Email:
qiuxu@model.chem.umbc.edu or bush@model.chem.umbc.eduDepartment of Chemistry and Biochemistry
University of Maryland Baltimore County
Baltimore, MD 21228
Aug. 27, 1996
Research supported by NSF Grant DMB-91-05586
A number of strains of oral streptococci have been identified which coaggregate specifically both with other streptococcal species and with representatives of other genera such as actinomyces, (Kolenbrander and London, 1993). These coaggregations between strains which are known to be early colonizers of the freshly cleaned tooth surface are thought to play an important role in the first stages of formation of dental plaque. Many of these coaggregations are inhibited by sugars and there is considerable evidence that the species specificity of the coaggregation results from interaction of a lectin on the surface of one cell with a cell surface polysaccharide on the coaggregation partner.
Our research group has participated in studies of the chemical structure of the cell wall polysaccharides of S. mitis J22 (Abeygunawardana et al., 1990) as well as of a number of other streptococci (Abeygunawardana and Bush, 1993; Reddy et al., 1993). These studies have resulted in the identification of two distinct carbohydrate epitopes in these polysaccharides, one which is the receptor for lectin binding and the other which is the antigenic epitope (Cisar et al., 1995). The complete chemical structure of the cell wall polysaccharide of S. mitis J22 was found to contain seven sugar residues in the repeating unit which are linked through phosphate diester bonds as shown in
The lectin binding site is associated with residues d, e and f while residues a, b, c and g make up the antigenic site. All the sugar residues are of the D configuration except for the two rhamoses, which are L.
Cassels et al., (1995) have reported molecular modeling studies of this polysaccharide and have proposed a model in which strong hydrogen bonding interactions contribute to a folded structure. Our group has reported molecular modeling and NMR data on the conformation of antigenic tetrasaccharide site which indicated that the polymer is highly flexible with no single conformation which is consistent with the experimental data, (Xu et al., 1996 a). The model proposed in our work features three conformers which coexist in solution in fast exchange on the NMR chemical shift time scale. Subsequent NOE studies on the conformation of a heptasaccharide isolated from the polysaccharide were found to be consistent with the model for the polymer (Xu et al., 1996 b). This result indicates NOE intensities from the polymer and oligomer can both be accounted for with the same combination of conformations based on a simple isotropic model. Therefore we have concluded that effects of anisotropic motion are not likely to be significant for interpretation of NOE of the polysaccharide.
Most lectin-carbohydrate interactions are relatively weak with the dissociation constant in the high micromolar to millimolar range for binding of a single carbohydrate epitope by the lectin. For such weak interactions to give rise to the observed interactions between cells, the binding must be cooperative with multiple lectin binding epitopes of the polysaccharide on the surface of one cell binding to multiple lectin binding sites on the partner cell. Cooperative interactions of this type are important in many other biological phenomena. As a result of the cooperativity, the binding depends not only on conformation of the binding epitope itself but also on the conformation and dynamics of the links between the lectin-binding sites. The conformation of small oligosaccharide lectin-binding epitopes has been extensively studied. The results and the significance of flexibility of lectin-binding sites on the thermodynamics of binding have been discussed in numerous reviews, (Carver, 1993; Bush, 1992; Homans, 1993). But the flexibility of the entire polymer including the dynamics and conformation of the polysaccharide parts which link the lectin binding epitopes together has received less attention.
There has been the considerable discussion in the literature about the structural factors of oligosaccharides which influence flexibility. In general, the crowded stereochemistry of glycosidic linkages leaves a rather limited amount of conformational space available and a conformational map for a disaccharide often shows only a few minimum energy conformations. But it is clear that in some types of disaccharide linkage there are several accessible conformations conferring flexibility, (Rice et al., 1993). It appears that there are some comparatively rigid epitopes, such as those of the blood group oligosaccharides. For these types of oligosaccharides, it is possible to construct a simple conformational model based on NOE data and molecular modeling, (Miller et al., 1992). In our group we have used an NOE mapping procedure originally proposed by Brisson and Carver (1983) which can be adapted to NOESY data (Bush 1994). But other procedures have been used to find the conformations of relatively rigid oligosaccharides including restrained molecular dynamics simulation, (Rutherford et al., 1995). On the other hand, many oligosaccharides are comparatively flexible, with two or more distinct conformations in exchange. For those cases, it can be shown that no single conformation will fit the NOE data (Cumming and Carver, 1987; Poppe and van Halbeek, 1992; Poppe et al., 1994; Lommerse et al., 1995; Xu et al., 1996 a,b).
