Structural basis for KIT receptor tyrosine kinase inhibition by antibodies targeting the D4 membrane-proximal region

Significance

The receptor tyrosine kinase KIT is aberrantly activated primarily by somatic mutations in gastrointestinal stromal tumors and in a subset of acute myeloid leukemia, melanoma, and other cancers. Treatment of these cancers with tyrosine kinase inhibitors shows durable clinical response, but drug resistance and disease progression eventually occur in all patients. Here we describe monoclonal antibodies that block the activity of KIT and its oncogenic mutant. Structural and biochemical analyses of anti-KIT antibodies in complex with a KIT fragment demonstrated that KIT antibodies bind to a critical Achilles heel region that is essential for receptor activation. These antibodies may provide a potentially unique therapeutic approach for the treatment of tumors driven by WT or oncogenically mutated KIT.

Abstract

Somatic oncogenic mutations in the receptor tyrosine kinase KIT function as major drivers of gastrointestinal stromal tumors and a subset of acute myeloid leukemia, melanoma, and other cancers. Although treatment of these cancers with tyrosine kinase inhibitors shows dramatic responses and durable disease control, drug resistance followed by clinical progression of disease eventually occurs in virtually all patients. In this report, we describe inhibitory KIT antibodies that bind to the membrane-proximal Ig-like D4 of KIT with significant overlap with an epitope in D4 that mediates homotypic interactions essential for KIT activation. Crystal structures of the anti-KIT antibody in complex with KIT D4 and D5 allowed design of affinity-matured libraries that were used to isolate variants with increased affinity and efficacy. Isolated antibodies showed KIT inhibition together with suppression of cell proliferation driven by ligand-stimulated WT or constitutively activated oncogenic KIT mutant. These antibodies represent a unique therapeutic approach and a step toward the development of “naked” or toxin-conjugated KIT antibodies for the treatment of KIT-driven cancers.

The receptor tyrosine kinase (RTK) KIT is a transmembrane protein that plays crucial roles in mediating diverse cellular processes including cell differentiation, proliferation, and cell survival, among other activities. These processes occur through activation of KIT upon binding by stem cell factor (SCF), a ligand found in membrane-anchored and soluble forms (1, 2) in a variety of cell types, including hematopoietic stem cells, germ cells, vascular endothelial cells, and the mesenchymal cells with uniquely neuromuscular differentiation known as the interstitial cells of Cajal (3, 4). KIT belongs to the type III subfamily of RTKs (5), a family composed of an extracellular region that includes five Ig-like domains (designated D1–D5), a single transmembrane domain (TM), a juxtamembrane region (JM), a tyrosine kinase domain split by a kinase insert, and a C-terminal tail (6) (Fig. 1A).

Fig. 1.

Structure of the Fab19–KIT_D4-5 complex. (A) Schematic representation of WT KIT as well as the KIT_D4-5 fragment that was used in this study. Glu381 and Arg386, residues important for KIT activation, are marked on KIT_D4-5 fragment. D1 to D5 are Ig-like domains 1 to 5 respectively. JM, juxtamembrane; KI, kinase insert; PTK, protein tyrosine kinase domain; TM, transmembrane. (B) Surface representation of Fab19–KIT_D4-5 complex. (Left) View following 90° rotation along the shown axis. D4 and D5 of KIT are colored with orange and pink, and heavy and light chains of Fab19 with blue and green, respectively. The regions of interactions between Fab19 and KIT_D4-5 are visualized in “open-book” format for D4 (C) and Fab (D). The buried surface area on the D4 side is shown in red in C and is in the same scale as in B. The interactions between Fab19 and D4 are mediated by CDR loops and surrounding residues. The buried interface surface of the Fab19 is color coded by CDR: H1, cyan; H2, magenta; H3, gray; L2, bright green; L3, yellow (D). Fab19 in D is enlarged compared with images in B and C.

Based on the determination of the crystal structure of the complete extracellular region of KIT before and after ligand stimulation (7), and the tyrosine kinase domain (8, 9), a mechanism has been proposed whereby activation of KIT is initiated through receptor dimerization (10). Dimerization of the KIT ectodomain is ligand-driven and initiated through high-affinity binding of an SCF dimer to the membrane distal Ig-like domains (D1–D3) of the receptor (11, 12). Cross-linking of the membrane distal Ig-like domains with their ligand dramatically increases local concentration of the membrane-proximal (D4 and D5) and TMs to enable homotypic contacts by weak D4–D4 and D5–D5 interactions between neighboring KIT molecules. These homotypic associations promote conformational rearrangements that permit correct association between neighboring cytoplasmic regions of KIT dimers resulting in autophosphorylation, tyrosine kinase stimulation, recruitment of signaling proteins, and cell signaling.

