Crystal Structure of the Tyrosine Kinase Domain of Colony-stimulating Factor-1 Receptor (cFMS) in Complex with Two Inhibitors*

The cFMS proto-oncogene encodes for the colony-stimulating factor-1 receptor, a receptor-tyrosine kinase responsible for the differentiation and maturation of certain macrophages. Upon binding its ligand colony-stimulating factor-1 cFMS autophosphorylates, dimerizes, and induces phosphorylation of downstream targets. We report the novel crystal structure of unphosphorylated cFMS in complex with two members of different classes of drug-like protein kinase inhibitors. cFMS exhibits a typical bi-lobal kinase fold, and its activation loop and DFG motif are found to be in the canonical inactive conformation. Both ATP competitive inhibitors are bound in the active site and demonstrate a binding mode similar to that of STI-571 bound to cABL. The DFG motif is prevented from switching into the catalytically competent conformation through interactions with the inhibitors. Activation of cFMS is also inhibited by the juxtamembrane domain, which interacts with residues of the active site and prevents formation of the activated kinase. Together the structures of cFMS provide further insight into the autoinhibition of receptor-tyrosine kinases via their respective juxtamembrane domains; additionally the binding mode of two novel classes of kinase inhibitors will guide the design of novel molecules targeting macrophage-related diseases.

Csf1 op mice are resistant to collagen-induced arthritis and show a reduced rate of mammary tumor progression into metastasis (6). Similar, but more pronounced effects, were observed with a mouse model in which the cFMS gene had been inactivated (Csfr Ϫ /Csfr Ϫ ) (7). CSF-1 has been shown to exacerbate collagen-induced arthritis in mouse models, an effect that could be suppressed with M-CSF-blocking antibodies (8). In another study, CSF-1 and GM-CSF (granulocyte macrophage colony-stimulating factor) induced prolonged inflammation and recruitment of macrophages in an mBSA-induced arthritis model (9). These studies demonstrated that CSF-1 and GM-CSF could exacerbate and prolong the histopathology of acute inflammatory arthritis and lend support to the hypothesis that monocytes/ macrophages are a driving influence in the pathogenesis of inflammatory arthritis. The data shown in these studies suggest that either CSF-1 or its cognate receptor cFMS are suitable targets for treating arthritis or other macrophage-induced inflammatory diseases.
In a recent study, expression of cFMS in various tumors was linked to poor survivability and increased tumor size (10). CSF-1 and cFMS are also expressed in some types of breast carcinomas and other epithelia of the female reproductive tract, where activation of the receptor by ligand, produced either by the tumor cells or by stromal elements, stimulates tumor cell invasion through a urokinase-dependent mechanism (11). These results support other preclinical findings that the CSF-1-signaling pathway may be involved in local invasion and metastasis, rendering cFMS a putative anti-cancer target.
In this study, we present the structures of two chimeric kinase domains of non-phosphorylated cFMS in complex with two inhibitors. The kinase domain is arranged in the bi-lobal kinase fold typical for protein kinases. Both compounds are competitive inhibitors and occupy the ATP pocket of cFMS. They form a typical kinase binding motif, in which one or more hydrogen bonds to backbone atoms of the hinge region of cFMS are formed. The compounds also prevent cFMS from switching into an active state by preventing the conserved kinase Asp-Phe-Gly (DFG) motif from reaching a productive conformation. To facilitate crystallization, we performed extensive studies on designing the most suitable construct for crystallization, which is outlined in detail in the accompanying article (Schalk-Hihi et al., 33). Additionally, we investigated the structural basis for the partial autoinhibition of cFMS via its juxtamembrane (JM) domain and compared it to the autoinhibitory mechanisms employed by the homologous kinases cKIT (12) and FLT3 (13).

