Filler size effect in an attractive ﬁbrillated network: a structural and rheological perspective †

The effect of the ﬁller size on the structural and mechanical properties of an attractive ﬁbrillated network composed of oxidised cellulose nanoﬁbrils (OCNF) in water was investigated. Silica nanoparticles, SiNp 5 and SiNp 158 which were ≈ 5 and ≈ 158 nm diameter, respectively, were chosen to be of a similar and a greater dimension of the network mesh size. Contrast matched Small Angle Neutron Scattering (SANS) experiments revealed that the presence of the ﬁllers (SiNp 5 and SiNp 158 ) did not perturb the structural properties of the OCNF network at the nm-length scale. However, the ﬁller size difference strongly affected the mechanical properties of the hydrogel upon large amplitude oscillatory shear. The presence of the smaller ﬁller, SiNp 5 , preserved the mechanical properties of the hydrogels while the larger ﬁller, SiNp 158 , allowed a smoother breakage of the network and low network recoverability after breakage. This study, show that the ﬁller-to-mesh size ratio, for non-interacting ﬁllers, is pivotal for tailoring the non-linear mechanical properties of the gel such as yielding and ﬂow.


Introduction
The growing need for sustainable materials has promoted interest in the use of cellulose nanofibrils (CNF) as building blocks for renewable products. CNF belong to the colloidal domain and are usually characterized by a radius in the order of few nm and a length of 100-1000 nm depending on the preparation methods and source. 1,2 TEMPO-mediated oxidation of cellulose has been, for instance, a successful route for the production of oxidised cellulose nanofibrils (OCNF) on a large scale. [2][3][4] Importantly, the carboxylate groups on the OCNF surface yield a surface negative charge which allows the preparation of stable aqueous dispersions. 3 Interfibrillar repulsive/attractive forces have been reported to be strongly dependent on the pH 5,6 and ionic strength [7][8][9] of the aqueous media. 10 Charge screening due to counterion binding to the carboxylate group has been shown to be an effective gelation mechanism. 7,9 Structural investigation of OCNF dispersions, via small angle X-ray scattering, revealed that, for concentrations above the overlap threshold, 11 a clear transition from a repulsive to an attractive fibrillated network occurs upon increasing ionic strength. 9 This finding is directly related to the sol-gel transition 9 and to the theoretical expectations based on the Derjaguin-Landau-Verwey-Overbeek (DLVO) for two cylindrical rods. 7 To broaden the understanding of interfibrillar assembly in systems closer to industrially relevant scenarios, the use of additives, such as surfactant 11,12 and polymers, 13,14 in CNPbased hydrogels has been explored. 1 Nevertheless, knowledge of the interplay between non-interacting colloidal fillers and a fibrillated network is lacking. The filler surface chemistry, size and volume fraction have been mainly investigated in protein-based hydrogels and referred to as "active" or "inactive" fillers based on their ability to strengthen or weaken the gel, respectively. [15][16][17] In principle, fillers could be employed to modulate other mechanical properties such as structural breakage (yielding), flow, structural recoverability and plasticity; important parameters to account for processing operations and/or customer appreciation. 18 Such mechanical properties often occur upon large deformations and high shear rates, where the mechanical response of the material depends on the applied strain and shear rate, identified as non-linear viscoelastic region (NLVR). On this ground, we investigated the filler size effect on the linear and non-linear viscoelastic response of an attractive fibrillated network composed of OCNF. The interfibrillar attractive forces were promoted via addition of 100 mM NaCl, leading to an OCNF network with a mesh size, defined as the average distance between the nearest fibrillar junc-tion zones, of 20-40 nm as recently reported by our group. 9 Silica nanoparticles (SiNp) with an average dimension similar to the mesh size of the network and a SiNp larger (ca. by a factor of 5) than the network mesh size were chosen as non-interacting fillers in the OCNF-based gel. We show that the filler-to-mesh size ratio, for non-interacting fillers, is of pivotal importance for tailoring non-linear mechanical properties of the gel such as yielding and flow.

