Could perfusion microbioreactors bring more agility to biomanufacturing?
By Rajeev J. Ram
In semiconductor manufacturing, for example, a thorough understanding of process variation allows companies to manufacture circuits with billions of transistors at high yields. These variations are translated into a set of design rules, which help ensure that designs will be manufactured successfully and meet safety and other regulatory requirements.
The ability to codify manufacturing into robust design rules has had a tremendous impact on the semiconductor industry, and enabled it to move from a simple vertically integrated model where manufacturers designed and manufactured product, to a model in which a few global suppliers outsource all manufacturing.
Would this structure be desirable for biopharmaceutical manufacturing? Although their respective regulatory environments are radically different, the two industries share some common elements: use of contract manufacturers, for instance, and a manufacturer focus on discovery and validation. A key difference is that biopharmaceuticals are manufactured using living organisms, which have yet to be fully characterized, designed, and produced with precision genetic modifications. Therefore, the equivalent design rules and computer-aided design tools that are used in the electronics industry cannot be applied and the manufacturing process must be developed, empirically, for each new product. Bioprocess development requires interaction between manufacturer and creator because biopharmaceutical product characteristics and quality are affected by the manufacturing process.
Reliable and standardized scale-down models would allow large-scale manufacturing and product development to be decoupled. If the large-scale bioreactor is well characterized, a small-scale system can be used to replicate all of its operating conditions and potential variations. Proper scale down requires good simulation of large-scale bioreactor conditions. The small-scale reactor needn’t have the same shape or geometry.
The industry needs technologies that can enable upstream processes to move quickly and seamlessly between partners, whether internal or external. Currently, cultural challenges and established ways of working pose challenges in moving to this model. For example:
• Upstream process performance (as measured by doubling time, viability, and product quality) is sensitive to a large number of parameters (e.g., pH, gas concentration, shear, nutrient feed rates, temperature, etc.).
• Bioreactors themselves, which control this large number of parameters, are large, complex to operate, and have not been standardized.
• Standardization is hampered by use of various platforms and lack of design rules.
For instance, in cell culture, the typical biopharmaceutical company uses at least a dozen different platforms, including shallow-well microtiter plates; deep-well microtiter plates; shake flasks; benchtop bioreactors of various impeller and sparger designs with various culture volumes and larger bioreactors of various impeller and sparger designs with various culture volumes ranging from 100 L–15,000 L.
Generally, production teams choose the platform that affords them the greatest control of the various process parameters. Without design rules for processes, this inevitably results in choosing platforms that are “familiar,” even if they might not be the best choice for the given process. In addition, discovery teams tend to select platforms that are easiest to use and that offer the possibility of high-throughput development.
As a result, biopharmaceutical process development and scale up are rarely linear. Fundamental understanding of the cellular and biochemical processes involved is often sacrificed because it isn’t always clear whether or not the metabolism of the production cells remains the same across various platforms.
In extreme cases, product quality can be affected. Consider the case of Myozyme (alglucosidase alfa), a lysosomal glycogen-specific enzyme indicated for use in patients with Pompe disease. In 2008, FDA rejected Genzyme’s application to produce Myozyme in a 2000-liter-scale facility under the same approval authorization given for its 160-liter-scale plant. The FDA ruling stated that the glycosylation of the products manufactured at each scale differed, and, thus, the 2000-liter product required a new biologic license application (1).
Why is process transfer hard?
Scale up and scale down are well-known barriers to biologics production. Because volume and surface area scale differently with length, even in similar benchtop bioreactors, the physical and chemical environment seen by the cells will be different from what is seen on the industrial scale. The physical and chemical environment of the cells can strongly affect the cells’ physiology and productivity and, hence, must remain constant or above critical values during scaling.
First, the gas transfer rate of O2 and CO2 must be suffciently high so that the dissolved oxygen (DO) level remains above the oxygen uptake rate of the cells, and waste gases such as carbon dioxide are effciently removed.