EXPERIMENTAL PROCEDURES
Cloning/Protein Purification-The construct design and purification of cFMS have been described in more detail elsewhere (Schalk-Hihi et al., 33). Briefly, all constructs used in the crystallization of cFMS encompass the JM domain and the kinase domain, beginning at amino acid 538 of cFMS and ending at amino acid 922. The chimeras were created by replacing the cFMS KID with the KID sequences from the FGF receptor and TIE2 receptor. For expression in Sf9 cells, a recombinant baculovirus was generated by subcloning the cFMS chimeras into a modified Invitrogen GATEWAY pDEST8 vector. Further steps followed the protocol for baculovirus expression according to the Bac-to-Bac (Invitrogen) manual. cFMS was purified by affinity chromatography on BD Talon resin, TEV protease cleavage to remove the histidine tag, and gel filtration chromatography. The apo form was concentrated to ϳ10 mg/ml and subjected to crystallization trials. For crystallization trials with inhibitors present, the inhibitors (10 mM in Me 2 SO) were added to a diluted solution of cFMS (1 mg/ml) and co-concentrated to 7-11 mg/ml.
Crystallization and Structure Determination-In a typical crystallization experiment, 1-2 l of cFMS protein complexed with the inhibitor of interest and concentrated to 7-10 mg/ml was mixed in a 1:1 ratio with well solution (15-28% PEG3350 (Hampton Research), 100 mM sodium acetate, pH 5.0 -5.6, 200 mM Li 2 SO 4 , 5 mM dithiothreitol, 0 -3% glycerol) and placed on a glass coverslip. The coverslip was inverted and sealed over a reservoir of 500 -1000 l of well solution and incubated at 22°C. Crystals of the complex with compound 1 (arylamide) usually appeared spontaneously overnight; for other complexes, -seeding was used to induce crystallization. For data collection, the crystals were quickly transferred to cryo-solution (27% PEG 3350, 100 mM sodium acetate pH 5.5, 200 mM Li 2 SO 4 , 5 mM dithiothreitol, 15% glycerol) and frozen by immersion in liquid nitrogen. Data were collected at 100 K on a Bruker AXS MO6XCE rotating anode and a SMART 6000 CCD detector or at the IMCA-CAT ID-17 beamline at the Argonne National Laboratory. The diffraction data were processed with the Bruker Proteum suite or the HKL suite. The initial cFMS structure was solved by molecular replacement using the structure of the FGFR kinase domain (1FGK, (14)) as a search model in CNX. Density modification as implemented in CNX was used to reduce model bias. Structure refinement and model building were carried out according to standard protocols using CNX (15) and O (16). In subsequent cFMS inhibitor structures, model bias was removed through simulated annealing and the inhibitor placed in the resulting 2F o -F c map ( Table 1). The following residues have been omitted from the final structure because of insufficient density; for the cFMS-TIE2-Arylamide complex: 535-546, 557-558, 686 -692, 922; for the cFMS-FGFR-Arylamide complex: 535-547, 558 -559, 565-572, 680 -695, 814, 917-922; for the cFMS-FGFR-Quinolone complex: 535-543, 680 -695, 814, 917-922.

RESULTS AND DISCUSSION
Overview of the cFMS Structure-This structure of cFMS was obtained by utilizing a construct of cFMS, which included the whole JM domain and the catalytic kinase domain minus the C-terminal tail of cFMS. Early protein constructs that represented the kinase domain, including the full KID (residues 680 -751), did not yield any crystals. This is probably because the KID is mostly unstructured and interferes with crystallization. A design strategy in which the kinase insert domain was replaced with the significantly shorter KIDs of previously solved RTKs (FGFR, TIE2, and IRK) proved to be successful. We were able to obtain crystals from the cFMS-FGFR and the cFMS-TIE2 chimeras. The cFMS-IRK chimera did not express well and was not pursued further. Choice of the cFMS ligand also proved to be critical for crystallization, because neither apo-cFMS nor nucleotide-complexed cFMS yielded usable crystals. Co-crystallization with the arylamide series inhibitor 1 proved to be a reliable tool for obtaining diffraction quality crystals.
