Optimizing Cell Line Development for High-Quality Biologics

View PDF

For a host-cell system to generate high yields of recombinant proteins and other entities, cells must be derived from optimized and stable cell lines. However, cell line development (CLD) can be tedious and time-consuming work, and every stage in the CLD workflow has its limitations and challenges. Researchers are creating advanced strategies and tools to overcome those challenges, especially for complex biologics such as bispecific antibodies (BsAbs) and difficult-to-express (DTE) proteins. Online presentations from the CLD track of the BioProcess International Europe conference, which was held 13–17 July 2020, addressed current implementations of high-throughput screening, synthetic biology, genetic engineering, next- generation sequencing, and automated analysis. And new approaches are helping scientists to study genes that are directly related to cell production, quality, and viability.

Current Trends
Monday’s presentations began with two keynote addresses, the first of which focused on current CLD trends. Ulrich Goepfert (principal scientist, cell technologies, at the Roche Innovation Center) noted that cell line development has become increasingly challenging with the emergence of sophisticated biologics such as complex engineered BsAbs, trivalent T-cell engagers, tumor-targeted checkpoint modulators, and antibody cytokine fusions. Biopharmaceutical developers also are facing increasing pressure to proceed quickly to clinical trials. New technologies and approaches are being used to meet those challenges.

Targeted Integration: Goepfert pointed out that the most significant change made to Roche’s cell line development process in the past three years was switching from random to targeted integration. He presented a study in which the company used an engineered Chinese hamster ovary (CHO) cell line that harbors a single landing cassette comprising three nonpromiscuous recombination sites. That enabled simultaneous integration of plasmids using recombination mediated cassette exchange and the placement of certain promoters and markers in selective locations on different regions of the plasmids. The most significant advantage of targeted integration is the fast and reliable generation of stable cell pools, easing production of significant amounts of representative material for early process development and good laboratory practice (GLP) toxicity studies. The pools were more homogenous than random pools with regard to productivity, product quality, and stability, so less screening could be performed without compromising outcomes.

Automation, Digitalization, and Data Integration: Goepfert said Roche is creating a digital environment and a “laboratory of the future” consisting of a device layer, workflow and data layer, system layer, integration layer, and an analysis and visual layer. The goal is to have a fully automated CLD workstream. All levels are fully functional except for the workflow and data level, which is in the planning phase. Currently, the company still has isolated software solutions and manual data transfers.

Cloning: Demonstrating that cell lines are monoclonal is a regulatory requirement. Plate imaging and visual tests have become the standard for demonstrating clonality. But that approach can have significant probabilities of error as a result of mixed populations. Goepfert presented results of a study in which two different cell lines were mixed and recloned. Retrieved colonies were analyzed, and researchers found that cell lines that passed their clonality assessment were >99% monoclonal.

Several strategies can be used for single-cell cloning, including single-cell printing, microfluidics, and fluorescence-activated cell sorting (FACS) analysis. Other devices also deliver images or videos of the cloning process or the opportunity to measure reduction soon after single-cell isolation. Method selection must meet key criteria in terms of automation and robustness. The selection also should not increase the level of hands-on work, which compromises throughput.

Machine Learning for Single-Cell Imaging: Goepfert noted that the large number of clones that need to be sorted out when you visually examine cell images could be problematic. “On average, we sort out about 20% of the clones because we believe that the software may be wrong, or the quality is insufficient. So we tend to check quite a lot of images, which can be a big effort.” Roche worked with its information technology (IT) group to improve that process. The resulting deep-learning model detected 90% of the true non-monoclonal antibody cells and 75% of the true monoclonal cells, with an overall accuracy of >95%. Roche currently is benchmarking the model with possible implementation to reduce the need for visual inspections.

The company also is investing in improving screening and analytics. One example is sequence variance (mutation and process-induced variance). The traditional approach uses liquid chromatography and tandem mass spectrometry (LC-MS/MS), but that method has limitations. The company switched to next-generation RNA/DNA sequencing.

Synthetic biology concepts for cell line engineering have the potential to become a large part of cell line development work. The use of gene switches to control transgene expression is “the most obvious application.” Common approaches are ligand-dependent transcription or regulators.

Another field of interest is host-cell engineering. “With next-generation sequencing, proteomics, targeted integration, and various CRISPR [clustered regularly interspaced short palindromic repeats] technologies, we have all the tools at hand to characterize, modify, and regulate the genome of our host cells broadly, in such a way that they are focused on production.”

