Research ArticleBiochemistry

A Kinetic Test Characterizes Kinase Intramolecular and Intermolecular Autophosphorylation Mechanisms

See allHide authors and affiliations

Science Signaling  02 Jul 2013:
Vol. 6, Issue 282, pp. ra54
DOI: 10.1126/scisignal.2003910


Many protein kinases catalyze their own activation by autophosphorylation. The mechanism of this is generally considered to be intermolecular and similar to that used in substrate phosphorylation. We derived the kinetic signatures of the four simplest autophosphorylation reactions and developed a test to determine the autoactivation mechanism of individual kinases. Whereas autophosphorylation of Nek7 and Plk4 occurred through an intermolecular mechanism, the kinases Aurora-A and Chk2 followed an intramolecular mechanism. Autophosphorylation of Aurora-A was accelerated in the presence of its protein activator TPX2. Nek9, the binding partner for Nek7, had a concentration-dependent effect such that low amounts enhanced autoactivation of Nek7 and high amounts were inhibitory. A structural model of Aurora-A undergoing autophosphorylation confirmed that an intramolecular mechanism is physically possible, and provided an explanation for how TPX2 could stimulate both autophosphorylation and substrate phosphorylation. The distinct mechanisms of autoactivation have consequences for cellular regulation because each molecule of a kinase that undergoes intramolecular autophosphorylation is activated individually, whereas the activity of kinases that undergo intermolecular autophosphorylation can be rapidly self-amplified in the cell. Local control of individual molecules, such as Aurora-A, may be particularly advantageous for a kinase with multiple, distinct cellular roles.


Protein kinases regulate many critical cellular processes and are a major class of drug target for many human diseases. Kinase activity is controlled through multiple mechanisms and commonly includes a requirement for phosphorylation of a regulatory serine, threonine, or tyrosine in a region of the kinase known as the activation loop for full catalytic activity. This phosphorylation is often mediated by a second kinase, itself regulated by phosphorylation, and control of cellular processes can be obtained by a network of kinases acting together in feedback loops, each regulating its partners. Many kinases catalyze the phosphorylation of their own activation loops (autophosphorylation), and the mechanism of this reaction also provides insights into kinase regulation in the cell.

Perhaps surprisingly, the mechanism of autophosphorylation is known for very few kinases. Autophosphorylation has often been assumed to proceed by an active, phosphorylated kinase molecule phosphorylating the loop of an inactive, unphosphorylated substrate kinase molecule in the same manner as the phosphorylation of any other substrate (1). Consequently, detailed studies of autoactivation have focused on those kinases for which there is an apparent conflict between the sequence of the activation loop and the substrate consensus sequence of the kinase in question. The mechanisms of autoactivation determined have been deemed exceptional because they are unlike the mechanism by which the kinase phosphorylates other substrates (14).

We investigated the mechanism of autoactivation of Aurora-A and Nek7, two mitotic serine/threonine kinases that we have characterized with respect to other aspects of regulation (58). Aurora-A has several cellular roles in mitosis, additionally functions in disassembly of cilia and in neurite extension, and has emerged as a promising target for cancer drug discovery with one inhibitor in phase 3 clinical trials (9). Aurora-A activation involves the autophosphorylation of Thr288 within its activation loop, and this phosphorylation event is controlled by binding partners such as TPX2 and the cellular phosphatase PP6 (1012). Because the primary sequence context of Thr288 is a good match for the consensus substrate recognition sequence of Aurora-A, it has generally been assumed that autophosphorylation is intermolecular and is essentially the same reaction that occurs for the phosphorylation of other substrates by Aurora-A (13). Nek7 and the closely related Nek6 are both essential for mitotic spindle assembly (14). Their encoding genes are highly expressed in human cancers, and therefore these proteins are potential cancer drug targets (15). Both are regulated through phosphorylation on their activation loops (Ser195 on Nek7), and this can occur by autophosphorylation or through transphosphorylation by an upstream kinase such as Nek9 (16, 17). There is insufficient information on Nek7 substrate recognition to predict its autophosphorylation mechanism.

