key: cord-0998012-che0kz2y authors: Coffman, Jon; Bibbo, Kenneth; Brower, Mark; Forbes, Robert; Guros, Nicholas; Horowski, Brian; Lu, Rick; Mahajan, Rajiv; Patil, Ujwal; Rose, Steven; Shultz, Joseph title: The design basis for the integrated and continuous biomanufacturing framework date: 2021-05-11 journal: Biotechnol Bioeng DOI: 10.1002/bit.27697 sha: e6765c47fb5fffcb070471e2d04e93edcbbd70b3 doc_id: 998012 cord_uid: che0kz2y An 8 ton per year manufacturing facility is described based on the framework for integrated and continuous bioprocessing (ICB) common to all known biopharmaceutical implementations. While the output of this plant rivals some of the largest fed‐batch plants in the world, the equipment inside the plant is relatively small: the plant consists of four 2000 L single‐use bioreactors and has a maximum flow rate of 13 L/min. The equipment and facility for the ICB framework is described in sufficient detail to allow biopharmaceutical companies, vendors, contract manufacturers to build or buy their own systems. The design will allow the creation of a global ICB ecosystem that will transform biopharmaceutical manufacturing. The design is fully backward compatible with legacy fed‐batch processes. A clinical production scale is described that can produce smaller batch sizes with the same equipment as that used at the commercial scale. The design described allows the production of as little as 10 g to nearly 35 kg of drug substance per day. facilities for their antibody treatments. Expanding these facilities will take about five years. In addition, these facilities cost $400-800 M (Jagschies, 2020) . Finally, due to supply chain concerns, many countries and regions want to manufacture these COVID-19 treatments within their political sphere of influence (Hafner et al., 2020) . The common framework enables single-use equipment to be as productive as larger stainless-steel plants. By enabling cost-effective use of single-use technologies, the common framework allows the fast, and relatively inexpensive, building of large-capacity biomanufacturing in nearly any locality. Examples are provided that include a 500 L bioreactor with a low productivity cell line and 2000 L bioreactors with a high productivity cell line. The 500 L bioreactor makes about 0.5 kg of material in the bioreactor per day, which results after 20 days in a 6 kg batch. The 2000 L bioreactor makes about 10 kg of material per day, which results after 20 days in a 120 kg batch. A four 2000 L process is described that results in as much as a 500 kg batch over 20 days, or about 8 ton per year in 17 batches. These upstream options are integrated with a downstream design that is flexible enough to allow an almost 10,000-fold productivity range. The design of the upstream and downstream was based on common and well-known engineering principles. The design examples are shown in Table 1 . The justification for the design, and a more detailed materials and methods, is in the Supporting Information. In addition, a spreadsheet is enclosed in the Supporting Information that can enable other manufacturing scenarios. A spreadsheet is located online, which will allow further development of new technologies (Coffman, 2020) . The high-level consequences of the common framework ICB are discussed in the accompanying article . Here we discuss the details of the design to demonstrate the flexibility and productivity of the framework ICB. The examples used are shown in Figure 1 and Tables 1 and 2. They are described in more detail below. The common framework bioreactor is designed to support high cell densities. The largest design challenges deal with supplying cells with O 2 and stripping cells of CO 2 . The mass of nutrients required by high cell densities requires the development of media concentrates, as well as the possibility of several feed solutions. The high density of cells also requires cooling to maintain the temperature. While these challenges impact the bioreactor design in several ways, they do not preclude the bioreactor for standard fed-batch operation. Dynamic, or non-steady-state perfusion, where the cell density is not directly controlled by a bleed, can reach cell densities of 100-150 Mcells/ ml (Wolf et al., 2020) . These cell densities would require oxygen transfer rates between 15 and 50 mM/h, as shown in Table 2 (Jorjani & Ozturk, 1999) . The gas/liquid transfer coefficient k L a between 15 and 70 h −1 would be required (Moutafchieva et al., 2013) . The cells consume oxygen so fast that they become hypoxic in less than 10 s without sparging (see supplemental information). Further, there are a few lab or pilot scale examples of perfusion cell culture at 240 M cells/ml (Clincke, Molleryd, Samani, et al., 2013; Clincke, Molleryd, Zhang, et al., 2013; Zamani et al., 2018) . If the industry is to expand to this cell density, a k L a of over 100 h −1 , or lower cell-specific oxygen uptake rates, would be required. The gassing strategy for a perfusion culture is a three-fold balancing act between control of dissolved oxygen (DO), pCO 2 accumulation (i.e., pH and subsequent base utilization), and foam