Monitoring by Raman spectroscopy of rabies virus-like particles production since the initial development stages

Abstract

BACKGROUND This study aimed to establish chemometric models using Raman spectroscopy data for biochemical monitoring of rabies Virus-Like Particles (VLP) production based on baculovirus/insect cell system. The models were developed using fresh and stored samples from the initial development stages (Schott culture flasks). The following modeling techniques were assessed: partial least squares (PLS) and artificial neural networks (ANN). The effects of spectral filtering approaches, spectral ranges (400–1850 cm−1; 100–3425 cm−1), and sample cryopreservation were also considered. The applicability of the models was evaluated using experimental data from assays carried out in a benchtop bioreactor.

RESULTS The results showed that the prediction capacity of the chemometric models was negatively impacted when samples from rabies VLP production were cryopreserved. Further studies are needed to confirm the maximum storage time for samples (< 4 months) without a significant difference in model predictions compared to those from an in line database. The dilution of the sample should be kept constant throughout the rabies VLP development stages. A nonlinear correlation was observed between dilution and the predicted values of biochemical parameters from Raman spectral data. The choice of spectral filtering has a major impact on the prediction accuracy of chemometric models.

CONCLUSION The optimal filtering approach should be individually optimized for each biochemical parameter. The ANN models were significantly more suitable for biochemical monitoring than the PLS approach. The 400–1850 cm−1 Raman shift range is recommended for biochemical monitoring of rabies VLP using a baculovirus/insect cell platform when samples are cell-free.