Carbohydrate Analysis of a Chimeric Recombinant Monoclonal

Anal. Chem. 1999, 71, 5185-5192

Carbohydrate Analysis of a Chimeric Recombinant Monoclonal Antibody by Capillary Electrophoresis with Laser-Induced Fluorescence Detection Stacey Ma† and Wassim Nashabeh*,‡

Departments of Analytical Chemistry and Quality Control Biochemistry, Genentech, Inc., South San Francisco, California 94704

A general method for the analysis of asparaginyl-linked (N-linked) carbohydrate moieties of an IgG1 monoclonal antibody is described here. The antibody, rituximab, is a mouse/human chimeric antibody to human CD20 antigen. The glycans present on rituximab are neutral complex biantennary oligosaccharides with zero, one, and two terminal galactose residues (G0, G1, and G2, respectively). To monitor the variation of the glycosylation during manufacture, the glycans were first enzymatically released from the antibody via digestion with peptide-N-glycosidase F, then derivatized with a charged fluorophore, 8-aminopyrene-1,3,6-trisulfonic acid and further separated by capillary electrophoresis with laser-induced fluorescence detection. All observed glycans were fully resolved, including the positional isomers of G1. The exact nature of the isomers in terms of the location of the terminal galactose was further characterized via multiple enzymatic digestion steps including mannosidase with activity toward specific Man(r1,3) linkage. The optimization of several key parameters, i.e., enzymatic digestion and derivatization, in the assay development will be discussed. Moreover, to ensure that the assay can be used in routine lot release testing, the assay was validated and found to be accurate and precise. The analytical approach described is suitable for characterization as well as routine testing of the N-linked glycan content in any IgG1 monoclonal antibody and glycoproteins in general. Recently, there has been an increasing interest in the fundamental understanding of the biological roles of the carbohydrate moieties on glycoproteins. In contrast to the protein amino acid backbone which is genetically coded with high fidelity, the cDNA sequence can only predict the position of potential glycosylation sites and gives no information about the actual glycosylation patterns. In fact, the biosynthesis of glycans on any glycoprotein is largely influenced by the cell line in which the glycoprotein is produced, the conformation of the protein, and cell culture conditions.1 Even under a defined set of expression and culture * Corresponding author. Tel. (650) 225-7213. Fax: (650) 225-8220. E-mail: [email protected]. † Department of Analytical Chemistry. ‡ Department of Quality Control Biochemistry. (1) Patel, T.; Parekh, R.; Moellering, B.; Prior, C. Biochem J. 1992, 285, 839845. 10.1021/ac990376z CCC: $18.00 Published on Web 10/08/1999

© 1999 American Chemical Society

conditions, further processing of the carbohydrate chains by enzymes present in limited quantities results in microheterogeneity of structurally related oligosaccharides at any of the glycosylation sites on a given glycoprotein. Such alterations in protein glycosylation with variable site occupancy or changes in oligosaccharide structure often result in variations in the biological activity of glycoproteins, such as the recombinant monoclonal antibodies.2-4 The importance of the glycan structures in the therapeutic use of monoclonal antibodies has been well-documented.5 Studies have shown that glycosylation of IgG influences both its physiochemical properties and, more importantly, its cellmediated effector functions such as complement binding and activation.6-9 These biological functions are dependent not only on the presence or absence of N-linked oligosaccharides but also on the specific structure of the oligosaccharides. Clearly, in the manufacturing of therapeutic recombinant monoclonal antibody, the assessment of oligosaccharide microheterogeneity and its batch-to-batch consistency are of utmost importance.10,11 Consequently, there is a demand for developing highperformance analytical techniques to bring about the structural characterization and routine analysis of closely related glycans derived from glycoproteins. While nuclear magnetic resonance and mass spectrometry12,13 are indispensable tools for the structural elucidation of carbohydrates, routine profiling and quantitative analysis of glycans are accomplished mostly by chromatographic14-18 and planar electrophoretic techniques.19 Two of the (2) Varki, A. Glycobiology 1993, 3, 97-130. (3) Lis, H.; Sharon, N. Eur. J. Biochem. 1993, 218, 1-27. (4) Rasmussen, J. R. Curr. Opin. Struct. Biol. 1992, 2, 682-686. (5) Parekh, R. B.; Dwek, R. A.; Sutton, B. J.; Fernandes, D. L.; Leung, A.; Stanworth, D.; Rademacher, T. W.; Mizuochi, T.; Taniguchi, T.; Matsuta, K.; Takeuchi, F.; Nagano, Y.; Miyanato, T.; Kotata, A. Nature (London) 1985, 316, 452-457. (6) Winkelhake, J. L. Immunochemistry 1978, 15, 695-714. (7) Rademacher, T. W.; Parek, R. B.; Dwek, R. A. Annu. Rev. Biochem. 1988, 57, 785-838. (8) Tsuchiya, N.; Endo, T.; Matsuta, K.; Yoshinoya, S.; Aikawa, T.; Kosuge, E.; Takeuchi, F.; Miyamoto, T.; Kobata, A. J. Rheumatol. 1989, 16, 285-290. (9) Wright, A.; Morrison, S. TIBTECH 1997, 15, 26-32. (10) Olden, K.; Bernard, B. A.; White, S. L.; Parent, J. B. J. Cell. Biochem. 1982, 18, 313-335. (11) Kobata, A. Eur. J. Biochem. 1992, 209, 483-501. (12) Lee, K. B.; Loganathan, D.; Merchant, Z. M.; Linhardt, R. J. Appl. Biochem. Biotechnol. 1990, 23, 53-80. (13) Chaplin, M. F.; Kennedy, J. F. Carbohydrate Analysis: A Practical Approach, 2nd ed.; IRL Press: Oxford, 1994. (14) Honda, S.; Suzuki, S. Anal. Biochem. 1984, 142, 167-174.

