Characterization of Higher Order Structural Changes of a Thermally Stressed Monoclonal Antibody via Mass Spectrometry Footprinting and Other Biophysical Approaches

Abstract

Characterizing changes in higher order structure (HOS) of monoclonal antibodies (mAbs) upon stressed conditions is critical to gain a better understanding of the product and process. One single biophysical approach may not be best suited to assess HOS comprehensively; thus, the synergy from multiple complementary approaches improves characterization accuracy and resolution. In this study, we employed two mass spectrometry-based footprinting techniques, namely fast photochemical oxidation of proteins (FPOP-MS) and hydrogen deuterium exchange (HDX)-MS, supported by dynamic light scattering (DLS), differential scanning calorimetry (DSC), circular dichroism (CD), and nuclear magnetic resonance (NMR) to study changes to the HOS of a mAb upon thermal stress. The biophysical techniques report a nuanced characterization of the HOS in which CD detects no changes to the secondary or tertiary structure, yet DLS measurements show an increase in the hydrodynamic radius. DSC indicates that the stability decreases, and chemical or conformational changes accumulate with incubation time according to NMR. Furthermore, whereas HDX-MS does not indicate HOS changes, FPOP-MS footprinting reveals conformational changes at residue resolution for some amino acids. The local phenomena observed with FPOP-MS indicate that several residues show various patterns of degradation during thermal stress: no change, an increase in solvent exposure, and a biphasic response in solvent exposure. All the evidence shows that FPOP-MS efficiently resolves subtle structural changes and novel degradation pathways upon thermal stress treatment at residue-level resolution.

Graphical Abstract

Introduction

Monoclonal antibodies (mAbs) have become critical therapeutics for a variety of diseases owing to their unique binding specificity and efficacy.¹ They are, however, more expensive than their small-molecule counterparts and present unique challenges for storage and administration given their high molecular weight, complexity, and structural stability.² The structural and conformational changes can affect the quality, safety, and efficacy of drug products, urging the implementation of effective structural measurement strategies during therapeutic development and lifecycle management. The emphasis on structural analysis methods is further underscored by the high influx and introduction of biosimilars as many patents expire.³ The guidelines defined by the United States Food and Drug Administration (FDA) and other regulatory agencies outline the important quality attributes that pharmaceutical companies must report for approval of antibody therapeutics and determine the product shelf life for administering physicians.⁴ Higher order structure (HOS) is an important quality attribute for protein therapeutics. HOS may be perturbed in forced degradation studies that use extreme temperatures, pH, light exposure, and chemical oxidants for heightened characterization assessment and greater process and product understanding.⁵ Product shelf life can be determined from real-time stability studies with the product held at preferred storage conditions.

Biophysical approaches for HOS measurements traditionally include far- and near-ultraviolet circular dichroism (far-/near-UV CD), dynamic light scattering (DLS), and differential scanning calorimetry (DSC), which are effective for characterizing the folded protein structure.⁶ Despite their low structural resolution, these traditional techniques are routinely used because most laboratories have fully integrated them. Their associated data acquisition and processing are straightforward, and modern instrumentation is amenable to a high-throughput format.⁷ On the contrary, X-ray crystallography (XRC) can provide high resolution characterization of HOS, but some analytes are difficult to crystallize, structural artifacts from solid phase crystallization can occur, and the process is tedious and time-consuming especially during process and product development when structural information is frequently assessed.⁸ NMR offers liquid-phase alternatives for atomic or residue-level HOS characterization in clinical formulation, but NMR has protein size limitations and still require experts for data acquisition, processing, and analysis.⁹ A lack of isotopic labeling schemes for mAbs and the requirement for high concentration samples further disadvantage NMR spectroscopy as a precise method for HOS assessment of mAbs. NMR has become an effective tool for the comparative analysis of mAbs and mAb HOS characterization owing to recent advances in NMR methods and software.¹⁰ Indeed, methods to probe HOS with relatively high resolution and high efficiency are desired, and the use of NMR is becoming more prevalent in the pharmaceutical and academic sectors.