Dysregulation of KIT, by a variety of different somatic as well as rare germ-line mutations, has been associated with numerous hematopoietic and other cancers, including gastrointestinal stromal tumors (GIST), a subset of melanomas, systemic mastocytosis, and acute myeloid leukemia. Most reoccurring activating mutations map to the cytoplasmic JM region (exon 11) and to the membrane-proximal Ig-like domain D5 (exon 9) of the extracellular region (13, 14). Currently, initial treatment for patients with GIST involves the small-molecule kinase inhibitor Gleevec (imatinib), which can efficiently block kinase activity of JM and D5 mutants of KIT. Unfortunately, most patients with GIST develop resistance to the drug within 2 y of treatment by acquiring additional mutations usually mapped to exons 13 and 14 (V654A and V670I, respectively). Patients resistant to Gleevec are usually treated with the kinase inhibitor Sutent (sunitinib), which can more efficiently inhibit WT KIT protein as well as many of the mutations that confer imatinib resistance. In addition to Gleevec-resistant mutations in exons 13 and 14, kinase domain mutants (exon 17) that are resistant to Gleevec and Sutent are also seen in patients with GIST (15).

Several mAbs targeting RTKs have shown promising results as anticancer therapies, including mAbs against members of the EGFR family of RTKs, which were approved for clinical use, including cetuximab and panitumumab (anti-EGFR mAbs), as well as trastuzumab and pertuzumab (anti-ErbB2 mAbs) (16, 17). Antibodies can be highly specific for their targets, and therefore off-target side effects are reduced compared with small-molecule kinase inhibitors. Specific targeting of oncogenic RTKs by inhibitory mAbs may allow the resistance that frequently occurs in patients treated with kinase inhibitors to be surmounted.

We have shown previously that homotypic interactions between the membrane-proximal domains of KIT (D4 and D5) are critical for receptor activation and that disruption of the D4–D4 interface strongly compromises receptor activation (7). As KIT oncogenic mutants located in D5 are dependent on homotypic contacts between neighboring ectodomains, it is reasonable to expect that activation of these mutants could be inhibited by monoclonal antibodies directed against KIT domains D4 or D5. Here, anti-KIT antibodies were isolated from a naive, phage-displayed synthetic antibody library. Crystal structures of a fragment antigen-binding (Fab) in complex with KIT membrane-proximal domains D4 and D5 (KIT_D4-5) revealed binding to D4 that overlapped significantly with an epitope required for homotypic interactions essential for SCF-dependent KIT activation. Furthermore, information obtained from the structure of antibody in complex with KIT was applied to guide the design of affinity maturation libraries, enabling isolation of antibody variants with increased binding affinity equating to increased efficacy. These antibodies were capable of efficient inhibition of KIT activity that led to suppression of cell proliferation and provide a potentially unique therapeutic approach for the treatment of tumors driven by WT or aberrantly activated KIT mutants.

Results

The two membrane-proximal Ig-like domains of KIT (KIT_D4-5 fragment; Fig. 1A) determined structurally to be critical for KIT activation (7) were used as an antigen to isolate binding Fabs from a naive phage-displayed library (library F; Fig. S1) containing more than 10¹⁰ unique clones (18, 19). The binding properties of the phage-derived Fabs were compared with the binding properties of a murine monoclonal anti-KIT antibody designated KTN37 that was obtained by immunization of mice with the same antigen. A variety of in vitro binding experiments using 3T3 cells ectopically expressing WT KIT demonstrated that the phage-derived Fabs and the KTN37 mAb bind specifically to D4 of recombinant isolated KIT or to native KIT molecules expressed on the cell surface of live cells. The structure of one of the most potent phage-derived Fab, designated Fab19 (Fig. S1), in complex with KIT_D4-5 (Fig. 1A), was further analyzed by X-ray crystallography.

The Structure of Fab19–KIT_D4-5 Complex.

A purified complex composed of Fab19 together with KIT_D4-5 was subjected to extensive screening for crystal growth and further optimization. We obtained crystals that belong to the C2 space group with a single 1:1 complex of KIT_D4-5 and Fab19 in the asymmetric unit. The structure of this complex was determined to 2.4-Å resolution (SI Materials and Methods, Fig. 1B and Table S1).

