Date of Graduation

8-2022

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Engineering (PhD)

Degree Level

Graduate

Department

Chemical Engineering

Advisor/Mentor

Ranil Wickramasinghe

Committee Member

Xianghong Qian

Second Committee Member

Daniel Strauss

Third Committee Member

Bob Beitle

Fourth Committee Member

Mathias Ulbricht

Keywords

Bioreactor clarification, Fouling, Perfusion cell culture, Tangential flow filtration

Abstract

Tangential flow filtration has many advantages for bioreactor harvesting as the permeate could be introduced directly to the subsequent capture step, the process is easy to scale up, and fouling of the filter is limited by the cross flow. However, membrane fouling has limited its widespread use. This is particularly problematic given the high cell densities encountered today. Here a reverse asymmetric commercial membrane, BioOptimal™ MF-SL (Asahi Kasei), where the more open surface faces the feed stream, and the tighter barrier layer faces the permeate stream, has been investigated for bioreactor harvesting. The open surface contains pores up to 40 µm in diameter, while the tighter barrier layer has an average pore size of 0.4 µm.

The filtration performance, including fouling analysis conducted in this dissertation, involves using different feed streams, comparison of the filter performance with other filters possessing different membrane structures, mathematical modeling to predict the flux and fouling, fouling visualization using confocal laser scanning microscopy, and fouling identification using liquid chromatography-mass spectrometry.

For the feed streams studies, filtration of yeast suspensions and Chinese hamster ovary cell culture has been conducted under various conditions. The yeast cells are trapped in the open pore structure, while CHO cells are more externally deposited. The membrane stabilizes an internal porous cake that acts as a depth filter. This stabilized cake layer removes particulate matter that fouls the barrier layer, protecting the fine pores from the large aggregates. As filtration continues, a cake layer forms on the membrane surface.

Resistance-in-series model has been developed to describe the permeate flux during tangential flow filtration. The model contains three fitted parameters, which can easily be determined from constant pressure normal flow filtration experiments and total recycle constant flux tangential flow filtration experiments. The model can be used to estimate the filter's capacity for a given feed stream. Our results suggest that using a reverse asymmetric membrane could avoid severe flux decline associated with fouling of the barrier during bioreactor harvesting.

Laser scanning confocal microscopy is used to observe the location of particle entrapment. The throughput of the reverse asymmetric membrane is significantly greater than the symmetric membranes. The membrane stabilizes an internal high permeability cake that acts as a depth filter. Confocal imaging helps visualize the secondary membrane directly by staining the DNA and membrane proteins using fluorescent dyes.

Host cell proteins are the most challenging impurities for downstream purification processes. In order to investigate the fouling during cell clarification, HCPs in the bioreactor, harvest, and backwash are identified and quantified using different methods. A dataset is established using the identified HCPs and used to train the deep learning model. The model predicts unknown HCPs on fouling potential with an accuracy of 76%. The dataset of identified HCPs in this study provides insights into the characterization of membrane fouling, membrane selection, and process development. This approach could be used to screen cell lines or hosts to select those with reduced HCP profiles or identify HCPs that are problematic and difficult to remove.

Share

COinS