(582ac) Application of Multivariate Analysis of Metabolic Models for Predictability of Cell Culture Performance and Quality Attributes; An Industry Prospective

Authors: 
Migliore, N., Janssen Pharmaceutical Companies of Johnson and Johnson
Tomlin-Verni, K., Janssen Pharmaceutical Companies of Johnson and Johnson
Mehrman, S., PDMS
Bruwer, M. J., ProSensus Inc.
Schaefer, E., Janssen Pharmaceutical Companies of Johnson and Johnson



Cell culture processes are highly multivariate in nature yet to date much of the analysis and optimization of these processes has been univariate.  Traditional multivariate techniques can be applied to analyze and optimize cell culture processes as well as increase the predictability of product quality attributes by correlating the outcome of a process to multiple inputs and process parameters as they vary over the course of the batch.   PLS models can be built using primary input and output parameters or as combinations of primary parameters and parameters derived from unique fundamental principles such as engineering correlations or metabolic relationships.   In this presentation the principles of PLS modeling have been applied to multiple comprehensive datasets with the following objectives:

  1. Combine a comprehensive dataset for fed-batch CHO processes including in-line measurements of pH, dissolved oxygen, and temperature averaged at hourly rates; at-line measurements of cell density and extra-cellular metabolites performed daily; off-line measurements of complex intra and extra cellular metabolites performed at critical points of inflection as defined by the process; off-line quality analysis of the mAb expressed including glycosylation, charge variants, and purity at completion of the process.
  2. Look at variations to this dataset which would include derived parameters from both engineering calculations and metabolic signatures of the culture.  The goal of developing metabolic signatures is to be able to use a minimum number of measured metabolites which can identify different cell metabolic states as they relate to the PLS model.
  3. Determine a predictable set of latent variables from in-process data that correlate with variation in the process and predict variability in product quality attributes namely glycosylation, charge variants, and purity.
  4. Identify the key individual measurements driving the model in predicting the quality attributes named above.
  5. Implement in-line or at-line cell culture measurement techniques, to collect more frequent data points for the key individual measurements correlating to predictability of quality attributes and implement at-line cell culture measurement techniques, such as amino acid analysis and TCA cycle intermediates, where in-line solutions are not practical

This presentation will discuss how to determine key cell culture measurements required to build a predictable model for quality attributes.   This presentation will also discuss which approaches may be robust for routine process development and which assays are most critical to be run for routine process development, scale up and troubleshooting.