Materials
An OCNF dispersion was prepared as previously described, 19 using a TEMPO/NaOCl/NaBr oxidation. 20 A 2 wt% OCNF stock was re-dispersed from freeze-dried OCNF in ultrapure deionised (DI) water (18.2 MΩcm) and stirred overnight, followed by a mild sonication process (Ultrasonic Processor FB-505, Fisher, 200 W cm −2 , equipped with a 1 cm probe, using a series of 1 s on 1 s off in pulse mode for a net time of 120 s at 30% amplitude on ca. 45 mL dispersion contained in an ice bath). A 10 wt% SiNp 158 dispersion was prepared dispersing silica nanopowder (718483 Sigma Aldrich R ) in ultrapure DI water followed by sonication (as for the OCNF indicated above) and used without further treatment. A ca. 9 wt% aqueous SiNp 5 dispersion, provided by Geo40 TM (sodium stabilised Colloidal silica of geothermal origin), was dialysed using cellulose dialysis tubing (Sigma-Aldrich R cellulose dialysis tubing, molecular weight cut-off of 12400 Da) against DI water for 3 days, refreshing the DI water twice per day. The dialysed SiNp 5 dispersion was pH adjusted to 7 using HCl (aq) and further dialysed against DI water, as previously described. Larger aggregates were removed using a syringe filter unit with a cutoff size of 0.22 µm (Millex R -GS) and the solid content (wt%) of the filtered dispersion obtained gravimetrically. Specific OCNF, OCNF-SiNp 5 and OCNF-SiNp 158 concentrations were obtained via dilution of the stock dispersions described above, except where otherwise stated, followed by the addition of a 2 M NaCl solution to achieve a final NaCl concentration of 100 mM in all samples.

Methods
Small angle X-ray scattering (SAXS) measurements were performed on an Anton-Paar SAXSpoint 2.0 equipped with a copper source (Cu K-α, λ = 1.542) and a 2D EIGER R series Hybrid Photon Counting (HPC) detector. The sample-to-detector distance was 556 mm covering a range of the scattering vector (q) of about 0.01 < q < 0.4 Å −1 . Samples were loaded into 1 mm quartz capillaries and the scattering intensity (I(q)) collected in three frames, with 300 s exposure per frame. Temperature was kept at 25 • C via a Peltier unit (± 0.1 • C). Small angle neutron scattering (SANS) experiments were conducted at the Institut Laue Langevin (Grenoble, France) on the D22 SANS beamline using a wavelength of 6 Å and a sample-to-detector distance of 2.8 and 17.6 m to yield a q-range of 0.003 < q < 0.4 Å −1 . Temperature was kept at 25 • C via a Julabo circulating waterbath (±0.5 • C). Particles which have the same scattering length density (SLD) as the continuous phase do not contribute to the scattering intensity (I(q) = 0) and they are said to be contrast matched. 21 We determined the contrast match point of the SiNp particles, using different H 2 O/D 2 O ratios as the continuous phase. This is possible due to the different SLD between H 2 O (−0.5 · 10 −6 Å −2 ) and D 2 O (6.3 · 10 −6 Å −2 ) which allow tuning of the SLD of the continuous phase upon changing H 2 O/D 2 O ratios. Samples were prepared with a continuous phase composed of 60 vol% D 2 O (99.9 atom % D, Sigma Aldrich R ) and 40 vol% H 2 O (DI water) which was experimentally determined to be the contrast match point for SiNp ( † Electronic ESI Fig. S1) in agreement to what was previously reported. 22 For this experiment, OCNF, OCNF-SiNp 5 and OCNF-SiNp 158 concentrations were obtained via dilution of a 2 wt% freeze-dried OCNF dispersed in pure D 2 O (prepared as previously described) and SiNp stock dispersions described above (SiNp 5 and SiNp 158 (in H2O)) with the required H 2 O/D 2 O ratio, followed by the addition of a 2 M NaCl solution (in H 2 O) to achieve a final NaCl concentration of 100 mM in all samples. Samples were measured in 2 cm wide optical quartz cells with 1 mm path-length. For SAXS and SANS measurements background subtraction and data treatment were performed using the Irena package 23 whilst data analysis was done using the NIST SANS Analysis package (models were used