The structure of the cFMS kinase domain closely resembles other receptor-tyrosine kinase domain structures in the inactive form determined so far (12,13). The protein is organized in a two-lobe structure (Fig. 1A), whereby the N-lobe, comprised of five twisted ␤-sheets and a single ␣-helix, is connected to the mostly ␣-helical C-lobe by a hinge region. The N-lobe and hinge regions are mainly responsible for nucleotide or inhibitor binding and provide part of the catalytic residues, whereas the C-lobe is responsible for substrate binding and catalysis.
Nucleotide or inhibitor binding takes place in a deep cleft between the N-and C-lobe. Further important structural motifs are the glycine-rich nucleotide binding loop or P-loop (residues 590 -594), the activation loop (residues 796 -825), and the catalytic loop (residues 776 -783). The native cFMS kinase insert domain (residues 680 -751), which has been replaced by the TIE2-KID or FGFR-KID, is located between ␣-helix-D and ␣-helix-E and is mostly disordered. The juxtamembrane domain (residues 538 -581) loops from the cleft between the N-and C-lobe over the ␣-helix C to the N-lobe portion of the molecule.
Activation Loop-The activation loop is an essential element for the regulation of kinase activity. In RTKs the activation loop is ϳ22 amino acids long, begins with a conserved Asp-Phe-Gly (DFG) motif, and ends with a Pro that is conserved among tyrosine kinases (17). Autophosphorylation of tyrosines present in the activation loop has been shown to be essential for stimulation of activity for RTKs. In the absence of phosphorylation, the activation loop is not properly positioned for catalysis and prevents binding of ATP and/or substrate. Phosphorylation events in the activation loop cause a significant structural change and stabilize a conformation in which the active site is accessible to substrates, and residues important for catalysis are positioned properly.
The cFMS activation loop is present in the canonical inactive or closed conformation, examples of which can also be found in FLT3 (PDBID 1RJB) or IRK (PDBID 1IRK). A typical short antiparallel ␤-strand, observed in many inactive RTKs, is also present in the structure.
Tyr 809 is the single tyrosine in the cFMS activation loop and is one of several tyrosines that are phosphorylated in response to ligand binding. Tyr 809 is bound in the active site in a manner very similar to that of Tyr 1162 of the inactive form of IRK (18) and Tyr 842 of FLT3 (13). The phenol group of Tyr 809 forms hydrogen bonding interactions with Asp 778 where ͉F o ͉ and ͉F c ͉ are the observed and calculated structure factor amplitudes, respectively.
and Arg 782 of the catalytic loop, which stabilize the inactive conformation of the activation loop. The effect of phosphorylation on this critical residue in the activation loop has been established by several mutagenesis studies. For instance, a Y809F mutation prevents differentiation of CSF-1-dependent bone marrow macrophages into osteoclasts (19). In a rat cell line model, Y809G abolished kinase activity and Y809F reduced kinase activity by 40 -60% (20). In a previous study, the same Y809F mutation was shown to retain activity as a tyrosine kinase in vitro and in vivo, was able to undergo CSF-1-dependent association with a phosphatidylinositol 3-kinase, and induced expression of the proto-oncogenes c-Fos and JunB, underscoring its ability to trigger some of the known cellular responses to CSF-1 (21). On the other hand, the mutated receptor failed to induce mitogenesis.
The DFG motif in cFMS (Asp 796 , Phe 797 , Gly 798 ) signifies the beginning of the activation loop. Asp 796 is conserved in kinases, and serves as the catalytic base in the phosphotransfer reaction. Depending on the activation state of the kinase, the DFG motif can be present in two distinct conformations. Activated kinases in a catalytically competent state display the so-called "DFG-in" conformation, in which the aspartic acid points toward the ␤-phosphate of the bound ATP and helps to coordinate a catalytically important Mg 2ϩ ion. The phenylalanine points away from the ATP and is tucked away under ␣-helix C. In the "DFGout" conformation, which is exhibited by cFMS, the aspartic acid and phenylalanine switch sides, i.e. the aspartic acid is pointing away from the location where ATP would be bound. This conformational change renders the kinase inactive.
