Challenges and Emerging Technologies in Biomanufacturing of Monoclonal Antibodies (mAbs)

Abstract

Therapeutic antibodies dominate the biopharmaceutical market with continual innovations being made to provide novel and improved antibody treatment strategies. Speed to-market and cost-efficiency are of increasing importance due to the changing landscape of the biopharmaceutical industry. The increasing levels of competition from biosimilars, the increase in small volume products and political and social pressure to reduce the cost of treatments are some of the challenges currently being faced. Chinese hamster ovary (CHO) cells have been the workhorse in the production of therapeutic antibodies over the last 36 years due to the robust nature and high productivity of these cell lines. However, there are many biomanufacturing challenges remaining. The aim of this review is to examine the current biological, and engineering challenges facing the biomanufacturing of antibodies and to identify the mitigations and emerging technologies that can be employed to overcome them. Developments in cell line engineering, intensified processing, continuous manufacturing, automation and innovations in process analytical technologies and single use technology will be discussed with regard to their ability to improve the current performance of mAb production processes.

1. Introduction

Therapeutic monoclonal antibodies, referred to as mAbs throughout this chapter, have emerged as the dominant player in the Pharmaceutical/Biopharmaceutical sector and are extremely effective agents for the treatment of cancer, inflammatory disorders and infectious diseases [1]. The effectiveness of mAbs as therapeutics is due primarily to their specificity in recognizing and binding to specific antigens through the antigen binding site. The efficacy of full length mAbs as anti-cancer agents is due to their ability to activate both complement-dependent cytotoxicity (CDC) and antigen-dependent cell cytotoxicity (ADCC) [2].

Worldwide, the mAbs market represents approximately 50% of the biotherapeutics market and according to Global Market Insights Inc., the global antibody therapy market is projected to reach $445 billion by 2028 which represents a 13.2% compound annual growth rate (CAGR) [3]. The first mAb, Muromonab-CD3 (proprietary name rather than brand name will be used throughout this chapter) was approved in 1986 and in June 2021 a landmark achievement was reported with the approval of the 100th mAb by the FDA [4]. As of May 2022 there are currently over 111 mAbs approved by either the European Medicine Agency (EMA) or the US Food & Drug Administration (FDA) [5]. In addition, there are over 15 mAbs pending confirmation of approval by either one or both regulatory agencies. Interestingly, several of the products pending approval are targeting the Covid-19 spike protein [6]. Over the same time period marketing authorisation for several products including Daclizumab, a humanized mAb for the treatment of multiple sclerosis, has been withdrawn and several products including the first approved mAb Muromonab-CD3 have been discontinued. Biosimilars are generic versions of biologics and thus far, over 20 mAb biosimilar products have been approved by the EMA or FDA [7]. Humira, a human antibody targeting tumor necrosis factor alpha used to treat rheumatoid arthritis and related disorders, is the top selling biopharmaceutical drug with a market value of $20 billion in 2021 [8]. Currently at least 8 biosimilar versions of Humira have been approved in various markets [7].

Immunoglobulin G (IgG) the dominant type of immunoglobulin manufactured is composed of two heavy (H) chains and two light (L) chains and has a molecular weight ranging from 140 to 160 kDa depending on the type of IgG subtype. Each of the light chains contain a variable (VL) and constant (CH) domain and the heavy chains contain one variable (VH) and 3 constant domains (CH1, CH2 and CH3). mAbs contain a number of glycosylation sites in the CH3 region. There is a huge variety in the type of mAb products approved with the major mAb formats being: canonical (full-length antibodies), antibody drug conjugates (ADCs) and antigen-binding fragments (Fabs). Figure 1A shows the distribution of the major types of mAbs currently approved by EMA or FDA [5].

Figure 1.

A: Pie chart shows number of different mAbs format approved B: Bar chart shows the number of products produced in a range of host cells.

Currently there are 92 canonical antibodies approved (83% of total) which are typically of the IgG1 subtype, have a molecular weight of ~150 kDa and are subdivided into chimeric, humanized or human antibodies. The majority of approved canonical mAbs target autoimmune diseases and many different types of cancers. Most canonical mAbs are monospecific with 3 bispecific antibodies (bsAbs) which recognize two different antigens (Emicizumab, Amivantamab and Faricimab) approved. ADCs are full-length antibodies with the addition of a highly toxic molecule usually attached to cysteine residues via a linker molecule. Upon arrival at specific cancer cells, targeted by the mAb component, the ADC is internalized by the cell and the toxin released through enzymatic cleavage [9]. ADCs represent highly specific and targeted therapies due to the specificity of the mAb component and currently 10 ADCs (9% of total) have been approved, of which two are targeting Her-2 positive breast cancer, ado-trastuzumab emtansine and trastuzumab deruxtecan [9]. Antibody fragments come in a variety of formats with molecular weights ranging from 12 to 50 kDa and contain a wide range of heavy and light variable and constant domains [10]. Currently, 4 Fab antibodies (4% of total) which contain the antigen binding site (i.e. one heavy and one light chain each with a variable and constant domain) have been approved. Fabs have a molecular weight between 40 and 50 kDa which improves tissue penetration due to their smaller size. The lack of the Fc domain means that these mAbs are not glycosylated so they can be produced in bacterial expression systems. There are 4 (4% of total) scFv products approved which are typically 25 kDa in molecular weight and are all produced in E.coli [10]. Interestingly, the scFv mAbs have the most diverse range of functionality: one is bispecific, one is a fusion-protein and the other is linked to an immunotoxin. The smallest of all therapeutic antibodies is the Nanobody with a molecular weight of 12 kDa. The first and currently only approved Nanobody is Caplacizumab which targets von Willebrand factor and is being used for the treatment of acquired thrombotic thrombocytopenic purpura [11].

