Interrogating heterogeneity of cysteine-engineered antibody-drug conjugates and antibody-oligonucleotide conjugates by capillary zone electrophoresis-mass spectrometry

ABSTRACT

Production of site-specific cysteine-engineered antibody-drug conjugates (ADCs) in mammalian cells may produce developability challenges, fragments, and heterogenous molecules, leading to potential product critical quality attributes in later development stages. Liquid phase chromatography with mass spectrometry (LC-MS) is widely used to evaluate antibody impurities and drug-to-antibody ratio, but faces challenges in analysis of fragment product variants of cysteine-engineered ADCs and oligonucleotide-to-antibody ratio (OAR) species of antibody-oligonucleotide conjugates (AOCs). Here, for the first time, we report novel capillary zone electrophoresis (CZE)-MS approaches to address the challenges above. CZE analysis of six ADCs made with different parent monoclonal antibodies (mAbs) and small molecule drug-linker payloads revealed that various fragment impurities, such as half mAbs with one/two drugs, light chains with one/two drugs, light chains with C-terminal cysteine truncation, heavy chain clippings, were well resolved from the main species. However, most of these fragments were coeluted or had signal suppression during LC-MS analysis. Furthermore, the method was optimized on both ionization and separation aspects to enable the characterization of two AOCs. The method successfully achieved baseline separation and accurate quantification of their OAR species, which were also highly challenging using conventional LC-MS methods. Finally, we compared the migration time and CZE separation profiles among ADCs and their parent mAbs, and found that properties of mAbs and linker payloads significantly influenced the separation of product variants by altering their size or charge. Our study showcases the good performance and broad applicability of CZE-MS techniques for monitoring the heterogeneity of cysteine-engineered ADCs and AOCs.

KEYWORDS:

• Antibody-drug conjugates
• antibody-oligonucleotide conjugates
• capillary zone electrophoresis-mass spectrometry
• fragment impurities
• oligonucleotide to antibody ratio

Introduction

Antibody-drug conjugates (ADCs) are a critical class of biotherapeutics that utilize monoclonal antibodies (mAbs) to deliver drugs for targeted cancer treatment.¹ In recent years, site-specific conjugated ADCs have been developed by incorporating cysteine residues on mAbs and conjugating payloads (e.g., small cytotoxic molecules, oligonucleotides) site-specifically.^2–5 Cysteine-engineered ADCs typically well maintain the antibody’s structural disulfide bonds and carry homogenous drug loadings.^6–8 Therefore, they might provide better stability in plasma, developability, therapeutic index, and pharmacokinetics profiles compared to conventional lysine or interchain cysteine-conjugated ADCs.⁶ Production of cysteine-engineered ADCs requires monitoring of undesired product variants, including fragments, variants with different post-translational modifications (PTMs), or drug-to-antibody ratios (DAR), and other impurities, to ensure product quality and safety. High-resolution mass spectrometry (MS) enables comprehensive interrogation of ADC samples via intact analysis.^9,10 Due to the heterogeneity of the biotherapeutics, efficient and high-resolution separation is needed to better resolve ADCs and their heterogeneous fragments prior to MS.

Different liquid phase chromatography coupled with MS (LC-MS) techniques have been frequently used to characterize variants in ADCs.^11–14 Size exclusion chromatography (SEC)-MS is a common approach for evaluating size variants (e.g., aggregates, fragments) in ADCs.^11,15 While SEC is effective for separating ADC aggregates and monomers, only a relatively low resolution can be achieved when the technique is used for low molecular weight fragments (~several kDaltons) and hydrophobic interactions between ADCs and the column stationary phase can cause problems.¹¹ Alternatively, reverse-phase liquid chromatography (RPLC)-MS facilitates better assessment of mAb fragments and can potentially be applied to ADC samples.¹⁶ Hydrophobic interaction chromatography (HIC)-MS and RPLC-MS are popular for resolving the drug-to-antibody ratio (DAR) species and determining drug load distribution, as the number of drug loading on ADCs correlates with their hydrophobicity in most cases.^11,17 In addition, native SEC-MS was previously implemented for determining the DAR of various interchain cysteine-linked ADCs.¹⁸ Unlike HIC, SEC in the study functioned for buffer exchange instead of separating the DAR species. Moreover, for antibody-oligonucleotide conjugates (AOCs), a special type of ADC, resolving oligonucleotide-to-antibody ratio (OAR) species using the HIC or RPLC could be very challenging, due to their minor differences in hydrophobicity and secondary interactions with the column.¹⁹

Capillary electrophoresis (CE) is a complementary technique to LC for investigating the heterogeneity of biotherapeutics.^11,20,21 CE-sodium dodecyl sulfate (SDS) presents an extraordinary advantage over LC-based methods in the separation of size variants.¹¹ In addition, imaged capillary isoelectric focusing (icIEF) allows charge variants of mAbs and ADCs to be monitored with ultrahigh resolving power on isoelectric points.^20,21 These methods, however, cannot determine the accurate molecular weights of separated species due to MS-incompatible buffers used in separation. Recently, two-dimensional CE platforms have been designed by coupling CE-SDS/icIEF with CZE-MS to reduce the influence of interfering substances on MS analysis.^22–24 Alternatively, CE-SDS offline fractionation combined with RPLC-tandem mass spectrometry (MS/MS) offers another solution for the identification and quantification of various antibody fragments.²⁵ In addition, cIEF platforms have greatly advanced on separation buffers, automation, and sensitivity of CE-MS interface, enabling direct coupling of cIEF to MS for characterizing mAbs and cysteine-linked ADCs with variations on PTMs.^26–29 In addition to the CE techniques mentioned above, capillary zone electrophoresis (CZE) is also attractive due to its ability to differentiate biotherapeutics variants based on their charge-to-size ratios. Although CZE-MS was found to present a relatively lower separation resolution compared to cIEF-MS in our previous study,²⁹ it requires simpler sample preparation, easier separation operations and offers much higher sensitivity and throughput, making the technique well suited for routine monitoring of product variants and impurities. A variety of CZE-MS studies were performed previously to study charge variants with PTMs, structure changes (e.g., conformational isomers, aggregates), digestion fragments, and biotransformation products of mAbs.^30–36 To the best of our knowledge, only one work previously used CZE-MS in a microfluidic device for resolving the DAR species of a lysine-linked ADC.³⁷ The capability of CZE-MS for studying cysteine-engineered ADCs thus remains unexplored. CZE-MS is potentially useful for separating size or charge variations of cysteine-engineered ADCs. It is also a highly promising method to differentiate OAR species for cysteine-engineered AOCs, considering that the loading of negatively charged oligonucleotides significantly alters the net charge of the conjugates.

