Discovery of Novel Fibroblast Growth Factor Receptor 1 Kinase Inhibitors by Structure-Based Virtual Screening

Abstract

Fibroblast Growth Factors (FGFs) play important roles in embryonic development, angiogenesis, wound healing, and cell proliferation and differentiation. In search of inhibitors of FGFR1 kinase, 2.2 million compounds were docked into the ATP binding site of the protein. A co-crystal structure, which shows two alternative conformations for the nucleotide binding loop, is reported. Docking was performed on both conformations and, ultimately, 23 diverse compounds were purchased and assayed. Following hit validation, two compounds 10 and 16, a benzylidene derivative of pseudothiohydantoin and a thienopyrimidinone derivative, were discovered that inhibit FGFR1 kinase with IC50 values of 23 and 50 µM. Initial optimization of 16 led to the more unsaturated 40, which has significantly enhanced potency, 1.9 µM. The core structures represent new structural motifs for FGFR1 kinase inhibitors. The study also illustrates complexities associated with the choice of protein structures for docking, possible use of multiple kinase structures to seek selectivity, and hit identification.

Introduction

Protein tyrosine kinases play an important role in the signaling pathways that control cell proliferation and differentiation. Enhanced protein kinase activity due to activating mutations or overexpression has been implicated in many cancers.^1–3 The Fibroblast Growth Factor Receptors (FGFR1 through FGFR4)⁴ play an important role in embryonic development, angiogenesis, wound healing and malignant transformation.⁵ In response to growth factor stimulation, these transmembrane receptors undergo ligand dependent dimerization, which activates their intracellular tyrosine kinase domains, resulting in autophosphorylation and subsequent interaction with and recruitment of downstream cellular target proteins.⁶ Inappropriate activation of FGF receptors have been implicated in several angiogenic pathologies including diabetic retinopathy, rheumatoid arthritis, atherosclerosis, and tumor neovascularization.^7,8 Aberrant FGFR kinase activity has been implicated in different cancers including breast cancer,^10–15 human pancreatic cancer,¹⁶ astrocytomas,^17,18 salivary gland adenosarcoma,¹⁹ Kaposi’s sarcoma,²⁰ ovarian cancer,²¹ and prostate cancer.^22,23 In addition, activating mutations in FGFR genes have been associated with various human skeletal disorders such as Crouzon syndrome,^24,25 achondroplasia,^26–29 and thanatophoric dysplasia.^29–31 Therefore, discovery of inhibitors of FGFR kinases has substantial potential therapeutic value.^32,33

Kinase inhibition can be achieved by competition with the substrate, with ATP, or by locking the kinase into an inactive state.^34,35 The human genome encodes at least 518 protein kinases.³⁶ All protein kinases share common sequences and structural homology in their ATP binding sites making selectivity an issue in the development of kinase inhibitors. However, the less well conserved areas of the ATP binding site can be exploited to increase selectivity, if desired. Important classes of FGFR1 inhibitors presently known include indolinones³² such as SU4984 (1) and SU5402 (2) in Figure 1, substituted pyrido[2,3-d]pyrimidines such as PD173074 (3),^33,37 and the closely related 3-aryl-1,6-napthyridine-2,7-diamines.³⁸ These compounds show varying kinase inhibitory strengths and selectivities. 1 inhibits the kinase activities of FGFR1, PDGFR, and insulin receptor (InsR), but it does not inhibit the kinase activity of EGFR.³² 2 is more selective. It inhibits the tyrosine kinase activity of FGFR1, it is a weak inhibitor of PDGFR, and it does not inhibit the activity of InsR and EGFR.³² 1 and 2 inhibit the activity of FGFR1 kinase with IC₅₀ values of 10 – 20 µM.³² However, 3 shows high selectivity for FGFR1 inhibiting its activity with nanomolar potency while inhibiting Src, InsR, EGFR, PDGFR, and several other kinases with 1000-fold or higher IC₅₀ values.³³ Several FGFR kinase inhibitors, particularly in the indolinone and 1H-quinolin-2-one classes, are currently in clinical trials. Like the highly successful indolinone sunitinib, they are multikinase inhibitors.³⁹

Figure 1.

Examples of inhibitors of FGFR1 kinase with available crystal structures; indolinones 1 and 2, the pyridopyrimidine 3, and 4.

FGFR1 kinase consists of two subdomains enclosing the ATP binding cleft.^32,33 This cavity, which is occupied by adenine of ATP or the core of the inhibitors, is lined by hydrophobic residues. When bound, 1 forms two hydrogen bonds with the backbone carbonyl oxygen of Glu562 and nitrogen of Ala564, which belong to the hinge region connecting the two lobes (Figure 2). Crystal structures of the inhibitors 1, 2, and 3 bound to FGFR1 kinase^32,33 indicate that they reside in the ATP binding site and have at least one of the two hydrogen bonds with the hinge region.

