Low pK_a of Lys promotes glycation at one complementarity-determining region of a bispecific antibody

Abstract

Protein glycation is a common, normally innocuous, post-translational modification in therapeutic monoclonal antibodies. However, when glycation occurs on complementarity-determining regions (CDRs) of a therapeutic monoclonal antibody, its biological activities (e.g., potency) may be impacted. Here, we present a comprehensive approach to understanding the mechanism of protein glycation using a bispecific antibody. Cation exchange chromatography and liquid chromatography-mass spectrometry were used to characterize glycation at a lysine residue within a heavy chain (HC) CDR (HC-CDR3-Lys98) of a bispecific antibody. Thermodynamic analysis revealed that this reaction is reversible and can occur under physiological conditions with an apparent affinity of 8–10 mM for a glucose binding to HC-CDR3-Lys98. Results from kinetic analysis demonstrated that this reaction follows Arrhenius behavior in the temperature range of 5°C–45°C and can be well predicted in vitro and in a non-human primate. In addition, this glycation reaction was found to be driven by an unusually low pK_a on the ε-amino group of HC-CDR3-Lys98. Van't Hoff analysis and homology modeling suggested that this reaction is enthalpically driven, with this lysine residue surrounded by a microenvironment with low polarity. This study provides, to our knowledge, new insights toward a mechanistic understanding of protein glycation and strategies to mitigate the impact of protein glycation during pharmaceutical development.

Significance

The significance of this manuscript is to understand the mechanism of protein glycation—a common post-translational modification—on therapeutic antibodies that impacts drug safety and efficacy—and to develop formulation strategies to mitigate the impact of protein glycation. Using first-principle analysis, biophysical models were developed to monitor and predict the level of protein glycation in bioreactors and in monkey serum. The framework established in this manuscript opens the possibility to leverage mathematical models to understand and predict protein post-translational modification during drug development.

Introduction

Glycation, a non-enzymatic reaction of a reducing sugar with an amine group, is the first step toward a more complex Maillard reaction (1, 2, 3). When the reaction occurs in a protein, it requires the condensation between an aldehyde group on the reducing sugar and a de-protonated amine group on the amino acid to form a Schiff base intermediate (imine), followed by the Amadori rearrangement to establish a more stable Amadori product (ketoamine) (2,4). Protein glycation is a naturally occurring process in vivo. For example, human serum albumin and hemoglobin were found to have varying degrees of glycation, depending on the blood glucose level and the amount of time in circulation (3,5,6). Similarly, proteins or peptides with susceptible amine groups can undergo glycation in vitro (7,8). Studies from the food industry have indicated that glycation from reducing sugars and subsequent reactions can impact the flavor, color, texture, and nutritional value of the affected food (9, 10, 11, 12, 13). Under stress conditions, such as higher temperature and/or an oxidizing environment, the Amadori products can slowly undergo further reactions, generating reactive carbonyl and dicarbonyl compounds that then react with proteins to form more stable and irreversible adducts, known as advanced glycation end products (2,4,14).

Early studies focused on the characterization of protein glycation began with the discovery of glycated hemoglobin, where a preferential glycation site was identified at the amino group of the terminal valine residue in the hemoglobin β chain (5,15). In addition, a subsequent study examining RNase A identified specific glycation sites, where glycation reduces enzymatic activity (16). In both cases, phosphate or carbonate ions were reported to influence the kinetics and specificity of glycation, possibly acting as a catalyst. In the postulated mechanism, the phosphate ion binds to the basic microenvironment adjacent to the glycated amino acid, stabilizing the Schiff base intermediate via acid-base catalysis (17). Importantly, protein glycation has also been reported in biotherapeutics, primarily recombinant monoclonal antibodies (mAbs) with the ε-amino group of lysine as the major glycation site (18, 19, 20, 21). In these mAbs, a nearby aspartic acid or histidine was proposed as the catalyst to stabilize the Schiff base intermediate.

mAbs have become increasingly important for therapeutic applications (22). All mAbs currently approved by the US Food and Drug Administration are produced from mammalian cell culture (20), during which mAbs are secreted into the culture medium, where glucose is present as an energy source and has a potential to interact with amino acids susceptible to glycation. In addition, during mAb purification, formulation, packaging, and storage, the extent of glycation may change, depending on the conditions and the nature of the glycation sites (19). Currently, the mechanisms and extent of glycation are assumed to be unpredictable from in silico analysis and in vitro experiments due to the complexity of this reaction (7,23).

Using cation exchange chromatography (CEX) and liquid chromatography-mass spectrometry (LC-MS), we identified a lysine residue that is sensitive to glycation in the complementarity-determining region (CDR) of one heavy chain (termed HC-CDR3-Lys98) of a bispecific mAb (termed mAb-1). Kinetic analysis demonstrated that the glycation reaction at this site can be well predicted and controlled in vitro. The glycation can also be predicted in vivo in a non-human primate. Thermodynamic analysis indicated that the ε-amino group of HC-CDR3-Lys98 has a substantially reduced pK_a, thus increasing the susceptibility to glycation. Finally, structural modeling showed the reduced pK_a is likely depressed by the low polar microenvironment of HC-CDR3-Lys98. Together, this study determined the mechanism of mAb-1 glycation and provided insights for strategies to mitigate protein glycation.

Materials and methods

General material details

A monoclonal, immunoglobulin-G4-based, bispecific antibody also referred to as mAb-1 and the anti-human antibody used in this study was manufactured by Regeneron Pharmaceuticals (Tarrytown, NY) and used throughout this study. mAb-1 has two different heavy chains termed HC and HC^∗, respectively. Concentrated stock mAb-1 solution was stored frozen before use. All other chemicals used were of analytical grade.

General experimental details

All reactions were carried out in type-1 glass vials, sealed with elastomeric stoppers and aluminum seals with a plastic flip-off cap, stored in temperature-controlled incubators. Four different buffer agents were used for different pH ranges: acetate for pH 5.0–5.5; histidine for pH 5.0–6.5; HEPES for pH 6.5–8.0; and phosphate for pH 6.0–8.0. All protein samples were prepared by directly mixing stock solutions of mAb-1 with other buffer components. The pH of each protein sample was monitored during the incubation to ensure no pH shift occurred. Protein concentrations were determined by the UV absorption at 280 nm using an extinction coefficient of 9.97 × 10⁻⁶ M⁻¹∙cm⁻¹.

