Anurag S. Rathore is a professor in the Department of Chemical Engineering at the Indian Institute of Technology Delhi and a member of BioPharm International's Editorial Advisory Board, Tel. +91.9650770650, asrathore@biotechcmz.com.

Anurag S. Rathore

Process analytical technology (PAT) is an increasingly important aspect of biopharmaceutical manufacturing processes. FDA in 2004 described a regulatory framework for PAT and has been urging the industry to voluntarily develop innovative tools and techniques for quality control and assurance in biopharmaceutical manufacturing unit operations (1). The quality-by-design (QbD) framework encourages the use of PAT tools to “build quality into the process” rather than “testing quality into the product” (2). The importance of PAT tools is amplified in the case of continuous manufacturing processes, where individual batches are not well-defined and critical quality attributes (CQAs) of the therapeutic must be consistently maintained, monitored, and controlled over months of operation, rather than simply tested at the end of each unit operation or each batch (3–4). The need for real-time monitoring and control of CQAs and critical process parameters (CPPs) in continuous manufacturing has led to the adaptation of traditional end-of-batch quality testing techniques, such as analytical high-performance liquid chromatography (HPLC) and ultraviolet-280 (UV-280) spectroscopy, into in-line at-line modalities with periodic sampling from continuous unit operations (5–7). It has also spurred development of new analytical tools that provide information in near real-time and can be placed in-line in process flow streams.

In-line spectroscopic sensors using infrared, near-infrared, or Raman spectroscopy are one such new class of tools finding multiple PAT applications in bioprocessing (8). Spectroscopic PAT applications typically consist of in-line sensors combined with statistical or modeling methodology to monitor, control, and/or predict biotechnology processes. The key advantage of spectroscopy is the ability to gain information about the presence and quantity of multiple analytes in the process mixture simultaneously in a few seconds or less, overcoming a major analytical bottleneck and facilitating real-time control decisions (9). Spectroscopic techniques are also non-degradative and provide a wealth of spectral information that can be used in tracking process trajectories and flagging deviations (10,11). Thus, spectroscopic probes and flow cells are well-suited to PAT applications, particularly in tandem with multivariate data analytics (MVDA) tools, which are critical for reducing the dimensionality of large spectral datasets and extracting statistically significant quantitative information (12).

 Summary of spectroscopy-based downstream process analytical technology (PAT) applications in recent literature.

 summarizes the recent literature of the past five years on implementing spectroscopy-based analytical tools in downstream bioprocessing (13–29). The use of spectroscopic immersion probes as PAT tools in upstream microbial and mammalian bioreactors has been extensively reported in the recent literature for identification and quantification of proteins, by-products, and substrates (30–32). However, their use in downstream processing is a more recent development. Near infrared spectroscopy (NIRS), in particular, has been demonstrated to have a wide variety of potential applications in different downstream unit operations, including capture chromatography (24), protein PEGylation reactions (25), and tangential flow ultrafiltration (26). These three case studies demonstrate the versatility of NIRS as a PAT tool and showcase the common underlying framework of the different applications, namely the collection of real-time spectra followed by comparison with spectral calibration libraries using multivariate statistical techniques. NIRS is demonstrated to be a reliable tool for acquiring rapid quantitative information in a range of downstream bioprocesses, and a key enabler for real time control.

Case study 1: NIRS for controlled loading.

 Protein A capture chromatography is typically the first downstream step in manufacturing of monoclonal antibodies (mAbs). One of the key difficulties in continuous Protein A chromatography is handling variability in the titer of the upstream material. The product titer is expected to change over time in the case of upstream perfusion cell culture systems (33). Even in the case of fed-batch processes, batch-to-batch variability necessitates flexibility in the downstream capture step (34). An increase or decrease in the protein concentration in the load stream affects the dynamic binding capacity of the Protein A resin, leading to changes in process performance (35). Over-loading the Protein A column with high-titer feed can lead to the loss of expensive mAb product, while under-loading decreases the resin utilization of the Protein A resin, one of the most expensive consumables that has been reported to contribute to up to 60% of the costs of the downstream processing train (36).

CLICK FIGURE TO ENLARGE. All figures are courtesy of the authors. 

 Near infrared spectroscopy (NIRS) as a process analytical technology tool for control of loading in continuous Protein A chromatography by monitoring load titer and column breakthrough percentage in real time. MVDA is multivariate data analytics.

 has been used to address this challenge of handling potential titer variability in continuous capture chromatography while maintaining resin utilization and preventing both under- and over-loading (24). NIRS flowthrough cells placed in the loading stream and the outlet of the load column were used to collect spectra of the harvest and column flowthrough every three seconds. These spectra were passed to online MVDA models calibrated with reference spectra to determine the concentration of mAb in the harvest and flowthrough to within ± 0.05 mg/mL. The real-time concentration data were used to calculate both the total mg of mAb loaded on the column as well as the percentage breakthrough from the column, and these were used to make control decisions to start and pause the loading in a three-column periodic counter current Protein A process on a continuous chromatography system (Cadence BioSMB, Pall). The normal operating range of the control system allowed for concentration variations in the harvest between 3–8 g/L, achieving optimal resin utilization as well as process scheduling within this range.

The NIRS-based PAT control strategy was tested by inducing a range of linear and non-linear deviations in the load titer in real time. The deviations were designed to closely simulate those which would potentially occur during a perfusion process. The NIR flow cells provided the advantage of monitoring not only loading but also changes in the column binding capacity in real time, which is useful for providing early warning of resin degradation or column quality issues and facilitating optimal loading without relying on adsorption isotherm models, which may not be valid after multiple cleaning cycles. The system allowed resin utilization to be maximized, lowering consumable costs by ensuring that the total protein processed per mg of resin was consistently maintained. The NIRS flow cells enabled online measurement of concentration and facilitated real-time control decisions for increased efficiency, flexibility, and agility of the continuous chromatography process in the face of unexpected deviations.

