The discovery of 2-substituted phenol quinazolines as potent RET kinase inhibitors with improved KDR selectivity

Deregulation of the receptor tyrosine kinase RET has been implicated in medullary thyroid cancer, a small percentage of lung adenocarcinomas, endocrine-resistant breast cancer and pancreatic cancer. There are several clinically approved multi-kinase inhibitors that target RET as a secondary pharmacology but additional activities, most notably inhibition of KDR, lead to dose-limiting toxicities. There is, therefore, a clinical need for more specific RET kinase inhibitors. Herein we report our efforts towards identifying a potent and selective RET inhibitor using vandetanib 1 as the starting point for structure-based drug design. Phenolic anilinoquinazolines exemplified by 6 showed improved affinities towards RET but, unsurprisingly, suffered from high metabolic clearance. Efforts to mitigate the metabolic liability of the phenol led to the discovery that a flanking substituent not only improved the hepatocyte stability, but could also impart a significant gain in selectivity. This culminated in the identification of 36; a potent RET inhibitor with much improved selectivity against KDR.


Introduction
RET (REarranged during Transfection) is a receptor tyrosine kinase (RTK) that is required for normal development, maturation and maintenance of several tissues and cell types [1]. Gain of function mutations in RET are implicated in several human cancers, e.g. medullary thyroid cancer (MTC) and lung adenocarcinoma (LAD). The identification of these mutations and rearrangements in RET which lead to constitutive activation, together with convincing preclinical data validating RET as a classical oncogene, make this kinase an attractive target for cancer therapy. At present, there are no known specific RET inhibitors in clinical development, although many potent inhibitors of RET have been opportunistically identified through selectivity profiling of compounds initially designed to target other RTKs. The small molecule inhibitors vandetanib 1 and cabozantinib 2 ( Fig. 1) exemplify this approach.
Although both have been approved for the treatment of advanced metastatic MTC [2,3] and are undergoing Phase II trials in LAD [4,5], RET inhibition is a secondary pharmacology of these drugs, which were initially developed as inhibitors of other receptor tyrosine kinases. Both agents target KDR (VEGFR2), whilst 1 has additional activity versus EGFR and 2 versus MET. Although it is possible that KDR activity may contribute to their clinical efficacy [6], the results of a large Phase III trial of 1 [2] showed a significantly better hazard ratio for RET-positive patients compared with RETnegative, suggesting that its efficacy is strongly related to its RET inhibitory activity. The EGFR activity of 1 is unlikely to significantly contribute to its efficacy since it has been demonstrated that selective inhibition of EGFR with gefitinib did not yield clinical responses in MTC [7]. Significant toxicity (e.g. rash, diarrhoea, hypertension) resulting from inhibition of these off-target kinases, particularly KDR, compromises the use of 1 and 2 in clinical settings [8]. Thus, there is a clear need for selective RET inhibitors which do not display the toxicities associated with the current treatments and enable more potent and sustained inhibition of RET signaling. These agents may offer greater clinical benefit for patients with RET mutant cancers and widen the scope for the clinical use of RET inhibitors. Although the initial clinical line of sight for this target was MTC, recent identification of RET fusions (e.g. KIF5B-RET and CCDC6-RET) present in approximately 1% of LAD patient samples offer an important disease segment in which a specific RET inhibitor would also offer clinical benefit. Additionally, RET has also recently been implicated in the progression of both breast and pancreatic tumours [9].
Our initial focus was to identify novel, low molecular weight, ATP-competitive inhibitors of the RET kinase domain with improved selectivity for RET against KDR in cellular assays, relative to 1. The availability of high quality X-ray crystal structures of RET and KDR in the public domain enabled us to pursue a structureguided medicinal chemistry approach to design and optimise novel and selective inhibitors. Hit identification employed a focused medicinal and computational chemistry programme to explore structure-activity relationships (SAR) around known RET scaffolds and to determine whether selectivity could be improved by targeting regions of the binding site that differ in sequence between RET and KDR. Although comparison of X-ray crystal structures of RET and KDR revealed that the ATP-binding pockets are generally very similar in structure, we hypothesized that there was potential to improve selectivity by targeting a number of specific residues readily accessible to the ligand. The anilinoquinazoline core of 1 was selected as a scaffold expected to impart good cellular activity and permeability, which allowed us to target binding site residues of interest using established medicinal chemistry methods.
The binding mode of 1 to the RET kinase domain has been reported previously [10]. The quinazoline core binds to the hinge region (Glu805-Ala807), with the anilino ring enclosed within the hydrophobic pocket between the gatekeeper residue Val804 and the catalytic lysine Lys758 (Fig. 2). As the gatekeeper pocket is well known to contribute to the affinity and selectivity of many series of kinase inhibitors, we decided to focus the initial optimization of the quinazoline series on exploring the SAR around the anilino ring, with a view to identifying substituents that improved selectivity towards RET. Given that the 6 and 7 positions on the quinazoline ring point primarily towards solvent, we expected these substituents to have a less pronounced role on selectivity, and hence fixed both positions as methoxy groups for the initial round of optimization.

