Introduction

Highly fibrillated cellulose is attracting attention as a sustainable “cellulose nanofiber (CNF)” with excellent mechanical and thermal properties (Nishiyama 2009; Solhi et al. 2023; Heise et al. 2022). CNFs are generally produced as aqueous dispersions from bleached wood pulps through wet disintegration. Controlling the dehydration of CNF dispersions results in various bulk structures, including aerogels with a sparse CNF network and films with a dense CNF network. These CNF aggregates exhibit functionalities depending on the network structure, such as heat insulating and gas barrier properties (Fukuzumi et al. 2009; Kobayashi et al. 2014).

The material properties of networked CNF structures correlate with the crystallinity of CNFs (Iwamoto et al. 2007; Uetani et al. 2015; Ishioka et al. 2023). The crystallinity of cellulose is significantly decreased when microfibrils tightly bundled in a pulp disperse as CNFs (Daicho et al. 2018). However, the crystallinity is recovered when CNFs are densely assembled through the dehydration of the dispersion (Daicho et al. 2021). This recovery of the crystallinity correlates with inter-CNF interactions, and the crystallite widths of CNFs are significantly enlarged through assembly with strong inter-CNF interactions. In this process, multiple CNFs, or cellulose crystallites, are assumed to partially fuse into single crystalline domains (Daicho et al. 2021). This is probably the same phenomenon as the “cocrystallization” or “twinning” of crystallites proposed by Newman et al. (2004, 2013). The mechanism of this crystallite fusion using CNFs as the structural unit is, however, not yet well understood, and neither is when or how it occurs during the dehydration of CNF dispersions.

Herein, we report changes in the crystallinity of CNFs when transitioning from a dilute dispersion to a dense aggregate by condensation at rest. The CNF used in this study was TEMPO-oxidized, and some of the surface C6 hydroxy groups were converted to sodium carboxylates (Saito et al. 2006). CNF samples with different solid contents, sampled in the process of condensation, were subjected to supercritical drying, followed by wide-angle X-ray diffractometry (WAXD) and solid-state cross-polarization magic angle spinning (CPMAS) 13C nuclear magnetic resonance (NMR) spectroscopy analyses. A portion of the CNF samples, without drying, were also subjected to small-angle X-ray scattering (SAXS) and viscoelasticity analyses to trace the networked CNF structure in water. Finally, we proposed a mechanism of the fusion phenomenon.

Materials and methods

Materials

A TEMPO-oxidized pulp in a wet state was kindly provided by DKS Co. Ltd. and used as the starting material. The carboxylate content of the pulp was determined to be 1.78 mmol g−1 by conductivity titration. Ethanol (99.5%) and distilled water were purchased from Fujifilm Wako Pure Chemical Co., Ltd. and used as received.

CNF dispersion

The TEMPO-oxidized pulp was suspended in distilled water at approximately 0.5% w/w and disintegrated into CNFs by passing it through a high-pressure water jet system (HJP-25005X, Sugino Machine Limited, Japan) at 150 MPa twice. The resulting CNF dispersion was passed through a nylon mesh filter with a mesh size of 5 μm to remove an unfibrillated fraction. A part of the dispersion was diluted to 0.1–0.4% w/w by adding distilled water.

Condensation

A 0.5% CNF dispersion (15 g) was poured into a polystyrene dish (44 × 44 × 15 mm) and condensed at rest in a thermohygrostat at 40 °C and 80% relative humidity (RH) until the solid content reached a plateau at approximately 85%.

Supercritical drying

A portion of the starting pulp and the condensed CNF samples were soaked in ethanol (200 mL) and gently shaken for two days using a rotary shaker at 50 rpm. The ethanol bath was replaced with fresh ethanol four times a day. The obtained samples were placed in the chamber of a supercritical dryer (Sanyu-Gijutsu, SYGLCP-8, Japan) under a liquid CO2 flow (4 mL min−1) at 15 °C for 9 h. After the flow of liquid CO2 was stopped, the chamber temperature was increased to 40 °C to reach the supercritical phase. After maintaining the supercritical phase for 1 h, the chamber was gradually depressurized at 40 °C for 1 h. The resulting supercritical-dried samples were conditioned at 23 °C and 50% RH for more than 1 day before analysis. The moisture contents of the supercritical-dried samples were approximately 5%.

