As modern medicines, biopharmaceuticals are characterized by high functional specificity and offer innovative solutions for the prevention or treatment of complex diseases, including autoimmune, infectious, genetic diseases, and cancer. The functionality of biopharmaceuticals is rooted in the product’s structure, whereby its quality attributes are intrinsically linked to the production process, making process understanding the key to ensuring product quality. Thus, the overarching objective of biopharmaceutical process development is to enhance process understanding through methodological and systematic approaches.
Beyond monoclonal antibodies representing the core platform in the biopharmaceutical industry, vector technologies have emerged as versatile nanocarrier platforms, particularly for vaccines and gene therapeutics. Nowadays, the leading non-viral vector classes comprise lipid nanoparticles (LNPs) and virus-like particles (VLPs), which are associated with lower safety concerns in comparison to viral vectors. LNPs are utilized for the protected delivery of nucleic acids and are composed of a defined ratio of distinct lipid types. ... mehrIn contrast, VLPs are formed through the spontaneous assembly of recombinant capsid proteins, thereby mimicking the structure of the native virus. VLPs can be used to present heterologous surface proteins or for the protected delivery of a post-inserted payload. As VLPs are recombinantly produced, a cascade of downstream processing (DSP) steps is required for their purification, involving the removal of process- and product-related impurities. However, the nanocarriers’ overall larger size compared to monoclonal antibodies, as well as their structural diversity within and between those non-viral vector classes, pose new challenges in the development of purification processes and analytics.
Recent advancements in nanocarrier purification promise a broader application of size-selective separation techniques and elevate the potential of precipitation and cross-flow filtration (CFF) for standardized purification processes. In contrast to size exclusion chromatography, precipitation and CFF offer universal applicability alongside fundamental process requirements covering reproducibility, scalability, and efficiency, but without being restricted by limited loading capacities. Presumably due to the recent emergence of the LNP technology, CFF process development for LNPs has not yet reached a mature state. The present limitations in process understanding are to be addressed through dedicated process parameter studies and analysis of process intermediates. In contrast, CFF is well established for VLP purification and has recently been explored for the processing of VLP precipitates, as selective precipitation is commonly employed to capture VLPs from clarified lysate. Consequently, ongoing efforts in process development for VLPs are already directed towards process integration, intensification, and standardization, with the objective of streamlining their DSP.
Beyond standard analytics, the concept of process analytical technology (PAT) enables real-time monitoring and control of critical process parameters and quality attributes by combining advanced sensor technologies and chemometrics. As with the DSP development, the maturity of sensor-based PAT tool development for monitoring varies across different separation techniques and biopharmaceuticals. PAT tools relying on ultraviolet-visible (UV/Vis) spectroscopy are leading for monitoring chromatographic purification of monoclonal antibodies, with the approach undergoing expansion to encompass filtration processes, including those for VLPs. Shortcomings persist in the development of PAT tools for the monitoring of precipitation processes, including VLP precipitation, as well as LNP purification in general. These shortcomings are accompanied by considerable challenges, requiring the utilization of alternative sensor technologies. In the field of optical spectroscopy, scattering-based sensor technologies such as Raman and dynamic light scattering (DLS) hold promise in addressing the individual challenges associated with the monitoring of VLP precipitation and LNP purification, respectively.
Supported by references, Chapter 1 provides a more detailed introduction that summarizes the theoretical background, recent advancements, and remaining research gaps in the field. As outlined in Chapter 2, the objective of this thesis was to advance the primary purification of the non-viral vector classes LNPs and VLPs through (I) size-selective separation and (II) sensor-based analytics and PAT tools. Aiming for standardized, scalable, size-selective processing, innovations in primary purification were driven through the integration of CFF. In order to obtain further insights into these purification processes and thereby enhance process understanding, tailored analytics and PAT tools were developed based on charged aerosol detection (CAD), DLS, and Raman spectroscopy. As demonstrated in diverse, consecutive studies, the integration of CFF enabled efficient processing. Moreover, analytics and the implementation of process monitoring tools yielded valuable insights into these processes.