Analytical Chemistry, Vol. 71, No. 22, November 15, 1999 5185

most widely used techniques are high pH anion-exchange chromatography with amperometric detection (HPAEC-PAD)15,20 and fluorophore-assisted carbohydrate electrophoresis (FACE).21,22 More recently, capillary electrophoresis (CE)13,23 has emerged as a powerful tool in carbohydrate analysis with enhanced separation efficiencies and shorter analysis time than the traditional methods. Rapid acceptance of CE in carbohydrate analysis was largely due to the fact that CE is simply the instrumental version of planar electrophoresis employing detection systems adapted from HPLC. As such, many of the electrolyte systems that were originally tested in traditional electrophoresis as well as precolumn derivatization schemes which afford sensitive detection of carbohydrates separated by HPLC have been readily adapted for carbohydrate analysis by CE. Further progress in CE of carbohydrates was facilitated by the introduction of on-column laser-induced fluorescence detection (LIF), the most sensitive detection method employed to date. This detection methodology triggered the development of several derivatization reagents with superior fluorogenic properties. Such reagents include 8-aminopyrene-1,3,6trisulfonic acid (APTS),24 8-aminonaphthalene-1,3,6-trisulfonic acid (ANTS),25,26 2-aminopyridine (AP),27 and 3-(4-carboxybenzoyl)-2quinolinecarboxaldehyde (CBQCA).28 Among them, APTS24,29 has demonstrated numerous advantages over other commonly used reagents in terms of its detection sensitivity and speed of analysis. With the 488-nm argon-ion laser, the limit of detection for APTSderivatized sugars was estimated to be 0.4 nM.24 Therefore, APTS allows analysis of even minute carbohydrate content in therapeutic biomolecules of interest. Rituximab is a genetically engineered chimeric monoclonal antibody approved for the treatment of low-grade non-Hodgkins lymphoma.30 The antibody contains a glycosylated human IgG1 κ constant region (Fc domain) and murine light- and heavy-chain variable regions (Fab domain). The murine sequences contain the complementary determining regions (CDR) that bind to the CD20 receptor on the surface of B cells and are responsible for the specificity of the antibody. The human sequences contain the Fc domain which is responsible for recruiting the patients’ effector functions to mediate targeted cell killing once the antibody is (15) Hardy, M. R.; Townsend, R. R.; Lee, Y. C. Methods Enzymol. 1989, 179, 65-76. (16) Tandai, M.; Endo, T.; Sasaki, S.; Masuho, Y.; Kochibe, N.; Kobata, A. Arch. Biochem. Biophys. 1991, 291, 339-348. (17) Lipniunas, P.; Gro ¨nberg, G.; Krotkiewski, H.; Angel, A.-S.; Nilsson, B. Arch. Biochem. Biophys. 1993, 300, 335-345. (18) El Rassi, Z. Carbohydrate Analysis: HPLC and CE; Elsevier: New York, 1995. (19) Morell, L.; Plotkin, L.; Leoni, J.; Fossa, C. A.; Margni, R. A. Mol. Immunol. 1993, 30, 695-700. (20) Spellman, M. Anal. Chem. 1990, 62, 1714-1722. (21) Jackson, P. Anal. Biochem. 1991, 196, 238-244. (22) Starr, C.; Masada, R.; Hague, C.; Skop, E.; Klock, J. J. Chromatogr., A 1996, 720, 295-321. (23) Honda, S.; Makino, A.; Suzuki, S.; Fujiwara, S.; Kakehi, K. Anal. Biochem. 1989, 176, 72-77. (24) Evangelista, R. A.; Liu, M.-S.; Chen, F.-T. A. Anal. Chem. 1995, 67, 22392245. (25) Jackson, P. Biochem. J. 1990, 270, 705-713. (26) Oefner, P.; Chiesa, C.; Bonn, G.; Horva´th, Cs. J. Capillary Electrophor. 1994, 1, 3-12. (27) Hase, S.; Hara, S.; Matsushima, Y. J. Biochem. (Tokyo) 1979, 85, 217220. (28) Liu, J.; Shirota, O.; Novotny, M. Anal. Chem. 1992, 63, 64-78. (29) Chen, F.-T. A.; Evangelista, R. A. Anal. Biochem. 1995, 230, 273-280. (30) Reff, M.; Carner, K.; Chambers, K.; Chinn, P.; Leonard, J.; Raab, R.; Newman, R.; Hanna, N.; Anderson, D. Blood 1994, 83, 435-445.

