A simplified techno‐economic model for the molecular pharming of antibodies

By the end of 2017, the Food and Drug Administration had approved a total of 77 therapeutic monoclonal antibodies (mAbs), most of which are still manufactured today. Furthermore, global sales of mAbs topped $90 billion in 2017 and are projected to reach $125 billion by 2020. The mAbs approved for human therapy are mostly produced using Chinese hamster ovary (CHO) cells, which require expensive infrastructure for production and purification. Molecular pharming in plants is an alternative approach with the benefits of lower costs, greater scalability, and intrinsic safety. For some platforms, the production cycle is also much quicker. But do these advantages really stack up in economic terms? Earlier techno‐economic evaluations have focused on specific platforms or processes and have used different methods, making direct comparisons challenging and the overall benefits of molecular pharming difficult to gauge. Here, we present a simplified techno‐economic model for the manufacturing of mAbs, which can be applied to any production platform by focusing on the most important factors that determine the efficiency and cost of bulk drug manufacturing. This model develops economic concepts to identify variables that can be used to achieve cost savings by simultaneously modeling the dynamic costs of upstream production at different scales and the corresponding downstream processing costs for different manufacturing modes (sequential, serial, and continuous). The use of simplified models will help to achieve meaningful comparisons between diverse manufacturing technologies.


| INTRODUCTION
The first monoclonal antibodies (mAbs) were produced in 1975 using mouse hybridoma cells (Köhler & Milstein, 1975). More than a decade later, the first therapeutic mAb was approved by the US Food and Drug Administration (FDA). This was muromonab-CD3 (Orthoclone OKT3, Janssen-Cilag), a murine IgG2a recognizing the T-cell antigen CD3, which was approved in 1986 for the control of acute allograft rejection in renal, cardiac, and hepatic transplant patients (Becker, 2007). This mAb was withdrawn in 2010 due to the availability of other treatments with similar efficacy and fewer side effects, and consequent declining sales.
The development pipeline for mAbs initially filled slowly, often reflecting the lack of suitable myeloma cell lines (Li, Vijayasankaran, Shen, Kiss, & Amanullah, 2010). However, the development of mAbs and other biopharmaceuticals expanded with the introduction of more amenable heterologous expression systems.
The earliest platforms (bacteria and yeast) were suitable for the expression of simple proteins, including antibody fragments such as the scFv, Fab, and dAb formats, but not generally for full-length antibodies, which require folding, assembly and posttranslational modifications such as glycosylation. Such complex proteins produced in bacteria tend to form inclusion bodies that need to be solubilized and refolded in vitro, and the lack of glycans means they often do not function in the same way as native mammalian antibodies. These issues were addressed by the development of mammalian host cells for the expression and secretion of mAbs and other complex glycoproteins, with Chinese hamster ovary (CHO) cells and the mouse myeloma lines NS0 and Sp2/0 achieving dominance and now used for all approved mAb products (Dumont, Euwart, Mei, Estes, & Kshirsagar, 2015;Kunert & Reinhart, 2016).
CHO cells are the most widely used platform due to the availability of a gene amplification system that achieves high yields, and the ability of CHO cells to grow in chemically defined medium (Kelley, 2009).
The advent of production platforms based on mammalian cells led to a surge in the development pipeline and rapid expansion of the global market for therapeutic mAbs, which was valued at $90 billion in 2017 and is projected to reach $125 billion by 2020 (Ecker, Jones, & Levine, 2015). Despite the success of CHO cells and their current hegemonic position as a production platform for mAbs, the production of biopharmaceutical proteins in plants (often described as "molecular pharming") offers a number of advantages over fermenter-based systems and has the potential to reduce the costs of manufacturing.
Although the mAb titers achieved in plants as a proportion of biomass still lag behind those of cell-based systems, the economic advantages of plants include the following: • Less expensive infrastructure for upstream production: Whereas cell-based systems require fermenters installed in dedicated facilities and all the associated sterile equipment for buffer and media preparation, plants can be grown in greenhouses with nonspecialized equipment. The upfront capital expenditure (CA-PEX) costs are therefore much lower.