Mass spectrometry (MS)-based approaches can also provide insights into HOS with higher structural resolution compared to traditional biophysical tools such as CD and DSC that assess HOS at the intact protein level. The approaches include protein footprinting, native MS, and native MS couple with ion mobility ^{11, 12}. A commonly used MS-based footprinting is hydrogen deuterium exchange (HDX)-MS where protein solvent accessibility is probed by allowing backbone hydrogens to exchange with deuterons supplied by a deuterated incubation buffer (D₂O).¹³ Most HDX-MS studies report the degree of exchange at the peptide level.¹⁴ To improve the resolution to the residue level, many small, overlapping peptides may be generated during protein digestion by employing multiple proteases. MS/MS fragmentation such as electron transfer dissociation (ETD), electron capture dissociation (ECD), or ultraviolet photon dissociation (UVPD) may also be adopted. These techniques induce fragmentation of the peptide without major scrambling of hydrogen and deuterium, permitting the location of a deuterium to be accurately determined.¹⁵ A recent study by the FDA¹⁶ monitored using HDX-MS changes in solvent accessibilities relative to an unstressed control of several mAbs incubated at 25 ± 2°C, 60 ± 5% RH for 6 months. Although this study provided a broad survey across several antibodies, there are still challenges to improve the spatial resolution (i.e., peptide length or number of unique peptides) and to employ robust statistical analysis for accurate and reliable measurements of subtle structural changes using HDX-MS methods. HDX-MS is very time consuming in an industry setting relative to the fast timelines at process and product development.

An emerging complementary technique to HDX-MS is fast photochemical oxidation of protein (FPOP)-MS and related covalent protein footprinting approaches (e.g., glycine ethyl ester labeling of Asp and Glu). In an FPOP experiment, the solvent-accessible protein side chains are labeled by a hydroxyl radical, generated by irradiating H₂O₂ with a 248 nm laser light beam.¹⁷ This labeling scheme mainly leaves stable mass tags in multiples of +16 Da, corresponding to the mass of an O atom, on the protein¹⁸, allowing easy detection by MS. Hydroxyl radicals have broad reactivity towards all amino acids except glycine. In practice, cysteine, methionine, phenylalanine, tryptophan, tyrosine and histidine will undergo most of the hydroxyl radical modifications and, therefore, the latter four amino acids are usually used for quantification.¹³ Given the irreversibility of the labeling, FPOP-MS can sometimes accurately report up to residue-level resolution by performing bottom-up analysis with traditional collision induced dissociation (CID) MS/MS of the footprinted samples. More importantly, the modified peptide typically can be separated from the unmodified peptide using the analytical liquid chromatography timescale, which enables more accurate and precise chromographic quantification owing to reliable UV signals. Recent review articles highlight the applications of FPOP-MS ¹⁹. A growing application of FPOP-MS for mAb HOS characterization is epitope mapping where FPOP-MS has been employed for the epitope characterization of anti-VEGF (vascular endothelial growth factor)²⁰, anti-thrombin²¹, anti-interleukin-23²², and the anti-NKG2A (NK group 2 member A)²³ antibody, among others. FPOP-MS can also capture mAb conformational dynamics and show an aggregation interface.²⁴ Despite its potential, the capability of FPOP-MS to study more subtle HOS changes after stressed treatment has not yet been investigated.

In this study, we address the challenges in acquiring high-resolution HOS data in a biopharmaceutical laboratory setting and explore the sensitivity of different HOS characterization methods by investigating thermally stressed bevacizumab as a case study. We incubated samples of bevacizumab at 40 °C and 75% relative humidity (RH) and froze aliquots after one (1), two (2), four (4), and eight (8) weeks for further analysis. We measured the progression of HOS changes as a function of time-of-stress relative to an unstressed, control sample (referred to as “0 week”) and determined the changes by using several biophysical approaches including NMR, CD, DSC, DLS, and the MS-based methods of HDX-MS and FPOP-MS. We found that the complementary nature of the methods out-performs any single technique when HOS characterization is required, but FPOP-MS provides the most well-resolved, residue-level structural information for HOS characterization. We demonstrate that FPOP-MS is efficient and can measure small, local changes in a mAb at relatively high speed.

Experimental Section

Material and Sample Preparation

Chemical regents were purchased from MilliporeSigma (St. Louis, MO) unless specified. Stress treatments were performed using bevacizumab (470332) in commercial formulation (51 mM sodium dihydrogen phosphate (NaH₂PO₄), 6% trehalose, and 0.04 PS 20 at pH 6.2) at a concentration of 25 mg/mL. Upon receipt, 14 mL of the protein sample was sterile filtered through a 0.2 um syringe filter. The filtered material was then aliquoted into glass vials, stoppered, crimped shut, and held at 40 °C for up to 8 weeks. Materials were further aliquoted for each experiment described below at each timepoint (0, 1, 2, 4, and 8 weeks), and frozen immediately upon aliquoting. Experimental details of far- and near-UV CD, DSC, and DLS can be found in Supporting Information.

