You are on page 1of 11

Article

Cite This: Macromolecules XXXX, XXX, XXX−XXX pubs.acs.org/Macromolecules

Quantifying Localized Macromolecular Dynamics within Hydrated


Cellulose Fibril Aggregates
Pan Chen,†,§ Camilla Terenzi,∥ Istvań Furo,́ †,‡ Lars A. Berglund,† and Jakob Wohlert*,†

Wallenberg Wood Science Center and ‡Department of Chemistry, KTH Royal Institute of Technology, SE-10044 Stockholm,
Sweden
§
Beijing Engineering Research Center of Cellulose and its Derivatives, School of Materials Science and Engineering, Beijing Institute
of Technology, 5 South Zhongguancun Street, Haidian District, Beijing 100081, China

Laboratory of Biophysics, Wageningen University and Research, Stippeneng 4, 6708 WE Wageningen, The Netherlands
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

*
S Supporting Information

ABSTRACT: Molecular dynamics (MD) simulations of 13C


Downloaded via LINKOPING UNIV on October 1, 2019 at 08:23:47 (UTC).

NMR longitudinal relaxation (T1) distributions were recently


established as a powerful tool for characterizing moisture
adsorption in natural amorphous polymers. Here, such
computational−experimental synergy is demonstrated in a
system with intrinsically high structural heterogeneity, namely
crystalline cellulose nanofibrils (CNFs) in highly hydrated
aggregated state. In such a system, structure−function
properties on the nanoscale remain largely uncovered by
experimental means alone. In this work, broadly polydispersed
experimental 13C NMR T1 distributions could be successfully reproduced in simulations and, for the first time, were
decomposed into contributions from distinct molecular sources within the aggregated CNFs, namely, (i) the core and (ii) the
less-accessible and accessible surface regions of the CNFs. Furthermore, within the surface groups structurally different sites
such as (iii) residues with different hydroxymethyl orientations and (iv) center and origin chains could be discerned based on
their distinct molecular dynamics. The MD simulations unravel a direct correlation between dynamical and structural
heterogeneity at an atomistic-level resolution that cannot be accessed by NMR experiments. The proposed approach holds the
potential to enable quantitative interpretation of NMR data from a range of multicomponent high-performance nanocomposites
with significantly heterogeneous macromolecular structure.

■ INTRODUCTION
High-strength fibers constitute one of the most important
The structural heterogeneity of native cellulose is carried over
to synthetic cellulose-based biomaterials, both where the
contributions of polymer science to the development of hierarchical structure of the woody tissue is conserved, such as
lightweight macromolecular structures for industrial use. in transparent wood,5,6 or where the CNFs have been
Poly(p-phenylene terephthalamide) (Kevlar) is the most disintegrated from the plant cell wall.
well-known example,1 and poly(p-phenylene-2,6-benzobisox- The strong nanostructural heterogeneity that exists in both
azole) (Zylon) possibly has the best mechanical properties man-made and natural fibers is undoubtedly a key factor
among commercially available fibers.2 Interestingly, cellulose, responsible for the complex structure−function relationships of
the linear β-(1,4) linked homopolymer of glucose, has similar the final polymeric product. Hence, being able to quantitatively
intrinsic mechanical properties as Kevlar but is biologically describe these materials with nanoscale spatial and chemical
synthesized and can be disintegrated into its nanoscale resolution would be of great practical importance for a better
components for subsequent engineering into man-made control of the design of new functional materials with finely
materials.3 Both natural and synthetic functional polymers tuned responses. The physicochemical structure of fibers has
typically exhibit strongly heterogeneous nanostructures due, been studied by using advanced structural characterization
for instance, to the coexistence of ordered and disordered techniques, such as synchrotron X-ray and neutron scattering.
regions that react differently to the ingress of moisture or to These methods provide valuable structural information about
chemical treatments. Cellulose is no exception to this. The the crystalline regions but are poorly informative about
elementary units of its native structure are micrometer long disordered domains that yield excessively broad diffraction
cellulose nanofibrils (CNFs) that are packed into larger fibril peaks.7 Particularly in the case of highly heterogeneous
aggregates (FA). The CNFs are structurally heterogeneous
such that they have a highly crystalline core surrounded by Received: March 8, 2019
noncrystalline surface polymers as well as less ordered domains Revised: September 9, 2019
occurring intermittently along the length of the nanofibril.4

© XXXX American Chemical Society A DOI: 10.1021/acs.macromol.9b00472


Macromolecules XXXX, XXX, XXX−XXX
Macromolecules Article

Figure 1. Hierarchical structure of the cellulose fibril aggregate (FA) model. (A) The four nanofibrils are arranged in antiparallel configuration, with
the direction defined by the direction of the glucan chains within each fibril. (B) The model assumes a square cross section made up of 36 glucan
chains. It distinguishes between chains located in surfaces and in the nanofibril core. Surfaces are further classified as either accessible or less
accessible, with respect to water, and chains located at fibril corners. (C) The cellulose Iβ unit cell contains two nonequivalent chains, named center
and origin. This distinction is inferred also on the noncrystalline surfaces, which further can be classified as (110) or (1−10) depending on which
crystal plane they reside upon. (D) Molecular graphics showing the hydrated FA model used in the MD simulations. Surfaces are red, the core is
black, and water molecules are shown in gray. Accessible and less-accessible surfaces are fully and partially hydrated, respectively. (E) Atomic
structure of the glucan chain, going from the nonreducing (left) to reducing end (right), with the atom naming used herein. Note that the 2-fold
screw symmetry makes the exocyclic hydroxymethyl group (C6−OH6) point in alternating directions.

materials like cellulose, 13C solid-state NMR spectroscopy has slow dynamics were also observed by Fernandes et al.13 in
contributed significantly to supramolecular structure elucida- isolated cellulose: ring carbons were reported to be the most
tion, thanks to its specific capability to partially resolve ordered rigid ones, and C6 signals yielded several groups of 13C T1
and disordered regions.8−12 In addition, spectrally resolved 13C values that were attributed to varying polymer mobility over
longitudinal relaxation time (T1) measurements have enabled different hydroxymethyl conformers (see the Results and
qualitative assessment of the local reorientational dynamics of Discussion section). Yet, linking the complex macromolecular
the 13C−1H bonds.8,13,14 dynamics with local structural heterogeneity by experimental
Recently, Terenzi et al.15 have experimentally studied by 13C means only remains somewhat conjectural and is practically
CP-MAS T1 measurements the effect of varying the moisture still an open challenge. To address this challenge, Chen et al.18
content on the polymer dynamics in cellulose and xyloglucan. recently demonstrated that MD simulations of 13C T1
Other earlier similar 13C CP-MAS NMR works8,9,16,17 typically distributions of the amorphous polysaccharide xyloglucan
reported only one single average T1 value per carbon site and/ enable reproducing experimental NMR results and interpreting
or focused only on C4 atoms by means of measurements that them at an atomistic level that cannot be accessed by NMR
were often undersampled due to excessively long experiment measurements alone.15
duration typical of 13C CP-MAS. A remarkable exception is the Inspired by that significant advance, the same methodology
work of Fernandes et al.,13 who observed a bimodal relaxation is here further developed with the aim to study the more
behavior in SP-MAS measurements that, however, are not
complex hydrated structure of CNFs. Specifically, we set out to
designed for probing long T1 ranges and could not be further
demonstrate the capability of MD simulations in disentangling,
decomposed into multiple dynamic modes due to low signal-
with unprecedented nanoscale spatial and chemical resolution,
to-noise ratio. As a significant step forward, Terenzi et al.15
introduced site-specific 13C T1 distributions, obtained by localized macromolecular motions within core or surface
inverse Laplace transformation (ILT) of raw data acquired regions of the heterogeneous CNF structures at high degree of
with sufficient signal-to-noise ratio and over long enough spin hydration. To this end, 13C T1 distributions are calculated
evolution times (see Figure S1) under CP-MAS conditions directly from MD simulation trajectories and compared to the
that enable sampling both short and long T1 ranges. In full experimental ILT results. Significant differences in the
analogy to widely explored water 1H T1 distributions, under macromolecular dynamics could be observed among structur-
suitable experimental conditions the polymer 13 C T 1 ally distinct regions of CNFs, such as locations of individual
distributions enable conjointly probing both local macro- polymer chains, surfaces, and interfaces. In a longer term
molecular dynamics and its heterogeneity,15 which cannot be perspective, the present methodology will enable a more
described adequately in the framework of just a single-8 or two- quantitative understanding of structure−function relationships,
component13 dynamic model per carbon site. Indeed, at each in both natural and synthetic high-strength fibers with well-
hydration level, the experimental 13C T1 distributions of CNFs controlled hydration conditions. Ultimately, this robust
covered a very broad range, namely between 10−2 and 103 s, approach will contribute to support future efforts toward the
suggesting the existence of distributions of both fast and slow development of improved high-performance polymeric materi-
molecular motions on the nanosecond time scale. Fast and als.
B DOI: 10.1021/acs.macromol.9b00472
Macromolecules XXXX, XXX, XXX−XXX
Macromolecules Article

