Uncertainty quantification for gene delivery methods: A roadmap for pDNA manufacturing from phase I clinical trials to commercialization

The fast‐growing interest in cell and gene therapy (C>) products has led to a growing demand for the production of plasmid DNA (pDNA) and viral vectors for clinical and commercial use. Manufacturers, regulators, and suppliers need to develop strategies for establishing robust and agile supply chains in the otherwise empirical field of C>. A model‐based methodology that has great potential to support the wider adoption of C> is presented, by ensuring efficient timelines, scalability, and cost‐effectiveness in the production of key raw materials. Specifically, key process and economic parameters are identified for (1) the production of pDNA for the forward‐looking scenario of non‐viral‐based Chimeric Antigen Receptor (CAR) T‐cell therapies from clinical (200 doses) to commercial (40,000 doses) scale and (2) the commercial (40,000 doses) production of pDNA and lentiviral vectors for the current state‐of‐the‐art viral vector‐based CAR T‐cell therapies. By applying a systematic global sensitivity analysis, we quantify uncertainty in the manufacturing process and apportion it to key process and economic parameters, highlighting cost drivers and limitations that steer decision‐making. The results underline the cost‐efficiency and operational flexibility of non‐viral‐based therapies in the overall C> supply chain, as well as the importance of economies‐of‐scale in the production of pDNA.

clinical trials of ATMPs at the end of 2021, 60% of which target prevalent diseases, the number of approved ATMP products is only expected to grow. [6,7]cent trials indicate that Chimeric Antigen Receptor (CAR) T-cell therapy is showing promising results compared to existing first-line cancer treatments (e.g., chemotherapy, radiotherapy and immunotherapy treatments) for the treatment of large B-cell lymphoma. [8]As the cell and gene therapy (C&GT) landscape becomes more mature, manufacturers, regulators, and suppliers need to develop strategies that will enable future success and grant access to these innovative therapies to a larger portion of the patient population, given clinical approval. [9]However, the high selling price of ATMPs (US$18,950-US$93,432 for tissue-engineered products, US$110,920-US$814,780 for cell therapy products and US$357,309-US$1,206,751 for gene therapy products [1] ) may limit the availability of these breakthrough therapies to a wide range of patients.
C&GT products use either viral or non-viral vectors as carriers of the gene of interest.As illustrated in Figure 1, the effector molecules expressing the gene of interest can be DNA (e.g., plasmid DNA, viral transgenes, transposons), mRNA, antisense oligonucleotides (e.g., siRNA) or gene editing nucleases (e.g., CRISPR). [10,11]Viral vector gene delivery systems are used in all commercial C&GT products as well as most of those in clinical trials and are considered the current state-ofthe-art as they allow for efficient and long-term gene expression. [12]wever, they pose clinical challenges such as unwanted immunogenicity, mutagenesis and cytotoxicity. [13,14]][17] To overcome these challenges, attention has shifted to non-viral delivery methods as an attractive alternative to viral vectors, [18][19][20][21][22][23][24][25][26][27] offering several advantages such as simpler manufacturing processes, increased yields, and reduced costs, [12,17,28] while also addressing safety and immunogenicity concerns that are associated with viral vectors. [29,30]Non-viral methods include the delivery of naked pDNA as well as physical and chemical pDNA delivery methods. [31]However, these systems have certain drawbacks that need to be overcome prior to their commercialization, such as lower efficiency of gene delivery compared to viral methods. [15]Non-viral methods also have a higher risk of off-target effects, as the gene delivery process may disrupt the normal functioning of the host genome. [12,31]Additionally, non-viral methods can be less stable and have a shorter duration of gene expression compared to viral methods. [12,31] 2021, adeno-associated viruses (AAVs) were used in 46% of all C&GT trials, while at least 8% used non-viral delivery methods, mostly plasmid DNA (pDNA) and nanoparticles. [6]As the C&GT industry evolves, a shift has been observed in clinical trials focusing now more on non-viral delivery systems, [6] which are expected to be complementary to viral vector systems in the near future. [29,30]While viral vectors will continue to be the main delivery strategy for treating rare diseases, non-viral vectors are expected to outperform them for the treatment of prevalent diseases such as cancer, infectious diseases and other chronic diseases. [30]As the field of C&GT continues to advance, it is increasingly evident that viral and non-viral delivery methods each have unique advantages and challenges.Therefore, it is clear that none of the approaches discussed here are mutually exclusive, and we can expect to see a diverse range of C&GT products utilizing different delivery methods as new strategies to optimize their safety, efficacy, and scalability are developed, paving the way for more effective and accessible therapies in the future.
Plasmid DNA (pDNA) is a critical starting material, intermediate, drug substance and/or drug product for the C&GT products supply chain as it carries the gene of interest. [32]Some of the main ATMPs that use pDNA include DNA and mRNA vaccines and viral-based and non-viral-based cell and gene therapies. [32]As illustrated in Figure 1, no matter what the effector molecule or delivery method is, pDNA is always the starting material that characterizes the nature/function of the final product.Thereby, both viral and non-viral delivery systems require the production of pDNA for the development of C&GT products; however, the amount of pDNA required per product differs. [33,34]ral vectors require ultra-low temperature storage on-site, which can impose constraints on inventory capacity and result in increased logistic costs. [35]Storage time for viral vectors typically ranges from 6 to 12 months at temperatures between −70 • C and −80 • C, [36] although certain product formulations may extend the shelf life up to 24 months. [37]pDNA is easier, faster and cheaper to produce, ship and store compared to viral vectors and RNA. [38]Furthermore, pDNA has a much longer shelf life and can be stored for up to 20 years at −20 • C. [39,40] As the C&GT market rises, the demand for pDNA will increase rapidly, mainly due to its key role in the manufacturing of C&GT products as well as mRNA therapeutics. [41,42]Given the ever-increasing C&GT product clinical trials, the demand for small quantities of pDNA for phase I/II trials has been growing in the past years. [28,43]Simultaneously, with more products being approved by regulatory agencies, it is forecasted that hundreds of grams or even kilograms of industrial scale pDNA will be required. [44,45]For example, to produce a lentiviral vector-based CAR T-cell therapy, third-generation lentiviral vectors are transfected with 4 different plasmids -three standard off-the-shelf plasmids (1 packaging plasmid, 1 envelope plasmid, 1 regulatory plasmid) and the transfer plasmid carrying the gene of interest. [46,47]This translates to 4 different batches of pDNA to accommodate one clinical trial.
Cell and gene therapies are often expensive to manufacture and distribute, posing challenges to reimbursement procedures.Plasmid costs are one of the main cost drivers for viral vector manufacturing. [56]esently, the supply of pDNA is limited and costly, considering that none of the pharma and biotech companies has the capacity to produce pDNA in the scale of kilograms. [57]Scaling up both pDNA and viral vector manufacturing processes to meet current and future demand is one of the main bottlenecks of the C&GT industry. [16,34,56,58,59]In many instances, C&GT products are highly customizable and patientspecific, leading to 1:1 business models and bespoke manufacturing F I G U R E 1 Different ways pDNA can be used in cell and gene therapy products.The vector carrying the genetic material can be either administered directly to the target organ (in vivo) or used to genetically engineer cells taken from the host that are then re-administered to the patient (ex vivo). [48]For tissue-engineered and cell-engineered products, the tissue/cells can be derived from the patient (autologous) or another healthy donor (allogeneic).Viral vector gene delivery systems are the current state-of-the-art and are based on the ability of an inactivated virus to inject and replicate its genetic material into a host cell. [49]The main viruses used for gene delivery are based on retroviruses, adenoviruses, lentiviruses, adeno-associated viruses (AAVs) and herpes simplex viruses (HSVs). [13,14,46,50]To date, all the approved C&GT products use viral vectors as a gene delivery method.Most commercially available gene therapies rely on AAVs, [14,46] whilst most CAR T-cell therapies rely on γ-retroviral (Yescarta and Tecartus) and lentiviral vectors (Kymriah, Breyanzi, Abecma, Carvykti). [10,47,51]Physical methods consist of electroporation, microinjection, sonoporation, magnetofection, optoporation, microfluidics, and gene gun. [10,31,49,52]On the other hand, chemical methods are mostly based on nanoparticle (NP) carriers and can be organic such as lipid-based NPs, polymeric NPs, cell-penetrating peptides or inorganic. [10,11,15,31]Generally, inorganic NPs are advantageous compared to other non-viral delivery methods, as they tend to be more stable in biological fluids. [52]Chemical methods also include liposome-and exosome-mediated delivery. [53]Regarding non-viral CAR T-cell therapies, a variety of transposon-based systems, that provide safe and reliable pDNA transfer, have been reported. [54]][21][22]55] The most commonly used method for the delivery of transposon-derived plasmid vectors is electroporation. [33,54]The main benefits of non-virally modified CAR T-cells are the shorter upstream process owing to the simplicity of the transfection process and the overall decrease in cost compared to the viral-based counterpart. [54]d distribution lines.Such an example is CAR T-cell therapies that are intended for critically ill cancer patients.The product lifecycle and associated risks of manufacturing and logistics failures or delays are among the most important key performance indicators and are critical for the patient's life quality. [60]The orchestration of manufacturing and distribution decisions, in such cases, takes place under high demand uncertainty.It is therefore important to build a robust and agile supply chain between plasmid manufacturers, viral vector manufacturers and C&GT manufacturers to ensure the smooth operation of the supply chain.
Centralized manufacturing has been the norm for large-scale, efficient and standardized production of pharmaceuticals, as it benefits from more efficient resource planning, easier monitoring, and overall lower cost per dose. [60]Given the bespoke nature of autologous cell therapies, they cannot benefit from economies of scale. [61]Decentralized or distributed small-scale manufacturing facilities, on the other hand, can be located closer to the end users and offer more operational flexibility. [62]Decentralized manufacturing of raw materials can potentially be a more attractive alternative to the centralized system as it can reduce lead times, batch failure rates and transportation-based costs and risks that derive from the sensitive nature of the raw materials (e.g., specific handling and cryopreservation/thawing cycles) but can also lead to higher production costs. [60,62,63]Furthermore, it may be possible to lower the risk of short-term supply chain failure due to raw material shortages by using multiple suppliers. [63]A shift towards decentralized and point-of-care manufacturing has been discussed in the literature [60,62,63] although not in terms of raw material acquisition.
Accelerating the production of C&GT products requires an industrial understanding of the different strategies available to address raw material shortages, delays and disruptions, but also identification of cost-effective solutions with the scope of advancing the future of regenerative medicine.In this work, we employ Process Systems Engineering (PSE) tools as an approach to strategic and operational planning and uncertainty propagation from the manufacturing process to the supply chain logistics.The study presents a novel model-based methodology for the identification of process uncertainties from clinical to commercial scale in the emerging C&GT industry based on techno-economic analysis (TEA) and global sensitivity analysis (GSA).
TEA combined with GSA can be used to verify the feasibility of a manufacturing process, compare various technologies, and as a means of risk analysis by the quantification of process uncertainties. [64]Our analysis is focused on the production of pDNA and lentiviral vectors for the development of ex vivo autologous CAR T-cell therapies, which are chosen for their manufacturing and supply chain complexities.
Specifically, we identify key process parameters for pDNA produc-  In this way, we use GSA to reduce the dimensionality of the design space by only considering the impact of the most influential process parameters.