Glycosylation plays an essential role in the biological efficacy of antibodies and is one of the critical quality attributes of mAbs [12]. N-linked glycosylation occurs in the Fc domain of the antibody. Canonical antibodies require glycosylation so production of full length mAbs is in mammalian cells, primarily CHO and murine myeloma cells (Figure 1B) as they have the biological capability to make these types of post-translational modifications (PTMs). CHO cells have been the workhorse in the production of antibody products over the last 36 years. CHO cells currently act as hosts for approximately 69% of all mAbs approved. Approximately 23% of approved products are produced in mouse myeloma cells lines with the predominant type being the non-secreting NS0 cell line followed by Sp2/0. Interestingly, the first mAb product approved Muromonab-CD3 was produced in a murine hybridoma cell line. A small number of Fab and scFv products that do not require glycosylation have been produced in bacterial cells, Escherichia Coli (E.coli). Thus far, just 1 product, Eptinezumab which is a humanized antibody has been produced in yeast cells Pichia pastoris [13]. Biomanufacturing of mAbs in CHO cells will be the focus of this chapter and looking at alternative expression systems is beyond the scope. Several comprehensive reviews of expression in bacterial, insect and yeast cells have been published [14, 15, 16].

Figure 2 shows a typical process flow diagram for the production of mAb drug substance, divided into upstream and downstream processing and formulation and fill-finish. Upstream processing encompasses the steps from vial thaw through inoculum and cell expansion with the aim of generating sufficient cells for the production stage bioreactor where the mAb is produced. Downstream processing encompasses all the operations from product capture, purification to formulation and fill finish.

Figure 2.

Process flow diagram of mAb biomanufacturing process. Typical fed-batch process duration indicated.

The biomanufacturing process is initiated with a vial of cryopreserved cells from the working cell bank (WCB) which is thawed into a small volume of media and then expanded through a series of shake flasks and bioreactors of increasing size, in order to generate sufficient cell numbers to inoculate into the production scale bioreactor. The final production vessel referred to as the N-stage bioreactor, is in the order of between 1000 and 20,000 L depending requirements. A fed-batch upstream production bioreactor step typically takes 10–20 days [17]. The production stage bioreactor duration can increase up to approximately 60 days if operated in perfusion [18]. The upper limit is determined by the validated Limit of In Vitro Cell Age (LIVCA) which is the number of generations for which the cell line has been demonstrated to be genetically stable.

The downstream process begins with recovery of the product from the bioreactor. Since CHO based systems secrete the product into the media, the product is usually harvested through centrifugation and/or depth filtration. Purification generally involves several chromatography steps for capture, intermediate purification and polishing. Affinity chromatography using Protein A which specifically binds to the Fc domain of the mAb is used to capture full length antibodies. Protein L can be used for purification of Fab fragments as it binds specifically to kappa light chains [19]. Anion and cation-exchange chromatography are typically used for intermediate purification and hydrophobic interaction chromatography can be used for final polishing. In addition to purification to >99.9%, the downstream process must include a viral removal and inactivation step before final formulation and finishing [19].

mAb processing methods have become the model for the production of both therapeutic proteins and emerging products like cell and gene therapy-based products as these have matured and been optimized over the last 36 years. However, with the increasing focus on speed-to-market and cost efficiency, it is necessary to continue to innovate and improve mAb biomanufacturing. The aim of this chapter is to examine the current challenges facing the industry and to discuss the mitigations and emerging technologies that have the potential to address them. This chapter is presented in 2 sections and will focus on the biological and engineering aspects associated with the manufacture of mAbs in CHO cells. The main challenges and mitigations within these themes are discussed and future directions and innovations for the biopharmaceutical industry are presented.

2. Biological and bioprocessing challenges and mitigations involved in biomanufacture of mAbs

Mammalian expression systems are chosen for full-length mAb production as they have the necessary cell machinery required to facilitate the correct protein-folding and glycosylation. Advances in genetic engineering and cell line development methods have been used to improve cell productivity and glycosylation control [20]. However, cell-based expression systems are challenging due to their inherent biological nature which results in variations in product yields and protein quality inconsistencies which can lead to challenges in manufacturing and increased production costs.

2.1 Production cell lines

CHO cell lines are the preferred host expression system in mAb production due to their capacity for complex human-like post translational modification, ease of genetic manipulation, lack of susceptibility to human viral infection and regulatory approval [21]. In terms of bioprocessing, CHO cells are robust and are easily adapted to grow in suspension in chemically defined and serum-free media making them readily amenable to process scale-up. CHO cell line development has generated cells with specific productivities (Qp) in the range of 5-20 pg/cell/day [22] and advances in single cell sequencing could increase this further [23]. Other mammalian cell lines like baby hamster kidney (BHK) cells and murine lymphoma cells lines such as NS0 and Sp2/0 also provide human-like glycosylation patterns but productivity levels can be significantly lower [24]. In addition, both NS0 and Sp2/0 have been shown to express two predominant glycan epitopes, galactose-alpha 1,3-Gal (alpha-gal) and N-glycolylneuraminic acid (Neu5Gc), which are immunogenic in humans and can lead to deleterious side effects [24].

2.2 Genetic engineering of cells

The first step in developing a biomanufacturing process is to generate a cell line producing the desired product, referred to as a ‘production cell line’. Generation of production cell lines involves many steps as outlined in Figure 3 and includes transfection of the gene of interest into the cells, followed by selection and screening to generate the optimal clonal production cell line for manufacturing. Once a production cell line has been generated there are regulatory guidelines in place governing the use of cells in production processes (Q5D ICH, 1998) that must be followed. The most common industrial approach for cell line development is transfection of the gene of interest using a non-targeted plasmid delivery system that contains a selection marker (Section 2.3) to enable the selection of stably transfected cells. Ideally, a high-quality production cell line should demonstrate high stability, scalability and high titer levels, in order to provide reproducible results with consistent product quality attributes.

Figure 3.

Cell line development process showing the steps required for the generation of production cell lines. Created with BioRender.com.

CHO cell lines used commercially were derived from the original spontaneous immortalized culture established in 1956 by Dr. Theodore Puck [25]. Several variants referred to as subclones were generated using both chemical and radiation mutagenesis. The most frequently used variants are CHO-K1 (single copy of dihydrofolate reductase (DHFR)), CHO-DG44 (both copies of DHFR missing) and CHO-DUBX [26]. The plasmids typically used in the transfection of the genes encoding the product are not designed for integration at specific chromosomal sites so this step generates a population or pool of cells with a wide range of expression levels that reflects the gene copy number and the transcriptional activity of the locus based on the integration site [27]. The transfection efficiency can range between 15 and 80% depending on the system used [28]. This means that if 1 million cells were transfected, potentially between 10,000 and 100,000 potential new cell lines can be generated. This random integration can provide a diverse range of expression levels but can also result in integration into unstable areas of the genome [29], resulting in varying levels of expression and lack of genetic stability.

