6.3.4 Decrease enzymatic reaction rate and reduce reaction time
Lowering pH and temperature of HCCF or reducing HCCF storage time can slow down the enzymatic reaction. These approaches can be coupled with other methods to further eliminate disulfide bond reduction. It should be noted that lowing pH is less frequently used since it changes microenvironment around mAb and may cause mAb aggregation (Trexler-Schmidt et al., 2010).
While there are different approaches to minimize disulfide bond reduction in manufacturing process, these approaches have some limitations: (1) inhibition of enzyme overexpression genetically in cells may increase the mAb development time, also this may increase the risk of not overexpressing the mAb correctly since the genetically modified mutant clone may not survive in antibody selection process at the first stage; (2) the type and the concentration of the inhibitors that eliminate disulfide bond reduction may vary with each mAbs, thus optimizing the inhibitor selection could be time-consuming; (3) the final inhibitor concentration present in drug substance needs be monitored and controlled, and it may require extra manufacturing steps to remove the inhibitors to meet the drug substance GMP release specifications for clearances; (4) air sparging, lowering the storage temperature and reducing the storage time may require extra financial investment for equipment thus increase the cost of the manufacturing process and reduce the flexibility of manufacturing process. Consequently, developing approaches to recover (or “savage”) the reduced disulfide bond should be considered in the manufacturing development process.
6.4 Rescue reduced mAb during downstream processing
Due to the limitations of operation flexibility and efficiency of controlling disulfide bond reduction, saving and recovering the already reduced mAb becomes a compelling alternative approach. Compared to the elimination approach, the recovery approach has advantages including: (1) saving the “waste” reduced mAbs, (2) further lowing the risks of mAb disulfide bond reduction in downstream process steps, (3) more flexible and economically favorable since this approach requires neither extra equipment nor extra operation steps.
Reduced mAb recovery involves the thiol-disulfide exchange reaction between the reduced mAb and a redox reagent (Bulleid & Ellgaard, 2011; Mamathambika & Bardwell, 2008). The thiol-disulfide exchange reaction includes two steps (Figure S4). In the first step, the nucleophilic thiolate group (S-) formed by free thiol deprotonation attacks one of the sulfur atoms of the redox reagent, and a disulfide bond forms between the redox reagent and the mAb (Figure S4A); in the second step, the remaining thiol group in mAb attacks the newly formed disulfide bond, releases the redox reagent and forms disulfide bond in mAb (Figure S4B). While the thiol-disulfide exchange reaction has been widely studied to provide deeper understanding of disulfide bond formation relations in cells (Cappel & Gilbert, 1988; Østergaard, Tachibana, & Winther, 2004; Tu, Ho-Schleyer, Travers, & Weissman, 2000), there are very few research in literature that applied this reaction to recover reduced mAb in the manufacturing process. In our recent in-house studies, we successfully recovered the reduced mAb by using redox reagent buffer cysteine/cystine as wash buffer in Protein A chromatography step (Figure S5, data not shown) (Tan et al., 2020; Tang et al., 2020). Based on the kinetic model developed in this study, the cysteine/cystine concentration, pH and mAb/redox buffer contact time were optimized to recover the disulfide reduced mAb. The results showed that mAb purity was improved from <5% in the load to > 90% in the elution. Cystine concentration and pH were found to be the most critical factors that affected reduced mAb recovery. To achieve a high mAb recovery ratio, buffer pH 8 to 10 and a minimum of 60 minutes mAb/redox buffer contact time was recommended from this study. The recovered mAb showed comparable properties to original intact mAb based on a variety of biochemical and biophysical analytical characterization. The studies demonstrated the feasibilities of applying recovery methods in downstream process to produce acceptable quality mAb product for potential clinical uses.