Computational methodology overview
4,096 quasi-random input scenarios are generated according to the process parameter ranges in Tables S1-S3, assuming different probability distributions (Tables S4-S6).The probability distributions are based on available data.The Sobol quasi-random sequence is selected as the sampling method to ensure a more efficient exploration of the model input space.The input samples are then simulated with Super-Pro Designer and the target output metrics are obtained.Afterwards, GSA is performed in the SobolGSA software to determine whether the behavior of the selected KPIs is the result of individual effects or interactions between uncertain inputs. [65,66]To do so, the first-order, second-order and total-order sensitivity indices are derived.The GSA algorithm considers the SuperPro Designer model as a 'black box' as it considers only input and output sample values.In this study, we use emulator-based GSA based on metamodeling methods in order to reduce the number of function evaluations needed to achieve convergence.9] A further 1125 input samples are simulated in SuperPro Designer to test the validity of the RS-HDMR surrogate model.The uncertain parameters whose total-order indices exceed the specified threshold (S T ≥S T,min ) for at least one of the output metrics, are defined as critical parameters as described in the "Results" section.The specified threshold, typically ranging between 5%-10%, is determined by the user based on the process nature and the perceived significance of F I G U R E 2 Main steps of the computational methodology for pharmaceutical process uncertainty quantification.
the inputs. [70]This enables the classification of the critical inputs by considering a lower-dimensional projection of the design space.

Plasmid DNA process description
The pDNA production process follows the steps of cell cultivation, plasmid recovery, purification, formulation and fill & finish as shown in Figure 3.The SuperPro Designer flowsheets are built based on the process description and parameter nominal values presented below.

