Genetically engineered deglycosylation of the variable domain increases the affinity of an anti-CD33 monoclonal antibody

0161-5890/93 $6.00 + 0.00 0 1993 Pergamon Press Ltd

Molecular Immunology, Vol. 30, No. 15, pp. 1361-1367, 1993 Printed in Great Britain.

GENETICALLY ENGINEERED DEGLYCOSYLATION VARIABLE DOMAIN INCREASES THE AFFINITY ANTI-CD33 MONOCLONAL ANTIBODY MAN SUNG

OF THE OF AN

Co,*$ DAVID A. SCHEINBERG,~ NEVENKA M. AVDALOVIC,* MAX VASQUEZ,* PHILIP C. CARON~ and CARY QUEEN*

KIMBERLY MCGRAW,?

*Protein Design Labs, Mountain View, CA 94043, U.S.A.; and TMemorial Sloan-Kettering Center, New York, NY 10021, U.S.A.

Cancer

(Received 9 March 1993; accepted 15 April 1993) Abstract-Ml95 is a murine monoclonal antibody that binds to the CD33 antigen and is being tested for the treatment of myeloid leukemia. Surprisingly, a complementarity determining region (CDR)-grafted, humanized Ml95 antibody displayed a several-fold higher binding affinity for the CD33 antigen than the original murine antibody. Here we show that the increase in binding affinity resulted from eliminating an N-linked glycosylation site at residue 73 in the heavy chain variable region in the course of humanization. Re-introducing the glycosylation site in the humanized antibody reduces its binding affinity to that of the murine antibody, while removing the glycosylation site from the murine MI95 variable domain increases its affinity. The removal of variable region carbohydrates may provide a method for increasing the affinity of certain monoclonal antibodies with diagnostic and therapeutic potential.

INTRODUCTION

of immunoglobulins has been shown to have significant effects on their effector functions, stability and serum half-life (Nose and Wigzell, 1983; Tao and Morrison, 1989). The carbohydrate groups responsible for these properties are generally attached to the C regions of the antibodies. For example, glycosylation of IgG at asparagine 297 in the Cr.,2 domain is required for the ability of IgG to activate the classical pathway of complement-de~ndent cytolysis (Tao and Morrison, 1989). Glycosylation of immunoglobulins in the V region has also been observed. In fact, about 30% of the human antibodies examined were glycosylated in the V domain (Abel et al., 1968). However, glycosylation of the V domain is believed to arise from fortuitous occurrences of the N-linked glycosylation signal Asn-XSer/Thr in the V region sequence (Sox and Hood, 1970), and is not generally thought to play an important role in antibody function. Ml95 is a murine monoclonal IgG2a antibody reactive with the CD33 antigen (Tanimoto et al., 1989; Scheinberg ef al., 1989). CD33 is expressed on early myeloid progenitor cells, some monocytes, and the cells of most myeloid leukemias, but not on the earliest hematopoietic stem cells (Griffen et al., 1984; Andrews et al., 1986). The efficient cellular binding and internalization of Ml95 has allowed use of the radiolabeled Gly~osylation

SAuthor to whom correspondence should be addressed at: Protein Design Labs, 2375 Garcia Ave, Mountain View, CA 94043, U.S.A. Abbreviations: AML, acute myelogenous leukemia; HAMA, human anti-mouse antibody; CHO, carbohydrate; CDR, complementarity-determining region.

antibody in clinical trials for acute myelogenous leukemia (AML) (Scheinberg et al., 1991). The murine Ml95 antibody, however, does not kill leukemic cells by complement-dependent cytotoxicity with human complement, or by antibody-dependent cellular cytotoxicity with human effector cells. The human anti-mouse antibody (HAMA) response may also preclude long term use of the murine antibody in patients. To increase the effector function and reduce the immunogenicity of the Ml95 antibody in human patients, we constructed chimeric and humanized IgGl versions of the antibody (Co et al., 1992). The chimeric antibody combines the murine Ml95 V region with a human C region, while the humanized antibody combines the CDRs of murine Ml95 with a human antibody V region framework and human C region. In addition, a computer model of the Ml95 V domain was used to identify several residues in the murine framework that may interact with the CDRs, and these residues were also retained in the humanized antibody. The chimeric and humanized Ml95 antibodies exhibited improved effector functions, as expected, but the humanized antibody also showed an unexpected increase in binding affinity to the CD33 antigen, measured in various assays as three- to eight-fold higher than the murine and chimeric forms (Co et at., 1992; Caron et al., 1992). Here we determine the cause for the increased affinity of the humanized Ml95 antibody.