Although the original CHO cell line contained 22 chromosomes, the subclones used commercially contain a range of chromosome numbers, e.g. CHO-K1 typically contains between 18 and 21 chromosomes [30]. CHO cells are inherently genetically unstable which has resulted in a genetically and phenotypically diverse lineage, manifested by many single nucleotide polymorphisms (SNPs), copy number variations, and karyotypes [31]. A recent review by Wurm and Wurm described CHO cell lines as having a “quasi species” genome [32]. Studies have shown that over the course of passaging of CHO cells, the DNA is unstable. The notion that CHO-K1 cells were truly clonal in origin has also been questioned [32].

This random integration of plasmid DNA approach has remained relatively unchanged over the last 36 years but the availability of advanced gene-editing tools now makes transfection at targeted chromosomal sites more feasible. These include CRISPR/Cas9, zinc finger nucleases (ZFNs), and transcription activator-like effector nucleases (TALENs), with alongside RNAi (RNA interference) and ribozyme engineering which facilitates “multiplexing gene-editing approach[es]” allowing precision and acceleration of genomic rearrangements for enhanced generation of successful clones with improved product yield and quality [33]. It is expected that future cell line development will include vectors targeted to stable areas of the CHO genome and will also be capable of expressing not only the transgene for the product required but additional genes involved in regulatory cell processes like apoptosis [34].

2.3 Selection, cloning and screening

In order to identify the transfected cells that have stably integrated the plasmid into the host genome, a selective marker is used for selection of cells which have acquired the highest expression of the associated product gene. First generation products were produced in CHO-K1 and CHO-DG44 cells that incorporated DHFR as the selection marker on the plasmid [35]. Transfected cells could be selected in media that did not contain nucleotides required for growth in the absence of DHFR activity. One advantage of the DHFR selection system was that gene expression and ultimately product production could be amplified using methotrexate (MTX). Antibiotic resistant genes like neomycin (selected with geneticin), are frequently used. The use of glutamine synthetase (GS) enzyme as a selection marker allows selection of cells growing in media without glutamine. The GS system has been successfully employed for the production of several mAbs. CHO cells produce low levels of GS activity and require selection in methionine sulfoximine (MSX). Interestingly, several studies have shown that amplification methods using DHFR and methotrexate may be susceptible to instability of the transgene that results in a decrease of recombinant gene copies in long-term culture [36]. For this reason, the MSX-GS selection system tends to be the most widely used alternative. NS0 cells do not express endogenous GS therefore can be selected without using MSX, thereby creating a simpler selection process [37].

The guidelines for using cells to manufacture products must be strictly adhered to and a key regulatory requirement is that production cells must originate from a single clone i.e. individual cell. Following the selection process, cells must be separated or cloned into single cells using limited dilution in suspension cultures or clonal rings if cells have been adapted to grow as adherent cultures. Single cell limited dilution relies strictly on a probabilistic approach which is time consuming and offers low throughput as a process platform. The use of plasmids incorporating green fluorescent protein (GFP) and the application of fluorescence-activated cell sorting (FACS) has helped simplify this process. Following transfection, GFP positive cells can be selected as pools of transfected cells and can be automatically separated into individual cells using single-cell isolation technologies such as FACS, magnetic activated cell sorting (MACS), microfluidic and manual cell picking [38]. Advancements in automated screening and selection systems such as Clonepix and CellCelector offer multitude cell screening through immobilization of cells within a semi-solid media matrix and relies with a fluorescent identifier in the vicinity of a resulting colony [39]. Additionally, these automated cloning systems are combined with imaging analysis to provide compelling visual evaluation that can be used in real-time.

Selection and cloning generate a large number of individual candidate cell lines that require screening and evaluation to select the optimal production cell line which is challenging with regard to throughput and creates a major bottleneck in upstream process development. A key requirement is a high specific productivity level and evaluation of mAb production by ELISA or ELISA spot assay is the first step in the screening process. A typical screening strategy will involve multiple evaluation stages in which a proportion of cell candidates are discarded after each stage to attain a small number of highly ranked possible cell line candidates [40]. A typical screening scenario involves initial screening of ~300 individual clones for productivity. Following elimination of low producing cell lines, the candidate cell line panel of ~50 is screened for proliferation capacity. Cells with high productivity (2-7 g/L) and specific growth rates (0.010-035 h-1) are then expanded into shake flasks and screened for their ability to adapt to production conditions [41]. The final step will generally involve ~5–10 cells lines grown under different bioreactor conditions in order to finally select the optimal cell line for commercial production. The introduction of miniature bioreactors such as the Ambr system allows for significant scale-down of this step and increased through-put significantly accelerating the final evaluation step.

2.4 Cell line characterization

A key requirement and challenge for manufacturing is to maintain optimal productivity and consistent quality between batches of product. As part of the regulatory requirements, a master cell bank (MCB) must be created from the initial cell line generated. A WCB is generated from the MCB and a new vial of cells is used to initiate each batch. As mentioned previously there are specific guidelines that must be followed when working with cell lines and these guidelines include specific tests that must be undertaken to characterize and authenticate cells within the MCB and WCB to ensure purity, identity and stability of the cells. In terms of purity the lack of microbial or viral contamination must be confirmed. The identity of the cells can be confirmed using STR profiling. Maintaining the genetic stability of the cells for more than 60 generations at high cell density must be demonstrated [42].

2.5 Media selection and optimization

Cell culture media are complex mixtures of nutrients, hormones, growth factors, salts, trace elements and buffers. Early media formulations used serum but regulatory concerns about possible prion and viral contamination led to the development of serum-free, protein-free and chemically defined media formulations that were free from animal-derived material. Serum-free media is widely used in CHO biomanufacturing and the basal media consists of 50–100 components. The optimal composition and amounts of these components must be determined to be capable of supporting cell growth and production of product with acceptable quality attributes. Overall, media development is an multiple process with iterative rounds of performance testing. Spent media analysis allows the development of feeding strategies for cells in fed-batch and nutrients like glucose and specific amino acids are routinely added to extend the lifetime and productivity of cells in culture [43]. Design of Experiments (DoE) has been used for development of cell culture media to evaluate both component concentrations and component interaction [44]. Many companies use a platform based approach and use the same media formulation for several different products.