Cell cultivation
Pharmaceutical pDNA fermentation is carried out by cultivation of recombinant Escherichia coli in fed-batch mode. [71]Fed-batch fermentation begins with a batch phase. [72]A seed train composed of three growth steps increased by a factor of 10× is employed. [73][76] The inoculum is then transferred to a bigger shake flask, where growth medium is added, and the fermentation is carried out in batch mode for 12 h. [74,77]Then, the culture is transferred to two staggered seed fermentors working in fed-batch mode for 30 h. [45] Finally, the culture inoculates the two main staggered fermentors operating in fed-batch mode and grows for 40 h. [72,73] The culture is grown at 37 • C and pH 7.0 and aqueous ammonia is added as a supplementary nitrogen source. [73,76]The fed-batch process provides a moderate pDNA yield of 0.25 g⋅L −1 . [45,72]The growth medium contains 20 g⋅L −1 glucose, 7 g⋅L −1 K 2 HPO 4 , 7 g⋅L −1 K 2 HPO 4 , 6 g⋅L −1 (NH 4 ) 2 SO 4 , 2 g⋅L −1 MgSO 4 . [73]Fed-batch medium of 50% w/w glucose is added to the seed and main production fermentors. [73]e growth media are supplemented with 0.05 g⋅L −1 kanamycin as selection pressure. [74,76,78]2.2 Plasmid DNA recovery Downstream processing of pDNA starts with cell recovery followed by cell lysis.[79] After fermentation, the culture is harvested in a solidbowl centrifuge [77,80] for 60 min [74,76,81] and 97% of the biomass is recovered.[73] This step not only provides a concentrated bacterial cell slurry but also removes the majority of the fermentation medium.2] The E. coli biomass paste then undergoes alkaline lysis to disrupt the cells and release pDNA and other intracellular components Process flow diagrams for the production of plasmid DNA and lentiviral vectors. [27]NA, genomic DNA, endotoxins and proteins). [45,72,78]Firstly, the bacterial cell paste is resuspended in Tris-EDTA buffer (50 mM Tris, 10 mM EDTA, pH 8.0) by stirring the mixture constantly at 4 • C until a homogeneous suspension is achieved. [72,74,44,83]Afterwards, lysis is performed for 10 min by rapidly adding the lysis buffer (0.2 M NaOH, 1% SDS). [43,74,44,84]The next step is neutralization for 30 min at 4 • C with the addition of a prechilled high-salt neutralization buffer (3 M potassium acetate, pH 5.5) that promotes the formation of genomic DNA (gDNA), proteins and large RNA molecules aggregates. [73,76,44]e ratio of the resuspension, lysis and neutralization buffers is 1:1:1. [43,75,83]After neutralization, the addition of 5 M calcium chlo-ride solution leads to the precipitation of the remaining RNA, gDNA and proteins. [73,76]e precipitated material containing cell debris and most of the gDNA, high molecular weight RNA, and proteins are removed by centrifugation for 30 min. [83,85]The pDNA supernatant is further clarified through a 0.22 µm depth filter. [75,76,85]To minimize plasmid losses, the depth filter is flushed with a Tris-EDTA buffer. [73]After the cell debris and most of the gDNA and proteins are removed, the majority of impurities in the process are now RNA, gDNA fragments, proteins, endotoxins and plasmid variants.The resulting mixture has a pDNA concentration of 0.10 g⋅L −1 .0]

Purification
The clarified lysate undergoes an ultrafiltration/diafiltration (UF/DF) step, to remove a large portion of the remaining low molecular weight RNA. [74]The UF/DF membrane employed has a molecular weight cutoff of 100-kDa. [73]The solution is concentrated 20× and then diafiltered with 50 volumes of Tris-EDTA buffer containing 0.5 M of NaCl. [73,76,81,83]The buffer exchange in this step accommodates the next step of anion-exchange chromatography.After the solution is diafiltered, the membrane is flushed with extra diafiltration buffer to maximize pDNA recovery. [73]The pDNA concentration after the UF/DF step is 0.18 g⋅L −1 .This solution is expected to still contain some residual RNA and undesired pDNA isoforms such as open-circular pDNA.
Subsequently, two steps of chromatography will be employed to further isolate and purify pDNA from impurities. [78,86]Anion-exchange chromatography (AEC) is the most used and well-suited method for the primary purification of supercoiled pDNA, taking advantage of its polyanionic nature. [72,79,87]A monolithic AEC column is used because monoliths have emerged as an excellent alternative to porous particles, mainly due to their high mass transfer rates, high binding capacities and low void volumes making them more suitable for large molecules such as pDNA. [72,79,87,88]AEC is performed in five cycles per batch with a binding capacity of 10 g⋅L −1 . [73,85,88]After loading, the column is equilibrated with a Tris-EDTA buffer with no salt and pH 7.2, and then a Tris-EDTA buffer with NaCl 0.5 M is employed for the washing step. [73,83]RNA is eluted with a buffer gradient starting with 0.5 M NaCl, followed by plasmid elution in a buffer gradient ending in 1.0 M NaCl. [74,76]Elution is achieved with 10-bed volumes (BV) [73] and the output pDNA concentration is 1.43 g⋅L −1 .
After AEC, the solution undergoes hydrophobic interaction chromatography (HIC), which exploits the hydrophobic nature of singlestranded nucleic acid impurities such as RNA, denatured gDNA, denatured pDNA and endotoxins and hence can isolate supercoiled pDNA. [72,87]The principle of HIC is based on the requirement of high salt concentrations; thus, it can be conveniently used after AEC. [72,79,88]The plasmid binding capacity of HIC resins is low and here it is assumed to be 3 g⋅L −1 . [72,73,79]Therefore, HIC can be considered a polishing step used such that pDNA complies with regulatory standards and is of pharmaceutical grade. [72,73]Three monolithic HIC columns are used in tandem with five cycles per batch each. [73]monium sulphate is added to the plasmid solution to a final concentration of 3.0 M. [73,74,83,84] The supernatant is then loaded onto the HIC column.The column is first equilibrated with a Tris-EDTA buffer and afterwards washed with a Tris-EDTA buffer containing 1.7 M (NH 4 ) 2 SO 4 and then eluted with 10 BV of a Tris-EDTA buffer containing 0.4 M (NH 4 ) 2 SO 4 . [73,74,83,84]The output stream contains a highly pure solution of supercoiled pDNA at a concentration of 0.29 g⋅L −1 .

Formulation and fill & finish
The final pDNA solution is subjected to another UF/DF step using a 100 kDa membrane, in order to exchange the buffer solution used in HIC with an appropriate Tris-EDTA formulation buffer and also concentrate the plasmid to the desired formulation concentration of 2 g⋅L −1 . [28,45,73,74]The solution is first concentrated to a pDNA titer of 10 g⋅L −1 , then diafiltered with 40 volumes of the formulation buffer, and finally flushed to maximize pDNA recovery. [73]e final pDNA product is then sterilized with a 0.22 µm filter [71,74,84] before it goes to fill & finish.The filling formulation is the same as the final formulation obtained after the second UF/DF step.In fill & finish, 1 mL vials are filled with the maximum volume of 1.35 mL, [34] capped, sealed, passed optical or visual quality check, labelled, stored at −20 • C [28,71,75] and finally packaged. [89]

Lentiviral vector process description
The lentiviral vector production process follows the steps of cell cultivation, vector concentration, purification, formulation and fill & finish as shown in Figure 3, and it is based on the model presented in Sarkis et al. ( 2023). [90]

Cell and viral growth
The upstream lentiviral vector production process starts with the working cell bank (HEK 293 cells) expanded in a series of 4 culture steps followed by viral propagation in the main bioreactor. [91,92]At the beginning of each step, the culture is diluted by a factor of 6× via the addition of serum-free medium to an initial cell density of 0.2⋅10 6 cells⋅mL −1 .The serum-free medium contains 6.5 g⋅L −1 glucose, 5 g⋅L −1 amino acids, and 10.5 g⋅L −1 solids in H 2 O. [91] A cell density of 1.4⋅10 6 cells⋅mL −1 is reached after 96 h of culture and the culture is transferred to the subsequent step.The main bioreactor operates at a 2,000 L scale, where cells are first expanded in an aerated broth at 37 • C to a target cell density of 1.9⋅10 6 cells⋅mL −1 . [91]Viral growth is then initiated by transient transfection of the cell culture with 2.5 µg pDNA⋅10 −6 HEK293 cells −1 [34,93] and proceeds for 48 h until a target titer of 1⋅10 7 ][96][97][98] It is assumed that the transfecting pDNA mixture contains the required self-inactivating transfer plasmid with the transgene of interest, the packaging plasmids carrying the gagpol gene and the rev gene, and an envelope plasmid containing the envelope protein. [99]

Primary recovery and purification
Lentiviral vectors are extracellular products and enveloped viruses.
In the downstream process, the harvested culture is mixed with 25⋅10 6 TU⋅mL −1 feed of Benzonase to degrade DNA. [34,93]DNA lysis continues for 1 h at 25 • C. [100] This is followed by a microfiltration step (0.45 µm) which is assumed to remove 100% cells and 80% precipitated nucleic acids, recovering 80% of the lentiviral product. [34]The filtration time for the step corresponds to 4 h, the mixture is concentrated by a factor of 20× and the resulting filtrate flux corresponds to 15 L⋅m −2 ⋅h −1 . [100]The clarified lentivirus mixture is sent to an ion exchange (IEX) chromatography step with a binding capacity of 5⋅10 8 TU⋅mL −1 , which recovers 40% of the lentiviral product and removes >70% of protein and nucleic acids impurities. [34,93,101]First, the product mixture is loaded onto the column at a rate of 30 BV⋅h −1 for 7.5 02] 2.