Juxtamembrane Domain-JM domains in RTKs constitute important regulatory elements. The cFMS JM domain corresponds to residues 538 -581 3 and contains two tyrosines, which upon phosphorylation mediate association of cFMS with partner proteins. Tyr 546 was shown to be a major autophosphoryl-ation site and binds to a yet unidentified 55-kDa phosphoprotein (22). A phosphopeptide modeled on the sequence of Tyr 561 and surrounding residues competed with the association of Fyn with cFMS (23). Furthermore, mutational analysis demonstrated that this and other sequences were required for the efficient association of Src family kinases with activated cFMS in vivo.
In addition to their role as substrates for phosphorylation, the JM domains also serve as autoinhibitory elements for RTKs. Mutations in the JM domain of cKIT and FLT3, for instance, are associated with loss of autoinhibitory function and lead to tumorgenesis. Details of this function were revealed by the recent structures of cKIT and FLT3 (12,13), which belong to the same family of type III receptor-tyrosine kinases. In both structures, aromatic residues of the JM domain protrude into the interface between the N-and C-lobe of the kinase and prevent the DFG motif and the activation loop from achieving a catalytically active conformation. This mode of inhibition differs from the one observed in the case of EphB2 (24), where the JM domain forms an ␣-helix, which distorts the conformation of the catalytically important ␣-helix C, and prevents activation of the enzyme.
Structurally, the cFMS JM domain adopts a very similar arrangement as the JM domains observed in the autoinhibited FLT3 and cKIT kinase structures (Fig. 2). It forms a hairpin-like loop, interrupted in part by disordered residues, and is tightly connected to the interface between the N-and C-lobe. In all complex structures of cFMS with various inhibitors, twelve residues N-terminal to Gln 547 do not show any electron density and are disordered. In some of the structures presented in this report, intermediate parts of the JM domain are also disordered; this is in contrast to the FLT3 and cKIT structures in which the corresponding residues could be traced without interruptions.
It is unlikely that the individual lengths of the JM domains are responsible for the observed difference in structural order between the JM domains, because all of them share a similar length. The cFMS JM domain is 44 amino acids long, whereas the FLT3 JM domain (residues 564 -609) and cKIT JM domain (residues 547-588) are 46 and 42 amino acids long, respectively. More likely in the case of cFMS, the difference in order can be explained by slight variations of crystal packing from structure to structure. On the other hand, we cannot exclude the possibility that the variations of structural order are of biological significance. For instance, one could speculate that the more ordered JM domains bind tighter to the respective kinases and therefore have higher autoinhibitory ability.
Residues 548 -552 are wedged between the catalytically important ␣-helix C and a ␤-sheet like loop region (residues 772-776) just preceding the catalytic loop. Trp 550 serves as the main anchor for the JM domain, and its indole sidechain is inserted deep into a cleft under ␣-helix C. The residue is located in a hydrophobic pocket formed by (Ile 636 , Met 637 , Leu 640 , Ile 646 , Leu 769 , Cys 774 , and Ile 794 ). The indole ring of Trp 550 forms a face-to-edge interaction with His 776 . The backbone participates in hydrogen bonding interactions with the sidechains of Arg 777 and Asp 796 . Asp 796 is part of the DFG motif and signifies the start of the activation loop. The DFG motif is present in the catalytically inactive DFG-out conformation, in which the aspartate points away from the active site and is unable to coordinate the catalytically important Mg ϩ2 ion. The backbone interaction of the carbonyl group of Asp 796 with Trp 550 stabilizes the DFG-out conformation of the activation loop, and prevents cFMS from obtaining a catalytically competent conformation.
Superposition of cFMS onto the structure of the auto-inhibited form of cKIT (PDB-ID 1T45) shows that these structural motifs are closely shared between these two proteins, and to a lesser extent with the homologous FLT3 (Fig. 2). By comparing the autoinhibited form of cKIT with the activated form of cKIT (12), the authors were able to show that the equivalent residue to Trp 550 in cKIT (Trp 557 ) prevents the phenylalanine of the DFG motif to switch into a catalytically active conformation by occluding the necessary space. A similar structural comparison for c-FMS is not possible, since no structure of cFMS in the activated form has been reported. On the other hand, based on the close structural similarities between cKIT and cFMS, one can reasonably assume that Trp 550 plays the same role as its counterpart in cKIT, and is therefore a crucial element in the autoinhibitory mechanism of cFMS by ultimately preventing cFMS from switching into the active conformation (Fig. 3).