without further modification), 24 both within IGOR Pro (Wavemetrics, Inc.). Data are presented as I(q) vs q and I(q) vs d, where d is the d-spacing (d = 2π q ). Details regarding the model of non-interacting flexible cylinders with an elliptical cross-section are described by Schmitt et al. 9 Dynamic light scattering (DLS) was performed on diluted samples (0.01 wt% in 100 mM NaCl) using a Malvern Zetasizer Nano ZSP R (Malvern, UK). The samples were loaded in disposable polystyrene cells with a path length of 1 cm and measured as an average of 4 measurements from 100 scans each. The values are reported as normalised intensity and obtained from the average of three separate samples. Temperature was kept at 25 The rheological measurements of the OCNF-based hydrogels were performed using a stress-controlled rheometer (Discovery HR3, TA instruments R ) equipped with a sandblasted plate-plate geometry (40 mm). To avoid evaporation, the edge of the samples was covered with low viscosity mineral oil and further covered with a solvent trap to ensure constant temperature within the chamber (25 • C ± 0.1 • C via a Peltier unit). After loading, the gel was exposed as following to i) 30 s pre-shear at 300 s −1 to ensure equal sample history, ii) time sweep employing small strain amplitudes (γ 0 = 0.05 (%)) at constant angular frequency (ω = 1 rad · s −1 ) for 50 · 10 3 s, iii) frequency sweep at constant strain amplitude, γ 0 = 0.1%, being in the linear viscoelastic region (LVR) ( † Electronic ESI Fig. S2) and iv) strain sweep at constant, ω = 1 rad · s −1 . The storage and loss modulus, G' and G" respectively, were computed by the TRIOS software and used to obtain tanδ = G /G . For iv), raw data were acquired as stress (σ ) as a function of intracycle shear rateγ (s −1 ) and the instantaneous intracycle strain (γ) and presented in the form of Lissajous-plots as σ vsγ and σ vs γ, known as the viscous and the elastic projection, respectively.

Results and discussion
Gelation of the 1 wt% OCNF was achieved upon addition of NaCl (100 mM), resulting in an attractive fibrillated network with a defined mesh size of 20-40 nm. 9 Silica nanoparticles, with a dimension similar and greater than the mesh size of the network were incorporated in the 1 wt% OCNF gel and their effect on structural and mechanical properties of the OCNF-based gel investigated. Although silica nanoparticles were chosen to avoid strong interactions with OCNF (e.g. electrostatic, hydrophobic), experimental evidence for this was sought. At first, the filler sizes were evaluated via SAXS using conditions of ionic strength as the one employed in this study to induce gelation of OCNF (Fig.  1a). Contrarily to SiNp 158 , the SiNp 5 displayed an approaching plateau at low-q, indicating that the characteristic size of the object is probed (radius of gyration). 25 The data of the SiNp 5 were fitted to a model of spheres with a log-normal size distribution, yielding a particle mean size of 5.2 ± 0.2 nm and a size distribution as shown in Fig. 1b. Since SAXS measurements did not access the larger size of SiNp 158 , DLS measurements were used to determine the hydrodynamic size of the particles (Fig. 1b). This accounts for the hydration shell of the particles which is expected to be in the order of few nm at this ionic strength. 26 Overall, SiNp 158 displayed a greater mean size (158.0 ± 1.3 nm) and a narrower size distribution compared to SiNp 5 . To explore the effect of the filler on the structural properties of the OCNF network, SANS measurements were performed under conditions where the SiNp scattering contribution was matched to the solvent. This results in a dominant OCNF scattering intensity, allowing us to probe, solely, the structural properties of the OCNF network in the presence of SiNp. In practice, this was achieved using a continuous phase composed of a mixture of 60 vol% D 2 O and 40 vol% H 2 O (see Fig S1 for contrast match point determination for SiNp). According to previous reports, the scattering pattern of 1 wt% OCNF at 100 mM NaCl (Fig. 2a) contains information regarding the shape of the fibrils in the high-q and intermediate-q region, which probe the radius and the larger dimension of the fibril, respectively. 