Scale-up economics for cultured meat

2.1 Model cell and growth characteristics

Different animal cell types have been proposed for cultured meat production: embryonic or pluripotent stem cells, adult or mesenchymal stem cells, and primary cells (Ben-Arye & Levenberg, 2019). Each of these would have characteristic proliferation and/or differentiation phases, each with its own media composition and bioreactor design. These details are presently unknown in the public sphere. Much more is known about mammalian cell lines used in biopharmaceutical manufacturing, most of which derive from Chinese Hamster ovary (CHO) cells. This analysis therefore draws on the CHO literature for guidance on cellular metabolism, growth inhibition, bioreactor design, and other aspects.

This analysis considers an abstract mammalian cell line adapted for suspension culture. Cells are assumed to be spherical with 70% intracellular water and a hydrated mass of 3000 pg. Cell growth proceeds with a maximum growth rate µ _max = 0.029/h, equivalent to a doubling time of 24 h. A CHON formula for animal dry cell mass (DCM_a) of CH_1.68O_0.34N_0.21 derives from an average composition of 15% lipid, 10% carbohydrate, 5% RNA/DNA, and 70% protein (Alberts, 2002). Sulfur, phosphorous, and metals are ignored and the CHON formula for protein (CH_1.57O_0.31N_0.28) derives from an average amino acid profile of 207 cellular proteins (Xie & Zhou, 2005). Formation energies of ∆H _f = −73.0 kJ/mol and ∆G _f = −31.9 kJ/mol are estimated with correlations (Battley, 1999; Burnham, 2010).

An anabolic growth reaction for DCM_a is constructed from the macromolecular composition: Lipid and carbohydrate are synthesized from glucose (Glc); nucleotides are synthesized from glucose and one of the N atoms on glutamine (Gln), rejecting glutamate (Glu); and protein is synthesized from its individual amino acid residues (Xie & Wang, 1994). Guan and Kemp (1999) characterized catabolic stoichiometry for CHO cultures by matching catabolite formation rates with observed heat dissipation. Reaction 1 was deduced at the end of the batch, after cell growth had ceased:

(1)

This reaction is a superposition of respiration, glycolysis, and glutamine catabolism, and it can be generalized with two degrees of freedom, for example, lactate/glucose ratio (Lac/Glc) and glutamine/glucose ratio (Gln/Glc). As written, these ratios are relatively high (Lac/Glc = 1.77 and Gln/Glc = 0.13), indicating significant metabolic inefficiency consistent with the so-called Warburg effect, which is frequently observed in rapidly proliferating animal cells (Hosios et al., 2016). Catabolism proceeds at a rate that meets an observed heat dissipation or “metabolic power.” West et al. (2002) showed that in vitro metabolic power P _M (in pW) can be related to hydrated cell mass M _c (in pg) with . The metabolic power of 3000 pg cells is 60 pW/cell and Reaction 1 (∆H _r = −681 kJ/mol) proceeds at 0.0077 mol/mol DCM_a-h. With µ = 0.029/h, anabolism and catabolism can be combined into an overall reaction:

(2)

Given the metabolic inefficiency inherent to Reaction 2, it will stand in for an unoptimized “wild-type” cell line. It will be demonstrated below, however, that its Lac and NH₃ generation rates preclude it from reaching an economically high cell density. Growth optimization strategies include: extensive characterization; selection for lactate reuptake; transfection of glutamine synthetase enzyme; and feedback control of glucose and pH (Freund & Croughan, 2018; Pereira et al., 2018). To permit a 20 m³ fed batch carried out with this metabolism to remain under likely inhibition limits of Lac and NH₃ (to be discussed shortly), the maximum inefficiencies are rather Lac/Glc = 0.50 and Gln/Glc = 0.025. Reaction 3 therefore represents this “metabolically enhanced” cell line:

(3)

At food scale, plant protein hydrolysates may be more cost-effective and sustainable than amino acids produced individually by fermentation. Figure 2 compares the amino-acid profile implied in Reaction 1 to that of U.S. soybean meal (U.S. Soybean Export Council, 2015). The essential amino acid (EAA) profiles are similar enough that if a quantitative hydrolysate of soybean meal were fed at 1.36 mol per mol protein (to match on threonine), all EAA requirements could be met except for glutamine and about 75% of tyrosine. With aggregate compounds standing for the soy hydrolysate and the unused amino acid fraction (UAA), the following enhanced-metabolism reaction can be derived:

