Surfactant induced gelation of TEMPO-oxidized cellulose nanofibril dispersions probed using small angle neutron scattering

As the most naturally abundant polymer,¹ cellulose has roused the interest of scientists in order to design eco-compatible materials.² In plants, cellulose is found as tightly packed bundles of fibrils. Strong hydrogen bonds between fibrils maintain a hierarchical organization preventing their dispersion in aqueous dispersions.³ The hydrophilicity of cellulose fibrils can be enhanced by chemical modification, such as TEMPO-mediated oxidation.^3,4 This method consists of the selective oxidation of the glucosyl C6 primary hydroxyl group, using NaOCl and mediated by NaBr and (2,2,6,6-tetramethylpiperidin-1-yl)oxyl (TEMPO). Negatively charged carboxyl groups are hence introduced on the surface of cellulose fibrils, and water-dispersible oxidized cellulose nanofibrils (OCNFs) are obtained. OCNFs are characterized by a cross-sectional diameter of typically 5–10 nm and a length ranging from 100 nm to a few μm and exhibit high surface charges (ζ-potential < −30 mV).⁴ 

OCNF can form hydrogels in aqueous suspensions, with rheological properties highly dependent on the concentration of nanofibrils,^5–7 the pH of the suspension,⁸ the temperature,⁹ the nature of the solvent¹⁰ or the presence of additives such as salt,^7,8,11 polymers^12,13 or surfactants.^5,14 For example, addition of 0.1M NaCl triggers gelation of OCNFs.¹⁵ It was shown both theoretically¹¹ and experimentally¹⁵ that addition of counterions to a suspension results in a charge screening of OCNF, reducing fibril–fibril electrostatic repulsion, which in turn encourages their aggregation into a percolated network. Typical rheological fingerprints of OCNF-based hydrogels are the presence of a yield stress (i.e., the stress above which the materials flow), a pronounced shear-thinning behavior, and frequency-independent elastic and viscous moduli (G′ and G″, respectively) in oscillatory shear experiments performed within the linear viscoelastic regime. Addition of polymers such as carboxymethyl cellulose was also found to induce gelation, thanks to depletion–flocculation mechanisms.¹² Similar rheological behaviors are observed for cellulose nanocrystals (CNCs), negatively charged rod-like nanoparticles that are generally smaller in length when compared to OCNFs.^16,17 The behavior of these OCNF hydrogels, along with the biocompatibility of cellulose, makes them promising materials for cosmetics,¹⁸ food¹⁹ or healthcare applications²⁰ (among others), for example as rheology modifiers,¹⁹ scaffolds for tissue engineering,²¹ or drug-delivery carriers.^22,23

Among the type of additives that can be added to those hydrogels, surfactants play an important role in a large range of industrial applications. Understanding the effect of surfactant addition to OCNF suspensions and how to exploit such understanding to induce formation of hydrogels is key to designing novel soft materials. We previously showed that the addition to a 2 wt. % OCNF suspension of a negatively charged surfactant, sodium lauroyl sarcosinate (SLS), commonly used in shampoo formulation thanks to its high capacity to foam, led to the gelation of the suspension (for concentrations of surfactants ranging from 1 to 5 wt. %, ∼30–180 mM).⁵ Quennouz et al. observed a similar behavior using 8 wt. % (∼175 mM) of sodium lauryl ether sulfate (SLES) to a 0.6 wt. % OCNF suspension.¹⁴ Interestingly, however, sodium dodecyl sulfate (SDS) at 4 wt. % (∼145 mM) led to a loss of sample stability. Moreover, the authors showed gelation could be obtained by adding nonionic surfactant (Triton X-100 and Pluronic F68) at high concentration (8 wt. %, ∼140 and 10 mM respectively), while even small quantities of dodecyl trimethyl ammonium bromide (DTAB) resulted in an unstable sample. The authors claimed that gelation of OCNF/surfactant mixtures occurred due to the affinity between the cellulose nanofibril surface and surfactant headgroups. Work on cationic surfactants such as C_nTAB with OCNF or CNC showed indeed a clear affinity between the positively charged headgroup and the negatively charged surface of the nanocellulose.^24,25 Regarding nonionic or anionic surfactants, other hypotheses such as depletion–flocculation, where surfactant micelles act as depletants, have also been suggested.⁵ Recently, we studied the effect of the shape of ionic surfactant micelles on the rheological properties of OCNF suspensions in saline conditions.²⁶ We compared spherical, rod-like, and worm-like micelles made by mixing cocamidopropyl betaine (CapB) and SLS; cocamidopropylamine oxide (CapOx), and SDS or CapB and SDS, respectively. Addition of salt is a requirement to form worm-like micelles when CapB and SDS are used. We showed that in those saline conditions, where a stiff gel is already formed by OCNF aggregation, addition of micelles leads to an increase of tan δ = G″/G′, acting as a plasticizer. Moreover, we saw that the addition of worm-like micelles, acting as a secondary entangled network, interferes with the OCNF percolated network.

To shed some light on the gelation mechanisms between OCNF and surfactants without salt, we compare in this paper the rheological and structural properties of OCNF mixtures with a nonionic (hexaethylene glycol mono-n-dodecyl ether, C₁₂EO₆), anionic (SDS), zwitterionic (CapB), or cationic (DTAB) surfactant. The four surfactants were chosen as they are all composed of the same C₁₂H₂₅ hydrophobic tail, with only the nature of the polar head varying. To probe the structural properties of the suspensions, we used small angle neutron scattering (SANS). SANS is a powerful tool to probe structural changes in these multicomponent systems. Indeed, the technique is noninvasive and in situ. It gives information not only about the shape and size of the particles in suspension but also on the interactions between particles. Moreover, using neutrons, contrast variation techniques allow focusing either on the scattering from the cellulose nanofibrils or on the surfactant aggregates in the mixtures.