WAXD

The conditioned samples were pressed at approximately 500 MPa for 1 min using a pelletizing device for WAXD. WAXD measurements were performed using a Rigaku Mini Flex diffractometer with a Ni-filtered Cu Kα beam (λ = 0.1542 nm) at 40 kV and 15 mA in the diffraction angle 2θ range of 3°–45° in reflection mode. The diffraction peaks corresponding to the (2 0 0) and the sum of the (1 1 0) and (1 −1 0) planes were deconvoluted with pseudo-Voigt function after subtracting a linear background from the profile, according to a previous report (Daicho et al. 2018) (see Fig. S1 for the peak separation of the WAXD profiles). Note that amorphous halo scattering is negligibly small in the WAXD reflection profiles of pelletized CNF samples, and the linear background includes other scattering contributions from environmental sources such as air molecules and moisture (Daicho et al. 2018). The crystal size perpendicular to the (2 0 0) plane was calculated using Scherrer’s equation with a shape factor of K = 0.9. The peak broadening due to the optical system of the instrument was corrected using an annealed plate of silicon monocrystal powder.

CP-MAS 13C NMR spectroscopy

The conditioned samples were packed into a ZrO2 rotor and analyzed by CP/MAS 13C NMR spectroscopy. CP/MAS 13C NMR measurements were performed using a JEOL JNM-ECAII 500 spectrometer equipped with a 3.2 mm HXMAS probe and rotor operating at 125.77 MHz for 13C. The samples were spun at 15,000 Hz and 298 K. The 90° proton decoupler pulse width, contact time, and relaxation delay were set to 2.5 µs, 2 ms, and 5 s, respectively (Daicho et al. 2020). Adamantane was used as an external standard for calibration of the chemical shift. The signals in the regions of 92–87 and 87–80 ppm, corresponding to the crystalline and noncrystalline C4 carbon atoms of cellulose, respectively, were separated, and the integral ratio of the C4 crystalline signal to all C4 signals was calculated as the crystallinity index (CI) value.

Recovery rate of crystallinity

The recovery rate of the crystal size and CI of a condensed sample was defined as follows:

$$\text{R}\text{e}\text{c}\text{o}\text{v}\text{e}\text{r}\text{y}\, \text{r}\text{a}\text{t}\text{e} \left({\%}\right)=\frac{{X}_{\text{c}\text{o}\text{n}\text{d}}-{X}_{\text{d}\text{i}\text{s}}}{{X}_{\text{p}\text{u}\text{l}\text{p}}-{X}_{\text{d}\text{i}\text{s}}}\times 100$$
(1)

where Xcond, Xdis, and Xpulp are the crystal sizes or CI values of the condensed sample, the 0.5% CNF dispersion, and the starting pulp, respectively. The crystal size and CI value of the pulp were 3.0 nm and 41%, respectively.

Scanning electron microscopy (SEM)

SEM images of the supercritical-dried samples were captured with a Hitachi S-4800 field-emission microscope at 1 kV. The samples were pretreated with a Meiwafosis Neo osmium coater at 2.5 mA for 10 s.

SAXS

SAXS with synchrotron radiation was carried out on dilute dispersions (0.1–0.5%) at the BL40B2 beamline of SPring-8 (Hyogo, Japan). The dispersions were set into a quartz-windowed metallic folder and mounted on a goniometer head. The X-ray (λ = 1.0 Å) irradiation was maintained for 100 s, and the diffraction pattern was recorded using a photon-counting pixel detector (PILATUS3 S 2 M, Dectris, Switzerland). The distance between the sample and the imaging plate (3193 mm) was calibrated using silver behenate powders (d = 5.838 nm) (Stephens et al. 2012).

SAXS measurements of the condensed samples (0.6–1.0%) were performed using a Rigaku NANOPIX SAXS system with monochromatized and collimated Cu Kα radiation (λ = 0.1542 nm) at 40 kV and 30 mA and a camera length of 920 mm. The condensed samples were set into a quartz-windowed metallic folder and mounted on a goniometer head. The 2D SAXS pattern was recorded using a hybrid pixel 2-dimensional detector (HyPix-3000, Rigaku, Japan) with an exposure time of 30 min. The camera lengths were calibrated using silver behenate. The recorded patterns were converted to a 1D scattering angle q-intensity profile using Rigaku 2DP software. The cross-sectional radii of gyration, Rc, for CNFs in the dispersions and condensed samples were calculated from the profiles in the q range of 0.50–0.94 nm-1 using the following equation (Glatter et al. 1982):

$${ln}\left(I\left(q\right)\cdot q\right)={ln}\left(I\left(0\right)\cdot q\right)-\frac{1}{2}{{R}_{c}}^{2}\cdot {q}^{2}$$
(2)