Figure 2.

The indolinone 1 bound in the ATP binding site of FGFR1 kinase. Hydrogen bonds formed with Glu562 and Ala564, which are in the hinge region, are highlighted.

The crystal structure of FGFR1 kinase bound to 1³² shows the nucleotide binding loop in a disordered conformation, whereas the loop is in an extended conformation in the crystal structures of FGFR1 kinase complexed with 2³² and 3.³³ At the outset of the present work, we determined a crystal structure of FGFR1 kinase with the 5-thiophen-2-yl derivative of nicotinic acid, 4, (Figure 1) at a resolution of 2.2 Å.⁴⁰ This small inhibitor resides in the ATP binding site and forms a hydrogen bond with the nitrogen of Ala564. Notably, the crystal structure also reveals two conformations in the same crystallographic unit, one with the nucleotide binding loop down (brown in Figure 3) and the other with the loop extended, forming part of a β-strand, and pointing upwards (green in Figure 3). An issue with structure-based inhibitor design for FGFR1 kinase is, thus, the flexibility of the protein in this region.

Figure 3.

Crystal structures of FGFR1 kinase with 4 illustrating the two conformations for the nucleotide binding loop: extended or up (blue) and down (brown).

Consequently, in view of the potential therapeutic importance and available crystallographic data, FGFR1 kinase is a compelling target for structure-based inhibitor design. However, no prospective virtual screening studies have been reported previously for discovery of FGFR kinase inhibitors. Thus, the present study was carried out (a) to seek novel leads as inhibitors of FGFR1 kinase by virtual screening, and (b) to test the viability of current docking methodology for this purpose. High potency and selectivity were not expected at this stage; rather, a foundation for subsequent lead optimization was sought. The study illustrates complexities associated with the choice of protein structures for docking, possible use of multiple kinase structures to seek selectivity, practical restrictions on numbers of compounds that can be purchased and assayed, and hit identification. In particular, we have docked a large database of molecules into the ATP binding site of FGFR1 kinase using both conformations for the nucleotide binding loop in the co-crystal structure for 4. Compound selection was also influenced by docking results for 5 additional kinases. Subsequent in vitro assaying followed by validation of active compounds and initial optimization did lead to the discovery of several compounds with new core structures that exhibit low-micromolar inhibition of FGFR1 kinase.

Methods

Virtual Screening

The ZINC database⁴¹ was docked into the two conformations from the FGFR1-4 structure⁴⁰ using Glide 4.0.^42,43 The ZINC collection contains about 2.2 million small molecules that have been filtered to be reasonable starting points for drug design and that are commercially available. The molecules are created, as appropriate, with multiple protonation and tautomeric states. The FGFR1 kinase conformations were prepared using standard Glide protocols.⁴² This includes addition of hydrogens, restrained energy-minimizations of the protein structure with the OPLS-AA force field,⁴⁴ and finally setting up the Glide grids using the Protein and Ligand Preparation Module. Crystal structures of kinases with inhibitors bound in the ATP site reveal that the inhibitors form at least one hydrogen bond with backbone amide or carbonyl groups in the hinge region. Thus, imposition of pharmacophore constraints with the hinge region has been routine in docking of molecules into the ATP sites of kinases to improve the chances of success in finding active compounds.^45–47 This practice was adopted here such that any acceptable protein-ligand complex was required to have at least one hydrogen bond with Glu562 or Ala564 (Figure 2). All 2.2 million compounds were first docked and ranked using standard precision (SP) Glide.⁴² The resultant top 40,000 compounds were then docked using the more accurate and computationally intensive extra-precision (XP) mode.⁴³ 41 compounds⁴⁸ known to exhibit activity against FGFR1 kinase including 15 indolinones such as 1 and 2, 15 pyrido[2,3-d]pyrimidines such as 3, and 10 napthyridines, were also included. The processing of the known active compounds was intended to help gauge the effectiveness of the screening strategy. The top 1000 compounds for both conformations from the XP docking were also docked into 5 additional kinase structures, EGFR,⁴⁹ InsR,⁵⁰ VEGFR2,⁵¹ Src,⁵² and MEK,⁵³ in order to see if the current scoring is accurate enough to provide some kinase selectivity. Compounds that were eventually purchased ranked in the top 100 for FGFR1, but not in the top 100 for any of the other 5 kinases.