9,19,27 The upturn in the low-q region has been instead associated with the attractive interfibrillar interactions as described by Schmitt et al. 9 The data were fitted, in the q-region where interfibrillar interactions are not detectable (high-q and intermediated-q range), to a model of noninteracting flexible cylinders with an elliptical cross-section from which the minor radius (R min ), major radius (R ma j ) and the Kuhn length (b Kuhn ) were obtained; where the b Kuhn is indicative of the OCNF mesh size. 9 The data fitting was carried out using a fixed and arbitrary contour length (L) of 500 nm as the average length was not accessible in the probed q-range, as indicated by the lack of a plateau in the low-q region. In addition, the interfibrillar interactions, in the low-q region, would not allow extrapolation of the OCNF contour length. Since the lower threshold for the data fitting was q min = 0.06 −1 , using values of L 2π q min , kept the fitting unchanged. The model yielded values of R min = 1.1 ± 0.1 nm, R ma j = 4.7 ± 0.1 nm and b Kuhn = 21.4 ± 0.1 nm, in good agreement with previous SAXS measurements 9,19,27 and imaging analysis of OCNF. 9 Upon SiNp addition, the SANS patterns had a similar trend to that of the pure OCNF gel, indicating that, in the probed q-range, neither of these fillers strongly alter the structure of the fibrillated network. As such, the SiNpcontaining samples were fitted using fixed values of R min = 1.1, R ma j = 4.7 and b Kuhn = 21.4 nm as found for OCNF, allowing only scale and background to vary. The data were satisfactorily fitted using these constrained values with exception of the sample containing 2.5 wt% SiNp 5 which displayed some difference in the intermediate-q region as shown by the residual plot (Fig. 2b). The poorer fitting in the intermediate-q range could be due to a small residual unmatched scattering contribution from the SiNp 5 . Analysis of the low-q region revealed a slope of ≈ 2.5 for all the samples, indicating that neither of the fillers alters the attractive interactions between OCNF at this length scales. Overall, both fillers preserved the architecture of the fibrillated OCNF network at the nm length scale, enabling comparision of the rheological properties of the OCNF-based gels in conditions where the network architecture is the same.
The equilibration of the rheological behaviour of the OCNFbased gels was evaluated post breakage (30 s atγ = 300 s −1 ) (Fig. 3). After the imposed breakage, both G' and G" of the 1 wt% OCNF gel followed a rapid change (up to ca. 10 · 10 3 s) succeeded by a less pronounced evolution, indicating the dynamic nature of the OCNF network. Both filler-loaded gels showed greater values of G' and G" compared to the pure OCNF gel indicating an overall augmented toughness of the gels (Fig. 3a-b). Nevertheless, it must be noticed that the magnitude of G' and G" does not necessarily describe the elastic/viscous-like balance of materials, but, for instance, its increase may be simply due to the increase in volume fraction upon filler addition. On the other hand, tanδ better correlates with the elastic/viscous-like contributions of materials despite changes in volume fraction (Fig. 3c-d). The tanδ of the 1 wt% OCNF gel containing 2.5 and 5 wt% SiNp 5 displayed a very similar tanδ profile to the 1 wt% OCNF gel up to 10 · 10 3 s, whilst, at longer times, tanδ showed greater values, indicating that the addition of SiNp 5 increases the viscous contribution of the gel without evident change of the elasticity. Contrarily, the gel containing the 2.5 and 5 wt% of SiNp 158 had a more pronounced decrease in tanδ compared to the OCNF profile alone, indicating augmented elasticity of the gel.
