Two-Step Size-Exclusion Nanofiltration of Prothrombin Complex Concentrate Using Nanocellulose-Based Filter Paper.

Coagulation Factor IX-rich protrhombin complex concentrate (FIX-PCC) is a therapeutic biologic product that consists of a mixture of several human plasma-derived proteins, useful for treating hemophilia B. Due to its complex composition, FIX-PCC is very challenging to bioprocess through virus removing nanofilters in order to ensure its biosafety. This article describes a two-step filtration process of FIX-PCC using a nanocellulose-based filter paper with tailored porosity. The filters were characterized with scanning electron microscopy (SEM), cryoporometry with differential scanning calorimetry, and nitrogen gas sorption. Furthermore, in order to probe the filter's cut-off size rejection threshold, removal of small- and large-size model viruses, i.e., ΦX174 (28 nm) and PR772 (70 nm), was evaluated. The feed, pre-filtrate, and permeate solutions were characterized with mass-spectrometric proteomic analysis, dynamic light scattering (DLS), sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE), and analytical size-exclusion high-performance liquid chromatography (SEHPLC). By sequential filtration through 11 μm pre-filter and 33 μm virus removal filter paper, it was possible to achieve high product throughput and high virus removal capacity. The presented approach could potentially be applied for bioprocessing other protein-based drugs.


Introduction
Replacement therapy using plasma-derived Factor IX (FIX) products is a life-saving treatment for patients with hemophilia B. Both recombinant and plasma-derived FIX show high efficacy in clinical trials [1]. Production of FIX normally involves multiple steps. High purity FIX is obtained from prothrombin complex concentrate (PCC), which is a mixture of vitamin K-dependent clotting factors, e.g., factor II (prothrombin), V, VII, IX, and X, and clotting inhibitors, e.g., protein C, Z, and S [2]. PCC preparation is a highly complex mixture of proteins and may contain up to 50% of proteins other than FIX [3]. Both highly purified FIX and PCC can be used for hemophilia B treatment [4]. Also, PCC preparation may be useful for prevention of bleeding due to overdose of oral anticoagulants and liver dysfunctions [2,3].
In this article, for the first time the filtration of FIX-rich PCC using a nanocellulose-based virus removal filter paper is described. Furthermore, a two-step size-exclusion nanofiltration process is developed to remove foulants and ensure efficient virus removal filtration of FIX-rich PCC using nanocellulose-based virus removal filter paper. FIX-rich PCC was used as a model for a highly complex plasma-derived product to simulate industrial bioprocesses where impurities may greatly affect product yield and biosafety.

Filter Preparation
Filters of different thickness were prepared from Cladophora cellulose dispersion (0.1 wt.%) made by microfluidization with 200 µm (twice) and 100 µm hole sized chambers at 1800 bar using LM20 Microfluidizer (Microfluidics, Westwood, MA, USA). Furthermore, the wet cake was made by draining the dispersion over a membrane (Durapore, 0.65 µm hydrophilic PVDF "DVPP", Merck Millipore, Burlington, MA, USA) in a funnel, driven by vacuum. Obtained cellulose cakes were dried at 140 • C to produce pre-filters and 80 • C for filter papers using hot-press (Carver Model 4122CE, Carver, Wabash, IN, USA).

Dissolution of Factor IX-rich PCC
Lyophilized FIX-rich PCC samples were reconstituted by dissolving in phosphate-buffered saline (PBS). No visible particles could be seen after reconstitution, and the solution was clear and transparent. Upon dissolution, the conductivity and pH values were 15.4 mS cm −1 and 7.4, respectively.

Filtration Setup
Pre-filtration and filtration steps were performed in a 47 mm diameter Advantech KST 47 filter holder. Prior to filtration, the pre-and filter papers were wetted in order to extrude the air by running 20 mL of PBS. The pre-filtration steps with 6 and 11 µm pre-filters were performed at 1 bar transmembrane pressure, and the filtrations with 33 µm filters were carried out at 1 or 3 bar. The permeate solutions were collected and for filtrations of the larger volume, permeate was collected in one or three fractions and saved.