Docking with Glide effectively performs a systematic search of the conformational, orientational, and positional space for the docked ligand in the binding site of the protein. This is followed by torsionally flexible energy optimization on an OPLS-AA⁴⁴ potential energy grid to arrive at a few final candidate poses. The lowest energy poses are further refined by Monte Carlo sampling. The scoring function used in (SP) Glide includes terms for ligand-protein interaction energies, hydrophobic interactions, hydrogen bonds, internal energies, and desolvation.⁴² The more sophisticated XP scoring function also employs terms that account for ligand-protein structural motifs that lead to enhanced binding affinity.⁴³ This includes hydrophobic enclosure, where lipophilic ligand atoms are enclosed on opposite faces by lipophilic protein atoms, and single or correlated hydrogen bonds in hydrophobic environments, which are relevant for kinase inhibitors in the ATP site. Thus, XP Glide is expected to be well suited for the study of ligand-kinase complexation.

The resultant most desirable compounds were also evaluated for conformational strain in the bound conformation. One measure of this strain is the root-mean-square deviation between a molecule’s docked structure and its structure after energy minimization in vacuum. Another measure that was applied is the difference in energy after minimization and the lowest energy conformation resulting from a 200 conformation Monte Carlo search with GB/SA hydration using the program BOSS.⁵⁴ These measures revealed that most compounds of interest were in reasonably favorable conformations in the protein binding site. Visual inspection of poses was also essential in arriving at the final selections, which were then purchased from commercial vendors.

MC/FEP Simulations

Some Monte Carlo/free-energy perturbation calculations were executed to compute relative free energies of binding in order to guide initial modifications of active compounds. The calculations were performed following standard protocols with the program MCPRO.⁵⁴ The models were built starting from the crystal structure for the complex with 4 and included the 170 amino acid residues nearest the ligands. Short conjugate-gradient minimizations were carried out on the initial structures for all complexes to relieve any unfavorable contacts. Coordinates for the unbound ligands were obtained by extraction from the complexes. The unbound ligands and complexes were solvated in 25-Å caps with ca. 2000 and 1250 water molecules. The FEP calculations utilized 11 windows of simple overlap sampling. Each window covered 10–15 million (M) configurations of equilibration and 20–30 M configurations of averaging. The ligand and side chains with any atom within ca. 10 Å of the ligand were fully flexible, while the protein backbone was kept fixed during the MC simulations. The energetics were evaluated with the OPLS-AA force field for the protein, OPLS/CM1A for the ligands, and TIP4P for water.⁴⁴

Experimental

Purchased compounds were initially assayed as received. Structures of active and new compounds were validated through ¹H and ¹³C NMR and high-resolution mass spectrometry, as detailed in the Supplementary Information. The purity of all active and new compounds was demonstrated to be >95% by high-performance liquid chromatography. Synthetic schemes for new compounds are provided below, while the synthetic details are provided in the Supplementary Information.

Biological assaying was performed using ALPHAScreen (Amplified Luminescent Proximity Homogeneous Assay) from Perkin Elmer. FGFR1 kinase, the biotinylated peptide, ATP, and the potential inhibitor were added to each of 10 wells with inhibitor concentrations typically ranging from 0.01 µM to 250 µM. Phosphorylation of the peptide was allowed to proceed for 30 minutes, and then the reaction was stopped by adding EDTA. The resultant differential phosphorylation of the peptide by FGFR1 kinase in each well depends on the effectiveness of the inhibitor. Acceptor beads coated with anti-phosphotyrosine antibodies and donor beads coated with streptavidin conjugates are then introduced into each well. Biotinylated peptide is efficiently captured by the donor bead. The FGFR1 kinase domain was obtained and purified as previously described.⁵⁵ Other kinases used in this study, EGFR, InsR, and Src, were obtained from Millipore Corp.

Purified FGFR1 kinase domain (20 mg/ml) was co-crystallized with 2-thienyl nicotinic acid, 4, (1mM) by vapor diffusion in 0.1 M MES-NaOH (pH 6.5), 14% PEG 4000, 0.2 M (NH₄)₂SO₄, and 5% glycerol at 4 C. The crystals were flash frozen in 0.1 M MES (pH 6.5), 25% PEG 4000, 0.2 M (NH₄)₂SO₄, and 10% glycerol. Diffraction data were collected on beamline X29 at the National Synchrotron Light Source. They were processed by using HKL2000, and a molecular replacement solution for the complex was found by using the structure of inactive FGFR1 (PDB ID: 1FGK).⁵⁶ The resultant structure has been deposited in the Protein Data Bank (PDB ID: 3JS2).⁴⁰