HC-CDR3-Lys98 glycation analysis and sample preparation

The glycation reaction at the side chain of HC-CDR3-Lys98 was measured by two methods, CEX and LC-MS peptide mapping.

CEX

The CEX method was run on a Waters Acquity ultra performance liquid chromatography (UPLC) system with a YMC-BioPro SP-F column controlled at 25°C. Ten to fifty micrograms of mAb-1 was loaded to the CEX column in the binding buffer, 50 mM 2-(N-morpholino)ethanesulfonic acid (pH 6.5) with 20 mM NaCl, and eluted with a linear concentration gradient of 0.5 mM NaCl/min from 20 to 150 mM NaCl, in a 0.5 mL/min flow rate. The elution was monitored at a wavelength of 280 nm, and the chromatogram was processed by Empower 3 data software (Waters, Milford, MA), where the relative fraction of each charge variant was calculated from the total area of the chromatogram.

LC-MS peptide mapping

mAb-1 samples were denatured and reduced with 5 mM Tris(2-carboxyethyl) phosphine hydrochloride (Thermo Fisher Scientific, Waltham, MA) at 80°C for 10 min. The reduced cysteine residues were alkylated with 5 mM iodoacetamide (Thermo Fisher Scientific) at room temperature for 30 min in the dark. The urea concentration was diluted to 1.25 M before digestion. Trypsin (Promega, Sunnyvale, CA) or Endoproteinase Asp-N (New England Biolabs, Ipswich, MA) was added with a 1:20 enzyme:substrate ratio and incubated at 37°C for 4 h. Digestion was terminated by adding 20% trifluoroacetic acid (TFA) (Thermo Scientific, San Jose, CA). The digested samples were stored at −80°C until analysis. The digested samples were loaded and separated using an Acquity UPLC BEH C18 column (Waters, Milford, MA) on a Waters Acquity UPLC system, coupled to a Q Exactive plus mass spectrometer (Thermo Fisher Scientific, San Jose, CA). The mobile phase A was 0.05% TFA in water and mobile phase B was 0.1% TFA in acetonitrile. A gradient from 0.1% mobile phase B to 35% mobile phase B for 75 min at a flow rate of 0.25 mL/min was used for peptide separation. The MS acquisition consisted of a full mass scan followed by tandem mass (MS/MS) scans of the top five most intense ions of each full scan. The peptide identification was determined by Byonic (v.2.16.11; Protein Metrics, San Carlos, CA) and verified manually. The peptide quantification was determined by integration of the extracted peptide ion peaks using Skyline daily (v.4.1.1.18151; MacCoss Lab, University of Washington, Seattle, WA) with a mass window of 5 ppm.

Sample preparation of mAb-1 with different levels of glycation at HC-CDR3-Lys98

mAb-1 with different levels of glycation at HC-CDR3-Lys98 was purified by CEX method. Fully glycated mAb-1 was purified by collecting and pooling fractions that eluted between 8.6 and 9.1 min on CEX. Similarly, non-glycated mAb-1 was purified by collecting fractions that eluted between 9.3 and 10.1 min on CEX. The purified glycated and non-glycated forms of mAb-1 were each dialyzed against 10 mM histidine (pH 6.0). LC-MS peptide mapping confirmed the level of glycation at HC-CDR3-Lys98. The glycated and non-glycated species were mixed at different ratios to create mAb-1 solutions with varying percentages of glycation at HC-CDR3-Lys98.

In vitro bioassay of mAb-1 activity

Serial dilutions of mAb-1 were added in triplicate to the first target cells (Jurkat/NFAT-Luc, 10,000 cells/well) in the presence of the second target cells (OVCAR-3 cells, 10,000 cells/well). The antibody dilutions and bioassay were performed in Jurkat complete media (RPMI medium 1640 supplemented with 10% fetal bovine serum and 1× penicillin-streptomycin L-glutamine). Wells containing no antibody were used as a control. Plates were incubated at 37°C, 5% CO₂ for 4–6 h and then brought to room temperature for 15 min. One-Glo luciferase substrate (100 μL) was added to each well, and the plates were incubated for 3–5 min. The luminescence signal was measured using a PerkinElmer Victor X5 plate reader, and measured values were analyzed by a four-parameter logistic equation over an 11-point response curve using GraphPad Prism (v.8, GraphPad Software, San Diego, CA).

Kinetic and thermodynamic analysis

The apparent kinetic analysis was performed using the following two models: glycation reaction (i.e., the increase in the percent of glycated HC-CDR3-Lys98 over time),

$$\%glycation=A_1\left(1-e^{\left(-k_{app,1}\cdot t\right)}\right)+b,$$ (1)

and de-glycation reaction (i.e., the decrease in the percent of glycated HC-CDR3-Lys98 over time),

$$\%glycation=A_2\cdot e^{\left(-k_{app,2}\cdot t\right)}+b,$$ (2)

where A₁ and A₂ are the amplitudes, k_app,1 and k_app,2 are the apparent reaction rate constants, t is the time, and b is the baseline.

In the data-fitting procedure, t was fixed as the independent parameter and A₁, A₂, k_app,1, k_app,2, and b were allowed to vary to find the best fit values. For conditions without glucose, b was fixed at 0 to avoid overparameterization. For a glycation reaction, the equilibrium level of glycation (i.e., plateau) can be determined from the sum of A₁ and b. For a de-glycation reaction, the equilibrium level of glycation is determined from b.

The Arrhenius equation was applied to calculate the apparent activation energy, E_a, from de-glycation kinetics at different temperatures,

$$Ln\left(k_{app,2}\right)=Ln\left(A\right)-\frac{E_a}{RT},$$ (3)

where A is a constant, R is the gas constant, and T is the absolute temperature in Kelvin. The linear relationship between Ln(k_app,2) and 1/T indicated that the reaction follows Arrhenius behavior, where E_a can be determined from the slope.

