Dynamic Equilibrium of the Aurora A Kinase Activation Loop Revealed by Single‐Molecule Spectroscopy

Abstract The conformation of the activation loop (T‐loop) of protein kinases underlies enzymatic activity and influences the binding of small‐molecule inhibitors. By using single‐molecule fluorescence spectroscopy, we have determined that phosphorylated Aurora A kinase is in dynamic equilibrium between a DFG‐in‐like active T‐loop conformation and a DFG‐out‐like inactive conformation, and have measured the rate constants of interconversion. Addition of the Aurora A activating protein TPX2 shifts the equilibrium towards an active T‐loop conformation whereas addition of the inhibitors MLN8054 and CD532 favors an inactive T‐loop. We show that Aurora A binds TPX2 and MLN8054 simultaneously and provide a new model for kinase conformational behavior. Our approach will enable conformation‐specific effects to be integrated into inhibitor discovery across the kinome, and we outline some immediate consequences for structure‐based drug discovery.

Protein kinases are essential for the regulation and signaling of eukaryotic cells and are important drug targets in cancer and inflammatory disease. [1] Many kinases are regulated by phosphorylation of ar egulatory Ser/Thr/Tyr residue on ar egion of the kinase known as the activation loop or T-loop.T he influence of phosphorylation and interactions with small-molecule inhibitors on kinase conformation can be summarized by two models.I nt he first model, phosphorylation achieves activation by "locking" the activation loop in ac onformation where the catalytic residues are aligned ( Figure 1a). [2] In the second model, an inactive-conformation kinase bound to at ype II inhibitor (an inhibitor whose binding site extends into aspecific allosteric pocket adjacent to the ATP-binding site) is in equilibrium with the ligand-free kinase in an active conformation (Figure 1b). In the context of these models,t he active conformation is typified by the activation loop being oriented to form the protein substrate binding site and the aspartic acid of the conserved DFG motif at the beginning of this loop pointing into the ATPb inding site to coordinate Mg 2+ /ATP (DFG-in). In the classical inactive conformation, both the activation loop and the DFG motif are rotated by 1808 8,and aphenylalanine replaces the aspartic acid in the ATP-binding pocket (DFG-out). Both models have been supported by X-ray crystallography but direct experimental testing of kinase activation loop mobility has proved impossible to date as activation loop dynamics occur on the same timescale as NMR intermediate exchange. [3] To test these models directly,w eu sed single-molecule fluorescence spectroscopy to monitor the conformation of the kinase activation loop.W ei mplemented our assay in Aurora Ak inase,amitotic kinase whose catalytic activity can be increased in the presence of its protein binding partner TPX2 [4] and inhibited with numerous drug-like small molecules. [4d, 5] Interest in Aurora Aa nd its conformational plasticity has recently increased owing to the discovery that inhibiting Aurora Awith MLN8054 or CD532, both observed to bind Aurora Ainaninactive T-loop conformation, disrupts the interaction of Aurora Aw ith the proto-oncogenic transcription factor N-Myc. This leads to degradation of N-Myc and offers an alternative therapeutic strategy for N-Mycdriven tumors (currently under clinical evaluation with MLN8237/alisertib). [5d, 6] These factors make Aurora Aa n excellent clinically relevant system for studying the heterogeneity and dynamics of T-loop conformation underlying the observed structural and catalytic properties of protein kinases.
Forclarity,weuse the terms "active T-loop" and "inactive T-loop" to refer to the two orientations of the activation loop (Figure 1c,d;s ee also the Supporting Information, Figure S1a). We reserve the terms DFG-in and DFG-out for discussions of the detailed orientation of the DFG motif (Figure S1 b).
We labeled ap seudo-wildtype construct of Aurora A (C290A/C393A) [7] with tetramethylrhodamine (TMR) on cysteine residues introduced into the activation loop (S283C) and the Nlobe of the kinase (K224C;F igure 1d). Thef luorescence of TMR is quenched when two molecules are closer than about 15 ,a nd this phenomenon has previously been used to probe small distance changes at the single-molecule level. [8] Control reactions confirmed that only two sites were available for coupling ( Figure S2).