(4)

2.2 Cell density limitations

Figure 3 presents a sketch of a stirred-tank bioreactor (STR) with two impellers, jacket heating/cooling, and a final working volume of 80%. Sparged gas bubbles are used to transfer O₂ into solution and strip CO₂ out. The stirring action of the impeller enhances this gas-liquid mass transfer. Sparging is quantified as superficial velocity u _s (m/s) and agitation as power input per unit of liquid volume (P/V, in W/m³). The maximum cell density supported in a STR may be limited by culture viscosity, gas-liquid mass transfer rates, mixing time, catabolite accumulation rates, and other factors. As these limits are exceeded, the growth rate will drop precipitously due to inhibition. At the already-low growth rate considered here, any inhibition is economically unacceptable with respect to the accumulation of bulk cell mass. In general, it will be more cost-effective to stop the batch and begin a new one at the uninhibited growth rate.

2.2.2 O₂ mass transfer

The volumetric oxygen transfer rate (OTR) is the product of a mass transfer coefficient k _L a and a driving force: the deviation of the dissolved oxygen concentration [O₂] from its Henry's Law saturated concentration [O₂*], expressed as a log-mean difference over the bioreactor top and bottom:

(5)

To determine k _L a, the correlation of Xing et al. (2009) was developed for CHO cell-culture media at 37°C:

(6)

STR design rules for animal cell culture limit the gas sparge rate to u _s = 0.006 m/s, which is equivalent to 0.1 vvm in a 20 m³ bioreactor (Ozturk, 1996). Agitation is limited to a power input that creates turbulent eddies λ _K on the length scale of a single cell (∼20 µm):

(7)

where ν = η/ρ is the kinematic viscosity of the medium fluid and the factor of 50 corrects for the power input close to the impeller relative to the bulk P/V (Nienow, 2010). With sparging and agitation both at their recommended limits, a maximum attainable OTR can be estimated from Equations 5-7. Cell density is thus limited to the point where the culture's oxygen uptake rate is equal to this maximum OTR. To increase OTR (and thus cell density), oxygen or O₂-enriched air is typically sparged instead of air. This analysis assumes on-site production of 90% O₂ in a vacuum pressure-swing adsorption unit.

2.2.4 Mixing

Mixing time in the STR can be estimated with Equation 8 (Nienow, 2014). Here, T is the bioreactor diameter, D is the impeller diameter, and H _L is the liquid height. Mixing time should be less than 1/k _L a to ensure that dissolved O₂ is quickly transported away from the bubble (Van't Riet & Van der Lans, 2011).

(8)

2.2.5 Catabolite inhibition

In fed-batch cell culture for biopharmaceuticals production, the accumulation of toxic and growth-inhibiting catabolites is a far more frequently encountered limit than the physical limits discussed so far. Inhibiting concentrations of 2–10 mM NH₃ have been reported for mammalian cells; lactate inhibition is an order of magnitude higher (Xie & Zhou, 2005). For modeling purposes, the limits of 5 mM NH₃ and 50 mM lactate are considered here. Figure 4a (dashed lines) presents a fed-batch simulation using Reaction 2 in a 20 m³ bioreactor sparged with 90% O₂. After two cell mass doublings, the batch ends with 5 mM NH₃ and a cell density of only 7.0 g/L. In fact, there is no practical way to reach an economically high cell density with a metabolism as inefficient as Reaction 2. At any appreciable starting density, NH₃ inhibition would occur in a matter of hours. In a fed-batch simulation with Reaction 3 (solid lines), a final cell density of 110 g/L is reached before NH₃ inhibition.