While it is possible with the NOE and molecular modeling methods to show that an oligosaccharide is not rigid, it has proven considerably more difficult to construct realistic models of flexible oligosaccharides on the basis of NOE data. One approach is the virtual conformation model of Cumming and Carver (1987) in which one that assumes a small number of discrete conformations exchange on time scale which is slow compared to the overall tumbling time. For this model, one calculates average values of the elements of the NOE relaxation matrix, rho and sigma. This procedure accounts for the r-6 weighting of dipolar relaxation which strongly weights short proton distances and also accounts for multiple spin relaxation pathways. But it has proven difficult to apply this model directly since the statistical weights of the conformations contributing to the average are not determined with sufficient accuracy by the molecular modeling methods, (Rutherford and Homans, 1994; Xu et al., 1996a). An additional uncertainty arises regarding the dynamics of conformational exchange which may not fit the assumptions of this model. The kinetics of conformational exchange in oligosaccharides which are complex and not well understood, may not be much slower than tumbling rates, (Hricovini et al., 1992). Lommerse et al., (1995) have discussed models for calculation of NOE averaged over multiple conformations of oligosaccharides for which internal motions are fast compared to the tumbling time. The number of parameters needed to define such models is much greater than the experimental data available, requiring either severe approximations or heavy reliance on molecular dynamics simulations for which the detailed reliability can be questioned.
Bonvin and Brunger (1996) have addressed the question of whether it is even possible to deduce conformations of flexible biopolymers from NOE data. They conclude that construction of models for flexible molecules will require additional experimental data to define the different contributing conformations and their statistical weights. A possible source of additional data for oligosaccharides is long range heteronuclear coupling constants across the glycosidic linkage which can be correlated with glycosidic dihedral angle ( Hamer et al., 1978). An advantage of the coupling data is that the measured parameter, 3JCH, is correlated directly with the key internal coordinate which determines conformation, the glycosidic dihedral angle. An additional advantage for molecules in conformational exchange, is that the appropriate time scale for averaging of coupling constants is more certain than for NOE. Since conformational exchange is certainly fast on the time scale of milliseconds, a simple average over the conformations is appropriate.
There have been several problems preventing extensive use of 3JCH data in determining oligosaccharide conformations. The primary problem is the difficulty of accurate measurements. But the measurement of coupling constants has been facilitated recently by the development of NMR techniques to accurately measure hetero-nuclear coupling constants, (Zhu et al., 1994; Biamonti et al., 1994). We have recently reported 3JCH data in oligosaccharides and polysaccharides which have been carefully checked against reliable values measured for model compounds (Gitti et al., 1994; Xu et al., 1996b). The availability of polysaccharides with uniform enrichment in 13C greatly facilitates these measurements. The interpretation of the data requires a precise correlation of 3JCH with glycosidic dihedral angle. Although detailed correlation functions have been proposed, (Mulloy et al.,1988; Tvaroska et al., 1989), their accuracy has been criticized since they fail to account for the positions of electronegative substituents on the coupling pathway, (Podlaske et al., 1995).
In this paper we report measurements of all the relevant coupling constants for a 13C enriched bacterial polysaccharide from S. mitis J22 by a 13C resolved 3-d NOESY method. We show that the coupling data can be incorporated in molecular modeling calculations to derive a flexible model for the heptasaccharide repeating unit of the polymer. We also report extensive NOE data for the labeled polymer derived from decoupled 3-d 13C resolved HMQC-NOESY spectra and we show that the NOE data fit the flexible model derived from the coupling data. We will also report in a separate paper heteronuclear 13C relaxation data showing internal motions both on the picosecond and on the nanosecond time scale which can be reconciled with our molecular model, (Xu and Bush, accompanying paper).
The polysaccharide sample from Streptococcus mitis J22 was 13C enriched to 95% according to the procedure described previously (Gitti et al., 1994). The sample was dissolved in 99.96% D2O at a concentration of 7-10mg/ml at neutral pH without any buffer.
1H-decoupled 3D HMQC-NOESY spectra were acquired on a GN-500 NMR instrument with a pulse sequence based on that described by Fesik and Zuiderweg (1988) A delay of 3.4 ms was employed in the HMQC part to develop the heteronuclear anti-phase magnetization and mixing time was 350 ms. The spectral width of the proton dimensions (t2 and t3) was 2500 Hz. The acquisition dimension (t3) contains 512 complex data points with eight transients. The evolution 1H dimension (t2) was obtained by acquiring 64 complex 2D planes. The 13C pulses were applied through f3 channel with carrier frequency in the middle of saccharide anomeric and ring 13C region with spectrum width of 13158 Hz. Therefore, methyl carbons were folded into the down-field side of the 13C dimension. The 13C evolution dimension was obtained by acquiring 64 complex blocks. The decoupling of 13C along acquisition dimension (t3) was accomplished through WALTZ-16 with a decoupling RF field strength of 4.0 KHz and 13C decoupling along t2 was realized with a 13C composite 180° pulse in the middle of proton evolution period (t1). The total acquisition time was about 64 hours for this 3D experiment.
