Bioprocess Development and Bioreactor Scale-Up for the Production of Recombinant Lentiviral Viral Vectors in HEK293 Suspension Cell Culture

Abstract

Therapeutic applications of viral vectors that initially targeted rare monogenic diseases have now grown to a broader set of indications including cell and gene therapy applications and vaccines. This has prompted the need to increase biomanufacturing capacities, which will require adjustments in the biomanufacturing space to increase yield and lower cost of goods of large-scale productions. HEK293 cells have been widely used for the production of viral vectors because they can grow rapidly in suspension and allow for different modes of production: batch, fed-batch and perfusion. Here we review methods and platforms for producing lentiviral vectors in HEK293 cells grown in serum-free media and the principles and challenges of optimizing and scaling up of bioprocesses in various bioreactors. Lentiviral vectors are particularly difficult to manufacture due to their labile nature. These challenges will be considered in view of current processes and future trends emerging to resolve bottlenecks and existing limitations.

1. Introduction

Through evolution, viruses have become highly efficient natural vehicles for the transfer of genetic material into living cells. This property has been exploited to develop recombinant viral vectors for R&D and therapeutic applications. Modern viral recombinant vectors are based on non-enveloped adenoviruses (Ads) and adeno-associated viruses (AAVs), or enveloped retroviruses such as lentiviruses (LVs), each bearing unique properties and requirements for engineering and production. Viral vectors are designed for efficiency and safety. Viral elements that permit carrying a genetic cargo encoding one or many factors for efficient delivery in targeted cells are preserved, while elements necessary for the virus’ replication are removed to improve their safety profile and increase payload size. Each class of vectors has limitations and advantages that, when judiciously selected, best serve targeted applications from vaccine to cell and gene therapy. Globally, there are currently more than 1000 ongoing clinical trials making use of viral vectors [1].

The demand for development of novel and improved viral vectors has reached a critical point where demand for efficient and cost-effective manufacturing is increasing faster than current technology is improving. Across vector design, production, purification and characterization, several bottlenecks need to be solved to improve targeting and infectivity; increase titers, batch to batch reproducibility, yields and purity; as well as expand in-process and post-process analytics tools for release and stability monitoring. The development of a viral vector production platform begins upstream with the selection of host cell, also defined as, the manufacturing unit. Once the host cell is selected, the process development is initiated at small scale. Viral vectors can be produced transiently by transfecting host cells with plasmids encoding viral vector elements or with engineered producer cells that stably express all viral vector components including the transgene(s). For most stable expression systems, producer cells are derived from packaging cells, which express all viral vector genes but the transgene(s).

Once a mammalian cell is selected, its productivity will also depend on the process developed at scale. The production process can be initiated using pre-established parameters and off-the-shelf consumables or by customizing an entire solution including expression system, culture media, supplements, etc. In both cases, a process intensification phase followed by scale-up will be needed to improve viral titers. The production process development and scale-up will significantly differ between expression systems and from one vector to another. Different modes of production such as batch, fed-batch and perfusion exhibit different levels of complexity to set up, and have been adapted for transient or stable production. With suspension cells maintained under constant agitation, batch and fed-batch production can be tested at small scale in shake flask, whereas perfusion usually requires a bioreactor. As production is scaled up, a stepwise progression going from smaller to larger bioreactors allows one to set and adjust control parameters with the goal of replicating original conditions and expected titers. Each mode will require its own bioreactor-controlled parameters and configuration. While pH, dissolved oxygen (DO) and agitation rate qualify as bioreactor-controlled parameters, other parameters will impact the process including bioreactor geometry, impeller type and tip speed, head space and mixing capacities affecting mass transfer. This chapter will focus on the production and scale-up of recombinant lentiviral vectors (rLVVs) using serum-free suspension-adapted HEK293 cells.

2. Characteristics and applications of rLVVs

In addition to the natural features borrowed from viruses, molecular engineering of viral vectors has allowed the design of fit for purpose vectors. For example, mutagenesis and pseudotyping is used to change the tropism of a virus for narrowing [2, 3] or increasing the host range [4]. The range of host cells targeted by a lentiviral viral vector can be broadened by incorporating protein G from the vesicular stomatitis virus (VSV-G) in the envelope [5]. The VSV-G pseudotyped vectors will enter the cells via an interaction with the widely distributed low density lipoprotein receptor (LDL-R) [6, 7]. The most used rLVV in gene therapy is derived from human immunodeficiency virus 1 (HIV-1) although several non-primate alternatives have been developed such as feline immunodeficiency virus (FIV) or equine infectious anemia virus (EIAV) [8]. Different promoters can be introduced to regulate tissue-specific expression of the payload [9]. In theory, the modifications that can be done on viral vectors are limited by their intrinsic properties including capsid/envelope composition and assembly, cargo size, targeted cells, etc. In practice, viral vector engineering not only has an impact on the vector’s properties but also may significantly alter expression system productivity, purification and quality attributes.

Viral vectors are used to efficiently transfer a nucleic acid cargo to their natural target, usually a mammalian cell. These nucleic acid cargos encode information for the expression of one or more transgenes. The transgene(s) can function as a reporter, as a therapeutic, for gene editing or to up/downregulate gene expression. They can be used in vitro, ex vivo or in vivo on a large array of human and animal cells and tissues.