Scanning Electron Microscopy (SEM)
For top-view images samples were fixed onto aluminum stubs with double-adhesive carbon tape, and for cross-section images the samples were mounted onto aluminum sample holders with screw. Imaging was performed using Zeiss Merlin FEG-SEM instrument (Jena, Germany). To reduce charging effects samples were sputtered with Au/Pd with a sputter coater (Polaron, Ashford, UK) was used. The sputtering settings were 4 × 10 2 mbar and 35 mA, and the sputtering time was 30 s.

Cryoporometry by Differential Scanning Calorimetry
Filter paper samples (1.5-2 mg) were soaked into deionized water overnight at room temperature. Water was decanted, and the samples were placed in aluminum crucibles with a lid. Samples were cooled down to 248.15 K (−25 • C) at a rate of 10 K min −1 followed by heating to 277.15 K (4 • C) at a rate of 0.7 K min −1 . Measurements were performed in five replicates.
The pore size was calculated according to Landry [41]: where r p is the radius of pore (nm) and ∆T is the difference between the peak maximum for melting of pore-confined water and peak value for melting of bulk water, experimentally determined at 0.6 ± 0.01 • C.

Dynamic Light Scattering
Particle size distribution was obtained from dynamic light scattering (DLS) using a Zetasizer Nano ZS (Malvern, UK) instrument. All experiments were performed in triplicates.

Polyacrylamide Gel Electrophoresis
Protein separation was performed by reducing polyacrylamide gel electrophoresis (SDS-PAGE). Samples were diluted (1:20 v/v) with PBS and Laemmli buffer, and boiled for 10 min. Electrophoretic separation was carried out at 270 V with Mini-PROTEAN Tetra Vertical Electrophoresis Cell (Bio-Rad, Hercules, CA, USA). Protein bands were detected with Gel Doc™ EZ System (Bio-Rad, Hercules, CA, USA), and quantified using Image Lab 6.0 analysis software (Bio-Rad).

Analytical SEHPLC
Samples were analyzed by size-exclusion high-pressure liquid chromatography using Hitachi Chromaster HPLC-UV system with bioZen 1.8 µm SEC-3 (Phenomenex, Torrance, CA, USA) analytical column. Chromatography was performed with 100 mM sodium phosphate, pH 6.8 mobile phase at 0.3 mL min −1 flow rate for 20 min.

LCMS
Equal amounts (20 µg) of protein samples were taken out for digestion. After reduction and alkylation, the proteins were on-filter digested by trypsin using 3 kDa centrifugal filters (Millipore Tullagreen, Ireland) according to a standard operating procedure. Obtained peptides were dried using a speedvac system. Pellets were resolved in 60 µL of 0.1% formic acid and further diluted four times prior to nano-LCMS/MS. Tandem mass spectrometry was performed by applying HCD in the QEx-Orbitrap mass spectrometer (Thermo Finnigan, San Jose, CA, USA), equipped with a reversed-phase C18-column by 35 min long gradient.
Database searches were performed using the Sequest algorithm, embedded in Proteome Discoverer 1.4 (Thermo Fisher Scientific, Waltham, MA, USA) against Homo Sapience proteome extracted from Uniprot, Release June 2019 with 95% confidence level per protein.

Bacteriophage Filtration and Titration
Coliphages PR772 and ΦX174 were spiked to the pre-filtered solutions to obtain final titer about 10 6 plaque forming units (PFU) mL −1 before filtration was performed. Bacteriophage titer was determined by PFU assay by double agar overlay method. Briefly, ten-fold serially diluted bacteriophage samples were mixed with host E. coli strains and melted soft agar, and poured on the surface of prepared hard agar plate, followed by incubation at 37 • C for 5 h.
Bacteriophage titer was calculated using Equation (1): where N is the number of plaques, V is the volume (typically 0.1 mL) of added virus and d is the dilution factor. The virus retention was expressed as log 10 reduction value (LRV):

One-Step 33 µm Filtration of FIX-Rich PCC
When the FIX-rich PCC at 20 L m −2 volumetric load was filtered through the 33 µm mille-feuille filter paper, a rapid flux decline was observed, e.g., from about 80 L m −2 h −1 to about 10 L m −2 h −1 at 3 bar overhead pressure. DLS analysis of the feed and permeate samples revealed that the feed sample showed widely distributed fraction of protein impurities above 70 nm, which were not detectable after filtration as shown in Figure 1. Notably, these large-size impurities could not be detected in the volume distribution profiles of the feed sample but only in the intensity distribution plots, which suggests that the original amount of the aggregates is small. In the permeate sample, no particle fractions above 40 nm were detected by DLS. Bacteriophage titer was calculated using equation 1: where N is the number of plaques, V is the volume (typically 0.1 mL) of added virus and d is the dilution factor.
The virus retention was expressed as log10 reduction value (LRV):

