New technical concept for alternating tangential flow filtration in biotechnological cell separation processes

Robust cell retention devices are key to successful cell culture perfusion. Currently, tangential flow filtration (TFF) and alternating tangential flow filtration (ATF) are most commonly used for this purpose. TFF, however, suffers from poor fouling mitigation, which leads to high filtration resistance and product retention, and ATF suffers from long residence times and cell accumulation. In this work, we propose a filtration system for alternating tangential flow filtration, which takes full advantage of the fouling mitigation effects of alternating flow and reduces cell accumulation. We have tested this novel setup in direct comparison with the XCell ATF® as well as TFF with a model feed comprising yeast cells and bovine serum albumin as protein at harsh permeate to feed flow conditions. We found that by avoiding the dead‐end design of a diaphragm pump, the proposed filtration system exhibited a reduced filtration resistance by approximately 20% to 30% (depending on feed rate and permeate flow rate). A further improvement of the novel setup was reached by optimization of phase durations and flow control, which resulted in a fourfold extension of process duration until hollow fiber flow channel blockage occurred. Thus, the proposed concept appears to be superior to current cell retention devices in perfusion technology.


| INTRODUCTION
Perfusion processes have gained in importance in biopharmaceutical cellular fermentation processes and research alike during the last few decades 1,2 due to increased productivity, improved product quality and higher batch-to-batch homogeneity in comparison to fed-batch processes. 3,4 The critical point is generally seen in the perfusion or cell retention device, 5,6 which is designed for separating the produced biological therapeutic substance in continuous production mode over longer periods of processing time, while retaining the producing cells.
Among the membrane-based perfusion devices, conventional tangential flow filtration (TFF) and alternating tangential flow filtration (ATF), that is, crossflow with periodically changing flow direction, are commonly used. In membrane-based perfusion processes, a key process performance indicator is the transmission of the target molecule (also known as product sieving). Deposits formed by retained cells and macromolecules can add an additional retention effect in comparison to the clean membrane. This deposit formation often results in reduced transmission, as it acts as a secondary membrane with its own and often unpredictable retention characteristics. Published work shows the superiority of ATF over TFF in terms of higher cell viability and lower product retention. 7,8 The ATF concept has become widely applied in the biopharmaceutical industry through the development of the commercially available XCell ATF ® device. Besides its application as perfusion device in the production of monoclonal antibodies, 8,9 it has also been studied for the production of virus particles, 10 to intensify N-1 seeding fermentation 11 and for harvesting of biopharmaceuticals. 12 The superior performance of the XCell ATF ® device versus conventional TFF has been attributed to three main effects: Firstly, the feed pump used in this system, a diaphragm pump, is considered a low-shear pump. The use of low-shear pumps was reported to result in low cell damage and therefore a lower release of DNA, RNA and other intracellular substances into the fermentation broth. 13 This reduces the complexity of substances in the aqueous phase, thus reducing fouling propensity, which otherwise increase filtration resistance and product retention. Secondly, these fouling effects were reported to be under better control by flow reversal and the associated pressure pulsations, which promote fouling mitigation and thus enhance filtration performance. 14 Thirdly, the changing pressure conditions of the unsteady flow were reported to cause backflushing of filtrate back to the retentate, also described as Starling flow phenomena, 15 which contributes to the removal of deposited material from the membrane surface. The fouling mitigation effects of alternating crossflow 14,[16][17][18] and other hydrodynamic fouling mitigation techniques have been extensively described in several works. [19][20][21] The remaining critical point, though, in our eyes is that the diaphragm pump applied in the XCell ATF ® device is limited in its operational flexibility, long-term processing and technically feasible range of processing conditions. Also, scaling up beyond XCell ATF ® 10, the largest commercially available device, is currently only possible by operating multiple units in parallel. The diaphragm pump is pneumatically actuated by supplying pressurized air and vacuum from the reverse side of the diaphragm, which results in a flow from the diaphragm pump back to the feed vessel (pressure phase) and from the feed vessel to the diaphragm pump (exhaust phase), respectively. 22 The use of pressurized air and vacuum, however, leads to small achievable flow velocities, for instance a maximum of 10 L min À1 in an XCell ATF ® 4 device, which results in a crossflow velocity of only 0.25 m s À1 in the attached hollow fiber module. Therefore, the wall shear stress along the membrane or deposit surface is limited to levels where only marginal deposit removal and fouling mitigation occurs. In perfusion processes of shear sensitive cells or mycelium-like aggregates, low crossflow velocities are well justified or even preferred, but for other processes working with more robust cells like yeasts, for instance Pichia pastoris, higher crossflow velocities could be desirable to enhance deposit layer removal. Several patents propose the advancement of the XCell ATF ® device by employing two diaphragms instead of one, motorized actuators, pistons or combinations thereof. [23][24][25] However, these technical developments appear to be mechanically complex to implement and they are not commercially available so far.
Another important aspect is that the diaphragm pump volume limits the pump's maximum displacement volume per stroke and cycle.
Therefore, the duration of each forward and backward cycle of alternating flow depends on the targeted flow rate. This interdependence limits the options for process optimization, as both frequency and crossflow velocity have an impact on the extent of fouling and fouling mitigation. 16,26 If, for instance, the crossflow velocity is reduced in order to decrease the shear stress acting on the cells, also the frequency will be reduced, which may have a negative impact on effective fouling mitigation. Additionally, the ratio of hold-up volume in the transfer line to the bioreactor and filter module relative to the fixed pump displacement volume can also be seen as unfavorable. This is because it provokes long residence times of cells in the device, which can lead to oxygen depletion, increased lactate production and reduced growth rate, impaired viability, and lower productivity. 27,28 Under certain processing conditions, cells even accumulate in the diaphragm hold-up volume, which leads to increased fluid viscosities and intensification of all aforementioned disadvantages, which are residence time related. That is why another patent proposes to replace the diaphragm pump with a bidirectional peristaltic pump. 28 The use of a peristaltic pump to transport the cell broth, however, poses an unwanted shear stress on the cells and is therefore not optimal for mammalian cell perfusion cultures.
Considering the reported advantages of alternating tangential flow filtration and to overcome the issues described above, a mechanically simple and more versatile alternating flow setup, capable of generating alternating flow within a wide range of flow rates and flow reversal frequencies would be desirable. At the same time, only low shear stress on the cells should be applied, residence times outside the bioreactor should be reduced and accumulation of cells in the external loop should be avoided. Therefore, a newly developed alternative alternating flow concept, mainly based on applying another pump concept with rapidly reacting centrifugal pumps acting in opposite flow direction (denominated setup II in the following), was studied in this work and compared with the XCell ATF ® device (setup I in the following). The detailed technical features of the used pumps are described in detail in the methods section.
The focus of this work was on the hydrodynamic conditions, that is, flow rates, pressure conditions, filtration resistances and cell accumulation effects of both filtration setups. A practicable model feed system was designed comprising yeast cells as a representative of producing cells and bovine serum albumin (BSA) as a substitute for produced biological substances meant to pass through the membrane. The application to a cell culture perfusion process must be the ultimate goal when developing new cell retention devices. At this stage of the work, the focus was on the hydrodynamic characterization of the proposed concept. However, the low-shear design of the pumps employed in setup II were already reported to have no significant damaging effect on mammalian cells. 13 The next step following this work therefore is the transfer of this concept to a mammalian perfusion culture and assessing the impact on cell viability, including cell size and cell metabolism, as well as product sieving and product quality.