A challenging aspect of kinase inhibitor design is ensuring selectivity, and autophosphorylation offers the potential of a reaction mechanism, distinct from other substrate phosphorylations, to target with selective kinase inhibitors (2, 3, 1820). Modern high- and medium-throughput assays enable easy and accurate measurement of kinase activity in the context of inhibitor programs (21). For example, an automated mobility shift assay (MSA) system measures the progression of the reaction as a ratio of phosphorylated product to substrate, separated by capillary electrophoresis, producing precise data with a high signal-to-noise ratio. We have previously used this system to probe the energetic contributions of phosphorylation and TPX2 binding to Aurora-A activation and the mechanism of inhibition of Aurora-A by the inhibitor MLN8054 (6, 7). Antibody-based assays, such as DELFIA (dissociation-enhanced lanthanide fluorescent immunoassay), can be used to monitor the accumulation of phosphorylation on the activation loop directly but have not been used to characterize the underlying mechanism of autoactivation, perhaps because the necessary kinetic schemes for modeling the results are not available. Kinetic analysis has the potential to answer questions about autoactivation decisively, and a simple method for distinguishing the different kinase autophosphorylation mechanisms would have applications in basic research and in drug discovery. To develop a robust platform to interpret autoactivation mechanisms, we built upon previous work (1, 3, 22) and modeled the expected behavior of the four simplest kinetic schemes. Two of these schemes are intermolecular (the action of one kinase molecule acting on a second), and two are intramolecular (a single kinase molecule acting on itself).


Substrate phosphorylation by kinases undergoing autophosphorylation exhibits a lag phase

We monitored the phosphorylation of substrate peptide by initially unphosphorylated Aurora-A as the enzyme itself underwent autophosphorylation (Fig. 1A). The experimental progress curves for peptide substrate phosphorylation showed a lag phase in the appearance of phosphorylated product as expected for an autoactivation reaction in which the concentration of active enzyme increases over time. The lag phase did not occur when the enzyme was preincubated with adenosine 5′-triphosphate (ATP) before addition to the substrate (Fig. 1B). Changing the length of the preincubation from 2 to 6 hours did not change the rate of product formation, indicating that autoactivation was complete within the first 2 hours (fig. S1). We postulated that analysis of the lag phase might yield insights into the mechanism of autophosphorylation.

Fig. 1 Substrate phosphorylation by initially unphosphorylated kinase exhibits a lag phase.

Progress curves for Aurora-A phosphorylation of peptide substrate. (A) Substrate conversion to product over time with initially unphosphorylated Aurora-A. The different colored lines represent twofold increasing enzyme concentrations (blue to red). (B) Substrate conversion to product over time with Aurora-A that had been preincubated with ATP for 2 hours.

Mathematical modeling of the four basic mechanisms of autophosphorylation reveals that a combination of graph shape and data normalization can distinguish the different mechanisms

To analyze the lag phase of autoactivation further, we built a kinetic model of the reactions taking place in our assay as the kinase autoactivates and phosphorylates substrate peptide (Fig. 2A). In this model, an unphosphorylated enzyme molecule, E, is first phosphorylated to an active kinase, E*, which subsequently acts upon a substrate peptide according to Michaelis-Menten kinetics.

Fig. 2 Cartoons and schemes for the autoactivation reaction.

(A) Underlying model for phosphorylation of a substrate peptide by an activated enzyme. (B) Two-state intramolecular autoactivation. (C) Three-state intramolecular autoactivation. The central enzyme state, E′, is unfocused to reflect the fact that there is no structural information available for this state, which has been termed the transitional intermediate elsewhere (2, 3). (D) Two-state intermolecular reciprocal autoactivation. (E) Two-state intermolecular substrate-like activation. Mechanisms in parentheses are not illustrated with cartoons and cannot be distinguished from the illustrated mechanism by the techniques shown in this paper.

We next considered the autoactivation reaction itself, EE*, which may progress according to either an intermolecular (trans) or an intramolecular (cis) mechanism. In an intramolecular reaction, one molecule of E will react upon itself to give one molecule of E*, either directly (two-state; Fig. 2B) or through a kinetic intermediate E′ (three-state; Fig. 2C). In the simplest intermolecular reaction, two molecules of E will come together with each molecule phosphorylating the activation loop of the other to generate two molecules of E*. We have termed this reciprocal autophosphorylation (Fig. 2D). Alternatively, one molecule of E* can act upon one molecule of E to give two molecules of E*, which we called substrate-like autophosphorylation (Fig. 2E). This latter mechanism requires the initial generation of a single molecule of E* as a priming event, either by an upstream kinase or by autophosphorylation through one of the other three mechanisms.

We derived equations to describe the changes in concentration of different enzyme forms over the course of autoactivation for each of these four mechanisms. We then applied these to the situation found in our experimental assay in which autoactivation and substrate peptide phosphorylation proceeded alongside one another in the same reaction mixture, but only the accumulation of phosphorylated peptide was monitored (see Materials and Methods and text S1 to S3). We have illustrated our simulations graphically by plotting the concentrations of phosphorylated peptide and phosphorylated kinase as a function of time for three different concentrations of enzyme (Fig. 3). We have also illustrated the rate of formation of phosphorylated enzyme (dE*/dt).