One-Step 33 μm Filtration of FIX-Rich PCC
When the FIX-rich PCC at 20 L m −2 volumetric load was filtered through the 33 μm mille-feuille filter paper, a rapid flux decline was observed, e.g., from about 80 L m −2 h −1 to about 10 L m −2 h −1 at 3 bar overhead pressure. DLS analysis of the feed and permeate samples revealed that the feed sample showed widely distributed fraction of protein impurities above 70 nm, which were not detectable after filtration as shown in Figure 1. Notably, these large-size impurities could not be detected in the volume distribution profiles of the feed sample but only in the intensity distribution plots, which suggests that the original amount of the aggregates is small. In the permeate sample, no particle fractions above 40 nm were detected by DLS. To investigate if significant changes were recorded in the protein molecular weight distribution in the permeate sample, SDS-PAGE analysis was performed, as shown in Figure 2. Additional proteomics analysis of the detected bands was not performed as it was outside of the scope of the present work. It is seen in Figure 2 that all major fractions in the permeate sample were reduced compared to the feed. The observed decrease in total protein fraction is concordant to that reported earlier for PCC product filtered through Planova 15N filter [14]. In all, it appears that the large molecular weight protein fractions are the main reason for the observed fouling. To investigate if significant changes were recorded in the protein molecular weight distribution in the permeate sample, SDS-PAGE analysis was performed, as shown in Figure 2. Additional proteomics analysis of the detected bands was not performed as it was outside of the scope of the present work. It is seen in Figure 2 that all major fractions in the permeate sample were reduced compared to the feed. The observed decrease in total protein fraction is concordant to that reported earlier for PCC product filtered through Planova 15N filter [14]. In all, it appears that the large molecular weight protein fractions are the main reason for the observed fouling.

Development and Validation of Two-Step Size-Exclusion Bioprocess for FIX-Rich PCC Nanofiltration
The nanocellulose-based filter paper platform provides the possibilities to relatively easily tailor the pore-size distribution of the filter paper to a specific cut-off value. This could be achieved for instance by varying the thickness of the filter paper. Figure 3 shows the SEM images of the filters with varying thickness, including their top-view and cross-section. It is seen from the images of the cross-sections of the filters that they indeed feature varying thicknesses.

Development and Validation of Two-Step Size-Exclusion Bioprocess for FIX-Rich PCC Nanofiltration
The nanocellulose-based filter paper platform provides the possibilities to relatively easily tailor the pore-size distribution of the filter paper to a specific cut-off value. This could be achieved for instance by varying the thickness of the filter paper. Figure 3 shows the SEM images of the filters with varying thickness, including their top-view and cross-section. It is seen from the images of the cross-sections of the filters that they indeed feature varying thicknesses.