| Analytical methods
The dry matter content of feed and retentate samples was determined by a microwave assisted drying balance SMART 6 (CEM Corporation,

| Filtration system
All filtration trials were conducted employing either of the two different filtration systems capable of generating alternating flow (see

| Flow profile characterization
As a first step for comparing setups I and II, the flow profiles generated by setup I were recorded for several flow rates and reproduced with setup II. For this purpose, the bioreactor was filled with desalted water and kept at the filtration temperature of 15 C. The permeate line was closed throughout these preliminary tests. The feed rate was set on the ATF controller, which finds the right pre-pressure and orifice size in an iterative process. When the set feed rate was reached and the pre-pressure did not further change over time, the flow profile was recorded. It should be noted that it can take several minutes for the ATF controller to reach the flow set point. Therefore, the related pre-pressures and orifice sizes were recorded to be used as starting point in the following filtration experiments. Flow rates higher than 5 L min À1 could not be sustained in setup I.
The flow profiles were analyzed in JMP ® Pro 14.1 software and the actual flow in each phase and the phase duration were noted.
These values were then used to program two-phase recipes on the console to set the duration of each phase and pump speed of the corresponding pump as input parameters for setup II.

| Filtration trial and data handling
Filtration experiments were conducted with both setups presented in for concentration processes. 12 However, harsh conditions up to a ratio of 1:3 were chosen in previous studies in order to provoke faster and more obvious fouling rates. 29 The chosen set point feed and permeate flow conditions of setup I are given in Table 1. It should be noted that the measured flow rates deviate from the set flow rates due to the indirect controlling strategy of the ATF controller, measuring cycle times instead of flow rates. The flow rates of setup II were set in order to match the actual flow rates of setup I, as can be seen from Figure 2. Shear rates τ w were calculated according to Equation (2) as a function of crossflow velocity v crossflow and hollow fiber inner diameter d i .
Additionally to the experiments presented in Table 1, single experiments with forward flow only (i.e., conventional non-alternating crossflow filtration) with the same actual feed flow rate and permeate flow rate were conducted using only the inlet centrifugal pump of setup II, while the other centrifugal pump was set inactive and was flown-through by the retentate.
Prior to each filtration experiment, the module was flushed with desalted water and the pure water permeability was measured at 15 C. Afterwards, the water was removed from the vessel and the filtration setup. While setup II was fully drainable, setup I had some remaining water in the diaphragm pump, which was not drainable without disassembling, which will be important when discussing the measured dry matter contents. Subsequently, 8 L of the pre-cooled feed suspension was given to the tank, the stirrer was set to 150 rpm and one initial feed sample was drawn. Afterwards, the feed flow was started, followed by the permeate flow induced by peristaltic pumps.  Table 1 were conducted in a randomized order. The optimization experiments were performed afterwards. Therefore, the T A B L E 1 Overview over filtration trial set points in setup I. The set points of setup II were chosen to match the flow rates of setup I (see Figure 2)  The time-resolved inline data for pressure and flow values were averaged in order to obtain data representing mean processing performance. The data processing was done according to Weinberger and Kulozik. 30 From these averaged flux (J) and transmembrane pressure (Δp TM ) data, the filtration resistance (R filtration ) was calculated according to Equation (3), considering the permeate viscosity η. The permeate viscosity of some permeate samples was measured using a MCR302 rheometer (Anton Paar GmbH, Graz, Austria) equipped with a double gap geometry; it was similar to pure water viscosity.
All data collected during the filtration trials were evaluated and plotted using JMP ® Pro 14.1.

| Flow profile
In order to directly compare both setups, flow profiles from setup I were recorded and reproduced with setup II. Figure 2 shows the flow profiles for the two feed flow rates chosen for this study, 2 L min À1

| Comparative assessment of filtration performance
The filtration trials were conducted at different feed flow rates and different permeate flow rates according to Table 1. Figure 3 shows process performance indicators of filtration trials conducted at equal feed flow rate but varying permeate flow rates (and thus varying permeate to feed ratio), while Figure 4 shows process performance indicators of filtration trials with equal permeate to feed ratio, but varying feed flow rates. This differentiation allows for the separate evaluation of the role of permeate to feed ratio on cell accumulation and of the crossflow velocity on fouling mitigation.
