Structure of Aurora B–INCENP in complex with barasertib reveals a potential transinhibitory mechanism

Two structures of Aurora B kinase bound to the specific inhibitor barasertib and to a nonhydrolyzable nucleotide provide evidences for a potential transinhibitory mechanism.


Introduction
The Aurora proteins are a small family of serine/threonine kinases that are expressed during mitosis and have been suggested to be attractive drug targets (Carpinelli & Moll, 2008;Andrews, 2005). Mammalian cells possess three Aurora kinase isoforms (Aurora A, B and C) that exhibit different subcellular localizations and have distinct functional roles (Carmena & Earnshaw, 2003). Together with INCENP, borealin and survivin, Aurora B kinase forms the chromosome passenger complex (CPC; Carmena et al., 2012). The most conserved C-terminal region of INCENP (the so-called IN-box) is sufficient to bind and stimulate the catalytic activity of Aurora B (Honda et al., 2003;Sessa et al., 2005). The CPC complex is involved in numerous mitotic functions, and the timing and mechanisms of each of these events are carefully regulated by its localization within mitosis (Carmena et al., 2012). The selection of Aurora substrates is also aided by a precise and exclusively unique phosphorylation motif (Alexander et al., 2011).
As expected, deregulation of Aurora kinases has drastic intracellular consequences (Meraldi et al., 2004), rendering these kinases attractive drug targets for small-molecule inhibitor development, targeting the active-site binding pocket which binds the ATP substrate (Andrews, 2005). Several small-molecule inhibitors of Aurora kinases have been developed for use as anticancer agents, some of which have progressed to early clinical evaluation (Carpinelli & Moll, 2008). Barasertib is a highly potent and selective inhibitor of Aurora B (K i = 1 nM) in comparison with Aurora A (K i = 1.4 mM) and has high specificity versus a panel of 50 other kinases (Mortlock et al., 2007). Barasertib is a prodrug of a pyrazoloquinazoline Aurora kinase inhibitor (Fig. 1a) and is rapidly converted to the active drug in plasma (Mortlock et al., 2007;Fig. 1a). Consistent with the inhibition of Aurora B kinase, the addition of barasertib to tumour cells in vitro induces chromosome misalignment, prevents cell division and consequently reduces cell viability and induces apoptosis (Yang et al., 2007).
Here, we report the crystal structure of the Xenopus laevis Aurora B-INCENP complex bound to the active form of barasertib. We also report the structure of the same complex bound to a nonhydrolyzable nucleotide that induces the formation of a crystal contact mediated by the substrate-recognition region of Aurora B and a neighbouring INCENP molecule.

Materials and methods
2.1. Protein expression, crystallization and structure determination 2.1.1. Aurora B-INCENP-barasertib complex. The conditions for the expression, purification and structure determination of the X. laevis Aurora B-INCENP complex have been described previously (Sessa et al., 2005). Crystals obtained by microseeding were gradually transferred into cryo-buffer (19% PEG 400, 100 mM bis-tris propane pH 6.5, 2 mM TCEP) and then incubated with 1/100 (v/v) of a 10 mM solution of barasertib (AZD1152-HQPA, Selleck Chemicals) dissolved in DMSO. After 16 h incubation with the inhibitor, the crystals were flash-cooled. X-ray diffraction data were collected from a single crystal on beamline PXIII at the Swiss Light Synchrotron (SLS), Villigen, Switzerland. Data processing was carried out with XDS (Kabsch, 2010) using the xia2 automated datareduction platform (Winter et al., 2013). To refine the orientation and location of the molecules, rigid-body refinement with REFMAC (Murshudov et al., 2011) was employed using the Aurora B-INCENP-hesperadin complex (PDB entry 2bfy; Sessa et al., 2005) as a model. Iterative model building was carried out with Coot (Emsley et al., 2010) and REFMAC (Murshudov et al., 2011), resulting in a good final model (Table 1).
2.1.2. Aurora B-INCENP-AMP-PNP complex. Co-crystals of Aurora B-INCENP in complex with AMP-PNP (Roche) were obtained by incubating the protein in the presence of an excess of the nucleotide analogue (1 mM final concentration) for 1 h at 4 C prior to crystallization. Crystals grew in a similar condition and crystal handling was carried out as described above. X-ray diffraction data were collected from a single crystal on beamline PXIII at the SLS.
Data processing and refinement were carried out as described for Aurora B-INCENP-barasertib, resulting in a good final model (Table 1).
Aurora B-INCENP complex as a structural template (Sessa et al., 2005). The sequence identity between the catalytic kinase domains of X. laevis and Homo sapiens Aurora B is significantly high. We determined the crystal structure of the X. laevis Aurora B-INCENP complex at 1.5 Å resolution, highlighting the binding of the active form of barasertib to the ATP active site (Fig. 1). Barasertib occupies the deep ATP-binding cleft at the interface between the small and the large lobes of the kinase (Fig. 1b), and its binding does not result in any significant conformational changes relative to the unbound kinase, which crystallizes in a partially active state (Sessa et al., 2005). Barasertib is bound at the ATP-binding site of the Aurora B kinase domain, where it spans the entire length of the cavity from the hingeloop to the C helix (Fig. 1b). Stacking interactions with the side chains of Leu99, Phe172, Leu223, Leu154 and Val107 hold the quinazoline and anilino moieties of barasertib in place (Fig. 1c). Furthermore, the carbonyl group and the amino group of the 3-fluoroaniline are engaged in hydrogen bonding to Lys122 and Gln145, respectively (Fig. 1c). The 3-fluoroaniline group is housed in a hydrophobic pocket formed by Leu138 and Leu168, disrupting the ion pair formed between the conserved Lys122 and Glu141 residues (Fig. 1c).