Table 1. Average 13C T1 Values (in seconds) Obtained from Simulated and Experimental T1 Distributions as Well as from
Fitting of Experimental Raw Data before ILT Processing (See Also Figure S1)a
C1b C235c C4 (core)d C4 (surface)e C6 (core)f C6 (surface)g
total
experimental (ILT) 50 34 105 25 34 1.6
experimental (fit) 49 33 106 20 28 1.8
MD 77 69 161 49 20 2.1
core (simulated values)
inner 167 148 193 27
intermediate 136 121 152 18
surface (simulated values)
accessible 52 50 58 1.8
less accessible 61 53 64 4.0
corner 22 20 21 0.8
accessible surfaces (simulated values)
center 36 39 45 1.4
origin 73 64 76 2.2
110 57 54 60 2.3
1−10 46 47 57 1.4
C6 exterior 39 40 43 0.2
C6 interior 68 63 79 14
a
The average spread of simulated T1 values is 2%, and in all cases it is <6%, while the error in the experimental values from ILT, due to
measurement reproducibility, is within 10%; the corresponding fitting error on the raw data is within 7−15%. Simulated values are further
decomposed into average contributions from various distinct sites within the model (see Figure 1E and text). b105.5 and 102.6 ppm. c75.6 and 72.9
ppm. d89.5 ppm. e84.3 ppm. f65.6 ppm. g62.5 ppm (see 13C CP-MAS spectrum in Terenzi et al.15).

■ METHODS
The presence of cellulose FA in the secondary cell wall and in
manner (Figure 1A). In reality, the situation is complicated by the
possibility of having a mixture of both antiparallel and parallel fibrils
as well as other factors that may influence the properties of the fibril−
chemically processed wood fibers (pulp) is well established.19 A fibril interface. Such factors include twisting of individual nanofibrils,
model FA was therefore built based on the X-ray and neutron
which would prevent aggregation of crystallographic planes13 and,
diffraction results for highly crystalline cellulose Iβ samples.20 In the
furthermore, based on the observation that native aggregate sizes
present model, the elementary unit within the structure, the CNF,
depend on both lignin and hemicellulose concentration,31 the
contained 36 chains, each 40 anhydroglucose units long, arranged in a
possibility of having cell wall matrix material also inside the FA.
square cross section with the crystallographic (110) and (11̅0)
The FA was initially positioned in a periodic simulation box with
surfaces exposed. Recent advances in both experimental character-
dimensions 10 × 10 × 22 nm3, with a distance between two adjacent
ization of fibril dimensions and understanding of plant cellulose
synthesis show that the wood nanofibril may contain as few as 24, or fibrils of 0.3 nm. A total of 14585 water molecules, corresponding to
even 18, cellulose chains. The chains can further be arranged in the moisture content15 measured experimentally at a relative humidity
various cross sections, commonly assumed to be either hexagonal or (RH) of 92%, were subsequently inserted in the system; out of these,
rectangular, that differ in which crystallographic planes are exposed to around 1500 water molecules were added in between the fibrils, at the
the surroundings.13,21,22 Although this can be assumed to affect the less-accessible surfaces. This amount corresponds to experimental
physicochemical properties of the fibrils, regardless of which structural indications.32
model is assumed, they all share the same basic structural hierarchy: MD simulations were run for 300 ns using GROMACS version
partially disordered surface-accessible chains surrounding a crystalline 2016.3.33 Potential parameters for cellulose were taken from the
core, albeit in slightly different ratios. They also share the other GLYCAM06 force field,34 and water was modeled by using TIP3P.35
structural features considered here (Figure 1), which means that the Distributions of 13C NMR relaxation times and their means were
36-chain model used in this work in this respect is qualitatively similar calculated from the Fourier transform of the rotational autocorrelation
to any of the smaller fibril models. function of the corresponding C−H vector,36 as described in more
The model fibril was subsequently used to build a model FA. Here, detail in the Supporting Information. Relaxation contributions from
one faces the choice of aligning the CNFs in either parallel or dipole−dipole interactions other than that with directly bonded
antiparallel configuration. There are a number of experimental hydrogens or from any spin couplings other than dipole−dipole
observations in favor of an antiparallel arrangement. Swelling of interactions are neglected. Enhanced relaxation effects for carbons
wood pulp fibers in concentrated NaOH aqueous solution followed having more than one directly bound hydrogens (nH > 1) are
by regeneration in water (mercerization) inevitably produces cellulose represented by simply multiplying by nH the relaxation rate yielded by
II. In this crystal allomorph, the individual glucan chains are known to the behavior of a single C−H bond. This is a conventional
be antiparallel.23,24 These facts can hardly be reconciled without approximation,37,38 well suited for experiments with strong dipolar
having antiparallel entities in close proximity in the fibril aggregate. In decoupling18 and carbon sites, such as the exocyclic carbon C6, where
fact, partial conversion was also observed in intact wood cell walls,25 there is no symmetry between the two involved hydrogens.
which agrees with the direct observation of cellulose synthases moving All production runs were performed in the NVT ensemble, with
in opposite direction in native Arabidopsis.26 Furthermore, antiparallel the temperature maintained at 300 K using the velocity-rescaling
arrangement of fibrils was also observed in Valonia by using method.39 The cutoff for Lennard-Jones interactions and the real-
transmission electron microscopy.27−29 The notion of antiparallel space part of the electrostatic interactions was 1 nm. Long-range
fibrils was also recently used to explain the conversion of wood electrostatics was included using PME.40 The equations of motion
cellulose into a pseudo-orthorhombic structure during hydrothermal were integrated using a leapfrog algorithm with a basic time step of 2
treatment.30 Based on these observations, the model FA used here fs. Constraints were applied to all bonds involving hydrogen atoms by
consisted of four cellulose nanofibrils assembled in an antiparallel using P-LINCS.41

C DOI: 10.1021/acs.macromol.9b00472
Macromolecules XXXX, XXX, XXX−XXX
Macromolecules Article

Figure 2. Comparison of experimental (dashed line) and simulated (solid line) distributions of 13C T1 relaxation times for selected carbon atoms.
The experimental curves were adapted from Terenzi et al.15 Within each graph, simulated and experimental distributions are normalized to the
same value.