Development and Validation of Two-Step Size-Exclusion Bioprocess for FIX-Rich PCC Nanofiltration
The nanocellulose-based filter paper platform provides the possibilities to relatively easily tailor the pore-size distribution of the filter paper to a specific cut-off value. This could be achieved for instance by varying the thickness of the filter paper. Figure 3 shows the SEM images of the filters with varying thickness, including their top-view and cross-section. It is seen from the images of the cross-sections of the filters that they indeed feature varying thicknesses.   To derive information about the pore size of the filter cryoporometry analysis was performed. Figure 4 shows the typical CP-DSC curves of the studied samples and the boxplots of the derived pore width modes. Cryoporometry analysis has the benefit that it probes the pores in the wet state, and it is a relatively quick and highly automated and reliable method. In this method, the samples are first frozen to −40 • C and then slowly thawed. As the ice crystals start to melt, there is a detectable endotherm peak. When the water is located inside mesopores (i.e., 2-50 nm pore width), there will be a melting point depression as opposed to bulk water, present outside pores or in macropores (above 50 nm). In our experiments, bulk water melts at around 0.6 • C. The larger the melting point depression, the smaller are the pores. As seen from Figure 4 there is a trend of decreasing pore width mode with increasing thickness. To derive information about the pore size of the filter cryoporometry analysis was performed. Figure 4 shows the typical CP-DSC curves of the studied samples and the boxplots of the derived pore width modes. Cryoporometry analysis has the benefit that it probes the pores in the wet state, and it is a relatively quick and highly automated and reliable method. In this method, the samples are first frozen to −40 °C and then slowly thawed. As the ice crystals start to melt, there is a detectable endotherm peak. When the water is located inside mesopores (i.e., 2-50 nm pore width), there will be a melting point depression as opposed to bulk water, present outside pores or in macropores (above 50 nm). In our experiments, bulk water melts at around 0.6 °C. The larger the melting point depression, the smaller are the pores. As seen from Figure 4 there is a trend of decreasing pore width mode with increasing thickness. To assess the particle rejection cut-off for each filter, model probes with 2 different particle sizes were used in the form of bacteriophages, i.e., PR772 (70 nm) and ΦX174 (28 nm) phages, see Table 1 and 2, respectively. These probes provide a highly sensitive tool for assessing the size-dependent rejection capability of the filters with varying thickness, i.e., 6, 11, and 33 μm. The 33 μm mille-feuille filter paper shows the lowest hydraulic flux, i.e., 38 L m −2 h −1 bar −1 , and the highest virus removal capacity for both small-and large-size viruses, i.e., LRV ≥5.7. The 6 μm filter in the series exhibits the fastest flux, i.e., 405 L m −2 h −1 bar −1 , but poor virus removal capacity, i.e., LRV <1 and <2 for 28 nm and 70 nm model phages. The flux and virus removal properties of 11 µ m filter are intermediate to the other two filters, wherein the 11 µ m filter paper shows high clearance towards 70 nm virus, i.e., >5.7, and moderate clearance toward 28 nm one, i.e., LRV 3.5-4.5, and hydraulic flux of 125 L m −2 h −1 bar −1 . Interestingly, the small-size virus removal capacity of 11 µ m filter decreased with increasing load volume, whereas that of 33 µ m filter remained unaffected under the experimental conditions. The latter could probably be due to redistribution of flow through the larger pores when the smaller pores become clogged in 11 µ m filter paper.  To assess the particle rejection cut-off for each filter, model probes with 2 different particle sizes were used in the form of bacteriophages, i.e., PR772 (70 nm) and ΦX174 (28 nm) phages, see Tables 1 and 2, respectively. These probes provide a highly sensitive tool for assessing the size-dependent rejection capability of the filters with varying thickness, i.e., 6, 11, and 33 µm. The 33 µm mille-feuille filter paper shows the lowest hydraulic flux, i.e., 38 L m −2 h −1 bar −1 , and the highest virus removal capacity for both small-and large-size viruses, i.e., LRV ≥5.7. The 6 µm filter in the series exhibits the fastest flux, i.e., 405 L m −2 h −1 bar −1 , but poor virus removal capacity, i.e., LRV <1 and <2 for 28 nm and 70 nm model phages. The flux and virus removal properties of 11 µm filter are intermediate to the other two filters, wherein the 11 µm filter paper shows high clearance towards 70 nm virus, i.e., >5.7, and moderate clearance toward 28 nm one, i.e., LRV 3.5-4.5, and hydraulic flux of 125 L m −2 h −1 bar −1 . Interestingly, the small-size virus removal capacity of 11 µm filter decreased with increasing load volume, whereas that of 33 µm filter remained unaffected under the experimental conditions. The latter could probably be due to redistribution of flow through the larger pores when the smaller pores become clogged in 11 µm filter paper. Table 1. LRVs for 70 nm (PR772) bacteriophages filtered through 6, 11, and 33 µm filter papers. The results represent the virus clearance data of virus-spiked PBS. Green color code denotes high virus clearance LRV > 5; Yellow denotes moderate virus clearance (2 < LRV < 5); and pink denotes low virus clearance (LRV < 1). Table 2. LRVs for 28 nm (ΦX174) bacteriophages filtered through 6, 11, and 33 µm filter papers.
The results represent the virus clearance data of virus-spiked PBS. Green color code denotes high virus clearance LRV > 5; Yellow denotes moderate virus clearance (2 < LRV < 5); and pink denotes low virus clearance (LRV < 2).