Case study 2: NIRS for monitoring/control.

 Protein modification with biocompatible polymers, such as polyethylene glycol (PEG), is often used to improve the pharmacological properties and stability of biotherapeutics (37). Manufacturing of PEGylated drugs requires a PEG conjugation reaction to be carried out on the purified drug substance at the end of the downstream train (38). Proper control of the PEGylation reaction is critical to maximize PEGylation efficiency while minimizing the presence of over-PEGylated variants. Various critical process parameters affect PEGylation quality, including pH, reaction time, and the concentration and order of addition of the reactants (39). In the present case study, the PEGylation reaction of recombinant human granulocyte colony stimulating factor (rh-GCSF) was considered. The challenge was to monitor the PEGylation process trajectory and reaction kinetics to flag potential deviations in real time, as well as to control the reaction quenching to optimize the production of the desired monoPEGylated variant before its further conversion into undesired diPEGylated and multiPEGylated forms.

A few different PAT tools have been explored in the literature for monitoring and control of PEGylation. Some researchers used at-line size exclusion chromatography to monitor the progress of the PEGylation reaction, though this was not suited for real-time control due to the long method run times of >30 minutes for an overall reaction time of one to two hours (40). Other researchers used at-line matrix assisted laser desorption ionization–time of flight (MALDI–TOF) mass spectrometry for tracking the kinetics of PEGylation, though the time scale of analysis was again too long at over one hour (41). The use of NIRS as a PAT tool in the present case study overcame the bottleneck of analysis time and was able to track the conversion of rhGCSF into its monoPEGylated and multiPEGylated forms on a time scale of a few seconds (25). The NIRS spectra were acquired every three seconds and compared against a calibrated MVDA regression model, built using control runs of the reaction with orthogonal HPLC-based quantification of the PEGylated variants. The spectra acquired during the control runs were also used to develop a multivariate batch evolution model of the ideal reaction trajectory. A flow chart of the NIRS-based PAT tool is shown in 

 Near infrared spectroscopy (NIRS) as a process analytical technology tool in PEGylation reaction of recombinant human granulocyte colony stimulating factor (rhGCSF) by monitoring concentrations of mono- and di-PEGylated species over time and tracking deviations in the process trajectory.

The PAT tool was demonstrated in various control and deviated reaction runs in which online spectra were acquired by the NIRS immersion probe and used not only for quantification of the PEGylated variant, but also for comparison against the ideal reaction trajectory. This allowed process deviations to be identified and flagged in case a statistically significant difference was found between the trajectories of the current process versus the ideal one. A range of deviations were found to be identifiable by the PAT tool, including incorrect concentration of chemical additives, incorrect ratio of PEG to rhGCSF, incorrect order of addition of the reactants, and incorrect quenching time. The impact of each deviation on CQAs of the PEGylation reaction product was characterized, and each resulted in lower product purity and process yield. The NIRS probe was demonstrated to provide critical real-time information and form the basis of a robust PAT tool for monitoring and control of the PEGylation reaction of rhGCSF. The overall approach is generalizable to other downstream reaction steps, such as enzyme reactions or esterification.

Case study 3: NIRS for control of retentate concentration.

 Tangential flow ultrafiltration (UF) is a key unit operation in downstream processing of biotherapeutics, used for concentrating and buffer exchanging the in-process drug substance into the desired target concentration and formulation of the final drug product (42). This is typically the last unit operation prior to fill/finish and packaging. The concentration of the biotherapeutic in the final drug product is a CQA determined solely by this unit operation, necessitating robust control. In batch mode, concentration is achieved by recirculating the drug substance held in a large tank through the UF membrane until the desired volume reduction is achieved (43). However, batch-mode recirculation is not possible in the case of constantly incoming flow streams in continuous processing. The solution is to use single-pass UF in which the membrane module has a larger area and a long flow path, facilitating volume reduction of the feed stream in a single pass without the need for recirculation (44).

There are several operational challenges that must be overcome, however, to ensure that the retentate stream emerges from the single-pass tangential flow filtration (SPTFF) module consistently at the fixed target concentration. Reversible and irreversible membrane fouling as well as deviations in the concentration of the incoming flow stream from prior unit operations, such as polishing chromatography, lead to changes in the concentration factor achieved in a single pass (45,46). Therefore, a PAT strategy is needed to monitor the concentration of the incoming feed stream and control the flux across the membrane, to ensure that the retentate concentration does not vary over the course of the process. Researchers demonstrated a strategy leveraging in-line NIRS flow cells in the feed and retentate streams of an SPTFF module to make flux-based control decisions and ensure consistency in the retentate concentration, as shown in the schematic in 

 Near infrared spectroscopy (NIRS) as a process analytical technology tool for control of retentate concentration in single pass tangential flow ultrafiltration in continuous formulation of monoclonal antibodies (mAbs). UF is ultrafiltration.

The NIRS flow cells were able to measure the concentration of mAb in the range of 0.5–200 g/L, a significant improvement over UV-based quantification methods, using suitably calibrated spectral libraries. The control decisions were made on the basis of the real-time NIRS data as well as a pre-characterized design space for the limits of maximum flowrates and concentration factors achievable for a given feed concentration and flow rate within the pressure limits of the membrane module. The NIRS data were analyzed, and closed-loop control established with a permeate pump and retentate valve, allowing flux control on the time scale of <1 second. Control and scheduling decisions were made for integrating the SPTFF step with the rest of the continuous downstream train and for ensuring timely cleaning. The overall PAT strategy was demonstrated over two 12-hour case studies. The NIRS monitoring sensors enabled the development of a robust PAT control strategy to ensure that retentate concentration targets were consistently met over long continuous campaigns.

