## Introduction

#### Effect of hornification

Hornification refers to the physical change that happens in cellulosic materials after drying and that has been proposed to be caused by lactone bridge formation60. Pulp hornification upon air-drying can reduce enzymatic digestibility61. In industrial processes pulps would be enzymatically hydrolyzed in a wet state, so avoiding hornification is preferable to adding a DMSO-washing step. Therefore, pulps pretreated at 50 wt% loading were saccharified on a wet basis, which increased the glucose yield by 70%, releasing 66 wt% of glucose (Figure ESI-17). Yields were further improved to 74% with addition of a DMSO washing step before saccharification of the wet pulp.

### Effect of particle size

Comminution is highly energy-intensive and costly. Pretreating larger particles would reduce energy costs during biomass processing25. The particle size has a direct effect on the contact and diffusion of chemicals into the complex interior of the lignocellulose structure34. To evaluate the effect of particle size on pretreatment, three different particle size fractions (coarse, medium and fine) of Miscanthus × giganteus were investigated at 20 wt% loading (Table 2). Pretreatment with dilute acid (3 wt% H2SO4, 120 °C, 1.5 h) was also compared62.

Chemical pretreatments, including ionoSolv, soften biomass structure by partially removing and modifying lignin and hemicellulose and reducing the particle size23,66,67. They are able to reduce size down to a certain “boundary”, which may be linked to the diameter of cellulose microfibril bundles released during hemicellulose and lignin solubilization67,68. Fiberization is reflected in the pulp images for Miscanthus, which had similar features regardless of initial particle size (Fig. 4).

The efficacy of pretreating different particle size fractions was evaluated by comparing the pulp and lignin recoveries, pulp delignification, glucose yields released by saccharification, and particle size distribution (PSD) of the pulps. While the enzyme-accessible surface area of the pulp is of most interest for explaining saccharification yields, its measurement is far more complicated than sieving, which was used as the preferred method to assess PSD changes in this study36. Miscanthus pulps were size-reduced during pretreatment (30–68%, based on the D50 values), due to its low recalcitrance and density (Figure ESI-18)67,69. Greater size reduction was observed for larger particles, though the change in D50 for the largest size fraction could not be quantified (Table 3). The medium and fine fractions were also pretreated using dilute sulfuric acid (3 wt%, 120 °C, 1.5 h). In this case, the recovered pulps closely resembled the starting materials and D50 values showed low size reduction for medium particles and virtually none for fine particles (Table 3, Figure ESI-19). On the other hand, ionoSolv pretreatment resulted in a dramatic degree of particle size reduction, with a more visible effect for the coarse chips (Fig. 4). This suggests that for low density grassy feedstocks, it may be possible to use pretreatment for initial size reduction followed by post-pretreatment size reduction to produce the fine particles needed for efficient hydrolysis and fermentation36, with added energy reduction due to the presence of the ionic liquid70.

Glucose yields were determined after grinding and sieving all pretreated pulps to 0.18–0.85 mm. No significant difference was seen in the glucan recovery values for coarse and medium particles (all ~ 90%), while a slightly greater degree of glucan loss was noted for fine particles (Fig. 5). This, previously observed for rice straw pretreatment34, suggests excess acidity in the IL2, which would have a greater effect for fine particles, with higher volumetric surface area exposed to the H+ protons. Hemicellulose extraction was more effective for finer particles due to greater sugar accessibility. Delignification showed a different trend, being greatest (60%) for medium particles. Coarse particles were less delignified due to the lower surface area to volume ratio; while fine particles also showed lower delignification than expected, presumedly due to re-precipitation of lignin onto the large pulp surface area. The lignin recovery, which was highest for fine particles, provides further support to this hypothesis. As a result, saccharification yields were higher for medium (77%) than for fine particles (67%). This suggests that the presence of lignin has a stronger negative effect on glucose yields than the positive effect of hemicellulose removal, as was previously noted57. These saccharification results were obtained for size-reduced pulps, which adds extra energy penalty but may be unnecessary for scale-up studies.

Based on this, the optimum size for unstirred Miscanthus pretreatment appears to be medium sized particles (~ 3 × 0.02 × 0.01 cm), in line with the proposed particle size range of 2–6 mm recommended by Cadoche and López to minimize comminution energy21.

Dilute acid treatment removed the majority of hemicellulose and small amounts of lignin (Figure ESI-21), as reported in literature71,72. However, delignification (~ 10%) and glucose yields (~ 28%) were very low for all size fractions. IonoSolv pretreatment offers many advantages over dilute acid pretreatment such as greater reduction in particle size, higher volumetric surface areas of the pulps, lower lignin contents, and higher lignin recoveries.

### Effect of 100-fold scale-up

Here, we demonstrate the 100-fold scale-up of the ionoSolv deconstruction of Miscanthus, relative to the bench scale (1 L vs 0.01 L) under the same conditions (120 °C, 6 h) and with solid loadings between 10–20 wt% (> 20 wt% could not be attempted due to stirrer motor limitations), using different particle sizes and stirring speeds. Insights into the effect of scale-up were garnered by comparing pulp and lignin recoveries, pulp composition and saccharification and lignin characteristics.

