AT9283

Fragment-Based Discovery of the Pyrazol-4-yl Urea (AT9283), a Multitargeted Kinase Inhibitor with Potent Aurora Kinase Activity†

Here, we describe the identification of a clinical candidate via structure-based optimization of a ligand efficient pyrazole-benzimidazole fragment. Aurora kinases play a key role in the regulation of mitosis and in recent years have become attractive targets for the treatment of cancer. X-ray crystallographic structures were generated using a novel soakable form of Aurora A and were used to drive the optimization toward potent (IC50 ≈ 3 nM) dual Aurora A/Aurora B inhibitors. These compounds inhibited growth and survival of HCT116 cells and produced the polyploid cellular phenotype typically associated with Aurora B kinase inhibition. Optimization of cellular activity and physicochemical properties ultimately led to the identification of compound 16 (AT9283). In addition to Aurora A and Aurora B, compound 16 was also found to inhibit a number of other kinases including JAK2 and Abl (T315I). This compound demonstrated in vivo efficacy in mouse xenograft models and is currently under evaluation in phase I clinical trials.

Introduction

Fragment-based drug discovery is a rapidly growing technique in medicinal chemistry that differs from more traditional approaches based on chemical leads derived from high through- put screening. The field has been reviewed extensively over recent years.1-6 Fragments are low molecular weight compounds (typically 100-250 Da) with generally low binding affinities that, as a result, typically require very sensitive biophysical screening methods such as X-ray crystallography,7,8 nuclear magnetic resonance spectroscopy (NMR),9 and surface plasmon resonance (SPR).10 An advantage of starting with fragments is that despite their often low affinity, fragments generally possess good ligand efficiency (LEa)11,12and as such form a small number of very high quality binding interactions with their protein target. LE is the ratio of free binding affinity to molecular size. It is possible to optimize fragments to relatively low molecular weight leads, and this can be achieved with a limited number of molecules, particularly if good structural data are available. Fragment-based drug discovery has been used in several laboratories to generate leads that have been progressed into clinical trials.6 From our own laboratories fragment screening and subsequent structure-based drug design have led to three compounds currently in clinical trials in cancer patients, including a CDK inhibitor,13 an HSP90 inhibitor, and the Aurora kinase inhibitor 16 described herein.

The Aurora enzymes are serine/threonine kinases that are known to be important regulators of mitosis.14 In recent years, two isoforms, Aurora A and Aurora B, have been of consider- able interest as targets in the discovery of new anticancer drugs.15 This interest stems from the fact that both isoforms are found overexpressed in solid tumors and leukemias.16-19 Moreover, experimental data suggest that inappropriately high levels of Aurora kinase activity are linked to genetic instability and cancer.20

Small molecule leads for Aurora kinases have been identified from both high throughput screening and, more recently, from fragment-based approaches.21-23 In the latter case, Erlanson et al. employed a fragment-based DCC (dynamic combinatorial chemistry) strategy while Warner and Hurley used computa- tional docking in combination with a structural model of Aurora A in order to identify fragments. The effect of small molecule Aurora kinase inhibitors on cells has been well described in the literature.24-28 Typically, dual Aurora A/Aurora B inhibitors result in cellular phenotypes predominantly related to Aurora B inhibition. This is characterized by rapid inhibition of serine- 10 phosphorylation on histone H3 and aberrant mitosis leading to failed cytokinesis and endoreduplication. Cells adopt a polyploid morphology and ultimately undergo apoptosis. To date, several potent inhibitors of Aurora kinases such as 1 (VX- 680/MK-0457) and 2 (PHA739358) have been progressed into clinical trials (Figure 1).26,29-31 Interestingly, in addition to inhibition of both Aurora A and Aurora B isoforms, these compounds also inhibit a number of other kinases including Flt-3 and Abl (compound 1) and FGFR1, RET, and Abl (compound 2), all of which have been suggested as targets for cancer drug discovery. More recently, isoform selective com- pounds have been described including 3 (AZD1152, Aurora B selective) and 4 (MLN8054, Aurora A selective).32-34 However, it remains to be seen precisely what type of molecular profile will be most desirable in a safe and effective Aurora kinase inhibitor.

In this paper we describe the discovery of a multitargeted kinase inhibitor with potent Aurora kinase activity. Compounds were demonstrated to have potent cellular activity and to induce phenotypical changes that were consistent with Aurora kinase inhibition being the predominant effect.