Unravelling barasertib selectivity towards the Aurora B isoform
Of the many Aurora B inhibitors that have been reported in the literature, in recombinant enzyme assays barasertib shows a higher selectivity, with more than a 1000-fold greater potency, for Aurora B than for Aurora A, with IC 50 values of 1 nM and 1.4 mM, respectively (Mortlock et al., 2007). In order to elucidate the structural basis that underlies this selectivity, we superimposed the Aurora B-INCENPbarasertib complex with the Aurora A-TPX-2 complex (Bayliss et al., 2003; Fig. 1d). Although the ATP-binding cleft is highly conserved between the two enzymes, the biochemical data clearly show strong selectivity of barasertib towards Aurora B (Mortlock et al., 2007;Fig. 1d). The X. laevis Aurora B residues engaged in hydrogenbonding to barasertib within the active site are well conserved in the ATP-binding site of H. sapiens Aurora A (Fig. 1d) and are therefore likely to interact with the small molecule in a similar manner. The only observed difference is the hydrogen-bonding of a negatively charged residue (Glu177 Aurora B ) to barasertib through a well defined water molecule (Fig. 1d). In Aurora A, this residue is replaced by a smaller threonine residue (Thr217 Aurora A ) that is too distant to influence the inhibitor binding (Fig. 1d). Both Glu177 Aurora B hydrogen-bonding and van der Waals contributions might justify the selectivity of barasertib towards Aurora B. Previous work has also established the key role of Thr217 Aurora A in determining the selectivity of the small molecule MLN8054 towards Aurora A (Dodson et al., 2010). Interestingly, a well positioned arginine residue (Arg97 Aurora B , Arg137 Aurora A ) is ready to stabilize the hydroxyl functional group of the active molecule in plasma in both Aurora isoforms (Fig. 1d).