For a more detailed kinetic analysis, the reaction between glucose and mAb-1 can be expressed in Fig. 1 A, where P is the non-glycated mAb-1, G is glucose, PG is the glycated mAb-1, and k₁ and k₋₁ are the association and dissociation rate constants, respectively. In the condition where the initial concentration of free glucose, [G₀], is much larger than the initial concentration of non-glycated mAb-1, [P₀], a kinetic description of the percent of glycation at time, t, can be described as follows:

$$\%glycation=\left(\frac{k_1\left[G_0\right]}{k_1\left[G_0\right]+k_{-1}}\left(1-e^{-\left(k_1\left[G_0\right]+k_{-1}\right)t}\right)+e^{-\left(k_1\left[G_0\right]+k_{-1}\right)t}\cdot \frac{\left[P{G}_0\right]}{\left[P_t\right]}\right)\cdot 100,$$ (4)

where [P_t] is the total protein concentration of mAb-1 and [PG₀] is the initial protein concentration of the glycated mAb-1. In the data-fitting procedure, [G₀], [PG₀], and [P_t] were fixed although k₁ and k₋₁ were allowed to vary to find the best fit values.

Figure 1.

Glycation reaction. (A) Macroscopic glycation reaction. (B) Microscopic de-protonation and glycation reactions are shown.

Based on Fig. 1 B, an equilibrium description of the percent of glycation can be described as follows:

$$\%glycation=\left(\frac{K\left[P_e\right]\left[G_e\right]}{\left[P_e\right]+K\left[P_e\right]\left[G_e\right]}\right)\cdot 100,$$ (5)

where [G_e] and [P_e] are the concentrations of free glucose and non-glycated mAb-1, respectively, at equilibrium. In the data-fitting procedure, [G_e] and [P_e] were implicitly solved over the range of 0 < [G] < [G_t] and 0 < [P] < [P_t], where [G_t] and [P_t] are the total concentrations of glucose and mAb-1, respectively, and the equilibrium constant, K, was allowed to vary to determine the best fit value.

Structure modeling for mAb-1

Predicted structure models for mAb-1 were generated using the molecular operating environment (Chemical Computing Group, Montreal, Canada). A database of antibody structures was used to generate the homology model for individual dimers of heavy and light chains. The Fab domains were built by grafting the appropriate framework and loop templates followed by energy minimization. Docking experiments were performed by modeling the covalent interaction between the ε-amino group of lysine side chain and the aldehyde group of glucose, where the local environment was refined by energy minimization.

Calculation of uncertainties in non-linear least square analysis

During the data-fitting procedure, all equations were programmed in Scientist Software (Micromath, St. Louis, MO). The best fit parameters were determined from the non-linear least square (NLLS) analysis with uncertainties reported in parenthesis. For symmetric errors, the uncertainties were reported as ± error under 95% confidence intervals. For asymmetric error, the uncertainties were reported as (lower limit, upper limit) under 95% confidence intervals.

HC-CDR3-Lys98 glycation in in vivo preclinical monkey serum

The preclinical serum samples were obtained from single-dose cynomolgus monkey pharmacokinetics studies. mAb-1 was administered to subjects intravenously. The cynomolgus monkey was dosed at 0.5 mg/kg, and serum samples were collected at designated time points (pre-dose, 5 min, 5 h, 1 day, 3 days, 7 days, 14 days, 28 days, and 42 days). The serum samples were stored at −80°C until analyses. The mAb-1 serum concentration at each collected time point was measured using an ELISA. In brief, the mAb-1 was captured on a microtiter plate coated with drug target. The mAb-1 captured on the plate was detected using biotinylated mouse anti-human immunoglobulin G4 monoclonal antibody, followed by NeutrAvidin conjugated to horseradish peroxidase. A luminol-based substrate specific for peroxidase was then added to achieve a signal intensity that is proportional to the concentration of mAb-1. The serum glucose levels were measured using a freestyle lite blood glucose monitoring system (Abbott Laboratories, Chicago, IL). All animal studies were approved by the appropriate institutional animal use committee.

Affinity purification of mAb-1 from serum samples

mAb-1 was purified from the collected monkey serum samples by affinity purification (24,25). In brief, a biotinylated anti-human antibody was conjugated to Dynabeads MyOne Streptavidin T1 magnetic beads (Invitrogen, Carlsbad, CA) at room temperature for 10 min. The conjugated beads were then incubated with serum samples at room temperature for 30 min. The beads were washed with HBS-EP buffer (GE Healthcare, Pittsburgh, PA) and then eluted with 0.1% formic acid (FA) and 50% acetonitrile.

Tryptic digestion

The purified mAb1 samples were dried down using a vacuum concentrator (LABCONCO, Kansas City, MO). The dried samples were re-suspended in 100 mM Tris-HCl containing 8 M urea and 10 mM Tris(2-carboxyethyl) phosphine hydrochloride and then incubated at 37°C for 30 min. The reduced cysteine residues were alkylated with 10 mM iodoacetamide at room temperature for 30 min in the dark. After alkylation, the urea concentration was diluted to 1.25 M before digestion. Trypsin (Promega) was added to the samples at an enzyme:substrate ratio of 1:10 and incubated at 37°C for 4 h. Digestion was terminated by addition of 20% FA. The digested samples were stored at −80°C until analysis.

LC-MS/MS and data analysis

Peptides generated by trypsin digestion were separated using an Acquity UPLC CSH C18 1.7 μm, 2.1 mm × 150 mm column (Waters) on an Acquity I-Class UPLC system (Waters) coupled to a Q Exactive Plus mass spectrometer (Thermo Fisher Scientific). Mobile phase A was 0.1% FA in water, and mobile phase B was 0.1% FA in acetonitrile. A gradient increasing from 2% mobile phase B to 30% mobile phase B over 56 min at a flow rate of 0.25 mL/min was used for peptide separation. The MS acquisition consisted of a full mass scan followed by MS/MS scans of the top five highest intensity ions from each full scan. Peptide and post-translational modification (PTM) identification was determined by Byonic (v.2.16.11; Protein Metrics) and verified manually. To quantify relative abundance of PTMs, the extracted ion chromatograms, based on the m/z of the first isotope peak of both the modified peptide and native peptide, were generated and the extracted peak areas were integrated using Skyline-daily (v.4.1.1.18151; MacCoss Lab) using a mass window of 5 ppm. The percentage of each PTM variant was calculated using the extracted ion chromatogram peak area of the modified peptide relative to the sum of the peak areas of the modified and native peptides.