Secondly, the maximum shear rate seen by the cells must remain the same or below the critical value that affects productivity during translation. This is especially important for mammalian cells such as Chinese hamster ovary (CHO) cells due to their shear sensitivity.
The circulation time is also an important parameter, since it affects the frequency at which the cells see the high shear. The repeated deformation of the endoplasmic reticulum can be detrimental to protein glycosylation. Bioreactors with different chamber volumes will have different circulation times, and hence, some benchtop bioreactors are equipped with a circulation line that allows the physical environment of the cells to mimic the circulation time seen in large industrial-scale bioreactors. When designing scale-down models of bioreactors, the energy dissipation rate has to be maintained constant so that the transfer of internal energy to the cell remains constant.
Decades of research have established these guidelines for process transfer from one platform to another. While not complete, they provide a set of recommendations that are grounded in an understanding of cellular processes and biochemistry. Many important principles of process transfer are rarely followed, however, for the following reasons:
• Microtiter plates and shake flasks are incapable of reproducing more than one (gas transfer) of these parameters (e.g., shear, circulation time, and energy dissipation rate) relative to a production bioreactor.
• Most companies strive to keep only a subset of these parameters constant as they scale upstream processes. A presentation by Biogen illustrated using either the kLa or the power dissipation per unit volume as scale-down parameters (2).
• Maintaining geometric similarity over-constrains the problem so that all of the relevant parameters cannot be held constant across scale. A paper from ETH Zurich (funded by Novartis) demonstrated that geometric similarity fundamentally does not allow for scale invariance (from 3 L to 15,000 L) of kLa, shear stress, and mixing time simultaneously (3).
In response to the increasing need for parallelization and miniaturization of controlled and monitored bioreactors, commercial and academic research groups have developed microbioreactors with working volume below 1 L to:
• Deliver high-throughput, easy-to-use platforms that can replace the microtiter plates and shake flasks used by discovery teams
• Achieve the same gas-transfer rates, shear, circulation, and energy dissipation of 15,000-L production reactors in the high-throughput platform
• Provide backward compatibility to all existing in-line and final assays/instruments performed by process development teams.
State of the art for scale-down
Approaches to miniaturization vary greatly, due to the many different technologies currently being developed. However, advances in microbioreactors are occurring, as the following examples demonstrate:
• pH and DO monitoring are increasingly being integrated at microscale (4,5), as disposable and non-invasive optical pH and DO sensors are validated against the electrochemical probes used by large-scale bioreactors.
• Optical DO sensors have been implemented in shake flasks as a ï¬rst step to allowing the monitoring of conditions in these containers (6, 7).
• pH and feed control are being implemented in shake flasks with integrated syringe pumps for liquid injections and pH probes (8), as shown by Weuster-Botz et al.
• Equipment is being miniaturized for controlling DO and pH (9, 10) by integrating optical pH and DO sensors with 24-well plates and sparging oxygen and carbon dioxide through a permeable membrane at the bottom of the well, as seen with Pall Corporation’s Micro-24 (M24) reactor. The reactor has been demonstrated in microbial cultivations (10), and has also been shown to work for chinese hamster ovary (CHO) cell cultures as well (11).
• Microfluidics is being used in well plates to improve the microtiter plate format in applications such as m2p-labs’ BioLector Pro. This tool enables automated liquid injections to be performed using pneumatic valves in the microfluidic section under the well plate.
• Microbioreactors are also being designed with optical density (OD), conductivity, and pH sensors (12). SimCell’s solution for high-throughput miniaturization is to design 700-μL microbioreactors in the form of cassettes agitated by a rotator. The SimCell model uses a robotic arm to transfer the cassette from the rotator to a sensing platform to measure pH, DO, and OD. These online measurements are supplemented by sample removal for offine measurements of glucose, viability, titer, and product quality (13). The device can use up to 324 cultures simultaneously, allowing it to be used for process optimization, screening, and media optimization (14, 15).
