Biopharmaceutical drugs such as antibodies or vaccines are produced in complex industrial bioprocesses. Following upstream production in genetically modified cells, the target molecule must be purified in several downstream process steps. Depending on the molecule type and intended use, different downstream steps may be required to this.
For drugs, that shall be injected subcutaneously, the target molecule must be highly concentrated. This dosage form is very beneficial to the patient but poses significant challenges for the downstream development and operation: At the end of downstream processing, a combined ultrafiltration and diafiltration (UF/DF) step is commonly used to achieve the desired protein concentration, pH, and composition of product excipient. At this stage, the product shall be concentrated, and the active ingredient shall be transferred into the final formulation buffer.
Unfortunately, the product also interacts with other solutes and dissolved excipients such as small ions. This interaction may lead to an uneven concentration distribution of solutes across the membrane, e.g. a depletion of desired excipients and an accumulation of undesired excipients. During a UF/DF process, volume exclusion effects and interactions between the drug substance protein and excipients can lead to discrepancies in pH and excipient composition between diafiltration buffer and product pool. In practice, the so-called Donnan effect becomes increasingly critical for high product concentrations and can cause a major deviation in pH and excipient concentrations between the desired values and the actual outcome of the UF/DF step.
These discrepancies make it extremely challenging to identify a suitable diafiltration buffer so that pH and excipient concentration in final pool fulfill the desired specifications. The identification of the right diafiltration buffer is often performed iteratively using lab experiments. This iterative process is constrained given the large quantity of drug substances typically required for a lab experiment. Further it is time intensive and provides only a rough, empirical understanding.
Mechanistic models to digitalize downstream process development
Traditionally, downstream process development fully relied on time and resource intensive lab experiments – not only for UF/DF processes. During the last decade, computer simulation has established itself as a pioneering approach to speed up the development of chromatographic purification steps, by replacing real wet lab experiments with predicted, simulated computer experiments. Digital process twins based on mechanistic models have thereby shown great advantage over statistical models, as they do not only reduce the experimental effort, but further enable a profound understanding of the effects causing deviations or failures, while disclosing causalities.
Simulating the Donnan effect – yes we can!
Digital twins based on mechanistic models may also offer substantial benefits for the development of UF/DF processes: An improved process understanding, shorter development times and less material consumption amongst others. At GoSilico, we have therefore developed a prototypical tool to simulate the Donnan effect as present in UF/DF processes, based on a mechanistic model. The model is created purely in-silico or calibrated using pre-existing datasets. Once set-up, the model is able to predict the pH and concentrations of excipients.
The Donnan effect on the pH and excipient concentrations in the final diafiltration pool is visualized in the Figure below. The Figure shows the diafiltration of a positively charged (blue) and negatively charged (orange) mAb using a 20 mM histidine buffer (pH 7.0) as diafiltration buffer. For the positively charged mAb, the final DF pool is characterized by an increased pH and a decreased histidine concentration. For the negatively charged mAb, on the other hand, an accumulation of histidine in the retentate and a lower pH can be observed. The experimental data previously described by Baek et al. can be adequately reflected by the developed model.
Digital UF/DF twins based on mechanistic models
With the UF/DF simulator we have developed, GoSilico paves the path towards digital twins for UF/DF processes and to roll out simulation to the entire downstream process chain. To our opinion, the potential use cases for such a simulation tool are massive and manifold:
- Development of formulation (specification of diafiltration buffer)
- Reduce experimental effort, safe drug substance
- Prediction of results of UF/DF (prediction of pH and excipient concentrations during UF1, DF, UF2)
- Donnan effect in dialysis experiments
- In silico process characterization
- Improved process understanding: which effects lead to which result?
Want to give it a try?
Our UF/DF simulator was developed with industrial partners and carefully tested and validated with industrial data. It is currently available as a beta version. What is your use case and do you want to challenge our simulator with your own filtration task?