• Less expensive raw materials for upstream production: Whereas cell-based systems require complex media, and either single-use bioreactors or rigorous cleaning-in-place between production campaigns for stainless steel facilities, plants require only light, water, and soil/hydroponic medium and essentially they are selfcontained individual single-use bioreactors that are discarded and destroyed by incineration or composting after each production campaign. The operational expenditure (OPEX) costs are therefore much lower.
• Greater scalability: Cell-based systems are restricted by the capacity of the fermenters available on site. In contrast, the scalability of plant-based systems can be increased simply by growing more plants.
• Speed of production and upscaling: The speed of production in cellbased systems is limited by the turnaround time for a single fermenter run, whereas transient expression systems in plants can achieve very short production campaigns. The speed of scale-up is also limiting for cell culture systems. It takes many months to install and commission new fermenters, whereas if a transgenic plant line is already available, the production campaign can be scaled up in one generation, simply by planting more seeds.
• Intrinsic safety: Mammalian cells can host adventitious or intrinsic human viruses. All biopharmaceutical manufacturing processes must, therefore, include rigorous orthogonal virus-removal and virus-inactivation steps, and even with such steps in place occasional contamination issues still occur (Goldblatt, Fletcher, McGill, Szer, & Wilson, 2011;Hollack et al., 2010). Plants do not support the replication of human viruses, so even if contamination does occur, the virus cannot replicate. This simplifies the purification of products intended for parenteral administration and also allows plants to be used directly for the preparation of oral or topical formulations without purification, reducing or even eliminating the downstream processing costs.
Molecular pharming began with the production of a recombinant antibody in transgenic tobacco plants (Hiatt, Cafferkey, & Bowdish, 1989) and this initial success led to many proof-of-principle studies in the 1990s and early 2000s involving diverse production platforms based on different plant species, whole plants as well as a range of tissue and cell-culture systems, and various strategies for expression including transgenic plants and transient expression (Twyman, Stoger, Schillberg, Christou, & Fischer, 2003). Few of these studies translated to commercial products despite initial interest from the pharmaceutical companies (Fischer, Buyel, Schillberg, & Twyman, 2014) but the use of plants to produce nonmedical proteins gained traction and gave rise to the first techno-economic models (Hood, Kusnadi, Nikolov, & Howard, 1999;Kusnadi, Hood, Witcher, Howard, & Nikolov, 1998), which have been developed into commercial success by companies such as Ventria Bioscience (Fort Collins, CO) and ORF Genetics (Kopavogur, Iceland). The use of plants to produce pharmaceutical proteins such as mAbs has enjoyed something of a Renaissance more recently with the development of processes that comply with current good manufacturing practice (cGMP). This MIR-ARTIGUES ET AL.

| 2527
initially involved fermenter-based systems analogous to mammalian cells (at least in regulatory terms) such as plant cell suspension cultures and clonally-propagating aquatic plants (moss and duckweed). In 2012, carrot cells were used by Protalix BioTherapeutics (Karmiel, Israel) to manufacture the first plant-derived pharmaceutical protein approved by the FDA for human use (Mor, 2015).
Transgenic tobacco plants were used by the Pharma-Planta consortium to produce the first mAb made in whole transgenic plants approved for Phase I clinical trials under cGMP regulations . The third major platform is transient expression, in which leafy plants are infiltrated with bacteria carrying the pharmaceutical transgene, resulting in the expression of the corresponding protein at high levels for a few days. Several companies have developed this system to produce vaccine candidates, and, in a well-publicized case, an experimental chimeric cocktail of three monoclonal antibodies (ZMapp) was produced in plants by transient expression as an emergency response to the 2014 West Africa Ebola virus outbreak (Na, Park, Yeom, & Song, 2015).
However, in most cases, the authors focused on a particular type of platform (plant cells or whole plants), a particular mode of expression (transient expression or transgenic plants), and/or a particular host species (e.g. tobacco or rice). The various models are not universally applicable, are generally unsuitable for direct comparison to other models, and the all-important comparison to CHO cells is challenging because of the different manufacturing scales, consumables, labor resources, and timelines. What is needed is a simplified model that allows the various parameters involved in the manufacturing process to be included as uniformly applicable variables that can be matched to different manufacturing scales and modes (batch, semi-batch, and continuous manufacturing).