Dynamic viscoelasticity

Dynamic viscoelasticity measurements were conducted using an MCR 302 (Anton Paar GmbH, Graz, Austria) at 25 °C. A cone-plate geometry (plate diameter 50 mm, cone angle 2°) and a parallel-plate geometry (plate diameter 25 mm) were used for the dispersions (< 0.5%) and condensed samples (> 0.5%), respectively. Strain sweep measurements were carried out in advance at strains γ = 0.01–100% and a frequency ω = 10 rad s−1 to estimate the linear viscoelastic region, and frequency sweep measurements were conducted at angular frequencies ω = 0.1–100 rad s−1 within the linear viscoelastic region (γ = 1%).

Results and discussion

Condensation

The CNF dispersion was slowly condensed at rest in a thermohygrostat set to 40 ℃ and 80% RH until the desired solid contents were attained. Figure 1a shows the appearances of the 0.5% dispersion and two condensed samples at the intermediate (1%) and final (85%) stages of condensation. The starting dispersion had fluidity. As the solid content increased to approximately 1%, the CNFs lost fluidity and formed a self-standing gel. Finally, the condensed sample formed a film with a solid content of 85%.

Fig. 1
figure 1

a Appearances of the CNF dispersion (0.5%) and two condensed samples at the intermediate (1%) and final (85%) stages of condensation. b A supercritical-dried sample prepared from condensed CNF samples. c Cross-sectional SEM images of the supercritical-dried sample prepared at a solid content of 0.5%

Full size image

A series of condensed CNF samples was then subjected to solvent exchange with ethanol, followed by supercritical drying (Fig. 1b). Supercritical drying is regarded as the best method for suppressing the dry agglomeration of dispersed CNFs in wet samples (Kobayashi et al. 2014). The wet CNF samples showed no obvious shrinkage in the supercritical drying process, and the resulting dried products or “aerogels” appeared to maintain their original volumes in the wet state. The SEM analysis of the sample supercritical-dried at a 0.5% concentration showed a network structure of well-individualized CNFs (Fig. 1c). Hereafter, the dried samples are referred to by their solid contents in the wet state.

Crystallinity

The crystallinity of the dried samples was analyzed by WAXD and CP/MAS 13C NMR spectroscopy. In the WAXD analysis, the crystal size in the direction perpendicular to the (2 0 0) plane was evaluated as a measure for the regularity of molecular sheet stacking of cellulose. In NMR analysis, the CI value was calculated as a ratio of the constituent carbon atoms having the crystalline conformation, which reflects the degree of hydrogen bonding within molecular sheets. These two values represent different scales of information for the crystallinity of cellulose.

The samples were deformable aerogels and were pelletized by compression for WAXD (see Fig. S2 for the appearance of the pellets). The lack of change in the crystallinity after this compression process was confirmed in advance by NMR (Fig. S3).

Figure 2a shows the reflection WAXD profiles of the samples. As the solid content increased, the diffraction peaks became sharper. We assumed that the (0 1 2) and (1 0 2) diffractions at approximately 20°–21° were negligibly small in the reflection profiles, considering that the CNFs in the pelletized samples were sufficiently oriented to the in-plane direction of the pellets (see Fig. S4 for transmission WAXD profiles of the pellets set parallel to the incident beam), so that no (0 0 4) diffraction at approximately 35° was observed in the reflection profiles. According to a simulation study by French and Santiago Cintrón (2013), cellulose I crystals with the preferred orientation show no (0 1 2) and (1 0 2) diffractions in their reflection profile, which is also accompanied by a significant reduction in the (0 0 4) diffraction.

Fig. 2
figure 2

a WAXD profiles, b (2 0 0) crystal sizes, c NMR spectra, and d CI values of a series of samples. e Recovery rates of the crystal size and CI.

