Glycobiology vol. 13 no. 4 pp. 255±264, 2003
DOI: 10.1093/glycob/cwg031
Predicting the molecular shape of polysaccharides from dynamic
interactions with water
Andrew Almond1,2 and John K. Sheehan3
2
Department of Biochemistry, University of Oxford, South Parks Road,
Oxford OX1 3QU United Kingdom; and 3 Department of Biochemistry
and Biophysics, University of North Carolina at Chapel Hill, Chapel Hill,
NC 27599, USA
Received on July 29, 2002; revised on October 30, 2002; accepted on
November 21, 2002
How simple monosaccharides, once polymerized, become the
basis for structural materials remains a mystery. A framework is developed to investigate the role of water in the
emergence of dynamic structure in polysaccharides, using
the important b(1 ! 4) linkage as an example. This linkage
is studied within decasaccharide fragments of cellulose, chitin,
mannan, xylan, and hyaluronan, using molecular simulations
in the presence of explicit water solvent. Although cellulose,
mannan, chitin, and xylan are chemically similar, their intramolecular hydrogen-bond dynamics and interaction with
water are predicted to differ. Cellulose, mannan, and chitin
favor relatively static intramolecular hydrogen bonds, xylan
prefers dynamic water bridges, and multiple water configurations are predicted at the b(1 ! 4) linkages of hyaluronan.
With such a variety of predicted dynamics, the hypothesis that
the b(1 ! 4) linkage is stabilized by intramolecular hydrogen
bonds was rejected. Instead, it is proposed that favored molecular configurations are consistent with maximum rotamer
and water degrees of freedom, explaining observations made
previously by X-ray diffraction. Furthermore, polysaccharides predicted to be conformationally restricted in simulations
(cellulose, chitin, and mannan) prefer the solid state in reality,
even as oligosaccharides. Those predicted to be more flexible
(xylan and hyaluronan) are known to be soluble, even as high
polymers. Therefore an intriguing correlation between chemical composition, water organization, polymer properties, and
biological function is proposed.
Key words: cellulose/hyaluronan/polysaccharide/water/
xylan
Introduction
Polysaccharides are the most abundant macromolecules in
the living world (Aspinall, 1985), and constitute a major
store of carbon, nitrogen, and photosynthetic potential
energy within the Earth's dynamic ecosystem (Haigler
et al., 2001; Karnezis et al., 2000). They are employed in
1
To whom correspondence should be addressed; e-mail:
[email protected]
diverse roles, as structural molecules in plants (cellulose)
and arthropods (chitin) or as dynamic lubricating components of the mammalian extracellular matrix (hyaluronan,
proteoglycans) and cell surface coatings (heparan sulfate
proteoglycans) on which protein recognition and cell±cell
adhesion can occur (Sugiyama and Imai, 1999; Muzzarelli
and Muzzarelli, 1998; Lee and Spicer, 2000; Iozzo, 1998;
Lyon and Gallagher, 1998). A molecular appreciation of
how these molecules perform their various functions is
potentially forthcoming from studies of three-dimensional
structure and dynamics. The current understanding of polysaccharide shape and form is largely based on models constructed from in vacuo calculations of internal energy
(molecular modeling), X-ray diffraction, and nuclear magnetic resonance (Duus et al., 2000). A compendium of this
work has recently been brought together and summarized
(Rao et al., 1998). Such information has allowed a reasonable representation of the average structure of polysaccharides, but it has not explained why some polysaccharides are
the basis of highly insoluble crystalline complexes whereas
others, only subtly different in chemical structure, form
gums, gels, and viscoelastic solutions. For example, in the
case of cellulose, it is completely obscure how one of the
most soluble biomolecules on the planet (glucose) can, after
knitting some half a dozen residues together, become the
basis of the most insoluble (Tonnessen and Ellefsen, 1971).
It is suggested here how these properties are emergent from
their interactions with water.
Although adiabatic in vacuo energy calculations have
played an important role in providing intuition about the
likely conformations of polysaccharides, included in this
approach is the smearing out of water into a smooth fluid
of constant dielectric. The assumption that molecular water
plays a minor role in the emergence of the shape and
dynamics of such polymers is therefore implicit. The main
impetus for their use was to allow predictions to be made
with modest computational resources. However, during the
development of molecular mechanics simulations, it was
recognized that the conformational structure and dynamics
of biopolymers was strongly influenced by the presence of
water (Franks, 1983; Brady, 1990). This resulted in important simulations of carbohydrates with explicit water when
computing resources were extremely limited. The first simulations included glucose (Brady, 1989), the sugar alcohols
sorbitol and mannitol (Grigera, 1988), and a-cyclodextrin
(Koehler et al., 1988).
Advances in computational power are removing the need
to make the severe approximations associated with treating
water as a continuum. Supercomputers can now perform
simulations of biomolecules with explicit solvent out to the
microsecond timescale (Duan and Kollman, 1998), to
Glycobiology vol. 13 no. 4 # Oxford University Press 2003; all rights reserved.
255
A. Almond and J.K. Sheehan
explore kinetic phenomena, such as protein folding. However, relatively extensive simulations of polysaccharide
fragments in the presence of water molecules are possible
without the need for supercomputers. The results of such
simulations are confirming that water has important effects
on the nature of carbohydrate dynamics, which are more
substantial than simply introducing an increased dielectric
or viscosity (Kirschner and Woods, 2001). These studies
give considerable insight into why relatively small changes
in polysaccharide chemical structure can result in large
changes in their physical properties, which go far beyond
predictions based on simple in vacuo modeling.