OCNFs were synthesized via the TEMPO/NaOCl/NaBr oxidation route followed by high-pressure homogenization, as described previously,^3,4 using wood pulp as cellulose source, resulting in an 8 wt. % solid paste in water. We previously measured the degree of oxidation by conductometric titration, found to be 25%.^21,27,28 After synthesis, OCNFs were purified in order to remove salts and preservatives using several dialysis steps, following a protocol previously described.^10,15 After purification, the OCNF solution was freeze-dried. Hexaethylene glycol mono-n-dodecyl ether (C₁₂EO₆), sodium dodecyl sulfate (SDS), and dodecyltrimethylammonium bromide (DTAB) were purchased from Sigma-Aldrich and used as received. Commercial grade cocamidopropyl betaine (CapB, Crodateric CAB 30-LQ-(MH), 30% aqueous solution, batch No 1189504) was provided by Croda. CapB was then freeze-dried before use. The critical micelle concentration (CMC) at room temperature of each surfactant can be found in the literature: 0.085 mM for C₁₂EO₆,²⁹ 8.20 mM for SDS,³⁰ 14.6–16.0 mM for DTAB,³⁰ and 0.28 mM for CapB.³¹ A stock suspension of OCNF at 2 wt. % and stock suspensions of surfactant at 200 mM in de-ionized water (DI, 18.2 MΩ.cm) were prepared and homogenized with a probe ultrasonicator (Vibracell VC300) using a tapered titanium microprobe (6.5 mm diameter) at an intensity of 10 W cm⁻² (applied power determined by heat balance), alternating 1 s sonication with a 1 s standby for 2 min. Then, samples were prepared by mixtures of stock suspensions and DI water to reach 1 wt. % OCNF and 20, 40, 60, or 80 mM surfactant (SDS and CapB). We note that for DTAB, addition of 20 mM of surfactant induces a phase separation due to fibril–fibril aggregation, hence concentrations of 1, 5, and 10 mM, below the CMC of this surfactant (∼15 mM³⁰), where the sample is homogeneous, were chosen. Finally, for control, OCNF (1 wt. %)-C₁₂EO₆ suspensions were studied at 1, 5, 10 mM (matching the concentrations used for the DTAB system) and 50 mM (to compare with SDS and CapB). Note that no extra salt was added to those suspensions. Hence, the change of ionic strength is solely dependent on the concentration in ionic surfactant (up to 10 mM for DTAB and 80 mM for SDS/CapB).

For neutron analysis, deuterated version of SDS and C₁₂EO₆ (labeled d-SDS and d-C₁₂EO₆ in the following; while hydrogenated version of the surfactants will be labeled h-SDS and h-C₁₂EO₆), both exhibiting a deuterated carbon tail, were supplied by the ISIS, Deuteration Facility (Rutherford Appleton Laboratory, Didcot, UK). Suspensions of OCNF (1 wt. %) and surfactant mixtures for neutron measurements were prepared following the protocol described above, using D₂O, H₂O, or mixtures depending on the contrast studied. This allowed performing some contrast matching measurements, i.e., matching the scattering length density of one of the components (OCNF or surfactant micelles) with the solvent to focus solely on the other component. For SDS, four contrasts were studied: h-SDS in D₂O (where both OCNF and SDS are visible); d-SDS in D₂O (contrast matching of SDS with the solvent to probe only OCNF in the mixtures); d-SDS in 50%D₂O (contrast matching of OCNF with the solvent to probe only the surfactant micelles); and finally d-SDS in H₂O (a check contrast, where both OCNF and micelles are visible). For C₁₂EO₆, due to the limited synergistic effect between OCNF and nonionic micelles, only two contrasts were studied: h-C₁₂EO₆ in D₂O (to see both OCNF and surfactant) and d-C₁₂EO₆ in D₂O (to focus on the OCNF). For CapB, only a hydrogenated version of the surfactant was accessible, hence measurements were primarily done in D₂O. Nonetheless, to further discriminate OCNF contribution from that of micelles, a contrast variation study was made for CapB 40 mM with the following percentage of D₂O in the solvent: 100, 70, 50 (OCNF contrast matching point), 40, 20 (theoretical contrast matching point of CapB), and 0%. Finally, for DTAB, only one sample at 5 mM was studied using neutrons in D₂O. For the DTAB sample, no contrast matching measurements were performed as the DTAB concentration is under the CMC. Thus, a negligible scattering contribution to the overall scattering pattern is expected from the DTAB. Surfactants (h-SDS, h-C₁₂EO₆ and CapB) were also measured in D₂O in the absence of OCNF at the same molar concentrations as the mixtures, for comparison.

Rheological properties of OCNF–surfactants hydrogels and suspensions were measured at room temperature using a stress-controlled rheometer (Discovery HR-3, TA Instruments) equipped with a sandblasted plate–plate geometry (40 or 12 mm in diameter depending on the measurements) with a ∼1 mm gap. The edge of the geometry was covered with a thin layer of mineral oil to prevent evaporation of water. First, amplitude strain sweep measurements of the storage and loss moduli G′ and G″ were conducted at ω = 10 rad s⁻¹ to assess the amplitude of the linear viscoelastic region. Then, a strain amplitude of 0.1% was chosen for frequency sweep measurements to characterize the possible gel-like behavior of the suspensions. Finally, steady flow measurements were performed to study the viscosity response of the sample to shearing, with a shear rate $γ̇$ ranging from 10⁻² to 10² s⁻¹.

Structural properties of the hydrogels were studied using SANS at the Larmor beamline at the ISIS Pulsed Neutron and Muon Source at the Rutherford Appleton Laboratory (RAL, Didcot, UK).³² A typical q-range of 4.5 × 10⁻³ ≤ q ≤ 0.7 Å⁻¹, with q being the scattering vector, was chosen. Measurements associated with the C₁₂EO₆ system were carried out at the Sans2D beamline at RAL.³³ In all cases, the samples were poured into 1 cm wide, 1 mm thick neutron cells and measured at room temperature. From the raw data, background subtraction and normalization of the scattering intensity were conducted using the Mantid software to give the scattered intensity I(q) in absolute scaling (cm⁻¹).³⁴ 

SANS data were fitted using a homemade Fortran software, via a least-squared fitting procedure. For isotropic suspensions, the normalized scattered intensity I(q) can be described as follows:³⁵ 

with $Iscatterersq=APqS(q)$ being the contribution from the scattering objects, A = nΔρ²V² = φ_pΔρ²V the scaling factor depending on the number of scatterers per unit volume n in cm⁻³ or the volume fraction of particles in the sample φ_p, the contrast between the scatterers and the solvent Δρ = ρ_s−ρ₀ (ρ_s and ρ₀ the scattering length densities of the scatterers and solvents respectively in cm⁻²) and V the volume of the scatterers (in cm³), P(q) the normalized form factor of the scatterers [P(q = 0) = 1] and S(q) the structure factor of the scatterers. B is a constant background parameter in cm⁻¹.