Thec atalytic activity of TMR-labeled phosphorylated Aurora Aissimilar to that of the unlabeled pseudo-wildtype protein ( Figure 1e). We tethered labeled protein to ac over slip via aH is tag ( Figure S4) and imaged the fluorescence intensity by total internal reflection fluorescence (TIRF) microscopy.Single-molecule fluorescence traces from doublelabeled Aurora Am olecules were identified by the presence of two-step photobleaching at their end. [9] Before photobleaching,t hese molecules exhibited as ingle high fluorescence intensity and transiently entered al ow-intensity, quenched state (example trace in Figure 1f), which we interpret to indicate changes in the position of the Aurora A T-loop:h igh fluorescence indicating an active-conformation T-loop,and quenched fluorescence an inactive T-loop (see the Experimental Methods Section in the Supporting Information).
Thef luorescence intensity histogram for phosphorylated Aurora Ashows two peaks,indicating that the kinase adopts both active and inactive T-loop conformations (Figure 2a). Ther elative areas of the peaks indicate that in solution the majority of molecules adopt an active T-loop conformation and am inority an inactive conformation (Table 1). This is consistent with the active T-loop conformation observed in X-ray structures of uninhibited kinase and with the high catalytic activity of phosphorylated Aurora A. [4b] Contrary to the model of al ocked phosphorylated T-loop,o ur data Active T-loop:b lue, inactive T-loop:p ink. Activation loop shown in cyan/bright pink. Labeling sites (K224 in orange and S283 in pink/cyan) shown as filled spheres.P DBs 1OL5 and 2WTV.e)Kinase activity assay for unlabeled pseudo-wildtype Aurora A( &)a nd TMR-labeled K224C/S283C (~). Activity shown as the [ADP] produced over the course of a1hreaction. The differencein activity between pseudo-wildtype and labeled protein cannot be accountedf or by incomplete protein labeling (see the SupportingI nformationf or labeling efficiency). f) Example trace of ad ouble-labeled K224C/S283C Aurora As ingle molecule showing the background-subtracted fluorescence intensity over time.
indicate that the activation loop of phosphorylated Aurora A exists in adynamic structural equilibrium (K eq = 0.3 AE 0.1).
It has been suggested that the free energy penalty of interconverting between active and inactive conformations of the tyrosine kinases Src and Abl underlies the selectivity of type II inhibitors such as Imatinib,a lthough this has been challenged. [10] Consequently,t here have been efforts to calculate the free energy difference between these conformations.N one of these studies explicitly distinguished between the orientation of the DFG motif and the overall orientation of the T-loop (usually assumed to be coupled);h owever, the experimental protocols suggest that they report on the interconversion between DFG-in active T-loop and DFG-out inactive T-loop conformations. [11] Our experimental measurements report directly on the interconversion between active and inactive T-loop conformations (without measuring the orientation of the DFG motif), and the free energy difference that we obtained for phosphorylated Aurora Ai s0 .7 AE 0.1 kcal mol À1 (DG inactive-active ). This is very similar to the value of 0.8 kcal mol À1 that has been measured for the two unassigned conformations of phosphorylated Erk2 [12] and considerably less than the calculated values for Src. [11a,c] We also measured the microscopic rate constant for adopting an active T-loop conformation, k active (Figure 2b; k active = 2.3 AE 0.2 s À1 ). We were unable to measure k inactive directly as the observation time of individual molecules was limited by photobleaching, but we calculated this to be 0.7 AE 0.2 s À1 from K eq .
To determine how kinase substrates influence the conformation of the activation loop,w em easured fluorescence intensity distributions of Aurora Ai nt he presence of saturating levels of ATP, kemptide (a 7-residue peptide substrate), and AMP-PNP (a non-hydrolysable analogue of ATP; Figure 2c-f). Neither kemptide nor AMP-PNP changed the position of the equilibrium from that of unliganded kinase while,surprisingly,A TP alone and AMP-PNP/kemptide both slightly increased the population of the inactive T-loop conformation (Table 1). We next measured the intensity distribution of Aurora A in the presence of the activator TPX2 (Figure 2g). Occupancy of the active T-loop conformation was increased (Table 1), consistent with the increased catalytic activity of the enzyme. To build apicture of the enzyme poised for maximal activity, we measured the distribution of the Aurora A-TPX2 complex bound to AMP-PNP and kemptide (Figure 2h). This adopted apredominantly active T-loop conformation, similar to that of the Aurora A-TPX2 complex alone.