Formulation and fill & finish
The purified lentiviral eluted solution is ultrafiltered-diafiltered (UF/DF) (0.005 µm).First, 40 L⋅h −1 of PBS-based buffer for 15 min are used for the flush step. [100]Secondly, the product is diafiltered, with a product recovery of 80% and titer of 9.8⋅10 8 TU⋅mL −1 . [34]The concentration factor for this step is 6×, with a process time of 4 h and a resulting filtrate flux of 5 L⋅m −2 ⋅h −1 . [100]The drug substance is sterile-filtered (0.2 µm), cryofrozen to −70 • C with liquid N 2 and stored.In the fill & finish process, the drug substance is thawed, formulated to a target concentration of 2⋅10 9 TU⋅mL −1 and filled in 50 mL vials, to be capped, labelled, cryofrozen and boxed. [34]

Data sources and assumptions
Information regarding pDNA and lentiviral vector production processes (Sections 3.3-3.4)and costs is obtained from the scientific literature, from cGMP grade manufactures and the SuperPro Designer's database.Media and key raw materials used in the processes are based on SuperPro Designer's example library. [73]A batch failure rate of 5% is assumed for these processes.
Equipment cost equations are developed based on SuperPro Designer's database and actual prices of laboratory equipment.Information on fill & finish processes and economics are obtained from the literature [34,89] and equipment suppliers. [103,104]The amount of pDNA required for the different scales in Table 1 is calculated based on VGXI's white paper and considers several additional factors such as release testing quantities and stability testing quantities. [105]It is assumed that the pDNA quantity required in each demand scenario is fulfilled in one batch.The lentiviral vector production process uses pDNA for HEK293 cell transfection and is assumed to require 2.5 µg pDNA⋅10 −6 HEK293 cells. [34]Furthermore, the dose size of lentivirus for CAR T-cells transfection is calculated based on the following information found in the literature: [34]  We modelled the different demand scale pDNA flowsheets based on linear scaling with the assumption of a moderate pDNA yield of 0.25 g⋅L −1 . [43,72]Most cGMP suppliers report higher yields of up to 0.55 g⋅L −1 . [45,72]However, achieving such high yields requires cell line optimization and gene sequence optimization, which is highly unlikely in the case of a new process.
More than 25 uncertain inputs are tested based on the proposed methodology (Figure 3).The uncertainty distribution for the input parameters is considered to be either uniform, with equal probability for every value in the range, or triangular with the highest probability for the nominal/peak value (Tables S4-S6).Triangular distribution is chosen for the diafiltration volumes in the 1st UF/DF step and the 2nd UF/DF step of the pDNA process because there is a small uncertainty in those parameters based on the literature. [73,74,76,81,87]For the remaining inputs, a uniform distribution is assumed.The ranges are assumed to vary ± 50% from the nominal value for all inputs in the nonviral-based scenario except for fermentation duration and diafiltration volumes in the pDNA process because the lower and upper bounds for these cases are more certain based on the literature.The ranges for all inputs in the viral-based pDNA production scenario are obtained from the literature.In addition, the input range for the pDNA cost in the lentiviral vector process is given by the results obtained from the pDNA process.
Sensitivity and uncertainty analyses are not performed for the anion exchange and hydrophobic interaction chromatography steps in the pDNA process.These are particularly complex systems, best described by dynamic models that detail the principles of separation.
Chromatographic separation is well-established and standardized in the case of pDNA, while uncertainties in its performance are reported for lentivirus purification. [101]For this, parameters related to chromatographic separation are included in the sensitivity and uncertainty analyses of the lentiviral process only.
For the purposes of this study, we assume processes that have already been optimized and approved by regulatory authorities and therefore key process parameters, such as temperatures, pressures, medium and buffer compositions are fixed.
TA B L E 1 Different demand scale scenarios for pDNA production.

Software
The bioprocess simulation platform SuperPro Designer Version 12, Build 3 from Intelligen, Inc (Scotch Plains, NJ, USA) is used to perform the techno-economic analysis.SuperPro Designer calculates material and energy balances and performs cost analysis with its built-in process and economic models based on sets of algebraic and differential equations.SuperPro Designer sizes the equipment, computes labour requirements, schedules operations and procedures, and conducts economic calculations for both capital and operating expenses.
Furthermore, apart from utilizing user-specified inputs, SuperPro Designer is linked to chemicals, consumables, equipment and other databases. [106,107]SobolGSA version 3.1.1 is used to perform the global sensitivity analysis and the user-defined equations are built in MATLAB R2022a.SobolGSA is a general-purpose global sensitivity and metamodeling software. [65,66]SuperPro Designer and SobolGSA are connected using the Component Object Model (COM) feature available in SuperPro Designer and linking it with Excel Visual Basic for Applications (VBA).

Case study overview
The computational methodology described in Section 2.1 is used to explore the acquisition of critical raw materials -pDNA and lentiviral vectors -for the production of CAR T-cell therapies.In this study, we examine the manufacturing of the raw materials for both the forwardlooking non-viral and the current state-of-the-art viral production of CAR T-cells.Table 1 displays the four demand scales considered herein.
][23][24]55] The key starting material for non-viral cell therapy is pDNA and as of yet, there is no commercial product avail- demand scenarios (Table 1).
In parallel, we investigate the process uncertainties in the produc-

Uncertainty and sensitivity analysis metrics
A variance-based global sensitivity analysis is performed to quantify how model input uncertainty propagates to output metrics and allocate output variability to specific uncertain inputs and their interactions.
For each scenario (Table 1)  4 Violin plots for all the key output metrics under the non-viral production of CAR T-cell therapy demand scenarios presented in Table 1.The black bar in the center of each plot is the interquartile range, the white dot inside the black bar is the median value, and the black lines stretched from the bar are the lower/upper adjacent values.Observations outside of the black lines can be considered outliers -cf. Figure S1 for violin plot comparisons on different y-axis scales.least 2% of the output variability for at least one output, and their corresponding ranges are summarized in Tables S1-S3.
In this section, the total output variance, the explained variance and the contributions of the individual critical input parameters and their interactions are examined.Violin plots in Figures 4 and 6 depict the total variance of the output metrics when all critical uncertain parameters are simultaneously varied based on specific ranges and distributions.Figures 5 and 7  Accordingly, Figure 5 shows the first, second and total order indices for all KPIs for those scenarios.

