Spectroscopy integration to miniature bioreactors and large scale production bioreactors–Increasing current capabilities and model transfer

Abstract

Spectroscopy techniques are being implemented within the biopharmaceutical industry due to their non-destructive ability to measure multiple analytes simultaneously, however, minimal work has been applied focussing on their application at small scale. Miniature bioreactor systems are being applied across the industry for cell line development as they offer a high-throughput solution for screening and process optimization. The application of small volume, high-throughput, automated analyses to miniature bioreactors has the potential to significantly augment the type and quality of data from these systems and enhance alignment with largescale bioreactors. Here, we present an evaluation of 1. a prototype that fully integrates spectroscopy to a miniature bioreactor system (ambr ® 15, Sartorius Stedim Biotech) enabling automated Raman spectra acquisition, 2. In 50 L single-use bioreactor bag (SUB) prototype with an integrated spectral window. OPLS models were developed demonstrating good accuracy for multiple analytes at both scales. Furthermore, the 50 L SUB prototype enabled on-line monitoring without the need for sterilization of the probe prior to use and minimal light interference was observed. We also demonstrate the ability to build robust models due to induced changes that are hard and costly to perform at large scale and the potential of transferring these models across the scales. The implementation of this technology enables integration of spectroscopy at the small scale for better process understanding and generation of robust models over a large design space while facilitating model transfer throughout the scales enabling continuity throughout process development and utilization and transfer of ever-increasing data generation from development to manufacturing.

A Novel Approach for Non-Invasive Continuous In-Line Control of Perfusion Cell Cultivations by Raman Spectroscopy

“…The good model performance of glucose agrees with previous findings with a similar spectrometer/probe/measurement chamber combination, which was integrated into an Ambr®15 system ( Rowland-Jones et al, 2020 ). The in-line application of a Raman probe into the cell-free harvest stream of a perfusion process allowed for an excellent signal to noise ratio as no cells (i.e., scattering particles) are present in the sample.…”

“…First, the concentration range is relatively low. Experience gathered in previous trials has shown that at least ten times the model error (of a valid model) is needed for model building, so at least 2 g/L ( Rowland-Jones et al, 2020 ). Second, most data-points are roughly between 0.4 g/L and 0.6 g/L.…”

“…In this case, the dataset was split into five groups, each consisting of one batch, thus mimicking the model’s later use to predict future perfusion runs. For further information on the data-pretreatment, modeling procedure and especially the cross validation, please refer to Rowland-Jones et al (2020 ).…”

Online 2D Fluorescence Monitoring in Microtiter Plates Allows Prediction of Cultivation Parameters and Considerable Reduction in Sampling Efforts for Parallel Cultivations of Hansenula polymorpha

“…The implementations of PLS regression used for the online monitoring and control of bioprocesses in stirred tank reactors are numerous and have extensively been reviewed in the literature [ 26 , 39 , 40 ]. Yet, to date, the means for the online recording of high-dimensional data in microbioreactors are limited, and applications of PLS modelling are mostly based on spectral at-line measurements [ 19 , 20 , 41 , 42 , 43 ]. Recently, Sawatzki et al [ 44 ] used PLS regression to predict the endopolygalacturonase activity during fed-batch cultivations of Saccharomyces cerevisiae in 8–12 mL stirred cultures.…”

Advancing Raman model calibration for perfusion bioprocesses using spiked harvest libraries

“…[13] Furthermore, a miniatured bioreactor system allowing up to 48 simultaneous cultivations at different conditions was shown to facilitate Raman model calibration for reliable glucose and lactate predictions. [14] Spiking strategies to augment analyte variability in the calibration data set were demonstrated in situ, [12,15] at-line, [16] and offline. [17] In contrast to the standard cali- of product titer in perfusion harvest using a stainless-steel flow cell was used to optimize an integrated continuous chromatography step, [18] whereas others developed a more sophisticated flow cell intensifying Raman signals to detect product breakthrough in a continuous protein A capture step.…”