formation. Achieving desired control in any one aspect can result in an undesirable result in another aspect. The supply of oxygen to the cells can easily be the limiting factor in achieving high cell densities (Ozturk, 1996; Zhu et al., 2017) . The design of the mass-flow controller and sparger require consideration. Generally, mass-flow controllers delivering air and O 2 for perfusion processes requires significantly higher flow capacity than fed-batch processes, ranging from 0.05vvm to 0.2vvm. With respect to sparger design, micro-spargers have been demonstrated to improve the oxygen transport capacity needed to achieve a desired DO at perfusion-scale cell densities (Diekmann et al., 2011; Dreher et al., 2014) . However, microspargers raise several concerns, including their potential to cause cell damage due to shear when the microscopic bubbles burst (Wolf et al., 2020) . Microspargers have also been observed to yield higher pCO 2 accumulation than traditional macro, i.e. ring or drilled-hole, spargers (Dreher et al., 2014) , which lowers culture pH, and increases base utilization, which can be increase cell stress and the rate of cellular apoptosis. Additionally, microsparging creates a thicker layer of foam than traditional spargers. This thick foam requires enhanced control methods, and is a concern with respect to fouling bioreactor exhaust. A dual-sparger design that is composed of a microsparger and a macrosparger permits the most flexibility in configuring the gas flows to balance the requirements for dissolved oxygen, dissolved carbon dioxide, and foam management in a perfusion bioreactor. In this design, the microsparger component generally provides the required dissolved oxygen to a culture via relatively smaller bubbles and the macrosparger component generally provides a means to manage pCO 2 accumulation via stripping carbon dioxide from the system with relatively larger air bubbles. The dual-sparger design can also provide an additional benefit of reducing the amount of foam generated via destruction interactions between the two bubble sizes (Karakashev et al., 2012) . Even with a dual sparger design, foam and aerosol management is more difficult in perfusion cultures due to relatively higher agitation and gas flow rates than fed-batch cultures. The first line of defense is to manage foam generation in the bioreactor itself through means such as mechanical disturbance, adjustment of overlay humidity, and The last consideration for the perfusion bioreactor gassing strategy is the control strategy. Essential to maintaining the balance between DO, pCO 2 accumulation, and foam generation is establishing bioreactor-specific relationships for the parameters that govern this balance, including the gas transfer (i.e., k L a) for a given gas species and sparger design. Therefore, it is essential for perfusion bioreactor control architecture to permit the use of reasonably complex algorithms (Abbate et al., 2020) to manage these three control elements, as well as the ability to incorporate so-called offline, at-line, and online measurements such as cell density, temperature, lactic acid concentration, glucose concentration, and pH. Media delivery is also different for high-cell density perfusion compared to fed-batch. The media is concentrated and can be in multiple streams rather than one stream. In the scenarios shown in Table 1 , the media concentrates are made up of a ×5, ×12, and ×20 concentrates, which, taken as a whole, are effectively a ×3 concentrate. Multiple media feed pumps sized from 0.02 vvd to 0.3vvd would be needed. For a 2000 L bioreactor, this is a surprisingly small 20-400 ml/min. A glucose concentrate and antifoam will also need to be added with even smaller flow rates. The pump rates should be capable of being set by complicated calculations that may rely on PAT or on mechanistic models, similar to the ones that might be required for gassing. Overall bioreactor level control is required to control the feeds or the permeate rate. The media concentrates may be added through a single line with a static mixer. Typically, a line of water enters the static mixer, and a series of T-connections allow the mixing of each concentrate in a serial fashion. The use of media concentrates greatly reduces the size of the media preparation and media storage area. In the 4 × 2000 L scenario shown in Tables 1 and 2, the media can be stored in sixteen 5000 L single use containers. The ×20 media would require an additional 2000 L container. These bioreactors may become exothermic and require cooling due to the exceptional quantity of cell mass, a novel concern for perfusion as compared to batch mammalian cell culture. Cells generate about 28 pW/cell (Kemp & Guan, 1997) The common framework bioreactor supports both ATF and TFF operation. The TFF and ATF design is shown in Tables 1 and 2 Single-use bioreactors over 500 L do well to have 1.5-to 2-in. ports to allow unrestricted flow for each pump system. Connecting the SUB to the TFF requires a 1.5-in. sanitary connection, which is not known to be commercially available. Instead, two 1-in. ports are often used for each