Bioreactor systems like the Ambr systems have benefitted the media development process as it allows for much greater throughput, smaller volumes and less labour-intensive experimentation. Dynamic flux balance analysis (DFBA) is a new approach which elucidates the relationship between media supplementation with amino acids, feeding strategy, increased product yield, an extended growth phase and increased density [44]. DFBA illustrates that metabolic state varies more at the beginning of the culture and rather less in the middle of culture [45]. Other advancements are towards multiplexed at-line and operator-independent analytics in benchtop bioreactors such as use of multi-analyte analyzers including NIR, Raman and 2D-fluorescence spectrometry which provide useful process measurements for feed media optimization [44, 45].

3. Engineering challenges and mitigations

Commercial manufacture of mAbs has evolved significantly since the first approval almost 40 years ago. The unit operations and technology routinely used are mature and well-understood. However, the landscape and business considerations associated with their production continue to evolve and drive new innovations and approaches. Increased expectations on the part of regulatory authorities, a burgeoning biosimilars market, ever-growing numbers of approved mAbs and wide ranging volume requirements have resulted in new challenges that must be addressed in order to successfully meet the patients’ needs while maintaining a viable business. Efforts typically focus on accelerating the time to market and reducing CoGs. There are several engineering-related challenges to achieving this, namely: process variability, sub-optimal volumetric productivity, long cycle-times, and the complexity of managing multi-product facilities [46, 47]. In this section, a number of technologies and strategies capable of addressing these challenges are presented alongside the opportunities that they provide the industry with as it moves towards the paradigm of Industry 4.0.

3.1 Bioprocessing considerations

The challenges facing mAb production include producing the product with a tight product quality profile, maintaining the biophysical properties of protein and reducing product and process-related impurities below the acceptable levels to meet quality specifications and ensure safety, efficacy and stability [48]. The first step is to produce a product of acceptable quality within the N stage bioreactor. Many of the critical quality attributes (CQAs) including glycosylation profile can be affected by operating conditions within the bioreactor including pH, temperature, dissolved oxygen (DO), media formulation and metabolites. Real time monitoring and control of CQAs is still aspirational and currently these attributes are at best managed through a QbD-based approach and automated control of basic parameters such as temperature, pH and DO as well as procedural controls such as raw material specifications which will be further discussed in later sections.

Impurities typically fall into two categories: product-related and process-related. Product-related impurities occur where the biophysical properties of the product are compromised and a portion of it degraded. This can occur by multiple mechanisms including aggregation, oxidation, fragmentation, deamidation and denaturation. These degradative mechanisms are the result of the conditions to which the protein is exposed such as high mechanical or physical stress due to agitation or gassing in vessels or fluid flow through piping, filtration and chromatography skids. Enzymatic reactions during cell cultivation can result in cleavage of the protein backbone and the generation of mAb fragments [49]. Aggregation where protein is converted from the desired monomers to dimers, trimers or even larger structures is one of the most common product-related impurities that results from physical stress [50]. Chemical degradation occurs predominantly during downstream processing as a result of changes in pH or buffer solutions. This generally occurs during the viral clearance, chromatography and ultrafiltration/diafiltration (UF/DF) steps. Mitigation strategies are focussed on reducing the residence time at low pH conditions and gentle pH shifts in UF/DF can circumvent the inherent critical effects of these processes in the final mAb product [51].

Process-related impurities are introduced as a result of the materials used in the process and include growth selection agents, surfactants, purification column binding agents and viral inactivation agents. Cell lysis within the bioreactor and harvest equipment results in the release of host cell proteins (HCP) and DNA. The final product must contain less than 100 pg of cellular DNA per dose, and ppm or ng quantities of HCP per mg of antibody product [52]. Conventionally, intact cellular components are removed during the harvest step by centrifugation and depth filtration, but additional purification steps are required to remove contaminants resulting from cell lysis. Whilst Protein A affinity chromatography is the gold standard for the capture step in the mAb purification process, it can leach from the column matrix and contaminate the product and so must be removed at a later stage with additional chromatography steps like ion exchange (IEX) and hydrophobic interaction to levels less than 1 ppm or 1 ng Protein A per mg product [52].

Effective purification is challenging due to the similarities between many of the contaminants and the product of interest in conjunction with the extremely low acceptable levels of contaminants, due to the parenteral nature of mAb products. Therefore, the best approach is to minimize the generation or introduction of these species. QbD, process monitoring and control and process modeling discussed in Section 3.2 and 3.3 can be leveraged to support this goal.

3.2 Quality-by-design (QbD), process monitoring and control

Much greater process variability is observed for mAb biomanufacturing processes as compared to traditional small molecule pharmaceutical processes. The main sources of variability fall into four main categories: biological factors, raw materials and consumables, operational inputs (measurements, methods, personnel and equipment) and environmental conditions [53]. Ultimately, the result of this is variable productivity and product quality, both of which must be controlled and optimized in order to minimize the CoGs. Additionally, from a regulatory perspective, the FDA’s guidance document on process validation recommends that manufacturers understand and control the sources of variation [54]. Understanding and reducing variability can be achieved by effective application of QbD, process monitoring and control.

Traditionally, biomanufacturing companies have maintained that “the process is the product” or the quality-by-testing (QbT) approach. Under QbT, the product quality attributes are empirically linked with the manufacturing process and its inputs, both materials and process parameters, during the clinical phases of development, with little or no mechanistic understanding. One of the many disadvantages to this approach is that Proven Acceptable Ranges (PARs) for process parameters are extremely narrow, limiting both the opportunity for post-approval optimization and the capacity for the process to be adjusted in response to process variability [55].

QbD overcomes the limitations of QbT by taking a science and risk-based approach to drug development in order to ensure process and product understanding and to implement effective control strategies. According to the European Medicine Agency (EMA): “Quality by design (QbD) is an approach that aims to ensure the quality of medicines by employing statistical, analytical and risk-management procedures in the design, development and manufacturing of medicines”. Summarily, this methodology is focused on the 1) identification of each source of process variability, 2) understanding of their effects on product’s critical quality attributes and 3) control strategies by applying statistical inferences, such as multivariate analysis. Additionally, according to EMA, process analytical technology (PAT) is commonly part of QbD methodologies and it is defined as the system of integrated technologies and methods for control of critical quality attributes of raw and in-process materials [56]. An excellent overview of the QbD approach is provided by Yu et al. [57].