For OCNF, previous studies have shown they could be described as rod-like particles, either with a parallelepiped or elliptical cross section.^15,36 We selected the elliptical cross-section model to fit the data. We previously demonstrated that the intensity associated with OCNF, $Iscatterersq=IOCNF(q)$⁠, can be written as follows:¹⁵ 

with A_OCNF the scaling factor associated with OCNF; P_rod(q, R_max, ε, L) the form factor for rods with a length L ≥ 100 nm (not measurable in the given q-range and later fixed at 160 nm as determined by transmission electron microscopy in our previous study¹⁵) and a cross-section described by the larger radius R_max (in nm) and ellipticity ε < 1; and S_PRISM(q, ν_RPA, R_Cq) the structure factor for rods using the PRISM model,³⁷ with ν_RPA representing the strength of the interactions between rod-like particles (ν_RPA < 0 for attractive interactions, ν_RPA > 0 for repulsive interactions) and R_Cq the radius of excluded volume along each point of the rod. The complete description of this model can be found in Ref. 15 and is not repeated here.

Similarly, the intensity associated with micelles of surfactants, $Iscatterersq=Imicelles(q)$⁠, can be written as

with A_micelles the scaling factor of the micelles and P_micelles(q) = P_spheres(q, R, σ) the form factor of polydisperse spheres of radius R (in nm) and polydispersity index σ (in %) valid for micelles of SDS or CapB. For micelles of C₁₂EO₆, an elliptical model is preferred with P_micelles(q) = P_ellipses(q, R, ε_p) the form factor of ellipsoids of dimensions (R, R, ε_pR) with radius R and ε_p the ellipticity parameter (ε_p > 1 for prolate-type ellipsoids). The complete model description can be found in Ref. 38. Structure factors used for sphere and ellipsoid-like micelles depend on the type of interactions between micelles (steric or electrostatic).

For the nonionic C₁₂EO₆ micelles and the zwitterionic CapB micelles, a hard-sphere model is enough to describe the interactions between micelles. The hard-sphere potential of interaction can be written as

with r the distance from the particle center and R_HS the hard-sphere radius of interaction (R_HS> R). The associated structure factor was calculated by Percus and Yevick.^39,S_micelles(q) = S_{Percus-Yevick}(q, R_HS, φ_HS), it depends not only on R_HS the hard-sphere radius of interactions but also φ_HS the volume fraction of interacting micelles.³⁸ 

For the negatively charged SDS micelles, a Coulomb interaction has to be taken into account in the form of a Coulomb or Yukawa potential,⁴⁰ 

with r the distance from the particle center. Analytical versions of the structure factor were given by Hayter and Penfold via the mean spherical approximation⁴¹ or the Yukawa structure factor.⁴⁰ The Yukawa structure factor can be written as S_micelles = S_Yukawa(q, R, φ, U, 1/κ) with R the radius of the particles, φ the volume fraction of interacting particles, U the strength of the Coulomb interaction, and 1/κ the Debye–Hückel length characterizing the length of electrostatic interaction.⁴⁰ We note that Hayter and Penfold have shown previously that for dilute micellar solution, the mean spherical approximation structure factor leads to unphysical results. A similar issue is observed with the Yukawa structure factor. This problem is circumvented by defining an effective radius R_eff ≫ R for the interacting particles, which reaches a volume fraction of interaction φ_eff large enough to properly fit the data.⁴² This rescaled mean sphere approximation structure factor hence gives R_eff, φ_eff, and (1/κ)_eff. By denoting x = R/R_eff the rescaling factor, the real values of these dimensions are given by φ = φ_effx³ and (1/κ) = x(1/κ)_eff. Note that in the solution provided by Hayter and Penfold, the surface charge on the micelles is extracted using the dielectric permittivity of the solvent and the salt concentration of the suspension.⁴⁰ 

For mixtures of OCNF and micelles, we have simply summed both contributions, neglecting possible cross terms in the calculation of the scattered intensity. We have

This simplification is enough to describe the data and evidence whether interactions arising between OCNF and micelles are present.

We first studied the addition of C₁₂EO₆ to a 1 wt. % OCNF suspension. In our conditions, a 1 wt. % OCNF suspension (in the absence of surfactants) presents a fluid-like behavior at rest (i.e., in the limit of small amplitudes in a classical oscillatory experiment), while gelation is only observed for concentrations ≥2 wt. %.¹⁵ The concentration of gelation is highly dependent on the type of OCNF used and notably their overall dimensions.^5,7,14 C₁₂EO₆ was chosen as nonionic surfactant to see if addition of non-electrostatically interacting micelles can influence the rheological properties at the probed concentrations (1–50 mM, above the CMC of the surfactant of 0.085 mM²⁹), purely due to the excluded volume effect. This will help decorrelate further effect of the charge of micelles for the other systems.

Figure 1(a) presents the viscosity curves of suspensions made of OCNF at 1 wt. %, without or with different concentrations of C₁₂EO₆. The presence of micelles of C₁₂EO₆ has little influence on the viscosity of the suspension, neither acting as a thickening nor thinning agent. At all concentrations, suspensions exhibit a shear-thinning behavior already observed for OCNF-based suspensions and hydrogels.⁵ Moreover, the suspensions still present a fluid-like behavior (G″ > G′), as observed in the frequency and amplitude sweeps in Fig. S1 in the supplementary material, with an important angular dependency of both G′ and G″ in the frequency sweeps (Fig. S1.a) and a large linear viscoelastic region (Fig. S1.b). These results suggest that at those concentrations, the addition of micelles of C₁₂EO₆ has little influence on the overall rheological properties of the sample, which are instead governed by the cellulose nanofibrils.

Structural information can be obtained using small angle neutron scattering (SANS). Figure S2 in the supplementary material gives the SANS pattern of an OCNF suspension at 1 wt. %, the fit associated, and the parameters from the fitting procedure. The SANS pattern is characteristic of rod-like structures, with a broad signal at large angles from the elliptical cross section and a q⁻¹ behavior (seen as a −1 slope in log-log scale) at small angles. Interactions would induce a deviation from this q⁻¹ behavior (a less pronounced slope would be associated with repulsion between fibrils, a more pronounced one with attraction). OCNF can hence be fitted as fibrils of length fixed at 160 nm, in agreement with previous work¹⁵ (as it is not accessible in the probed q-range) with an elliptical cross section of maximum radius R_max = 5.0 ± 0.1 nm and ellipticity ε = 0.22 ± 0.01. This model corresponds well to previous TEM and cryo-TEM measurements of the nanofibrils.¹⁵ At this concentration, no interaction between fibrils is required to fit the data (ν_RPA = 0). The results are consistent with previous SAXS and SANS studies.^15,43