In this study, for the first time, we demonstrated the use of CZE-MS technique for better characterization of fragment product variants in cysteine-engineered ADCs and OARs of AOCs, which were previously considered highly challenging using LC-MS. A CZE-MS approach with high throughput (separation in 15 minutes) and improved separation resolution was developed using one tool cysteine-engineered ADC sample (ADC S1). To investigate the wider utility of the method, CZE-MS was further used to characterize fragment product variants of five different cysteine-engineered ADCs (ADC1a-c, ADC2, and ADC3) that were made from different parent mAbs and linker-payloads. To decipher the source of the product variants, CZE-MS analyses of the parent mAbs were also performed to compare differences with the results of the ADCs. Furthermore, we developed a novel CZE-MS approach by modifying sheath liquid, desolvation parameters, and background electrolyte (BGE) to facilitate the separation and characterization of OAR species of two cysteine-engineered AOCs, which were found to be challenging using HIC/RPLC-MS.

Results

CZE-MS setup and workflow

In this study, CZE-MS analyses of mAbs, ADCs, and AOCs were carried out on Agilent CE and Agilent time-of-flight (TOF) instruments equipped with a sheath-flow interface and using the workflow shown in Figure 1. Before analysis, the samples were buffer exchanged to the desired sample buffer for CZE separation using 10 kDa cutoff centrifugal filter units. The separation parameters and MS parameters were optimized accordingly for ADC/AOC samples and more details were disclosed in the following sections. The masses of product variants were determined by averaging across the selected peak followed by deconvolution. The species of product variants were estimated by comparing the detected masses with their theoretical masses.

Figure 1. Workflow of CZE-MS analysis of mAbs, ADCs, and AOCs.

CZE-MS analysis of mAbs, ADCs, and AOCs were performed on an Agilent CE-MS system equipped with a sheath-flow interface. The samples were buffer exchanged using 10kDa cut-off centrifugal filters. CZE enables rapid separation (~15 min) of impurities and variants of the samples based on their difference in charge-to-size ratio. For mass determination, the MS spectra of the selected peak were averaged followed by deconvolution.

Development and optimization of CZE-MS method for characterization of fragment product variants of cysteine-engineered ADCs

Cysteine-engineered ADCs are produced via the reduction of cysteines at specific sites of mAbs, followed by covalent drug conjugation. This type of ADC conserves antibody structure integrity, and therefore they are well compatible with denaturing conditions. For CZE-MS analysis, 5% acetic acid was used as BGE based on the previous optimization work with a standard protein mixture,³⁸ and 30 kV voltage was used to facilitate efficient separation. In addition, a sheath liquid containing 0.2% formic acid (FA) and 10% methanol (MeOH) was filled in the emitter to assist the ionization of the analytes migrated out of the capillary. We investigated the separation resolution and signal intensity of a tool ADC (ADC S1, ~2 mg/mL) with different sample buffer compositions [5% acetic acid (AA, pH ~ 2.2), 10 mM NH₄Ac (pH ~ 6.7), 10 mM NH₄HCO₃ (pH ~ 7.8)], sample buffer concentrations (10 mM, 50 mM, 100 mM NH₄HCO₃), and the sample injection volumes (10 nL, 25 nL, 50 nL, 100 nL). As shown in Figure S1, using salt buffer (NH₄Ac or NH₄HCO₃) as a sample buffer greatly benefited the separation of fragment product variants of ADC S1 by the sample preconcentration effect.³⁹ Further evaluation of different sample buffer concentrations revealed that using 10 mM and 50 mM NH₄HCO₃ slightly better separated peaks 2 and 3 than 100 mM NH₄HCO₃ (Figure S2). We prefer 50 mM NH₄HCO₃ to 10 mM NH₄HCO₃ in later experiments, as higher salt concentration presented higher signal intensity for fragment species (peaks 3–5), probably by enhanced sample preconcentration. Moreover, we carefully balanced the separation and protein signals with different sample injection volumes. Ideally, a smaller sample loading volume can effectively reduce the band broadening caused by sample charge repulsion, especially for large protein molecules. However, it also sacrifices the protein signal, leading to signal loss of very low abundant species. We found that decreasing the sample injection from 100 nL to 10 nL improved the separation of two intact ADC species with conformational heterogeneity (peak 1 and 1*), but the intensity of fragments (peak 2–5) also decreased around 10-fold (Figure S3). We considered 50 nL as a suitable sample injection volume since both a reasonable separation and peak intensities were observed under this condition.

Taken together, we chose 50 mM NH₄HCO₃ as sample buffer, 50 nL sample injection volume in the following CZE-MS analysis of mAbs, ADCs, and AOCs. CZE-MS characterization of the ADC S1 under this condition resolved six peaks in total (Figure 2). The species of peaks 1 and 1* presented the same deconvoluted masses, corresponding to intact DAR2 of ADC S1 and its glycoforms. However, the charge distribution of peak 1* significantly shifted to a high m/z range compared to peak 1, indicating that DAR2 in peak 1* has a more compact structure relative to peak 1. In addition, peak 2 showed a mixture of intact DAR2 and half mAb conjugated with one drug (half mAb S1_DAR1), whereas peaks 3, 4, and 5 were mainly low molecular weight fragments, including a light chain conjugated with one drug (Lc_DAR1, peak 3), as well as light chains with C-terminal cysteine clipping (peak 4), and a PTM (mass shift of 32 Da, probably a persulfidation or dioxidation on cysteine, peak 5).

