Techno‐economic analysis of a plant‐based platform for manufacturing antimicrobial proteins for food safety

Abstract Continuous reports of foodborne illnesses worldwide and the prevalence of antibiotic‐resistant bacteria mandate novel interventions to assure the safety of our food. Treatment of a variety of foods with bacteriophage‐derived lysins and bacteriocin‐class antimicrobial proteins has been shown to protect against high‐risk pathogens at multiple intervention points along the food supply chain. The most significant barrier to the adoption of antimicrobial proteins as a food safety intervention by the food industry is the high production cost using current fermentation‐based approaches. Recently, plants have been shown to produce antimicrobial proteins with accumulation as high as 3 g/kg fresh weight and with demonstrated activity against major foodborne pathogens. To investigate potential economic advantages and scalability of this novel platform, we evaluated a highly efficient transgenic plant‐based production process. A detailed process simulation model was developed to help identify economic “hot spots” for research and development focus including process operating parameters, unit operations, consumables, and/or raw materials that have the most significant impact on production costs. Our analyses indicate that the unit production cost of antimicrobial proteins in plants at commercial scale for three scenarios is $3.00–6.88/g, which can support a competitive selling price to traditional food safety treatments.

treatments are largely effective, yet may still present foodborne disease vulnerability in key processing steps for many products. For example, recent literature highlights the challenge of the "viable but nonculturable" (VBNC) state of microorganisms in these foodsanitizing treatments. 6 One or more of the current food sanitizing treatments have been shown to induce a VBNC state from which reversion to a culturable state is possible for major foodborne disease-associated microorganisms such as Escherichia coli, 7 Salmonella enteritidis, 8 Listeria monocytogenes, 9 and Shigella flexneri. 10 Biotic approaches to food sanitization have high potential as supplementary treatments to de-risk the supply chain by employing efficacious and orthogonal protection against high-risk pathogens. Food safety applications of bacteriophages (viruses capable of killing bacteria), endolysins (antibacterial proteins derived from bacteriophages), and bacteriocins (antimicrobial proteins produced by bacteria for ecological dominance) have already been approved for commercial use in the United States. For example, Intralytix Inc. offers a suite of FDAapproved bacteriophage-based antibacterial food safety products (ListShield™, EcoShield™, SalmoFresh™, and ShigaShield™). Human exposure to large numbers of bacteriophage and bacteriocin is likely in a typical diet as well as from commensal microflora in the gastrointestinal tract. Therefore, there is a strong and intuitive case for acceptance of certain bacteriophage-and bacteriocin-derived antimicrobial treatments for food safety applications. 11 In fact, various preparations of bacteriophages, such as the Salmonella-specific bacteriophage cocktail SalmoFresh™, 12 endolysins, 13,14 and bacteriocins, such as colicins 15,16 and nisins, 17 have already been granted Generally Recognized as Safe (GRAS) status as food antimicrobials by the US Food and Drug Administration (FDA). It is anticipated that similar antimicrobial preparations will be granted GRAS status by FDA in the future, as the popularity of these technologies grows and additional regulatory notices are filed.
The costs of standard food sanitizing treatments are as low as $0.01-0.10/kg food. 18 In the cost-constrained markets of food additives and processing aids, these new biotic approaches to food sanitation will need to be accessible at the low selling prices that the food industry is accustomed to, or gain market entrance as a luxury good on the basis of their differentiating features, including worker safety in the preparation and handling of the products, environmentally friendly disposal, nonimpact on the organoleptic properties of food, and no or minimal food matrix alteration. 19 Strategies to meet low cost of use can be broadly classified as either pertaining to molecular engineering of the treatment agent or manufacturing science and technology. Substantial research has been done to employ genetic engineering to alter the action of native antimicrobial proteins. 20 For example, the modular structure of the bacteriophage class of enzymes known as endolysins provides a perfect "Lego ® block"-like molecular engineering platform to swap the N-terminal catalytic domain or the C-terminal binding domain to create novel hybrid moieties. 21 Although molecular engineering approaches possess substantial potential for human therapeutics, changes to the native structure of antimicrobial proteins for food safety applications bar them from taking advantage of the expedited GRAS marketing allowance pathway.
For antimicrobials that are novel, or altered, and hence not "generally recognized" as safe, the alternative marketing approval route (food additive petition) requires a full preclinical safety data package, which is a costly and time-consuming process that creates a significant barrier to entry for new food safety interventions, given the abovementioned current pricing structures, regulations, and public perception.
Consequently, biotic food safety approaches are more amenable to cost containment through manufacturing science and technology.
The cost sensitivity of the food industry is the most significant barrier to the adoption of new food sanitizing treatments, such as antimicrobial protein (AMP) preparations. Plant-based platforms have the potential for producing market-relevant volumes of AMPs at competitive costs, because they do not require expensive bioreactors and culture media. In recent studies, we have shown that plants such as Nicotiana benthamiana, spinach, and leafy beets are an attractive and scalable production platform for production of AMPs, including antibacterial colicins, salmocins, and bacteriophage endolysins. We have previously reported expression levels as high as 3 g/kg plant fresh weight (FW). 11,14,15,[22][23][24] In this study, we address cost sensitivity with a comprehensive techno-economic analysis of plant-based production of AMP for food safety applications. We used laboratoryscale results and working process knowledge from pilot and commercial processes to develop a process simulation model using SuperPro Designer ® to assess the commercial viability of the production platform and to identify economic "hotspots" to help guide future research and development.
A selection of recently published studies on the techno-economics of N. benthamiana plant-based production of a variety of recombinant proteins is summarized in Table 1. [25][26][27][28] To our knowledge, this study is the first techno-economic analysis of a plant-based production platform for AMPs as food safety additives.