Results

Charge variant analysis of mAb-1 by CEX and LC-MS peptide mapping revealed a species with reversible glycation at HC-CDR3-Lys98

Protein charge variants are frequently identified as critical quality attributes for biotherapeutics and typically require extensive characterizations (26, 27, 28). Here, CEX was used to separate mAb-1 charge variants with two major species eluted at 8.9 and 9.5 min and a number of minor species eluted earlier than 8.6 min and later than 10 min (Fig. 2 A). After incubating mAb-1 in 10 mM histidine (pH 6.0) and at 37°C for 14 days, the relative amount of the species eluting at 8.9 min decreased and the relative amount of the species eluting at 9.5 min increased by approximately the same amount (Fig. 2 A), suggesting an ongoing reaction causing the interconversion of these two species.

Figure 2.

CEX and LC-MS peptide mapping of mAb-1. (A) Charge variants of mAb-1 separated by CEX. Black: mAb-1 in 10 mM histidine (pH 6.0) without treatment is shown; green: mAb-1 in 10 mM histidine (pH 6.0) and incubated at 37°C for 14 days is shown; red: mAb-1 in 10 mM histidine (pH 6.0) with 0.1 M glucose and incubated at 37°C for 14 days is shown. Glycated mAb-1 elutes between two gray dashed lines (8.6 and 9.1 min) are shown. Non-glycated mAb-1 elutes between two gray dashed lines (9.3 and 10.1 min) are shown. (B) The UV chromatograms of the peptide maps of the neat mAb-1 (black) and the mAb-1 incubated at 37°C for 14 days (green) in the absence of glucose. The glycated lysine residue is highlighted in red. To see this figure in color, go online.

To understand the chemical nature of these two mAb-1 charge variants, they were purified using CEX and assessed using LC-MS peptide mapping analysis to identify and quantify the PTMs (29). This analysis revealed a mass shift of +162 Da at HC-CDR3-Lys98 in the species eluting at 8.9 min, but not in the species eluting at 9.5 min. Quantitative analysis indicated that the levels of glycation in the species eluting at 8.9 min and 9.5 min species were 88% and 1%, respectively. No other PTMs were identified that differed significantly within these two species. Similarly, in the other heavy chain (HC^∗) CDR3, there is a lysine at position 98 (known as HC^∗-CDR3-Lys98) with no detectable level of glycation (data not shown).

After incubating mAb-1 at 37°C for 14 days, LC-MS peptide mapping analysis revealed a decrease of the glycated HC-CDR3-Lys98 peptide peak and a concomitant increase of the native (i.e., non-glycated) HC-CDR3-Lys98 peptide peak compared with the neat mAb-1 (Fig. 2 B). The levels of glycation in neat mAb-1 and 37°C-incubated mAb-1 were 32% and 2%, respectively, confirming the reversible nature of this reaction and the consistency of glycation levels quantified by CEX and LC-MS peptide mapping.

To investigate whether this glycation is driven by glucose, mAb-1 was incubated with 0.1 M glucose at 37°C for 14 days. An increase in the glycated species with a concomitant decrease in the non-glycated species was observed (Fig. 2 A). Similarly, the level of glycation increased as determined by LC-MS peptide mapping (data not shown). Together, the CEX and LC-MS peptide-mapping analyses indicated that the glycation reaction is reversible and that the level of glycation depends on the presence of glucose in solution.

The level of glycation at HC-CDR3-Lys98 is inversely correlated with potency

To understand whether the glycation at HC-CDR3-Lys98 interferes with biological activity (e.g., potency), glycated and non-glycated forms of mAb-1 were purified by CEX chromatography, mixed at various ratios to generate mAb-1 samples with glycation levels ranging from 1% to 88% (determined by LC-MS peptide mapping), and subjected to a cell-based potency assay. The results indicated a linear and inverse correlation between the level of glycation at HC-CDR3-Lys98 and the potency, with a higher level of glycation associated with a lower level of drug potency (Fig. 3). In addition, the binding affinity to the corresponding antigen, as measured by surface plasmon resonance, indicated that mAb-1 with 88% glycation had approximately a twofold decrease in association rates compared with mAb-1 with 1% glycation (6.54 × 10⁴ Ms⁻¹ compared with 1.10 × 10⁵ Ms⁻¹), resulting in approximately twofold lower binding affinity (9.34 × 10⁻⁹ M compared with 5.13 × 10⁻⁹ M). Mutating this HC-CDR3-Lys98 to arginine reduced the binding affinity by over 20-fold (data not shown). Together, these results suggested that the HC-CDR3-Lys98 is critical for mAb-1 binding to the antigen and its glycation can compromise binding affinity.

Figure 3.

The correlation between the level of glycation at HC-CDR3-Lys98 in mAb-1 and potency by cell-based potency assay is linear and has a correlation coefficient of 0.97. Error bars represent the standard deviation of triplicate potency results.

HC-CDR3-Lys98 glycation reaction depends on the concentration of glucose in solution

To understand the impact of glucose on the glycation at HC-CDR3-Lys98, 6.9 μM mAb-1 (∼1 mg/mL) was incubated with different concentrations of glucose ranging from 0.3 to 111 mM at 37°C for 28 days. The levels of glycation at HC-CDR3-Lys98 slowly decreased in solutions with ≤3 mM glucose, increased in solutions with ≥6 mM glucose, and reached steady states (i.e., equilibrium) in all conditions over time (Fig. 4 A).

Figure 4.

Kinetic and thermodynamic analyses of mAb-1 glycation reaction in the presence of different concentrations of glucose. (A) Glycation and de-glycation kinetics of 6.9 μM mAb-1 in 30 mM HEPES (pH 7.4) with different concentrations of glucose when incubated at 37°C over the course of 28 days: no glucose (closed black circle); 0.3 mM glucose (closed teal circle); 0.6 mM glucose (closed blue circle); 3 mM glucose (closed red circle); 6 mM glucose (closed pink circle); 10 mM glucose (closed dark green circle); 28 mM glucose (closed light green circle); 56 mM glucose (closed olive circle); and 111 mM glucose (closed burgundy circle). Solid lines are from the global analysis using a kinetic model with the best fit k₁ = 8.89 (±1.79) M⁻¹ day⁻¹ and k₋₁ = 0.09 (±0.02) day⁻¹. (B) Thermodynamic analysis of glycation reaction is shown. The data were determined from the best fit of baseline and amplitude using Eqs. 1 and 6. The solid line is from the thermodynamic analysis with the best-fit association constant K = 112 (±40) M⁻¹. To see this figure in color, go online.