As can be seen from Figure 3, all filtration trials with a feed flow rate of 4 L min À1 could be sustained for at least 5 h. The process performance was similar for both alternating setups I and II with only minor differences. For the filtration trial with conventional nonalternating crossflow, the process performance was worse than for the alternating crossflow conditions, as can be seen from an up to ten- In comparison to the major difference between conventional non-alternating and alternating crossflow filtration, the effects of varying permeate flow rates and the filtration setup were comparably small, but yet observable. Figure 3a shows that the transmembrane pressure was higher for trials with higher permeate flow rate. This increase of transmembrane pressure was proportional to the increase in permeate flow rate, since the difference between each pair of trials vanishes when considering the filtration resistance (see Figure 3b).
The filtration resistance, however, reveals a minor difference between the two filtration setups, where the resistance using setup II was approximately 30% lower than for setup I. Also the pressure profiles, as exemplarily shown in Figure S2 for setup I and setup II, show only minor differences. The fluctuation of absolute pressures is stronger pronounced for setup I and all local pressures cyclically reach negative values due to the acting diaphragm pump. But the transmembrane pressure is slightly positive for both setups with only single outliers, which are probably just artifacts due to the high data acquisition rate.
The lower fluctuation of pressures in setup II might be beneficial as it results in a more even transmembrane pressure distribution across the membrane module and better module usage. 15 The BSA transmissions, as depicted in Figure 3c, for all alternating trials were comparable and scattered at about 100%. This high transmission value can be attributed to the rather large pore size of 0.5 μm and the overall low transmembrane pressure, which prevented an undesirable compaction of fouling material. The fact that BSA transmission values were partially above 100% might be due to analytical variations and possibly due to the approximation of the real transmission, by taking the feed BSA concentration and not the retentate BSA concentration inside the filtration device into account (see Equation (1)).
Lastly, the dry matter content in the retentate was determined as a measure for cell accumulation (see Figure 3d). Cell accumulation can be observed for both alternating flow setups. This is due to the hold-  When comparing setup I and II, the filtration resistance in experiments using setup II was 20 to 30% lower than in experiments using setup I (as can be seen from Figure 4b), which is comparable to the observation from Figure 3b. The BSA transmission was not significantly different for most trials shown, with only a minor reduction after 3 h of filtration for the trial with 2 L min À1 feed flow rate conducted with setup I (gray open circles in Figure 4c). This trial also stands out in terms of accumulated dry matter in the retentate (see Figure 4d). Whereas all other trials with a permeate to feed ratio of 1:10 showed similar dry matter contents in the retentate, the dry matter content in the retentate was approximately doubled for the 2 L min À1 trial conducted with setup I due to cell accumulation over time. Considering the difference between 2 L min À1 and 4 L min À1 feed flow rate, on the one hand, it seems that the lower feed flow rates resulted in drag forces too low to transport the easily sedimentable cells against gravity back into the feed tank. Considering the differences between the filtration setups, on the other hand, setup II has no dead-end, as it is characteristic for the diaphragm pump of setup I. Setup II thus draws fresh medium from the feed tank also during the backwards flow phase, which reduces or even avoids accumulation of cells during filtration even at low feed flow rates. Hence, it can be said that even when setup II was operated with similar flow profiles as setup I, which is unfavorable in terms of the insufficient exchange volume, the issue of cell accumulation was less severe and filtration resistances were thus reduced.  Figure S1). Figure 5 shows the process performance indicators for filtration trials conducted with 2 L min À1 feed flow rate and 400 ml min À1 F I G U R E 5 Process performance indicators of filtration runs with a nominal crossflow of 2 L min À1 and a nominal permeate flow rate of 400 ml min À1 : (a) Actual mean feed flow rate, (b) dry matter content in the retentate, (c) mean transmembrane pressure, (d) mean filtration resistance. Dark gray circles refer to setup I, black diamonds setup II and light gray squares represent a run with conventional non-alternating crossflow conducted with one centrifugal pump of setup II being active. Black rectangles represent filtration trials with setup II, but prolonged forward and backwards phases and black asterisks represent filtration trials with setup II, where the pumps were flow controlled instead of speed controlled. The error bars indicate the range of a randomized duplicate. Lines are given as a guide to the eye permeate flow rate. It can be seen that, due to the deliberately chosen extreme permeate to feed ratio, none of the trials could be sustained for the filtration time of 5 hours. The high permeate rate led to the concentration of the retentate (see Figure 5b), which results in an increased retentate viscosity and an impaired pumpability. Note that the controller of setup I cannot satisfactorily cope with increased fluid viscosities. 31 As a result of cell accumulation and increased fluid viscosity, the feed flow decreases over time (see Figure 5a). The insufficient volume exchange in setup I and the non-optimized setup II, as discussed in chapter 3.2, aggravates this issue. Also, the high permeate to feed ratio, as intended, led to severe deposit formation, which can be concluded from increasing transmembrane pressures and filtration resistances (see Figure 5c,d).