#### Pulp washing protocol optimization

The volume of solvent employed for pulp washing at the bench scale becomes impractical at larger scales. Material handling and pulp washing at larger scales is a critical operation that requires improvement37,59. Three methods were investigated to adapt the protocol for ~ 100 g of pulp: muslin cloth straining, centrifugation and vacuum filtration. Each washing step used 1 mL of ethanol per g of IL. Multiple washing steps with smaller volumes each should lead to better IL removal from the pulp with lower solvent requirements, as the constant partition coefficient for IL between the wash solvent and pulp is multiplied geometrically over multiple cycles. To assess this, the IL content in the solid fraction was measured to track the efficiency of washing steps, and the pulp samples were subjected to saccharification to evaluate the degree of inhibition by residual IL (Figure ESI-15).

Straining using a muslin cloth rapidly reduced IL content, giving acceptable digestibility after only 4 washes. However, further washing could not reduce the pulp IL content to below 6%, which could be problematic for downstream processing. Centrifugation required up to 16 washing steps to reduce the IL content to 8%. Also, glucose yields increased slowly and the pulps appeared clumpy and compressed (Figure ESI-16), which may limit washing efficacy. Vacuum filtration was the most effective, producing fluffy pulp that gave high glucose yields after only 6 washes. A steady decrease in IL content to 4% after 8 washes and < 0.5% after 12 washes was noted. Vacuum filtration with 10 washing steps was found optimal, giving high glucose digestibility, saving ~ 25% ethanol compared to bench scale and eliminating the need for a Soxhlet extraction step.

#### Product recoveries and pretreatment effectiveness

Pulp yield recoveries for finely ground Miscanthus at 10 wt% and 20 wt% loadings (49.7% and 52.7%, respectively, Fig. 6a) were slightly lower at the 1 L scale than at the 10 mL scale (51.6% and 52.9%). Higher lignin precipitate yields, slightly higher delignification, hemicellulose removal, glucan recovery and improved glucose release were also seen at the 1 L scale. The exception being the hemicellulose extraction at 20 wt% loading (Fig. 6a, Table 4). These effects were attributed to improved heat and mass transfer upon the introduction of stirring.

At the 1 L scale, higher solids concentrations (20 wt% loading) increased the surface area for lignin re-precipitation onto the pulp, resulting in lower delignification, lignin and saccharification yields. However, this decrease in delignification and saccharification yield at 20% loading at the 1 L scale (9% and 16% lower, respectively) was less significant than at 10 mL scale (11% and 19% lower, respectively). Quantitative IL liquor recovery was also obtained.

The effect of different particle sizes was also compared for both scales. After scaling up more pronounced differences in product yields were found with increasing particle size (Fig. 6b, Table 4). The decrease in pulp yield upon up-scaling was more pronounced for medium and coarse fractions (around 10% drop at the 1 L compared to the 10 mL scale in both cases) than for fine particles (2% decrease). Lower pulp yields can be explained by improved lignin and hemicellulose extraction, as evidenced by compositional analysis. Stirred scale-up experiments also gave higher lignin precipitate yields, particularly for larger particle sizes.

All the pulps were subjected to saccharification without further size reduction (Table 5). Glucose yields for fine particles (80%) compared closely to those from 10 mL scale (79%). For medium particles a slight increase in glucose yield was found (68% vs 66%). Tellingly, the more drastic improvement was seen for coarse pulps (59% vs 45%), likely due to the improved mass transfer and size reduction during pretreatment with stirring. High uncertainty (~ 18% error) in the glucose yield for coarse particles was attributed to particle size heterogeneity. IL liquor recoveries were improved at the 1 L scale due to the larger solvent volumes used; recoveries exceeding 100 wt% are due to lignin fragments and other non-volatile extractives remaining in the liquor.

#### Insights from lignin HSQC NMR and GPC

The lignin precipitates obtained from stirred scale-up experiments for different loadings and particle sizes were analysed by HSQC NMR (Figures ESI-4, ESI-5 and ESI-6) and GPC. When increasing loading from 10 to 20 wt% at 1 L scale, precipitated lignins showed similar ether cleavage and condensation, as observed by levels of β-aryl ether, G2 and G2,cond sub-units (Figure ESI-9). Different particle sizes, however, resulted in lignin with more distinct properties. For the largest particle sizes with lower volumetric surface area, IL diffusion into the particles and lignin diffusion out of the particles is slower, with lower proton/IL concentration at the particle core, reducing the rate of lignin extraction from the core and producing less depolymerized and less condensed lignins. This was seen from increasing β-O-4′ linkage abundance, increase in S and decrease in Scond sub-units, and increase in S/G ratio as determined by HSQC NMR analysis of lignins extracted from coarse Miscanthus chips (Table 5, Figure ESI-9).