For LNPs, standard analytical procedures are primarily applied to the microfluidic mixing step and typically cover particle size, surface charge, and encapsulation efficiency. Although lipid composition directly affects the aforementioned LNP attributes, the quantification of lipids remains an uncommon practice. Reversed-phase (RP)-high performance liquid chromatography (HPLC) in conjunction with CAD previously demonstrated a high degree of selectivity and sensitivity for lipids. Thus, a RP-HPLC-CAD method was developed to quantify lipids in LNP process intermediates (Chapter 3). The method enabled precise lipid quantification of former LNPs, providing the lipid molar ratio as an additional LNP attribute and the lipid recovery at each process step as a key process performance parameter. The method’s applicability and its benefits were demonstrated in a process parameter study involving microfluidic mixing and dialysis, with variations in the mixing total flow rate (TFR). LNP size exhibited a TFR-dependency, which is in accordance with the results of previous studies, whereas all novel findings derived from lipid analytics were TFR-independent. The lipid concentrations showed a batch and process step dependency, while the lipid molar ratio was batch-independent and constant throughout the processing. Another batch-dependency was found for the performance of the dialysis step, with lipid loss-associated decreases in lipid recovery. In summary, utilizing lipid analytics supports the characterization of LNPs and processes. To contribute to new perspectives on process development for LNPs and to enhance process understanding, LNP purification by CFF may also benefit from its application.
Although CFF offers a scalable, time- and cost-efficient alternative to dialysis for buffer exchange to remove residual solvent and raise the pH, with seamless, integrated LNP concentration through volume reduction, CFF-based LNP purification remains unexplored. This includes the effects of membrane characteristics and CFF process parameters on LNP attributes, as well as CFF monitoring. Thus, a CFF parameter study was performed, with at-line monitoring of the LNP size using DLS (Chapter 4). Four experiments were conducted, in which LNPs were purified by dialysis and CFF for side-by-side comparison, with the latter as constant-volume diafiltration (DF) followed by ultrafiltration (UF). No consistent impact on LNP attributes was observed from the systematic variations of distinct membrane characteristics and CFF process parameters. This included persistent lipid molar ratios and lipid recoveries, as revealed by the developed RP-HPLC-CAD method (cf. Chapter 3). However, time-resolved analysis supported by at-line monitoring using DLS revealed a linear dependency of LNP size on CFF processing time. Differences in particle size alternation between dialyzed and CFF-purified LNPs led to the hypothesis of an interplay among pump-induced shear forces, LNP fusion, and LNP equilibrium size. In summary, this study provides novel insights into LNP purification by CFF. Alongside the underlying mechanisms driving these increasing particle sizes, which still need to be explored, these findings point towards future size-controlled CFF-based LNP purification.
With respect to the size-selectivity of CFF, precipitation also exploits the size difference between the product and impurities. For the primary purification of VLPs from clarified lysate, the conjunction of selective precipitation and CFF-based microfiltration (MF) for precipitate processing previously demonstrated improvements in productivity, purity, and yield compared to VLP recovery by centrifugation, thereby addressing scalability, process integration, and intensification. Specifically, two consecutive constant-volume DF steps on one CFF unit enable precipitate washing to remove soluble impurities, followed by VLP re-dissolution. During the latter step, the VLPs pass through the MF membrane and are thereby separated from the irreversibly precipitated species. However, VLP concentration is constrained by DF-induced permeate dilution, and purity is limited by the presence of residual precipitant. To improve VLP concentration and purity through process integration, a novel dual-stage CFF set-up was developed for the integrated, seamless recovery of VLP precipitates (Chapter 5). The dual-stage CFF set-up was equipped with a MF/UF membrane configuration. With this set-up, DF-based VLP recovery involved VLP re-dissolution and their isolation through the simultaneous depletion of residual precipitant under constant volume in the second membrane stage, and their subsequent concentration using an integrated UF step. Alongside precipitant depletion, the dual-stage CFF set-up allowed for the effective removal of co-redissolved impurities. In summary, the dual-stage CFF setup with the MF/UF membrane configuration establishes a foundation for standardized precipitate processing and has promising future applications for diverse VLPs and biopharmaceuticals, as well as potential transferability to crystallization processes.