MATERIALS AND METHODS Construction of antibody variants

The construction of chimeric and humanized Ml95 antibodies has been reported (Co et al., 1992). To

1361

MAN SUNGCo et al.

1362

78

70 Chlmerlc

Thr

Val

Asp

lian

Ser

Sgf. Ser

Thr

Ala

Chlmerlc - CHO

Thr

Val

Asp

Gln

Ser

Ser

Ser

Thr

Ala

Humanized

Thr

Ala

Asp

J&n

Ser

Ser

Ser

Thr

Ala

Thr

Ala

Asp

Glu

Ser

Thr

Am

Thr

Ala

+ CHO

Humanized

Fig, 1. Amino acids 70-78 of the V, domain of the chimeric and humanized M 195 antibodies, with and without N-linked glycosylation sites (CHO = carbohydrate). The amino acids are numbered according to Kabat ef al. (1991). The glycosylation sites at ammo acids positions 73-75 are underlined.

into the Vu region of the humanized antibody, the sequence in position 73-76 was changed from Glu-Ser-Thr-Asn to the sequence Asn-Ser-Ser-Ser that occurs in the chimeric Vi.,region (Fig. 1). Residues 73-75 represent the Asn-X-Ser/Thr glycosylation signal, while residue 76 was replaced in case the amino acid immediately after the glycosylation site affects the extent of glycosylation. These amino acid alterations were achieved by site-directed mutagenesis of the respective genes. Genetic engineering was used in preference to enzymatic deglycosylation or metabolic inhibitors of glycosylation, because the putative V region carboAJinity measurements hydrate could be removed completely without also removing the C region carbohydrate or otherwise altering Murine M 19.5antibody was labeled with 12’I-Na using the antibody. chloramine-T, to 2-10 pCi/,ug protein. Relative affinity The purified antibody variants were analyzed by of the various Ml95 constructs was measured by comSDS-PAGE under reducing conditions {Fig. 2). The petitive binding with the ‘25I-M195 antibody. Specifi25 kD light chains of the chimeric and humanized antically, increasing amounts of cold competitor antibody bodies migrate slightly differently because of the differwere incubated with 2 x lo5 HL60 cells and 50 ng lz51ing compositions of their V, domains. The heavy chains Ml95 in 200 ,ul RPM1 plus 2% human serum for 1 hr at of the forms of the chimeric and humanized antibodies 0°C. Cells were washed three times in RPMI and with potential Vu glycosylation sites (Fig. 2, lanes 2 and counted. 4) corn&rate with the murine heavy chain (lane I), while the chimeric and humanized Ml95 antibody variants with the putative V, glycosylation site respectively removed and added, the genes for their respective V, regions were modified by site-directed mutagenesis using the polymerase chain reaction. After the modifications were verified by sequencing, the modified genes were inserted in the pVg1 expression vector and transfected into Sp2/0 cells together with the respective light chain containing vectors, as described {Co et al., 1992). Antibody-producing clones were selected, and antibody purified by protein A chromatography. construct