Full size image

The (2 0 0) diffraction was then separated from these WAXD profiles by peak deconvolution (Fig. S1), and their crystal sizes were calculated using Scherrer’s equation (Fig. 2b). The crystal size increased from approximately 1.8–2.4 nm as the solid content increased from 0.5 to 85%. Interestingly, this crystal-size enlargement showed two steps: a rapid increase in the range of 0.5–1%, followed by a gradual increase in the range of 1–85%. The slight shift of the (2 0 0) peak position from 21.9° to 22.3° with the solid content can be explained by this crystal-size enlargement (Wada et al. 1997; Huang et al. 2018). The shape and position of the combined (1 1 0) and (1 −1 0) peak for the sample with a solid content of 85% were also varied when compared to those for the starting pulp, suggesting that the configuration of CNFs assembled by dehydration is different from that of microfibrils assembled in a pulp fiber.

Figure 2c shows the NMR spectra of the samples. In the C4 region, the crystalline signal at 87–92 ppm increased little by little relative to the noncrystalline signal at 80–87 ppm with increasing solid content. A similar change was also observed in the C6 region; the crystalline tg signal at approximately 65 ppm gradually increased relative to the noncrystalline gt and gg signals at approximately 60–63 ppm. The CI values were calculated as the integral ratio of the crystalline C4 signal to all C4 signals in the NMR spectra. The CI value hardly changed in the range of 0.5–30% but gradually increased from approximately 23–28% in the range of 30–85% (Fig. 2d).

The recovery rates for the crystal size and CI value at each solid content were calculated based on the decrement of these values upon disintegrating the starting pulp into CNFs with a 0.5% concentration (see Methods section for details). The crystal size and CI value were recovered by 41% and 28%, respectively, throughout the whole range of 0.5–85% (Fig. 2e). The reason for the higher recovery rate of crystal size is discussed later.

CNF assembly in water

The fact that the obvious enlargement in crystal size occurred for the supercritical-dried samples with low solid contents below 1% implies that the adjacent CNFs dispersed in water started to assemble at such a low solid content. To verify this hypothesis, the cross-sectional radii of gyration, Rc, of the CNFs in the wet samples were analyzed by SAXS measurements.

Figure 3a shows the 1D SAXS profiles of the wet CNF samples with solid contents of 0.1–1% (see Fig. S5a for the SAXS patterns of all the samples). The shape of the profile changed with the solid content. The Rc of the CNFs at each solid content was calculated from the profiles in the q range of 0.50–0.94 nm−1 (Fig. 3b). The resulting Rc value indeed increased with increasing solid content. This result indicates that the CNFs assembled in water, and the inter-CNF contact points increased upon condensation up to 1%. Assuming a cylindrical model, the cross-sectional diameter was estimated to be approximately 3.8‒4.1 nm using the Rc values and scaled to a similar extent with the crystal size shown in Fig. 2b (~ 0.3 nm). Note that the CNFs with these low solid contents were randomly oriented on average (see Figure S5b for 2D SAXS patterns).

Fig. 3
figure 3

a 1D SAXS profiles of the wet CNF samples and b the radii of gyration, Rc, calculated from the 1D profiles

Full size image

Network growth

The dynamic viscoelasticity of the wet samples with low solid contents was measured to investigate the network formation of CNFs in water. Figure 4a shows the frequency dependency of storage modulus, G′, and loss modulus, G″, of the wet samples with solid contents of 0.13–0.47%. At 0.13%, the G′ values were lower than the G″ values in the frequency range of 0.1–2 rad s−1; the sample was in a sol state and had fluidity. At 0.25%, the G′ values were nearly equal to G″ in the frequency range of 0.1–20 rad s−1, suggesting the network formation of dispersed CNFs. With a further increase in solid content to ~ 0.5% or above, the G′ values became clearly higher than the G″ and were almost constant over the entire frequency range (Fig. 4b). These results show that the wet CNF samples with solid contents of ~ 0.5% or above behaved as elastic bodies.

The G′ values at 1 rad s−1 in the plateau region were taken as the plateau moduli, G′p, and were plotted against the solid content in a double logarithmic plot (Fig. 4c). There were two clearly distinguished regions with different slopes in the plot. The following relationship between the G′p values and solid contents has been proposed for semiflexible fiber networks (Tatsumi et al. 2002; Saito et al. 2011):

$${{G}^{{\prime }}}_{p}=A{c}^{\alpha }$$
(3)

where A is a constant that reflects individual fiber properties, such as the aspect ratio and elastic modulus, and the power, α, is related to the network structure and corresponds to the slope of the double logarithmic plots. The α value for the wet CNF samples in this study changed from 3.5 to 1.3 at the threshold of approximately 1%; at this threshold, the contribution of CNFs to the network growth changed as the solid content increased. The threshold value was consistent with the solid content at which the trend in the crystal-size enlargement clearly changed (see Fig. 2b), indicating a correlation between the crystal-size enlargement and networked CNF structure in water.