Figure 2. CZE-MS analysis of ADC S1 using 50 mM NH₄HCO₃ as sample buffer and 50 nL sample injection volume. (a) Base peak electropherogram of ADC S1; (b) Averaged mass spectra and deconvoluted mass spectra (inserted figures) of peak 1–5 separated from ADC S1.

CZE-MS analysis of ADC S1 separated and identified six species from ADC S1, corresponding to different intact conformational isomers, half mAb and light chain-related fragments.

CZE-MS for tracking the nature of fragment product variants and to guide optimization of ADC molecules

The development of cysteine-engineered ADCs requires complicated processes, including mAb expression, purification, deblocking/decapping to remove cysteinylation and glutathionylation from the culture medium, and drug payload conjugation.⁴⁰ It is important to investigate whether any of these steps introduces product heterogenicity and instability. Herein, we applied the optimized CZE-MS method to characterize five different cysteine-engineered ADCs (ADC1a-c, ADC2, ADC3) and their parent mAbs (mAb1–3). More detailed information (parent mAbs, linkers, and payloads) about ADCs was disclosed in Table S1 (supporting information). A comparison of fragment product variants between ADCs and mAbs could potentially allow optimization of conjugation site, the decapping process, and the chemistry of the conjugation in ADC molecules.

ADC1a-c share the same parent mAb (mAb1, engineered with two cysteines, deblocked) but were conjugated with different non-charged drug molecules (Table S1). CZE-MS characterization of mAb1 determined a mixture of intact mAb1 and half mAb1 in peak 1, a mixture of conformational isomers of intact mAb and half mAb in peak 1* (shown as the shoulder peak of peak 1), and a mixture of light chain and low abundant light chain with C-terminal cysteine clipping in peak 2 (Figure 3a,b). In contrast, the result of ADC 1a (Figure 3c,d) showed a mixture of intact ADC (DAR2 as a major DAR species) and half mAb1_DAR1 (peak 1), the conformational isomers of the species in peak1 (peak 1*), a light chain with C-terminal cysteine clipping (peak 2), and an Lc_DAR1 (peak 3). Similar intact and fragment species had been found in the results of ADC 1b and ADC 1c (Figure S4). The mAb1 in Figure 3a was “deblocked” by reducing the disulfide bonds, followed by oxidating endogenous disulfide to generate free thiol groups on the engineered cysteine for site-specific conjugation. Therefore, the half mAb presented in the mAb1 sample is most likely produced from the incomplete reoxidation process. By further comparing the results of mAb1 and ADC1, we can conclude that all fragment species (half mAb1_DAR1, light chain with C-terminal cysteine clipping, and Lc_DAR1) in ADC1a-c originated from the fragments (half_mAb1, C-terminal cysteine clipping, and light chain) of parent mAb1 sample. Interestingly, we found that the light chains with C-terminal cysteine clipping were not shown as drug-conjugated form in ADC1a-c samples (e.g., peak 2 in Figure 3c), but the intact light chains were (e.g., peak 3 in Figure 3c), which indicates that the drug conjugations of the free light chains likely occur on the C-terminal cysteine.

Figure 3. CZE-MS analysis of parent mAb1 and ADC1a. (a) Base peak electropherogram of mAb1; (b) Averaged mass spectra and deconvoluted mass spectra (inserted figures) of peaks 1, 1*, and 2 separated from mAb1; (c) Base peak electropherogram of ADC1; (d) Averaged mass spectra and deconvoluted mass spectra (inserted figures) of peaks 1, 1*, 2, 3 separated from ADC1.

The CZE-MS results of parent mAb1 and ADC1a were compared to determine any differences in impurities and identify the possible source of product variants in ADC1a.

ADC2 was developed from site-specific conjugation of four non-charged drugs on mAb 2 (engineered with four cysteines, deblocked, Table S1). As shown in Figure 4, CZE-MS analysis of mAb2 resolved an intact mAb2 (peak 1), a mixture of conformational isomers of intact mAb2 and half mAb2 (peak 1*), a half mAb2 (peak 2), and an light chain (peak 3), whereas ADC2 presented an intact ADC (DAR4 as a major DAR species, peak 1), a mixture of conformation isomers of intact DAR4 and half mAb2_DAR2 (peak 1*), a half mAb2_DAR2 (peak 2), a mixture of half mAb2_DAR1 and Lc_DAR1 with a PTM (~32 Da mass shift, potentially a persulfidation/dioxidation) (peak 3), and an HC S222C223 clipping (peak 4). Interestingly, the free light chain carrying the PTM (persulfidation/dioxidation) in peak 3 was conjugated with one drug, not two drugs. The observation provided further evidence that the PTM was on the cysteine that was blocked from conjugation. In addition, potential HC S222C223 clipping appears at the hinge region. Similar clipping between serine and cysteine at the HC was reported in another study.²⁵ A comparison of the results of mAb2 and ADC2 revealed that the half mAb_DAR2, half mAb_DAR1, Lc_DAR1 were generated from incomplete reoxidation in the mAb deblocking process followed by drug conjugation. However, the HC S222C223 clipping and the potential PTM (persulfidation/dioxidation) on Lc_DAR1 were present in ADC2, but not mAb2, suggesting the two impurities were induced during or after the drug conjugation process. All the drug-conjugated species in ADC2 were found to contain additional mass shifts associated with the hydrolysis of the succinimide ring of maleimide linker (18 Da mass shift from each hydrolysis). The maleimide hydrolysis is common to see in cysteine-engineered ADCs and can promote the stability of the drug conjugation and ADCs.^4,6 The susceptibility of hydrolysis can be influenced by the structure of linkers and pH conditions during conjugation and storage.^41,42 In our study, hydrolysis was observed in ADC2 but not ADC1a-c, due to the different linker structure and additional post-conjugation treatment for ADC2. ADC1a-c and ADC2 used maleimidocaproyl (MC) linker and 4-(maleimidylmethyl) cyclohexane-1-carboxyl (MCC) linker for drug conjugation, respectively. The ADC2 consisting of MCC linker is typically unstable and prone to drug deconjugation. A similar instability concern about MCC-based ADC was reported in another study previously.⁴³ To improve the stability, ADC2 was incubated at pH 9 under 37°C for two days to promote the maleimide ring opening of MCC linker. This process could be a potential source of other variants, such as potential HC S222C223 clipping, because of high pH stress.