RESULTS AND DISCUSSION Since the only difference between the chimeric and humanized antibodies is the amino acid sequence of the V domain, we focused our analysis of the affinity difference on this region. Examination of the sequence of the murine (or chimeric) heavy chain V region revealed that it contains the amino acid sequence Asn-Ser-Ser starting at position 73 (Fig. 1), which is an example of the Asn-X-Ser/Thr recognition sequence for N-linked glycosylation. Position 73 is located in the V, framework between the second and third CDRs. In contrast, the humanized Vu region, which utilizes the framework of the human Eu antibody (Co et al., 1992), does not have this or any Asn-X-Ser/Thr ~ycosylation site (Fig. 1). Most, but not all, occurrences of Asn-X-Ser/Thr in glycoproteins are actually glycosylated (Gavel and von Heijne, 1990). To verify that glycosylation at Asn 73 occurs and to determine whether it affects the antibody binding affinity, we decided to remove this potential glycosylation site from the chimeric Ml95 antibody and to introduce a similar site into the humanized antibody. To remove the site from the Vu region of the chimeric antibody, the Asn codon at position 73 was changed to a Gln codon. To introduce a potential glycosylation site

50 kD

25 kD

Fig. 2. SDS-PAGE analysis of purified Ml95 antibody variants. Lane 1, murine antibody; lane 2, humanized antibody with glycosylation site added; lane 3, humanized antibody; lane 4, chimeric antibody; lane 5, chimeric antibody with glycosylation site removed. The antibodies were analyzed by 10% SDS-PAGE after reduction.

1363

Increase of antibody affinity by V region deglycosylation

-

Murine

-t-

Chime&

-

Chime& - CHO

-

Humanized + CHO

-

Humanized

01

.l

.Ol

1

10

Competitorconcentration(us/ml)

loo-

ih* B

so Z i

60-

E e40 $ 20

0

.Ol

.l

Competitor

1

concentration(@ml)

loo-

D

20 -

\ 01 .Ol

.l

1

Competitor concentration (ug/ml)

.Ol

.l

1

10

Competitor concentration (@ml)

Fig. 3. Competitive binding of Ml95 antibodies to HL60 cells. (A) Chimeric and murine, (B) humanized and murine, (C) humanized with glycosylation site added, and murine, (D) chimeric with glycosylation site removed, and chimeric, (E) chimeric with glycosylation site removed, and humanized. Increasing amounts of cold competitor antibody were incubated with HL60 cells and L251-labeledmurine Ml95 Percent bound is relative to iZ51-Ml95 bound to the cells in the absence of competitor antibody. Each point is the average of duplicate samples. Panel B is based on data previously described (Caron et al., 1992).

the heavy chains

of the forms without potential V, glycosylation sites migrate slightly faster (lanes 3 and 5). The only amino acid differences between the two forms of the chimer-k antibodies, and respectively between the

two forms of the humanized antibodies, are the changes we introduced at the glycosylation site. Since these minor amino acid changes would not in themselves be expected to alter the apparent molecular weights, the slower mobility of the forms with the glycosylation sites must be due to increased molecular weight provided by an

attached carbohydrate. Moreover, for the three heavy chains with the V, glycosylation site (lanes 1, 2 and 4), there is a faint lower band co-migrating with the heavy chains without the site (lanes 3 and 5), suggesting that a fraction of the heavy chain in these antibodies (l&20%) is not glycosylated at Asn 73. Heavy chain heterogeneity in SDS-PAGE analysis of monoclonal antibodies has been observed previously, so our results show that in at least some cases such heterogeneity is due to glycosylation of the V, domain.

