Process intensification for lentiviral vector manufacturing using tangential flow depth filtration

Abstract

For gene therapies to become more accessible and affordable treatment options, process intensification is one possible strategy to increase the number of doses generated per batch of viral vector. Process intensification for lentiviral vector manufacturing can be achieved by enabling perfusion in the production bioreactor when applied in tandem with a stable producer cell line, allowing for significant expansion of cells and production of lentiviral vectors without the need for transfer plasmids. Tangential flow depth filtration was used to achieve an intensified lentiviral vector production by enabling perfusion to expand cell density and allow for continuous separation of lentiviral vectors from producer cells. Hollow-fiber depth filters made of polypropylene with 2- to 4-μm channels demonstrated high filter capacity, extended functional life, and efficient separation of lentiviral vectors from producer cells and debris when used for this intensified process. We anticipate that process intensification with tangential flow depth filtration at 200-L scale from a suspension culture can produce on the order of magnitude of 10,000 doses per batch of lentiviral vectors required for CAR T or TCR cell and gene therapy that would require approximately 2 × 10⁹ transducing units per dose.

Process intensification was achieved using tangential flow depth filtration (TFDF) and stable producer cells to produce an order of magnitude more lentiviral vectors compared with the standard batch process. TFDF membranes demonstrated high filter capacity and efficient separation of lentiviral vectors from producer cells for up to 96 h postinduction.

Graphical abstract

Introduction

The use of retroviruses for gene delivery began in the 1980s and has continued to be a useful tool in part due to the ability of retroviruses to be pseudotyped with vesicular stomatitis virus envelope glycoprotein (VSVg) to enhance stability and provide broad tropism.^1,2,3 Lentiviral vectors (LVVs), specifically, have become powerful tools for gene therapy because they can effectively integrate proviral elements into the host genome, can transduce non-dividing cells, and have low immunogenicity.^3,4,5 LVVs can be used to transduce patient T lymphocytes to drive expression of genes for T cell receptors (TCRs) or chimeric antigen receptors (CARs) to treat different cancers.

As of the beginning of 2022, approximately 3,000 clinical trials had been completed, with another 597 active, recuriting, or soon to be recruiting clinical trials using LLVs, a testament to the significant interest in LVVs in cell and gene therapy.^6,7 However, the cost of these therapies is a hurdle for many prospective patients in part due to the costs associated with LVV manufacturing resulting from modest process yields and costly supply of plasmid DNA needed for transient transfection processes.^{8,9,10,11,12,13,14} These manufacturing challenges affect not only the cost of the therapy but also the ability to produce a sufficient number of doses required to meet potential demand for prevalent diseases. For example, developing a cell therapy that is both available and affordable for a large fraction of those afflicted by lung cancer, for which there were about 131,880 deaths alone in the United States in 2021,¹⁵ would not be feasible without increasing the number of doses obtained per batch of LVVs or producing an inordinate number of batches.

Cost modeling analyses indicate that the cost of goods for LVV manufacturing can be reduced through the use of single-use stirred-tank bioreactors (STRs) as well as utilizing stable producer cell lines (SCLs), which removes the reliance on plasmid DNA in LVV manufacturing.⁹ The use of SCLs presents the opportunity to enhance production by producing more LVVs per bioreactor volume, achieved by increasing cell density through a perfusion system. The utilization of these technologies has the potential to decrease costs and reduce manufacturing footprint while supplying sufficient dosage for cell and gene therapies. It is worth mentioning that there are other strategies to increase generated doses per batch, for example, improving downstream recoveries, improving transduction efficiency, and designing improved transgene cassettes aimed at increasing gene expression.

Manufacturing LVVs can be done through a transient transfection process, typically achieved through the use of four plasmids (env, gag-pol, rev, and the gene of interest) complexed with polyethylenimine (PEI) to transfect HEK293T cells.^12,16,17 Transient processes can produce clinical-grade LVVs quickly; however, these processes can become costly if using cGMP-grade plasmid DNA in manufacturing.⁹ On the other hand, while SCLs require technical expertise and potentially longer timelines for process development, once generated they can be used for LVV production without the reliance on plasmid DNA and transfection.¹⁸ In addition, and most relevant to this study, is the benefit of scalability of SCLs. Inherently stable producer cells can be expanded in culture without loss of the vector genome and induced with a Tet-On system, thereby eliminating the need for transfer plasmids for production.¹⁸

Manufacturing of LVVs—from either transient transfection or a SCL—can be achieved using an adherent cell culture or suspension-adapted cell culture. Adherent cells require a substrate for growth to allow for adhesion and spreading. Scalable manufacturing platforms for LVVs using adherent cultures include layered vessels (Cell Factory),^19,20,21 fixed-bed bioreactors (iCELLis),^14,22,23 hollow-fiber bioreactors (Quantum),²⁴ and microcarriers such as Fibra-Cel used in conjunction with substrate-lacking bioreactors such as rocking motion reactors (WAVE) or STRs.^25,26,27 Suspension-adapted cultures are typically paired with STRs as they are not limited by the surface area of the substrates for previously listed systems.

Lesh et al. conclude that the iCellis (Pall Corp.) fixed-bed bioreactor is a favorable option for adherent manufacturing due to scalability and ease of harvest, but note that it is not possible to directly measure cell density after the initial inoculation.²⁸ Powers et al. inoculated an iCellis at 0.15 × 10⁵ cells/cm² and estimated harvest cell density to be approximately 6.8 × 10⁵ cells/cm² by indirect measurement using an Aber Futura biomass probe on the 2.65-m² bioreactor.²² Pall Corp. offers the iCellis system at up to 500 m²; this can support up to 3.4 × 10¹² cells using the harvest density measured by Powers et al. In terms of cell density, this would be approximately double that of a 500-L STR supporting a suspension cell culture when performed in a batch mode with a harvest density of 3 × 10⁶ cells/mL,¹⁸ which would yield 1.5 × 10¹² cells per batch. There exist further pros and cons to each system—LVV output in adherent culture versus suspension, purity of the bulk harvest material, ease of use, cost, and manufacturing footprint—yet clearly both systems offer the capability to support a large quantity of cells and consequently produce large amounts of LVVs; Powers et al. estimated a single iCellis run using the 300-m² low-compaction system can produce 6,490 doses of LVV for X-linked severe combined immunodeficiency (X-SCID) from a single batch, assuming production comparable to their 2.65-m² small-scale model.²²