Nuclear magnetic resonance (NMR) spectroscopy

All NMR experiments were performed at 25 °C on an 800 MHz Bruker Neo spectrometer on the provided samples (see Materials and Sample Preparation) with 10% D₂O added as a deuterium lock. Processing, including baseline correction, phase correction, and preprocessing for the chemometrics analysis (i.e., broadening factor, binning, normalization, filtering, and scaling) of the NMR spectra were performed using MNova and MBioHOS plugin (MestreLab Research, Santiago de Compostela, Spain). The ¹H one-dimensional (1D) NMR experiments were collected with six replicates per sample using the pulsed field gradient echo (PGSTE) pulse program as described in the literature²⁵ with 1024 scans, spectral width of 15.6 ppm, acquisition time of 1.31 s, diffusion delay of 60 ms, diffusion gradient strength of 98% for 2 ms, and relaxation delay of 2.5 s.

The thermally stressed bevacizumab was assessed relative to the unstressed (0 week) material using protein fingerprinting by line shape enhancement (PROFILE) and principal component analysis (PCA) as chemometric methods. PROFILE analysis was performed based on previous studies by using a broadening factor of 100 Hz.²⁶ Intra-sample comparisons are defined as comparisons between replicate spectra of a sample whereas inter-sample comparisons are of spectra between different samples. The standard deviation (SD) of the analysis is determined using the within-group variance of the intra-sample comparisons through a one-way analysis of variance (ANOVA) with the assumption that these comparisons have equal instrumental variability. The statistical limits are set at 2 SD and 3 SD from the mean of the intra-sample comparisons of the control to represent different levels of spectral differences, where points beyond 3 SD of the mean have the greatest differences, and within 2 SD from the mean have the least differences. Additional processing for the PCA is included in the Supporting Information.

Hydrogen Deuterium Exchange Mass Spectrometry

The HDX experiment was carried out on a LEAP Technologies HDX PAL robot. For the mapping experiment, elevated temperature and a longer than usual quench incubation time were used to improve peptide identification. In the mapping experiment, 7 μL of 33 μM protein solution, diluted by 63 μL of PBS buffer, was added to 140 μL of quench solution containing 6 M Urea and 1 M TCEP (pH 2.5). After 3 h of incubation at 20 °C, 10 uL of 0.33 μg/μL protease type XIII from Aspergillus saitoi (FPXIII) in PBS (pH 2.5) was added and incubated for 2 min. In the HDX experiment, labeling was carried out by adding 7 μL of 24 μM protein solution into 63 μL (90%) of PBS-D₂O buffer for 20, 200, 2000 and 14400 s. A nondeuterated control sample was prepared by replacing PBS D₂O buffer with PBS buffer. The resulting labeled and control solutions were mixed with 140 μL of quench solution containing 6 M Urea and 1 M TCEP (pH 2.5). After 4 min of incubation at 4 °C, 10 uL of 0.33 μg/μL Protease type XIII from Aspergillus saitoi (FPXIII) in PBS (pH 2.5) was added and incubated for 2 min. For both mapping and the subsequent HDX experiments, 215 uL of solution (protein + quench + FPXIII) was submitted to sequential digestion via a custom-packed pepsin column and a FPXIII columns in series at a flow rate of 200 μL/min in 0.1 % formic acid in water. The resulting peptides were trapped by ORBAX Eclipse XDB C8 column (2.1 × 15 mm, Agilent Technologies, Santa Clara, CA). Digestion and desalting were performed for 3 and 1 min, respectively. Desalted peptides were separated on a Hypersil Gold C18 column (2.1 × 50 mm, Waters, Milford, MA) by using a gradient from 10% B at 0.3 min to 50% B at 11 min to 100% B at 11.5 min with a flow rate of 100 μL/min. The data were collected on a Bruker Maxis qTOF mass spectrometer (Bruker, Billerica, MA). Measurements were performed in triplicate. MS/MS mapping data, taken prior to the HDX data, were searched by using Byonic (Protein Metrics, Cupertino, CA), and the peptide list was generated by Byologic (Protein Metrics Inc., Cupertino, CA). The peptide list and HDX data were submitted to HDExaminer (Sierra Analytics, Modesto, CA) for analysis.