5186

Analytical Chemistry, Vol. 71, No. 22, November 15, 1999

bound to the target. Glycosylation at the conserved Asn-301 site in each of the heavy chains is required for the Fc effector functions, but not the CD20 binding. The glycans found on rituximab are asialo-N-linked complex biantennary structures with a core fucose, typical of a monoclonal antibody produced in chinese hamster ovary (CHO) cells. During the cell culture fermentation, significant microheterogeneity of glycosylation arises and yields molecules with variations in galactosylation patterns. The variations in the terminal galactose occupancy were shown to affect the biological activity of rituximab as determined by a complement-dependent cytotoxicity assay.31 Hence, a CE method was developed for the rapid and quantitative analysis of the glycans found on rituximab. In this method, the carbohydrates are first removed enzymatically using peptide-N-glycosidase F (PNGase F), a highly specific endoglycosidase that hydrolyzes the β-aspartylglycosylamine bond between the asparagine residue and the innermost N-acetylglucosamine of the glycan.32-34 The released glycans are then derivatized with APTS and further separated by capillary electrophoresis with laser-induced fluorescence detection. In addition, as the FDA requires all assays for the release of commercial products to be validated,35 a set of experiments was also conducted according to the International Conference on Harmonization (ICH) guidelines35 to ensure that the assay is appropriate for its intended use and to determine that the assay is suitable for routine analysis. The results for various aspects of the assay development and its performance are reported in this paper. EXPERIMENTAL SECTION Reagents and Chemicals. Rituximab was produced by Genentech Inc. and IDEC Pharmaceuticals Corp. (San Diego, CA). The primary vendor for the recombinant Peptide-N-Glycosidase F (PNGase F) was Oxford Glycosystems (Bedford, MA). Alternative vendors included Boehringer Mannheim (Indianapolis, IN), New England Biolabs (Beverly, MA), Calbiochem (La Jolla, CA), and Glyko (Novato, CA). The G0 standard and the enzyme β-Nacetylhexosaminidase were purchased from Oxford Glycosystems. The R1-2,3 mannosidase was supplied by New England BioLabs. The internal standard, maltoheptaose (G7), as well as maltopentaose (G5) and maltohexaose (G6) were obtained from Sigma (St. Louis, MO). 8-Aminopyrene-1,3,6-trisulfonic acid, trisodium salt (APTS), and the background electrolyte, carbohydrate separation gel buffer, were supplied by Beckman (Fullerton, CA). A solution of 1 M sodium cyanoborohydride in tetrahydrofuran was purchased from Aldrich (Milwaukee, WI). PNGase F Digestion. Typically, 300 µg of rituximab in the formulation buffer was buffer-exchanged into 40 µL of the PNGase F digestion buffer (20 mM sodium phosphate, pH 7.5, containing 50 mM EDTA, and 0.02% (w/v) sodium azide using a Microcon30 concentrator. Five units of PNGase F (10 µL) was then added (31) Gazzano-Santoro, H.; Ralph, P.; Ryskamp, T.; Chen, A.; Venkat, M. J. Immunol. Methods 1997, 202, 163-171. (32) Chiesa, C.; O’Neill, R.; Horva´th, Cs.; Oefner, P. Capillary Electrophoresis in Analytical Biotechnology; Righetti, P., Ed.; CRC Press: Boca Raton, 1996; pp 277-430. (33) O’Neill, R. A. J. Chromatogr., A 1996, 720, 201-215. (34) Takahashi, N.; Muramatsu, T. CRC Handbook of Endoglycosidases and Glycosamidases; CRC Press: Boca Raton, 1992. (35) International Conference on Harmonization, Guideline on the validation of analytical procedures: methodology. Fed. Regis. 1997, 62(96), 27464-27467.