The need for universally applicable TEA models for mAb manufacturing is rooted in the concept of integrated process development, in which the process is designed based on the optimum arrangement and juxtaposition of different upstream production and downstream purification units to reflect the overall economy, duration, and safety of production (Kelley, 2017). To accommodate new innovations, TEA models must encompass the entire process, otherwise, the introduction of a technology that improves the efficiency of one process unit could have the unintended consequence of reducing overall productivity or economy if its impact on the entire process is not evaluated. It is therefore beneficial to use multiobjective optimization strategies to identify the best process options, for example, to optimize for mAb titer, purity and costs simultaneously while accommodating any process uncertainties (Liu & Papageorgiou, 2018;Liu, Farid, & Papageorgiou, 2016). One example of process-wide or facility-wide modeling is the integration of continuous manufacturing concepts. Traditionally, mAbs have been produced using a batch manufacturing strategy, but following the lead from other manufacturing industries, the benefits of continuous production have been considered, particularly with the adoption of continuous fermentation, continuous centrifuges for clarification, and various forms of continuous chromatography (such as simulated moving bed chromatography) for downstream processing (Bisschops, 2017;Patil & Walther, 2018). The economic ramifications of continuous manufacturing in the field of biopharmaceuticals have therefore been evaluated in detail (Pollock, Ho, & Farid, 2013;Stock, Bisschops, & Ransohoff, 2014) and robust TEA models have been developed, which allow the assessment of process changes for biopharmaceuticals as well as conventional smallmolecule drugs (Hassan, 2012;Jolliffe & Gerogiorgis, 2017).

| A SIMPLIFIED ECONOMIC MODEL FOR THE MANUFACTURE OF MONOCLONAL ANTIBODIES
The stages of the product cycle and the regular processing activity in the biomanufacturing sector are summarized in Figure 1. Our TEA model focuses on the regular production process for mAbs, that is, the main economic features of the various manufacturing platforms rather than the preceding research, development, and innovation, which differ from case to case (Grupp, 1998). By comparing the manufacturing platforms, we can focus on the cost of goods sold (COGS), which is the most reliable comparative measure of manufacturing costs. The following key stages of the process must be included (Holtz et al., 2015;Nandi et al., 2016;Petrides, Carmichael, Siletti, & Koulouris, 2014;Prado et al., 2014;Tusé et al., 2014;Wilken & Nikolov, 2012): 2. The upstream production phase. In the case of CHO cells, the cells would be inoculated into a bioreactor and cultivated for 7-14 days with the initial (batch), periodic (fed-batch), or continuous addition of nutrients. A titer of 1-5 g/L is commonly reported for mAbs, rising to 13 g/L in exceptional cases (Kelley, 2017). The bioreactor scale ranges from 5,000 to 25,000 L, with the output of each fermentation considered as a batch (Kelley, 2009;Trexler-Schmidt et al., 2010;Kunert and Reinhart, 2016). 3. Once the upstream phase has been completed, the recombinant proteins are recovered by downstream processing (DSP), which includes primary recovery/clarification, capture, and purification (polishing). Primary recovery generates a crude feed stream that facilitates product capture, with clarification to remove particulates. CHO cells secrete the product into the medium so primary recovery is a simple process in which the cells and debris are removed by filtration and/or centrifugation, whereas most platforms based on whole plants involve the intracellular accumulation of the product, making primary recovery and clarification much more complex. The capture phase for mAbs involves the selective retention of the product, typically by Protein A affinity chromatography, which increases its purity and reduces the feed volume. Further purification by orthogonal chromatography steps removes residual impurities and contaminants, including viruses.
4. The final output from the manufacturing facility is the bulk drug substance, which is shipped to the final drug manufacturing site to be formulated as the drug product.
The efficiency of DSP depends on the concentration of the recombinant protein, the intrinsic difficulty of extraction, and the required level of purity. Therefore, when comparing mammalian cells to plant-based systems, the initial product concentration in the feed stream is an important factor, and a key difference is that mammalian cells secrete recombinant proteins into the medium, whereas whole plants usually retain the product as an intracellular fraction that must be liberated by shredding and macerating (leaves) or grinding (seeds).
