Robert
Pendrill
a,
Olof
Engström
a,
Andrea
Volpato
b,
Mirco
Zerbetto
b,
Antonino
Polimeno
*b and
Göran
Widmalm
*a
aDepartment of Organic Chemistry, Arrhenius Laboratory, Stockholm University, S-106 91 Stockholm, Sweden. E-mail: goran.widmalm@su.se
bDipartimento di Scienze Chimiche, Università degli Studi di Padova, Padova 35131, Italy. E-mail: antonino.polimeno@unipd.it
First published on 18th December 2015
The monosaccharide L-rhamnose is common in bacterial polysaccharides and the disaccharide α-L-Rhap-α-(1 → 2)-α-L-Rhap-OMe represents a structural model for a part of Shigella flexneri O-antigen polysaccharides. Utilization of [1′-13C]-site-specific labeling in the anomeric position at the glycosidic linkage between the two sugar residues facilitated the determination of transglycosidic NMR 3JCH and 3JCC coupling constants. Based on these spin–spin couplings the major state and the conformational distribution could be determined with respect to the ψ torsion angle, which changed between water and dimethyl sulfoxide (DMSO) as solvents, a finding mirrored by molecular dynamics (MD) simulations with explicit solvent molecules. The 13C NMR spin relaxation parameters T1, T2, and heteronuclear NOE of the probe were measured for the disaccharide in DMSO-d6 at two magnetic field strengths, with standard deviations ≤1%. The combination of MD simulation and a stochastic description based on the diffusive chain model resulted in excellent agreement between calculated and experimentally observed 13C relaxation parameters, with an average error of <2%. The coupling between the global reorientation of the molecule and the local motion of the spin probe is deemed essential if reproduction of NMR relaxation parameters should succeed, since decoupling of the two modes of motion results in significantly worse agreement. Calculation of 13C relaxation parameters based on the correlation functions obtained directly from the MD simulation of the solute molecule in DMSO as solvent showed satisfactory agreement with errors on the order of 10% or less.
Whereas the monosaccharide D-mannose is present in both human and bacteria, L-rhamnose (6-deoxy-L-mannose) is absent from the former, but quite common among bacteria of different species. It is a constituent of the core region of lipopolysaccharides (LPS),6 the backbone of O-antigen polysaccharides of LPS,7–11 and also in side-chains of polysaccharides,12,13 thereby leading to branched structures. Furthermore, it is not uncommon to find it as the terminal residue in an LPS14 thereby being responsible for some of the cross-reactivities between different bacterial serogroups (which are defined by their respective O-antigen polysaccharides).
When an L-rhamnose residue is joined to another L-rhamnose sugar this often occurs as either an α-(1 → 2)- or α-(1 → 3)-linkage. Oligosaccharides having these structural elements have been investigated in detail by different research groups15–18 and conformational descriptions with major and minor states have resulted from these studies. The dynamics of L-rhamnose-containing oligosaccharides have also been examined employing 13C NMR spin relaxation studies,19 analyzed by the model-free approach.20 Another parameter utilized to obtain detailed information on conformational preferences is heteronuclear three-bond coupling constants, 3JCH, but also homonuclear 3JCC, readily available using site-specifically 13C-labeled compounds, are likewise very useful when interpreted via Karplus-type relationships.21 Furthermore, the interpretation of different NMR observables is often aided by molecular dynamics (MD) simulations, from which a number of parameters can be calculated and compared to the experimentally measured ones.22,23
The disaccharide α-L-Rhap-α-(1 → 2)-α-L-Rhap-OMe (R2R), the [1′-13C]-isotopologue24 of which is shown in Fig. 1, is a model for one of the commonly observed structural elements described above. In a previous 13C NMR spin relaxation study of four different disaccharides25 it was observed that when DMSO-d6, in contrast to D2O, is used as solvent, the correlation time (τc) of the molecule becomes such that measurements of the 13C{1H} nuclear Overhauser effect outside the extreme narrowing regime are possible. Also, the NOE factor (1 + η) is <2.99 and magnetic field dependent, ranging from ∼2.8 at 7.0 T to ∼2.1 at 14.1 T for the four disaccharides studied, but >1.15 (i.e., the limit for slowly reorienting molecules). It should be noted that τc was dependent on the number of hydroxyl groups in each disaccharide, i.e., the higher the number of HO-groups interacting via hydrogen bonding to DMSO as an acceptor, the longer the correlation time. We have previously been able to show that 13C NMR spin relaxation data are reproduced by a combination of MD simulations and a stochastic approach, the diffusive chain model (DCM).26–28 In this model the global reorientation dynamics of the molecule is considered in conjunction with the dynamics of the flexible molecule which was represented by rigid units connected by joints, where significant torsional motions were possible. Herein we describe the conformational preferences of α-L-Rhap-α-(1 → 2)-α-L-Rhap-OMe (R2R) in water and dimethyl sulfoxide (DMSO) solutions based on MD simulations in the respective solvents, supported by transglycosidic 3JCH and 3JCC coupling constants. The 13C NMR spin relaxation data of [1′-13C]-R2R in DMSO-d6 at two magnetic fields are subsequently reproduced using the DCM in which the relevant degree of freedom, i.e., the ψ torsional angle is treated as the pertinent variable.