Direct binding of iodinated antibodies to determine affinity constants may be inaccurate, due to iodine atoms introduced into the binding region or denaturation during radiolabeling. Therefore, to compare the binding affinities of the various antibody constructs, the unlabeled antibodies were allowed to compete with iodinated murine Ml95 for binding to HL60 cells, which express the CD33 antigen. Human serum, containing human IgG, was present in the reactions to inhibit non-specific and Fc receptor binding. The chimeric M 195 antibody competes with the same efficiency as murine Ml95 (Fig. 3A), showing that it has the same affinity for antigen. This is consistent with expectation, since the chimeric antibody has the same V domain as the murine antibody. However, the humanized Ml95 antibody competed much more effectively than the murine antibody (Fig. 3B), as has been seen previously (Caron et ul., 1992). When a glycosylation site was introduced into the humanized Ml95 V, region, its increased affinity was lost, as the humanized antibody with glycosylation site competed no better than the murine antibody (Fig. 3C). Conversely, the chimeric antibody from which the V, glycosylation site had been removed competed more efficiently than the original chimeric Ml95 antibody (Fig. 3D). Consistent with these results, the chimeric antibody without the V, glycosylation site competed about as well as the humanized antibody (Fig. 3E), showing that elimination of glycosylation at Asn 73 is sufficient to account for the increased affinity of the humanized antibody. The results here contrast with a recent report that glycosylation at CDR2 of the heavy chain, in the antigen binding site, of a murine antibody specific for a(l-6)dextran increases its affinity for dextran (Wallick et al., 1988; Wright et al., 1991). To determine how glycosyfation of the V, region of the Ml95 antibody might decrease binding affinity, we constructed a molecular model (Fig. 4). For this purpose, a model of the murine M 195 V domain without glycosylation was first created using the ABMOD and ENCAD computer programs (Zilber et al.,1990). A carbohydrate group was then added to the model at Asn 73, assuming that the composition and conformation of the group is the same as that of the carbohydrate group attached to C,2 of human IgGl, which has been determined by X-ray crystallography (Deisenhofer, 1981). From the model, it is clear that the bulky carbohydrate group could easily interfere with binding of the antibody to antigen by steric hindrance (Fig. 4), and this interference is largely independent of the exact composition and conformation of the carbohydrate group. However, it cannot be ruled out that the carbohydrate group acts by altering the conformation of the nearby CDRs, which in turn reduces affinity for antigen. The murine Ml95 antibody has been tested in two clinical trials for AML. An initial trial showed that small doses of Ml95 localized rapidly and specifically to the leukemic cells in the blood and bone marrow (Scheinberg et al., 1991). However, the mouse antibody did not kill the leukemic cells and provoked a HAMA response

in several patients. More recently, in a second trial. 13’I-labeled Ml95 was capable of killing large numbers of leukemic cells (greater than 99%) in relapsed and refractory AML patients (Schwartz et al., 1991). Destruction of bystander bone marrow cells because of the long path length of the 13’1emissions remains a problem for this approach to specific leukemia therapy, however. The increased affinity of the humanized Ml95 antibody, due to removal of the glycosylation site, as well as its improved effector function (Caron et al., 1992) and potentially lower immunogenicity relative to the murine Ml95 antibody, may eliminate the need for radiolabeling and allow repeated doses in future therapeutic trials. Computer search of a data base (Kabat et al., 1991) revealed that 141 out of 808 essentially complete murine V, sequences, or 17.5%, contained an Asn-X-Ser/Thr recognition site for N-linked glycosylation. Many members of the murine antibody V, subgroup IIIB encoding a glycosylation signal in the second CDR are specific for carbohydrate (Wright et al., 1991). Many others such as M 195 contain a glycosylation signal in the framework region. The significance of such glycosylation is not clear. Here we show that, for the example of M195, removal of the glycosylation site can increase the binding affinity of such an antibody, although this will not be true in all cases. Increased affinity may in turn enhance sensitivity in a diagnostic assay, improve the effector functions of an antibody, or reduce the concentration needed to inhibit function of a target such as a cellular receptor, thereby reducing the dose needed to achieve desired targeting and saturation in human patients. Our work therefore provides a potential method to increase the efficacy and reduce the required dose of other antibodies of medical importance. Acknowledgements-D.

A. Scheinberg is a Lucille P. Markey Scholar. This work was supported in part by the Lucille P. Markey Trust and NIH ROlCA55349.

REFERENCES Abel C. A., Spiegelberg H. L. and Grey H. M. (1968) The carbohydrate content of fragments and polypeptide chains of human ?G-myeloma proteins of different heavy-chain subclasses. ~~oc~ernistr~ 7, 1271-1278. Andrews R. G., Takahashi M., SegaI G. M., Powell J. S., Bernstein I. D. and Singer J. W. (1986) The L4F3 antigen is expressed by unipotent and multipotent colony-forming cells but not by their precursors. Blood 68, 1030-1035. Caron P. C., Co M. S., Bull M. K., Avdalovic N. M., Queen C. and Scheinberg D. A. (1992) Biological and immunological features of humanized Ml95 (Anti-CD33) monoclonal antibodies. Cancer Res. 52, 6761-6767 Co M. S., Avdalovic N. M., Caron P. C., Avdalovic M. V., Scheinberg D. A. and Queen C. (1992) Chimeric and humanized antibodies with specificity for the CD33 antigen. J. rmmunoI. 148, 1149-l 154. Deisenhofer J. (1981) Crystallographic refinement and atomic models of a human Fe fragment and its complex with fragment B of protein A from Staphylococcus aureus at

