FGF receptors (FGFR) are attractive candidate targets for cancer therapy because they are dysregulated in several human malignancies. FGFR2 and FGFR3 can be inhibited potentially without disrupting adult tissue homeostasis. In contrast, blocking the closely related FGFR1 and FGFR4, which regulate specific metabolic functions, carries a greater safety risk. An anti-FGFR3 antibody was redesigned here to create function-blocking antibodies that bind with dual specificity to FGFR3 and FGFR2 but spare FGFR1 and FGFR4. R3Mab, a previously developed monospecific anti-FGFR3 antibody, was modified via structure-guided phage display and acquired additional binding to FGFR2. The initial variant was trispecific, binding tightly to FGFR3 and FGFR2 and moderately to FGFR4, while sparing FGFR1. The X-ray crystallographic structure indicated that the antibody variant was bound to a similar epitope on FGFR2 as R3Mab on FGFR3. The antibody was further engineered to decrease FGFR4-binding affinity while retaining affinity for FGFR3 and FGFR2. The resulting dual-specific antibodies blocked FGF binding to FGFR3 and FGFR2 and inhibited downstream signaling. Moreover, they displayed efficacy in mice against human tumor xenografts overexpressing FGFR3 or FGFR2. Thus, a monospecific antibody can be exquisitely tailored to confer or remove binding to closely related targets to expand and refine therapeutic potential. Mol Cancer Ther; 14(10); 2270–8. ©2015 AACR.

This article is featured in Highlights of This Issue, p. 2165

FGFs and their tyrosine kinase receptors (FGFR) play key roles in regulating specific pathways during embryonic development, as well as homeostasis of diverse tissues, wound-healing processes, and certain metabolic functions in the adult animal. In humans, there are 4 highly homologous FGFRs (FGFR1–4) and 22 FGFs (FGF1–14 and FGF16–23; refs. 1–4). The FGFRs comprise an extracellular region with 3 immunoglobulin domains (D1, D2, and D3), a single-pass transmembrane region, and a split cytoplasmic kinase moiety (1, 5). Alternative splicing gives rise to 2 major variants of FGFR1–3, termed isoforms IIIb and IIIc, which differ in the second half of D3 and consequently in ligand-binding specificity (6).

Dysregulated signaling by FGFR1–4 is associated with pathogenesis in several cancer types (2, 3). Genomic FGFR alterations, which include gene amplification, chromosomal translocation, and activating mutations, can drive aberrant activation of the FGF pathway and promote neoplastic transformation of normal cells. FGFR2 gene amplification occurs in about 10% of gastric and about 4% of triple-negative breast cancers (7–9), while FGFR3 amplification is associated with specific subsets of bladder cancer (9, 10). Missense FGFR mutations are also found in multiple types of cancer (2, 11). Specifically, amino acid substitutions in the linker region between D2 and D3, for example, S252W in FGFR2 and S249C in FGFR3, augment FGF-driven signaling and tumor cell proliferation and represent hotspots for somatic mutation (12, 13). Activating mutations also occur in the tyrosine kinase region of FGFRs (14).

Targeting the FGF/FGFR pathway has been a major area of focus for cancer drug development. This effort has included small-molecule tyrosine kinase inhibitors (TKI), blocking antibodies, as well as ligand traps (9). Current high-potency FGFR TKIs have limited selectivity for different FGFRs (9), which may affect their therapeutic window. For example, disruption of FGF23 signaling through heterocomplexes of FGFR1 and the coreceptor Klothoβ can lead to hyperphosphatemia and soft-tissue calcification in patients (15, 16), whereas blockade of FGF19 signaling through FGFR4 heterocomplexes with Klothoβ can disrupt bile acid metabolism (17). More selective antibodies have been developed to antagonize ligand signaling through individual FGFRs, including FGFR1 (18), FGFR2 (19), and FGFR3 (20). However, antibodies recognizing more than one FGFR have not yet been reported.

The previously described monospecific anti-FGFR3 antibody R3Mab effectively blocks binding of FGF1 and FGF9 to both the IIIb and IIIc isoforms of wild-type FGFR3, as well as to certain cancer-associated mutant forms of FGFR3 (20, 21). X-ray structural analysis revealed that R3Mab binds to a specific epitope on FGFR3 that is required for ligand binding. R3Mab displayed potent antitumor activity in mice against human bladder cancer and multiple myeloma tumor xenografts. In this study, structure-guided phage display was used iteratively to engineer R3Mab into derivative antibodies that carry dual specificity for FGFR3 and FGFR2 while sparing FGFR1 and FGFR4. The aim of this study was to broaden the potential therapeutic scope beyond that of the parent molecule while avoiding added safety risks. The engineered antibodies displayed inhibition of FGF-stimulated tumor cell growth in vitro and significant efficacy against human cancer xenografts overexpressing FGFR2 or FGFR3 in vivo.

Generation of FGFR2-binding R3Mab variants by phage library selection

Random mutations were incorporated into each of the CDR loops H1, H2, H3, or L2 (Supplementary Table S1) using the method of Kunkel and colleagues (22). Purified phage suspensions from each library were panned separately against immobilized FGFR2-IIIb proteins for the first round of panning. Eluted phage particles were then pooled together and propagated for subsequent rounds of panning. Ninety-six randomly picked colonies were individually cultured and assayed by phage ELISA to screen for FGFR2 binders. Meanwhile, phagemid DNA was extracted and sequenced from the cultures.

Library construction and selection for FGFR2 binders that do not bind FGFR4

The phage display libraries were constructed on the basis of the phagemid displaying the Fab fragment of antibody 2B.1.3. Selected positions in CDR H1, H3, or L2 loops were subject to random mutagenesis (Supplementary Table S2). For selection of clones that have reduced FGFR4 binding while retaining FGFR2 specificity, in the first round, 1.5 OD of each phage library was mixed with 0.5 nmol/L FGFR4-Fc protein. The mixture was then incubated overnight at 4°C with plate-immobilized FGFR2-IIIb. The second round of panning was similar to the first round except that 1.5 OD of phage preparations were mixed with 10 nmol/L FGFR4-Fc. For the third and fourth rounds, 0.5 OD of phage preparations were mixed with 460 nmol/L FGFR4-Fc protein and shaken at room temperature (RT) for 20 minutes before being incubated with coated FGFR2-IIIb. Randomly picked clones were cultured for phage ELISA assays and DNA sequencing as described above.