• Most recently, Sartorius Stedim’s TAP ambr 15 and ambr 250 have been gaining traction for scale down. In addition to optical pH and DO sensors, the system combines miniaturized stirred tanks with a pipetting robot and automated feed pumps, greatly improving automation of the fermentation process. By utilizing a biosafety cabinet, the system ensures that sterility is maintained during automated pipetting operations required for feeding, inoculation, and sampling. Both microbial and mammalian processes have been studied in the systems and have shown reasonable correlation with bench-scale stirred tanks (16, 17). These studies have also shown how certain parameters, such as agitation and power dissipation, do not scale linearly with bioreactor size, especially when the stirred-tank form factor is maintained (18).
Nearly all of these microbioreactors are capable of both batch and fed-batch operation. For the needs of conventional biomanufacturing, these cultivation modes are sufficient. However, continuous culture operation, in all its forms--chemostat, turbidostat, and perfusion--can support metabolic flux analysis for increased process understanding (see sidebar). Continuous perfusion holds the promise of improved product quality and consistency because of steady-state operation and because the secreted molecule does not have to be held until the end for harvest.
Realizing such continuous perfusion operation can be challenging for microbioreactors, however. It requires long-term cultures, and places greater stress on sterile interfaces for the continuous in-flow and out-flow of medium and cells.
For continuous perfusion processes to be efficient, the cell density must be high, up to 100 million cells/mL, to convert the nutrient-rich media into protein product, efficiently (19). Such high cell densities can be demanding for gas transport and fluid addition. In addition, removal with cell retention is a challenge. In fact, operation in continuous perfusion culture mode has not been reported for any of the microbioreactors systems described in previous passages.
Scale-down for continuous bioprocessing
For continuous perfusion bioprocessing, a lack of scale-down technology is a major barrier to entry (20). Because typical perfusion process experiments consume 1–3 working volumes per day and can run for 20–60 days, culture media costs alone contribute to prohibitively expensive experimental campaigns. For example, perfusion process development experiments (21) using a 4-L working-volume Wave bag and alternating tangential flow (ATF) or tangential flow filtration (TFF) cell-retention filter, consumed between 55 L and 717 L, with a median of 105 L. With a typical medium cost of approximately $50/L, media costs alone contribute to $5,000 per data point. The high cost associated with running perfusion experiments is mainly due to the working volumes available for current perfusion technology.
Hollow-fiber bioreactors are currently used for small-scale perfusion processing. These consist of semipermeable tubing that allows nutrients to diffuse to cultured cells and removes waste products. FiberCell Systems, for example, provides disposable perfusion cartridges with volumes as low as 2.5 mL. Typical usage of the FiberCell system involves seeding cells in the space outside of the fibers and pumping fresh media through the fibers to maintain a proper environment for cell growth (22). Since cells are typically seeded in the housing of the FiberCell system, it is difficult to measure parameters of the culture environment such as dissolved oxygen and lack of mixing makes cell sampling difficult. External hollow-fiber systems such as the ATF and TFF couple hollow-fiber filters to a conventional bioreactor. These systems circulate cells between the conventional bioreactor and the hollow fibers to allow for more environmental control representative of large-scale production. The tangential flow geometry of these filters, in particular, the alternating tangential flow of ATF systems, helps to prevent fouling of the hollow fiber filters and prolong filter life in high cell density applications. However, these systems are currently unable to operate at scales as small as the FiberCell System. For example, both Pall Corporation’s iCELLis, used for adherent perfusion culture, and Repligen Corporation’s ATF, used for suspension culture, have minimum working volumes of 1 L. Therefore, there is a real need for small-volume perfusion bioreactors capable of supporting the high cell densities desired for continuous perfusion processes.
The primary technology hurdle for existing microbioreactors is the development of high-throughput means for continuous fluid addition and harvesting with cell retention capable of supplying up to 100 working volumes over 30 days. This is difficult for well-plate technologies due to the close proximity of growth chambers, which constrains external fluid connections. In addition, for robotic pipetting architectures, fluid removal with cell retention cannot be accomplished with standard pipette tips.