The need for rapid, non-degenerative, and robust analytical tools will continue to grow as the biotherapeutic industry embraces the PAT and QbD paradigms. Spectroscopic sensors are uniquely suited to fulfill this need as spectral information can be collected in a few seconds and analyzed in milliseconds using multivariate techniques. The spectra can yield a wealth of information about the in-process sample, including quantification of multiple analytes and the overall health of the process compared to an ideal trajectory. Spectroscopic applications in upstream microbial and mammalian cell cultures have grown exponentially in the past 10 years, and this can be expected to happen in the case of downstream unit operations in the near future. The rapidly growing interest in continuous downstream processes in both academia and industry will amplify this effect due to the unique real-time PAT control challenges that arise when batch-mode product quality testing is no longer an option. To truly bring spectroscopic applications into a manufacturing setting, however, the industry needs to gain confidence in the use of multivariate calibration models, which can only be done by having robust guidelines and frameworks for spectroscopic model calibration, validation, and periodic checking to account for shifts in spectroscopic sensor data over months or years. More fundamental studies are also needed to draw clear relationships between biotherapeutic CQAs and their effect on spectral features, as has been done in the case of generic pharmaceuticals. Finally, the ability of spectroscopic sensors to collect time-stamped spectral process signatures at high frequency over months or years is also a potential advantage that can be leveraged to create large-scale manufacturing datasets providing deep process history information, which can be of use to regulatory bodies.

Guidance for Industry, Q8 (R2) Pharmaceutical Development

 (CDER, CBER, November 2009).
2. A.S. Rathore and H. Winkle, 

 27 (1) 26–34 (2009).
3. J.Glassey, et al., 

 6 (4) 369–377 (2011).
4. A.S. Rathore, R. Bhambure, and V. Ghare, 

 398 (1) 137–154 (2010).
5. A. Tiwari, et al., 

 90 (13) 7824–7829 (2018).
6. S. Huffman, K. Soni, and J. Ferraiolo, 

 12 (8) 66–72 (2014).
7. M. Jiang, et al., 

 114 (11) 2445–2456 (2017).
8. K. A. Bakeev, 

Process Analytical Technology: Spectroscopic Tools and Implementation Strategies for the Chemical and Pharmaceutical Industries

 (John Wiley & Sons, West Sussex, UK, 2nd Ed. 2010).
9. M. Rüdt, T. Briskot, and J. Hubbuch, 

 1490, 2–9 (2017).
10. R.W Kessler, W. Kessler, and E. Zikulnig-Rusch, 

 88 (6) 710–721 (2016).
11. J. ClaSSen, et al., 

 409 (3) 651–666 (2017).
12. L. Eriksson, et al., 

Multi- and Megavariate Data Analysis: Basic Principles and Applications

 (MKS Umetrics AB, Malmö, Sweden, 2013).
13. N. Brestrich, et al., 

 112 (7) 1406–1416 (2015).
14. S. Hansen, et al., “Mid-UV Protein Absorption Spectra and Partial Least Squares Regression as Screening and PAT Tool,” in 

Preparative Chromatography for Separation of Proteins

, A. Staby, A.S. Rathore, and S. Ahuja, Eds. (John Wiley & Sons, 2017), pp. 501-536.
15. M. Rüdt, et al., 

 114 (2) 368–373 (2017).
16. N. Brestrich, T. Hahn, and J. Hubbuch, 

 1437, 158–167 (2016).
17. M. Ghodbane, et al., “Analytical and Data Strategy for Continuous Downstream Manufacturing,” in 

Integrated Continuous Biomanufacturing IV

, V. Warikoo, et al., Eds, ECI Symposium Series, 2019.
18. L. Rolinger, 

 412 (9) 2123–2136 (2020).
19. S. Großhans, et al., 

 1547, 37–44 (2018).
20. A. Sanden, et al., 

 1608, 460410 (2019).
21. D.G. Sauer, et al., 

 116 (8) 1999–2009 (2019).
22. C. Walther, 

 49 (7) 1113–1121 (2014).
23. N. Walch, et al., 

 14, 1800521 (2019).
24. G. Thakur, V. Hebbi, and A. S. Rathore, 

, 117 (3) 673-686 (2020).
25. V. Hebbi, G. Thakur, and A. S. Rathore, 

 325 (10) 303–311 (2020).
26. G. Thakur, S. Thori, and A. S. Rathore, 

, 613 (1) 118492 (2020).
27. F. Feidl, et al., 

 35 (5) e2847 (2019).
28. S. Goldrick, et al., 

 8 (9) 1179 (2020).
29. D. Yilmaz, et al., 

 36 (3) e2947 (2020).
30. S.M. Mercier, et al., 

 16 (1) 25–35 (2015).
31. N.R. Abu-Absi, et al., 

 2 (3) 267–284 (2014).
32. J. Claßen, et al., 

 409 (3) 651–666 (2017).
33. L. Chemmalil, et al., 

 17 (12) 3757–3765 (2020).
34. F. Li, A. Shen, and A. Amanullah, 

Pharmaceutical Sciences Encyclopedia: Drug Discovery, Development, and Manufacturing

 2, 1–38 (2010).
35. S. Hober, K. Nord, M. Linhult, 

 848 (1) 40–47 (2007).
36. A. Rathore, et al., “Re-use of Protein A Resin: Fouling and Economics,” 

 28 (3) 2015.
37. D. Pfister and M. Morbidelli, 

 180, 134–149 (2014).
38. S. Jevsevar, M. Kunstelj, and V.G. Porekar, 

 5 (1) 113–128 (2010).
39. R. Shekhawat, et al., 

 14 (3) e0212622 (2019).
40. J. Behi, et al., 

International Journal of Biological Macromolecules

 109, 888–895 (2018).
41. D.H. Na, Y.S. Youn, and K.C. Lee, 

 17 (19) 2241–2244 (2003).
42. H. Lutz, 

, (Woodhead Publishing, Cambridge, UK, 2015).
43. L.J. Zeman, and A.L. Zydney, 

Microfiltration and Ultrafiltration: Principles and Applications

 (CRC Press, Boca Raton, FL, 2017).
44. E. Thomas, et al., “Application of Single Pass TFF to Enable Intensified and Continuous Biological Manufacturing,” in 