Figure 1. Structure of the N-linked glycans present on the rituximab.

to the sample, which was then incubated for approximately 15 h at 37 °C. After the digestion, 50 µL of the 50 µg/mL internal standard solution was added to each sample. The deglycosylated protein was heated at 95 °C for 5 min and then precipitated by centrifugation at 10 000g for 10 min. The supernatant containing the oligosaccharides was dried in a centrifugal vacuum evaporator to a translucent pellet. APTS Derivatization. The pellet was reconstituted in 15 µL of 19.1 mM solution of APTS in 15% acetic acid, following which 5 µL of sodium cyanoborohydride solution was added to it. The labeling solution was kept at 55 °C for 2 h and then diluted approximately 25-fold with water prior to CE analysis. Capillary Electrophoresis. A P/ACE 5000 CE system (Beckman) equipped with a 3 mW argon-ion laser with an excitation wavelength of 488 nm and an emission band-pass filter of 520 ( 10 nm was used in the study. Coated capillaries with reduced electroosmotic flow, the eCAP N-CHO capillaries from Beckman (Fullerton, CA) or the BioCAP LPA and BioCAP XL capillaries from Bio-Rad (Hercules, CA) were used in this study. All capillaries were of 50-µm i.d. and 27/20 cm length. The capillary temperature was set at 20 °C throughout the study. The samples were introduced hydrodynamically for 8 s at 0.5 psi. The separation was performed at a constant electric field of 740 V/cm. Between the runs, the capillary was rinsed with the buffer at 20 psi for 1 min. Characterization of the Positional Isomers. Six rituximab samples were digested with PNGase F as previously described without the addition of the internal standard (IS). The supernatants containing the released oligosaccharides were pooled and divided into two parts. About 40% of the material was derivatized with APTS and further analyzed by CE as described above. The rest of the material was dried in a centrifugal vacuum evaporator and reconstituted with 40 µL of 100 mM sodium citrate phosphate, pH 5.0. Fifteen units of β-N-acetylhexosaminidase reconstituted in 10 µL of water were added to each sample and incubated at 37 °C for 18 h. Following the digestion, the solution was heated

at 95 °C for 5 min, following which it was centrifuged at 10 000g for 10 min. Again, the supernatants containing the oligosaccharides were pooled and further divided into two parts. About half of the material was derivatized with APTS and further analyzed by CE as described above. The rest of the pooled oligosaccharide solution was dried to completion in a centrifugal vacuum evaporator. One hundred and thirty five units of the enzyme R1-2,3 mannosidase was buffer-exchanged into 50 mM sodium citrate/5 mM CaCl2, pH 6.0, supplemented with 100 mg/mL of BSA; a total volume of 10 µL was added to the sample and incubated at 37 °C for 60 h. Following the digestion, the solution was heated at 95 °C for 5 min, following which it was centrifuged at 10 000g for 10 min. The supernatant containing the oligosaccharides was dried and derivatized with APTS prior to CE analysis as described above. Data Analysis. The corrected peak area (CPA) for each glycan was calculated by dividing the peak area by its own migration time. The corrected peak areas of the two G1 positional isomers were then added, and the sum was used for all subsequent calculations. The percent relative corrected peak area for each glycan is determined by dividing its corrected peak area by the sum of the peak areas of all glycans. RESULTS AND DISCUSSIONS N-Linked Glycan Distribution on Rituximab. A schematic illustration of the glycans observed on rituximab is presented in Figure 1. As seen from the figure, all glycans share the same fucosylated, branched core structure but vary in their terminal galactose occupancy: (a) a degalactosylated glycan (G0), composed of eight sugar units with no terminal galactose; (b) a partially galactosylated glycan (G1), containing nine sugar units with one terminal galactose and hence two positional isomers; and (c) a fully galactosylated glycan (G2), containing 10 sugar units having galactose on both termini. To monitor the glycan distribution in the production of rituximab, the glycans are first released enzymatically using PNGase F. The released glycans are subsequently derivatized with APTS at the reducing terminus. Analytical Chemistry, Vol. 71, No. 22, November 15, 1999

5187

Figure 3. Electropherograms of the glycans obtained from the rituximab after sequential enzymatic digestion steps: (a) PNGase F; (b) β-N-acetylhexosaminidase; (c) R 1-2,3 mannosidase. Samples were derivatized before CE analysis.

