Biosimilar or Not: Physicochemical and Biological Characterization of MabThera and Its Two Biosimilar Candidates

Abstract

The development of therapeutic biosimilar antibodies has become an important driving force of the modern biopharmaceutical industry. In this study, physiochemical characteristics (amino acid sequence, intact/subunit molecular weight, isoelectric point, post-translation modification, and disulfide linkage pattern), purity (charge variants, high and low molecular weight variants), antigen binding activity, Fc receptor binding affinity and Fc-effector function (CDC and ADCC) were analyzed by using an extensive set of state-of-the-art and orthogonal analytical technologies to provide a comprehensive characterization of the innovative product rituximab and two biosimilar candidates. The similarity study showed that biosimilar candidate 1 (BC1) and the reference product (RP) MabThera had an identical protein amino acid sequences and highly similar primary structures along with similar purity, heterogeneity profiles, antigen binding activity, Fc receptor binding affinity, and Fc-effector functions. Biosimilar candidate 2 (BC2), which had an amino acid replacement at a constant region, a different N-glycosylation profiling, and purity, was not analytically similar to RP. Although BC2 showed improvement such as an increased level of afucose, another IgG1 allotype, and similar biological activities, it was not recommended to be applied as a biosimilar compound in drug registration because the biosimilar manufacturer must first show that its primary structure was identical to that of RP. Our physicochemical characterizations and bioassay comparability study provided a deepened understanding of the structure–function relationship of quality attributes.

1. Introduction

Monoclonal antibodies (mAbs) have become an important class of therapeutic agents for numerous indications, including cancers and infectious and autoimmune diseases. However, mAbs are very costly, with annual treatments costing up to tens of thousands of dollars. Given that the patents of some best-selling mAbs (adalimumab, rituximab, bevacizumab, trastuzumab, and infliximab) have expired, biosimilars, which are designed as biologically similar alternatives that are similar in quality, safety, and efficacy to the approved reference products (RPs), have led to significant expansion of biopharmaceutical industry.¹ The appearance of biosimilars has resulted in affordable biological treatments and thus increase patient access to otherwise expensive therapies.²

Rituximab is a chimeric human–murine IgG1 mAb targeting CD20, a surface marker that is highly expressed on most malignant B cells.³ Anti-CD20 antibodies are categorized as Type I or II in accordance with their mode of CD20 binding and their primary mechanism for killing CD20-positive cells. The primary mechanism of action (MOA) of rituximab, which is classified as an anti-CD20 IgG1 type I antibody, is based on its capacity to bind CD20-positive human cells to induce cell death via antibody-dependent cell-mediated cytotoxicity (ADCC), complement-dependent cytotoxicity (CDC), and apoptosis.⁴ Rituximab is currently indicated for the treatment of patients with various autoimmune disorders such as rheumatoid arthritis, Wegener’s granulomatosis, and microscopic polyangiitis.⁵ The patent of innovator rituximab, which is sold as MabThera (EU) and Rituxan (USA), has expired in Europe and the USA. As a consequence, the development of its biosimilar has substantially expanded in China, Europe, South Korea, and other countries.⁶⁻⁸ At present, the European Medicines Agency (EMA) and US Food and Drug Administration (FDA) have approved four rituximab biosimilars, which are produced by Celltrion Healthcare, Sandoz, and Pfizer. The biosimilar of MabThera, HLX01, which was developed by Shanghai Henlius Biotech, is the first MabThera biosimilar approved by China’s National Medical Product Administration (NMPA).⁹ A total of 18 and 35 proposed rituximab biosimilars remain in the clinical development phase in China and the world, respectively. The Center for Drug Evaluation (CDE) of NMPA has provided the proposed analytical similarity criteria for rituximab biosimilars to improve the practice of biosimilar evaluation; these criteria are based on the clinical risk of quality attributes (QAs).¹⁰

The FDA, EMA, and NMPA have issued technical guidelines for research, development, and evaluation of biosimilars to meet the fast development of biosimilar. The principle of biosimilars development and evaluation is to establish quality similarity between a proposed biosimilar candidate (BC) and its RP. A thorough physicochemical and functional comparability exercise serves as the first priority for the subsequent demonstration of biosimilarity in further nonclinical and clinical trials. According to these guidelines, the amino acid sequence of biosimilar products should be the same as that of the reference product. The post-translation modifications (PTM), such as terminal amino acid variants, charge variants, and oligosaccharide profiles, that may present different profiles, need to be further evaluated.¹¹

However, the analytical comparability assessment requires a tremendous amount of work. mAbs are large (∼150kD) proteins with inherent heterogeneity due to PTMs, biosynthetic processes, and their subsequent manufacturing and storage. The heterogeneous isoforms of mAbs include glycoforms, charge, cysteine-related, oxidized, size, and low-level point-mutation variants.¹² Inherent complexity can result in a large number of QAs, some of which are critical QAs (CQAs). A battery of state-of-the-art and orthogonal technologies for the comparison of physicochemical and bioactivity are applied to assess the innovator and two biosimilar candidates thoroughly. These technologies include those based on physiochemical characteristics (amino acid sequence, intact/subunit molecular weight, isoelectric point (pI), PTMs, and disulfide linkage patterns), purity (charge variants, high- and low-molecular-weight variants), antigen binding activity, Fc receptor binding affinity, and Fc-effector functions. The sensitivity, selectivity, specificity, and resolution of liquid chromatrography tandem mass spectrometry (LC-MS) have allowed this method to become an essential tool in the analytical characterization of biosimilar. Therefore, LC-MS and LC-MS/MS are also used in the characterization and identification of QAs such as primary structure, PTMs, and sequence variants.