Microfluidic technology has been applied to continuous culture over the past decade. This technology has been successful primarily for the culture of attachment-based cells and tissue culture (23–25). The large surface area to volume ratio of microfluidic devices facilitates efficient diffusion of nutrients while providing mechanical support for the cells. Unlike hollow-fiber-type perfusion bioreactors, microfluidics allows for the integration of various sensors and also facilitates the precise control of environment. There are a few early examples of continuous suspension culture using microfluidics (26, 27); these examples do not incorporate control of pH and dissolved gases, which is typically required in a bioreactor.
A few commercial microfluidic perfusion systems without active mixing are currently available. The Millipore CellASIC traps cells in a closed chamber and uses microfluidic channels smaller than the size of the cell to deliver media and reagents. The small 2.8-mm diameter culture chambers can be observed using microscopy and can be used for screening and gene-expression experiments (28).
New advancements in the field of microbioreactors
Recently, a microfluidics-based bioreactor system was introduced by Pharyx—a recent start-up out of MIT—for continuous and perfusion processes (Figures 1 and 2). This microbioreactor’s smaller 1–2 mL volume enables long-term continuous experiments with neglibile media usage. In addition, all pumps and control valves are integrated on-chip as disposable elements, greatly simplifying setup and operation. Mixing is performed by moving silicone diaphragms to push the fluid between mixing chambers. The relatively large surface area of the fluid also allows for bubble-free gas transfer through the silicone diaphragms, with kLas greater than 30 h^-1. Bubble-free gas transfer enables online optical density measurements as well, which provides new functionality, such as automated cell bleeding and on-line cell density control. The form factor of the microbioreactor also enables integration of a perfusion filter membrane directly inside the growth chamber, where constant mixing of the culture provides passive filter cleaning, prolonging the lifetime of the perfusion filter.
The original experiments with a similar microfluidic bioreactor focused on the demonstration of various cell-culture modes and on the development of metabolic models (29). Due to the ability to switch between multiple input streams and accurately measure the optical density, pH, and oxygen, a variety of functions were possible. Multiple experiments in chemostat and turbidostat modes with different media compositions were demonstrated in a single device. These experiments proved that modulation of input sources was possible, high-performance liquid chromatography (HPLC) sample collection times could be fast enough to look at dynamics, and control of oxygen during continuous culture could be implemented. Additionally, operation for three weeks without evaporation was demonstrated, all while maintaining sterility.
Modrirez et al. have used microfluidic perfusion devices to study therapeutic protein production using Pichia pastoris (30). A perfusion filter (polyethersulfone) with a 1-cm diameter was incorporated underneath the growth chamber to allow for fluid flow-through while maintaining all of the cells inside the growth chamber and enabling the switching of induction media. The ratio of the filter surface area to bioreactor volume was 0.758 - a factor of three greater than in previously reported surface area measurements in high-performance, bench-scale perfusion bioreactors (21). A combination of perfusion flow and cell density control enabled process optimization of various parameters, such as inducer concentration and perfusion flow rate. Cultures were also shown to be stable, producing consistent levels of both hGH and interferon alfa 2b for more than 10 days.
In-line cell density and precise flow-rate control enabled Han et al. to study genetic switching of Saccharomyces cerevisiae during exponential growth in a steady-state turbidostat (31). Cells were genetically modified to produce different recombinant proteins in response to different chemical inducers and it was shown that by testing the synthetic circuits in turbidostatic steady state, circuit induction increased over 8x and the standard deviation of activation decreased by more than 50% when compared to growth in test tubes.
While continuous microbioreactors are early in their development and commercial deployment, the preliminary data with microbial cell lines demonstrate the capabilities for high gas-transfer rates--supporting the high cell densities required for scale-down models of continuous biomanufacturing, the ability to maintain sterile conditions with continuous in-flow and out-flow, and the ability to support high perfusion rates over a wide range of operating conditions.
A mature perfusion microbioreactor platform has the potential to fill a crucial need in the evolving landscape of biomanufacturing. As with earlier microbioreactors for batch bioprocessing, perfusion microbioreactors could lower the cost for bioprocess development. More importantly, such devices can fulfill multiplexing performance and ease-of-use demands in early-stage development and discovery. Providing development and discovery teams for robust scale-down models in biomanufacturing has the potential to help the industry advance tremendously.