Perfusion—the continuous exchange of fluid noted by nutrient addition and by-product removal—is another factor in considering bioreactor options for LVV manufacturing. The iCellis system is inherently a perfusion system as a consequence of the adherence of cells to the fixed-bed membrane and the ability to exchange media without disturbance of the cells. However, it is not clear if the iCellis system can support higher cell densities than those previously mentioned by increasing the rate of medium exchange after inoculation. Hollow-fiber membranes (HFs) have been used with STRs and rocking-motion bioreactors to exchange media and support higher cell densities than would normally be achieved in a batch or fed-batch bioreactor. These membranes can be operated in either a tangential flow filtration (TFF) or an alternating tangential flow filtration (ATF) mode and have been shown to successfully generate cell cultures as high as 2.14 × 10⁸ cells/mL.^29,30,31

The benefits of perfusion in mammalian cell culture can be found in several different areas of manufacturing. In biopharmaceutical manufacturing processes, perfusion has been utilized for the preparation of high-cell-density seed banks, which can be frozen and thawed when needed to inoculate a bioreactor, reducing the overall seed train time. In addition, perfusion can be utilized for an N-1 bioreactor, which reduces the reactor volume required prior to inoculation of the production (N-stage) bioreactor. Perfusion in the N-1 bioreactor can also allow for inoculation at higher cell densities, providing additional time savings in the plant. Last, perfusion can be utilized in the N-stage reactor to increase product yield.^29,31 An increase in production of monoclonal antibodies (mAbs),^30,32,33,34 adenoviral vectors,³⁵ and LVVs has been demonstrated by the utilization of perfusion systems in combination with closed-system bioreactors.³⁶

ATF technology is an established option for achieving perfusion in STRs for mammalian cell culture processes. Currently available through Repligen, the ATF consists of many tubular hollow-fiber filters encased in a cylindrical shell. Filter pore size will determine the categorization of the system as ultrafiltration (50 kDa) or microfiltration (0.2 or 0.45 μm). The filters are connected to a dip tube within the bioreactor and a diaphragm pump at the bottom of the system. The direction and rate of flow are determined by a controller dictating air pressure in the diaphragm, creating an alternating upstroke of cell culture into the bioreactor and downstroke back into the device; this alternating tangential flow facilitates self-cleaning by removing impurities collected on the surface of the filter. A permeate line is located on the outside of the encased hollow-fiber bundle. The rate of removal from the reactor is either set to a constant based on a predetermined perfusion rate, dictated by a cell-specific perfusion rate (CSPR), or set to maintain a constant weight in a bioreactor system while medium is added at a constant rate as determined by the CSPR.

A new filter technology being explored is Repligen’s Tangential Flow Depth Filtration (TFDF), which has several key differences compared with the ATF. The TFDF is a tubular polypropylene filter with a wall thickness of about 5 mm. In contrast, the ATF hollow-fiber polyethersulfone membranes are about 0.075–0.2 mm in thickness. The TFDF lumen is made from isotropic material; in combination with its thickness, this filter allows for an extended tortuous path of channels approximately 2–4 μm in diameter, which particles must traverse before clearing into the permeate. The ATF hollow fibers are an anisotropic material functioning as an absolute cutoff filter, either 50 kDa, 0.2 μm, or at the largest offering, 0.45 μm. The combination of depth filtration with TFF allows the TFDF to exhibit higher filter capacities while avoiding retention of biological products inside the filter. The TFDF system functions without alternating flow: a low-shear pump recirculates the culture through the filter in one direction by two dip tubes within the bioreactor, and the permeate is similarly removed from the filter casing by a peristaltic pump based on a constant rate of removal or by weight control.¹⁸

The material differences between the two perfusion technologies allow them to be applied in different modes and for different product types. Monoclonal antibodies such as IgG1 can be retained inside ATF hollow fibers when ultrafiltration is utilized, but if the system is switched to use microfiltration hollow fibers, the product can be collected continuously in the permeate.³² From our experience, microfiltration ATF membranes retain LVVs (data not shown), approximately 100–120 nm in diameter,³⁷ while the TFDF’s significantly larger pore structure has been shown to allow LVVs to pass freely into the permeate.^38,39

Oxford Biomedica has demonstrated that TFDF technology can be utilized for the separation of HEK293T cells from LVVs and scaled up for batch filtrations. Initial experimentation demonstrated that multiple 5-L batch harvests can be performed using a single 55-cm² TFDF filter and from a single bioreactor by resuspending the culture in fresh medium after the initial harvest. They also compared the microfiltration technique to commercially available depth filters, showing higher yields from the TFDF microfiltration harvest. Last, they successfully scaled up their process from 5- to 50-L STRs.³⁸

The benefits of the TFDF are intriguing for its potential application for a continuous harvest from a high-cell-density culture, supported with perfusion. The prior example effectively demonstrates separation of HEK293Ts from LVVs; however, the system was utilized for 1-h batch harvests, not continuous separation. It is likely that perfusion was not utilized prior to the harvest, and it is unclear what the harvest densities were, as they were undisclosed. We sought to test the usefulness of the filters for process intensification by way of facilitating both perfusion and continuous harvest filtration.

We report continuous production and harvest of LVVs from an SCL at high-cell-densities (>20 × 10⁶ cells/mL) in a perfusion system, achieved by the utilization of TFDF. In this study, filters were initially evaluated at low-cell-densities (LCDs) to verify findings reported by Oxford Biomedica. A perfusion process using ATF technology has been developed, which concentrated LVV in the bulk with producer cells and restricted the harvest to a single batch harvest performed at 48 h post-induction (hpi). Perfusion achieved using the TFDF allowed for continuous harvest and separation of LVVs from high-cell-density culture. Consequently, collection of LVVs from the bioreactor was successfully achieved up to 96 hpi, double the production time achieved using ATF technology with a single batch harvest.

Results

Our initial evaluation of TFDF technology was performed by executing a batch harvest using an SCL without perfusion. Process conditions and cell culture attributes for this experiment are listed in Table S1.

We sought to use the same exact filter for a 48- and 96-hpi harvest to demonstrate reusability. Both harvests used a C1/DF/C2 process (C1 or C2,– concentration; DF, diafiltration). Due to observed filter fouling for the 96-hpi harvest, the second concentration step was not performed. In addition, the inlet flow rate was doubled to increase shear rate and self-cleaning across the membrane to extend the life of the filter.