In our study, the innovator rituximab (MabThera) and its two biosimilar candidates were collected. Physicochemical properties including primary structure, molecular size variant, charge variant and N-glycosylation, amino acid modification, CD20 binding activity, Fc receptor binding affinity, and Fc-effector function (CDC and ADCC) were compared to elucidate the differences between the originator and biosimilar candidates. The extended characterization of BC1 demonstrated its physicochemical and biological similarities to the RP. However, the assessment of BC2 and the RP indicated that BC2 was not similar to the RP in terms of primary structure, N-glycosylation profile, and purity. These results would contribute to the development and evaluation of other biosimilar drugs.

2. Materials and Methods

2.1. Materials

China-sourced Rituximab RP solution (100 mg/10 mL, MabThera, Roche, Lot H0182) was purchased from Roche. Two biosimilar candidates, BC1 (100 mg/10 mL) and BC2 (100 mg/10 mL), were obtained from two companies in China, respectively. A genetically modified Chinese hamster ovary (CHO) cell line K1, cultured in the commercial serum-free medium, was used to produce BC1 and BC2. The monoclonal antibodies were captured from harvest using protein A chromatography media and purified by a combination of chromatography unit operation. All samples were stored at −80 °C.

2.2. Peptide Mapping and PTM Identification

The amino acid sequence and CEX fraction with PTMs, including deamidation, oxidation, and glycosylation, were assessed by using reduced tryptic peptide mapping followed by RP-UPLC-MS/MS (Waters ACQUITY UPLC-Xevo G2 QTof system and Vanquish UHPLC Q Exactive plus). Sample preparation followed product monograph (USP Medicines Compendium). RP and two biosimilar candidates were processed by denaturation with guanidine hydrochloride, reduction with dithiothreitol (DTT), and alkylation with iodoacetamide (IAM). The reduced alkylated samples were then desalted and buffer exchanged to 50 mM Tris-HCl 10 mmol/L CaCl₂ (pH 7.8) with a NAP-5 column (GE Healthcare) and then digested with trypsin (Promega) or chymotrypsin (Roche). For N-glycosylation site profiling, the samples were further digested with PNGase F (New England Biolabs).

The resulting peptides of the standard solution as well as the sample solutions were separated by UPLC (Waters) and UPLC-QDa (Waters) using a BEH C₁₈ column (1.7 μm, 2.1 mm × 100 mm, Waters). Absorbance was monitored at 214 nm with a UV detector and QDa detector. Meanwhile, the resulting peptides were subsequently injected onto a Waters Acquity UPLC BEH C18 column (1.7 μm, 2.1 × 150 mm²) coupled on-line to Waters Xevo G2-S Q-TOF MS system for separation and detection.

Disulfide bond connectivity was analyzed by nonreduced peptide mapping. The samples were digested by trypsin in 50 mM Tris-HCl (pH 8.0) and 1 M urea. The resulting disulfide bond linked peptides were identified by LC-MS/MS analysis as mentioned above. The data were analyzed by UNIFI 1.9 software.

2.3. Molecular Weight Determinations of Intact and Reduced Proteins

The molecular weights and modification (sialylation and glycation) of intact, reduced, and IdeS-cleaved RP and two biosimilar candidates were determined by UPLC (Acquity C4 BEH column, 1.7 μm, 2.1 × 50 mm², Waters) coupled online to Waters Xevo G2-S Q-TOF MS. For IdeS digestion, the protein/enzyme mixture, 1:50 (w/w) was diluted with a digestion buffer (50 mM NH₄FA, pH 6.6) to a final concentration of 1 mg/mL and incubated at 37 °C for 30 min. The samples were then reduced with 20 mM DTT in 50 mM NH₄HCO₃ (pH 8.0) at 37 °C for 10 min and then analyzed by RP-UPLC-MS. UNIFI 1.9 software was used to process and interpret the collected LC-MS data.

2.4. N-Glycan Profiling and Sialic Acid Assay

Two N-glycan profiling techniques with different separation mechanisms, including CE and hydrophilic interaction liquid chromatography (HILIC) coupled with QTof, were used to elucidate N-glycoprofiles of RP and its two biosimilar candidates. The oligosaccharides mixture A (USP, Lot. F00150) was analyzed as the glycoform standard for CE.

N-Glycans were released by PNGase F digestion and further labeled with 2-aminobenzamide (2-AB) followed by HILIC-FLD-QTof analysis. N-Glycans labeled with 2-AB were separated on Waters BEH Glycan Amide column (1.7 μm, 2.1 × 150 mm²) with a UPLC system (Waters H Class) equipped with a fluorescence detector (excitation/emission light of 330 nm/420 nm) connected on-line to a Xevo G2-S Q-Tof MS apparatus. For CE separation, the released N-glycans were labeled with 8-aminopyrene-1,3,6-trisulfonic acid trisodium salt (APTS) and analyzed via a 40 cm separation length capillary with a fluorescence detector (excitation/emission light of 488 nm/520 nm).

The contents of sialic acid in rituximab (RP) and two biosimilar candidates were analyzed by RP-UPLC after labeling with DMB (Ludger). Briefly, sialic acid was released by 2 M acetic acid and incubated at 80 °C for 2 h. Reagent and coupling solution were then added, and the mixture was incubated at 50 °C for 3.5 h, protected from light. The sample was then separated on BEH C18 column (1.7 μm, 2.1 × 100 mm², Waters) and Waters H Class equipped with a fluorescence detector (excitation/emission light of 373 nm/448 nm).