Synthesis of anilinoquinazolines
Quinazolines 4e45 were prepared using the commercially available 6,7-bismethoxychloroquinazoline 3 as a simplified analogue of 1, as illustrated in Scheme 1.

Synthesis of key intermediates
Many of the required nucleophiles were commercially available; the remainder was synthesized using known or modified chemistry. The routes used to prepare anilines 46aei are summarized in  Schemes 2e6. Common precursors to the desired anilinophenols were the requisite methyl ethers 47a-d or nitro compounds 48aec, which could be demethylated or reduced using standard procedures as shown in Schemes 2 and 3.
Alternative precursors to the anilinophenols were the corresponding bromides. These could be converted to the boronate esters 49hei, then oxidized to the required phenols 46hei as shown in Scheme 4.
Attempts to prepare 3-amino-2-(trifluoromethyl)phenol were unsuccessful. Instead, benzyl ether 50 was prepared from commercial starting material then converted to aniline 51. This was coupled directly with 3, then deprotected to return the desired quinazoline 17 as shown in Scheme 5.
N-Methylated phenol 53 was prepared by reduction of the corresponding N-Boc material 52 (Scheme 6).

Biochemical evaluation
Our initial aim was to prepare to number of anilinoquinazolines to explore the affinity and selectivity towards RET as compared to 1 [11,12]. The biochemical data for selected compounds are shown in Table 1. First, we ascertained that switching from the more decorated quinazoline core present in 1 to the bismethoxy scaffold present in 4 showed only a 2-fold drop in affinity (data for 1 not shown). Therefore, we opted to use this simplified scaffold for further exploration, given it was commercially available. Of the first swathe of approximately 30 simple anilinoquinazolines prepared, the most interesting observation was that phenol 6 resulted in a significant gain in affinity towards RET. The same level of affinity was not maintained in the isomeric compounds 5 and 7, or for the analogous aniline 8 or the corresponding methoxy ether 9.

Structural considerations
The boost in RET affinity from the R 2 hydroxyl group may be rationalised by consideration of the hydrogen bonding contacts formed in the gatekeeper pocket. Initial modelling of 6, subsequently confirmed by the determination of the X-ray structure bound to RET (Fig. 3), highlighted a pair of hydrogen bonds from the phenol to the side-chain of Glu775 (the conserved glutamate located on the aC-helix) and the backbone NH of Asp892 (from the conserved DFG motif). It is likely that this specific network of hydrogen bonds is responsible for the gain in affinity and the tight SAR for the R 2 hydroxyl, with alternative polar R 2 substituents or hydroxyls in the R 1 or R 3 positions all demonstrating reduced affinity Comparison of the X-ray structures of 6 and 1 reveals that the ligands share very similar binding modes. The anilino ring of 6 binds between the gatekeeper Val804 and the catalytic lysine Lys758 and, in comparison to 1, shows a small rotation towards Asp892, with minor re-orientation of neighbouring side-chains consistent with the formation of hydrogen-bonding interactions between the phenol and Glu775/Asp892. The most substantial structural change is the displacement of the phosphate-binding loop away from the ATP binding site in the X-ray structure of 6; this may be a consequence of a degree of induced fit in the ligand binding site or may simply reflect the inherent flexibility of this loop. Comparison with X-ray structures of 6 bound to other kinases, e.g. to CDK2, CDPK1 and TTBK1 (PDB codes 1DI8, 3NYV and 4BTK, respectively), reveal a variety of different conformations and hydrogen bonding interactions for the phenol moiety, distinct from that observed for 6 bound to RET. Thus it should be possible to modulate the selectivity of this series by appropriate substitution around the anilino ring.

