Composite membranes of cellulose–mesoporous silica: optimization of membrane fabrication and adsorption capacity

In this work, we produced composites of mesoporous silica synthesized in-situ on never-dried bleached eucalyptus kraft pulp fibers with the aim of providing cost-effective depth filters, having high flux, and adsorptivity. The mesoporous silica loading for the produced samples was in the range of 12–35 wt%. The performance of double-layer membranes was studied for the adsorption of charged molecules. The best nanofibrillated cellulose-pulp-mesoporous silica membrane adsorbed 1160 mg/m2 of methylene blue and had a flux of 10 L m−2 h−1 bar−1. The nanofibrillated cellulose layer supported the pulp-mesoporous silica layer and improved the adsorptivity of the pulp-mesoporous silica depth filter layer by controlling flux. The membranes showed non-linear-pseudo-first-order adsorption kinetics and non-linear Freundlich isotherm for methylene blue adsorption. The nanofibrillated cellulose-pulp-mesoporous silica membrane was modified for metanil yellow adsorption by adding polyamide amine-epichlorohydrin resin. The best metanil yellow saturated adsorption capacity was 9400 mg/m2. The nanofibrillated cellulose-pulp-mesoporous silica depth filter without modification with a polyelectrolyte also had 92% and 94% heavy metal removal of 20 mg of Cu2+ and Pb2+ ions, respectively. The novel pulp-mesoporous silica composite membrane, with high adsorption capacity and manufactured by lower embodied energy of cellulose fiber, can significantly lower large-scale depth filter production costs due to the elimination of cellulose pre-treatment for the depth filter layer. The reusability performance in the fifth cycle, after four cycles of metanil yellow adsorption and desorption, was 5.2 mg/g, which was stabilized from the 3rd to 5th cycles. This suggests the suitability of these membranes for industrial applications.

they cause severe environmental issues and have a low degradation rate. Therefore, wastewater treatment has been developed to overcome water pollution (Cheng et al. 2020a, b). Common water treatment methods are adsorption (Beyki et al. 2016), filtration (Gebru 2017), precipitation, and photocatalytic degradation to eliminate water contaminants (Dong et al. 2021). However, each of these processes has certain constraints, such as high operating costs and energy requirements (Mansoori et al. 2020). Adsorption using cellulose-based materials is a promising approach for the removal of different water pollutants (Beyki et al. 2014). Cellulose, as the most abundant natural polymer, is an ideal material for use in water treatment material due to its low cost, versatile modifiability (Miri et al. 2021), hydrophilicity, and biodegradability (Nadeem et al. 2020). However, cellulose has low adsorption capacity and needs modifications to improve it (Onur et al. 2019a, b), including charge modification (Tao et al. 2017) and using nanoparticles with high adsorptivity (Gebru 2018). But, prolonged operative duration (Geng 2018a, b), use of toxic reagents (Geng 2018a, b), and high costs (Xu et al. 2018) are the main downsides of cellulose modification on large scales.
Combined adsorption and filtration via cellulosebased depth filters is a facile approach for water treatment (Onur et al. 2018), which performs via both size exclusion and adsorption (Onur et al. 2019a, b). Therefore, using materials with high specific surface area and functional groups (Wang et al. 2019) in cellulose depth filters can be a versatile and environmentally benign approach to enhance adsorption (Onur et al. 2019a, b). Mesoporous silica (MS) has been widely applied in composites for water treatment due to its high saturated adsorption capacity (SAC) and specific surface area (Miri et al. 2021). Mesoporous silica can be considered an attractive material for cellulose depth filter production, adding unique characteristics to the cellulose composite. They have higher SAC than pure cellulose and can easily be modified using polyelectrolytes to allow adsorption of charged molecules (Tao et al. 2017). Following synthesis, a cellulose-mesoporous silica composite can simply be collected via filtration. This contrasts with the powder form of mesoporous silica, which requires additional operations such as centrifugation to concentrate mesoporous silica. This has limited the sample capacity for large-scale composite production (Zhou 2012).
In our previous research, we have investigated the in-situ precipitation of mesoporous silica nanoparticles onto the surface of nanofibrillated cellulose (NFC) fibers, which resulted in the formation of nanoparticles with the median size and surface area in the range of 23-30 nm and 210-567 m 2 /g, respectively (Miri et al. 2021). We then investigated two applications. Firstly, NFC-mesoporous silica nanoparticles aerogels were used for adsorption. However, these were fragile and not scalable due to the high cost of freeze-drying (Miri et al. 2021). The second research area is the combination of an NFC barrier and depth filter layers (Miri et al. 2022). The NFC-NFC-mesoporous silica nanoparticles doublelayer membrane showed excellent adsorption, but the material had low flux hence limiting the grams per square meter (gsm) of the adsorptive layer. Moreover, cellulose mechanical pre-treatment via multi-pass homogenization to produce NFC is an expensive and energy-intensive process for large-scale depth filter manufacturing (Nadeem et al. 2020).
This paper aims to overcome the high cost of using NFC in the depth filter layer by replacing it with conventional wood pulp (BEK) fibers. BEK is a commodity product available worldwide. We describe the implementation of in-situ synthesized BEK composite with mesoporous silica nanoparticles to fabricate double-layer membranes, including the BEKmesoporous silica depth filter and the NFC barrier layer. The structural characteristics and specific surface area of film and aerogel samples were measured to evaluate the impact of the NFC layer on the porosity of the BEK-mesoporous silica layer in the film form. The flux and SAC of NFC-BEK-mesoporous silica membranes were investigated via different gsm of BEK-mesoporous silica samples for methylene blue (MB) adsorption as the cationic model pollutant. The impact of the NFC layer on the flux and adsorption kinetic of depth filters was evaluated. In addition, BEK-mesoporous silica composites modified by polyamide amine-epichlorohydrin resin (PAE) were synthesized to examine the efficiency of modified membranes for adsorption of metanil yellow (MY) molecules with a negative charge. Moreover, the metal removal efficiency of NFC-BEK-mesoporous silica membranes with and without modifications with polyelectrolytes was investigated. The reusability of the NFC-BEK-mesoporous silica membrane was evaluated toward MY solution over five cycles. The embodied energy for producing BEK and NFC samples required for loading 1 kg mesoporous silica was also evaluated.

