Creating a testing control strategy and limits associated with the tests chosen should consider data beyond just that provided by clinical trials alone. This presentation describes how to use a patient-centric philosophy in understanding what is critical to the patient utilizing information such as prior knowledge, in vitro testing, in vivo studies, and molecular and in silico modeling to create an appropriate control strategy.

Continuous process improvement is an inherent and integral part of a commercial product lifecycle. Such process lifecycle changes often require updates to control strategy and a subsequent PPQ. Scenarios differing in scope of process change will be discussed to highlight the need to have a case-by-case approach for defining the scope and extent of characterization and PPQ, such that the journey from characterization to successful PPQ is efficient and effective.

The scope of this study was feasibility assessment of Near Infrared (NIR) Spectroscopy to monitor bacterial fermentation. Initially, calibration models were built using NIR spectra collected off-line. Models for optical density prior to media inoculation, as well as for protein concentrations and pH during fermentation, showed that NIR spectra can be used to predict process parameters.

This presentation describes the design and assembly of an end-to-end continuous monoclonal antibody manufacturing testbed at MIT. The testbed is fully automated and includes extensive inline and at-line process analytical technology, including Raman spectroscopy, high-performance liquid chromatography (HPLC), and liquid chromatography-mass spectrometry (LC-MS). The presentation describes use of the testbed for the evaluation of data analytics and machine learning methods, validation of mechanistic models, and real-time feedback control of CQAs.

Analytical assays for residual host cell proteins (HCPs) characterization are currently considered as one of the weakest analytical links in the total characterization and control strategy for biotherapeutics development. The applications of quantitative LC-MS-based proteomics and targeted quantification by MRM (multiple reaction monitoring) approaches for upstream clone selection, downstream process development, and root cause investigation will be highlighted for biologics processing and product characterization.

A quantitative MS-based method was developed to identify variants and PTMs in biotherapeutics down to 1% without prior knowledge of the modification. The method utilizes stable isotope-tagged reference standard, a version of the analyte protein that is uniformly labeled with 13C6-arginine and 13C6-lysine. SITRS serves as an internal control that is trypsin-digested and analyzed by LC/MS with the analyte sample. The method is useful in understanding protein degradation sites.

Freezing and thawing are common unit operations in the production of biologics that potentially impact CQAs of such molecules. Stability aspects of freezing during processing and storage must be evaluated based on knowledge of underlying mechanisms to design robust freeze/thaw unit operations. Challenges and potential approaches related to instability of antibody-based molecules during freezing storage and kinetic aspects of freeze/thaw on product quality are highlighted through two case studies.

Multivariate data analysis is essential in process analytical technology to maintain unit operations that meet quality control, and glean process knowledge. A novel projection pursuit machine learning method is presented that analyzes three-dimensional datasets comprising N-batches of experiments that measure p-variables over a sample time, T. Without unfolding the data format, feature extraction, hypothesis testing, and identifying critical process parameters are learned in a single-step automated workflow.