The kinetics of glycation were analyzed with first-order models (Eqs. 1 and 2) by resolving three fitting parameters—the apparent rate constant (k_app), amplitude (A), and baseline (b) for each solution condition (Table S1), where the level of glycation at equilibrium was determined from A + b for glycation reactions and from b for de-glycation reactions. A thermodynamic model (Eq. 5), based on a simple chemical reaction: $P+G\leftrightarrow PG$ (G, free glucose; P, non-glycated mAb-1; PG, glycated mAb-1; Fig. 1 A), was used to directly fit the equilibrium level of glycation at various glucose concentrations, returning the best fit of association constant K = 112 (±40) M⁻¹ (Table S2). In addition, a kinetic model based on the same chemical reaction was applied to determine the glycation (k₁) and de-glycation (k₋₁) rates (Eq. 4). The global analysis of data from different glucose concentrations returned the best fit k₁ = 8.89 (±1.79) M⁻¹ day⁻¹ and k₋₁ = 0.09 (±0.02) day⁻¹ (Table 1). The association constant, K, calculated from k₁ and k₋₁ is 99 M⁻¹, statistically identical to the K obtained from thermodynamic analysis. The dissociation constant (K_D = 1/K) of this reaction is approximately 8–10 mM, suggesting the level of glycation at HC-CDR3-Lys98 depends on the actual glucose concentration in the cell culture medium at any given time, typically ranging from 5 to 55 mM. Therefore, to produce mAb-1 with a consistent charge variant profile, it is important to maintain a consistent level of glucose during the cell culture process.

Table 1.

Summary of kinetic and thermodynamic information from Fig. 1 and Van't Hoff analysis for the mAb-1 glycation reaction at 37°C

	Kinetic and thermodynamic analysis			Thermodynamic analysis		Van't Hoff analysis
Kinetic and thermodynamic information	k1 (M−1 day−1)	k−1 (day−1)	K (M−1)	K1 (M)	K2 (M−1)	ΔH (kcal/mol)	−T∗ΔS (kcal/mol)	ΔS (kcal/mol∙K)
Best fit parameters	8.89 (±1.79)	0.09 (±0.02)	112 (±40)	2.03 (±0.76) × 10−7	120 (±26)	−4.0 (±0.8)	2.2 (±0.9)	−0.007 (±0.003)

HC-CDR3-Lys98 de-glycation kinetics depends on the temperature and pH in solution and follows Arrhenius behavior

Temperature is among the most important factors influencing the Maillard reaction (1). The impact of temperature on the de-glycation kinetics was examined by incubating 13.8 μM mAb-1 in solutions without glucose (10 mM histidine, 0.05% polysorbate 20, and 292 mM sucrose) and at temperatures ranging from 20°C to 45°C for up to 35 days. At all temperatures tested, mAb-1 underwent de-glycation over time. This reaction is specific to the reducing sugar, as the presence of an excess amount of non-reducing sugar (e.g., sucrose) does not impact the level of glycation. Another critical factor that impacts glycation kinetics is pH, where an acid in solution was suggested to catalyze the Amadori rearrangement (2). As a result, the rate of de-glycation on different proteins was found to increase at higher pH (2,12,30). Indeed, this was observed for mAb-1: the apparent de-glycation is faster at pH 6.0 than at pH 5.0 (Fig. 5 A).

Figure 5.

De-glycation kinetics analysis of mAb-1. (A) mAb-1 de-glycation kinetics at different temperatures and pH. mAb-1 in 10 mM histidine (pH 6.0), 292 mM sucrose, and 0.05% polysorbate 20 was incubated at pH 5.0, 5.5, and 6.0, at 20°C (closed burgundy circle), 30°C (closed orange circle), 35°C (closed light green circle), 40°C (closed dark green circle), and 45°C (closed teal circle). The data were fitted to Eq. 2. (B) Arrhenius analysis of mAb-1 de-glycation at pH 5.0 (closed black circle), pH 5.5 (closed green circle), and pH 6.0 (closed blue circle) is shown. The lines are the linear fit of each data set at designated pH with correlation coefficients of 0.98. The apparent activation energy was determined from the slope of the fit. (C) Comparison of the predicted and real time de-glycation kinetics at 5°C at pH 5.0 (closed black triangle), pH 5.5 (closed green triangle), and pH 6.0 (closed blue triangle) is shown. The curves are the predictions of percent glycation generated from Arrhenius analysis. To see this figure in color, go online.

We applied a first-order kinetic model to determine the de-glycation rates under different solution conditions and temperatures. The natural log of the de-glycation rates linearly decreases with decreasing temperatures, suggesting the de-glycation kinetics follow Arrhenius behavior in the range of 20°C–45°C (Fig. 5 B). Using the Arrhenius equation (Eq. 3), the apparent activation energy, E_a, was calculated to be 23.5 (±2.1), 23.1 (±1.8), and 21.4 (±1.7) kcal/mol, at pH 5.0, pH 5.5, and pH 6.0, respectively. The lower E_a at higher pH suggests the energy barrier of de-glycation is lower, as a faster reaction rate was observed at higher pH.

Next, we extrapolated Arrhenius behavior to determine the de-glycation rates at 5°C. These rates allow the simulation of the de-glycation profiles using Eq. 2, where A₂ was fixed at the level of glycation at t = 0 and k_app was fixed at the extrapolated de-glycation rate (0.0011, 0.0007, and 0.0004 day⁻¹ at pH 6.0, 5.5, and 5.0, respectively) at 5°C. As shown in Fig. 5 C, the simulated profiles and real-time de-glycation data at 5°C are in excellent agreement for at least 36 months, indicating that the de-glycation kinetics (1) follow Arrhenius behavior, even at a lower temperature, and (2) can be well predicted from the short-term incubation experiments at higher temperatures.

pH dependency of the glycation equilibrium revealed a low pK_a of the amine on the HC-CDR3-Lys98 side chain

To understand the pH dependence of glycation, 6.9 μM mAb-1 was incubated in different buffers with pH values ranging from 5 to 8 and with glucose concentrations varying from 3 to 11 mM until the glycation reached equilibrium. The equilibrium level of glycation was determined using first-order kinetic models and plotted as a function of pH (Fig. 6). At a constant pH (e.g., pH 7.0), a higher equilibrium level of glycation was observed at the higher glucose concentrations, consistent with previous results (Fig. 4). At constant glucose concentrations, the level of glycation increased non-linearly with increasing pH. It has been suggested that phosphate and/or carbonate ions can facilitate protein glycation of lysine residues via the preferential interaction with the local environment of lysine (2,7). However, this was not observed for mAb-1. For all conditions that were tested, the buffer components had no apparent effect on the glycation equilibrium (Fig. 6).