The overall structure of KIT_D4-5 bound to Fab19, is very similar to the structures of these two Ig-like domains observed previously as part of the structures of full-length extracellular region of KIT alone, or in complex with SCF [Protein Data Bank (PDB) ID codes 2EC8 and 2E9W; ref. 7]. Superposition of individual D4 and D5 from Fab19–KIT_D4-5 complex structure with corresponding domains of KIT ectodomain structure (PDB ID code 2EC8) revealed rmsd values of 0.65 Å for 96 and 59 Cα residues in D4 and D5, respectively. The structure revealed Fab19 binding exclusively to D4 of KIT with a buried surface of 1,029 Ų on the D4 side of the interface (Fig. 1C and Table S2). Nearly the entire β-sheet of D4 (one of two β-sheets in Ig-like domain), including βA, βB, β, and βD, as well as the AA′, A′B, EF, and DE loops, was buried under the Fab19 surface (Fig. 1D and Fig. S2).

The majority of the contacts were made by the heavy chain of the Fab (800 Ų vs. 283 Ų for the light chain; Fig. 1D, Fig. S2, and Table S2), with most key interactions mediated by the complementarity-determining region (CDR) loops of the Fab, including all three CDRs of the heavy chain and L2 and L3 of the light chain. The majority of the specificity-determining contacts came from CDRs H2 and H3 (additional details about Fab19–D4 interface is described in SI Appendix).

Fab19 Contacts Critical for KIT Receptor Inhibition.

We have shown previously that D4–D4 homotypic interactions mediated by two salt bridges between Arg381^D4 and Glu386^D4, both located in the EF loop (Fig. 2A), are critical for proper ligand dependent receptor activation (7). Analysis of the Fab19–KIT_D4-5 complex structure revealed that Arg381^D4 makes contact with CDR L2 of Fab19 (Fig. 2B); the Arg381^D4 side chain makes hydrogen bonds with the side chain of Tyr49^L and the main chain of Leu54^L. In addition, Tyr101^H of CDR H3 makes contact with Thr380^D4 (Fig. S3) of the EF loop, which is involved in mediating D4–D4 homotypic interactions. The structure revealed that contacts between Fab19 and D4 of KIT occluded Arg381^D4 from forming a salt bridge with Glu-386 of D4 from a neighboring KIT receptor, thereby preventing proper lateral association between membrane-proximal domains. As a consequence, homotypic contacts between the cytoplasmic domains, transphosphorylation and KIT activation were inhibited (as detailed later).

Fig. 2.

D4–D4 homotypic interactions are occluded by Fab19 binding. (A, Left) Surface representation of KIT–SCF 2:2 complex. D4 and D5 are colored as in Fab19–KIT_D4-5 complex in Fig. 1; D1, D2, and D3 are in red, and SCF is in brown. (A, Right) Cartoon representation of D4–D4 homotypic interactions, mediated by two salt bridges between Arg381 and Glu386. Arg381 and Glu386 side chains are shown with sticks, and hydrogen bonds are shown as dashes. (B, Left) Surface representation of Fab19–KIT_D4-5 complex, with the same color code as in Fig. 1. Residues Arg381 and Glu386, which mediate D4–D4 homotypic interactions, are indicated by arrows. (B, Right) Close-up view of the interactions between Arg381 and CDR L2 of Fab19; cartoon representation. Side chains for Arg381, Glu386 (D4), Tyr49 (light chain), and main chain of Leu54 (light chain) are shown with sticks, and hydrogen bonds are shown as dashes.

Affinity Maturation of Fab19.

To improve the binding affinity of Fab19 to D4, affinity maturation was performed in a two-step process. The first step was performed before solving of the Fab19–KIT_D4-5 complex structure and involved soft randomization of individual CDRs wherein targeted residues were likely to be kept parental, thereby minimizing the likelihood of changing epitopes during Fab maturation. The four CDRs into which diversity was introduced in library F, CDR L3 and all heavy-chain CDRs, were targeting individually in affinity maturation library design and resulted in the isolation of numerous variants from all four libraries. Variants from the CDR H1 library showed strong improvement over the parental Fab19, and, in particular, Fab12I, estimated to have one of the strongest binding affinity (Fig. S4 A–C) and KIT receptor inhibition effect, was chosen for further affinity maturation with the aid of the newly solved Fab19–KIT_D4-5 complex.