■ RESULTS AND DISCUSSION


The 13C NMR spectrum of CNFs is partially resolved with
NMR data on known crystal structures of cello-oligomers,46 it
was shown that tg has higher chemical shift than both gg and gt,
which was assumed to be the cause for splitting of the C6
respect to signals from the chemically nonequivalent carbons
signal also in the 13C NMR spectrum of cellulose. This
C1, C4, and C6 (Figure 1E), while signals from C2, C3, and
assignment was recently corroborated by density functional
C5 atoms exhibit significant overlap.42 The 13C chemical shift theory calculations, where it was also noted that the
values for cotton cellulose have been reported and assigned in hydroxymethyl conformation affects the chemical shift of C4
the literature as follows:10,43 C1 (106.3, 105.2 ppm), C4 (89.8, in a similar manner as for C6, which can explain the splitting of
85.1 ppm), C2, C3, C5 (72.4, 73.5, 75.3 ppm), and C6 (66.2, that peak as well.47 Thus, since tg dominates in the crystalline
63.8 ppm). These chemical shift values are consistent with the regions of native cellulose,20 the gg/gt signals for both C6 and
experimental results of Terenzi et al.,15 here summarized in C4 become significative of fibril surfaces. However, the
Table 1. We note that both the signals of C4 and C6 partially assignment of individual chemical shifts to location within
split into two peaks: these are assigned to glucan chains that the fibril assembly is complicated by the fact that there is no
belong either to nanofibril surfaces or to the crystalline unambiguous one-to-one mapping between hydroxymethyl
nanofibril interior.16,44 conformation and structural entity (e.g., surface or core), as
The reason for the splitting of the C4 and C6 peaks can be they all possess a distribution of both chemical shifts48 and
traced to the conformations of the hydroxymethyl groups. The conformation.49
hydroxymethyl group in cellulose, as in all hexopyranoses, By performing spectrally resolved 13C T1 measurements, it is
adopts three more or less stable conformations, named tg, gg, possible to obtain information about local C−H dynamics at
and gt. These letters stand for trans and gauche, respectively, each carbon site. This has, for example, unravelled that the
where the first letter refers to the rotation of the C5−C6 bond exocyclic carbon C6 is more mobile; i.e., it has shorter 13C T1
with respect to O6 and O5 and the second one refers to the than the ring carbons C1 to C5.15 This can be explained by the
C5−C6 bond rotation with respect to O6 and C4.45 From 13C larger conformational degree of freedom of C6 as compared to
D DOI: 10.1021/acs.macromol.9b00472
Macromolecules XXXX, XXX, XXX−XXX
Macromolecules Article

the ring carbons and the resulting faster rotational dynamics of focuses solely on the different behavior of C1, C4, and C6,
the hydroxymethyl group, whereas relaxation times for C1-to- while ring carbons C2, C3, and C5 were all found to follow a
C5 sites to a larger extent reflects the more restricted glucan similar behavior as C1. Figure 2 shows the simulated and
backbone dynamics. Furthermore, the shorter T1 for C4 and experimental T1 distributions for C4 and C6, separately for
C6 surface chains, when compared to those for the respective core and surface chains. The first important evidence is that
crystalline peaks, can be understood in terms of the less both experimental and simulated distributions are very broad,
constrained mobility of surface chains14 as compared to that even after core and surface contributions were disentangled,
for the chains in the fibril interior. and cover, in each plot, approximately the same number of
Such link between dynamics and structural properties has decades along the T1-axis. Specifically, the simulated C4
certainly been instrumental for improving our understanding of distributions cover the ranges 10−2000 s and 1−1000 s, for
the nature of cellulose supramolecular organization. However, core and surface respectively. The simulated C6 distributions
the resolution of 13C NMR spectroscopy is limited by chemical are respectively about 2 and 1 decade broader toward shorter
identity of carbon sites and by the need for multicomponent T1 values, covering about 0.2−1000 s (core) and 0.4−500 s
spectral deconvolution to resolve partially overlapping (surface). The T1 ranges covered by the experimental
contributions from surfaces and core regions.8,16 MD distributions are basically the same as those seen in the
simulations do not suffer from such limitations and can respective simulations, with excellent agreement at short T1
therefore be used as a powerful tool to probe dynamics in the values, while a small narrowing effect, by a factor of up to 2−
range from picoseconds up to hundreds of nanoseconds, with a 2.5 for surface chains, is observed at long T1 values. We also
resolution down to individual atoms. A crucial element of the note that the polymers underlying the surface chains, termed
work here is to compare detailed experimental 13C relaxation “intermediate core” in Table 1 and identified as having
time distributions, to corresponding distributions obtained conformations distinguishable from that of the chains in the
from simulations. Specifically, 13C T1 and their distributions innermost core,48 exhibit dynamics that is similarly distinguish-
were simulated for the different chemically nonequivalent able from the dynamics in the innermost core.
carbons (Figure 1E) in a hydrated cellulose FA, at conditions This substantially good agreement between NMR and MD
closely corresponding to those in the experimental study of distributions, in terms of their overall T1 range, enables safely
Terenzi et al.15 The good quantitative match between neglecting possible artifactual excessive broadening due to ILT
experimental and simulated mean relaxation times and processing52,53 of natural abundance 13C NMR data if, for
distributions lends credibility to the simulation results, in example, those are acquired with too low signal-to-noise ratio
terms of both model and interaction parameters. Furthermore, and/or insufficient data sampling. This proved not to be the
the excellent consistency between the mean experimental T1 case of our similar measurements of amorphous xyloglucan,15
values obtained either from the T1 distributions, after ILT where a comparably good agreement with MD simulations was
processing, or from the original raw data, by biexponential previously observed.18 For completeness, we also note that in
fitting (see Figure S1 and Table 1), confirms the reliability of the experimental data and for both C4 and C6 sites there is
the experimental data here used as a reference for the MD always some degree of spectral overlap between peaks
simulations. Based on this method validation, our combined belonging to crystalline (interior) and surface chains, which
NMR-MD approach is here exploited to interpret the results does not exist in the MD simulations. Yet, by varying the
based on the structural details revealed in the simulations. integration range of the NMR spectral peaks (data not shown
Mean 13C T1 Values: Site-Specific Macromolecular here), no appreciable variation in the resulting T1 distributions
Dynamics. Outside the extreme narrowing regime, such as in was observed.
the present case, short T1 values correspond to short motional We can thus conclude that an intrinsically large variation in
correlation times and thus to high C−H mobility.50 Here, the terms of polymer dynamics exists in CNFs, even over moieties
simulations reproduce the important trends in the mean that are identical, both chemically and in their basic spatial
experimental 13C T1 relaxation times. Indeed, we find (see classification. We also note that the dynamical heterogeneity of
Table 1) that C6 has shorter T1 values than the ring carbons, C6 core and surface sites, as seen by both NMR and MD
and surface chains have shorter T1 than the nanofibril interior. simulations, is much more pronounced than in the respective
We also find that the C−H bond mobility for C2, C3, and C5 C4 domains.
is higher than for C1 (and C4, average not shown in Table 1). Yet, some important differences between MD and NMR in
This can be rationalized by considering that the effect on C−H the shape of the respective distributions are noticeable in
bond orientations from ring puckering motions is larger for C2, Figure 2, and to explain those discrepancies the main factor to
C3, and C5. This particular contribution to the relaxation is be taken into account is the role of cellulose processing and
relatively more important in the core than in surfaces, where sample preparation. Indeed, a biexponential fitting analysis (see
the dynamics around glycosidic bonds is less restricted. text and Figure S1) of experimental raw data, before the ill-
Simulations resulted in average 13C T1 values that were posed ILT procedure was applied, confirmed the non-
systematically larger, by up to a factor of 2, than the monodispersed character observed in the experimental
experimental ones (Table 1). Although this discrepancy may distributions.
seem large, it is in reality a good agreement, considering how In the case of C4, the experimental distributions present, at
sensitive the dynamics is to the exact parametrization of, for short T1 values, a higher signal density population than the
example, the dihedral potentials involved. Those are usually simulated ones and a correspondingly lower intensity at long
optimized to reproduce equilibrium distributions,51 with little T1 values. It is likely that the real sample contains a fraction of
attention paid to the kinetics. This may not be the only reason, highly mobile glucan chains54 that could, for instance, be
however, and this point will be further discussed below. connected to naturally occurring disordered regions in wood
13
C T1 Distributions: Intrinsic Dynamical Heteroge- CNFs or to defects induced during processing, e.g., during
neity. In the following, the discussion of T1 distributions disintegration from the wood fiber cell wall, that are not
E DOI: 10.1021/acs.macromol.9b00472
Macromolecules XXXX, XXX, XXX−XXX
Macromolecules Article