MLN8054 and CD532 are both nanomolar inhibitors of Aurora A, and X-ray structures show that each binds in an inactive T-loop conformation (Figure S1 c, d). [5b,d,g] Although both are referred to as type II inhibitors,neither extends into the allosteric hydrophobic pocket, and neither has been captured binding the kinase with acanonical DFG-out motif.
Aurora Ab ound to MLN8054 adopts an unusual DFG conformation previously termed DFG-up, [5b] and Aurora A bound to CD532 is DFG-in. [5d] To determine the effect of these inhibitors on the activation loop of Aurora Ai ns olution, we repeated our assay in their presence (Figure 2i,j). Each inhibitor resulted in al arge increase in the population of the inactive T-loop conformation, which is consistent with the crystal structures (Table 1). Control measurements showed that neither inhibitor affected the peak fluorescence intensity of TMR, although peak broadening was observed, particularly for CD532 ( Figure S5 and Table S1). Then umber of fluorescent molecules in the field of view remained constant within each control (35 AE 6k inase alone,3 4AE 5w ith MLN8054; n = 12).
As TPX2 and MLN8054 move the position of the conformational equilibrium in opposing directions,w ew ondered how they interacted when present simultaneously.This is ap hysiologically relevant scenario at the mitotic spindle, and kinetic studies have shown that the presence of TPX2 increases the K i value of MLN8054 by afactor of greater than four. [5b] Similar changes have been observed for VX680 and GSK623906A. [4d] In the presence of both MLN8054 and TPX2, Aurora Aa dopted ap redominantly active T-loop conformation, similar to that for TPX2 alone (Figure 2k and Table 1).
To determine whether this result represents am ixture of binary Aurora A-MLN8054 and Aurora A-TPX2 complexes or population of an Aurora A-TPX2-MLN8054 triple complex, we calculated the expected experimental result for the mixture of binary complexes based on the published affinities of the two ligands [4b,5b] (see the Supporting Information). A mixture of binary complexes would result in an inactive T-loop population of 43 %, which is inconsistent with the experimental result (15 %; Table 1). Our experimental samples must thus contain at riple complex of Aurora A-TPX2-MLN8054. We were unable to quantify the extent of triple complex formation, but to account for the experimental results,itmust be the major species and must occupy amainly active T-loop conformation.
Our measurements indicate that phosphorylated Aurora Ai sn ot locked into as ingle conformation, and that [h] K m measured in this study ( Figure S3) with an error of AE 70 mm.ND= not determined.
it spontaneously interconverts between active and inactive T-loop conformations in solution, even in the presence of saturating quantities of the small-molecule inhibitors MLN8054 and CD532 or of TPX2. To explain our results, we hypothesized that either the ligand residence time is short [and rearrangement of the activation loop occurred in the unliganded kinase;E quations (1), (2), and (3)] or that Aurora Ab ound to saturating quantities of ligand can interconvert between T-loop conformations.
Active Ð InactiveG þinh HInactive-inh Ð Inactive * -inh ð2Þ We tested this by using the K eq value of the kinase alone and the known ligand K d to predict the observed K eq value in the presence of ligand (see the Supporting Information). Equation (1), an equivalent scheme for TPX2, and Equations (2) and (3) (which postulate the presence of as econd sub-conformation, inactive*, within the inactive T-loop conformation) all predicted K eq values that were 2-4 orders of magnitude different from those measured. We therefore concluded that ligand-bound Aurora Ai nterconverts between T-loop conformations and modelled our data by using Equation (4). This surprising conclusion is supported by X-ray crystallography.T hree PDB structures (2WTV,3 H10, and 2X81) show Aurora bound to MLN8054 or asimilar compound in an inactive T-loop conformation, while in af ourth, low-resolution structure (2WTW), the crystal packing is incompatible with an inactive T-loop,a nd the MLN8054-bound kinase adopts an active T-loop conformation. [5b] All four Aurora-MLN8054 structures superpose within the Na nd Clobes of the kinase,v arying only in the orientation of the activation loop and in the exact angle between the two lobes.T he position of MLN8054 relative to the Nlobe is identical across all structures,implying that no change in the binding mode of the inhibitor is required for interconversion between active and inactive T-loop conformations.W ep ropose that 2WTW and 2WTV represent two snapshots of the Aurora A-MLN8054 conformational ensemble.I nterconversion between these two conformations is brought about by movement of the activation loop without the inhibitor leaving the binding site.