SANS patterns of C₁₂EO₆ micelles in D₂O are provided in Fig. S3. They can be fitted using a model of prolate ellipsoids of radius R and ellipticity ε_p > 1. The higher the concentration of surfactant, the more anisotropic the ellipsoids become while keeping R constant. Hence, in this case, the change of shape with concentration induces an increase of the volume of the micelle $V=43πεpR3$⁠, which goes from 96.7 nm³ at 1 mM to 260.0 nm³ at 50 mM. The highest concentration (50 mM) was even fitted using interactions between micelles via the Percus–Yevick model for hard spheres. The fit parameters are given in Table I. This behavior is consistent with the results of previous SANS studies of C₁₂EO₆ micelles in solution.⁴⁴ 

Figures 1(b) and 1(c) give the SANS patterns for mixtures of OCNF and C₁₂EO₆ using either a deuterated version of the surfactant (d-C₁₂EO₆) to focus on OCNF or a hydrogenated version of the surfactant (h-C₁₂EO₆) to study both OCNF and micelles. Using d-C₁₂EO₆ (partial deuteration as only the carbon tail is deuterated), we can fit the data using the signal of OCNF nanofibrils only for [d-C₁₂EO₆]≤10 mM [see Fig. 1(b), and Table II for the main parameters extracted from of the fits]. For 1 and 5 mM, the dimensions of the nanofibrils are fixed according to results from the SANS pattern of OCNF at 1 wt. %. At 5 mM, we need to add repulsive interactions between nanofibrils: with ν_RPA = 2.66 ± 0.01 the strength of the repulsion and R_Cq = R_max = 5.0 nm (fixed during the fitting) the excluded volume radius around the fibril cross section. For 10 mM, we have also to adjust slightly the nanofibril dimensions (R_max = 4.1 ± 0.1 nm and ε = 0.44 ± 0.02) while maintaining repulsion between nanofibrils (ν_RPA = 1.76 ± 0.01 and R_Cq = 8.6 ± 0.1 nm). We explain this apparent change in nanofibril cross section by the rise of an extra contribution from the d-C₁₂EO₆ micelles in the SANS pattern, neglected during fitting. Indeed, as the polar head is not deuterated, the micelles are not exactly contrast matched with the solvent, hence the micelles shell can still contribute to the SANS pattern, especially in the q-range associated with the nanofibrils cross section. This is easily seen at 50 mM, where several oscillations are observed at q ≥ 0.05 Å⁻¹. The signal cannot be fitted using the model of rod-like nanofibrils. This SANS pattern is probably the sum of nanofibrils and shell-contrasted micelles. Due to the complexity of this system, fitting was not attempted.

When OCNF and h-C₁₂EO₆ are mixed [Fig. 1(c)], the SANS patterns can be fitted well by simply summing the contributions of OCNF and C₁₂EO₆, without any change in either the micelle or the fibril form factors. The only adjustable parameters are the scaling factors of both OCNF and micelles contributions (A_OCNF and A_micelles respectively) and the fibril interaction parameters (ν_RPA and R_Cq), all of which can be found in Table II. We note that adding repulsion between nanofibrils has an effect only at the smallest angles, as can be seen in Fig. S4, which compares the fits obtained for OCNF+h-C₁₂EO₆ at 50 mM with and without repulsion between nanofibrils. Reasonable fits can already be obtained without repulsion for all concentrations, but the presence of repulsion when d-C₁₂EO₆ is used and smaller values for the least-square fitting parameter χ² (which estimates the deviation between the experimental scattering intensity and the fitted intensity in each data point, see S.I.) when repulsion is added (4.33 vs 4.99 for 50 mM) indicate that repulsive interactions between nanofibrils must be accounted for. The values of A_micelles and A_OCNF are not affected by the addition of a repulsive structure factor between nanofibrils. Furthermore, the values of A_micelles found in the mixtures are of the same order of magnitude as for the samples without OCNF at the same concentration, while A_OCNF strongly increases with the micelle concentration. A similar trend for A_OCNF is observed for the samples with d-C₁₂EO₆ (see Table II).

This behavior is unexpected as the scaling factor depends on the number of OCNFs per unit volume and the contrasts, both of which are expected to be unchanged by the addition of surfactants. For comparison, OCNF at 2 wt. % without surfactant was also measured in SANS (see Fig. S5 for the SANS patterns and fitting parameters), and the scaling factor was found to be A_OCNF = (1.5 ± 0.1) × 10⁻² cm⁻¹ at this nanofibril concentration (∼2.34 × A_{OCNF at 1 wt. %}). Results with OCNF 1 wt. % and 10 mM h-C₁₂EO₆ hence give the same order of magnitude for A_OCNF. Moreover, in both cases, OCNFs experience repulsion (ν_RPA = 1.62 ± 0.01 for the system OCNF 2 wt. %, >2 for OCNF + C₁₂EO₆ as shown in Table II). These results suggest that adding C₁₂EO₆ may result in concentrating the nanofibrils in the water phase due to an excluded volume effect. The change in scale would hence be an increase of the apparent crowding of OCNF, with adding 10 mM of C₁₂EO₆ being roughly equivalent to doubling the concentration of nanofibrils. A similar effect of suspension crowding was previously investigated with CNC in the presence of polymer. It was shown that adding polymer had the same effect on the alignment of CNC under shear as self-crowding induced by increasing the concentration, notably, on the value of the shear rate for the onset of particle alignment.⁴⁵ 

Nonetheless, we showed in a previous study that increasing the concentration of nanofibrils from 1 to 2 wt. % in aqueous solutions results in a liquid-to-gel phase transition observed in rheology, which we attributed to the emergence of repulsive electrostatic interactions between nanofibrils seen in small angle x-ray scattering.¹⁵ Similarly, SANS fitting of the suspension of OCNF at 1 wt. % alone does not require the addition of interactions between nanofibrils (ν_RPA = 0), while repulsive interactions (ν_RPA = 1.62 ± 0.01 and R_Cq = R_max = 5.0 nm) are needed at 2 wt. % (see Figs. S2 and S5). Yet, in mixtures of OCNF and C₁₂EO₆, we see an emergence of repulsive interactions between nanofibrils, consistent with the increase in nanofibril crowding by excluded volume effect, but no gel is observed in rheological studies. An interesting feature is the excluded radius around the nanofibrils cross section R_Cq. For nanofibrils alone (OCNF at 2 wt. %), R_Cq is fixed at R_max, the radius of the nanofibrils. Letting free this parameter only results in that case in its decrease, yet, by definition R_Cq ≥ R_max (the excluded volume radius cannot be smaller than the particle radius). Similarly, when micelles are “contrast matched” (OCNF 1 wt. % + d-C₁₂EO₆), we observe the same results. This indicates that, indeed, the presence of micelles induces a concentration of the nanofibrils that hence experience repulsion. Nonetheless, when micelles are “visible” (OCNF 1 wt. % + h-C₁₂EO₆), although repulsion between nanofibrils is present, we find R_Cq ≫ R_max, which suggests that micelles also act as a steric barrier between nanofibrils, disturbing the overall network and preventing gelation from occurring macroscopically. A crude geometrical calculation supports this theory. Defining x = R_Cq − R_max, one can note that for 5 mM, x = 5.3 nm ≈ 2R = 4.8 nm with R the radius of the micelles. The increase of x with micelle concentration could be related to the elongation of micelles.