present in the model. Furthermore, in the real CNF sample Nanoscale Structural Heterogeneity in CNFs. To gain
measured by NMR a fraction of residual hemicellulose was insight into the structural effects associated with the
present54 in the amount of 12.3 wt %. Both defects and polydispersed dynamical modes within CNFs, we resort to
residual hemicellulose are expected to behave like a hydrated simulations and to their ability to decompose the 13C T1
amorphous material. As previously seen in hydrated distributions, calculated on a per-atom basis, into basically any
xyloglucan, which has mean T1 values at 92% RH that are imaginable deconvolution criteria. One illustrative example of
even lower than for cellulose surface chains (e.g., 8 s in the case this is shown in Figure 2 where in full analogy to C4 and C6
of xyloglucan C4 compared to 25 s in cellulose),18 such a data the T1 distribution for C1 is decomposed into its
mobile polymer phase would exhibit a relatively higher signal respective contributions from fibril interior and surface chains.
density population at short T1 values. As fast motional modes We note that such a distinction could not be achieved for
become more abundant, slow modes at long T1 values are NMR experimental data due to the lack of sufficient chemical
expected to become correspondingly less populated, as shift contrast between the overlapping core and surface
compared to a system devoid of defects and residual populations. As expected, the width, shape, and T1 location
hemicellulose such as the model used here for the simulations. of the core and surface distributions for C1 resemble both the
This is indeed what is observed in Figure 2. simulated and experimental results for C4 (Figure 2).
In addition, we note that the 13C NMR chemical shifts for Nanofibril Core: The Crystalline Region. As demonstrated
the implicated disordered and/or defect-rich regions in C4 are in Figure 2, the 13C T1 distributions remain broad even when
expected to be closer to those of surface chains than to those the core regions are investigated separately. This is surprising
for the crystalline core chains. Hence, any contribution from since the core region has a high degree of order, as also
disordered regions or hemicellulose in the real sample should indicated by its far slower average mobility (see Table 1). The
add to the experimental signal assigned to surface chains. source of that broadening can be investigated by, first, further
Indeed, the existence of a larger difference between the shapes dividing the core region into (i) an intermediate layer, which
of the NMR and MD distributions for the surface chains than contains the chains immediately below the surface chains, and
between those for the interior chains supports this hypothesis. (ii) the inner core. The former, intermediate layer, which could
Based on this interpretation, the presence of disordered/ be expected to have properties in between that of a highly
defective regions and residual hemicellulose, alongside the crystalline and a disordered surface layer, has been proposed to
choice of potentials in MD calculations, can also explain well be the source of the paracrystalline signal that was used in the
the larger difference between experimental and simulated mean spectral decomposition of the 13C spectrum.8,16 Although the
T1 values (see Table 1) for the surface chains (on average, by distribution for the intermediate layer is shifted to shorter T1
more than a factor of 2) as compared to those for the interior values as compared to that for the inner core (see Figure S2),
chains (∼40% average difference). this shift is not significant compared to the width of the
The simulated T1 distributions of C6 are bimodal, with two individual distributions, whose shape is also quite similar.
well-separated peaks positioned around 0.4−1 and 40 s toward Hence, the question remains: what is the molecular origin of
the edges of the T1 axis. This is most clearly visible for the case the wide T1 distribution for highly ordered regions of CNFs,
of surface chains, where the short-T1 peak is more intense. The such as their crystalline core? A preliminary indication is
experimental distributions do not display such neat bimodality, provided by a spatial mapping of high/low T1 values as a
although they cover the same decades along the T1-axis. function of glucose positions within the FA model (Figure 3).
In the case of C6, we cannot interpret differences between As expected, a higher concentration of high-mobility sites in
experiments and simulations in terms of defects because those the intermediate layer (the outermost layer in Figure 3) is
affect mainly the backbone carbon sites; instead, the effect of visible, which indicates that the inner core is the most
residual hemicellulose is expected to be similar to that constrained part of the crystalline region. But it also shows that
observed for C4. Other sources of artifacts, e.g., due to ILT the more mobile regions are not homogeneously distributed
processing, can be safely discarded here, as it was anticipated along the length of the fibrils, but instead exhibit a degree of
above. We then note that in our previous work on hydrated clustering. Two possible structural causes for this heterogeneity
xyloglucan18 we argued that the inherently longer time scales are stresses either caused by (i) nanofibril−nanofibril contacts
for experiments, as compared to simulations, could lead to a and/or (ii) varying degree of hydration over the less-accessible
more efficient motional averaging of dynamical differences in surfaces. As such stresses may influence the response of
the random and loose molecular network of xyloglucan, which cellulose to, for example, chemical modification, this feature
was detected as a symmetric narrowing of the experimental deserves further studies in the future.
relaxation distributions. However, in cellulose the chain As indicated in Table 1, the exocyclic C6 is more mobile
positions are fixed, and thereby such an effect is expected to than other carbons within the core. Rotational mobility around
be negligible. On the other hand, the high-amplitude molecular the C5−C6 bond involves transitions between the three
motions of C6, associated with conformational transitions as conformational states discussed above. In the core (as opposed
described in detail in the Results and Discussion section, are to the surfaces), most (>96%) hydroxymethyl groups are in the
expected to be strongly sensitive to the accuracy of the tg state, as expected for crystalline cellulose Iβ.20 Yet, our
representation of dihedral potentials in the simulations.51 This simulations also indicate rare conformational transitions into
effect could explain the pronounced discrepancy observed for the minor gg and gt states with a mean transition time of about
core C6 sites in the long-T1 range. 4 ns. Although due reservations must be retained considering
Having demonstrated the intrinsic dynamical heterogeneity the large sensitivity of those transition times to the potentials
in cellulose, in the next section we shall attempt to identify the involved, the existence of such conformational transitions can
underlying structural causes for it by performing selective certainly cause appreciable broadening of the C6 T 1
decomposition of the MD results. distributions.
F DOI: 10.1021/acs.macromol.9b00472
Macromolecules XXXX, XXX, XXX−XXX
Macromolecules Article