References
1. G. Mack, Nature Biotechnol. 26, pp. 592 (2008).
2. V. Janakiraman et al., “Application of Multivariate Analysis Tools and Design of Experiments (DOE) to Model the Design Space for Characterization of a Mammalian Cell Culture Process,” presentation from the AIChE Annual Meeting (2012).
3. M. Soos et al., “Characterizing Heterogeneity of Environmental Conditions in Various Bioreactor Scales Used for Cell Cultivation,” presentation at the AIChE Annual Meeting (2012).
4. H.R. Kermis et al., Biotechnol. Prog. 18 (5), pp.1047–1053 (2002).
5. M.A. Hanson et al., Biotechnology 97 (4), pp. 833–841 (2007).
6. A. Gupta and G. Rao, Biotechnol. Bioeng. 84 (3), pp. 351–358 (2003).
7. C. Wittmann et al., Biotechnol. Letters 25 (5), pp. 377–80 (2003).
8. D. Weuster-Botz, J. Altenbach-Rehm, and M. Arnold, Biochem. Eng. J. 7 (2), pp. 163–170 (2001).
9. A. Chen et al., Biotechnol. Bioeng. 102 (1), pp. 148–160 (2009).
10. K. Isett et al., Biotechnology 98 (5), pp. 1017–1028 (2007).
11. S.R.C. Warr, Anim. Cell Biotechnol. (1104), pp. 149–165 (2013).
12. A. Buchenauer et al., Biosensors Bioelectronics 24 (5), pp. 1411–1416 (2009).
13. R. Legmann et al., Biotechnol. Bioeng. 104 (6), pp. 1107–1120 (2009).
14. A.P. Russo et al., “Multi-Parameter Process Optimization Using the SimCell System,” in the Proceedings of the 21st Annual Meeting of the European Society for Animal Cell Technology, N. Jenkins, N. Barron, and P. Alves, Eds. (Springer, Dublin, Ireland, 5th ed., 2009), pp. 515–518.
15. Z. Xiao et al., Methods Mol. Biol. (1104), pp. 117–137 (2014).
16. V. Janakiraman et al., Biotechnol. Prog. doi: 10.1002/btpr.2162 (2015).
17. M. Tai et al., Biotechol. Prog. (5), pp. 1388–1395 (2015).
18. A.W. Nienow et al., Biochem. Eng. J. (76), pp. 25–36 (2013).
19. M.S. Croughan, K.B. Konstantinov, and C. Cooney, Biotechnol. Bioeng. 112 (4), pp. 648–651 (2015).
20. E.S., Langer and R.A. Rader, BioProcess. J. (13), pp. 43–49 (2014).
21. M.F. Clincke et al., Biotechnol. Prog. 29 (3), pp. 754–767.
22. W.G. Whitford and J.J.S. Cadwell, Bioproc. Int., (10), pp. 54–63 (2009).
23. M.J. Powers et al., Biotechnol. Bioeng. 78, pp. 257–269 (2002).
24. S. Ostrovidov et al., Biomed. Microdev. 6, pp. 279–287 (2004).
25. M. Reichen et al., PLoS ONE 7 (12):e52246 (2012).
26. F.K. Balagadde et al., Science 309, pp. 137–140 (2005).
27. References 5–9 as originally cited in reference 28.
28. P. Lee, T. Gaige, and P. Hung, Methods Cell Biol. (102), pp. 77–103 (2011).
29. K.S. Lee et al., Lab on a Chip (11), pp. 1730–1739 (2011).
30. N.J. Mozdzierz at al., Lab on a Chip (15), pp. 2918–2922 (2015).
31. N. Han et al., “Microfluidics for Control in Synthetic Biology,” presentation at the 18th International Conference on Miniaturized Systems for Chemistry and Life Sciences, MicroTAS 2014, pp. 312–314 (2014).
Rajeev J. Ram is professor of electrical engineering at MIT.