Two-
Step 6 µm/33 µm Filtration of FIX-Rich PCC Figure 5 shows the permeate flux through the 6 µm/33 µm filtration sequence at 1 bar. Rapid flux decline was observed for the permeate after initial plateau. Please note that the flux of pre-filtrate was so fast that it was not recorded.   Figure 5 shows the permeate flux through the 6 μm/33 μm filtration sequence at 1 bar. Rapid flux decline was observed for the permeate after initial plateau. Please note that the flux of pre-filtrate was so fast that it was not recorded.  Figure 6 shows the DLS results for pre-filtrate and permeate samples for the 6 μm/33 μm filtration sequence at 1 bar. It was observed the fraction of large colloids was not removed by 6 μm pre-filtration. However, no aggregates were observed in the permeate sample after filtration through 33 μm filter.  Figure 6 shows the DLS results for pre-filtrate and permeate samples for the 6 µm/33 µm filtration sequence at 1 bar. It was observed the fraction of large colloids was not removed by 6 µm pre-filtration. However, no aggregates were observed in the permeate sample after filtration through 33 µm filter.  Figure 6 shows the DLS results for pre-filtrate and permeate samples for the 6 μm/33 μm filtration sequence at 1 bar. It was observed the fraction of large colloids was not removed by 6 μm pre-filtration. However, no aggregates were observed in the permeate sample after filtration through 33 μm filter.  Figure 7 shows the results of SDS-PAGE analysis of the collected samples. It is seen from the graph that all bands showed decreasing intensity. Even after 6 µm pre-filtration, some decline in the band intensity was observed. The bands for lower Mw fractions, i.e., bands 4-6, were reduced to a greater extent after pre-filtration than those of the larger Mw, i.e., bands 1-3. In the permeate sample all band intensities were further decreased. LCMS analysis suggested that key coagulation factors IX, X, V as well as prothrombin were not removed following the two-step 6 µm/33 µm filtration sequence, as shown in Appendix Tables A1-A3.  Figure 7 shows the results of SDS-PAGE analysis of the collected samples. It is seen from the graph that all bands showed decreasing intensity. Even after 6 μm pre-filtration, some decline in the band intensity was observed. The bands for lower Mw fractions, i.e., bands 4-6, were reduced to a greater extent after pre-filtration than those of the larg Overall, the results from the filtration with 6 μm/33 μm sequence suggest that the large Mw impurities were not removed during the pre-filtration step and, subsequently, caused filter fouling and thereby low product yield during the second step.

Two-Step 11 μm/33 μm Filtration of FIX-Rich PCC
In another set of experiments, the pre-filtration was performed using 11-μm filter paper followed by filtration with 33 μm filter at 1 bar. Figure 8 shows the flux data of permeate for 11 μm/33 μm filtration sequence. Increasing the thickness of the pre-filter from 6 to 11 μm significantly affected the results. The flux values for pre-filtration indicated rapid fouling as observed above for 33 μm filtration. However, in the second step of 11 μm/33 μm filtration sequence, i.e., through 33 μm filters, stable flux was observed for the entire processed volume, Figure 7. The results contrast starkly those observed for 6 μm/33 μm filtration sequence as shown in Figure 4. Overall, the results from the filtration with 6 µm/33 µm sequence suggest that the large Mw impurities were not removed during the pre-filtration step and, subsequently, caused filter fouling and thereby low product yield during the second step.