These simulations can be extended to find the maximum fed-batch cell density as constrained by O₂, CO₂, NH₃, and mixing. As shown in Figure 4b, NH₃ inhibition limits the cell density to 110 g/L in bioreactors <20 m³. At 20 m³, the NH₃-limited density is coincident with the pCO₂-limited density (the catabolic parameters of Reaction 3 were selected to cause this coincidence). Bioreactors >20 m³ have a lower maximum cell density due to CO₂ inhibition. While NH₃ inhibition can be mitigated to a degree with enhanced metabolic efficiency, CO₂ inhibition cannot. Regardless of the metabolic parameters, the respiratory quotient (CO₂/O₂) of the growth reaction will remain ∼1 and the CO₂ stripping rate will remain equal to (or less than) the O₂ transfer rate. The only way to circumvent CO₂ inhibition in a sparged bioreactor is to sparge harder, possibly to the point of cell death (Al-Rubeai et al., 1995). This may preclude the scale-up of animal cell culture into extremely large bioreactors.

CO₂ and NH₃ inhibition can also be mitigated, to a degree, with perfusion. In perfusion culture, the contents of the bioreactor are continuously cycled through a cell retention device, which continuously removes extracellular products and inhibitors and generally permits higher cell densities than fed-batch cultures. At steady state, cells are bled from the bioreactor to maintain growth rate; in principle, this cell bleed could be a harvest stream for bulk cell culture. Perfusion culture volumes are limited by the capacity of the cell retention device. The alternating tangential-flow (ATF) filter is commonly used in high cell-density applications and the largest available (e.g., the Xcell ATF 10) can run at perfusion rates up to 1000 L/d (Pollock et al., 2013). In dual-ATF configurations, 1 m³ bioreactors can thus be perfused at up to 2 reactor volumes per day (RV/d), and 2 m³ bioreactors at up to 1 RV/d. Figure 4c presents curves of attainable cell density as a function of perfusion rate. The wild-type Reaction 2 generates 2 mmol NH₃/mol DCM_a-h. If NH₃ is removed via the perfusate stream at a steady-state concentration of 5 mmol/L, then at a perfusion rate of 2.0/d it can be computed that inhibition is reached at a cell density of only 20 g/L (6.8 × 10⁶/ml). With the more efficient Reaction 3, an O₂-limited cell density of 195 g/L (65 × 10⁶/ml) can be achieved at a perfusion rate of 1.0/d.

2.3 Capital costs

The capital cost of a conceptual bulk animal cell-culture process is developed from the bare-equipment costs of its most important items. From this purchased equipment cost, a total capital investment (TCI) is obtained through the application of cost escalation factors, which are understood to be rather high for biopharmaceutical cell-culture processes (Petrides, 2015). Compared to existing biopharmaceutical processes, future bulk cell-culture processes for cultured meat would probably have similar safety and sterility requirements to protect the culture from being overcome by contaminating microorganisms or infected by adventitious viruses (Moody et al., 2011). Equipment and facility design, however, would necessarily be more commoditized. Process-industry estimation software can thus be leveraged for capital cost development.

Figure 3 presents a sketch of a sterile STR suitable for animal cell culture: sterilizable inputs, jacket heating and cooling, CIP/SIP, automation, and so forth. The ASME standard for bioprocessing equipment (ASME, 2016) dictates full-vacuum design and 316 L stainless-steel construction. At 20 m³, Aspen Capital Cost Estimator (ACCE) estimates the cost of the bare vessel and agitator as ∼$330k, as indicated in Table 1. Piping and instrumentation costs can also be estimated with ACCE. With additions for surface treatment (electropolishing, passivation), internals (sparger, spray balls), and externals (exhaust heater, sterile impeller seal), the estimated total direct cost (TDC) of a 20 m³ system is $1.5 M. Estimated bare equipment (vessel and agitator) and total system costs for bioreactors of 1–200 m³ are further shown in Table 1, along with direct-cost factors. While installation costs dominate at all volumes, a strong economy of scale is noted.