The result of using QbD during drug development is a strong understanding of the link between process inputs and product quality within the design space mapped. The benefits of this are manifold and include:

• Product defects, rejections and recalls are reduced due to an increase in process capability and decrease in process variability.

• The increased process understanding allows for enhanced root cause analysis, resulting in reduced deviations and batch failures.

• Post-approval process optimization is allowed to occur without regulatory notification if it lies within the design space filed.

Once a mAb moves from development into production, it is necessary to implement a control strategy. There are three levels of control possible [57]:

• Level 1 uses automated control in order to maintain the CQAs of the product at the desired value in real-time. Raw material attributes and process conditions are monitored and the process parameters are automatically adjusted in order to control the CQAs. Level 1 represents a highly adaptive form of control which requires process analytical technology (PAT), process models and advanced control algorithms to be implemented and can enable real time release (RTR).

• Level 2 is a mixture of automated and procedural controls which leverage the QbD design space in order to reduce end-product testing and allow for more flexibility in raw materials and process parameters.

• Level 3 is the traditional QbT-based control strategy where raw materials and process parameters are tightly controlled and there is extensive end-product testing.

Moving from Level 3 to 1 is desirable, due to regulatory pressure and the business benefit of reducing the CoGs through reduced variability, batch failures, process optimization and the removal of the time and cost associated with product quality testing. In order to reach Level 1, it is essential to be able to quantify both the control targets and the associated process responses. This may be done by direct measurement or with the use of predictive models coupled with indirect measurements.

Currently, a few variables (e.g. DO, pH and temperature) are measured routinely with in-line probes, however, most key process parameters (e.g. cell density and viability etc.) are measured off-line and the delay between sample extraction and analytical results can hinder the process productivity and the ability to implement adaptive closed loop control [58]. Much work has been done to address the gap in process measurement through the development of various sensors.

Sensors can be classified by their principle and their structure:

• Structure - hard vs. soft sensors: A hard sensor is based on equipment capable of direct measurement of the required attribute whilst a soft sensor is a combination of equipment and a mathematical or data-driven model which infers the measurement required [58]. For instance, Narayanan et al. [59] investigated the use of a hybrid model coupled with a Kalman filter (EKF) for real-time monitoring and control of mammalian cell culture processing. The hybrid model coupled with EKF was applied over a data set with 15% added Gaussian noise and the approach improved predictive accuracy for process variables (e.g. cell density, titer, lactate concentration and glucose processes. concentration) by about 35%, when compared to partial least square (PLS) algorithm.

• Principle - refers to the scientific principle on which the measurement is based. For instance, biosensors (ligand base), sensors based on optical absorption (e.g. spectroscopy – UV/VIS, fluorescence, infrared, near-infrared, mid-infrared), sensors based on light scattering and optical density, etc. [60]. As an example, the work of Whelan et al. [61] highlighted the application of Raman spectroscopy for the measurement of cell density and metabolite concentration in a CHO production bioreactor. They found an acceptable agreement between the in-line Raman measurement and off-line samples across different process scales.

The availability of real time process measurements is a prerequisite for the implementation of automated closed-loop feedback control which can adapt and respond to process variability in order to maintain a steady CQA profile in the final product. Automated control also reduces the risk of human error in the process. There have been reports in the literature demonstrating the capability and benefit of such control. Craven et al. [62] applied a nonlinear model predictive controller NMPC to a 15 L CHO fed-batch bioreactor to control glucose concentration at a defined set-point (11 mM) by adjusting the feed rate to the bioreactor. The substrates (glucose and glutamine) and byproducts (lactate and ammonia) were measured by in-situ Raman spectroscopy and the concentration values were determined by a partial least squares (PLS) calibration model analysis. The determined metabolite concentrations were inputted to the NMPC algorithm which used a first principles mechanistic process model in conjunction with an optimization algorithm to determine the optimal control output or feed rate. Both simulated and real-time application of the NMPC showed similar performance and the results highlighted the feasibility and capability of NMPC for bioreactor control. While it was shown that process model inaccuracy could hinder NMPC performance, the controller showed good ability to function with both noisy and non-continuous process measurements. While there are multiple reports of the benefits of this type of control in literature, currently, it is not implemented in commercial manufacture due to the conservative nature of the biopharmaceutical industry. It is however routinely employed in other sectors such as commodity chemicals and petrochemicals e.g. Quin & Badgell [63] reported 93 industrial applications of NMPC due to the improved safety, quality and efficiency that it enables [64]. As such, it presents a future opportunity for mAb production.

In summary, innovations in and implementation of QbD, process monitoring and control in mAb biomanufacturing can address the key challenge of reduction of CoGs by increasing process capability and robustness, reducing process variability and failures and facilitating post-approval optimization.

3.3 Process modeling

Process modeling or process simulation can be applied in order to address the challenges of speed-to-market and the reduction in CoGs. There are numerous types of models that may be deployed depending on the objective. These range from mechanistic to data-driven models with reports of hybrid model approaches for bioprocesses increasing in the literature [65, 66].

Mechanistic models, also known as mathematical, first principle or white-box models depend on the laws of nature to describe a specific phenomenon. They require fundamental understanding of the phenomenon being studied. This is often difficult for bioprocesses due to the complexity of the system and the strong level of interaction and dependency between multiple parameters. They have high extrapolation capacity but are limited by the degree of accuracy of the model equations describing the behavior. Conversely, data-driven or black-box models rely on larger data sets and greater computational efforts in order to predict process responses, with no reference to the underlying mechanisms. This limits the ability to extrapolate predictions to unseen scenarios but significantly reduces the complexity of the model [65]. Hybrid models have elements of both mechanistic and data-driven models, circumventing the challenges of both by reducing the amount of data and level of process knowledge required. It should be noted that regardless of the type of model used, it is crucial to verify and validate the model outputs using independent or unseen datasets before using them.