Figure 4. CZE-MS analysis of parent mAb2 and ADC2. (a) Base peak electropherogram of mAb2; (b) Averaged mass spectra and deconvoluted mass spectra (inserted figures) of peak 1, 1*, 2, 3 separated from mAb2; (c) Base peak electropherogram of ADC2; (b) Averaged mass spectra and deconvoluted mass spectra (inserted figures) of peak 1, 1*, 2, 3, 4 separated from ADC2.

The CZE-MS results of parent mAb2 and ADC2 were compared to determine any differences in impurities and identify the possible source of product variants in ADC2.

Moreover, we were interested in ADC3, which was conjugated with four of the same non-charged drugs as ADC2 but on a different parent mAb (mAb3, engineered with four cysteines, deblocked, Table S1). The mAb3 was modified from mAb1 by engineering two additional cysteines. In CZE-MS analysis, mAb3 showed similar separation profiles as mAb1, presenting three separated peaks, including a mixture of intact mAb3 and half mAb3 (peak 1), a mixture of conformational isomers of mAb3 and half mAb3 (peak 1*), and a light chain (peak 2) (Figure 5a,b). The CZE-MS analysis of ADC3 detected a mixture of intact ADC (DAR 4 as a major DAR species) and half mAb3_DAR2 (peak 1), Lc_DAR2 (peak 2), and potential HC S222C223 clipping (peak 3) (Figure 5c,d). Based on the results, we speculate that half mAb3_DAR2 and Lc_DAR2 were originally from half mAb and light chain produced in the mAb deblocking process. Peptide mapping analysis (Figure S5 and S6) also suggested that Lc_DAR2 was produced from an expected engineered cysteine conjugation and an unexpected free cysteine conjugation from an intrachain disulfide bond introduced during reduction/reoxidation step (see Materials and Method). In addition, all the drug-conjugated species in ADC3 were presented with hydrolysis (+18 Da mass shift), due to the same post-conjugation treatment for maleimide ring opening as ADC2. Potential HC S222C223 clipping was also observed in ADC3, which could be introduced during post-conjugation treatment.

Figure 5. CZE-MS analysis of parent mAb3 and ADC3. (a) Base peak electropherogram of mAb3; (b) Averaged mass spectra and deconvoluted mass spectra (inserted figures) of peak 1, 1*, 2 separated from mAb3; (b) Base peak electropherogram of ADC3; (b) Averaged mass spectra and deconvoluted mass spectra (inserted figures) of peak 1, 2, 3 separated from ADC3.

The CZE-MS results of parent mAb3 and ADC3 were compared to determine any differences in impurities and identify the possible source of product variants in ADC3.

RPLC-MS is a classic and widely used technique for characterizing product variants of ADCs. To have a better understanding of the features of CZE-MS for studying biotherapeutics, we carried out RPLC-MS analysis of mAbs and ADCs studied above and compared the results with CZE-MS. The RPLC with a fast gradient (10 min) showed limited separation resolution for fragment product variants (Figure S7-S9). For example, RPLC-MS analysis of ADC1a-c showed only two separated peaks, corresponding to a mixture of intact ADC (DAR 2 as a major DAR species) and half mAb_DAR1 and a light chain (Figure S7). The low abundant light chain with C-terminal cysteine clipping showing in CZE-MS was not found in RPLC-MS, most likely due to signal suppression by coelution or sample loss in the RPLC column. In addition, RPLC-MS analysis of ADC2 detected only one peak associated with coeluted intact DAR4 and half mAb_DAR2 (Figure S8). Other variants discovered by CZE-MS, such as half mAb_DAR1, Lc_DAR1 with a PTM (persulfidation/dioxidation), and HC S222C223 clipping, were missing in RPLC-MS data. To have a fair comparison, we also selected one sample (ADC1a) and performed RPLC separation with an extended gradient (30 minutes) to compare the separation with CZE-MS (Figure S10). Despite the improved separation for intact DAR2 and light chain, the separation resolution of RPLC remains two times lower than CZE (1.5 vs. 3.2) based on the resolution equation (Equation 1, supporting information). Our results highlight the potential of CZE-MS to serve as a complementary approach for better characterization of fragment-related impurities.

CZE-MS for characterization and quantification of OAR species of AOCs

Composed of mAbs conjugated to oligonucleotides such as small interfering RNA (siRNA), and anti-sense oligonucleotides,^19,44–47 AOCs have been reported as promising gene-silencing cancer therapeutics and enhancing reagents in radiotherapy.^44,45 While most AOC drugs are in the preclinical stages, clinical trials for several (DYNE-101, DYNE-251, AOC 1001, AOC 1044, AOC 1020) have recently started. The analyses of AOCs also face challenges, particularly for OAR species characterization.^19,47 The conventional chromatographic methods developed for resolving DAR variants of ADCs based on hydrophobicity, such as HIC and RPLC, do not apply to AOCs.¹⁹ Most recently, native SEC was coupled with Fourier transform MS to characterize AOCs, but the direct resolution of DAR variants is difficult with this method.⁴⁷ Despite the challenges, the OAR species of AOCs are potentially resolvable based on their heterogeneity in charge, given that oligonucleotides carry various negative charges and can significantly modify the net charge of AOCs. Previously, a study successfully applied anion-exchange chromatography to separate the DAR of AOCs.⁴⁴ However, the method relied on a high concentration of sodium chloride as a mobile phase for separation and cannot be directly coupled to MS for mass characterization of individual species. Other than ion exchange chromatography, CZE can separate AOC variants with charge heterogeneity. Herein, we purpose a CZE-MS method for resolving and characterizing OARs of cysteine-engineered AOCs.