The structure of the Fab19–KIT_D4-5 complex was used to guide affinity maturation library design after analysis of the antibody–antigen interface revealed that, whereas the heavy chain of Fab19 makes extensive contacts with D4, light-chain interactions were few and weak (Fig. 1D). In particular, no contacts were observed between CDR L1 and D4 (Fig. 3A). Considering that the light chain makes most of the contact with Arg381^D4, a residue that is important for formation of D4 homotypic contacts for receptor activation, affinity maturation libraries targeting CDR L1 were designed. Length variation between four and seven residues was incorporated into library design in addition to allowing for possible CDR H1 S31V and/or S33M substitutions found in the parental Fab12I CDR H1 template. Eight unique Fabs (Fab79A–H) were isolated with CDR L1 lengths ranging from four to seven residues, suggesting that increased length allowed additional contacts with D4 (Fig. S4D). Interestingly, all clones were found to contain double S31V S33M substitutions in the CDR H1, indicating the importance of these nucleophilic to hydrophobic substitutions (detailed in SI Appendix) (Fig. S9).

Fig. 3.

Comparison of Fab19–KIT_D4-5 and Fab79D–KIT_D4-5 complex structures. (A and C) Cartoon representations of contacts between CDR L1 of Fab19 (A) and Fab79D (C) and βD of D4. Side chains are shown as sticks, and hydrogen bonds are shown with dashes. (B) Structural alignment of VL, VH, and D4 domains from Fab19–KIT_D4-5 and Fab79D–KIT_D4-5 structures (ribbon representation). Fab79D complex is colored as follows: D4, orange; VL, green; VH, cyan. Fab19 complex is colored as follows: D4, red; VL, yellow; VH, dark blue. CDRs L1 are highlighted with magenta for Fab79D and with black for Fab19.

Binding of Anti-D4 Fabs to KIT_D4-5 Fragment.

We used surface plasmon resonance (SPR) analysis to quantitatively characterize the kinetics and dissociation constants of different generations of affinity-matured anti-D4 Fabs (Fab19, Fab12I, and Fab79D). As a control, we used the Fab of the murine antibody KTN37 (as detailed later). Purified KIT_D4-5 fragment, was covalently attached onto a CM5 biosensor chip and serial dilutions of each Fab were flowed over the biosensor surface to reveal the binding kinetics (Table 1). Values for the association and dissociation rates showed that the initial phage-derived synthetic Fab19 possessed high binding affinity with a K_d value of 0.63 nM, similar to the affinity of the mouse-derived KTN37 Fab, which had a K_d value of 0.25 nM (Table 1). Although Fab19 and KTN37 Fab had similar affinities, the kinetics were very different: KTN37 bound to KIT with a high association rate but relatively fast dissociation rate whereas Fab19 showed slower association and dissociation rates (Fig. S5). In two steps of affinity maturation, improved binding affinity by two orders of magnitude were observed, bringing binding affinity from the subnanomolar (K_d = 0.63 nM for Fab19) to the picomolar range (K_d = 6.4 pM for Fab79D).

Table 1.

Antibody binding kinetics to KIT D4-D5

Fab	Ka, 1/Ms	Kd, 1/s	Kd, M
KTN37	(1.5 ± 0.4) × 107	(3.6 ± 0.5) × 10−3	(2.5 ± 0.3) × 10−10
Fab19	(1.0 ± 0.7) × 106	(6.6 ± 0.2) × 10−4	(6.3 ± 0.5) × 10−10
Fab12I	(1.4 ± 0.3) × 106	(2.7 ± 0.3) × 10−5	(2.0 ± 0.6) × 10−11
Fab79D	(2.4 ± 0.7) × 106	(1.5 ± 0.5) × 10−5	(6.4 ± 0.6) × 10−12

Structure of Fab79D–KIT_D4-5 Complex.

To understand how affinity maturation improved the binding properties of Fab79D compared with Fab19 and to confirm that the binding epitope remained unchanged, we performed crystallization of the Fab79D–KIT_D4-5 complex. The structure was solved by molecular replacement using the Fab19–KIT_D4-5 complex as a search model. Data collection and refinement statistics are presented in Table S1.

