− group with the protein-binding site), other effects could also influence C-4 chemical shift, the most likely being the conformational change at the glycosidic linkage GlcN,6-SO3–GlcA. In fact, such an interpretation agrees well with the variations in two other NMR parameters, 1JC-H and transferred NOEs.
The change in conformation of the IdoA residue in the presence of antithrombin was analysed considering the variations in 3JH-H and 1JC-H (Tables 3 and 4). In both cases, minor changes were detected in one-dimensional and two-dimensional spectra in the free and bound state. As mentioned previously, 3JH1-H2 in the free state was determined by computer fitting of the amplitude of the I1 signal, and the computed value was 2.51 Hz. However, an analogical procedure could not be used in the presence of antithrombin because of strong HDO signal interference. Therefore, only a qualitative comparison was made based on the measured splittings (Table 4). As the linewidth of the I1 signal varied negligibly in both the free and bound state (ΔΔν1/2 = 0.05 Hz) the measured change from 3JH1-H2 = 1.82 Hz to 1.73 Hz appears to be due to the variation in conformational equilibrium, not the shift of doublet maxima caused by the linewidth. A similar decrease in J-coupling magnitudes was also detected for 3JH4-H5 (from 2.23 Hz to 1.95 Hz). The smaller values of 3JH1-H2 and 3JH4-H5 suggest that the equilibrium is shifted towards the 1C4 chair form in the presence of antithrombin [35,36]. It should also be noted that the experiment was carried out with protein albumin which does not specifically interact with AGA*IM; in this case the measured couplings were the same in both cases, both free and in the presence of protein. Minor changes in 1JC1-H1 and 1JC5-H5 (Table 3) also support the prevalence of the chair form in this residue in the presence of antithrombin.
Variations in some other 1JC-H values were also detected (Table 3). The largest changes in couplings measured in the free and bound states were found for 1JC1*-H1*, 1JC2*-H2* and 1JC3*-H3* (Δ1JC-H was 1.1 Hz to 1.3 Hz), Δ1JC1-H1 in the GlcN,6-SO3 residue was 2.2 Hz. This last relatively large change is in accordance with the indicated (based on chemical shift changes) conformational change at the glycosidic linkage GlcN,6-SO3–GlcA. The lower value of 1JC-H is consistent with φ1 = –45° (free state), and the increase in the magnitude agrees qualitatively  with the change in the dihedral angle to φ1 = 30° (see later analysis of transferred NOEs). The second independent support for the conformational change is based on the same dependence of 1JC4-H4 in GlcA unit; this value could not be measured reliably because of the interference of protein signals in coupled two-dimensional HSQC spectra. The differences in 1JC1*-H1* in the A* residue could also be partially caused by a conformational change at the glycosidic linkage. The variation in electron density caused by strong electrostatic interactions of the SO3− groups with antithrombin amino acids could, presumably, be the reason for the observed changes in 1JC2*-H2* and 1JC3*-H3*.
Transferred NOEs and ROEs. Two-dimensional transferred NOEs at 25 °C and 40 °C were large and negative in the presence of antithrombin (Tables 9 and 10), those at 25 °C being considerably larger. The two-dimensional T1ρ-filtered transferred NOE spectrum at 40 °C is shown in Fig. 7 and selected cross-sections along the ω2 axis in Fig. 8. In fact, the same type of cross-peak pattern was observed in this spectrum as in ROE in the free ligand, thus no additional peaks appeared to be due to spin diffusion. Cross-peak intensities from two-dimensional transferred ROESY spectra, collected at 40 °C, are given in Table 11. These spectra showed neither cancellation nor considerable change in cross-peak volume compared with two-dimensional NOESY. The observed data support the lack of spin-diffusion effects in the transferred NOESY spectra of the AGA*IM–antithrombin complex. The magnitude of the transferred NOEs depends on various variables (in addition to geometry of the molecular complex) such as Kd, off-rate (koff) and the molecular correlation times of ligand (τlig) and protein (τantithrombin). The present interpretation of the experimental data