To determine whether the reported change in K i of MLN8054 upon addition of TPX2 [5b] reflects atrue change in the K d value of MLN8054, or whether it can be accounted for by changes in the position of the conformational equilibrium alone,w ep artitioned the K i value for MLN8054 into conformation-specific dissociation constants (Equation (4) and the Supporting Information for modeling). In the absence of TPX2, K d,active = 1.0 nm and K d,inactive = 0.4 nm,w hile in the presence of TPX2, these values are equal and remain unchanged (within experimental accuracy) at 3.3 nm,indicating agenuine change in K d for the Aurora A-TPX2 complex.
Our analysis (see the Supporting Information, Equation (S18)) also provides insight into the drivers of inactive and active T-loop conformations.T he position of the equilibrium for any binding partner depends on the K d,active /K d,inactive ratio and the intrinsic position of the equilibrium for the kinase (K eq,free ,which is aconstant for each protein). In other words,the proportion of ligand-bound kinase molecules in an inactive T-loop conformation is driven by the degree to which an inhibitor can discriminate between inactive and active T-loop conformations,n ot by the overall inhibitor-kinase affinity.
Our reported measurements derive from in vitro experiments.W hile the exact values that we have determined may change in the cellular environment, we expect the principle of conformational equilibria and the models of inhibitor binding that we have established to translate into cell-based contexts. Our results thus have an umber of consequences for drug discovery: 1) At least two Aurora Ai nhibitors bind ac onformation of the kinase (the active T-loop conformation) that had previously not been expected. It is now possible to measure (and thus develop validated prediction algorithms for) the effect of an inhibitor on kinase conformational equilibria in solution. 2) Potent conformation-independent inhibitors need to bind both active and inactive T-loop conformations,a nd structure-based drug design may need to focus on common features of these. 3) Forinhibitors binding to the inactive T-loop conformation, differential binding to this conformation should be maximized. This is particularly important for inhibitors designed to induce aspecific conformation of the kinase in order to disrupt aphysiological interaction (e.g., Aurora A with N-Myc). While the affinity for the active T-loop conformation is retained, we expect to find as mall proportion of the complex (e.g., Aurora A-CD532-N-Myc) in the active T-loop conformation, even at saturating concentrations of inhibitor. 4) Fort hose kinases where K eq,free ! 1, an inactive T-loop inhibitor must achieve greater discrimination between active and inactive T-loop conformations than an inhibitor for ak inase where K eq,free ! 1. Some inhibitors may thus appear to be inactive T-loop inhibitors when bound to one target and active T-loop inhibitors when bound to another, potentially contributing to unexpected cellular phenotypes.W ep redict that the kinase phosphorylation state will also affect the value of K eq,free . 5) Modification of the target protein (e.g., by binding to aphysiological protein partner such as TPX2) can change the binding affinity of an inhibitor beyond what would be predicted solely from ac hange in the position of the conformational equilibrium. Distinguishing between allosteric partners such as TPX2 and scaffolding partners may be possible from X-ray structures,f rom enzyme activity assays,o rp otentially by inference from physiological function (e.g., catalytic activation vs.s ubstrate recruitment). This means that early decisions over the best form of the enzyme to target are still important. 6) Success in allosteric disruption by using an ATP-competitive small molecule to displace ac onformation-specific physiological ligand will depend on the properties of the protein-protein interaction to be disrupted ( Figure S6). By carefully matching inhibitors and interactions,i tm ay be possible to achieve specificity between different binding partners of the same kinase.I np ractice,t he limits of this approach will need to be found experimentally.