Crystallite fusion in nanocellulose aggregates


 Crystallite refers to a single crystalline grain in crystal aggregates, and multiple crystallites form a grain boundary or the inter-crystallite interface. A grain boundary is a structural defect that hinders the efficient directional transfer of mechanical stress or thermal phonons in crystal aggregates. We observed that grain boundaries within an aggregate of a-few-nanometers-wide fibrillar crystallites of cellulose were crystallized by enhancing their inter-crystallite interactions; multiple crystallites were coupled into single fusion crystals without passing through a melting or dissolving state. Accordingly, the crystallinity of naturally occurring cellulose, which has previously been considered irreversible once decreased, was recovered, and the thermal energy transfer in the aggregate was significantly improved. Other fibrillar crystallites of chitin also showed a similar fusion phenomenon by enhancing the inter-crystallite interactions. Crystallite fusion in aggregates may occur for other biopolymers.


Introduction
Mechanical and thermal energy transfers in crystal aggregates are governed by a grain boundary or the interface where multiple crystallites meet 1 . Proper crystallite interactions at the grain boundary can allow mechanical stress or thermal phonons to directionally transfer in the aggregate; however, a grain boundary is a structural defect that deforms under stress or scatters phonons 3,4 . This feature becomes pronounced when the crystallites have nanoscale dimensions and form a large area of the grain boundary.
In the field of nanotechnology, it is currently a challenge to exploit the potential of such nanoscale crystallites, including biopolymer fibrils and clay platelets, in bulk aggregates or composites by tailoring the interactions between crystallites or with other components 5-7 . Ideal energy transfer in crystal aggregates must be realised by crystallization of the grain boundary. If multiple crystallites can be coupled into single fusion crystals by forming a bulk aggregate from their dispersions, scalable polycrystalline materials with more efficient mechanical and thermal energy transfers will be produced. However, crystallites often have disordered structures or defects at their surfaces 5,8 , hindering efficient energy transfer.
Herein, we report that grain boundaries within an aggregate of a-fewnanometers-wide crystalline cellulose nanofibers (CNFs) were crystallized by enhancing the inter-CNF interactions. CNFs are stainable materials with excellent mechanical and thermal properties, which are produced as a water dispersion using wood pulps as the raw material 9 . Their elemental unit is a crystallite consisting of uniaxially oriented molecular chains, known as a cellulose microfibril 10,11 . In the dispersion, the surface molecules of CNFs have a thermodynamically stable, noncrystalline conformation and are uniaxially oriented, similar to the inner crystalline molecules [11][12][13] .
We previously showed that the crystallinity of CNFs significantly decreased when aggregated microfibrils in a pulp dispersed as CNFs or separated into individual crystallites 12 . This phenomenon was interpreted to result from the exposure of the grain boundary that was partially crystallized in the microfibril aggregates. This study was aimed at recovering the decreased crystallinity of the CNFs by assembly. We also verified if the crystallization of the grain boundary led to greater efficiency in the thermal energy transfer of bulk crystal aggregates.