For the 48-hpi harvest, results can be found in Figures 1A–1E. Transmembrane pressure (TMP) was low (<1.5 psi) throughout the process, and permeate turbidity decreased during the initial concentration and diafiltration stages. Permeate turbidity increased slightly during the second concentration stage, from 21.6 to 24.7 nephelometric turbidity units (NTU). Viable cell density (VCD) increased during both concentration stages, as anticipated, and was observed to decrease slightly during the diafiltration, which is attributed to minor mismatches in pump flow rate between the fresh medium input and the permeate output during this stage. p24 (Figure 1A) and infectious titer (Figure 1B) for both reactor and permeate samples trend closely to one another, indicative of near-complete recovery of LVV through the TFDF filter for the 48-hpi harvest.

Figure 1.

Low-cell-density TFDF harvest results

A low-cell-density (LCD) experiment was conducted with (A–D) 48 h post-induction (hpi) and (F–I) 96 hpi TFDF harvests from the same reactor and using the same filter after resuspending the cells in fresh medium following the initial harvest. (A and F) mean p24 (±SEM), (B and G) mean ddPCR infectious titer (±SEM), (C and H) VCD and viability, and (D and I) turbidity are plotted versus throughput for a 3-cm² TFDF filter. The 48-hpi harvest demonstrated high filter throughput (>5,000 L/m²) without filter fouling and near full recovery of the LVV through the TFDF membrane. The 96-hpi harvest culture contained more debris as indicated by the higher turbidity and lower viability. As a result, filter fouling (increasing TMP) was observed, yet the filter still displayed a high throughput (>5,000 L/m²). (E and J) At 48 hpi LVV recovery through the membrane was high (88% by p24 and 99% by ddPCR) with minimal loss to the filter; at 96 hpi, LVV was noticeably lower (70% by p24 and 50% by ddPCR) and more LVV was unaccounted for, likely present in the partially clogged filter at the conclusion of the harvest. End C1, end of first concentration step and beginning of diafiltration; end DF, end of the diafiltration and beginning of the second concentration step.

The batch was resuspended in fresh induction culture medium following the 48-hpi harvest and left to continue LVV production for another 48 h. The 96-hpi harvest results can be found in Figures 1F–1J. The starting TMP value was greater (>3 psi) for the 96-hpi harvest compared with the 48-hpi harvest. This required a change to the inlet flow rate from 1 to 2 L per minute (LPM), which temporarily helped reduce the TMP to 2.3 psi; however, the TMP gradually increased to 5.6 psi by the end of the harvest. Unlike the 48-hpi harvest, an increase in turbidity was observed during the concentration stage, decreasing again only once the diafiltration stage began. VCD increased during the concentration and, once again, a minor decrease was observed during the diafiltration. p24 (Figure 1F) and infectious titer (Figure 1G) were observed to trend close together; however, some sample measurements indicated that a complete recovery of LVV was not achieved for the 96-hpi harvest. In particular, the mid-diafiltration sample for p24 and ddPCR infectious titer indicated lower values in the permeate relative to the bioreactor sample.

Indeed, the recovery of the 96-hpi harvest appears to have been less efficient compared with the 48-hpi harvest. In Figures 1E and 1J, based on volumes at the time of sampling, we calculated the total p24 and transducing units (TUs) in the bioreactor at the beginning and end of the experiment, the total in the collected permeate bulk, and the unaccounted amount at the end of the experiment, which is likely to have been LVV bound within the depth filter. The yield was calculated as a theoretical value that could have been collected, comparing the collected permeate bulk with the difference between the amounts in the bioreactor at the beginning and the end of the experiment. The 48-hpi harvest demonstrated a high overall yield and minimal material unaccounted for, while the 96-hpi harvest was shown to have markedly lower yields and larger unaccounted totals for p24 and TUs. After confirming high throughput and high recovery of LVVs using TFDF membranes for LCD batch harvest, we performed a perfusion-facilitated continuous harvest to target approximately 10-fold the cell density that was obtained through the batch harvest process.

Table S2 displays the VCD at inoculation, induction, and 48 hpi for three TFDF runs and one control using ATF to enable perfusion. With regard to the ease of the process setup, an optimal target inoculation density of 2 × 10⁶ cells/mL was determined by run 3. Lower target inoculation densities required longer expansion times, and higher target inoculations made it challenging to prime the TFDF recirculation loop prior to inoculation due to insufficient medium volume.

The VCD and viabilities of each run from inoculation to reactor termination can be found in Figure 2. To allow for easier comparison in Figure 2, the x axis (days) was normalized such that day 0 is approximately 3.3–4.3 × 10⁶ cells/mL and day 4 is the day of induction for all runs. Each run followed a similar trend for growth, with different rates pre- and post-induction. Double times pre-induction were comparable across all runs (Table S3). Viability also remained high (e.g., >95%) pre-induction and gradually decreased with each day post-induction, which was expected due to the presence of cytotoxic VSVg protein from the pseudotyped LVV.

Figure 2.

Viable cell densities (VCDs) and viabilities for all runs are plotted against elapsed time (days)

To allow for easier comparison, time was normalized such that the VCD for all reactors is approximately 3.3–4.3 × 10⁶ cells/mL on day 0 and induction starts on day 4. Inoculation for TFDF runs 1, 2, and 3 falls on days −3, −1, and 0, respectively. All runs used Tangential Flow Depth Filtration (TFDF) to enable perfusion except the final run, labeled as ATF (alternating tangential flow). HEK293T producer cells were successfully expanded by both ATF and TFDF filter membranes.

Turbidity measurements taken from the bioreactor, inline permeate samples, and each bulk collection bag (approximately every 24 h or ∼6.2 L) can be found in Figures 3 and S1. Cell culture turbidity increased linearly throughout the run, with an average increase of 1,274 NTU per day. The highest reactor turbidity measured was during run 2, with a measurement of 4,158 NTU at 24 hpi and ending at 7,614 (96 hpi). In Figure 3B, the bulk collection increased linearly and remained <100 NTU by 96 hpi for all runs. Permeate bulk turbidity measurements remained <1.6% relative to the reactor turbidity measurement at the end of each collected fraction. The inline permeate turbidity measurements and relative inline turbidity ratios can be found in Figure S1, which further supports the data in Figure 3.

Figure 3.

High-cell-density perfusion using TFDF: turbidity and transmembrane pressure

(A and B) Turbidities for reactor (e.g., cell culture) and collected bulk permeate samples. Turbidity in the collected permeate was <1.6% of the starting cell culture turbidity. (C) TMP remained low with a linear trend up to approximately 3.5 days post-induction. Both observations indicate that the TFDF filter is successfully separating cellular debris from the permeate without filter clogging.