Phage ELISA-binding assay

A 384-well MaxiSorp plate was coated E25 (control antibody), FGFR2-IIIb-His, FGFR2-IIIc-His, or FGFR4-His in each quadrant. Phage supernatant was added into quadrant after blocking with BSA. Bound phage particles were detected with HRP-conjugated anti-M13 monoclonal antibody (GE Healthcare).

Surface plasmon resonance assay

The binding affinities of R3Mab variants for FGFR antigens were determined using a Biacore T100 (GE Healthcare). Anti-human Fc monoclonal antibody was immobilized onto a CM5 biosensor chip. FGFR antigens of various concentrations were injected over captured R3Mab-derived antibodies. Kinetic analyses were performed using the T100 evaluation software to obtain the kinetic and affinity constants.

Protein expression, purification, and structure determination

The human FGFR2-IIIb ECD (residues 140–369) was expressed as inclusion bodies in Escherichia coli BL21(DE3)pLysS cells. The inclusion bodies were washed and dissolved in 6 mol/L guanidine HCl, 20 mmol/L Tris, pH 8, 10 mmol/L TCEP for in vitro folding using the rapid dilution method. The refolding mixture was concentrated and purified through a heparin affinity column (GE Healthcare), followed by ion exchange chromatography. The 2B.1.3 Fab was expressed and purified as described (20). The FGFR2 and Fab proteins were mixed together at a molar ratio of 1:1 and diluted to 2 mg/mL for crystallization. Crystals were grown at 20% (w/v) PEG 3350, 0.1 mol/L sodium citrate, pH 5.5, and 0.2 mol/L ammonium sulfate using the vapor diffusion method. Diffraction data were collected with a beam wavelength of 1 Å at the ALS and processed to 2.36 Å. Two complexes were found in an asymmetric unit cell. The final model was validated using the program MolProbity (23). Rwork and Rfree values are 19.8% and 24.4%, respectively. No Ramachandran outliers were detected.

Cell lines

SNU16 and MFM-223×2.2 cell lines were obtained from an internal cell bank. The cell line RT112 was obtained from ATCC. The cells were cultured in RPMI medium supplemented with 10% FBS. All cell lines are tested for mycoplasma, cross-contamination, and genetically fingerprinted when new stocks are generated to ensure quality and confirm ancestry.

FGF ligand-blocking ELISA

A 96-well MaxiSorp plate was coated with anti-human Fc antibody (Jackson ImmunoResearch Lab). FGFR-Fc fusion proteins were incubated after blocking. The plate was washed before being added with the antibody and FGF ligand mixtures. Bound ligand was detected by subsequent incubations of biotinylated anti-FGF antibodies (R&D Biosystems, #BAF273, #BAF251, #BAF969), streptavidin HRP (Invitrogen), and the TMB substrate.

Immunoblotting

Tumor cells were seeded on tissue culture plates for 24 hours, pretreated with 10 μg/mL anti-FGFR antibodies or control anti-gD antibody, and then stimulated with 25 ng/mL FGF7 (R&D Systems) in the presence of 20 μg/mL heparin (Sigma) for 15 minutes. Total cell lysates were blotted with primary antibodies recognizing various signaling molecules, including phospho-FGFR (Y653/654), FGFR2, phospho-FRS2 (Y196), FRS2 (Santa Cruz Biotechnology); phospho-ERK1/2 (T202/Y204), ERK1/2, phospho-AKT (S473), AKT, phospho-HER3 (Y1289), HER3, phospho-PLCγ1 (Y783), PLCγ1 (Cell Signaling); and β-actin (Sigma).

Xenograft experiments

All procedures were approved by and conformed to the guidelines and principles set by the Institutional Animal Care and Use Committee of Genentech and were carried out in an AAALAC-accredited facility. SNU-16 tumor fragments of about 15 to 30 mm3 were implanted subcutaneously into the right flanks of 6- to 8-week-old female Balb/c nude mice (Shanghai Laboratory Animal). Seven million RT-112 bladder carcinoma cells suspended in HBSS with Matrigel were inoculated subcutaneously in the 6- to 8-week-old female C.B-17 SCID mice (Charles River Laboratories). When the mean tumor volume reached 100 to 200 mm3 (day 0), mice were randomized into groups of 10 and treated starting on day 1 with twice weekly intraperitoneal injections of 2B.1.3.10, 2B.1.3.12, or R3Mab (30 or 50 mg/kg). Control groups were treated with a control human IgG1 antibody diluted in PBS (30 mg/kg). The tumor volumes were measured twice a week.

Statistical analysis

Xenograft data are expressed as mean tumor volumes ± SEM. Unpaired, two-tailed t tests were performed to assess the statistical significance between the experimental and control groups. Time points with P < 0.005 are considered as significant and marked with asterisks.

Complete details of Materials and Methods are provided in Supplementary Data.

Broadening the binding specificity of an anti-FGFR3 antibody

The objective of this study was to develop an antibody with dual specificity for FGFR3 and FGFR2 but sparing the highly related receptors, FGFR1 and FGFR4, as a potential cancer therapeutic. The starting point for this study was the monospecific antibody R3Mab, which binds to the FGFR3-IIIb and IIIc isoforms with sub-nanomolar affinities (20). R3Mab shows robust inhibition of FGFR3 signaling and tumor growth in vivo (20) and has been studied in phase I clinical trials.