Batch size
The batch size (i.e., productivity) is firstly examined for the three different pDNA demand scenarios.The median values reported in the violin plots in Figure 4A are very close to the target batch sizes presented in Table 1 for each scale.There is similar batch size variability in all three scales (Figure S1 that the realized batch size for all scales might be substantially lower than the median value.Figure 5A suggests that the highest contributors to batch size variability are the fed-batch medium volume (35%), the initial medium working volume (21%) and the diafiltration volumes in UF/DF 2 (24%), with a relatively significant combined second-order interaction for the former two input volumes (8%).In phase I/II clinical trials, the fermentation duration in the main fermentor accounts for 87% of the variability (Figure 5B).In phase III clinical trials and commercial scale, 56% of the batch time variability is attributable to the main fermentation duration and 37% to the seed fermentation duration (Figure 5B).The model appears to behave linearly in terms of the batch time with all interaction terms being approximately zero.

Batch time and cycle time
Cycle time variability distribution is identical at all scales (Figure 4C). Figure 5C shows that the main fermentation duration explains 91%-93% of the variability in cycle time in all cases.The

CapEx, OpEx, and production cost
Figure 4D shows that the capital investment cost does not scale linearly with batch size.The phase I/II clinical trials scale has a very high capital cost with a median value only 4× lower than the commercial scale.
In comparison, the batch size is 158× lower on phase I/II clinical trials scale compared to the commercial scale.The major contributors to the variability in capital costs across all scales are fed-batch medium volume and initial working volume, with respective first-order interactions (36%-40%) and (20%-25%) as well as second-order effects (5.5%-7%) (Figure 5D).Varying working volumes in fact require equipment resizing, which directly impacts expenditure.Hence, first-order sensitivities are also seen for inputs related to volumes of buffer added to the process (6.5%-22%).
Similarly, operating cost is not significantly different across demand scales (Figure 4E).This is because pDNA variable costs are governed by high fixed costs such as labour and facility-dependent costs, which is typical for cell-based and biopharmaceutical processes (Figure S2). [110] observed in Figure 5E, there are some key differences in the apportionment of variance to individual input parameters across the scales.The most important contributor to operating expenditure is the main fermentation duration, with high first-order contributions across all scales (65%-89%).The main fermentation duration is expected to impact labour costs (Figure S1) by varying the required working hours.Furthermore, at commercial scale, 11% of the OpEx variability is attributable to the fed-batch volume, as the equipment is resized accordingly and therefore causes some variation in the computed equipment depreciation cost.
It is apparent that the production cost is inversely proportional to the batch size (Figure 4A,F) proving that economies of scale appear to be the main driver for lowering pDNA production costs.The apportionment of production cost variability in terms of the individual parameters is similar across scales and mirrors that of the batch size with the addition of fermentation duration as a moderate input uncertainty (Figure 5F).Significant first-order effects are in fact attributed respectively to fed-batch volume (38%-44%), diafiltration volumes (30%-36%) and initial working volumes (19%-22%).

Plasmid DNA and lentiviral vector uncertainty and sensitivity analysis for viral-based CAR T-cells
The manufacture of lentiviral vectors for CAR T-cell therapy transduction uses pDNA to initiate viral propagation in the main bioreactor.
The integrated pDNA and viral vector manufacturing system for viralbased CAR T-cell therapy is therefore examined, and GSA is implemented to quantify uncertainty around techno-economic KPIs.As viral vector-based CAR T-cell therapies are already commercially available, the two manufacturing platforms are assumed to be fully developed, validated and implemented at a large scale.This entails considering the pDNA commercial scale (40,000 patients) (Table 1) and the commercial scale for lentiviral vectors. [34,101]Equipment size is fixed, and a range of operational inputs is considered, including key process steps and yield for both platforms.Additionally, the computed pDNA unit production cost is integrated as an input in the lentiviral vector analysis.The violin plots in Figure 6

Batch time, cycle time and throughput
Information regarding the intrinsic production rate of the two integrated platforms can be obtained by analyzing the batch time and cycle time of the processes.As can be seen in Figures 7B,H, the time required to process a batch is mainly impacted by the variability around the longest tasks in the process.These correspond to task durations in the seed and main fermentor for the pDNA process (Figure 7B) and result in a range of batch times (M = 132 h, IQR = 126-138 h) (Figure 6B), whereas the lentiviral vector manufacturing batch time 6H) is influenced mostly by the duration of cell expansion steps (22.5%), and the viral growth step duration (10%) (Figure 7H).The duration of these steps determines the location of bottlenecks of the processes in the USP, with cycle time for pDNA (M = 36 h, IQR = 32-41 h) (Figure 6C) remaining significantly shorter than in the lentiviral vector process (M = 6.8 d, IQR = 6.1-7.6 d) (Figure 6I).Cycle time variability in the lentiviral vector process is dictated by the longest task durations, which take place in the main bioreactor, namely cell expansion (68%) followed by viral growth (28%) (Figure 7I).  1.The black bar in the center of each plot is the interquartile range, the white dot inside the black bar is the median value, and the black lines stretched from the bar are the lower/upper adjacent values.Observations outside of the black lines can be considered outliers.
coefficient in UF/DF 2 (78.2%) (Figure 7F) as these directly impact the batch size.Second-order interactions between the main fermentation duration and the concentration rejection coefficient in UF/DF 1 are also important (8.1%) (Figure 7F).
In lentiviral vector manufacturing, capital cost variability (M = M$128, IQR = M$127.4-128.5)(Figure 6J) corresponds to the working capital output variance, and it is mostly explained by the uncertainty in the cost of the most expensive raw material, namely pDNA ( S pDNA = 50%) (Figure 7J).Sensitivity of the capital cost to and viral growth task lengths (13% and 5%) is also observed, as these inputs impact the number of batches processed within the first 30 days of annual operation and the required working capital.The varying cost of pDNA also results in a variance of the operating cost per batch (M = 835,000$⋅batch −1 , IQR = 751,000-919,000$⋅batch −1 ) with a sensitivity of 99.6%, highlighting the contribution to materials expenditure per batch compared to other operating cost components (Figures 6K and 7K).Finally, the cost of producing a lentiviral vector dose for CAR T-cell transduction (M = 3,400$⋅dose −1 , IQR = 2,403-4,901$⋅dose −1 ) is highly sensitive to input parameters that impact the batch size, namely USP titer (53.1%),IEX recovery (32.8%) and to the cost of pDNA (4.6%) (Figures 6L and 7L).

DISCUSSION
The computational methodology presented in this study starts with  a IQR(M).
Manufacturing Organizations (CDMOs).The computational methodology presented here can therefore accelerate the approval and  These cost metrics play a crucial role in determining decisions regarding in-house or outsourced raw material manufacturing at different stages of scale-up.

