In biopharmaceutical manufacturing, optimizing chemically defined media (CDM) is critical for enhancing cell growth, productivity, and product quality in mammalian cell cultures. Traditional screening methods, such as one-factor-at-a-time or factorial designs, tend to be labor-intensive. Media blending offers a simpler alternative—creating concentration variations through liquid mixing of pre-existing “mother” media. Blending, however, is rarely used because it lacks a standardized workflow for experimental design.
In a recent paper, Hirotaka Kuroda and Kazuya Sorado, of Osaka University and Shimadzu, and Noriko Yamano-Adachi and Takeshi Omasa, of Osaka University and the Manufacturing Technology Association of Biologics, address this gap with a mathematically precise approach to media blending that has yielded a systematic workflow for initial component screening.
Multicollinearity is a key challenge, they note, because the number of mother media typically falls below the number of components (typically more than 50 in CDM), causing rank deficiency and correlated variables that hinder regression model interpretation. To overcome this, their workflow uses principal component analysis (PCA) to reduce dimensionality, projecting data into an orthogonal space that retains maximum variance while eliminating redundancy. It then selects media blends using D-optimal design principles, maximizing the determinant of the information matrix to minimize correlations in non-redundant space.
Three-step workflow
“The workflow consists of three steps: the experimental design, the cell culture experiment, and regression modeling and interpretation,” they point out.
Kuroda and colleagues assessed 120 experimental conditions using 11 different CDM that are relevant for Chinese hamster ovary cells (CHO), culturing IgG1-producing Chinese hamster lung (CHL-YN) cells, which grow faster than CHO cells. Their results accounted for variations in viable cell concentrations that ranged from 5.8 to 19.4 x106 cells/mL and enabled them to design dedicated media sets for blending, as well as a systematic workflow that addresses optimization from blending design to analysis. In the end, they identified a dozen key components and their individual effects on cells.
Such insights, they suggest, may help manufacturers better optimize specific concentration ranges for key components in commercial media, even without extensive prior knowledge.
The study also simulates dedicated mother media sets with disclosed, uncorrelated compositions, dramatically reducing multicollinearity and enhancing interpretability. With Python code for executing this workflow available on GitHub, this framework democratizes media blending, enabling rapid screening without custom media preparation.
Kuroda and colleagues caution that results need careful interpretation, however. Their next step is to automate the pipetting used in this workflow.
