General workflow for the scale-up of perfusion processes based on tangential flow filtration

Tangential flow filtration (TFF) perfusion technology utilising hollow-fibre modules offers significant advantages over other mammalian cell retention techniques. In recent years, a growing understanding of Starling flow has enabled filtration performance to be enhanced to levels comparable with, or even exceeding, those of alternative technologies. Nevertheless, the large-scale application of TFF perfusion technology continues to face challenges and typically requires considerable time spent on trial and error. This paper proposes a systematic, data-driven four-step workflow for the design and scale-up of TFF perfusion systems, which integrates hydrodynamic factors, bubble removal studies and pump characteristic analysis. Whilst filter selection in TFF perfusion is typically a major focus, the design of the cell retention circuit is often overlooked. However, this element is critical to system performance and may lead to suboptimal pump operation, excessive residence time and bubble accumulation. The data-driven workflow developed in this paper helps to optimise the design of general-purpose TFF perfusion systems utilising hollow-fibre modules, effectively bridging the gap between small-scale and large-scale production applications.

Introduction

Perfusion cell culture plays a key role in current process intensification strategies; it can both enhance the productivity of existing fed-batch production facilities through N-1 perfusion and serve as the primary production bioreactor in continuous production processes. Among the various cell retention technologies available, membrane systems comprising hollow fibres (HF) have become the industry’s preferred choice. This preference stems primarily from its compatibility with single-use (SU) technology, its ability to achieve complete cell retention via particle size exclusion at the membrane, and its good scalability.

Typically, two distinct cross-flow filtration systems (commonly referred to as tangential flow filtration, or TFF) are employed in perfusion processes: one commonly used TFF perfusion setup is a single-connection configuration, such as the alternating tangential flow (ATF) operating mode (Figure 1A). This alternating flow mode has demonstrated advantages in laboratory and pilot-scale studies, with improved product sieving efficiency, which may be attributed to the backwash effect occurring on both sides of the HF module. However, at production scale, pneumatic diaphragm pumps suffer from operational instability and cross-flow inconsistencies; these issues not only hinder scale-up but also pose challenges at high cell densities due to increased culture medium viscosity. Although the plug-and-play nature of ATF systems is considered advantageous for accelerating the development of early-stage clinical trials or commercialisation processes with low material requirements, their capacity for large-scale production is limited. Specifically, the ‘one pump, one filter’ design and the limited volume of diaphragm pumps necessitate the placement of multiple HF modules in close proximity to the bioreactor to achieve the required filtration area. This results in a significant footprint at production scale, restricts the number of available bioreactor ports, and requires the provision of additional utility infrastructure within the facility, such as vacuum and compressed air.

Figure 1. This paper provides an overview of various perfusion and filtration system configurations suitable for benchtop bioreactors. (A) Schematic diagram of an ATF perfusion system using a diaphragm pump and a filter, representing a ‘dead-end system’; (B) A simple TFF system utilising a single pump with unidirectional flow through the filter; (C) An rTFF system employing dual pumps to achieve alternating flow; (D) An HPTFF setup utilising dual pumps, with one pump in the feed loop and the other in the filtrate recirculation loop, thereby ensuring uniform pressure distribution across the membrane; (E) The figure illustrates Stirling flow within a hollow-fibre filter and proposes strategies to mitigate its effects.

Another common configuration is the recirculating loop system, commonly referred to in the industry as tangential flow filtration (TFF) perfusion. These systems typically comprise a pump integrated into the piping loop, which draws cell culture medium from the bioreactor, directs it through the HF module at one end, and then returns it to the bioreactor via the module outlet (Figure 1B). This recirculating loop configuration enables continuous, unidirectional cross-flow through the HF module. The main pump types reported in the literature as having been successfully applied in high-density perfusion TFF are rotary pumps, which are positive-displacement pumps, and low-shear centrifugal pumps, which are dynamic pumps. Unlike positive-displacement pumps, centrifugal pumps are designed to provide pressure rather than flow rate; consequently, their operating speed depends largely on the fluidic design of the flow path. However, poor flow path design and the resulting excessive hydraulic loads are also considered key challenges for rotary pumps. High backflow within the pump head generates excessive shear forces, making the fluid system design a critical parameter independent of the pump itself. As most tangential flow filtration (TFF) systems are currently custom-built, fluid path design may present challenges in large-scale applications. For N-1 perfusion at a scale of several thousand litres, multiple pump technologies have been successfully employed to realise several unidirectional TFF units. For N-stage perfusion, some literature reports indicate that, when operated at similar high cross-flow rates, unidirectional TFF systems exhibit a faster decline in filtration efficiency than alternating-flow systems due to customised design and suboptimal operation.

Although TFF filtration systems have reportedly exhibited certain limitations in terms of filtration performance, there is a clear trend within the industry towards adopting TFF systems due to their operational advantages, particularly in perfusion processes involving high cell densities and volumes exceeding 500 litres. This trend has prompted a significant amount of research aimed at narrowing the performance gap between ATF and TFF systems; the main findings are as follows:

1. Reduction of Stirling flow in unidirectional tangential flow filtration systems: The main advantage of the unidirectional tangential flow filtration mode lies in the simplicity and low complexity of its technical and mechanical design, as well as the implementation of its control system. Alternating-flow systems benefit from Stirling flow due to the backflushing effect at both ends of the filter, which aids in the removal of substances accumulated within the concentration boundary layer; unidirectional tangential flow filtration systems, however, do not possess this advantage, as the Stirling flow is always unidirectional. Consequently, minimising Stirling flow is crucial for avoiding excessive and unnecessary filtration flux. This reduction typically results in wall shear rates below 1000 s⁻¹, which can be achieved by lowering the pressure drop across the hollow fibre lumens. Other strategies for reducing Stirling flow include increasing the inner diameter of the hollow fibre lumens, shortening the filter length, or increasing membrane resistance (Figure 1E).