Integrated Continuous Biomanufacturing III

 (ECI Symposium Series, 2017) pp. 1.
45. C. Casey, et al., 

 384, 82–88 (2011).
46. A. Arunkumar, et al., 

Garima Thakur is a graduate student in the Department of Chemical Engineering at the Indian Institute of Technology New Delhi, and Anurag S. Rathore*, asrathore@biotechcmz.com, is a professor in the Department of Chemical Engineering at the Indian Institute of Technology New Delhi and a member of 

*To whom all correspondence should be addressed.


Vol. 45, No. 3
March 2021
Pages: 32–40

When referring to this article, please cite it as G. Thakur and A.S. Rathore, “Near Infrared Spectroscopy as a Versatile PAT Tool for Continuous Downstream Bioprocessing,” 

Articles

The need for real-time monitoring and control has spurred the development of new analytical tools.

Near Infrared Spectroscopy as a Versatile PAT Tool for Continuous Downstream Bioprocessing

Biotherapeutics are an emerging class of treatment modalities that are manufactured using living cells. They have been successfully used for treating many life-threatening and chronic diseases. Compared to the traditional small-molecule (pharmaceutical) drugs, biotherapeutics are complex and have the ability to bind to more than one target molecule. Biosimilars provide a more affordable treatment option, and this is likely to become more relevant in the future as the affordability of these products remains critically poor in emerging and underdeveloped economies (1–3). The biopharmaceutical industry faces an increasing demand to accelerate process development while saving on cost and time. A possible solution to alleviate this is by using high-throughput potential tools such as surface plasmon resonance (SPR) to measure biomolecular interactions in real-time and in a label-free environment (4, 5).

	Biomolecular interaction analysis (BIA) can be used in the following ways (3):

		To monitor the level of interactions between two or more species

		To determine the affinity of the interactions

		To estimate the actual association and dissociation rates

		To measure the concentration of one of the species.

	SPR is a rapidly emerging tool for studying ligand binding interactions with membrane proteins, which are the major molecular targets for biopharmaceutical products (6). In SPR, one of the interactants is immobilized to the sensor surface and the other is kept free in solution and passed over the surface. Association and dissociation are measured and displayed in the form of a sensorgram. In this 41st article in the Elements of Biopharmaceutical Production, the authors present the basics of SPR as well as the various applications it offers in biopharmaceutical analysis. 

	Detection of specificity is comparably easy to perform in SPR. An analyte is injected and, after a certain amount of time, the response is measured. Plotting the response curve against sample number indicates the relative strength of binding. Bulk response caused by the medium in which the analyte is dissolved is the major drawback but can be easily avoided with proper reference measurements (7, 8).

	Analyte concentrations are measured at very high ligand densities on the sensor surfaces. The rate of the binding constant is proportional to the concentration of the analytes. After a standard curve has been created, unknown samples can be measured quickly.

	Equilibrium analysis is used to determine the strength of the binding. The first type is performed by using several analyte concentrations, which involve flowing the analyte over the ligand until the signal levels out and the net association is equal to the dissociation. By plotting the maximum response versus the analyte concentration, a line can be fitted to estimate the affinity constant (KD). The second type of experiment involves putting the two interactants together and incubating them until equilibrium. One of the interactants is to be constant, and the other is varied over a range of concentration. The concentration of the free analyte can then be determined after equilibrium.

	Kinetic rate analysis is used to investigate the behavior of the system. Interaction kinetics describes the interaction between one or more components. After the interaction, the components leave each other unchanged as opposed to enzyme kinetics.  The rate of association is determined in real time when the analyte flows over the ligand. Over time, the buffer replaces the analyte and the dissociation rate of the analyte is monitored. Both the association and dissociation curve can be fitted to one of the chosen models. In addition, the equilibrium constant can be calculated (9, 10). There are two types of experimental approaches available for kinetics experiment: multi-cycle kinetics (classical “standard” approach) and single-cycle kinetics (see 

	Various techniques are used to study structure-function affinity. The function is measured in terms of specificity (affinity), rate and equilibrium constants, as well as thermodynamic properties. While with the majority of SPR experiments the interaction conditions are held constant, varying these conditions (e.g., the temperature) can reveal important thermodynamic properties. SPR systems are capable of measuring the specific ligand–analyte interaction in real-time, which enables the researcher to simultaneously estimate both the rate and equilibrium constants (11, 12).

	Calibration free concentration analysis (CFCA) has been developed for measuring the active concentration of a ligand without a calibration curve (9). The method makes use of the mass transport limitation, which occurs when high-density ligand surfaces are used. By injecting the analyte at two different flow rates (e.g., 10–90 µl min-1), the active analyte concentration can be calculated from the slopes of the curves (13).

 shows the ligand immobilization on the sensor chip are as follows:

		Kinetics should be done with the lowest ligand density, which gives a good response without being disturbed by secondary factors such as mass transfer or steric hindrance.

		For specificity measurements, almost any ligand density will do as long as it gives a good signal.

		Concentration measurements need the highest ligand density to facilitate mass transfer limitation.

		Affinity ranking can be done with low to moderate density sensor chips.

		Low molecular mass binding should be done with high-density sensor chips to bind as much as possible of the analyte to gain proper signal.