TMP is plotted in Figure 3C; TMP remained low (<0.5 psi) throughout the entirety of all runs. TMP for all runs exhibited similar linear trends up to approximately 3.5 days post-induction. Run 3 was terminated due to the feed pump unexpectedly turning off, resulting in massive and rapid fouling of the filter. This can be seen from the pulse-like spike in pressure at 3.5 days. Run 1 and 2 pressure data both show features of the beginning of an exponential increase in pressure at this point (∼3.5 days post-induction), more so for run 1, possibly due to the rapid syringe sample withdrawals made, which may have caused minor TMP excursions and turbidity breakthrough. The 30-cm² filters used for 40–60 × 10⁶ cell/mL perfusion cultures at 2-L scale with low flux (∼90 liters per square meter per hour, or LMH) did not approach maximum pressures previously observed from our prior demo experiment (5.6 psi for 96 hpi harvest); therefore, further analysis is needed to determine the maximum capacity in terms of pressure and loss of LVV due to filter fouling.

Particle concentration in the permeate and bioreactor was measured using FlowCam 8000 for TFDF run 2 to quantify breakthrough of different diameter bins of particles. These particle data found in Figure 4 supplement turbidity, a measurement that provides a general assessment of cloudiness without direct information on the size of the particles in suspension. A linear trend was observed for the permeate from day 4 (induction) to day 8 (termination), indicative of minimal breakthrough. However, the reactor particle concentration increased non-linearly, likely due to the retention of larger particles by the TFDF membrane. To identify particle permeability as it relates to diameter, concentrations within 1-μm bins were compared for the permeate and reactor samples. This comparison shows that as particle size decreased, the percentage of particles that pass freely from the reactor into the permeate increased, as anticipated. Miniscule amounts between 3 and 4 μm (<1%) passed through the TFDF membrane, yet only 5.1%–6.1% of the smallest particles detectable by FlowCam (1- to 2-μm bin) passed through the membrane. This observation is another indication of minimal breakthrough; as shown in Figure 4, a significant portion of particles retained in the bioreactor are <5 μm. Additional particle size analysis was performed and can be found in Figure 5.

Figure 4.

High-cell-density perfusion using TFDF (run 2 only): particle concentrations and percent permeability

Particle concentrations for (A) permeate and (B) bioreactor samples taken for TFDF run 2. (A) Concentration in the permeate remained linear over time, indicative of adequate filter performance. (B) mean concentrations (±SD) of all sizes, <10 μm, and <5 μm particles; a large majority of particles are <5 μm, which are likely to cause filter fouling. Percentage permeability is calculated by taking the ratio of particle concentration between the permeate and the bioreactor within 1-μm bins. While there is an observed increase in percentage permeability from day 4 to 8 post-inoculation, the increase is miniscule (0.3%, 1.1%, 0.6%, and 0.21% for the four 1-μm bins, respectively), indicating the filter is performing consistently and without breakthrough of larger particulates over the life of the run.

Figure 5.

Particle histograms for permeate and bioreactor samples measured using FlowCam 8000 and taken from TFDF run 2

(A–D) Permeate particle diameter increased minimally (<0.2 μm increase) from day 4 to day 8 post-inoculation, indicating continued particle retention by the TFDF membrane. Negligible amounts of particles are observed at >4 μm, known to be the upper size cutoff of the TFDF filter, demonstrating the absence of particle breakthrough throughout the run. (E) Particle size within the bioreactor (when examining only particles <10 μm) increased over the same duration (0.56 μm).

Particle histograms display the diameter from permeate and bioreactor samples from run 2. For reactor samples, our analysis was performed on filtered data restricted to particles that were less than 10 μm, as these are more likely to contribute to filter fouling than larger particles and cells. Average diameter in the reactor increased gradually from 2.98 μm on day 4 (induction) to 3.54 μm on day 8 (termination), characteristic of an increase in cellular debris with a decrease in viability over time. While both the concentration of particles and the average diameter increased with each day, the increase in diameter in the permeate changed minimally. The largest difference in particle diameter in the permeate was determined to be between day 5 and day 8 samples for a difference of only 0.23 μm. This minimal increase in particle diameter in the permeate is yet another indication that there was favorable retention of cellular debris in the reactor and minimal breakthrough.

LVV production was quantified as seen in Figure 6 for all three TFDF runs using quantitative techniques, one each for a physical and infectious marker of LVV presence; the same concentration data are presented per million viable cells in Figure S2. The p24 concentration and ddPCR infectious titer are plotted per day following induction of the culture, starting at day 1 (24 hpi). Production of LVV was delayed from the onset of induction, and consequently the concentration of LVV gradually increased, reaching a maximum after 48–72 hpi. This is best depicted in Figure 6D, where it is observed that day 2 samples (48 hpi) all had the highest infectious titer. The peak for p24 production in Figure 6A is not directly aligned with the peak for TUs; the producer cells continue to produce LVV, yet the ratio of non-functional to functional LVVs increases with time. Infectivity (infectious titer normalized by p24 concentration) was plotted for each run per day post-induction. Infectivity did in fact decrease gradually from the onset of induction. In short, the optimal collection time for the highest titer and quality LVV was between 24 and 72 hpi, with the last final fraction at 72–96 hpi containing lower total TUs and with lower infectivity relative to the prior two 24-h fractions.

Figure 6.

High-cell-density (HCD) p24 (pg/mL), ddPCR infectious titer (TU/mL), host cell protein (HCP) (ng/mL), and double-stranded DNA (ng/mL) are plotted against time (days) for reactor, inline permeate, and bulk bag samples

(A–M) Physical and infectious titers for LVVs are present in the permeate at concentrations comparable to those of the reactor, resulting in the ability to collect LVV in the permeate gradually. The largest fractions of LVV are at 24–72 hpi (days 1–3 post-induction); infectivity (TU/p24) is also shown to decrease as time progresses. For A–L, values are mean (±SD).

Host cell impurities were quantified by measuring double-stranded deoxyribonucleic acid (dsDNA) and host cell protein (HCP). Impurity measurements in Figures 6G–6L (and per total cells in Figure S3) indicate that minimal clearance of HCPs or dsDNA occurs by the TFDF filter, as anticipated due to the large pore size of the membrane.

An attempt at determining recovery through the TFDF filters was made, as a traditional measure of inputs and outputs cannot be made for a dynamic continuous process. The bioreactor samples were used to determine a theoretical maximum total p24 and total TUs that can be collected in the permeate bulk fractions. The area under the curve in Figures 7A and 7B represents this maximum, as the total p24 and TUs collected are a function of the changing concentrations within the bioreactor and the volume of permeate collected. Empirical values for the collected permeate fractions were determined from the bulk p24 concentrations and infectious titers (Figures 6C and 6F) and measured volumes of the bags. The total permeate p24 and TUs collected are the sums from individual bags from the 24- to 96-hpi fractions. Similarly, the theoretical maximum excludes the area under the curve from the first 24-hpi fraction. The first 24 hpi was excluded because in practice this fraction contains an order of magnitude less LVV than the following fraction (24–48 hpi).

