Design of approach for modular media blending
We sought to develop an approach capable of identifying important, beneficial modifications for media tailored to a given phenotype of interest. We reasoned that key requirements for such an approach would be that it is fast and automatable, with minimal dependence on complex analytical assay development. Such features would enable routine application to any measurable phenotype of interest. In general, media blending allows both speed and low analytical complexity. We aimed to retain these features while minimizing the labor and constraints on compositions imposed by linear combinations of fully formed and unique media. We reasoned that diverse and flexible blends of media could be created by defining simple concentrated stock solutions as basic modules to combine further. These modules would comprise individual components or common subsets of components with compatible solubilities (e.g. YNB). If media components could be formulated in concentrated stock solutions that could be stored stably over time, then the components could be routinely and interchangeably combined and diluted to the desired final concentrations. This approach would yield a broadly applicable modular strategy for media blending amenable to conventional liquid handling automation.
To test the feasibility of this approach, we first assessed whether many common media components could be formulated in concentrated stable aqueous stock solutions. Using the CHO medium eRDF as a reference, we estimated the solubility of each component of this medium, using data from AqSolDB as well as other online sources (Combs, 2012; FSA Panel on Additives and Products or Substances used in Animal Feed (FEEDAP), 2011; Ritacco et al., 2018; Schnellbaecher, Binder, Bellmaine, & Zimmer, 2019; Sorkun, Khetan, & Er, 2019; Yamamoto & Ishihara, n.d.). We compared the estimated solubility of each media component to its concentration in eRDF and found that, individually, most media components are soluble at levels >10x higher than their eRDF concentration (Figure 1A ). The existence of a wide range of commercially available concentrated supplements further supports this result: >50x concentrated solutions of amino acids, vitamins, lipids, and trace metal supplements are common and commercially available.
Next, we used the product information of commercially available supplements, literature sources, and inspection to estimate the percentage of eRDF media components that could be stored in stable solutions for >6 months. We estimated that over 75% of eRDF components met this criterion (Figure 1B ). To address stability challenges caused by less stable components, we reasoned that less stable components or supplements, such as vitamins, could be prepared, aliquoted, and stored frozen for long-term storage (Schnellbaecher et al., 2019); these aliquots could then be thawed and used within a defined period to mitigate component instability and enable their integration into our modular blending strategy. Together, these solubility and stability data suggested that a modular approach to media development could be defined in this way to accommodate a range of new formulations easily.
We next automated the process for constructing media, using the Opentrons OT-2. We chose this liquid handler due to its low cost, reliability, and compatibility with simple formats for data input, such as Excel spreadsheets. We then created an open-source Python package, named Openblend, which simplified the media construction process by handling routine experimental design and execution steps (Figure 1C). Openblend creates an experimental design spreadsheet, specifying the number of 24 well plates, the desired media composition of each well, and stock solution names and concentrations. The script then checks the feasibility of the experimental design, ensuring that the total volume of each well will not exceed the target volume and avoiding the addition of sub-microliter stock solution volumes. If the design passes this assessment, the script then outputs a new spreadsheet containing the setup for the OT-2 deck and required volumes of stock solutions, providing a user with instructions on how to setup the OT-2 liquid handler. We found that our typical time to execute this script, setup the OT-2 and initiate plate building was ~15 minutes, and the time for the automated steps was about two hours.
Finally, we defined a modular approach for optimization to effectively leverage the Openblend tool (Figure 1D ). Beginning from an initial basal medium, improved media are constructed through successive rounds of optimization. In each round, a library of media components and supplements are screened to identify beneficial additives. These additives are then screened in combination and over a range of concentrations to further optimize the performance of the medium. Each modular addition and optimization of additives can be guided simply by measurements of the phenotype of interest (e.g. biomass accumulation). This greedy approach to multi-dimensional optimization could continue iteratively until the resulting media met desired specifications, all available media components were explored, or no additional gains in performance realized.