Comparison between individual domains of the two structures of Fab19 and Fab79D in complex with KIT_D4-5 fragments showed them to be highly similar with the biggest difference in the elbow angle between D4 and D5 and the elbow angle between the variable and constant domains of the Fabs. Superposition of the constant domains of Fab19 and Fab79D and KIT D5 domains within complex structures revealed rmsd values of 0.41 Å and 0.74 Å for Cα residues, respectively. Similarly, variable domains and D4 were not altered significantly, with rmsd values of 0.55 Å for 279 Cα residues of the VL and VH domains and D4 (Fig. 3B). As expected, the biggest difference between the two complexes was found in the CDR L1 loop, which was targeted with length diversity during affinity maturation. Fig. 3B shows that the L1 loop of Fab79D moved toward the D4 domain within the Fab79D–KIT_D4-5 complex structure, and, unlike Fab19, made contact with βD of D4 (Fig. 3C and Fig. S6); Arg31^L and Asn32^L of Fab79D were located within hydrogen bonding distance of the main chain of Pro363^D4 and side chain of Glu360^D4, respectively. This CDR L1 loop extension, so evident upon complex structure comparison, appears to be responsible for the increased binding affinity of Fab79D toward KIT D4.

KTN37–Murine Anti-D4 mAb.

As a positive control in our experiments, we used KTN37 mAb, a monoclonal antibody obtained by immunization of mice with the KIT_D4-5 fragment. It was shown that KTN37 IgG bound D4 of human KIT with high affinity, and was a very potent antagonist of the KIT receptor (as detailed later). As we were not able to obtain diffraction quality crystals of KTN37 in complex with KIT D4 and D5 fragment, molecular details of the complex could not be obtained. However, to shed light on the binding epitope of KTN37, we compared the KTN37 IgG binding to the ectodomain of KIT from different species (Fig. S7A). Our results showed that this mAb was cross-reactive with KIT from human, monkey, dog, and cat, but not mouse or rat. Sequence alignment of D4 of KIT for these six species revealed one major spot with a few residues spread around D4, which are different in the rodent sequences (Fig. S7B). We mapped all residue differences onto the D4 structure and found out that most of these residues cover two continuous areas on the surface of D4 (Fig. S8), suggesting that these two regions may represent the binding epitope of KTN37 Fab to D4. It is worth noting that these regions are located right on top of the D4 homotypic contact interface (Fig. S8), and, most likely, binding of KTN37 significantly overlaps with the region essential for mediating D4–D4 homotypic interactions.

KIT Autophosphorylation Is Inhibited by Anti-D4 Antibodies in Living Cells.

The crystal structure of Fab19–KIT_D4-5 and Fab79D–KIT_D4-5 complexes revealed significant overlap between the Fab binding epitope and the D4–D4 homotypic interface (Fig. S8), important for KIT activation, suggesting that binding of these Fabs likely interferes with KIT receptor autophosphorylation. To test for KIT inhibition, National Institutes of Health (NIH) 3T3 cells expressing WT KIT receptor were incubated for 5 h with varying concentrations of Fabs or IgGs before SCF stimulation. All three generations of synthetic Fabs, together with Fab KTN37, were tested, along with the IgG version of Fab79D (IgG 79D) and IgG KTN37. As predicted from the structure, SCF-stimulated autophosphorylation of KIT was strongly inhibited upon binding of anti-D4 Fabs and IgGs (Fig. 4A).

Fig. 4.

WT KIT activation is inhibited by anti-D4 antibodies. (A) NIH 3T3 cells expressing WT KIT were incubated for 5 h with the indicated concentrations (Upper) of Fabs or IgGs following stimulation with 25 ng/mL of SCF for 5 min at 37 °C. Lysates of unstimulated or SCF-stimulated cells were subjected to immunoprecipitation (IP) with anti-KIT antibodies followed by SDS/PAGE and immunoblotting (IB) with anti-KIT or anti-phosphotyrosine (pY) antibodies. (B and C) Ba/F3 cells expressing WT KIT were plated in six-well plates and grown in the presence of 250 ng/mL SCF and different concentrations as indicated (Lower) of Fabs (B) or IgGs (C). Cell numbers were measured after 72 h relative to 0 h time points.

During the two steps of affinity maturation, the inhibitory properties of the anti-D4 Fabs were improved significantly; KIT inhibition by Fab19, the parental synthetic Fab, occurred at 50 nM, whereas the last-generation Fab, Fab79D, inhibited KIT at 5 nM, suggesting that increased affinity correlated to increased KIT inhibition (Fig. 4A and Table 1). Consistent with this, Fab12I appears to be more effective at KIT inhibition than Fab19 but weaker than Fab79D.