6.5 Disulfide bond reduction prediction and monitoring during manufacturing process
Besides disulfide bond reduction elimination approaches and recovering already reduced mAb approaches summarized above, proactively applying the process analytical technology (PAT) for disulfide bond reduction prediction and on-line/at-line monitoring during the manufacturing development process is also important (Chemmalil et al., 2020). PAT enables the detection of the disulfide bond reduction in early process steps thus allows the proper monitoring, and the prevention methods or rescue methods to be applied in time to ensure final product quality. In addition, the real-time analytical results of disulfide bond reduction can provide a quantitative understanding of the root cause of the disulfide bond reduction in manufacturing process and allow more efficient control to improve process design (Read, Park, et al., 2010; Read, Shah, et al., 2010; Simon et al., 2015).
It was also reported that, the correlation between cell culture redox potential and antibody reduction can be applied to predict sulfide bond reduction levels in early manufacturing process steps (Handlogten et al., 2020). Through online cell culture redox potential measurements and analyzing the amount of intact mAb at late stage of cell culture such as on Day 12 and Day 14, a redox potential of -70mV was found to be used as a cell culture redox potential threshold for an IgG2. Above this redox potential, there was minimal level of disulfide bond reduction while below this redox potential, the disulfide bond reduction level could vary significantly. These researchers further designed a redox reduction control system to maintain the cell culture redox potential higher than the threshold by adding CuSO4 and /or improving DO level, and demonstrated the success in keeping the disulfide bond reduction level of cell culture at a minimal level.
In addition to using redox potential as a predictive tool during cell culture step, free thiol levels in HCCF can also be controlled to predict disulfide bond reduction level and appropriate measures may be taken to mitigate the risk. In a typical monoclonal antibody manufacturing process with the Chinese Hamster Ovary (CHO) cell line, we have established three critical thresholds of free thiols in HCCF based on the disulfide reduction risk: < 100 μM (low risk); 100 - 200 μM (medium risk) and > 200 μM (high risk) (Du et al., 2018). For the high risk HCCF, in addition to low temperature storage and maintaining aeration, the harvested material is processed through Protein A chromatography as early as possible. To further de-risk the possible disulfide reduction, each elution pools are kept separately if multiple Protein A run cycles are performed.
For the downstream process steps such as Protein A chromatography and ion exchange chromatography, currently there are no reported research on applying on-line analysis methods for disulfide bond reduction. However, the success of incorporating on-line liquid chromatography (LC) for charge variant (Alvarez et al., 2011) and size variant analysis (Chemmalil et al., 2020) showed the potential of on-line disulfide bond reduction monitoring and analysis. Alvarez et al designed an in-house on-line liquid chromatography (LC)–mass spectrometry (MS) system that can directly trap fractions of interests from ion exchange chromatography (IEC) and size exclusion chromatography (SEC) to the reversed phase (RP) trap cartridges for desalting then subsequently analyze the fraction sample in MS (Alvarez et al., 2011). They successfully identified mAb charge variants due to different levels of glycosylation with this system. In recent years, commercial on-line process analysis systems have been developed. Patel and coworkers developed an on-line ion exchange LC using commercial PATROL® UPLC Process Analysis System (Patel et al., 2017), where sample pooling during downstream operations were guided based on near real time charge variant analysis. The commercial PATROL UPLC Process Analysis System used in Patel’s study can also be applied to develop the on-line SEC detection. In our recent in-house study an on-line SEC setup showed the capability of investigating IEC fraction size variant compositions (LMW, monomer, and HMW) under both native condition and denaturing condition (Chemmalil et al., 2020), and may be developed for disulfide reduced mAb identification in the future.
7 CASE STUDY ON DISULFIDE BOND REDUCTION ELIMINATION AND DISULFIDE REDUCED MAB RECOVERY
This case study illustrates an example of controlling disulfide bond reduction by using both a preventive strategy and a rescue strategy in manufacturing process, as detailed in Section 6 and illustrated in Figure 5. In brief, the preventive strategy essentially applied the temperature control as well as oxygen control to prevent disulfide reduction, and the rescue strategy demonstrated that the “waste” disulfide reduced antibody could still be rescued and recovered by re-oxidizing the reduced disulfide bonds during Protein A chromatography.
