Discussion

Applying co-current filtrate flow represents a promising tool to alleviate product retention in mammalian perfusion processes by achieving uniform TMP conditions along the filter length. This decreases (or eliminates) the Starling recirculation flow, reducing the filter load by using the entire membrane surface (Radoniqi et al., 2018). In this study, pressure characterization experiments demonstrated that HPTFF operation is possible for a wide range of perfusion relevant crossflows using levitated centrifugal pumps that provide uniform (non-pulsatile) flow (Figure 2B ). By matching the inlet retentate pressure \(\text{PT}_{R1}\ \)and the inlet filtrate pressure\(\text{PT}_{F1}\) with a simple delta pressure control to 0 mbar, a uniform TMP was achieved along the lab-scale filter (Figure 2C ). Interestingly, the outlet filtrate pressure \(\text{PT}_{F2}\)showed increasing discrepancy from the retentate outlet pressure (\(\text{PT}_{R2}\)) with increasing co-current filtrate flow. This discrepancy may come from increasing turbulence at the filtrate outlet, although it had no effect on the filtrate pressure profile (\(\text{PT}_{A1-5}\)). A reduced setup consisting of two pressure sensors (\(\text{PT}_{R1}\) and \(\text{PT}_{F1}\)), a retentate centrifugal pump (\(\text{CD}_{R1}\)) and a filtrate centrifugal pump (\(\text{CD}_{F}\)) are therefore sufficient to operate the HPTFF system (Figure 1D ).
Large-scale pressure characterization revealed that HPTFF can also be achieved with manufacturing scale filters (Figure 7 ). With the filtration module used in this study, the inlet filtrate pressure\(\text{PT}_{F1}\) had to be increased by 6 mbar compared to the inlet retentate pressure \(\text{PT}_{R1}\) to match the filtrate pressures\(\text{PT}_{A1-5}\) with the retentate pressure drop. A pressure decrease from \(\text{PT}_{R1}\) positioned in the inlet retentate tubing compared to the pressure sensor located in the adapter piece connecting tube was observed (Figure 7EF ). This offset might be due to the change in tube diameter from the inlet tubing to the much wider adapter piece connecting to the hollow fiber module. Further, pressure \(\text{PT}_{F1}\) had to be controlled higher than expected to achieve HPTFF. As already observed in lab-scale, the outlet pressure sensor in the filtrate loop \(\text{PT}_{F2}\) was lower than anticipated. These findings might be explained by a combination of a relatively smaller filtrate inlet adapter diameter than in the lab-scale and perturbation of the flow pattern at elevated co-current filtrate flows in the large-scale module. Nevertheless, determination of the offset by pressure characterization allowed us to achieve HPTFF operation across the entire tested crossflow range from 0 - 45 L/min by only measuring pressures \(\text{PT}_{R1}\) and \(\text{PT}_{F1}\).
In perfusion cell culture, a uniform TMP, as per definition in HPTFF, is not necessarily the highest priority as in protein separations (van Reis, 1993; van Reis et al., 1997). The main objective in perfusion processes is to avoid filter clogging and reduce product retention. Therefore, a membrane sweep from time to time in the form of a backflush can be beneficial to remove some deposited material, but avoiding intense backflushing as attributed to irreversible fouling (Weinberger & Kulozik, 2022). A novel operating mode was designed in this study named stepping co-current TFF (scTFF). scTFF can be operated with the same hardware setup as described for the HPTFF (Figure 1D ). By lowering and subsequently increasing the co-current filtrate flow rate compared to HPTFF operation, a TMP gradient was achieved along the filter length resulting in a Starling recirculation (Figure 4A ). The Starling recirculation changed direction upon switching from scTFF phase 1 to phase 2, generating a backflush on the first half of the filter and then on the second half of the filter similar to what occurs in ATF or rTFF operation (Pappenreiter et al., 2023; Radoniqi et al., 2018; Weinberger & Kulozik, 2022). However, in contrast to ATF and rTFF, where the strength of the Starling recirculation is a function of crossflow velocity and filter length, the strength of the Starling recirculation in scTFF can be tuned independently of both crossflow velocity and fiber length. For demonstration, a TMP of ± 10 mbar was targeted (Figure 4C ), but any other TMP larger or smaller can be achieved just by varying the co-current filtrate flow rates (Figure 4B ). Furthermore, scTFF can either be operated by switching between scTFF phase 1 and phase 2 (Figure 4D) , or by operating at HPTFF conditions and integrating a membrane sweeping from time to time by lowering or increasing the co-current filtrate flow (Figure 4E ). The duration of each phase can thereby freely be chosen, giving even more operational flexibility. scTFF operation was demonstrated at lab-scale, and pressure characterization experiments revealed applicability at manufacturing scale without changing the system setup (SI Figure 2 ).