Next to the high impurity levels in primary purification processes, selective precipitation inherently yields particulate precipitates and turbidity through the addition of precipitant, rendering direct quantification of the precipitated product impossible and posing distinct challenges for developing sensor-based PAT tools. In such environments, widely applied monitoring tools based on UV/Vis spectroscopy and chemometrics reach their limits, while Raman spectroscopy represents a promising sensor alternative. However, in general, in-depth insights into data generation and spectral preprocessing for the knowledge-driven development of Raman-based PAT tools remain relatively rare. Thus, a systematic approach was pursued for the development of a Raman-based PAT tool for VLP precipitation monitoring (Chapter 6). To ensure data diversification, batch and fed-batch precipitation experiments were conducted using conditioned lysate, resulting in variations in the precipitation dynamics and background composition. A systematic combination of spectral preprocessing operations and spectral feature analysis was used to evaluate the effectiveness in eliminating differences in spectral backgrounds as well as interferences caused by the presence of precipitates and turbidity. The concentration of precipitated VLPs was predicted with preprocessed Raman spectral data and partial least squares (PLS) regression, and effects of individual preprocessing operations on PLS model performances were evaluated. In summary, Raman-based PAT enables VLP precipitation monitoring in particulate, turbid lysate. The gained insights through a deliberately systematic development approach underscore the importance of data diversification and targeted spectral preprocessing for future PAT tool development.
Compared to the controlled addition of precipitant in batch and fed-batch precipitation processes, precipitant depletion in dynamic processes should be monitored through precipitant quantification, such as during the filtration-based VLP recovery using the presented dual-stage CFF set-up (cf. Chapter 5). Furthermore, the quantification of the re-dissolved VLPs accumulated in the second membrane stage is of particular interest; however, sampling during the DF process leads to product loss and perturbs its dynamics. PAT tool development for monitoring two components simultaneously from the same acquired spectral data clearly benefits from a systematic development approach, such as the one presented for VLP precipitation monitoring (cf. Chapter 6). However, a limitation observed was the transfer from batch precipitation-derived Raman-based PAT tools for monitoring fed-batch precipitation, which, however, might be more straightforward in the processes where fewer spectral interferences are expected. Under the scope of model transfer, Raman-based PAT tools were developed from stock solutions of pure components for simultaneous monitoring of VLP accumulation and precipitant depletion in the filtration-based VLP recovery step (Chapter 7). Investigating spectral contributions of VLPs and precipitant using pure component spectra revealed attribute-specific spectral features as well as a higher inherent sensor sensitivity towards the precipitant than VLPs. Accordingly, attribute-specific preprocessing operations were selected to account for these differences as well as for detector saturation effects, and several regression models were built with preprocessed spectral data from pure component solutions. Concurrently, the dual-stage CFF set-up was equipped with a Raman spectrometer within an on-line monitoring loop in the second membrane stage, and process spectral data were acquired from three diversified CFF experiments. From these CFF-derived spectral data, VLP and precipitant concentrations were predicted by application of the transferred models. The fact that the final prediction accuracy was higher for the precipitant than VLPs despite attribute-specific selections of preprocessing operations and models unraveled the importance of attribute-specific sensor selection. In summary, model transfer is demonstrated through the systematic development of Raman-based PAT tools, thereby emphasizing the critical role of attribute-specific sensor selection for future PAT tool development.
This thesis presents advances in DSP development through the integration of size-selective separation techniques and sensor-based PAT tools, tailored to the primary purification of the current leading non-viral vector classes, namely LNPs and VLPs. Despite the differences in developmental maturity between LNP and VLP process development, the tailored integration of CFF enabled efficient processing, and analytics and process monitoring tools provided valuable insights into these processes. Overall, the methodological and systematic approaches presented demonstrate the potential to be transferred to other biopharmaceuticals or processes. Collectively, these advances highlight the potential of size-selective purification and scattering-based sensor technologies to address the specific challenges associated with novel vector technologies, while ultimately supporting the overarching goal of enhancing process understanding in biopharmaceutical process development. Such developments lay the foundation for future standardized and controlled DSP of vectors in the biopharmaceutical industry.
1