connection. This arrangement is suboptimal. The 2000 L bioreactor requires four 1.5-to 2-in. ports to support the ATF and two 1.5-to 2-in. ports to support the TFF operation. Dual chromatography is the simplest and perhaps easiest form of ICB that can be implemented (Angarita et al., 2015) . The high-level chromatogram skid design shown in Figure 1 allows flexibility for ICB and batch processing. The system uses two pump sizes. A smaller pump is used for the solution concentrates. Larger pumps are used for the diluent and the load. This design has a large processing range. The overall accurate flow rate range is 0.3 to 13 L/min. Although these skids are used by many for all three chromatography steps, the Protein A capture step determines the productivity and timing of the process. High productivity processes use solution concentrates and dilute with water. The largest processing possible on this system is two 25 L columns with a 10 cm bed depth operated in series, allowing for 57.6 kg/day. The smallest mass that can be processed uses only the smaller pump with no dilution at the lowest accurate flow rate, allowing the use of a 0.6 L column. This set up processes as little as 36 g/day. Two columns operated in parallel (a standard batch operation) can process only as much as 48 kg/day, due to the decreased effective capacity of the column. Because these columns are never operated in series, they can be 20 cm in length, and 50 L each. A onecolumn capture step can also be considered (Kamga et al., 2018) . It would allow only about 24 kg/day production on a 43 L column (data not shown) and can support significantly less bioreactor volume. This mass is less than half of that for the dual column system for about the same Protein A volume. Since these last two options do not overload the Protein A column and run a second column in series to capture the product that flows through the column, they are different from a regulatory perspective than the dual-column chromatography operation. They would likely require a different process characterization and process performance qualification package. The downstream process can adjust the mass throughput in two ways. Firstly, the capture columns can be underloaded. The load range of the Protein A is assumed to be 10-72 g/L. The second is that the columns can be cycled as few as once per day, to about 15 times each per day, depending upon the titer. The lower limit is set, in part, by the bioburden control strategy, in that we prefer to load a column less than 12 h before sanitizing it. The total in-batch range of productivity is therefore over 200-fold. Between batches, the columns can be changed in size from 0.6 to 25 L. Thus, the overall batch-tobatch productivity range for the downstream is an astonishing 9000fold range. The polishing steps can be operated in a similar cadence. The average number of cycles per day is set to avoid changing columns during the 20-day run for commercial processes. Each Protein A column is cycled on average 10 times per day, for a total cycling of 200 cycles per lot. The polishing steps are cycled 12.5 times per day for the maximum productivity, for a total cycling of 250 cycles per lot. While column lifetime may vary from product to product, targeting 200 for the Protein A step and 250 cycles for the polishing steps is not overly burdensome for commercial processes. Poorer column lifetime will require larger columns and fewer cycles per day, or changing the columns during the lot. The framework also recommends ultraviolet (UV) absorbance detection with two path lengths to allow both high and low con- (Brestrich et al., 2014 (Brestrich et al., , 2015 (Brestrich et al., , 2016 . Additional data connections should be available for expanding PAT methods that are not described here. This UV detection system does not exist commercially. The dual column chromatography skid may not benefit from being single use, depending upon the application. Small lot sizes will have an increased burden on the COGM of the single use assemblies. This burden is less important in clinical manufacturing, where flexibility and changeover time might be more important. This burden is also less important for very large batches, such as those derived from the 4 × 2000 L example. The dual column system is backward compatible with batch processing. The downstream can process 48kg/day, or from a fedbatch with 24 g/L titer, from each of the four bioreactors if they are harvested on different days. This titer is well in excess of any standard fed batch process. The framework virus inactivation step supports both batch and continuous virus inactivation. The skid performs on-line titration of the peak as it elutes from the Protein A step. The acidified peak then goes into a tank for the batch operation or into a plug flow reactor (PFR) for continuous inactivation. At the end of the incubation time, the product stream is neutralized before passing to the next step. Those processes that use a surge tank before the virus inactivation step will have a homogeneous product stream entering and exiting the PFR. This homogeneity allows for either a