Finally, Goepfert outlined new opportunities and challenges, including performing CLD technology at the drug discovery phase and the stable production of cell lines for adenoassociated viral (AAV) vectors for in vivo gene therapy.

Understanding Protein Synthesis
The second keynote on 13 July was presented by Ioanna Tzani (postdoctoral researcher at the National Institute for Bioprocessing Research and Training). Her presentation was titled “Understanding Protein Synthesis in CHO Cells Using Next-Generation Sequencing.” Although high-titer recombinant protein production has been observed for some proteins, others remain DTE, so protein production of economically viable quantities remains challenging. The study of protein synthesis can lead to better characterization of host-cell proteins and more efficient production of recombinant proteins. Next-generation sequencing provides a “snapshot” of the current state of cells. It allows the sequencing of millions of DNA and RNA molecules.

Tzani focused on transcriptomics, both RNA sequencing and ribosome profiling through sequencing. She showed how such methods have been used in her laboratory to gain a better understanding of protein production in bioprocessing.

Tzani demonstrated how RNA sequencing detects changes in alternative splicing and facilitates characterization of gene expression in Chinese hamster ovary (CHO) cells under bioprocess conditions. The method was applied to CHO cells at mild hypothermic conditions to characterize the molecular events underlying protein production. Tzani’s laboratory observed a 25% reduction in cell density and used those cells for gene-level differential expression. More than 2,000 genes were expressed differentially between 37 °C and 31 °C. Researchers used StringTie assembly, which increased the number of transcripts available. Results showed that RNA sequencing can provide information for gene-level abundance, isoform identity and abundance, abundance of noncoding RNAs, and differential analysis of RNA expression level. This level of detail is used to improve transcriptome annotation of CHO cells for genetic engineering.

Finally, Tzani demonstrated how ribosome profiling can be applied for the study of translations for temperature-shifted CHO cells. Ribosome profiles can be applied for translation localized in the endoplasmic reticulum and at the mitochondrion or can be used to identify the proteins synthesized at those sites.

High-Throughput Screening
Jason Wade (principal scientist at Pfizer) presented “A High-Throughput Antibody Production Process: Generating Thousands of Proteins and Bispecific Antibodies at the 1–2 mg Scale.” He explained that protein production at the company first was performed by individual scientists, but that approach had a number of limitations. Expression and purification also were first performed by individual scientists. But there was variability in quality, and often observations by individual scientists didn’t transfer to larger scales. Individual scientists also were using limited data (with variability in the data captured). So the company’s High-Throughput Protein Expression and Purification Group developed automated operations. Wade reviewed the benefits of automation and highlighted the final antibody production process. It includes microbiology plating; automated DNA production and normalization; robotic mammalian transfection, feed, and harvest; semiautomated cell dispensing; robotic purification and buffer exchange; and automated analysis and downstream assays, data analysis, and curve fitting.
Pfizer also developed an automated process for producing and screening thousands of BsAbs in a matrix format at the 100-μg scale. For example, the approach to creating BsAbs is to produce homodimers (one with positive charge and the other with negative charge) in separate cell lines. The cells are purified separately, and an oxidation-reduction reaction is used to make the BsAbs. That process yields tens of thousands of antibodies at the 1–2 mg scale annually, and when one monospecific antibody is fixed, the process produces BsAbs at the 1-mg scale.

Viral Vectors
On 14 July, Conrad Vink (director of vector development, cell and gene therapy, GlaxoSmithKline, GSK) presented “Rapid Generation of Lentiviral Vector Producer Cell Lines Using a Single DNA Construct.” Lentivirals typically are produced by transient transfection using four plasmid expression vectors: most commonly, Rev, vesicular stomatitis virus G (VSV-G), group antigens (Gag)-Pol, and one that includes the gene of interest. Those plasmids are transfected into human embryonic kidney (HEK) 293T host cells. The current practice is to switch production from cell trays (for early development) to suspension cultures in bioreactors.

Traditional approaches for generating high-titer suspension–adapted producer cells for lentiviral vectors required a sequential introduction of vector components. But that approach is slow (up to a year for four or more rounds of transfect-select single-cell clone screens) and inflexible (no choice of host cell line). The integration of components at multiple genomic loci also increases the risk of genetic instability or transcriptional silences at one or more loci.