Plant cells can also secrete the product into the medium, but the presence of the cell wall often causes large proteins such as antibodies to accumulate in the compartment beneath the wall, known as the apoplast (Schillberg, Raven, Fischer, Twyman, & Schiermeyer, 2013). Interestingly, the FDA-approved product taliglucerase alfa/Elylyso, a recombinant form of human glucocerebrosidase, is produced in carrot cells but is deliberately targeted to accumulate in the intracellular vacuole so that the correct glycan structures are formed, and the cells must therefore be disrupted to release the product (Shaaltiel et al., 2007). The crude extract from plants, and in many cases also plant cells, therefore, contains a much larger proportion of particulates and host cell proteins than the supernatant of mammalian cells, placing a greater burden on the clarification steps, which often require multiple depth filters in series (Buyel, Fischer, & Twyman, 2015). The clarified feed stream from mammalian cells contains the recombinant protein product and a small number of host cell proteins and metabolites, whereas the clarified feed stream from whole plants usually contains high concentrations of host cell proteins in addition to the product.
Extra processing steps can be used during clarification to achieve the precipitation of as many of these contaminants as possible, for example, blanching, flocculation, or the use of filter aids (Buyel et al., 2015). Some whole plant systems overcome this issue by secreting proteins into the medium, as is the case for the duckweed platform developed by Biolex Therapeutics (Everett, Dickey, Parsons, Loranger, & Wingate, 2012) and the rhizosecretion of proteins by tobacco plants grown in sterile hydroponic medium (Madeira et al., 2016).
We have developed a simplified TEA model for the manufacture of mAbs based on the multiple-pass regeneration process proposed by Smith (1966), which considers the resin as a central element of chemical processing. This is also the case for mAb manufacturing, where the resin is almost universally Protein A, although the model is also suitable for any other product as long as the primary DSP step is used as the central element. By focusing on this, we can modify variables related to surrounding parameters (upstream and downstream) to identify areas where cost savings can be made. The purification phase requires consumables and raw materials, some of which are very expensive. This is particularly the case for Protein A resins, which are used to capture mAbs from the feed stream and account for more than 25% of downstream processing costs (Pathak, Ma, Bracewell, & Rathore, 2015). Moreover, additional ion exchange and hydrophobic interaction resins are used for polishing (Wilken & Nikolov, 2012). These resins must be regenerated after a certain number of batches, so the DSP phase can be considered to be organized in cycles, which contain a certain number of batches. The batch can be considered in terms of the volume of fermentation broth from a CHO or plant cell fermenter (measured in L) or fresh weight (FW) of plant biomass (measured in kg). The TEA model can then be described using the following notation: B batch biomass, expressed as the volume of fermentation broth (measured in L) or kg FW of plant biomass. The units are different but can be considered equivalent because each liter of fermentation broth will (before clarification) have contained a certain amount of cell biomass, which secreted the product into the medium. This biomass is measured directly in the case of plants because the product remains inside the plant cells.
n number of batches in a cycle.
α average titer or concentration of the product (g/L or g/kg FW) in the bioreactor or installation, α > 0.
β average percentage recovery of the product with fresh resin (0 < β ≤ 1). It can be also understood as the typical purification yield. q outflow (kg or g) of a given bulk drug or intermediate product in a given cycle.
Q annual amount of outflow (kg or g) of a given bulk drug or intermediate product obtained at the end of the manufacturing process, B>>Q.
When processing one batch after another, the resin capacity falls slightly with each batch until after a certain number of batches, the capacity is low enough to justify regeneration. If this number of batches is defined as a production cycle (nB), then regeneration is needed after each cycle. Let us assume that after each regeneration, the capacity declines by a constant percentage for each batch (τ, 0 < τ < 1). Therefore, given that the output of the cycle using new resins is αβB, the output of the subsequent cycles will decline to ταβB, then τ 2 αB, and then τ 3 αB, and so on. The output of these successive cycles before resin replacement can therefore be defined as shown in Equation (1): Based on the above, each new resin allows a total of u + 1 cycles.