1H-13C coupled 3D HMQC-NOESY was carried out on a GE-Omega 500 spectrometer using a pulse sequence based on that described by Fesik and Zuiderweg (1988). Since there was no 13C decoupling during proton evolution and acquisition periods, proton peaks doubled along each proton dimension due to coupling to their bonded 13C. The NOE mixing time was 200 ms. The spectral width for the proton dimensions (t2 and t3) was 2500 Hz. The acquisition dimension (t3) contained 512 complex data points with eight transients. The evolution the 1H dimension was obtained by acquiring 64 complex 2D planes. The spectral width of 13C was 14493 Hz with 64 complex blocks.
Felix 2.05 software was used to process the 3D HMQC-NOESY data. 3D matrices of 1024x128x1024 (t3 x t1 x t2) were generated in real data type. Linear prediction was used for the first point of FID's along the 13C evolution dimension to provide a flat baseline along the 13C dimension of the spectrum. FID's along the two evolution dimensions (t1 and t2) were extended from 64 points to 128 points with linear prediction before apodization with a cosine bell function. FID's along the 1H evolution dimension were zero filled to 1K data points. Fourier transformation was applied to each dimension followed by phase correction. The baselines along the two proton dimensions were flat and needed no further smoothing. All the cross peaks in 3D spectra were assigned according to previously reported chemical shift data, (Abeygunawardana et al., 1990). Each cross peak area on NOESY planes was picked out manually, and integrated with the Felix software according to assigned peak area.
Heteronuclear coupled HMQC-NOESY was processed with Felix 2.30 software (Biosym) on an SGI workstation. The data along proton and carbon dimensions were zero-filled to 1024 points and 128 points respectively with linear prediction for the first point of each FID along the 13C dimension. All the FID's were apodized with a cosine bell function before Fourier transformation. All the cross peaks in heteronuclear coupled HMQC-NOESY were assigned by matching chemical shifts of middle points of 13C-coupled protons and previously assigned data. The 3JCH values were obtained from NOESY by a shift and match scheme as in E.COSY. On a NOESY plane of carbon labeling Cx, two sets of row vectors containing cross peaks of Hx and Hy were selected and the vectors were truncated to contain only the cross peak of interest (Hx/Hy). After reverse Fourier transformation, the new time domain data were zero filled and Fourier transformed with a resolution of 0.32 Hz. The two new vectors were saved in ASCII. Their amplitudes and frequencies were matched with a self-coded program based on the singular value decomposition method (SVD) (Press et al., 1992). The frequency shift which gave minimum root mean square difference for the above SVD match of the two vectors is the long range heteronuclear coupling constant, 3JCH.
Molecular modeling for the polysaccharide was done with CHARMm version 22 (Polygen). The force field parameters were those of Polygen which are based on CHARMm parameters as modified for carbohydrates by Ha et al., (1988). The residue topology file (RTF) was modified to include L-rhamnose for which charges were assigned from fucose. It was also necessary to add beta galacto furanose to the RTF which was based on alfa arabinofuranose.
The dielectric constant was set as distance dependent (RDIE) in these calculations. The non-bonded interaction energy (van der Waals and electrostatic energy) was calculated with a smooth- switch function, (Brooks et al., 1983) with a cutoff distance of 15.0 Å. A visual check on conformations was carried out with the QUANTA molecular graphics software(Biosym).
Starting conformations for molecular modeling of the heptasaccharide subunit of the polysaccharide were selected as follows. The dihedral angles across the glycosidic linkages of the segment, a-b(-g)-c were taken from the previously established conformation for this antigenic site (Xu et al., 1996a). The remaining dihedral angles across the glycosidic linkages for the repeating heptasaccharide were selected to give reasonable starting conformations based on 3JCH data as discussed below. Selected conformations were then energy minimized with 300 steps of steepest descent followed by 500 steps of conjugate gradient minimization.
None of these individual starting conformations for this flexible polysaccharide was adequate to model the experimental NMR data. Therefore, linear combinations of individual starting conformations were sought which would reproduce the experimental 3JCH data and eventually the NOE data as well. The search was based on a linear combination of 3JCH values of individual models using the singular value decomposition (SVD) method (Press et al., 1992).
where SIGMA pi = 1 and pi is a statistical weight of conformer i. 3JoCH refers to observed experimental value of the coupling constant and 3JcCHi refers to the values calculated from a single conformation of the model. 3JcCHi was calculated with the following equation, (Tvaroska et al., 1989)
where theta = PHIH is defined by the four atoms, H1-C1-O-Cx or PSIH is defined by the four atoms, C1-O-Cx-Hx. According to the IUPAC approved definition, the glycosidic dihedral angle PHI is defined by O5-C1-O-Cx (O4-C1-O-Cx for furanose) and PSI is defined by C1-O-Cx-Cx-1. The two definitions are correlated depending on the configuration at C1 or Cx. For example, the relationship PHI = PHIH + 120° holds for an alfa 1-> 4 linkage of a D sugar, and PSI = PSIH - 120° for a beta 1-> 4 linkage of a D sugar. For 1->6 linkages, an additional dihedral angle omega is defined by the four atoms O6-C6-C5-O5. For the 1->6-linked galactofuranose residue, an additional conformational dihedral angle, gamma is defined by the four atoms C6-C5-C4-O4 in addition to omega for the 1 -> 6 linkage.