Recombinant LVs (rLVV) are derived from HIV-1, and thus share several common physico-chemical properties. They are enveloped viral vectors with a spherical shape of 80–120 nm in diameter. They are sensitive to several factors such as pH, osmolarity, shear stress, freeze-thaw cycles and temperature [10] and their half-life is in the range of only 3–18 h at 37°C [11]. Because of all the above, rLVVs are considered to be labile and bring a unique set of challenges in terms of bioprocessing. rLVVs have been made replication defective to increase their safety profile, and overall, rLVV safety and efficacy have been improved since their first developments in the early 1990s [12]. This was made possible by transferring only the essential rLVV elements on four separate plasmids used in co-transfection for production [13].

rLVVs are used in a multitude of applications. They can deliver genetic material of up to 11 kilobases (kb) in size with high efficacy in both dividing and non-dividing cells, including difficult to transfect and transduce cells such as human neuron, primary and stem cells. rLVVs were employed in more than 100 clinical trials in 2018 [10]. Lentiviral-based gene therapy has successfully treated multiple genetic blood cell diseases including Wiskott-Aldrich syndrome, X-linked severe combined immunodeficiency (XSCID), X-linked adrenoleukodystrophy, b-thalassemia and metachromatic leukodystrophy [8].

Another key use and success with rLVVs in therapies tested in the clinic is in the field of cancer treatment. Lentiviral vectors are used to deliver Chimeric Antigen Receptors (CAR) into patients’ T cells ex vivo. The cells expressing the modified receptor are re-introduced into patients where they can recognize the antigen of interest, e.g. CD19 or BCMA and attack tumor cells. Since 2017, six CAR-T cell therapies have been approved for hematological cancers by the Food and Drug Administration (FDA) [14].

3. Cultivation of cells for viral vector production

The most common mammalian cells for the production of viral vectors are human embryonic kidney cells (HEK293), human lung adenocarcinoma cells (A549) and kidney epithelial cells from African green monkey (Vero). These cells can be grown as adherent or in suspension in a bioreactor. Adapting cells from adherent to suspension is challenging, and while the adaptation of HEK293 cells has been achieved by many groups, only a few laboratories have succeeded in adapting Vero cells in suspension [15, 16]. Suspension grown HEK293 cells are the most popular cells for the production of viral vectors in stirred tank bioreactors. They are easy to grow; they can be engineered to develop packaging and producer cell lines and are permissive to the production of many types of viral vectors. In addition, culture media, supplements and vector systems are commercially available.

3.1 Cultivation of cells in two-dimensional (2D) systems

The cells used to make viral vectors can be grown in a two-dimensional (2D) system to provide a surface for the cells to adhere. Several types of vessels can be used in such static systems including tissue culture flasks (T-flasks), cell factories or cell stack. The main advantage of 2D systems is that they are relatively easy to implement. The productions are scaled up by multiplying the number of vessels. However, these systems have a significant footprint and the overall process is labor-intensive [17, 18].

Adherent cells can also be cultivated in stirred tank bioreactors on microcarrier beads made of glass, plastic or other material that provide a surface for the cells to grow. This combination significantly reduces the footprint but the process development is not straightforward and requires substantial expertise. A single-use fixed-bed bioreactor is the most recent device developed to scale up adherent cell expression platforms. It reduces physical space requirements; it simplifies process development and is offered by several manufacturers. While these systems can support the growth of 30–200 M cells per mL of fixed bed, the cells cannot be counted when transfection or infection is performed. The cells are concentrated in a proportionally smaller bioreactor volume, which increases the demand for nutrients. The nutrients can be supplied throughout the production phase either by perfusion or by a medium re-circulation strategy. Despite these challenges, some robust processes have been developed using fixed-bed bioreactors [19]. The largest fixed-bed bioreactors, such as the iCellis 500 from Pall or the Scale-X from Univercells [20], offer surface areas for 2D culture to up to 500 m² and 600 m² respectively. To obtain a similar growth surface using cell factories, approximatively 950 units would be required [19] while similar cell mass could be grown in 100–1000 L stirred tank bioreactor, depending on the achieved cell density [21]. iCEllis bioreactors have been successfully used to produce lentiviral vectors in chemically defined media [22].

3.2 Cultivation of suspension-adapted HEK293 cells in serum-free medium

The adherent human cell line HEK293T remains the host of choice for the production of viral vectors because it is well characterized and safe for clinical use. However, most adherent cell systems rely on serum supplemented media, generally fetal bovine serum (FBS). This increases the production costs, risk for contamination with adventitious agents and production lot-to-lot variability. While fixed-bed bioreactors have improved production with 2D systems, investments and efforts are being made to replace traditional 2D systems with large-scale productions that offer more practical options for high-titer viral vector–based productions in adherent mode. Preferred host cells for the development of expression systems for viral vector production are robust, have short doubling times and grow in suspension at high cell density in chemically defined media absent of serum or supplements derived from an animal source. Cells that grow in suspension can be cultivated in stirred-tank bioreactors to reduce the footprint and process labor.

The most common systems for viral vector production use mammalian cells that are permissive to vector production while limiting the production of replication competent particles. The human embryonic kidney cell HEK293 is one commonly used cell line for production of viral vectors [23]. Of note, HEK293 cells stably express the adenoviral E1A and E1B-55 k genes that support a helper function necessary for the propagation of AAV and Ads vectors [24].

An example of suspension HEK293 cells is the HEK293SF-3F6 cell line. It was adapted to suspension in 1998 in serum-free media [25] through a series of adaptation steps while reducing calcium concentration. The final clone with demonstrated monoclonality, namely HEK293SF-3F6, was selected on the basis of its doubling time of approximately 24 hours. Since its establishment, the cell line has been used for bioreactor productions of adenovirus [26], AAV [27], lentivirus [28], an Ebola vaccine [29] and influenza virus and viral-like particles [30].