Mitigation of DMPK concerns and delivery of unanticipated selectivity
The presence of a phenol was some cause for concern given it was anticipated to undergo phase II metabolism. Measurement of in vitro DMPK properties (data not shown) of the initial compound 6 showed that solubility and CYP inhibition were acceptable at this stage, although the permeability and efflux needed improvement. Observation of metabolism in microsomes, albeit to a 3-fold lesser extent than in hepatocytes, indicated phase I metabolism was occurring in addition to phase II.
In order to mitigate phase II metabolism, we explored further substitution on the phenolic aniline. Our goals here were two-fold. First, it was speculated that the presence of flanking substituents might attenuate the propensity of the phenol to undergo conjugation, thereby increasing hepatocyte stability. Second, it would allow a more general SAR exploration to determine what functionality could be tolerated around the phenyl ring.
We first prepared a number of mono-substituted anilinophenols. Most of these bore a flanking substituent either at the R 1 or R 3 position, primarily to investigate whether the hepatocyte stability could be mitigated through steric encumbrance. However, it soon became apparent that a suitable R 1 substituent also had a considerable influence on KDR selectivity (10e13); albeit with some loss of RET affinity for 11e13. This selectivity enhancement was less evident with fluorine (14), whereas larger substituents were not tolerated, as indicated by the large drop in RET affinity (15e18). Conversely, a flanking substituent at R 3 (19e22) was tolerated with respect to RET affinity but selectivity was comparable to, or worse than, the unsubstituted phenol 6. Although we anticipated that any improvement in hepatocyte stability would be most likely achieved by exploiting substitution at the flanking R 1 or R 3 positions, limited examples of substitution at the R 4 or R 5 position were also explored. Halogens were tolerated at R 4 and R 5 (23, 25e26) but a methyl group resulted in a modest drop in affinity, especially at the R 4 position (24 and 27).
In addition to these mono-substituted examples, we also prepared a number of di-substituted anilinophenols. Appropriately positioned halogens generally retained or enhanced affinity but  selectivity was modest in the absence of a substituent larger than either H or F at the R 1 position (28e34).
Clearly, incorporation of a suitable group at the R 1 position was beneficial for selectivity, although affinity was somewhat diminished, especially when R 1 was Me. Given that halogens, especially fluorine, at R 5 had been seen to generally enhance affinity, it was encouraging to see that combining these two observations resulted in compounds which now exhibited both improved affinity and selectivity (35e37). When R 1 was fixed as chloro, substituting at R 3 with a methyl group was detrimental with regard to selectivity (compare 12 with 38). Further substitution at R 5 with chlorine (39) was tolerated whereas fluorine (40) appeared to be beneficial, both in terms of affinity and selectivity.
Modelling of this series in RET suggests that the R 1 methyl substituent is positioned close to the side-chain of Ser891 (immediately preceding the DFG motif), and this potentially disfavoured contact could account for the observed reduction in RET affinity compared with 6 ( Fig. 4). In KDR, Ser891 is replaced by a bulkier cysteine (Cys1045). Hence, although the R 1 methyl is somewhat disfavoured in RET, there is a larger drop-off in affinity against KDR, leading to an improved selectivity profile towards RET overall.