Interestingly, previous work by Quennouz et al. showed that gelation of OCNF suspensions could be obtained by adding nonionic surfactants, in their case Triton-X-100 or Pluronic F68.¹⁴ They required a large amount of surfactant (1 wt. %) to see a mild increase in the rheological properties of the hydrogels. We can hypothesize that those surfactants induce gelation by concentrating OCNF in the water phase—in a similar fashion as C₁₂EO₆ here seen in SANS or even by acting as depletants to induce gelation by depletion–flocculation—but that they do not disturb the OCNF network as strongly as C₁₂EO₆ micelles do. Indeed, Triton X-100 and Pluronic F68 are known to form oblate⁴⁶ and spherical⁴⁷ micelles, respectively. It is then possible that the shape variation of C₁₂EO₆ micelles with concentration is what is preventing gelation in our system. Indeed, we saw previously that having elongated micelles can more efficiently disturb the OCNF network.²⁶ Moreover, we also showed that the OCNF network integrity could really be affected by the presence of nanofillers with a size larger than the mesh size of the network.⁴⁸ Both phenomena may explain the incapacity of C₁₂EO₆ to induce gelation by depletion forces or simply by crowding.

Hence, we can conclude that addition of C₁₂EO₆ at relatively low concentration (<100 mM) is enough to influence crowding of OCNF in suspension thanks to the excluded volume effect but that it has little influence on the overall suspension rheological properties, probably as it also acts as steric barriers between nanofibrils.

As OCNFs are negatively charged particles, the addition of a charged surfactant may have a larger influence on the properties of the suspension. We hence studied the addition of SDS to the system.

Figure 2 gives the rheological properties of mixtures of OCNF at 1 wt. % and SDS at various concentrations. Contrary to C₁₂EO₆, the addition of this anionic surfactant has a clear effect on the rheological properties of the suspensions, which are forming hydrogels. Indeed, they exhibit an increased viscosity [Fig. 2(a)], a dominant gel-like behavior with G′ > G″ over a wide frequency range [Fig. 2(b)], and a large linear viscoelastic regime [Fig. 2(c)]. The gel-like behavior is characterized by G′ > G″, and a modest frequency dependency even for the highest SDS concentrations indicates a weak physical gel. Importantly, with increase in the SDS concentration, the G′ and G″ dependency as a function of angular frequency (ω) flattens out, indicating the slowing down of OCNF dynamics due to the presence of SDS. A similar trend was observed by Quennouz et al. with SLES.¹⁴ Hence, the addition of SDS triggers the liquid-to-gel phase transition while maintaining relatively low shear viscosity and storage modulus compared to other methods of gelation. Indeed $η(γ̇=0.1)$ reaches ∼40 Pa.s for the mixture of OCNF 1 wt. % with [SDS] = 80 mM, while it is >100 Pa.s for a suspension of OCNF at 2 wt. %, see Ref. 15. Similarly, G′(ω = 1) = 2 Pa for OCNF(1 wt. %)+SDS(80 mM), but it reaches 70 Pa for OCNF(2 wt. %).

As salt is known to induce gelation by charge screening of nanofibrils, it is interesting to consider the influence of ionic surfactants on the ionic strength I of the suspension. It could help determine whether the gelation is related to the presence of more counterions in the solution that would screen the surface charges of OCNF. The variation of ionic strength is directly related to the concentration of SDS within the sample. It can be written as follows:⁴⁹ 

With c the concentration of SDS in mM, CMC is the critical micellar concentration (8.2 mM) and α = 0.567 the degree of micelle ionization.⁴⁹ For c = 20 mM, the increase in ionic strength is hence I = 11.5 mM while it reaches I = 28.6 mM at c = 80 mM. For comparison, a gel-like behavior was clearly observed for OCNF at 1 wt. % and [NaCl] = 0.1M only (i.e., a ionic strength of 100 mM).¹⁵ Hence, gelation in the presence of SDS is triggered at much lower ionic strength than in the presence of NaCl.

A second hypothesis is that the addition of SDS may alter the pH of the solution, which in turn could drastically change the ζ-potential of OCNF nanofibrils. pH measurements of SDS suspensions showed little variation as a function of concentration (pH = 6.6 at c = 0 mM, 6.0 at c = 20 mM and 6.0 at c = 80 mM). We previously showed that OCNFs exhibit a clear stable ζ-potential of ca. −60 mV for pH ≥ 4.⁵⁰ This invalidates this hypothesis.

Moreover, the absence of yield strain on the amplitude sweeps [see Fig. 2(c)], even at strains reaching 100%, suggests a homogeneous microstructure with a network that can withstand large deformations. This behavior is different from the attractive colloidal gel behavior observed when the charges between nanofibrils are screened by the addition of salt or change in pH.^15,50

A study of SANS data of both pure SDS micelles and mixtures of OCNF and SDS can help unravel the mechanisms of gelation between the negatively charged nanofibrils and the anionic micelles (see Fig. 3). The SANS patterns for SDS micelles alone (red curves in Fig. 3) are characteristic of interacting spherical objects, with an oscillation at ca. q = 0.1 Å⁻¹ associated with the size of the objects, and a broad peak at ca. q = 0.05 Å⁻¹, increasing with micelles concentration, due to electrostatic repulsion between the micelles. At 80 mM, the increase observed at small angles is probably an artifact from the subtraction. The patterns can be fitted using the model of charged spheres presented in the section titled Materials and Methods. Fitting of the micelle dimensions was done with the sample at 40 mM to have a clear signal-to-noise ratio; then, the radius was fixed for all other concentrations, with only the interaction parameters adjustable. Micelles are characterized by their radius R = 1.8 ± 0.1 nm and their polydispersity σ = 11% ± 1% at 20 mM or 17% ± 1% (for the other concentrations). Miura et al. evidenced a second CMC via conductivity measurements at 65 mM, usually associated with micelle elongation.⁵¹ Nonetheless, as satisfactory fits were obtained with spherical micelles even at 80 mM, we privileged this simpler model. Fitting of the micelle–micelle interactions was made using the Yukawa structure factor. Table III gives the values of U, φ, and (1/κ) the strength of interaction, the volume fraction of interacting micelles, and the Debye–Hückel length of interaction, respectively.

