Report

Variants of the Antibody Herceptin That Interact with HER2 and VEGF at the Antigen Binding Site

See allHide authors and affiliations

Science  20 Mar 2009:
Vol. 323, Issue 5921, pp. 1610-1614
DOI: 10.1126/science.1165480

Abstract

The interface between antibody and antigen is often depicted as a lock and key, suggesting that an antibody surface can accommodate only one antigen. Here, we describe an antibody with an antigen binding site that binds two distinct proteins with high affinity. We isolated a variant of Herceptin, a therapeutic monoclonal antibody that binds the human epidermal growth factor receptor 2 (HER2), on the basis of its ability to simultaneously interact with vascular endothelial growth factor (VEGF). Crystallographic and mutagenesis studies revealed that distinct amino acids of this antibody, called bH1, engage HER2 and VEGF energetically, but there is extensive overlap between the antibody surface areas contacting the two antigens. An affinity-improved version of bH1 inhibits both HER2- and VEGF-mediated cell proliferation in vitro and tumor progression in mouse models. Such “two-in-one” antibodies challenge the monoclonal antibody paradigm of one binding site, one antigen. They could also provide new opportunities for antibody-based therapy.

The binding of antibodies to specific single antigens has prompted their use for numerous targeted therapies (1). However, the notion of an antibody that recognizes more than one antigen is intriguing. It has been suggested that multi-specificity may evolve and play a role in the highly efficient antibody repertoire for immune protection with one antibody performing more than one task (2, 3). However, the few examples of such multi-specificity are limited to antibodies that bind small haptens (4), and a strategy to generate a single antigen binding fragment (Fab) capable of recognizing two unrelated proteins has not yet been reported.

We set out to explore whether dual specific antibodies can be derived from a monospecific antibody with the following approach: A repertoire of Herceptin (Genentech, South San Francisco, CA) antibody variants with mutations in the light chain (LC) complementarity determining regions (CDRs) were generated, and Fabs that can bind a new protein antigen while maintaining human epidermal growth factor receptor 2 (HER2) binding were identified. The approach is based on the understanding that modifications of the LC sequence can modulate the binding specificity of antibodies (5, 6). In addition, many antibodies, with Herceptin as a prime example (7, 8), bind the antigens by using mainly the heavy chain (HC) CDRs, suggesting that mutations in the LC CDRs might allow preservation of the original antigen binding specificity.

The framework regions of Herceptin variable domains (VH and VL, the variable domain of HC and LC, respectively) belong to subtypes that are prevalent in the human antibody repertoire (VH3, VLkappa1). Thus, mutations within the antigen binding site of Herceptin that confer a second specificity may indicate a potential for a dual specific antibody to evolve from the natural repertoire. As Herceptin is a validated therapeutic for breast cancers that overexpress HER2 (9), recruitment of a second binding specificity to Herceptin may add to it a distinct pharmacological activity.

We constructed the repertoire of Herceptin LC variants by randomizing a subset of solvent exposed LC CDR positions to mimic the natural diversity in amino acid composition and length (tables S1 and S2) (10). Selection of the Herceptin variant library against vascular endothelial growth factor (VEGF), death receptor 5 (DR5), and the complement binding fragment of immunoglobulin G (IgG) (Fc) generated over one hundred specific clones containing 3 to 17 amino acid substitutions and/or insertions compared with Herceptin (tables S3 and S4). Some of the clones lost binding affinity for HER2 and switched binding entirely to the new antigen, whereas others maintained the ability to bind HER2 and thus were dual specific.

Representative VEGF and DR5-binding clones were expressed as Fab and IgG proteins to assess their specificity and affinity (Table 1 and fig. S1). VEGF and DR5 are therapeutic targets of known structure. Equilibrium binding affinities (Kd) of the LC library-derived monospecific antibodies ranged from 15 to 150 nM (Table 1). The dual specific antibodies bound the new antigens (i.e., VEGF or DR5) with high nanomolar to micromolar affinity while preserving HER2 binding in the low nanomolar affinity range. The antibody bH1 exhibited the highest dual affinity for the two completely different protein antigens VEGF (Kd = 300 nM) and HER2 (Kd = 26 nM).