Recently, we reported the results of long simulations on a
range of oligosaccharides of hyaluronan, a polymer of alternating b(1 ! 3) and b(1 ! 4) linkages (Almond et al., 2000),
in the presence of explicit water molecules. These simulations were carefully compared against X-ray diffraction,
nuclear magnetic resonance, circular dichroism, and hydrodynamic data (Almond et al., 1988a,b). From the large
agreement between simulation and experiment it was concluded that simulations can provide new insights into the
unexplored microscopic dynamics of the hyaluronan polymer. Important microscopic information was forthcoming
from these studies, providing a new understanding of both
the physiochemical and biological properties of hyaluronan. Contrary to previous hypotheses of hyaluronan secondary structure (Atkins et al., 1980), the molecule was
predicted to have intrinsic flexibility at the b(1 ! 4) linkage.
This is exemplified in Figure 1a, showing glycosidic data
from the four b(1 ! 4) linkages in the decasaccharide structure simulated in water for 5 ns, overlaid on the same plot
(see Materials and methods for angular definitions). Specific
conformations at the linkage were occupied for periods on
the nanosecond timescale, whereas transitions occurred
Fig. 1. The chemical structure of hyaluronan, showing the repeating
disaccharide, and the hydrogen bonds that have been proposed to stabilize
the b(1 ! 4) linkage (Atkins et al., 1980). (a) The glycosidic angles (see
Materials and methods for definitions) predicted from a simulation of a
hyaluronan decasaccharide (Almond et al., 2000). The data for four
linkages has been overlaid onto the same plot. (b) A plot of the y-angle for
a single b(1 ! 4) linkage as a function of time from an aqueous simulation
of hyaluronan.
256
rapidly on the subnanosecond timescale (Figure 1b). The
predicted stability and instability of particular conformations, the transition time between them, and the resulting
effect on structure were in no way intuitive and raised
questions about the behavior of the b(1 ! 4) linkage in
other carbohydrate polymers.
In this article the consequences of some of these findings
for other b(1 ! 4) linkages found in cellulose, mannan,
chitin, and b(1 ! 4)-linked xylan are explored. It is predicted that apparently minor modifications of side groups
and hydroxyl positions result in effects that would not be
expected from nonaqueous models and are curiously mirrored by previous experimental observations. Important
new insights into the role of water in organizing polysaccharide structure are emergent from these studies, which
point the way toward creating a new dynamic language
for visualizing the effects of water on both oligosaccharide
and polysaccharide structure.
Results
The unexpected rigidity of cellulose in aqueous solution
Dynamic simulations of length 5 ns containing explicit
water molecules were performed on decasaccharides of
cellulose, mannan, chitin, and b(1 ! 4)-linked xylan (see
Materials and methods). They were examined in a similar
way to that described previously for a hyaluronan decasaccharide (Almond et al., 2000). Glycosidic torsion angles for
every linkage were extracted at 100-fs intervals during each
simulation, using definitions described in Materials and
methods. In Figure 2 the torsion angles for all nine linkages
have been overlaid onto individual plots for cellulose,
xylan, mannan, and chitin. As exemplified by hyaluronan
previously (Figure 1), the b(1 ! 4) linkage can occupy relatively large areas of its conformational space. However,
comparison of the simulation of cellulose against that for
hyaluronan produced an immediately interesting prediction; namely, that the glycosidic linkages in cellulose are
more tightly restricted than the hyaluronan equivalent
(Figure 2a) even though the cellulose linkage is paradoxically less sterically obstructed. Such an observation is
preliminary evidence that forces other than steric repulsion
are at work around the glycosidic linkages, perhaps involving water molecules.
As a simple parameter for visualizing linkage dynamics
and water interaction, intramolecular hydrogen bonds were
documented at regular intervals throughout the cellulose
simulation. Their definition was initially based on angle
and distance criteria (see Materials and methods). As in a
previous article (Almond et al., 2000), the number of
observed hydrogen-bond pairs was calculated as a function
of time (Figure 3a). This graph was used, as previously, to
ensure that adequate conformational sampling had taken
place. After a period as short as 500 ps, all of the persistent
intramolecular hydrogen-bonds had been observed (lower
part of the graph); implying that these conformations were
highly sampled during the simulation. Furthermore, few
new interactions were seen after a period of 2 ns had
elapsed, indicating that a large percentage of conformational space had been sampled during this period and
Predicting molecular shape from interactions with water
Fig. 2. The chemical structures and predicted glycosidic populations for
the b(1 ! 4) linkages of (a) cellulose, (b) xylan, (c) mannan, and (d) chitin.
The data were extracted from 5-ns molecular dynamics simulations with
explicit water molecules. See Materials and methods for angular definitions.
confirming that simulations of length 5 ns were adequate to
ensure statistical sampling of the major conformers. This
was also evident from the observation that each of the
internal linkages in the decasaccharide populated space in
an identical way. If this were not true, then inadequate
conformational sampling would be suspected. Similar
checks were made for all of the decasaccharide simulations,
with similar results, but the graphs are not shown.
The nature of these hydrogen bonds can be elucidated by
separating them into categories. In the case of cellulose,
three major intramolecular hydrogen-bonding interactions
were observed, and these are labeled A, B, and C in
Figure 3b. Their presence during the simulation, at each
linkage, is detailed in Figure 3c. Hydrogen bond A corresponds to the localization of OH3 with respect to O5 and is
predicted to be a particularly persistent interaction in the
cellulose structure as seen in Figure 3c. The other two
interactions involve the hydroxymethyl moiety; B represents the interaction from OH3 to O6, and C represents
the interaction from OH6 to O3.