2.5. Charge Heterogeneity Assays

Two capillary electrophoretic methods, capillary isoelectric focusing (cIEF) and whole column imaging detection-capillary isoelectric focusing (icIEF), were applied for the determination of pI values of MabThera and two biosimilar candidates. cIEF analysis followed the product monograph (USP Medicines Compendium). In icIEF analysis, rituximab was mixed with Pharmalyte 3–10, Pharmalyte 8–10.5, 1% methyl cellulose, pI marker (7.0/10.0), and distilled water. The mixture was loaded onto an iCE3 icIEF instrument (Protein Simple) and resolved by prefocusing for 1 min at 1500 V and focusing for 4.5 min at 3000 V. The pI of the sample peaks were determined using a linear regression between two or three pI marker peaks.

Rituximab samples were treated with carboxypeptidase B (CpB) (100:1 w/w) for 2 h to remove the unclipped Lys residue at the C terminus of the heavy chain. The distribution of charge variant was evaluated by cation exchange chromatography (CEX) analysis using ProPac WCX-10 (10 μm, 4.0 × 250 mm², Thermo Scientific) on Agilent 1290 HPLC system. UV detection was carried out at a wavelength of 280 nm. For fraction collection, samples were separated on ProPac WCX-10 Semiprep cation exchange column (9 × 250 mm², Thermo Scientific). Chromatographic was executed on an Äkta Explorer 10S apparatus equipped with UV detection at 280 nm and Fra-950 fraction collector (GE Healthcare).

2.6. Molecular Size Variants Assay

Size-exclusion chromatography (SEC) and capillary electrophoresis sodium dodecyl sulfate (CE-SDS) following product monograph (USP Medicines Compendium) were used to separate lower and higher molecular mass variants. In brief, SEC was carried out on Tosoh Biosciences G3000SWXL column (5 μm, 7.8 mm × 30 cm), with the UV detection wavelength of 280 nm. CE-SDS under reducing and nonreducing conditions was carried out on a Beckman PA800 Plus (Beckman Coulter). mAbs were denatured using SDS and incubated at 70 °C for 10 min before injection. For reduced CE-SDS, samples were denatured, and any free sulfhydryl group from the cysteine side chain was alkylated with IAM to prevent disulfide shuffling before analysis. Samples were pressure-injected, and the separation was carried out with 15.0 kV. UV absorbance was recorded at 214 nm.

2.7. C1q Binding Assay

C1q binding activity was determined by ELISA. Serial antibody dilutions were coated onto a 96-well plate. After blocking with bovine serum albumin (BSA), 0.5 μg/mL C1q (Abcam) was added and incubated for 2 h at room temperature. Then 2 μg/mL anti-C1q-HRP antibody conjugate and TMB, which generates a colorimetric reaction, were added to complete the reaction. The C1q binding activity of mAbs was determined by EC50. The relative C1q binding activity of BC1/2 were determined from comparison with the EC50 value of RP.

2.8. FcRn Receptor Binding Assay

The affinity of mAb samples to human FcRn receptors was determined by Biolayer Interferometry (BLI) using a ForteBio (Octet Qke). Each of the mAb samples was biotinylated, respectively. Seven SA sensors were used to load biotinylated mAb samples at the concentration of 5 μg/mL until to reach about 1.5 nm. The sensors were moved to PBST buffer wells (PBS Ph 6.0, 0.1% BSA, 0.02% Tween 20) for baseline generation and subsequently moved to FcRn wells (concentration range is 4000, 2000, 1000, 500, 250, 125, and 0 nM) for 300 s. Then they were dipped in PBST buffer wells for dissociation for 60 s. The sensors were used just once. The results of FcRn receptor binding activity were analyzed by the multicyclic kinetics strategy and reported by comparing the K_D ratios of BC1/2 to that of RP.

2.9. CD20 Binding Assay

Antibody binding to membrane-bound CD20 was determined by surface plasmon resonance (SPR) using a Biacore T200 (GE Healthcare). His-tagged human CD20 (ACROBiosystems) was captured by the anti-histidine antibody immobilized on CM5 sensor chips (0.05 μg/mL CD20-His-tag, 10 μL/min, 120 s). Each test sample was serially diluted (18.75, 9.375, 4.688, 2.344, 1.172, and 0 μg/mL) with HEPES buffer (10 mM HEPES, 150 mM NaCl, 3 mM EDTA, with 0.005% Tween-20, 0.05% DDM, and 0.01% CHS, pH 7.4) and injected (30 μL/min, 120 s association, 300 s dissociation). The chip surface was regenerated after each cycle by injecting 10 mM Glycine-HCl, pH1.5 (30 μL/min, 60s). The results of CD20 binding activity were analyzed by the multicyclic kinetics strategy and reported by comparing the KD ratio of BC1/2 to that of RP.

2.10. Bioassay

In order to assess biological activity, ADCC and CDC bioassays were conducted. ADCC was evaluated by luciferase release assay.¹³ WIL2-S cells were used as target cells, and Jurkat/NFAT-luc+FcγRIIIa, which constitutively express FcγRIIIa receptor and nuclear factor of activated T cells (NFAT) luciferase, were used as effector cells. Briefly, 25 μL of WIL2-S cells (1 × 10⁶/mL) (Promega) and Jurkat cells (6 × 10⁶/mL) (Promega) in logarithmic growth phase were seeded in a 96-well plate with 5-fold serially diluted test sample at a concentration from 18000–0.01 ng/mL. After incubation at 37 °C, 5% CO₂ for 6 h, cell cytotoxicity was measured by quantification of released luciferase-detecting luminescence with Bio-Glo Luciferase Assay Reagent (Promega).