Stochastic approaches, when properly validated, can be a fast and accurate way to simulate and gather significant insight from NMR relaxation data. The stochastic model employed here for the interpretation of NMR observables is based on the choice of a set of relevant (slow) coordinates of the R2R molecule. This suggests that a posteriori validation is desirable. We have compared results, i.e., relaxation times and NOE, obtained from the model, which exhibit excellent agreement of calculated (without fitting parameters) and experimental NMR relaxation data, with a direct evaluation of the same observables obtained from MD simulations. The latter is nowadays an approach of choice, whenever long enough trajectories are available, which is not often the case for larger molecular systems. In this case, since R2R saccharide is a relatively small molecule, it is possible to access NMR relaxation times (agreeing to ∼10% error to experimental ones) directly from the MD simulations.
Finally, we have investigated the role of the coupling between the internal motion (i.e., rotation about ψ) and the global tumbling of the molecule. As discussed in the following, we show that the slow dynamics of the system arises from a strong interplay between these two degrees of freedom, suggesting that separation of time of scales between internal and global motions is not always valid and should be invoked with care.
The stochastic process associated with the relevant dynamics of R2R is represented by the general set Q = (Ω,θ), where Ω is the set of Euler angles transforming from the laboratory frame (LF) to a molecule-fixed frame (MF), cf.Fig. 1, and θ = (ϕ,ψ) describes all the torsion angles included explicitly in the (supposedly Markovian) set Q.28 Assuming that a diffusive Fokker–Planck description is acceptable, one can write the following time evolution equation for the conditional probability P(Q,t|Q0,t0) of the system29
![]() | (1) |
To proceed, we need to evaluate the diffusion tensor and the POMF. The diffusion tensor is obtained by a hydrodynamic approach, which considers the molecule as a flexible set of linked spheres. Details of the procedure are described elsewhere.30 Parameters entering the calculation are (i) the molecular geometry and (ii) the translational friction coefficient of a single sphere, given by Stokes's equation ξ = CRπη, as a function of the temperature-dependent bulk viscosity η, the effective radius of the spheres R and a constant C depending on the assumed boundary conditions.
In an isotropic medium, the POMF does not depend on the global orientation of the molecule. Therefore the equilibrium distribution depends only on the internal conformation, i.e. Peq(Q) = Peq(θ)/8π2. An estimate of Peq(θ) can be provided by unbiased or biased MD simulations, depending on the complexity of the system under study. Eqn (1) can be solved numerically using linear algebra methods employing a representation with a proper basis set spanning the vector space of functions defined in Q.31,32 NMR relaxation data are evaluated directly, in the Redfield limit, from second rank correlation functions. Explicit expressions for the calculation of T1, T2, NOE are given elsewhere.28,33 The whole procedure is implemented in the C++OPPS software package34 available for download under the GPL v2 license at the URL http://www.chimica.unipd.it/licc/software.html.