Observation of the molecular length as a function of time
permits a visualization of the overall polymer dynamics.
This parameter is documented from the simulation of a
cellulose decasaccharide in Figure 3c. On this figure the
predicted length for a decasaccharide based on the axial
rise observed in the cellulose I crystal (Gardner and
Blackwell, 1974) has been marked (5.15 nm) and is close
Fig. 3. Intramolecular hydrogen bonds found in a cellulose decasaccharide
from a 5-ns simulation in the presence of explicit water molecules.
(a) Shows all of the hydrogen bonds numbered in order of observation,
irrespective of the type of interaction. (b) Shows the important interactions
(A±C) found across the b(1 ! 4) linkages of cellulose. (c) Shows the
presence of the interactions detailed in (b) at each of the linkages in the
decasaccharide and the molecular length as a function of time. Marked
over this is the molecular length which would be predicted if the
molecule took up a conformation as in the cellulose-I crystal.
to the maximum length obtainable for cellulose. The
predicted dynamic length stayed close to this value for the
majority of the cellulose simulation, although excursions to
more contracted conformations (~4.4 nm) were observed
ephemerally. These excursions correspond to unusual
conformations, and presumably water structure, at one or
more of the polymer linkages, which last for only a few
picoseconds. Thus the emergent average polymer conformation is highly extended and very close to that observed in
crystal structures.
Chitin and mannan are cellulose-like; xylan is not
Use of the analysis strategies already applied to hyaluronan
and cellulose allows comparison to be extended to a larger
variety of polysaccharides. The most frequently observed
257
A. Almond and J.K. Sheehan
conformations and extent of glycosidic motion can be compared between hyaluronan, cellulose, xylan, chitin, and
mannan using the (f, y) plots of Figures 1a and 2. The
similarity of cellulose, chitin, and mannan is noticeable,
but in xylan there is a distinct difference. Also, the hyaluronan data highlights that different polymers have specific
and different preferences around the b(1 ! 4) linkage.
Hence the structure and dynamics of these polymers is
predicted to be sequence specific.
The occurrence of intramolecular hydrogen-bonds at
each linkage during simulation are shown in Figures 3c
and 4 for cellulose, chitin, mannan, xylan, and the
b(1 ! 4) linkage of hyaluronan. It can be seen that cellulose,
chitin, and mannan share similar intramolecular hydrogenbond dynamics, coincident with their similar glycosidic
conformational preference. In all of these structures the
interaction between OH3 and O5 is present for greater
than 80% of the time. However, in the xylan structure this
interaction is almost completely absent and instead moves
away into another region. The consequence of this can be
seen in Figure 4a, where the predicted molecular length for
xylan has been plotted as a function of time. Also on this
diagram the predicted length of a decasaccharide, based on
the xylan left-handed three-fold X-ray fiber diffraction
structure (Nieduszynski and Marchessault, 1972), has been
marked (4.95 nm) as compared with that for the cellulose
structure. For mannan and chitin the predicted length variation is shown in Figures 4b and 4c, together with the values
observed in crystal fibers of mannan-I/II (Atkins et al.,
1988) and a/b-chitin (Gardner and Blackwell, 1975), again
in good agreement. Similarly for hyaluronan, crystal structures are known (Atkins et al., 1972; Winter et al., 1975;
Sheehan and Atkins, 1983) that would predict lengths of 4.9
nm and 4.2 nm as upper and lower limits, but 4.75 nm is
most frequently observed (Figure 4d). This is, intriguingly,
in excellent agreement with the simulated dynamics.
Fig. 4. Intramolecular hydrogen bonds and molecular length as a function
of time in 5-ns molecular dynamics simulations of decasaccharides with
explicit water molecules. (a) Xylan and the interaction is OH3±O5. (b)
Mannan, the interactions are as in cellulose (see Figure 3b). (c) Chitin, the
interactions are as in cellulose (see Figure 3b). (d) Hyaluronan, interactions
are described in Almond et al. (2000). Dotted lines denote length
predictions based on previous X-ray fiber diffraction studies; xylan 3-fold
symmetry, mannan I/II, chitin a/b and hyaluronan in K ‡, Na ‡, and Ca2 ‡
salts and a variety of pH values (see text).
Observing the subtle but dominating effects of water
Comparison of the dynamics of several different b(1 ! 4)
linkages suggested that the hydrogen bond between OH3
and O5 may not be a criterion for structuring but simply
consistent with it. To examine this idea, the hydrogen bond
in question was examined in more detail. The distance
between O3 and O5 and the angle between O3, OH3, and
O5 was extracted across one of the linkages and is plotted in
Figure 5a and b for both cellulose and xylan because the
notion of a hydrogen bond is related to distance and angular criteria. In cellulose the region around the distance 0.28
nm and angle 150 is highly populated and corresponds to
what was originally called an intramolecular hydrogen
bond. However, in xylan the system bifurcates. Two new
regions are populated at 0.41 nm, 110 and 0.42 nm, 20 .
They correspond to water-bridged linkages and differ simply by rotation of the OH3 hydroxyl group, a rotation not
favored in the cellulose structure. The dynamic water structure in xylan results in a shift in the glycosidic conformation
to (f, y) ˆ (50 , 50 ), as compared with cellulose conformation at (60 , 0 ) (Figure 5c). On this diagram iso-lines are
drawn that correspond to linkage geometries that produce
exact two- and threefold helices for the polymers. It can be
seen that the cellulose linkage localizes to the twofold isoline and xylan is close to the threefold iso-line. Typical
predicted conformations and local water structure for the
linkages of cellulose and xylan are shown in Figure 6. This is
in exact agreement with the known crystal structures. Thus,
differences between computer predictions are reflected in
solid experimental data.