An increase in the micelle concentration is associated with an expected increase in φ, the volume fraction of interacting micelles, and a decrease in the Debye–Hückel length (1/κ), consistent with a reduced intermicellar distance, while the overall potential of repulsion, U, slightly increases.

Regarding mixtures of OCNF at 1 wt. % and SDS micelles, the contrast made using d-SDS and D₂O (green curves in Fig. 3) allows studying only the nanofibrils. Contrary to the signal of OCNF 1 wt. % alone, this time the slope deviates strongly from the q⁻¹ behavior, with a much more pronounced slope, indicating attraction between fibrils. Indeed, they were fitted using the same OCNF form factor as previously, but all concentrations required the addition of an attractive structure factor (ν_RPA < 0). No clear tendency regarding the strength of the attraction can be drawn from the SANS fits though, as both samples made with 20 and 60 mM SDS exhibit an unusually high attraction (ν_RPA ≤ −5), while suspensions at 40 and 80 mM present a “weaker” attraction (−2 ≤ ν_RPA ≤ −1).

Another contrast made using d-SDS in 50% D₂O (blue curves on Fig. 3) focuses only on SDS micelles. The signal resembles the one of pure SDS micelles, but it can be clearly observed that the strength of micellar repulsion is lowered in the mixture, with a broader peak found at larger angles compared to the system made purely of SDS. Table IV gives the values of ϕ, U and (1/κ) obtained from the fits of the mixtures at this contrast. When comparing the values found for micelles alone in Table III and mixtures in Table IV, one can note a decrease of the Debye–Hückel length of interaction (1/κ) at all concentrations. Hence, similar to the nanofibril, the micelle–micelle electrostatic repulsion is lowered in the mixtures.

The last two contrasts (yellow and pink curves in Fig. 3) can be fitted by summing the contribution of both attractive OCNF and SDS micelles, fixing both nanofibrils and micelles form factor parameters. Nonetheless, while we could fit the data by fixing the strength of nanofibril attraction according to the results found at the contrast d-SDS/D₂O, we had to make the micelle interaction parameters adjustable for each contrast for satisfactory fitting of the data. These differences are probably the signature of nanofibril–micelle interactions that are neglected in our fitting model. Table V gives the results of the fits for the system OCNF+SDS at 40 mM and for all contrasts, while Tables S1, S2, and S3 in the supplementary material give the results for 20, 60, and 80 mM, respectively. The variation of scaling factors with contrasts can be easily explained by differences in deuterium/hydrogen ratios, as A ∝ Δρ², with Δρ the SLD contrast between scatterers (nanofibrils or micelles) and solvents. We note that for the last two contrasts, the signal from micelles is highly dominating the SANS patterns, and it would not be possible to extract the fibril structure factor solely from those two contributions. Actually, satisfactory fits of those two contrasts can already be obtained assuming ν_RPA = 0 (no interactions between nanofibrils). Nonetheless, as the contrast using d-SDS in 50% D₂O, focusing only on OCNF, highlights that attractive interactions between nanofibrils are at play, we decided to keep this attractive interaction contribution for the contrasts h-SDS in D₂O and d-SDS in H₂O, where both OCNFs and micelles are visible.

Analysis of SANS data reveals that in mixtures of nanofibrils and SDS micelles, OCNF experiences attractive interactions, while the overall repulsion between micelles is reduced. Moreover, results also hint at the presence of nanofibril–micelle interactions. A similar behavior was previously observed by Kline and Kaler when studying mixtures of Ludox silica spheres and SDS.⁵² In their case, the essentially noninteracting Ludox spheres experienced attraction in the presence of SDS, while the screened Coulomb repulsion between micelles was reduced. The authors attributed these results to depletion attraction of Ludox spheres by the presence of SDS micelles and vice versa, while the Ludox–SDS interaction is probably related to steric/electrostatic repulsion. Associated with the difference in size between nanofibrils (length > 100 nm) and micelles (typically 1–2 nm in radius), the presence of attraction between nanofibrils may hint at depletion–flocculation as the main mechanism for gel formation. Nonetheless, this mechanism would give an attractive colloidal gel, which contradicts the amplitude sweep measurements discussed earlier. A more nuanced view is that the gelation is probably due to a balance between nanofibril–nanofibril depletion attraction and nanofibril–micelle electrostatic repulsion that tune the depletion-induced gelation mechanism, resulting in a more stable microstructure. This may explain why the gel properties are highly sensitive to the amount of SDS added and notably why Quennouz et al. observed a loss of stability at high SDS concentrations due to fibril aggregation.¹⁴ 

Our measurements were carried out in the absence of salt to probe only the effect of SDS micelles on the hydrogels. As salt and anionic surfactant induce gelation via seemingly different mechanisms, it would be interesting to probe a hydrogel in the presence of anionic surfactant and salt. Previous work carried out on a mixture of OCNFs at 1 wt. % and N-lauroyl sarcosine sodium salt (SLS), another anionic surfactant, in the presence of 1 wt. % NaCl, showed weaker values of G′ and G″ when the surfactant was added than in its absence.²⁶ This suggests that the gelation mechanism triggered by anionic surfactants and the OCNF charge screening in the presence of salt are competitive phenomena. Moreover, the addition of salt can also influence the micelles by notably screening their Coulomb repulsion. A hypothesis is that in the presence of salt, charge screening of OCNFs induces the formation of a stiff network that is the main mechanism for gelation in that case, while anionic micelles act solely as steric barriers disturbing this attractive colloidal network. This could be further studied in the future.

Due to the positive charge of the polar head, DTAB is expected to strongly interact with the negatively charged nanofibrils. Indeed, even at concentrations below the CMC, addition of DTAB to a 1 wt. % OCNF suspension results in a stiff gel, as evidenced by Figs. 4(a)–4(c).

