Improving Biotherapeutic Production Using Parallel Combinatorial Screening

Home » Improving Biotherapeutic Production Using Parallel Combinatorial Screening

Improving Biotherapeutic Production Using Parallel Combinatorial Screening

Schematic representation of a CombiCult experiment to find protocols for differentiation of stem cells.

Schematic representation of a CombiCult experiment to find protocols for differentiation of stem cells.

The challenges facing the industrial production of any medicinal product are substantial. For biologics in particular, the process defines the product.

As with any process, a series of unit operations must be completed, controlled and analysed; for each unit operation, a number of steps exists and at each one of these steps, a number of possible variables may be present that impact the critical quality attributes (CQA) of the product and, therefore, its clinical safety and efficacy.

Approaches that build quality into a process through the use of Quality by Design (QbD) are favoured by the regulatory agencies.1 However, the number of experiments required to build quality in to a process can be vast. In theory, it should be possible to test all the possible combinations of variables for a given process; but, in practice, such undertakings become too large (and expensive) to be performed. Currently the most widely used approach to circumvent this problem is the Design of Experiment (DOE) approach. DOE is a statistical-based tool that reduces the number of experiments needed to model the responses of multifactorial experiments by testing a subset of values for each variable and using these values to predict the impact of changes to one variable on other variables and the product.

Ultimately, however, the DOE approach is still limited by the throughput of the experimental method available to analyse the outputs, and it doesn’t test all possible combinations of variables. The amount of experimentation needed should not be underestimated as the process optimization challenge lies in the combinatorial nature of the problem: how can multiple variables be tested in parallel?

Plasticell’s platform technology, CombiCult, has been designed with that challenge in mind. Originally used in combinatorial chemistry, the concept was applied to the paradigm of stem cell differentiation, upon which Plasticell was founded and where most of its efforts have been focused. As with any process, the differentiation of stem cells (both in vivo and in vitro) is determined by the concerted action of a series of combinations of cell signalling molecules acting on the cell at specific times in a specific order.

As there are multiple steps in the differentiation process, and multiple possible variables (ligands, cytokines, growth factors, etc.) at each stage, the number of possible combinations can be vast. For example, if a process has four stages, and at each stage there are 10 possible variables, a total of 10,000 variable combinations are available. Testing 10,000 combinations in parallel using standard methods (even in a miniaturized form) can be time consuming, expensive and difficult to analyse.

CombiCult offers precisely that level of parallelization in a format that is easy to handle without the need for specialized equipment or large expensive experiments. Furthermore, Plasticell has been able to show that a 50-fold increase in efficiency, along with a cost reduction, can be achieved using CombiCult to substitute expensive growth factors with small molecules for the differentiation of embryonic stem cells.2

CombiCult is a high throughput platform that uses combinatorial cell culture technology to screen tens of thousands of protocols in one experiment. Combicult combines the miniaturization of cell cultures on microcarriers, a pooling/splitting protocol and a unique tagging system to allow multiplexing of experiments. Cells grown on microcarrier beads are randomly shuffled through multiple conditions using a split-pool method. The iterative process of splitting, culturing and pooling systematically samples all possible combinations of conditions in a predetermined matrix.

If X conditions are tested on each of Y number of cycles, XY protocols are tested simultaneously. Each condition is spiked with a unique fluorescent tag that attaches to the beads. At the end of the process, beads bearing cells with the desired phenotype are identified by a screening assay (such as immunostaining or reporter gene expression) and individual positive beads are isolated. The cell culture history of each positive bead is then deduced by analysis of the fluorescent tags attached to the bead. Protocols are analysed using bespoke bioinformatics software (Ariadne), which uses criteria such as hierarchical clustering and probability analysis to select the optimal protocols for further validation.

Additionally, Plasticell has shown that it can multiplex different cell lines into CombiCult screens. This would prove an efficient way to discover protocols that are applicable to many different lines or to investigate differences between cell lines.