Plots of the distance between O3 and O5 and the angle
between O3, OH3, and O5 for chitin and mannan show
very similar behavior to cellulose; compare Figure 7a and b
with Figure 5a. Again, the region around distance 0.28 nm
and angle 150 is highly populated, indicating stability of
the hydrogen bond between OH3 and O5. The emergent
preferred structures for chitin and mannan are shown in the
disaccharides of Figure 7c and d. Certainly, as exemplified
by their variation of molecular length with time (Figures 3c,
4b, and 4c) these molecules appear to be very constrained,
remarkably rigid, and almost fully elongated.
258
Predicting molecular shape from interactions with water
Fig. 6. Stereo views of the predicted favorable conformations for cellulose
and xylan. (top) Shows the direct hydrogen bond between OH3 and O5
predicted in cellulose and free water structure bridging the O6 and O3
moieties. (bottom) Shows the most highly populated water-bridged
structure observed in the xylan polymer, allowing free rotation of OH3,
and thus differing from cellulose.
Fig. 5. Predictions for the most populated conformations of (a) cellulose
and (b) xylan. On these plots the distance O3 to O5 is plotted against
the angle O3, OH3, O5. (c) Shows the effect of changes in water structuring
on the predicted linkage conformation using population contour
maps. Marked on this plot are the iso-lines where linkage geometries
result in particular crystal morphologies for the polymer.
So why is the OH3±O5 hydrogen bond highly favored in
some cases of the b(1 ! 4) linkages and not in others? In the
case of hyaluronan, the cellulose-type interaction is certainly present but not dominant. It is the interaction of the
amide proton with carboxyl moiety that dominates, and
thus the linkage is partially drawn toward another conformation, not sampled in cellulose. In this case there are two
water environments to be considered, one at either side of
the linkage. Plotting the y torsion angle against the angle
between O3, OH3, and O5 for a particular b(1 ! 4) linkage
demonstrates these two major conformations and their
relative populations (Figure 8). One is similar to that
found in cellulose and corresponds to y ˆ 0 . The other is
Fig. 7. Identifying the most populated regions for the linkages of (a)
mannan and (b) chitin. On these plots the distance O3 to O5 is plotted
against the angle O3, OH3, O5 and can be compared against cellulose and
xylan in Figure 5. The predicted conformations of (c) mannan and (d)
chitin, which are cellulose-like.
where a favorable interaction between carboxyl and acetamido occurs and involves water bridges. In this conformation the O3 group rotates into bulk water. Again
consideration of the molecular length as a function of time
259
A. Almond and J.K. Sheehan
Fig. 8. Identifying the conformational attractors in hyaluronan. Two
regions are observed with differing linkage geometry. On this plot the
distance O3 to O5 is plotted against the angle O3, OH3, O5 and can be
compared against cellulose and xylan in Figure 5 and mannan and chitin in
Figure 7. Exchange between them results in polymer chain dynamics,
possibly leading to its high solubility.
(Figure 4d) reveals how this microscopic instability of interactions carries through to polymer dynamics. As compared
with cellulose (Figure 3c) and even xylan (Figure 4a) there
are significant fluctuations in the molecular length
predicted for hyaluronan.
Discussion
Water interactions at linkages dominate the dynamics
Three distinctively different kinds of dynamic behavior for
the b(1 ! 4) linkage have been highlighted in five different
polymers. Based on these observations, relationships
between dynamic conformation and primary chemical
structure are evident. First, it is predicted that modification
of position C2 of glucose has little effect on the dynamics as
compared with modification at position C5. Second, the
data indicate there is something special relating to the substitution at this position that has not previously excited
attention. The xylan and hyaluronan data also undermine
the idea that the O3±O5 hydrogen bond (A in Figure 3b), is
in some way special or strongly stabilizing at b(1 ! 4) linkages. Therefore, an alternative explanation for the presence and absence of hydrogen bonds than those typically
offered must be sought.
That there is a very high degree of restriction predicted by
the simulations at the linkages of cellulose can be exemplified by comparison with molecular length predictions based
on steric hindrance alone. In Figure 9 the distribution of
molecular lengths for cellulose has been plotted from the
simulation as compared with distribution of molecular
lengths predicted by allowing the linkages to explore all of
the sterically accessible b(1 ! 4) space with equal probability. Thus in water there appears to be a strong force acting
to elongate the molecule. If there is, how is water implicated in the emergence of such a force? It is also remarkable that the predicted dynamic structure for the
260
Fig. 9. Binned molecular lengths for random configurations of cellulose
generated by exploring the whole sterically allowable space with equal
probability (see Materials and methods). The most frequently observed
length is distant from the theoretical maximum extension, as anticipated.
Binned molecular lengths for cellulose, xylan, and hyaluronan
extracted from aqueous molecular dynamics simulations are shown for
comparison.
oligosaccharides in water is so similar to that observed for
the high polymer fibers in X-ray diffraction studies (Gardner and Blackwell, 1974); the consequences of this will be
discussed later.
It would not be expected a priori that the polymers, such
as cellulose, would intrinsically prefer such extended conformations, so why are they observed? Detailed comparison
of the cellulose and xylan simulations appears to give a clue.
Absence of the hydroxymethyl moiety in xylan promotes a
shift in the preferred glycosidic linkage population from
conformation in which the hydrogen bond (A in Figure 3b)
is present, as found in cellulose, to conformational states
involving water bridges (described for xylan in Figure 5b).