High viscosities (with a viscosity at 10⁻² s⁻¹ in the range of 1000 Pa s) and storage and loss moduli (with values almost 100 times higher than for samples made with SDS at 80 mM) are observed for the samples containing DTAB. Moreover, the frequency sweeps show little frequency dependency, indicating that the OCNFs are in an arrested state within the probed time frame (i.e., 1/ω). The strain required for the material to yield is usually defined in amplitude sweep measurements by the point in strain at which G′ = G″. We note that for the samples containing DTAB (especially at 10 mM), G′ decreases and approaches the value of G″. Thus, the amplitude sweeps indicate the approaching of yielding of the gel due to the dislodging of the nanofibrils composing the network. In contrast, for the OCNF + SDS samples, G′ does not converge toward G″ within the probed strains, indicating that larger values of strains are required for the material to yield.

The elasticity (captured by G′) and viscosity of the samples containing DTAB at concentrations of 5 and 10 mM are superior to those obtained for a suspension of OCNF at 1 wt. % with 100 mM NaCl,¹⁵ although in the latter a higher amount of positive charges were added. This strongly indicates that not only the charged polar head but also the hydrophobic tail of the surfactant has an influence on the macroscopic properties. A possible mechanism of gelation could be the adsorption of the polar head of the surfactant onto the nanofibrils surface due to electrostatic attraction. This would render the nanofibril not only less charged but also more hydrophobic, which could encourage fibril–fibril aggregation between “hydrophobic patches” of two nanofibrils. We hence monitored in SANS the system at 5 mM of DTAB [see Fig. 4(d)]. Interestingly, the systems exhibit a SANS pattern largely modified compared to the signal of pure OCNF at 1 wt. %. To simulate fibril side–side aggregation, we fitted this pattern using a signal of fibrils with the same length L fixed at 160 nm but a larger cross section with a radius R_max = 32.0 ± 0.1 nm and ε = 0.11 ± 0.02 (much larger than the typical dimensions of OCNF). Moreover, a small oscillation at q = 0.15 Å⁻¹ requires the addition of a signal from small spheres of radius R = 1.5 ± 0.1 nm to be fitted. This may indicate that some DTAB micelles are present, although the solution concentration is expected to be below the CMC (i.e., ∼3× below the CMC).³⁰ The model used here to describe the data may be inadequate, hence, great caution should be taken regarding the results from the fit. Nonetheless, it is still clear that when DTAB is added, OCNFs experience strong charge screening and aggregation explaining the rheological properties observed.

Furthermore, the system is highly sensitive to the amount of DTAB added, as 1 mM is not enough to trigger this aggregation phenomenon, while extremely large amounts of DTAB (20 mM and above) result in macroscopic phase separation, in agreement with previous work by Quennouz et al.¹⁴ 

Finally, CapB was chosen as it is composed of the same C₁₂ hydrophobic tail as the other surfactants studied, but its betaine polar head exhibits both positive (thanks to the quaternary ammonium group) and negative charges (via the final carboxylic group). This zwitterionic surfactant also strongly influences the rheological properties of OCNF suspensions, as seen in Fig. 5. The viscosity of the suspensions are of the same order of magnitude as for mixtures of OCNF and SDS, with a shear viscosity at 0.1 s⁻¹ of ∼10–50 Pa.s for the highest surfactant amounts [60 and 80 mM, see Fig. 5(a)]. A fluid-to-gel phase transition is also observed with increasing concentration of the surfactant. Interestingly, the gel properties at 80 mM of CapB are ∼10 times higher than for its SDS counterpart [see Figs. 5(b) and 2(b), respectively], with a weaker frequency dependency. Similar to the samples with SDS, yielding of the gel is not observed in the probed strains in the amplitude sweeps, although a deviation from the linear viscoelastic regime is observed for G′ and G” at ∼50% and 30% for concentration of 60 mM and 80 mM, respectively [Fig. 5(c)]. Moreover, similar to anionic surfactants, we showed in a previous publication that the addition of CapB to a OCNF suspension in the presence of salt leads to a reduction of both G′ and G″, suggesting that when a stiff network is formed, CapB micelles act also like a steric barrier.²⁶ 

Similar to the case of SDS, we can calculate the increase in ionic strength with surfactant concentration (with α = 0.88 for CapB).⁵³ For the CapB concentration c = 20 mM, the increase in ionic strength is I = 9 mM and reaches I = 35.4 mM at c = 80 mM. They are hence of the same order of magnitude as when SDS is used. Again, pH measurements showed no effect on adding CapB to the pH (with values of 6.6 at c = 20 mM and 6.4 at c = 80 mM).

From the rheological properties, we can hypothesize that similar gelation mechanisms between OCNF and CapB are at play when compared with SDS. Nonetheless, the presence of positive charges within the polar head might be responsible for the much higher storage and loss moduli of the samples compared to the system with SDS. This hypothesis is again verified using SANS data. Figure 6 gives the SANS patterns of pure CapB suspensions [Fig. 6(a)] and mixtures of OCNF and CapB [Fig. 6(b)] in D₂O. As we only have access to a hydrogenated version of the surfactant, we did some contrast variation studies at [CapB] = 40 mM by varying the D₂O/H₂O composition of the solvent [Fig. 6(c)].

At all concentrations, CapB micelles can be fitted as spheres of radius R = 2.4 ± 0.1 nm and polydispersity σ = 15–17%. To properly model the signal at small angles, weak steric repulsion between micelles, modeled using the Percus–Yevick structure factor, must be taken into account for concentrations ≥40 mM. The results from the fits are given in Table VI.

Surprisingly, for mixtures of OCNF and CapB (with the exception of the sample at 40 mM), the SANS data can satisfactorily be fitted by simply summing the contributions from micelles and nanofibrils, without further interactions between nanofibrils or nanofibrils and micelles. Moreover, except for the sample at 80 mM, the scaling factors A_OCNF and A_micelles are found to be similar in these mixtures compared to suspensions containing only OCNF or CapB micelles (see Table VII). At 80 mM, the signal of OCNFs seems weaker than expected [A_OCNF = (0.41 ± 0.01) × 10⁻² cm⁻¹ vs (0.64 ± 0.01) × 10⁻² cm⁻¹ for a suspension of OCNF at 1 wt. % only], which may be explained by the dominance of the signal from micelles at this concentration. This is a striking difference compared to the case with nonionic C₁₂EO₆ where A_OCNF increased with concentration. We note that the C₁₂EO₆ micelles present a much lower CMC (0.085 vs 0.28 mM) and undergo a clear change in shape with concentration (from quasi-spherical to highly elongated). We expect elongated objects to more drastically affect the OCNF network²⁶ and hence concentrate the nanofibrils. On the contrary, CapB micelles remain spherical at all concentrations, with an occupied volume fraction below 3% as seen from the fitting of pure micelle suspensions (see Table VI). This should explain why the overall volume fraction occupied by nanofibrils (and hence the scaling factor A_OCNF) is unaltered by the micelle’s presence.