	Analytical characterization and comparability of biotherapeutics 

	Biotherapeutics are complex products and, therefore, require characterization using numerous orthogonal analytical tools. Biosimilars undergo limited clinical examination prior to approval under the premise that comparability to the corresponding innovator product using an exhaustive analytical characterization exercise has been demonstrated (14). The comparability exercise involves characterization and analysis by a platform consisting of numerous, orthogonal analytical tools (

). Failing to demonstrate similarity can come at a significant cost and can trigger a more extensive (and expensive) clinical examination prior to receiving regulatory approval. Guidance documents from regulatory health authorities, such as FDA and the European Medicines Agency (EMA), emphasize the importance of extensive analytical characterization in showing the similarity between a biosimilar and its reference product based on a comprehensive assessment of protein structure and function (15, 16). As such, developers of biosimilars must put together a toolbox of state-of-the-art technologies. Companies such as Sandoz and Celltrion, which have both received FDA approvals for biosimilars, have had success in using real-time, label-free biophysical analytics as a platform approach for product and process support, evaluating target and receptor binding, and for measuring immunogenicity and specificity. Another significant challenge with biosimilars as well as other biological products is that of standardizing the Fc receptor binding. Fc receptors can be activating, or inhibitory, or without any effect in antibody-dependent cellular cytotoxicity (ADCC) and complement-dependent cytotoxicity (CDC). Characterizing these highly complex products and their interactions, however, is extremely challenging (13).

	Monoclonal antibodies (mAbs) are at present the fastest growing subclass of biopharmaceuticals and have been successfully demonstrated to treat a variety of diseases, mainly in oncology and auto-immune and infectious disease segment. Antibodies recognize their antigen through the variable regions of the antigen-binding portion (Fab). As a result, they may interfere with one or several functions of this antigen, leading to the therapeutic effect. On the other hand, through the constant regions (Fc), they may interact with Fc-binding molecules and recruit patient immune effector function to destroy the marked target. The ADCC is triggered by an interaction between the Fc region of an antibody bound to, for example, a tumor cell and the Fcγ receptors on immune effector cells, leading to the elimination of the tumor cell by phagocytosis or lysis, depending on the type of mediating effector cell. CDC is initiated by complement component C1q binding to the Fc region of the antibody, triggering activation of the complement that leads to cell death by phagocytosis, lysis, or disruption of the cell membrane (16). SPR is a well-established technique for detection and monitoring of biomolecular interactions in real time. It has been formatted as a parallel line assay and researchers have demonstrated high levels of accuracy, precision, and linearity with the assay, making it useful for establishing comparability, estimating potency and examining stability (17–20). SPR has also shown greater precision and reproducibility than the traditional cell-based assays such as ADCC and CDC.

	SPR applications are not only limited to ligand–receptor interaction kinetics dynamic analyses, but they are also used for drug discovery and development. There are several different formats of SPR biosensors, including the array format, multi-channel unit format, and SPR imaging format, which allow simultaneous and continuous detection to analyze the performance of hundreds to thousands of affinity binding events on a chip surface. In SPR imaging, the incidence angle remains fixed and the binding of biomolecules on a gold surface is measured as the change in reflectivity in relation to the incident ray intensity, unlike SPR sensors that depend on the measurement of the absorption dip in the SPR angle or SPR wavelength. Despite the excellent benefits inherent in SPR technology, conventional propagating SPR biosensors have a serious limitation due to their inability to support multiplex analysis, as less than four analyses with conventional SPR instruments make such parallel operations feasible. In contrast, SPR imaging technology uses a multi-analyte biosensor that permits a high-throughput approach and achieves a similar degree of sensitivity that is achieved by conventional SPR biosensors. Therefore, SPR imaging systems without any labeling requirements are more suitable for high-throughput screening (HTS), particularly in drug discovery, than any other optics-based detection techniques (21, 22).

	Antibody applications of SPR broadly fall into two categories, such as screening and characterization. Antibody screening often involves capture of an antibody followed by a single injection of antigen. The kinetics of binding can be estimated directly from a screening experiment and is of major interest for antibodies intended for diagnostic or therapeutic purposes. Antibodies with slow off rates are often selected as candidates to provide prolonged drug residence time and also potentially higher efficacy (23). A complicating factor in the selection procedures is that antigen binding may be heterogeneous, and such antibodies are often deselected in favor of monophasic binders that may be more specific. The selection is, therefore, focused not only on stable binding but also on monophasic binders. In antibody characterization, SPR is typically used for epitope binding, specificity, concentration, and kinetic/affinity analysis. When using SPR for establishing analytical comparability, the biosimilar product is compared with the reference product.

	Antibodies, cytokines, and hormones typically interact with their receptors. While cytokines and hormones retain their natural sequence and folding, antibody therapeutics are engineered to interact with target molecules (including antigens, Fc receptors, and complement factors) based on their intended mechanism of action. SPR data experiments were performed using SPR based sensor system (GE Healthcare) with analysis temperature set to 25 °C. Series S Sensor Chip CM5, Series S Sensor Chip NTA, Series S Sensor Chip Protein A HBS-EP buffer (10 mM Hepes, 0.15 M NaCl, 3 mM ethylene diamine tetra acetic acid [EDTA], and 0.05% [v/v] surfactant P20, pH 7.4), Amine Coupling Kit, Human Fab Capture Kit, NTA Reagent Kit, Protein L, and Biotin Capture Kit were all obtained from GE Healthcare.

	Case study I: Binding kinetics of other rituximab biosimilars and Ristova to human FcRγIIIa (CD16a)

	The binding kinetics of Ristova, a rituximab biosimilar, and other rituximab biosimilars to human FcRγIIIa (CD16a) receptor (R&D systems) were determined by SPR-based sensor system (GE Healthcare). Kinetic analysis of FcRγIIIa (CD16a) binding was performed by injecting different known concentrations of aggregates (from 0.25 to 4 µM) onto immobilized carboxymethyl dextran-coated CM5 sensor chips, which were used with His-coupling chemistry (13). All measurements were performed at 25 

C with a flow rate of 30 µL/min using HBS-EP buffer with association time 60 s followed by 60 s dissociation phase. KD were calculated from the sensorgrams using the 1:1 fit model using SPR-based sensor evaluation software (GE Healthcare) (

	Figure 2. Binding kinetics of different rituximab biosimilars and Ristova (rituximab) to human FcγRIIIa (CD16a). The color key (-4 µM, -2 µM, -1 µM, -0.5 µM, -0.25 µM) is same for all the sensorgram indicates known concentrations of analytes.