Fast Photochemical Oxidation of Protein

The FPOP experiments were performed similarly to previously published reports.^{17, 18} Briefly, 20 μM bevacizumab and 1 mM histidine was prepared in PBS solution. H₂O₂ was added immediately before the introduction to the flow capillary to give a final concentration of 20 μM. The resulting sample solution was introduced to the flow and past an irradiation window via a syringe pump and a 150-μm i.d. silica capillary. The flow rate was adjusted to accommodate an exclusion volume of 20%. A 248-nm KrF excimer laser was used to irradiate the sample solution at the capillary transparent window. The laser energy was adjusted to 21 mJ/pulse, and the frequency was set to 7.4 Hz. The sample following FPOP footprinting was collected in an Eppendorf tube containing 70 mM methionine and 0.25–0.63 units/μL of catalase. The resulting samples were flash-frozen and stored at −80 °C until analysis.

FPOP Proteolysis, Sample Handling, and LC-MS/MS Instrumentation

The proteolysis protocol was described previously. Briefly, ~80 ug of the sample following footprinting was denatured by 4.2 M GdnHCl in 100 mM Tris-HCl (pH 8.2). The reduction and alkylation of disulfide bonds were carried out by first incubating with 9.5 mM DTT for 30 min at 37 °C and then with 32 mM iodoacetic acid in the dark for 30 min at room temperature. The resulting sample was buffer exchanged into 100 mM Tris-HCl (pH 8.2) using Bio-Spin 6 columns (BioRad, Hercules, CA). The concentration of the protein in the flow-through was measured by a Nanodrop spectrometer (Thermo Fisher, Waltham, MA). Trypsin/LysC (Promega, Madison, WI) digestion (1:10 enzyme:protein (w/w)) was carried out for 60 min at 37 °C and was quenched by adding TFA to give a final concentration of 1%.

Chromatographic separation of peptides was carried out on a Waters Hclass UPLC system (Waters, Milford, MA) equipped with a 150 mm x 2.1 mm, 2.5 mm Xselect XP CSH C18 column (Waters, Milford, MA) with UV detection at 214 nm. The mobile-phase A consisted of water with 0.1% formic acid, - and mobile phase B was acetonitrile with 0.1% formic acid. Mobile phase B was changed from 1% to 30% between 5-150 min.

LC-MS/MS experiments were performed using an Orbitrap Lumos Tribrid mass spectrometer (Thermo Fisher, Waltham, MA). Peptide accurate masses and sequences were obtained by MS and MS/MS, respectively. Both MS and MS/MS data were analyzed by Protein Metrics (Protein Metrics Inc., Cupertino, CA) according to the workflow reported previously.¹⁸

Results

HOS assessment by CD, DSC, and DLS

Far- and near-UV CD were applied for secondary and tertiary structure characterization of the intact bevacizumab, respectively. ²⁷ Far-UV CD (190-250 nm) is sensitive to changes in the secondary structure, and near-UV CD (250-360 nm) is optimal to monitor any changes in tertiary structure. We recorded far-UV CD spectra (Figure S1A) in triplicate for samples at 0 week and, after thermal stress for 1, 2, 4, and 8 weeks. The spectra of all the samples show a negative peak at approximately 217 nm and a positive peak at approximately 202 nm, indicating that the secondary structure primarily comprises β-sheet, which is consistent with an IgG1 antibody such as bevacizumab. The averaged spectra for each time point match closely at all wavelengths from 195 to 250 nm, confirming the secondary structure is maintained in the stressed sample relative to the control.

Near-UV CD spectra for all bevacizumab samples (Figure S1B) showed a few broad positive bands and several distinct, negative and alternating peaks. These spectral features result from multiple Trp, Tyr, and disulfide bonds from the constrained conformations in folded sample molecules. The average of triplicate spectra for the samples incubated at elevated temperatures for 1-8 weeks match closely at all wavelengths from 250 nm to 350 nm to that of the control sample. This indicates that there are no significant changes to the tertiary structure of the stressed materials relative to the control.

For assessment of the conformational stability changes, we also applied DSC to bevacizumab. DSC accurately measures the stability of intact protein by analyzing the changes in the heat capacity of the protein relative to a buffer blank as the temperature is linearly increased.²⁸ The unfolding thermograms for bevacizumab are shown in Figure S2 over a temperature range of 30 to 95 °C. Two apparent thermal transition peaks representative of 50% unfolded bevacizumab can be extracted from the thermograms: T_m1 (~ 74 °C) and T_m2 (~ 84 °C). The average of triplicate thermograms for the control and the stressed samples were plotted and overlaid. The peak for T_m2 appeared relatively stable across all samples. The stressed samples, however, showed detectable changes compared to the control for the T_m1 peak; the T_onset dropped from 64 °C for the control to 60 °C for the eight-week sample. Although the T_m1 and T_m2 for the unfolding transitions remained nearly the same, the peak area, which represents the domain unfolding energies, decreased for the T_m1 peak. The observed decrease in heat capacities of the stressed samples indicate decreases in the thermal stability of HOS for these samples, trending with the thermal stress incubation time.