Figure 2. Typical electropherogram obtained showing (a) the glycan distribution on a sample of rituximab; (b) an expanded-scale view showing the glycans of interest.

Without further purification, the derivatization mixture containing the excess reagents and the APTS-glycan derivatives are simply diluted with water and analyzed by CE with on-column laserinduced fluorescence detection (LIF). A typical electropherogram for the analysis of the N-linked glycan distribution of rituximab is shown in Figure 2a, and the section containing the glycans of interest is expanded in Figure 2b. The multiple peaks observed between 1.5 and 2.5 min, also found in the APTS blank, are the excess APTS reagent and its impurities.24 The APTS-glycan derivatives and the internal standard (IS), migrate much later, between 3.0 and 4.5 min. The separation is performed in the negative polarity mode (anode at the detection end) using fusedsilica capillaries with a variety of hydrophilic polymeric coatings (e.g, poly(vinyl alcohol), polyacrylamide). The only requirement of the surface modification is to effectively suppress the electroosmotic flow so that the negatively charged glycans migrate with high mobility toward the anode resulting in a short analysis time. Since the APTS-glycans carry the same net charges, the separation is based on the differences in their apparent hydrodynamic sizes. Therefore, the glycans migrate in the order of increasing size (G0 < G1 < G2) and are baseline resolved from each other. In addition, the two G1 positional isomers are separated with baseline resolution. The underlying mechanism for the separation of the two G1 positional isomers with the same molecular weight is based on the difference in their apparent hydrodynamic sizes resulting from the asymmetrical nature of the fucosylated branched core structure. Similar observation was reported for the APTS deriva5188 Analytical Chemistry, Vol. 71, No. 22, November 15, 1999

tives of two positional isomers of maltotetraose24 and two isomers of trisialylated triantennary glycans from bovine fetuin.36 Characterization of Positional Isomers. The two G1 positional isomers were identified on the basis of CE profiles obtained from sequential enzymatic digestions. First, the rituximab was incubated with PNGase F to release the N-linked oligosaccharides as shown in Figure 3a. The underivatized glycans were further incubated with β-N-acetylhexosaminidase to remove the terminal N-acetylglucosamine (GlcNAc) residue(s) on the G0 and G1 isomers as illustrated in Figure 3b. Since there is no exposed GlcNAc on G2, its migration time remained the same as in Figure 3a. On the other hand, the migration time change was bigger for the G0 than for the G1 isomers as expected because of the removal of two sugar units from G0 as opposed to one from G1. Finally, the exposed mannose residues on the G0 and G1 isomers, which are linked to the mannose in the core structure through either a Man(R1,3) or Man(R1,6) linkage, were subject to enzymatic cleavage by incubating with R1-2,3 mannosidase. This enzyme37 has been shown to specifically catalyze the hydrolysis of the terminal Man(R1,3) linkage and not the Man(R1,6) linkage. The glycans digested with this mannosidase were labeled with APTS and analyzed by CE as shown in Figure 3c. Only the exposed mannose with a Man(R1,3) linkage can be removed from the glycan; hence, this structure has a change in its migration time when compared with the profile in Figure 3b. Therefore, the predominant form of the N-linked oligosaccharides on rituximab has a galactose on the mannose (R1,6) arm. Mobility Shift. It is well-known that CE is a separation technique with high resolution because the electrophoretic mobil(36) Guttman, A.; Chen, F.-T. A.; Evangelista, R. A. Electrophoresis 1996, 17, 412-417. (37) Wong-Madden, S. T.; Landry, D. Glycobiology 1995, 5, 19-28.