Table 1.

The representative antibodies from the LC library of Herceptin. Mutations from Herceptin are shown in italics. Dashes indicate positions where no residue is present. Antigen binding affinity (Kd) was determined by surface plasmon resonance using Fab. NB indicates that no binding is detected. Single-letter abbreviations for the amino acid residues are as follows: A, Ala; C, Cys; D, Asp; E, Glu; F, Phe; G, Gly; H, His; I, Ile; K, Lys; L, Leu; M, Met; N, Asn; P, Pro; Q, Gln; R, Arg; S, Ser; T, Thr; V, Val; W, Trp; and Y, Tyr.

View this table:

To understand the molecular basis of the dual specificity, we determined the crystal structures of the bH1 Fab bound to the extracellular domain of HER2 (residues 1 to 624) and to the VEGF receptor-binding domain (residue 8 to 109) at 2.9 and 2.6 Å resolution, respectively (Fig. 1A and table S5). The structure of the bH1/HER2 binding domains (bH1 Fv and domain IV of HER2) superimposes onto the Herceptin/HER2 complex with a root mean square deviation of 0.8 Å, showing that the Herceptin binding epitope on HER2 is retained (Fig. 1A) (7). The HER2 binding sites on bH1 and Herceptin differ only in the CDR-L1 and -L2 regions where the bH1 sequence differs from Herceptin (Fig. 1B).

Fig. 1.

The crystal structures of bH1 Fab bound to HER2 or VEGF. (A) The bH1 Fab (gray)/HER2 (gold) superimposed on to the Herceptin (pink)/HER2 (red) complex (left), and bH1 Fab (light blue)/VEGF (green, teal) complex (right). (B) Fab surface residues are colored according to the extent buried in the complex (red, >75%; orange, >50 to 75%; yellow, >25 to 50%). The underlined amino acids differ between bH1 and Herceptin. The white dotted line separates the LC and HC. (C) Superposition of the CDR loops of VEGF and HER2-bound bH1 (blue, gray) and HER2-bound Herceptin (pink) in the same orientation as in (B). (D) CDR-L1 regions of the two bH1 complexes shown in the same orientation. The residues with temperature factors higher than average are shown in red and orange. VEGF would clash with Tyr32 of bH1 in its HER2-bound conformation.

In the structure of the bH1/VEGF complex, the Fab recognizes an epitope that overlaps with the binding sites of the VEGF receptors and other VEGF binding antibodies (1113). Consistent with this observation, bH1 blocks the interaction between VEGF and its receptors (fig. S2).

Comparison of the bH1/VEGF and the bH1/HER2 complexes shows an extensive overlap of the VEGF and HER2 binding sites on bH1 (Fig. 1B and table S6); 11 out of the 13 bH1 residues that are in close contact with HER2 also make contact with VEGF. In the HER2 complex, the LC and HC CDRs contribute approximately equal antigen contact area (53% and 47%, respectively), whereas in the VEGF complex, the LC CDRs constitute nearly 70% of the buried surface (fig. S3).

With the exception of CDR-L1, the CDR conformations of the bH1 Fab bound to VEGF are markedly similar to those observed in bH1 or Herceptin bound to HER2 (Fig. 1C). The capability of CDR-L1 to rearrange appears to be necessary for the dual specificity of bH1 (Fig. 1D). In the HER2 complex, CDR-L1 is minimally involved in antigen interaction, and part of the loop appears to be flexible, as evidenced by poorly defined electron density (Fig. 1D). In contrast, in the VEGF complex, this loop is well structured and forms extensive contacts with VEGF. Similar conformational flexibility of CDR-L1 has been reported to play a role in antigen recognition of natural antibodies (14, 15).