Materials and method
BEK is never-dried unrefined industrial virgin hardwood pulp ) that was supplied by Australian Paper, Maryvale, Australia with an approximate solid content of 14.65 wt%. The approximate width and length of BEK fibers were 14.5 μm and 1220 μm, respectively (Ang et al. 2019). The freeness of BEK pulp was 630 ml Canadian Standard Freeness (CSF) (de Assis 2019). The content of cellulose, hemicellulose, and lignin in the BEK pulp were approximately 80%, 17%, and 3%, respectively (Ang et al. 2020). We did not measure the change of hemicellulose in BEK when it was soaked at pH = 11.4.
The bulk density ( bulk ) of NFC and BEK composite films was calculated by measuring the weight, length, width, and thickness of films. The skeletal density ( skeletal ) of samples was measured using the helium gas pycnometer, Micromeritics AccuPyc 1330 pycnometer (Norcross, GA, USA). The porosity ε (%) was calculated as follows: Preparation of NFC and mesoporous silica nanoparticles Nanofibrillated cellulose (NFC) was homogenized according to the procedure described in our previous report (Miri et al. 2021). The NFC (1.5 wt%) was prepared by homogenizing the microfibrillated cellulose suspension (1.5 wt%) at 1000 bar pressure for five passes using a Pony homogenizer, GEA, Italy. The median size of NFC fibers diameter was 34.7 nm (Miri et al. 2021). The NFC (1.5 wt%) was diluted to 0.2 wt% with deionized water for paper making. The mesoporous silica nanoparticles sample was synthesized separately (for the X-ray photoelectron spectroscopy (XPS) analysis and zeta potential measurement) according to the procedure outlined in our previous report (Miri et al. 2021).
In-situ precipitation for preparation of BEK-mesoporous silica and NFC-mesoporous silica nanoparticles composites The suspension of BEK was prepared by adding 10 g (dry weight) of BEK in 5 L of deionized water and mixing it with 10 g of CTAB solution as the pore-directing agent. The pH was adjusted to 11.4 by adding 2 M NaOH. Then, 50 ml of TEOS was added to the suspension dropwise for 30 min while stirring, followed by vigorously stirring for another 30 min. Afterward, the prepared composite was filtered through a Buchner funnel, rinsed three times with deionized water, and heated to reflux with ethanolic HCl (1.5% v/v of HCl 32% in EtOH) for 4 h to extract CTAB. The suspension was filtered through a Whatman filter paper on the funnel and rinsed with deionized water to neutralize its pH and eliminate all unbound chemicals. The sample was stored at 4 °C and labelled the B1 composite. The same procedure described above for the B1 composite was adapted to prepare two other types of BEK-mesoporous silica composites (Table 1), differing in the amounts of TEOS and CTAB that were added to the reaction medium. The B2 sample was dried in an oven at 105 °C overnight for the BET specific surface area (S BET ) measurement. The NFC-mesoporous silica (%) = 1 − bulk skeletal nanoparticles composite preparation was carried out as outlined in our previous report (Miri et al. 2021) ( Table 1). The synthesis conditions for N1, N2, and N3 samples were the same as BEK-mesoporous silica samples, except NFC was used instead of BEK (Miri et al. 2022). By adding more amount of TEOS, the yield of mesoporous silica did not change significantly until 50 ml of TEOS was added to BEK. This suggests saturation of the surface of BEK with mesoporous silica by adding 50 ml of TEOS, which led to the lower yield of mesoporous silica.