Results and Discussion

Hit Generation. Pyrazole-benzimidazole fragment 5 was previously identified in our laboratories as a starting point for the development of CDK inhibitors (Scheme 1).13 Subsequently a structure-based approach using CDK2 crystallography resulted in the identification of the benzamide analogue 7. During the course of this work it was established that these pyrazole- benzimidazoles also had good activity and ligand efficiency for Aurora A. Moreover, fragment 5 demonstrated superior ligand efficiency (LE ) 0.59) for Aurora A compared to CDK1 and CDK2 (Table 1). Unlike some fragment hits, 5 demonstrated sufficient potency to allow detection in a conventional enzyme bioassay.

Compounds 5 and 7 were selected for further characterization with a view to optimization of their Aurora kinase activity. X-ray crystallographic structures for each compound were solved using a soakable crystal form of activated Aurora A.35 In the case of fragment 5 the X-ray crystal structure showed the ligand sitting deeply in the ATP-binding site of Aurora A (Figure 2a). The ligand hydrogen-bonds to backbone carbonyl of Glu211 and also to the backbone NH and carbonyl of Ala213 of the protein hinge region. In Aurora A, the benzimidazole motif binds in the cleft defined by Ala213, Pro214, Leu215, and Gly216. This cleft has significant differences compared to the comparable region in CDK2 (defined by Leu83-Gln85). In particular, this region in Aurora A is largely defined by an extra glycine residue (Gly216) that is absent in CDK2. Consequently, the H-bond that exists in Aurora A between the carbonyl of Ala213 and of the DFG motif that lies within the activation loop of serine and threonine kinases. Comparison of the Aurora A and CDK2 crystal structures revealed that the DFG motif adopts different conformations in the two proteins. In contrast to CDK2, the conformation of the DFG loop seen in Aurora A structures prohibits interaction between the ligand 7 and the corresponding residue Asp274.

Hit-to-Lead Phase. In contrast to the pyrazole-based CDK inhibitors described in our previous publication,13 the aim of this work was to identify compounds that intervene in the cell- cycle process primarily via inhibition of Aurora kinases A and B. Although a measure of potency and selectivity for these enzyme targets was available from in vitro recombinant enzyme assays, a functional indication of selective Aurora B inhibition in cells was provided by the well described polyploid phenotype described in the Introduction. The optimization of the pyrazole- benzimidazole series focused on maximizing the in vitro potency for both Aurora A and Aurora B isoforms in addition to routinely screening for a polyploid phenotype in HCT116 cells.36

As a starting point for optimization, it was considered that the modest cellular potency of 7 (polyploidy observed at 3 µM) could be improved by introducing a basic motif. This strategy has been successful for other classes of kinase inhibitors.13,37 An examination of the binding mode of 7 suggested that the 5- or 6-position of the benzimidazole could be further functionalized without causing any clashes with the protein. Given that the series already possessed three hydrogen bond donors, a weakly basic morpholine group (pKa ≈ 7) was chosen in order minimize the increase in effective hydrogen bond donors. This initial strategy, as exemplified by 8 and 9, was successful in increasing cellular potency (Table 1). The increase in cellular potency (comparing 8 with 7) was most likely due to the associated 10-fold increase in affinity for Aurora B. Of these two compounds, the p-fluorophenyl analogue 9 was found to have encouraging pharmacokinetic properties in mouse (Clp ) 43 (mL/min)/kg, Fpo ) 26%), although plasma protein binding was found to be high (PPB ) 99.5%). Subsequent optimization was therefore aimed at identifying compounds with reduced plasma protein binding and that maintained a similar in vitro profile to amides 8 and 9.

Lead Optimization. A successful strategy to reduce plasma protein binding was conceived on the basis of making broad changes to the molecular pharmacophore. In support of this, the X-ray crystal structure of 9 complexed with Aurora A was solved in order to guide compound design (Figure 3a). Analysis of the structure revealed that the pyrazole-benzimidazole motif has excellent complementarity with the narrow region of the ATP pocket. This supported the decision to retain this motif in future designs as a key driver for potency and selectivity. Likewise, it was considered important to retain the 5-morpholi- nomethyl motif, as it had been shown to increase affinity for Aurora B. In contrast to the pyrazole-benzimidazole motif, the p-fluorophenyl group only partially fills the surrounding region of the ATP pocket and leaves a significant amount of space to accommodate molecules with modification in this region. Furthermore, the glycine loop is known to be flexible and therefore may tolerate significant modifications to the benzamide motif. Replacing the amide linker with a urea motif offered an opportunity to significantly modify the molecular shape without adding excessive molecular weight. This strategy was successful in significantly reducing plasma protein binding while maintain- ing a comparable in vitro profile as exemplified by urea 10 (mouse PPB ) 95.3%) (Table 2). Analysis of the crystal structure of 10 complexed with Aurora A showed a similar binding mode to the hinge region of Aurora A kinase as seen with previous compounds (Figure 3b). The urea linker adopted a cis/trans configuration resulting in the molecule having a folded conformation with the phenyl ring placed in proximity to the benzimidazole. This binding mode necessitates the phenyl ring to adopt a partly twisted conformation relative to the plane of the pyrazole-benzimidazole scaffold.