Our antibody redesign strategy was guided by the previously determined crystallographic structure of the R3Mab Fab fragment in complex with FGFR3-IIIb (PDB 3GRW; ref. 20). This structure indicates that R3Mab interacts with both the D2 and D3 domains of FGFR3-IIIb. Although D2 was subsequently found here to be sufficient for R3Mab binding (see below), initial analyses were based on the contacts within this original structure. Most of the contact surface on the FGFR3-IIIb antigen was contributed by the antibody complementarity-determining regions (CDR) H3 (46%), H1 (23%), and L2 (22%), with small contributions from CDR H2 and framework region (FR) residues (ref. 20; Supplementary Fig. S1A). The similarity between FGFR3-IIIb and the intended additional FGFR2-IIIb antigen were compared. Of the total 1,421-Å2 binding area of R3Mab on FGFR3-IIIb, 399 Å2 (28%) accounts for the residue difference between FGFR2-IIIb and FGFR3-IIIb (Supplementary Fig. S1B). In addition, the D2D3 regions of these 2 homologs share 68% of protein sequence identity, whereas their D2 domains share 76% identity (Supplementary Table S3). Because D3 of the R3Mab-bound FGFR3-IIIb had a different geometry as compared with all other FGFR structures (20), the structures of FGFR2-IIIb and FGFR3-IIIb were superimposed on their D2 regions, which yielded a calculated root mean squared deviation (RMSD) of α-carbon atoms of 0.78 Å. This high degree of structural similarity suggested that it might be feasible to engineer R3Mab to bind and inhibit FGFR2 as well.

To construct a phage display library, mutations were designed that cover most residues in each of the individual heavy-chain CDRs and a selection of the contact residues on all CDRs (Supplementary Table S1). R3Mab variants displayed as Fab fragments on phage particles were selected for binding to FGFR2-IIIb. Selection on FGFR3 was not performed at this stage to keep the selection stringency low when recruiting binding to FGFR2. After the first round of panning, the phage outputs from the individual libraries were combined and subjected to 3 further rounds of selection. Ninety-five antibody clones, designated as the 2B.1 series, were screened by phage ELISA-binding assay. Among these, 81 clones, representing 32 unique sequences, bound to FGFR2-IIIb. All binding clones were apparently derived from the H2 library because they contained mutations in CDR H2 but not elsewhere (Table 1). The 32 unique antibodies were expressed and purified as corresponding IgG1 molecules. All of them showed substantially improved binding to FGFR2-IIIb relative to R3Mab, with KD values ranging from 0.3 to 17 nmol/L (Table 1). Remarkably, the mutated H2 sequences contain significant variation, lack clear consensus, and differ from R3Mab at 4 or 5 positions (Table 1, Supplementary Fig. S2). Thus, there appear to be multiple possible solutions to conferring high-affinity binding of FGFR2-IIIb onto R3Mab.

Table 1.

Binding affinities of 2B.1 antibodies for FGFR2-IIIb

Variant IDCDR-H2 sequenceTimes found (n)FGFR2-IIIb KD, nmol/LVariant IDCDR-H2 sequenceTimes found (n)FGFR2-IIIb KD, nmol/LVariant IDCDR-H2 sequenceTimes found (n)FGFR2-IIIb KD, nmol/L
R3Mab IYPTN ND 2B.1.14 LWYFD 1.6 2B.1.72 VHPFE 3.5 
2B.1.1 YWAWD 0.29 2B.1.71 VWMFD 1.6 2B.1.44 WWSWG 3.6 
2B.1.88 IWMFT 0.64 2B.1.28 FWAWS 1.8 2B.1.52 FSLGD 3.9 
2B.1.38 FWAYD 1.1 2B.1.95 LIFFT 1.8 2B.1.30 VSFFS 4.1 
2B.1.20 LDVFW 1.2 2B.1.50 LNFYS 2.0 2B.1.82 INFFS 4.9 
2B.1.32 WVGFT 1.2 2B.1.81 VNNFY 2.1 2B.1.93 IDNYW 13 5.1 
2B.1.49 LSFFS 1.3 2B.1.25 WHPWM 2.3 2B.1.55 VDVFW 5.9 
2B.1.86 LSFWT 1.3 2B.1.3 THLGD 2.6 2B.1.35 WHPFR 9.4 
2B.1.9 YHPYL 1.4 2B.1.65 YNAYT 2.7 2B.1.33 YHPFH 15 
2B.1.73 MIFYN 1.4 2B.1.94 LVFFS 3.1 2B.1.80 YWAFS 17 
2B.1.74 YHPFR 1.4 2B.1.78 LSFYS 3.2 2B.1.92 WVAFS NA 
Variant IDCDR-H2 sequenceTimes found (n)FGFR2-IIIb KD, nmol/LVariant IDCDR-H2 sequenceTimes found (n)FGFR2-IIIb KD, nmol/LVariant IDCDR-H2 sequenceTimes found (n)FGFR2-IIIb KD, nmol/L
R3Mab IYPTN ND 2B.1.14 LWYFD 1.6 2B.1.72 VHPFE 3.5 
2B.1.1 YWAWD 0.29 2B.1.71 VWMFD 1.6 2B.1.44 WWSWG 3.6 
2B.1.88 IWMFT 0.64 2B.1.28 FWAWS 1.8 2B.1.52 FSLGD 3.9 
2B.1.38 FWAYD 1.1 2B.1.95 LIFFT 1.8 2B.1.30 VSFFS 4.1 
2B.1.20 LDVFW 1.2 2B.1.50 LNFYS 2.0 2B.1.82 INFFS 4.9 
2B.1.32 WVGFT 1.2 2B.1.81 VNNFY 2.1 2B.1.93 IDNYW 13 5.1 
2B.1.49 LSFFS 1.3 2B.1.25 WHPWM 2.3 2B.1.55 VDVFW 5.9 
2B.1.86 LSFWT 1.3 2B.1.3 THLGD 2.6 2B.1.35 WHPFR 9.4 
2B.1.9 YHPYL 1.4 2B.1.65 YNAYT 2.7 2B.1.33 YHPFH 15 
2B.1.73 MIFYN 1.4 2B.1.94 LVFFS 3.1 2B.1.80 YWAFS 17 
2B.1.74 YHPFR 1.4 2B.1.78 LSFYS 3.2 2B.1.92 WVAFS NA 

NOTE: Residues that match those in R3Mab are underlined.

Abbreviations: NA, not available due to protein aggregation; ND, not detectable.

Next, 6 variants were selected for measurements of binding to FGFR3 on the basis of their affinities (KD < 3 nmol/L) for FGFR2 and sequence diversity. All the variants showed improved affinities for FGFR3-IIIb (Supplementary Table S4). To further assess their ability to inhibit receptor-dependent cell growth, proliferation of MCF7 breast carcinoma cells was assayed either with or without FGF7—a specific ligand for FGFR2-IIIb (1, 19). Variant 2B.1.3 exhibited the greatest antagonistic activity, as compared with other variants, which showed less or no inhibition, or even displayed stimulatory effects (Fig. 1). Hence, 2B.1.3 was carried over as a functional antibody for further characterization.