Preparation of single-layer membranes
To form the pristine NFC membrane using a British hand sheet maker via vacuum filtration, the NFC suspension (0.2 wt%, including 1.2 g of the dried NFC sample) was poured into the hand sheet maker chamber on a filter paper (Whatman 542 Hardened Ashless) placed on a 150-mesh screen in the chamber. After draining the suspension, the wet film was placed between two blotting papers and two steel plates. Next, the membrane was obtained after pressing at 3.5 bar pressure for 7 min and then drying using an automatic sheet dryer at 112 °C. The B3 singlelayer membrane (6.75 dry g, 340 gsm) was diluted to 2 L, poured into the hand sheet maker chamber, and drained through the filter paper placed on the micrometer-sized mesh to form sheets uniformly. The B3 membrane was prepared by placing it in an oven at 105 °C for 2 h. The B1 single-layer membrane was also produced by following the procedure for the B3 membrane preparation. Drying films at 112 °C using the automatic sheet dryer is related to preparing the single-layer pristine NFC membrane. Using the automatic sheet dryer created some cracks in the B1 composite films by rolling them during the drying process due to the high gsm of the films and high content of mesoporous silica. Thus, drying of BEK films was performed at the oven under the common temperature for drying of cellulose samples at 105 °C. Single-layer and double-layer BEK composite films were dried at the same condition and temperature using the oven at 105 °C. Drying BEK composite films at room temperature took several days, which was not suitable in terms of the sample preparation duration. In addition, the porosity of the B3 film dried in the oven at 105 °C and 112 °C led to the same porosity of films at both temperatures. The lower temperature of 105 °C was used for drying of BEK composite films as this is the standard temperature to measure the moisture content of cellulose.

Preparation of double-layer membranes
A total of 400 gsm was selected for the NFC-BEKmesoporous silica double-layer membrane preparation. Since pouring NFC first can increase the draining step duration, the B3 suspension (340 gsm) was poured into the hand sheet maker first to prepare the NFC-B3-340 membrane. The B3 sample (6.75 dry g, 0.2 wt%) was provided (340 gsm) and poured into the chamber on the filter paper placed on the mesh. After draining the B3 suspension, the NFC suspension was poured into the chamber on top of the drained B3 composite film. Consequently, pressing and drying steps of the membrane were carried out as described for the B3 membrane production. The NFC-B3-370 membrane was produced by pouring 370 gsm and 30 gsm of the B3 and NFC suspensions into the vacuum chamber. Figure 1 displays the scheme of an NFC-BEK-mesoporous silica double-layer membrane. Details for preparing the NFC-N3-18 double-layer membrane via sequential pouring of 60 gsm and 18 gsm of NFC and N3 suspensions were provided in our previous report (Miri et al. 2022). The prepared  Table 2. Some samples were also modified with PAE or PEI.

Additional chemical treatments
Additional treatments were used for some samples to improve the adsorption of MY molecules and metal ions. NFC (1.2 dry g) and BEK (6.75 dry g) suspensions were blended with 5 mg and 30 mg dry amounts of PAE, respectively. The ratio of PAE to NFC and B1 suspensions was equal to 1.4 mg/g and 1.5 mg/g, respectively. The B1 sample (340 gsm) was mixed with 1.5 mg/g PAE (the B1-340PAE1.5 sample) and poured into the British hand sheet maker chamber. After draining the B1-340PAE1.5 sample, the NFC suspension was mixed with 1.4 mg/g PAE and added to the top of the B1-340PAE1.5 film in the vacuum chamber. Draining and pressing samples were followed as outlined for single-layer membrane production. Next, it was placed between blotting papers and  placed in an oven at 105 °C for 2 h. The sample was labelled the NFC-PAE-B1-340PAE1.5 membrane ( Table 2). The B1 suspension (340 gsm) was mixed with 1.5 and 4.4 mg/g PEI and labelled B1-340PEI1.5 and B1-340PEI4.4 composites. Next, NFC-B1-340PEI1.5 and NFC-B1-340PEI4.4 double-layer membranes were prepared as outlined for single-layer membrane production ( Table 2).
The nomenclature for NFC-BEK-mesoporous silica double-layer membranes is of the form NFC-BX-Z-OY, where X is the conditions for preparing the BEK-mesoporous silica composite layer, with X = 1, 2, or 3 from Table 1 and Z gives the BEKmesoporous silica composite layer gsm. The terms O and Y provide additional treatments that were applied for some of the samples and the amount of the polyelectrolyte per gram of the composite layer. Thus, we denoted the NFC-B1-340PAE0.5 specimen to a sample containing a 60 gsm NFC layer with the second layer of the BEK-mesoporous silica suspension, where 340 gsm of the B1 suspension was mixed with 0.5 mg/g of PAE.