Two-Step 11 µm/33 µm Filtration of FIX-Rich PCC
In another set of experiments, the pre-filtration was performed using 11-µm filter paper followed by filtration with 33 µm filter at 1 bar. Figure 8 shows the flux data of permeate for 11 µm/33 µm filtration sequence. Increasing the thickness of the pre-filter from 6 to 11 µm significantly affected the results. The flux values for pre-filtration indicated rapid fouling as observed above for 33 µm filtration. However, in the second step of 11 µm/33 µm filtration sequence, i.e., through 33 µm filters, stable flux was observed for the entire processed volume, Figure 7. The results contrast starkly those observed for 6 µm/33 µm filtration sequence as shown in Figure 4.  Figure 9 shows the results of DLS analysis of the pre-filtrate and permeate samples. It is seen that the fraction of large-size impurities, which was clearly visible in the feed solution, was absent both in the pre-filtrate and permeate fractions of 11 μm/33 μm filtration sequence. The latter suggests that pre-filtration with 11 μm filter paper efficiently removes the large-size impurities, unlike prefiltration with 6 μm filter paper. Additional SEHPLC analysis was performed on these samples as shown in Figure 8. It is seen in the graph that the peak retention times and relative intensities are similar in all three samples except for the early peak at 0.5 min in the feed sample. This peak, which corresponds to the largest protein fraction was not detectable in pre-filtrate and permeate samples. The results of the SDS-PAGE analysis for 11 μm /33 μm filtration sequence are summarized in Figure 10. It is seen from the graph that the band intensities were reduced in the pre-filtrate and permeate samples as compared to the feed. It should be noted that in general the band intensities were reduced to a greater extent after pre-filtration with 11 μm filter than with 6 μm filter. The decrease of band intensity levels in the permeate sample passed through the 33 μm filter after 11 μm filtration was much less drastic than that for 6 μm/33 μm filtration sequence. In particular, no significant changes were observed for bands 1, 2, and 4. For bands 3, 5, and 6 some intensity reduction was further detected in the permeate sample. LCMS analysis suggested that key coagulation factors IX, X, V as well as prothrombin were not removed following the two-step 11 μm/33 μm filtration  Figure 9 shows the results of DLS analysis of the pre-filtrate and permeate samples. It is seen that the fraction of large-size impurities, which was clearly visible in the feed solution, was absent both in the pre-filtrate and permeate fractions of 11 µm/33 µm filtration sequence. The latter suggests that pre-filtration with 11 µm filter paper efficiently removes the large-size impurities, unlike pre-filtration with 6 µm filter paper. Additional SEHPLC analysis was performed on these samples as shown in Figure 8. It is seen in the graph that the peak retention times and relative intensities are similar in all three samples except for the early peak at 0.5 min in the feed sample. This peak, which corresponds to the largest protein fraction was not detectable in pre-filtrate and permeate samples.  Figure 9 shows the results of DLS analysis of the pre-filtrate and permeate samples. It is seen that the fraction of large-size impurities, which was clearly visible in the feed solution, was absent both in the pre-filtrate and permeate fractions of 11 μm/33 μm filtration sequence. The latter suggests that pre-filtration with 11 μm filter paper efficiently removes the large-size impurities, unlike prefiltration with 6 μm filter paper. Additional SEHPLC analysis was performed on these samples as shown in Figure 8. It is seen in the graph that the peak retention times and relative intensities are similar in all three samples except for the early peak at 0.5 min in the feed sample. This peak, which corresponds to the largest protein fraction was not detectable in pre-filtrate and permeate samples. The results of the SDS-PAGE analysis for 11 μm /33 μm filtration sequence are summarized in Figure 10. It is seen from the graph that the band intensities were reduced in the pre-filtrate and permeate samples as compared to the feed. It should be noted that in general the band intensities were reduced to a greater extent after pre-filtration with 11 μm filter than with 6 μm filter. The decrease of band intensity levels in the permeate sample passed through the 33 μm filter after 11 μm filtration was much less drastic than that for 6 μm/33 μm filtration sequence. In particular, no significant changes were observed for bands 1, 2, and 4. For bands 3, 5, and 6 some intensity reduction was further detected in the permeate sample. LCMS analysis suggested that key coagulation factors IX, X, V as well as prothrombin were not removed following the two-step 11 μm/33 μm filtration sequence (for details see Appendix Table A1, A4-A5). The results of the SDS-PAGE analysis for 11 µm /33 µm filtration sequence are summarized in Figure 10. It is seen from the graph that the band intensities were reduced in the pre-filtrate and permeate samples as compared to the feed. It should be noted that in general the band intensities were reduced to a greater extent after pre-filtration with 11 µm filter than with 6 µm filter. The decrease of band intensity levels in the permeate sample passed through the 33 µm filter after 11 µm filtration was much less drastic than that for 6 µm/33 µm filtration sequence. In particular, no significant changes were observed for bands 1, 2, and 4. For bands 3, 5, and 6 some intensity reduction was further detected in the permeate sample. LCMS analysis suggested that key coagulation factors IX, X, V as well as prothrombin were not removed following the two-step 11 µm/33 µm filtration sequence (for details see  Appendix Tables A1, A4 and A5). Based on the above results, it was concluded that pre-filtration with 11 µ m pre-filter removes the aggregates, which in turn greatly enhances the yield of the 33 µ m filtration. To confirm the high virus removal capacity of 33 µ m filter, the filter paper was loaded with much larger volume than that tested earlier, i.e., 175 L m −2 . Figure 11 shows the result of the large load filtration. Following the filtration, no abrupt filter fouling was detected for the entire processed volume, although some flux decline could be observed (Figure 11a). Under the experimental conditions, it is estimated that Vmax of the process will be roughly around 500 L m −2 , which is a drastic improvement from 20 L m −2 when filtering in a single-step process through 33 µ m filter paper. Furthermore, the filter paper showed high model small-size virus removal capacity, wherein LRV was ≥ 5 in all collected fractions ( Figure  11b). In particular, no detectable PFUs were observed at all up to 90 L m −2 load volume. In the last fractions only residual breakthrough (1-2 PFUs per agar plate, corresponding to 0.7 PFU mL −1 ) was detected. Thus, it was confirmed that the two-step 11 µ m/33 µ m filtration provides enhanced throughput and good capacity to remove small-size virus without abrupt fouling even when challenged with a relatively large load.

