Downstream simulation of biomolecules
The ever-growing diversity of innovative biomolecules is leading to a stunning range of different molecules, which also require different production processes. The increasing complexity and diversity of the pipeline also poses new challenges for product manufacturing. Mechanistic modeling plays a key role in coping with these challenges. Mechanistic model applications are thereby not limited to a certain class of biomolecules but are broadly applicable. Depending on the specific molecule and the manufacturing process used, different modeling strategies may be required.
Today, the biopharmaceutical market is still dominated by monoclonal antibodies (mAbs). The structural similarity of pharmaceutical mAbs have led to the development of platform processes. A platform process is a production process suitable for the manufacture of a group of related products enabling rapid and cost-effective process development. Platform processes enable process understanding to be generated rapidly, as many products use the same process. Process understanding acquired with one product can be used for a similar product. Also, the risk associated with developing a completely new process is decreased.
However, each molecule comes with individual product-related impurities like aggregates and charge variants, as well as process-related impurities such as host cell proteins (HCP) and DNA. Platform processes are therefore not completely rigid. Their process parameters still need to be optimized for each product and optimized process conditions need to be validated for robustness. This procedure is based on an extensive use of experiments which makes process development time-consuming and costly.
Additionally, more and more new modalities have been developed in recent years as an alternative to natural mAbs. Many of these new modalities are non-platform molecules which cannot be covered by classical platform processes and require additional development effort.
Downstream simulation of monoclonal antibodies
Simulating the commonly used downstream processes of a mab platform, such as protein A, ion-exchange, mixed-mode and hydrophobic interaction chromatography or filtration is very much possible. The accuracy of simulating product size and charge heterogeneity, as well as HCP, DNA and leached protein A is only limited by the data quality of the company-specific analytical assays.
With the current biopharmaceutical market dominated by antibody platforms, multi-specific antibodies and especially bispecific antibodies (BsMAbs) represent key components of next-generation antibody therapies. In contrast to natural antibodies, BsMAbs can target two different antigens at the same time, allowing more tailored immunogenic targeting. However, unlike conventional monoclonal antibodies, great production challenges arise with respect to product quality.
While the purification of monoclonal antibodies can be streamlined by using platform approaches, the downstream processing of BsMAbs is not always as straightforward. Due to the inherent diversity of this class of molecules, different BsMAbs need different purification solutions, hampering the development of a platform approach. Apart from the lack of a platform process, product-related impurities present an even larger problem for BsMAbs in comparison to natural mAbs. Besides the desired heterodimerization of heavy chains, multiple product-related impurities like homodimer forms can be formed during product expression. This results in a lower yield for the molecule of interest and increased effort to achieve the required product quality.
Process understanding for bispecific antibodies using mechanistic models
The increased production challenges for BsMAbs are accompanied by an increased experimental effort during process development. Mechanistic modeling can help to decrease the experimental effort during the process development of BsMAbs and to find robust process conditions to consistently achieve the required product quality. In addition to the species modeled for a natural mAb, the different homo- and heterodimer and other multimers are modeled explicitly. The increased process understanding provided by mechanistic models can also help to find synergies between BsMAbs which can finally lead to the development of platform processes for this type of new modality.
Another new modality in the diverse pipeline of targeted cancer therapies are antibody-drug conjugates (ADCs). Combining the advantage of the selectivity of an mAb with the potent cell-killing activity of a small cytotoxic molecule, ADCs can provide high selectivity, safety and efficacy. One important property of an ADC is the drug-to-antibody ratio (DAR) which is the average number of drug molecules conjugated to the mAbs. It is a critical quality attribute affecting the pharmacokinetics, efficacy and safety of the product. Initially, the DAR is a result of molecular engineering and the conjugation reaction, where the drug molecules are covalently attached to the mAb via a hydrophobic linker. Depending on the conjugation strategy, the reaction leads to a rather broad or narrow distribution of components with different drug loading. To achieve a predefined and narrow DAR in the final drug substance, a robust combination of molecular engineering, conjugation and subsequent purification is mandatory.
Downstream processing (DSP) is commonly used to remove ADCs with an unfavorable DAR, as well as the unconjugated payload. Due to the hydrophobic properties of the payload-linker-conjugate, hydrophobic interaction chromatography (HIC) is frequently used. Variations in DAR after the conjugation can therefore be corrected by DSP.
However, to achieve this goal, the downstream process needs to be well-understood. Finding robust operating conditions in DSP development can be challenging and requires significant experimental effort. As handling ADCs is associated with a higher risk for the workforce, replacing lab experiments with computer simulations comes with an increased safety at work.
In silico process development for antibody drug conjugates
Establishing mechanistic models for ADC processes leads to more efficient process development, optimization and a higher process robustness. In addition to the species modeled for an mAb, the ADC species with different DAR as well as the unconjugated payload are modeled explicitly.