4. Production of recombinant viral vectors by plasmid transfection

When optimizing a transfection process, a Design-of-Experiment (DoE) approach is preferred because of the large number of parameters that could possibly interact and affect the result. DoE is now well established for improving the production of recombinant proteins, especially with the growing implementation of miniaturized parallel automated bioreactors such as the Ambr®15 and Ambr®250 from Sartorius [31, 32]. In general, parameters targeted when optimizing transfection of rLVV in serum-free medium include relative ratio of each plasmid, total quantity of DNA, PEI to DNA ratio, cell culture medium composition and cell density at transfection [33]. As for other biologics applications, we foresee a broader use of the DoE approach, combined with the use of high-throughput parallel bioreactors for rapid and efficient optimization. Improving viral vector titers is multi-factorial and the transfection method alone cannot adequately explain differences in production yields. For rLVV, size and nature of the gene of interest (GOI), the mode of production and the expression system all influence vector quantities and quality. In general, transfection methods are efficient at small scale but costly and difficult to scale up, especially with GMP quality plasmids.

4.1 Transfection reagents

The production of recombinant viral vectors by plasmid transfection is relatively simple and accessible at small scale. Several non-viral transfection methods using cationic reagents have been developed to deliver the plasmids to the host cells. The most common methods include calcium phosphate precipitation, polyethyleneimine (PEI) or cationic lipid such as lipofectamine.

The most affordable transfection method using calcium phosphate co-precipitation with plasmid DNA has been used for decades in HEK293 adherent cells, but also with suspension cells in stirred tank bioreactor [34, 35]. The main drawback however is the sensitivity of the precipitation process to pH or agitation force [36]. Furthermore, calcium phosphate co-precipitation usually requires the presence of animal-derived components in the medium such as serum or albumin [37, 38] that is frowned upon by regulators. In comparison, the transfection efficacy of cationic lipids has been reported to be similar to or even better than calcium precipitation [39] or polyethylenimine (PEI) [40, 41]. However, their use is limited due to their cytotoxicity [10] and higher costs [10, 36, 42]. Non-chemical non-viral methods, such as flow electroporation, are also available for the transfection of HEK293 cells in suspension for the production of viral vectors such as lentiviral vectors [43]. However, the scalability of this method is limited by the need to exchange medium and concentrate the cells.

4.2 Production of rLVVs by plasmid transfection

rLVVs have been produced by transfection in HEK293 cells adapted to grow in suspension in serum-free media using four plasmids. One plasmid contains the Gag/Pol genes that encode key enzymes: a reverse transcriptase, integrase and protease. These genes also mediate crucial steps in vector assembly including the recruitment of Viral Env protein, packaging of the RNA payload and binding to the plasma membrane. A second plasmid encodes the Rev gene essential for transporting the unspliced RNA into the cytoplasm by interacting with the cis-acting Rev response element (RRE). For pseudotyping the vector particles, a third plasmid contains VSV-G gene. The fourth plasmid encodes the GOI with its specific promotor and regulatory elements [8, 10, 42, 44]. The development of lentiviral vectors is now at its third generation, which presents additional safety features over previous generations [45].

4.3 Production of rLVVs by plasmid transfection: challenges and opportunities for improvement

The reproducibility of transfection methods can be variable, especially at large scale. PEI and DNA are mixed together in specific ratios usually ranging between 2:1 and 3:1 [11, 46, 47]. The relative ratios of the different viral vector plasmids used for transfection might result in poor packaging and the accumulation of empty vectors and batch-to-batch variability [18] . Because PEI alone is relatively cytotoxic to the cells [36, 48], the PEI:DNA ratios must be carefully optimized to maximize transfection efficacy and minimize cytotoxicity. Once mixed, the polyplexes (complexes formed between PEI and DNA) grow in size over time [49, 50, 51]. Efficient polyplexes for transfection have a net positive charge and a time-sensitive size distribution. Polyplexes that are too small or too large will not transfect cells efficiently [52]. In most optimized protocols for HEK293, PEI and DNA are mixed together, and incubated (Figure 1) for a maximum of 30 minutes (usually 10–15 minutes) in a total volume accounting for 5–10% of the cell culture volume to be transfected [46, 47, 53, 54]. Other process-related factors, such as the medium and ion concentration used in the plasmid DNA mix, will have an impact on particle formation and transfection efficacy [49, 50, 55, 56, 57]. Therefore, the optimal incubation time may vary between processes and products. Given the relatively short mixing time and size of transfection mix, it becomes technically challenging to complete transfection within these parameters at increasing scales. Indeed, the time for mixing large volumes of DNA and PEI and transferring the complex to the bioreactor at a given transfer rate significantly increases with the scale. This also narrows the time window of operation, depending on whether the mix is transferred by gravity or pumped in the bioreactor. When adding DNA to PEI mix or vice versa, addition rates and shear stress introduced by mixing velocity are also important parameters to be considered [55, 58].

Figure 1.

Example of a polyethyleneimine (PEI)-mediated LVV transfection process for large-scale production. (A) The four plasmids i.e., the gene of interest (GOI), Gag/Pol, Rev and the glycoprotein G of vesicular stomatitis virus (VSV-G) encoding plasmid, are diluted in buffer (Fresh PBS or cell culture medium) and mixed together. (B) PEI, diluted in buffer (Fresh PBS or cell culture medium), is added to the mix via pumping, while agitation allows for constant mixing of the solution. (C) Agitation is stopped, and the mixture is incubated for 10–15 minutes to allow for the PEI-DNA complex to form. (D) The PEI-DNA complex mixture is added to the bioreactor via pumping, with special attention to the pumping rate and tubing size to minimize the shear stress applied to the complex

The optimal HEK293 cell density for transfecting plasmids for producing rLLVs usually ranges between 1.0 and 2.0E6 cells/mL, when cells are still in the exponential growth phase. While higher densities can easily be achieved in batch cultivation, the transfection process becomes suboptimal. For HEK293 cells, the transfection efficacy and the specific productivity, that is, the number of viral genome copies produced per cells, decreases significantly when cells density is above 1.0E6 cells/mL [27]. Even when the quantities of plasmid and PEI are proportionally increased at higher cell densities, the titers do not significantly increase and cost-effectiveness is reduced. The mechanism for this is not well understood, but it may be a result of the accumulation of an inhibitory metabolite or secreted protein (shedding of HSPG), or to a change in the physiological state of cells grown at higher densities [59].