2. Reverse cross-flow in TFF systems: Alternating flow can be introduced into a unidirectional TFF system by adding a reverse pump; this configuration is known as reverse TFF (rTFF) (Figure 1C). Although system complexity is increased due to the addition of a pump and control strategy, this setup significantly enhances separation performance whilst retaining most of the core advantages of a unidirectional TFF system. Unlike ATF systems, where cycle times are limited by the flow rate of the diaphragm pump and are typically only a few seconds, rTFF systems offer greater flexibility. Cycle times can range from a few seconds to several hours, opening up new opportunities for optimising process performance. This approach is particularly suitable for processes involving large cell aggregates that are prone to blocking the hollow fibre inlets.

3. Eliminating Stirling flow via co-current filtrate: Another filtration method involves introducing co-current filtrate along the membrane on the outside of the hollow fibres in a unidirectional tangential flow filtration (TFF) system, thereby eliminating Stirling flow. This method, known as high-performance tangential flow filtration (HPTFF) (Figure 1D), has been shown to deliver significant improvements in product separation. However, the increased system complexity requires more sophisticated control strategies and necessitates design modifications to existing hollow fibre (HF) modules to ensure scalability.

3. Eliminating Stirling flow via co-current filtrate: Another filtration method involves introducing co-current filtrate along the membrane on the outside of the hollow fibres in a unidirectional tangential flow filtration (TFF) system, thereby eliminating Stirling flow. This method, known as high-performance tangential flow filtration (HPTFF) (Figure 1D), has been shown to offer significant improvements in product separation. However, the increased system complexity requires more sophisticated control strategies and necessitates design modifications to existing hollow-fibre (HF) modules to ensure scalability.

Although the industry has identified effective methods to address performance-related shortcomings, an increasing number of tangential flow filtration (TFF) perfusion processes are gradually moving towards commercial-scale implementation. The main challenge at present lies in translating performance improvements—most of which have been developed on a small-scale or pilot scale—into reliable and scalable production solutions. Regardless of which TFF perfusion mode is selected for a specific process (unidirectional TFF, rTFF or HPTFF), all configurations share a common recirculation loop comprising pipework, connectors, pumps and one or more HF modules (Figures 1B–D).

It is worth noting that filtration performance is often limited by process-specific fouling—an accumulation of a ‘gel layer’ comprising genomic DNA, cell debris and hydrophobic defoamers. Consequently, HF filters are generally regarded as the most critical component in the cell retention circuit, whilst the design of the liquid circuit has become a blind spot in the implementation of TFF perfusion systems. Although process-specific fouling is highly dependent on proprietary processes, hydrodynamic conditions can be compared across different processes and scales; however, a trial-and-error approach is still frequently employed, which is both time-consuming and labour-intensive.

Consequently, this study aims to summarise the industry’s current understanding of the most critical parameters for the scale-up of TFF perfusion systems and to propose a generic workflow to facilitate the transition from empirical, trial-and-error scale-up to a faster, more reliable, computationally and data-driven approach. Whilst this workflow is broadly applicable to various pump technologies, this paper uses practical guidelines and data for suspended centrifugal pumps as examples, given their widespread use in the industry.

Results and Discussion

The design of an effective TFF cell retention solution can be divided into two distinct phases. In the first phase, process-specific design inputs must be defined to ensure that the equipment is correctly matched to the intended perfusion process. These inputs vary significantly across the industry, depending on the cell line, product and operational strategy. The second phase, which is process-independent, requires the incorporation of scale-related design inputs to ensure that the TFF cell retention system operates reliably under the specified process conditions. Given the high variability of specific process requirements and the extensive literature on this subject, this paper provides only a brief overview of key parameters and typical industry ranges, which may serve as a starting point for evaluation. The primary objective of this paper is to propose a clear, structured workflow for addressing the challenges of scaling up TFF cell retention systems, which can be applied to any specific process environment. Figure 2 outlines the proposed workflow for scaling up TFF perfusion.

Figure 2. Schematic diagram outlining the proposed TFF scale-up workflow. The process-specific design inputs and characterisation data provided in this paper will be used for the proposed four-step TFF scale-up method. This scale-up method comprises the following four steps: 1. process scale-up; 2. filter scale-up; 3. fluid circuit design; 4. pump selection.

Process-specific design inputs for TFF systems

The process-specific design inputs for TFF cell retention systems can be broadly categorised into the following groups: cell culture parameters, fluid handling parameters and filter-related parameters (including filter characteristics and operating conditions). These inputs are crucial for ensuring compatibility between the process and the equipment configuration, and contribute to the robust implementation of process control strategies. The following summarises the ranges of these parameters commonly found in the industry, providing a useful reference for initial system sizing and evaluation.