Evaluation in cellular assays
A number of these compounds (selected on the basis of affinity and/or selectivity) were progressed into BaF3 cellular assays for RET and KDR, the results of which are shown in Table 2. Disappointingly, the affinity and selectivity observed for this phenolic series in the biochemical assay did not transfer well to the cellular context. Only 4 compounds (25, 26, 28 and 30) were <100 nM against RET in the cellular assay and only one compound (40) showed >10-fold selectivity versus KDR. In contrast, 17 compounds were 10 nM in the biochemical assay and 22 compounds showed >10-fold selectivity. For reasons not fully understood, the disconnect from the biochemical to cellular assay is much greater for RET than KDR, in effect compressing the selectivity margins, often from >100-fold in the biochemical assay to parity (or worse) in the cellular assay. Permeability was not believed to be the cause for the disparity as the observed reduction in affinity was not of the same magnitude for both RET and KDR in a matched cell line. The affinity for ATP in the biochemical assay is also unlikely to explain the difference in the reduction as the K m values for both proteins (RET 9 mM and KDR 8 mM) were similar. Interestingly, the non-phenolic quinazoline 4 does not appear to suffer to the same extent.
The discrepancy between biochemical and cellular selectivity may be related to the binding mode of the R 2 phenol. Modelling of 6 in various published X-ray structures of KDR suggested that the hydrogen bonding network around the phenol in RET may be less readily achievable in KDR. This is a consequence of KDR X-ray structures displaying a conformation in which the aC-helix is displaced away from the ATP binding site in comparison to RET. The associated movement of the conserved glutamate located on the aChelix (equivalent to Glu775 in RET) would likely compromise the formation of the pair of hydrogen bonds to the phenol observed crystallographically in RET. This "aC-helix-out" motif is typically characteristic of an inactive kinase conformation. It may be that the difference between biochemical and cell selectivity profiles is related to different populations of active/inactive RET conformations under the different assay conditions. Thus, under the conditions of the biochemical assay, RET may exist predominantly in the active conformation characterized in the available X-ray structures, whereas in the cellular context there may be a larger population of inactive conformations, similar to the "aC-helix-out" conformation observed in KDR X-ray structures, to which binding of the phenol is less favoured. Regardless of the explanation for the observed differences, the decision was taken that the more physiologically relevant cellular data would drive subsequent progression of the project.

Investigation of alternate chemotypes
Different linkers were investigated as an alternative to the aniline but showed no improvement in comparison to 6. As shown in Table 3, it can be seen that the N-methyl analogue 41 is significantly less potent than the NH-, O-and S-linked analogues (6, 42 and 43). Although 42 and 43 retain biochemical affinity as compared to 6, they too suffer from a disproportionate reduction in affinity against RET compared to KDR in the cell assay.
Despite the drop in affinity of the N-methylated analogue 41, it retained some activity. Based upon this observation, the tethered compounds 44 and 45 were synthesized to test the hypothesis that these may deliver the same gain in selectivity as did the 2-  substituted analogues described earlier. However, although this was observed biochemically for 44, these biochemical activities again did not translate to cellular potency and selectivity, as shown in Table 4.

Further biological profiling
As mentioned previously, the initial interest in preparing phenolic anilines with flanking substituents was to mitigate the anticipated metabolic liability of the phenol. Table 5 shows the human hepatocyte stability data for selected compounds, ranked from most to least stable. Although compounds 10, 11 and 12 benefitted from an improved selectivity profile compared to unsubstituted 6, the R 1 groups in these examples were detrimental to hepatocyte stability. Pleasingly, 13 did show the desired improvement in stability, albeit with a 20-fold reduction in affinity. Inclusion of fluorine at either the R 4 or R 5 positions went some way toward recovering the affinity but it can be seen that 36 is preferred over 35 with regard to hepatocyte stability. Interestingly, the isomer 30, bearing the flanking substituent in the R 3 position rather than the R1 position as in 36, is significantly less stable. Unfortunately, 33, 34 and 40 (the most promising compounds with regard to affinity and selectivity in the cell assay) all suffered from high clearance.
There does not appear to be a clear correlation between stability and phenol pK a , indicating other effects (e.g. steric influence of the flanking substituent or contributions from other substituents in the case of the di-substituted examples) are contributing to the observed clearance across the series. Looking at those compounds mono-substituted at R 1 (13, 10, 11 and 12), only the Me-substituted compound 13 shows an improvement in clearance when compared to the unsubstituted phenol 6. Although it is true that, for this particular set of compounds, a lower pK a results in higher clearance, this is not observed in all cases. Isomeric pair 30 and 36 have identical calculated pK a values yet differ only by the position of the flanking Me substituent, indicating that substitution with a Me group at the R 3 position is detrimental to metabolic stability. Comparison of 30, 35 and 36 shows the inclusion of F at R 5 is beneficial, although the effect on pK a is negligible, whereas inclusion of F at R 4 resulted in a considerably lower phenol pK a and higher clearance. Compounds containing a halogen at both R 1 and R 5 displayed lower clearances than those only substituted with a halogen at R 1 , despite similar pK a values (compare 33, 34, 40, 11 and 12), again indicating that blocking R 5 is beneficial. There does not appear to be any correlation between clearance and predicted logP (XLogP).
These compounds were also tested for non-specific cellular toxicity, and, with the possible exception of 30, all were found to be devoid of non-specific toxicity in a wild-type BaF3 cell line, the parental cell line used to prepare the RET and KDR driven cell lines used in our routine screening assays. This pleasing result further suggests that the compounds display meaningful kinase selectivity in the cellular context and do not promiscuously inhibit off-target kinases responsible for cell proliferation and survival.
On the basis of these data, 36 was selected for further in vitro and in vivo pharmacokinetic assessment. In terms of metabolic stability, intrinsic clearance was higher in human hepatocytes than in human microsomes (CL int 6.2 mL/min/mg), indicative of phase II metabolism. Metabolism was more rapid in mouse in both microsomes and hepatocytes (CL int 28.2 mL/min/mg and 38.1 mL/min/ 10 6 cells, respectively). In terms of physical properties, 36 showed good aqueous solubility (in excess of 100 mM) but only moderate permeability in Caco-2 cells (P app 8.2 Â 10 À6 cm s À1 , efflux ratio 4.9). Pharmacokinetics were measured in the mouse via intravenous and oral routes of administration. Total blood clearance was low (<10% LBF) and bioavailability was approximately 35%. Oral half-life was measured at approximately 2 h.