Alternative production modes such as perfusion applied to a process intensification approach may improve the physiological state of the cells at high densities and allow for higher volumetric titers and cost-effective transfection. Current literature on production of viral vectors by transfection indicate that cell engineering, vector and capsid design as well as novel and low-cost reagents producing stable complexes and using less plasmid DNA could contribute to increase vector quality, batch-to-batch consistency and cost-effectiveness [60].

5. Development and production of rLVVs with packaging and producer cell lines

Although the production of viral vectors by transfection is relatively simple and accessible, it scales poorly, it is susceptible to batch-to-batch variability and the need for GMP-grade plasmids and PEI to produce clinical material is cost prohibitive. One attractive approach to offset some of these cost- and scale-related challenges is to develop packaging cell lines that stably express the genes essential for viral particle formation. The production of rLVVs in packaging cells necessitates only one plasmid that carries the GOI. The GOI is stably integrated in the genome of producer cells, where rLVV production is controlled by an inducible system [23].

Stable packaging cell lines for lentiviral vectors are challenging to engineer because some viral vector proteins are toxic to producer cells, most notably VSV-G and the protease encoded by the Gag/Pol gene. Inducible systems have been used to control the expression of toxic viral proteins with some success. However, many of the tested inducible systems were derived from microbial operons including tetracycline, cumate and coumermycin that can be leaky [8]. Broussau et al. have developed a dual control system by combining tetracycline and cumate operons to regulate the expression of cytotoxic vector elements to produce a stable packaging HEK293 cell line [61] and later used a similar strategy by combining coumermycin and cumate to generate a producer cell line [62].

In producer cell lines, the GOI is also under the control of a molecular switch as part of the inducible system. The production of the rLVV is initiated either directly by inducers in producer cell lines or after transfection of the plasmid with the GOI combined with inducers in packaging cell lines (Figure 2). The lowest overall manufacturing costs are obtained with suspension producers [63] where the product is not contaminated by residual plasmids and transfection reagents. This simplifies the subsequent purification steps. However, generating such a cell line requires 6–12 months for an experienced team, which can be difficult to accommodate with aggressive timelines in competitive commercial landscapes and to address the needs of patients for therapies of last resort. The packaging cell line offers the versatility and speed of execution of the production with the four-plasmid transfection method. In addition, optimizing the process is easier with packaging cell lines.

Figure 2.

Evolution of the rLVV expression systems towards the reduction of cost of goods (COGS). Recombinant lentiviral vectors (rLVV) can be produced with adherent or suspension-adapted cell lines. Although fully developed, packaging and producer cell lines represent a significant investment they simplify the production and reduce the costs of plasmids; only the gene of interest (GOI) plasmid is needed for the production with packaging cells. Despite that process development and scale-up for suspension cells can be less strait forward than adherent cells it will in the long term significantly reduce the costs of production. The least cost effective method most labour intensive to produce rLVV is by transfecting adherent cells with 4 plasmids respectively encoding, Gag/Pol, Rev, the glycoprotein G of vesicular stomatitis virus (VSV-G) and the GOI. The most cost effective platform for the production of large quantities of rLVV combine a suspension-adapted producer cell line and a production process in a stirred-tank bioreactor.

Generating stable packaging and producer cell lines requires a high level of expertise. A recent review reports a total of 10 stable producer cell lines for the same viral vector (Green Fluorescent Protein) yielding titers varying from 10⁵ to 10⁷ TU/mL [10]. Of those, only one grew in suspension in serum-free condition, which is the most desirable option for large-scale production [61].

The development of stable packaging or producer cell lines is a major step toward the development of cost-effective and scalable production platforms. The development of molecular switches that are tighter in mammalian expression systems would allow for better control of cytotoxic viral genes and in turn increase the stability of packaging and producer cell lines. High-throughput automated cloning systems and improvements in methods and strategies for selecting high producing clones should allow, in a near future, the development of rLVV producer cell lines capable of producing high titers on a large scale for clinical application.

6. Modes of production: batch, fed-batch, perfusion

rLVV can be produced using transient and or stable systems with HEK293 suspension cells. The production processes supporting each system can be developed with different cultivation modes, batch, fed-batch or perfusion and adapted for each system to consolidate a production platform (Figure 3). The decision to produce using one mode or another rests with numerous factors linked to the viral vector characteristics and its corresponding expression system, the targeted titers and batch size needed, downstream processing needs and the availability of appropriate consumables and equipment. Scaling up a process using equipment from different manufacturers with specific characteristics and configuration is not straightforward and will impact the process development and timelines.

Figure 3.

Process steps for batch, fed-batch and perfusion processes for the production of viral vectors.

6.1 Batch mode

The batch mode is a simple and rapid production process initiated with bioreactor seeding. Depending on the seed train strategy, the bioreactor can be seeded at full bioreactor working volume, or at 25–50% of the final process volume. The seed train strategy is defined by the amplification of cells to obtain a defined volume and cell density to inoculate a bioreactor that is aligned with the targeted final cell density, working volume and production schedule. Under some processes, cells are diluted to the final processing volume after seeding. This simulates cell growth up to the density where the production is initiated either by transfection for HEK293 and packaging cells or by induction of stable producers. The addition of histone deacetylase inhibitors, sodium butyrate or valproic acid on the following day has shown to significantly increase titers [64]. Typically, the harvest takes place when maximum titers are obtained, while at the same time aiming for high cell viability to minimize cell debris and other contaminants that reduce yield at subsequent clarification and purification steps.