To develop capital costs for specific configurations, TDCs for bioreactors are estimated with Equation 9. This piecewise correlation combines the ACCE estimates in Table 1 with estimates from SuperPro Designer at small volume, where ACCE is less accurate:

(9)

Costs for minor process equipment items (media tanks, sterilizers, filters, etc.) are estimated with ACCE, SuperPro Designer, or the correlations in Couper et al. (2012). Process support equipment (utilities, O₂ generation, etc.) is estimated independently or represented with an operating cost. To all equipment except bioreactors, an installation factor of 1.3× is applied to their purchased cost (Petrides, 2015). Building costs are computed from an estimate of equipment footprint and areal costs taken from ACCE or Petrides. To the TDC of the facility, an indirect cost factor of 0.6 is applied for engineering and construction fees, giving a total plant cost (TPC). An additional contingency factor of 0.15 is applied to the TPC to compute the TCI. Finally, the TCI is represented as an annual charge ($/y) by applying a capital charge factor (CCF; see Equation 10) of 15%/y to the TCI. Here, i is taken as 7.5% and n as 10 years; these are common values for food manufacturing facilities including plant-based meat replacements (Damodaran, 2020; Maroulis & Saravacos, 2003).

(10)

2.4 Raw material costs

Animal cell-culture media generally contains a defined composition of sugar (glucose), up to 20 essential and non-EAAs, fatty acids, phosphate, trace minerals, and various vitamins, hormones, and cytokines (collectively known as growth factors). Many of these components are not currently produced at scales consistent with food production. This section discusses where these raw materials might come from and how demand levels are likely to influence future price.

2.4.2 Amino acids

Each amino acid is currently produced at some commercial scale ranging from thousands of kTA for animal-feed supplements to <1 kTA for aminos with primarily pharmaceutical uses. Figure 5a presents available price-volume data for individual amino acids (BCC Research, 2017; IHS Chemical, 2019; Sanchez et al., 2018). Although Figure 5a clearly indicates that aminos with smaller market volumes cost more, the high-volume/low-cost data points for the major amino acids reflect feed-grade formulations unsuitable for cell-culture media. Suitably pure formulations cost more. Recognizing that the price-volume relationship observed across all amino acids must also exist across formulations of an individual amino acid, the market volume of a novel “cultured meat-grade” formulation is probably a more reliable predictor of price at scale than any specific details of its manufacturing process. These will certainly influence price, but not by multiple orders of magnitude. Prices for individual amino acids at scale are therefore correlated to estimated production volume by the following equation:

(11)

2.4.4 Protein micronutrients

Protein growth factors are provided in media to regulate growth and metabolism. Animal-free sources of these proteins include highly processed extracts of plant proteins and recombinant proteins made by fermentation. Commercial recombinant proteins (therapeutics, industrial enzymes) have a price-volume relationship similar to amino acids, as presented in Figure 5b and regressed in Equation 12 (Arbige, 1989; Gotham et al., 2018; Kelley, 2009).

(12)

Micronutrient usages are estimated on a loss basis: in perfusion culture, losses occur over time; in fed-batch culture, losses occur when the bioreactor is emptied. Four distinct micronutrients are considered in the analysis: insulin at 19.4 mg/L, transferrin at 10.7 mg/L, fibroblast growth factor at 0.1 mg/L, and transforming growth factor β (TGF-β) at 0.002 mg/L (Chen et al., 2011). Characteristic losses are, for example, 10–40 MT/y of insulin and 1–4 kg/y of TGF-β. Although the unit prices estimated with Equation 12 are relatively high, growth factors only contribute $3–4/kg of wet cell mass at 100 kTA.

2.5 Fixed costs

Facility overhead is scaled to CAPEX and is here taken to comprise 4% TCI/y for maintenance (including CIP) and 5% TCI/y for insurance. Labor is quantified in terms of operator attention per batch, based on a series of task/time assumptions. A salary of $50,000/y for regular FTEs is taken from the U.S. Bureau of Labor Statistics—Chemical Plant Operator; the supervisor's salary is 140% of this. A labor burden of 100% is added to the total labor cost.

3.1 Fed-batch case study

A process flow diagram (PFD) of a conceptual fed-batch cell-culture process is given in Figure 6a. For the detailed example given below, the model cell-culture facility is designed with 24 × 20 m³ production bioreactors and produces 6.8 kTA of wet cell mass. Cells are propagated from the lab through a seed train to the production bioreactors. Upon harvest, the cell mass is dewatered to 20% solids in a disk-stack centrifuge. Two large media tanks attached to HTST sterilizers are shared between bioreactors to provide pre-inoculation fill. A smaller, dedicated tank attached to sterile retention filters contains the pre-mixed media that will be added during the batch.