Table 1. Electrophoretic Mobility for Different Oligosaccharide Structures

ity is influenced not only by molecular charges but also by molecular sizes. Hence, various carbohydrate structures with differences in their branching or positional isomers are readily resolved by CE.24,36,38 Under a given set of experimental conditions, the electrophoretic mobility is a function of the specific carbohydrate adduct. In addition, the mobility differences among various carbohydrate structures were found to be related to the exact nature of the individual monosaccharide.38 In our case, the electrophoretic mobilities of the glycans observed from various enzymatic digestions shown in Figure 3a-c were calculated and listed in Table 1. Also listed are the number of sugar units and molecular weight of each glycan. As the molecular charges on all APTS derivatized glycans are about the same, the mobilities of the glycans should decrease with increasing size or molecular weight. As seen from Table 1, indeed the electrophoretic mobilities decrease with increasing numbers of monosaccharide units or molecular weights of the glycans. Moreover, glycans with the exact same number of monosaccharide units or even the exact same molecular weight have different mobilities due to their different hydrodynamic sizes. For instance, the two G1 isomers have the exact same molecular weight, yet they have very different mobilities as shown earlier. In the cases of G0, G1(R1,3)-Gn, and G1(R1,6)-Gn, these structures all have eight monosaccharide units with the exact same core structure and about the same molecular weight. Among the three glycans, G0 has the most compact structure with one sugar residue on both antennae; whereas G1(R1,3)-Gn and G1(R1,6)-Gn both have sugar residues on the same antennae, resulting in a more elongated structure. As, G1(R1,3)-Gn has the extended antennae (38) Nashabeh, W.; El Rassi, Z. J. Chromatogr. 1992, 600, 279-289.

on the opposite side of the fucose, it has the largest hydrodynamic size. Therefore, G0 has the largest mobility and G1(R1,3)-Gn has the smallest mobility. Table 1 also lists the mobility differences between G2 and G0 and between G0 and G0-2Gn as 0.97 and 1.11 × 10-9 m2/Vs, respectively. The 10% difference is believed to be caused by the molecular sizes of the GlcNAc and Gal residues. Since GlcNAc is slightly larger in its hydrodynamic size than Gal, the removal of two terminal GlcNAc residues from G0 is expected to cause a larger increase in the mobility than that of two terminal galactose residues from G2. The examples illustrated here suggest that the electrophoretic mobility is an effective parameter in the characterization of carbohydrate structures typically observed in the biosynthesis of therapeutic biologics in CHO cells. Assay Development. The assay described here is intended for the quantitative determination of the relative distribution of N-linked glycans on rituximab. Therefore, two essential assumptions need to be verified. First, the extent of enzymatic release of different glycans is the same, i.e., the enzyme shows no bias in the release of the various glycans. Second, the reductive amination of the various glycans with APTS has the same absolute yield under the experimental conditions used, that is, the derivatization shows no bias toward a specific glycan. The validity of these assumptions is discussed in the following sections. PNGase F Digestion. PNGase F hydrolyzes the β-aspartylglycosylamine bond between the Asn-301 residue and the innermost N-acetylglucosamine of the glycan. The released N-acetylglucosamine is spontaneously hydrolyzed to form ammonia and free oligosaccharide chains with intact reducing termini. One of the key parameters in ensuring a complete homogeneity of glycan release is the enzyme digestion time. Rituximab samples were digested with PNGase F as described before for various times to as much as 24 h. A constant amount of the internal standard (IS) was added to each digest before labeling to correct for slight variation in the labeling and injection. The corrected peak area (CPA) ratios of each the three glycans to that of the IS are plotted in Figure 4a as a function of the digestion time. It can be seen that the CPA increases initially with digestion time as more glycans are being released from the protein. After about 8 h of incubation, the release reached a plateau. The addition of a second enzyme spike or the use of an increased initial enzyme concentration did not further increase the amount of released glycans. The digested rituximab was further analyzed by a peptide map to verify the extent of the reaction and indeed there is no oligosaccharide left at the site (Mike Mulkerrin, personal communication). Therefore, the digestion was believed to be complete after 8 h under the conditions investigated here. A more pertinent issue for this study is that the enzyme exhibits no bias toward the release of any glycan structures. The relative glycan distributions in terms of the percent relative corrected peak areas of G0, G1, and G2 were also plotted versus the digestion time in Figure 4b. It is clear from the figure that the percent relative corrected peak areas for all glycans remained constant over the entire digestion time course. Further analyses by MALDI-TOF were conducted to verify that the PNGase F activity indeed gives a constant rate of cleavage for the glycans.39 (39) Papac, D.; Briggs, J. B.; Chin, E. T.; Jones, A. J. S. Glycobiology 1998, 8, 445-454.

Analytical Chemistry, Vol. 71, No. 22, November 15, 1999

5189

Figure 4. Time course of PNGase F digestion of rituximab. (a) The corrected peak area ratios of (0) G0, (b) G1, and (3) G2 to the IS at various digestion times. (b) The percent relative corrected peak areas of (0) G0, (b) G1, and (3) G2 obtained at the digestion times evaluated.