To assess CDC activity, WIL2-S cells (2 × 10⁶/mL) were incubated with 3-fold serially diluted mAbs (30–0.014 μg/mL) in the presence of normal human serum as a complement source. After incubation for 0.5 h at 37 °C and 5% CO₂, CDC activity was monitored by determining cell viability, reading the absorbance at 450 nm (OD450) with CCK8 reagent (Dojindo). The ADCC and CDC activity were deduced from a parallel logistic assay (4-parameter fit) and reported as EC₅₀. The relative potencies of both ADCC and CDC of BC1/2 were determined from comparison with the EC₅₀ value of RP.

3. Results

3.1. Primary Structure and Disulfide Linkages

The function of a therapeutic mAb is fundamentally determined by its primary amino acid sequence and disulfide linkages. The primary structure of MabThera and its two biosimilar candidates were investigated using multiplex methods, including intact, reduced, and IdeS-cleaved protein mass analysis, and reduced peptide mapping. The predominant peaks in the intact protein MS analysis of RP and BC1 closely matched the theoretical molecular weights of rituximab with different glycan forms (mass error < 5 Da). As shown in Figure 1A, the four highest MS peaks represent the 2 heavy chain (HC) glycoforms/modifications G0F/G0F/-2K/PyroQ(4), G0F/G1F/-2K/PyroQ(4), G1F/G1F/-2K/PyroQ(4), and G1F/G2F/-2K/PyroQ(4) of rituximab, respectively, in which “-2K” indicates C-terminal lysine truncation on both HCs and “PyroQ(4)” indicates the N-terminal glutamate-to-pyroglutamate conversion on both HCs and both light chains (LCs). The molecular masses of reduced BC1 further proved that the LC and HC primary sequences of the biosimilar were similar to those of the RP (Figure 1B,C).

Figure 1.

Comparison of intact protein and subunit masses of RP, BC1, and BC2. (A–C) Intact protein, HC and LC mass of the RP and BC1. (D–G) HC, Fd′, Fc/2, and LC mass of RP and BC2.

The intact mass of BC2 was not consistent with that of the RP (Figure S-1). The difference in mass might be caused by the glycosylation distribution and amino acid sequence. The LC/HC and Fc/Fd regions were further identified after reduction and sequentially IdeS digestion to reveal the mass differential region.¹⁴ Although the main peak of the BC2 LC was consistent with that of the RP, these differences include different abundances of major N-glycans (G0, G0F, G1F, and G2F; Figure 1D). The mass variant was also observed in the BC2 HC with a mass shift of approximately +28 Da from the G0F and G1F glycan forms. This result suggested that a sequence variant or some new modifications might exist in the HC. Fc/2 and Fd′ from HC, digested with IdeS, were further identified to confirm the mass shift region. For the mass consistency of Fc/2, a mass shift of +28 Da occurred in Fd′ (Figure 1E,F). Additional mass variation details were further investigated by using peptide mapping because it is an important technique for characterizing the proteins amino acid sequences.

As shown in Figure 2, the peptide mapping of the RP and its biosimilar candidates were compared via UPLC. The complementarity-determining regions (CDRs) and peptides separated from the RP were assigned by UPLC-QDa. The retention time and masses of the two biosimilar candidates’ peptides, including the HC and LC CDRs, were obtained as assigned by the QDa detector (Table 1) and were consistent with those of the RP. A data acquisition C18 LC-MS with alternate changeable energy full scanning (LC-MS^E) was used to analyze the peptides digested by trypsin and chymotrypsin to reveal the mass difference. Almost 100% sequence coverage, including all the CDR regions, was obtained in peptide mapping, and the tryptic peptide mapping chromatographic profiles of the RP were consistent with that of BC2. However, a new peak with m/z 717.3581 6+, eluted at 38.28 min, was detected in the chymotryptic peptide mapping chromatographic profiles of BC2. This peak did not match the desired rituximab sequence according to the data analyzed by UNIFI 1.9. The new peak molecular weight was 28 Da heavier than the semi-digested chymotryptic C17′ peptide of Fd′ (ICNVNHKPSNTKVDKKAEPKSCDKTHTCPPCPAPELL, m/z 712.6864, 6+) (Figure 3A). The increased mass of the C17′ peptide complied with the mass shift in the subunit analysis. The MS/MS data analysis of the collision-induced dissociation (CID) of the new peak is shown in Figure 3B. The MS/MS spectrum of the C17′ chymotryptic peptide of BC2 showed a mass shift of +28 Da in the b18 ion relative to the theoretical b18 ion mass. The C17′ chymotryptic peptide was collided with higher energy to improve the fragmentation. The identifaction of the b15, b16, and b17 ions illustrated that alanine (A) at 219 of HC was replaced by valine (V). The A219V mutation, which is located in the constant region 1 of HC, was an allotype of IgG1.

Figure 2.

Comparison of the peptide map profiles of the RP, BC1, and BC2 by UPLC. Due to coelution with other peptides, HC CDR3 was not labeled in peptide mapping.

Table 1. Retention Times and Masses of CDRs for the RP and Its Biosimilar Candidates Detected by UPLC-QDa.

	heavy chain								light chain
	CDR1		CDR2		CDR3				CDR1 + 2		CDR3
samples	t/min	m/z	t/min	m/z	t/min	m/z	t/min	m/z	t/min	m/z	t/min	m/z
RP	44.01	598.27	54.13	1092.50	50.38	893.06	62.81	963.79	55.02	787.79	46.20	960.45
BC1	43.64	598.28	53.82	1092.55	50.36	893.05	62.81	963.79	54.47	787.80	45.91	960.47
BC2	43.43	598.27	53.67	1092.51	50.38	893.06	62.60	963.76	54.37	787.82	45.73	960.35

Figure 3.

Extracted ion chromatograms (XIC) spectrum and MS/MS spectrum of the C17′ peptide precursor ion (m/z 717.3581, 6+). (A) Comparison of the C17′ peptide XIC spectra of RP and BC2. (B) MS/MS spectrum of the C17′ peptide with A219 V mutation.