Conclusion
A structure-based drug design programme led to a series of phenolic anilinoquinazolines showing high affinity for RET in the biochemical context. Concern over the metabolic liability of phenol 6 prompted exploration of flanking substituents to attenuate the propensity of the phenol to undergo phase II metabolism. Pleasingly, incorporation of Me at R 1 not only resulted in improved metabolic stability but also in an unexpected gain in selectivity over KDR, which could be rationalised by modelling. The improved selectivity was accompanied by some reduction in affinity but this could be recovered to some extent by inclusion of fluorine at the R 5 position, resulting in 36; a potent and selective RET inhibitor. However, for reasons not fully understood, the translation of biochemical potency to cellular potency was disproportionate when comparing RET and KDR, in effect compressing the apparent selectivity observed in the biochemical assay. Further efforts to improve both the cellular affinity and selectivity and the ADME properties of 36 are underway in our laboratory.

Chemistry
All reagents obtained from commercial sources were used without further purification. Anhydrous solvents were obtained from the Sigma-Aldrich Chemical Co. Ltd. or Fisher Chemicals Ltd. and used without further drying. Solutions containing products were either passed through a hydrophobic frit or dried over anhydrous MgSO 4 or Na 2 SO 4 , and filtered prior to evaporation of the solvent under reduced pressure. Thin layer chromatography (TLC) was conducted with 5 cm Â 10 cm plates coated with Merck type 60 F 254 silica gel to a thickness of 0.25 mm. Chromatography was performed on Biotage SNAP HP-Sil cartridges using a Combi-Flash Companion machine. Proton ( 1 H) NMR spectra were recorded on a 300 MHz Bruker spectrometer at ambient temperature. Solutions were typically prepared in either deuterochloroform (CDCl 3 ) or deuterated dimethylsulfoxide (DMSO-d 6 ) with chemical shifts referenced to deuterated solvent as an internal standard. 1 H NMR data are reported indicating the chemical shift (d), the multiplicity (s, singlet; d, doublet; t, triplet; q, quartet; m, multiplet; br, broad; dd, doublet of doublets, etc.), the coupling constant (J) in Hz and the integration (e.g. 1H). Deuterated solvents were obtained from the Sigma-Aldrich Chemical Co., Goss or Fluorochem. LCÀMS spectra with UV detection were recorded on a Waters Acquity UPLC. Mass spectrometry was performed on a Waters Acquity SQD quadrupole spectrometer running in dual ESþ and ESÀ mode. High pH runs were conducted at pH 10 and low pH runs were conducted at pH 4, with a run time of 2 min. The column temperature was 40 C, and the flow rate was 0.6 mL/min. Further details, including solvent gradients, are given in the Supporting Information. Details of the preparative HPLC instrument and the solvent gradient used to purify compounds are also given in the Supporting Information. All compounds were !95% purity as determined by examination of both the LCeMS and 1 H NMR spectra unless otherwise indicated. When Cl or Br was present, expected isotopic distribution patterns were observed.