Process models can be used to accelerate process development. Typically bioprocess development has a heavy reliance on experimentation under a traditional Design of Experiment (DoE) framework, augmented with prior expert knowledge. Moller et al. [67] reported the use of model-assisted DoE for bioprocess development. A mathematical model was developed that described cell growth, metabolism and antibody production for a CHO DP-12 cell line under both a batch and fed-batch mode of operation. The model was then used to reduce the boundary values for the experimental DoE. It was found that the same optimal conditions were identified for both the traditional and model-assisted approaches with a reduction in the number of experiments required from 16 to 4 in the case of batch and 29 to 4 in the case of fed-batch. Given the time-consuming, expensive nature of bioprocess experiments, this represents a significant potential to accelerate development timelines.

A key enabler for commercial production is the scale up of the bioprocess from small laboratory scale equipment sets through an intermediate scale required to supply the clinical trials to large scale commercial manufacturing plants. As discussed in Section 3.1, both the cell and the mAb product can be affected by factors such as shear stresses, extremes of temperature and pH among other things. As an example, as the production scale increases, it becomes more difficult to maintain a fully homogenous environment and so a balance needs to be identified and maintained between effective mixing and exposure of the cell and product to damaging conditions. This may be achieved empirically, however, the conditions identified by such means are potentially sub-optimal and the opportunity to develop fundamental process understanding for use subsequently for troubleshooting and tech transfer is missed. As an alternative, computational fluid dynamics (CFD) can be used to support and enhance process scale up.

CFD mathematically models fluid flow and its interactions with solid bodies by numerically solving systems of partial differential equations governing fluid dynamics problems (e.g. Navier-Stokes equations) [68, 69]. CFD can be used to derisk and support process scale up by predicting the conditions created under a range of agitation and aeration rates. For example, Mishra et al. [70] studied the effect of agitation and aeration rates on the mass transfer of oxygen and shear stress in the liquid phase of a 10 L single-use bioreactor. Their approach combined computational fluid dynamics with species transport and population balance models in order to predict the maximum total stress and energy dissipation rates that the cell culture would be exposed to. The simulations were performed for stirring speeds between 50 and 300 rpm and aeration rates between 0.1 and 0.2 LPM, considering 45% fill volume. The results indicated a maximum total stress of 34.17 Pa and energy dissipation rate of 1.352 m²/s³ (at 300 rpm and 0.2 LPM), which are unlikely to affect mammalian cells according to the author’s consulted literature. The simulations were validated by experimental determination of the oxygen’s mass transfer coefficient (stirring speed range: 50 rpm - 200 rpm, 0.2 LPM and 6.75 L of liquid volume) and the observed errors (predicted vs experimental) were between 0.47% and 10%.

Process modeling can support a reduction in CoGs via a number of approaches. One such approach is the use of multivariate data analysis to generate a “golden batch trajectory”, a form of data-driven process model, against which real time production plant data is compared. The process insight gained can be used to rapidly identify root causes for batch failures and deviations and to prevent them reoccurring, hence reducing the rate of batch failures and process variability experienced. The knowledge gained can also be used to fine-tune the Normal Operating Ranges (NORs) for a process within its design space and maximize the productivity or particular CQA of the product. Sokolov et al. [71] applied multivariate analysis in the form of partial least square regression - PLSR to predict mAb-based product quality attributes (aggregates, fragments, charge variants and glycan profile) from process data (media supplements, pH and temperature shifts). The data set was obtained from a 91 run DoE at milliliter scale and the model performance was evaluated using the root mean square error in cross-validation (RMSECV). They firstly used principal component analysis (PCA) to analyze the correlation among 14 product quality attributes (QAs) and, since their findings indicated a strong correlation among QAs, they concluded that the variables should be treated as one characteristic. Additionally, PLSR1 and PLSR2 were used to predict product quality attributes and provided comparable prediction accuracy. The PLSR2 models were further investigated by the addition of genetic algorithm (GA), in which the results became more accurate and, in complex cases, the GA was able to remove noise, inconsistency and redundancy in the data set. Availability of such models can also enable statistical model predictive control the benefits of which were be discussed further in Section 3.2.

Digital twins are another approach to supporting a reduction in CoGs. Essentially, a digital twin is a digital representation of a physical process. The application is similar to that of the golden batch trajectory, however, digital twins are typically hybrid process models which may leverage PAT and other process monitoring (see Section 3.2). They are used to identify process bottlenecks, key engineering targets and identify operational strategies that improve the reliability and productivity of their physical twin. As an example, a digital twin of a production bioreactor could represent the physiology and metabolism of the cell culture by applying genome-scale metabolic models (e.g. Flux Balance Analysis - FBA, with appropriate objective function and adequate constraints) alongside the process kinetics obtained from in-line monitoring (e.g. Raman-based monitoring system) [72, 73]. Digital twins are key enablers of Industry 4.0, which seeks to revolutionize how industry operates through the use of smart, autonomous systems running on data and machine learning and have huge potential to improve the biomanufacturing of mAbs [66].

3.4 Process intensification

An intensified process can be defined as one that increases productivity (e.g. per batch, per facility) and/or reduces environmental impact (energy, waste, materials), facility footprint (smaller equipment, shorter process streams), manufacturing costs, process times, or process bottlenecks. It should be noted that having a genetically stable cell line with good growth and productivity characteristics and media systems capable of supporting the increased nutrient demands are prerequisites for upstream process intensification.

The pre-production stage encompasses all process steps before the production bioreactor, i.e. cell revival from the working cell bank (WCB) and the inoculum and seed train (Figure 2). These steps represent a significant portion of the process cycle time and typically require highly skilled labour and expensive equipment and infrastructure in order to be reliably executed. Two main approaches to circumventing the time and resources required at this stage have been reported in the literature, namely the modification of the cell banking approach to provide a greater number of cells upon thaw by either increasing volume or cell density or a combination of both and the use of the perfusion mode of operation at the N-1 seed bioreactor stage.