APTS Labeling. The coupling of the carbonyl group, available only at the reducing end of the glycans, with the primary amine on APTS renders the stoichiometry of the labeling reaction such that only one fluorophore molecule is attached to each glycan. This makes the different glycans equally sensitive to optical modes of detection such as UV or fluorescence and is a prerequisite for the accurate estimation of the molar proportions of these species. The validity of this assumption was assessed using a set of commercial maltooligosaccharide standards, G5, G6, and G7. Each standard with concentrations at 0.25, 0.50, 1.0, and 2.0 nmol was derivatized with APTS as described in the Experimental Section. As expected, equimolar amounts of all standards gave equivalent fluorescent responses. However, there was a constant 3-fold reduction in the absolute response when the labeling reaction was done in the protein digest matrix versus in neat water over the 10-fold concentration range investigated here. This reduction in the labeling yield is likely due to interferences from the salts accumulated through the various sample treatment steps prior to labeling. These salts may affect, among other things, the effective pH of the labeling reaction. To further assess this effect, rituximab samples were digested with PNGase F and labeled with APTS for various times from 0.5 to 2.5 h. The CPA values of G0, G1, and G2 were plotted as a function of the labeling time in Figure 5a. Clearly, the absolute fluorescent response of the glycans increased with time, and under the conditions studied here the reaction did not reach completion. Reductive amination is generally a slow reaction, and ways to 5190 Analytical Chemistry, Vol. 71, No. 22, November 15, 1999

Figure 5. Time course of APTS labeling of rituximab. (a) The corrected peak area of (0) G0, (b) G1, and (3) G2 released at various labeling times. (b) The percent relative corrected peak areas of (0) G0, (b) G1, and (3) G2 obtained at the labeling times evaluated.

improve the reaction yields, as have been extensively studied in the laboratories of Hase27,40,41 and Honda,42,43 include reducing the reaction volume (i.e., increasing reactant concentration), increasing the reaction time, and increasing the temperature. Whereas increased labeling efficiency was observed by reducing the labeling reaction volume from 20 to 5 µL, the manipulation of such small volumes lead to greater variability in the fluorescent response and thus was deemed unacceptable for routine lot release testing. The other means to improving the labeling efficiency is overnight labeling. Yet again, this was not considered a viable option in practice as it will significantly lengthen the assay to more than 48 h. These studies serve as a caveat to the fact that the reductive amination of carbohydrates is sensitive to various interfering matrix components, and there is definitely the need to control for these parameters especially when absolute quantitation is desired. Nevertheless, as the intended purpose of this assay is to determine the relative rather than the absolute molar concentration of the N-linked glycans on rituximab, only equivalence in the labeling efficiency of the glycans at a given time point is necessary regardless of the absolute yield. This is demonstrated (40) Hase, S.; Ibuki, T.; Ikenaka, T. J. Biochemistry 1984, 95, 197-203. (41) Hase, S. J. Chromatogr., A 1996, 720, 173-182. (42) Honda, S.; Iwase, S.; Makino, A.; Fujiwara, S. Anal. Biochem. 1989, 176, 72-77. (43) Kakehi, K.; Honda, S. J. Chromatogr., A 1996, 720, 377-393.

Table 3. Recovery by Sample Mixinga

Table 2. Recovery by G0 Standard Addition theoretical total amt of G0 recovered G0 measured over background amt of G0 a added (nmol) by CE (nmol) (nmol) 0 0.25 0.75 1.00 1.50

2.35b 2.61 3.07 3.37 3.85

percent recovery

0.258 0.716 1.02 1.49 mean RSD (%)

103 95.5 102 99.5 100 3.4

a The total amount of G0 was determined from a calibration curve of G0/IS (internal standard) versus G0 standards established in the same protein matrix, with G0 ranging from 0.25 to 4.0 nmol. b This value is the mean result from two sample preparations.

