Lentiviral vectors (LV) are the most common delivery method for transducing T cells for CAR T cell therapies. LV production and purification are major cost drivers in CAR T manufacturing, with an industry ‘gold standard’ of 20% to 40% recovery. This poor overall recovery leads to oversized and expensive production batches. The first step of purification, physically segregating LV from the cells from which they bud, remains a serious challenge, especially in bioreactor harvests. Legacy methods such as batch centrifugation lead to poor recoveries, high cost, and labor-intensive processes.
In this study, we chose filtration as a scalable, closed alternative to centrifugation to remove producer cells, debris, and large aggregates from the LV feed. Here we discuss development of a primary clarification protocol to separate LV from bioreactor-derived suspension cells and debris using a scalable filter train. Producer cells and debris were cleared from LV with minimal to no loss of infectious titer.
Lentiviral vectors (LV) are the most common delivery method for transducing T cells for CAR T therapies, and producing and purifying them are major cost drivers in manufacturing. An industry gold standard of 20% to 40% recovery results in oversized and expensive production batches (1). Effective and consistent purification strategies for LV are a serious challenge for four reasons: the labile nature of the virus; the need to physically segregate LV from the cells from which they bud; the need to remove host cell DNA (HCD) and protein (HCP); and the use of 0.2 µm sterile filtration for high concentrations of 0.08 to 0.12 µm particles. Although unit operations from monoclonal antibody (mAb) and vaccine bioprocessing are well developed, they have yet to be successfully adapted and commercialized for LV (e.g., affinity chromatography) or are detrimental to infectivity (e.g., anion exchange). Thus, a host of wildly divergent, open, and nonscalable schemes are being developed across industry and academia. These schemes result in poor recoveries, inconsistency of product, and risk of contamination.
The objective of our overall process development (PD) project was to develop a closed, single-use downstream lentiviral vector purification process workflow that could be achieved in a single day at ambient temperature (Fig 1) to:
- robustly and consistently purify and concentrate LV to a minimum of 1 × 108 TU/mL with 25% recovery and 2 to 3 log reduction of host cell protein and DNA
- replace nonscalable ultracentrifugation with filtration
- replace the capture step from bioprocessing with tangential flow filtration (TFF) due to a lack of commercially available affinity chromatography solutions for LV
- apply a novel multimodal chromatography approach for purification
- develop effective methods for sterile filtration
The combination of single-use consumables, reduced processing time, and increased recovery will greatly reduce cost of goods (COGs) for the entire CAR T manufacturing process and thus increase access to these powerful new treatments.
Here we focus on the second bullet point, using a scalable filter train to replace legacy methods, which are not scalable.
See tab at bottom of page for detailed Materials and methods.
Fig 1. Developed LV production and purification process. LV are produced by cells grown in suspension in serum- and animal-derived component free (ADCF) media in a stirred-tank bioreactor (STR). DNA digestion is completed in the bioreactor, followed by harvest and clearance of cells and large debris by clarification. The diagram also shows the subsequent steps in the purification workflow.
The first step of purification, physically segregating LV from the cells from which they bud, remains a serious challenge, especially in bioreactor harvests. Nonscalable legacy methods such as batch centrifugation can lead to poor recoveries, high cost, and labor-intensive processes. Filtration was chosen as a scalable, closed alternative to centrifugation to remove producer cells, debris, and large aggregates from the LV feed. A comparison of the two methods is provided in Table 1.
Table 1. Comparison of legacy centrifugation vs modern filtration methods for clarification of LV from cells
|Simple at small scale (< ~ 5 L)||Operation similar at any scale|
|Easier with adherent cells||Works with adherent or suspension cells|
|Not truly scalable (many batches), or require cleaning, sanitization, and open manipulation||Scalable (increase filter surface area and flow rate for same process time), disposable single-use consumables, no sanitization required|
|Labor intensive||Minimal manipulation required|
|High equipment cost||Simple, low-cost equipment for basic operation|
In order to minimize the risk of contamination during processing, we designed the clarification rig to be functionally closed. This was accomplished by using single-use ReadyMate aseptic connectors and/or weldable tubing to connect the various components of the system. In a PD laboratory this allows clarification in a Biosafety Level 2 (BSL-2) environment, instead of the confined space of a biological safety cabinet (BSC), which is restrictive to the filter capsules required at larger scales. Process closure is especially important for manufacturing LV in a good manufacturing practices (GMP) facility, where large batch sizes require a cleanroom with ample space. By using a closed system for clarification, the risk of losing a batch to contamination is greatly reduced, resulting in potential savings in both manufacturing costs and time.