Adsorption and flux experiments
The membranes were cut into circular discs and put on the bottom of a dead-end cell with an effective membrane area of 0.0045 m 2 (Merck Millipore dead-end stirred cell, model UFSC40001). The NFC barrier layer was used as the top layer for filtration to support the BEK-mesoporous silica composite layer during filtration (Fig. S1). Experiments were done at room temperature. The UV-Vis flow cell connected to Agilent Cary 60 UV-Vis spectrophotometer (Santa Clara, CA, USA) and the dead-end cell were used to measure the UV absorbance of the filtrate within specific time intervals at 664 nm and 450 nm for MB and MY solutions, respectively. The balance was connected to a computer and recorded the weight of the filtrate at specific time intervals via LabX Direct Balance Software. The flux was measured by the permeated volume per area and time normalized by the applied pressure during filtration (L m −2 h −1 bar −1 ).
Metal adsorption tests were performed for 50 mg L −1 of Cu 2+ and Pb 2+ ions. Ionic solutions were prepared in a volumetric flask. The membrane was set as described above in a dead-end cell. A 400 ml solution of either Cu 2+ or Pb 2+ ions was added to the dead-end cell and filtered under 1 bar pressure. The filtrate was diluted to 1-4 mg L −1 using a volumetric flask (the dilution is required for Inductively Coupled Plasma-Mass Spectroscopy (ICP-MS) experiments). Afterwards, the concentration of ions was measured using the ICP-MS instrument performed using the triple quadrupole instrument iCAP TQe ICP-MS (Thermo Fisher Scientific, Bremen, Germany).
Regeneration of the NFC-B1-340PAE0.5 membrane Adsorption-desorption experiments were carried out five times in total by MY filtration to investigate the regeneration performance of the NFC-B1-340PAE0.5 membrane. The SAC of membranes after each cycle was measured. The same procedure described for filtration through double-layer membranes was used for MY adsorption through filtration with the initial concentration of 5 ppm. The membrane with MY adsorbed was immersed into a 200 ml solution containing 0.5 M HCl (EtOH/H 2 O 1:1 v/v) for 1 h. Afterward, the membrane was removed from the solution and rinsed with deionized water. The same approach mentioned above was used for the next regeneration cycles of the membrane for MY filtration.
Desorption of MB from the NFC-B1-340 membrane Desorption of MB (5 ppm) from the NFC-B1-340 membrane was examined after its first adsorption run by filtering 0.5 M HCl (EtOH/H 2 O 1:1 v/v) from the membrane under 2 bar pressure in the dead-end cell.

Results and discussion
Characterization of composites Several techniques were used to characterize BEKmesoporous silica composites and their membrane form. SEM was applied to visualize mesoporous silica precipitated on BEK fibers. Sample comparisons were made using atomic concentration percentage (at%) of chemical bonds from the X-ray photoelectron spectroscopy (XPS) survey scan. In addition, nitrogen adsorption-desorption isotherms estimated mesoporosity of composites and film samples. The porosity of membranes was investigated using the helium pycnometer.

SEM micrographs of composite fibers
The BEK fiber is presented in Fig. 2a with a median fiber diameter of 11 µm, which is significantly larger than NFC fibers with a 34.7 nm median diameter (Miri et al. 2021). Zoomed-in SEM micrographs of BEK fibers in B2 and B3 samples illustrate that MS particles are bonded on their surface (Fig. 2b-f). ImageJ software was used to estimate the median particle size of 90 nm in the B3 composite ( Fig. 2e-f). However, particles in the B2 sample ( Fig. 2b-d) were significantly larger compared to the MS size of the B3 sample ( Fig. 2e-f) due to the higher TEOS concentration in the suspension for in-situ reaction. The median size of particles in the B3 sample was 3.67 times larger than those of the N3 composite (Miri et al. 2021), while the added TEOS amount in both samples was the same. The higher aggregation and coating of mesoporous silica on fibers was observable on BEK fibers in the B2 sample. Therefore, we use the term mesoporous silica for the mesoporous silica particles on BEK samples. The mesoporous silica nanoparticles term was used for the NFC composites due to the formation of the mesoporous silica nanostructure in NFC samples. Furthermore, some small particles were observable inside BEK fibers in the B2 sample (Fig. S3).

The porosity of samples and MS loading
Loading results from the calcination of BEKmesoporous silica composite samples prepared directly after precipitation and washing are presented in Fig. 3a. To evaluate the difference in MS loading using BEK and NFC samples, BEK-mesoporous silica composites were prepared using the same molar  (Miri et al. 2021). MS wt% of B3, B2, and B1 composites were 12%, 23%, and 35%, respectively, while the MS loading of N3, N2, and N1 composites were 20%, 37%, and 46%, respectively (Miri et al. 2021). The mesoporous silica content of BEK-mesoporous silica specimens was lower than corresponding NFC-MSN composites synthesized.
The MS wt% ratios of BEK-mesoporous silica composites to NFC-mesoporous silica nanoparticles ones synthesized by the same cellulose: TEOS molar ratios were in the range of 62-76%. The lower MS loading of BEK-mesoporous silica composites compared to NFC-mesoporous silica nanoparticles is attributable to the shrinkage of BEK fiber with large diameters that decreased the retention of small particle sizes hydraulically on fibers, the much lower surface area of the BEK fibers, compared to the NFC fibers and dissociation of larger mesoporous silica particles (Fig. 2c-d). The difference in loading of mesoporous silica on BEK and NFC was relatively small, while the fiber diameter of NFC is significantly smaller than BEK fibers. The high content of mesoporous silica on the BEK fibers despite the much lower surface area is because the MS must be precipitated inside of BEK fibers as well, which is observable in Fig. S3.
The correlation of porosity of membranes to their flux is a critical factor to investigate for filtration processes. In addition, the porosity of the B3 film dried in the oven at 105 °C and 112 °C led to the same porosity of films at both temperatures. The lower temperature of 105 °C was used for drying of BEK composite films to reduce the cost of energy for film making. The porosity of the BEK-mesoporous silica single-layer membrane is significantly higher than the NFC membrane because of the larger fibers of BEK. The higher porosity of the B1 sample compared to the B3 specimen is attributed to the higher MS amount in the B1 sample, which reduces the packing of the BEK fibers. NFC-B3-340 and NFC-B1-340 membranes displayed slightly lower porosity compared to the single-layer BEK-mesoporous silica membranes (B1 and B3 membranes, Fig. 3b). Since we added a high porosity layer to the low porosity NFC layer, the composite will have high overall porosity. In addition, the unmodified B1 film displayed 1.2 times higher porosity compared to the NFC-PAE-B1-340PAE sample because of cross-linking of fibers in the NFC-PAE-B1-340PAE sample. The BEK-mesoporous silica membrane characteristics