was based on the corcema program  using full relaxation and conformational exchange matrix analysis, and the values of Kd, koff and τantithrombin were estimated; τlig was determined independently (0.32 ns and 0.47 ns at 40 °C and 25 °C, respectively) as described in the previous section. A search was carried out in which all three parameters, Kd, koff and τantithrombin were varied systematically, computed NOEs were compared with the experimental values and R-factors were calculated. NOEs computed for fixed distances, with the values Kd = 1.8 mm, koff = 60 s−1 and τantithrombin = 38 ns, agreed well with the experimental NOEs (A1)–(A2) with R = 0.06 and (A1*)–(A2*) with R = 0.02 at 40 °C whereas Kd = 0.47 mm, koff = 20 s−1 and τantithrombin = 52 ns interpreted experimental NOEs (A1)–(A2) and (A1*)–(A2*) at 25 °C (Tables 9 and 10). However, the geometry of the tetrasaccharide used to interpret ROEs in the free state could not interpret all the transferred NOEs in the presence of antithrombin. The main difference between the theoretical and the experimental transferred NOEs was between A1 and G4 protons, thus on the GlcN,6-SO3–GlcA glycosidic linkage. The computed interglycosidic NOEs were smaller than the measured values using the conformation with φ1 = –45°, ψ1 = –30° (A1–G4 distance is 2.548 Å) whereas intraunit (A1)–(A2) NOEs were larger. In addition, differences in other NOEs were also observed, particularly (A1*)–(A2*) and (A1*)–(I3). Satisfactory agreement (R = 0.07 at 40 °C and R = 0.18 at 25 °C) was obtained with the geometry of minimum 4 (with φ1 = 39°, ψ1 = 12°; φ2 = 43°, ψ2 = 14°) with, however, the conformation on the GlcN,3,6-SO3–IdoA linkage as found in conformer 2 (φ3 = –27°, ψ3 = –48). By using the geometry of the latter conformer, better agreement was found not only for (A1)–(G4) NOEs, where larger NOEs were computed (A1–G4 distance is 2.295 Å), but also for (A1*)–(A2*) and (A1*)– (I3) NOEs (Tables 9 and 10). It should be noted that the tightly coupled protons G3 and G4 could affect NOE magnitude to some extent. However, the difference in chemical shift between G3 and G4 in the free state and in the presence of antithrombin (20 : 1) remained the same (thus the ratio δ/J was not changed in the free state and in the complex). A small difference was found for (A1*)–(I4) NOEs as well. It is of interest that in the free state the computed ROEs were smaller than the experimental (A1*)– (I4) values at both temperatures (Tables 5 and 6), thus indicating that the interatomic (A1*)–(I4) distance is slightly larger than one would expect from the experiment. The opposite situation was observed in the bound state. The computed (A1*)–(I4) NOEs were somewhat larger than the experimental values, therefore the geometry at the GlcN,3,6-SO3–IdoA linkage seems to be partially biased in both the free and the bound state. The fact that the contribution of the 2S0 conformer (only the 1C4 conformer was used to compute transferred NOEs) and that of the protein protons was neglected in these calculations could, to some extent, contribute to the observed discrepancy. The different conformation at the GlcN,3,6-SO3–IdoA linkage could be another reason for such discrepancies. As mentioned, the ROE intensities were found to be comparable (although slightly smaller) with the transferred NOEs, suggesting that no strong indirect transferred NOEs are present in the spectra. The present data thus suggests variation in the interresidue dihedral angles, particularly between GlcN,6-SO3–GlcA units, during the binding of tetrasaccharide AGA*IM to antithrombin. This observation is in agreement with the changes in 13C C-4 chemical shift in the GlcA residue and 1JC1-H1 in the GlcN,6-SO3 residue.
Table 9. Experimental and computed transferred NOEs (in percentage with respect to A1 diagonal peak intensity) in tetrasaccharide AGA*IM in the presence of antithrombin at 40 °C (20 : 1, AGA*IM/antithrombin) at three different mixing times (100, 150 and 300 ms). Computed NOEs were obtained with koff = 60 s−1, Kd = 1.8 mm, τlig = 0.32 ns and τantithrombin = 38 ns. The geometry φ1, ψ1 (39°, 12°), φ2, ψ2 (43°, 14°) and φ3, ψ3 (−27°, −48°) was used for the calculation of theoretical NOEs. The R factor is given in the last column.