6.2 Fed-batch mode

A process developed in a fed-batch mode will, for the most part, follow the same steps as a batch mode and be complemented with a feed regimen. The feed regimen consists of one or more additions of feed at a given time and intervals. Feeds are concentrated cocktails of nutrients, and commercially available feeds have been specifically designed for most producer cell lines, including HEK293 cells. The feeds support cell growth and production and help prevent rapid nutrient depletion. For each cell line, several feeds are available and the selection of the best feed and regimen for each process is determined in small-scale studies. Best results usually occur when feeding during the growth phase, but can be done any time during the process to maximize titers. Fed-batch processes typically result in higher cell densities and may also improve the cell’s specific productivity and overall volumetric titers. However, the impact of the presence of feed in the harvest on downstream processing and final yield must be evaluated before scaling up the process. Both batch and fed-batch modes are well adapted for the production of viral vectors by plasmid transfection as well as stable producers.

6.3 Perfusion mode

In a production using perfusion mode, a flow system with an in-line separation device is interconnected between the bioreactor and a reservoir for harvesting cell-free spent media pumped out of the bioreactor. The working volume is maintained throughout the process by pumping in equal volumes of fresh media. The separation device is designed to keep the cells in the bioreactor and let through the smaller viral vector particles. The perfusion or media exchange is initiated during the growth phase of the cells, when the final process volume is attained and potentially after dilution. The renewal of media during a perfusion process allows for a constant supply of fresh nutrients, feed and inducers if needed and the removal of metabolites that may become cytotoxic when accumulating. The media exchange rate can vary from 1 to 3 vessel volumes per day (vvd), depending on the expression system.

A perfusion process will allow to increase the cell density by several fold in comparison to a batch or fed-batch process. Depending on the perfusion device and the presence or absence of cell-free media, the viral vectors in the harvest can be purified in batches. This is desirable for unstable vectors such as rLVVs. Stable producer cell-lines are the preferred expression systems to couple to a perfusion mode. In addition, continuous bioprocessing can be implemented by directly feeding clarified permeate into the DSP flow path in a setting where refrigerated storage capacity for the collected permeate may be reduced or even eliminated. Especially in the later scenario, perfusion systems enable the reduction of the production scale and equipment footprint.

There are several types of perfusion systems and retention devices commercially available with different functional designs, including alternating tangential flow filtration (ATF), tangential flow filtration (TFF) and acoustic devices such as the BioSep. ATF and TFF are filter-based devices composed of different types of fibers and variable pore sizes that need to be carefully selected to ensure the membrane is permeable to the viral vector produced [65].

The ATF system (Figure 4, C-1) (XCell^® ATF System, Repligen) consists of a dip tube, a hollow-fiber module and a diaphragm pump. Culture flows from the bioreactor through the dip tube and hollow-fiber filter when a vacuum is applied to the diaphragm pump. To reverse the direction of the flow, a pressure is applied. The alternating diaphragm movement creates a low shear flow for the cell suspension as well as a backflushing effect across the filter surface of the hollow-fiber filter, in contrast to traditional TFF methods. Filter fouling and retention of the product of interest are thereby reduced, increasing the lifespan of the filter and allowing for longer perfusion processes. The filtered media or permeate is collected in a refrigerated container using a peristaltic pump. Hollow-fiber modules made from different materials and with a range of pore sizes are available from several manufacturers. In general, permeates from ATF systems are cell-free and contain fewer cell debris. However, some devices are subject to fouling depending on the bioprocess parameters (cell density, viability, DNA, feed and protein content, etc.) and process parameters (backflush, permeate flow rates). ATF devices and filters are currently available for bioreactor sizes of up to 1000 L (Table 1). ATF perfusion technology has been successfully used for viral vectors such as adenovirus, rAAV and rLVVs [71, 72, 73].

Figure 4.

Seed train (A) and process set-ups (B and C) for viral vector bioreactor production. (A) Cells are thawed and maintained in shake flask, amplified in (a) shake flasks for small-scale bioreactor production (e.g., up to 10 L production volume) or in (b) wave bioreactor to constitute the seed train. (B) Bioreactor set-up for batch or fed-batch production (“feed” is added during the fed-batch process). (C) Bioreactor set-up for perfusion production (C-1: ATF, alternating tangential flow; C-2: TFF, tangential flow filtration; C-3: acoustic retention device). A higher media exchange or perfusion rate in general results in higher sustained cell density and faster removal of product, while a lower vessels volume per day (vvd) causes less strain on filter-based retention devices such as (C-1) or (C2).

Technology	Example	Bioreactor scale	Pore size/MWCO1	References
ATF2	XCell^® ATF System,; Repligen	0.5–1000 L	0.2–0.5 μm/50 kDa	[66]
TFDF3	KrosFlo^®; TFDF^® System, Repligen	1–2000 L	2–5 μm	[67]
TFF4	Xcellerex™ APS, Cytiva	50–500 L	0.1–0.45 μm; 10–750 kDa	[68]
TFF4	Cellicon^® cell retention; solution, Millipore Sigma	3–2000 L	5–15 μm	[69]
ATF2 or TFF4	VHU^® System, Artemis BIOSYSTEMS	0.5–1000 L	5–15 μm	[70]

Table 1.

ATF and TFF perfusion technologies.

¹

MWCO: molecular weight cut-off.

²

ATF: alternate tangential flow filtration.

³

TFDF: tangential flow depth filtration.

⁴

TFF: tangential flow filtration.