Figure 1.

Inhibitory effects of engineered 2B.1 antibodies for FGF7-stimulated MCF-7 cell proliferation. Error bars, SEM.

Figure 1.

Inhibitory effects of engineered 2B.1 antibodies for FGF7-stimulated MCF-7 cell proliferation. Error bars, SEM.

Close modal

Because all FGFR homologs share nearly 70% sequence identity between each other (Supplementary Table S3), the engineered variant 2B.1.3 was analyzed for binding to other FGFRs. Mab 2B.1.3 bound FGFR2-IIIc with similar affinity as FGFR2-IIIb (Table 2). Mab 2B.1.3 also showed several fold higher affinity for FGFR3-IIIb and FGFR3–IIIc than did R3Mab, even though the selection strategy used was based on binding to FGFR2-IIIb. Moreover, Mab 2B.1.3 also bound to FGFR4, with a KD value of 32 nmol/L, yet showed no detectable binding to FGFR1 (Table 2). Therefore, variant 2B.1.3 is trispecific, binding to FGFR2, FGFR3, and FGFR4, but not FGFR1.

Table 2.

Binding affinities of R3Mab and its variants to all human FGFR homologs

KD, nmol/L
FGFR1-IIIbFGFR1-IIIcFGFR2-IIIbFGFR2-IIIcFGFR3-IIIbFGFR3-IIIcFGFR4
R3Mab ND ND ND ND 0.24 0.61 ND 
2B.1.3 ND ND 2.6 2.0 0.09 0.07 32 
2B.1.3.10 ND ND 2.9 1.1 0.11 0.25 ND 
2B.1.3.12 ND ND 3.0 6.1 0.50 0.72 ND 
KD, nmol/L
FGFR1-IIIbFGFR1-IIIcFGFR2-IIIbFGFR2-IIIcFGFR3-IIIbFGFR3-IIIcFGFR4
R3Mab ND ND ND ND 0.24 0.61 ND 
2B.1.3 ND ND 2.6 2.0 0.09 0.07 32 
2B.1.3.10 ND ND 2.9 1.1 0.11 0.25 ND 
2B.1.3.12 ND ND 3.0 6.1 0.50 0.72 ND 

Abbreviation: ND, not detectable at 500 nmol/L.

Structural determination of the complex between Mab 2B.1.3 and FGFR2

To obtain direct insight into how the engineered variant 2B.1.3 acquired specificity for FGFR2, the crystal structure of its complex with FGFR2 was determined (Fig. 2, Supplementary Table S5). FGFR2-IIIb D2D3 was first generated by expression in E. coli and refolding from inclusion bodies and judged to be intact by SDS-PAGE and mass spectrometry. However, in crystals, this protein contained only the isoform-independent D2 domain, suggesting proteolysis between D2 and D3 during the crystallization process. The previously determined FGFR3-IIIb:R3Mab complex structure contains both the D2 and D3 domains of FGFR3-IIIb. The whole complex of FGFR2-D2:Mab 2B.1.3 superimposed closely onto the FGFR3-IIIb:R3Mab structure (Supplementary Fig. S3), with an overall α-carbon RMSD of 1.4 Å, indicating that the engineering retained the same binding geometry as the original antibody R3Mab. The FGFR3:R3Mab crystal structure suggests considerable interactions between FGFR3 D3 and the CDR H1 loop. Therefore, to investigate the involvement of D3 in binding, proteins of the D2 domains of FGFR2 and FGFR3 were prepared and their binding affinity to R3Mab and Mab 2B.1.3 measured. Only very minor differences in binding affinity between D2 alone and the D2D3 domains were observed for both receptors (Supplementary Table S6). Thus, D2 is primarily responsible for binding of R3Mab and its derivatives, whereas D3 plays a minimal role.

Figure 2.

Crystal structure of the complex between FGFR2 D2 domain and the Fab fragment of Mab 2B.1.3. A, FGFR2-D2 (magenta) in complex with Fab 2B.1.3 heavy (green) and light chains (blue). B, overlay of the structures of the complex between FGFR2-D2 and 2B.1.3 (colored as in A) and the complex between FGFR3-D2D3 (yellow) and R3Mab (gray). C, zoom-in representation of the boxed area in B showing the structural differences between the 2 complexes in the same color scheme.

Figure 2.

Crystal structure of the complex between FGFR2 D2 domain and the Fab fragment of Mab 2B.1.3. A, FGFR2-D2 (magenta) in complex with Fab 2B.1.3 heavy (green) and light chains (blue). B, overlay of the structures of the complex between FGFR2-D2 and 2B.1.3 (colored as in A) and the complex between FGFR3-D2D3 (yellow) and R3Mab (gray). C, zoom-in representation of the boxed area in B showing the structural differences between the 2 complexes in the same color scheme.

Close modal

The CDR H2 sequence in Mab 2B.1.3, THLGD, is completely different from the parental H2 sequence in R3Mab, IYPTN. As expected, the conformations of the CDR H2 loops in the 2 Mabs differ substantially (Fig. 2C). Upon aligning the variable domains of Mab 2B.1.3 onto those of R3Mab (Fig. 2B), the H3 loop also appears twisted by a few degrees, resulting in a distance of 2.6 Å between the Cα atoms of the H3 tip residue Y100b in both structures (Fig. 2C). Accordingly, the position of the FGFR2 D2 domain overall is shifted by about 3 Å from that of the FGFR3 D2 domain. Comparison of the interface between the variants and the FGFR antigens revealed that such reorganizations of the H2 and H3 CDR loops in Mab 2B.1.3 significantly improved packing against the FGFR2 surface. In the parental structure, the shape complementarity score between R3Mab and FGFR3-D2 is 0.731. If the D2 domain of FGFR2 is aligned onto and replaces FGFR3 D2, the shape complementarity score between R3Mab and FGFR2 D2 drops to 0.685. This might explain the lack of R3Mab binding to FGFR2 (Table 2). However, in the new crystal structure, the shape complementarity score between 2B.1.3 and FGFR2-D2 dramatically increased to 0.768, which is consistent with the gain of high-affinity binding to FGFR2 through engineering of R3Mab.