Figure 7.

High-cell-density perfusion using TFDF: estimated lentiviral vector production and yields

(A) Reactor mean p24 concentration (±SD) and (B) mean ddPCR infectious titer (±SD) versus permeate volume, the area under the curve representing the maximum available LVV for collection. (C) Total p24 and (D) transducing units plotted for each run and split into each collected fraction (bulk bags 2, 3, and 4 are the permeate collection taken approximately every 24 h, excluding the first 24 hpi), the sum of all the fractions, and the area under the area (theoretical maximum) under the curve from (A) and (B). For all runs, the estimated recovery is ≥100%.

The empirical totals in the bulk permeate from 24 to 96 hpi were greater than the calculated theoretical maximum for all runs for both p24 and TUs. While the calculated yield is greater than 100%, the true value is likely to be lower; of course, greater than 100% recovery is not feasible. This is likely due to the lack of sampling between 24 and 48 hpi (or ∼6 and ∼12 L in Figures 7A and 7B); the slope of our concentration-volume line in this range may be greater than estimated as a consequence of cellular processes beginning to ramp up LVV production. Additional sampling per unit time during the initial 48 hpi would help provide a more accurate representation of yield throughout each run. Yield is not expected to be dramatically lower than estimated values. For example, day 4 average p24 and ddPCR yields were 95% and 83%, respectively. Taking day 4 as a worst case scenario due to increased TMP and buildup of debris, these recoveries are rather reasonable.

Finally, the total production of the two perfusion formats was compared with a standard batch process. By utilizing perfusion, a nearly 10-fold increase in cell density was obtained for perfusion formats over the traditional batch. It comes as no surprise that the ATF-enabled perfusion process produced 9.7 and 13.3 times the p24 and TUs, respectively, relative to batch, as these processes follow comparable 48-h harvest times post-induction. The TFDF continuous harvest format collected permeate up to 96 h, and consequently we observed that 22.4 and 25.4 times the p24 and TUs were collected relative to the batch format. Comparing TFDF to ATF, 2.3 and 1.9 times the p24 and TUs were collected, which demonstrates again that output can be increased by extending the window of production of the producer cells.

Discussion

Initial evaluation of the TFDF membrane was performed as a proof of concept to answer several questions using a single 2-L bioreactor. The primary focus was to understand and quantify the yield of LVV through the filter, determine membrane capacity at a typical process flux, and assess the possibility of performing multiple harvests from a single bioreactor. To accomplish this, we used an undersized filter (3 cm²) to reach a process endpoint based on a maximum load.

Previous studies reported favorable LVV yields from a two-batch harvest using 55-cm² TFDF membranes and 5-L bioreactors with undisclosed cell density and LVV titer. These conditions were initially selected as they represented optimal process parameters, resulting in a harvest time just under 1 h and 909 L/m² throughput for each harvest.³⁸ In our demonstration we report high filter capacities as a result of sizing a 3-cm² filter with a 2-L bioreactor. The combination of the two batch harvests totals just over 11,000 L/m², approximately 6,170 L/m² for the 48-hpi harvest and 5,200 L/m² for the 96-hpi harvest. Although TMP approached 5 psi for the 96-hpi harvest, indicative of a gradually fouling membrane, the second harvest was successfully complete, which otherwise would not be possible with normal flow depth filtration. In practice, these high pressures can be avoided by sizing a larger filter area per unit volume; however, understanding the filter capacity is critical when scaling up to manufacturing to limit cost and equipment size in the manufacturing plant.

The assessment of LVV yield was also encouraging. For the 48-hpi harvest, nearly full LVV recovery was determined, an improvement over commercially available depth filters, which typically result in clarification yields between 75% and 90%.^14,38 While some loss to the filter membrane was found to have occurred for the 96-hpi harvest, this is not an entirely surprising result given that the membrane was pushed to its maximum filter capacity. We would expect improved yields for a second harvest, again, if one increased the filter area per harvest volume. In addition, we decided to disconnect the filter and drain its contents into the reactor after the 48-hpi harvest and reprime the filter just before initiating the 96-hpi harvest. After observing the ability of the filter to operate for extended times in the latter perfusion experiments, we would have likely decided against draining and repriming the membrane if repeating a multi-batch harvest. Alternatively, we would have left the membrane to recirculate from 48 to 96 hpi without the permeate pump running and then simply turned on the pump when ready for the second harvest. We believe this may be a better approach due to the benefit of the self-cleaning effects of tangential flow on the membrane; in addition, it is physically less work. Additional experiments would be required to determine the benefit for scaling up a process with multiple batch harvests.

For LCD cell cultures (e.g., non-perfusion batch process) the TFDF appears to be an effective membrane for separation of LVVs from host producer cells. While we purposefully overloaded a 3-cm² filter, a more practical batch process would likely operate with a safety factor of 2–2.5 and attempt to keep total processing time around 1–2 h. If developing a process within these limits, one could use 50-cm² membranes for up to 10 L of culture, 450-cm² for 100 L, and 2,100-cm² for 500 L, filter areas that are, respectively, paired with the lab-, pilot-, and production-scale KrosFlo offerings by Repligen. Theoretically, it is possible to perform a two-harvest process at 2,000 L using their largest 6,000-cm² filter, but this would push processing time to approximately 3 h and utilize a low safety factor of about 1.3, based on our results. Nevertheless, detailed evaluation should be done on a case-by-case basis for each producer cell line to determine an acceptable process tolerance.

After the initial demo, we sought to evaluate the TFDF for the purpose of enabling perfusion in conjunction with our SCL to achieve an intensified version of the traditional batch format. The observed capacities suggest that the filters may also work for this application and support higher-cell-density cultures. Past experiments have been performed by utilizing 0.2- and 0.45-μm ATF hollow fibers to enable perfusion; however, a negligible amount (<1% by p24 and ddPCR) of the LVVs was quantified in the permeate. With ATF-enabled perfusion, LVVs must be separated in a batch format from the producer cells after a period post-induction; therefore, the ATF’s purpose is to facilitate cell expansion to significantly higher densities than would otherwise be possible without perfusion. By utilizing TFDF, we anticipated being able to continuously isolate LVVs from producer cells in addition to the ability to expand the culture to higher densities prior to induction.