Figure 4. Decomposition of simulated 13C T1 distributions of surface


chains into contributions from accessible surfaces, less-accessible
surfaces, and corner chains.

consider that local mobility in an amorphous polysaccharide,


such as xyloglucan,18 has been shown to be very strongly
dependent on the global level of hydration, a parameter that
locally differs a lot between accessible and less-accessible
surfaces. The local order imposed by the nanofibril surfaces,
which is lacking in hydrated xyloglucan, presumably dampens
the sensitivity of chain dynamics to the presence of water.
Wickholm et al.16 reported surface-specific estimates of C4
T1 values based on a small (0.2−0.8 ppm) difference between
the C4 chemical shifts for accessible and less-accessible
surfaces in combination with spectral deconvolution. They
indicated mean T1 values for less-accessible surfaces in cotton
linters that were roughly twice as long (39 s) as for the
accessible surfaces therein (11−18 s). This result is not
reproduced in the present model, which gives approximately
Figure 3. Simulation snapshots depicting C1 atoms whose T1 values identical mean values for those two surfaces. We have two
are (i) <50 s and are therefore in the most mobile segments (red possible explanations to offer. First, all nanofibril−nanofibril
spheres), (ii) >400 s are therefore in the most rigid segments (blue contacts are antiparallel in the model used here, separated by at
spheres), and (iii) small black dots indicating C1 atoms in segments most a single layer of water molecules, whereas in the
with roughly average mobility. The FA is viewed along the fibril experimental system there is certainly a possibility for parallel
direction (A) and from the side (B). arrangement as well, which would permit cocrystallization.
Second, when two fibrils come into contact, the process
Accessible and Less-Accessible Surfaces. Native and involved for attaining the lowest energy conformation may be
processed cellulose typically exists as FA. As shown in Figure slow compared to the MD time scales. Thus, there is an
1B, aggregation gives rise to a possible distinction between inevitable risk that the FA model may not have reached
nanofibril surfaces located at the exterior of a FA and those equilibrium in this regard. In both cases, fibril−fibril
within. In wet systems, the former ones are readily accessible to interactions at the less-accessible surfaces could be under-
water, whereas the latter ones, although still hydrated to some estimated in the model, yielding a too-high mobility close to
extent, are in very slow exchange with bulk water and exhibit that for accessible surfaces. To investigate this, the number of
more restricted and slower molecular dynamics.16,32,55 In the hydrogen bonds formed between fibrils was calculated based
following, these distinct regions are termed accessible or less- on standard geometrical criteria: a donor−acceptor distance
accessible surfaces, respectively. <0.3 nm and a hydrogen donor−acceptor angle <35°. In
Interestingly, as can be seen in Figure 4, the C6 T1 equilibrium, this number is expected to be constant, though
distributions for both surfaces retain a bimodal character, fluctuating, over long time scales. However, the simulations did
although as expected the short-T1 part of the distribution is indeed exhibit a small drift in fibril−fibril hydrogen bonds,
more suppressed for the less-accessible regions. This part of increasing by ∼5% during the full 250 ns run (Figure S3). In
the distribution is likely to have its origin in the large- addition, as mentioned in the Introduction, there is a range of
amplitude motions connected to rotation of the hydroxymethyl other factors that could potentially contribute to the
group, which is naturally more restricted at the less-accessible heterogeneity in real FAs. This includes twisting of nanofibrils
surfaces. Nevertheless, the effect is small, particularly so if we as well as the possibility of having hemicellulose present inside
G DOI: 10.1021/acs.macromol.9b00472
Macromolecules XXXX, XXX, XXX−XXX
Macromolecules Article

Figure 5. Decomposition of simulated 13C T1 distributions of accessible surfaces into contributions from (A) glucose residues with hydroxymethyl
orientations either exterior or interior and (B) chain identity (center or origin).

the aggregates, both of which would lead to increased disorder 62:37:1, which is in qualitative agreement with earlier
in the less-accessible surfaces. simulations using the same force field.57 On the basis of
Regardless of which fibril cross section is assumed, there will NMR measurements on CNF that was repeatedly TEMPO-
be chains at the corner between different surfaces, which makes oxidized followed by alkali extraction, Funahashi et al.58
the assignment of these surfaces with respect to their concluded that gg was the dominating conformer of the
accessibility to water ambiguous (see Figure 1B). These chains exterior surface C6 groups, while gt was dominating for the
are also the least constrained by neighboring chains. From the interior ones. The same assignment was also made by
simulations it was found that corner chains exhibit a broad T1 Fernandes et al.13 However, Phyo et al.49 reported based on
distribution, shifted toward short T1 values as compared to the 2D 13C NMR that the gt conformer is the dominating one in
total distribution, meaning that corner chains are more mobile surface chains. Their conclusion correlates well with the results
than the average surface chain. This is an illustration of the presented here for the exterior surface C6, although the
level of detail possible with the present methodology. simulated population in gg is higher than what was detected in
Role of Surface Orientation for Hydroxymethyl Groups. the experiments. The distribution for interior groups, on the
Throughout surface chains, the glucose unit alternatively has other hand, is closer to what is expected for the core, with tg
its hydroxymethyl group pointing outward, to the exterior of being the dominant conformer.
the fibril, or inward, to the interior of the fibril (Figure 1E).
The mean transition time between conformations is 0.5 ns
When the distributions for surface C6 carbons were
for the exterior C6, 2 ns for interior, and, as mentioned above,
decomposed based on this distinction, a clear trend was
4 ns for the crystalline core. Obviously this is much shorter
found. For C6, the population of interior groups accounted for
than the NMR time scales and also sufficiently fast to
the long T1 values, while exterior groups accounted for the
short T1 values (Figure 5A and Table 1). It comes as no contribute to 13C relaxation times. This contributes to the
surprise that this orientation influences the T1 of C6, since the complexity of the C6 region of the 13C NMR spectrum,
motions of hydroxymethyl groups oriented toward the fibril presumably consisting of three populations which are all
interior are obviously restricted compared to the exterior ones. substantially overlapping: one almost exclusively in tg (core),
As noted above, crystalline cellulose have its hydroxymethyl one with almost no tg at all (surface exterior), and one in
groups almost exclusively in tg, whereas disordered chains (e.g., between these two (surface interior). This distinction has to be
surfaces) primarily exhibit gt and/or gg. This distinction accounted for a correct interpretation of the relative signal
between core and surfaces with respect to the hydroxymethyl intensities.59 Importantly, the short transition time among
conformation was seen in MD simulations already 20 years different conformers indicates that experimental NMR results
ago.56 It is also reproduced in the present work, where the core that have been obtained in a given finite chemical shift region
chains exhibits >96% tg, the rest having gg conformation. cannot be assumed to arise from unique distinct conformers.
For surfaces, the relative population in tg is 37%, but when Instead, also considering that line widths are quite significant
also the orientation is taken into account, it is found that for as compared to the chemical shift differences between distinct
exterior surface hydroxymethyl groups the population is conformers, one cannot exclude a significant admixture of
distributed as (tg:gg:gt) 7:27:66, and for interior groups other conformational states in any data recorded in a set
H DOI: 10.1021/acs.macromol.9b00472
Macromolecules XXXX, XXX, XXX−XXX
Macromolecules Article