Table 10. Experimental and computed transferred NOEs (in percentage with respect to A1 diagonal peak intensity) in tetrasaccharide AGA*IM in the presence of antithrombin at 25 °C (20 : 1, AGA*IM : antithrombin) at three different mixing times (100, 150 and 300 ms). Computed NOEs were obtained with koff = 20 s−1, Kd = 0.47 mm, τlig = 0.47 ns and τantithrombin = 52 ns. The geometry φ1, ψ1 (39°, 12°), φ2, ψ2 (43°, 14°) and φ3, ψ3 (−27°, −48°) was used for the calculation of theoretical NOEs. The R factor is given in the last column.
Table 11. Experimental ROEs (in percentage with respect to A1 diagonal peak intensity) in tetrasaccharide AGA*IM in the presence of antithrombin at 40 °C (20 : 1, AGA*IM : antithrombin) at different mixing times.
The interaction between antithrombin and tetrasaccharide AGA*IM has been analysed using three NMR parameters: chemical shifts, coupling constants and NOEs. Each depends on the structure of the ligand and complex in a different way, thus a detailed picture of this interaction could be obtained. The origins of the effects, which lead to variations in both 1H and 13C shieldings, can be divided into two groups. The major effect is due to changes in electron density caused by the ‘direct’ shielding/deshielding influence of the protein. As seven charged groups (SO3−, COO−) are present in the structure of AGA*IM, strong electrostatic interaction with protein amino acids that take part in the binding process can be expected. Recently, a crystal structure of the complex of a structurally similar pentasaccharide, AGA*IAM, with antithrombin was analysed . The data indicate that the conformational rearrangement in antithrombin causes a significant redistribution of positive charges in the binding site of the protein upon the adoption of a low-affinity latent conformation. In contrast, in the ligand molecule considerable variations in electron density can be expected, particularly in the central A* residue (bearing three charged groups), as the result of electrostatic interaction with antithrombin amino acids. In fact, the largest 1H chemical shift changes were observed for this residue, with practically all the shifts varying upfield or downfield in the presence of antithrombin. The differences in the 1H shifts of A6 and A2 are probably from the same origin in the non-reducing GlcN,6-SO3− residue. The second effect is due to induced conformational change in the ligand molecule upon binding to antithrombin. Both 1H and 13C shieldings depend strongly on the glycosidic linkage conformation, and such changes could be correlated with torsion angles . Thus, chemical shifts of atoms at the glycosidic linkage, such as A1, G4, A1* and A4*, could be affected by conformational change due to variations in the φ, ψ dihedral angles. In particular, 1H A1 and 13C G4 chemical shift variations agreed well with the differences in other NMR parameters, and suggested conformational change at the GlcN,6-SO3–GlcA glycosidic linkage during the binding process.
In several NMR studies dealing with protein–carbohydrate interactions [20,48–50], considerable spin-diffusion effects were observed to influence the magnitude of the transferred NOEs. Such mediated effects were caused by magnetization transfer via intramolecular (carbohydrate) protons as well as by protein protons. However, in the present study, no mediated effects were observed, suggesting that the protein protons are not in close contact with the protons in the AGA*IM molecule that were analysed by transferred NOEs. This seems to be incompatible with the observed relatively large variations in proton chemical shift, which would indicate quite close antithrombin protons. The above contradictory evidence may stem from the bulky charged SO3− and COO− groups of AGA*IM. Because of these relatively large groups in the tetrasaccharide, the protein protons remain close to the surface of sulfo and carboxyl groups but relatively distant from the anomeric and ring protons (those monitored in two-dimensional transferred NOEs) owing to steric effects. Thus, antithrombin could cause variation in electron density in AGA*IM as the result of electrostatic interaction, with a consequent large variation in proton shieldings. However, the transfer of magnetization during the NOE experiments was not greatly influenced by protein protons. Such electrostatic interactions and the effect of the sulfate and carboxylate groups confirms the importance of these groups on activity of heparin, the pentasaccharide AGA*IAM and other heparin-derived oligosaccharides [2,4,10].