Results

Docking Known Inhibitors

The top 40,000 compounds emerging from the SP Glide calculations plus the 41 compounds possessing known activity were processed with XP Glide. For both loop conformations, 8 of the known active compounds ended up in the top 1000. There were 4 active compounds in common, and thus 12 unique active compounds were retrieved. Assuming equal distribution of the 41 compounds in 40000, there is roughly 1 in every 1000. Thus, the finding of 8 in the top 1000 reflects significant enrichment. If the top 10,000 compounds are considered, 17 and 11 of the known active compounds were identified using the protein conformation with the binding loop down and up, respectively. The performance of SP Glide alone can also be noted. When the 41 known active compounds are added to the 2.2 million library compounds, 9 and 10 are ranked in the top 10,000 using the protein conformations with the binding loop down and up, respectively, and 12 and 11 compounds are among the top 50,000. The random result would be retrieval of roughly 1 known active compound per 50,000. Thus, docking with Glide in both SP and XP modes demonstrated success in identifying known FGFR1 kinase inhibitors. Additional benefits of the XP mode over SP were not clearly apparent in this exercise. As detailed below, the performance of XP Glide for the known inhibitors was also class specific.

Superposition of the crystal structures with the poses from Glide reveals good correspondence for the positioning of 1 and 2 (Figure 4). In general, among the known active compounds, ones with the indolinone core show the correct orientation when compared to the crystal structures of the complexes for 1 and 2.³² There is also very good overlap of the positions of 4 in the crystal structure⁴⁰ and in the docked complexes for both conformations of the nucleotide binding loop. However, the docked structure for FGFR1 kinase with the nucleotide binding loop up complexed with 3 has incorrect orientation of the ligand (Figure 4). Superposition of structures of complexes with compounds having the pyrido[2,3-d]pyrimidine or napthyridine cores from the XP docking and the crystal structure of 3 reveal a 180° flip in the binding site (Figure 4). The crystal structures used for the docking were the ones for 4, as this was the most complete structure with no missing residues. However, most of the pyridopyrimidine inhibitors have large side chains, which clash sterically with the side chain of Lys514 when the inhibitor’s core is properly oriented. Thus, incorrect docking of the pyrido[2,3-d]pyrimidine class could be improved by screening against additional conformations of FGFR1 kinase. This would hopefully lead to additional compounds with very low docking scores that could be further scrutinized. Alternatively, the conformational problem argues in favor of fragment screening to discover viable core structures without the need to accommodate large substituents. Owing to these considerations, 8 of the 12 unique active compounds in the top 1000 from the XP docking with the nucleotide binding loop in either orientation belong to the indolinone class. The correct docking of these compounds presumably accompanies their good rankings. The compounds in the pyrido[2,3-d]pyrimidine and the napthyridine classes are incorrectly docked and rank poorly.

Figure 4.

Overlays of the crystal structure (grey) with the XP GLIDE pose (green) for 1, 2, 3, and 4. Hydrogen bonds with Glu562 and Ala564 are highlighted.

The distribution of XP scores for the 40,000 compounds from the ZINC library, when docked into either FGFR1 kinase conformer, covers a 16-kcal/mol range (Figure 5). Many library compounds have scores as low as those for the best-ranked known actives (in red in Figure 5). Though it is encouraging that many compounds from ZINC yield such favorable scores, many known active compounds have poor scores in view of the incorrect docking.

Figure 5.

XP Glide score frequency for 40,000 compounds from the ZINC collection in black and known active compounds in red (scaled by 300) for the conformation with (A) the nucleotide binding loop up and (B) with the nucleotide binding loop down.

Compound Selection and Activity

The 2000 library compounds, 1000 for each FGFR1 conformer, with the best XP scores were processed further. For both conformers, in order to favor compounds that could selectively inhibit FGFR1 kinase, the top 1000 compounds were docked into the ATP binding sites for the 5 other kinases, EGFR, InsR, VEGFR2, Src, and MEK. While the kinome has roughly 500 members,³⁶ this subset was chosen based on potential availability of the protein for assaying and, of course, availability of a co-crystal structure of the kinase with a ligand in the ATP site. Compounds ranked within the top 100 for FGFR1 kinase, but not in the top 100 for any of the other 5 kinases, were retained for further consideration. Their docked protein-ligand complexes received extensive visual inspection. Ligands that possessed inappropriate geometries such as twisted amides or esters were rejected.⁵⁷ The compounds were also checked for excessive conformational strain in the bound conformation, as described above.

At this point, for the set of compounds from the screening of the conformation with the binding loop up, a total of 31 desirable compounds remained. To promote structural diversity, this set was further partitioned into 14 chemical classes. Eventually, 11 compounds, which were mostly the top-ranked in their class, were purchased. These are shown in Figure 6. In one case, 7, structural ambiguity arose, and it was subsequently determined that the purchased compound was actually the isomer 10. As described below, this was demonstrated through synthesis and assaying of both 7 and 10. The docked structures and scores for 7 and 10 are essentially identical as the structural difference is in the edge that is solvent-exposed in the complexes. Thus, the structures of the two isomers, 7 and 10, and the 10 other purchased compounds are shown in Figure 6. This set has a good representation of 5- and 6-membered heterocycles including 6 with fused polycyclic ring systems. Not surprisingly, in the docked structures, the polycyclic cores overlap the position of adenine of ATP when bound. All computed structures of the complexes feature hydrogen bonds in the hinge region and extensive overlap with the positions of known inhibitors in crystal structures for complexes with FGFR1 kinase.