Surface morphology
The SEM micrographs of the B1 membrane from surface and cross-section views were provided. The high porosity among BEK fibers and MS particles is noticeable (Fig. 4Aa-c). In addition, the particles on the surface of the BEK fibers of the B1 sample revealed a broad size distribution of particle sizes (Fig. 4Ad). Cross-linking of fibers with PAE reduced the thickness of the BEK-mesoporous silica film correlated to the lower roughness of the B1-340PAE0.5 film. The roughness of the NFC layer (air side or the opposite side of the composite layer) and composite layers were investigated by a profilometer. The average roughness of the B1 film (r a = 14) was higher than the B3 film with r a = 5. The presence of the higher MS amount contributes to the higher roughness of the B1 film than the B3 one. However, the air side of r a of the NFC layer of the NFC-B1-340 membrane was 1.7 times lower than the BEK-mesoporous silica side, which is due to the highly uniform surface of the NFC layer compared to the surface of the BEKmesoporous silica layer with a high mesoporous silica amount (Fig. 4B). This result is consistent with the porosity of BEK-mesoporous silica and NFC films (Fig. 3).
The drainage time for BEK-mesoporous silica film fabrication and thickness and weight loss of BEKmesoporous silica films during filtration.
The drainage of the pristine NFC film took 7 min, while the drainage time of the B1-340 was performed for one min, which was much quicker in comparison to NFC. In addition, the drainage of the B1 composite on the NFC layer to form the NFC-B1-340 composite film increased to 16 min, suggesting the impact of the NFC layer on the drainage time of the double-layer membrane. However, there was not an observable difference between the drainage times of the B1 and B3 composites on the NFC layer to fabricate their corresponding double-layer membranes. The thickness of the NFC-B1-340 membrane was 875 µm, while the thickness of the B1 layer (340 gsm) was 815 µm. The weight loss of the NFC-B1-340 membrane was 5 wt% after the MB filtration. The weight loss of the NFC-B1-340 membrane after five times filtration increased from 5% after the 1st run to 7% after the 5th run. The weight loss of 5 wt% after the first run of filtration suggested the separation of large particles in the BEK-mesoporous silica structure during the filtration. However, the remained particles for the next runs of filtration can be in the smaller median particle size. This suggests the NFC-B1-340 membrane was stable during filtration and had a slight change in the membrane weight after more runs of filtration. The weight loss of the NFC-BEK-mesoporous silica membranes can be explained by the competition of hydrogen bonds of water with hydrogen bonds among MS structures and BEK fibers, which led to the detachment of larger particle sizes in the B1 composite.