To eliminate cleaning validation and minimize turnaround time between batches, we used single-use consumables and processes for the entire process. Designing a clarification rig composed of single-use bags, filters, tubing, and pressure sensors eliminated the need to clean lines and systems between batches and to validate cleaning efficiency. For the clarification filter trains, we selected the ULTA series, as these capsules are designed for single-use and are available in ReadyToProcess versions, which are sterilized by gamma irradiation and provided with ReadyMate aseptic connectors and/or weldable tubing to maintain a closed system.
The final important design aspect of this clarification workflow is to ensure process scalability. It is very important to determine the final scale required for the process at an early stage, so that the chosen filter family can be applied to both your scale-down model as well as the final manufacturing skid. Although small filter discs can be useful to determine product compatibility, it is very difficult to determine capacity, and, in turn, scaling factors, since the flat shape does not mimic the structure of the larger capsules. Similarly, it is important to ensure that the filter system chosen for lab-scale/scale-down experiments is available in the larger surface areas required for manufacturing. These choices ensure the ability to scale up the process instead of scaling out to multiple filters in a manifold. The ULTA filter range allows for direct scalability. Thus, all the work presented here was performed with ULTA capsule filters, which have a direct path from lab to manufacturing.
Results and discussion
Filter preconditioning with basal media
Early experiments showed that approximately 40% of both physical (p24) and infectious (TU/mL) titer was lost at this step (see Fig 2). These results indicate that LV particles were potentially trapped in the filters or were adsorbing to the filter surface.
Based on a literature review (2, 3) and experimental evidence, we hypothesized that LV were adhering to the glass fibers, the polyethersulfone (PES) membrane of the filters, or both, leading to a decision to ‘coat’ (or condition) the filter surfaces with the same basal medium used in the bioreactor. This medium is HyClone HyCell TransFx-H, supplemented with 20 mM Tris-HCl pH 7.4. By filling the filters with this medium and allowing them to soak for at least one hour under static conditions, followed by flushing a volume of 5 L per m2 of filter surface area (typically 1 L per 0.2 m2), recovery of physical and infectious titer from the input material exceeded 90% (Fig 2).
Further optimization of the preconditioning step could lead to alternate strategies to achieve the same recovery. For example, the soaking time and/or flushing volume need to be studied in more detail to determine which factors influence the observed increase in recovery (or, alternatively, whether they can be minimized or eliminated if not significant). The preconditioning method will also depend on the cell line to be clarified. All of the following factors could affect the preconditioning regimen: cell type (suspension vs adherent); concentration; viability; amount and size of debris; production medium composition (especially presence or absence of serum and other additives); amounts of HCP and DNA (both HCD and from plasmids); and LV titer. For example, a low LV titer at harvest could theoretically result in the vast majority of viral particles adhering to the filter surfaces, which may require ‘rescuing’ by a more stringent preconditioning regimen involving a blocking agent such as human serum albumin.
Fig 2. Preconditioning of filters affects performance. Bars represent averages of three technical replicates from each run. Error bars represent the standard deviation of the three technical replicates.
Clarification of lentiviral vectors from suspension cell material
Filter train selection
Filters of various materials, pore sizes, train combinations (number of filters, pore sizes, and relative surface area ratios), and filters from several manufacturers were tested for LV compatibility (data not shown). Based on these initial data and the required specifications of scalability, system closure, and single-use consumables, we selected the ULTA filter line. We chose the ULTA GF (glass fiber) 5 µm filter capsule as a prefilter, because the average size of the harvested cells of the stable inducible LV producer cell line is approximately 14 ± 9 µm, and this filter will retain most intact single cells and aggregates. This retention is especially important when LV are produced in suspension cells.