Production and investment costs
The production of pDNA benefits from economies of scale, however, to a smaller extent than lentiviral vector production. [90]Manufactur- Sarkis et al. (2023) [90] quantified the benefit of economies of scale in lentiviral vector production, revealing that small-scale manufacturing translates into a 100-fold increase in cost per dose.This highlights the significance of maximizing the volumetric scale in lentiviral vector manufacturing, as it substantially reduces costs.In this study,  1.The plots are presented on a logarithmic scale to accommodate the wide range of cost values.The violin plots serve to summarize and compare the cost variabilities between the two gene delivery methods, providing insights into the cost efficiency associated with each approach.The black bar in the center of each plot is the interquartile range, the white dot inside the black bar is the median value, and the black lines stretched from the bar are the lower/upper adjacent values.Observations outside of the black lines can be considered outliers.
corresponding selling price of raw materials from a Contract Manufacturing Organization (CMO). [34,93]ese production costs also apply to in-house manufacturing of raw materials, which entails additional investment risks for capacity (2022). [17]These findings emphasize that the adoption of non-viral vectors in CAR T-cell therapy manufacturing could make in-house manufacturing of raw materials an attractive option, even in the early stages of clinical trials.and pDNA. [37,39]In current industrial scenarios, viral vector CMOs typically serve a variety of clinical and commercial applications and strategically allocate resources within production campaigns to fulfil target mid-term orders from CAR T-cell therapy manufacturers.The extended shelf life of lentiviral vectors allows large inventories to be depleted according to the incoming patient schedule.Similarly, pDNA is procured and stored in advance for lentiviral vector transfection. [39,40]e resilience of this make-to-stock strategy employed by supply chain agents, based on mid-term orders, can be challenged by demand uncertainty.Manufacturing slots with CMOs are typically booked more than 1 year in advance for lentiviral vectors [111][112][113] and 3 months for pDNA, [112] resulting in lengthy order delivery times.Therefore, an unforeseen increase in demand that necessitates additional orders submitted to CMOs may lead to backlogs and delays.This puts pressure on manufacturers to develop a robust raw material acquisition plan a priori, capable of addressing various demand scenarios.Additional complexities emerge at the CMOs level, as they ought to cope with demand uncertainty across multiple markets, including potential surges in demand for vector-based vaccines in the event of future pandemics.These circumstances emphasize the need for a proactive approach to capacity planning under uncertainty for raw material suppliers. [90]spite the time invested in process development, plant installa-

Outlook
The The analysis presented in this work highlights the complex tradeoffs and interdependencies with respect to costs and responsiveness which emerge during supply chain planning activities for C&GT products.Given the company's objectives and resources available, the methodology can be complemented with Mixed Integer Programming (MIP) and multi-objective optimization approaches to assist decisionmakers in the C&GT sector with identifying the most-cost effective and responsive solution. [114,115]

A
model-based methodology is developed to address the challenge of quantifying key process uncertainties that heavily influence the C&GT manufacturing and supply chain.landscape.The main steps of the developed methodology are illustrated in Figure 2. The methodology is based on the analysis of detailed process flowsheets to produce pharmaceutical-grade pDNA at four different scales and the largescale production of lentiviral vectors in SuperPro Designer (Intelligen, Inc).A variance-based GSA is performed based on user-identified key economic and process parameters.GSA is used to quantify the contribution of uncertain input parameters to the output variance.Uncertain parameters based on literature, cGMP manufacturers, suppliers and expert feedback on the design of specific processes are considered.The distribution for each uncertain parameter is chosen based on literature or empirical knowledge.Parameters that mostly account for the variance in the selected output metrics -batch size, batch time, cycle time, capital investment cost, operating cost, and production cost -are regarded as critical parameters.The non-influential parameters, whose total sensitivity indices do not exceed a user-specified threshold ( S T,min = 2%), are then fixed and kept constant in the process flowsheet for the subsequent global sensitivity runs.This procedure is repeated until all the remaining parameters are critical and the algorithm converges.
2⋅10 9 TU⋅dose −1 , 20 TU⋅cell −1 HEK293.Both processes comprise unstaggered equipment prior to formulation, with resulting batch times of approximately 5 and 22 days for pDNA and lentivirus, respectively.A seed fermentor is not used for the phase I/II clinical trials pDNA production process because of the small scale.In the pDNA for viral vector-based CAR T-cell therapy scenario, staggered equipment is used for the whole fill & finish section because of the magnitude of the scale.Specifically, the largest available fill & finish equipment cannot accommodate the large scale and fill & finish ends up being the time-limiting step.To overcome this, a second parallel line of fill & finish is utilized.Parallel lines are not required in lentiviral product filling steps as products are highly concentrated and can be filled in 50 mL vials at rates of 10 vials⋅min −1 .
able.Therefore, we develop the SuperPro Designer process flowsheet for pharmaceutical pDNA in design mode, which computes equipment sizes as model outputs alongside the KPIs of interest, namely batch size, batch time, cycle time, capital investment cost, operating cost, and production cost per gram.This enables an analysis of how process uncertainties for processes under development may impact the costrelated and throughput-related KPIs.We developed three different flowsheets for the different demand scales.Specifically, the production of pharmaceutical-grade pDNA for phase I/II clinical trials, phase III clinical trials, and commercial use was modelled based on future of viral CAR T-cell therapies for commercial products based on lentiviral vectors (i.e., Kymriah, Breyanzi, Abecma, Carvykti).In this process, pDNA is the starting material for the transfection of lentiviral vectors, which are then used in CAR T-cell manufacturing.We examine only the commercial demand scenario since these products are already commercially marketed.Hence, the SuperPro Designer flowsheet for both pDNA and lentiviral vectors for the viral gene delivery route is in rating mode.This translates into an already established process where the equipment size is fixed, and we investigate operational uncertainties and performance.

, 4 ,
096 quasi-randomly generated samples of model input variables based on Sobol sequence, within the ranges shown in Tables S1-S3, are simulated to obtain the corresponding outputs.The chosen output metrics quantify the techno-economic performance of the processes in terms of batch size, batch time, cycle time, capital cost, operating cost and unit production cost.Batch time is defined as the total time required to manufacture a batch of product, while the cycle time corresponds to the length of the bottleneck process step and becomes the minimum time required between two consecutive batches within a manufacturing campaign.The aforementioned metrics are crucial for quantifying manufacturing and supply chain uncertainty.Manufacturing uncertainty is visualized as a violin plot, which presents the median value of the set of output samples (M) and the interquartile range (IQR), which represents the spread of the middle 50% of the output samples.Quantifying the uncertainty distribution provides more confidence about the trends and comparisons required for decision-making across scales and production routes.The initial screening of influential parameters started with various parameters including task lengths, conversions, buffer volumes, working volumes, rejection and concentration coefficients, failure rates, etc.The inputs that have little or no effect on all output metrics are fixed to their nominal values.Finally, the global sensitivity analysis is repeated until all the remaining inputs are influential.The final critical input parameters for non-viral and viral gene delivery, which account for at Figure4present the total variance of the KPIs for all non-viral production scale scenarios (also shown on different y-axis scales in FigureS1).