Cell Culture Parameters

In modern perfusion culture processes, cell density typically ranges from 50 to 100 M cells/mL, with some studies reporting densities as high as 200 M cells/mL. Whilst live cell density (VCD) is a recognised metric, the live cell volume fraction (VCV) more accurately reflects biomass (Equation (2)) and may therefore provide a more meaningful basis for comparing different perfusion culture processes. Typical VCV values can reach 30 per cent, whilst recent studies have demonstrated even higher values. In addition to VCD and VCV, the viscosity of the cell culture medium is also an important design consideration. Viscosity reflects changes in cell concentration and cell viability. For most perfusion processes, viscosity is maintained within the range of 1 to 2 cP, but may rise to 5 cP in high-viability cultures. Lower cell viability towards the end of the culture period leads to a further sharp increase in viscosity, which, in the authors’ experience, typically exceeds 5 cP.

Fluid Management

The cell recovery rate in a perfusion process is a key input parameter for the rational design of the cell retention system, as it directly influences the flux through the filter. In most steady-state perfusion processes, the perfusion rate ranges from 1 to 2 vvd, with a waste rate typically accounting for 10% to 20% of the perfusion rate. Consequently, the cell recovery rate usually accounts for 80% to 90% of the perfusion rate. In dynamic perfusion processes, the cell recovery rate is equal to the perfusion rate.

Characteristics of HF Components

Hollow-fibre membranes used in perfusion cell culture are primarily made from polyethersulphone (PES), polysulphone (PS) or polyvinylidene fluoride (PVDF). The internal diameter of the fibre lumen is one of its design parameters, which influences both hydraulic performance and filtration performance. A larger lumen diameter typically reduces the pressure drop across the fibres, thereby lowering the Stirling flow rate and, due to reduced hydraulic resistance, decreasing the required pump speed. However, increasing the lumen diameter also reduces the effective surface area of the membrane, as the number of fibres that can be accommodated within a given filter housing is reduced due to the lower surface-to-volume ratio. The most commonly used lumen diameters in industrial perfusion applications range from 1.0 to 1.4 mm. Filter length is another parameter that affects the hydraulic pressure drop. Longer filters increase the pressure drop across the membrane, leading to higher Stirling flow rates. Modules 60 to 100 centimetres in length are typically selected, although shorter modules of 20 to 40 centimetres are also frequently used in laboratory-scale and automated reactor systems. Although larger pore sizes have been explored to increase membrane flux and reduce product retention, a pore size of 0.2 μm is the standard for most perfusion processes. This allows the harvested material to be fed directly into downstream processing without the need for additional filtration steps. In processes where the product is retained within the bioreactor, such as monoclonal antibody production, membrane modules with smaller pore sizes and a molecular weight cut-off of less than 50 kDa are typically used.

HF Operating Parameters

In TFF perfusion systems, membrane flux is typically expressed in L/m²/h (abbreviated as LMH), a value normalised by the membrane surface area relative to the flow rate. For N-stage perfusion processes, where product sieving is critical, lower LMH values of 2 to 6 LMH are typically employed. Conversely, in N-1-stage perfusion processes where product retention is not a primary consideration, higher LMH values of up to 20 are typically employed. Another key parameter is the effective cross-flow velocity, which directly influences the sizing of tubing and components in the cell retention circuit. The cross-flow velocity is defined by the target wall shear rate within the hollow fibre lumen. Early studies typically operated systems at shear rates of 1,500–2,000 s⁻¹ or higher, whereas current TFF perfusion processes generally employ lower shear rates, ranging from 400 to 1,000 s⁻¹. This shift has been driven by improvements in filtration performance under low-shear conditions, as well as the need to reduce shear stress on cell cultures.

Scale-related design inputs for TFF systems

Once reasonable process-specific design inputs have been established—either based on prior process knowledge or through dedicated development work—the next key challenge is how to scale up the TFF cell retention system whilst maintaining process performance. To address this issue, the authors have developed a systematic workflow based on current industry practice and supplemented by novel characterisation data. This workflow is designed to guide users through the key steps required for reliable and predictable scale-up. It comprises four main stages: 1. Process scale-up, 2. Filter scale-up, 3. Fluid circuit design, 4. Pump selection (Figure 2).

Process Scale-up

When scaling up a TFF perfusion process, key process input or output parameters—such as the cell concentration distribution expressed as VCD or VCV, and the relative perfusion rate distribution expressed as vvd—typically remain constant across different scales. Furthermore, it is assumed that the distribution of cell viability and the viscosity of the culture medium remain comparable across different scales. Assuming that the distribution of waste rates (where applicable) is comparable, and that the process is robust with cell doubling time being scale-independent, the absolute yield increases linearly with bioreactor scale.

Filter Scaling

The fundamental principle in selecting filters is to maintain consistency in filter characteristics to ensure that filters of different sizes retain good filtration performance. This is relatively easy to achieve for parameters such as membrane composition, lumen diameter and pore size, as most filter manufacturers maintain consistency in these characteristics when supplying filters in a range of sizes.

A key scaling guideline commonly used in the industry for operational filters is to maintain a standardised filtrate flux (measured in LMH) across different sizes. As the size increases, the required absolute yield also increases; consequently, a larger filtration area is needed to achieve the same LMH. Altering the filter length to increase the filtration area during scaling will change the magnitude of the Stirling flow, which may have unpredictable effects on filtration performance. Consequently, for downscaling models used to evaluate the performance of cell retention devices, it is strongly recommended that a uniform HF component length be used across all sizes to ensure consistency and predictability of performance. In addition to filter length, this paper also recommends maintaining a constant wall shear rate within the HF lumen across different sizes. Combined with a consistent filter length, this ensures that flow conditions within each fibre remain similar, thereby supporting consistent performance across different sizes. It is worth noting that the effects of Stirling flow diminish when HF components with larger internal diameters are used. The filtrate flux can be further fine-tuned by adjusting the bioreactor’s liquid volume or introducing a recirculation loop (returning a portion of the filtrate to the bioreactor), which artificially increases the LMH at the laboratory scale. In practice, during scale-up, the filter length is typically kept constant as far as possible. At smaller scales, such as the AMBR 250 mL perfusion system, maintaining the filter length during production may be impractical or unfeasible. Conversely, at larger scales, limiting the filter length may restrict the available filtration area, resulting in an excessively wide filter.