The variety of peptides, proteins and enzymes for therapeutic applications is stunning. Ranging from small peptides and hormones to proteins such as insulin, erythropoietin, blood factors and enzyme replacement therapies. The major drawback of these classes of molecules compared to, for example antibodies, is their molecular diversity. This requires discovery, design and manufacturing processes which must be developed on a case-by-case basis and which come with a significantly higher risk of failure. In downstream processing, some of the main differences to antibody processes are the lack of platform knowledge, the absence of an affinity capture, the comparably low product concentrations and sometimes an increased instability of the molecules.
Mechanistic process models for peptides, proteins and enzymes
Using a mechanistic process model for downstream processing compensates for most of those drawbacks and model-based process development workflows from the antibody world can be transferred. The mechanistic process simulation accounts, of course, for the low abundance of proteins in a feedstock, simply being limited by the analytical assays of a company. Developing non-affinity capture chromatography is probably one of the most complex DSP tasks, and the related skills have not been widely practiced over the last couple of decades as the focus was mostly on antibodies. Mechanistic models allow a highly efficient development of non-affinity capture processes, as well as optimizing these processes to handle a variable feedstock coming from the bioreactor.
The market entry of generic molecules with lower prices after patent expiry is a well-known phenomenon for small pharmaceuticals. As the development and manufacture of completely identical biomolecules by another company is not possible, generic versions of biopharmaceuticals are called biosimilars or biobetters.
Biosimilars are highly similar to an already licensed biological product, showing no meaningful differences in purity, potency and safety when compared with the originator product. Developing biosimilars is associated with numerous challenges, including the proprietary nature of the manufacturing process of the reference product and the need to demonstrate product similarity. As the goal of a biosimilar is to provide a cheaper alternative to the innovative reference product, there is also a higher cost pressure on pharmaceutical companies and costs for biosimilar manufacturing therefore becomes more important.
Biosimilars – an even faster development through computer simulation
The process development of biosimilars is usually an iterative process to select the process configuration that will deliver target product quality attributes similar to those of the reference product. This task becomes non-trivial for biological products as they typically possess several quality attributes like charge variants and glycosylation variants that are affected by numerous process parameters and raw material attributes.
The focus of ensuing biosimilarity lies on molecular and cell line engineering as well as upstream processing. Although it is well known that the patterns of product-related impurities and product variants can be modulated with downstream processing, it is commonly neglected – simply because it would be too complex when executed experimentally.
With a mechanistic process model, developing such a sophisticated downstream process is not more challenging than any other PD task. The model can simplify and streamline the process development of biosimilars as they describe the relationship between product quality attributes, process parameters and material attributes. They can be used to identify process conditions that consistently achieve the desired product quality profile of the reference product without the need for elaborated experimental studies.
Virus-like particles (VLPs) are an effective complement to traditional vaccines against infectious pathogens. In addition, VLPs are key to some of the most promising concepts of gene therapies and CAR-T cell therapies. The VLP assembled from capsid proteins can serve as delivery vectors for genes or other therapeutics.
Despite their therapeutic potential, there are still technical challenges in the manufacture of VLPs, in particular regarding product purification. The presence of the host cell-derived membrane of VLPs provides the possibility to integrate antigens and adjuvants but, at the same time, poses challenges for downstream process development due to the existence of multiple undesired process-related contaminants. They are predominantly attributed to host cell impurities such as cell debris, host cell proteins, DNA and lipids. Developing a purification process that safely removes these process-related impurities can be very complicated. Depending on the application, the separation of empty/full VLPs is a critical element of downstream processing.
Downstream simulation of VLPs
Although the underlying theory of mechanistic chromatography simulation was originally developed for protein separations, it has been successfully applied to industrial VLP and vaccine purification as well. Like the modeling workflow for modeling the product heterogeneity of therapeutic proteins and antibodies, the different VLP variants are simulated as separate species. Empty/full separation are, for example, possible as the encapsulated DNA alters the surface charge of the VLP.
Finally, mechanistic modeling can be applied in the same way as for other biomolecules to reduce the experimental effort in process development and find robust process conditions to maximize product purity and yield.
Nucleic acids such as DNA and RNA are the most disruptive class of biomolecules for pharmaceutical applications. From a patient perspective, gene therapies based on nucleic acids promise to actually cure serious diseases instead of just temporarily restoring a healthy condition. From an industrial perspective, nucleic acid based therapies can be engineered and produced quickly allowing the implementation of industrial concepts such as rapid prototyping. As the synthesis of nucleic acids can be easily scaled down, personalized therapies become feasible.
Techniques to produce DNA and RNA for research purposes have also been established for decades, but when it comes to industrial production, the development of safe and efficient processes is still an ongoing task. Safety measures, e.g. to ensure a homogenous and well-defined product, are even more important compared to protein therapeutics, as nucleic acids potentially integrate into the patient’s genome and negative side effects might last for the patient’s entire life.
Downstream process simulation of nucleic acids
Although the underlying theory of mechanistic chromatography simulation was originally developed for protein separations and has been frequently used to simulate nucleic acids only as trace impurities, accounting for plasmid DNA or RNA as a bulk component is possible as well. Typical model applications for nucleic acid are the separation of different plasmid conformations or sequence variations.