Disulfide bond formation is critical for protein folding, assembly, structure, stability, and function. Rituximab consists of a total of 32 cysteine residues that form 16 disulfide bonds.¹⁵ Twelve of them are intrachain disulfide bonds that connect the two layers of antiparallel β-sheets of each domain, and four of them are interchain disulfide bonds that connect the LCs and HCs. Nonreducing LC-MS/MS peptide maps exhibited the expected disulfide bond connectivity, indicating the presence of identical disulfide structures in the RP and BC1/2. The identified disulfide bridges were: intrachain [LC:C²³–LC:C⁸⁷, LC:C¹³³–LC:C¹⁹³, HC:C²²–HC:C⁹⁶, HC:C¹⁴⁸–HC:C²⁰⁴, HC:C²⁶⁵–HC:C³²⁵, and HC:C³⁷¹–HC:C⁴²⁹] and interchain [LC:C²¹³–HC:C²²⁴, HC:C²³⁰–HC:C²³⁰, and HC:C²³³–HC:C²³³]. The retention time and masses of peptides containing disulfide bonds were obtained as shown in Table S-1.

3.2. Charge Variants and Percentages of PTMs

The charge variants of mAb typically resulting from PTMs might influence antibody structure, long-term stability, biological activity, and pharmacokinetics. These charge variants are generally referred to as acidic or basic in comparison with the main species. cIEF represents one of the highest resolution methods for separating proteins in accordance with their pI. The assessment of charge variants by cIEF revealed that the pI values of the main peak obtained from the RP, BC1, and BC2 were 9.29, 9.36, and 9.35, respectively, which satisfied the criteria of 9.1–9.5 in accordance with the USP MC (Figure 4A). The main peak pI values from the two biosimilars differed by no more than ±0.1 pI units from the corresponding peak of the RP. Compared with that in the RP, a higher level of basic variants was found in BC1, whereas slightly higher levels of acidic and basic variants were observed in BC2.

Figure 4.

cIEF (A) and icIEF (B) profiles of MabThera and its two biosimilar candidates for the determination of their pIs and the charge variants.

For the further elucidation of the charge heterogeneity, the RP and its biosimilar candidates were analyzed by using icIEF as well. As shown in Figure 4B and Table S-2, the percentage of basic charge variants, including two peaks at pI 9.33 and 9.43, was higher in BC1 than in the RP, whereas the percentage of acidic variants was higher in BC2 than in the RP in compliance with the results from cIEF.

CEX is a technique orthogonal to cIEF, given that only the solvent-exposed part of the protein can interact with the stationary phase, whereas the pI value in cIEF described the pH at which the net charge is equal to 0.¹⁶ In CEX analysis, charge variants were successively eluted in acidic peaks, main peak, and basic peaks (Figure 5). After the removal of lysine with CpB, BC1 exhibited similar levels of basic and acidic variants compared with the RP, while BC2 had a higher level of acidic and basic variants (Table S-3). In contrast to the high level of basic variants in BC1 observed with cIEF and icIEF analysis, the result obtained from CEX indicated that BC1 might have a higher level of C-terminal lysine.

Figure 5.

Cation exchange chromatograms of the RP and its two biosimilar candidates with CpB digestion.

Acidic and basic variants can be formed via multiple chemical and enzyme modifications, with the acidic fraction typically containing a variety of protein species, including deamidated, sialylated, and glycated forms, whereas the basic fraction contains oxidated and C-terminal lysine variants.¹⁷ Five fractions of the RP and BC1 were collected, and six peaks of BC2 were obtained to reveal these modifications. The fractions of the peptides digested with trypsin were analyzed by C18 Vanquish UHPLC Q Exactive plus for the elucidation of charge heterogeneity (Table S-4). The LC-MS/MS analysis of acidic or basic CEX fractions revealed that the acidic variants were due to a slightly increased level of deamidation, whereas basic peaks resulted from the unprocessed C-terminal lysine on the HC and N-terminal Q (not pyroglutamate) on the LC. The higher level of C-terminal lysine residue observed in BC1, could account for the high level of basic variants found through cIEF and icIEF analyses (Figure 6A). Considering that differences in C-terminal lysine truncation have a negligible effect on structure, biological activity, pharmacokinetics, and safety, this attribute could not affect the potential clinical outcome.

Figure 6.

Percentage of lysine truncation, deamidation, and oxidation at HC N⁵⁵, C-terminus, HC M²⁵⁶, and HC M⁴³². (A) C-Terminal lysine truncation. (B) Deamidation of HC N⁵⁵. (C–D) Oxidation of HC M²⁵⁶ and HC M⁴³².

Furthermore, the fragments of mAbs produced by IdeS were analyzed through high-resolution MS to reveal the sialylation of Fc/2 and the glycation of LC and Fd′ in charge variants (Table S-5). Given that sialylation and glycation have been identified as two important factors that contribute to the formation of acidic species,¹⁸ relatively high levels of sialylation and glycation were also observed in acidic peaks of the RP, BC1, and BC2. Therefore, the acidic peaks were mainly formed via deamidation, sialylation, and glycation, whereas basic peaks were due to a C-terminal lysine on the HC and no pyroglutamate of the N-terminal Q on the LC.

