Search

Antibody-drug conjugates (ADCs) have been designed as a combination of monoclonal antibody (mAb) therapy and chemotherapy. From this fact, they draw their potential of uniting the advantages of both strategies in one molecule. mAbs have the ability to specifically bind their target antigen, thus focusing the effect on the target site of action. Due to their size and other biochemical properties, they have a good circulation half-life in the body, which is an important pharmacokinetic property. While mAbs are applied in various therapeutic fields, they form a highly important part of modern oncology. Here, mAbs are used to target antigens that are highly expressed on cancer cells, exhibiting different modes of action to fight the cancer. In order to increase their capacity of killing cancer cells, small cytotoxic molecules, as applied in chemotherapy, can be covalently attached to the mAbs, forming ADCs. Due to the decreased systemic exposure, drug molecules with higher cytotoxicity can be used. Motivated by this potential and the market approval of the first successful products in 2011 and 2013, ADCs gained a lot of attention. By the end of 2019, there were already six products on the market and over 60 candidates in clinical trials. Substantial progress has been made in areas like the development of new cytotoxic drugs, linker chemistries, and conjugation strategies. Despite these successes, the development of new ADCs remains challenging. Unfavorable pharmacokinetic profiles caused by the hydrophobic nature of the drugs and heterogeneity in the degree and site of conjugation are factors which are being improved for current ADCs. Solutions include, for example, site-specific conjugation strategies. Still, the number of parameters for optimization is high for these complex hybrid molecules. Issues range from antibody, drug, and linker over attachment chemistry to the optimal drug-to-antibody ratio (DAR). In order to unlock the full potential of ADCs, efficient, knowledge-based process development is necessary. Also looking at the current landscape of biopharmaceutical development, it is evident that there is high pressure on process developers to efficiently deliver robust processes while gathering enhanced knowledge on process and product. One reason is the diversification of the product pipeline caused by emerging new modalities like ADCs and other antibody formats or cell and gene therapy. It increases development efforts and hinders the use of platform approaches. In addition, time to market gets more crucial with rising development costs and growing global competition, for example by producers of so-called biosimilars. Finally, it is promoted by regulatory agencies like the U.S. Food and Drug Administration or the European Medicines Agency that the concept of quality by design (QbD) is implemented in pharmaceutical development. Its goal is for processes to be designed in a way that the desired product performance is robustly achieved in a controlled fashion. It requires increased process understanding and the thorough characterization of the relationship between critical process parameters and critical quality attributes of the product. The goal of this thesis is to advance the process development of ADCs in the direction of more efficient, systematic, and knowledge-based approaches. As a strategy for the realization of this objective, the establishment of high-throughput, analytical, and digital tools for ADC processes was investigated. High-throughput tools, especially in combination with design of experiments (DoE), can lead to a strong increase in efficiency regarding time as well as material consumption. In order to prevent an analytical bottle neck, high-throughput compatible analytics are crucial. Also analytical techniques for the on-line monitoring of processes have great benefit. They are the basis for implementing process analytical technology (PAT) tools, which give the opportunity for real-time monitoring and control of product quality attributes. Digital tools, such as methods for the mechanistic modeling and simulation of processes, offer many advantages for process development. Apart from granting a deeper understanding of the process fundamentals, mechanistic models can be efficient tools for process optimization and characterization of the design space. The methods for ADC process development applied or developed in this work did not rely on the highly toxic drugs used in ADCs. Instead, nontoxic surrogate drug molecules, similar in relevant properties like size and hydrophobicity as commonly used cytotoxic drugs in ADCs, were employed. The applied combination of cysteine-engineered mAb and maleimide conjugation chemistry is a strategy for site-specific conjugation with high relevance for ADC development. In the first part of this thesis, a high-throughput process development platform for site-specific conjugation processes was developed1. The multi-step process of making ADCs from cysteine-engineered mAbs was successfully transferred to a robotic liquid handling station. This included a high-throughput buffer exchange step using cation-exchange batch adsorption and the subsequent automated protein quantification with process feedback. As high-throughput compatible analytics, a reversed-phase ultra-high