Lastly, DLS was employed to report changes in the hydrodynamic radius of bevacizumab at two different protein concentrations after thermal stress. The average from triplicate measurements at both protein concentrations shows that the changes are not concentration-dependent (Figure S3). There is a clear increase in the average diameter of bevacizumab from the control to the 8-week stressed material by 3-4 nm, which is significant and represents a large change in relative diameter. Although there are no changes to the secondary or tertiary structure based on the CD and DSC data, there are more subtle structural changes that are causing a decrease in conformational stability and an increase in hydrodynamic radius based on DLS data. Consequently, we chose to apply NMR and MS to investigate further the nuanced structural changes.

Characterization by 1D ¹H Nuclear Magnetic Resonance (NMR) Spectroscopy

We focused the NMR experiments on 1D ¹H methods because it has enhanced sensitivity compared to 2D heteronuclear methods. 1D ¹H NMR is more informative on chemical modifications and local conformational changes, addressing potential gaps in structural characterization differences of the other biophysical methods²⁹. Regardless, the 2D ¹H-¹³C NMR spectra further support the results from near-UV CD in that the overall global structure is maintained (Figure S4 and Figure S5). The representative 1D PGSTE ¹H NMR spectra collected for samples at each incubation time show differences compared to the spectrum of the control (Figure 1A). Although the PGSTE pulse program does well to suppress most excipient peaks of low molecular weight, there are still some unsuppressed peaks from polysorbate 20 (PS 20) in the aliphatic region of the spectra (0 to 2.5 ppm) at 0.76, 1.17, 1.47, and 2.21 ppm, which are excluded from the chemometric analyses (shown as gray regions in the spectra). Alternatively, the amide and aromatic regions of the spectra (6.3 to 8.5 ppm) are free of excipient peaks and appear highly comparable except for the resonances at 6.78, 7.06, and 8.17 ppm, which linearly decrease in intensity with increasing incubation time (Figure S6). These visual differences in the 1D ¹H spectra of bevacizumab result from thermal stress, consistent with chemical modifications or local conformational changes. The extent and the trend of the differences in the spectra, however, are difficult to determine accurately from visual inspection. Therefore, we applied chemometric analyses to quantify the change in structure with incubation time.

Figure 1: 1D ¹H NMR spectra and PROFILE analysis.

A. 1D ¹H NMR spectra of bevacizumab samples where the control is in red, one week stressed sample in yellow, two weeks in green, four weeks in blue, eight weeks in purple. Gray shading indicates regions of the spectra that were excluded from analysis because of unsuppressed excipient peaks. Arrows indicate peaks of interest where the greatest changes in the spectra occur and are addressed in the text. B. PROFILE analysis of the intra-sample comparisons for the control and the inter-sample comparisons between the control and the four stressed samples at different incubation times. The top and bottom dotted lines correspond to 2 SD and 3 SD from the mean of the intra-sample comparisons of the control, respectively.

The PROFILE analysis (Figure 1B) indicates that the similarity score for the inter-sample comparisons (between the control and the stressed material at different incubation times) is inversely proportional to incubation time. In fact, there are fewer differences between the control and the stressed sample incubated for one week than the other stressed samples because the inter-sample comparison shows that the majority of data points (25 points) are within 2 SD of the mean, eight points within 3 SD of the mean, and only three points beyond 3 SD of the mean. The differences compared to the control sample increase for the stressed sample incubated for two weeks because all the points are beyond 3 SD of the mean. Lastly, the stressed samples incubated for four and eight weeks are the most different from the control as shown by significantly lower similarity scores than found in the control intra-sample comparisons. The PROFILE analysis quantifies the changes in the spectra and supports the conclusion that the changes are accumulating with incubation time.