Figure 6.

Structures of the purchased compounds obtained by docking into the conformation of FGFR1 kinase with the nucleotide binding loop up. Though 7 was ordered, it turned out to be the isomer 10. Samples of both 7 and 10 were synthesized.

In the same manner, the top 1000 compounds from the docking calculations with the FGFR1 binding loop in the down conformation were narrowed to 37 compounds, which were assigned to 15 classes. Among these, the 12 high-ranking ones in Figure 7 were ultimately purchased. In both cases, compounds with cores not previously reported as FGFR1 kinase inhibitors were sought; indolinones, naphthyridines, and compounds with 2-aminopyrinidine fragments were avoided. A total of 23 compounds was purchased and tested in vitro for inhibition of FGFR1 kinase using the AlphaScreen assaying system. The results of the assaying along with Glide scores and ranks are shown in Table 1. Two of the compounds were found to inhibit the activity of FGFR1 kinase. Though follow-up of these hits is enough to consume our available resources, purchase and assaying of at least all 68 of the compounds deemed most desirable would likely have generated additional alternatives. It is also expected that many of the core structures for the inactive compounds among 5 – 28, in fact, provide viable platforms for discovery of active inhibitors; small changes are often all that is needed to turn an inactive compound into an active one.^58,59

Figure 7.

Structures of the purchased compounds obtained by docking into the conformation of FGFR1 kinase with the nucleotide binding loop down. Compounds with stereocenters are racemic mixtures.

Table 1.

Experimental IC₅₀ Values for Inhibition of FGFR1 Kinase, and Glide Scores and Rankings.

compd	XP rank	XP score	SP rank	IC50 (µM) a
5	1	−16.18	112	na
6	7	−14.84	11628	na
7	16	−14.29	33929	na
8	19	−14.20	22167	na
9	25	−13.98	550	na
10	38	−13.79	61627	23
11	46	−13.71	8409	na
12	51	−13.66	13883	na
13	52	−13.64	6377	na
14	74	−13.49	36663	na
15	77	−13.48	396	na
16	84	−13.45	20052	50
17	2	−18.53	429	na
18	10	−17.38	10443	na
19	29	−17.01	12782	na
20	37	−16.96	7676	na
21	41	−16.93	14137	na
22	45	−16.89	19473	na
23	55	−16.84	9711	na
24	56	−16.84	15905	na
25	64	−16.78	6525	na
26	72	−16.75	4238	na
27	93	−16.62	27217	na
28	97	−16.60	22135	na

^ana indicates not active in the assay.

7, 10, and Analogues

One of the purchased compounds from ChemBridge Corp., presumed to be 7, had an IC₅₀ value of 23 µM. For initial follow-up, 7 and several analogues, 29 – 34, were synthesized as summarized in Scheme 1. The intention was exploration of modifications to the methoxy group in 7 and a chlorine scan for the terminal phenyl ring. MC/FEP results indicated that chlorine and methyl substitution should be most favorable at the meta positions in this ring. The benzylidene derivatives were prepared from either commercially available p-hydroxybenzaldehyde or 3-alkoxy-4-hydroxybenzaldehydes. In the case of 29, first p-hydroxybenzaldehyde (41) was converted to 3-methoxymethyl-4- hydroxybenzaldehyde (42c). The aldehydes 43a–c resulted from O-alkylation of 42a–c with 1,3-dibrompropane. The O-alkylation of alkyl bromides 43a–c with phenols 44a–d then provided aldehydes 45a–g, and Knoevenagel condensation of 45a–g with thiohydantoin afforded 7 and 29–34.

Scheme 1^a.

