Introduction
Achieving high volumetric productivities of biologic drugs in cultivation is a key step in advancing candidate biologic drugs. The outcome of this effort ultimately impacts manufacturing costs as well as readiness for transitioning clinical-stage development (Love, Love, & Barone, 2012). The development of standard, chemically defined media for established manufacturing hosts, such as CHO, has made such transitions efficient for monoclonal antibodies by achieving high biomass accumulation, cell viability, operational consistency, and specific productivities, streamlining development efforts (McGillicuddy, Floris, Albrecht, & Bones, 2018; Rodrigues, Costa, Henriques, Azeredo, & Oliveira, 2012). Nonetheless, optimizing productivity or quality attributes for a specific product often still requires further refinement of media (Ritacco, Wu, & Khetan, 2018). Such development may require evaluating dozens of variants derived from a common standard formulation to address the specific challenges encountered (Gagnon et al., 2011; Loebrich et al., 2019). Media development for entirely new biomanufacturing technologies, such as alternative hosts (Matthews, Kuo, Love, & Love, 2017a) or new product modalities (Lu et al., 2016), may also require new formulations or extensive optimizations due to limited prior knowledge.
Common approaches to develop a medium to optimize a phenotype of interest are often labor intensive, low throughput, or rely heavily on extensive analytical capacity (Galbraith, Bhatia, Liu, & Yoon, 2018). For example, analysis of residual media after cultivation requires extensive capabilities for analytical characterization and prior experience with the manufacturing host to identify potentially limiting or toxic media components (Mohmad-Saberi et al., 2013; Pereira, Kildegaard, & Andersen, 2018). As a result, optimizations can be slow and iterative. Furthermore, for an alternative host such asKomagataella phaffii (formerly known as Pichia pastoris ), there is substantially less, if any, prior knowledge available to establish profiles for residual components in media after fermentation. Other analytical techniques like RNA-seq combined with methods for reporter metabolite analysis can guide media optimization, to generate testable hypotheses regarding beneficial modifications to media (Matthews et al., 2017a). Such genome-scale approaches, however, require prior host-specific knowledge, such as well-annotated genomes, and are still limited by slow iteration and labor-intensive preparations of new media to test the hypotheses generated from computational analyses.
Alternative strategies for blending basal components for media allow linear combinations of existing media to explore many variations rapidly (Jordan, Voisard, Berthoud, & Tercier, 2013). While this approach avoids slow iterative analyses, the typical experiment is labor intensive to perform, often requiring independent preparations of over a dozen stock media to combine (Rouiller et al., 2013). Similar to analytical-based approaches for optimization, the selected variations of media are simultaneously guided and constrained by prior experience and media designs, which may limit the breadth of components examined (Kennedy & Krouse, 1999). For less established hosts with fewer available formulations of media, media blending may also require fullyde novo formulations for initial studies. Further complicating such designs, different and new components for media can present challenges in solubility or unanticipated interactions with other elements in the formulations (Ritacco et al., 2018). New approaches to blending could, however, enable fast, flexible experimentation and minimize the time, labor, and analytical development needed initially to optimize media for new applications and phenotypes.
Here, we present a novel and generalizable approach for the modular development of media and demonstrate its use to create optimized media for two different phenotypes—cellular growth and recombinant expression of a protein (as measured by the secreted heterologous protein titer) from Pichia pastoris . Our approach comprises two modular parts for blending and optimization. We determined that a set of simple concentrated stock solutions constructed in defined modules could generate many media by blending or dilution. We then automated a simple, inexpensive liquid handling system (Opentrons OT-2) to enable high-throughput screening for the effects of diverse media on a phenotype of interest in milliliter-scale batch cultures. To maximize the benefit of this automated blending, we also developed an algorithmic framework for systematic modular media optimization, beginning from a simple minimal media (here a YNB-based one). This framework provides insights pertaining to key media components during stages of optimization, as well as overall mapping of the design space for the media. In the examples presented here, the resulting defined media developed with this strategy outperformed commonly used BMGY and BMMY complex formulations for biomass accumulation and secreted heterologous protein production.