| Process simulation
The plant-based AMP production and purification process was modeled using SuperPro Designer ® Version 10 (Intelligen, Inc., Scotch Plains, New Jersey; http://www.intelligen.com), a computer modeling tool capable of sizing equipment, performing material and energy balances, developing flowsheets, scheduling operations and debottlenecking. SuperPro Designer ® built-in unit models include a suite of manufacturing unit operations (>140) that can be configured to represent a manufacturing process flow diagram for the biotechnology, pharmaceutical, and food industries. The software uses these process flows and unit operations to then generate process and economic reports, including annual operating expenditures (OPEX) and capital expenditures (CAPEX). All currency is listed in USD.
The manufacturing process flow (e.g., unit operations, materials, process parameters) was developed using working process knowledge, unpublished lab-, pilot-and commercial-scale data, and data published in the literature. Built-in SuperPro Designer ® equipment design models were used for equipment sizing.

| Host selection
Nicotiana benthamiana is used as the plant host organism in the base case scenario. Nicotiana benthamiana is used extensively for indoor plant molecular farming applications based on its rapid growth, genetic tractability, susceptibility to agrobacterium transformation, and high expression levels of recombinant proteins. [29][30][31] The species is used in the commercial scale production of therapeutics and vaccines by companies such as Kentucky BioProcessing Inc. (Owensboro, Kentucky), 32 Medicago Inc.
(Québec, Quebec, Canada), 33 and iBio CMO (Bryan, Texas). 34 The modeled facility is designed to accommodate a previously reported process using transgenic N. benthamiana featuring a doubleinducible viral vector, developed by Icon Genetics GmbH (Halle/Saale, Germany). Published results demonstrate minimal background expression of recombinant protein until the induction of deconstructed viral RNA replicons from stable DNA proreplicons is triggered by 1-20% (v/v) ethanol applied as a spray on the leaves and/or a drenching of the roots, to achieve expression levels as high as 4.3 g/kg plant FW. 35 Although the more common Agrobacterium-mediated transient expression production platform enables rapid production of recombinant target molecules, 36 this transgenic system obviates the need for additional expenses associated with Agrobacterium tumefaciens preparation, vacuum infiltration, and agrobacterium-introduced endotoxin removal. 37

