(513f) Design of Experiments Compared to Artificial Neural Network Approach – How Can We Use Them to Understand Spray Drying of Biologics?

The therapeutic potential of biologics with the advances made in biotechnology leads to a wide variety of biologicals on the pharmaceutical market and in the pipeline. Most of these valuable and complex entities are currently formulated as a liquid final dosage form. However, the transformation into a solid-state offers the benefits of longer storage times and higher storage temperatures. The gold standard process for this liquid to solid transformation is lyophilization. Lyophilization offers advantages, such as low temperatures during drying, large knowledge available in the literature, and filling and drying in the final containment. However, lyophilization is a batch process, with flexible processing and up-scaling being major issues, combined with the risk of activity loss for the biologic itself through, e.g., pH shift, cold denaturation, ice-liquid interfacial denaturation, etc. (Sharma et al., 2021). One of the most promising alternatives is spray drying. In this process, the feed solution is atomized into small droplets. Then, the solvent is evaporated gently through mass transfer from the drop to the surrounding drying gas stream. Finally, the dried powder can be collected after a separation step (e.g., cyclone). This drying process can be performed continuously, resulting in dried active biologics of therapeutic interest (e.g., Interleukins (Fiedler et al., 2021)) with more flexibility in throughput and upscaling.

Process development of a spray drying process requires several grams (to kilograms) of the active pharmaceutical ingredient (i.e., biologic). However, on the one hand, their prices can be in the range of several hundred dollars per milligram, and on the other hand, the production of biologics may be in the range of several kilograms per year (Pennington et al., 2021). Hence, developing a spray drying process can be unfeasible due to the lack of available material and high development costs. Thus, a cheap surrogate substance was investigated in our approach to overcome this hurdle. Clearly, the most critical part of such an approach is finding a suitable surrogate substance with a similar drying performance as the protein. However, identifying relevant material attributes for a drying process is challenging due the complex phenomena observed, including skin formation on the droplet and water diffusion in the particle and through the skin, which significantly influences drying performance. Therefore, we started with Lactose as surrogate substance as it widely used in spray drying for process validation runs during equipment setup and testing. Process data from such tests are often available.

Hence, a first Design of Experiments (DoE, 2⁴ full-factorial) was performed with Lactose. The data gained from this DoE reflects the drying kinetics and evaporation capacity of the spray dryer (Buchi B-290) at different process settings. Eventually, only three spray drying experiments were performed with the protein (human serum albumin, HSA) to transfer the process dynamics including the protein. Therefore, we used the DoE data to develop an Artificial Intelligence Approach (using artificial neural networks ANN), which we compared to and combined with the classic DoE approach.

From the lactose DoE the derived models primarily depend on the number of investigated factors and the factor combinations and the number of experiments performed, resulting in linear and quadratic models (as well as considering the interaction of factors). The capability of the resulting models to predict the responses is limited, especially when the underlying physical correlations are complex and interdependent. Moreover, the Critical Quality Attributes (e.g., residual moisture, particle size distribution, yield, etc.) were investigated and examined concerning their ability to predicted reliably. Additionally, the HSA was analyzed concerning its secondary structure to derive an understanding of the limits of proteins processed at mild temperature spray drying conditions (i.e., outlet temperature less than 100°C). When it comes to the drying of biological entities, their influence on the drying kinetics (e.g., the binding capacity of water, diffusion in the liquid drop) is not well understood. Therefore, we investigated the influence of protein concentration on the evaporation kinetics by means of thermogravimetric analysis.

This talk will highlight the intrinsic opportunities of an ANN approach compared to a DoE approach for developing spray drying processes for biologics. With this, it is possible to fully exploit the advantages of a continuous drying process. Furthermore, we show a method that allows the determination of the drying kinetics of different protein concentrations at mild temperatures, including first results for HSA. This method can also be used to test surrogate substances and create a database for which will in the future allow even better fitting of the developed models. Initial experimental results highlight the potential and benefits of a combinatory planning approach, by requiring fewer experiments, and thus significantly reducing the material and time during process development.

Literature:

Fiedler, D., Hartl, S., Gerlza, T., Trojacher, C., Kungl, A., Khinast, J., Roblegg, E., 2021. Comparing freeze drying and spray drying of interleukins using model protein CXCL8 and its variants. Eur. J. Pharm. Biopharm. 168, 152–165. https://doi.org/10.1016/j.ejpb.2021.08.006

Pennington, M.W., Zell, B., Bai, C.J., 2021. Commercial manufacturing of current good manufacturing practice peptides spanning the gamut from neoantigen to commercial large-scale products. Med. Drug Discov. 9, 100071. https://doi.org/10.1016/j.medidd.2020.100071

Sharma, A., Khamar, D., Cullen, S., Hayden, A., Hughes, H., 2021. Innovative Drying Technologies for Biopharmaceuticals. Int. J. Pharm. 609, 121115. https://doi.org/10.1016/j.ijpharm.2021.121115