ϕ = H1′–C1′–O2–C2 and ψ = C1′–O2–C2–H2. |
An NMR sample of [1′-13C]-R2R24 was prepared by freeze-drying from D2O and dissolving it in DMSO-d6 (6.4 mg in 0.5 mL, 26 mM). The translational diffusion coefficient was measured using 1H pulsed-field-gradient (PFG) spin-echo experiments on a Bruker AVANCE III 600 MHz spectrometer (14.09 T) equipped with a 5 mm inverse Z-gradient TXI (1H/13C/31P) probe. In each experiment the PFG strength was increased linearly 32 times between 2% and 95% (100% = 55.7 G cm−1), with a PFG duration (δ) of 4 ms and a diffusion time (Δ) of 100 ms. The spectra were recorded with 32 scans using 16k data points for the acquisition, a spectral width of 14 ppm and an inter-scan delay of 5 s. The free induction decays (FIDs) were zero-filled twice and a 5 Hz exponential line broadening function was applied prior Fourier transformation. Translational diffusion coefficients were calculated using a protocol developed by Damberg et al.36 The PFG calibration was performed using a doped water sample (1% H2O in D2O + 1 mg mL−1 GdCl3) and a literature value of Dt = 1.90 × 10−9 m2 s−1 for the HDO resonance.37 The diffusion coefficients of [1′-13C]-R2R and DMSO-d5 were fitted to the integral decay of the resonances of the sugar bulk region (3.0–4.0 ppm) and to the solvent signal, respectively, and averaged values were calculated from nine independent measurements.
The 13C NMR relaxation experiments were performed on a Bruker AVANCE III 700 MHz spectrometer (16.44 T) equipped with a 5 mm TCI Z-Gradient Cryoprobe and a Bruker AVANCE III 600 MHz spectrometer (14.09 T) equipped with a 5 mm BBO probe. The experiments were recorded with the carrier set in the region (100–102 ppm) near the 13C-labeled anomeric resonance of [1′-13C]-R2R and with inter-scan delays of ≥5 s. Longitudinal relaxation times (T1) were measured with the fast inversion recovery experiment38 using 14 different relaxation time delays ranging between 0.01 and 4 s in a shuffled manner for each experiment. Spectra were recorded with either 32 or 16 scans, 8k or 16k data points and spectral widths of 70 or 190 ppm, at 14.09 T and at 16.44 T, respectively. An exponential window function of 3 Hz was applied prior Fourier transformation whereupon peak intensities were extracted. The relaxation times were then fitted based on the peak intensities using an in-house MATLAB (MathWorks, R2012a) script and average values were calculated from 17 and 24 independent experiments, at the lower and higher magnetic field strengths, respectively. Transverse relaxation times (T2) were measured using a Carr–Purcell–Meiboom–Gill (CPMG) pulse sequence.39 The delays between 13C refocusing pulses in the pulse trains were set to 0.25 or 0.50 ms and the total length of the pulse-train was varied ten times in a shuffled manner for each experiment with a total pulse-train length from 5 to 300 ms. Spectra were recorded with 16 scans, using either 16k or 32k data points, and spectral widths of 100 or 120 ppm; an exponential window function of 5 Hz was applied prior Fourier transformation. The T2 relaxation times were then calculated from 20 and 15 independent experiments, respectively, using the TopSpin software. The NOE enhancement was calculated from the anomeric intensity ratio of a steady state NOE experiment with one long 1H radiation time (5 s) and a short one (1 ms). Spectra were recorded with either 64 or 32 scans, 8k or 16k data points and spectral widths of 70 or 190 ppm; an exponential window function of 3 Hz was applied prior to Fourier transformation. Average NOE enhancement values were calculated from 18 and 24 experiments, respectively. The standard deviations for all the NMR relaxation data were ≤1%.
The 3JC1′,H2 coupling constant was measured from the H2 multiplet (δH 3.65) in a 1D 1H NMR experiment, using the J doubling methodology40,41 implemented in-house by a MATLAB script. The spectrum was recorded on the 700 MHz NMR spectrometer with 64 scans and an acquisition time of 2.3 s. Resolution enhancement was achieved by applying a Gaussian window function centered at 0.9 s in the FID together with an lb = −2 Hz. The 3JC1′,C1 and 3JC1′,C3 coupling constants were measured employing the J doubling methodology to the peak-to-peak separation of the C1 resonance (δC 99.90) and the C3 resonance (δC 70.43) in a 13C spectrum, recorded with 1k scans and an acquisition time of 4.0 s. Resolution enhancement was achieved by using a Gaussian window function centered at 1.2 s in the FID together with an lb = −2 Hz.