performance liquid chromatography (RP-UHPLC) method without sample preparation was developed, focusing on a short runtime for high efficiency. The final platform was used in a conjugation DoE, showing the capacity of the method for efficient process characterization. Finally, the comparability of the high-throughput results with experiments in a larger scale was demonstrated. The second part describes the establishment of an on-line monitoring approach for ADC conjugation reactions using UV/Vis spectroscopy2. First, a spectral change caused by the conjugation of the maleimide-functionalized surrogate drug to the thiol group of the engineered cysteines was detected. Spectra were recorded during the reaction in two setups with different detectors. Subsequently, the spectral change was correlated to off-line concentration data measured by RP-UHPLC using partial least-squares (PLS) regression. The calibrated PLS models enabled the prediction of the amount of conjugated drug directly from UV/Vis spectra. Both external validation data sets as well as cross-validation were used for model validation. The successful prediction of the reaction progress was shown with two different surrogate drugs in both setups. After covering high-throughput tools, analytics, and process monitoring in the first and second parts, the third part focuses on applying mechanistic understanding towards conjugation process development. In this section, a kinetic reaction model for the conjugation of ADCs was established and the application of the mechanistic model to process development was investigated3. Before model calibration, six model structures were set up based on different assumptions regarding the binding to the two available cysteines. All six models were fit to a calibration data set and the best model was selected using cross-validation. The results suggest that the attachment of a first drug to the mAb influences the attachment to the second binding site. An external data set including data outside the calibration range was used for the successful validation of the model. The validated model was then applied to an in silico screening and optimization of the conjugation process, enabling the selection of conditions with efficient drug use and high yield of the target component. Additional process understanding was generated by showing a positive effect of different salts on the reaction rate. Finally, a combination of the kinetic model with the monitoring approach of the second part was investigated. While the previous parts are primarily concerned with the conjugation reaction itself, the fourth part deals with the subsequent purification of the ADCs. A mechanistic model was established for the separation of ADC species with different DAR using hydrophobic interaction chromatography (HIC)4. This separation allows to set the target DAR also post-conjugation. For modeling the transport of solutes through the column and the adsorption equilibrium, the transport-dispersive model and a suitable adsorption isotherm were applied. First of all, a detailed characterization of the chromatography system and column was conducted, which served the calculation of a number of model parameters. The rest of the model parameters were determined by parameter estimation using numerical simulations. For the calibration, nine experiments with different linear and step gradients were run with varying load compositions. Peak positions as well as peak shapes were accurately described by the model for all components. Applying the final model to process optimization gave step gradients with improved yield, DAR, and concentration in the pool. The successful prediction of yield and DAR in the pool of the optimized gradients was validated with external data. In a first in silico study, model-based process control was used to react to variations in the preceding unit operation, ensuring a robust achievement of a critical quality attribute, the target DAR. A second in silico study shows that a linkage of the HIC model with the kinetic reaction model developed in the third part of this thesis can be profitably applied to process development. This ‘digital twin’ widens the system boundaries over two adjacent unit operations, which could enable the establishment of a flexible design space over more than one process step. In conclusion, the present thesis helps to shape the ADC process development of the future, able to cope with the challenges of a transforming biopharmaceutical industry. The whole process from the preparation of the conjugation sites over the conjugation reaction through to the purification of the conjugates was covered. Efficient characterization of the design space was demonstrated by incorporating tools like high-throughput experimentation combined with DoE, and mechanistic modeling techniques. The implementation of QbD relies on the establishment of suitable tools for acquiring enhanced process knowledge and for process monitoring and control. To this end, a PAT method for conjugation monitoring based on multivariate data analysis, and mechanistic models for conjugation and purification were developed. The presented studies showcase the realization of new ideas for exploiting the potential of digital tools for the specific challenges of ADC process development.