(15b) Deploying Automation to Execute Real-Time Feed Strategies in Bench-Scale Bioreactor Systems

Kristin O’Neill, William Tran, Linda Hoshan, Sen Xu, and T. Craig Seamans

Biologics Process Development and Clinical Manufacturing, Merck Research Laboratories,

2000 Galloping Hill Road, Kenilworth, NJ 07033

Key to effective upstream process development is the presence of resource-efficient systems for examining complex process options. In this talk, we report on the progress of a completely automated sampling and feed control system, capable of simultaneously evaluating different cell density and metabolic-based feed strategies for bench scale cell culture processes. Though there are automated feed systems available for micro-scale vessels, to date there is not a complete 3-L scale sample and feed automation solution as described here, with capability to transfer small volume samples from multiple bioreactors (single-use or glass) to a suite of independent analyzers with concurrent feedback control of nutrient feed pumps.

The automated on-line sampling and feed system is capable of feedback control for a variety of complex feed strategies that incorporate data inputs from integrated third-party analyzers. Automated feeds can be delivered based on time as well as calculated with inputs from metabolite and cell density measurements. The completely automated system has supported experiments with two to eight 3-L bioreactors running in parallel, sampling each tank one to six times per day, and executing up to 4 independent feed pumps per vessel depending upon experimental needs.

In a recent typical experiment, eight bioreactors were sampled three times per day with feedback control for two separate nutrient feed solutions per vessel, based upon independent metabolite triggers. Over the course of the 14 day run, each bioreactor triggered feeds 35-40 times with feed volumes ranging from 2-47 mL, for a total of over 300 automated feeds across the eight vessels. In another case study, two bioreactors were sampled twice per day with feedback control for four separate nutrient feed solutions per vessel. In this study three different feed algorithms were employed with calculations utilizing inputs from metabolite and cell count analysis as well as real-time working volume data from integrated balances. Each vessel received ~80 automated feeds over the two week culture duration.

Increased use of automation for physical operations can increase efficiency in executing experiments but can shift the bottleneck to data capture. This issue was addressed with refined data extraction templates which streamlined data imports into an electronic notebook system and PI integration which enabled remote monitoring of sample analysis and feeding.