TFF perfusion systems (Figure 4, C-2) are available from different suppliers. With this perfusion technology, the cell culture is pumped through a tubular filter in tangential flow mode, driven by a low-shear peristaltic pump. Most of the feed, including the cells, is redirected back to the bioreactor. Part of the clarified culture and components smaller than the pore size of the filter pass through as permeate and are pumped into a refrigerated harvest container. For viral vectors with 20–100 nm diameters, TFF filters with an effective pore size rating of 2–5 μm are recommended for filtering out cells and most cell debris. Repligen has developed a TFDF perfusion device that combines the principles of tangential flow and depth filtration while following the same technical principles as a standard TFF perfusion process. TFDF perfusion has been successfully used for AAV and LV processes [65, 72, 74]. In general, the higher pressure in TFF systems is more likely to cause shear stress than ATF perfusion and is subject to filter fouling [72]. The TFF perfusion technology is currently available at up to 2000 L bioreactor scale (Table 1).

An acoustic device (Figure 4, C-3), such as the Applikon BioSep (Applikon) uses ultrasonic waves to separate cells from culture medium in a resonator chamber. The chamber is composed of two opposed glass surfaces. When specific frequencies are selected, an acoustic standing field is generated between the glass walls. Culture that is pumped from the bioreactor into the chamber forms loose cell aggregates in the acoustic field, which then sediment out of the field, disaggregate and return to the bioreactor through a dip tube. A second pump removes harvest from the exit port of the resonator chamber. The combined flow of harvest and culture returned to the bioreactor is equivalent to the culture circulated from the bioreactor to the resonator chamber. While the harvest contains some dead cells and debris, an acoustic device is gentle and non-fouling and can operate continuously for up to thousands of hours. Acoustic perfusion technology has been described in the literature for different viral vectors [11, 28, 75, 76]. The Applikon BioSep is currently scalable from 0.1 to 1000 L/day) [77].

7. Scaling up the production of viral vector

For suspension cell platforms such as the HEK293 cells, many options are available for large-scale manufacturing. The stirred tank reactor (STR) remains the gold standard for manufacturing when large quantities of material for clinical trial and commercial production are needed. Other alternatives are available for suspension cell lines, such as rocking motion bioreactors [47] and orbitally shaken bioreactors [78] for viral vector production. Most suppliers only provide these alternatives in a volume range from 25 L to 100 L [79]; however their use at a volume of 200 L and beyond has been tested [80]. The production of both rLVVs and rAAVs by transfection is challenged by the limited scalability of transfection processes and COGS, especially for GMP quality plasmids.

7.1 Principles of cell culture scale-up

While stirred tank bioreactors are the preferred model for large-scale manufacturing, they do present unique scale-up challenges to maintain favorable oxygenation and mixing conditions that can support high viable cell densities. One key objective of scale-up strategies is to limit shear stress linked to hydrodynamic conditions to which HEK293 and other mammalian cells are sensitive. Several factors may influence the hydrodynamic conditions in a given stirred tank bioreactor: the dimension and geometry of the bioreactor, the operating cell culture volume, the gas flow rates (air and O₂) delivered by different types of spargers mixed in the liquid phase and the presence of baffles or other obstacles. Finally, the moving force of the impeller that is dependent on impeller size, type and setup is crucial. A bioreactor can be configured with one or more impellers positioned at variable distances from the bottom of the bioreactor and from each other. The impeller shaft can be positioned within the central axis or off-centered, or parallel to the side wall. Whether the impeller setup creates an upward or downward flow pattern will also affect the hydrodynamic conditions.

Formulas to calculate the agitation rate of the impeller have been developed and are used in scale-up processes (Table 2). A typical scale-up strategy consists in selecting one physical property that is kept constant across the different scales. For example, to maintain the volumetric power input and volume of gas per volume of liquid per minute (vvm) constant, the agitation and sparging rates are calculated from their respective equations using equipment dimensions from each scale. The drawback of this approach is that it is physically impossible to keep all the impact parameters constant during scale-up, and some must be prioritized. Bioreactor scale-down models are useful tools to generate large datasets to explore trends that link different operation conditions and physical constraints to the cell culture performance and productivity [83, 84]. Based on the data generated at smaller scale, a design space for large-scale manufacturing can be established, where all the relevant physical parameters are within acceptable boundaries for cell growth and viral vector production.

Parameter	Equation
Volumetric power input P/V (W/m^3)	$\frac{N_p{}^{\ast }{\rho }^{\ast }{N}^{3\ast }{D}^5}{V}$ Eq. (1)
Tip speed (m/s)	${\pi }^{\ast }{N}^{\ast }D$ Eq. (2)
Reynold number (−)	$\frac{{\rho }^{\ast }{N}^{\ast }{D}^2}{\mu }$ Eq. (3)
Volumetric oxygen mass transfer coefficient (kLa, h^−1)	$A^{\ast }{\left(\frac{P}{V}\right)}^{\alpha \ast }{\left(v_s\right)}^{\beta }$ Eq. (4)

Table 2.

Scale-up criteria equations for agitation rate in reusable bioreactors.

N refers to the agitation rate (s⁻¹), D to the impeller diameter (m), ρ the density of the liquid (kg/m³), V to the volume (m³), μ to the viscosity of the liquid (Pa*s) and v_s to the superficial velocity of the gas into the vessel (m/s). N_p, the dimensionless power number of the impeller often referred as Newton number [81], is an impeller-specific and vessel-specific characteristic of the mixing system. A, α and β are coefficient that might be either estimated from values in the literature [82] or estimated experimentally for each vessel for more accurate results.