Expansion of the culture to higher densities by perfusion was in fact possible with the TFDF and did not appear to be of any difference compared with ATF hollow fibers. Prior to induction, cultures reached an average of 37.1 × 10⁶ cells/mL with viabilities ≥95%; in our case higher densities could likely have been achieved but would have required more growth medium per day. Our process was limited to approximately 3 vessel volumes per day (VVD) to avoid impractical medium requirements for a potential scaled-up version of this process. As a result, further expansion would have risked deficiencies in essential nutrients, potentially affecting LVV production from producer cells.

We set out to characterize the performance of the filter by quantification of permeate and reactor turbidities, TMP, and particle size and concentration. All techniques have indicated that TFDF membranes provide an efficient separation of LVV from producer cells and cell debris. Turbidity in the permeate was consistently less than 2% of the measured reactor turbidity, apart from run 1 inline permeate samples, as seen in Figure S1. For run 1 measurements, we strongly suspect the act of sampling by rapid withdrawal using syringes on the provided one-way valve sample ports—located on the permeate side of the TFDF filter casing—contributed to an artificially higher flux than the set point (90 LMH). Inline permeate turbidities for run 1 at 2, 3, and 4 days post-induction were all higher than any of the bulk collection turbidities. If the inline turbidities were in fact representative, we would have expected larger turbidities in the bulk collection. For this reason, we would advise against using syringe ports for inline sampling and to use a Y connector to collect into a sterile sample bottle as done for our second and third TFDF runs. Overall, turbidity reduction was consistent, and no breakthrough was observed up to 4 days post-induction, when defining turbidity breakthrough as an increase in relative permeate turbidity to reactor turbidity.

The low TMP in Figure 3C and linear increase of this measurement additionally support our observation that minimal particle breakthrough was occurring. For a breakthrough event, we may expect TMP to transition into a non-linear phase, indicative of significant blockages within the filter, which can often result in a higher fraction of particles breaking through. Conversely, a rapid decrease in TMP could indicate a breakthrough by a compromised filter, no longer able to retain large particulates. Depth filtration provides a unique benefit that typically provides higher capacities compared with sheet filters by allowing particles to become captured within its channels without completely occluding liquid flow through that same channel. The TFDF, as noted by its fitting name, combines this benefit with that of TFF, helping shear the membrane surface to minimize deposition of particulates within the filter. As a result, the gradual linear increase in TMP, not exceeding 0.5 psi for all runs up to 4 days post-induction (>8,000 L/m²), is a positive indication of successful separation of LVVs from cell debris.

Our last method to characterize the filter was to quantify particle concentration and size in the reactor and permeate directly using flow imaging, which has a purported imaging range of 2–50 μm (although some particles are detected at <2 μm). We found that even the smallest particles detectable by FlowCam are effectively blocked by the TFDF (Figures 4 and 5), and there was no dramatic increase in the observed size of particles in the permeate up to 4 days post-induction. This observation again supports our conclusion of efficient separation of LVVs from cell debris using TFDF membranes.

Thus far in our discussion, we have highlighted data supporting that cell and cell debris remain on the retentate side of the membrane. To conclude efficient separation of LVVs from producer cells and debris, we quantified p24 concentration and infectious titer by ddPCR in Figures 6 and 7 to demonstrate that high LVV recoveries were achieved. First, in Figure 6, it is clear that comparable assay markers are measured in the permeate relative to the reactor; our yield calculations in Figure 7 quantify these results and conclude that nearly full recovery of LVVs was obtained in our collection. Thus, we have demonstrated that TFDF membranes do perform well for the separation of LVVs from producer cells and a vast majority of the debris present in the reactor.

One observed drawback of the process is the apparent decrease in infectivity of LVVs as time progresses post-induction; however, this issue is independent of the TFDF filter and can be attributed to the SCL. LVVs pseudotyped with VSVg are known to be cytotoxic to producer cells.⁵ It is shown in Figure 6 that a noticeable decrease in both infectious titer and infectivity occurs. The former decreases after an inflection in LVV production, approximately 48 hpi, while the latter decreases from the onset of induction. For our producer cell line, on average 42.5%, 37.4%, and 17.2% of total TUs was collected from days 1 to 2, 2 to 3, and 3 to 4 post-induction, respectively. If scaling up a process with these characteristics, a cost-benefit analysis should be performed to determine the harvest window and whether a collection past day 3 is advantageous. In addition, further characterization would be needed to assess if the purified LVV product from collections long after induction has quality attributes sufficient for use in patients relative to collections directly after induction with higher infectivity. Fractions with lower infectivity may necessitate higher multiplicity of infection (MOI), necessitating a higher dose for a cell and gene therapy. We would like to mention that we suspect further optimization to the upstream process, through enhancement of production medium, vector construct design and pseudotyping, and clone selection may result in an improved producer cell line with enhanced ability to produce LVVs with more consistent infectivity for extended periods of time post-induction.

Our comparison of intensified perfusion processes, enabled by ATF or TFDF, with a standard batch process demonstrates the significant production gains that are feasible with currently available filtration technologies. As seen in Table 1, ATF or TFDF processes both allow for expansion of producer cells to an order of magnitude higher density compared with non-perfusion batch production, and practically equivalent increases in LVV output can be obtained. Having an SCL allows for these intensified processes to be feasible and ultimately cost effective; since transfer plasmids are not required, one needs only to induce the culture with a switch from growth medium to production medium to turn on gene expression from the SCL, leading to assembly and budding of LVVs from producer cells. TFDF provides an additional benefit of continuous separation of LVVs from the producer cells, which resulted in nearly 2-fold more TUs per batch compared with an ATF intensified process. This production gain from TFDF intensification is a benefit due to the fact that the culture need not be terminated to isolate LVVs, which was required for ATF intensification.

Table 1.

Comparison of 2-L perfusion TFDF and ATF processes with batch using equivalent producer cell line

Process (n = 3)	Induction VCD (×106 cells/mL)	Total p24 (×108 pg)	Collection volume (L)	Total TUs (×1011)	Total p24/batch p24	Total TUs/batch TUs	Total p24/ATF p24	Total TUs/ATF p24
Batcha	4.41	7.74	2.0	4.82	–	–	–	–
ATF	45.5	78.5	1.9	64.0	10.2	13.3	–	–
TFDF (day 1)	–	2.69	5.9	3.87	0.4	0.8	–	–
TFDF (day 2)	–	49.5	6.2	51.9	6.4	10.8	–	–
TFDF (day 3)	–	71.8	6.2	45.6	9.3	9.5	–	–
TFDF (day 4)b	–	49.6	6.4	21.0	6.4	4.4	–	–
TFDF total	37.1	174	24.6	122	22.4	25.4	2.2	1.9

Intensified (ATF and TFDF) reactors produce an order of magnitude more LVV compared with a batch process. TFDF intensified reactors produced approximately 2-fold LVV relative to ATF. All TU and p24 totals are permeate for TFDF and centrifuged for ATF.