If R resin changes are required per year, and assuming that all resins have the same quality, the annual output can be calculated using Equation (2): Equation (2) can be simplified to Equation (3) to show the average capacity over all regenerations: Equation (3) can be simplified further by simultaneously multiplying and dividing it by (u + 1) to give Equation (4): where the annual output (Q) is equal to the volume of product per cycle (αnB) multiplied by the number of cycles (u + 1), the purification yield of the cycles across all resin changes ( ̅ ), βτ and the number of these changes (R). This expression can be simplified further given that (u + 1)nR = N, where N is the number of batches in a year, thus giving Equation (5): Equation (5) mirrors the basic techno-economic features of downstream processing, but it also includes a reference to process efficiency. Indeed, the volume αβB is associated with a given number of resins. If γ* represents the optimized relationship between the volume of bulk drug produced in a given cycle and the mass of resins used (both in kg), then when γ falls below the value γ*, the cycle is operating below potential. This could reflect a lack of operator experience or problems during purification. On the other hand, if ̅ ⁎ τ indicates the expected filtering recovery after regeneration, a current ̅ < ̅ ⁎ τ τ implies that this task is inefficient.

| FACTORS THAT REDUCE THE COST OF GOODS SOLD
By studying Equation (5), we can identify different strategic economic factors that reduce the COGS, including: • The economy of scale as the upstream production volume increases.
• The increase in product concentration (titer) during upstream production.
• The increase in downstream yield.
• The timing of the batches.
• External costs, such as raw materials, equipment, and labor.
In Equation (5), B reflects the potential for economy of scale, which is geometric in nature: As the capacity of any container increases in cubic terms, the required investment, which mostly depends on the perimeter of the container, increases in squared terms. However, economies of scale are usually less in practice than their theoretical level because the walls of the container may need to be coated, or internal agitators may be required to ensure that the reaction is spatially homogeneous. As a result, the impact on COGS is proportionally less than the increase in capacity. For example, Kelley (2009) suggests that a 10-fold increase in bioreactor capacity reduces the COGS by only five-fold. Walwyn et al. (2015) use the typical expression for economies of scale, which is shown in Equation (6): where c 0 is the COGS before the increase in capacity from B 0 to B t , c t is the COGS after this increase, and δ is a multiplier based on the sixth-tenths factor rule (Peters, Timmerhaus, & West, 2002). For pharmaceutical manufacturing, δ = 0.6 is considered standard.
For the production of mAbs and other biopharmaceuticals in CHO cells, the upstream titers (α) increased steadily from~1 g/L in the 1980s and 1990s to~2 g/L in the 2000s (with some processes achieving up to 5 g/L). Nowadays, 5 g/L has become routine and some processes have achieved > 10 g/L (Kelley, 2017). The same can be α . The two factors considered above can, for example, be incorporated into Equation (7): where t represents time.
Another important aspect of mAb manufacturing is scalability. A system with this quality can accommodate an expanding or contracting workload without great expense or inconvenience.
Formally, the production process should be able to maintain almost the same level of efficiency when the number of batches (N) becomes larger or smaller in Equation (5).
Any discussion about scalability must start from a manufacturing facility which operates at capacity, that is, its output matches the capabilities of the equipment and infrastructure according to its planned design and no bioreactors stand idle. Furthermore, the product is not stockpiled, so a higher demand cannot be met by  (5), this infers that the variables B, α, β, and ̅ τ cannot be changed. Any increase in B would require the construction of new production facilities, which may be appropriate for long-term increases in demand but not for short-term fluctuations. Accordingly, only the number of batches (N) can respond to fluctuating demand and this can be managed in two ways: • First, by changing the temporal relationship between batches, for example, by moving from a sequential process to a serial one, which is only possible if the appropriate equipment is available at an acceptable cost.
• Second, by increasing the number of shifts.

Both solutions are examples of process intensification.