Figure 6.

The pH dependence of mAb-1 glycation at equilibrium. We incubated 6.9 μM mAb-1 in 30 mM of different buffers with varying concentrations of glucose, (A) 3 mM, (B) 6 mM, and (C) 11 mM, at 37°C for 28 days. The equilibrium levels of glycation were determined from the apparent kinetic analysis. All three data sets were analyzed globally by Eq. 8. The results of this analysis demonstrated the pK_a of Lys98 to be 6.7 (6.6, 6.9) and the microscopic association constant, K₂, for glycation to be 120 (±26) M⁻¹. To see this figure in color, go online.

A simple equilibrium model was applied to quantitatively analyze the pH dependence of glycation equilibrium. This model constitutes two reactions: the de-protonation of the amine on HC-CDR3-Lys98 side chain and the binding of glucose to the de-protonated amine, as shown in Fig. 1 B, where PH⁺ and P are mAb-1 with the protonated and de-protonated amine on the HC-CDR3-Lys98 side chain, respectively; G is free glucose; H⁺ is free proton; PG is mAb-1 glycated at HC-CDR3-Lys98; K₁ is the equilibrium constant for the de-protonation reaction; and K₂ is the equilibrium constant for the glycation reaction. The overall observed fraction of glycated mAb-1 (% glycation) can be written

$$\%glycation=\left(\frac{\left[PG\right]}{\left[P_t\right]}\right)\cdot 100,$$ (6)

where [PG] is the concentration of glycated mAb-1 and [P_t] is the concentration of total mAb-1. According to mass conservation in Fig. 1 A, [P_t] is the sum of [P], [PH⁺], and [PG], where [P] is the concentration of de-protonated mAb-1 and [PH⁺] is the concentration of protonated mAb-1. Hence, Eq. 6 can be expressed as

$$\%glycation=\left(\frac{\left[PG\right]}{\left[P\right]+\left[P{H}^{+}\right]+\left[PG\right]}\right)\cdot 100=\left(\frac{K_2\left[P\right]\left[G\right]}{\left[P\right]+\left(\frac{\left[P\right]\left[H^{+}\right]}{K_1}\right)+K_2\left[P\right]\left[G\right]}\right)\cdot 100$$ (7)

followed by the rearrangement to obtain Eq. 8,

$$\%glycation=\left(\frac{K_1{K}_2\left[G\right]}{\left[H^{+}\right]+K_1+K_1{K}_2\left[G\right]}\right)\cdot 100,$$ (8)

where [G] and [H⁺] are the concentrations of free glucose and free protons in solution, respectively. The pK_a of the ε-amine group of HC-CDR3-Lys98 can be determined from pK_a = −log (K₁), and K₂ is a pH-independent, microscopic association constant for the glycation reaction.

Equation 8 was used to globally fit three equilibrium data sets shown in Fig. 6. In the NLLS analysis, [H⁺] was determined from the measured pH; [G] was implicitly solved over the range of 0 < [G] < [G_t] ([G_t] is the total glucose concentration determined from the added and bound glucose concentrations) with [P_t] fixed at 6.9 μM for each condition. This analysis returned the best-fit global parameters K₁ = 2.03 (±0.76) × 10⁻⁷ M and K₂ = 120 (±26) M⁻¹ (Table 1), by which the pK_a of the ε-amino group of the HC-CDR3-Lys98 was estimated to be 6.7 (6.6, 6.9). This pK_a is lower than the typical lysine ε-amino pK_a value (10.4) in water (31).

Thermodynamic analysis indicated the glycation reaction is enthalpically driven

To understand the driving force of the glycation reaction, the apparent association constant, K, of the glucose binding to HC-CDR3-Lys98 was measured by incubating 6.9 μM mAb-1 with different concentrations of glucose and at temperatures ranging from 15°C to 45°C. The equilibrium level of glycation was determined for each condition, followed by NLLS analysis to determine K at each temperature (Fig. S1; Table S2). K slightly decreases with increasing temperature, suggesting the reaction is not driven by the change of entropy. Using Van't Hoff analysis (plotting ln [K] against 1/T) (Fig. 7), we determined the changes in enthalpy (ΔH = −4.0 [±0.8] kcal/mol) and entropy (ΔS = −0.007 [±0.003] kcal/mol∙K) (Table 1), thereby indicating that the glycation at HC-CDR3-Lys98 is predominantly driven by enthalpy. The small and negative entropy change suggests that the overall microenvironment may be more constrained upon glycation.

Figure 7.

Van't Hoff analysis of mAb-1 glycation equilibrium. The association constant, K, was determined from the thermodynamic analysis of mAb-1 glycation at different temperatures and glucose concentrations. The line is the linear fit of all data with a correlation coefficient of 0.89. Error bars represent the upper and lower values from thermodynamic analysis under 95% confidence intervals.

Structural modeling suggested the HC-CDR3-Lys98 is surrounded by environment with low polarity

Lysine residues with lowered pK_a values are typically found buried inside the protein or surrounded in the environment with low polarity. The shift of pK_a may be used to calculate the apparent dielectric constant, ε_app, using Born formalism (31),

$$1.36z\left(p{K}_{a,ref}-p{K}_a\right)=\frac{332\cdot Z^2}{2r_{cav}}\left(\frac{1}{{\epsilon }_{app}}-\frac{1}{{\epsilon }_{H_{2}O}{e}^{\kappa r_H}}\right)+\frac{332\cdot Z^2}{2r_H}\left(\frac{1}{{\epsilon }_{H_{2}O}{e}^{\kappa r_H}}-\frac{1}{{\epsilon }_{app}}\right),$$ (9)

where z is the charge number of the lysine side chain, pK_a,ref is the reference pK_a for lysine in water, Z is the valence of the transferred ion, r_cav is the cavity radius of the ionizable part of lysine, r_H is the hydrodynamic radius of mAb-1, ε_H2O is the dielectric constant of water, and κ is the inverse Debye length of mAb-1. This method assumes the self-energetic difference between the charged lysine in water and in HC-CDR3-Lys98 local environment is the sole contributor to the shift of pK_a. Using the following parameters: pK_a,ref = 10.4; Z = 1; r_cav = 2Å; r_H = 49Å; ε_H2O = 74.2 (at 37°C); and κ = 8.7 × 10⁻⁴ (1/Å), the estimated apparent dielectric constant, ε_app, is 13.0, which is substantially lower than the corresponding dielectric property in water (32), indicating that HC-CDR3-Lys98 is surrounded in a relatively low hydrated environment.