Figure 6A illustrates a comprehensive study plan, in which HCCFs of three mAbs (mAb-1, mAb-2, and mAb-3) were used for the study (Tan et al., 2020). Each HCCF was divided into two pools, which underwent two treatments and storage conditions: “good HCCF” (air sparging + 4 °C) and “bad HCCF” (nitrogen sparging + room temperature (19~25 °C)). Both HCCFs were processed through Protein A purification using two Wash 2 buffers (Control buffer and Redox buffer), respectively. Figures 6B and 6C compare product purity and aggregation using two Protein A wash arms for the three mAbs in both intact and reduced forms. It was observed that “good HCCFs” for all three mAbs maintained high product purity using both wash arms, suggesting that the preventive strategy (chilled storage temperature of harvested bulk with dissolved oxygen) was able to prevent the disulfide bond reduction prior to Protein A step. In contrast, “bad HCCFs” using the control wash condition (arm 1: high pH without redox wash) showed low purity (< 50%) for all three mAbs. However, high product purity was obtained by using the redox wash (arm 2), which demonstrated the effectiveness of redox wash to enhance the disulfide bond reformation on the Protein A column.
Implementation of either preventive strategy or rescue strategy or both clearly demonstrated that disulfide reduction issue could be resolved to achieve high-purity antibody product. Furthermore, this case study demonstrated that the redox wash has no negative impact on process yield or product quality. Extensive characterization of the recused antibody confirmed a complete formation of interchain disulfide bonds and comparable biophysical properties to the reference material. The detailed study was published in a separate paper (Tan et al., 2020).
8 CONCLUSIONS AND FUTURE PERSPECTIVES
The disulfide bond reduction during the biologics manufacturing process are getting more and more attention currently in the biotech industry as it is a frequently encountered technical challenge and its effective management will help meet critical requirement for protein quality specifications and therapy efficiency. In this paper, we reviewed the root cause of disulfide bond reduction, analytical methods for disulfide bond reduction analysis, as well as disulfide bond reduction elimination and recovery approaches. Furthermore, we suggest additional aspects of disulfide bond reduction issue worth being addressed in future antibody manufacturing development:
8.1 Minimizing disulfide bond reduction and rescue disulfide bond in purification platform
While it is difficult to adapt a truly pre-defined process for different individual mAb purification, the platform development approaches based on common unit operations have been proposed by many biopharmaceutical companies (Kelley, Blank, & Lee, 2009; Shukla, Hubbard, Tressel, Guhan, & Low, 2007). Since disulfide bond reduction mainly happens at the beginning of the purification process, the cell culture harvest and Protein A chromatography could be the main unit operations to align the strategy for disulfide bond reduction elimination and recovery approaches. For the harvest step, disk-stack continuous centrifuge is applied as a standard operation for large scale cell culture harvest and clearance. However, as discussed in previous sections, the cell lysis during the cell culture harvest may release intracellular enzymes, and lead to disulfide bond reduction. To address this issue, fully hermetic centrifuge may be used as platform harvest equipment. Compared with the disk-stack continuous centrifuge, the hermetic centrifuge machine can be fully filled with liquid to eliminate the air-liquid surface, which is known to be one of the root causes for cell lysis (H. F. Liu, Ma, Winter, & Bayer, 2010). Hermetic centrifuge machine is commercially available (Rose, 2008), however, operation parameters such as throughput, process speed and time need to be further optimized from facility fit perspective. Another direction to align the strategy of the disulfide bond reduction elimination approaches during cell culture harvest is using computational fluid dynamic simulation (CFD) to optimize centrifuge parameters. CFD can simulate the shear stress and support the estimation of cell lysis level under certain centrifuge operation conditions (Boychyn et al., 2004; Megson, Wilson, & MacGregor, 2002; Molina-Miras, Sánchez-Mirón, García-Camacho, & Molina-Grima, 2018). For Protein A chromatography step, using redox reagent buffer during wash step can successfully recover the reduced mAb. This could be aligned with platform wash buffer to recover the disulfide-reduced mAb or further reduce risk of the disulfide bond reduction for the intact mAb (Tan et al., 2020; Tang et al., 2020).