Additional NMR measurements were carried out on a sample of [1′-13C]-R2R in D2O (30 mg mL−1)24 at 14.09 T using a 5 mm BBO probe. The NOE experiments were recorded with 128 scans in addition to four dummy scans and with an inter-scan delay of 20 s. FIDs were acquired with 64k data points and were multiplied with an exponential window function of 3 Hz prior the Fourier transformation. The NOE enhancement was calculated as the intensity ratio of the anomeric resonance from C1′ using an experiment with a long (2 s) and with a short (1 ms) mixing-time. An average NOE value was calculated from ten independent measurements. Translational diffusion measurements, using the BBO probe, were performed as described above.
![]() | ||
Fig. 2 Probability distribution functions of the glycosidic torsion angles in R2R from the MD simulations in (a) water and (b) DMSO; ϕ (dashed line) and ψ (solid line). |
Water | DMSO | |||||||
---|---|---|---|---|---|---|---|---|
Expta | MD | ψ + (MD) | ψ − (MD) | Expt | MD | ψ + (MD) | ψ − (MD) | |
a Heteronuclear coupling constants are averages from different methods for their measurements according to Jonsson et al.24 b Root-mean-square deviations are given in parenthesis. c Calculated using eqn (9) from Säwén et al.59 d Calculated using eqn (10). e Calculated using eqn (8), with a constant-in-plane (CIP) addition.59 f Calculated using eqn (8), without CIP effect. g Hardy et al.60 h Populations derived using 〈3JC1′,H2〉, 〈3JC1′,C1〉 and 〈3JC1′,C3〉 in the ψ+ and ψ− states from the MD simulations. | ||||||||
Population MD | 0.412 | 0.588 | 0.680 | 0.320 | ||||
ϕ/° | 42.1 (11)b | 45.4 (12) | 39.9 (10) | 44.8 (12) | 46.4 (12) | 41.4 (12) | ||
ψ/° | −16.9 (29) | 13.3 (13) | −38.1 (16) | −0.1 (26) | 15.4 (13) | −33.1 (15) | ||
3 J H1′,C2 /Hz | 4.2 | 4.07 | 3.77 | 4.28 | 4.0g | 3.81 | 3.66 | 4.10 |
3 J C1′,H2 /Hz | 4.6 | 4.58 | 5.66 | 3.83 | 5.04 | 5.16 | 5.57 | 4.29 |
3 J C1′,C1 /Hz | 0.70 | 1.48 | 2.36 | 0.86 | 0.96 | 1.95 | 2.49 | 0.79 |
3 J C1′,C3 /Hz | 1.90 | 2.07 | 0.42 | 3.23 | 0.99 | 1.17 | 0.33 | 2.96 |
Population 〈3J〉h | 0.36 | 0.64 | 0.56 | 0.44 |
![]() | ||
Fig. 3 Newman projections of the ψ torsion angle showing the major conformational state (a) in water having 〈ψ〉 = −38° and (b) in dimethyl sulfoxide (DMSO) having 〈ψ〉 = 15°. |
From the MD simulation translational diffusion coefficients and DMSO solvent viscosity were also calculated. The disaccharide had D298t = 2.16 × 10−10 m2 s−1 whereas DMSO showed D298t = 7.24 × 10−10 m2 s−1 for a box of pure solvent. Additionally, the shear viscosity of the solvent was calculated as η298 = 2.32 cP, closely similar to that of DMSO-d6, η298 = 2.19 cP52 which was the solvent used for the NMR spin relaxation experiments. Thus, as the viscosity derived from the MD simulation agrees very well with that of the solvent used experimentally, the MD simulations should be suitable for calculation of NMR observables.
Spatial distribution functions (SDFs)53–55 were used to investigate the solvent structure around the disaccharide. For the major conformation of R2R in water and in DMSO regions with higher solvent density than the bulk were possible to identify. In water solution two conspicuous regions were evident, viz., (i) a water molecule acting as a hydrogen bond acceptor to the HO3 hydroxyl group of the rhamnosyl residue carrying the O-methyl group and (ii) a water molecule donating a hydrogen bond to the ring oxygen of the same sugar residue (Fig. 4a). It can be noted that for the R2R molecule inter-residual hydrogen bonding is not present, in contrast to e.g. the disaccharides cellobiose56 or sucrose;57 thus, its conformational preferences at the glycosidic linkage are indeed governed also by solvent cage interactions. In DMSO solution the hydroxyl groups of R2R are involved in hydrogen bonding to the oxygen atoms of the solvent acting as acceptors (Fig. 4b).