| Facility design
The simulated manufacturing facility is composed of two separate process models/flowsheets: (1) the upstream processing models the plant growth, ethanol induction, and product generation, which feeds into (2) the downstream processing model for purification of the product from the process and product impurities to meet food processing aid specification. Quality assurance (QA), quality control (QC), and laboratory costs associated with good agricultural and collection practices (GACP) for upstream processing and FDA food industry current good manufacturing practice (cGMP) for downstream processing are included in the design. Equipment, materials of construction, and prices are also modeled on food cGMP standards. 38 The location of commercial-scale plant molecular farming operations of Kentucky BioProcessing Inc. A compilation of facility and process parameter inputs is presented in Tables S1-S4 or in the base case model itself, which is publicly available at http://mcdonald-nandi.ech.ucdavis.edu/tools/techno-economics/.

| Upstream processing
The upstream processing model flowsheet is graphically depicted in Tusé et al. 26 Walwyn et al. 27 Nandi et al. 28 Alam et al. 25 (1) (2) processing are generated in-house from validated Working Seed Banks, which were in turn generated from validated Master Seed Banks. The seed bank release testing includes germination efficiency >95%, confirmation of growth kinetics, and viral testing. CAPEX related to seed generation are excluded, but associated seed production costs are included in the estimate of $9.50/g seed (1 g of seed is approximated as 9,500 seeds).

| Downstream processing
The downstream processing model flowsheet is graphically depicted in  Table S5. A plant-made AMP purification protocol uses similar acidic extraction to remove N. benthamiana host proteins. 11 The plant extract is clarified using tangential flow microfiltration.
The clarified stream is then ultrafiltered with additional tangential flow filtration using a 10 kDa molecular weight cutoff to a concentration factor of 20.
The AMP in the retentate stream is then purified with cation exchange column chromatography in a bind-and-elute mode of operation. The AMP is eluted isocratically in elution buffer (50 mM sodium di-hydro phosphate, 1 M NaCl). The purified stream is subjected to one final tangential flow filtration procedure for buffer exchange into phosphate-buffered saline (PBS) with a diafiltration factor of

| Scenario analysis
Base case scenario outputs were used to identify parameters with significant impact on process economics. We focused the scenario analy- Resource purchase costs are defined as inputs that directly control the economic impact of resource utilization for outputs of the model.
For the purpose of this analysis, purchase price parameters are contained to cost items within OPEX.

| Alternative scenarios
Alternative facility design scenarios were developed as comparative models to more broadly explore the context of the base case scenario process economics. The alternative scenario models were designed in alignment with base case scenario inputs unless otherwise noted; each alternative scenario was chosen to isolate the impact of a key facility design assumption. however, their expression levels were approximately 10 times lower than that in N. benthamiana so additional research is needed to increase production levels. 11,14,22 Several salmocins and lysins can be expressed at high levels in spinach, which is comparable to expression levels in N. benthamiana. 17,20 The primary distinction in this alternative plant host organism is the lack of nicotine, the major alkaloid in Nicotiana species. In the base case scenario, significant downstream processing emphasis is placed upon nicotine removal. The upstream and downstream processing model flowsheets are graphically depicted in Figure 1 and Figure 2. A complete list of changes to the base case scenario inputs can be observed in Table S4.
The second scenario investigates outdoor field-grown transgenic ethanol-inducible Nicotiana tabacum as an alternative to an indoor plant growth facility. Large Scale Biology Corporation previously investigated N. tabacum outdoor field-grown production of recombinant proteins and personnel involved in that work recommended pursuit of this agronomic approach, with special consideration of field condition variability on product consistency. 40 N. tabacum is used instead of N. benthamiana for its increased resilience to agricultural pathogens and weather fluctuation. 40 The upstream processing model is adapted from a techno-economic analysis of plant-made cellulase produced in the field. 26 The upstream and downstream processing model flowsheets are graphically depicted in Supplementary Information, Figure S1 and Figure S2. A complete list of changes to the base case scenario assumptions can be viewed in Table S6.

