Véronique Chotteau

The popularity of rAAV vectors in gene therapy are placing a burden on current production systems. To improve the accessibility of these life changing treatments, increases in production yields and a reduction in the cost-of-goods are needed. Transient transfection is the most common way to introduce rAAV-encoding plasmids to producer cells but it suffers from significant drawbacks such as low and inconsistent yields as well as high cost due to its need for plasmid DNA. This study aims to address the low yield of transient transfection-based rAAV production through advanced methods in process characterization. Adherent and suspension cultures of a HEK293T cell line were triple-transfected for rAAV9 production using polyethylenimine (PEI). Samples were taken at various times post-transfection for analysis with bulk and single-cell transcriptomics. It was revealed that 46% of the cells lacked transcripts of genes from at least one plasmid, indicating that a significant proportion of the cells did not have the genes necessary for rAAV9 production. Among the remaining 54% of the cells expressing genes from all three plasmids, only 8% showed high plasmid gene expression. Flow cytometric analysis of intracellular rAAV9 confirmed these results by showing that only ~ 3% of cells contained assembled rAAV9 capsids. Titre analysis by qPCR of the supernatant and lysate of the producer cells indicated an average culture performance of 10 ¹³ vg/L. Analysis of the single-cell transcriptomic data showed that a significant proportion of cells that had high plasmid gene expression were in the S-phase. Trajectory inference highlighted that genes involved in the G2-M phase transition, immune response, and protein unfolding were differentially expressed at the branch point between high and low plasmid expression. This study reveals a significant bottleneck in the transient transfection-based production of rAAV. With less than 5% of cells producing rAAV, significant improvements in titres can be achieved if this fraction can be increased. Moreover, regulation of the cell-cycle, inhibition of the immune response, and alleviating protein misfolding all potentially offer the key to enabling these life changing treatments to reach a wider audience.

Natural killer (NK) cell therapies hold great promise for cancer treatment; however, donor-to-donor heterogeneity in the ex vivo expansion process remains a critical bottleneck in their supply. This study aimed to identify factors influencing donor variability in a two-week long ex vivo NK cell expansion from PBMCs, analyzed across three donors. Single-cell transcriptomics was applied to investigate the distribution of cell types and phenotypes, as well as trajectory inference and differential gene expression. Our results identified that several factors were associated with the variability in the final NK cell fraction and expansion, and that their influence was prevalent between culture day 3 and 8. Compared to high final NK cell fraction, a culture with low final NK cell fraction exhibited an upregulation of some stress and inflammatory genes and an increase in one specific subcluster of the NK cells already on culture day 3. It showed a low score of CD56 ^Bright CD16 ⁻ phenotype and high score of CD56 ^Dim CD16 ⁺ phenotype. It had also an increased presence of cytotoxic CD8 ⁺ Tm cells. Among the observed subclusters of CD8 ⁺ Tm cells, it exhibited a higher presence of a subcluster associated with a less differentiated and less cytotoxic phenotype as well as a lower prevalence of a subcluster associated with chemokine and cytotoxic genes. Finally, it had a major expansion of one of the CD8 ⁺ Tm cells subclusters annotated as NK-like T cell and characterized by a high CCR5 mRNA expression while the levels of CCL3, CCL4, and CCL5 mRNA were downregulated. The present findings point towards a potential link between CCL signaling and improved NK cell expansion performance, including possible markers for further investigations, and suggest future strategies to increase the final NK cell fraction and expansion based on donor-specific markers.

We present a novel workow for feed-medium optimization in CHO-K1 perfusion cultures, based on mechanistic macroscopic modeling using Elementary Flux Modes (EFMs). Our approach aims to maximize steady-state monoclonal antibody (mAb) production by integrating transcriptomics data and cell-specic rate measurements into the EFM selection process. This integration enhances the biological relevance of the model and addresses the inherent undetermination of metabolic networks. The resulting transcriptomics-guided model signicantly outperforms baseline models in predicting measured cell-specic rates. Building on these kinetic models, we formulate a model-based feed optimization problem that incorporates constraints on osmolarity and steady-state operation. The non-convexity of this problem is handled by a multistart strategy informed by experimental feed compositions. Additional criteria ensure that optimized feeds remain close to the training data and are robust to process perturbations. Experimental validation conrms that the optimized feed increases harvested mAb concentration by 9.9% compared to the best design-of-experiment (DoE) optimized feed. These ndings highlight the potential of transcriptomics-informed mechanistic modeling and optimization as a powerful alternative to traditional data-driven methods in bioprocess development. This work introduces the rst mechanistic model-based feed-medium optimization framework utilizing a genome-scale metabolic network for CHO