The pucker of the galactofuranose residue d is described by a single variable called phase angle, P. It is defined as (Saenger, 1983):
In this equation, nu0 is defined by the four atoms C4-O4-C1-C2, nu1 is defined by O4-C1-C2-C3, nu2 is defined by C1-C2-C3-C4, nu4 is defined by C2-C3-C4-O4 and nu4 is defined by C3-C4-O4-C1.
The statistical weights, pi, extracted from eq. 1 by the method described above were used to calculate the expected NOE for the model for comparison with experimental data. For this purpose, the 'virtual conformation' model was used in which the NOESY relaxation rate matrix elements were weighted and the average matrix was used to calculate the NOE for the model, (Keepers and James, 1984; Borgias and James, 1989). The quality of simulated NOE intensities was judged by calculating the crystallographic-type R-factor and sixth-root weighted Rx factor (James, 1991):
In order to probe the dynamics of this flexible polysaccharide and for mapping the dihedral angle space accessible to those multiple conformers, dynamics simulation was run for the conformers derived from the best 3JCH data fitting. Dynamics simulations were run both with and without dihedral angle constraints derived from the experimental coupling constants. The constraint for 3JCH was constructed as a parabolic function (Mierke et al., 1994; Eberstadt et al., 1995)
In this equation,3JoCHi is an observed 3JCH value and 3JcCHi is a calculated value from the model with Equation 2. The force constant KJ of the constraint was set as 1.0 kcal/mol Hz2. The atomic force from the above penalty function (Equations 2 and 6 ) required for molecular dynamics simulation is obtained from the partial derivatives: (Mierke et al., 1994; Eberstadt et al., 1995).
where ri is the distance between C and H. The penalty energy and atomic force was incorporated in the "user subroutines" in the CHARMm source codes.
The dynamics simulation was carried out in vacuum. All the bonds with hydrogens were constrained with SHAKE using 1 × 10-10 tolerance. The energies of the starting conformers were minimized with 200 steps of steepest descent followed by 200 steps of conjugate gradient minimization. Then the conformer was heated from 0K to 300K in 3 ps steps with a 5K increment. The conformer was later equilibrated at 300K for 1ps before 1.0 ns dynamic simulation. The Verlet algorithm was used to compute the molecular dynamics trajectories. The constrained dynamics simulation at 300K was run with a constant temperature method at an equilibrium rate of every 5 fs with a coupling decay time of 0.001 ps.
The long range heteronuclear coupling constants 3JCH across the glycosidic linkages of this polysaccharide obtained from 3D H1-C13 coupled HMQC-NOESY spectrum are given in Table 1.
Table 1 Experimental and calculated 3JCH of polysaccharide S. mitis J22
Cal.(HZ) Exp.(Hz) Cal.(Hz) Exp.(Hz) 3JCH(phi)ab 1.70 1.50 3JCH(psi)ab 2.31 1.50 3JCH(phi)bc 1.73 2.00 3JCH(psi)cb 2.10 2.10 3JCH(phi)gb 1.48 1.50 3JCH(psi)gb 3.39 4.10 3JCH(phi)cd 1.46 1.50 3JCH(psi)cd 2.18 1.90 3JCH(phi)de 2.04 1.50 3JCH(psi)de 1.40 1.30 3JCH(phi)ef 2.71 2.40 3JCH(psi)ef 2.36 1.30 standard deviation = 0.5 (Hz)
In the 13C decoupled NOESY spectrum, cross peak volumes are related to proton proximities providing additional information on the polysaccharide conformation. 2D NOESY spectra of the polysaccharide provide a few cross peaks across the glycosidic linkages, but many inter-residue cross peaks overlap because ring protons tend to appear within a one ppm chemical shift region. The NOESY cross peak separation along the carbon dimension in 3D HMQC-NOESY as discussed above makes peak integration accurate, providing many crucial long range proton cross peaks. The cross peaks reported in Table 2 are an extension of the data reported previously by Xu et al., (1996a) in a paper which treated the conformation of only the antigenic site of this polysaccharide.