8.2 Identifying disulfide bond reduction risks using machine learning algorithms
In recent years, machine learning algorithms have been developed rapidly and started to gain more and more attention in the pharmaceutical industry. Several machine learning algorithms have been applied to predict mAb pH stability, thermal stability and chemical stability (Jia & Sun, 2017; King et al., 2011; Sankar et al., 2018). For instance, Sankar et al predicted mAb methionine oxidation risk based on a random forest prediction model (Sankar et al., 2018). In their studies, mAb structure-based features, sequence-based features and dynamic-based features were extracted as input data, and mAb peptide mapping results on oxidation were extracted as output data for model training. Similar machine learning algorithms were also developed recently for mAb deamidation risks prediction (Delmar, Wang, Choi, Martins, & Mikhail, 2019; Jia & Sun, 2017). The success of machine learning in predicting the mAb biophysical and biochemical stabilities provides the possibilities of applying machine learning algorithm to predict mAb disulfide bond reduction risks. One proposed approach is to use different mAb sequence-based features, 3-D structure-based features as data input, and CE NR-SDS analytical results as output features for model training. One of advantages of the machine learning prediction is to identify, in very early development stage, the mAbs that show high disulfide bond reduction risks. If the “easily reduced” mAb can be recognized early, an alternative mAb candidate can be selected to avoid the disulfide bond reduction, or the downstream development researchers can start to optimize the manufacturing process early enough to address the challenge. Another advantage of the machine learning approach is that it could identify the features such as structure-based features, sequence-based features and dynamic-based features that have higher impacts on mAbs disulfide bond reduction, and provide insights on the disulfide bond reduction level differences observed in different molecules. However, machine learning algorithm has its own challenges at this early stage of its development, and people should be aware of: (1) the data could be unbalanced, for instance, if there are more mAbs that show low disulfide bond reduction risk than mAbs that show high risks. This could increase the false positive and false negative prediction rate; (2) the data sets may be not large enough. The number of data sets are around 100~200 in existing research mentioned earlier. Thus, some advanced machine learning algorithms such as neural network or gradient boosting may not be able to be applied, otherwise it may lead to overfitting issues.
8.3 Disulfide bond reduction during multi-specific antibody purification
Though multi-specific antibodies (BsAbs and TsAbs) have not been as widely manufactured as monoclonal antibodies, there is a growing interest in multi-specific antibody purification in the biopharmaceutical industry in recent years (Brinkmann & Kontermann, 2017; Y. Li, 2019; Swope et al., 2020; Tustian et al., 2016; X. Yang et al., 2015). Taking BsAb as examples, during bispecific antibody purification process, the inter-chain disulfide bond between the heavy chains in two parental mAbs need to be reduced first to generate half-mers and followed by free thiol oxidization to form bispecific antibodies. For instance, one method for bispecific antibody purification is to capture parental mAb using Protein A chromatography separately followed by incubating mixed parental mAbs in phosphate-buffered saline (PBS) buffer that contains 2-MEA (reducing agent) for disulfide bond reduction. Then the mixture is re-oxidized by dialysis against PBS buffer without 2-MEA (Paul et al., 2016). In addition to purification process, disulfide bond reduction and re-oxidization also influence bispecific antibody stabilities and impurities level. Researchers found that intact bispecific antibody are susceptible to the reduction condition. Compared with mAbs, disulfide bond reduction for BsAbs is much more severe, since it could lead to both aggregation and mispairing. Higher HMW levels have also been observed for the bispecific antibody when the bispecific antibody and the parental mAb were reduced and re-oxidized under similar conditions. In addition, disulfide bond mispairing may lead to homodimer formation(Kuglstatter et al., 2017). Thus, both disulfide bond reduction elimination and its recovery are critical for bispecific antibody processing, and the learnings on mAb disulfide bond reduction could also benefit bispecific antibody development.