The effect of flexibility and the conformational equilibrium in IdoA on the biological activity of heparin and heparin-derived compounds is a subject of interest [10,37,38]. Both synthetic and crystallographic studies have analysed the possible active conformers of this residue in the bound state. For example, a synthetic analogue of pentasaccharide AGA*IAM, in which an additional OSO3− group was introduced at the reducing-end GlcN,6-SO3 residue and the conformational equilibrium was shifted almost completely towards the 2S0 conformer in the IdoA residue, showed a biological activity that was about twice as high as that of natural heparin . Furthermore, a recent analysis of a derivative of pentasaccharide with an L-IdoA residue in the fixed 1C4 chair form showed very low activity in an antithrombin-mediated anti-Xa assay . This evidence led to the conclusion that the 1C4 conformer is unlikely to be the active one. However, the crystal structure of the heparin-derived hexasaccharide complexed to the basic fibroblast growth factor showed that the IdoA ring adopted multiple conformations in the presence of the protein: one IdoA residue in the hexasaccharide was in the 1C4 chair form, the other one was in the skew 2S0 form , indicating that an even energetically less favourable conformer of the IdoA residue can be the active one. In the present case, 3JH-H coupling constants measured in the IdoA residue in the tetrasaccharide in the free solution suggested that the conformational equilibrium is shifted towards the 1C4 conformer (population of about 75%). A small decrease in coupling constants was observed in the complex with antithrombin, which may indicate a further shift towards the chair form of the pyranose ring. The shift in conformational equilibrium towards the chair form in the IdoA residue suggests a stabilization of the energetically more favoured conformer in the presence of the protein. However, as already mentioned, the exclusive presence of a more stable conformer in the bound state is not always the case, even in structurally very similar systems. Thus, a direct comparison of the present data or extrapolation with other heparin-derived oligosaccharides may not be the most appropriate approach.
The analysis of transferred NOEs was focused mainly on the characterization of glycosidic linkage conformation. The geometry of the tetrasaccharide, found in the Monte Carlo conformational search, was used to interpret NOEs using the full relaxation matrix approach. In the free state (Fig. 9), the best fit to experimental NOEs was obtained with geometry of φ1, ψ1 (−45°, −30°), φ2, ψ2 (42°, 18°) and φ3, ψ3 (−27°, −48°). The geometry at the glycosidic linkages is comparable with that computed for the pentasaccharide AGA*IAM. This evidence is not surprising as the structure of the A-G-A* part of the molecule is identical. However, in the bound state (Fig. 10), the present data suggest that the receptor probably selected a different conformation (with φ1 = 39°,ψ1 = 12°) at the A-G glycosidic linkage. The difference in conformation at this linkage is also supported by the variation in the other two NMR parameters, the chemical shifts and 1JC-H. Protein-induced conformational changes in the ligand molecule were also observed in other protein–carbohydrate systems [54–56] and support the idea that the bound conformation may not correspond to the one in the free state. In fact, the indications of conformational changes in two other glycosidic linkages, G-A* and A*-I, were observed as well, as discussed in connection with the differences in (A1*)–(I4) and (A1*)–(I3) NOEs as well as chemical-shift changes of A1* and A4*. Detailed quantitative analysis of the mentioned NOEs is not straightforward owing to the complex internal dynamics at the A*-I linkage (the different correlation times for A1*–I3 and A1*–I4 relaxation vectors because of pseudo-rotation of the IdoA residue and the flexibility of the A*-I glycosidic linkage) and will be the subject of further studies. Moreover, the interference of the residual HDO resonance with the G1 proton signal precluded a precise evaluation of transferred NOEs at the G–A* glycosidic linkage. Quantitative analysis using the corcema program allows estimation of Kd and koff. The computed value for Kd was 0.47 mm at 25 °C, which was somewhat larger than that (17 μm) determined by fluorescence emission spectroscopy for the AGA*IM–antithrombin complex, indicating weaker binding . The difference in the computed and experimental Kd values is probably due to the simplified two-state model used to characterize the AGA*IM–antithrombin interaction in the present interpretation of the NOE experiments . The same is valid for the computed values of off-rates which were, respectively, 20 s−1 and 60 s−1, depending on temperature, and are higher than those obtained recently .