Table 2. 13C enriched polysaccharide of J22 with 350 ms mixing at 24.0 C
Peaks Cal. Exp. Peak Cal. Exp. aH1-bH1 0.0053 0.0052 bH5-cH2 0.0007 0.0070 aH1-bH2 0.0488 0.0340 gH2-cH2 0.0001 0.0003 aH1-bH3 0.0237 0.0159 gH2-cH5 0.0004 0.0019 aH1-bH4 0.0030 0.0051 gH3-cH5 0.0042 0.0004 aH1-bH5 0.0019 0.0010 gH5-cH5 0.0169 0.0081 aH1-gH1 0.0285 0.0130 cH1-dH6 0.0301 0.0313 aH1-gH4 0.0007 0.0008 cH1-dH6'0.0240 0.0226 aH1-gH5 0.0040 0.0017 cH1-dH5 0.0040 0.0143 aH2-bH1 0.0006 0.0005 cH1-dH4 0.0046 0.0069 aH2-bH2 0.0050 0.0062 cH2-gH2 0.0001 0.0003 aH2-bH3 0.0031 0.0018 cH2-dH6 0.0019 0.0015 aH2-gH1 0.0050 0.0038 cH2-dH6'0.0020 0.0023 aH2-gH5 0.0005 0.0020 cH2-dH5 0.0006 0.0022 aH3-bH3 0.0032 0.0037 cH2-dH4 0.0018 0.0020 aH3-bH1 0.0019 0.0011 cH2-eH3 0.0001 0.0037 aH3-gH1 0.0144 0.0063 dH5-fH5 0.0001 0.0010 aH4-bH3 0.0033 0.0051 dH4-dH1 0.0038 0.0026 aH5-bH1 0.0035 0.0023 dH4-eH5 0.0013 0.0006 aH5-bH3 0.0374 0.0256 dH4-fH5 0.0012 0.0094 aH5-bH4 0.0127 0.0033 dH1-dH2 0.0289 0.0147 aH5-bH5 0.0032 0.0014 dH1-dH3 0.0035 0.0057 aH5-gH1 0.0081 0.0038 dH5-dH1 0.0016 0.0017 aH6-bH3 0.0025 0.0054 dH1-eH1 0.0005 0.0003 aH6'-bH30.0038 0.0045 dH1-eH4 0.0019 0.0035 bH1-cH1 0.0013 0.0019 dH1-eH5 0.0037 0.0035 bH1-cH2 0.0025 0.0079 dH1-eH6 0.0241 0.0129 bH1-cH3 0.0130 0.0197 dH1-eH6'0.0269 0.0099 bH1-cH4 0.0196 0.0437 dH1-fH4 0.0012 0.0031 bH1-cH5 0.0026 0.0040 dH2-eH4 0.0030 0.0038 bH1-gH1 0.0020 0.0084 dH2-fH2 0.0001 0.0002 bH1-gH5 0.0039 0.0053 dH2-fH4 0.0006 0.0010 bH2-gH1 0.0174 0.0253 dH2-fH5 0.0001 0.0043 bH2-gH5 0.0283 0.0120 eH1-fH2 0.0025 0.0077 bH2-cH1 0.0013 0.0004 eH1-fH3 0.0501 0.0282 bH2-cH3 0.0164 0.0043 eH1-fH4 0.0194 0.0058 bH2-cH4 0.0029 0.0066 eH1-fH5 0.0041 0.0018 bH2-cH5 0.0025 0.0027 eH2-fH3 0.0025 0.0017 bH3-gH1 0.0049 0.0029 eH4-fH3 0.0008 0.0083 bH4-gH1 0.0118 0.0034 eH5-fH3 0.0047 0.0084 bH4-gH2 0.0074 0.0034 eH5-fH4 0.0081 0.0012 bH4-cH2 0.0002 0.0011 eH5-fH2 0.0004 0.0007 R = 0.01b Rx = 0.002b ------------------------------------------------ a. NOE are calculated from the statistical weights of Table 4 with an effective correlation time of 1.4 ns b. R and Rx are crystallographic-type R-factor and sixth-root weighted Rx-factor (James,T. L. 1991)
In order to interpret the NOE and 3JCH data in terms of molecular conformations, computer generated molecular models were built. We have previously reported modeling studies on a fragment of four residues ( a, b, c and g) which make up the antigenic site. With the additional experimental data reported in this paper, the modeling is extended to include the complete heptasaccharide repeating unit including the lectin binding site. The phosphate is not included in the modeling since we lack data which determine the conformation of that linkage. Our recently reported NMR data on a dephosphorylated heptasaccharide indicate that the phosphate does not have a great effect on the conformation (Xu et al., 1996 b) and preliminary calculations indicate that the phosphate group neutralized by a H+ ion exerts little influence on the conformation of the seven residues. (Data not shown).
Several lines of evidence indicate that the heptasaccharide is likely to be quite flexible. First, the NMR lines for this polysaccharide are relatively narrow indicating that there must be substantial internal motion. Second, even though the branched structure of the antigenic site might be expected to be somewhat rigid, nevertheless, our previous studies (Xu et al., 1996a) suggest that multiple conformations exist. The lectin binding site, with a furanoside and 1->6 linkages is expected to be even more flexible.
Building conformational models for flexible oligomers or polymers is a challenging task for which no clear general solution exists in spite of many efforts devoted to flexible peptides,( Bruschweiler et al., 1991) and nucleic acids,( Ulyanov et al., 1995; Gonzalez et al., 1995) as well as carbohydrates (Rutherford and Homans, 1994; deWaard et al., 1992; Brisson et al., 1992; Lommerse et al., 1995). Using NOE intensities as distance constraints for flexible molecules requires careful consideration to avoid physically unrealistic "virtual conformations"(Jardetzky, 1980). The NOE mapping procedure, which is reasonably effective for rigid oligosaccharides has been shown not to be satisfactory for more flexible molecules (Xu et al., 1996a).