	SPR was used to provide a comparison of the binding kinetics of rituximab biosimilars to human FcγRIIIa with respect to Ristova. As given in 

, the KD value of Rituximab biosimilars to FcγRIIIa was calculated to be within the same order of magnitude as Ristova. KD values were compared, wherein Ristova exhibited higher binding affinity than the other rituximab biosimilars. Biosimilar two exhibited comparable affinity for binding to FcγRIIIa with respect to Ristova, but others yielded lower values.

	Table III: The binding kinetics of different rituximab biosimilars and Ristova (rituximab) to FcγRIIIa were estimated by using SPR-based sensor system.

	Case study: II Binding kinetics of GCS-F biosimilars and Neupogen (filgrastim) to GCS-F-R (CD114)

	The binding kinetic interactions of five different biosimilars samples of granulocyte-colony stimulating factor (GCSF) to human CD114-R receptor (R&D systems) was determined by Biacore X100 plus biosensor (GE Healthcare). For GCSF-R affinity analysis, CD114-R receptor was immobilized on a CM5 sensor chip surface according to the manufacturer’s recommendation with a level of ∼200 response units (RU) reached. Samples were injected in a series of concentrations ranging from 2–32 nM (

) with association time 120s followed by 120s dissociation phase. All measurements were performed at 25 

C with a flow rate of 30 µL/min using HBS-EP buffer according to the manufacturer’s protocol. Kinetic constants were calculated from the sensorgrams using the 1:1 fit model using SPR-based sensor evaluation software (GE Healthcare) (

	Figure 3. Comparative analysis of granulocyte-colony stimulating factor biosimilars and Neupogen (filgrastim) for affinity to CD114-R by surface plasmon resonance. The color key (-4 µM, -2 µM, -1 µM, -0.5 µM, -0.25 µM) is same for all the sensorgram indicates known concentrations of analytes.

 represents the KD value of GCSF biosimilars to CD114-R was calculated to be within the same order of magnitude as Neupogen. Biosimilar four exhibited little higher affinities than Neupogen for binding to CD114R, but others yielded lower values. Because nearly all mAb biosimilars are produced in mammalian cells, they exhibit different post translational modifications (PTMs), which results in heterogeneity. This could be attributed to differences in expression system, culture conditions, purification processes, formulations, or storage conditions of the product and need to be suitably addressed as per the “similar but not identical” paradigm.

	The equilibrium constant determines the ratio of the antibody association rate (kon) of the antibody, how quickly it binds to its antigen, and antibody dissociation rate (koff), how quickly it dissociates from its antigen. Chi

 and residual values were used to evaluate the quality of fit between the experimental data and individual binding models. Plots of residuals indicate the difference between the experimental and reference data for each point in the fit. The Chi

 value represents the sum of squared differences between the experimental data and reference data at each point. Lower Chi

 values indicate a better fit; however, it is difficult to recommend absolute values for acceptance limits for Chi

. U-value indicates the uniqueness value for the kinetic rate constants (19). Lower values indicate greater confidence in the results. Prior to analysis, sensorgrams are double referenced by first subtracting data from a reference flow cell and then subtracting a blank cycle where the buffer is injected instead of protein sample. Sensorgram comparison requires that standards, controls, and samples have been generated in the same way using identical association and dissociation times (20, 21). The assay format must be the same, and it is not possible to mix multi-cycle kinetic data and single-cycle kinetic data in the same session.

	SPR provides a meaningful insight into the characterization of biotherapeutics and comparability of biosimilars present in the market.

	This review has demonstrated that SPR is a rapidly developing technique typically used for characterization of protein interactions and in screening for selection of antibodies or small molecules with preferred binding properties. In characterization, complete binding curves are normally fitted to defined interaction models to provide affinity and rate constants, whereas report points indicative of binding and stability of binding are often used for analysis of screening data. It is a high-throughput technique that can potentially serve as a useful tool to evaluate the higher-order structural integrity of proteins for characterization as well as comparability purposes. As an outcome of this review, SPR provides a meaningful insight into the characterization of biotherapeutics and comparability of biosimilars present in the market.

	This work was funded by the Center of Excellence for Biopharmaceutical Technology grant under Department of Biotechnology, Government of India.

	1. R. Karlsson, “Surface Plasmon Resonance in Binding Site, Kinetic, and Concentration Analyses,”  in 

, D. Wild, Ed. (Elsevier, San Diego, CA, Fourth ed., 2013), pp. 209-222.
	2. A.S. Rathore, 

 32 (10) 992–1000 (2014).
	4. D. L. M. Rupert, et al., 

 88, 9980–9988 (2016).
	5. O. K. Kari, et al., 

 7, 228–240 (2017).
	6. B.D. Brooks, et al., 

 19, 1040–1044 (2014).
	7. P. Safsten, 

 323, 325–340 (2000).
	9. R. Karlsson and A. Falt, 

 200, 121–133 (1997).
	10. R. L. Rich, and D. G. Myszka,

 14, 273–294 (2001).
	11. R. Karlsson, 

 8 (4) 347–358 (2016).
	12. A. S. Rathore, 

 27, 698–705 (2009).
	13. N. Nupur, et al., 

 10 (1) 143–158 (2018).
	14. M. Federici et al., 

Guidance for Industry: Scientific Considerations in Demonstrating Biosimilarity to a Reference Product 

(Rockville, MD, April 2015).
	16. M. Schraeml and M. Biehl, 

. 901, 171 –181 (2012).
	17. P. Safsten,

 524, 67–76 (2009).
	18. Å. Frostell, et al., 

. 477, 1–9 (2015).
	19. Accurate comparability assessment of a biosimilar interferon in process development, GE Healthcare, Application Note 29-1154-78 AC  (2014).
	20. A. Moberg, et al., 

78,  224–232 (2013).
	21. D. G. Myszka, 

323, 325–340 (2000).
	22. I. Navratilova, et al., 

 355, 132–139 (2006).
	23. T. Kitazawa et al., 

The ability of SPR-based analysis to estimate association and dissociation rate constants is a major advantage over the traditional steady-state analysis of biomolecules.