When there is very low demand, the production process is usually sequential. 1 Setting aside common measures that guarantee process flexibility, 2 the sequential deployment of upstream batches (each lasting T hours) coupled to a purification phase with duration h p T (defined as a fraction of T, 0 ≤ h p ≤ 1), gives rise to the number of batches shown in Equation (8) The number of shifts (maximum three in a working day) can be increased if more upstream bioreactors are installed. Whether the process is sequential or serial, the premise should be replicated according to the number of shifts. The impact on COGS will depend F I G U R E 3 Schematic representation of a sequential manufacturing process. The numbered section in the upstream phase refers to the number of days, each of 8 hr, required to produce a single batch, and the downstream processing begins on the 6th day, and only requires 6 hr. The hatched boxes show the idle time of the downstream processing equipment F I G U R E 4 Schematic representation of a serial manufacturing process. The numbered section in the upstream phase refers to the number of days, each of 8 hr, required to produce a single batch, and the downstream processing begins on the 6th day, and only requires 6 hr. The hatched boxes show the idle time of the downstream processing equipment 1 Every batch exiting the upstream phase is processed without delay one after another, but a new batch is prepared only when the previous one is complete. If the capacity of DSP is exhausted by simultaneously processing several batches in parallel, all will be in the same growth phase and the DSP equipment remains idle most of the time while waiting for the next set of batches.
2 Measures such as ensuring there is reserve capacity in case of equipment failure. 3 We assume that 40 hr includes loading/unloading and batch preparation. Only working hours (normal shift) are considered, even though cells and plants will continue to grow throughout the day. The cost of supervision outside of normal shift hours must be included. We now consider the COGS and how it relates to the market price of the manufactured output, that is, the bulk drug product (p b expressed in $/g). This can be expressed in Equation (9), which is based on the model proposed by Mir-Artigues and González-Calvet (2007), assuming that different preparation, processing, and finishing tasks have been optimized: The following new terms are introduced: A annuity for depreciation of capital investments. 4 W wages paid for workers and employee services.
pk price of the k inflows f, given in units of B, which are consumed during upstream production (u), such as energy and raw materials.
p k' price of the k' inflows f consumed during DSP (d), with the exception of resins.
f R quantity of Protein A chromatography resins. 5 p R price of Protein A chromatography resins. 5 m gross margin of the manufacturing process.
Equation (9) is defined in annual terms. The COGS correspond to the items between square brackets. This can be rearranged as shown in Equation (10): where the first term represents the weight of depreciation and labor cost per unit of output. This element could encompass two thirds of the total costs (see below). The second term is the burden of the upstream operating cost, which is strongly influenced by the average percentage of recovery during DSP. The last term in parentheses is the operational cost of purification, where the impact of resins has been highlighted.
There is little information available about the composition of the COGS (Walther et al., 2015). An exception is Nandi et al. (2005), who reported that the cost of equipment depreciation is approximately two-thirds of the total (20-25% during upstream production and 65-90% during DSP), and that three-quarters of the operational costs reflect material inputs, energy, and maintenance associated with purification tasks. Therefore, if these data are considered representative, more than two-thirds of overall manufacturing costs correspond to the first term of Equation (10) (wages are included), a maximum of 1/12th of costs are associated with the second term (upstream operational costs) and, finally, a maximum of one-quarter of the COGS is related to the third term (downstream operational costs). There is no doubt that the main source of cost reduction is capital savings, which also depend on the production level (Q).
The most complex element of Equation (9) is the gross profit margin (m). It turns out that the difference between the COGS and p b is usually very large. In 2008, the average bulk product price for the 15 most important mAbs was $8,000/g, with market prices from $2,000/g to $20,000/g (Kelley, 2009). This large margin (i.e., m>>1), This factor can also be used for other capture resins in processes that do not include Protein A chromatography. to a given drug have expired, competitors can manufacture generic versions and the market price generally falls due to the competition.

| COMPARISON OF PRODUCTION COSTS IN DIFFERENT PLATFORMS
In the absence of a universal model, it is difficult to compare TEA studies in molecular pharming even if they show itemized production costs.
