Support Vector and Locally Weighted regressions to monitor monoclonal antibody glycosylation during CHO cell culture processes, an enhanced alternative to Partial Least Squares regression

Daniel Arturo Zavala-Ortiz,Bruno Ebel,Meng-Yao Li,Dulce María Barradas-Dermitz,Patricia Margaret Hayward-Jones,María Guadalupe Aguilar-Uscanga,Annie Marc,Emmanuel Guedon

doi:10.1016/j.bej.2019.107457

Abstract

Since monoclonal antibodies (mAb) are sensitive to the manufacturing process, several mAb variants can be the result of a single batch production. The most critical source of heterogeneity is glycosylation which has a profound impact on safety and efficacy of the final product. Implementation of monitoring and control of the process using the Quality by Design (QbD) approach may help to ensure mAb specifications, although its implementation is limited by the availability of real-time specific measurements. All current approaches to elucidate mAb glycoforms require sampling and labour-intensive efforts. Thus, glycosylation analysis is often performed with the objective of detecting quality defects at the end of the culture process. In this work, the capability of Near Infrared spectroscopy and chemometric treatment to accurately monitor mAb glycosylation during CHO cells cultures using in situ probes is shown for the first time. Real-time monitoring of glycosylation, in terms of high mannose isoforms, fucosylated, sialylated and galactosylated isoforms as well as non-glycosylated mAb, has been successfully performed by the novel use of Locally Weighted Regression (LWR) and Support Vector Regression (SVR). These encouraging results open the way for the implementation of control systems on the impact of cell culture operating parameters on mAb heterogeneity, particularly glycosylation, during CHO cell culture processes through the QbD approach.

Highlights

Monoclonal antibodies produced in animal cell culture processes represent a success in terms of clinical benefit for patients and revenue generated by biopharmaceutical industries
Such molecules are quite sensitive to changes in manufacturing processes and several monoclonal antibodies (mAb) variants could be produced within a single batch due to posttranslational modifications
The monitoring of both macro- and micro-heterogeneity of glycosylated mAb was improved by the novel use of sample space-based regression methods, Support Vector Regression (SVR), that could handle non-linear relationships between glycoforms and spectra

Summary

Introduction

Monoclonal antibodies (mAb) produced in animal cell culture processes represent a success in terms of clinical benefit for patients and revenue generated by biopharmaceutical industries. Such molecules are quite sensitive to changes in manufacturing processes and several mAb variants could be produced within a single batch due to posttranslational modifications. The control of glycosylation specific profiles of mAb during the process is critical for therapeutic efficacy and patient safety For this purpose, regulatory agencies proposed the Process Analytical Technology (PAT) strategy to control pharmaceutical manufacturing processes through the continuous adjustment of Critical Process Parameters (CPP) which affect Critical Quality Attributes (CQA) of the product. Continuous monitoring of CQA, such as mAb glycosylation, is required to establish advanced retro-control systems to guarantee mAb specifications [1]

Methods

Results

Conclusion