Kinase Insert Domain-The kinase insert domain is an additional loop region found in a subset of RTKs, which is located between ␣-helix D and ␣-helix E. It can vary in length from a dozen to almost 80 residues. The cFMS KID (residues 680 -751) contains several phosphorylation sites (Tyr 699 , Tyr 708 , and Tyr 723 ), which have been implicated in mediating protein-protein interactions and downstream signaling upon phosphorylation. The analysis of these results is complicated by the fact that the outcome of similar experiments is dependent on the context of the cell lines used. Deletion of the entire KID completely abrogated signal transduction by murine cFMS expressed in Rat-2 fibroblasts (25,26). On the other hand, expression of a triple mutant (Y697F, Y706F, Y721F) of murine cFMS in FDC-P1 cells slightly increased proliferation (27). Specific interactions have been mapped to individual tyrosines. For instance, Grb2 associates with Tyr 699 upon receptor stimulation with CSF-1 in Rat-2 fibroblasts, whereas Tyr 721 was found to be a binding site for the p85 subunit of PI3-kinase (26). Another study in FDC-P1 cells identified Tyr 706 as a binding site for the transcription factor STAT1 (28). Mutation of either Tyr 697 or Tyr 721 (Tyr 699 and Tyr 723 in human cFMS) compromised signal transduction by cFMS, and the receptor lost all ability to induce changes in morphology or to increase cell growth rate in response to CSF-1.
With ϳ73 residues, cFMS possesses one of the larger KIDs in the RTK superfamily. Early cFMS constructs comprising the kinase domain including the full-length KID did not express well and did not yield crystals. This is probably because the KID is mostly unstructured and tends to interfere with crystallization. Instead of testing out various deletion mutants, our design strategy for new constructs focused on replacement of the KID with KIDs from previously determined receptor-tyrosine kinases. We chose to replace the cFMS KID with the corresponding counterparts from the FGF receptor, TIE2 receptor and IRK, because these are considerably shorter (between 13 and 20 residues) and were at least partially ordered in their respective crystal structures. Both FGFR and TIE2 chimeras yielded crystals, whereas the IRK chimera did not express well and was not further pursued.
Ligands of cFMS-In this report, we used two compounds to study the specific interactions between cFMS and inhibitory molecules. Our study focused on two series of inhibitors, arylamides and quinolones, which were obtained through optimization of initial leads identified via the proprietary Ther-moFluor assay (29). The arylamides (Compound 1, Fig. 1B) are derived from ortho-piperidin-1yl-phenylamines that have been capped with 5-membered heteroaromatic carboxylic acids via amide linkages. The quinolones (Compound 2, Fig. 1B) are bicyclic heteroaromatic systems, comprised of 6-substituted-2quinolone core scaffolds that are further substituted at the 3and 4-positions with heterocyclic and carbocyclic rings, respectively. Both compounds examined in this study are potent Almost all ATP competitive kinase inhibitors for which x-ray structures have been published mimic in part the binding of ATP. The adenine ring of ATP is involved in a tridentate hydrogen-bonding interaction with the backbone of the hinge region, which connects the N-and C-lobe of the protein. N1 of the purine ring acts as a hydrogen bond-acceptor, whereas N6 and C2 function as donors, the later as an example of an unconventional CH⅐⅐⅐OH-bond (30). This H-bonding pattern is at the heart of the interaction of ATP with kinases, and has been used extensively to design and optimize ATP competitive inhibitors.