Thus the average emergent conformation in xylan is a lefthanded threefold helix just as found in the hydrated
fiber X-ray diffraction structure (Nieduszynski and
Marchessault, 1972). This has been something of a mystery
and has previously been commented on by Rao et al. (1998)
in their book on carbohydrate conformation. We quote: ``In
terms of intra-molecular forces, it is difficult to understand
why (1 ! 4)-xylan favors a threefold screw symmetry
instead of twofold as in cellulose. It is tempting to attribute
the preference to interactions between chains and water
molecules in the lattice (since there is evidence of at least
one water molecule per xylose residue in the unit cell).''
From these simulations it is apparent that water interactions may underlie these changes in conformation and that
qualitatively at least the simulations mirror the experimental observations. However, a significant challenge remains:
If one now assumes that the hydroxymethyl group is implicated in defining the restricted cellulose conformation,
as opposed to xylan, how is this to be explained at the
microscopic level?
Predicting molecular shape from interactions with water
A dynamic description of polysaccharide±water
interaction
To offer a reductive explanation for the data, the nature of
interactions of molecular water with the linkage must be
reconsidered and compared with intramolecular hydrogen
bonds. In a polymer, such as cellulose, all of the side groups
can interact with water, and so in each molecular conformation distinct possibilities for interactions with local water
molecules will be available. Without considering loss of
degrees of freedom, there is negligible energy difference
between a directly intramolecular hydrogen-bonded conformation and a water-bridged conformation (Williams and
Westwell, 1988). Therefore, rather than using intramolecular hydrogen bonds as the causal explanation of structuring,
it is proposed that geometries are preferred for the ensemble
as a whole, that is for molecule and water. In this interpretation cellulose has a single conformation consistent with
favorable polymer±water interaction (Figure 5a). This conformation implies a rigid and the fully extended molecule.
Similarly, xylan has a single favorable region that involves
water bridges at the linkage (Figure 5b). On the other hand,
the b(1 ! 4) linkage in hyaluronan is predicted to possess
two competing favorable water structures (Figure 8)
which have differing linkage conformations. The question
of structuring in these polysaccharides then reduces to
understanding how these polymers interact with water,
using statistical mechanics, rather than naively assuming
they are stabilized by intramolecular hydrogen bonds.
Here these arguments are made for cellulose and xylan. In
cellulose the argument revolves around the hydroxymethyl
group, which is common to pyranoses.
First attempt at a rational reductive explanation for
the extended nature of cellulose
Restricting the motion of the hydroxymethyl comes with a
relatively high penalty, compared to other parts of the
molecular assembly. This is because the hydroxymethyl
can potentially occupy a large number of conformations,
as compared to, say, a hydroxyl. Similarly, water molecules
in the region of O3 and O6 have many degrees of freedom.
Thus the preferred dynamic state will be that which preserves the maximum degrees of freedom in the hydroxymethyl group together with the maximum freedom of
movement of local water molecules in this region. In the
two-fold cellulose conformation, it is the OH3 rotamer that
has lost degrees of freedom (Figure 10a). It could occupy
another rotameric state if it was not involved in the hydrogen bond to O5. However, in this dynamic scenario the
hydroxymethyl occupies all rotameric states, except the tg
conformation (Figure 10b). Now the hydrogen bonds elucidated in Figure 3 can be seen to be part of the water
structure around the hydroxymethyl group. The gt and gg
states are stabilized by water bridges from O6 to O3 (in two
separate orientations Figure 10c and e). More importantly,
water molecules can exchange rapidly with these structures,
which preserves the number of degrees of freedom available
to water molecules while maintaining intramolecular
hydrogen-bond networks.
In large simulations, even unfavorable molecular configurations are observed, such as that shown in Figure 11.
Fig. 10. (a) Visualizing the conformational attractors at the linkages of
cellulose. (b) Preferred orientations for hydroxymethyl groups within
cellulose. (c) The most frequently observed cellulose linkage configuration
with a planar structure, with an intramolecular hydrogen bond across the
linkage and water bridge between O3 and O6. (d) Planar geometry, but
with the hydroxymethyl closing the linkage region totally to water
molecules. (e) Similar to (c) but with hydroxymethyl group on the opposite
side of linkage and still involved in a water bridge.
Fig. 11. Stereo image of a linkage geometry rarely observed in the cellulose
simulation, coincident with large conformational shifts in the cellulose
chain. The large, solvent-accessible surface, deep into the linkage
region, provides an environment for trapping water molecules and
hence is disfavored.
Static visualization reveals that such a molecular configuration is perfect for coordinating water molecules within the
b(1 ! 4) linkage. In this case the hydroxymethyl group and
OH3 become heavily implicated in water coordination.
Consequently, water molecules penetrate relatively deeply
into the structure and cannot easily exchange with the bulk.
This results in large reductions in the total numbers of
degrees of freedom for both water molecules and biomolecule. Hence these molecular conformations are strongly
disfavored and thus cannot be readily observed by experiments. Therefore, although the twofold conformation permits remarkably little flexibility in the chain, it maximizes
the total degrees of freedom of the water and rotameric
substituents (by shielding potential sites of water localization), a statistically emergent or entropic contribution to
261
A. Almond and J.K. Sheehan
conformational stabilization. Thus it is the twofold shape
that provides the minimum disturbance to the preferred
statistical state of water, and the emergence of polymer
order is in effect paid for by maximization of water disorder. Surprisingly, it is suggested that the extended cellulose
shape arises from avoidance of strong hydrophilic, watertrapping molecular conformations. The same argument follows for both mannan and chitin, which share a similar
environment with cellulose around O5.
