Chemometrics for etanercept bioprocess monitoring in the PAT context

CHIAPPINI F; ALCARAZ M R; FORNO G; GOICOECHEA H C

Congreso; 26th ESACT Meeting: Cell Cculture Technologies: Bridging Academia and Industry to Provide Solutions for Patients.; 2019

Background and noveltyIn the last years, regulatory agencies in biopharmaceutical industry have promoted the design and implementation of Process Analytical Technology (PAT) in order to improve quality monitoring of bioprocesses. In this context, the use of spectroscopic techniques for data generation in combination with chemometrics has enabled the design of alternative analytical methods for on-line critical process variables prediction. In this work, a novel multivariate strategy for Etanercept (Et) at-line concentration prediction in perfusion processes from spectral data has been developed.Experimental approachA sterile sample from standard Et process was acquired daily for viable cells count and determination of Et concentration in supernatant by standard off-line univariate techniques. Simultaneously, fluorescence Fluorescence excitation-emission matrices (EEM) were simultaneously obtained as (second second-order data). The EEM were alternatively analyzed modeled by different chemometric modelsalgorithms. Firstly, unsupervised decomposition methods Parallel Factor Analysis (PARAFAC) and Multivariate Curve Resolution (MCR) were considered to qualitatively analyse the spectral information. Afterwards, two multivariate regression strategies for Et prediction based on Unfolded Partial Least Squares (U-PLS) and Back Propagation Neural Network (BP-ANN) models were developed and compared.Results and discussionPARAFAC components were putatively related to biological fluorophores present in the culture media during fermentation of CHO cells (aromatic aminoacids, pyridoxine, flavin, folic acid and NAD). On the other hand, the spectral data showed good correlation with Et concentration after building the regression models. In particular, the prediction performance of BP-ANN-based model was much better due to the non-linear data structure, obtaining relative mean errors around 10%. This novel strategy would represent a faster and cheaper approach for Et monitoring, aiming to achieve PAT goals.Acknowledgements and FundingCONICET and ANPCyT.BibliographyMercier et al. Eng. Life Sci. (2016) 16, 25-35.