The determination of the three-dimensional structure of AGA*IM in the bound state is an important step toward understanding the interaction mechanism of heparin-derived oligosaccharides with antithrombin at the molecular level and the thermodynamics of the binding process. Both modelling and crystallographic studies have focused on analysing the interaction of pentasaccharide AGA*IAM with antithrombin [2,3,6,10,11,57,58]. One fundamental conclusion of these studies was the importance of specific SO3− and COO− groups in the structure of the oligosaccharide. The lack of a single group can considerably decrease the biological activity , underlining the importance of their involvement in the binding process, particularly the electrostatic interaction with amino acid residues in the binding site. As found in the present study, the 6-O-SO3− group in the GlcN,6-SO3− residue, the COO− group in GlcA,N-SO3− and 3-SO3− groups in GlcN,3,6-SO3− residues create a sequence of four negative charges in the molecule in the free state (Fig. 9), in which the distances between the sulfur atom in the 6-SO3− group in the GlcN,6-SO3− residue and sulfur atom in the 3-SO3− group in the GlcN,3,6-SO3− residue was 5.8 Å and the distance between the sulfur atom in the 6-SO3− group in the GlcN 6-SO3− residue and the carboxylic carbon in the GlcA COO− residue was 4.5 Å. However, in the bound state (Fig. 10), the distribution of negatively charged groups was different, a consequence of the variation in glycosidic-linkage conformation evident from the interatomic distances. The distance between the sulfur atoms in the 6-SO3− group in the GlcN, 6-SO3− residue and the 3-SO3− in the GlcN,3,6-SO3− was 7.5 Å whereas the distance between sulfur and carbon atoms (6-SO3− group in the GlcN,6-SO3− residue and carboxylate carbon in COO− in GlcA) was 6.7 Å. Thus, the orientation of the charged groups on the non-reducing end of AGA*IM is different in the presence of antithrombin. Such spatial distribution would appear to fit the structure of the antithrombin binding site better, probably increasing the strength of the electrostatic interaction with the amino acid residues. This picture of the binding process is compatible with the recently proposed two-step model [2,6] describing the interaction of heparin-derived oligosaccharides with antithrombin, with a minor extension: in the first step a complex between ligand and receptor is formed, accompanied by conformational changes in the tetrasaccharide AGA*IM, creating a complementary three-dimensional structure to fit the active conformation of the protein-binding site. During the second step, as observed in a structurally similar pentasaccharide , conformational changes in the binding site of the protein result in a relaxed state.
An analysis of the tetrasaccharide–antithrombin complex, using both 1H and 13C chemical shifts, homo- and heteronuclear coupling constants and relaxation measurements, led to a relatively detailed picture of this interaction.
The chemical shifts suggest considerable antithrombin shielding/deshielding effects, predominantly influencing atoms linked to charged groups, owing to strong electrostatic interaction. The effect was pronounced in GlcN,3,6-SO3 as well as in GlcN,6-SO3 residues, particularly for A1, A2, A6, A6′, A1*, A2*, A3* and A4*.
Protein-induced changes at the glycosidic linkage could be qualitatively monitored by the variations in chemical shifts and one-bond proton–carbon coupling constants.
An analysis of the magnitude of proton–proton and proton–carbon coupling constants in IdoA residue indicated that the conformational equilibrium is most likely shifted towards the 1C4 chair form in the presence of antithrombin.
NOE analysis in the free and the bound states revealed that the conformation at the glycosidic linkage between the GlcN,6-SO3 and GlcA residues changed during the binding process. This change was supported by the variations in two other NMR parameters. Changes in the three-dimensional structure also resulted in a different charge distribution in the AGA*IM molecule in the presence of antithrombin, which was due to the altered orientation of the SO3− and COO− groups.
The NMR experimental data confirmed the crucial role of charged sulfate and carboxylate groups in heparin-derived tetrasaccharide during the binding process with antithrombin.
The data presented are compatible with the two-step model [2,6] describing the complexation of heparin-derived saccharides with antithrombin. However, a minor modification in the first step is introduced: a complex between the ligand and the receptor is formed, accompanied by a conformational change in the ligand molecule. The driving force for such a change in ligand conformation appears to be a strong electrostatic interaction between AGA*IM and antithrombin. During the second step conformational changes [2,11] in the binding site of the protein result in latent conformation of antithrombin.
The authors are indebted to Dr M. Petitou from Sanofi Recherche for the tetrasaccharide sample and Professor B. Casu for useful discussions. Part of this work was supported by EC TMR Project CARENET-2 (ERB-FMRX-CT96–0025) and by the Italian Ministry of Research (Piano Nazionale Ricerca Farmaci, 199401–1023/518).