The power input per unit of volume (P/V) (W/m³) is a very common parameter to scale-up agitation rate, as it allows keeping the quantity of energy transferred to the cell culture via the impeller constant (per unit of volume) during the scale-up. HEK293 suspension cell culture requires volumetric power input values between 13 and 44 W/m³ [85, 86, 87, 88, 89]. HEK293 cells in small-scale benchtop vessels have been shown to tolerate specific power input of much greater intensity, up to 451 W/m³ in [90]. Using specific power input for scale-up also correlates well to the oxygenation rate. It has been demonstrated that the coefficient of volumetric oxygen transfer is strongly correlated to the power input volume via the Van’t Riet correlation [91]. Indeed, increasing agitation tends to break larger bubbles into smaller ones, thus increasing the surface available for oxygen transfer.

The tip speed (m/s, Table 2, Eq. (2)) is a parameter linking shear stress applied to the cells and agitation rate. Tip speed of ~1–2 m/s is generally acceptable, although there is no consensus on how to interpret this parameter [92]. Reynold number (Re, Table 2, Eq. (3)) is a parameter linked to the agitation rate that represents the ratio of the momentum force of the flow compared to the viscous shear forces. While not used directly for agitation scale-up calculations, the evaluation of the Reynold number allows to establish the flow regime of the bioreactor [82]. A Re value greater than 10,000 corresponds to a turbulent flow regimen, which is what is typically observed in a large-scale bioreactor.

The agitation in the bioreactor has for a long time been considered as the main contributor to shear stress leading to cell damage. However, bubble fragmentation in bioreactors was more recently identified as a highly probable contributor to shear stress [93]. Bubble fragmentation is induced by a combination of impeller velocity and sparging rate. Importantly, very high specific energy dissipation rates are associated with every step of the gas sparging, from bubble formation to bubble fragmentation at the surface. While some bubble fragmentation seems to be well tolerated by robust CHO cells, HEK293 cells could be more susceptible to this phenomenon. The sensitivity of HEK293 and CHO cells to bubble fragmentation was compared using a small-scale perfusion process. The study concluded that the agitation and sparging conditions were harmful to HEK293 but not CHO cells [94].

Pitch blade/elephant ear impellers [92] are commonly used for mammalian cell culture applications. This type of impeller allows for a good compromise between radial and axial pumping, thus allowing for both sufficient oxygen transfer by effective bubble breaking and dispersion, and bulk mixing to maintain a homogeneous mixture. In addition to shear stress, the agitation rate may influence the tendency of suspension adapted HEK293 cells to aggregate [90, 95]. The cell aggregation can affect the growth properties and productivity of the cells and is very difficult to control at larger scales. Sparging is critical to monitor and control during process scale-up. Over- or under-sparging will induce either foaming and cell damage, or oxygen limitations and CO₂ accumulation respectively. One way to fine-tune the bioreactor’s oxygenation consists of using a combination of gases (nitrogen, air, oxygen). Air sparging supplemented by oxygen on-demand is the preferred approach used as a scale-up strategy.

The oxygenation rate for a specific bioreactor is linked to fraction of oxygen in the gas composition and the volumetric mass transfer coefficient k_La (Table 2, Eq. (4)). The k_La value depends on cell culture medium properties and the surface area available for oxygen transfer [96]. Oxygen transfer from the bubble to the liquid is a limiting factor because the solubility of oxygen in aqueous solutions is low, particularly at 37°C. In bioreactor operation, the most practical way to increase oxygenation is to increase the surface area of the bubbles by increasing the sparging rate and/or reducing the size of the bubbles. Different sparger types will have different pore sizes as well, resulting in variable bubble sizes that will also influence the oxygenation rate greatly. The characterization of the stirred tank bioreactor oxygen transfer coefficient k_La is an important criterion for selecting and operating a bioreactor. The gassing-out method based on removing dissolved oxygen by gassing nitrogen is recommended for k_La determination [81]. Bioreactor suppliers will often provide specific values and guidelines for their equipment. A k_La value for oxygen in the order of 1–10 h⁻¹ should cover most standard cell culture applications for viral vector production [97].

Lastly, CO₂ is also as a potential limiting factor for high cell density culture when scaling up. Larger bioreactors tend to accumulate CO₂ at a higher rate than smaller bioreactors because of the difference in mass transfer mechanism between oxygen supplementation and CO₂ removal [98, 99, 100]. It has rarely been reported as an issue in scaling up viral vector bioprocesses, likely because viral vectors are still produced at low cell densities. Still, the type of sparger and the sparging rate scale-up strategy must be tested and selected with care for a specific cell line to obtain a good balance between oxygenation, CO₂ removal and shear stress applied to the cells.

7.2 Large-scale production in reusable bioreactors

Reusable bioreactors made of glass and stainless steel such as clean in place (CIP) steam in place (SIP) stirred tank bioreactors have a strong track record for the production of biologics at various manufacturing scales. These bioreactors have been used for decades for monoclonal antibody production up to the 25,000 L scale [101]. Their scalability, while still challenging, is very well documented and studied. They can be used from process intensification to scale-up development, from small-scale screening (0.25 L) to the commercial scale. Mixing of the bulk liquid phase is done by one or more impellers, and gassing is done via surface aeration and sparging at the bottom of the vessel. Gas sparging with oxygen-enriched air ensures appropriate dissolved oxygen (DO) levels. CO₂ sparging is used to lower pH value to physiological levels. Both dissolved oxygen and pH are monitored continuously during the production. Electrochemical glass sensors for pH monitoring and amperometric Clark-type sensors for dissolved oxygen levels are commonly used [102]. Both types are available in steam-in-place, clean-in-place versions, suitable for repeated use in bioreactors.

7.3 Single-use bioreactors: configuration and costs of goods matters

One of the heavy trends in the cell culture industry over the last decade is the adoption of single-use bioreactors (SUBs) to replace reusable stainless-steel ones. Benchtop bioreactors up to 50 L are typically made of rigid polymers while pilot scale bioreactors of 50 L–6000 L are inflatable bags that can be inserted in temperature-controlled jackets. Both bench top and pilot scale bioreactors are fitted with their own pre-sterilized impeller, sparger, inlets, outlets and probes [103]. SUBs do not pose any risks of carry-over contamination from one batch to another, or from one product to another between different batches or campaigns, respectively [104].