Note: studies showed that ULTA GF 10 µm and 5 µm have equivalent performance (data not shown).
In initial studies we paired this prefilter with a Whatman Polycap TC 0.45 µm filter capsule. A third, intermediary, filter was added to the system to reduce excessive pressure on the Polycap and thus increase capacity. We studied various pore sizes and compared performance. Figure 3 shows the pressure difference in a filter train with 0.6 or 2 µm intermediary filters. The ULTA GF 10 µm filter was loaded with 6.60 × 1010 total cells/m2 in (A) and 6.72 × 1010 total cells/m2 in (B). We eventually selected the ULTA GF 0.6 µm filter because of its ability to dramatically reduce pressure on the Polycap filter (< 0.1 bar vs > 0.5 bar) and reduce clogging, allowing significantly more LV to pass through into the product.
Reducing the amount of material that reaches the microfilter is especially important as the LV particles are approximately 0.08 to 0.12 µm in size (1) and thus are very close to the size of the 0.45 µm pores. LV production is very stressful to cells and may cause excessive cell death and debris formation. Therefore, excess material that can be blocked by the 0.6 µm filter results in less pore clogging in the microfilter and greater product recovery.
Fig 3. Pressure comparison on system when using a 2 µm vs 0.6 µm intermediary filter. Two 1 L clarifications were conducted using an ULTA GF 10 µm filter, a Polycap TC 0.65 + 0.45 µm, and either an ULTA GF 2 µm (A) or 0.6 µm (B) as an intermediary filter.
The modified filter train produced excellent results but needed further refinement as the Polycap filter is not scalable. To identify a scalable replacement, we compared filter trains with ULTA GF 10 µm, ULTA GF 0.6 µm as intermediate filter, and a 0.2 µm filter ‒ either Polycap TC or ULTA CG. Infectious titer results were statistically equivalent for both filter trains (see Fig 4). Physical titer recovery, pressure traces, and subsequent ultrafiltration step recovery were also equivalent (data not shown). Based on these results, we replaced Polycap with an ULTA CG 0.2 µm nominal PES filter.
Fig 4. Equivalency test with Whatman Polycap and ULTA CG filters. Two replicate 1 liter LV harvests were clarified using filter trains consisting of an ULTA GF 10 µm, an ULTA GF 0.6 µm, and either a Polycap TC or ULTA CG 0.2 µm filter. Each dot represents one technical replicate of the gene transfer assay. For each biological replicate (replicates 1 and 2), three technical titering replicates were completed.
To obtain the results in these studies, we chose a filter surface area ratio of 2:1:1 (5 µm:0.6 µm:0.2 µm); however, this will require further study and optimization for other LV production systems. The harvest viability of HEK293 cells that produce LV can vary from approximately 60% to more than 99% based on culture conditions, additives, and harvest timing (data not shown). This variability means the ratio of intact cells and aggregates to cell debris can also vary widely. Therefore, it may be necessary to alter the surface area of filters designed to retain these different-sized particles. Similarly, cell lines and production conditions that produce more or less LV, HCP, DNA, or all three may affect the size of filters required, as will the concentration of cells present at transfection, induction, and harvest. Additives may also affect the capacity by blocking nonspecific binding (e.g., serum) or by causing premature fouling.
Infectious Titer recovery – stable inducible LV producer cell line
With the 5 (or 10) µm - 0.6 µm - 0.2 µm filter train, we completed several clarifications using a range of inputs between 0.5 and 2.1 liters of harvested bioreactor material with the stable inducible GFP LV producer cell line. For 0.5 to 1.1 L, LV recovery was 110.2% ± 10.5% (n=14), while for 1.5 to 2.1 L (where the size of each filter was doubled), LV recovery was 100.4% ± 5.8% (n=8), equaling a 106.6% ± 10.1% recovery for all runs combined (n=22) (see Fig 5). Thus, across 22 biological replicates minimal to no loss of LV was seen during clarification using this filter train. A visual comparison of the harvested material prefiltration and the clarified material post-filtration demonstrated the effectiveness of this process in removing cells and debris (see Fig 6).