Upstream Protein/Antibody Production
The efficient and economical production of therapeutic proteins, antibodies or other biologics depends on the ability to create cell lines that can synthesize and secrete the protein of interest in high amounts in a reasonable timeframe using processes and materials that are inexpensive and easy to reproduce. The most widely used mammalian expression systems are CHO cells and hybridomas. These cells are usually genetically engineered to overexpress the protein of interest and to secrete this into the media.

Years of research have improved dramatically the yields of proteins from mammalian systems, but there is still room for further enhancement. Optimizing the genetic manipulation of the cells is outside the scope of the technology at present; nevertheless, clones produced in different ways may be tested simultaneously for other variables.

The first stage in a CombiCult screen is to define the desired cell phenotype and develop an assay to detect it. The most desirable phenotypic characteristic of the cell may be the protein yield; protein secretion can easily be detected using FLSS (fluorescent labelling in semi-solid media) if an appropriate antibody for the protein of choice is available. Using a device such as ClonePix (Molecular Devices) beads bearing high producing cells can be automatically detected by the extent of the halo they produce in the semisolid media containing a fluorescent antibody against the secreted protein, then picked and dispensed in 96 well plates. Further cellular phenotypes can be additionally assessed and detected using immunostaining with other fluorophores.

Once an assay has been established, a matrix with the variables to be tested can be designed. Even though the total number of combinations can be large, there are some practical considerations in defining the number of stages to test and the number of conditions to be tested at each stage. In practice, we routinely test 10–20 variables per stage and 3–4 stages.

The CombiCult workflow.

The CombiCult workflow.

For stage 1, for example, a number of different clones (say 10) produced using different vectors/cell lines or transduction techniques can be used; these are seeded onto Plasticell microcarrier beads and tagged with a unique tag that identifies each individual clone (there is an inbuilt excess of microcarrier beads/pathway, so that if 10,000 possible combinations are tested, at least 300,000 beads are seeded). After an overnight incubation to allow cell and tag attachment, the beads undergo the first round of pooling and are then split into a further 10–20 conditions, in which a range of media compositions can be tested.

Each media condition can have either a simple composition with one or two components or a complex one containing a number of small molecules/growth factors/lipids and hormones. Again, each condition is individually marked with a unique fluorescent tag that will allow its identification. The cells can be incubated in these conditions for a predetermined period of time and then another round of pooling and splitting can be done. During this stage, another set of variables can be tested; these can be other media conditions, supplements or physical parameters such as pH, temperature and/or oxygen tension.

At the end of the culture period, each pool can be either fixed and immunostained, or plated in semisolid media containing a fluorescently conjugated antibody against the expressed protein. After 3–5 days, halos of antibody-antigen precipitates can be visualized under epifluorescence, and beads with the largest halos can be picked either manually or using the ClonePix (Molecular Devices).

Once beads are picked, they can be analysed for tag content and their culture history can be deduced. This data is fed into Plasticell’s bespoke software Ariadne, which uses clustering algorithms and probability analysis to rank the most successful combination of variables that give the desired outcome. Depending on the number of positive beads and the degree of clustering, a set of 10–20 possible pathways (or sets of combinations) can then be validated, thus reducing the total number of combination of variables from thousands to just and handful.

Mapping the journey taken by positive beads will elucidate a combination of process steps that best produces the desired cell phenotype, providing the basis for the subsequent step of process refinement to define the final protocol.

1. A.S. Rathore and H. Winkle, “Quality by Design for Biopharmaceuticals,” Nat. Biotechnol. 27, 26–34 (2009).
2. M. Tarunina, et al., “Directed Differentiation of Embryonic Stem Cells Using a Bead-Based Combinatorial Screening Method,” PLoS One 9, e104301 (2014).

Dr Diana Hernandez
Principal Scientist
Plasticell Ltd
Stevenage Bioscience Catalyst