The bivalent IgG format confers avidity effects to a Fab that are evident upon testing IgG KTN37, whose effectiveness at blocking KIT autophosphorylation could be seen even at 0.5 nM, compared with the 50 nM level required for Fab KTN37 (Fig. 4A). Interestingly, much less gain in KIT inhibition from avidity could be seen upon conversion of Fab79D to IgG 79D (5 nM for Fab79D and 1 nM for IgG 79D).

Anti-D4 Antibodies Efficiently Inhibit Proliferation of KIT-Dependent Ba/F3 Cells.

We next examined the effect of the Fabs and IgGs on KIT-mediated cell proliferation. We constructed stable Ba/F3 cell lines expressing WT KIT (Ba/F3 KIT^wt) or an oncogenic AY502,503 duplication KIT mutant (Ba/F3 KIT^AY502,503dup). The parental Ba/F3 cells are an IL-3–dependent murine pro-B cells lacking endogenous KIT expression. Upon exogenous WT KIT expression, Ba/F3 cells become dependent on SCF stimulation of these cells. Moreover, expression of constitutively active KIT (such as the AY502,503 duplication in D5) induces transformation of these cells (20). Incubation of these lines with anti-D4 Fabs and IgGs revealed varying cell proliferation inhibition of the D4 binders that correlated with results seen in the KIT autophosphorylation inhibition experiments (Figs. 4 and 5). Like in the inhibition of KIT autophosphorylation experiment, Fab19 could inhibit Ba/F3 KIT^wt cell proliferation only at the highest concentration (500 nM), whereas subsequent generations of anti-D4 Fab showed significant improvements to phosphorylation and proliferation inhibition; indeed, Fab79D showed significantly improved inhibitory properties that could inhibit cell growth at 50 nM. KTN37 Fab showed a relatively weak inhibitory effect (200 nM); however, avidity effects contributed significantly upon testing of IgG KTN37, whereupon cell proliferation inhibition could be seen at 1 nM. Conversely, conversion of Fab79D to IgG 79D led to decreased efficacy in cell proliferation assays.

Fig. 5.

Proliferation of Ba/F3 cells expressing A502,Y503 duplication oncogenic KIT mutant is inhibited by anti-D4 antibodies. Ba/F3 cells expressing the dupA502,Y503 KIT mutant were plated in six-well plates. Cells were grown in the presence of different concentrations of Fabs (A) or IgGs (B) in the absence of IL-3 or SCF (as indicated). Fold increase of cell number was measured after 72 h.

As with the Ba/F3 KIT^wt cells overexpressing WT KIT, significant improvements to cell proliferation inhibition of unstimulated Ba/F3 KIT^AY502-3dup could also be seen with successive generations of affinity-matured Fabs (Fig. 5). As observed with Ba/F3 KIT^AY502-3dup cells, Fab79D (5 nM) proved to be a more effective inhibitor than Fab KTN37 (50 nM); however, IgG KTN37 (1 nM) showed marked improvement whereas the efficacy of IgG 79D (10 nM) was less than that of Fab79D (5 nM). Although IgG KTN37 was a more sensitive cell proliferation inhibitor, able to inhibit proliferation of Ba/F3 KIT^AY502-3dup at concentrations lower than that of IgG 79D, overall levels of proliferation were lower in IgG 79D-treated cells than in those treated with IgG KTN37, revealing IgG 79D potency in inhibition of cell proliferation (Fig. 5).

Discussion

Tyrosine kinase inhibitors have been applied successfully in the clinic for treating patients with cancer whose tumors were driven by activated RTKs. Indeed, Gleevec and Sutent have been successfully developed as important therapeutic options for the treatment of GIST driven by activated KIT (14). Although most patients with KIT-driven GIST respond well when treated with tyrosine kinase inhibitors, sometimes with several years of remission, eventually most cancers relapse because of drug resistance (13, 14). An alternative and complementary approach is to use therapeutic monoclonal antibodies that target the extracellular region of KIT, including mutants that become resistant to tyrosine kinase inhibitors. Moreover, recent advances in the generation of potent and selective toxin conjugates of RTKs may provide an opportunity to actually kill the tumor cells and eliminate the tumor entirely. In this report, we present a step toward accomplishing this goal.