As one of the common chemical modifications in IgG1, asparagine (N) deamidation or isomerization in the CDRs of the HC causes the loss of activity in IgG1.¹⁹ In quantitative LC-MS analysis, all fractions at the N⁵⁵ (located in the HC CDR 2) were 1.5–2% deamidated and might affect the antigen-binding activity of rituximab (Figure 6B). Methionine (M) oxidation is a very common PTM that can influence the bioactivity and half-life of an antibody and potentially induce an immunogenic response. The oxidation of two conserved HC M residues, M²⁵⁶ and M⁴³² (located at the interface of the C_H2 and C_H3 domains), were observed in the RP, BC1, and BC2. Compared with the RP, BC1/2 had slightly higher level of M²⁵⁶ and M⁴³² oxidation, which might decrease thermal stability, protein A binding, FcRn binding, and the circulation half-life of IgG1 antibodies (Figure 6C,D).²⁰⁻²² The glycation level of the RP was slightly higher than that of BC1 and BC2. Therapeutic mAb glycation is a nonenzymatic glycosylation on protein amine groups and has a potential effect on biological functions, such as blocking the biologically functional site or further degradation that induces aggregation.²³

3.3. N-Glycosylation Profiling and Sialic Acid Content

N-Glycosylation patterns play an important role in the function, efficacy, in vivo half-life, and immunogenicity of mAbs. Oligosaccharides affect the binding affinity and ADCC and CDC activities, which are the modes of action of rituximab.²⁴ LC-MS/MS peptide mapping confirmed the absence of O-linked glycosylation modifications in the RP, BC1, and BC2. As expected, one N-linked glycosylation site existed at N³⁰¹ of the HC. Consistent with the result of LC-MS/MS analysis, the reduced CE-SDS results indicated that 99.6, 99.8, and 99.5% of HC N³⁰¹ were glycosylated (Table S-8) for the RP, BC1, and BC2, respectively.

Here, we applied two techniques with CE and HILIC-MS to elucidate the N-glycan profile of the RP and its two biosimilar candidates. The retention time of oligosaccharides mixture A in CE, including G0, G0F, G1F(1,6), G1F(1,3), G2, G2F, G2FS1, G2FS2, and Man-5, was used for the identification of N-glycans. In HILIC-MS analysis, the N-glycans were determined in glucose units through comparison with a standard dextran in HILIC and their masses detected in MS, which were compared with reference values in the “Glycobase” database for preliminary structural assignment. As shown in Tables 2 and S-6 and Figure S-2, the major N-glycans of the RP and BC1 were G0F, G1F(1,6), G1F(1,3), and G2F, which were consistent with the glycosylation distribution of intact mass. Similar N-glycan species and abundance distribution were detected in the RP and BC1. The average percent galactose values of the RP and BC1 were 50.6 and 48.6%, respectively. Although CDC activity was enhanced in the presence of galactose residues, a difference of 2% in galactose between the RP and BC1 would not cause noticeable changes in their CDC activities as has been confirmed by the cell-based bioassay for CDC. High mannose content was relatively high but was lower in BC1 than in the RP. High mannose residues bearing IgG molecules exhibited shortened circulatory half-life and enhanced ADCC activity. The afucosyl levels of the RP and BC1 were similar and were correlated with ADCC and FcγRIIIa-binding activities.

Table 2. Glycan Profile of Rituximab and Its Two Biosimilar Candidates by HILIC-MS.

sample	G0F%a	man%b	sialylation%c	gal%d	afuc%e	NANA mol/mol	NGNA mol/mol
RP	42.3 ± 0.03	5.2 ± 0.06	2.4 ± 0.03	50.6 ± 0.10	1.9 ± 0.02	0.038 ± 0.005	ND
BC1	42.7 ± 0.05	3.0 ± 0.02	2.3 ± 0.03	48.6 ± 0.00	1.7 ± 0.02	0.045 ± 0.003	ND
BC2	61.8 ± 0.06	8.1 ± 0.34	0.6 ± 0.02	21.1 ± 0.18	7.4 ± 0.13	0.076 ± 0.008	ND

^aG0F and G0F-GN.

^bMan-5, Man-6, Man-7, an Man-8-type N-glycans.

^cG1FS1-, G1FS1-GN-, G2FS1-, and G2FS2-type N-glycans.

^dG1(1,6)-, G1(1,3)-, G1F(1,6)-, G1F(1,3)-, G1F-GN-, G1FB-, G2-, and G2F-type N-glycans.

^eG0-, G0-GN-, G1(1,6)-, G1(1,3)-, G1-GN-, and G2-type N-glycans.

The major N-glycans of BC2 were G0F, G1F(1,6), G1F(1,3), and G0, which were not consistent with those of the RP. BC2 had considerably higher relative G0F% content (61.8%) and lower relative content of galactose-containing glycans (21.1%), compared to the RP. The average G0% of BC2 was 7.7%, accounting for the considerably high level of afucose. Biological activities were further evaluated by taking the significance of glycosylation into account.

Sialylation can affect antibody effector functions and safety, especially the half-life of protein drugs in the human body.²⁵ The absolute amount of sialic acid was determined. In line with the results of N-glycan analysis, BC1 and BC2 exhibited a content of N-acetylneuraminic acid (NANA) (0.045 and 0.076 mol/mol, respectively) slightly higher than that of the RP (0.038 mol/mol). N-Glycolylneuraminic acid (NGNA), a cause of potential immunogenicity in humans, was not detected in the RP and its biosimilar candidates.

3.4. High- and Low-Molecular-Weight Variants and Purity

High-molecular-weight (HMW) species have the potential to increase product immunogenicity. Another important category of mAb size variants comprises low-molecular-weight (LMW) species, which may have decreased activity or a reduced serum half-life because of missing Fab or Fc fragments. The molecular size variants of the RP and its biosimilar candidates were examined by applying SEC and CE-SDS.