Intensification of cell banking eliminates multiple expansion steps which reduces the resources and time required to inoculate the production reactor, an example of which is described by the work of Seth et al. [74]. Their work investigated the cryopreservation of CHO cells at low density (LD, ~30x10⁶ cell/mL), mid density (MD, ~70x10⁶ cells/mL) and high density (HD, ~110x10⁶ cells/mL) in single-use cryopreservation bags. They named the strategy the Frozen Accelerated Seed Train for Execution of a Campaign (FASTEC) which allowed the seed train to be bypassed. Seth et al. [74] evaluated the FASTEC approach in an 80 L seed bioreactor by comparing two processes: i) an inoculum seed train with 3 passages; ii) an inoculum seed train with only 1 passage. The processes were able to produce final cell densities of approximately 7x10⁶ viable cells/mL in 10 days and 4 days, respectively. Additionally, after inoculating the production bioreactor (400 L) operated in fed-batch, product titres (1.1 g/L & 1.2 g/L), triple-light chain impurities (3LC, 1.7% & 2.6%) and aggregates (3.1% & 2.2%) were comparable to the seed train control (1.1 ± 0.2 g/L; 3.7 ± 1.0%; 4.3 ± 1.0%, respectively). Seth et al. [74] concluded that, despite cells displaying lower growth rate and viability immediately post-inoculation, the FASTEC process was able to produce comparable titer and quality of mAbs than the standard process whilst significantly reducing the duration of the upstream process.

The second strategy commonly reported in the literature is the use of a perfusion N-1 seed bioreactor. The benefit of this approach is that a significantly higher cell density is achieved in the seed train which allows a fed-batch production reactor to be inoculated at a higher initial cell density. As a result the duration of the production reactor can be reduced by a number of days and/or higher titres can be achieved due to the increased cell time within the reactor. Xu et al. [75] applied an N-1 perfusion seed step in conjunction with media enrichment to a CHO fed-batch production reactor for four processes producing four different mAb products. They reported that increasing the initial seed density from 0.3–1.2 x 10⁶ cells/ml to 10–20 x 10⁶ cell/ml resulted in an up to 10 fold increase in titer in addition to up to a 4 day decrease in production reactor duration while maintaining product quality.

The benefits of continuous processing have been discussed extensively in the literature for a range of industries including the small molecule pharmaceutical sector [76]. They are generally smaller, faster and cheaper with greater levels of flexibility and quality assurance achievable. As such there is much interest in the application to mAb production. However, fully integrated end-to-end continuous processes in this space are not yet a reality in commercial manufacture although research and development of such systems is ongoing and has demonstrated feasibility [77]. As a first step, continuous unit operations, primarily perfusion bioreactor steps and continuous chromatography have been implemented.

Many authors have discussed the advantages and disadvantages of perfusion and fed-batch operations for mAb production. The discussion is complex as the best choice is dependent on the specific scenario i.e. the quantity of material required, the productivity of the cell line, the nature of the protein, the cycle times of the processes etc. Direct comparison is often hampered by differing scales and definitions of productivity [78]. For instance, Lee et al. [79] studied the change from perfusion to fed-batch for expression of biosimilar monoclonal antibody A (CR-mAb-A) by recombinant Sp2/0 mouse myeloma cells. The fed-batch operation was evaluated in 8 runs at the following bioreactor scales: 3 L (3×), 100 L (2×) and 12,500 L (3×). The results indicated that, although the perfusion mode provided higher volumetric productivity, the fed-batch operation showed increased total productivity (7.5 fold increase) due to its higher volume capacity. Lee et al. [79] also investigated the mAb-based product quality by measuring oligosaccharide profiles and charge variants of mAbs expressed by fed-batch and perfusion. They observed slight differences in heavy chain glycoforms (G0, G1 and G2) between fed-batch and perfusion, while different scales of fed-batch provided comparable proportions. Additionally, by performing capillary electrophoresis sodium dodecyl sulfate (CE-SDS), they observed a slightly lower amount of intact IgG (4%, normalized) obtained in fed-batch in comparison to perfusion.

It is important to highlight that, in order to benefit from upstream intensification, one must guarantee that downstream operations are capable of handling high volumes and titres while maintaining the product quality. To this end, efforts to improve resin capture and capacity in chromatography operations have been documented in addition to continuous processing. Gerstweiler et al. [80] highlighted applications to enable continuous processing, such as: periodic countercurrent chromatography (PCC), simulated moving bed chromatography (SMB), continuous flow-through chromatography and multi-column designs (e.g. continuous multicolumn countercurrent solvent gradient purification). Moreover, in regard to Protein A ligand-based columns, Somasundaram et al. [81] stated that the dynamic binding capacity and resin reusability are important aspects to be considered in continuous processing.

3.5 Platform processes

Producers of mAbs typically have platform production processes that span the majority of their portfolio. A platform process comprises the expression system, typically a suspension CHO cell line, the associated basal and feed media formulations and the series of unit operations used to produce and purify the mAb. The platform may be fine-tuned for each product, for example, the media formulation may be slightly modified for the particular nutrient requirements of a specific clone in order to boost productivity or ensure product quality, but there are no major fundamental changes unless essential for a given product. The benefits of utilizing a platform approach are: faster process development, easier facility fit for scale up and tech transfer, greater process knowledge and more robust processes, all of which contribute to delivering a faster time to market and lower CoGs.

The majority of manufacturers rely on a CHO fed-batch upstream process platform with serum-free media although a significant number have opted for perfusion. There is also a general move away from using undefined media components such as hydrolysates to fully chemically defined media formulations in order to reduce raw material and hence, process variability. The downstream process generally follows a platform process flow of one or two harvest steps (centrifugation/depth filtration) followed by a Protein-A chromatography step for product capture, a low pH hold for viral inactivation, two polishing chromatography steps, a virus filtration and a final ultrafiltration/diafiltration step to the final concentration in the formulation buffer [82].

Efforts to improve the existing platform processes focus on process intensification strategies such as high cell density cell banks and N-1 perfusion steps as discussed in Section 3.4 as well as the potential to replace the Protein-A capture step. Protein-A is the single biggest contributor to the Op-Ex costs associated with mAb production. It is widely used as it is extremely effective and reliable. Both chromatographic and non-chromatographic options have been explored. These include: precipitation, crystallization, cation exchange chromatography and multimodal chromatography [82, 83]. Other approaches seek to improve the efficiency of Protein-A usage. Typically, Protein-A is used in a single product, packed bed format. There is work currently underway exploring resin use across multiple products, particularly useful for small volume products, as well as alternative formats such as membrane chromatography and monolithic chromatography which allow for higher flow rates and hence, throughput [84].