When the same CZE-MS method for ADCs was applied to study a cysteine-engineered AOC1 sample spiked with 20% of unconjugated mAb, no AOC1 signal was observed. We speculate that the ionization efficiency of AOCs is much lower than standard ADCs because fewer basic residues are available to capture protons during electrospray ionization (ESI). To address this issue, we modified the sheath liquid from 0.2% FA and 10% MeOH to 25% acetonitrile (ACN) and 20% AA and adjusted the drying gas flow rate from 1 L/min to 2 L/min to improve ionization and desolvation. Finally, we obtained a clear signal of AOC1 (Figure 6). However, the use of the previous separation method for ADCs did not resolve the OAR species of AOC1 (Figure 6a,b). It is known that the pH of BGE can affect separation by altering the structure and the surface charge of biomolecules. We investigated the separation of the AOC1-mAb mixture using BGEs with different concentrations of AA [20% (pH 1.9), 10% (pH 2.1), 5% (pH 2.3), and 1% (pH 2.7)]. We noticed that lowering the concentration of AA resulted in a shorter migration time and a narrower separation peak due to the reduced viscosity of BGE and less protein diffusion (Figure 6a). When the AA concentration was decreased to 1%, the OAR0 and OAR2 were nearly baseline resolved in 15 minutes. The mass spectra of OAR0 (parent mAb) and OAR2 showed different charge state distributions under this condition, suggesting significant net charge variations induced by oligonucleotide loading (Figure 6c). We also tested an even lower AA concentration (0.5% AA, pH 2.9), but observed severe peak tailing (data not shown). Based on the results, we consider 1% AA to be the ideal BGE condition for separating OAR species of AOCs. We evaluated the relative abundance of OAR species by integrating and comparing the peak areas of the electropherogram. Our quantification result showed that DAR0 accounts for 23% of the abundance of OAR2, which is very close to our theoretical abundance of 20%.

Figure 6. Optimization of background electrolyte (BGE) in CZE-MS analysis to separate OAR species of AOC1 sample. (a) Base peak electropherogram of AOC1 incorporated with 20% of parent mAb under 20% AA, 10%AA, 5% AA, and 1% AA conditions. (b) Averaged mass spectra and deconvoluted mass spectra (inserted figures) of the single peak (a mixture of OAR0 and OAR2) under 5% AA condition. (c) Averaged mass spectra and deconvoluted mass spectra (inserted figures) of the OAR0 (peak 1) and OAR1 (peak 2) separated from the AOC1-mAb mixture under 1% AA condition.

The BGE for CZE-MS was optimized to achieve baseline separation of OAR species of AOC1. The OAR0 and OAR2 have a significant difference in their charge distributions due to different numbers of ONs loading.

To examine the suitability of CZE-MS for analyses of different AOC samples, we further applied the optimized MS and separation conditions for characterizing cysteine-engineered AOC2. Other than an OAR2, the AOC2 sample intrinsically contains very low abundant OAR1 (Figure S11). The two OAR species coeluted in RPLC-MS (Figure S11), but they could be resolved by our CZE-MS method (Figure S12). Furthermore, we incorporated 1% of parent mAb into the AOC2 sample to create a more complex mixture that contains OAR0 (parent mAb), OAR1, and OAR2. Our CZE-MS method still exhibited good separation for these OAR species (Figure 7a). Similar to AOC1-mAb, MS characterization of the AOC2-mAb mixture showed significantly different charge distributions among OAR0, OAR1, and OAR2 (Figure 7b,c).

Figure 7. CZE-MS analysis of OAR species of AOC2 sample using 1% AA as BGE. (a) Base peak electropherogram of AOC2 (a cysteine-engineered AOC) incorporated with 1% parent mAb. (b) Averaged mass spectra and deconvoluted mass spectra (inserted figures) of the OAR0 (peak 1), OAR1(peak 2), and OAR2 (peak 3) separated from the AOC2-mAb mixture.

CZE-MS analysis of the AOC2 incorporated with 1% parent mAb achieved separation of OAR0, OAR1, and OAR2. Different charge distribution of the three OAR species was shown in the mass spectra.

The quantification of OAR species revealed that the abundance of DAR0 and OAR2 are 2.3% and 3.1% relative to OAR2. The relative abundance of OAR0 has a deviation compared to the theoretical abundance of 1%, but the result is still in an acceptable range. HIC-MS is a gold standard approach for evaluating DAR species of ADCs. Thus, we performed HIC-MS analysis on AOC2 and an AOC2-mAb mixture for comparison. However, as shown in Figure S13, the OAR variants were not separated by HIC-MS and severe potential secondary interaction was observed in both AOC2 and the AOC2-mAb mixture. Our study of different AOC-mAb mixtures further demonstrated that CZE-MS is a simple, rapid, and high-resolution technique for the characterization and quantification of OARs of AOCs.