volumebased titration, or one based on a closed-loop pH control. Those steps that titrate the peak in-line without a surge tank will require some feed-forward control based on UV. The UV signal from the Protein A skid will require integration with the virus inactivation skid, or the VI skid will require a UV sensor on the inlet. The flow rate of acid, base, and water required to titrate the product stream will be linear with respect to UV: where A, C, and E are proportionality constants related to the amino acid composition of the product, and b, d, and f are offsets largely related to the steady-state buffer concentration. These constants can be easily determined experimentally. The flow rate of water would be required to maintain a constant ionic strength if needed for the subsequent step. Dilution would not be needed if there was a surge tank after the virus inactivation that averages out the ionic strength variation across the peak. Since the titration of the product stream changes with concentration, the overall flow rate changes during the elution. This variability can be accounted for in the PFR volume, as well as the load of the subsequent step, by sizing each to the largest expected flow rate. High-density fed-batch processes precipitate HCP and other impurities after the low pH virus inactivation process. The particulates are removed typically by a depth filter. Perfusion based processes do not precipitate with the same frequency or degree as fed-batch processes (data not shown). Even so, the framework process requires the flexibility of adding a depth filter. The depth filter has challenges for continuous processing. There are not many depth filters that arrive sterile or have a low bioburden specification. This means that depth filtration is an opportunity for bioburden to enter the process. Some depth filters can be sanitized with various solutions, hydroxide being the most typical. Since the depth filter capacity is assumed to be low (200 L/ m 2 ), it will require frequent changing. The framework presumes daily changing, largely to avoid bioburden in depth filters that cannot be sanitized. The filter area is proportional to the Protein A column volume, and varies from 0.3 m 2 for the clinical process to 6 m 2 for the 500 kg process. While 6 m 2 is seen commonly in small manufacturing plants, it is best flushed with its own pump system (not shown). Virus filtration provides a robust and scalable solution for viral clearance in biologic manufacturing. The choice of a suitable virus filter is governed by a combination of feed characteristics and process parameters (Bohonak & Zydney, 2005; Bolton & Apostolidis, 2017; Bolton et al., 2005; Rathore et al., 2014; Syedain et al., 2006) . Low flow rates, and concomitant pressure, have been reported to reduce viral retention (Strauss et al., 2017) . This effect is dependent on factors such as pH, ionic strength, and membrane composition. This loss of retention under continuous operation has, for many, restricted the VRF process to a batch operation. Tables 1 and 2 depict various scenarios for a constant flow-rate VRF process following the second polishing step. As shown in Table 2 , the low productivity scenario would necessitate a membrane area of 0.5 m 2 . Each VRF sub-batch is expected to be operated at 24 h intervals. Here, each sub-batch will take roughly 3 h, yielding the final product pool of 277 L at 24 g/L ( Table 2 ). The concomitant adjustments in membrane area and other operating parameters towards changes in capacity and productivity are mentioned in the supplementary information (Table 2 and Supporting Information Table) . The common framework process enables both batch and single-pass (SP) ultrafiltration and diafiltration (UFDF). The framework uses a flexible single-use skid with the same pumps and different flow path to accommodate both operations. The batch UF/DF step is shown in Figure 1 and involves batch tangential flow filtration (TFF) and a recirculation skid to attain the desired product concentration. The framework process performs a batch UFDF frequently to avoid the need for a large system. The entire UF/DF process will be split into as many as 20 sub-batches that are expected to be in-sync with the VRF pro- Each are sub-batched one time and 20 times, respectively. The volume of DF buffer required for a batch with high productivity can be prohibitively high (Table 2 and SI Table) . For the 500 kg batch, the DF buffer volume is 66,000 L, which is 25% of the total downstream solution volume. This issue can be partially miti- Recent reports also successfully demonstrated the use of hollow-fiber dialyzers for buffer exchange (Yehl et al., 2019) . The framework skid would also enable the operation of this step, which has the potential of reducing the total DF buffer required by a factor of 8 from the batch-wise UFDF option. The common ICB framework enables the design of equipment flexible enough to process 6-500 kg lot sizes. The design also allows flexibility for some batch operations and is compatible with legacy processes. This flexible, mostly single-use design has particular relevancy today during