As previously pinpointed, no evidence of attraction between nanofibrils (contrarily to the mixtures with SDS) is seen from the SANS data, except possibly for the sample at 40 mM where a slight increase is seen at the smallest angles [cyan curve in Fig. 6(b)]. As already demonstrated with the other surfactants, contrast matching studies are crucial to isolate the contribution from OCNF, especially at high surfactant concentrations. Hence, we studied OCNF 1 wt. % + CapB 40 mM at different D₂O/H₂O ratios for the solvent [see Fig. 6(c)]. 20% D₂O (pink curve) corresponds to the matching point of CapB micelles. The data were fitted using attraction between nanofibrils, with ν_RPA = −3.10 ± 0.01 and R_Cq fixed at R_max. Then, all the other contrasts can be fitted summing the contributions of CapB micelles and attractive fibrils. Nonetheless, in contrast to SDS, no extra contribution or alteration of OCNF/CapB structure factor associated with nanofibril–micelle interactions (neglected in our model) is required to give a satisfactory fit of the data. Scaling parameters are given in Table S4 of the supplementary material.

The evidence of fibril–fibril attraction with the contrast variation studies suggests that, as with SDS, the fibrils evidence a slight depletion attraction in the presence of CapB. The CapB structure factor seems unaffected by the presence of OCNF; nonetheless, one must remember interactions between CapB are modeled with only a very weak hard-sphere potential instead of a strong Coulomb repulsion with SDS. The absence of nanofibril–micelle interactions here is more puzzling but may be related to the absence of net charge for the micelles. The absence of such strong nanofibril–micelle repulsion (that is expected to oppose the rising depletion attraction between nanofibrils) would explain the higher values of G′ and G″ for samples with CapB than SDS at the same concentration but also the deviation from the linear viscoelastic regime observed in the amplitude sweeps at the highest CapB concentrations. Hence, these results suggest that mixtures of OCNF and CapB exhibit similar gelation mechanisms as those containing OCNF and SDS. It is highly possible that OCNFs in suspensions with 60 and 80 mM CapB experience fibril attraction as well, but due to the strength of the signal of micelles in pure D₂O, this is not detected in this contrast. For 20 mM, the signal from OCNF is still clearly high compared to the contribution of the micelles, with a q⁻¹ slope visible at small angle, which suggests that attraction between nanofibrils is not present or rather weak at this concentration. Further contrast studies, preferentially with a deuterated version of the surfactant to avoid a strong incoherent scattering signal from hydrogens would help ensure that the results observed at 40 mM can be applied to the other concentrations.

In this work, we studied the effect on OCNF suspensions of surfactants bearing a C₁₂H₂₅ tail and a nonionic (C₁₂EO₆), anionic (SDS), cationic (DTAB), or zwitterionic (CapB) polar head. We compared the rheological properties of these OCNF–surfactants suspensions with their structural properties measured in SANS. We showed that the polar head of the surfactant had a strong influence on the rheological properties of the suspensions, with formation of a gel observed for surfactants that have ionic polar headgroups. Indeed, we showed that, at the concentrations probed, the nonionic surfactant does not affect the rheological properties of a 1 wt. % OCNF suspension, which exhibits a fluid-like behavior. SANS results suggest that if nonionic micelles induce crowding of OCNFs in the water phase due to an excluded volume effect, they also act as a steric barrier between nanofibrils, preventing gelation. Their change in shape with concentration, going from quasi-spherical to elongated micelles, might prevent them from acting as depletants. On the contrary, addition of SDS induces a fluid-to-gel phase transition, with a weak physical gel behavior observed in rheology. SANS measurements made at different contrasts show that nanofibrils exhibit attraction in the presence of SDS, while electrostatic repulsion between micelles decreases. Moreover, results suggest a nanofibril–micelle interaction is also arising. These results hint at a competition between depletion attraction and steric stabilization as the main force driving gelation. Similarly, CapB also induces gelation, with slightly tougher gels compared to SDS when added at the same concentration. SANS data for the sample at 40 mM hint at depletion-induced gelation, notably as fibril–fibril attraction is evidenced while no fibril–micelle interaction is observed. Results at other concentrations are less clear and future measurements using contrast matching techniques would help confirm this result. Finally, DTAB had the strongest effect on the dispersion, with tough and stiff gels obtained at concentrations below the CMC (5 and 10 mM). SANS data suggest fibril aggregation, probably as DTAB adsorbs at the surface of nanofibrils to screen the negative charges, making them more hydrophobic. At higher concentrations of DTAB, the gels obtained are unstable and phase separation occurs. Understanding the nature of the interactions between OCNF and surfactants is a key step in using these bio-based materials for commercial applications.

See the supplementary material for the definition of χ² in SANS fitting, rheological data for OCNFs (1 wt. %) + C₁₂EO₆, SANS data for individual components OCNF, C₁₂EO₆, and their mixtures, and the fitting parameters for OCNF with SDS and CapB.

The authors thank EPSRC for funding this project (Grant No. EP/N033310/1). Dr. Vincenzo Calabrese thanks the University of Bath for PhD studentship funding. Experiments at the ISIS Pulsed Neutron and Muon Source were supported by beam time allocations for Larmor from the Science and Technology Facilities Council under Proposal No. RB1720220. Measurements on C₁₂EO₆ systems were made on Sans2D (Proposal No. RB1720206).

The authors have no conflicts to disclose.

Julien Schmitt: Conceptualization (equal); Data curation (equal); Formal analysis (lead); Investigation (lead); Methodology (lead); Software (lead); Writing – original draft (lead); Writing – review & editing (lead). Vincenzo Calabrese: Investigation (supporting); Methodology (supporting); Writing – original draft (supporting); Writing – review & editing (supporting). Marcelo A. da Silva: Investigation (supporting). Kazi M. Z. Hossain: Investigation (supporting). Peixun Li: Resources (equal). Najet Mahmoudi: Investigation (supporting). Robert M. Dalgliesh: Investigation (supporting); Software (supporting). Adam L. Washington: Investigation (supporting); Software (supporting). Janet L. Scott: Funding acquisition (equal); Project administration (supporting); Supervision (equal). Karen J. Edler: Conceptualization (equal); Funding acquisition (equal); Methodology (supporting); Project administration (lead); Supervision (equal); Writing – review & editing (equal).

The data that support the findings of this study are openly available from the University of Bath data archive at: https://doi.org/10.15125/BATH-01237.⁵⁴