When the rLVV productivity is similar in suspension vs. adherent HEK293 cells, large scale production in STR is usually more cost-effective. Comisel et al. [105] compared the cost of lentivirus production in packed-bed versus stirred tank bioreactors and concluded that the single-use bioreactor was the most cost-effective means of culturing cells for rLVV production at a scale of 200 L with suspension cells, compared with cell factories or packed-bed bioreactors and adherent cells. Using a suspension cell line for rLVV in a stirred tank bioreactor reduced 95% of the COGS when comparing STR and cell factory, or 17% when comparing STR and packed bed reactor. While the difference is less significant in earlier clinical trials, it does become more and more significant according to their finding as the scale of production increases during the product development process cycle.

Because of these characteristics, SUBs are broadly used for clinical trials, since they do not pose any carry-over risks between campaigns. In addition, their size (up to 2000–5000 L depending on the manufacturer) is perfectly suitable for most current and future viral vector applications. While an attractive solution for new viral vector products, single-use bioreactors do have some limitations. There are some concerns over the potential risks with leachable and extractable compounds from the single-use plastic surfaces [104]. These compounds can both affect the growth of the cell lines and contaminate the final product. Another concern is the lack of uniformity between the different single-use product manufacturers and the reliance of the customer toward its supplier for single-use products [106]. For bioreactors, each supplier has its own exclusive single-use bioreactor design that fits with the equipment, with its geometric characteristics often unique to the system. Unlike traditional stainless-steel vessels with well-established designs, single-use bioreactors come in a variety of designs affecting impeller position and angle, as well as the shape of the vessel itself. These designs have been adopted to increase mixing efficacy in the absence of baffles and multi-impeller agitation shaft (Figure 5). These systems increase complexity of scale-up and technical transfer due to geometric dissimilarities across different scale and limited availability of small-scale bioreactor models for scaling-down processes.

Figure 5.

Bioreactor design for single use bioreactors (SUBs). Comparison of bioreactor design for single-use bioreactors compared with traditional, multi-impeller baffled STR (System A). First row represents the profile view and second row the top view. In order to compensate for the lack of baffles and to prevent vortex formation, asymmetric bioreactor design have been developed for single-use bioreactors, such as with an eccentric position (B), an offset with agitation shaft (C), or by adopting a cuboid design for the vessel instead of a cylindrical one (D). Arrows are an approximation of the movement of the impeller on the flow pattern (axial) for each design [107].

8. Conclusions

Decades of research and progress in the development of safer viral vector and biomanufacturing processes has allowed the commercialization of cell and gene therapies. Despite these major accomplishments, there are still numerous obstacles and remaining challenges to produce and commercialize cost-effective cell and gene therapies. There will always be a need for new and improved reagents and analytical methods that reduce the COGs of viral vector biomanufacturing, such as transfection reagents and media that support high viable cell densities. The increase in manufacturer’s offering of SUBs with various configurations, cell culture media, and feeds with new formulations has increased complexity in the selection of equipment and consumables to develop optimized expression systems and processes. Further, the COVID pandemic has highlighted the unresolved challenges for low-cost, consistent, traceable and well-characterized raw material and consumables and emphasized the vulnerability of globalized supply chains.

The reduction of the dependency of rLVVs biomanufacturing to high COGs, GMP-grade plasmids and transfection reagents in particular is being addressed by the development of packaging and producer cell lines. One of the main engineering challenges is to tighten the expression systems that regulate on demand the expression of toxic proteins such as VSV-G used in rLVVs pseudotyped particles. Ideally, a packaging or producer cell line should be robust, stable and highly scalable and produce high titers to facilitate GMP compliant processes. While titers are often used to decide on the best clones early in cell line development, all the above criteria should be considered in the evaluation. As such, it is worth considering engineering or subcloning parental and packaging cells to select host cells for their robustness and scalability into which a packing or producer cell line would be developed.

Notwithstanding that rLVVs are from design to production at scale under continuous improvement, the relatively limited information available in the public domain for the biomanufacturing and scale-up of rLVVs, and the lack of reference material and platforms that can be used to benchmark process development, is slowing down progress. The biomanufacturing of rLVVs is further challenged by the fact that it is a labile vector that needs to be processed with relative care. The nature of lentiviral vectors makes them compatible with the development of stable producers, perfusion processes and continuous manufacturing. However, the need for stable producers and complex biomanufacturing processes can only be justified by the need for large quantities of rLVVs. The emergence of rLVVs as powerful vectors for vaccination could be the driver that will accelerate the development of large-scale perfusion processes and continuous manufacturing.

Batch and fed-batch production are more appropriate for cell and gene therapy applications for rare diseases, personalized medicine and production in SUBs where pilot scale productions may be needed. A better understanding of SUBs design space and hydrodynamic environment would benefit their use in various bioprocesses. This knowledge combined with the current scale-up principles should help define optimal operating conditions and configurations in bench top bioreactors and develop reliable and predictive scale-down models for rLVVs production that are representative of manufacturing scale bioreactors. The potential of rLVVs to address unmet medical needs is very promising. To support this, access to more robust cell lines for transient and stable production, improved transfection methods and better scale-up tools and models will contribute to accelerate the development of rLVVs and reduce the COGs for biomanufacturing.

Acknowledgments

The authors would like to thank the human health therapeutic research center of the national research council for supporting the publication of this book chapter. We would also like to thank Dr. Mauro Acchione for reviewing the text.

Conflict of interest

The authors declare no conflict of interest.