A strain sweep spacing from small up to large deformation was employed to reveal how the filler size affects the OCNF network (Fig. 4). At small deformations, the 1 wt% OCNF gel showed a clear LVR whilst at higher deformation both G' and G" showed a pronounced strain overshoot (signal increase followed by a decrease). Similar G' and G" overshoots have been associated with the balance between breakage and regeneration of the network junctions. 28 Specifically, G' local maxima could arise by the increased connectivity between the fibrils occurring upon deformation, increasing the elastic contribution of the network. However, the presence of a local maxima in G" would further indicate a cooccurring energy dissipation process, consequent to the network breakage. This is depicted by several network models as the balance between the formation and loss of the network junctions upon large deformations. 29,30 The G' and G" overshoots have been classified by Hyun et al. as a strong strain overshoot. 28 The tanδ profile only displayed the upturn corresponding to the onset of yielding due to the greater G" overshoot compared to the G' overshoot. When either 2.5 or 5 wt% SiNp 5 were added into the OCNF gel, G' overshoots were still appreciable suggesting that SiNp 5 has little influence on the gel mechanics. On the other hand, the G" profile of the SiNp 5 containing gel, showed a similar overshoot as the pure OCNF gel although with an extra contribution appearing at lower values of γ 0 . The G' overshoot for the OCNF gel containing 2.5 wt% SiNp 158 is strongly smoothed out, and was completely absent for the 5 wt% SiNp 158 . The respective G" overshoots became remarkable, indicating a pronounced energy dissipation process upon deformation due to the loss of network junctions. The pronounced dissipative process occurring for the SiNp 158 containing gel is well captured by tanδ , where the abrupt upturn occurs at lower values of γ 0 compared to the OCNF gel. Differently, for the SiNp 5 , the overshoots are still visible in the Fig. 4 Strain sweep for the OCNF-based hydrogels. G' and G" normalized by the respective modulus in the LVR (G' 0 , G" 0 ) as a function of strain (γ 0 ) are shown in the left and central panels, respectively. Right panels display the tanδ profile as a function of strain (γ 0 ). The darkening of the background refers to the transition from solid-like (G' > G") to liquid-like (G' < G") for the 1 wt% OCNF gel and is drawn to guide the eye. The green filled symbols are used to highlight the smaller overshoots. On the right, schematics of the OCNF-based hydrogels upon deformation.The displayed data are obtained from a set of duplicate samples; the respective error is within the size of the symbol.
tanδ profile due to the stronger G" overshoot compared to the G' overshoot. Both fillers, SiNp 5 and SiNp 158 , showed a shift of the G" overshoots towards lower values of γ 0 at greater filler concentrations. This can be explained by the greater number of particles which would require less deformation to cage, allowing an earlier onset of the dissipative energy. SiNp 5 and SiNp 158 belong respectively to a similar and a greater (ca. by a factor 5) length scale compared to the mesh size of the OCNF network. 9 This information coupled with the herein described rheological measurements indicate that, when strong attractive interactions between the network and the filler could be ruled out (e.g. electrostatic attraction, hydrophobic interactions), the filler-to-mesh size ratio is of great importance to induce specific rheological responses. Specifically, it is possible to attribute the unchanged elastic fingerprint (G' profile) of the OCNF-SiNp 5 gels to the possibility of the small filler particles to freely move within the aqueous phase of the OCNF network; only mildly affecting the breakage dynamics of the OCNF network upon oscillatory strain sweep. By contrast, the larger filler, SiNp 158 , would not have available free space, leading to a completely different breakage dynamic where the G' overshoot disappears and the dissipative contribution displayed by G" increases substantially (Fig. 4, schematics). Moreover, in agreement with the lack of strong OCNF-SiNp interactions, the addition of SiNp 158 in to the OCNF gel resulted into a transition from a strong strain overshoot (G' and G" increase followed by a decrease) to a weak strain overshoot (only G" increase followed by a decrease) due to the weakening of the associative forces between the building blocks of the network, leading to a smoother structural breakage (yielding) upon deformation. 