Rational design of a drug that targets the extracellular region of KIT became possible upon solving of the crystal structure and elucidating the mechanism of ligand-induced or oncogenic KIT activation. Crystal structures of the extracellular regions of KIT before and after ligand stimulation (7) strongly emphasized the role of membrane-proximal domain homotypic interactions in receptor activation. Upon ligand binding, the weak homotypic interactions between membrane proximal D4 and D5 allow precise positioning of the two C-terminal regions of the receptor ectodomains in a manner and distance important for positioning of the TM domains in the correct orientation that enable activation of the cytoplasmic tyrosine kinase domain (21–24). Moreover, disruption of homotypic contacts in the membrane-proximal domains of KIT (7), PDGF receptor (22) and VEGF receptor 2 (21) strongly impairs receptor activation. The importance of these domains in activation strongly suggested specific antagonists to the KIT receptor could be generated through membrane proximal domain targeting. Specificity in drug design can be achieved through antibodies that bind with high affinity to their target in a manner necessary to avoid unwanted side effects caused by off-target interactions. Indeed, it was shown previously (25) that monoclonal antibodies against D4 domain of KIT can inhibit receptor activation. Furthermore, the advent of fully human synthetic antibody libraries allows rapid selection of binders specific to their target, coupled with an ability to affinity-mature initially isolated variants for desired attributes such as increased affinity and specificity (18). Here we describe the isolation and maturation of an antibody directed against KIT D4 that strongly impairs receptor activation and cell proliferation to a level that could be potentially used in cancer therapy.

We combined phage display with structural guidance to develop anti-KIT antibodies that hold therapeutic promise. Structural analysis of the Fab19–KIT_D4-5 complex allowed suboptimal contacts to be determined, thereby permitting facile and focused affinity maturation library design. Indeed, structural analysis revealed that most contacts were mediated by the heavy chain of Fab19, whereas contacts between the light chain and D4 of KIT were crucial for inhibition of KIT receptor. Considering the importance of the light chain contribution to KIT inhibition, affinity maturation libraries were directed toward CDR-L1, lacking any contact between parental Fab19 and KIT D4, that included length variation in attempt to create contacts. This structure-guided strategy proved successful, as significant improvement to inhibitory properties of the final variant (Fab79D) could be seen, especially compared with those of the affinity-matured variant isolated in the absence of structural information (Fab12I). The crystal structures of these inhibitory Fabs in complex with KIT_D4-5 revealed clear steric blocking of homotypic contacts between the membrane proximal domains of KIT by the Fabs that led to inhibition of receptor autophosphorylation and activation. Considering the conservation of the mechanism of activation of RTKs and the central role played by homotypic contacts between membrane proximal regions in RTK activation similar strategies could be applied to design inhibitors of most, if not all, RTKs.

Despite the affinity of the designed Fab fragment, Fab 79D, being nearly twofold higher than that of Fab KTN37, it was surprising to see such pronounced inhibition by IgG KTN37. Because of the fast association and dissociation rates of Fab KTN37, which result in relatively low KIT binding affinity, a pronounced increase in the binding affinity and inhibitory capacity of IgG KTN37 is achieved by the strong impact of increased valency resulting from an avidity effect. By contrast, the high binding affinity of Fab 79D is primarily caused by the very slow dissociation rate of the monovalent 79D antibody toward KIT binding, and therefore the impact of increased valency (avidity effect) on the binding and inhibitory capacity of IgG 79D toward KIT is rather minimal. Although the two antibodies use different mechanisms for KIT binding, each antibody binds to KIT with high affinity and exerts strong inhibition of KIT activity and function. Comparison of the binding properties of KTN37 to KIT from different species suggests that the binding epitope of Fab KTN37 lies directly on top of the region responsible for mediating the D4–D4 homotypic contacts. However, in the absence of structural information about the exact contact region of Fab KTN37 with KIT D4, it is impossible to compare the exact molecular mechanism of inhibition of the two antibodies. Both antibodies, IgG 79D and IgG KTN37, inhibit proliferation of Ba/F3 cells expressing WT KIT or an oncogenic mutant located in D5 (exon 9) associated with GIST. These two antibodies affect proliferation at nanomolar concentrations and can therefore potentially be used for anticancer therapy and may help to overcome resistance that frequently occurs in patients treated with tyrosine kinase inhibitors. Recent studies have shown that, although mutations to KIT exon 11 (JM domain) are highly sensitive to Gleevec, sensitivity of KIT exon 9 mutants (D5) is at least 10 times lower (14). It is thus not surprising that patients with exon 11 mutations have significantly better disease-free progression and survival than patients with mutations in exon 9, and has led to the idea that resistance to the drug might be caused by subtherapeutic Gleevec dosing in patients with a mutation to KIT exon 9 (13, 14). To address this insensitivity and resistance to Gleevec, the antibodies developed in the present study have the potential to be used as specific inhibitors of KIT exon 9 mutants. Furthermore, these antibodies can be used in combination with tyrosine kinase inhibitors to allow the use of lower doses to delay or prevent the occurrence of resistance. Finally, the advances in development of potent and selective toxin conjugates may provide an opportunity to develop anti-KIT toxin conjugates that will kill and eliminate the tumor entirely. Here we present a step toward reaching this goal.