3.6 Facility design and single use technology (SUT)

The portfolio of mAb products currently approved ranges from large volume blockbuster therapies to low volume products that address orphan indications. Therefore, the scale of commercial manufacture varies considerably across products. For a given product, the annual requirement can also vary significantly. It may increase as a new product gains market share or is approved for additional indications or may decrease if a competing product, a new alternative therapy or a biosimilar version of the product in question, is launched. Biosimilar manufacturers target high volume products coming off patent, supplying a comparable product at a reduced cost. This results in a significant drop in the volume required for the original product, as observed for granulocyte colony stimulating factor (G-CSF). According to an IMS report [85], in 2016 within the EU, biosimilars accounted for 88% of the market as compared with the reference product. This resulted in a 37% reduction in price as compared with the year prior to entry of the biosimilars into the market.

As a result, where once a dedicated high volume production plant using stainless steel equipment was the norm, there has been a shift towards flexible multi-product facilities in order to accommodate the ever-changing numbers and volumes of mAbs to be supplied to meet patient needs. SUT and modular or ballroom style facilities help to satisfy these requirements.

In SUT, all surfaces that come in contact with the process are disposable and are replaced after a single batch. This includes the vessel itself which is typically a bag supported externally by a metal exoskeleton and fabricated from FDA (Food and Drug Administration) approved polymers such as polyethylene (PE), polytetrafluoroethylene (PTFE) and polypropylene (PP) supplemented with additives to enhance performance and/or extend useable life as well as impellers, probes, resins, filter cartridges etc. SUT eliminates the need for the validation requirements associated with cleaning and sterilization of equipment, reduces the turn-around time between batches and reduces the risk of both microbial contamination and cross-product contamination in multi-product facilities [86]. The cap-ex investment required to establish a single use facility is significantly lower than the stainless steel equivalent and the utility requirements particularly for steam and water for injection (WFI) are massively decreased [87]. Studies have also reported that despite the increased plastic waste produced from SUT, overall they are environmentally less impactful than stainless steel [87]. Currently, there are commercially marketed SUT solutions for each unit operation typically used to produce a mAb [86].

There are however some disadvantages and challenges that remain to be overcome. Firstly, leachables and extractables are a concern due to the material of construction. These substances may be detrimental to process performance and/or human health [86]. Typically this is overcome by performing studies on the material to prove suitability. Secondly, scale is limited. There are mechanical challenges in producing what are essentially plastic bags with sufficient strength to withstand the loads associated with large volumes. The largest volume routinely seen at commercial scale is 2000 L as opposed to 25,000 L in stainless steel [86, 88, 89]. There is one system currently on the market at 6000 L made by ABEC [90]. Other challenges include extremely long lead times of 9–12 months currently for the sterile consumables required in addition to high op-ex costs associated with them. The SUT available for downstream is less mature than in the upstream space and as such is less likely to be adopted. This is evolving over time especially when considering the increase of demand by industry. According to American Pharmaceutical Review [91], 46.9% of a survey’s respondents (12th Annual Report and Survey of Biopharmaceutical Manufacturing) had claimed to investigate single-use technologies in downstream bioprocessing to improve purification operations, in contrast to 36.8% in 2012. Despite these challenges however, many manufacturers have adopted either fully SU equipment trains or a hybrid approach where the upstream process up to a volume of 2000 L is SU and the remainder of the upstream and the downstream processes are stainless steel.

Facility design and construction takes an average of 1 year to design and 3–4 years to build and costs several hundred million to over one billion euros depending on the size of the facility. To maintain strategic relevance in the current market, biopharmaceutical companies must design these facilities to be flexible and multi-product while still maintaining a high standard of product safety and efficacy. Modified-ballroom or dance-floor facility design integrated with closed systems and SUT is the most common approach to achieve these objectives while managing the risks associated with large integrated production spaces. In this type of design, a series of rooms that meet the Clean-Not-Classified (CNC) criteria are interconnected through wall panels. Within each room, single use, closed systems are operated. This equipment can be based on modular skids that can be changed if requirements change in the future. This approach reduces the footprint of the facility by removing the need for personnel and material airlocks to a large extent as well as decreasing the both cap-ex and op-ex costs associated with graded cleanroom environments. Time for construction is also reduced. It eliminates the need for gowning and simplifies installation, maintenance and operation of equipment as there are less restrictions on activities in the production space [92].

4. Conclusion

mAbs represent the largest category of biopharmaceuticals on the market and the number of approved products continues to grow. The commercial production of these products has evolved significantly since the first approval in 1986. Expression systems, once only capable of titres in the mg/L range, are now routinely producing 5–10 g/L due to advances in cell line development. Efforts have moved from establishing reliable robust platform processes to optimization of product quantity and quality. New technologies and approaches are being adopted in order to achieve this despite the mAb specific challenges associated with processing protein molecules and controlling biological processes. QbD, process monitoring, and control can be harnessed to manage the inherent variability associated with the raw materials and biological expression system. Process modeling, particularly hybrid models, can mitigate the expensive, time-consuming nature of experimental approaches and the empirical approach taken to process development historically in order to accelerate time to market and optimize and troubleshoot the manufacturing process. There are currently multiple strategies for process intensification being adopted in order to reduce cycle time and increase productivity. New approaches to facility design coupled with SUT reaching greater levels of maturity have reduced the risk and complexity associated with multi-product facilities. Alternative technologies on the horizon such as greater offerings for SUT in the downstream space and cheaper alternatives to Protein-A packed bed chromatography are opening up new avenues for significant cost reduction. Over the coming decades, mAb production will continue to evolve. There are many promising technologies and approaches to address the existing challenges. While adoption is slow due to the regulated, conservative nature of the biopharmaceutical industry, where strong business drivers exist, this will be overcome and, in the future, integration of these technologies will become widespread. It is exciting to consider the next evolution of mAb production.

Acknowledgments

The authors acknowledge receipt of funding to Raymon Floyd Principe From the UCD Advance PhD Core Scheme and funding for Maycou Soares Zamprognio from SSPC, the Science Foundation Ireland Research Centre for Pharmaceuticals (12/RC/2275_P2).

Conflict of interest

The authors declare no conflict of interest.