the pandemic. While most COVID-19 mAb projects are made in fed-batch right now, converting from fed-batch to ICB is possible. While converting to ICB will not help in the near term, it would allow manufacturing expansion within about two years. Converting to a perfusion bioreactor would be a relatively large change from a regulatory perspective, and would require strong comparability package. The alternative is a capacity crunch and investment in billions of dollars of stainless-steel infrastructure in countries all over the world. The pandemic exemplifies the uncertainty in manufacturing: the antibody manufacturing may be required for many years to come or it may not. In addition, fast development of scalable mAb manufacturing will help in the next pandemic. The framework ICB allows companies to harmonize their platform around off-the-shelf equipment. Also, it's possible for contract manufacturers to buy equipment that would be backward compatible with batch processes, as well as future compatible for ICB. This flexibility has not been realized to date. Many companies think of ICB processes as a step forward from which there is no turning back. This paper demonstrates that the common framework ICB can be designed, purchased, and installed with an eye toward the future, and yet without worry about the past. Experimental validation of a cascade control strategy for continuously perfused animal cell cultures Twin-column captureSMB: A novel cyclic process for protein A affinity chromatography Compaction and permeability effects with virus filtration membranes Normal-flow virus filtration: Detection and assessment of the endpoint in bio-processing Mechanistic modeling of the loss of protein sieving due to internal and external fouling of microfilters A tool for selective inline quantification of co-eluting proteins in chromatography using spectral analysis and partial least squares regression Application of spectral deconvolution and inverse mechanistic modelling as a tool for root cause investigation in protein chromatography Advances in inline quantification of co-eluting proteins in chromatography: Process-data-based model calibration and application towards real-life separation issues Very high density of Chinese hamster ovary cells in perfusion by alternating tangential flow or tangential flow filtration in WAVE Bioreactor-Part II: Applications for antibody production and cryopreservation Very high density of CHO cells in perfusion by ATF or TFF in WAVE bioreactor. Part I. Effect of the cell density on the process Common framework spreadsheet U1eSSDQhbQ/edit?usp=sharing A common framework for integrated and continuous biomanufacturing Single use bioreactors for the clinical production of monoclonal antibodies-A study to analyze the performance of a CHO cell line and the quality of the produced monoclonal antibody Design space definition for a stirred single-use bioreactor family from 50 to 2000 L scale COVID-19 and the cost of vaccine nationalism Hierarchy of high impact improvements in bio manufacturing. Paper presented at Workshop on Innovations in Pharmaceutical Manufacturing Effects of cell density and temperature on oxygen consumption rate for different mammalian cell lines Integrated continuous biomanufacturing platform with ATF perfusion and one column chromatography operation for optimum resin utilization and productivity Foams and antifoams Industrialization of mAb production technology: The bioprocessing industry at a crossroads. mAbs Heat flux and the calorimetricrespirometric ratio as measures of catabolic flux in mammalian cells Exploration of synthetic depth filtration applied to mammalian cell harvest Experimental determination of the volumetric mass transfer coefficient Engineering challenges in high density cell culture systems Evaluating disposable depth filtration platforms for mAb harvest clarification Introduction on foam and its impact in bioreactors Computational fluid dynamic modeling of alternating tangential flow filtration for perfusion cell culture Mechanistic modeling of viral filtration Characterizing the impact of pressure on virus filtration processes and establishing design spaces to ensure effective parvovirus removal Protein fouling of virus filtration membranes: Effects of membrane orientation and operating conditions Shear contributions to cell culture performance and product recovery in ATF and TFF perfusion systems Perfuson cell culture process for biopharmaceuticals Hollow fiber countercurrent dialysis for continuous buffer exchange of highvalue biotherapeutics High cell density perfusion culture has a maintained exoproteome and metabolome Industrial production of therapeutic proteins: Cell lines, cell culture, and purification The authors wish to thank the many participants who participated in the discussions and interviews. The authors are paid employees of their affiliated biopharmaceutical companies. None have received any compensation for the research reported in this article beyond that from their affiliated employer. Additional Supporting Information may be found online in the supporting information tab for this article.