| Optimization concepts using setup II
During conventional non-alternating crossflow filtration, cell accumulation issues due to insufficient fluid exchange did not occur.
However, the sharp increase of fouling resistance after 15 to 20 min of conventional non-alternating crossflow filtration at similar permeate to feed flow ratio (see Figure 5d), hints at severe deposit layer formation due to the high drag forces toward the membrane and insufficient fouling prevention. Obviously, alternating crossflow was able to mitigate fouling in an efficient way, resulting in a longer feasible filtration time of alternating crossflow for both setups I and II, despite the occurrence of cell accumulation. Therefore, an alternating flow filtration process, which efficiently mitigates fouling and avoids cell accumulation, increasing retentate viscosity and feed flow reduction (as occurring for setup I and non-optimized setup II) is desired.
In setup I, the exchange rate can only be increased by constructional means, such as a higher diaphragm pump volume (not in the operator's hands) or a shorter transfer line to the feed tank. 27 Using two counteractive centrifugal pumps in setup II, the fluid exchange can be easily improved by increasing the volume transported in each phase, that is, by increasing the phase duration (as shown in Figure S1C) and/or feed flow rate, or by counteracting the increasing retentate viscosity by implementing a feedback loop (as shown in Figure S1D). Both optimization options are addressed in the following.
By reducing cell accumulation, prolonged phases (compare phase duration from lower panels in Figure S1) not only led to a longer feasi- • higher crossflow velocities and higher frequencies to increase fouling mitigation; • higher crossflow velocities and longer phases to reduce cell accumulation at high permeate to feed ratios; • longer forward phases with shorter intermittent backwards phases to combine the advantages of conventional and alternating crossflow; • flow profiles with less steep ramps to protect shear-sensitive cells from turbulences; • combination of alternating and pulsatile flow, where short flushing phases of higher flow rate might be used as inline cleaning technique.
However, these options are yet to be systematically investigated.

| CONCLUSION AND OUTLOOK
In this work, a new technical concept for alternating crossflow filtration was proposed and its hydrodynamic performance investigated in direct comparison with the state-of-the-art XCell ATF ® device and Open Access funding enabled and organized by Projekt DEAL.