Therefore we have taken a novel strategy to building the model in which the 3JCH values are used to guide the conformational search. We had previously observed that the 3JCH data for the antigenic tetrasaccharide provided a useful source of information which could be easily translated into starting conformations by means of eq. (2). (Xu et al., 1996a). Therefore, three conformers of the antigenic site determined in that earlier study were combined with conformations of the remaining residues ( d, e and f) which were consistent with the measured 3JCH. The dihedral angle values in the antigenic segment, a-b(-g)-c, were taken from conformers 3, 4 and 8 of Xu et al., (1996a) and the remaining glycosidic dihedral angles for the starting heptasaccharide conformations were set at values calculated from the measured values of 3JCH according to the calibration curve of eq. 2. Since the relationship between 3JCH and dihedral angles is trigonometric, there are four possible dihedral angle values for each measured 3JCH. In addition, because H6 and H6' of residues d and e were not specifically assigned as ProS or ProR, it was necessary to set starting conformations with all four possible combinations of these assignments for the two 1->6 linkages in the heptasaccharide as illustrated in Tables 3 and 4.
Table 3. The Number of conformers before energy minimization
Conformation Conformation Conformation H6's assignment on d, e family 3 family 4 family 8 H6R on d & H6R on e 233 234 234 H6R on d & H6S on e 275 275 275 H6S on d & H6R on e 230 230 230 H6S on d & H6S on e 271 271 271
Table 4. The Number of conformers after energy minimization
Conformation Conformation Conformation H6's assignment on d, e family 3 family 4 family 8 H6R on d & H6R on e 134 155 48 H6R on d & H6S on e 106 84 70 H6S on d & H6R on e 83 53 48 H6S on d & H6S on e 103 57 75
After energy minimization, the conformers listed in Table 4 within a minimum energy of 20 kcal/mol were selected as the conformation pool for matching with the experimental data, 3JCH and NOE. As might be anticipated for a flexible molecule, no single conformer from this pool matches the 3JCH data and therefore a linear combination of conformations from this pool was sought which could reproduce the 3JCH data. A realistic model for this flexible polysaccharide segment presumably requires many conformations in exchange, but with the limited number of experimental 3JCH data given in in Table 1, we must use a limited number of conformations to avoid an under determined system. We have chosen the number of conformations as six, a compromise value based on the number of experimentally observed 3JCH in Table 1.
The strategy for selecting from the pool of low energy conformations in Table 4 is as follows. From the entire pool of conformations, we pick one from each of the three most important conformations of the antigenic site, model 3, 4 and 8 from the study of Xu et al., (1996a). Next are selected combinations of conformations of the remaining residues ( d, e and f) along with combinations of the unknown dihedral angles and the assumptions concerning the assignments for the H6's as indicated in the rows of Table 4.
With the conformation pool shown in Table 4, a selection of six conformers from three families requires at least two billion possible combinatorials for one row in Table 4. In order to obtain the best match in reasonable time, the searching process was carried out by first fixing conformers from family 8 followed by searching conformers from families 3 and 4 to best fit experimental data. Then conformers were fixed from family 3 and conformers from families 4 and 8 were searched to best fit the experimental data. The statistical weights of conformers for each linear combination were obtained with singular value decomposition with the minimization of the root mean square difference of 3JCH between experiment and the model, (Press et al., 1992). This strategy lacks rigor in that it does not search all possible combinations of the conformers of Table 4. However, alternative strategies which fixed the conformation of either family 3, 4 or 8 give essentially the same statistical weights of the major conformers if the search time or the number of search steps is made sufficiently long. Elimination of either family 3 or 4 does not satisfactorily reproduce the coupling constants within experimental error but elimination of family 8 has a less important influence on the fitting.
In the best matching combination of conformers, which is listed with statistical weights in Table 5, both H6 protons (the upfield resonance) were assigned as H6S.