Next, we performed shotgun scanning mutagenesis studies to assess the energetic contributions of individual CDR residues to VEGF and HER2 binding (S11 and S12 in the supporting online material) (10). We determined the change in free energy (ΔΔG) as a result of mutation to alanine (alanine scan) or a homologous residue (homolog scan) (table S7), revealing the CDR residues that make up the functional paratopes of the Fab in the two complexes (Fig. 2A).

Fig. 2.

The distinct interactions of bH1 with HER2 and VEGF. (A) The ΔΔG values (y axis, kilocalories per mole) of each mutation to alanine (black) or a homologous amino acid (white) for VEGF or HER2 binding based on shotgun scanning mutagenesis (10). The extent of the bH1 residues buried upon VEGF or HER2 complex formation is indicated (single asterisk, >25 to 50% buried; double asterisk, >50 to 75%; triple asterisk, >75%). The dagger symbol represents a lower limit (table S7). (B) The residues that make structural contacts (>25% buried) or energetic interaction (ΔΔG > 10% total binding energy) with HER2 (pink), VEGF (green), or both (shared, blue) are mapped on the surface of HER2-bound bH1.

In contrast to the extensive overlap between buried surfaces in the interfaces of bH1 bound to either VEGF or HER2, the two functional paratopes show only limited overlap (Fig. 2B and fig. S4). Energetically, the VEGF binding interaction is mediated primarily by the LC CDRs, whereas HER2 binding is dominated by HC CDRs. Compared to Herceptin, bH1 maintains the same core hotspot residues for HER2 binding (Arg50, Trp95, and Tyr100a of HC) (fig. S4).

The structural and functional studies show that the interactions between bH1 and the two entirely unrelated large proteins are distinct (Fig. 2 and fig. S5) and are characterized by conformational adaptation of the antigen binding site (table S8) and differential engagement of VL and VH (Fig. 2). The molecular versatility observed in the dual specific bH1 is reminiscent of other antibodies binding multiple small haptens or peptides (4, 16) or of an antibody binding its cognate antigen and an anti-idiotype antibody (17, 18). Further, the differential engagement of residues within a common binding site also echoes the plasticity observed in other protein-protein interactions, such as the recognition of distinct antigens (major histocompatability complex–bound peptides) by T cell receptors (19) or growth factor binding to different receptors (20).

We further examined whether the dual binding specificity can translate into dual activity in vivo. To facilitate the investigation of the in vivo activity of bH1, we generated versions of bH1 with high affinity and specificity for HER2 and VEGF (figs. S1, S2, S6, S7, and S8 and table S9). bH1, the affinity-improved bH1-81 (Kd = 58/6 nM VEGF/HER2), and bH1-44 (Kd = 3/0.2 nM VEGF/HER2) inhibit VEGF-induced proliferation of human umbilical vein endothelial cells (HUVECs) and the growth of the HER2 overexpressing breast cancer cell line, BT474 (Fig. 3A). The potencies of the bH1 variants correlate with their relative affinities. The bH1-44 variant inhibits the growth of HUVEC and BT474 cells with potencies similar to bevacizumab or Herceptin, respectively.

Fig. 3.

Inhibition of VEGF and HER2 function in vitro and in vivo. (A) bH1 and the affinity improved variants bH1-81 and bH1-44 as IgGs inhibit VEGF-stimulated proliferation of HUVECs and human breast cancer cell BT474 in a dose-dependent manner. Herceptin and bevacizumab (anti-VEGF) were used as controls. Error bars represent the average deviation of triplicates (n = 3). (B) Tumor growth inhibitions of bH1-44 in Colo205 and BT474M1 xenografts in immunocompromised mice. The graph shows tumor volume as a function of time. The arrows indicate the time points for antibody (human IgG) administration. Error bars indicate SEM (10). An asterisk indicates the time at which the effects on tumor growth inhibition were compared. For the statistical analysis, we used one-way analysis and Student's t tests. For BT474M1, 5 out of the 8 mice (5/8) dosed with bH1-44 at 10 mg/kg had partial responses (PR, 50 to 99% regression from initial volume), similar to Herceptin (PR = 6/8) or the anti-VEGF/Herceptin combination (PR = 5/8). Anti-VEGF treatment alone did not yield any partial response in this model.