Aggregating CNFs
Scheme 1 shows the procedure for enhancing the inter-CNF interactions in a bulk aggregate. The CNF dispersion was prepared by wet disintegration of a TEMPOoxidised wood pulp 14 . In the TEMPO-oxidation reaction, the C6 hydroxy groups exposed on the microfibril surfaces are regioselectively oxidised, such that the resulting CNFs possess a high surface density of carboxy groups (up to ~1.7 groups per square nanometer). The carboxy groups are Na + type at the initial step. By the dissociation of the carboxy groups, the CNFs are stably dispersed in water and spontaneously form a nematic liquid-crystalline phase where the CNFs are uniaxially oriented in places (Scheme 1a) 15 . In this study, the nematic-ordered CNF dispersion was assembled into dry sheets by two processes. One process was freeze drying from a 30% tert-butyl alcohol-containing diluted dispersion, followed by pressing of the dried aerogel-like product (sample i) 16 . The other process was evaporative drying via solvent casting of the CNF dispersion at 40 °C under a high relative humidity of 80%, resulting in the formation of transparent and flat dry sheets (sample ii) 17 . Sample ii was further processed with a dilute acid solution to convert the carboxy groups from sodium salt into protonated acid form, H type (sample iii, see Figure S1a for the evidence of protonation) 18 , followed by a hydrothermal treatment using a common autoclave at 135 °C and 212 kPa (sample iv). The hydrothermal treatment was adopted to enable the CNFs to rearrange and stabilise their interaction. The CNFs remained solid, without melting or dissolving, under the conditions adopted in this study.
The assembled state of the CNFs in each sheet is illustrated in Scheme 1b: roughly oriented in sample i, and highly oriented in samples ii-iv. Meanwhile, the CNFs of samples i-iv were on average packed parallel to the sheet surface (see Figures S2 for X-ray diffraction (XRD) diagrams) 17 . The CNFs in samples iii and iv were also hydrogen-bonded with one another via the surface carboxy groups, as analysed by Fourier-transform infrared (FTIR) spectroscopy (Figure S1a), and those in sample iv were more strongly hydrogen-bonded ( Figure S1b,c). Accordingly, the porosity of the samples decreased in the order of sample i > ii > iii > iv, at 16%, 10%, 9%, and 4%, respectively. The specific surface area (SSA) of sample i was ~6.3 m 2 g -1 , as measured by nitrogen adsorption analysis ( Figure S3), whereas those of samples ii, iii, and iv were all below the limit of detection (~2 m 2 g -1 ). Therefore, the CNFs in each sample were considered to interact more closely or strongly in the order of sample i < ii < iii < iv. Figure 1a shows powder XRD profiles of the oxidised starting pulp and CNF samples iiv. The XRD profiles of the pulp and sample i notably differed; all the peaks broadened and their intensities decreased. From sample i to ii, the peak widths narrowed. From sample iii to iv, the peaks further sharpened and intensified, and the (2 0 0) peak shifted slightly to the high-angle side. Finally, sample iv exhibited a distinct XRD profile rather than the starting pulp.

Recovery of Crystallinity
The nuclear magnetic resonance (NMR) spectra for the samples are shown in Figure 1b. In the C4 and C6 regions of the NMR spectra, the crystalline signals centred at 88 ppm (C4) and 65 ppm (C6) remarkably decreased at the first step of the pulp-sample i conversion and increased in the order of samples ii, iii, and iv. Meanwhile, the non-crystalline signals at 84 ppm (C4) and 62 ppm (C6) increased at the first step and then decreased in the order of samples ii, iii, and iv.
As measures of the crystallinity, the Scherrer's crystal sizes of the (2 0 0) plane and crystallinity indices were calculated from the XRD profiles and NMR spectra, respectively (Figure 1c,d) 12 . The crystallinity indices were expressed in two different ways as the area ratios of the crystalline and non-crystalline signals in the C4 and C6 regions. Note that the crystalline C6 signal arises from the C6 carbon atoms turning the C6−O6 bond to the "trans−gauche (tg)" configuration against the C5−O5/C4−C5 bonds in anhydroglucose units 11 , and that the crystalline C4 conformation is fixed with the intra/inter-molecular hydrogen bonds via the tg-configurational C6 hydroxy groups. The crystallinity indices reflected the degree of intra-/inter-molecular hydrogen bonds.
The (2 0 0) crystal size decreased from 3.5 nm to 2.0 nm by the pulp-sample i conversion, and increased up to 3.8 nm in the order of samples ii, iii, and iv ( Figure 1c). This increase in the crystal size explained the slight shift of the (2 0 0) peak position to the high-angle side, according to a previous report 19 . Also, the two crystallinity indices of the C4 and C6 carbon atoms decreased by ~30% from 40-50% on the conversion to sample i, and then increased by ~30% for samples ii, iii, and iv ( Figure 1d). For both the crystal size and crystallinity index, the initial decrease was interpreted to result from the increase in SSA by disintegration of the pulp into the CNF dispersion (see Introduction) 12 . The successive increase was thus attributed to assembly of the CNFs in dry sheets; the inter-CNF interaction was dominant in the recovery of crystallinity.
Interestingly, the final values for sample iv reached or even surpassed the crystal size and crystallinity indices of the starting pulp; the crystallinity of the CNFs was reversible (see Figure S4 for the repeatability data). The recovery of crystallinity was also demonstrated for other type of CNFs produced from chemically unmodified, raw pulps solely by wet disintegration ( Figure S5). Furthermore, other fibrillar crystallites of α-chitin showed the recovery of crystallinity through the same process as the dispersion and assembly of CNFs in this study ( Figure S6). 6 The contribution of the inter-CNF interactions to the recovery of crystallinity was investigated in more detail using two additional surface-modified CNFs ( Figure 2a):