Further optimization was aimed at improving the affinity for both Aurora kinases and increasing cellular potency. Considering the physicochemical properties of 10 (MW ) 417, clogP ) 3.7 and 4 HBD), care was taken to keep increases in clogP and MW to a minimum. Furthermore, an increase in the hydrogen bond donor (HBD) count was considered unacceptable in order to maintain cellular penetration.

An initial strategy to improve enzyme potency was aimed at reinforcing the twisted conformation of the phenyl ring by introducing o-fluoro substituents, e.g., 11 (Aurora A IC50 ) 2.8 nM) and 12 (Aurora A IC50 ) 1.5 nM) (Table 2). This correlated with improved activity in the HCT116 cellular assay (polypolidy observed as 0.1-0.3 and 0.03 µM, respectively). An alternative strategy replaced the phenyl ring with a saturated cyclohexyl group as exemplified by 13. As with the o-fluorophenyl urea 11, the cyclohexyl group would be expected to adopt a conformation with the plane of the ring being twisted relative to the pyrazole-benzimidazole scaffold. Compared to 10, compound 13 demonstrated a marginal increase in enzyme affinity for Aurora A and 10-fold increase for Aurora B. Of these compounds, the 2,6-difluorophenyl analogue 12 was chosen for further profiling and demonstrated potent inhibition of HCT116 colony formation (IC50 ) 17 nM)36 in the secondary cellular assay. Compound 12 was also found to have reduced plasma protein binding (PPB ) 83.1%).

Further work then focused on identifying compounds with an attractive in vitro profile similar to 12, but with reduced lipophilicity (log D7.4 of 12 ) 3.1). Although minimizing lipophilicity offers no guarantees in terms of avoiding poor ADMET properties, it is widely accepted that minimizing log P is broadly associated with reduced attrition rates during clinical development.38,39 Analogues containing more polar heterocycles retained respectable activity in the enzyme assay but demon- strated reduced cellular potency, as exemplified by the 3-pyridyl analogue 14 (polyploidy at 1-3 µM) and the 4-tetrahydropy- ranyl analogue 15 (polyploidy at 1 µM). An alternative strategy to reduce lipophilicity, based on cyclohexyl analogue 13, relied on reducing the size of the hydrophobic cyclohexane motif rather than introducing heteroatoms. Gratifyingly, this approach demonstrated that high enzyme and cellular potency could be can be considered to be efficient ligands that make strong interactions with the target protein. This case adds to the growing number of literature examples where the fragment-protein interactions are maintained throughout the optimization process and are ultimately retained in final compounds.6 In the case of compound 16, the good ligand efficiency of 5 (Aurora A IC50) 0.91 yM, LE ) 0.59) and the decision to retain the pyrazole- benzimidazole motif created a high probability of achieving low nanomolar potency while keeping well within the molecular weight and lipophilicity guidelines for drug-likeness.38-40

Biological Profile of 16. In a cross-screen against other kinases, compound 16 demonstrated activity against other enzymes that have been suggested as targets for anticancer therapy, including JAK2, Flt-3, and Abl (T315I) (Table 3). This can be understood structurally. It is proposed that the benzimi- dazole motif of 16 binds to a region of these kinases that shares close structural similarity with Aurora A. This was confirmed in the case of JAK2 by the X-ray crystallographic structure of this enzyme complexed with compound 16 (Figure 6). As in Aurora A, the hinge region of JAK2 contains an extra glycine residue (Gly935), relative to the CDKs, which leads to the pocket having a particular affinity for flat aromatic heterocyles such as benzimidazole. This glycine residue is common to other kinases with a strong affinity for compound 16 such as Flt-3 and Abl (T315I) (Table 4).