Because of the remarkable similarity among FGFRs, 2B.1.3 cross-reacts with multiple homologs in the family. Although FGFR1 binding was not acquired along with FGFR2 binding, FGFR4 interaction was. Considering that FGFR4 inhibition carries an increased risk of toxicity (17), a second round of engineering was undertaken to eliminate FGFR4 binding.

Further engineering to remove FGFR4 binding

To generate a Mab 2B.1.3 derivative that binds FGFR2 and FGFR3 but not FGFR4, it seemed useful to identify antigen residues that likely interact with the antibody but differ between the various FGFRs (Supplementary Table S7), assuming that Mab 2B.1.3 recognizes all FGFRs in an analogous mode to its interaction with FGFR2. Three phage display libraries were constructed on the basis of the 2B.1.3 Fab template, with random mutagenesis at selected positions on the contacted CDRs H1, H3, and L2 (Supplementary Table S2). During engineering, the focus was on binding to FGFR2 instead of maintaining both FGFR2 and FGFR3, as in the previous engineering. Therefore, selection was undertaken with immobilized FGFR2-IIIb alone during panning. To counterselect FGFR4 binders, phage particles were incubated with large amounts of soluble FGFR4-Fc protein. The concentrations of FGFR4-Fc were increased up to 0.46 μmol/L for successive rounds of selection (see Materials and Methods). Individual clones from round 4 (n = 96) were assayed by ELISA with FGFR2-IIIb and FGFR4 and ranked by the ratio of FGFR2- to FGFR4-binding ELISA values. Six clones with the highest FGFR2/FGFR4-binding ratios were sequenced, expressed as IgG, and characterized for binding to FGFR2-IIIb and FGFR4 (Supplementary Table S8). Characterized clones from the H3/L2 libraries 2B.1.3.2, 2B.1.3.4, and 2B.1.3.6 contained mutations only in CDR H3, not CDR L2, whereas characterized clones from the H1/H3 library 2B.1.3.8, 2B.1.3.10, and 2B.1.3.12 contained mutations in both CDR H1 and H3. Although the 4 residues in H3 from L100a to D100d were fully randomized, Y100b remained unchanged, suggesting that the interaction of Y100b with FGFR2 is crucial for binding. In addition, L100a was conservatively mutated to Thr or Ile and V100c mostly to Asp. The H1/H3 mutants containing an additional H1 mutation of T28P displayed slightly higher affinities for FGFR2. These antibodies bound FGFR2 with KD values of 1.4 to 6.6 nmol/L but showed minimal binding to FGFR4 when using concentrations as high as 1 μmol/L for measurements, except that clone 2B.1.3.8 still retained detectable yet weak affinity for FGFR4 (Supplementary Table S8). The convergence in both sequences and affinities of the 2B.1.3 variants indicated that the last rounds of phage selection had reached the limit of enrichment for binders with desired functions, that is, diminished FGFR4 binding and retention of tight FGFR2 binding.

Considering that greater differential in binding to FGFR2 and FGFR4 as well as fewer mutations are preferable, Mab 2B.1.3.10 and 2B.1.3.12 were selected for further characterization. Both antibodies showed no binding to FGFR1 and retained strong binding to FGFR3 with affinities slightly weaker than 2B.1.3 (Table 2). Therefore, after the second-step engineering, the 2B.1.3 derivatives Mab 2B.1.3.10 and 2B.1.3.12 cross-react with FGFR2 and FGFR3 but show no detectable binding to FGFR4.

Next, the abilities of the R3Mab variants to block FGF ligand binding to the specific FGFRs were evaluated. R3Mab blocks FGF ligand binding to both the FGFR3-IIIb and FGFR3-IIIc isoforms. Owing to their different specificities for different FGFRs, the blocking spectrum of each of the new antibodies varied (Fig. 3). All the engineered antibodies showed blocking activities for both FGFR2 and FGFR3, whereas R3Mab did not inhibit FGF7 binding to FGFR2-IIIb or FGF1 binding to FGFR2-IIIc. Whereas 2B.1.3 strongly inhibited FGF19 binding to FGFR4, 2B.1.3.10 and 2B.1.3.12 did not block the latter interaction, due to substantially diminished FGFR4 affinity.

Figure 3.

Differential blocking of FGF ligands by R3Mab IgG1 variants. A, blocking of FGF7 binding to human FGFR2-IIIb. B, blocking of FGF1 binding to human FGFR2-IIIc. C, blocking of FGF1 binding to human FGFR3-IIIb. D, blocking of FGF1 binding to human FGFR3-IIIc. E, blocking of FGF19 binding to human FGFR4.

Figure 3.

Differential blocking of FGF ligands by R3Mab IgG1 variants. A, blocking of FGF7 binding to human FGFR2-IIIb. B, blocking of FGF1 binding to human FGFR2-IIIc. C, blocking of FGF1 binding to human FGFR3-IIIb. D, blocking of FGF1 binding to human FGFR3-IIIc. E, blocking of FGF19 binding to human FGFR4.

Close modal

Redesigned Mab variants inhibit FGFR2- or FGFR3-dependent tumor cell growth

The newly redesigned variants 2B.1.3.10 and 2B.1.3.12 display dual specificity for FGFR2 and FGFR3. To investigate their biologic activities, their effects on receptor-dependent signaling and proliferation in different types of tumor cells were examined. First, the new variants were assessed for inhibition of growth of FGFR2-overexpressing tumor cells in vitro. Both the SNU-16 gastric carcinoma and MFM-223×2.2 triple-negative breast carcinoma cell lines have amplification of FGFR2, evident by increased FGFR2 gene copy numbers and protein overexpression (7). In SNU-16 cells, 2B.1.3.10 and 2B.1.3.12 substantially suppressed FGF7-induced FGFR2 phosphorylation. In addition, the two 2B.1.3 variants markedly reduced phosphorylation of the downstream signaling molecules FRS2α, MAPK, PLCγ1, and AKT (Fig. 4A). Similarly, both variants diminished phosphorylation of FGFR2, FRS2α, MAPK, and HER3 in FGF7-treated MFM-223×2.2 cells (Supplementary Fig. S4A).