Conclusions from the visual inspection and PROFILE analysis are further supported by PCA (Supporting Information). The clustering of data points in the scores plot and increasing Mahalanobis distances³⁰ of clusters from the control dataset demonstrate that the differences from the 0-week control accumulate with incubation time, which is consistent with the PROFILE analysis (Figure S7). Furthermore, the clearest trend and greatest differences among all the samples are along principal component 1 (PC1). The values in the loading matrix (components or weights that determine the principal components) confirm that PC1 is dominated by the changes in the resonances at 6.78, 7.06, and 8.17 ppm. Although extensive investigation is needed to determine the location on the protein where these differences occur and whether the change represents a specific chemical or conformational change, PCA confirms that the majority of the differences in the spectra are concentrated in the amide region of the spectra, and the results are consistent with the visual inspection and PROFILE analysis. These NMR experiments provide useful insight to more subtle HOS changes of bevacizumab, but greater structural resolution can be achieved by some MS-based methods, as discussed in the following sections.

Characterization by Hydrogen Deuterium Exchange MS

HDX-MS reports nonspecifically on amide backbone dynamics, yielding insights about secondary and tertiary structure. Building on the 1D NMR findings, we next evaluated whether the local changes in HOS of thermally stressed bevacizumab can be elucidated by HDX-MS. Although HDX may respond to all residues of a protein (except proline), incomplete digestion is a common challenge that prevents complete sequence coverage and limits data interpretation.³¹ For example, another analysis of bevacizumab by HDX-MS reported 119 peptides, 86.7% sequence coverage of the heavy chain (HC) and 28 peptides, 59.8% sequence coverage for the light chain (LC).¹⁶ In this study, we fine-tuned our method by focusing on HDX mapping to improve the sequence coverage for the HC to 93% with 173 peptides and to 96% coverage with 84 peptides for the LC (Figure S8; details about mapping optimization can be found in SI). The improved peptide mapping coverage lays the foundation for describing a largely complete picture of the whole protein sequence.

To compare the HOS of stressed and control bevacizumab, we measured the peptide-level deuteration of bevacizumab stored for 1-8 weeks and compared it to that of the control as a reference state. For each sample, we performed HDX in triplicate at 20, 200, 2000 and 14400 s to record a range for the kinetic phenomena. The differential deuteration of each peptide for each stress incubation time compared to the control are shown as a global Woods’ plot display with global significance thresholds shown horizontally at ± 0.71, which corresponds to a 99% confidence interval (Figure S9). Differences in the relative HDX levels (i.e., Δ # of D) beyond the significance threshold (99% confidence interval) are considered significant. At 99% confidence, there are no significant Δ # of D measurements observed for the stressed samples with respect to the control state, indicating that any HOS changes of bevacizumab upon thermal stress treatment are beyond the detection capacity for HDX-MS.

We hypothesize that there are three reasons why no significant differences can be seen by HDX-MS. First, the HDX experiment was carried out with peptide-level resolution, which may impede detection of local, tightly focused conformational changes that are undetectable owing to dilution by other nearby amino acid residues that do not change. Second, HDX-MS probes protein HOS by interrogating the protein backbone environment and may miss changes involving the side chains. Finally, our experimental design prioritized sequence coverage possibly at the expense of dynamic range. Therefore, in an effort to detect change, we continued the investigation by using FPOP-MS. It is complementary to HDX with enhanced sensitivity to changes in side chains, possibly supplying a different view of the stressed antibody.

Characterization by Fast Photochemical Oxidation of Protein MS

We chose FPOP to continue our search for changes in HOS and protein side chain environment of thermally stressed bevacizumab samples for residue-level resolution in structure. The level of modification (% modification) is plotted as a function of incubation time for the thermal stress at 40 °C (Figures 2, 3, and 4). The % modification (black open square) is a composite measurement of the (i) hydroxyl radical labeling, (ii) H₂O₂ induced oxidation, and (iii) oxidation occurring in the thermal stress treatment. Among these three sources of oxidation, the modification level from hydroxyl radical labeling informs on the mAb HOS whereas the other two are independent of structure. Therefore, we introduced a no laser control (NLC) sample (blue open square) where oxidation that is H₂O₂-induced and/or thermal stress-induced are observed. The % modification contributed from only hydroxyl radical labeling (magenta, closed square) is calculated by the difference between the footprinted and the NLC sample.

Figure 2.

Type I FPOP response at residue level: extent of footprinting remains constant as a function of stress time. Delta modification/oxidation (magenta close square) is determined by the difference between the footprinted sample (black open square) and the no laser control (NLC, blue open square). The side chain of LC Y32 is shown as magenta stick in Beva Fab crystal structure (PDB 6BFT). The crystal structure of Beva Fc is not available.