Next, the three
different scales are compared in terms of batch and cycle time.As illustrated in Figure 4B, pDNA production for phase III clinical trials and commercial use has an identical batch time variability distribution (M = 109 h, IQR = 102.9-115h).On the other hand, pDNA for phase I/II clinical trials is predicted to be the fastest due to the lack of the seed fermentation step (M = 88 h, IQR = 82.5-93.6 h).
broad cycle time variability (M = 36.2h, IQR = 32.4-41.1 h) as well as the median could be reduced by using staggered equipment (scale-out) for the time-limiting steps.Staggered equipment would increase the number of batches produced per year (throughput) as well as capital expenditure.Throughput is directly influenced by cycle time as the latter determines the frequency of a new batch.
present the variability of KPIs for the production of pDNA and lentiviral vectors used in the manufacturing of viral-based CAR T-cell therapies.Figure 7 illustrates the underlying 1st, 2nd and total order sensitivity indices for all KPIs.Results are compared to the KPIs of pDNA production for non-viral-based CAR T-cell therapies.3.4.1 Batch size The amount of pDNA required per dose for commercially available viral vector-based CAR T-cell therapies is more than 12x higher than the not-yet clinically approved non-viral CAR T-cells (Figures 4A and 6A), resulting in a proportionally larger batch size.The requirements for pDNA for the viral pathway are increased as they factor in transfection efficiency for viral propagation in the producer cell line as well as transduction efficiency of the viral vectors in the CAR T-cells during cell activation and growth.In pDNA production, the batch size (M = 55.2 g⋅batch −1 , IQR = 40.6-75.4g⋅batch −1 ) is mostly influenced by the pDNA diafiltration rejection coefficient in UF/DF 2 (75%) (Figure 7A).This is attributed to the fact that the target stream is stripped of most of the impurities after chromatography and is more concentrated in pDNA compared to UF/DF 1, thus any losses in pDNA at UF/DF 2 result in significantly lower batch size.Simultaneously, there are strong interactions between the seed fermentation duration and the pDNA rejection coefficient in the concentration step of UF/DF 1 (10%) resulting in a non-additive model.In the lentiviral manufacturing process the batch size (M = 244 doses⋅batch −1 , IQR = 171-344 doses⋅batch −1 ) strongly depends on the viral titer and the recovery of the IEX separation step and their interactions with respective individual contributions of (55%) and (38.9%) (Figure 7G).

3. 4 . 3
CapEx, OpEx, and production cost Key inputs that impact capital costs, operating cost per batch and unit production costs are also identified.Capital and operating costs for the pDNA process demonstrate similar variability to the smaller non-viral scales (Figure6D,E and Figure4D,E).Capital cost variability is apportioned to centrifugation duration in the second centrifugation step (78%), the first centrifugation step (8.4%) and the main fermentation duration (11.5%) (Figure7D).Specifically, the second centrifugation step handles 9.55x higher volumetric flows compared to the first, which explains the high apportionment of CapEx variability to the second centrifuge.The variability in this KPI (M = M$638, IQR = M$633-644) is less significant when compared to the non-viral case studies as equipment size is fixed in the viral-based flowsheet simulations, which eliminates variability in equipment purchased cost.The operating cost (M = 46 M$⋅batch −1 , IQR = 42.5-48M$⋅batch −1 ) mostly varies due to the task length in the main fermentor (86%) (Figure 7E), which impacts the required working hours and correlated labour cost contribution.The production cost per gram of pDNA for the viral pathway results in an order of magnitude less (M = 3,900$⋅g −1 , IQR = 2,800-5,300$⋅g −1 ) than its non-viral counterpart (Figure 6L, Figure 4F), which highlights that pDNA production benefits from economies of scale and a larger batch size lowers unit production costs.The production cost per gram is mostly influenced by the diafiltration pDNA rejection F I G U R E 6 Violin plots for all the key output metrics under theviral-based production of CAR T-cell therapies demand scenarios as presented in Table the development of process flowsheets to produce pharmaceutical plasmid DNA (pDNA) and lentiviral vectors.Both are critical starting materials, intermediates, drug substances and/or drug products for the C&GT supply chain.The model-based methodology then continues with the identification of the relevant process uncertainties and the assignment of a probability distribution to each uncertain input.Subsequently, uncertainty and global sensitivity analysis is conducted, and the estimates are assessed based on KPIs relevant to the C&GT supply chain.The goal is to provide manufacturers, regulators, and suppliers with useful insights that will enable them to develop strategies for the establishment of a robust and agile supply chain in the otherwise empirical field of C&GT.Applying a systematic global sensitivity analysis to apportion the uncertainty propagation to individual input parameters can help gain more insight into the manufacturing process by ranking parameters in terms of sensitivity to specific KPIs.These insights can then steer decision-making and may involve batch sizes, inventory management, in-house manufacturing or outsourcing to Contract Development andF I G U R E 7Global sensitivity analysis of multiple uncertain input parameters on key performance indicators (KPIs) for the production of pharmaceutical pDNA for lentiviral vector transduction and the production of lentiviral vectors.The uncertain input parameters for the pDNA process are seed fermentation task length (t seedfermentor ), fermentation task length (t mainfermentor ), centrifugation 1 task length (t1 centrifugation ), centrifugation 2 task length(t2 centrifugation ), concentration pDNA rejection coefficient in UF/DF 1 (x1 conc-pDNA ), concentration pDNA rejection coefficient in UF/DF 2(x2 conc-pDNA ), diafiltration pDNA rejection coefficient in UF/DF 2 (x1 dia-pDNA ).The uncertain input parameters for the lentiviral process are seed bioreactor 1 task length (t1 cellexpansion ), seed bioreactor 2 task length (t2 cellexpansion ), seed bioreactor 3 task length (t3 cellexpansion ), main bioreactor task length (t4 cellexpansion ), viral growth task length(t4 viralgrowth ), USP viral titer (x conc-virus ), IEX elution recovery (x iex-virus ), and pDNA cost (c cost-pDNA ).The KPIs are productivity (batch size), batch time, cycle time, CapEx, OpEx, and production cost.First-order effects (S i ) are depicted in the diagonal and second-order effects (S ij ) are presented in the upper and lower triangular.Total first-order effects (∑S i ), total second-order effects (∑S ij ), and total effects (∑S T ) are also displayed.TA B L E 2 Summary of key output metric variability, which is represented as the interquartile range (IQR) and the median value of the set of output samples (M).
production of C&GT products by ensuring efficient timelines, scalability, and cost-effectiveness of the production of key raw materials (pDNA and lentiviral vectors).The proposed methodology is applied to the industrially relevant case studies of pDNA manufacturing for the non-viral Sleeping Beauty transposon system for the production of CAR T-cell therapies, as well as pDNA and lentiviral vector manufacturing for the already commercially available CAR T-cell therapies.Non-viral-targeted pDNA is examined from clinical to commercial scale, as practical considerations and careful planning are needed during the early stages of development up to commercialization when a new technology is established.On the other hand, viral-targeted pDNA is only analyzed on a commercial scale.With pDNA's central role in the C&GT market and with numerous C&GT products in the pipeline from discovery to market launch, the importance of identifying uncertainties in the manufacturing process qualitatively and quantitatively from clinical to commercial scale is significant.As seen in FigureS1, the variability distribution patterns for all metrics are almost identical throughout the scales and thus uncertainties quantified early on can be reduced before product commercialization ensuring cost-effectiveness and efficient timelines.