General procedure for synthesis of quinazolines
4-Chloro-6,7-dimethoxyquinazoline 3 and the required nucleophile were heated in solvent either thermally or using microwave heating until no further reaction was observed. On cooling, the hydrochloride salt was isolated by filtration. Alternative isolation procedures were employed if precipitation did not occur. Additional purification by preparative HPLC or flash column chromatography was employed in some cases. Spectroscopic data for compounds 4 [13], 6e9 [14e16], 20e21 [13], 25 [13], 28 [17] and 30 [18] are in agreement with those reported in the literature.  a Most of these data are expressed as the geometric mean of at least four independent determinations and standard deviations are quoted in parentheses.  a Biological values are expressed as the geometric mean of four independent determinations and standard deviations are quoted in parentheses. b Most of these data are expressed as the geometric mean of at least four independent determinations and standard deviations are quoted in parentheses.

3-Amino-2,4-difluorophenol (46c).
A mixture of 2,6difluoro-3-methoxyaniline (1.0 g, 6.28 mmol) and pyridinium chloride (2.1 g, 17.79 mmol) was heated at 200 C for 1 h then allowed to cool to room temperature. The reaction mixture was diluted with water (25 mL) and then neutralized with Na 2 HCO 3 . The resultant precipitate was isolated by filtration, washed with water and dried overnight at 40 C under vacuum to return 46c

Biochemical assay
Kinase activity was detected using CisBio HTRF kinEASE kit based on time-resolved fluorescence transfer (FRET). The assay was performed in 384-well white plates (Corning #3574) in a reaction volume of 10 mL containing 1Â CisBio enzymatic buffer supplemented with a final concentration of 5 mM MgCl 2 , 1 mM DTT, 10 nM SEB and 0.01% Triton X100 for RET. The same buffer was used for the KDR biochemical assay with the addition of 2 mM MnCl 2 .
The reaction was initiated by the addition of 5 mL ATP and substrate at 2Â final reaction concentrations. For RET, this was 18 mM and 2 mM; for KDR, this was 16 mM and 1 mM, respectively. Reactions were performed at ATP K m for each target. The assay was allowed to proceed at room temperature for 20 min before terminating with the addition of 10 mL HTRF detection buffer containing EDTA supplemented with TK-antibody labelled with Eu 3þ -Cryptate (1:100 dilution) and streptavidin-XL665 (128 nM). Following incubation at room temperature for 1 h, FRET signal was measured using the Pherastar FS Microplate Reader.

BaF3 cellular assay
The system originally developed by Daley and Baltimore [26] was used, whereby IL3-dependent BaF3 cells are modified to express an activated recombinant kinase. Following removal of IL3, the modified cells are dependent on the activity of the recombinant kinase for survival and proliferation. The BaF3 cell lines, expressing KIF5B-RET (gift from Pasi Janne [27]) and KDR (Advanced Cellular Dynamics, San Diego) were maintained in RPMI-1640 media containing 10% FBS and appropriate antibiotics. Non-modified BaF3 cells (WT) were maintained in RPMI-1640 media containing 10% FBS and supplemented with 10 ng/mL recombinant mouse IL3 (R&D systems). For assessment of compound IC 50 , cells were plated into 384-well plates at 1500 or 3000 cells per well in 30 mL culture medium and compounds dispensed using an acoustic liquid handling platform (LABCYTE). Following incubation of the cells for 48 h at 37 C in a humidified 5% CO 2 atmosphere, viability was determined by addition of 10 mL CellTiter-Glo reagent (Promega) and measurement of luminescence.

Pharmacokinetics
All studies were conducted after review by the Animal Welfare and Ethical Review Body at CRUK:MI and in accordance with the University of Manchester Policy on the use of animals in research.
All work was carried out in compliance with the Animals (Scientific Procedures) Act 1986. Pharmacokinetics were studied in male CD-1 mice following single intravenous or oral administration. Blood samples were collected as dried blood spots and assayed following solvent extraction through a phospholipid removal plate followed by LC-MS/MS analysis. The resulting concentration-time data were analysed by non-compartmental methods (PK Solver, Excel Add-In).