Figure 4d shows the correlation lengths, L, in the polymer networks, or the segment lengths between entanglements, estimated using the G′p values (Ferry 1961). These L values correspond to the average distances between the inter-CNF contact points in this study and were estimated by the following formula:

$${{L}} = \rho{{RT/}}{{G}^{{\prime }}}_{p}$$
(4)

where ρ, R, and T are the density or CNF weight per volume of the wet sample, gas constant, and temperature, respectively. The L value drastically decreased with increasing solid content up to 1% but became almost constant at a threshold of approximately 1%.

Fig. 4
figure 4

a G′ and G″ values as a function of frequency for the CNF dispersions with solid contents of 0.13–0.47%, and b G′ values for the condensed samples with solid contents of 0.47–3.3%. c Relationship between the G′p values and solid contents of the condensed samples. d Change in the correlation length, L, estimated from the G′p values and solid contents

Full size image

Mechanism

Figure 5 illustrates our interpretation of the mechanism of network formation or growth by CNF assembly. Taking into account the results shown in Figs. 3 and 4, we assumed the following mechanism (Fig. 5a): (1) the inter-CNF contact points significantly increase until reaching a solid content of 1%, at which point the basic skeleton of the CNF network in a wet sample is set, and (2) at a solid content higher than 1%, the inter-CNF contact areas gradually expand in a zip-up manner with the solid content. This mechanism corresponds to the two-step enlargement of the crystal size shown in Fig. 1b.

Another mechanism of CNF assembly, shown in Fig. 5b, was assumed to explain the preferential recovery of the crystal size at low solid contents in.

Fig. 5
figure 5

a Mechanism of network formation or growth by CNF assembly during condensation, and b the inter-CNF interactions in water

Full size image

comparison with the CI (see Fig. 2e). In this assumption, we used an 18-chain model with molecular sheet stacking of 2/3/4/4/3/2 as a single CNF. At low solid contents up to 1%, the CNFs are assumed to preferentially assemble by hydrophobic interactions, where the (2 0 0) surfaces of the CNFs are stacked on top of each other. This CNF assembly explains the crystal size enlargement or “crystallite fusion” observed at low solid contents. At higher than 1%, we assumed that gradual assembly between the hydrophilic surfaces is induced by condensation, exceeding the repulsive force of electric double layers. This latter step of assembly explains the gradual increases in the CI as well as the crystal size. Note that some of the hydroxy groups on the hydrophilic (1 1 0) and (1 −1 0) surfaces were oxidized to carboxy groups in this study. The CI recovery of such surface-oxidized CNFs was also observed in our previous study (Daicho et al. 2021). The most plausible explanation for the CI recovery was that the residual surface C6 hydroxy groups took the crystalline tg conformation from noncrystalline gt and gg states. However, the contribution of this conformational change to the CI recovery was small considering the residual ratio of surface C6 hydroxy groups. Our tentative interpretation is thus that the CI recovery is a result from conformational changes of not only the residual surface C6 hydroxy groups but also the interior C6 hydroxy groups within each CNF.

Conclusions

The recovery of the crystallinity of CNFs transitioning from a dilute dispersion to a dense aggregate by condensation was monitored in this study. The condensed CNF samples were subjected to supercritical drying, followed by WAXD and NMR analyses. In the WAXD analysis, a two-step enlargement in the (2 0 0) crystal size was observed; the first step was a rapid increase in the solid content range of 0–1%, followed by a gradual increase in the range of 1–85%. The CI value estimated by the NMR analysis hardly changed in the range of 0.5–30% but gradually increased in the range of 30–85%. The wet CNF samples were also subjected to SAXS and viscoelasticity analyses without supercritical drying. These analyses of wet samples indicated that the inter-CNF contact points significantly increased until reaching a solid content of 1%, and then at solid contents higher than 1%, the contact areas of each point gradually expanded. Taking into account the preferential recovery of crystal size at low solid contents, we further assumed a mechanism of crystallite fusion; at low solid contents, the hydrophobic (2 0 0) surfaces of CNFs are preferentially stacked on top of each other, followed by gradual assembly between not only the hydrophobic (2 0 0) but also the hydrophilic (1 1 0)/(1 −1 0) surfaces at high solid contents. The findings in this study will contribute to diverse material designs involving CNF assembly.