Assuming a consistent product or product class such as mAbs, the studies nevertheless feature many different upstream production platforms and an inconsistent set of DSP unit operations (although helpfully they at least tend to feature a central Protein A capture step or some functional equivalent). Furthermore, the costs vary depending on the production scale and the use of fixed equipment versus disposables (Shukla & Gottschalk, 2013), and the parameters included in different breakdowns differ considerably, with some grouping aspects such as capital expenditure and depreciation but others separating these factors. A unified model, which works regardless of platform technology and production scale, requiring a standardized set of input data, would facilitate interstudy comparisons that are currently impossible. Even so, it is possible to compare the bottom-line numbers for parameters such as CAPEX, OPEX, and COGS, especially if the TEA calculations have been presented using a widely-used modeling platform such as SuperPro (Intelligen, Inc., Scotch Plains, NJ). The following studies, considering the costs of CHO cells and molecular pharming as platforms for the production of mAbs and technical enzymes, are therefore discussed in terms of the COGS and the data from these studies and others are summarized in Figure 6. The costs of each process are adjusted to reflect the combined effect of (a) inflation in the interval between the publication of each TEA study and the preparation of the current article and (b) incremental increases in the efficiency of equipment and reagents (especially filter media, chromatography apparatus, and resins/membranes), which leads to a reduction in the COGS over time. Given that more recent studies would generally use the latest available equipment and reagents, the COGS in older studies has been adjusted downward to compensate for anticipated improvements and then corrected for inflation. Kelley (2009) reported that a COGS of $300/g in 2,000 for mAbs produced in CHO cells had fallen to $100/g in 2008 (or as low as $20/g for output volumes exceeding 10 tonnes) but these were based on published estimates rather than known costs, varied according to the mAb titer, and are likely to reflect ideal scenarios with a plant operating at full capacity. We have applied a more conservative estimate where inflation-adjusted efficiency results in a cost reduction of 1.5% per year.
Among CHO facilities, comparisons are fairly straightforward because production technologies for both the upstream and downstream phases are well established and standardized, as required under cGMP regulations, so the most significant confounding factor is the production scale. Werner (2004) estimated that the OPEX and COGS for a 250 kg/year CHO facility producing a mAb, assuming a 1 g/L titer and 70% purification yield, were $65 million/year and $260/g, respectively ($202/g in 2019 dollars, adjusted as described above). Kelley (2009) describes a model factory producing mAbs in CHO cells using batch mode, with an estimated COGS of $134/g at 1,000 kg/year and a titer of 0.5 g/L, and $26/g at 10,000 kg/year and a titer of 5 g/L ($114/g and $22/g, respectively, in 2019 USD). Petrides et al. (2014) proposed a representative case of high volume mAb production. The estimated COGS was $86/g ($80/g in 2019 USD) at 1,580 kg/year and a CAPEX of $512 million, including startup and validation. According to the model built by Walther et al. (2015), the estimated COGS of a mAb batch production process was $22/g at 1537 kg/year throughput, and $17/g if the process was continuous ($21/g and $16/g respectively in 2019 USD). In general, the COGS decreased by up to 55% when converting a batch process into a continuous process (Table 1). The relative benefits of different unit operations and their impact on COGS are discussed in detail by Xenopoulos (2015).
The comparison of molecular pharming studies is more difficult due to the differences in platform technologies, production scales, and the lack of standardization. In one of the earliest evaluations, Nandi et al. (2005) representing the US company Ventria Bioscience, Inc., considered rice grown in the open field (600 kg harvest) expressing recombinant human lactoferrin, and calculated a COGS of $382/g ($302/g in 2019 USD), a high value because the yield was only 50 mg/kg FW. However, due to the scalability of rice, if the titer increased by 100-fold to 5 g/kg, their discrete event modeling approach predicted that the COGS would fall to $5.90/g. Subsequently, the same company reported that such yields had been achieved and that the actual COGS was "comfortably" below the target threshold of $3.75/g (Broz, Huang, & Unruh, 2013). These figures compare well to earlier studies conducted on technical reagents produced in maize. For example, Evangelista, Kusnadi, Howard, and Nikolov (1998)  Note: Data reproduced from Nandi et al. (2016); COGS correspond to indicator N2 in Figure 6. produced using a conventional mammalian cell culture process in a high-capacity facility (1,000 kg/year) at a titer of 1 g/L, but even at a lower capacity (600 kg/year) the plant-based system had a COGS of $99/g ($95/g in 2019 USD), including both upstream production and DSP, representing a > 50% reduction in manufacturing costs. Based on data collected from a small sample of mAb production processes in 2009 and 2015 (Nandi et al., 2016), plants grown in confinement, with a yield of 1 g/kg FW and 70% recovery, allow a 50% reduction in COGS compared to mammalian cell production systems with a similar annual output (Table 1). However, the global production of 31 full-length mAbs produced in mammalian cells was only 8,182 kg in 2013 (Ecker et al., 2015), and there are no data for the cost of molecular pharming platforms at a very large scale, as would be the case for plants growing in an open field.