An unexpected crystal contact occupies the Aurora B substratebinding region
During refinement of the Aurora B-INCENP-barasertib structure, we noticed extra electron density corresponding to a well ordered arginine residue, Arg847 INCENP , in close proximity to the activation loop (data not shown). We sought to investigate this further, and grew crystals of Aurora B-INCENP in the presence of a nonhydrolyzable ATP analogue (AMP-PNP). The 1.7 Å resolution electron-density map revealed that the nucleotide phosphate groups stabilized an unexpected crystal-packing contact (Fig. 2a). The C-terminus of a neighbouring INCENP molecule occupies the Aurora B substratebinding region of an adjacient symmetry mate (Fig. 2). The INCENP C-terminal peptide was found in an extended conformation engaged with the large Aurora B C-terminal lobe (Fig. 2a). INCENP residues from Lys839 INCENP to Arg847 INCENP structurally mimic a substrate peptide bound to the large C-lobe of the catalytic domain of the Aurora B kinase domain (Fig. 2a). The well defined electron-density map clearly reveals contiguity of the main chain of the INCENP C-terminus up to the last residue of the crystallized construct (Arg847 INCENP ; Fig. 2b). The structure of the kinase domain was not perturbed and it adopts a highly similar overall conformation to the previous published structures of the catalytic domain both alone and in complex with small-molecule inhibitors (Andersen et al., 2008;D'Alise et al., 2008;Girdler et al., 2008;Sessa et al., 2005). Although this segment of INCENP does not perfectly match the Aurora B consensus motif (Alexander et al., 2011), it is highly enriched with positively charged residues promoting the docking of the INCENP stretch onto Aurora B (Figs. 2a and 2b). During the structure refinement, we were intrigued by the precise docking of the INCENP C-terminus onto the Aurora B substrate-recognition region. Multiple sequence alignment revealed high conservation of the residues preceding the well conserved Thr848-Ser849-Ser850 INCENP (TSS) motif (Fig. 2c), which is absent in our crystallization construct and has been shown to be crucial to achieve full Aurora B activity (Honda et al., 2003;Sessa et al., 2005). Of particular interest is the well ordered Arg847 INCENP residue deeply anchored into a pocket formed between pThr248 Aurora B and the C helix, which is able to displace the side chain of the Arg140 Aurora B residue that in turn is involved in pThr248 Aurora B phosphate-group coordination (Fig. 2b)  i jI i ðhklÞ À hIðhklÞij= P hkl P i I i ðhklÞ, where I i (hkl) is the ith observed intensity of a reflection and hI(hkl)i is the average intensity obtained from multiple observations of symmetry-related reflections. ‡ R conv = P hkl jF obs j À jF calc j = P hkl jF obs j, where F obs and F calc are the observed and calculated structure-factor amplitudes, respectively. § R free is equivalent to R conv for a 5% subset of reflections that were not used in the refinement.  Zheng et al., 1993) (e). Surfaces are coloured according to the electrostatic potential distribution of the kinases, ramped from blue (positive) to red (negative). ( f ) Schematic representation of an Aurora B potential transinhibitory model proceeding through an intermolecular mechanism, which may be enhanced by a high local protein concentration. This transinhibition mechanism may be relieved by a pTSS-pThr248 repulsion and lead to full activity of Aurora B. A Ser/Thr phosphatase can then revert the full Aurora B activity to a transinhibited state. which projects out from a neighbouring Aurora B molecule, coordinated by a negatively charged pocket (mainly Glu281 Aurora B ; Figs. 2b and 2d). Further interactions arise from the nucleotide phosphate group coordinating the positively charged residue Lys839 INCENP and Lys841 INCENP into the ATP-binding pocket (Figs. 2b and 2d), thus contributing with Arg847 INCENP to stabilization of the INCENP peptide on the Aurora B surface.

Comparison with protein kinase A in complex with the inhibitor peptide PKI
Although unexpected, the crystal contact favoured by the presence of a nucleotide within the Aurora B ATP-binding cleft may resemble the inhibitory mechanism observed for protein kinase A (PKA), in which an inhibitor protein (PKI) was originally co-discovered (Walsh et al., 1971). Interestingly, it was discovered to act by competitively binding with the protein substrate to the catalytic subunit of the protein kinase, binding with high affinity in an ATP-dependent manner (Ashby & Walsh, 1973). The structure of protein kinase A (PKA) was first reported in complex with the inhibitor peptide PKI, in which the position P(0) is occupied by an alanine (Knighton et al., 1991). The residues in positions P(À3), P(À2) and P(À1) of PKI bind to the C-lobe of the kinase in a similar orientation to that observed for the INCENP segment bound to the Aurora B C-lobe (Figs. 2d and 2e). In support of this hypothesis, it has been reported that high local concentrations of Aurora B are required to obtain transphosphorylation on the T-loop and thus full Aurora B activity (Kelly et al., 2007). A mechanism for modulating the catalytic activity of Aurora B through a mimicking substrate-inhibitory strategy in cellular compartments where the enzyme is highly enriched is therefore not unreasonable (Fig. 2f ). Furthermore, the adjacent phosphorylatable TSS segment may play a role in relieving this transinhibitory mechanism (Fig. 2f ). Phosphorylation of the TSS motif is indeed required for full Aurora B activity (Bolton et al., 2002;Bishop & Schumacher, 2002;Honda et al., 2003;Sessa et al., 2005). Our structural data predict that phosphorylation of the TSS motif would disfavour its proximity to the pThr248 Aurora B residue on the activation loop. In this model, the phosphorylated TSS motif of INCENP could reverse the inhibitory substrate-mimicking effects, generating a fully active kinase (Fig. 2f ). This model would explain the still unknown role of the phospho-TSS motif in achieving full Aurora B activity. Counteracting Ser/Thr phosphatases could then limit Aurora B activity when enriched at centromeres and/or kinetocores (i.e PP1 and PP2A phosphatases; DeLuca et al., 2011;Foley et al., 2011;Fig. 2f ). These data also provide useful insights for the design of Aurora-specific competitive synthetic peptide inhibitors. These pseudosubstrate inhibitor peptides could represent a valid and specific alternative to ATP-competitor small molecules.