On the other hand, the xylan conformation emerges
because the rotamer OH3 is maximally disordered and
water molecules can move more freely around the glycosidic
region and exchange with bulk water. The absence of a
hydroxymethyl group in the xylose sugars ensures there
are no water entrapment mechanisms available at the linkages. In xylan the most disorganized state is that in which
water bridges surround the linkage and thereby maintain in
exchange with bulk water. The result of these water bridges
is to change the average conformation of the linkage. This
has the effect of breaking the twofold symmetry predicted
for cellulose (see Figure 5c). Such a dynamic scenario is an
impossible balancing act for the cellulose decasaccharide,
and the extended planar structure emerges. Consequently
the molecular length of a xylan decasaccharide is on average
shorter, and the molecule is more dynamic as detected by
the fluctuations in the molecular length.
Physical and biological consequences
Where does the b(1 ! 4) linkage of hyaluronan fit in all of
this? This linkage is identical in geometry to that for chitin
except for substitution of a single carboxylate group for
hydroxmethyl at position C5. It could be anticipated that
the incorporation of a bulkier group would restrict the
dynamic space around the linkage. Paradoxically, the opposite is true; the dynamic space around the linkage is
extended and multiple conformations coexist. Why is this
so? The preferred explanation flows from the previous
argument. The presence of the two oxygen atoms on the
carboxyl group together with the actetamido group now
provides a strong water coordination site on both sides of
the linkage, and, in inhibiting one, the molecule opens the
other. This puts a strong intrinsic dynamic instability into
this linkage, as though it were see-sawing between two
strong conformational attractors. One attractor is the O3
and O6 environment already described, and the other
involves the carboxylate and acetamido. Again this shows
that there is nothing intrinsic about the stabilized conformation in cellulose at the b(1 ! 4) linkage. In the simulation
of hyaluronan it is present, but so are other favored geometries. The result of this degeneracy in conformations at the
b(1 ! 4) linkage of hyaluronan is to produce, on average, a
highly extended molecule, which is at the same time intrinsically unstable, thereby producing strong fluctuations in its
molecular length (Figure 4d).
Considering the simulation data, it is intriguing to propose that there is a strong correlation between the predicted
flexibility observed in these simulations and the solubility of
these molecules. Both cellulose and chitin become insoluble
beyond tetrasaccharides. It is proposed that this is resultant
from the conformational restriction observed in simulations
262
of single chains. Combined with their extended, flattened
shape, this drives the molecules to prefer chain±chain associations. Xylan, although not strongly soluble, can be dispersed as a polymer in water and makes tough gels. Its
predicted tendency toward a threefold shape favors hexagonal packing geometries that incorporate water. On the
other hand, hyaluronan is highly soluble over a very wide
molecular weight range and never forms cuttable solids
unless some cross-linking factor is added, which is consistent with the prediction of a highly dynamic chain.
The consequences of gaining a microscopic understanding of these molecules, as attempted here, is undoubtedly
important. Cellulose and chitin are among the most ubiquitous macromolecules on the planet, and much biological
and industrial activity is centered on their production, modification, and breakdown. Similarly, hyaluronan is a major
constituent of the mammalian extracellular matrix and has
attracted attention in recent years as a nonimmunogenic
pharmaceutical. As such, these results provide crucial new
insights into the relationship between polysaccharide structure and function and clarify why such molecules have been
selected as major structural molecules during evolution in
an environment of liquid water. Furthermore, the importance of the b(1 ! 4) linkage is not restricted to polysaccharides; it is found in a range of glycoproteins and
glycolipids, molecules that have been the subject of intense
study recently because they are at the center of a wide
variety of biological phenomena. Furthermore, extension
of these studies to other linkages is now important to gain
intuition into water interactions within larger classes of
carbohydrates and hence widen the hypotheses set forth
here. Such studies will continue to shed light on the diverse
functions of glycans within living organisms.
Materials and methods
Molecular dynamics simulations
Simulations used an all-atom approach in which the atoms
were represented as van der Waals spheres with a partial
charge. Bonds were represented using the approach of
molecular mechanics and a force field suitably modified
for carbohydrates (Woods et al., 1995). In this representation, torsional terms are calculated from ab initio molecular
orbital calculations to which force field parameters are
empirically fitted. The force field can then correctly reproduce torsional potentials for a- and b-glycosdic linkages
predicted by quantum calculations, using suitable partial
atomic charges. In this case, partial charges were precalculated by least squares fitting to quantum mechanical
electrostatic potentials, calculated with the HF/6-31G*
basis set for neutral sugars and HF/6-31G** for charged
sugars. Such modifications are necessary because the default
hydrocarbon-based force field does not take into account
electronic interactions, which are due to the presence of ring
and glycosidic oxygens in the linkage region of carbohydrates; no additional torsional bias was applied beyond
that predicted by ab initio calculations. These interactions, common in carbohydrate molecules, have been termed
the exo-anomeric effect to explain the conformational
Predicting molecular shape from interactions with water
preferences at glycosidic linkages. However, although it is
strictly correct to apply such modifications to the underlying carbohydrate force field, these electronic interactions
may be weakened or rendered insignificant by the presence
of water, as discussed previously (Almond et al., 1997).
Therefore, it is noteworthy that the true origin of conformational preference in aqueous solvated carbohydrate linkages, and thus the so-called exo-anomeric effect may be
primarily due to interactions with water molecules, rather
than electronic interactions.