Discussion
In this article, the filtration of a highly challenging hematologic product was investigated. Considering that FIX-rich PCC inherently consists of many bioactive components and some Based on the above results, it was concluded that pre-filtration with 11 µm pre-filter removes the aggregates, which in turn greatly enhances the yield of the 33 µm filtration. To confirm the high virus removal capacity of 33 µm filter, the filter paper was loaded with much larger volume than that tested earlier, i.e., 175 L m −2 . Figure 11 shows the result of the large load filtration. Following the filtration, no abrupt filter fouling was detected for the entire processed volume, although some flux decline could be observed ( Figure 11A). Under the experimental conditions, it is estimated that Vmax of the process will be roughly around 500 L m −2 , which is a drastic improvement from 20 L m −2 when filtering in a single-step process through 33 µm filter paper. Furthermore, the filter paper showed high model small-size virus removal capacity, wherein LRV was ≥ 5 in all collected fractions ( Figure 11B). In particular, no detectable PFUs were observed at all up to 90 L m −2 load volume. In the last fractions only residual breakthrough (1-2 PFUs per agar plate, corresponding to 0.7 PFU mL −1 ) was detected. Thus, it was confirmed that the two-step 11 µm/33 µm filtration provides enhanced throughput and good capacity to remove small-size virus without abrupt fouling even when challenged with a relatively large load. Based on the above results, it was concluded that pre-filtration with 11 µ m pre-filter removes the aggregates, which in turn greatly enhances the yield of the 33 µ m filtration. To confirm the high virus removal capacity of 33 µ m filter, the filter paper was loaded with much larger volume than that tested earlier, i.e., 175 L m −2 . Figure 11 shows the result of the large load filtration. Following the filtration, no abrupt filter fouling was detected for the entire processed volume, although some flux decline could be observed (Figure 11a). Under the experimental conditions, it is estimated that Vmax of the process will be roughly around 500 L m −2 , which is a drastic improvement from 20 L m −2 when filtering in a single-step process through 33 µ m filter paper. Furthermore, the filter paper showed high model small-size virus removal capacity, wherein LRV was ≥ 5 in all collected fractions ( Figure  11b). In particular, no detectable PFUs were observed at all up to 90 L m −2 load volume. In the last fractions only residual br

Discussion
In this article, the filtration of a highly challenging hematologic product was investigated.