Instead of having four or five different constructs, GSK researchers put them together as separate transcription units (each with its own promoter) all on a single large DNA construct (bacterial artificial chromosome, BAC), which encodes all vector components. The construct can be transfected into HEK293T cells in a single transfection to obtain a producer cell-line directly. The cell line then can be screened for product quality, productivity, and other traits. Once they have the desired BAC construct, scientists can make stable producer cell lines using a conventional CLD process. That consists of vial thawing of HEK293T cells, expanding them in shake flasks, and transfecting 40 μg of BAC DNA into tens of millions of cells using polyethylenimine. Then cells are then plated out in the presence of serum, selected with zeocin to make a stable pool, suspension adapted to remove the serum, and single-cell printed into 96-well plates. Single-cell clones then go through several stages of expansion up to 24-well plates, and finally back to shake flasks. That approach typically generates a few tens of clones for screening.

Vink presented analytical results of producer cell clones (BAC-integration profile) and compared transcription levels with transient transfection. Data showed high yield and infectivity relative to the transient system as well as functional and genetic stability of clones in extended culture over time. Vink also showed results of 50-L runs in stirred-tank bioreactors. Researchers obtained high functional titers (in the 107 TU/mL range) and infectivities at 40–50 TU/pg. So they were able to get industrial titers in a cell-culture format.

Optimizing Cell Lines
Clone Screening: Also on 14 July, Lena Thoring (scientist at Glycotope GmbH) presented “From Clone Selection to Perfusion Processes: Improved Clone Screening for Continuous Bioprocessing.” She addressed how clone selection can be improved to fulfill perfusion process criteria, then she highlighted the company’s GlycoExpress (GEX) production system, which can be used as an alternative host for a glycoprotein. Thoring also pointed out that perfusion processes have several stress factors, including high shear stress caused by the pumping out of cell suspension, long running times (40–50 days, so stable clones are needed), and high cell densities. Clones that can fulfill those process criteria are required.

Thoring highlighted her company’s two-step cloning procedure. It consisted of clone generation and screening, followed by process compatibility screening. The company performs sedimentation in an automated microbioreactor (SAM) scale-down perfusion in an Ambr 15 bioreactor system (from Sartorius Stedim) to mimic the alternating tangential-flow filtration (ATF) perfusion process. Thoring also presented a case study of clones selected to produce DTE BsAbs for perfusion cultivation. The process comprised four steps: clone selection, SAM perfusion for selection of 3–5 clones, stability testing, and 1-L ATF perfusion.

Protein Phosphorylation: On Wednesday, 15 July, the cell line development track focused on perfecting cell lines and producing DTE biologics. Paula Meleady (associate director at the National Institute for Cellular Biology in Dublin City University) presented “How Posttranslational Modifications of Cellular Proteins Influence Growth and Productivity of CHO Cells.” Because at least 70% of biologics are made using mammalian cell expression systems, it is important to improve understanding of CHO cell production. Meleady works in the proteomic aspects of CHO cells. She says systems biology has been used to understand CHO cells to develop the next generation of CHO cell factories. It involves the integration of many data sets from genome, transcriptome, microRNA (miRNA), epigenome, and similar studies. Such integrations can be used to identify targets to rationally engineer and improve cell-line hosts.

Meleady pointed out that little work has been performed on the phosphoproteome in CHO cells, which has a crucial role in regulating bioprocess-relevant processes such as transcription, growth, signaling, metabolism, and protein synthesis. Her work focuses on protein phosphorylation (an important event in cellular-signaling pathways), specifically for CHO cells under different culture conditions. It is thought that at a given time about 30% of proteins are potentially phosphorylated at multiple sites. But phosphoproteins present several analytical challenges, including their need to be enriched from a sample. So Meleady’s team used immobilized metal affinity chromatography (IMAC) and TiO2 to enrich phosphoproteins from samples and MS for protein identification and quantification of proteins and their modifications.

Meleady presented results of three case studies of using proteomic and phosphoproteomic approaches to understand different phenotypic traits within CHO cells. The first focused on reduction of culture temperature and how it influences culture conditions with regard to the proteome and phosphoproteome. The second was a differential proteomic and phosphoproteomic analysis of the growth phases of recombinant CHO (rCHO) cells in batch suspension culture. The third was a differential proteomic and phosphoproteomic analysis of adaption to growth in glutamine-free culture conditions.