Under native conditions, SEC separates monomeric mAbs from other variants of lower or higher molecular weight. As quantified via the area normalization method, the relative contents of monomers in the RP, BC1, and BC2 were 98.45, 98.22, and 96.44%, respectively (Figure 7). A shoulder peak after the main peak, which might be the Fab/c missing fragment, was observed in all samples.²⁶ Slightly higher levels of aggregates were observed in BC1 and BC2 compared to that in the RP. The relative content of HMW variants and the shoulder peak impurity in BC2 were higher than those in the RP and the BC1 (Table S-7). Given that achieving an ideal resolution of LMW species with SEC was difficult, CE-SDS was used as an orthogonal method to measure fragmentation levels.

Figure 7.

HMW species and LMW species profiles of the RP and its two biosimilar candidates by SEC.

Nonreducing CE-SDS was used to separate monomers from fragments and HMW variants, whereas reducing CE-SDS was utilized to resolve LC, HC, and nonglycosylated HC (NGHC) from impurities by size. The comparison of the reducing and nonreducing CE-SDS profiles revealed a similar pattern of purity and size heterogeneity between the RP and BC1 (Figure 8 and Table S-8). In nonreducing CE-SDS, fragment No. 5 (F5) of BC2, which might be missing one LC (HHL), was found to be much higher than that of the RP. Consistent with SEC spectra, the shoulder peak impurity (F6) of BC2, which was suspected to be the Fab/c missing fragment, was also higher than that of the RP. These peaks accounted for the considerably low relative monomer content in BC2. Reduced CE-SDS showed that the relative HWM species content was higher in BC2, resulting in the low relative content of HC.

Figure 8.

Nonreduced CE-SDS (A) and reduced CE-SDS (B) electropherograms of the RP and its two biosimilar candidates.

3.5. Biological Activities

Binding affinities to FcRn, C1q, and CD20 were determined through Fortebio, ELISA, and SPR, respectively. The participation of FcRn in the recycling of IgG can potentially affect the half-life of mAb drugs in serum. The in vitro FcRn-binding assay results shown in Table 3 suggested that BC1 and BC2 had similar binding capabilities. The binding of the Fc to C1q could initiate the classical pathway of complement activation and then lead to the dissolution of target cells. The CDC-related C1q-binding affinity was confirmed to be consistent between the RP and BC1/2. The binding capacity to the CD20 molecule on the surfaces of B cells triggers rituximab-induced cell death. Although CD20 binding affinity of BC2 seemed to be higher than that of RP, the relatively high variability was observed in the binding affinity assay.^8,9 Therefore, this result might indicate that the CD20 binding affinity of BC1/2 was similar to that of the RP.

Table 3. Relative Binding Activities between the RP and Its Two Biosimilar Candidates.

sample	FcRn-binding (%)	C1q-binding (%)	CD20-binding (%)
BC1	86.7 ± 10.7	110.1 ± 11.6	100.0 ± 8.5
BC2	81.3 ± 9.1	99.0 ± 12.3	126.5 ± 15.0

CDC and ADCC potencies were evaluated as tier 1 in the rituximab quality similarity assessment.¹⁰ CDC and ADCC activity, as the comprehensive effect of physicochemical characteristics, are important for the clinical efficacy of therapeutic antibodies. The relative activities of BC1 and BC2 were further compared with those of the RP in CDC and ADCC assays. These activities depend on a fully functional Fc domain in addition to target binding. Table 4 shows that the CDC effects induced by BC1 and BC2 were similar to those induced by the RP. The relative activities of ADCC were analyzed for the RP, BC1, and BC2 to further evaluate the Fc-effector function of rituximab. The result showed that RP, BC1, and BC2 induced similar ADCC responses.

Table 4. Relative Fc-Effector Function of the RP and Its Two Biosimilar Candidates^a.

sample	relative activity of CDC (%)	relative activity of ADCC (%)
BC1	106.2	105.9
BC2	100.6	95.1

^aThese methods have been well-validated and verified.

4. Discussion

On the basis of the risk-based ranking of rituximab’s QAs published by the CDE of NMPA, an extensive set of state-of-the-art and orthogonal analytical technologies were used to provide a comprehensive characterization of the RP and its two biosimilar candidates. The characteristics that were compared included primary structures and disulfide linkages, biological activities, purity/impurities, and N-glycosylation. Advanced analytical technologies, such as UPLC-MS, icIEF, and SPR, have provided the full characterization and improved the understanding of the structure of antibodies. In our study, LC-MS was used extensively in the characterization of rituximab and its biosimilar candidates to reveal the amino acid replacement, CDRs, glycosylation patterns, and disulfide bond characteristics of the compounds. The MS-based multi-attribute method, which allows for the direct measurement of multiple PQAs, was also used in PTM percentage analysis. icIEF, which allows whole-column real-time detection, achieves faster separation with higher resolution, better reproducibility, and reduced sample volume compared with cIEF, was utilized in charge variants assessment.²⁷ SPR-based analytical biosensors are extremely powerful tools for the characterization of molecular interactions. In our research, SPR was used in the analysis of CD20 binding affinity. As a result, most of the measured contents of the RP and two biosimilar candidates in our study were consistent with the reported quality range of rituximab.⁶⁻⁹

All the comparison results demonstrated that BC1 was analytically similar to the RP. First, a series of LC-MS analyses suggested that BC1 and the RP had identical amino acid sequences and highly similar PTMs, including the glycosylation species, disulfide linkages, and PTM sites, and most PTM percentages. Second, the purity and heterogeneity profiles (obtained via SEC, CE-SDS, CEX, and icIEF) indicated that BC1 had similarly high purity. Third, BC1 and the RP had similar antigen binding activity, Fc receptor binding affinity, and Fc-effector function (CDC and ADCC). The oxidation percentage of BC1 was slightly higher than that of the RP, and the effect of this difference could be further evaluated by the nonclinical and clinical studies. The totality of these results demonstrated the analytical similarity of BC1 to the RP, supplying fundamental information for a further nonclinical and clinical similarity evaluation.