Inhibitors of the arylamide series mimic this interaction in part (Fig. 4A). The bridging amide carbonyl accepts a hydrogen bond from the backbone nitrogen of Cys 666 . The five-membered ring together with the cyano group occupies the adenine pocket. Astacking interaction is formed with Phe 797 of the DFG motif. Other van der Waals interactions are mediated by the surrounding hydrophobic pocket formed by Val 596 , Ala 614 , Lys 616 , Val 647 , Thr 663 , Leu 785 , and Ala 800 . The para-methylpiperidine ring is located in the sugar pocket. This pocket is capped by parts of the activation loop, especially Arg 801 , which forms its bottom. The methoxy-aryl ring projects into the solvent area and interacts with some of the specificity surface residues, mainly Leu 588 and Gly 669 . A weak hydrogen bonding interaction between the methyl-hydroxyl group and the phenol-hydroxyl group of Tyr 665 can be observed as well. Although the hydroxyl methyl group can be omitted without great penalty toward affinity, it is nevertheless essential for obtaining crystals of cFMS, because the vast majority of co-crystallization attempts with similar inhibitors under development yielded no crystals (data not shown).
Inhibitors of the quinolone series also occupy the nucleotide binding pocket, with the chloro-aryl ring located in the adenine pocket (Fig. 4B). Both the amide oxygen and nitrogen of the quinolone ring form hydrogenbonding interactions with the backbone of the hinge residues Cys 666 and Glu 664 . The tautomeric state of the quinolone is presumed to be the keto form. Interestingly, the hydrophobic phenyl ring occupies the ribose binding pocket, which in this case is also capped by Arg 801 . The isoxazole ring is rotationally disordered, and a potential polar interaction of the oxygen with the backbone carbonyl of Cys 666 cannot be observed. The remainder of the interactions is mainly of hydrophobic nature and involves residues of the solvent interface and the sugar pocket (Leu 588 , Gly 589 , Leu 596 , Gly 669 , Asp 670 , Leu 785 , Phe 797 , Ala 800 , Arg 801 ).
During the binding of nucleotides to kinases, the activation loop is displaced from its canonical inactive conformation to the open conformation. This conformational rearrangement has to take place, because the activation loop is occupying and blocking parts of the nucleotide binding site. In contrast, small molecule inhibitors can accommodate various conformations of the activation loop. This has been illustrated in the case of cABL kinase with the inhibitors STI-571 and PD173955 (31,32). The inhibitor STI-571 binds to the inactive form of cABL (Fig. 5C); the activation loop is in the inactive conformation with the DFG motif in the catalytically incompetent DFG-out state. Tyr 393 , which becomes phosphorylated on activation of cABL, mimics a bound substrate and thereby prevents access to the binding site. In the PD173955 complex (Fig. 5D), the activation loop adopts the active conformation seen in many examples of activated kinases, where the activation loop is folded outwards and permits access to the binding site. The DFG motif, however, is still in the DFG-out position, which is why the conformation of the activation loop is more correctly described as being pseudoactive. The binding mode of STI-571 is very interesting, because the inhibitor not only prevents ATP from entering the active site, but it also locks the activation loop into an inactive conformation. STI-571 accomplishes this fact by two different means. First, the inhibitor clamps down onto Phe 382 of the DFG motif and forms a hydrogen-bond to the backbone of the corresponding Asp 381 , thus effectively preventing the DFG motif from shifting and the activation loop from adopting an active conformation. Second, the phenylpiperazine part of STI-571 occupies the same space that Phe 382 would if the DFG motif were to switch into the catalytically active DFG-in state. This property of the inhibitor also blocks the activation loop from switching into the active conformation.
The binding mode of our inhibitors is similar with respect to the clamping effect on the DFG motif (Fig. 5, A and B). In both series, the inhibitor shields the corresponding Phe 797 almost completely from the solvent. The accessible surface area for  A and B show cFMS complexed with the arylamide and the quinolone scaffold, respectively. C and D show cABL complexed with STI-571 and PD173955, respectively. The color-coding scheme is the same as in Fig. 1, with the protein backbone depicted as a ribbon diagram, and the DFG motif is displayed in stick representation. In the case of the arylamide series (A), the quinolone series (B), and STI-571 (C), the phenylalanine of the DFG motif is partially blocked by the inhibitors, and the activation loop is in the closed conformation. In the case of cABL in complex with PD173955 (D), the phenylalanine is exposed to the solvent, and the activation loop is folded into the open conformation.