Different NMR relaxation data are often measured at more than one magnetic field strength in order to obtain a sufficiently large number of observables to fit to motional parameters of dynamic models.61,62 For small molecules in low viscosity solvents the rotational reorientation is rapid and the NMR relaxation parameters are independent of the magnetic field employed, i.e., the extreme narrowing region prevails where ω2τc2 ≪ 1.63 For a sample of [1′-13C]-R2R in D2O at 298 K the heteronuclear NOE factor at 14.1 T was determined herein revealing NOE = 2.70, i.e., quite close to the theoretical maximum of the NOE factor, where it will be difficult to extract reliable information about internal motion from such data.64 For the translational diffusion measurements, D298t = 4.59 × 10−10 m2 s−1, the molecular global correlation time was deduced, τc = 90 ps, consistent with that determined at 310 K, viz., τc ≈ 65 ps.65 In order to utilize the magnetic field dependence of, in particular, the heteronuclear NOE66 for the molecule under study, it was decided that DMSO-d6 would be the solvent of choice (cf.ref. 25 where the suitability of the solvent in conjunction with available magnetic field strengths for disaccharides had been established). 13C NMR spin–lattice relaxation times T1, spin–spin relaxation times T2, and heteronuclear NOE factors (1 + η) were thus measured for R2R in DMSO-d6 at two magnetic field strengths, viz., 14.1 T and 16.4 T, and these are given in Table 2. Notably, due to the very high 13C-labeling of 99 atom% at the C1′-position24 the quality of the data was very high and the standard deviations for all the NMR relaxation data was ≤1%. Translational diffusion was also determined for the disaccharide which had D298t = 2.83 × 10−10 m2 s−1 whereas DMSO-d6 in the same sample showed D298t = 5.28 × 10−10 m2 s−1, confirming that the MD simulations gave reasonable results also for these physical properties.
1H frequency/MHz | Experimental | DCM exact | DCM exact |err|% | MD | MD |err| % | DCM decoupled | DCM decoupled |err| % | |
---|---|---|---|---|---|---|---|---|
T 1/ms | 600.1 | 440.0 | 429.5 | 2.4 | 487.1 | 10.7 | 316.4 | 28.0 |
T 2/ms | 402.6 | 396.0 | 1.6 | 452.4 | 12.4 | 302.2 | 24.9 | |
NOE | 2.361 | 2.314 | 2.0 | 2.381 | 0.85 | 2.620 | 11.0 | |
T 1/ms | 700.0 | 475.6 | 469.5 | 1.3 | 525.1 | 10.4 | 335.7 | 29.4 |
T 2/ms | 432.9 | 426.4 | 1.5 | 481.8 | 11.3 | 317.5 | 26.7 | |
NOE | 2.215 | 2.188 | 1.2 | 2.290 | 3.4 | 2.536 | 14.5 |
![]() | (2) |
n | Re![]() |
Im![]() |
---|---|---|
0 | 11.6 | 0 |
1 | 5.51 | 0.842 |
2 | 1.21 | 0.362 |
3 | −1.03 | 0.598 |
4 | −0.542 | 0.557 |
5 | 0.335 | −0.192 |
6 | 0.205 | −0.356 |
7 | −0.185 | 0.132 |
8 | −0.11 | 0.310 |
9 | 0.0628 | −0.0211 |
10 | 0.0089 | −0.204 |
11 | −0.0568 | 0.0297 |
12 | 0.0501 | 0.0948 |
The comparison of experimental NMR data with calculations performed with C++OPPS is presented Table 2. The program implements the solution of a Smoluchwski equation by projecting the diffusive operator on a set of orthonormal basis functions , followed by direct tridiagonalization of the resulting symmetric matrix via Lanczos algorithm.28,31 Matrices with dimensions of the order of 103 need to be treated.28 The numerical simulation shows excellent agreements with the relaxation parameters at the two frequencies of 600.132 and 699.973 MHz (Table 2).