Fluid Circuit Design

The fluid circuit of a TFF perfusion system comprises all components that come into contact with the cell culture medium, such as hollow-fibre filters, tubing, fittings and pump heads. Whilst the filter assembly is typically the focus of attention, other components of the fluid circuit are often overlooked. A robust and well-balanced fluid circuit design must meet the following requirements:

· Minimise shear stress on cells: Hydrodynamic shear stress is generated throughout the cell retention circuit, including within tubing and hollow fibre lumens, in transition zones between components (e.g. where the internal diameter changes abruptly without a smooth transition), and inside the pump head. The shear stress generated in centrifugal pumps is primarily driven by the impeller speed. Therefore, reducing the load on the liquid system helps to lower the required pump speed, thereby reducing shear stress. Accurate calculation or experimental determination of shear stress, particularly within the pump head, is highly complex, and its biological effects vary depending on cell type and cell line. However, to ensure scalability, minimising shear stress throughout the entire circuit is widely recognised as good practice.

· Minimise the residence time of cells outside the bioreactor: Cells circulating in the cell retention loop temporarily leave the controlled environment of the bioreactor, where key process parameters such as temperature, pH and dissolved oxygen (DO) are strictly controlled. At high cell densities, dissolved oxygen can be depleted within seconds, leading to hypoxia, which may adversely affect cell health and process performance. It is therefore essential to minimise the residence time of cells outside the bioreactor. Furthermore, the liquid circuit should be designed to eliminate dead zones or areas of low flow velocity, as these increase the residence time of cells outside the bioreactor.

· Minimise the ingress of bubbles into the cell retention circuit and ensure their effective removal: as aeration rates are high in most perfusion bioreactors, bubbles will intermittently enter the cell retention circuit. It is therefore essential to minimise the ingress of these bubbles and to effectively remove any that do enter. Bubble accumulation can destabilise flow, interfere with flow sensor readings and, in particular, may lead to pressure relief in the centrifugal pump or the entire cell retention circuit.

To meet these circuit design requirements, appropriate qualitative design choices (e.g. materials, connector design, component positioning) and quantitative design choices (e.g. internal diameter and length of tubing, internal diameter of connectors, pump size) must be considered.

Considerations for Qualitative Liquid Circuit Design

Qualitative design begins with the selection of component materials. Smooth surface finishes are preferred to minimise shear stress. Commonly used tubing materials include platinum-cured silicone rubber or weldable thermoplastic elastomers (TPEs) with similar surface properties. Biocompatible polypropylene (PP), polysulphone (PS) and polycarbonate (PC) are commonly used for connectors, sensors and pump heads. Stainless steel is also an option, particularly in large-scale production. Ideally, the fluid circuit should remain unobstructed, and the internal diameter of the entire cell retention circuit should be consistent, with the exception of HF filter assemblies and pump heads. This means that the internal diameter of connectors should match that of adjacent tubing, including the immersion tubes used in small bioreactors. Particular care should be taken to avoid bends in the tubing, as these reduce the internal diameter and affect flow characteristics. Where the internal diameter of a connector decreases significantly, the resulting transition leads to increased shear stress, which in turn places a greater load on the entire fluid system, forcing the pump to operate at higher impeller tip speeds. Conversely, a sudden increase in internal diameter may create areas where reduced fluid velocity leads to the accumulation of bubbles or cell sedimentation. In practice, maintaining a consistent internal diameter throughout the entire cell retention circuit is not always feasible. In such cases, reducer fittings with smooth transitions should be used to avoid turbulence and minimise shear stress.

Initial priming of the cell retention circuit and maintaining proper priming conditions are crucial for ensuring the stability of the perfusion operation. The centrifugal pump should be positioned below the liquid level in the bioreactor so that hydrostatic pressure can be utilised to prime the pump via the outlet at the bottom of the bioreactor. Minimising bubble ingestion is another important design consideration. The position of the outlet (leading to the cell retention device) should ensure that bubbles from the aerator are not directly drawn into the cell retention circuit. Taking the direction of agitation into account, it is best practice to position the outlet upstream of the aerator to minimise bubble ingestion. When using an immersion tube at laboratory scale, it is common practice to extend the tube with a silicone tube to a position where bubble ingestion is avoided, such as below the aerator.

Even with an optimal design, bubbles may occasionally enter the filter cartridge retention circuit. In some cases, these bubbles can cause pressure relief in the pump head or even the entire circuit. Pressure loss in the pump head typically occurs only in centrifugal pumps of the smallest size, as the bubble size is relatively large compared to the volume of the pump head. In such cases, bubbles may become trapped at the centre of the rotating impeller. Whilst temporarily increasing the pump speed to expel trapped bubbles is a viable solution, this method also increases the flow velocity and alters the wall shear rate within the HF components, thereby having an unknown impact on filtration performance. The strategy currently more widely adopted within the industry is to angle the pump head outlet at 45°, ensuring that the opening leading to the outlet is situated at the highest point of the pump head. During periodic pump shutdowns lasting a few seconds, rising air bubbles can be expelled through this opening, thereby preventing bubble entrapment without affecting the wall shear rate. In large-scale applications, the probability of pump depressurisation is reduced because the bubble size is negligible relative to the pump head, allowing the bubbles to be flushed out without the need for specific removal strategies. Depressurisation of the circuit can be mitigated through quantitative design measures, such as selecting an appropriate internal pipe diameter.