“The data generated emphasized the dynamic nature of the changing proteome and phosphoproteome during growth of rCHO cells in various culture conditions, with the inclusion of phosphoproteomic data contributed an adding layer of complexity to the proteomic analysis. Further work will need to be carried out to determine whether specific phosphorylation sites have a functional role. Approaches such as site-directed mutagenesis can determine [whether] those sites have a role in signaling pathways . . . . Such steps could enable rational design media and processes for improving CHO cell productivity.”

DTE Proteins: The session also included a presentation from Hirra Hussain (postdoctoral research associate at the Manchester Institute of Biotechnology, University of Manchester, UK) who discussed “Use of Protein Engineering Strategy to Overcome Limitations in the Production of Difficult-to-Express Recombinant Proteins.” In the expression pathway, a protein either is secreted into the culture medium, retained within cells, or degraded. Each step in the pathway (from DNA sequence to mature protein) has potential limiting steps and different posttranslational modifications can be a bottleneck. The efficiency of how proteins are produced depends on all steps.

Hussain introduced two model proteins that her team has been working on from the tissue inhibitors of metalloproteinase (TIMP) family, called TIMP 2 and TIMP 3. They regulate the activation of matrix metalloproteinases (MMPs) and remodeling of the extracellular matrix. Although both TIMP 2 and TIMP 3 were similar in sequence and structure, they showed significant differences in their production. The team expressed both in CHO and in HEK293T cells. Although TIMP 2 had good expression in both cell lines, TIMP 3 had no detectable expression, even after changes in culture temperature and in signal peptide, and even though analysis in transient CHO expression showed both TIMP 2 and TIMP 3 had similar growth profiles.

The team conducted a systematic screening of each stage of the protein-expression pathway to determine where the block in expression of TIMP 3 occurred and how it compared with that of TIMP 2. After conducting messenger RNA (mRNA) analysis and examining intracellular protein localizations and modifications to understand the sugars on TIMP 3, Hussain’s team concluded that the bottleneck occurred at the posttranslational stage. They implemented a protein-engineering strategy to overcome the block using a furin-cleavable TIMP-3 sequence and applied that method to a new TIMP 4 protein — a “difficult” protein that also had a poor secretion. TIMP 4 was modified to include the furin-cleavable prosequence (with linker) to test the ability of the protein-engineering strategy to overcome limitations in production. Results showed that the attachment of the prosequence led to increased secretion of TIMP 4. The team then evaluated the molecular basis of poor protein production between TIMP 2 and TIMP 3 using an algorithm to assess surface features of the proteins (looking at hydrophobicity and charge patches). Results showed that TIMP 3 had more nonpolar patches at the C-terminal. They also showed that the N-terminal sequence had a significant effect on TIMP-3 expression. The team engineered a form of TIMP 3 without the N-terminal. That form secreted successfully although the mechanism by which the sequence affects expression was unknown.

Cell Lines for BsAbs
Nathan Lewis (associate professor in the department of pediatrics and bioengineering at the University of California, San Diego) presented the keynote address on 15 July. It was titled “Insights into Cell Line Engineering from the Lewis Lab.” His team has worked on trying to turn CHO cells to engineerable systems. He highlighted recent results of studying how a cell responds when trying to engineer in improved product quality — specifically, glycosylation — and ensuring that the correct posttranslational modifications have been achieved. “Whereas 10 years ago manufacturers solely focused on process optimization, we are moving to a stage where we are starting to understand what is going on inside cells.” His laboratory works to turn CHO cells into engineerable systems by first sequencing the CHO-cell genome and the genome of Chinese hamsters to identify genetic elements. His team maps out biosynthetic pathways for protein synthesis and secretion, including translation.

The team also maps out pathways for posttranslational modifications (e.g., glycosylation) and different metabolic pathways and uses different gene-editing methods to rationally engineer cells. Lewis warns, “Care must be taken because even changing a single sugar molecule in a particular region of an antibody can change antibody activities, such as switching it from being proinflammatory to being antiinflammatory. So glycosylation must be controlled carefully.”

Lewis highlighted the steps of glycosylation, using alpha-1 antitrypsin (A1AT) as an example. He compared the glycoengineered CHO-produced A1AT to human-plasma–derived A1AT. His presentation then focused on studies to determine what happens to glycans and cells after glycoengineering and whether glycosylation is compromised. His team studied more than 180 glycoengineered clones, glycoprofiled them, measured the RNA sequence on them, and phenotyped the cells to study whether the glycosylation affected the cells, comparing a clone and a knockout. The challenge to finding the source of a change in glycosylation is that glycans are independent, so it is difficult to identify whether transcription factors or miRNAs are regulating individual glycosyltransferases or whether there are effects caused by metabolic regulations and gene mutations. Thus, a t-test cannot be used to identify glycans that change significantly.