Therapeutic proteins from mammalian cells, such as monoclonal antibodies produced by Chinese hamster ovary (CHO) cells, play a critical role in modern medicine. However, the inherent complexity of cellular metabolism poses signicant challenges for accurate kinetic modeling. In this work, we present a novel modeling framework that integrates data-driven Elementary Flux Mode (EFM) selection with macroscopic rate estimation based on Monod mechanistic laws. Leveraging a column generation algorithm, our approach indirectly ts macroscopic rate data, enabling parallel estimation and the development of complex Monod models that capture double-component eects of multiple extracellular metabolites. To prevent overtting, we incorporate an ℓ 1 -norm regularization with cross-validation and subsequently rene parameter estimates using Bayesian inference under qualitative constraints. Applied to CHO-K1 perfusion data using a reduced genome-scale metabolic network, our model robustly represents cell-specic kinetics and oers valuable insights into intracellular metabolic dynamics.

Recombinant adeno‐associated virus (rAAV) vectors are widely used in gene therapies, but the rapidly increasing global demand has created a significant challenge for rAAV manufacturing, where production capacity remains a critical bottleneck. To address this, strategies to enhance production yields are urgently needed. This study presents an innovative approach to rAAV production using high cell density (HCD) stirred tank perfusion culture. rAAV1 and rAAV9 vectors carrying GFP cargo were used as models, with triple‐plasmid transfection performed in suspension HEK293 cells at a high viable cell density of 50 million cells/mL in culture then maintained at ≥ 30 million cells/mL throughout production. Transfection and production parameters were first optimized in a 5 mL pseudo‐perfusion spin tube screening system at HCD. A proof‐of‐concept was then demonstrated by scaling up to a 200 mL stirred tank bioreactor (STR) in perfusion mode. This intensified process achieved rAAV9 production levels per cell comparable to those observed in reference shake flask cultures at 1 million cells/mL. By implementing transfection at very HCD in a perfusion‐based STR, this approach has the potential to significantly enhance rAAV volumetric production capacity, providing a promising solution to meet the growing demand for gene therapies.

The possibility to produce recombinant adeno‐associated virus (rAAV) by adherent HEK293T cells was studied in a stirred tank bioreactor (STR) culture of cell aggregates. A proof‐of‐concept of rAAV production was successfully demonstrated in a process where single cells were first expanded, then cell aggregates were formed by dilution into a different medium 1 day before triple plasmid transfection was conducted. An alternative approach for the STR inoculation using a seed taken from a high cell density perfusion (HCDP) culture was also investigated. It was, however, found that the spent medium of the HCDP inhibited the transfection of HEK293T cell aggregates, which was confirmed when testing with single‐cell suspension culture. The formation of aggregates in shaken multi‐well plates was also investigated to develop a screening system using the average power input as a scale‐down criterion, which revealed that cell aggregates could be generated in 12‐well plates, however with a larger size than in a STR. Taking into account the reported higher rAAV production of adherent cells in comparison with single cells for triple‐plasmid transfection, HEK293T cell aggregates can possibly surpass single‐cell suspension in space–time rAAV yield. The formation of HEK293T cell aggregates in a STR system offers a promising approach for scaling up and intensifying rAAV production by triple‐plasmid transfection, in comparison with traditional 2D scale‐up methods.

This study models two alternative downstream processes based on magnetic separation with the objective of understanding the economic feasibility of these processes compared to the traditional mAb process. The key focus lies in the economic understanding of the cell harvest and capture steps in the models. Here, the models revealed that integrating cell removal and product capture in a single operation is the main factor driving the unified productivity between USP and the magnetic bead-based processes. This results in significant economic benefits, such as savings in both the cost of goods per gram of mAb and fixed costs, as well as increasing annual facility output. The predicted savings potential approaches 38% for COGs, 17% for capital investment, and 40% for annual facility output. For mammalian cell-based manufacturing, the magnetic separation-based DSP provides a highly valuable option due to its integration of several individual unit operations compared to the traditional process both in reducing process time and cost and accommodating higher demands.