Figure 4.

2B.1 variants inhibit FGFR2 signaling in vitro and suppress xenograft growth in vivo. A, blocking of FGF7-stimulated FGFR2 signaling by 2B.1 variants in the gastric cancer cell line SNU-16. B, effects of R3Mab, 2B.1.3.10, and 2B.1.3.12 on the growth of FGFR3-dependent RT112 bladder cancer xenografts. ****, P < 0.0001 for 2B.1.3.10 or 2B.1.3.12 versus control MAb at day 22. n = 10 per group; error bars, SEM. C, effect of 2B.1.3.10 and 2B.1.3.12 on the growth of FGFR2-dependent SNU-16 xenografts compared to control antibody and R3Mab. ****, P < 0.0001 for 2B.1.3.10 or 2B.1.3.12 versus control MAb at day 28; ***, P < 0.001 for 2B.1.3.10 or 2B.1.3.12 versus R3Mab at day 31. n = 10 per group; error bars, SEM.

Figure 4.

2B.1 variants inhibit FGFR2 signaling in vitro and suppress xenograft growth in vivo. A, blocking of FGF7-stimulated FGFR2 signaling by 2B.1 variants in the gastric cancer cell line SNU-16. B, effects of R3Mab, 2B.1.3.10, and 2B.1.3.12 on the growth of FGFR3-dependent RT112 bladder cancer xenografts. ****, P < 0.0001 for 2B.1.3.10 or 2B.1.3.12 versus control MAb at day 22. n = 10 per group; error bars, SEM. C, effect of 2B.1.3.10 and 2B.1.3.12 on the growth of FGFR2-dependent SNU-16 xenografts compared to control antibody and R3Mab. ****, P < 0.0001 for 2B.1.3.10 or 2B.1.3.12 versus control MAb at day 28; ***, P < 0.001 for 2B.1.3.10 or 2B.1.3.12 versus R3Mab at day 31. n = 10 per group; error bars, SEM.

Close modal

Next, the ability of the dual-specific Mab 2B.1.3.10 and 2B.1.3.12 to inhibit in vivo FGFR2-dependent and/or FGFR3-dependent growth of tumor xenografts was investigated. The RT112 cell line expresses FGFR3 but not FGFR2. As anticipated, both Mab 2B.1.3.10 and 2B.1.3.12, which retained the parental specificity for FGFR3 after engineering, as well as the parental antibody R3Mab, suppressed the growth of FGFR3-overexpressing RT112 tumor xenografts (Fig. 4B). The engineered variants 2B.1.3.10 and 2B.1.3.12, with tumor growth inhibition (TGI) values of 48% and 64%, displayed weaker potency than the parental R3Mab (TGI 82%), which could be possibly due to modified pharmacokinetics. For FGFR2-based efficacy, we turned to the SNU-16 cell line, which expresses readily detectable FGFR2 along with very low FGFR3 levels. Mice bearing SNU-16 xenografts were dosed with nonspecific IgG control antibody, the parental R3Mab, or the engineered variants 2B.1.3.10 or 2B.1.3.12. The engineered variants displayed similar TGI values of 63% and 61%, respectively (Fig. 4C). Surprisingly, R3Mab, although not binding to FGFR2, also showed a measurable TGI of 44%. The tumor samples were then collected and analyzed for FGFR2 and FGFR3 expression (Supplementary Fig. S5). FGFR3 was upregulated in the SNU-16 tumor xenografts in vivo, which may explain the observed inhibitory effect of R3Mab in this model. Regardless, the engineered variants showed significantly stronger activity as compared with R3Mab (P < 0.001, day 31). In another experiment, 2B.1.3.10 and 2B.1.3.12 also retarded the growth of MFM-223×2.2 tumor xenografts in mice (Supplementary Fig. S4B). Thus the engineered antibodies can serve as dual agents to effectively inhibit both FGFR2- and FGFR3-dependent cancer cell growth.

The FGFR family is associated with versatile normal biologic functions and is additionally implicated in a number of cancer malignancies (2, 3). R3Mab, an antibody that binds monospecifically to FGFR3, was tailored here for binding to other FGFR family members through multiple rounds of engineering, including recruiting desired binding to FGFR2 and removing undesired binding to FGFR4. The first step of engineering was carried out to recruit FGFR2 binding. Each phage library constituted mutagenesis of one contacting CDR, and the range of mutagenesis covered as many residues in that CDR as allowed by library size. Choosing multiple consecutive positions for mutagenesis permitted significant freedom in the CDR backbones. Most of the resulting clones that were able to engage FGFR2 harbored all 5 mutations in CDR H2. The crystal structure demonstrated that the full range of mutagenesis was coupled with complete remodeling of the geometry of the CDR loop. The solutions to spatial reorganizations of a CDR are numerous, as evidenced by the identification of diverse H2 mutants that had gained binding to FGFR2. Such a large variety of solutions is not typically seen as outcomes from standard affinity maturation experiments, whereby the recovered sequences usually contain sparse positions on individual CDRs. Therefore, acquiring additional specificity for homologous antigens may require larger mutagenesis freedom than affinity maturation. As a result, undesired specificity may be acquired through the process, such as the FGFR4 binding in these studies.

The second round of engineering was aimed to refine specificity by removing unwanted FGFR4 binding. Detailed structural analysis of contact residues between the antibody CDR loops and the antigen surface was used to guide the design of phage display libraries. Selected antibody variants showed reduction in unwanted FGFR4 binding with retention of binding to FGFR2/3. The sequence solutions to this step were much more limited as compared with the first round of engineering. No large backbone conformational changes would be expected at this stage. The refinement step further demonstrated the ability to exquisitely differentiate binding specificities among closely related antigens during antibody engineering.