the lentiviral vector cost per dose at a 2 , 1 .
000 L scale factors in the pDNA cost variability and corresponds to M = 3,400$⋅dose −1 , IQR = 2,403-4,901$⋅dose −Switching from viral-based therapies to non-viral-based CAR T-cell therapies results in a significant reduction in raw materials cost.The pDNA cost per dose, which is now the key raw material in the non-viral pathway, is demonstrated to be up to 1000-fold lower for non-viralbased therapies compared to viral vectors (Figure 8).Even when pDNA is manufactured at a smaller scale (Phase I/II clinical trials scale), the cost per dose does not surpass 50$⋅dose −1 , indicating that the smallest scale of the non-viral pathway remains more cost-effective than the largest scale of the viral-based one (Figure 8).Hence, economies of scale are not as important for key raw materials of non-viral CAR T-cell therapies, making them an attractive alternative for rare diseases.It should be noted that the raw material production costs for both pathways provide insight into the order of magnitude of the F I G U R E 8 Violin plots for raw material production costs per dose under the non-viral-based and viral-based production of CAR T-cell therapies demand scales as presented in Table installation and further costs for personnel training to develop inhouse manufacturing expertise.We quantify the capital investment required for both viral-based and non-viral-based raw materials.For viral vector manufacturing, a single production line can yield a maximum of 15,000 doses per year across 50 batches per year.To fulfil a target commercial demand of 40,000 CAR T-cell doses annually, installing a capacity of more than 3 bioreactor lines would be necessary.Investments required for such capacity would include lentiviral vector capital costs (M = $128 M, IQR = $127.4M-128.5 M) per manufacturing line as well as potential in-house pDNA capital costs (M = $63.83M, IQR = $63.34M-64.46 M).In contrast, the nonviral pathway requires significantly lower investments (M = $18 M, IQR = $17.5 M-19.7 M), approximately 10-fold less than the viralbased raw materials, which aligns with estimates by Moretti et al.

From
an operational perspective, the successful delivery of CAR T-cell therapies requires seamless coordination among all parties involved and a supply chain that can quickly adapt to short-term demand fluctuations.In the CAR T-cell therapy space, demand can rapidly scale up or decline depending on therapy success (clinical trials).After market approval, the demand for cancer therapies would remain challenging to predict, with non-seasonal, day-to-day fluctuations possibly occuring.This places pressure on CAR T-cell therapy manufacturers to proactively develop supply chain strategies to cope with demand uncertainty, spanning from clinical to commercial stages.Making informed deci-sions regarding raw material procurement becomes crucial to mitigate manufacturing delays.The identified variability for batch and cycle times along with the corresponding dominating input parameters that contribute to this variability in both the non-viral and viral pathways provides valuable insights into the operational flexibility and how to ensure robustness and resilience of the overall CAR T-cell therapy supply chain.Specifically, this sheds light on the use of pDNA as a critical raw material for CAR T-cell therapies, as opposed to pDNA being a starting raw material used in the manufacturing of a critical raw material (i.e., lentiviral vectors).Additionally, these findings help manufacturers assess the challenges and opportunities of in-house versus outsourced raw material acquisition for each pathway.The involvement of CMOs offers CAR T-cell therapy manufacturers access to critical raw material manufacturing expertise and capacity.However, planning decisions in this context are often taken asynchronously, based on company-specific objectives.Outsourcing raw material production to specialized CMOs and accumulating raw materials stocks is a well-established strategy in the sector, driven by the long shelf lives of both lentiviral vectors tion and validation activities, in-house raw material manufacturing paves the way for on-demand raw material production, with potentially improved responsiveness to demand uncertainty.Specifically, in-house lentiviral vector manufacturing entails batch times of M = 21 d, IQR = 20-22 d of manufacturing and cycle times M = 6.8 d, IQR = 6.1-7.6 d.On the other hand, it would only require hours (M = 88 h, IQR = 82.5-93.6 h for a Phase I/II clinical trial batch and M = 109 h, IQR = 102.8-115.1 h for a commercial batch) to manufacture batches of the key raw material for non-viral CAR T-cell therapies (i.e.pDNA).Faster production timelines enable more reactive scheduling under demand uncertainty, batch failure and other disruptions.These benefits combined with the discussed reduced investment risk highlight the inherent flexibility and cost-competitiveness of an inhouse pDNA manufacturing set-up for non-viral CAR-T cell therapies.This paves the way for increasingly distributed raw material manufacturing, that moves away from the centralized counterpart.Raw material decentralization in this fashion holds the potential to significantly enhance the resilience of the end-to-end CAR-T cell therapy supply chain.This would represent a significant step towards CAR T-cell point-of-care manufacturing, which would offer a multitude of benefits, including improved supply chain responsiveness by providing access to treatments in underserved or remote areas and granting healthcare providers the flexibility to swiftly adapt to patient needs.
study presented herein compares the commercially established viral gene delivery with the potential of non-viral gene delivery for CAR T-cell therapy manufacturing, assuming the latter surpasses clinical bottlenecks and reaches commercialization.This study serves as a valuable tool to assess the cost-effectiveness and supply chain resilience of the two delivery modes.A computational methodology is developed to compare the manufacture of critical raw materials for both delivery modes and conduct a scalability assessment for pDNA manufacturing, aiding in the transition stages of introducing the pDNA-based non-viral technology to the markets.The proposed decision-support methodology can be applied to any C&GT candidate product or intermediate no matter which effector molecule carrier and delivery system are utilized.Four different scales for pDNA manufacturing were tested, ranging from batch sizes of 0.03 g to 60 g.Quantifying the uncertainty around KPIs, production scales and supply chain pathways reinforced the comparative analysis, by highlighting cost drivers, limitations, and key process parameters that will aid decision-making.Such decisions may include in-house or outsourced manufacturing of critical raw materials, centralized or decentralized manufacturing, batch sizes and inventory management decisions.Specifically, the study highlights the inherent cost-effectiveness of pDNA-based non-viral C&GT supply chains compared to the viral vector-based counterparts.It suggests that decentralized manufacturing of raw materials for non-viral therapies, closer to or co-located with the cell therapy facility, offers a significantly more cost-efficient alternative to the well-established centralized viral vector-based network.Notably, non-viral CAR T-cell therapies can achieve cost-effectiveness even when the key raw material (i.e., pDNA) is manufactured locally in smaller batches (5 g⋅batch −1 ) as opposed to centralized large-scale lentiviral vector manufacturing.From an operations perspective, distributed pDNA manufacturing would minimize logistics delays and maximize the responsiveness and flexibility of cell therapy manufacturers to incoming patient schedules.With raw materials available within shorter timeframes as patient samples arrive, cell therapy manufacturers can expedite the production process.Timelines for the manufacturing of additional batches of pDNA would be lowered (≈4 days) compared to the viral-based supply chain (≈22 days), which would improve the end-to-end supply chain responsiveness under demand uncertainty.
CONTRIBUTIONS N.T.: Conceptualization, methodology, formal analysis, visualization, writing -original draft, and writing -review & editing.M.S.: Conceptualization, methodology, formal analysis, writing -original draft, and writing -review & editing.A.K.: Methodology.N.S.: Conceptualization, supervision, and writing -review & editing.M.M.P.: Conceptualization, supervision, and writing -review & editing.C.K.: Conceptualization, supervision, and writing -review & editing.ACKNOWLEDGMENTS This research is funded by the Department of Health and Social Care using UK Aid funding and is managed by the Engineering and Physical Sciences Research Council (EPSRC; grant EP/R013764/1).The views expressed in this publication are those of the author(s) and not necessarily those of the Department of Health and Social Care.Funding from the UK Engineering & Physical Sciences Research Council (EPSRC) for the Future Targeted Healthcare Manufacturing Hub hosted at University College London with UK university partners is gratefully acknowledged (Grant Reference: EP/P006485/1).Financial and in-kind support from the consortium of industrial users and sector organizations is also acknowledged.NT is thankful for the Marit Mohn Scholarship awarded by the Department of Chemical Engineering, Imperial College London.