chemical shift range. Hence, the results of Phyo et al.49 may when the hydroxymethyl group in the same residue is free to
permit a small yet nonvanishing tg population in surface chains. rotate. This is further strengthened by the observation that
The hydroxymethyl populations in the accessible surfaces there is no significant difference in C1 and C4 relaxation
were stable in the simulations, but in the less-accessible between center and origin chains at less-accessible surfaces,
surfaces, just as for the fibril−fibril hydrogen bonds, the where also exterior hydroxymethyl groups are more con-
simulations exhibited a small drift, where gt was slowly strained. We tentatively identify center and origin chains as
increasing, at the expense of tg, a further indication of the those two distinct populations of surface chains that have been
extremely long time scales needed to bring the fibril−fibril termed f and g by Wang et al.,48 which have been observed to
interface to equilibrium (Figure S4). In addition, the exhibit different average conformations but to appear well-
hydroxymethyl orientation also affects the local mobility of mixed and hydrated.


the backbone carbons (C1 and C4) within the same glucose
unit. Although the distributions are overlapping to a large CONCLUSIONS
extent (Figure 4), their means differ by a factor of 2 (Table 1).
This result stands in contrast with a previous study where the In the present work, 13C NMR T1 distributions in a model
C6 orientation did not affect simulated mean T1 for C460 at all. cellulose fibril aggregate were calculated from MD simulation
In that same study, a significant difference was detected at full trajectories. Good agreement between simulated and exper-
hydration for mean C4 T1 values in surface chains on top of imental distributions was demonstrated for the spectrally
the crystallographic (110) or (11̅0) planes. This difference is resolved carbons C1, C4, and C6 separately. Both simulated
not reproduced here. As one explanation, one has to keep in and experimental distributions were broad, indicating a large
mind that the model used in that study was far less detailed variation in dynamics over chemically equivalent 13C groups.
and realistic both as concerning atomic (such as no explicit The results showed that this dynamical heterogeneity to a large
inclusion of aliphatic hydrogens) and nanofibril (a semi-infinite extent is a consequence of the structural heterogeneity in
crystal model instead of a nanofibril aggregate) arrangements. cellulose. Utilizing the full atomistic resolution of MD
Although equilibrium properties are less affected, coarse- simulations, it was possible to reach beyond the mere
grained schemes are known to influence dynamic properties, distinction between surface and core chains, which can also
primarily by smoothing out the potential energy surface, which be made in experiments, and investigate effects from more
in turn leads to a speed-up of the kinetics. subtle structural variation such as orientation of the
Center and Origin Chains. In addition to the structural hydroxymethyl group and differences between center and
distinctions discussed above, the cellulose Iβ structure contains origin chains. This demonstrated the strength of the present
two distinguishable molecules that are present in the crystal methodology since these are features that cannot be resolved
unit cell, termed center and origin chains, respectively,20 where by using experiments alone.
one chain is located at the origin of the unit cell and the other From a materials perspective, nanocellulose is unique
at its center (Figure 1C). This implies that half of the molecules compared with other high-strength fibers and other nano-
in CNF are origin chains and the other half are center chains. particles in that there is substantial dynamics in the reinforcing
Experiments indicate that these chains show differences in phase. Today, there are many types of nanocelluloses available.
terms of structural parameters, such as puckering of the They originate from different sources and disintegration
pyranose rings, conformation of glycosidic linkages, and procedures, which give them different fibril size distributions
orientation of the hydroxymethyl group as well as hydrogen- and degrees of order. These nanocellulose materials typically
bonding pattern.20,51 These structural differences impart small undergo various chemical modifications and also have different
variations in the 13C NMR chemical shifts of C1, C4, and C6 hygrothermal histories (e.g., in terms of drying and rewetting),
in highly crystalline samples.61 The most important geo- which severely affects the state of aggregation. Even common
metrical differences between center and origin chains are also chemical pulp consists of cellulose fibril aggregates, with a
reproduced in simulations with the GLYCAM06 force field.57 structural heterogeneity that is dependent on its history. Here,
Yet, within an infinite crystal one would expect the difference the combination of spectrally resolved 13C NMR relaxometry
in dynamics between center and origin chains to be rather and MD simulations is shown to be a excellent tool for learning
small, since the coordination of the glucose units in them is about supramolecular structure and organization, and its effect
very similar. However, this is not valid anymore when the on the state of macromolecular mobility in the presence of
chains are located at fibril surfaces. From the present moisture. Specifically, we show that the interplay between
simulations, we find that average 13C T1 values from surface structure and dynamics is highly complex and that even the
origin chains are consistently larger than the corresponding seemingly homogeneous crystalline core exhibits heteroge-
values from surface center chains (Table 1). The underlying neous dynamics that varies over an order of magnitude. The
reason can be seen in Figure 5B, which shows that for the case methodology can be easily adapted to study other natural
of C1 and C4 center and origin chains give rise to two distinct fibers, e.g., chitin, polymer matrices in nanocomposites, native
subpopulations in the T1 of accessible surfaces. For C6, on the hemicellulose/cellulose systems in plant cell walls, or synthetic
other hand, the differences are negligible, presumably because fibers, with a level of detail that goes far beyond the
faster dynamical modes dominate. experimental limit.


Interestingly, when considering also the orientation of C6,
one can see that for residues with C6 oriented to the exterior, ASSOCIATED CONTENT
the difference between center and origin chains observed for
C1 and C4 relaxation becomes stronger, while for those with *
S Supporting Information

C6 oriented toward the interior, the difference disappears The Supporting Information is available free of charge on the
(Figure S5). This means that there is a difference in backbone ACS Publications website at DOI: 10.1021/acs.macro-
dynamics between center and origin chains that is tangible mol.9b00472.
I DOI: 10.1021/acs.macromol.9b00472
Macromolecules XXXX, XXX, XXX−XXX
Macromolecules Article

Experimental site-resolved T1 buildup curves with (11) Š turcová, A.; His, I.; Apperley, D. C.; Sugiyama, J.; Jarvis, M. C.
Structural details of crystalline cellulose from higher plants.
respective biexponential fitting results; computational
Biomacromolecules 2004, 5, 1333−1339.
details for calculation of NMR relaxation times; (12) Thomas, L. H.; Forsyth, V. T.; Š turcová, A.; Kennedy, C. J.;
decomposed T1 distributions for the core, accessible May, R. P.; Altaner, C. M.; Apperley, D. C.; Wess, T. J.; Jarvis, M. C.
surfaces, and less-accessible surfaces; number of fibril− Structure of cellulose microfibrils in primary cell walls from
fibril hydrogen bonds; conformation of hydroxymethyl collenchyma. Plant Physiol. 2013, 161, 465−476.
groups in less-accessible surfaces (PDF) (13) Fernandes, A. N.; Thomas, L. H.; Altaner, C. M.; Callow, P.;