Characterizing Biotherapeutics and Determining Biosimilar Comparability

Ensuring the stability of biotherapeutic products, which has an impact on their safety and efficacy, continues to be a major challenge for the biopharmaceutical industry. This is particularly true for monoclonal antibodies (mAbs), which account for most commercial biopharmaceuticals today. Both physical effects such as aggregation and chemical modifications (e.g., fragmentation) are the two leading paths for mAb degradation. Kinetic modeling can be an effective tool for establishing product stability, offering a deeper understanding of these processes and allowing developers to mitigate the risk of product degradation during the early stages of process and product development.

	Monoclonal antibodies are currently the leading class of protein therapeutic molecules due to their ability to treat a wide range of lethal and life-threatening diseases, including cancer, Ebola (1, 2), multiple sclerosis, rheumatoid arthritis, psoriasis, and asthma. FDA has approved a number of these antibody drug molecules, which are in various stages of clinical and pre-clinical development. But since the major constituent of the final drug product is protein, there is a significant risk associated with the stability of these drug molecules (3–5). 

	Instability associated with various critical quality attributes (e.g., aggregation, fragmentation, charge variants, and glycosylation) continues to be a major concern for the biopharmaceutical industry. It can, either directly or indirectly, affect mAbs’ biological activity as well as toxicity (4). 

	The instabilities associated with mAbs can be broadly classified as either physical or chemical (3, 4). Aggregation is one of the major physical instabilities that is believed to impact the immunogenicity of a therapeutic (5).

	Different types of chemical instabilities include deamidation, oxidation, fragmentation, and hydrolysis. Although the effects of these instabilities (both physical as well as chemical) may differ among various classes of mAb molecules, biopharmaceutical manufacturers desire to minimize them (2). Aggregation and fragmentation are the two most significant degradation pathways for mAbs (6–7). They result in instabilities that can occur at various stages of product development, including upstream processing (during protein expression in cell culture), downstream processing (purification), product formulation, storage, and transport (6, 7).

	Factors that affect the stability of an mAb therapeutic can relate either to its structure (primary, secondary, or tertiary), or to its environment. In the primary structure, a protein is more susceptible to instability if there is change in its surface charge distribution or hydrophobicity. Alterations in the secondary structure, especially an increase in beta content of the protein molecule, can make it significantly more vulnerable to aggregation. Environmental factors that can have an impact on the protein include pH, temperature, salt concentration, buffer type, protein concentration, ionic strength, mixing, shear, metal ions, pressure, freeze thawing, freeze drying, and reconstitution (4, 8). 

	A protein may change in response to these factors, altering its physical structure either by physical association (aggregation) or chemical degradation (fragmentation). This article in the Elements of Biopharmaceutical Production discusses the kinetics of these degradation pathways as well as the insights they reveal that can help developers ensure the stability of biotherapeutic products. 

	Case study I: aggregation of mAb products

	Modeling the kinetics and thermodynamics of protein aggregation can prove valuable in understanding its mechanism (9). Mathematical modeling of protein aggregation correlated along with kinetic data can help developers gain both qualitative and quantitative insights into the mechanism behind aggregation. Armed with this knowledge, developers can better predict and control aggregation by optimizing timeframes and environmental conditions (10). 

	A number of protein aggregation models already exist (9), including the Lumry-Eyring model, which is one of the most commonly used models to predict aggregation kinetics. According to this model, aggregation occurs in two steps: conformational reversible unfolding of the protein molecule followed by irreversible assembly into aggregates, which are either physically or chemically linked to each other. 

	Many researchers have modified the Lumry Eyring model to account for nucleation as well as higher order aggregation (10). Examples are the Extended Lumry Eyring (ELE) Model and the Lumry Eyring Nucleated Polymerization (LENP) Model (11). In addition, alternative models (9) have proposed alternate aggregation models for proteins such as prions (12), amyloid β (13), and insulin (14).

	Lumry Eyring Nucleated Polymerization (LENP)

	The Lumry Eyring model introduces the concept of nucleation to aggregation. The kinetic reaction scheme presented in this model depends on:

		Size distribution amongst different aggregate species (6). 

	The parameters which are considered in this model are nucleus stoichiometry (x), monomers added in each growth step (δ), and the inverse rate coefficients for nucleation and growth (10). These rate coefficients denote the corresponding time scales (τ

) and the details of the model can be found in the literature (10). 

	Another model used for kinetic analysis of aggregation is the extended Lumry-Eyring (ELE) model. With this approach, unfolding is considered as a single reversible rate-limiting reaction with both folded native and reactive unfolded monomer species in thermodynamic equilibrium with each other (11, 15, 16). This model takes into consideration both reversible and irreversible conformational changes that occur during aggregation (17). It also considers the solution’s conformational and kinetic colloidal stability. Concentration-based experimental data are fitted into model equations to calculate the apparent rate constants and predict the monomer degradation rate (17).

	Aggregate sample prep and characterization

	To monitor and understand the aggregation mechanism and its behavior in proteins, a model protein (mAb) was chosen with an isoelectric point (pI) of 8.5. Buffer exchange was used to monitor the aggregation kinetics of this mAb under the conditions that are prevalent in the biopharmaceutical industry during downstream purification. 