The dual-specific antibodies generated here bind to 2 closely related antigens, namely FGFR2 and FGFR3. These 2B.1.3 antibody variants are regular IgG molecules in that they use identical heavy and light chains for the two antigen-binding fragments. They can potentially bind to 2 FGFR2 isoforms, 2 FGFR3 isoforms or 1 FGFR2 and 1 FGFR3 isoform in a bivalent or monovalent manner, respectively. This contrasts to conventional bispecific IgG, which commonly use two different heavy/light-chain pairs to bind to 2 different antigens in a monovalent manner. The dual-specific antibodies described herein share some similarities with “two-in-one” antibodies (24). Bostrom and colleagues randomized all 3 light-chain CDRs of herceptin and selected for a second specificity as well as the parental specificity. As expected, the second specificity comes from the dominant contributions of light-chain CDRs (24, 25). In one case, although EGFR and HER3 are homologous, the binding epitopes by an anti-EGFR/HER3 “two-in-one” antibody are different (25). The approach described here differs from “two-in-one” antibodies in that it appreciates the sequence and structure similarities between the 2 homologous antigens and focuses on a more limited set of mutagenesis so as to retain the parental epitope during engineering.

The antibody engineering presented here started from an existing and extensively characterized antibody, R3Mab, that has potential use for cancer therapy. Since introduction of the first therapeutic monoclonal antibody in the mid-1980s, there have been many clinically and commercially successful antibody drugs in different disease areas, including trastuzumab, cetuximab, adalimumab, bevacizumab, etc. These antibodies displayed exceptional activities in inhibiting their molecular targets. On the other hand, like the FGFR family, multiple homologous proteins are pursued as molecular targets for their various disease associations. Traditional discovery routes to obtain antibodies targeting a functional epitope, either animal immunization or other display-based library selections, are not guaranteed to be successful. The approach developed here to engineer a known antibody for acquiring specificity toward homologous targets provides an alternative route for antibody discovery. Moreover, it takes advantage of the favorable properties of previously developed antibodies by maintaining the functional epitopes and presumably the biologic functions as well. As the clinical antibody repertoire expands, more antibodies could be reengineered instead of being discovered ab initio. Potential applications may include protein families that comprise multiple members as disease targets, such as the EGFR (26), TNFR (27), TAM (28, 29), and Eph-Ephrin families (30). As in the traditional discovery processes, engineered antibodies toward homologs should be considered as new molecules and still need full characterization of their biochemical, biophysical, and biologic properties for any potential therapeutic applications.

R.M. Neve reports receiving commercial research support from and has ownership interest (including patents) in Genentech. M. Merchant has ownership interest (including patents) in Roche. P.J. Carter has ownership interest (including patents) in Genentech/Roche. No potential conflicts of interest were disclosed by the other authors.

Conception and design: Y. Yin, S. Djakovic, Y. Wu, M. Merchant, A. Ashkenazi, P.J. Carter

Development of methodology: Y. Yin, S. Djakovic, J. Tien, A. Ashkenazi, P.J. Carter

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Yin, S. Djakovic, S. Marsters, J. Tien, J. Peng, J. Tremayne, G. Lee, R.M. Neve, M. Merchant

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Yin, S. Djakovic, J. Tien, J. Peng, J. Tremayne, G. Lee, R.M. Neve, M. Merchant, A. Ashkenazi, P.J. Carter

Writing, review, and/or revision of the manuscript: Y. Yin, G. Lee, R.M. Neve, M. Merchant, A. Ashkenazi, P.J. Carter

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Marsters, J. Peng, J. Tremayne

Study supervision: J. Tien, J. Tremayne

The authors thank Charlie Eigenbrot, Ping Wu, and Weiru Wang for providing support on structural studies and Qing Jing, Mark Dennis, Wenwu Zhai, Isidro Hötzel, Thomas Hunsaker, and Kedan Lin for helpful discussions and collaborative studies. The Advanced Light Source (ALS) is acknowledged for providing a synchrotron X-ray source for collecting crystallographic data. The ALS is supported by the Director, Office of Science, Office of Basic Energy Sciences, of the U.S. Department of Energy under Contract No. DE-AC02-05CH11231.