Forsyth, V. T.; Apperley, D. C.; Kennedy, C. J.; Jarvis, M. C.
Nanostructure of cellulose microfibrils in spruce wood. Proc. Natl.
AUTHOR INFORMATION Acad. Sci. U. S. A. 2011, 108, E1195−E1203.
Corresponding Author (14) Viëtor, R. J.; Newman, R. H.; Ha, M.-A.; Apperley, D. C.; Jarvis,
M. C. Conformational features of crystal-surface cellulose from higher
*E-mail: jacke@kth.se.
plants. Plant J. 2002, 30, 721−731.
ORCID (15) Terenzi, C.; Prakobna, K.; Berglund, L. A.; Furó, I.
Pan Chen: 0000-0003-3794-717X Nanostructural effects on polymer and water dynamics in cellulose
Camilla Terenzi: 0000-0003-3278-026X biocomposites: 2H and 13C NMR Relaxometry. Biomacromolecules
2015, 16, 1506−1515.
István Furó: 0000-0002-0231-3970 (16) Wickholm, K.; Larsson, P. T.; Iversen, T. Assignment of non-
Lars A. Berglund: 0000-0001-5818-2378 crystalline forms in cellulose I by CP/MAS 13C NMR spectroscopy.
Jakob Wohlert: 0000-0001-6732-2571 Carbohydr. Res. 1998, 312, 123−129.
Notes (17) Earl, W. L.; VanderHart, D. L. High resolution, magic angle
sampling spinning carbon-13 NMR of solid cellulose I. J. Am. Chem.
The authors declare no competing financial interest.


Soc. 1980, 102, 3251−3252.
(18) Chen, P.; Terenzi, C.; Furó, I.; Berglund, L. A.; Wohlert, J.
ACKNOWLEDGMENTS Hydration-dependent dynamical modes in xyloglucan from molecular
This work was partially supported by the Wallenberg Wood dynamics simulation of 13C NMR relaxation times and their
Science Center (WWSC) funded by the Knut and Alice distributions. Biomacromolecules 2018, 19, 2567−2579.
(19) Fahlén, J.; Salmén, L. Pore and matrix distribution in the fiber
Wallenberg Foundation, the Swedish Research Council, VR
wall revealed by atomic force microscopy and image analysis.
(I.F.), and the Beijing Institute of Technology Research Fund Biomacromolecules 2005, 6, 433−438.
for Young Scholars (P.C.). Computational resources were (20) Nishiyama, Y.; Langan, P.; Chanzy, H. Crystal structure and
provided by the Swedish National Infrastructure for Comput- hydrogen-bonding system in cellulose Iβ from synchrotron X-ray and
ing (SNIC) at the PDC Center for High Performance neutron fiber diffraction. J. Am. Chem. Soc. 2002, 124, 9074−9082.
Computing. (21) Cosgrove, D. J. Re-constructing our models of cellulose and


primary cell wall assembly. Curr. Opin. Plant Biol. 2014, 22, 122−131.
REFERENCES (22) Daicho, K.; Saito, T.; Fujisawa, S.; Isogai, A. The crystallinity of
nanocellulose: dispersion-induced disordering of the grain boundary
(1) Magat, E. E. Fibers from extended chain aromatic polyamides, in biologically structured cellulose. ACS Appl. Nano Mater. 2018, 1,
new fibers and their composites. Philos. Trans. R. Soc., A 1980, 294, 5774−5785.
463−472. (23) Langan, P.; Nishiyama, Y.; Chanzy, H. X-ray structure of
(2) Hunsaker, M. E.; Price, G. E.; Bai, S. J. Processing, structure and
mercerized cellulose II at 1 Å resolution. Biomacromolecules 2001, 2,
mechanics of fibres of heteroaromatic oxazole polymers. Polymer
410−416.
1992, 33, 2128−2135.
(24) Langan, P.; Sukumar, N.; Nishiyama, Y.; Chanzy, H.
(3) Berglund, L. A.; Peijs, T. Cellulose biocomposites - from bulk
Synchrotron X-ray structures of cellulose Iβ and regenerated cellulose
moldings to nanostructured systems. MRS Bull. 2010, 35, 201−207.
(4) Nishiyama, Y.; Johnson, G. P.; French, A. D.; Forsyth, V. T.; II at ambient temperature and 100 K. Cellulose 2005, 12, 551−562.
Langan, P. Neutron crystallography, molecular dynamics, and (25) Revol, J.-F.; Goring, D. A. I. On the mechanism of the
quantum mechanics studies of the nature of hydrogen bonding in mercerization of cellulose in wood. J. Appl. Polym. Sci. 1981, 26,
cellulose Iβ. Biomacromolecules 2008, 9, 3133−3140. 1275−1282.
(5) Li, Y.; Fu, Q.; Yu, S.; Yan, M.; Berglund, L. A. Optically (26) Watanabe, Y.; Meents, M. J.; McDonnell, L. M.; Barkwill, S.;
transparent wood from a nanoporous cellulosic template: combining Sampathkumar, A.; Cartwright, H. N.; Demura, T.; Ehrhardt, D. W.;
functional and structural performance. Biomacromolecules 2016, 17, Samuels, A. L.; Mansfield, S. D. Visualization of cellulose synthases in
1358−1364. Arabidopsis secondary cell walls. Science 2015, 350, 198−203.
(6) Berglund, L. A.; Burgert, I. Bioinspired wood nanotechnology for (27) Revol, J.-F. On the cross-sectional shape of cellulose crystallites
functional materials. Adv. Mater. 2018, 30, 1704285. in Valonia ventricosa. Carbohydr. Polym. 1982, 2, 123−134.
(7) Nishiyama, Y.; Langan, P.; O’Neill, H.; Pingali, S. V.; Harton, S. (28) Chanzy, H. In Cellulose Sources and Exploitation: Industrial
Structural coarsening of aspen wood by hydrothermal pretreatment Utilization, Biotechnology and Physico-Chemical Properties; Kennedy, J.
monitored by small- and wide-angle scattering of X-rays and neutrons F., Phillips, G. O., Williams, P. A., Eds.; Ellis Horwood, Inc.:
on oriented specimens. Cellulose 2014, 21, 1015−1024. Chichester, 1990; pp 3−12.
(8) Larsson, P. T.; Wickholm, K.; Iversen, T. A CP/MAS 13C NMR (29) Ogawa, Y.; Chanzy, H.; Putaux, J.-L. Transmission electron
investigation of molecular ordering in celluloses. Carbohydr. Res. microscopy of cellulose. Part 1: historical perspective. Cellulose 2019,
1997, 302, 19−25. 26, 5−15.
(9) Whitney, S. E. C.; Brigham, J. E.; Darke, A. H.; Reid, J. G.; (30) Kuribayashi, T.; Ogawa, Y.; Rochas, C.; Matsumoto, Y.; Heux,
Gidley, M. J. In vitro assembly of cellulose/xyloglucan networks: L.; Nishiyama, Y. Hydrothermal transformation of wood cellulose
ultrastructural and molecular aspects. Plant J. 1995, 8, 491−504. crystals into pseudo-orthorhombic structure by cocrystallization. ACS
(10) Horii, F.; Hirai, A.; Kitamaru, R. Solid-state high-resolution Macro Lett. 2016, 5, 730−734.
13
C-NMR studies of regenerated cellulose samples with different (31) Donaldson, L. Cellulose microfibril aggregates and their size
crystallinities. Polym. Bull. 1982, 8, 163−170. variation with cell wall type. Wood Sci. Technol. 2007, 41, 443−460.