^aValue for non-perfusion batch production is average of four runs. Average p24 concentration = 4.0 × 10⁵ pg/mL. Average infectious titer = 2.5 × 10⁸ TU/mL for traditional non-perfusion batch format.

^bDay 4 for run 3 was not completed due to pump failure, average p24 and TUs include only runs 1 and 2.

Further benefits and drawbacks should be highlighted when comparing TFDF and ATF intensified processes. While we have highlighted that production can be enhanced by extending the usefulness of the cells post-induction using the TFDF format, an additional benefit is that no secondary separation step is required: the TFDF serves as the tool for facilitating perfusion while simultaneously serving as the first isolation step. For the ATF format, since LVVs remain in the retentate with the producer cells, a secondary unit operation would be required to perform a comparable isolation. For example, a centrifugation or depth filtration step could be used for this separation; however, it is a significant cost at scale and takes up additional footprint in a manufacturing setting. An ATF and centrifuge pairing could feasibly allow for multiple harvests from a perfusion intensified reactor; however, additional development would be required to determine how the ATF can remain primed and functional for more than one harvest and how to return the cell pellet from the centrifuge back into the bioreactor with medium resuspension.

While the above benefits hold true for TFDF, it is not without its drawbacks. The major disadvantage is the volumetric output in which the LVV is contained for the TFDF format. We produced approximately 3 VVD of permeate, which equates to either 6 or 9 total vessel volumes if collecting 2 or 3 days’ worth of permeate, respectively. Although the ATF format would require the same quantity of medium as the TFDF format, at the point of harvest the supernatant containing LVVs is just a single vessel volume. Therefore, their subsequent downstream process can look nearly equivalent to a non-perfusion batch process, while the TFDF format would require some form of volume reduction or collection of the permeate in a much larger holding tank relative to the volume of the bioreactor. The former would be preferred, as one who would even consider a TFDF for continuous manufacturing would likely be interested in the footprint reduction typically achieved when switching from batch to continuous manufacturing.

To further elaborate on that point, a true continuous process could be designed around the TFDF to further reduce the manufacturing footprint and isolate and purify LVVs from the bioreactor faster compared with batch processing. This offers potential to improve product quality and yield since LVVs are known to be temperature sensitive and thus cannot be held for long periods of time. Single-pass TFF (SPTFF) is one option that can be used to reduce the volume of the TFDF permeate to a practical level for subsequent purification steps. Another option could be multi-column chromatography (MCC); in this case, the TFDF permeate could be passed through a prefilter and loaded continuously onto columns that are alternating between load, elution, and regeneration stages. Both options require further evaluation for LVV manufacturing; however, they are of current interest for biopharmaceutical manufacturing to increase productivity and reduce footprint and cost; thus, the promise of these technologies extends to cell and gene therapy and the manufacture of LVVs.^40,41,42

The development of a perfusion intensified process for LVV manufacturing introduces the possibility of providing sufficient LVV doses for large patient populations. If we make several assumptions, we can begin to estimate the number of doses that can be produced at a commercial scale. These initial assumptions include a modest 20% downstream yield, that total TUs by ddPCR are approximately 5-fold the function titer, and that a 2-L scale process can be effectively scaled up to 200 L. If we further assume that a required dose per patient is 2 × 10⁹ TUs for hypothetical TCR and CAR-T therapies and 2 × 10¹⁰ TUs for hypothetical hematopoietic stem cell (HSC) gene therapies, totals previously estimated by Comisel et al.,⁹ then 200-L batch, ATF-enabled perfusion, and TFDF-enabled perfusion processes would yield 900, 12,000 and 24,000 doses, respectively, for TCR and CAR-T therapies and an order of magnitude less for HSC gene therapies. While each individual therapy will depend on an empirically determined MOI and each process will vary in production output, this example stands to demonstrate the possibility of treating large patient populations in the near future with cell and gene therapies using LVVs.

Materials and methods

Non-perfusion batch bioreactor runs

Two-liter single-use UniVessels (Sartorius, Göttingen Germany) were used in combination with Sartorius BioStat B benchtop controllers for monitoring dissolved oxygen, pH, and weight. Integrated PID controllers controlled dissolved oxygen (DO) to 50% using either filtered air or pure oxygen or a mixture of the two. pH was controlled to 7.1 ± 0.1 via filtered CO₂ sparging through the tube sparger or addition of sterile filtered 1.5 M sodium carbonate. Bioreactor temperature was maintained at 37°C and mixing was maintained at 225 RPM. The stable cell line in this study used the same technology and selection process as described by Chen et al.¹⁸; however, a different therapeutic transgene was utilized. A generic HEK293 cell culture medium, supplemented to 4 mM GlutaMAX (Gibco Thermo Fisher, Waltham, MA) and 0.1% Pluronic (Gibco Thermo Fisher) was used as growth medium. Induction was achieved by a bolus of sodium butyrate and doxycycline to achieve 5 mM and 2 μg/mL in the bioreactor, respectively.

For the proof-of-concept run completed and described in Figure 1, ProConnex TFDF Flow Paths (Repligen, Waltham, MA) were connected to bioreactors using AseptiQuik sterile connectors (Colder Products, Roseville, MN); pressure, permeate flow rate, and inlet flow rate were monitored and controlled using a KrosFlo TFDF Lab System (Repligen). A 3-cm² TFDF membrane was used and connected to the reactor 5 days after inoculation at 0.3–0.4 × 10⁶ cells/mL. Induction occurred on day 3 to allow for 48 h of LVV production. For the batch harvest, permeate flux was set to 1,300 LMH or 6.5 mL/min. LVV was separated from the culture by performing an initial concentration, diafiltration, and second concentration (C1/DF/C2) process. Repligen’s product webpage provides a visual of the setup and operation when used with an STR.⁴³ For the concentration steps, no additional culture medium was added. For the diafiltration, induction culture medium (growth medium supplemented with sodium butyrate and doxycycline) was added to replenish the reactor at the same rate as the permeate rate of removal. Two harvests were performed at 48 and 96 hpi using the same TFDF filter from the same reactor. Cell culture induction medium was added to the reactor to top up to 2 L after the 48-hpi harvest was completed.