Inter-CNF interactions
hydrophobic CNFs bearing bulky tetra-n-butylammonium (TBA) as the counterion of the surface carboxy group 20 ; and polymer-covered CNFs adsorbing an amorphous, hydroxyethyl cellulose by ~10% w/w 21 . The adsorption of 10% roughly corresponded to two HEC molecules per single CNF. These two CNF dispersions were assembled into transparent dry sheets through the same drying process as adopted in the preparation of samples ii and iii. surfaces (see inset in Figure 1a), such that the hydrophobic (2 0 0) surface was exposed.
Thus, the recovery of the crystal size indicated that TBA-bearing CNFs were stacked on the exposed (2 0 0) plane. Meanwhile, the low crystallinity index indicated that the inter-CNF hydrogen bonds between the hydrophilic surfaces were inhibited by the bulky TBA ions.
The HEC-covered CNFs showed different behaviours, in that both the crystal size and crystallinity index slightly increased (Figure 2b,c). HEC is amphiphilic, such that the whole surface of the CNFs were sparsely covered with HEC at a CNF/HEC ratio of 9:1. The results in Figure 2b,c for the HEC-covered CNFs were thus attributed to "partially-blocked" inter-CNF interactions.
These results supported that the inter-CNF interactions were dominant in the recovery of crystallinity, indicating that the inter-CNF (2 0 0) stacking and hydrogen bonding contributed mainly to the recovery of the crystal size and crystallinity index, respectively, which was also supported by the results for samples ii and iii in Figure   1c,d. From the Na + -type sample ii to the hydrogen-bonded H-type sample iii, the (2 0 0) crystal size was approximately constant, whereas the crystallinity index significantly increased.
It is well known as a phenomenon "hornification" in the field of wood pulp and paper sciences that the crystallinity of cellulosic samples including CNFs slightly increases by the repeating cycle of wet-dry states 22,23 . This phenomenon has been ambiguously interpreted to result from enhancement of the degree of hydrogen bonding in samples by drying, and its mechanism remains unclear. The results in this study suggested that hornification is based on inter-microfibril (2 0 0) stacking and hydrogen bonding.