Adsorption performance and flux
Adsorption of organic charged molecules and flux of composite membranes Figure 5 presented the data for flux and SAC of BEK-mesoporous silica membranes for MB and MY adsorption. Given that the samples have different sample gsm, we have chosen to present the data as flux per unit area per bar pressure and as the total amount of dye adsorbed per unit area. For comparison, the NFC-N3-18 double-layer membrane is also included in Fig. 5.
The B3 single-layer membrane had a flux of 211 L m −2 h −1 bar −1 due to the high porosity between the BEK fibers (Fig. 5). Adding a 30 gsm NFC layer in the NFC-B3-370 membrane reduced flux to 114 L m −2 h −1 bar −1 , while increasing the NFC layer to 60 gsm in the NFC-B1-340 membrane reduced the flux further to 10 L m −2 h −1 bar −1 . The flux of the NFC-B1-340 membrane was higher than that of the single-layer NFC membrane, which can be correlated to the BEK-mesoporous silica double-layer membrane fabrication since the NFC suspension was filtered on the BEK-mesoporous silica layer, which The saturated adsorption capacity (SAC) of different samples was investigated by measuring the amount of dye adsorbed per area (mg/m 2 ). The B3 and NFC-B3-370 membranes displayed lower SAC (143 and 480 mg/m 2 , respectively) than the NFC-B3-340 membrane (1016 mg/m 2 ), even though they had the same or slightly higher masses of adsorbent per unit area. The difference was not due to the addition of the NFC membrane, which had the lowest SAC value of all the membranes tested (31 mg/m 2 ), but rather was because the reduction in flux has reduced channeling through the filter and increased the efficiency of the adsorbent.
A comparison of the NFC-B1-340 and NFC-B1-370 results shows the versatility of the filters produced here. Changing the NFC layer from 60 to 30 gsm, while increasing the BEK-mesoporous silica depth filter from 340 to 370 gsm, has halved the SAC, but increased the flux by over a factor of 10. The filters produced without chemical modification have a good performance for the adsorption of cationic molecules. Moreover, the NFC-N3-18 membrane exhibited a SAC of 399 mg/m 2 , which was remarkably lower than the SAC of the NFC-B1-340 membrane.
PAE was added to NFC and B1 layers in the NFC-PAE-B1-340PAE1.5 membrane to investigate its impact on the adsorption of MY molecules with the negative charge. Since the BEK-mesoporous silica composite is negatively charged, the charge modification of the composite is required to adsorb anionic pollutants (Miri et al. 2022). The film modified with the positively charged PAE can adsorb molecules with negative charges, which improves the saturated adsorption capacity of films. In addition, using PAE in the BEK composite layer reduced the flux of films during filtration due to improving wet strength of films, which enhanced the adsorption efficiency of films. PAE has contribution to cross-link BEK fibers. Hence, it does not have an impact on fixing particles on BEK fibers.
The addition of PAE was investigated by its mixing with the B1 layer. Three samples prepared by adding three different amounts of PAE (0.5, 1.2, and 1.5 mg/g) to the B1 layer were provided. Since NFC layer has high wet strength due to the high surface area of its fibers, the addition of different amounts of PAE to the NFC layer was not investigated. There was one sample that adding PAE to two layers was investigated as per the method of our previous study (Miri et al. 2022). This decreased the porosity of the NFC-PAE-B1-340PAE1.5 membrane decreased by around 20% compared to the NFC-B1-340 membrane through PAE cross-linking. We found previously that adding PEI as the positively charged polyelectrolyte in the NFC membrane did not result in a high SAC toward MY (Miri et al. 2022). Therefore, PAE adsorbed on the BEK fiber surface does not contribute to MY adsorption due to the very low BEK surface area. The presence of polyelectrolytes on high specific surface area materials increased their electrostatic interactions with charged molecules due to more accessibility of their functional groups for electrostatic interactions (Miri et al. 2022;Zhao et al. 2018). The SAC of the NFC-PAE-B1-340PAE1.5 membrane for MY feed solution was 9400 mg/m 2 . Hence, higher adsorption of MY molecules in the NFC-PAE-B1-340PAE1.5 sample can be due to the presence of PAE on MS with a high specific surface area, which increased its electrostatic interactions with MY molecules. PAE also increased the positive charge (Fig. 6) of the B1 layer, which can provide approaching MY molecules to pores of MS via electrostatic interactions (Tao et al. 2017). However, the NFC-PAE-B1-340PAE1.5 membrane exhibited a noticeable lower SAC toward MB compared to the NFC-B1-340 membrane attributed to the electrostatic repulsion of PAE and MB molecules. These observations are confirmed in Figs. S4-S5 in the Supporting Information (Sect. S2.3), which shows pictures of different membranes after MB filtration. Fig. S5 S6). Decreasing the added PAE amount from 1.5 to 0.5 mg/g to the BEK-mesoporous silica layer also resulted in the low adsorptivity of MB molecules.

Comparison with other membranes
The NFC-N3-18 membrane had low water flux despite using a low gsm of the depth filter (Miri et al. 2022). Therefore, the amount of the depth filter that can be added is very limited in NFC-NFC-MSN double-layer membranes. In comparison, NFC-BEK-mesoporous silica membranes had higher flux, despite the much higher gsm of BEK-mesoporous silica depth filters that were used compared to NFC-NFC-MSN membranes. Therefore, the BEKmesoporous silica depth filter can be more effective for industrial applications due to having both higher flux and gsm. Using BEK also eliminates further processing of cellulose, such as refining and homogenization for NFC production (Miri et al. 2022;Sehaqui et al. 2018). Thus, using BEK in the depth filtration instead of NFC is cost-effective and requires the least energy input compared to NFC. Moreover, vacuum filtration of NFC-BEKmesoporous silica double-layer membranes was much quicker. In fact, NFC-NFC-mesoporous silica nanoparticles membranes were fabricated by employing the NFC layer as the substrate of the NFCmesoporous silica nanoparticles layer under vacuum (Miri et al. 2022). Small median NFC fiber diameters and the slow settling of the NFC-mesoporous silica nanoparticles layer on the drained NFC layer with its low porosity created long drainage of NFC and NFC-mesoporous silica nanoparticles layers. However, the drainage of the BEK-mesoporous silica suspension as the substrate layer to prepare the NFC-BEK-mesoporous silica membrane occurred quickly because of their large fiber diameter. The BEKmesoporous silica layer also had high porosity, which facilitated the drainage of the NFC suspension as the second layer. Thus, the production of NFC-BEKmesoporous silica double-layer membranes reduced time and energy consumption during vacuum filtration, which is significantly important for large-scale sustainable composite production (Shanmugam et al. 2021).
There have been a number of other cellulosebased filters described in the literature. Cheng et al. used electrospinning to produce the cellulose acetate nanofibrous membranes modified by polydopamine for MB adsorption (Cheng et al. 2020a, b), but the electrospinning method has drawbacks for large-scale film production (Bhagwan et al. 2019) and requires using organic solvents (Gopakumar et al. 2017). In addition, the bamboo-based membrane was produced by adding cellulose to synthesized 1-butyl-3-methylimidazolium chloride ionic liquid (Esfahani et al. 2020). The limitations of the bamboo-based membrane include the long-time reaction for the ionic  (Table 3). However, it is not applicable for a continuous filtration process. In comparison, the BEK-mesoporous silica composite membranes produced here are readily scalable without requiring the use of toxic solvents or requiring multi-step modifications (Tao et al. 2017;Chen et al. 2020). In addition, it can be collected easily after synthesis.