Analysis of Table 4 suggests that the binding of compound 16 is relatively insensitive to the nature of the gatekeeper residue. This is in contrast to other classes of compound such as the Abl kinase inhibitor 17 (imatinib, Figure 7). In Abl (T315I), a more sterically demanding isoleucine gatekeeper maintained while reducing both MW and lipophilicity as exemplified by the cyclopropyl urea 16 (log D7.4 ) 2.1, MW )
381). Although compound 16 (Aurora A IC50 ≈ 3 nM, LE ≈ 0.42) proved too potent for an accurate IC50 value to be measured in the available assay format, HCT116 cells treated with this compound (0.03 yM) demonstrated a clear polyploid phenotype. Furthermore, 16 demonstrated potent inhibition of HCT116 colony formation (IC50 ) 12 nM) in the secondary cellular assay, a clean CYP450 profile (IC50 > 10 yM for CYP3A4, 2D6, 1A2, 2C9, 2C19), acceptable mouse plasma protein binding (81.5%) and good thermodynamic solubility (2.0 mg/mL at pH7.0 and 13 mg/mL at pH5.5).

The crystal structure of compound 16 complexed with Aurora A is shown in Figure 4. As with the initial urea analogue 10, the urea linker adopts a cis/trans configuration that results in the molecule having a folded conformation. This same confor- mation was also observed in the crystal structure of 16 alone (Figure 5a) and in DMSO solution as confirmed by NMR (Figure 5b). An NOE was observed between H3b/H3b of the cyclopropyl ring and the H4 and H7 protons of the benzimi- dazole ring. These data, taken together, suggest that this folded conformation may be stabilized by a hydrophobic interaction between these two groups.

Figure 4 also serves to illustrate that the key hydrogen bonding interactions between initial fragment 5 and Aurora A are retained in the case of compound 16. This is consistent with the concept that fragments, while only having modest affinity,As a result of the complex signaling network associated with kinases, multitargeted kinase inhibitors may offer advantages in the treatment of cancer. The possibilities presented by the additional interesting activities of 16 will be the subject of future publications.

On the basis of the good balance between in vitro potency and physicochemical properties, compound 16 was chosen for further studies in vivo. Following intravenous (iv) dosing to HCT116 xenograft bearing mice, 16 demonstrated a significantly longer half-life in tumors compared with plasma (Table 5). The compound was also found to possess modest oral bioavailability in mice (Fpo ) 24%) in spite of relatively high plasma clearance. Interestingly, incubation of 16 with mouse hepatocytes in vitro (clearance ) 8.8 (yM/min)/million cells) was predictive of only modest clearance in vivo, suggesting that a mechanism other than hepatic clearance may be involved. Overall, this profile was considered suitable for development of 16 as an iv agent with the potential for oral administration.

Compound 16 was evaluated for its in vivo antitumor activity in immunocompromised BALB/c nude mice bearing early stage HCT116 human colon carcinoma xenografts (Figure 8). In this study, the hydrochloride salt of 16 was administered via the intraperitoneal route (ip) on an intermittent dosing schedule. Doses of 15 and 20 mg/kg were seen to produce a significant tumor growth inhibition of 67% (% T/C ) 33%) and 76% (% T/C ) 24%), respectively (on day 16 of the treatment). These doses were well tolerated with mean body weight being maintained above 90% relative to the starting weight. A more extensive in vivo efficacy evaluation, including PK/PD and cell cycle analysis, will be published separately.

Conclusions

The pyrazole-benzimidazole 5 was identified during the course of a CDK program as a ligand efficient fragment starting point for the discovery of Aurora kinase inhibitors. X-ray crystallographic structures were generated using a novel soak- able form of Aurora A and were used to drive the optimization toward potent (IC50 ≈ 3 nM) dual Aurora A/Aurora B inhibitors. These compounds inhibited growth and survival of HCT116 cells and produced the polyploid cellular phenotype typically associated with Aurora B kinase inhibition. Optimization of cellular activity and physicochemical properties ultimately led to the identification of 16 (AT9283). In addition to Aurora A and Aurora B, 16 was also found to inhibit a number of other kinases including JAK2, Flt3, and Abl (T315I) (IC50 ) 1-30 nM). Compound 16 demonstrated in vivo efficacy in mouse xenograft models and was selected for preclinical development. This compound has advanced into phase I clinical trials for the treatment of cancer.41