In the case of cellulose, chitin, xylan, and mannan the
molecular dynamics simulations were performed using the
CHARMm program (Brooks et al., 1983). For hyaluronan
the molecular dynamics simulation was performed using the
tinker program as a basis (Dudek and Ponder, 1995) but
using the same force field, as described previously (Almond
et al., 2000). In all cases explicit water was modelled using
2000 TIP3P water molecules (Jorgensen et al., 1983) equilibrated at 300 K. For cellulose, chitin, mannan, and xylan
the water molecules were packed into a hexagonal prism of
width 3.15 nm and length 7.0 nm. In the hyaluronan simulation a rectangular box of dimension 3.2 nm 3.2 nm
6.4 nm was used. Long-range electrostatics were treated
using periodic boundary conditions and a group based
cut-off switched between 0.8 and 1.2 nm. Following equilibration (for 100 ps) the simulation was continued for 5 ns
at constant temperature (300 K) and volume, by weak
coupling to a heat bath, to construct a canonical (NVT)
ensemble. Molecular dynamics integration was carried out
using the leap-frog formulation (Hockney, 1970) of the
Verlet algorithm (Verlet, 1967), and hydrogen covalent
bond lengths were kept constant using the SHAKE procedure (van Gunsteren and Berendsen, 1977). An integration
step size of 1 fs was used to provide precise trajectories. No
explicit hydrogen bonding function was used in the simulations, because it is assumed that they are well represented by
the partial atomic charges and van der Waals parameters
(Brady and Schmidt, 1993). Coordinates were written
every 0.1 ps, and the nonbonded lists were updated at
every step.
Analysis of molecular dynamics simulations
Glycosidic linkage conformation was represented by dihedral angles f (H1-C1-O4-C4) and c (C1-O4-C4-H4) defined
by the hydrogen atoms at b(1 ! 4) linkages. Intramolecular
hydrogen-bond interactions were calculated as in our previous work (Almond et al., 1997, 1988b) and were defined
by a distance of D (hydrogen donor) to A (hydrogen acceptor) is less than 3.5 , and the angle D-A±A is less than 60 .
This (arbitrary) definition of a hydrogen bond has been
used in the analysis of other carbohydrate simulations
(Brady and Schmidt, 1993). For every frame in each of the
simulations, all of the intramolecular hydrogen-bonds were
identified using these criteria. Following extraction they
were classified depending on their overall persistence during
the simulation. Only the most persistent were kept for later
analysis. The molecular length was defined between oxygen
atoms in the two extreme hydroxyls. In cellulose this was O1
at one end of the molecule and O4 at the other. The contour
plot in Figure 5c was created by placing all of the extracted
dihedral angles (every linkage) into bins on the twodimensional plane with resolution of 5 . These were then
converted to probabilities and transformed into an energylike function by taking the natural logarithm (Boltzmann
statistics). The resulting function was contoured at relevant
intervals for both cellulose and xylan simulations.
Calculation of sterically accessible conformations of cellulose
Random configurations of a cellulose decasaccharide were
generated. Each conformation was then tested for any
atomic overlap. In this process the effective carbon and
oxygen hard-sphere diameters were taken as 0.32 nm and
0.27 nm, respectively (Pauling et al., 1951). Conformations
without steric clashes were accepted, and their molecular
length was binned to produce the curve of Figure 9.
Acknowledgments
A.A. was funded by a Wellcome Trust Prize Travelling
Research Fellowship, grant reference number 058154. J.S.
was also funded by the Wellcome Trust on a Showcase
Award Scheme.
References
Almond, A., Sheehan, J.K., and Brass, A. (1997) Molecular dynamics
simulations of the two disaccharides of hyaluronan in aqueous
solution. Glycobiology, 7, 597±604.
Almond, A., Brass, A., and Sheehan, J.K. (1998a) Deducing polymeric
structure from aqueous molecular dynamics simulations of oligosaccharides: solution simulations of hyaluronan tetrasaccharides compared with hydrodynamic and X-ray fibre diffraction data from the
polymer. J. Mol. Biol., 284, 1425±1437.
Almond, A., Brass, A., and Sheehan, J.K. (1998b) Dynamic exchange
between stabilized conformations predicted for hyaluronan tetrasaccharides: comparison of molecular dynamics simulations with available NMR data. Glycobioogy, 8, 973±980.
Almond, A., Brass, A., and Sheehan, J.K. (2000) Oligosaccharides as model
systems for understanding water-biopolymer interaction: hydrated
dynamics of a hyaluornan decamer. J. Phys. Chem. B, 104, 5634±5640.
Aspinall, G.O. (1985) The polysaccharides. New York: Academic Press.
Atkins, E.D.T., Phelps, C.F., and Sheehan, J.K. (1972) The conformation
of the mucopolysaccharides. Hyaluronates. Biochem. J., 128, 1255±1263.
Atkins, E.D.T., Meader, D., and Scott, J.E. (1980) Model of hyaluronic
acid incorporating four intra-molecular hydrogen-bonds. Int. J. Biol.
Macromol., 2, 318±319.
Atkins, E.D.T., Farnell, S., Mackie, W., and Sheldrick B. (1988) Crystal
structure and packing of mannan I. Biopolymers, 27, 1097±1105.
Brady, J.W. (1989) Molecular dynamics simulations of a-D-glucose in
aqueous-solution. J. Am. Chem. Soc., 111, 5155±5165.
Brady, J.W. (1990) Molecular dynamics simulations of carbohydrate
molecules. Adv. Biophys. Chem., 1, 155±202.
Brady, J.W. and Schmidt, R.K. (1993) The role of hydrogen-bonding in
carbohydrates: molecular dynamics simulations of maltose in aqueous
solution. J. Phys. Chem., 97, 958±966.
Brooks, B.R., Bruccoleri, R.E., Olafson, B.D., States, D.J.,
Swaminathan, S., and Karplus, M. (1983) CHARMm: a program
for macromolecular energy minimization and dynamics calculations.