Materials and Methods

Proteins Expression and Purification.

Soluble KIT_D4-5 fragment (amino acids 308–514) was expressed in baculovirus Sf9 cells and purified by affinity, size-exclusion, and anion exchange chromatography. All Fab fragments were expressed in Escherichia coli and purified by affinity and cation-exchange chromatography. Further details are provided in SI Materials and Methods.

Crystallization and Data Collection.

Fab19–KIT_D4-5 and Fab79D–KIT_D4-5 complexes were crystallized by hanging-drop vapor diffusion methods at 21 °C. Single crystals for both complexes were obtained by macroseeding. For crystallization of Fab19–KIT_D4-5, crystallization buffer containing 13% PEG 3350, 0.5 M MgCl₂, and 0.1 M Tris⋅HCl, pH 9.0, was mixed with equal volume (0.6 μL) of protein solution (7 mg/mL). Single crystals were dehydrated by transferring into cryoprotectant solution containing 22% PEG 3350, 0.5 M MgCl₂, 0.1 M Tris⋅HCl, pH 9.0, and 30% ethylene glycol, and were incubated over the reservoir of this buffer for 2 to 3 d. Crystals were flash-frozen in cryoprotectant solution. Crystals of Fab79D–KIT_D4-5 were obtained by mixing crystallization buffer containing 20% to 24% PEG 400 and 0.1 M Tris⋅HCl, pH 8.2, with protein sample (6.5 mg/mL). Crystals were flash frozen in the reservoir solution supplemented with PEG 400 up to 35%. X-ray diffraction data were collected at the X25 beamline of National Synchrotron Light Source, Brookhaven National Laboratory. Data collection statistics are summarized in Table S1.

The structures of Fab19–KIT_D4-5 and Fab79D–KIT_D4-5 complexes were solved by molecular replacement using the PHASER program (26) under the CCP4 software suite (27) (SI Appendix).

Phage Display Selection and Characterization.

Phage pools consisting of a phage-displayed synthetic antibody library (library F) were cycled through five rounds of selections by using KIT_D4-5 immobilized on 96-well MaxiSorp immunoplates (Thermo Scientific) as antigen, as described previously (18). Culture supernatants of 96 clones from each of rounds four and five grown in 96-well format were used directly in phage ELISAs to identify clones binding KIT specifically (using BSA as a negative control). KIT-specific clones were subjected to DNA sequence analysis. Unique clones were subjected to competitive ELISAs that allow for affinity estimation and rank ordering among clones.

Fab Affinity Maturation.

KIT affinity maturation libraries were constructed as described previously (19). Briefly, affinity-maturation libraries targeting CDR-H1 were constructed by introducing TAA stop codons into CDR-H1 of phagemid Fab19. The resulting phagemid was used as a template for a mutagenesis reaction that replaced stop codons in CDR-H1 with oligonucleotides mixed in a 70%:10%:10%:10% ratio whereby parental nucleotides were represented at 70% and the remaining nucleotides at 10%. In this manner (soft randomization), targeted residues were biased toward parental but still allowed all 19 possible substitutions.

Cell Culture and Ba/F3 Proliferation Assay.

NIH 3T3 cells stably expressing WT-KIT were previously described (7). Further details are provided in SI Materials and Methods. Ba/F3 cells were grown in RPMI medium 1640 supplemented with 10% of FBS and 10 ng/mL recombinant murine IL-3. WT and A502,Y503 duplication KIT mutant were cloned into pMSCVpuro vector and transfected into Ba/F3 cells by using electroporation. Stable cell lines expressing WT or mutant KIT were selected in the presence of puromycin and IL-3. After establishing stable cell lines, IL-3 was withdrawn and cells expressing WT KIT were supplemented with 250 nM SCF.

Ba/F3 cells expressing WT or mutant KIT were plated in six-well plates at 400,000 per well in 2 mL media at day 0. Fab or IgG were added to each well at specified concentrations. At 72 h later, cell number was determined by using a Scepter Handheld Automated Cell Counter. Cell number increase is expressed as fold change compared with day 0.