Fig. 3 Modeled data showing the kinetic signatures of autoactivation.

(A) Intramolecular mechanisms showing multiple enzyme concentrations (top) and normalized data in which the values in the top row were divided by [enzyme] (bottom). (B) Intermolecular mechanisms showing multiple enzyme concentrations (top) and normalized data in which the values in the top row were divided by [enzyme] (bottom). Left graphs, formation of phosphorylated peptide substrate; center graphs, [phosphorylated enzyme] (the equivalent to differentiating the adjacent left graph); right, rate of formation of phosphorylated enzyme (the equivalent to differentiating the adjacent center graph). Light red, two-state intramolecular mechanism; dark red, three-state intramolecular mechanism; dark blue, two-state intermolecular mechanism (substrate-like autophosphorylation); light blue, two-state intermolecular mechanism (reciprocal autophosphorylation). Solid lines, high [enzyme]; dashed lines, medium [enzyme]; dotted lines, low [enzyme]. Schemes modeled using k0 = 0.05, k1 = 0.5, k2 = 0.2, kobs = 0.3, v′ = 1, α = 0.1, and [enzyme] = 1, 0.5, and 0.25. X-axis ticks are at identical time intervals in all graphs; y axes are of the same magnitude for each set of vertical panels.

All four mechanisms exhibit a lag phase in the production of substrate (Fig. 3, left), an indication that the enzyme undergoes activation in the course of the reaction. The normalized traces for different enzyme concentrations in the intramolecular mechanisms all overlay exactly (Fig. 3A, lower) because, physically, no complex of two enzyme molecules is formed, and thus none of the equations contains a term that is the product of two (or more) enzyme concentrations. This is not the case for the intermolecular reactions (Fig. 3B, lower).

The three-state intramolecular mechanism and the substrate-like intermolecular mechanism exhibit a lag phase in the production of active enzyme (Fig. 3, center). These mechanisms also show peaks in the rate of production of active enzyme, whereas the other two mechanisms generate smooth decay curves (Fig. 3, right). Combining this with the information from normalization, a combination of graph shape and dependence on total enzyme concentration provides a simple test to distinguish between the four simplest mechanisms of autoactivation (two intermolecular, two intramolecular).

To apply our test to experimental data, we needed to generate a panel of experimental curves equivalent to either Fig. 3A or Fig. 3B. This can be done from a single experimental run of the kind routinely carried out in drug discovery projects, in which the accumulation of phosphorylated substrate peptide is monitored over time at a number of different enzyme concentrations. Numerically differentiating these data (dP/dt, calculating the gradient between each successive pair of data points) produces a quantity that is directly proportional to the concentration of phosphorylated enzyme. Numerically differentiating these data again (d2P/dt2) generates the rate of formation of phosphorylated enzyme. Calculating the normalized data curves from this information is simply a case of dividing the value of each data point by the concentration of enzyme used in that experiment.

Autoactivation kinetics have previously been measured as the direct incorporation of radiolabeled phosphate into the enzyme (23) or the effect of autoactivation on the phosphorylation of peptide substrate using a coupled reaction in a continuous format (22, 24, 25). Whether the reaction is intermolecular or intramolecular has been determined from an analysis of the linear dependence of product formation on enzyme concentration: If double the enzyme concentration leads to double the product formed (with the same incubation time), then the reaction is intramolecular. This is analogous to normalization of our curves. Calculation of the intercept of the asymptotes to initial and steady-state rates (the induction time) or the x-intercept of the steady-state reaction as a function of enzyme concentration (22, 24) has also been used to infer reaction type. If the induction time or x-intercept (the time taken for phosphorylated enzyme to be formed) remains constant with varying enzyme concentration, then autophosphorylation is intramolecular. If it decreases with increasing enzyme concentration, then the reaction must be intermolecular. When we performed simulations and plotted the x-intercept of the steady-state reaction for each of the four activation mechanisms and compared those to the experimental data obtained for Aurora-A (fig. S2 and text S4), we found that these alternative approaches were unreliable.

Nek7 autophosphorylates through a substrate-like intermolecular mechanism

We monitored Nek7 autophosphorylation with an automated MSA system to measure the progress of substrate peptide phosphorylation, which revealed a prominent lag phase (Fig. 4A, left). Normalizing the progress curves by total enzyme concentration (T0) revealed curves that did not superpose (the duration of lag phase depended on enzyme concentration), indicating autophosphorylation by an intermolecular mechanism (Fig. 4B, left). Numerical differentiation of the progress curves (dP/dt, calculated by determining the gradient between each successive pair of data points) followed by normalization showed that the time taken to reach the maximum concentration of active enzyme was dependent on enzyme concentration (Fig. 4, center), which is also consistent with an intermolecular activation mechanism. The singly differentiated data revealed a lag phase, which translated into a series of peaks in the doubly differentiated curves (dET*/dt), both of which signify that the mechanism of activation was substrate-like and intermolecular (Fig. 4, right). This mechanism is consistent with the accepted model for most protein kinases.