To examine the two unique activities of bH1-44 in vivo, we studied mouse xenograft tumor models known to be responsive to treatment by anti-VEGF antibodies (Colo205, a human colorectal cancer cell line) or Herceptin (BT474M1, human breast cancer cell line). We compared the bH1-44 treated groups with those treated with anti-VEGF (B20-4.1) (21), Herceptin, or the combination (Herceptin + anti-VEGF) (table S10) (10). The results of the Colo205 model demonstrate that the VEGF binding component of bH1-44 is inhibiting tumor growth (P < 0.0001, n = 10 mice, compared with the control IgG treated group), as Herceptin alone has no effect (P = 0.12, n = 10). The BT474M1 model indicates that the HER2 binding component of bH1-44 is active (P < 0.0001, n = 7), because the effect observed with the anti-VEGF control antibody is not significant (P = 0.6, n = 7) (Fig. 3B).

Hence, bH1-44 appears to have the pharmacological activity of a VEGF blocking antibody and a Herceptin-like antibody, although we do not yet know whether bH1-44 is as effective as Herceptin or anti-VEGF administered separately or in combination. Whether there is a therapeutic benefit to co-targeting the tumor cell proliferation mediated by HER2 and tumor angiogenesis mediated by VEGF is a question currently being evaluated in clinical trials combining Herceptin and bevacizumab in breast cancer patients (2224). If the drug combination is found to be beneficial, then dual specific antibodies like bH1-44 may merit further exploration as potential therapeutics.

Various bi-targeting antibody formats that assemble two distinct antibody fragments into one molecule have previously been described (25, 26). In contrast to these bispecific formats, the dual specific “two-in-one” antibody we describe has the molecular structure of a regular IgG (or Fab). It has all the favorable attributes of an IgG for therapeutic development, such as predictable pharmacokinetic properties, well established manufacturing protocols, choice of Fc-mediated effector functions, and bi- or mono-valencies (25).

In summary, we have demonstrated that an antigen binding site is capable of interacting with two unrelated protein antigens with high affinity. The dual specific antibodies reported here are derived from a monospecific antibody through mutations in the periphery of the antigen binding site in the LC CDRs. This strategy is a general one and can be applied to create dual specific antibodies against two distinct antigens. The mutational analysis of bH1 and bH1-44 (Fig. 2 and fig. S8) suggested that the dual specificity could be switched to monospecific binding to either antigen (10). Indeed, bH1-44 lost binding to VEGF but retained HER2 binding when mutating two LC residues. Similarly, two alanine mutations in the HC drastically reduced the affinity for HER2 while preserving tight binding for VEGF (fig. S9). This finding highlights how a limited number of mutations in the antigen binding site can alter specificity or add a distinct specificity. During development of the natural antibody repertoire, the antigen binding sites often undergo diversification by exchanging the VL that pairs with a VH (6). Somatic mutations also occur frequently, in particular among the residues in the periphery of the antigen binding site (2730). Our studies reveal a mechanism by which one antibody can diverge into many antibodies with distinct specificity profiles. This mechanism may contribute to the large capacity of the natural antibody repertoire for diverse antigen recognition.

Supporting Online Material

www.sciencemag.org/cgi/content/full/323/5921/1610/DC1

Materials and Methods

Figs. S1 to S9

Tables S1 to S10

References

References and Notes

View Abstract

Navigate This Article