As was sketched in the Introduction, we employ the MD trajectory as a further way of validation of the model. To this purpose, we first test the validity of the MD simulation itself by calculating R2R NMR relaxation data. The R2R trajectory is referred to a reference frame, CF, centered on the center of mass of the molecule, and arbitrarily oriented over the molecule. Then, we extract the time series of the orientation of the 13C–1H bond, Ω(t), with respect to CF, and we calculate the Wigner matrix element D20,0(Ω(t)) function, which, in the secular approximation, is the only function used to define the dipolar–dipolar spectral density. In particular, given the time series of the Wigner matrix, we calculate the (normalized) autocorrelation function C(t) = 5〈D20,0(Ω(t))D20,0(Ω(0))〉. Finally, the spectral density for the 13C–1H dipolar interaction is calculated as the Fourier–Laplace transform of C(t). We stress here that in order to obtain a reasonable spectral density from the MD trajectory, the correlation function must firmly be at convergence since, among other information, the spectral density at zero frequency is required (which, in turn, requires that the long-time behavior of the autocorrelation function is calculated correctly). It is worthy to note that, despite of the apparent ‘simplicity’ of the system, a 100 ns long trajectory is barely enough for calculating a good correlation function. In practice, we split the trajectory into two trajectories of 50 ns each, which facilitated fitting of the correlation function to a bi-exponential decay. Autocorrelation functions are calculated for each of the two sub-trajectories, and then summed and renormalized. The resulting function is fitted with a bi-exponential functional form: C(t) = ae−ωat + be−ωbt. Best fitting parameters are: a = 0.471, ωa = 3.687 × 109 Hz, b = 0.529, and ωb = 1.456 × 1010 Hz. The spectral density is analytic and NMR data are easily accessed. Table 2 shows the results, obtained with a CH bond length of 1.111 Å (i.e., the value of the bond length in the MD force field, without the need of vibrational correction). The agreement with experiment is satisfactory, all data reproduced with an error of ∼10% or less.
From the MD simulations the T1 and T2 relaxation times and the NOE values are consistently longer and higher, respectively, than what are observed from experiments (Table 1). Thus, the correlation time τc from the MD simulation is shorter than from experiment, despite the fact that the shear viscosity of the solvent calculated from the MD simulation was slightly higher than that determined experimentally for DMSO-d6 as solvent (vide infra). This deviation in τc may be due to the fact that the potential describing the solute–solvent interaction is somewhat weaker than needed for a correct description of the rotational correlation time, i.e., its part in the ‘interaction triad’ would have to be optimized further.69
We also investigate the role of coupling between internal conformational motions and global tumbling. In our stochastic model, the global and internal motions are rigorously treated as coupled. We can easily explore what happens in absence of coupling between internal and global motions. In the DCM, the molecule is partitioned in two rigid fragments, which can rotate with respect to each other around the selected bond (ψ dihedral angle). When global tumbling is considered to be uncoupled from the torsional angle ψ, the decoupled correlation function reads
![]() | (3) |
NMR relaxation studies have been informative on the dynamics of several open-chain oligosaccharides built of two to five sugar units. Small, di- and trisaccharides, were extensively studied via MD simulations. Data on the flexibility at the glycosidic linkages can then be obtained through the calculation of mean-force potentials and correlation times. Information from MD can be merged with molecular rotational reorientation, deduced from experimental translational diffusion values, thus allowing the calculation of NMR relaxation data.55,70
In a recent study, an MD simulation was coupled to a model-free approach for the interpretation of 13C NMR T1 relaxation times of sucrose,64 thus assuming decoupling between global and internal dynamics. The D20,0 time correlation function for each 13C–1H pair was fitted to a double-exponential. The same approach was previously used in a study of trisaccharide dynamics.70 The calculated 13C R1 and R2 relaxation rates of sucrose were slightly underestimated in pure water, whereas in a water–DMSO mixture they were overestimated due to faster and slower reorientational dynamics, respectively. For these small saccharides, model-free analysis returned quite high ordering, that would suggest that the motion of the 13C–1H probe is confined in a quite narrow potential well. Also, τm and τe differed by just one order of magnitude. Comparing these conclusions to the disaccharide studied in this work (see Fig. 5), as well as the di- and tri-saccharides studied in ref. 26 and 28, respectively, an order parameter of 0.9 appears to be overestimated, since if one considers that thermal energy is about 2.5 kBT units at 298.15 K, the span at a ψ torsion likely covers a 120° range, while around a ϕ torsion, a 60° range is expected. Also, values of the rotational and internal parts of the diffusion tensors for the molecules suggest that motions may occur on similar time scales, requiring an explicit modeling approach, such as the DCM, for the coupling between motions occurring on closely related time scales, which was found to be the case in small oligosaccharide dynamics.26,28