J DOI: 10.1021/acs.macromol.9b00472
Macromolecules XXXX, XXX, XXX−XXX
Macromolecules Article

(32) Lindh, E. L.; Terenzi, C.; Salmén, L.; Furó, I. Water in (54) Prakobna, K.; Terenzi, C.; Zhou, Q.; Furó, I.; Berglund, L. A.
cellulose: evidence and identification of immobile and mobile Core-shell cellulose nanofibers for biocomposites − nanostructural
adsorbed phases by 2H MAS NMR. Phys. Chem. Chem. Phys. 2017, effects in hydrated state. Carbohydr. Polym. 2015, 125, 92−102.
19, 4360−4369. (55) O’Neill, H.; Pingali, S. V.; Petridis, L.; He, J.; Mamontov, E.;
(33) Hess, B.; Kutzner, C.; van der Spoel, D.; Lindahl, E. Hong, L.; Urban, V.; Evans, B.; Langan, P.; Smith, J. C.; Davison, B.
GROMACS 4: algorithms for highly efficient, load-balanced, and H.; et al. Dynamics of water bound to crystalline cellulose. Sci. Rep.
scalable molecular simulation. J. Chem. Theory Comput. 2008, 4, 435− 2017, 7, 11840.
447. (56) Heiner, A. P.; Kuutti, L.; Teleman, O. Comparison of the
(34) Kirschner, K. N.; Yongye, A. B.; Tschampel, S. M.; González- interface between water and four surfaces of native crystalline
Outeirino, J.; Daniels, C. R.; Foley, B. L.; Woods, R. J. GLYCAM06: cellulose by molecular dynamics simulations. Carbohydr. Res. 1998,
A generalizable biomolecular force field. Carbohydrates. J. Comput. 306, 205−220.
Chem. 2008, 29, 622−655. (57) Matthews, J. F.; Beckham, G. T.; Bergenstråhle-Wohlert, M.;
(35) Jorgensen, W. L.; Chandrasekhar, J.; Madura, J. D.; Impey, R. Brady, J. F.; Himmel, M. E.; Crowley, M. F. Comparison of cellulose
W.; Klein, M. L. Comparison of simple potential functions for Iβ simulations with three carbohydrate force fields. J. Chem. Theory
simulating liquid water. J. Chem. Phys. 1983, 79, 926−935. Comput. 2012, 8, 735−748.
(36) Abragam, A. The Principles of Nuclear Magnetism; University (58) Funahashi, R.; Okita, Y.; Hondo, H.; Zhao, M.; Saito, T.;
Press: London, 1961. Isogai, A. Different conformations of surface cellulose molecules in
(37) Kowalewski, J.; Effemey, M.; Jokisaari, J. Dipole-dipole native cellulose microfibrils revealed by layer-by-layer peeling.
coupling constant for a directly bonded CH pair − A carbon-13 Biomacromolecules 2017, 18, 3687−3694.
relaxation study. J. Magn. Reson. 2002, 157, 171−177. (59) Oehme, D. P.; Downton, M. T.; Doblin, M. S.; Wagner, J.;
(38) Brondeau, J.; Canet, D. Longitudinal magnetic relaxation of 13C Gidley, M. J.; Bacic, A. Unique aspects of the structure and dynamics
(or 15N) interacting with a strongly irradiated proton system. J. Chem. of elementary Iβ cellulose microfibrils revealed by computational
Phys. 1977, 67, 3650−3654. simulations. Plant Physiol. 2015, 168, 3−17.
(39) Bussi, G.; Donadio, D.; Parrinello, M. Canonical sampling (60) Bergenstråhle, M.; Wohlert, J.; Larsson, P. T.; Mazeau, K.;
through velocity rescaling. J. Chem. Phys. 2007, 126, 014101. Berglund, L. A. Dynamics of cellulose-water interfaces: NMR spin-
(40) Darden, T.; York, D.; Pedersen, L. Particle mesh Ewald: an N· lattice relaxation times calculated from atomistic computer
log(N) method for Ewald sums in large systems. J. Chem. Phys. 1993, simulations. J. Phys. Chem. B 2008, 112, 2590−2595.
98, 10089−10092. (61) Kono, H.; Numata, Y. Structural investigation of cellulose Iα
(41) Hess, B. P-LINCS. A parallel linear constraint solver for and Iβ by 2D RFDR NMR spectroscopy: determination of sequence
molecular simulation. J. Chem. Theory Comput. 2008, 4, 116−122. of magnetically inequivalent D-glucose units along cellulose chain.
(42) Atalla, R. H.; Vanderhart, D. L. Native cellulose: a composite of Cellulose 2006, 13, 317−326.
two distinct crystalline forms. Science 1984, 223, 283−285.
(43) Yamamoto, H.; Horn, F. In Situ crystallization of bacterial
cellulose I. Influences of polymeric additives, stirring and temperature
on the formation celluloses Iα and Iβ as revealed by cross polarization/
magic angle spinning (CP/MAS) 13C NMR spectroscopy. Cellulose
1994, 1, 57−66.
(44) Earl, W. L.; VanderHart, D. L. Observations by high-resolution
carbon-13 nuclear magnetic resonance of cellulose I related to
morphology and crystal structure. Macromolecules 1981, 14, 570−574.
(45) French, A. D.; Pérez, S.; Bulone, V.; Rosenau, T.; Gray, D.
Encyclopedia of Polymer Science and Technology; John Wiley & Sons,
Inc.: 2018; pp 1−69.
(46) Horii, F.; Hirai, A.; Kitamaru, R. Solid-State 13C-NMR study of
conformations of oligosaccharides and cellulose. Polym. Bull. 1983, 10,
357−361.
(47) Yang, H.; Wang, T.; Oehme, D.; Petridis, L.; Hong, M.;
Kubicki, J. D. Structural factors affecting 13C NMR chemical shifts of
cellulose: a computational study. Cellulose 2018, 25, 23−36.
(48) Wang, T.; Yang, H.; Kubicki, J. D.; Hong, M. Cellulose
structural polymorphism in plant primary cell walls investigated by
high-field 2D solid-state NMR spectroscopy and density functional
theory calculations. Biomacromolecules 2016, 17, 2210−2222.
(49) Phyo, P.; Wang, T.; Yang, Y.; O’Neill, H.; Hong, M. Direct
determination of hydroxymethyl conformations of plant cell wall
cellulose using 1H polarization transfer solid-state NMR. Biomacro-
molecules 2018, 19, 1485−1497.
(50) Kowalewski, J. In Nuclear Magnetic Resonance; Kamienska-
Trela, K., Ed.; Royal Society of Chemistry: 2015; Vol. 44, pp 235−
293.
(51) Chen, P.; Nishiyama, Y.; Mazeau, K. Torsional entropy at the
origin of the reversible temperature-induced phase transition of
cellulose. Macromolecules 2012, 45, 362−368.
(52) Istratov, A. A.; Vyvenko, O. F. Exponential analysis in physical
phenomena. Rev. Sci. Instrum. 1999, 70, 1233−1257.
(53) Berman, P.; Levi, O.; Parmet, Y.; Saunders, M.; Wiesman, Z.
Laplace inversion of low-resolution NMR relaxometry data using
sparse representation methods. Concepts Magn. Reson., Part A 2013,
42, 72−88.

K DOI: 10.1021/acs.macromol.9b00472
Macromolecules XXXX, XXX, XXX−XXX

You might also like