Discussion
Here we have implemented a novel and broadly applicable approach for media development that relies on rapid, automated construction of diverse media from defined modules of components. We demonstrated the utility of this approach by developing two new media for two phenotypes of interest in the heterologous production of proteins by yeast, namely biomass accumulation and secreted production. We systematically identified and optimized the concentration of media components important to each phenotype of interest. Importantly, defining these new formulations of media did not require advanced analytical capabilities and required minimal experimental time to assess more than 360 total formulations during two to three rounds of optimization for each.
Our optimized formulations affirmed the importance of lipid-related components for maximizing titers in Pichia pastoris cultivations. The importance of optimizing membrane fluidity or lipid metabolism has been well established in CHO and appears to be key to optimizing heterologous protein secretion in P. pastoris cultivation as well (Clincke et al., n.d.; Ritacco et al., 2018; Zhang, Wang, & Liu, 2013).
Modular media blending has four advantages over existing methods. First, the use of common stock solutions and supplements to formulate media reduces initial labor required for new experiments or optimizations ~15 minutes per experiment, making parallel testing of multiple hypotheses efficient and requires less resources overall. Here, we created 30 stock solutions, and evaluated >360 unique media compositions, without manual preparation of individual media or extensive blending calculations or planning. Most of these solutions could be readily reused in future experiments to optimize for new phenotypes of interest. Second, our method requires minimal knowledge of the host organism a priori and could, in principle, be applied to any measurable phenotype of interest. We anticipate that this method could be used to optimize other phenotypes of interest, such as glycosylation profiles. Third, our method provides certain practical advantages, including minimal requirements for analytical characterization and rapid identification of component interactions that lead to solubility challenges. These traits make it possible to learn about formulations that may lead to extensive precipitates like those encountered with our rich defined medium formulation (Figure 4A ). Finally, modularly constructed media, such as DM2, can be ~70% pure water with low osmolarity, leaving volumetric and osmotic space for future modifications to accommodate new or related phenotypes of interest (Figure 4B ).
We also acknowledge certain limitations in the present study that may be addressed in future work. First, while modular media development identifies components key to the optimization of the phenotype of interest, additional media optimization effort may be necessary to translate these learning in batch cultivations to scaled-up fed-batch or perfusion operation, where additional variables such as supplemental feed composition and feeding schedule must also be considered. In principle, modular media construction could be applied to high-throughput scale-down cultivation models, such as Ambr250s. Second, our approach for optimization relies on greedy algorithms tailored to create a new media for a single phenotype of interest; however, given the vast explorable media space it is possible to find a local optimum. Further metabolic or -omic modeling techniques could be employed to guide broader exploration of media space, co-optimize multiple phenotypes, or facilitate biologically informed optimization, albeit with more complex experimental and computational requirements (Matthews, Kuo, Love, & Love, 2017b; Mohmad-Saberi et al., 2013). Third, our current method used commercially available supplements, but in practice, beneficial supplements could be simplified by using individual components, to facilitate more biological inferences and aid development of improved host-specific supplements. Finally, initial screens to identify beneficial supplements rely on reasonable choices of initial concentrations for screening. These currently require prior knowledge from the literature or commercial sources; with further use in the community of the Openblend approach, it is possible additional sharing of knowledge could help inform further developments.
The improved speed and accessibility of in-depth media development experiments enabled by modular media construction could help improve expression of many classes of proteins in laboratories and discovery centers that have not traditionally had access to such capabilities. Since many lead candidates for new therapeutic proteins begin in small biotech firms and academic labs, early-stage improvements in productivity could help advance more proteins towards the clinic simply by facilitating access to larger quantities of proteins for initial research and non-clinical studies. In more established companies, the ability to make rapid improvements to existing media may enable faster product development timelines and could reduce manufacturing costs overall. Rapid identification and optimization of sensitive media components could also enable easier adoption of a range of industrially relevant alternative hosts, resulting in further manufacturing flexibility and potentially cost savings (Coleman, 2020).