in Figure 5b in which it can be seen that the percent CPA values of G0, G1, and G2 were indeed constant throughout the labeling time course regardless of the actual labeling yields. Assay Performance. This assay is intended for the quantitative determination of the relative distribution of N-linked glycans on rituximab in a routine lot release setting. Therefore, the assay performance requirements are stringent especially in terms of its accuracy and precision as defined by the ICH guidelines. Accuracy. The accuracy of the assay was evaluated by recovery of known amounts of G0 standard. A set of purified G0 standards were added to rituximab samples without PNGase F digestion to eliminate any matrix effects in the labeling with APTS prior to CE analysis. A calibration curve based on the corrected peak area ratio of G0/IS was established for G0 in the range from 0.25 to 4 nmol. Another set of rituximab samples were digested with PNGase F and spiked with G0 standards in the concentration range from 0.25 to 1.5 nmol. On the basis of the corrected peak area ratio of G0/IS released from the rituximab sample used in this study, the amounts of G0 added to the rituximab sample are approximately in the range from 10 to 60% of the actual amount of G0 present on that particular sample. The samples were analyzed as previously described and the recoveries, determined by standard addition, as listed in Table 2, range from 95.5 to 103% with a mean recovery of 100%. As the purpose of the assay is to quantitate the relative glycan distribution, the assay performance in terms of its ability to differentiate minute differences in the relative glycan distribution of rituximab is crucial. To address this, two samples of rituximab of the same protein concentration but different glycan distribution were used to prepare admixtures with percent G0 values ranging from approximately 45 to 80%. The protein mixtures were digested with PNGase F and analyzed accordingly. The recoveries determined from the %CPA ratio of the experimental to that of the theoretical are summarized in Table 3. The data demonstrated an excellent accuracy, as the recoveries ranged from 100 to 102%. The performance of the assay is further demonstrated by its accuracy in determination of the glycan distribution with different initial protein concentrations. Rituximab samples were bufferexchanged into the incubation buffer and then serially diluted to yield initial protein concentrations ranging from 0.25 to 4.0 nmol. The samples were analyzed, and the recoveries were calculated on the basis of the corrected peak area ratio of G0/IS obtained for the 2.0-nmol sample. As shown in Table 4, the mean recovery is 103% over the 16-fold protein concentration range.

total vol of protein in the mixture (µL) lot A

lot B

200 160 135 120 110 95 80 55 0

0 40 65 80 90 105 120 145 200

percent relative corrected peak area of G0

percent recovery

theoreticalb experimental 51.1 55.8 58.7 60.5 63.4 66.2 70.9

43.6c 52.2 56.5 59.6 61.2 63.9 66.8 71.1 81.2c

102 101 102 101 101 101 100 mean RSD (%)

101 0.5

a Mixtures of rituximab lots A and B were prepared. The protein concentration of both lots prior to mixing were determined spectrophotometrically. Lot B was diluted with formulation buffer to yield the same protein concentration as that of lot A. b The theoretical values for the mixtures were calculated on the basis of the values for each lot as explained in footnote c (below) and the appropriate volume. c Three replicates each of both rituximab lots were analyzed, and the mean value of the percent relative corrected peak area of G0 from the three sample preparations is reported.

Table 4. Recovery by Sample Dilution corrected peak area ratio, G0/IS

initial protein concna (nmol)

theoretical

experimental

0.25 0.50 1.0 2.0 4.0

0.105 0.210 0.420 0.841 1.68

0.116 0.221 0.434 0.841 1.60

percent recoveryb 110 105 103 mean RSD (%)

95.1 103 5.5

a Following buffer exchange, rituximab was serially diluted with the formulation buffer to yield the initial protein concentrations shown. b Recoveries were calculated on the basis of the value obtained with the 2.0-nmol sample.

Precision. Precision is another important parameter in judging the performance of the assay for routine lot release testing of a drug product. In this case, precision was evaluated in three parts, by repeatability, intermediate precision, and reproducibility as defined by the ICH guidelines. Since reproducibility, by definition, involves interlaboratory studies at different companies, it truly represent the utmost precision of any assay. In this case, the overall precision was better than 0.2, 0.1, and 0.9% for G0, G1, and G2, respectively. Throughout the precision studies, no difference in performance was observed when differently coated capillaries were used. Clearly, the assay described here is suitable for use in the characterization as well as routine lot release testing on the basis of its performance in terms of the excellent accuracy and precision demonstrated here. CONCLUSIONS A CE/LIF method was developed for determination of the relative glycan distribution of the monoclonal antibody, rituximab. The method is simple, accurate, precise, and suitably robust for routine bulk product testing. Although this method reported was developed and validated specifically for rituximab, it can be applied Analytical Chemistry, Vol. 71, No. 22, November 15, 1999

5191

for studies on N-linked carbohydrates in general, including other recombinant monoclonal antibodies or glycoproteins.

discussions. The authors would also like to acknowledge Mike Mulkerrin and Eleanor Canova-Davis for their comments.

ACKNOWLEDGMENT

Received for review April 9, 1999. Accepted August 4, 1999.

The authors gratefully acknowledge Lori Schalk for some of the experimental results and Glen Teshima for valuable

5192

Analytical Chemistry, Vol. 71, No. 22, November 15, 1999

AC990376Z