Investigation of the factors that influence electrophoretic mobility of ADCs and their product variants

Electrophoretic mobility dominates the migration of biomolecules in CZE separation in a neutrally coated capillary and is correlated with their charge-to-size ratios. We compared the migration time and separation profiles across different intact mAbs, ADCs, and their impurities to learn how the mAbs and drug loadings affect their electrophoretic mobility. We noted that the migration time of the main peaks (peak 1) of different ADCs are similar, suggesting that they have close charge-to-size ratios under our separation condition. Comparison of the migration time between ADCs and their parent mAbs showed that the conjugation of different linker-payloads (neutral, MWs: 600 ~ 1400 Da) also had limited alternation on the mobility of the intact biomolecules. In addition, conformational isomers (peak 1 and peak 1*) were widely observed in mAbs and ADCs and they showed different mobility in CZE separation. Conjugation of a high number of linker payloads to mAbs can modify the relative abundance of the two conformational isomers of ADCs. For example, the ADC2 contains four neutral linker payloads. The conformation in peak 1 was shown as major species in its parent mAb (mAb2), whereas the conformational isomer in peak 1* was presented with higher abundance in ADC2 (Figure 4a vs. Figure 4d). While peak 1* presented a charge state distribution at higher m/z than that in peak 1, ADC2 favors a more compact structure than its parent mAb, which could be due to the hydrophobic effects from the high number of drug loadings. Moreover, for AOC samples, oligonucleotides have a large molecular weight (~11kDa) and are negatively charged. More oligonucleotide loadings can contribute to slower mobility by decreasing the surface charge and increasing the size of the AOCs. As expected, CZE-MS analysis of the AOC2-mAb mixture under 1% AA BGE conditions showed that mobilities of OAR species were OAR0>OAR1>OAR2 (Figure 7).

Compared to intact ADCs, the electrophoretic mobilities of fragments were more extensively affected by the properties of mAb subunits/fragments and drug loadings. For example, ADC2 and ADC3 have four of the same drugs conjugated to mAb2 and mAb3, respectively. While the half mAb3_DAR2 was coeluted with DAR4 of ADC3, the half mAb2_DAR2 was capable of being separated from the intact DAR4 of ADC2 (Figure 5c vs. Figure 4c). This could be associated with different charge and size properties of the half mAb2 and the half mAb3. Moreover, drug loading impacts the mobility of fragments differently. For the ADC1 sample, the Lc_DAR1 showed slower mobility than the light chain only (Figure 3c vs. Figure 3a). As the linker-payload in ADC1 is neutral, the conjugation of one linker-payload on the light chain most likely reduces the charge-to-size ratio by increasing the size. However, in ADC2, loading two drugs on half mAb2 exhibited faster mobility than the half mAb conjugated with only one drug (Figure 4, peak 2 vs. peak 3). The more drug loading is more likely to cause compaction of half mAb structure through hydrophobic effects, leading to an increased charge-to-size ratio.

Discussion

Here, we report the development and application of CZE-MS methods for characterizing the sample heterogeneity of a variety of cysteine-engineered ADCs and AOCs. The CZE-MS system was constructed based on a sheath flow nano-spray interface, and the separation was largely optimized with a model cysteine-engineered ADC (ADC S1). We achieved resolution of six peaks from ADC S1 using CZE separation, representing conformational isomers of the intact ADC and different half mAb and light chain-related fragments. The method was used to study five additional cysteine-engineered ADCs (ADC1a-c, ADC2, and ADC3) and revealed various undesired fragment product impurities and variants, such as half mAbs with DAR1/DAR2, light chains with DAR1/DAR2, Lc/Lc_DAR1 with a potential PTM (persulfidation or deoxidation), light chains with C-terminal cysteine clipping, and HCs with S222C223 clipping (Table S2, supporting information). Peptide mapping was performed to verify the fragments in ADCs. We were able to identify two drug conjugation sites of Lc_DAR2 in ADC2 sample. However, peptide identification of other fragment species remains challenged by their low abundance, sharing sequence with intact ADCs, or low retention on RP column due to short peptide length after various enzymatic digestion.

Further comparison of fragment product variants between ADCs and their parent mAbs could be used to decipher the potential processing steps that introduced variations, which can guide ADC production quality control. Free light chains and Lc_DAR1/DAR2 were found in the parent mAbs and ADCs samples, revealing incomplete reoxidation after mAb deblocking. The data strongly suggests improvement is required for mAb conjugation procedure.

Moreover, we hypothesized that CZE-MS is well suited for separating OARs of cysteine-engineered AOC samples since the conjugation of oligonucleotides significantly influences the electrophoretic mobility of AOCs by changing their overall charges. For the first time, we report a novel CZE-MS method for the characterization and quantification of OAR species of AOC samples. In particular, we carefully modified the sheath buffer and drying gas conditions, which greatly benefited the ionization and desolvation of AOCs. Optimization of BGE conditions found the alteration of the pH of BGE influenced the separation of OARs by altering their surface charges and structures. Using 1% AA, nearly baseline separation of the OAR species of AOCs was achieved.

Finally, to investigate the factors that influence the electrophoretic mobility of ADCs/AOCs and their product variants, we compared migration time and separation profiles among parent mAbs, ADCs, and AOCs. Our result demonstrated that the structure and properties of mAbs and linker-payloads significantly influence the separation of the product variants by affecting their sizes or charges.

In general, our CZE-MS system showed higher separation resolution for fragment product variants of ADCs than RPLC-MS and also outperformed both HIC-MS and RPLC-MS for OARs analysis of AOCs. Our study highlights CZE-MS as a rapid and high-resolution analytical tool for monitoring the heterogeneity of cysteine-engineered ADCs and AOCs. It is crucial for the pharmaceutical industry to understand the capabilities of CZE-MS and its potential to solve sample characterization issues that are not addressed by traditional LC-MS. Our techniques can also be potentially applied to the broad characterization of similar biotherapeutics. It is important to note two additional discrepancies between the CZE-MS and RPLC-MS results. First, while Lc_DAR1/DAR2 were widely found in CZE-MS analysis of ADCs, RPLC-MS only detected free light chain that has no drug conjugation in ADC1a-c and ADC3 samples. Theoretically, the cysteine thiols of the free light chain are also reactive with maleimide-based linker-payload and free light chains were supposed to present in drug-conjugated form. The free light chain was determined as an unconjugated form in RPLC-MS analysis, probably because Lc_DAR1/DAR2 are inherently unstable and underwent deconjugation under a high RPLC column temperature condition or due to other factors. This discrepancy suggests that CZE-MS can reflect the impurities in ADCs more accurately than RPLC-MS. Second, when comparing the abundance of intact ADCs and the half mAb_DAR1/half mAb_DAR2 in mass spectra, intact ADCs showed much higher relative abundance in RPLC-MS than that in CZE-MS (e.g., Figure 3d vs. Figure S5). This is likely because the RPLC-MS and CZE-MS were performed on different mass spectrometers, 6545XT Q-TOF and 6230 TOF, respectively. The former MS instrument favors better transmission for large biomolecules, and therefore presents higher relative abundance for intact ADCs. CZE-MS analysis on high-end mass spectrometers is a better solution and will be used in our future studies to fix this bias.