To understand the local environment of HC-CDR3-Lys98, a homology model was built (Fig. 8). The predicted mAb-1 structure suggested the side chain of HC-CDR3-Lys98 is in a non-polar environment, surrounded by the aromatic rings of Tyr32, Phe27, and Tyr106 and the aliphatic side chain of Val2 (Fig. 8 A). In contrast, the HC^∗-CDR3-Lys98 is in the environment with high polarity, where the side chain is facing the hydroxyl group of Tyr32 and Tyr109, and in close proximity to the carboxyl group of the side chain of Asp112, which can potentially form an ionic bond, stabilizing the local environment of HC^∗-CDR3-Lys98 (Fig. 8 C). A local energy minimization was performed for the HC-CDR3-Lys98 with glucose adduct (Fig. 8 B). The ΔG for this interaction is estimated to be −2.8 kcal/mol, which is in excellent agreement with the experimental data (Table S2). Together, the results from in silico analyses are consistent with thermodynamic measurements, suggesting that HC-CDR3-Lys98 is in the microenvironment with lower hydration and low ability for ionization.

Figure 8.

Predicted structure of two mAb-1 arms. The heavy and light chains are shown in dark and light green, respectively. The CDR1 and CDR2 of the HCs are shown in brown. The CDR3 of the HCs is shown in red. The CDR of the light chains is shown in purple. Nitrogen atoms are colored blue. Oxygen atoms are colored red. Carbon atoms are colored dark gray. Hydrogen atoms are colored light gray. (A) Predicted local environment of HC-CDR3-Lys98 is shown, where this lysine is surrounded by the aromatic rings of Tyr32, Phe27, and Tyr106 and the aliphatic side chain of Val2. (B) Predicted local environment of the HC-CDR3-Lys98 with glucose adduct from local energy minimization is shown. (C) Predicted local environment of HC^∗-CDR3-Lys98 is shown, where this lysine is surrounded by Phe27, Tyr32, and Asp112. Unlike HC-CDR3-Lys98, which faces the aromatic rings, HC^∗-CDR3-Lys98 is oriented toward the side chain of Asp112 with predicted hydrogen bonds showing in turquoise. To see this figure in color, go online.

The level of HC-CDR3-Lys98 glycation decreased in a non-human primate before reaching an equilibrium level predicted from in silico analysis

Finally, we investigated the change in the level of HC-CDR3-Lys98 glycation in a non-human primate. A single dose (0.5 mg/kg) of mAb-1 was administered to a cynomolgus monkey followed by the collection of monkey serum samples at various time points (pre-dose, 5 min, 5 h, 1 day, 3 days, 7 days, 14 days, 28 days, and 42 days). Using immunoprecipitation followed by LC-MS peptide mapping to characterize mAb-1 in these serum samples (33), we revealed that the in vivo levels of glycation decreased from 34.4% in the 5-min post-administration sample to 23.7% in the 42-day post-administration sample (Fig. 9 A). We also measured the serum glucose concentrations, which fluctuated around 100 mg/dL (5.6 mM; Fig. 9 B). At this concentration range of glucose, we expected the level of glycation to slightly decrease over time before reaching an equilibrium. Indeed, after 14 days post-administration, the level of HC-CDR3-Lys98 glycation was stable at around 24%, indicating the in vivo level of HC-CDR3-Lys98 glycation was driven by the glucose concentrations in serum.

Figure 9.

Changes of HC-CDR3-Lys98 glycation in vivo. (A) The levels of mAb1 HC-Lys98 glycation in the monkey PK serum samples. The predicted glycation profile using Eq. 2 is shown as the red curve. (B) The serum glucose concentration in the monkey serum samples is shown. To see this figure in color, go online.

To compare the in vitro and in vivo kinetics, we generated a prediction profile using parameters determined in vitro. The level of glycation was simulated using Eq. 2 with b fixed at the equilibrium level of glycation (24.0%), A₂ fixed at the difference between the levels of glycation at equilibrium and pre-dose (10.5%), and k_app fixed at 0.14 (day⁻¹), as shown in Table S1. This predicted glycation profile was in excellent agreement with the in vivo level of glycation measured by LC-MS (red curve, Fig. 9 B), suggesting the mechanism of glycation and de-glycation determined in vitro and in silico could be applied to the prediction of in vivo glycation profile in non-clinical and clinical studies.

Discussion

We characterized the glycation on a lysine residue in one HC CDR of a bispecific monoclonal antibody, mAb-1. This reaction occurs between a reducing sugar, glucose, and the ε-amino group of the lysine side chain, resulting in a decreased antigen binding affinity and reduced mAb-1 potency. Kinetic analysis suggested the Arrhenius behavior of this reaction. Thermodynamic and structural modeling revealed that this lysine, despite being at the surface of the protein, has a low pK_a value depressed by being in the microenvironment with low polarity. Understanding the mechanism of glycation enables better control of this PTM during the manufacturing process and offers a precise prediction of this PTM in animal and human blood streams, both of which are critical for the safety and efficacy of mAb-1.

Based on the structural modeling, HC-CDR3-Lys98 is located at the center of one of the antibody-antigen-binding sites; therefore, a major concern from this glycation is the impact on potency. Whereas a high level of glycation (88%) on HC-CDR3-Lys98 reduces the drug potency by approximately twofold compared with essentially no glycation (1%), mutating this lysine to arginine reduced the binding affinity by over 20-fold, indicating that this lysine and its local environment are critical for the interaction with the antigen.