28 The dynamic moduli, G' and G" are calculated based on the assumption of the sinusoidal stress response of the material. 31,32 Although G' and G" provide a robust way to obtain structural information in the LVR, this assumption loses rigorous mathematical support in the NLVR, where the stress response deviates from being sinusoidal, as usually the case for large deformations. 31,32 In the case of the herein described gels, the richest rheological behaviour lies in the large deformation range, hence a waveform inspection was conducted via plotting the stress response (σ ) as a function of the instantaneous intracycle strain (γ) (Fig. 5), commonly known as the elastic projection of Lissajous-plot (for the viscous projection as σ vs shear rate (γ) see † Electronic ESI Fig.  S3, whilst for the applied sinusoidal deformation as γ vs time and the corresponding stress response as σ vs time see † Electronic ESI Fig. S4). 31,32 At low deformation (γ 0 = 5 %), the Lissajous-plots of the gels showed a perfectly elliptical shape which encloses a little portion of area, indicating the dominant linear response and the dominant elastic behaviour of the material, respectively (Fig.  5). By contrast, at greater deformations (γ 0 = 38 and 71 %) the Lissajous-plots acquired a more distorted shape indicating the deviation of the stress signal from linearity. At the largest deformation (γ 0 = 254 %) the stress response was predominantly viscous and non-linear as indicated by the squared shape of the Lissajous-plot. It is important to notice that this trend reflects that displayed by G' and G", indicating that the physical interpretation of G' and G" being proportional to the average energies stored and dissipated per cycle, respectively, is not violated. For γ 0 values of 38 and 71 % the stress response, starting at σ = 0, followed a linear increase with the strain (elastic straining), after which a more pronounced stress increase was observed, indicating strain stiffening at greater intracycle deformations (γ 0 = 71 % in Fig. 5) which well relates with the G' overshoot displayed in Fig. 4. Similar strain stiffening phenomena have been reported as ubiquitous in any network composed of semiflexible filamentous proteins 33 although found difficult to mimic using commercially available polymers. 34 The origin of strain stiffening in fractal aggregates formed by a diffusion-limited cluster aggregation process, as expected for the OCNF network herein reported, has been proposed to arise from the intrinsic stiffness of the cluster backbone. 35 For γ 0 = 38 and 71 %, where the solid-like behaviour dominates (G' > G"), the shape of the Lissajous-plots of the pure OCNF and the 5 wt% SiNp 5 loaded gel showed similar features. In the presence of the 5 wt% SiNp 158 the Lissajous-plots displayed a more opened structure, indicating a pronounced energy dissipation (proportional to the area enclosed in the Lissajous-plot in the σ vs. γ 0 projection), and a milder intracycle strain stiffening, underpinning the reorganization of the fibrillar network occurring upon deformation. At greater deformations, γ 0 = 254 %, the dominating liquid-like behaviour (G' < G") was underlined by the appearance of the stress shoulder associated with the static yielding prior to flow, where the static yielding is defined as the stress that needs to be overcome to make the material flow. 31 At γ 0 = 254%, the stress shoulder became broader for both filler loaded gels and, after the static yielding, displayed an abrupt stress decrease followed by an increase (indicated by the arrow). These similarities appeared solely in the Lissajous-plots of the filler-loaded gels, suggesting that at larger deformations, after the static yield point, the flow properties are characterised by the filler. To better interpret the non-linear behaviour, we employed the framework of Rogers et al. 31 , which decomposes the whole amplitude cycle into sep-arate sequences of events. Although different approaches have been developed to analyse Lissajous-plots, the one proposed by Rogers et al. 31 better links rheological events to structural properties. 36,37 In particular, we characterize the linear stress response starting at σ = 0 by the local cage modulus G cage as follows: In the framework proposed by Rogers et al., 31 starting at σ = 0, the Lissajous-plot at large deformations could be interpreted as a sequence of straining (where G cage is calculated), static yielding, flowing and structural reformation (labelled in Fig. 6b by 1,  2, 3 and 4, respectively). At small amplitudes, in the LVR, G cage reduces to G' as shown in Fig. 6c for the case of 1 wt% OCNF gel (See † Electronic ESI Fig. S5 for non-normalised values of G', G" and G cage ) whilst at greater deformations, G cage diverges from G' due to the structural breakage which is not fully recovered upon static yielding (2), flowing (3) and structural reformation (4) imposed by the sinusoidal cycle (Fig. 6a). 