It is also worth noting that CZE-MS methods still can be improved in the following aspects. First, both CZE-MS and RPLC-MS remain limited in resolving intact DAR species of ADCs because of the minor impact on the mobility of ADCs by the neutral and small payloads. For example, as shown in Figure S14, other than the major DAR4, several other DAR species (DAR2 and DAR3) of ADC3 were detected in the same mass spectra. The CZE separation resolution for DAR species can be further boosted by using a longer capillary, but this approach has a potential drawback because a longer capillary can sacrifice the throughput. Second, while TOF is used for mass characterization in this study, we expect that coupling CE to an advanced mass spectrometer with better transmission and higher resolution can greatly improve the quality of the characterization results (e.g., higher intensity and mass accuracy). Third, since the impurities in ADCs in our study were determined based on intact mass information, incorporating fragmentation techniques (AI-ETD, EThcD)^48,49 can facilitate direct sequence analysis and more confident identification of potential PTMs, which benefits confirmation of product liabilities, developability, and stability challenges in the early stages of biologics discovery. We expect this novel method to be transferable to development functions for the identification of ADC and AOC product variants and impurities and for the identification of product critical quality attributes in development. Fourth, although the CE-MS was used for the quantification of OARs of AOCs in this work, direct quantification of ADC fragments remains challenging because of the different ionization efficiency among size variants. The development of imaging CE-MS could potentially provide more accurate quantification of fragments through optical detection.

Materials and methods

Reagents

Ammonium acetate (NH₄Ac), ammonium bicarbonate (NH₄HCO₃), ammonium persulfate, Amicon Ultra centrifugal filter units (0.5 mL, 10 kDa cutoff size) were bought from Sigma-Aldrich (St. Louis, MO). Acetic acid (AA), FA, ACN, MeOH, and water are LC-MS grade and were purchased from Fisher Scientific (Pittsburgh, PA). Fused silica capillaries (50 mm i.d., 360 mm o.d.) were obtained from Polymicro Technologies (Phoenix, AZ). Acrylamide was purchased from Acros Organics (Fair Lawn, NJ).

Antibody production and ADC generation

mAbs containing one or more engineered cysteine were generated from transiently transfected Chinese hamster ovary (CHO)-Expi cells or from CHO pools. Antibodies were purified from culture supernatant by affinity chromatography using a Protein A coupled resin (MabSelect™SuRe™ (GE Healthcare Life Sciences, Pittsburgh PA)) followed by potential polishing steps to achieve > 95% monomer by UP-SEC prior to conjugation. The purified antibodies containing engineered cysteine were further deblocked by incubating with 40 equivalent dithiothreitol reduction for 40 minutes at room temperature, and 10 equivalent dehydroascorbic acid (DHAA) reoxidation for 1 h at room temperature to remove the cysteinylation/glutathionylation and conjugated with maleimide linker-payloads for ADC and AOC generation. Additional post-conjugation treatments were applied to ADC2 and ADC3 to improve their stability. ADC2 and ADC3 were incubated at pH 9 under 37°C for two days to promote the maleimide ring opening.

Sample preparation

100 μg each of mAb, ADC, and AOC samples were buffer exchanged to 50 mM NH₄HCO₃ by Amicon Ultra centrifugal filters according to the manufacturer’s instructions. The final concentration of the samples was around 2 mg/mL.

CZE-MS analysis

An Agilent 7100 CE system (Santa Clara, CA) was online coupled with an Agilent 6230 TOF mass spectrometer (Santa Clara, CA) via an electrokinetically pumped sheath-flow nanospray interface (EMASS-II CE−MS interface, CMP Scientific, Brooklyn, NY) for automated CZE-MS analysis (Figure 1). A 75 cm linear polyacrylamide (LPA)-coated capillary (50 µm i.d., 360 µm o.d.) was used for CZE separation. The LPA coating of the capillary was prepared using a recently optimized protocol.²⁷ The inlet and outlet of the capillary were fixed in the CE system and ESI emitter (25–35 µm orifice size) of the interface, respectively. The ESI emitter was positioned 4–6 mm away from the MS inlet. For the sheath liquids in the emitter, a solution of 0.2% FA and 10% MeOH was prepared for mAb and ADC analysis, and a solution of 25% ACN and 20% AA was applied for AOC analysis. The voltage of the power supply was 2.3 ~ 2.5 kV. The position and voltage of the power supply were slightly adjusted to achieve a stable electrospray.

To carry out CZE separation, the capillary was sequentially loaded with 50 nL of the sample (100 mbar pressure for 23 seconds), inserted into a background electrolyte (BGE), then applied with 30kV voltage. For mAb and ADC analysis, 5% acetic acid (AA) was used as BGE. For AOC analysis, the BGE was optimized to 1% acetic acid for better separation of DAR species.

The full scan mass acquisition was performed on the TOF mass spectrometer. A regular spray shield was installed on the MS inlet. The voltages for VCap, skimmer, and fragmentor were 0 V, 300 V, 380 V, respectively, based on the optimized conditions in previous works.²⁷ For MS analysis of mAb and ADC, the drying gas temperature and the flow rate were set to 365°C and 1 L/min, respectively. The full scan acquisition range was 1500 ~ 5000 m/z, and the scan rate was 0.5 spectrum/s. For analysis of AOC, the drying gas was adjusted to 2 L/min for achieving better desolvation. The mass spectra were collected at 1500 ~ 3200 m/z with a scan rate of 0.7 spectrum/s.