Table 5 Statistical weights of conformers from linear 3JCH fitting
Conformera coefficient pi conformer 322-7 0.4369 conformer 322-48 0.3247 conformer 422-82 0.2343 conformer 422-239 0.0009 conformer 822-71 0.0013 conformer 822-182 0.0019 a. Conformers are defined in Table 6
Table 6 Glycosidic dihedral angles of conformers of flexible polysaccharide from S. mitis J22
a-b b-c g-b c-d d-e e-f phi psi phi psi phi psi phi psi gamma omega Phase phi psi omega phi psi Conf. 322-7 52.2 71.1 44.9 67.6 -54.8 159.0 -64.7 -176.0 -87.0 65.5 -27.3 -29.7 180.0 -165.0 -68.2 -169.0 Conf. 322-48 52.9 66.3 49.2 69.8 -59.5 162.0 -49.1 98.5 -81.2 35.8 37.8 -169.0 -174.0 37.3 -80.2 175.0 Conf. 422-82 82.6 168.0 140.0 62.6 -66.7 154.0 -57.8 -160.0 -79.5 68.0 -42.4 -151.0 -176.0 68.0 -77.8 -158.0 Conf. 422-239 79.3 160.0 135.0 63.2 -83.4 143.0 -31.2 -65.5 -67.9 172.0 -28.0 -65.2 -93.3 60.9 -85.5 -148.0 Conf. 822-71 63.1 104.0 88.3 -46.5 -85.8 77.0 -67.3 -139.0 -78.6 65.3 -44.9 -47.6 172.0 170.0 -76.3 -166.0 Conf. 822-182 61.3 108.0 59.2 -78.4 -86.2 122.0 -3.48 75.6 -79.3 160.0 -29.0 -62.4 -81.3 75.9 -33.8 -149.0
While none of the individual conformers in Table 6 can reproduce the experimental 3JCH data of Table 1, the linear combination of these conformers with the statistical weights in Table 5 match experimental results with a root mean square difference between the calculated and experimental data of 0.50 Hz, which is essentially the experimental error of 3JCH measurement. We recognize that a more realistic model of the oligosaccharide subunit should contain more than six conformers in exchange but the model of Table 5 is adequate to reproduce the experimental data at hand. It also provides an instructive model for interpretation of the NOESY data as well as dynamic data on heteronuclear relaxation rates (Xu and Bush, accompanying paper).
Since this combination of six conformers fits the 3JCH data well, it was necessary to test whether the model could also reproduce the NOESY data. For this purpose, a complete relaxation matrix method was used, in which the average values of the matrix elements were calculated with the statistical weights given in Table 5, (Cumming and Carver, 1987; Bush, 1994). A value of an effective rotational correlation time, which is needed for this method, was taken to be 1.4 ns from our previous work (Xu et al., 1996a). The one-bond 1H-13C dipolar auto-relaxation rate was included in all the diagonal terms of the relaxation matrix. This effect was found to be significant to properly account for the proton auto-relaxation rates for a uniformly 13C enriched polysaccharide at this effective rotational correlation time. In Table 2, the quality of agreement with experiment of the simulated NOE intensities is evaluated with the two X-ray type indices (R and Rx) as described in Materials and Methods above. The index R reflects the match for short distance NOE data and RRx reflects the match for long distance NOE data. Although this model was constructed with 3JCH data and no NOE constraints were used in defining it, nevertheless simulation with this ensemble of conformers is quite satisfactory. The indices of quality are favorable in comparison with results reported for flexible DNA conformations (Ulyanov et al., 1995).
For the model of Table 5, which is based solely on 3JCH data, the statistical weights of conformations from family 8 are negligibly small. Our previous modeling of the antigenic site based on 3JCH data gave only conformers 3 and 4. Conformer 8 was included in that model only to match experimental NOE intensities of the cross peak observed for bH1-cH5. The statistical weight reported here for conformers from family 8 is based only on 3JCH data with NOE data used only as a check. In fact difference between experimental and simulated intensities of the cross peak between bH1-cH5 in Table 2 is approximately the experimental error of ± 0.01 suggesting that a model with only the three major conformers accounts for our data. The phase angle (P) of residue d, the galactofuranose, varies between -45° and 38°, a region corresponding to the sugar pucker described as C1-endo, C2-exo, and C3-endo for a furanose ring.
A model of the type proposed in Table 5 above based on a linear combination of a small number of distinct conformations can be called a space averaged model, (van Gunsteren et al., 1994). An alternative way of describing a flexible model is the time averaged method in which a molecular dynamics trajectory is used. Several methods have been described including free molecular dynamics (MD), restrained molecular dynamics (r-MD), or time averaged restrained molecular dynamics (tar-MD), (van Gunsteren et al., 1994). MD trajectories of our heptasaccharide model were started from each of the three major conformers of Table 5 and run for 1 ns both as free MD and restrained by 3JCH with constraints of 1.0 kcal/mol Hz2. A comparison of 3JCH and NOE calculated over these trajectories with experimental results is given in Table S1 in the Supporting Information. In general the values of 3JCH calculated over the free MD trajectories do not agree well with experimental values showing standard deviations of 1.3 to 1.7 Hz, well outside experimental error. The results for r-MD with 3JCH constraints are considerably better but still the agreement of calcualted 3JCH values for this time averaged model is not as good as with the space averaged model of Table 6. The agreement with experimental data of NOE values calculated from both MD and r-MD is not as good as that with the space averaged model.