Considerations for the Design of Quantitative Liquid Circuits

Scale adjustment is required to accommodate a specific fluid flow rate (cross-flow). This flow rate is determined by process-specific design choices based on the target shear rate at the inner wall of the HF filter assembly. When the fluid flow rate is fixed, the primary factors controlling the liquid system load, residence time and bubble removal are the internal diameter and length of the tubing. Whilst both the internal diameter and length of the tubing affect the liquid system load and increase the circuit’s dead volume, only the internal diameter affects the fluid velocity. To simplify the following analysis of the liquid system, it is assumed that the internal diameter of the connectors is the same as that of the tubing, and the tubing and connectors are collectively referred to as the circuit.

Minimising Hydraulic Load

The shear rate generated by high fluid velocities within the circuit can be reduced by increasing the circuit’s internal diameter and can be estimated using a formula. In contrast, the calculation of shear stress generated within the pump head is considerably more complex and falls outside the scope of this article. However, a pragmatic approach is to minimise the hydraulic load on the system as much as possible, enabling the pump to operate at the lowest possible speed whilst still meeting the requirements for residence time and bubble removal. To assess the impact of changes in circuit internal diameter or circuit length on the hydraulic system load, the Darcy–Weisbach equation can be applied. The total hydraulic system load for a cell retention device comprises the contributions from the HF filter, the circuit (pipes and fittings), and the pump. The Darcy–Weisbach equation can be used to evaluate each section of the circuit separately, and these contributions can be added together to obtain the total system load. Ideally, the HF component should account for the majority of the total fluid system load, whilst the contribution from the remaining circuit components should be as small as possible. The influence of the pump is generally negligible when a pump of sufficient capacity is selected so that the inlet and outlet fittings do not restrict flow. The influence of the HF component is fixed, as it is determined by the target wall shear rate specified in the process design and cannot be adjusted at this stage. By contrast, the impact of the circuit can be reduced by increasing the circuit’s internal diameter or shortening its length whilst minimising the number of bends.

Although minimising the circuit’s contribution to the overall hydraulic load is generally desirable, even optimised large-scale cell retention systems may still generate a high overall hydraulic load. Factors contributing to an increase in the overall hydraulic system load include: an increase in the length of the HF component following scale-up, leading to a higher proportion of the total load; increased pipework length due to bioreactor positioning constraints; and selection limitations imposed by the fixed dimensions of bioreactor ports. Maintaining a consistent overall hydraulic system load across different scales is considered good practice to ensure that pump operating conditions and impeller tip speeds remain consistent at different scales. Consequently, loop internal diameter and loop length are key parameters for regulating hydraulic load and can be adjusted in scaled-down models to better simulate the performance of large-scale systems. However, design changes that alter the system load will also affect fluid velocity and loop retention volume. Therefore, such adjustments must be kept within a range that meets retention time requirements and ensures effective bubble removal.

Residence Time Constraints

Increasing the internal diameter of the circuit and the length of the tubing will both result in an increase in the retained volume of cells within the circuit. Given that the fluid flow rate is preset according to the target wall shear rate within the HF component, any increase in retained volume will inevitably lead to an extension of the cells’ residence time within the circuit. Accurate estimation of residence time requires consideration of the entire retained volume, including the volumes of the tubing, pump head and HF filter. Oxygen is the limiting factor determining the residence time of cells in the retention circuit. Due to the high metabolic activity associated with high-cell-density processes, oxygen gradients or even oxygen depletion cannot be entirely avoided. However, it is worth noting that, at larger scales, cells have been subjected to repeated oxygen gradients (or even oxygen depletion) with mixing times of up to 100 seconds, without any reported adverse effects. Although a universally applicable critical residence time for cells in the retention loop has not yet been established, the residence time should be minimised as far as possible; most TFF perfusion systems typically operate within 30 to 60 seconds, which is shorter than the mixing times in many established large-scale processes.