| CONCLUSIONS
The analysis developed above indicates several major economic trends associated with mAb manufacturing that are relevant to molecular pharming: • Depreciation of equipment is an important component of COGS. However, its weighting depends on the volume of production, which, in turn, reflects the impact of economies of scale (through the well-known geometrical relationship between vessel capacity and perimeter, and titer). The volume of output per unit time also depends on the batch deployment pattern. Any benefit from such advantages requires that demand be sufficiently high.
• The operational cost during DSP is higher than that of upstream production. Both are affected by the cost of inputs and consumables, but upstream costs can also be reduced improving the purification yield (β).
In  (Buyel, Twyman, & Fischer, 2017). Regardless of the platform, bulk drug prices will probably remain high for new products due to the need to recoup the cost of clinical trials and regulatory approval. However, once the products come off patent, costs could be substantially reduced if changing the platform (from mammalian cells to plants) does not require a full series of new clinical tests to prove equivalence, and allows the better scalability of plants to be exploited in the generics market (Simoens, Jacobs, Popovian, Isakov, & Shane, 2017). There is also the possibility to reduce margins by free technology transfer to developing countries (Ma et al., 2013). In this case, the cost advantages of molecular pharming will reduce the final product costs provided there is enough purification capacity. Moreover, the enormous space available for agricultural pharming would further reduce production costs and the storage of seeds would create a buffer to facilitate short-term responses to peak demands. But this would be possible only if the restrictions on pharmaceutical GM crops are eased. 6 The main barriers preventing the adoption of molecular pharming are industry inertia (partly explained by the need to recoup the After the Prodigene incident, in which transgenic maize plants were found growing in a soybean field, such relaxation seems much less likely (Hundleby, Sack, & Twyman, 2018). investment in fermenter infrastructure) and the regulatory issues associated with the dissemination of transgenic plants. Compared to the mature status of CHO cells, molecular pharming is an emerging technology for the production of biopharmaceuticals and the industry is wary of switching from well-tested technologies to a relatively untested platform. Most of the therapeutic proteins produced in plants have intrinsic advantages over the equivalent product manufactured in mammalian cells, such as the lack of sialic acid residues on recombinant glucocerebrosidase produced in carrot cells (taliglucerase alfa, Elelyso) compared to its counterpart produced in CHO cells (imiglucerase, Cerezyme). ZMapp was the first product of molecular farming whose approval was dependent, at least in part, on the advantages of the production platform rather than the product. In this case, the rapid production achieved by In the future, most biopharmaceutical manufacturing will be carried out in small to medium-sized production facilities (from 10 to 100 kg/year), which is smaller than current facilities for blockbuster drugs. This will provide an opportunity for molecular pharming platforms to gain more of a foothold in the market. However, despite the claims by Tusé et al. (2014), molecular pharming is unlikely to disrupt the market extensively by displacing incumbent platforms such as CHO cells. In the case of nontherapeutic products, there is already a burgeoning industry for molecular pharming as a platform for the manufacture of cosmetic ingredients and laboratory reagents, where the "green credentials" offer a unique selling point for environmentally conscious consumers, and certified animal-product-free laboratory reagents are desirable.
In each of these cases, it would be valuable to evaluate the relative costs of new and established production platforms using a rational and transparent model that allows universal comparisons. Our proposed TEA model is the first step toward such a goal, breaking the costs of production into factors that can be used as inputs regardless of platform technology, manufacturing scale, or manufacturing mode.