Table S1 MD and r-MD time-averaged coupling constants (Hz)
MD | r-MD 322-7 322-48 422-82 | 322-7 322-48 422-84 3JCH(phi)ab 1.4 1.1 4.1 | 1.4 1.5 3.1 3JCH(psi)ab 1.9 2.0 2.3 | 1.7 2.2 2.2 3JCH(phi)bc 4.4 5.2 4.1 | 3.2 1.1 4.8 3JCH(psi)bc 3.8 3.2 3.6 | 2.9 2.5 11.9 3JCH(phi)gb 2.8 1.7 4.2 | 1.0 1.8 2.5 3JCH(psi)gb 3.6 2.8 4.6 | 3.2 3.9 4.0 3JCH(phi)cd 2.7 0.7 2.5 | 2.3 1.1 1.6 3JCH(psi)cd 2.5 3.9 2.1 | 0.7 3.9 0.5 3JCH(phi)de 1.5 1.9 1.0 | 0.6 2.1 2.9 3JCH(psi)de 3.2 4.7 2.1 | 1.8 0.9 1.4 3JCH(phi)ef 2.2 2.9 3.1 | 2.6 1.9 2.4 3JCH(psi)ef 3.6 3.2 2.7 | 2.2 3.2 2.6 sigma 1.3 1.7 1.5 | 0.77 0.94 1.2 R 0.003 0.05 0.20 | 0.26 0.06 0.34 Rx 0.0007 0.012 0.053 | 0.053 0.014 0.099
Sample dihedral angle plots from the MD and r-MD trajectories started from each of the three major conformers of Table 5 are shown in Figs. S1 and S2 of the supplementary material. While some transitions are seen among the low energy conformers, it is obvious that the full conformational space is not sampled uniformly in any of these 1 nsec trajectories. In the r-MD simulations, relatively weak constraints were used in an attempt to avoid forcing the trajectories into higher energy virtual conformations. The r-MD trajectories show more limited excursions than the free MD trajectories with a few transitions among local energy minima. The trajectories also show that the lectin binding site has greater mobility than the antigenic site with significant fluctuations in the furanoside puckering angle and also the dihedral angles OMEGA and GAMMA.
Although it has been reported that restrained MD simulations using NOE-based constraints leads to a satisfactory model for a large oligosaccharide, the penta-antennary 15 residue N-linked glycopeptide studied by Rutherford et al. (1995) is considerably more rigid than is the polysaccharide from S. mitis J22. NOE constraints did not lead to a good model for our polysaccharide while the approach based on 3JCH was successful. Bonvin and Bruenger (1996) have proposed that it may not be possible to uniquely and accurately describe flexible biopolymer models from NOE data. We conclude that NOE data are not adequate to determine models for flexible oligosaccharides and that the use of accurately determined 3JCH data across the glycosidic linkage will be extremely valuable for constructing models primarily because of the simplicity of the time averaging of these data compared to the difficulty of averaging NOE data in flexible polysaccharides, (Lommerse et al., 1995). We have used 3JCH data to generate starting conformations and also to judge the quality of a flexible model based on exchange among a modest number of distinct conformations. We have shown that molecular modeling constraints in MD simulations based on 3JCH data are also useful for refining structures. The models generated by these procedures fit the NOE data quite well. We conclude that NOE data are not so sensitive to the nature of a flexible model and that their value in describing flexible conformations is much less than their value for describing rigid structures.
We reported results for a space averaged model in which a modest number of conformers was chosen and statistical weights calculated to fit the experiment. While this model is not completely realistic since it does not have enough conformers to fully represent a flexible oligosaccharide, it provides a simple model for the flexible polysaccharide which reproduces both the 3JCH and the NOE data generally better than does the time-averaged model calculated over a 1 ns trajectory. While it is possible that restrained MD trajectories longer than 1 ns could also provide models which would reproduce both NOE and 3JCH data, detailed interpretations of long trajectories are cumbersome.
Cassels et al. (1995) have reported modeling of the conformation of the polysaccharide from S. mitis J22. Although their paper does not quote precise structural parameters, the description of structural features of their model suggests a hairpin bend at the lectin binding site somewhat similar to our major conformer 322-7 (Table 6 and Fig. 1a). But their model does not account for the flexible nature of this polysaccharide which is indicated by our experimental NMR data which include 3JCH, NOE and also 13C relaxation rates (accompanying paper). Their model implies that the residues around the lectin binding site form a hydrogen- bonded structure with limited internal motion of the dihedral angles of the glycosidic linkages. They also emphasize the role of the phosphodiester linkage in the formation of the folded hydrogen-bonded structure, a feature which is not supported by our studies on the conformation of the isolated heptasaccharide (Xu et al., 1996 b) nor by our calculations.
Our flexible model shows some fluctuation of the antigenic site conformation; we have previously compared data for this tetrasaccharide with those for "rigid" oligosaccharide epitopes such as blood group oligosaccharides (Xu et al. 1996 b). But the models of Figure 2 and Table 6 show that there is more variability in the conformation of the lectin binding site. This result, which is not surprising when one considers the influence of the 1->6 linkages and the furanoside puckering of residue d is also supported by the MD simulations. In the accompanying paper (Xu and Bush, accompanying paper) we report 13C relaxation data which show that there is a large amplitude of internal motion on the nanosecond time scale in the lectin binding site.