Figure 3.

Type II FPOP response at residue level: extent of footprinting increases as a function of incubation time. The side chains of HC W50 and HC W108/113 are shown as magenta stick in Beva Fab crystal structure (PDB 6BFT).

Figure 4.

Type III FPOP response at the residue level, showing a relatively sudden increase in oxidation extent at four weeks and a decrease post four weeks. The side chains of HC M34 and HC M83 are shown as magenta sticks in Beva Fab crystal structure (PDB 6BFT).

Three types of responses are observed regarding the extent of hydroxyl radical footprinting. The first is a constant or slightly increasing modification extent (magenta close square) as a function of stress time as observed for LC Y32, HC M258, and HC M365 (Figure 2). The constant level of footprinting indicates little or no HOS change in these regions upon stress treatment, despite the constant increasing modification observed in either the FPOP and the NLC experiments (especially for HC M258 and HC M365, which reside in the constant region for the HC). This category serves as a control for other categories because it indicates that the FPOP footprinting is not disrupting the antibody HOS.

The second type is exemplified by HC W50 and W108/W113 (the latter two W residues are indistinguishable by our HPLC and MS/MS). The modification level (blue markers) from hydroxyl radical footprinting of these residues clearly increases with the incubation time (Figure 3). Heavy chain W50, located in the CDR2 region, exhibits a constant increase between the control and the 8-week sample whereas the heavy chain W108/W113, located in the CDR3 region, undergoes increased footprinting after the 4-week incubation timepoint. We interpret the increase in modification from FPOP footprinting to indicate that these regions become exposed upon thermal stress. The DSC data also indicate that the stability of bevacizumab decreases with incubation time at elevated temperatures. The decrease in stability is consistent with the changes of HOS detected by FPOP-MS. Upon stress treatment, these regions become exposed and less structured, and, therefore, disrupt the intramolecular forces maintaining the HOS.

The third type of change is exemplified by the HC M34 and M83, which show nonlinear behavior where the % modification peaks at 4 weeks (Figure 4) and then decreases. Notably, the side chains of these two residues face each other in the available crystal structure and are separated by 16.7 Å. Rather than a single event, this behavior is more indicative of two events occurring simultaneously or in series where the second is triggered around 4 weeks. This behavior is not considered to be an artifact or measurement error because measurements of the previous types of behaviors on the same samples are linearly increasing with incubation time. The increase in % modification indicates an increase in instability or unfolding that maximizes at 4 weeks. After 4 weeks, the instability reaches a threshold followed by a collapse in the exposure of these residues. It is unlikely that this collapse in exposure is a result of a change in secondary or tertiary structure because this effect is ruled out by the CD results. Instead, this behavior likely results from protein aggregation and is supported by the significant increase the hydrodynamic radius at 8 weeks that was demonstrated by DLS. To our knowledge, this is the first reported observation by FPOP of such behavior for an antibody under thermal stress, and it shows the potential of using FPOP-MS to pinpoint residues involved in antibody degradation with high analytical sensitivity.

Discussion

The complementary nature of the biophysical techniques used here to characterize bevacizumab after thermal stress leads to a detailed and nuanced interpretation of the structural changes observed. Although changes to the stability of bevacizumab even after 1 week at elevated temperatures are evidenced by DSC, those changes do not seem to affect the secondary or tertiary structures at the protein level. Support for this conclusion comes from the CD results, which do not report any such changes. Additionally, HDX-MS shows that there are not any statistically significant changes in solvent exposure at the peptide level that might account for changes in stability. Instead, this evident change in stability likely results from subtle, local conformational changes or from chemical modifications that cannot be captured at the protein level but are observed by NMR spectroscopy. The NMR results correlate with the DSC data in showing an accumulation of changes (either local conformational changes or chemical modifications) as a function of incubation time.

We propose that these changes observed by NMR lead to subtle changes in the hydrogen bonding network and in the solvent exposure at the residue level that do not disrupt the secondary structure, tertiary structure, or solvent exposure at the peptide/protein level. This is well-demonstrated by the FPOP-MS results because only select residues localized in the HC CDRs (Table S1) exhibit changes in solvent exposure. This is consistent with the reported thermal stress study of bevacizumab where no secondary structure changes are detected by CD but subtle changes after thermal stress treatment are captured by ion mobility MS (IM-MS)¹². Moreover, not all the residues observed by FPOP-MS show linear responses as a function of incubation time. The biphasic behavior of certain residues captured by FPOP-MS indicate that as the subtle structural changes accumulate, the solvent exposure of samples incubated for 4 weeks triggers other changes in the structure, causing a collapse in the side chain exposure. As these subtle changes in the structure accumulate, the stability decreases and bevacizumab becomes more susceptible to aggregation, which is supported by the DLS results that show increasing average diameter with time. More intensive investigation is needed to determine whether these sites of interest are in fact related to aggregation.