The precipitated lignins also had higher Mw for coarse (6500) than for fine particles (4900, Table 5). The increase in Ð values with particle size (from 4.1 to 5.3) could be explained by less depolymerized lignins being released from the core while smaller lignins still being released from the surface of the particles. Ð values were also higher than at the bench scale (2.7–3.0). These differences highlight the improvement in ionoSolv processing upon scale-up with stirring. Better mass transfer due to mixing may facilitate extraction and dispersion of lignin fragments in the IL.

#### Effect of stirring

The power requirements of mixing are non-negligible and demand optimization of slurry density and viscosity, mixing velocity and agitator design7. Different designs of agitator were tested and the most effective for mixing IL-biomass slurries, an anchor-shaped one, was selected. The effect of stirring speed on pretreatment was assessed by comparing stirring at 150 rpm and the maximum possible speed for a given slurry at 10 wt% loading (450 rpm for fine particles, 250 rpm for the medium particles and 400 rpm for coarse chips). Despite different ‘maximal’ stirring speeds, the maximum stirrer power was employed in all cases, therefore, the amount of energy delivered per second to the IL-biomass slurry was assumed to be the same.

Increasing stirring speed resulted in slightly lower pulp yield and higher lignin precipitate yield, suggesting improved lignin extraction (Fig. 6c). This became more prominent with increasing particle size. However, pulp compositions and saccharification yields were not significantly affected, with the exception of the saccharification yield of the fine fraction, which dropped drastically upon increasing stirring speed from 150 to 450 rpm (from 79.1 to 66.1%, Table 6).

More rapid lignin extraction may take place with greater mixing due to improved heat and mass transfer. However, as fine particles have higher volumetric surface area, this increases the available pulp surface for lignin re-precipitation, worsened by particle size reduction during pretreatment. Consequently, overall delignification remained approximately constant with stirring speed (Table 6). However, for the coarse particle size fraction, increasing the stirring speed from 150 to 400 rpm improved delignification (from 69 to 87%) though glucose yields were very similar (59%). This indicates that mixing speed should be optimized depending on the particle size of the feedstock. The use of high stirring speeds at elevated temperatures for short residence times could help to maximize delignification and avoid prolonged contact of the pulp with the liquor to reduce lignin re-precipitation.

Faster mixing gave rise to greater in situ size reduction, producing pulps with lower average particle size and higher volumetric surface area (Figure ESI-20). D50 values showed a greater size reduction (40, 91 and 96% for fine, medium and coarse particles) than for unstirred experiments at the bench scale (30 and 68% for fine and medium particles).

The lignin precipitates recovered from 1 L scale experiments at different mixing speeds were analyzed by HSQC NMR and GPC. Only subtle differences in lignin characteristics were observed (Figure ESI-8). In all cases, faster mixing speeds produced lignins with lower abundance of β-O-4′ ether linkages, suggesting it is more cleaved. Also, a slight increase in signal intensity for G6 and G2,cond subunits, and lower amounts of G2 and were seen, suggesting that lignins extracted using faster mixing were also more condensed. The implication, based on analysis of the lignin yields and structure, is that faster mixing led to a transition from “diffusion controlled” to “kinetically controlled” reaction regimes, leading to more condensation.

GPC analysis of the lignins (Table ESI-5) showed that increasing mixing speeds produced lignin precipitates with lower molecular weight, which may be due to a greater likelihood of lignin macromolecules precipitating onto the (increased) pulp surface area, though this is unclear. The different results in lignin yield, delignification and MW depending on particle size illustrate the combined effects of lignin reactivity (producing more condensed lignins) and re-precipitation (reducing the proportion of high MW lignins remaining in solution until the water addition step), both of which appear to become enhanced with stirring speeds.

## Conclusions

Process intensification of the ionoSolv process using the low-cost PIL [TEA][HSO4] and the grassy biomass Miscanthus was investigated, including the effects of biomass loading, particle size, 100-fold scaling up and stirring speed.

Five-fold increase in biomass loading (from 10 to 50 wt%) at the 10 mL scale decreased glucose yields from 68 to 23% due to re-precipitation of extracted lignin onto the pulp surface. The use of medium particle sizes (3 × 0.02 × 0.01 cm) was preferable over fine powders (0.18–0.85 mm), due to reduced surface area available for lignin re-precipitation, giving higher glucose yields while minimizing grinding energy needs. Comparison with dilute acid pretreatment showed ionoSolv processing is far more tolerant to use of larger particle sizes.

100-fold scale-up (from 10 mL to 1 L) improved the efficacy of ionoSolv pretreatment, after optimizing pulp washing protocols and agitator design. The introduction of stirring allowed higher delignification, reduced lignin condensation and improved lignin mass balance, particularly for larger particles (~ 3 × 1 × 1 cm). At this scale, doubling the solid loading from 10 to 20 wt% reduced lignin re-precipitation onto the pulp surface, giving higher glucose yields. IL liquor recovery rates were found to be > 99.2% in all cases. Substantial cost savings could be achieved by tailoring the pre-processing size reduction stages. Minimal stirring is preferable to minimize energy requirements.

Process intensification of stirred ionoSolv pretreatment has been demonstrated, showing great potential for further scaling up and optimization towards industrial application. Pulp washing protocols need further improvement to reduce both lignin precipitation and water requirements.