Increase of antibody affinity by V region deglycosylation

Fig. 4. Computer model of the murine Ml95 antibody variable domain. The framework region is in white, the CDRs in red, the N-linked glycosylation recognition sequence in magenta, and the attached carbohydrate in blue.

1365

Increase of antibody affinity by V region deglycosylation 2.9- and 2.8-Angstroms resolution. Biochemistry 20, 2361-2370. Gavel Y. and von Heijne G. (1990) Sequence differences between glycosylated and non-glyosylated Asn-X-Thr/Ser acceptor sites: implications for protein engineering. Protein Engng 3, 433442. Griffen J. D., Linch D., Sabbath K., Larcom P. and Schlossman S. F. (1984) A monoclonal antibody reactive with normal and leukemic human myeloid progenitor cells. Leukemia Res. 8, 521-534. Kabat E. A., Wu T. T., Perry H. M., Gottesman K. S. and Foeller C. (1991) Sequences of Proteins of Immunological Interest, Vol. I, 5th U.S. Edn. Department of Health and Human Services, Washington. Nose M. and Wigzell H. (1983) Biological significance of carbohydrate chains on monclonal antibodies. Proc. natn. Acad. Sci. U.S.A. 80, 6632-6636. Scheinberg D. A., Tanimoto M., McKenzie S., Strife A., Old L. J. and Clarkson B. D. (1989) Monoclonal antibody M195: A diagnostic marker for acute myelogenous leukemia. Leukemia 3, 4W445. Scheinberg D. A., Lovett D., Divgi C. R., Graham M. C., Berman E., Pentlow K., Feirt N., Finn R. D., Clarkson B. C., Gee T. S., Larson S. M., Oettgen H. F. and Old L. J. (1991) A phase I trial of monoclonal antibody Ml95 in acute myelogenous leukemia: specific bone marrow targeting and internalization of radionuclide. J. Clin. Oncol. 9, 478490.

1367

Schwartz M. A., Lovett D. R., Redner A., Divgi C. R., Graham M. C., Finn R., Gee T. S., Andreeff M., Oettgen H. F., Larson S. M., Old L. J. and Scheinberg D. A. (1991) Leukemia cytoreduction and marrow ablation after therapy with ‘311-labeled monoclonal antibody Ml95 for acute myelogenous leukemia (AML). Proc. Am. Sot. Clin. Oncol. 10, 230. Sox H. C. and Hood L. (1970) Attachment of carbohydrate to the variable region of myeloma immunoglobulin light chains. Proc. natn. Acad. Sci. U.S.A. 66, 975-979. Tanimoto M., Scheinberg D. A., Cordon-Cardo C., Huie D., Clarkson B. D. and Old L. J. (1989) Restricted expression of an early myeloid and monocytic cell surface antigen defined by monoclonal antibody M195. Leukemia 3, 339-348. Tao M.-h. and Morrison S. L. (1989) Studies of aglycosylated chimeric mouse-human IgG. J. Immunol. 143, 2595-2601. Wallick S. C., Kabat E. A. and Morrison S. L. (1988) Glycosylation of a V, residue of a monoclonal antibody against ~(1-6) dextran increases its affinity for antigen. J. Exp. Med. 168, 1099-1109. Wright A., Tao M.-h., Kabat E. A. and Morrison S. L. (1991) Antibody variable region glycosylation: position effects on antigen binding and carbohydrate structure. EMBO J. 10, 2717-2723. Zilber B., Scherf T., Levitt M. and Anglister J. (1990) NMRderived model for a peptide-antibody complex. Biochemistry 29, 10,032-10,041.