| Upstream processing
To meet the yearly production demand of 500 kg AMP, upstream processing must produce 867 kg AMP to offset the 42% downstream processing loss. Each upstream processing batch yields 9,520 kg N. benthamiana plant FW containing 9.52 kg AMP, which represents 10% of the total soluble protein (TSP). 11

| Economic analysis of base case
The base case manufacturing facility requires $50.1 million CAPEX and $3.44 million/year OPEX. The AMPs' cost of goods sold (COGS) is calculated to be $6.88/g. Figure 3 shows an economic assessment of upstream and downstream processing. Upstream processing repre-

| Expression level and production capacity analysis
To evaluate the impact of AMP expression level and facility AMP production level, we developed models for a 500 kg AMP/year production level with different AMP expression levels ranging from 0.5 to 5 g AMP/kg FW (Figure 5a,b), and for an expression level of 1 g AMP/kg FW over a range of AMP production levels from 100 kg AMP/year to 1,000 kg AMP/year (Figure 5c,d). Note that in all cases, the unit operations were resized to meet the design requirements. COGS decreases with diminishing returns as a function of expression level, as can be seen in Figure 5. To illustrate this point, consider that an increase of expression level from 0.5 to 1 g/kg FW results in $4.43/g decrease in COGS, while an increase from 4 to 5 g/kg FW results in $0.22/g decrease in COGS. These changes are equivalent to 39% and 6% reductions, respectively. Also note that at low expression levels the upstream operating costs contribute more to the COGS, whereas at high expression levels downstream operating costs contribute more to the COGS. This is reasonable because the number of plants per batch will increase as expression level decreases, thus requiring more soilless growth media, seeds, and nutrients. CAPEX follows a similar trend with expression level; however, the downstream process is the main contributor to CAPEX, except for very low expression levels (less than 0.5 g/kg FW). The majority of COGS and CAPEX variation with expression level is attributable to upstream processing, with downstream process costs remaining fairly consistent over the range of expression levels considered.
COGS also decreases with diminishing returns as a function of yearly production capacity. Downstream processing is the main contributor to COGS at low production levels while upstream processing is the main contributor at high production levels; at 100 kg/year, downstream processing represents 64% ($8.51/g) of the COGS, while at 1,000 kg/year the contribution is reduced to 35% ($2.15/g) of the  At all yearly production levels, significant diminishing returns for increases to expression level are evident within the selected range expression level.

| Alternative scenario analysis
The nicotine-free S. oleracea scenario produces 500 kg AMP/year at 1 g AMP/kg FW with 66% product recovery and 63% purity formulation (   plants or a second round of seeding to generate new plants. In the former situation, manufacturing cost reductions would also include those associated with seeding and tray cleaning operations.

| Yearly production demand and expression-level analysis
Within the given parameter range for expression level and yearly production volume, COGS is more strongly impacted by the expression level. This behavior is specific to the defined parameter ranges, which were selected based on anticipated needs and expectations. In this study, we assumed that raw material and consumable resource purchase costs per unit are independent of yearly amount purchased. As yearly production increases, economies of scale dictate that the material unit price will decrease. This becomes a more important consideration when evaluating COGS over a wide yearly production range. of the approximately linear scalability of the production platform. This is a main advantage of plant-based production that makes the scale-up from lab to commercial scale considerably simpler and faster than traditional bioreactor-based production platforms. 43 As yearly production changes, the upstream processing scales in an approximately linear fashion for a given processing strategy. However, one could anticipate that scaling to even higher yearly production could enable higher efficiency upstream processing strategies and thus improve the scaling dynamics of upstream economic contributions.