However, qualitative assessment can be performed through manual analysis of biosynthetic pathways. Because doing so on 180 mutants would be too difficult and time-consuming, and clustering those glycoprofiles was unsuccessful, the team developed a comparison algorithm to identify glycoprofiles that do not cluster well and to enable clustering of different genotypes. With that approach, the team identified dominant isozymes and off-target effects from knocking out glycosyltransferases.
Lewis then discussed what happens to the cells after glycoengineering (because cells use glycans to interact with their environments). The team measured how engineering affected cell growth rate, viability, maximum cell density, and other traits for 180 clones using similarity network fusion to identify traits of three major phenotype groups associated with cell size, growth rate, viability, and other attributes. Lewis concluded his presentation by showing results of pathways with increased or decreased expression as a result of knocking-out genes from the mannosyl-glycoprotein N-acetylglucosaminyltransferases (MGAT) family.

Cell Line Engineering
Bjorn Voldborg (director of CHO cell line development at the Novo Nordisk Foundation Center for Biosustainability) presented the keynote on Thursday, 16 July. It was titled “Engineering CHO Cells for the Production of the Next Generation of Therapeutic Proteins.” His team has been working on “tailor-made” glycosylation of therapeutic proteins by making different cell lines for GMP manufacturing. His presentation centered on glycoengineering and glycosylation, which can affect significantly the efficiency, activity, stability, half life, and immunogenicity of proteins. His team glycoengineered CHO cell lines to make custom glycoprofiles by using information gained from systematic knocking out of glycogenes to elucidate the “glycomachinery” of CHO cells. His team generated a panel of different glycoengineered CHO cell lines for production of drug candidates (from mono- to tetra-antennary). The objective is to “glycol-optimize a therapeutic protein for optimal efficacy. That means that when you choose a cell line, you obtain a highly defined glycovariant of the protein you want to produce. Using the panel, you can produce different versions of the same protein with different glycans. So there is greater diversity across the CHO glycosylation space with more homogeneous glycosylation. That enables the optimal (high-producing) cell line to proceed to clinical trials.” Voldborg showed an example of how optimization of glycoforms was conducted using erythropoietin (EPO) model protein, liver-secreted plasma proteins, and A1AT. The goal was to design hard-to-produce (as opposed to DTE) proteins and build them into active therapeutics.

Next-Level Cell Lines
On 17 July, Nicolas Mermod, PhD (professor and director of the Institute of Biotechnology at University of Lausanne, Switzerland) presented “Genetic Engineering to Improve CHO Cell Metabolism and Therapeutic Protein Production.” He discussed the causes of clonal variability in expression as well as limitations of CHO metabolism. Good clone selection can improve cell metabolism because specific sets of CHO genes are associated with high production levels. For example, some genes contribute to proper protein folding and cell viability.

Because animal cells depend on vitamins to sustain bioenergetic and biosynthetic needs of cells, Mermod’s team assessed the need of CHO cells for certain vitamins. In particular, they examined whether the knowledge of vitamin B5 dependency might be used to select cells that express a transgene of interest efficiently. It was thought that cells expressing a vitamin transporter would grow faster and have a better metabolism than native CHO cells, and those cells with improved metabolism would make higher levels of therapeutic proteins. Mermod presented results showing that cells expressing the SLC5A6 transporter increased expression and had improved B5 metabolism. He also showed how clone selection could be improved with metabolic screening.

Mermod then highlighted studies that examined whether there are metabolic bottlenecks that could impair production irrespective of selection approach. That study involved identifying genes associated with high production. Specifically, his team focused on Erp27 (when coexpressed with Erp57) and the FOXA1 gene. The latter activates production-increasing CHO genes. Some FOXA1 effects were in part caused by cytoskeleton reorganization, and other effects were caused by other metabolic processes that lowered accumulation of toxic metabolic byproducts. Finally, Mermod showed that actin polymerization can be used as a selection procedure.

Join Us Online This Month
On behalf of the BPI team, I hope you join us online for the BioProcess International Conference, which will be presented 21–25 September 2020. For more information visit https://informaconnect.com/bioprocessinternational.

Maribel Rios is managing editor at BioProcess International, a part of Informa Connect; maribel.rios@informa.com.

Leave a Reply