However, some QAs of BC2 were dissimilar from those of the RP. First, the intact mass and peptide mapping data indicated that alanine at 219 of HC was mutated into valine, which was an allotype of IgG1. Although the allotype is designed to reduce the risk of the potential for immunogenicity, therapeutic IgG1 allotypes located in the constant region influence the pharmacokinetics through FcRn binding.²⁸ Therefore, an in vivo nonclinical study may be required to further evaluate the difference in amino acid sequence. Moreover, the technical guideline for the research, development and evaluation of biosimilars, issued by NMPA, excludes biologics that have a different amino acid sequence from their reference products as biosimilars. Although BC2 had the same CDRs as and modes of action similar to those of the RP, it might not be applied as a biosimilar in drug registration.

Second, the N-glycosylation distribution and relative contents of BC2 were not consistent with those of the RP. The lack of core fucose in rituximab improves binding to human FcγRIIIa and ADCC.^29,30 The terminal galactose residues of rituximab affect CDC activity by increasing C1q binding affinity.³¹ Recent studies further indicated that galactosylation, especially terminal galactose on afucosylated glycan chains, has a positive effect on the binding activity of the IgG1 to FcγRIIIa receptors and ADCC activity.^32,33 A type II humanized anti-CD20 antibody GA101 (Obinutuzumab) has been glyco-engineered to produce a nonfucosylated Fc region that substantially enhances the ADCC activity against lymphoma cells compared with that of rituximab.³⁴ BC2 processed considerably lower levels of fucose and galactose than those of the RP. However, the ADCC activity of BC1 was similar to that of the RP. One possible cause for this is as follows: although the high level of afucosylation might improve the ADCC activity, this positive effect might be weakened by the low level of terminal galactose. The most frequently observed notable difference between biosimilar candidates and the RP is that the difference in afucosylated glycans in some infliximab biosimilars (Flixabi and Zessly) also does not affect ADCC activity.³⁵ Interestingly, although BC2 had a lower Gal content than that of the RP, the C1q binding affinity and CDC activity of BC2 were consistent with those of the RP. Although galactosylated glycans could affect CDC activity, product knowledge and clinical experiments have shown that several mAbs have wide ranges of galactose content (10–40%).³⁶ Furthermore, in addition to N-glycans, the C_H2 and lower hinge region of the IgG Fc domain also play an important role in the binding affinity of FcγRIIIa and C1q.³⁷ Considering that higher-order structures affect the biological activities of mAb, BC2 might differ from the RP in terms of its C_H2 higher-order structure and hinge region. These differences compensated for the lack of galactose moieties.

Third, nonreduced CE-SDS and CEX detected the lower purity of BC2. Size variants including aggregates, monomers, and NGHC are considered as CQAs in rituximab biosimilar evaluation. The much lower relative content of the BC2 monomer found through CE-SDS might be attributed to the missing Fab/c and could affect the binding between BC2 and CD20 or the Fc-receptor function. Several studies indicated that the different proportions of acidic peaks and basic peaks between the RP and biosimilar do not influence their in vitro potency pharmacokinetics.¹⁶ However, the higher acidic and basic species levels of BC2, which were formed by deamidation, sialylation, glycation, unprocessed C-terminal lysine, and N-terminal Q (not pyroglutamate), still need to be further evaluated in nonclinical and clinical studies.

The most frequent notable differences encountered in the development of biosimilar candidates are increases in HMW species, modification at the CDR or Fc region, and levels of glycans chains (e.g., afucosylated glycans, galactose, and high mannoses); these differences are also observed in BC2.³⁵ The amino acid replacement of BC2 in the Fc region might further affect glycosylation, glycation, and charge heterogeneity. Furthermore, the different biomanufacturing processes and environmental conditions may cause the structural QAs of BC2 to exceed the acceptable limits of the RP. However, the RP and BC2 had similar functional QAs, including CD20 binding activity, FcRn binding affinity, C1q binding affinity, and Fc-effector function. Although the functional QAs reflect clinically relevant modes of action and provide useful information for predicting the outcomes of clinical studies, structural QAs are the cornerstone of comparability between the RP and its BC. The ultimate goal we are pursuing for biosimilars is the absence of clinically meaningful differences from the RP in terms of safety and effectiveness. Therefore, the similarity of quality is key to achieve this goal.

Our study provided the variability in the QAs of biosimilar candidates. Our results could further understand the development of biosimilars in the real word. In the future, BC2 may be developed with further improvement and applied as a novel anti-CD20 IgG1 type I antibody.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsptsci.0c00225.

• Intact protein mass of BC2; peptides containing disulfide bonds identified in RP and two biosimilar candidates; percentages of pI distribution in RP, BC1, and BC2; percentages of acidic peaks, basic peaks and main peak in RP, and two biosimilar candidates; percentages of post-translational modifications in RP, BC1, and BC2 CEX fraction; percentages of glycation and sialylation in RP, BC1, and BC2 CEX fractions; glycan chain percentages of Rituximab and its two biosimilar candidates by CE analysis; N-glycosylation profiling of RP and its biosimilar candidates by HILIC-MS; percentages of molecular weight variants identified by SEC analysis; purity levels identified by non-reduced and reduced CE-SDS (PDF)

Author Contributions

^∥ H.W. and L.W. contributed equally to this work. H.W. designed the project, acquired funding and carried out experiment. L.W., J.X., and C.W. carried out experiment. H.Y., H.G., and L.Z. provided technical guidance. H.S. and G.C. finalized the manuscript.

The authors declare no competing financial interest.