Kinetic study
The adsorption kinetics of samples were evaluated by recording adsorption capacity (Q t , mg/g) at different time intervals of dye filtration. The B3 membrane had negligible adsorption due to its high flux. The NFC-B3-370 membrane also revealed low SAC than the NFC-B3-340 one due to its 7.9 times higher flux. Hence, using the NFC layer is crucial to control the flux of BEK-mesoporous silica depth filters and reach higher SAC values. The NFC-B1-340 membrane reached the equilibrium adsorption with a higher kinetic rate than the NFC-B3-340 sample, which correlates to its higher mesoporous silica content. Hence, using a high mesoporous silica amount in the BEKmesoporous silica sample can lead to more efficient adsorption performance. Although the NFC-N3-18 membrane exhibited a high SAC value, it reached equilibrium adsorption slower than the NFC-B1-340 membrane, which is attributable to its lower flux. Thus, using the BEK-mesoporous silica depth filter is more efficient for industrial applications due to its faster kinetic rate, high flux, and application of higher gsm (Fig. 7). We fitted the experimental breakthrough curves using non-linear pseudo-first and pseudo-secondorder models to quantify changes in adsorption with time. As illustrated, the non-linear pseudo-first-order kinetic has higher R 2 and closer Q e values to the experimental Q e (Q e,exp ) (Table S4). Thus, the adsorption mechanism of charged molecules to the NFC-B1-340 membrane was physical. The equations of Non-inear pseudo-first order fit NFC-B1-340 (MB-5) desorption models and fitted parameters are presented in Sect. S1.6 and Table S4, respectively. Desorption of MB molecules from the NFC-B1-340 membrane after the first adsorption run was also investigated to compare it with the adsorption of MB molecules. Desorption of MB molecules after the first adsorption cycle occurred over 0.4 h, which was faster compared to its adsorption. The fast desorption of MB molecules after the first run of adsorption can be correlated to the electrostatically adsorbed MB molecules to MS hydroxyl groups (Danshvar et al. 2017).

Adsorption isotherm
The amount of adsorption at equilibrium (Q e , mg/g) of the NFC-B1-340 membrane from experimental results was examined by changing the initial concentration of MB solution ranging from 5 to 30 mg L −1 (Sect. S1.7). The experimental Q e (mg/g) versus the initial concentration of the MB solution was plotted in Fig. 8 using the non-linear Freundlich adsorption isotherm. As can be observed, the experimental Q e of the membrane increased significantly from 2.9 mg/g for 5 mg L −1 to 25.2 mg/g for 30 mg L −1 initial concentration. R 2 values of non-linear Freundlich and non-linear Langmuir models were 0.9941 and 0.7762, respectively. Fitted values are in section S2.4 (Table S5). Hence, the non-linear Freundlich model was a better fit for the experimental data and can describe the adsorption process. The Freundlich model suggested the heterogeneous adsorption sites and multilayer adsorption of the NFC-B1-340 membrane (Khani et al. 2018) (Fig. 8).

Metal adsorption and size exclusion
Unmodified and modified membranes were tested for Cu 2+ and Pb 2+ adsorption via filtration. The two unmodified membranes (NFC-B1-340 and NFC-B3-340 films) had the best metal absorption efficiency, with the NFC-B1-340 having an efficiency above 90% for both metal ions. Modifying the membrane with either PAE or PEI reduced the performance, as would be expected, since the metal ions and the polyelectrolytes are both positively charged. The higher adsorption efficiency was obtained by the unmodified film (NFC-B1-340 and NFC-B3-340 films), which revealed the importance of the surface charge of the film to reach higher metal removal efficiency. However, their removal efficiencies did not reach 100% due to the high flux of the unmodified films. Decreasing flux improved adsorption efficiency toward negative charged molecules by using PAE as a positive charged polyelectrolyte and cross-linker. However, the positive charged film can lead to the electrostatic repulsion of metal ions and the positive sides of the film. Hence, the films that are modified with negative charged molecules and polymers can improve metal ion adsorption.
The samples prepared by adding APTES had very poor performance (Table S1). The lower metal adsorption of the NFC-B1-340-AP0.6 film compared to other samples could either be related to the high  flux during filtration or the reduced performance of the surface chemistry of the APTES modified MS structure. Hence, adding APTES is not recommended for use in this system (Table 4). In our previous report, we indicated that the size exclusion of double-layer membranes was improved using the NFC layer with low porosity (Miri et al. 2022). We tested the NFC-B3-340 membrane for size exclusion of PEG (20 kDa) and found the size exclusion of the pristine single-layer NFC membrane and the NFC-B3-340 double-layer membrane was 13.7% and 47.8%, respectively. Since the BEK-mesoporous silica layer did not reduce the size exclusion performance of the NFC layer, using the NFC layer in BEKmesoporous silica depth filters can be beneficial to have size-exclusion performance for water treatment applications.