Discussion
In this article, the filtration of a highly challenging hematologic product was investigated. Considering that FIX-rich PCC inherently consists of many bioactive components and some impurities, the virus removal filtration of this product is difficult without fouling. The virus removal filtration of PCC was previously reported using Ultipor DV50 filters, which are dedicated for removal of large-size viruses but do not ensure viral safety against parvoviruses [2]. Filtration of PCC through small-size virus removal filters, e.g., Planova 15N, resulted in nearly 39% total protein loss and reduced FIX and FII activity, which was ascribed to presence of large-size complexes between clotting factors and high molecular weight impurities [14]. It was further reported in the same study that filtration of highly purified FIX through Planova 15N not only did not result in the decrease of FIX activity but also improved its purity [14].
In this work, in order to achieve high virus removal capacity combined with reduced fouling, a tailored two-step process of filtration with nanocellulose-based filter paper was developed. In particular, sacrificial pre-filters with a thickness of 6 and 11 µm were tested. The increased thickness of the filters resulted in tighter pore structure as detected by cryoporometry. The observed effect is explained as follows and illustrated in Figure 12. The mille-feuille filter paper consists of a stratified 3-dimensional network of cellulose nanofibers, producing a mesh-like stricture. The layered structure is illustrated in the side-view panel of Figure 12. Considering that the nanofibers are randomly distributed in each layer, the pores, which percolate throughout the entire depth of the filter, become tighter with increasing number of layers. The latter is reflected, e.g., in improved virus clearance properties with increased thickness or enhanced aggregate removal properties. Based on the results of PFU titrations of 27 and 70 nm phage particles, it was concluded that the tested filters show varying particle size rejection threshold as the thickness of the filter is increased. Thus, the observed effect is due to the combination of the receding pore size and depth effects (increased tortuosity). The latter enables using pre-filters with tailored cut-off to remove protein aggregates, which eventually results in improved flux through the dedicated virus removal filter. of large-size viruses but do not ensure viral safety against parvoviruses [2]. Filtration of PCC through small-size virus removal filters, e.g., Planova 15N, resulted in nearly 39% total protein loss and reduced FIX and FII activity, which was ascribed to presence of large-size complexes between clotting factors and high molecular weight impurities [14]. It was further reported in the same study that filtration of highly purified FIX through Planova 15N not only did not result in the decrease of FIX activity but also improved its purity [14].
In this work, in order to achieve high virus removal capacity combined with reduced fouling, a tailored two-step process of filtration with nanocellulose-based filter paper was developed. In particular, sacrificial pre-filters with a thickness of 6 and 11 μm were tested. The increased thickness of the filters resulted in tighter pore structure as detected by cryoporometry. The observed effect is explained as follows and illustrated in Figure 12. The mille-feuille filter paper consists of a stratified 3-dimensional network of cellulose nanofibers, producing a mesh-like stricture. The layered structure is illustrated in the side-view panel of Figure 12. Considering that the nanofibers are randomly distributed in each layer, the pores, which percolate throughout the entire depth of the filter, become tighter with increasing number of layers. The latter is reflected, e.g., in improved virus clearance properties with increased thickness or enhanced aggregate removal properties. Based on the results of PFU titrations of 27 and 70 nm phage particles, it was concluded that the tested filters show varying particle size rejection threshold as the thickness of the filter is increased. Thus, the observed effect is due to the combination of the receding pore size and depth effects (increased tortuosity). The latter enables using pre-filters with tailored cut-off to remove protein aggregates, which eventually results in improved flux through the dedicated virus removal filter. Overall, the two-step approach presented here is based on the size-exclusion principles and is therefore robust. It could thus further be adapted in the manufacturing of other protein-based pharmaceutics, too, including recombinant proteins wherein impurities in the form of host cell proteins may greatly affect the final yield of the biologics during virus removal nanofiltration. Illustration of the mechanism of virus removal with increased thickness of nanocellulose-based filter paper. Yellow symbols represent large-size model virus and red symbols represent small-size viruses. Increased thickness of the filter results in tighter pores and enhanced virus clearance.
Overall, the two-step approach presented here is based on the size-exclusion principles and is therefore robust. It could thus further be adapted in the manufacturing of other protein-based pharmaceutics, too, including recombinant proteins wherein impurities in the form of host cell proteins may greatly affect the final yield of the biologics during virus removal nanofiltration.

Conclusions
A two-step process was developed to both enhance filtration capacity (25-fold) and achieve high clearance of small-size viruses (LRV >5) using appropriate pre-filter paper. Large-size aggregates were the main foulants in the feed solution, and by tailoring the properties of the pre-filters the foulants were efficiently removed. In particular, 11 µm/33 µm filtration was found most suitable. The presented approach could potentially be applied for bioprocessing other protein-based drugs, both derived from plasma and produced by recombinant approaches. The article further provides new insights regarding the mechanism of virus removal in the nanocellulose-based filter paper, highlighting the combined effect of size exclusion and tortuosity of pore network.