	Detailed experimental procedures and conditions have been described elsewhere (6). In brief, buffers examined included those commonly used for Protein A chromatography, cation exchange chromatography, and anion exchange chromatography. 

	Samples were kept at three different temperatures (4 °C, 15 °C, and 30 °C) at a concentration of 10 mg/mL. Sampling was performed over 120 hours at intermittent time points. The samples were analyzed for aggregation content using size-exclusion high performance liquid chromatography (SE–HPLC) and types of oligomers using dynamic light scattering (DLS) (7) (

	All the experiments were done in triplicate to ensure reproducibility. Aggregation data obtained by SE–HPLC were analyzed using MATLAB R2011a and fitted into ELE and LENP models. Sets of ordinary differential equations (ODEs) were then solved using Gauss-Newtonian algorithm to estimate the model parameters (18).

	Aggregation was found to be at its maximum at low pH and accelerated upon increasing temperature and salt concentration (9). Aggregation at low pH has been primarily associated with changes in the fragment crystallizable (

domain of the antibody (19). These changes include partial unfolding of the 

domain of the mAb, which leads to the exposure of the hydrophobic residues. These hydrophobic residues, which were previously buried inside the molecule, became attracted to each other. This resulted in the enhancement of aggregation at low pH. 

illustrates that aggregation is minimal in cation and anion exchange (CEX and AEX) buffer conditions (high pH) as compared to protein A purification conditions (low pH). Further, higher temperature resulted in partial or complete unfolding of mAb structures and resulted in aggregation due to instability of the protein molecule at higher temperature (

	Presence of salt also affects protein stability and the extent depends on the type of salt, its concentration, protein charge, and protein and salt interaction (

) (6). An increase in the rate of aggregation is seen with increasing salt concentration. Apart from these factors, the type of buffer species also seems to have a profound effect on the stability of protein products. 

	Higher stability is seen in glycine buffer, as compared with citrate and acetate buffers (

) (6), governed by the interaction of the 

domain of the mAb molecule with different buffer species (20). Results obtained from circular dichroism (CD) spectroscopy and DLS measurements support these observations as well (6). To summarize the key findings, it can be said that pH is the most important factor in protein aggregation, followed by temperature, salt concentration, and buffer species.

	Data obtained from SE–HPLC analysis were fitted using both ELE and LENP models, and the accuracy of the models was judged by doing a comparative evaluation of the experimental values with the model predicted values. Curve fitting for computing the coefficient of regression (R

value) was performed using both models. A comparison of the models found that that the LENP model did a better job of predicting the aggregation kinetics than ELE (6). 

	Results were then used to compute the time scales of nucleation (τ

) in order to predict aggregation kinetics (

). The modeling parameters of nucleation and growth time scales were compared across different incubation conditions. It was noted that, as the temperature is increased from 4 ºC to 30 ºC (in the case of citrate buffer at pH 3.0 with 100 mM NaCl), the nucleation and growth time scales decrease, implying faster aggregation (6) (

	There is also a linear relationship between nucleation time scale and monomer half-life as temperature is increased from 4 ºC to 30 ºC. The steeper monomer loss at high temperature is attributed to the increase in the collision frequency due to an increase in protein diffusion. 

	A similar effect was observed upon increasing the salt concentration from 0 mM NaCl to 100 mM NaCl (

). This is likely due to salt ionization, which weakens protein structure and decreases stability because of hydrophobic interactions and weakening of electrostatic repulsions as the salt concentration increases. 

	Overall, for the system under consideration, the citrate buffer at pH 3.0 showed the steepest monomer loss as compared to acetate and glycine buffers (6) (

). There is a linear correlation between nucleation time scale and monomer half-life across the different conditions, with slope between 0.4–0.5, indicating that nucleation dominates aggregation in mAbs and hence that the LENP model offers a better fit than the ELE model (6) (

	Fragmentation involves the breaking down of the mAb into smaller units, fueled by thermal or chemical energy (21). Fragmentation results in the breakage of bonds between the amino acids, transforming the primary structure of the molecule and, ultimately, other higher order entities. Cleavage of the peptide bond (hydrolysis) at the mAb’s hinge region leads to the formation of 

	Responses from different classes of the mAb molecules vary with respect to the fragmentation phenomena, depending on each one’s inherited primary structure. Various contaminants (e.g., metal ions and proteases) in the formulation buffer systems significantly affect the extent of fragmentation (22), and degradation can occur at various stages of mAb production and processing (23). Size-based monitoring tools such as SEC, DLS, mass spectrometry (MS), and sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS–PAGE) are commonly used to detect fragmentation and identify the various species that are formed (24). 

	Fragmentation sample prep and characterization

	The mAb sample was diafiltered in the formulation buffer (15 mM phosphate, pH 6.5 with 150 mM NaCl, and 0.02% Polysorbate 80) at a final concentration of 5 mg/mL. This sample was then incubated at 50 °C, and the extent of fragmentation was monitored by SE–HPLC at regular time intervals (

	Various other analytical techniques (e.g., DLS and SDS–PAGE [7]) were used to confirm the sizes of the fragments that were formed (

). Reverse-phase high-performance liquid chromatography with mass spectrometry (RPHPLC–MS) was used to analyze the intact mAb samples as well as the fragmented mAb sample and the purified and enriched fragment species to locate the site of fragmentation in the mAb molecule (

	Results of the SEC, DLS, SDS–PAGE, and MS analysis found that two different fragment species were formed upon fragmentation of intact monomer (148 KDa) mAb molecule-Fragment 1 (96.9 KDa) and Fragment 2 (47.3 KDa) (

Ensuring the stability of biotherapeutics can be achieved through kinetic models that study aggregation and fragmentation.