Regeneration study
Regeneration of membranes is critical for the costeffectiveness of the water treatment processes. Desorption of MY molecules was investigated from the NFC-B1-340PAE0.5 membrane using ethanolic HCl solution over five cycles. Releasing MY molecules from the NFC-B1-340PAE0.5 membrane occurred over 1 h with 8.2 mg adsorbed per g of the total weight of membrane (mg/g) and 6.8 mg/g adsorption capacity in the first and second runs, respectively. Decreasing the adsorption capacity of the membrane toward MY molecules after the first regeneration cycle can be explained by the protonation of OH groups of MS particles by the acidic eluent during the regeneration of the membrane after adsorption into the acidic solution and the saturation of available binding groups of MS in the membrane by MY molecules (Daneshvar et al. 2017). In addition, 5 wt% of the weight of the membrane after the 1st run of filtration due to the loss of larger mesoporous silica particles could lead to the lower adsorption capacity of the film for the next runs of recycling. Since the weight loss from 2nd to 5th runs was 2%, this can also be related to the similar adsorption capacity of 3rd-5th runs.
The fast desorption of MY molecules suggests MY molecules could be electrostatically interacting with PAE molecules adsorbed onto the MS surface. Indeed, PAE adsorbed on MS mesopores can have channel structures (Miri et al. 2022). These formed channels of the polyelectrolyte provided high surface area (Zhao et al. 2018) and accessibility for its electrostatic interactions with MY molecules. This result suggests the efficient structure of the membrane for the regeneration of cationic molecules. The desorption capability of anionic MY molecules can be useful in utilizing BEK-mesoporous silica membranes for loading drugs, wound-healing applications, and adsorption of nutrients to release them as a fertilizer. The ratio of desorbed MY/adsorbed MY over five runs are consistent with decreasing the desorption efficiency from the second run (Fig. 9). The similar adsorption of runs 3-5 can be explained by decreased flux due to the organic load in the membrane (Cainglet et al. 2021). The low flux of the recycled membrane could provide MY adsorption to access binding sites in the membrane during filtration. The decreasing desorption of MY/adsorption of MY ratio is correlated to the remaining content of MY molecules in the membrane after the second run of adsorption. This also can explain the low ratio of desorption/ adsorption of cycles 3-5, while the adsorption of these cycles remained the same. Because the desorption process was via immersion for 1 h, longer desorption via immersion into the solvent can also enhance desorption efficiency after the second cycle.

Embodied energy
We investigated the energy consumption of producing NFC and BEK fibers to consider their environmental impact on composite production. The methods and materials to calculate the overall embodied energy for fabricating NFC and BEK fibers were tabulated in Table 5. We selected a functional of 1 kg of mesoporous silica in a BEK or an NFC composite. Setting a functional unit for comparison of embodied energy of BEK and NFC samples used in composite formation is difficult, since the same amount of MS in a BEK composite has slightly poorer adsorption performance than in an NFC composite. The different flux of NFC and BEK composite membranes also impacts the different performances of MS in BEK and NFC membrane samples. In addition, there will be a significant lower pressure drop of the BEKmesoporous silica composite membrane for a given level of adsorption compared to the NFC-MSN composite membrane. The embodied energy was calculated based on the amount of the cellulose fiber for loading of 1 kg mesoporous silica. The embodied energy of the BEK production was 72.7 MJ, which was 391.9 MJ lower than the embodied energy of producing the NFC sample. This suggests that the BEK-mesoporous silica composite production for membrane formation can be more environmentally friendly. The refining pre-treatment and homogenization of microfibrillated cellulose to produce NFC are energy-consuming processes for the NFC-mesoporous silica nanoparticles composite preparation, which can increase the total embodied energy of the NFC-mesoporous silica nanoparticles production compared to the BEK-mesoporous silica composite.

Conclusion
Double-layer membranes, including an NFC barrier layer and BEK-mesoporous silica composites as the depth filter layer, were fabricated using vacuum filtration. Different MS content in BEK-mesoporous silica samples was provided through the in-situ synthesizing. The BEK-mesoporous silica composite film displayed a noticeable S BET and pore volume compared to the NFC-MSN films in our previous report,