The cocrystal structure of Mab 2B.1.3 and FGFR2-IIIb has been deposited in the Protein Data Bank (PDB) under accession code 4WV1.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1.
Goetz
R
,
Mohammadi
M
. 
Exploring mechanisms of FGF signalling through the lens of structural biology
.
Nat Rev Mol Cell Biol
2013
;
14
:
166
80
.
2.
Turner
N
,
Grose
R
. 
Fibroblast growth factor signalling: from development to cancer
.
Nat Rev Cancer
2010
;
10
:
116
29
.
3.
Beenken
A
,
Mohammadi
M
. 
The FGF family: biology, pathophysiology and therapy
.
Nat Rev Drug Disov
2009
;
8
:
235
53
.
4.
Wesche
J
,
Haglund
K
,
Haugsten
EM
. 
Fibroblast growth factors and their receptors in cancer
.
Biochem J
2011
;
437
:
199
213
.
5.
Mohammadi
M
,
Olsen
SK
,
Ibrahimi
OA
. 
Structural basis for fibroblast growth factor receptor activation
.
Cytokine Growth Factor Rev
2005
;
16
:
107
37
.
6.
Johnson
DE
,
Lu
J
,
Chen
H
,
Werner
S
,
Williams
LT
. 
The human fibroblast growth factor receptor genes: a common structural arrangement underlies the mechanisms for generating receptor forms that differ in their third immunoglobulin domain
.
Mol Cell Biol
1991
;
11
:
4627
34
.
7.
Kunii
K
,
Davis
L
,
Gorenstein
J
,
Hatch
H
,
Yashiro
M
,
Di Bacco
A
, et al
FGFR2-amplified gastric cancer cell lines require FGFR2 and Erbb3 signaling for growth and survival
.
Cancer Res
2008
;
68
:
2340
8
.
8.
Turner
N
,
Lambros
MB
,
Horlings
HM
,
Pearson
A
,
Sharpe
R
,
Natrajan
R
, et al
Integrative molecular profiling of triple negative breast cancers identifies amplicon drivers and potential therapeutic targets
.
Oncogene
2010
;
29
:
2013
23
.
9.
Dieci
MV
,
Arnedos
M
,
Andre
F
,
Soria
JC
. 
Fibroblast growth factor receptor inhibitors as a cancer treatment: from a biologic rationale to medical perspectives
.
Cancer Discov
2013
;
3
:
264
79
.
10.
Nord
H
,
Segersten
U
,
Sandgren
J
,
Wester
K
,
Busch
C
,
Menzel
U
, et al
Focal amplifications are associated with high grade and recurrences in stage Ta bladder carcinoma
.
Int J Cancer
2010
;
126
:
1390
402
.
11.
Byron
SA
,
Gartside
M
,
Powell
MA
,
Wellens
CL
,
Gao
F
,
Mutch
DG
, et al
FGFR2 point mutations in 466 endometrioid endometrial tumors: relationship with MSI, KRAS, PIK3CA, CTNNB1 mutations and clinicopathological features
.
PLoS One
2012
;
7
:
e30801
.
12.
Dutt
A
,
Salvesen
HB
,
Chen
TH
,
Ramos
AH
,
Onofrio
RC
,
Hatton
C
, et al
Drug-sensitive FGFR2 mutations in endometrial carcinoma
.
Proc Natl Acad Sci U S A
2008
;
105
:
8713
7
.
13.
Cappellen
D
,
De Oliveira
C
,
Ricol
D
,
de Medina
S
,
Bourdin
J
,
Sastre-Garau
X
, et al
Frequent activating mutations of FGFR3 in human bladder and cervix carcinomas
.
Nat Genet
1999
;
23
:
18
20
.
14.
di Martino
E
,
L'Hote
CG
,
Kennedy
W
,
Tomlinson
DC
,
Knowles
MA
. 
Mutant fibroblast growth factor receptor 3 induces intracellular signaling and cellular transformation in a cell type- and mutation-specific manner
.
Oncogene
2009
;
28
:
4306
16
.
15.
Yanochko
GM
,
Vitsky
A
,
Heyen
JR
,
Hirakawa
B
,
Lam
JL
,
May
J
, et al
Pan-FGFR inhibition leads to blockade of FGF23 signaling, soft tissue mineralization, and cardiovascular dysfunction
.
Toxicol Sci
2013
;
135
:
451
64
.
16.
Brown
AP
,
Courtney
CL
,
King
LM
,
Groom
SC
,
Graziano
MJ
. 
Cartilage dysplasia and tissue mineralization in the rat following administration of a FGF receptor tyrosine kinase inhibitor
.
Toxicol Pathol
2005
;
33
:
449
55
.
17.
Pai
R
,
French
D
,
Ma
N
,
Hotzel
K
,
Plise
E
,
Salphati
L
, et al
Antibody-mediated inhibition of fibroblast growth factor 19 results in increased bile acids synthesis and ileal malabsorption of bile acids in cynomolgus monkeys
.
Toxicol Sci
2012
;
126
:
446
56
.
18.
Sasaki
S
,
Ishida
T
,
Toyota
M
,
Ota
A
,
Suzuki
H
,
Takaoka
A
, et al
Interferon-alpha/beta and anti-fibroblast growth factor receptor 1 monoclonal antibody suppress hepatic cancer cells in vitro and in vivo
.
PLoS One
2011
;
6
:
e19618
.
19.
Bai
A
,
Meetze
K
,
Vo
NY
,
Kollipara
S
,
Mazsa
EK
,
Winston
WM
, et al
GP369, an FGFR2-IIIb-specific antibody, exhibits potent antitumor activity against human cancers driven by activated FGFR2 signaling
.
Cancer Res
2010
;
70
:
7630
9
.
20.
Qing
J
,
Du
X
,
Chen
Y
,
Chan
P
,
Li
H
,
Wu
P
, et al
Antibody-based targeting of FGFR3 in bladder carcinoma and t(4;14)-positive multiple myeloma in mice
.
J Clin Invest
2009
;
119
:
1216
29
.
21.
Kamath
AV
,
Lu
D
,
Gupta
P
,
Jin
D
,
Xin
Y
,
Brady
A
, et al
Preclinical pharmacokinetics of MFGR1877A, a human monoclonal antibody to FGFR3, and prediction of its efficacious clinical dose for the treatment of t(4;14)-positive multiple myeloma
.
Cancer Chemother Pharmacol
2012
;
69
:
1071
8
.
22.
Kunkel
TA
,
Bebenek
K
,
McClary
J
. 
Efficient site-directed mutagenesis using uracil-containing DNA
.
Methods Enzymol
1991
;
204
:
125
39
.
23.
Chen
VB
,
Arendall
WB
 III
,
Headd
JJ
,
Keedy
DA
,
Immormino
RM
,
Kapral
GJ
, et al
MolProbity: all-atom structure validation for macromolecular crystallography
.
Acta Crystallogr D Biol Crystallogr
2010
;
66
:
12
21
.
24.
Bostrom
J
,
Yu
SF
,
Kan
D
,
Appleton
BA
,
Lee
CV
,
Billeci
K
, et al
Variants of the antibody herceptin that interact with HER2 and VEGF at the antigen binding site
.
Science
2009
;
323
:
1610
4
.
25.
Schaefer
G
,
Haber
L
,
Crocker
LM
,
Shia
S
,
Shao
L
,
Dowbenko
D
, et al
A two-in-one antibody against HER3 and EGFR has superior inhibitory activity compared with monospecific antibodies
.
Cancer Cell
2011
;
20
:
472
86
.
26.
Baselga
J
,
Swain
SM
. 
Novel anticancer targets: revisiting ERBB2 and discovering ERBB3
.
Nat Rev Cancer
2009
;
9
:
463
75
.
27.
Ashkenazi
A
. 
Targeting death and decoy receptors of the tumour-necrosis factor superfamily
.
Nat Rev Cancer
2002
;
2
:
420
30
.
28.
Bosurgi
L
,
Bernink
JH
,
Delgado Cuevas
V
,
Gagliani
N
,
Joannas
L
,
Schmid
ET
, et al
Paradoxical role of the proto-oncogene Axl and Mer receptor tyrosine kinases in colon cancer
.
Proc Natl Acad Sci U S A
2013
;
110
:
13091
6
.
29.
Verma
A
,
Warner
SL
,
Vankayalapati
H
,
Bearss
DJ
,
Sharma
S
. 
Targeting Axl and Mer kinases in cancer
.
Mol Cancer Ther
2011
;
10
:
1763
73
.
30.
Pasquale
EB
. 
Eph receptors and ephrins in cancer: bidirectional signalling and beyond
.
Nat Rev Cancer
2010
;
10
:
165
80
.