TFDF-enabled perfusion bioreactor runs

The same bioreactor equipment, producer cell line, and setup procedures were utilized for the TFDF perfusion experiments as previously described in the LCD batch harvest experiment. For perfusion only, the weight of the reactor was maintained at 2.0 kg by addition of cell culture growth medium (growth medium used prior to induction). The key difference between LCD non-perfusion TFDF clarification and HCD perfusion is that for perfusion, TFDF is run in a continuous medium replenishment mode (i.e., diafiltration) exclusively at significantly lower permeate flux: 90 LMH versus 1,300 LMH.

A 30-cm² membrane was utilized for all perfusion runs, and inlet flow rate was kept at a constant 1 L per minute (unless otherwise mentioned) using the KrosFlo’s integrated flow sensor. Flexible polyethylene foam pipe insulation (McMaster-Carr, Elmhurst, IL) was used to insulate ProConnex tubing. A CSPR of 0.18 nL/cell/day was used based on prior experimentation. Permeate flow rate was updated daily based on VCD and the CSPR. Permeate flow began when the VCD hit 2 × 10⁶ cells/mL (day 2 for run 1 and day 0 for runs 2 and 3) and was capped at 4.4 mL/min to avoid exceeding ∼3 VVD of permeate collection. Permeate bulk was collected continuously into Flexboy (Sartorius) bags located in refrigerators maintained at 6 ± 2°C.

For runs 1 and 2, induction was performed using a bolus of doxycycline and sodium butyrate to obtain 2 μg/mL and 5 mM, respectively, in the bioreactor. In addition, growth medium was switched to induction medium to maintain these concentrations. For run 3, the bolus was not performed, and the reactor was induced by simply switching the medium type from growth to induction.

All TFDF reactors were run in a continuous collection mode in which the permeate was saved for the following 4 days post-induction, excluding run 3, which was run for only 3 days due to a pump failure. In this article we refer to the permeate bulk bag 1 as 0–24 hpi, bag 2 as 24–48 hpi, etc.

ATF-enabled perfusion bioreactor runs

ATF-enabled perfusion reactors were run with the same bioreactor equipment and procedures as mentioned for TFDF-enabled perfusion reactors and used an XCELL ATF 2 (Repligen) with 0.20 μm polyethersulfone (PES) hollow fibers for permeate removal. As mentioned, these batches were harvested at 48 hpi via centrifugation.

Offline monitoring

Cell density and viability was monitored using Vi-CELL (Beckman Coulter, Brea, CA). Offline measurements of dissolved gases and key metabolites were performed using BioProfile FLEX2 (Nova Biomedical, Waltham, MA). Turbidity measurements were taken using a 2100Q portable turbidity meter (Hach, Loveland, CO).

Sampling

For LCD non-perfusion runs, samples were taken from the permeate, bioreactor, and final bulk at the end of processing. For perfusion runs, samples were taken from the permeate, bioreactor, and bulk permeate collection every 24 hpi. Bioreactor samples were removed via syringe withdrawal from the UniVessel dip tube. Cells were removed via centrifugation at 800g for 5 min and supernatant was filtered through 0.45-μm Minisart PES syringe filters (Sartorius) prior to freezing at −80°C. Inline permeate samples for run 1 were taken via a one-way T-junction valve positioned on the permeate line; for runs 2 and 3 this was changed by inserting a Y connector on the permeate line; samples were then removed by diverting flow from the collection bag into a 50-mL sterile bottle (SaniSure, Camarillo, CA) to avoid pressure fluctuations caused by rapid withdrawal of the syringe. Bulk permeate samples were taken directly from the bags after approximately 24 h of collection of permeate into a sterile 10-L Flexboy (Sartorius). Bulk and inline permeate samples were both filtered with 0.45-μm Minisart PES syringe filters for consistency with bioreactor samples. Samples pulled for particle size analysis using FlowCam were not filtered with additional 0.45-μm membrane, and bioreactor samples were not centrifuged or filtered.

Oxygen sparging and cascade

For TFDF runs 1 and 2, 50% DO was maintained by using filtered air, oxygen, or a combination of the two through the tube sparger fitted on the UniVessel. For perfusion runs, it was clear that oxygen transfer was limited by surface area of gas bubbles; however, a method for reducing bubble size was not implemented until the third run due to sourcing delays. For run 3, a 2-μm microsparger (Mott, Farmington, CT) was fitted on the reactor in addition to the tube sparger; as a result, pure oxygen was added through the microsparger and all other gases were added through the tube sparger.

LVV titer and host cell impurity quantification

p24 capsid protein was quantified using an ELLA Simple Plex immunoassay kit (Protein Simple Bio-Techne, Minneapolis, MN) after serial 4-fold dilution into DPBS (Gibco Thermo Fisher) with 0.5% Triton X-100 and 1-h incubation at 37°C.

Viral titer was determined by multi-step dilution of LVV and transduction of supT1 cells. Genomic DNA was extracted using KingFisher (Thermo Fisher) with a magnetic bead-based extraction. Droplets were prepared with primers and probes for HIV and human ribosomal protein lateral stalk subunit P0 (RPLP0) and ddPCR Supermix (Bio-Rad) and generated on an automated droplet generator (Bio-Rad). PCR was performed using a C1000 Touch thermal cycler (Bio-Rad) prior to reading the droplets. Concentrations of HIV and RPLP0 copies per 20 μL were determined by ddPCR using a QX200 droplet reader (Bio-Rad) and analysis was done using QuantaSoft Analysis Pro Software. Copy number variant for HIV/RPLP0 was calculated by the software, and viral titer was calculated by multiplication with transduced cells (5 × 10⁴) and vector dilution factor (50–400) and divided by transduction volume (0.1 mL).

Host cell impurities were determined using commercially available kits. Quant-iT PicoGreen (Thermo Fisher) was used for dsDNA quantification, HEK293 ELISA (Cygnus Technologies, Southport, NC) was used for HCP quantification, and plate measurements were made using a SpectraMax i3x Multi-Mode microplate reader (Molecular Devices, Downingtown, PA).

Particle size

Measurements were taken using a FlowCam 8000 (Yokogowa Fluid Imaging Technologies, Scarborough, ME) on unfiltered permeate and reactor samples. Permeate samples on day 4 were undiluted; on all other days (5–8) samples were diluted 1:10 in H₂O. Reactor samples were diluted 1:10 on day 1 and 1:100 on days 4–8. VisualSpreadsheet 4.19.3 software was used for processing images. Histograms were generated using GraphPad Prism 9 (GraphPad Prism Software, San Diego, CA).

Data availability

GSK is unable to provide materials, additional datasets, or protocols.