Precautions for Bubble Flushing

When the dead volume of the cell retention loop increases due to an increase in the loop’s internal diameter, whilst the volumetric flow rate remains constant, the resulting linear velocity decreases. Although a lower flow velocity can reduce shear stress, an excessively low flow velocity may impair the effective removal of bubbles from the cell retention loop into the bioreactor. Bubble accumulation occurs when the buoyancy-driven upward velocity exceeds the downward flow velocity. In such circumstances, bubbles cannot be effectively transported, particularly in vertically downward-flowing sections of the circuit, as buoyancy acts in the opposite direction to the flow (Figure 3A). Over time, inefficient bubble removal leads to gas accumulation, ultimately resulting in system depressurisation. Whilst larger bubbles typically accumulate at the highest point of the circuit—usually above the HF component—bubble accumulation has also been observed in horizontal sections. To overcome the resistance of trapped gas, the centrifugal pump must operate at a higher impeller speed, thereby increasing shear stress. Initially, fluid flow may be maintained, but continued accumulation will eventually lead to complete depressurisation of the system. In the event of degassing failure, the entire cell retention circuit, including the pump head, becomes filled with gas, resulting in a loss of flow and failure of the cell retention device, ultimately leading to the termination of the perfusion process. Determining the critical fluid flow rate (below which the flushing out of bubbles cannot be guaranteed) is crucial for the robust design of cell retention devices. To address this issue, we conducted screening experiments to estimate the critical bubble removal rate (Figure 3B). According to Archimedes’ principle, buoyancy and, consequently, the rate of ascent increase with bubble size; therefore, we used large bubbles generated through a 2.5 mm injection orifice for the screening experiments to determine the critical flow rate required to ensure the removal of bubbles of the most unfavourable size. All experiments were conducted at 37°C using water–glycerol mixtures of varying viscosities to simulate the rheological properties of cell culture media. The results indicate that the critical bubble removal flow velocity depends on the inner diameter of the loop and the fluid viscosity. As the loop inner diameter increased from 1/8 inch to 3/4 inch, the critical flow velocity for bubble removal rose sharply from 0.12 m/s to 0.47 m/s. When the loop inner diameter exceeded 3/4 inch, the critical flow velocity plateaued at approximately 0.55 m/s at a viscosity of 0.7 cP. Similar trends were observed at higher viscosities (2.5 cP and 5.0 cP), which are representative of high-density perfusion culture; an increase in viscosity led to a slight decrease in the critical bubble removal velocity for all loop inner diameters. The literature reports that for individual bubbles ranging in size from 1 to 10 mm, the terminal rise velocity is between 20 and 30 cm/s. The lower critical flow velocities observed in smaller-diameter tubing (1/8 inch and 1/4 inch) can be attributed to the reduced internal diameter; the presence of the tube wall restricts the relative motion of the fluid surrounding the bubble, thereby slowing its rise. Furthermore, the higher critical flow rates observed in larger-diameter tubing (1/2 inch and above) can be attributed to the hydrodynamic velocity distribution within the tube, where the fluid velocity is lower near the tube wall and the bubble traverses regions of varying local velocity. From a practical perspective, perfusion culture typically begins at a low cell density, and the fluid viscosity is close to that of water at 37°C. Consequently, designing the cell retention loop to maintain flow rates above the critical threshold derived from 0.7 cP provides a conservative safety margin, ensuring reliable bubble removal even if viscosity increases during perfusion. It is worth noting that the length of the cell retention loop does not affect bubble removal efficiency, as it does not alter the linear fluid flow velocity.

Figure 3. (A) A schematic diagram highlighting the opposing effects of downward fluid flow and upward buoyancy on bubble motion within a vertical pipe, which leads to bubble accumulation when the fluid velocity is insufficient. (B) Relationship between the critical bubble removal rate and the inner diameter (ID) of the silicone tubing, measured at three different fluid viscosities (0.7 cP, 2.5 cP and 5.0 cP), which correspond to the fluid viscosities associated with perfusion processes.

In summary, the key challenge lies in minimising shear stress on the cells whilst ensuring an adequate flow rate, thereby guaranteeing an acceptable residence time and effective bubble removal within the cell retention circuit. Achieving this balance requires combining qualitative design considerations with quantitative criteria to weigh up the relationship between reduced shear stress and the constraints of residence time and bubble flushing.

Pump Selection

In theory, an ideal scale-up strategy should involve selecting and operating a suitable pump such that the shear stress generated within the pump head remains constant across different scales, thereby promoting consistency in cellular responses and minimising unexpected outcomes during scale-up. However, in practice, accurately calculating or simulating the shear stress within a centrifugal pump head under specific operating conditions is highly complex and requires numerous assumptions. For example, factors such as the daily culture turnover rate, the residence time of cells in critical regions, and changes in medium composition and viscosity during cell culture all influence the cells’ response to shear stress. These variables are difficult to quantify precisely, and the uncertainties they introduce make accurate estimates of shear stress impractical. In light of these challenges, it is recommended that a pragmatic, experiment-driven approach be adopted to guide pump selection and operating conditions, thereby enabling a robust and scalable cell retention process. Specifically, risks during scale-up can be mitigated by designing large-scale systems that exert a comparable or even gentler shear stress on cells than small-scale systems. Validating the performance of centrifugal pumps under typical or worst-case conditions at a small scale ensures that the relevant pump parameters are compatible with cell viability and establishes a reliable baseline for scaled-up production. This strategy prioritises the maintenance or improvement of key process parameters of the pump during scale-up. These parameters include maintaining a constant impeller tip speed, ensuring that the culture turnover rate remains equal or is reduced, and maintaining or shortening the residence time within the pump head per turnover. These factors are crucial during scale-up for the following reasons:

1. The impeller tip speed is a key process parameter during the scale-up process. Although this paper does not provide absolute recommendations, given the significant differences in shear sensitivity amongst different cell types and cell lines, carefully designing a cell retention device to reduce the impeller tip speed can generally ensure that operating conditions do not adversely affect cell culture. During scale-up, the impeller tip velocity should not exceed the values encountered in the development experiments. Due to differences in impeller size, the rotational speed (in rpm) required to generate sufficient pressure to overcome the load in the fluid system (Figure 4A) varies with pump size. By carefully designing the head of the centrifugal pump, pumps of different sizes can achieve similar pressures at comparable impeller tip velocities (Figure 4B). The appropriate pump size can be determined from the pressure-flow characteristic curves specific to the pump head (Figures 4C–G), which contain contour lines representing a constant impeller tip speed. At low fluid flow rates, these contour lines are typically horizontal, indicating that the required pressure is essentially independent of flow rate; the same impeller tip speed is sufficient to generate the required pressure, and increasing the flow rate does not significantly raise the tip speed. However, as the fluid flow rate increases further, the contour lines begin to slope downwards or become inclined, reflecting the increase in internal fluid resistance within the pump at high flow rates. This resistance stems from the pump head and the restrictions at its inlet and outlet. In practice, beyond this point, a further increase in flow rate requires a higher impeller tip speed, which increases shear stress. When operation deviates from the region of near-horizontal isoclines, it is advisable to replace the pump with one offering greater head. Due to its larger internal flow passages and connections, the higher flow rate will once again fall within the more favourable, near-horizontal region of the pressure-flow curve for the larger pump, thereby keeping the impeller tip speed within an acceptable range. However, as the pump head volume increases, switching to a larger pump head will increase the residence time of cells within the pump head per cycle; this trade-off must be carefully weighed against the advantage of reduced shear stress resulting from lower impeller tip speeds. It is therefore generally recommended to select the smallest centrifugal pump that operates within the horizontal contour region at the target flow rate. Within certain flow rate ranges, both pump head sizes may be suitable: a smaller pump with a slightly higher impeller tip speed, or a larger pump with a lower tip speed but a longer cell residence time per cycle. In such cases, a case-by-case assessment is recommended. Based on these principles, the authors propose a matching scheme for appropriate flow rates and pump sizes to assist with pump selection (Figure 4H).

2. According to the design, the clearance between the impeller and the pump casing increases as the pump size increases. Although the use of a simplified model (treating the impeller as a moving wall and the pump casing as a fixed wall) to estimate shear stress does not fully reflect the complexity of shear forces generated within a centrifugal pump, it does indicate that a larger clearance reduces shear stress. From this simplified perspective, scaling up whilst maintaining a constant tip speed will naturally result in reduced shear stress due to the increased clearance between the impeller and the pump casing; this can be regarded as a conservative and robust strategy capable of reducing the risk of shear damage caused by scale effects.

3. Under the reasonable assumption that a reduced number of culture turnovers at the pump head is beneficial or has no effect on cell culture performance, the objective of scale-up should be to achieve a number of turnovers similar to or lower than that in small-scale culture. When the selected filter allows the filter length to remain constant across different scales, and both the luminal wall shear rate and flow rate are kept constant, the ratio of fluid flow rate to vessel volume remains unchanged; consequently, the number of culture turnovers at the centrifugal pump head also remains constant. If a longer filter is used during scale-up whilst maintaining constant LMH and wall shear rate within the fibre lumen, the fluid flow rate will increase, but the increase in vessel volume will be even greater. This will result in a reduction in turnover frequency as the scale increases. In both cases, the turnover rate remains constant or decreases, which is beneficial for scale-up.

4. Another factor that must be prioritised for maintenance or minimisation during scale-up is the residence time of the cells within the pump head during each cycle. This residence time depends on the fluid flow rate and the pump head volume (18). In large-scale production, maintaining or, ideally, reducing the residence time within the pump head helps to minimise the risks associated with the scale-up process. Residence time is typically longest in small-scale production, as the pump head size may be too large relative to the required flow rate, resulting in prolonged cell residence time within the pump. In large-scale production, a wider range of pump head sizes is available, allowing for more precise pump sizing and, consequently, a reduction in the residence time of cells within the pump head. Although the authors are unable to define a specific critical residence time, a practical design principle is to ensure that the maximum residence time occurs during small-scale production and to conduct targeted screening experiments at this stage to confirm that it does not adversely affect cell culture performance, thereby providing a sound basis for robust scale-up.

Figure 4. Figure (A) shows the performance data for the suspended centrifugal pump series in water at 37°C. The figure illustrates the relationship between the speed required for all pump models within the horizontal contour lines and the pressure generated by the pumps, whilst Figure (B) shows the results after converting the speed to impeller tip speed. The pressure-flow curves for the pumps are grouped by model: DCP-30.2 (C), DCP-200.2 (D), DCP-600.1 (E), DCP-2000.1 (F) and DCP-4000.1 (G). The colour of the contour lines is determined by the tip speed (TS [m/s]), with a uniform tip speed applied across all pump models. Figure (H) shows the recommended flow rates for each pump based on an analysis of the pump characteristics.

Multi-filter systems

When the bioreactor volume exceeds a certain threshold, a single filtration retention device may no longer be suitable. In such cases, a multi-filter configuration is required to increase the filtration surface area and maintain the target LMH value. In processes where product retention is critical (e.g. steady-state perfusion), a multi-filter configuration is typically required when the perfusion bioreactor volume exceeds 500 litres. By contrast, for processes with less stringent sieving requirements—such as N-1 perfusion, or processes where monoclonal antibodies (mAbs) are retained within the bioreactor—a single filter remains sufficient even at volumes of several thousand litres, as higher LMH target values can be achieved before multiple filters are required. A variety of different multi-filter configuration designs are feasible, each with its own advantages and disadvantages (Figure 5).

Figure 5. Schematic diagram of multi-filter configurations for production-scale bioreactors. These configurations include: (A) a scale-up scheme utilising two independent systems, each comprising a filter and a pump; (B) a parallel configuration, in which a single pump simultaneously distributes the cell culture medium to two filters; (C) a series configuration, in which a single pump sequentially delivers the cell culture medium to two filters.