| Alternative scenario analysis
The nicotine-free S. oleracea scenario provides insight into the manufacturing costs associated with nicotine clearance. There are minor differences in plant growth and harvest operations, but the majority of upstream COGS reduction is because of higher product recovery and thus lower biomass requirements for a given yearly pro- The field-grown N. tabacum scenario results in the lowest COGS of $3.00/g AMP, providing reasonable justification to pursue this manufacturing process. However, our assumptions do not account for potential upstream difficulties associated with product expression consistency, greenhouse growth, and transplantation of seedlings (direct field seeding is assumed) or crop loss because of adverse weather events throughout the growing season, nor do they account for the downstream difficulties associated with removal of the more viscous N. tabacum host leaf impurities. Future work to experimentally support key assumptions of field growth could add higher confidence and value to this alternative scenario. Additionally, the current growth strategy is based on tobacco production as a commodity good; there may be a different growth strategy that is optimal for recombinant protein production (e.g., increased planting density with reduced time to harvest and higher number of batches per year). It is worth noting that this manufacturing process is expected to scale especially well. In our model, we assume that dedicated personnel and upstream equipment are required for transgenic handling. At an annual production level of 500 kg AMP, this results in 17% upstream equipment utilization. This means that as the yearly production demand increases, we expect marginal increases to upstream CAPEX and OPEX. As such, we expect upstream-related COGS to reduce dramatically with increases in yearly production demand.

| Cost of use
Biotic food sanitizers can be used in a variety of applications to augment traditional food sanitizing treatments against specific high-risk pathogens. Given the differences in food safety practices among food products, it can be difficult to measure the cost of use as a single value. Instead, we focused our discussion on cost of use calculations with application rates representative of AMP use-colicins for control of E. coli on red meats. We chose to investigate this example at several points along beef processing: animal washing, post-slaughter carcass cleaning, and meat product protection. We anticipate an application rate of 2-10 ppm AMP in water for animal and carcass  Table S7 and Calculation S1, respectively. In all three points of intervention, AMP application cost ranges are below or overlapping those of standard treatments. Additional information is needed on application rates and spray volume used in animal washing to reduce AMP cost of use range and increase confidence in cost comparison to standard treatments. On the other hand, AMP cost of use ranges for treatment of meat product overlap significantly with standard interventions, indicating comparable costs. Finally, the AMP cost of use ranges for post-slaughter carcass cleaning suggest that the use of AMP at this beef processing juncture has the potential to be substantially lower in cost than standard treatments.

| CONCLUSIONS
Current food safety practices, although largely effective, result in foodborne illnesses that impose a $14 billion annual burden on the US healthcare system. As the looming prevalence of antibiotic resistance grows, so will the impact of foodborne illnesses. The need for protection against foodborne pathogens is only increasing.
Reports as far back as 20 years ago acknowledge that areas of the food industry like the meat sector will need to absorb additional costs to improve food safety levels. 45 We investigated bacteriophagederived lysins and bacteria-derived AMPs to explore the capacity of this class of biotic sanitizers to improve food safety levels in the costsensitive food industry. Although previous studies illustrate the efficacy of AMPs, in this study, we performed a techno-economic analysis of plant-based production of AMPs to better understand the commercialization potential of products produced using this platform.
Our analysis predicts a $6.88/g AMP COGS for the base case scenario, $4.92/g for the nicotine-free S. oleracea scenario, and $3.00/g for the field-grown N. tabacum scenario. We also evaluated the sensitivity of the base case COGS to changes in purchase price, expression level, and yearly production. In doing so, we identified economic "hot spots," which include the large contribution of the soilless plant substrate (41.2% of annual operating costs; Figure 4b) and downstream labor-dependent costs (18.5%). The cost of use analysis indicates that AMPs are projected to de-risk foodborne disease in beef processing as supplemental sanitizing treatments at only minor economic perturbation across several key processing junctures. It is expected that other food processing operations would yield similar benefits.
This techno-economic analysis of plant-based production of AMPs is focused on manufacturing costs and the implications for application costs. In developing this model and analysis, we have identified several areas of importance for future analysis, for example, consideration of avoided costs associated with the prevention of food disease and illness. An example of a major avoided cost is that associated with food recall, which includes impact to brand image and loss of sales. A cost-benefit model that includes these avoided costs may provide more complete insights into AMPs as a food sanitizing treatment. In addition, there are social, cultural, and behavioral factors that can impact food safety that are not considered in this economic analysis.
In our analysis, we describe plant-based production of AMPs as a food processing aid. A direct evaluation of traditional manufacturing platforms, such as mammalian cell suspension culture and bacterial fermentation, as alternative scenarios would be a valuable future con-