Similar arguments apply to larger saccharides composed of four or five sugar units. In the model-free analysis of the tetrasaccharide LNnT the two inner sugar residues showed more restricted motions, i.e., the generalized order parameters S2 were higher and closer to unity than for the terminal sugar residue.71 Similar results were also drawn from a study of a branched tetrasaccharide.72 The trisaccharide melezitose in which a fructose residue is disubstituted by two glucosyl residues exhibited a larger S2 value for the fructose residue,73 although it can be anticipated to show some pseudorotation for the 5-atom-membered furanose ring. However, like in LNnT the central residue exhibited the highest generalized order parameter. In a study of a mannosyl-containing pentasaccharide 13C NMR T1 relaxation times were measured and the MD simulation was utilized as the basis for comparison and evaluation of overall hydrodynamic behavior and internal motions.74 From the MD trajectory, the spectral density function J(ω) was calculated explicitly by Fourier transformation of the angular correlation function, and subsequently calculation of the T1 value from its dependence on the combination of spectral densities. It was concluded that, within experimental error, the computed T1 values based on the MD simulation were in agreement with the experimentally measured ones. In a pentasaccharide, being a model for part of N-glycan structures of glycoproteins, the disaccharide N-acetyllactosamine substituted positions 2 and 6 of a mannosyl residue, i.e., for the latter one an additional degree of freedom exists; the flexibility for the disaccharide substituting O6 was higher, consistent with the fact that the ω torsion angle (O5–C5–C6–O6) populates at least two conformational states.75 Furthermore, the stereochemistry and the substitution patterns at the glycosidic linkage influence the dynamics, e.g., in an α-(1 → 4)-linked fucosyl-containing disaccharide the flexibility was higher than in the β-(1 → 4)-linked methyl α-cellobioside,25 which can either be due to the inherent difference between α- and β-linked pyranosides or stemming from the presence of an intramolecular interresidual hydrogen bond stabilizing the molecular structure, thereby resulting in a more rigid disaccharide. Furthermore, the role of rotational anisotropy, in addition to flexibility, was thoroughly investigated for the pentasaccharide LNF-1, in which a differential and larger flexibility was observed toward the reducing end of the oligosaccharide.76 We described LNF-1 at the DCM level, using an initial MD simulation to identify the relevant degrees of freedom.28 Of the eight glycosidic torsion angles those toward the terminal part of LNF-1 were identified as ‘rigid’ (in the sense of having faster relaxation relatively to the other torsions – see Section S2 of the ESI of ref. 28) whereas the other ones toward the reducing end of the molecule were considered flexible. Once the DCM with the ‘proper’ complexity had been deduced from MD simulations, Brownian dynamics simulations were adopted to obtain correlation functions, spectral densities and eventually 13C NMR relaxation parameters. Notably, as in this work, no further refinement was needed in order to observe good agreement (±10%) between simulated and experimental 13C NMR parameters.
Finally, closed-chain geometries can be addressed. In particular, cyclodextrins have been studied27 using both the model-free approach and a two-body general approach (Slowly Relaxing Local Structure, SRSL).77 In a model-free analysis of the α- and γ-cyclodextrins that contain six and eight glucosyl residues, respectively, in a cyclic fashion, the flexibility was found being slightly larger (lower S2) in the latter compound,78 presumably due to less strain in the cyclic molecule.
We stress that while it is true that MD simulations offer the possibility to include trivially all the relaxation processes occurring in a molecule, slow processes are still difficult to sample. The larger the molecule, the longer (in a non-linear way) is the simulation time required to gain sufficient statistics. Also, ergodicity is an important problem; standard thermostatic MD simulations are not assured to allow a complete exploration of the phase space, no matter how long (realistically) the trajectory is protracted.
Conversely, a model approach can be efficiently employed. Validation can be based, for relatively small systems, on long enough MD trajectories. In this work, a 100 ns long trajectory of the R2R molecule was used and also employed to recover the mean force potential adopted in the stochastic model.
We have shown that the relevant dynamics of a ‘short linear chain’ of sugar rings can be safely described using a DCM model, where internal flexibility is given in terms of dihedral angles. Our results suggest that a protocol in which (i) short MD simulations are used on one hand to detect the most relevant dihedral angles, and on the other hand, to access the potential of mean force, and (ii) hydrodynamics modeling is employed for the friction, can provide a complete parametrization of a diffusive coupled approach accurate enough to reproduce long-time dynamics behavior of oligosaccharides.
This journal is © the Owner Societies 2016 |