The level of mAb-1 HC-CDR3-Lys98 glycation in vivo could impact the potency of mAb-1 and thus is another important attribute to monitor. The normal blood glucose level typically ranges between 4 and 7 mM (34). Based on the equilibrium constant resolved at pH 7.4 and 37°C, the level of mAb-1 glycation is expected to reach a steady state between 20% and 35% after circulating for approximately 14 days, which is similar to the level of glycation found in human serum albumin and hemoglobin (3,5). We characterized the level of glycation of mAb-1 in a cynomolgus monkey with a blood glucose level around 5.6 mM. As we expected, the level of glycation in circulating mAb-1 reached a steady state of ∼24% after 14 days post-administration. This level of glycation remained stable for at least another 28 days, indicating that the serum glucose has a direct impact on mAb-1 glycation. To maintain a consistent therapeutic effect, mAb-1 should be manufactured with a level of glycation approximated to the equilibrium level in the physiological concentration range of glucose.

During the cell culture process, glucose is an essential energy source for the growing cells. To maintain a high growth rate for cells, glucose is fed constantly into the bioreactor, where the unconsumed glucose impacts the glycation at HC-CDR3-Lys98. The kinetic analysis suggests that, once mAb-1 is glycated, the de-glycation is relatively slow as compared with the glycation reaction. The net result is that, during the purification process (without glucose present), the level of glycation is likely to be similar to the level of glycation before harvesting in the bioreactor. In addition, during storage at 5°C, our data showed that glycated mAb-1 undergoes a consistent and measurable rate of de-glycation.

The de-glycation kinetics follow Arrhenius behavior in the temperature range from 5°C to 45°C (Fig. 5). As a result, the data generated from the higher temperatures (20°C–45°C) can be used in conjunction with an Arrhenius model to predict the de-glycation behavior at lower temperatures, such as 5°C. The calculated activation energy from the Arrhenius equation indicated that the de-glycation is pH dependent with lower activation energy at higher pH, suggesting a faster de-glycation rate at higher pH. Experimental results confirmed this prediction and demonstrated that the apparent de-glycation rate constant, at 20°C, increased by three-fold when pH increased by one unit (Table S1). This observation is likely due to increased rate of the reverse reaction for the Amadori product (ketoamine) and a concomitant release of glucose from the Schiff base at higher pH.

Although the apparent de-glycation kinetics are faster at higher pH, at steady state, the equilibrium level of glycation for mAb-1 is greater at higher pH (Fig. 6), suggesting the glycated HC-CDR3-Lys98 is more stable at higher pH. In addition, our data showed that the presence of histidine and phosphate and carbonate ions does not promote the overall glycation reaction. This was unexpected because the previous literature indicated the histidine and phosphate and carbonate ions facilitate the Amadori rearrangement (8,17,21). An alternate explanation of this behavior is that the glycation at HC-CDR3-Lys98 is not driven by stabilizing the Schiff base intermediate or an Amadori product but rather by lowering the pK_a at HC-CDR3-Lys98. When the equilibrium data at different pH conditions were analyzed globally using a simple de-protonation model, a lower-than-normal pK_a value of 6.7 for HC-CDR3-Lys98 was resolved. This pK_a suggests that, at pH 7.4, 84% of HC-CDR3-Lys98 will be de-protonated. Analyzing the shift in pK_a with a Born formalism returned an apparent dielectric constant of 13.0. Although this value is slightly larger than the dielectric property observed inside the protein (31,35), it is significantly lower than the dielectric constant of water, implying the microenvironment of HC-CDR3-Lys98 is less hydrated than what might be expected at the surface of a protein. In addition, the thermodynamic analysis suggests that the overall glycation reaction is favored by enthalpy and unfavored by entropy, suggesting the following mechanism: upon glycation, there is a slight structural reorganization causing the local environment to be more constrained. During this reorganization process, the microenvironment with low hydration limits the extent of water rearrangement. Hence, the reaction is primarily driven by the change of enthalpy.

Most examples of lysine ε-amino groups with a reduced pK_a are found buried inside proteins, serving, for example, enzymatic or energy transduction functions (35,36). However, HC-CDR3-Lys98 is on the surface and critical for the interaction with the antigen. HC^∗-CDR3-Lys98 (the corresponding amino acid on the other HC CDR of mAb-1), on the other hand, is also exposed to the surface, and yet the level of glycation is below the level of detection, which suggests the local environment at HC-CDR3-Lys98 facilitates this reaction. The simulated mAb-1 structure indicated the microenvironment of HC-CDR3-Lys98 is less polar than HC^∗-CDR3-Lys98, not only due to the surrounding amino acids but also due to the orientation of their side chains. Within 4.5 Å, HC-CDR3-Lys98 is surrounded by aromatic rings from amino acids neighboring in space. The closest polar group is the side chain of Tyr32, approximately 4.7 Å away; however, lysine and tyrosine are unlikely to form an ionic interaction. On the other hand, HC^∗-CDR3-Lys98 is facing toward the side chains of Asp112 and can potentially form a salt bridge with the carboxyl group on Asp112 (only 1.8 Å away). Because of the nature of the local environment, it is possible that HC-CDR3-Lys98 may have more flexibility than HC^∗-CDR3-Lys98 before glycation. After glycation, the glucose-lysine adduct may undergo structural reorganization and become more constrained, as predicted by the negative entropy change.

Conclusions

A comprehensive approach for studying protein glycation by analytical, kinetic, thermodynamic, and structural approaches is presented. This glycation is reversible and depends on solution temperature, glucose concentration, and pH. The analytical methods indicated that the glycation occurs in only one CDR of mAb-1 and impacts its antigen binding affinity. Kinetic analysis suggested that this reaction follows Arrhenius behavior, and the level of glycation can be well predicted in vitro for at least 36 months at 5°C and in a non-human primate for at least 42 days. Thermodynamic analysis revealed the apparent affinity of this reaction, which aligns well with both in silico and in vivo data. A simple de-protonation model demonstrated that this glycation is driven by the low pK_a of the ε-amino group on HC-CDR3-Lys98. This is unusual considering the proximity of HC-CDR3-Lys98 to the surface of the protein. Furthermore, structural prediction suggested that this lysine is located in an environment less polar than the surrounding solvent, consistent with the low apparent dielectric constant calculated from thermodynamic analysis. Altogether, our results elucidate the mechanism of protein glycation and provide strategies to mitigate this PTM by optimizing solution conditions.

Author contributions

X.X., J.A.O., Z.G., T.P., D.E. K., and T.-C.Y. designed and performed the studies and analyzed the data. J.A.O., Z.G., and T.-C.Y. developed the biophysical models. X.X., D.E. K., and T.-C.Y. wrote the manuscript with help from H.Q., N.L., T.P., K.S. G., and M.S.

Editor: John Correia.