The bioreactor’s environment and media contents affect the metabolism and productivity of a cell line. Traditional analytics like pH, temperature, dissolved gas levels, and titer reveals useful but incomplete information about the cells’ metabolism. Extracellular metabolites and nutrients in the cell media can be monitored to provide some information about the health of the cells. Additionally, that info can assist with identifying molecules that may promote or inhibit the cells’ metabolism and productivity, yielding insight into how to optimize future processes. For instance, lactate has long been known to be an inhibitor to a culture’s health, and it is closely monitored to ensure that levels are kept within an appropriate range. Also, monitoring amino acid levels in basal media and feed supplements confirms that their concentrations are suitable to promote a healthy and productive process.
A collaboration between the biologics process development team at Bristol-Myers Squibb and Seongkyu Yoon’s group at UMass Lowell investigated how amino acid and metabolite levels monitored by LC-MS coupled with transcriptome analysis could be used to identify promoters and inhibitors early during process development. They sought to perform multivariate analysis to correlate the changes in metabolite levels to specific metabolic pathway regulation. This would allow for faster optimization of an appropriate media strategy for increased titers. The teams used a CHO-K1 GS knockout cell line fed the same basal media across four 5-L bioreactors in a 14-day fed-batch process. In those four reactors, two levels of a selection agent (methionine sulfoximine) and two different feeds (standard and fortified with serine, threonine, tyrosine, and lysine) were evaluated. Daily samples were frozen and run after the processes were completed with a combination of LC-MS and HILIC-MS for metabolomics analysis.
The metabolomics analysis tracked over 150 metabolites. Additionally, UPLC with derivatization measured the amino acid levels from the same samples. The combined data were imported into the SIMCA software package for the model generation to identify the most differentiating molecules. The top molecules were correlated to the CHO KEGG pathways to determine their role in metabolism. Finally, transcriptome analysis provided by RNAseq from day 6 and day 10 samples were used to correlate changes with the identified metabolic pathways.
The results from the initial assessment led to the discovery of 8 metabolites which exhibited promoter effects on titer. These metabolites included fatty-acid like molecules and derivatives of amino acids like homocysteine and prolyl-hydroxyproline. Transcriptome analysis revealed the upregulation of some amino acid, TCA cycle, and fatty acid pathways. When reviewed, these pathways were believed to be critical for cell growth and productivity during the stationary phase of cell growth. It was hypothesized that designing a media supplement to support these pathways may increase titer.
The findings were implemented in a new feed media designed with higher levels of histidine, glutamic acid, leucine, methionine, and tyrosine. These five amino acids were selected since they were believed to support the detected promoters, decrease inhibitors, or increase gene expression associated with higher protein production. To fully validate this approach, the team developed another feed media with higher levels of aspartic acid, cysteine, isoleucine, serine, threonine, and tryptophan, which were all believed to increase inhibitor effects which should lower protein production. When the two new media feeds were used in subsequent bioreactor runs, the hypothesis was validated. The “promoter” media increased titer by 8.2% while the “inhibitor” media decreased titer by 7.1% compared to the original run. In summary, the big data approach led to meaningful changes to cell media design after just a four parallel bioreactor runs.
As upstream process development teams continue to look towards improving predictability, incorporating new methods of lab informatics and faster optimization will continue to be a focus. Big data approaches, like the one discussed in this study, may accelerate drug-pipelines to ensure that once a biologic is moved into production, the process has already been engineered to deliver high yields and batch-to-batch consistency.
By, Glenn A. Harris