The fed-batch simulation in Figure 4a is used to size equipment and compute media and utilities usages. Bioreactor costs are estimated with Equation 9, while costs for the remaining equipment and buildings are estimated as described above. A Class 8 clean room is selected for cell-culture areas and Class 6 for laboratory areas. The sum of equipment and buildings gives a TDC of $94 M. Indirect costs are factored from the TDC to give a TCI of $328 M. As an annual capital charge, this TCI is equivalent to $48 M/y, or ∼$12/kg of wet cell mass. At a total production volume of 100 kTA, macronutrients (amino acids) contribute an additional $19/kg and micronutrients (growth factors) $3/kg. Consumables (filter membranes), utilities (clean room power) and labor (95 total FTE) contribute a combined $3/kg. These capital and operating costs are summarized in Table 2. The overall cost of production estimated for a fed-batch cell-culture process is $37/kg wet cell mass.

Figure 7 presents sensitivity analyses for the fed-batch production process. The chart is colored by the individual contributions of CAPEX, OPEX, and so forth, and the upper edge of the colored area represents the total estimated COP. In Figure 7a, nutrient costs vanish at extremely large production volume. An asymptotic cost of ∼$16/kg is predicted at 10⁵ kTA. At 100 kTA, substituting hydrolysate at $2/kg and repeating the fed-batch simulation with Reaction 6 reduces the macronutrient contribution by almost $16/kg, bringing the total cost to $22/kg. Further opportunities for cost reduction are limited. Figure 7b indicates that 24 production bioreactors are optimal in a single facility; the production cost increases at >24 bioreactors because the clean room area grows faster than the process volume it contains. Figure 7c presents the sensitivity associated with the production bioreactor size, indicating an optimal volume of 50 m³. At larger volume, the reduction in final cell density (due to pCO₂ limitations) outweighs the cost benefits of larger reactors.

3.2 Perfusion case study

A PFD of a conceptual cell-culture process based on perfusion technology is given in Figure 6b. The 2 m³ production bioreactors are inoculated by expanding cells from the lab through 125 and 500 L seed bioreactors in fed-batch mode. Upon inoculation, the production bioreactor also operates in fed-batch mode up to 120 g/L. At this density, perfusion begins at a rate of 1.0/d (requiring 2× ATF 10 filters) and the retained cell density rises to 195 g/L per Figure 4c. Each production bioreactor operates with continuous cell harvest except during 10 scheduled turnarounds (72 h each) per year. The average time at steady state is ∼91%, and 9 kg/h of wet cell mass is harvested from each bioreactor. To produce 6.9 kTA (to match the fed-batch process above), 96 bioreactors are required.

Equipment and building costs are estimated as described previously. Bulk amino-acid and glucose tanks are shared for initial filling of all bioreactors, and each production bioreactor gets two dedicated tanks to hold 24 h of makeup media each. A cost for the ATF 10 perfusion devices is taken from Pollock et al. (2013). The TCI developed in Table 2 is $663 M, giving an annual capital charge of $97 M/y or ∼$23/kg wet cell mass. Macro- and micronutrient costs are similar to the fed-batch process, while consumables costs are significantly higher due to the replacement of ATF membranes (20 per bioreactor per year at $16k each). The perfusion process is also slightly more labor-intensive than the fed-batch process (132 total FTE) and its overall cost of production is estimated as $51/kg wet cell mass.

Figure 8 presents sensitivity analyses for the perfusion process. The limited volume, relatively high bioreactor direct costs, and the CAPEX and consumables associated with the perfusion device present significant disadvantages. Nutrient costs can be minimized by assuming a much larger global production volume, as shown in Figure 8a, or reduced $16/kg by substituting low-cost hydrolysate. Figure 8b presents sensitivity to the facility production rate, noting a very weak minimum at 3.5 kTA. The issue with divergent clean room cost is also present for perfusion, but to a much smaller degree. In a 2 m³ bioreactor, a retained cell density of 195 g/L is attained at a perfusion rate of 1.0/d, which requires dual ATFs. With a single ATF and a perfusion rate of 0.5/d, 140 g/L of cells can be retained. As shown in Figure 8c, the capital and consumables costs associated with the second ATF offset the economic benefit of higher density, such that the production cost at 195 g/L is hardly any better than at 140 g/L.