Traditionally, the HOS characterization of therapeutic antibodies employs automated CD, DLS, and DSC. The advantage of these methods is that the stability and secondary, tertiary, and quaternary structure can be quickly and routinely characterized for the intact protein and confirmed. Unfortunately, they lack structural resolution and miss subtle changes in conformation, dynamics, and chemical modifications. In our study, the sensitivity of CD was insufficient to capture subtle local changes to the secondary and tertiary structure of bevacizumab caused by thermal stress. DSC detects changes in bevacizumab stability, and DLS shows that the molecular diameter of bevacizumab increases at elevated temperatures, but both techniques lack the resolution to locate these changes on the protein. We were able to detect chemical or conformational changes with 1D ¹H NMR and to quantify these changes by chemometric analysis, but also without spatial resolution. HDX-MS provides peptide-level resolution measurements on the solvent accessibility of bevacizumab, but it did not show differences when the results were submitted to stringent statistical analysis.

FPOP-MS delivered the highest-resolution measurements on differences in residue exposure and demonstrated three distinct types of behavior as a result of thermal stress. The most interesting behavior is the increase in footprinting occurring over 4 weeks of thermal stress followed by a decrease, indicating that several regions of the antibody first unfold and then become molecular contacts for intermolecular antibody interaction (aggregation). Moreover, FPOP-MS allows the location of residues showing HOS changes to be located within the HC CDR regions (Table S1), suggesting that subtle changes in the HC CDRs accumulate and may lead to the decreased stability detected by DSC. In this study, however, the footprinting coverage is incomplete. The correlation between residue HOS changes and the overall antibody stability is better considered with near complete footprinting coverage. One strategy to improve footprinting coverage we are pursuing is the development of complementary reagents as exemplified by other reactive species such as •CF₃ ³², sulfate radical cations ³³, carbenes³⁴, and carbocations³⁵ or possibly with slower reacting, more specific reagents¹³. Other ongoing research in our lab is to develop multiplex labeling methods where an aliquot of sample is labeled by different radical footprinting reagents simultaneously or sequentially. Compared with generating multiple datasets with different labeling reagents, multiplex labeling gains time efficiency.

In addition to reporting on changes in HOS, the bottom-up workflow of FPOP-MS affords insights into chemical modifications of the antibody primary sequence. We extracted the information on stress-induced oxidation (blue open squares in Figures 2, 3, and 4) and deamidation (Figure S10) from the FPOP no-laser control (NLC) data and found no consistent relationship between oxidation and changes in local HOS, as exemplified by HC M258 (correlated) and HC W50 (not correlated). The residues for which FPOP-MS reports HOS changes localize within the HC CDR1, CDR2 and CDR3 (Table S1), which may contribute to the destabilization detected by DSC. Similarly, residues with various levels of deamidation were observed across the LC and HC chains (Figure S11), further demonstrating that there was no consistent trend between chemical liabilities and HOS changes. With a comprehensive list of chemical modifications and near complete coverage, the relationship between chemical labilities and HOS changes can be evaluated systematically. Although others have investigated chemical modifications³⁶, we propose a comprehensive assessment as a subject of future research.

Conclusion

Our study illustrates the synergy of applying several HOS characterization approaches to detect HOS changes upon stress. So far, FPOP-MS remains the most sensitive as an emerging HOS characterization tool and can provide insight into HOS with spatial resolution, in some cases to the residue level as demonstrated by this study. Given the success of this study, we hope to further explore the capabilities and applications of FPOP-MS for structure characterization of mAbs and other biologics. As the demands for enhanced, high-quality characterization of biologics grow, the scope and ease of application FPOP should be extended, and other sophisticated approaches developed.

We suggest that combinations of technologies can benefit other important applications during discovery, development, and lifecycle management of protein-based biotherapeutic products and vaccines. The rich HOS information from these technologies may demonstrate molecular comparability between different processes or batches, afford insight of stability under intended storage conditions, enable enhanced understanding of degradation pathways upon various forced degradation conditions, and provide scientific-driven information for regulatory submissions.