Note: LV recovery appears > 100% in most replicates in these studies. This could be due to the inherent variability of the cell-based gene transfer assay, a result of removing transduction blocking compounds from the bioreactor harvest, filtering out large particles that cause LV aggregation, or a combination of these factors.
Fig 5. LV recovery after clarification. Infectious titer recovery of bioreactor harvest material post-clarification using primary and secondary filters in line with a tertiary microfilter. Infectious titer determined by gene transfer assay of HEK293T cells measuring GFP+ cells by flow cytometry. Aggregate data from 5 µm and 10 µm filters, based on internal data showing equivalent performance.
Fig 6. Pre- and post-clarification images. A sample of the bioreactor harvest before clarification (left) was imaged, showing transduced GFP+ suspension cells. A post-clarification sample (right) shows complete removal of intact cells and large cell debris.
Infectious titer recovery – transient transfection cell lines
The 5 µm - 0.6 µm - 0.2 µm filter train was also used to clarify LV from bioreactor harvest produced with a commercially available transient transfection system. The infectious titer recovery was 110.1% ± 9.0%. The same train was doubled in size to clarify two liters of material from another transient transfection cell line for a 97.0% ± 14.6% step recovery. These results show that this clarification filter train is amenable to both producer and transient transfection cell line production (see Fig 7).
Fig 7. Post-clarification recovery with transient transfection cell lines. Infectious titer recovery of bioreactor harvest material post-clarification using primary and secondary filters in line with a tertiary microfilter. ‘Producer’ refers to a stable inducible LV producer cell line. ‘Transient 1’ is a commercially available LV transient transfection production system. ‘Transient 2’ is a cell line optimized for LV production by transient transfection. Infectious titer was determined by gene transfer assay of HEK293T cells measuring GFP+ cells by flow cytometry.
Post-clarification recovery flush
A study with media flush showed that 95% of the LV in the clarification product is present in the initial flowthrough, with only an additional 5% recovered in the media flush (see Fig 8). This is an important trade-off to consider if there is a process prioritization between total LV recovery or TU/mL concentration. As the media flush could potentially dilute the product more than 20%, it could be eliminated in order to maintain the harvest titer, save time, and reduce cost of goods. However, if concentration will be done in a subsequent step, then the flush is recommended in order to collect as many LV particles as possible and therefore maximize concentration later in the process.
Fig 8. Recovery (%) of infectious titer after media flush. A 1 liter clarification was completed using a filter train consisting of an ULTA GF 5 µm, an ULTA GF 0.6 µm, and an ULTA CG 0.2 µm nominal filter. Clarified flowthrough was collected as one bulk, and then TransFx-H media was passed through the system and collected in fractions to be analyzed for infectious titer by gene transfer assay. DV is dead volume of system and consisted of the volume of all three filters and the tubing connecting them. Fractions were collected at one quarter of a dead volume (0.25 DV), a second quarter of dead volume (0.5 DV), and finally another half dead volume (1 DV).
Here we present a scalable, closed workflow for LV clarification by filtration. The three-stage filtration system allows for effective clearance of LV producing cells as well as debris and aggregates approximately 0.45 µm and larger with minimal to no loss of infectious titer (see Fig 5). The workflow was shown to be compatible with a stable producer line as well as two transient transfection lines (Fig 7). Also, a trade-off between total viral recovery and concentration was shown (Fig 8). Although this study indicates that recoveries were much higher after preconditioning filters with medium (Fig 2), future work is needed to reduce cost of goods (COGs) by minimizing the amount and/or time of filter soaking, or potentially switching to a more cost-effective presoaking material. This workflow was shown to be compatible with three different cell lines, but further process development will be required to fine-tune process parameters for ‘Transient 1’ (Fig 7), any additional cell lines used to produce LV, and for LV production with a different transgene.
All work was performed in collaboration with CCRM through funding from FedDev Ontario and Cytiva at the Centre for Advanced Therapeutic Cell Technologies (CATCT), Toronto, Ontario, Canada. The reporting and interpretation of the research findings are the responsibility of the author(s).
- Merten O-W, Hebben M, Bovolenta C. Production of lentiviral vectors. Mol Ther Methods Clin Dev. 2016;3:16017. https://www.cell.com/molecular-therapy-family/methods/fulltext/S2329-0501(16)30158-9.