^aReagents and conditions: (a) 1. HCHO, HCl, 50 °C, 3h; 2. MeOH, ref lux, 3h, 57% two steps (b) 1,3-dibromopropane, K₂CO₃, DMF, 80 °C, 1h (43–60%) (c) K₂CO₃, DMF, 80 °C, 1h (44–91%) (d) thiohydantoin, pipiridine, MeOH, ref lux (54–74%) (e) pseudothiohydantoin, AcONa, AcOH, 170 °C, 15 min, MW, 53%

The compounds were assayed (Table 2), and it was surprising to find that 7 was now inactive and that only 29 and 30 showed weak activity. A small molecule crystal structure was then obtained for 31, which confirmed the structures in this series to be as shown in Table 2, notably with the thiohydantoin substructure and Z-olefin geometry. Thus, the purchased compound displaying the 23-µM activity was not 7. Scrutiny of the NMR spectra for the purchased compound and the synthesized 7 led to the expectation that the purchased compound might be the pseudothiohydantoin isomer 10. This compound was then prepared by Knoevenagel condensation of aldehyde 45a with pseudothiohydantoin (Scheme 1). The AlphaScreen assay demonstrated that 10 was the active compound (Table 1), and the spectral data confirmed that 10 and the purchased compound were the same. Compound 10 was subsequently docked into FGFR1 kinase. The predicted pose is almost identical to that for 7, and it was found to rank well (Table 1). 10 also showed poor ranking against the other 5 kinases, thus meeting the computational selectivity criterion. In summary, extensive validation was needed to confirm that 10 rather than 7 was the active compound found in the screening.

Table 2.

Experimental IC₅₀ Values for Inhibition of FGFR1 Kinase by Analogues of 7.

compd	R1	R2	R3	R4	R5	IC50 (µM)
7	OMe		Me		Me	na
29	OEt		Me		Me	250
30	CH2OMe		Me		Me	3300
31	OMe	Cl				na
32	OMe		Cl			na
33	OMe			Cl		na
34	OMe		Me			na

16 and Analogues

Thienopyrimidinone 16 is the other purchased compound that was found to inhibit the activity of FGFR1 kinase. However, the IC₅₀ value of 50 µM was modest, so some initial lead optimization was pursued to see if a simple analogue in the low-µM range could be obtained. To examine the role of the carboxylic acid side chain of 16, which was predicted to be largely solvent exposed in the complex with FGFR1 kinase, compounds 35 and 36 were synthesized, as shown in Scheme 2. Relatives with the tricyclic core fully unsaturated such as 37 – 40, were also pursued. MC/FEP calculations were performed and indicated that replacement of H by methyl at R₁ and R₂ should improve the free energy of binding by 1–2 kcal/mol. However, introduction of a methyl group at the open positions adjacent to R₁ and R₂ was not pursued because it was computed to be unfavorable owing to steric clashes with the side chain of Lys514 or the backbone of Glu562, respectively. Finally, given the experience with 7 and 10, 16 too was synthesized.

Scheme 2^a.

^aReagents and conditions: (a) S, piperidine, EtOH, 50 °C, 4h (23–60%) (b) p-Chloranil, 1,4-Dioxane, 90 °C, 5h (c) 3,4-dimethoxybenzaldehyde or 3-methoxybenzaldehyde, conc HCl, n-BuOH, 80 °C, 8H (48–67%) (d) ethyl 2-(4-formyl-2-methoxyphenoxy)acetate (52), conc HCl, n-BuOH, ref lux, 2h (30–62%) (e) 5N NaOH, EtOH, reflux, 2h (30%-Quant.)

Preparation of thiouracil derivatives 16 and 35–40 commenced with 3- or 4-substituted cyclohexanones. Gewald reaction of cyclohexanones 46a–e with 2-cyanoacetamide provided 2-aminothiophenes 48a–e. Condensation of 48a with aldehyde 52 provided ester 49, which was hydrolyzed to afford 16. The spectra for the purchased 16 and synthesized 16 were the same. The reaction of 2-aminothiophene 48a with 3,4-dimethoxybenzaldehyde and 3-methoxybenzaldehyde resulted in the formation of 35 and 36. For the synthesis of benzothiophene derivatives 37 to 40, first, the 2-aminothiophenes 48a–e were converted to 2-aminobenzothiophenes 50a–e. Acid-catalyzed condensation of 50a–e with aldehyde 52 resulted in esters 51a–e, which were hydrolyzed to yield 37–40.

As summarized in Table 3, the decarboxylated 35 and 36 do not inhibit FGFR1 kinase. However, the unsaturated analogues, 37 and 40, provided significant advance (Table 4). Conversion of the cyclohexyl ring to phenyl in going from 16 to 37 lowered the IC₅₀ from 50 to 4 µM. This outcome was not obvious owing to the tradeoff between greater hydrophobicity with 16 and greater rigidity with 37. Replacing the methyl substituent by ethyl in proceeding from 37 to 38 yielded little change in activity, while introduction of trifluoromethyl at R₁ in 39 eliminated activity. Consistent with the MC/FEP results, a methyl group was found to have similar effect at R₁ and R₂ with 40 being the most active analogue with an IC₅₀ of 1.9 µM. Overall, the unsaturated analogues 37, 38, and 40 provide a novel core structure for FGFR1 kinase inhibition and a much improved starting point for full lead optimization.