LC-MS analysis

RPLC-MS analysis of mAb, ADC, and AOC samples were performed on an Agilent LC system integrated with an Agilent 6545XT Q-TOF mass spectrometer. The samples were diluted to 0.2 mg/mL and 5 µL of the sample was injected into an RPLC column (PLRP-S, 2.1 mm i.d. ×50 mm length, 5 µm particle size, 1000 Å pore size). The column temperature was set to 80°C. Mobile phase A was 0.1% FA in water and mobile phase B was 0.1% FA in ACN. For rapid RPLC separation (10 minutes), the gradient was maintained at 20% B for 2 minutes, then changed from 20% B to 75% B in 6 minutes, and finally stayed at 75% B for 2 minutes, with a flow rate of 0.5 mL/min. For RPLC with an extended gradient (30 minutes), the setting was 0% B to 10% B in 2 minutes, then increasing to 60% B in 20 minutes, 80% B in 3 minutes, and 90% B in 1 minute. The gradient was eventually maintained at 90% B for 6 minutes. The flow rate was 0.3 mL/min. The voltages for VCap, skimmer, and fragmentor of the Q-TOF were set to 0 V, 300 V, 380 V, respectively. The drying gas temperature was maintained at 365°C and the flow rate was 1 L/min. A 5 V collision-induced energy (CID) was added to benefit the transmission of large biomolecules in QTOF. Full-scan MS data were acquired across an 800–7000 m/z range, with a scan rate of 1 spectrum/s.

HIC-MS analysis was performed on an Agilent 1290 two-dimensional (2D) LC-MS system. 20 µg of AOC-mAb mixture samples were loaded onto HIC column (MAbPAC HIC-10, 4.6 mm i.d. ×100 mm length, 5 µm particle size, 1000 Å pore size) in the first dimension. Mobile phase A was 2.5 M of ammonium acetate and 0.1 M phosphate buffer (Na₂HPO₄) with pH of 7.0. Mobile phase B was 0.1 M phosphate buffer (Na₂HPO₄ and NaH₂PO_4, pH 7.0). The gradient was changed from 10% B to 100% B in 15 minutes and then maintained at 100% B for 10 minutes before switching back to 10% B for equilibrium until 35 minutes at the flow rate of 0.5 mL/min. The second dimension includes an RPLC column (Agilent PLRP-S, 2.1 mm i.d. ×50 mm length, 5 µm particle size, 1000 Å pore size) for desalting the sample eluted from first-dimensional HIC. Mobile phase A was 0.1% FA in water and mobile phase B was 0.1% FA in ACN. The gradient and MS settings remained the same as the rapid RPLC method described above.

Peptide mapping

ADCs (1 mg/mL) were denatured by mixing with 8 M Guanidine HCl/Tris buffer, followed by reduction and alkylation with DTT and IAM, respectively. The samples were then buffer exchanged into 10 mM Tris-HCl (pH 7.5) using Biospin 6 cartridges (Bio-Rad). Afterward, an enzyme for digestion (Trypsin, Glu-C, or Asp-N) was added to the sample with a ratio of 1:20 (w/w) and the mixture was incubated at 37°C overnight. Peptide mapping was performed on a Thermo Ultimate 3000 UHPLC coupled to a Thermo Q-Exactive HF-X mass spectrometer. Separation was carried out using a Waters CSH C18 column (2.1 mm i.d. ×150 mm length, 1.7 µm particle size, 130 Å pore size). Mobile phase A was 0.1% FA in water and mobile phase B was 0.1% FA in ACN. Peptides were eluted with a gradient from 2% B to 45% B in 50 minutes followed by washing with 90% B for 5 minutes. For the mass spectrometer, full MS scan settings were resolution of 60,000, AGC target of 3e6, and scan range of m/z 250 to 2000. The top 10 most abundant peaks (charge state of 2 to 7) were isolated with a window of m/z 1.2) for further fragmentation. MS/MS spectra were acquired with a resolution of 15,000, AGC target of 1e5, and maximum injection time of 100 ms. Dynamic exclusion was set to 6s.

Data analysis

Analyses of electropherograms, chromatograms, and mass spectra were performed using Agilent Mass-Hunter Qualitative Navigator. The devaluation of mass spectra was performed by averaging across a single peak in an electropherogram and processing by Agilent Mass-Hunter BioConfirm with the Maximum Entropy algorithm. In deconvolution settings, the mass step was set to 0.1 Da, and the mass range was adjusted accordingly for different analytes. Other parameters were maintained as default.

Abbreviations

AA	=	Acetic acid
ADC	=	Antibody-drug conjugate
AOC	=	Antibody-oligonucleotide conjugate
BGE	=	Background electrolyte
CE-SDS	=	Capillary electrophoresis-sodium dodecyl sulfate
CZE-MS	=	Capillary zone electrophoresis-mass spectrometry
DAR	=	Drug-to-antibody ratio
ESI	=	Electrospray ionization
HC	=	Heavy chain
HIC	=	Hydrophobic interaction chromatography
icIEF	=	Imaged capillary isoelectric focusing
Lc	=	Light chain
Lc_DAR1/DAR2	=	Light chain conjugated with one/two drug
LC-MS	=	Liquid phase chromatography-mass spectrometry
mAb	=	Monoclonal antibody
OAR	=	Oligonucleotide-to-antibody ratio
ON	=	Oligonucleotide
pCQA	=	Product critical quality attribute
PTM	=	Post-translational modification
RPLC	=	Reverse-phase liquid chromatography
SEC	=	Size exclusion chromatography

Acknowledgments

The authors would like to acknowledge the contributions of all the members of the Merck & Co., Inc., South San Francisco, CA, USA Protein Sciences Department within Discovery Biologics (protein expression, protein purification, and characterization groups) as well as the Discovery Chemistry group in South San Francisco for ADCs and AOCs conjugation. The authors also acknowledge Dr. Liangliang Sun at Michigan State University for support on method development.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/19420862.2023.2229102

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

The author(s) reported there is no funding associated with the work featured in this article.