- Manfredsson FP. Introduction to viral vectors and other delivery methods for gene therapy of the nervous system. Methods Mol Biol. 2016;1382:3-18. https://pubmed.ncbi.nlm.nih.gov/26611575/
- Heldt CL, Saksule A, Joshi PU, Ghafarian M. A generalized purification step for viral particles using mannitol flocculation. Biotechnol Progress. 2018;34:1027-1035. https://pubmed.ncbi.nlm.nih.gov/29717555/
- Manceur AP, Kim H, Misic V. et al. Scalable lentiviral vector production using stable HEK293SF producer cell lines. Hum Gene Ther Methods. 2017;28(6):330-339. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5734158/.
|HyClone HyCell TransFx-H transfection media||SH30939.02|
|ULTA Prime GF 5 µm, 4" step HB||KGF-A-0504GG|
|ULTA Prime GF 0.6 µm, 2" step HB||KGF-A-9602GG|
|ULTA Prime CG 0.2 µm, 2" step HB||KMP-CG9202GG|
|ULTA Prime GF 10 µm, 4" Step HB||KGF-A-1004GG|
|ULTA Prime GF 5 µm, 6" step HB||KGF-A-0506GG|
|ULTA Prime GF 0.6 µm, 4" step HB||KGF-A-9604GG|
|ULTA Prime CG 0.2 µm, 4" step HB||KMP-CG9204GG|
Related products for single-use assemblies
|ULTA Prime GF 5 µm 4" Capsule Filter, 6" of tubing||12410087|
|ULTA Prime GF 0.6 µm 2" Capsule Filter, 6" of tubing||12410065|
|ULTA Prime CG 0.2 µm 2" Capsule Filter, 6" of tubing||12410058|
|ReadyMate Disposable Aseptic Connector (DAC) 250 with 1/4" Barb||28936612, 28936691|
|Sensor Assembly, SciLog pressure sensor||12410233|
|ReadyCircuit 1 L Hanging/Pillow Bag, needleless CLAVE sample port, 3 ports, 1 ft of C-Flex® 374 0.25"||12410218|
Upstream production of lentiviral vectors
LV were produced using a stable inducible LV producer cell line (NRC; 4) in 1 to 3 L STRs using animal-derived component free (ADCF) HyClone TransFx-H media (Cytiva) without serum supplementation. Prior to harvest, the cell suspension was supplemented with Tris-HCl, MgCl2, and Benzonase™ (Millipore Sigma), and allowed to mix for a minimum of 15 min to digest host cell DNA. Bioreactor contents were harvested into a single-use bag.
C-Flex® Animal-Derived Component Free (1/4” internal diameter) (Cole-Parmer) and Masterflex™ L/S Precision Pump Tubing, BioPharm Plus Platinum-Cured Silicone size 17 tubing (Cole-Parmer) were cut to size, required connectors were inserted, and the assembly was autoclaved along with three SciLog™ SciPres™ single-use Biopharm pressure sensors (Parker) and a 0.2 µm air filter. Autoclaved C-Flex tubing was attached to single-use bags (harvest, media flush, waste, and product) in a BSC. The clarification assembly was prepared by attaching two lengths of autoclaved C-Flex tubing (with one end sealed on each prior to autoclaving) to a “Y” connector (Cole-Parmer), along with a length of autoclaved BioPharm Plus tubing, which was connected to the first pressure sensor.
The pressure sensor was connected to the hose barb inlet of the ULTA GF 5 µm capsule (Cytiva) with the outlet attached to the second pressure sensor. The second pressure sensor was connected to the inlet of the ULTA GF 0.6 µm capsule (Cytiva) with the outlet attached to the third pressure sensor. The third pressure sensor was connected to the inlet of the ULTA CG 0.2 µm nominal filter (Cytiva) while the outlet was connected to the single-use waste bag via a length of C-Flex tubing. All hose barb connections were zip-tied to ensure a tight connection. The assembled and closed rig was transferred to a lab bench, pressure sensors were connected to a SciLog single-use sensor monitor (Parker), and BioPharm Plus tubing was placed in a peristaltic pump. Figure 9 shows a diagram for the clarification rig setup.