Table 3.

Inhibitory Activities of FGFR1 Kinase by Analogues of 16.

compd	R3	IC50 (µM)
16	OCH2CO2H	50
35	OMe	na
36	H	na

Table 4.

Inhibitory Activities of FGFR1 Kinase by Tetradehydro Analogues of 16.

compd	R1	R2	IC50 (µM)
37	Me	H	4.0
38	Et	H	5.5
39	CF3	H	na
40	H	Me	1.9

Computed Properties, Structures, and Selectivity for 10 and 40

Thus, the screening and subsequent synthetic efforts delivered two principal compounds, which are suitable for further lead optimization, the pseudothiohydantoin derivative 10 with an IC₅₀ of 23 µM and the benzothienopyrimidinone derivative 40 with an IC₅₀ of 1.9 µM. To our knowledge, no compounds with these core structures have been demonstrated previously to be FGFR1 kinase inhibitors. Both compounds evolved from the docking calculations that used the conformation of FGFR1 kinase with the nucleotide binding loop in the extended conformation (Figure 3). The chosen compounds that came from the docking calculations with the nucleotide binding loop down (Figure 7) were all inactive. This should not be over-interpreted in view of the modest success in finding only one true active compound, 16, among the 23 compounds, which were purchased. More favorable XP scores were obtained for the conformation with the binding loop down (Table 1); however, a possible reorganization penalty for achieving this conformation is not included in the calculations.

In order to provide an initial sense of expected pharmacological properties, the program QikProp ⁶⁰ was used to make the predictions in Table 5. The selected properties are expected to influence bioavailability through dissolution, cell permeation, and metabolism. When QikProp is run for a set of 1700 oral drugs, 95% are predicted to have molecular weights between 130 and 500, log P values between −2 and 6, log S values between −6.0 and 0.5, PCaco2 values greater than 25 nm/s, and 7 or fewer primary metabolites.⁶¹ The predicted properties of the two key compounds compare favorably with these ranges, though poorer solubility needs to be avoided during further lead optimization. For 10, the predicted primary metabolites arise from ether cleavages, benzylic methyl oxidations, and possible sulfur oxidation. For 40, the predicted metabolic processes are for oxidation of the sulfur atom and the three side chains leading to possible catechol formation.

Table 5.

Some Properties of the Lead Compounds Predicted using QikProp 3.0.

compd	MWa	QP logPb	QP logSc	QP PCaco2d	N Metabol.e
10	412.5	4.49	−6.1	673	6
40	396.4	3.26	−5.6	36	5

^aMolecular weight.

^bLog of the octanol/water partition coefficient.

^cLog of the aqueous solubility S (mol/L).

^dCaco2 cell permeability in nm/sec.

^eNumber of primary metabolites.

The structures from the Glide XP docking for 10 and 40 are illustrated in Figure 8. Both ligands are predicted to bind in the hinge region and both feature two hydrogen bonds with Ala564 via the amido fragments (O=C−NH) in the pseudothiohydantoin and pyrimidinone rings. There is considerable overlap with the observed positioning of the indolinones, e.g., in Figure 2.³² However, the hydrogen-bonding motif is interestingly different, since for the indolinones the order of the amido fragments (HN−C=O) is reversed, and the complementarity is with the backbone carbonyl oxygen of Glu562 and the NH of Ala564. The bound 10 also extends more to the right towards Phe489 than for the indolinones. In addition, the complex for 10 has a hydrogen bond between the methoxy group on the ligand’s central ring and the side-chain ammonium group of Lys514. The assay results for 29 in Table 2 indicate that the methoxy to ethoxy change is beneficial for binding, perhaps owing to favorable additional hydrophobic interactions in the Val492 − Lys514 area, while change to methoxymethyl (30) is less productive. Another notable motif is the sandwiching of the dimethylphenyl ring of 10 between the side chains of Phe489 and Lys514, forming presumably constructive π-π and cation-π interactions. This does impose conformational restrictions on the 1,3-dioxypropyl linker. Returning to the 7 versus 10 conundrum, the computed structures do not provide an obvious reason for the inactivity of 7 since the S=C−NH and HN=C−S edges are predicted to be solvent exposed. Presumably, there is sensitivity of the critical hydrogen bonding with Ala564 to the geometrical and electronic differences between the isomeric rings or there are subtleties in their hydration; further computational investigation is warranted.

Figure 8.

Computed structures for the complexes of FGFR1 kinase with 10 (A) and 40 (B). Selected backbone and side-chain atoms of the kinase are shown; carbon atoms of the inhibitors are colored green. Hydrogen bonds are highlighted with black lines.