Simulation
To assess the major configuration of CNFs in the interactions, an XRD profile of sample iv was compared with simulated profiles of possible inter-CNF configurations (Figures 3 and S7). The experimental profile was obtained by azimuthally integrating an XRD diagram of a sample iv sheet set parallel to the beam. We assumed no specific orientation of the crystal planes to the sheet surface by considering: 1) the similarity of the experimental profile in Figure 3 with that obtained by the reflection method for sample iv in Figure 1a, and 2) the twisting structure of the CNFs around the crystallographic c axis 24-26 .
In the simulation, the structure of the single CNFs was assumed to be composed of 18 cellulose chains with a stacking mode of 2/3/4/4/3/2, based on previous reports on the morphological analyses of single CNFs 12,27 . This 18-chain model had flat two-molecule-wide (2 0 0) surfaces, enabling the CNFs to stably stack on the (2 0 0) plane (see the results for TBA-bearing CNFs in Figure 2). As shown in Figure 3, the CNFs assembled parallel to one another along the c axis. Simulations of antiparallel assembly are shown in Figure S7. Coupling of only two CNFs was simulated here for simplicity but the reality should be more complex.
The highest R 2 value (0.897) was obtained for configuration #12, where the modelled CNFs coupled facing their (1 1 0) surfaces 28 . This configuration allowed the CNFs to form inter-CNF hydrogen bonds, explaining the recovery of the crystal size and crystallinity index for sample iv. In addition, the (1 1 0) plane had a larger surface free energy than the other planes 29 . The CNF coupling for configuration #12 thus reduced the free energy in the system to a greater extent and was the most stable of the possible inter-CNF configurations. The XRD profile of the starting pulp (Figure 1a) also showed the best fit (R 2 = 0.883) with configuration #12. A similar result held for the antiparallel assembly shown in Figure S5; the best fit (R 2 = 0.875) was achieved at a configuration that significantly reduced the (1 1 0) surface. In the antiparallel assembly, there existed no configuration where the CNFs coupled by facing the same crystal plane.
Configuration #7 matched the experimental profile in terms of peak shape, especially at lower angles of ~15°. However, its coefficient of correlation, R 2 (0.826), was the lowest (see Methods for the R 2 calculation). This gap was because the intensity ratio of the (2 0 0) plane and combined (1 −1 0)/(1 1 0) peaks in the simulation was significantly different from the corresponding ratio for the experimental profile; on closer inspection, the position of the combined (1 −1 0)/(1 1 0) peak in the simulation was shifted by ~1° from the experimental peak.

Modelling
The recovery of the (2 0 0) crystal size was easily interpreted as the phase extension of the (2 0 0) plane by CNF assembly. However, the mechanism for the recovery of crystallinity index must be considered. In Figure 1d, the C4-and C6-derived crystallinity indices were coordinated and recovered from the lowest degree of 15% (sample i) by up to ~30% (sample iv). Considering that most of the C6 hydroxy groups exposed on the CNF surface were converted to carboxy groups, the recovery of crystallinity index was interpreted to result from conformational changes not only of the residual surface C6 hydroxy group but also of the interior of each CNF structure via inter-CNF hydrogen bonding. Figure 4 illustrates a model for the conformational changes induced by the inter-CNF hydrogen bonding. Three cases were assumed based on the hydrogenbonding mode in the cellulose I type structure 11 . In case 1, the residual surface C6 hydroxy groups take the tg conformation from other noncrystalline states (gt or gg, see the section "Recovery of Crystallinity"). This change was the most plausible, but the maximum contribution to the recovery of crystallinity index was estimated to be ~10%, which was insufficient for explaining the recovery of 30%. Cases 2 and 3 describe the conformational changes occurring at the interior of each CNF structure. In case 2, the C6 hydroxy groups in the surface molecules of each CNF facing inside adopt the tg conformation. In case 3, the C6 hydroxy groups in the interior molecules facing the surface molecules adopt the tg conformation. These two cases were assumed to result from some restraint imposed on the surface molecules via the inter-CNF hydrogen bonding.

Bulk properties
The recovery of crystallinity was expected to improve the bulk properties of the CNF structures. Figure 5 shows the thermal diffusivity, α, and conductivity, k, of samples i-iv as a function of the crystallinity index. With the recovery of crystallinity, both the diffusivity and conductivity significantly improved (see Figure S8 for the specific values of α and k divided by their bulk densities, showing the same trend). This trend is reasonable because heat transfers in a solid via phonon propagation. The phonon is an elastic wave and is often scattered at the grain boundary in a particle assembly. The recovery of crystallinity was caused by the inter-CNF interactions, which reduced the grain boundary and facilitated phonon propagation.