38 For the pure OCNF gel, a slight decrease in G cage is displayed at greater deformations as expected by structural breakage which is not fully recovered within the sinusoidal cycle. The presence of the SiNp 5 did not substantially affect the rate of change of G cage upon deformation, confirming that the structural recoverability of the gel is dominated by the OCNF network. However, the presence of SiNp 158 in the OCNF gel strongly affected the structural recoverability of the gel as displayed by the pronounced G cage decrease. Fig. 6 (a) G cage normalised by the respective modulus in the LVR (G cage0 = G' 0 ) as a function of strain (γ 0 ) for the OCNF-based gels. The 1 wt% OCNF gel is shown in both graphs as a reference. The darkening of the background refers to the transition from solid-like (G' > G") to liquid-like (G' < G") for the 1 wt% OCNF gel and is drawn to guide the eye. (b) The Lissajous-plot is drawn to depict the sequence of events which soft materials undergo upon LAOS: straining (1), static yielding (2), flowing (3) and reformation (4) as suggested by Rogers et al. 31 (c) Strain sweep for the 1 wt% OCNF gel showing non-normalised G' and G cage as a function of γ 0 . The displayed data are obtained from a set of duplicate samples; the respective error is within the size of the symbol for γ 0 < 100 (%).
The onset of G cage decrease, for the SiNp 158 containing gel, occurred at lower deformations than the OCNF gel, indicating that the network undergoes breakage at lower deformation and structural recoverability is not completed within the cycle (Fig. 6b). Although structural properties of the hydrogels, at the nanoscale, are almost unaffected by the different filler size, the mechanical properties of the gel upon breakage change substantially. This indicates that the relation between mesh size of the network and filler are of great importance to modulate the rheological properties of the hydrogel. Specifically, the presented rheological data point to a mechanism where the SiNp 158 , with a larger size compared to the mesh size of the OCNF network, actively dislodge the physical junctions of the network upon large deformations, leading to a smoother yielding (without G' overshoot). Contrarily, the smaller SiNp 5 does not dislodge the physical junctions of the network due to their ability to be accommodated within the mesh of the network upon large deformations, without disrupting the breakage dynamics of the network.

Conclusions
In this study we investigated the effect of the filler size on the structural and mechanical properties of an attractive fibrillated network composed of OCNF, in conditions where strong fillerfiller and filler-fibril interactions are absent. The two fillers, SiNp 5 and SiNp 158 were chosen to be of a similar and a greater dimension of the gel network mesh size, respectively. Small angle neutron scattering revealed that the OCNF network preserved its nanoscale architecture in the presence of both fillers whilst oscillatory shear rheology captured clear rheological differences. Large amplitude oscillatory shear (LAOS) displayed the richest rheological behaviour and allowed to access the impact of the filler-to-mesh size ratio on the yielding behaviour of the hydrogel. Our findings indicate that the presence of the smaller filler, SiNp 5 , in the OCNF network maintains the mechanics of the network almost unvaried. This phenomenon is associated with the ability of SiNp 5 to fit between the mesh size of the gel network without altering the dynamics of the network upon large amplitude oscillatory shear. In contrast, the presence of the larger filler, SiNp 158 , resulted in a gradual structural breakage and a low network recoverability after the breakage induced by large deformations, suggesting that the SiNp 158 could actively dislodge the junctions of the network. The fundamental understanding provided in this study has implications for industrially relevant formulations, where the efficacy of additives in primary matrixes is of main importance towards tunability of specific structural and mechanical properties on demand.

Conflicts of interest
The authors declare no conflicts of interest.