J. Comp. Chem., 4, 187±217.
Duan, Y. and Kollman, P.A. (1998) Pathways to a protein folding
intermediate observed in a 1-microsecond simulation in aqueous
solution. Science, 282, 740±744.
Dudek, M.J., and Ponder, J.W. (1995) Accurate modeling of the
intramolecular electrostatic energy of proteins. J. Comput. Chem.,
16, 791±816.
263
A. Almond and J.K. Sheehan
Duus, J.é., Gotfredsen, C.H., and Bock, K. (2000) Carbohydrate
structural determination by NMR: modern methods and limitations.
Chem. Rev., 100, 4589±4614.
Franks, F. (1983) Water. London: Royal Society of Chemistry.
Gardner, K.H. and Blackwell, J. (1974) The structure of native cellulose.
Biopolymers, 13, 1975±2001.
Gardner, K.H. and Blackwell, J. (1975) Refinement of the structure of
b-chitin. Biopolymers, 14, 1581±1595.
Grigera, J.R. (1988) Conformation of polyols in waterÐmolecular
dynamics simulation of mannitol and sorbitol. J. Chem. Soc., Faraday
Trans. I, 84, 2603±2608.
Haigler, C.H., Ivanova-Datcheva, M., Hogan, P.S., Salnikov, V.V.,
Hwang, S., Martin, K., and Delmer, D.P. (2001) Carbon partitioning
to cellulose synthesis. Plant Mol. Biol., 47, 29±51.
Hockney, R.W. (1970) The potential calculation and some applications.
Meth. Comp. Phys., 9, 135±211.
Iozzo, R.V. (1998) Matrix proteoglycans: from molecular design to
cellular function. Ann. Rev. Biochem., 67, 609±652.
Jorgensen, W.L., Chandrasekhar, J., Madura, J.D., Impey, R.W., and
Klein, M.L. (1983) Comparison of simple potential functions for
simulating liquid water. J. Chem. Phys., 79, 926±935.
Karnezis, T., McIntosh, M., Wardak, A.Z., Stanisich, V.A., and
Stone, B.A. (2000) The biosynthesis of beta-glycans. Trends Glycosci.
Glycotech., 12, 211±227.
Kirschner, K.N. and Woods, R.J. (2001) Solvent interactions determine
carbohydrate conformation. P. Natl. Acad. Sci. USA, 98, 10541±10545.
Koehler, J.E.H., Saenger, W., and van Gunsteren, W.F. (1988)
Conformational differences between a-cyclodextrin in aqueoussolution and in crystalline formÐa molecular dynamics study.
J. Mol. Biol., 203, 241±250.
Lee, J.Y. and Spicer A.P. (2000) Hyaluronan: a multifunctional,
megaDalton, stealth molecule. Curr. Opin. Cell. Biol., 12, 581±586.
Lyon, M. and Gallagher J.T. (1998) Bio-specific sequences and domains in
heparin sulphate and the regulation of cell growth and adhesion.
Matrix Biol., 17, 485±493.
264
Muzzarelli, R.A.A. and Muzzarelli, C. (1998) Native and modified chitins
in the biosphere. In Nitrogen-containing macromolecules in the bio- and
geosphere. ACS Symposium Series, 707, 148±162.
Nieduszynski, I.A. and Marchessault, R.H. (1972) Structure of b(1 ! 4)xylan hydrate. Biopolymers, 111, 1335±1344.
Pauling, L., Corey, R.B., and Branson H.R. (1951) Structure of proteins:
two hydrogen-bonded helical configurations of the polypeptide chain.
Proc. Natl Acad. Sci., 37, 205±211.
Rao, V.S.R., Qasba, P.K., Balaji, P.V., and Chandrasekhan, R. (1998)
Conformation of carbohydrates. Amsterdam: Harwood Academic
Publishers.
Sheehan, J.K. and Atkins, E.D.T. (1983) X-ray fibre diffraction study of
the conformational changes in hyaluronate induced in the presence of
sodium, potassium and calcium cations. Int J. Biol. Macromol., 5,
215±221.
Sugiyama, J. and Imai, T. (1999) Aspects of native cellulose microfibrils at
molecular resolution. Trends Glycosci. Glycotech., 11, 23±31.
Tonnessen, B.A. and Ellefsen, é. (1971) Investigations of the structure
of cellulose and its derivatives. F. Submicroscopical investigations.
In Bikales, N.M. and Segal, L. (eds.), Cellulose and cellulose
derivatives, Part 4. New York: Wiley Interscience, pp. 265±304.
van Gunsteren, W.F. and Berendsen, H.J.C. (1977) Algorithms for
macromolecular dynamics and constraint dynamics. Mol. Phys., 34,
1311±1327.
Verlet, L. (1967) Computer experiments on classical fluids. I. Thermodynamical properties of Lennard-Jones molecules. Phys. Rev., 159,
98±103.
Williams, D.H. and Westwell, M.S. (1998) Aspects of weak interactions.
Chem. Soc. Rev., 27, 57±63.
Winter, W.T., Smith, P.J.C., and Arnott, S. (1975) Hyaluronic acid:
structure of a fully extended 3-fold helical sodium salt and comparison with
the less extended 4-fold helical forms. J. Mol. Biol., 99, 219±235.
Woods, R.J., Dwek, R.A., Edge, C.J., and Fraser-Reid, B. (1995)
Molecular mechanical and molecular dynamical simulations of
glycoproteins and oligosaccharides. 1. GLYCAM-93 parameter
development. J. Phys. Chem., 99, 3832±3846.