Fig 9. Clarification rig set-up.
Tris-HCl (20 mM) pH 7.4 was supplemented to HyCell TransFx-H media, which was transferred into a single-use bag in a BSC. An autoclaved length of C-Flex tubing with one end sealed and an autoclaved length of BioPharm Plus with a 0.2 µm air filter inserted were attached to the bag. The media bag was connected to the clarification rig via sterile weld with the C-Flex tubing attached to the Y connector. The other arm of the C-Flex Y was clamped off, and the upstream vents on the three filters were all opened. The peristaltic pump was turned on at low RPM to begin flow from the media bag to the first filter. Flow was continued until the filter filled with media (the capsule was tilted to ensure the entire volume was filled), at which point the vent was closed.
This procedure was repeated for the second and third filters, leaving all tubing, pressure sensors, and filters filled with media in the absence of pockets of air. Tubing on the outlet of the third filter was clamped off to prevent any media from escaping. The system was left to rest for a minimum of one hour, after which the peristaltic pump was set to the RPM to be used in the experiment, the outlet clamp was removed, and 5 L of medium per m2 of total filter surface area (typically 1 L of medium for 0.2 m2 surface area in this study) was flushed through the system and collected in the waste bag.
The bag containing harvested cell/LV suspension was attached to the rig by sterile weld to the C-Flex in the Y connection. The clamp was removed from this line and transferred to the media bag line. A flux of 300 L/m2/h (LMH) was calculated by using the smallest surface area of the three filters (usually the ULTA CG) and was converted to mL/min or RPM. The peristaltic pump was started at this flow rate, and media was collected in the waste bag until approximately half of the system was filled with the cell/LV suspension.
Note: In this workflow, approximately half of the media left in the filters after preflush is collected in the waste bag. This is a compromise between discarding all the remaining media from the preflush, which could potentially lead to lost product, and not discarding any leftover media, leading to dilution of product. Thus, a trade-off between ease of operation (deciding when to switch from waste to product collection), time, and product recovery vs concentration is achieved here and can be altered to suit the needs and goals of each purification scheme.
Once half of the system was deemed to be filled with the cell/LV suspension, the peristaltic pump was stopped, and a clamp was attached to the outlet line of the final filter in order to prevent fluid from escaping. The waste bag was swapped out with the final product bag via a sterile weld. The clamp was removed, and the peristaltic pump was re-started at the same 300 LMH flow rate. Pressure was monitored using the SciLog monitor to ensure there were no sudden increases in pressure in any of the filters and to maintain pressure below a 0.8 bar safety limit.
Pressure readings were collected periodically during filtration to graph pressure vs time on all three filters. Flow was continued until the harvest bag emptied, at which point the pump was stopped, and the clamp from the media bag line was transferred to the harvest bag line. Flow was resumed at 300 LMH to push media through the filters in order to flush out LV. Once approximately half of the volume of the system (filters and tubing) was collected in the product bag, the media bag was inverted and the line to the attached 0.2 µm sterile air filter was opened to use air to drain the system of liquid.
Note: media flush is optional and may be skipped if LV product concentration is prioritized over total recovery. See Fig 8 for a study of volume of recovery flush vs percent recovery of infectious titer.
Once liquid was no longer collected from the outlet or pressure reached a safety cut-off of 0.8 bar, the pump was stopped, and the outlet line and product bag were clamped off. Finally, the product bag was removed from the system by using a tube sealer on the outlet line and cutting the sealed section of tubing.
Pre- and post-clarification samples were imaged using EVOS™ FL Auto (Thermo Fisher).
Infectious titer (IT) was determined by transducing HEK293T adherent cells with LV particles for 48 h and then measuring GFP transgene expression using flow cytometry (measured as % GFP+ expression).
Physical titer (PT) was measured using the HIV-1 p24 ELISA Kit (XpressBio).
Host cell protein
Host cell protein (HCP) was quantitated using the HEK 293 HCP ELISA Kit (Cygnus Technologies).
Host cell DNA
Host cell DNA (HCD) was quantitated using the Quant-iT™ PicoGreen™ dsDNA Assay Kit (Invitrogen).