Perfusion cell culture runs revealed significantly reduced product sieving below 60% for TFF operation (Figure 5F ). This agrees with the literature, where similarly reduced product sieving was reported (Karst et al., 2016; Pappenreiter et al., 2023; Wang et al., 2017a). It is worth mentioning that the average filtrate fluxes in this study were particularly low with 0.6 L/m2/h compared to commonly reported filtrate fluxes of 2-3 L/m2/h (Radoniqi et al., 2018; Romann et al., 2023).
Despite similar pressure drop and therefore comparable absolute Starling recirculation flow of rTFF compared to TFF, rTFF showed significantly improved product sieving above 90%, which is comparable to what has been observed in ATF systems (Pappenreiter et al., 2023). This confirms that with an identical pump system, rTFF clearly outperformed TFF. However, this study does not allow us to distinguish between the beneficial contributions of backflushing at both the inlet and outlet, utilization of the entire membrane surface, and / or relaxation of the fouling deposit when the crossflow direction changes. The rTFF_2 run with higher amount of cell culture debris showed lower product sieving than rTFF_1 (Figure 5D ), highlighting that rTFF is still prone to product retention which can likely be attributed to pronounced fouling at the inlet or exit of the hollow fiber modules at elevated debris levels (Sundar et al., 2023).
HPTFF operation entirely removing Starling recirculation due to a uniform TMP along the filtration module showed similar or even higher product sieving than rTFF operation. The HPTFF operation was interrupted every 3 minutes for 3 seconds by stopping the crossflow to release potentially trapped gas bubbles from the centrifugal pump head, which might even have had a beneficial impact on product sieving. The slightly delayed PI-response controlling the co-current filtrate flow resulted in a quick sweep of the membrane, initially backflushing the membrane on the first filter half, followed by backflushing of the second filter half upon crossflow re-activation (Figure 6 ). Whereas bubble trapping in the centrifugal pump head plays a minor role at larger scales, intentional pump stopping from time to time to generate a membrane sweep in HPTFF operation might still be an attractive option (SI Figure 2D ). It must be mentioned that a similar effect can be achieved by shortly increasing the crossflow by maintaining the co-current filtrate flow PI control active.
Similar but more controlled sweeping of the membrane was alternatively achieved by increasing or lowering the co-current filtrate flow at constant crossflow (Figure 4A ). Intensity and location of the backflush can be adjusted by changing the magnitude of the co-current filtrate flow (Figure 4B ), offering a wide range of possibilities not available in ATF or rTFF operation. scTFF allows the Starling recirculation flow to be adjusted independently of filtration module specifications or crossflow velocities without changing the hardware setup. This novel approach enables further research to evaluate the benefits of membrane sweeping in a controlled but flexible manner to define best operating conditions depending on process requirements.
A critical aspect of unidirectional crossflow systems remains filter inlet blocking (Weinberger & Kulozik, 2021a; Zydney, 2016). Cell clumps or aggregates getting into the cell recirculation loop can be trapped at the filter inlet blocking entire hollow fibers. In two unidirectional crossflow runs (TFF_1 and HPTFF_2) filter inlet blocking led to premature run termination. When working with cell lines prone to aggregation, rTFF should be the chosen cell retention operation mode to prevent inlet blocking by crossflow reversal. When aggregation is not an issue and unidirectional crossflow represents no risk to premature run termination, HPTFF or scTFF clearly outperform conventional TFF operation. Further, HPTFF and scTFF offer greater flexibility compared to ATF or rTFF systems by alleviating previously described restrictions on filter characteristics and operation parameters:
  1. Crossflow velocity: No restriction to low crossflows as strategy to avoid extensive Starling recirculation.
  2. Filtration module length: Enabling longer filtration modules due to TMP control and thereby reducing system complexity with multiple parallel modules.
  3. Inner fiber diameter: No need for increased inner fiber diameters to reduce pressure gradient at the cost of membrane surface area or greater hold-up volume.
  4. Pore size: Possibility to utilize larger pores sizes without increasing Starling recirculation caused by lower membrane resistance.