Single-component, multi-loop set-up

The simplest method of implementing a multi-filter system is scale-up (Figure 5A). This involves replicating the cell retention loop using the largest available filters. The primary advantage of scale-up is that it allows the replication of a successful small-scale loop design without the need for modification. This approach ensures a safe scale-up process and is particularly suitable where only a small number of filter components (e.g. two to three) are required. By duplicating the entire circuit, the system gains redundancy, as each filter operates independently and precise flow control can be exercised over each component. The scaling-up approach is limited when a large number of filters are required, as each circuit necessitates two bioreactor ports—one for feed and the other for the retentate from the cell retention device—and the positioning of these ports must ensure that the culture volume does not short-circuit. This approach is commonly used for steady-state perfusion in the 1,000–1,500 litre range, as well as in processes involving N-1 perfusion or monoclonal antibody retention within the bioreactor, where a single filter assembly is no longer suitable.

Parallel multi-filter configuration

Parallel hollow-fibre (HF) filter modules reduce the number of cell retention loops, thereby reducing the number of bioreactor ports required, whilst also increasing the total filtration area (Figure 5B). This enables the construction of very compact skid-mounted filtration systems, in which a single pump can supply fluid to two or more HF filter modules. From a fluidics perspective, parallel hollow-fibre (HF) filter modules do not increase the fluid contribution to the filter, as the cross-flow velocity and fibre length remain constant. The fluid paths are typically split upstream of the HF filtration modules via a Y-junction and then recombine on the return side of the circuit. As the fluid paths are split, the flow rate in each branch decreases significantly, which has major implications for pipework design. A clear distinction must be made between the ‘convergence’ and ‘filtration’ lines. The ‘convergence’ line connects from the bioreactor to the Y-type diverter fitting and returns to the bioreactor from the converging Y-type fitting; whereas the ‘filtration’ line connects directly to the HF filter assembly and conveys the fluid at a lower flow rate. When applying the scaling principles described in this paper, appropriate pipe diameters must be selected separately for the convergence line and the filtration line, based on their specific flow rate requirements. This ensures that all key design criteria are met, including minimum hydraulic load, a sufficiently short residence time in the cell retention loop, and effective bubble flushing. The main limitation of parallel configuration lies in the increased fluid flow rate within the convergence line, which places higher demands on the centrifugal pump. Although a two-module parallel configuration is feasible and widely used in industry, scaling up to a higher degree of parallelism presents challenges, particularly in single-use systems, due to limitations on available pipe diameters. A viable compromise is to employ a multi-loop parallel configuration, comprising multiple independent cell-retention loops, each containing two HF filter modules in parallel. The general trend towards reducing the shear rate on the inner walls of the fibre lumens makes it possible to operate at lower cross-flow rates; this, in turn, reduces the flow rate in the collector pipe, rendering the parallel connection of more than two filters more feasible. However, other trade-offs include reduced control over precise flow distribution between components and increased design complexity. Parallel HF filter configurations are most commonly used in perfusion processes with lower LMH requirements, such as steady-state perfusion with bioreactor volumes exceeding 1,000 L.

Series Multi-Filter Configurations

Although series multi-filter configurations can also increase the filtration area without the need for additional bioreactor ports, they differ fundamentally from the parallel approach (Figure 5C). Firstly, as the hollow fibre filter elements are arranged in series, their fluid flow rates are cumulative; this directly increases the power requirements of the pump and necessitates higher impeller tip speeds to overcome the additional pressure drop. Consequently, series configurations can be challenging when using small-bore filters, as these inherently impose a significant hydraulic load. By contrast, large-bore filters, which have inherently lower fluid resistance, are better suited to series configurations. Secondly, although the filtration area is increased, the total circuit flow rate remains the same as in a single-filter configuration. This can be advantageous as it avoids restrictions on pipe sizing and reduces the requirements for pump size and power. Although series multi-filter configurations are not common in the industry due to the increased hydraulic load and the associated higher impeller tip speed, they may be a viable option when using large-bore hollow fibre filters operating at lower wall shear rates. Overall, parallel, series and hybrid multi-filter configurations enable highly compact integrated filtration systems, resulting in significant savings in floor space. Whilst increasing the filtration area, these configurations require fewer bioreactor ports than single-component, multi-loop cell retention systems. Furthermore, they simplify start-up and integrity testing and offer greater flexibility in the positioning of cell retention devices, allowing the system to be located in more convenient positions without the need to be situated immediately adjacent to the bioreactor.

Conclusion

This paper proposes a structured, data-driven framework to address the industry challenge of scaling up small-scale tangential flow filtration (TFF) perfusion processes to production scale. We can provide high-performance TFF equipment and filters to support the full-chain implementation of the process from the laboratory to industrialisation.

We have established a four-step workflow encompassing process scale-up, filter scale-up, liquid circuit design and pump selection. Given the complex parameters inherent in the TFF process itself, this workflow—leveraging the stable flux performance, excellent fouling resistance and comprehensive performance database of our TFF equipment and filters—provides a practical, implementable solution for developing robust TFF perfusion systems. Rooted in established industry practices and incorporating data on fluid and pump characteristics, this method eliminates the need for traditional trial-and-error approaches, enabling predictable and reproducible scale-up design.

The solution is compatible with standard hollow-fibre modules and is suited to large-scale parallel multi-filter configurations; it broadly supports steady-state perfusion, dynamic perfusion, N-1 perfusion and various processes where the product is retained by hollow fibres, whilst also being extendable to emerging fields such as cell and gene therapies. This framework effectively bridges the technical gap between pilot-scale results and large-scale production. When paired with our highly consistent TFF equipment and filters, it assists the biopharmaceutical industry in establishing reliable, high-performance perfusion systems, whilst simultaneously constructing representative scaled-down models.