For 40, additional hydrogen bonding is indicated via salt-bridge formation between the ligand’s carboxylate group and the ammonium terminus of Lys482. These groups can also be fully solvent exposed. Thus, the energetic benefit of the salt bridge is not clear; however, the results in Table 3 indicate that the carboxylate group is making a positive contribution to the activity. The beneficial methyl groups at R₁ and R₂ in 37, 38, and 40 (Table 4) are inserted into the hydrophobic region near Val492. Comparison of the computed structures for the complexes of 10 and 40 suggests that lead optimization for 40 has opportunities in expansion towards Lys514 and Phe489 and truncation at the other end of the inhibitor.

Finally, testing of compounds 10 and 40 for activity against other kinases was performed using EGFR, InsR, and Src (Table 5). Compound 10 is an inhibitor of all four kinases with IC₅₀ values of 10 – 56 µM , while 40 does not inhibit InsR, but it is a 2-µM inhibitor for FGFR1, EGFR, and Src. Thus, the limited computational selectivity filter was not effective; however, fundamentally, the diverse results in Figure 5 for known inhibitors of FGFR1 kinase need to be noted along with the fact that the 2 active compounds from the present screening had essentially the two worst scores in Table 1. Under the circumstances, the current docking and scoring methodology does not appear to be accurate enough to guide economically viable compound acquisition in the absence of substantial human post-processing.⁶² It seems even less likely that the current methodology is accurate enough to successfully address kinase selectivity, which requires reliability of the results for multiple targets. The issue is complicated by the recognition of potential clinical benefits for compounds with multi-kinase activity.³⁹ Nevertheless, there is clear value in the docking as it did provide a key component of a compound selection protocol that enabled the discovery of two new series of FGFR1 kinase inhibitors. Selectivity is typically expected to be addressed during lead optimization through combination of more computational modeling, synthesis, assaying, and crystallography.⁵⁹

Table 5.

Inhibitory Activities for Four Kinases

IC50 (µM)
compd	FGFR1	EGFR	Src	InsR
10	23	56	10	47
40	1.9	2.4	1.9	na

Conclusions

The ZINC database of 2.2 million compounds was screened using two conformations of FGFR1 kinase. On docking with XP Glide, 8 of 41 known active compounds emerged in the top 1000 of 40,000 compounds, which were the best ranked ones using SP Glide. The indolinone class of inhibitors and nicotinic acid derivative 4 were handled well, while the docked structures and scoring for the pyrido[2,3-d]pyrimidine and napthyridine classes were inaccurate. The conformation of Lys514 in the utilized structure led to steric incompatibilities with many of the known inhibitors. The virtual screening progressed to the purchase of 23 structurally diverse compounds. Two compounds were initially found to be active. However, much effort was needed to demonstrate that the structure for one was incorrectly assigned and that the isomer 10 was the true active compound; synthesis of both isomers was required. 10, a benzylidene derivative of pseudothiohydantoin, and 16, and a thienopyrimidinone derivative, were found to show inhibitory activity towards FGFR1 kinase with IC₅₀ values of 23 and 50 µM. Modifications of 16 led to the more unsaturated 40, which showed a much improved IC₅₀ value of 1.9 µM. The predicted structures for the complexes of FGFR1 kinase with 10 and 40 appear reasonable in comparison to known crystal structures and they are generally consistent with the initial structure-activity results presented here. Both compounds are expected to form two hydrogen bonds with the oxygen and amide NH of Ala564; aryl-aryl, cation-π, and salt-bridge interactions are also represented. Finally, the selectivity of 10 and 40 for FGFR1 kinase received some analysis through assaying with three additional kinases, EGFR, InsR, and Src. Little selectivity was found except that 40 shows no inhibition of InsR kinase. Though the applied computational selectivity filter could be made more restrictive, it is unlikely that the accuracy of the current docking and scoring methodology is sufficient to provide a solid basis for this purpose. Optimization of the two new series of FGFR1 kinase inhibitors for both potency and selectivity is being pursued using a combined approach featuring free-energy perturbation calculations, organic synthesis, biological assaying, and protein crystallography.

Abbreviations

ATP

adenosine triphosphate

EDTA

ethylenediaminetetraacetic acid

EGFR

epidermal growth factor receptor

FEP

free energy perturbation

FGFR

fibroblast growth factor receptor

GB/SA

generalized Born/surface area

InsR

insulin receptor

MC

Monte Carlo

MEK

mitogen-activated extracellular signal-regulated kinase

MES

2-(N-morpholino)ethanesulfonic acid

OPLS-AA

optimized potentials for liquid simulations - all-atom

PDGFR

platelet-derived growth factor receptor

SP

standard precision

VEGFR

vascular endothelial growth factor receptor

XP

extra precision