Conclusions and perspectives
Enhancing the interaction between CNFs or a-few-nanometer-wide fibrillar crystallites of cellulose resulted in the coupling of multiple crystallites into single fusion crystals in a bulk aggregate without passing through a melting or dissolving state. The interaction was enhanced by the following steps: starting from a nematic liquid-crystalline dispersion where the CNFs were uniaxially oriented in places, forming a dense aggregate by condensation, and enabling the CNFs to bind with one another via hydrogen bonds at the grain boundary. These steps induced the conformational change of the constituent carbon atoms of the CNFs to be crystalline, and the phase extension of the crystal planes occurred. Accordingly, the lowered crystallinity of the CNFs, which was previously considered irreversible, was recovered, and thermal energy transfer in the aggregate was significantly improved. CNFs have recently been produced in industrial settings, e.g. capacity ~1000 tons in Japan in 2020 30 , and this finding will contribute to the building of the technical bases for exploiting the potential of CNFs in bulk materials. This finding offers a deeper understanding of "hornification", which has been ambiguously interpreted in the field of wood pulp and paper sciences. CNF sheets. The CNF dispersion was subjected to freeze drying or evaporative drying.
The dispersion (0.1%) was freeze-dried from a 30% tert-butyl alcohol-containing wet state according to a previously reported method 16 . The freeze-dried CNF was conditioned at 23 °C and 50% relative humidity, followed by pressing at ~750 MPa for 1 min to obtain an aerogel-like sheet (sample i). The dispersion was concentrated to 0.4% at 40 °C using an evaporator, and then poured into a 90-mm diameter polystyrene petri dish, followed by evaporative drying at 40 °C and 80% relative humidity for over a week (sample ii) 17 . The resulting CNF sheet was immersed in a 0.1 M HCl aqueous solution for 2 h, followed by washing with distilled water and drying at 40 °C and 80% relative humidity (sample iii). Sample iii was placed in a glass petri dish and further subjected to hydrothermal treatment using a LSX-500 autoclave at 135 °C and 212 kPa for 30 min. The resulting film was gently washed with distilled water, followed by drying at 40 °C and 80% relative humidity (sample iv). All samples were conditioned at 23 °C and 50% relative humidity before analysis.
TBA-bearing CNF. The oxidised pulp suspension (0.1%) was mixed with 0.1 M HCl for 2 h. After washing with distilled water, the pulp was neutralised with 10% tetra-nbutylammonium (TBA) hydroxide according to a previously reported method 31 . The TBA-bearing CNF dispersion was then dried at 40 °C and 80% relative humidity, and conditioned at 23 °C and 50% relative humidity before the analyses.
HEC-covered CNF. The Na + type of the 0.1% CNF dispersion was mixed with a 10% w/w hydroxyethyl cellulose solution (degree of substitution, 1.6-1.8) 21 , and subsequently dried and conditioned in the same manner as the TBA-bearing CNF.
XRD. XRD measurements were performed in the reflection and transmission modes. In the reflection mode, the XRD profiles were obtained at diffraction angles 2θ ranging from 3-45° using a Rigaku Mini Flex diffractometer with Ni-filtered Cu Kα radiation (λ = 0.1542 nm) at 40 kV and 15 mA. The crystal size was calculated from the XRD peak corresponding to the (2 0 0) plane using Scherrer's equation (with a shape factor K = 0.9). Peak separation was performed according to a previous report 12 . In the transmission mode, the XRD diagrams was recorded on a Fujifilm imaging plate (2540 × 2540 pixels, 50 × 50 μm 2 ) at room temperature using a Rigaku MicroMax-007 HF system operating at 40 kV and 30 mA with Cu Kα radiation (λ = 0.15418 nm). The samples were set parallel to the X-ray beam, and the distance between the sample and imaging plate was calibrated using NaF. The recorded diagram was read using a Rigaku RAXIA-Di system and converted to a one-dimensional 2θ-intensity and azimuthal profiles using a Rigaku 2DP software. The degree of orientation (DO) was calculated from the azimuthal profile of the (2 0 0) reflection using the following equation 17 : where FWHM is the full width at half maximum.  .
The true densities were measured using a BELPycno helium pycnometer according to a previously reported method 34

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. SIDaicho.docx