Fig. 4 Experimental progress curves for the Nek7 reaction measured by MSA.

(A) Measured curves. (B) MSA data normalized by total amount of enzyme (T0). Left graphs, product formation; center graphs, dP/dt first differential of the adjacent left graph (∝[ET*]); right graphs, d2P/dt2 first differential of middle graph (∝dET*/dt). Each curve is a twofold dilution of previous [enzyme] starting at 600 pmol enzyme (20 μM) and progressing red–orange–yellow–dark green–mid green–light green. Data are truncated at 20% substrate conversion to avoid artifacts from non–Michaelis-Menten kinetics as substrate becomes limiting. Data shown represent two replicate experiments.

Aurora-A autophosphorylates through a three-state intramolecular mechanism

We investigated autophosphorylation of full-length Aurora-A (fig. S3) and of the catalytic domain alone (amino acids 122 to 403) (Fig. 5). Both produced similar results, although the catalytic domain produced better quality data.

Fig. 5 Aurora-A activation measured by MSA.

(A) Experimental progress curves for the Aurora-A reaction measured by MSA. (B) MSA data normalized by total amount of enzyme (T0). Left graphs, product formation; center graphs, dP/dt first differential of the adjacent left graph (∝[ET*]); right graphs, d2P/dt2 first differential of middle graph (∝dET*/dt). Insets in right show smoothed curve. Each curve is a twofold dilution of previous [enzyme] starting at 12.5 pmol enzyme (500 nM) and progressing red–orange–yellow–dark green–mid green–light green–blue. Curves in right graph (main panel) are for illustration only and are generated using average fit values of the middle graph. All data are truncated at 20% substrate conversion to avoid artifacts from non–Michaelis-Menten kinetics as substrate becomes limiting. Data shown represent two replicate experiments.

Normalizing the Aurora-A progress curves by total enzyme concentration (T0) revealed curves that overlaid well, indicating autophosphorylation by an intramolecular mechanism (Fig. 5B, left). Numerical differentiation of the progress curves revealed a slight lag phase (Fig. 5, center). The lag phase for both product formation and enzyme formation was shorter for Aurora-A than for Nek7 (compare Fig. 4, A and B, left, to Fig. 5, A and B, left), indicating that autophosphorylation of Aurora-A was more rapid than that of Nek7. We expected the signal-to-noise ratio on the doubly differentiated data (dET*/dt) to be poor (Fig. 5, right) because relatively few points defined the early part of the curve. However, our kinetic-based test does not require these data to be of high enough quality to fit. Instead, we need only to distinguish between data that follow a smooth decay curve throughout and data that contain a peak. The Aurora-A data revealed a peak with an apex at ~10 min (arrow in Fig. 5, A and B, right inset), which was visible in the raw data for the enzyme concentrations shown (especially the normalized data) and was also apparent in the smoothed normalized curves down to a 15 nM enzyme concentration. The peaks for the normalized graphs of dET*/dt were aligned (compare Figs. 3 and 5, right graphs), and so we concluded that Aurora-A autoactivates according to a three-state intramolecular mechanism. This is distinct from the intermolecular reaction by which Aurora-A phosphorylates other substrate proteins.

To extract the rate constants of Aurora-A autophosphorylation, we fitted both the product and dP/dt data sets to the appropriate analytical equations (Table 1). Agreement between the data sets was good, with the differences indicating the likely degree of precision of the measurements. Comparison of the steady-state rate obtained after Aurora-A autoactivation in vitro (Table 1) with that for enzyme autophosphorylated during expression in Escherichia coli (7) suggested that either the activated enzyme was less active or only a small population (~20%) of Aurora-A was autophosphorylated in vitro. Both mass spectrometry and Western blot analysis supported the latter hypothesis (fig. S4 and table S1).

Table 1 Fitted kinetic constants for Aurora-A autoactivation reaction.

Values are averages of three independent measurements (each a run of duplicate measurements) ± SEM. To fit the graphs of phosphorylated peptide product against time (Fig. 5A, left), the data were truncated at 20% substrate conversion and fit globally to Eq. 21 with T0 constrained for each data set. In fitting these data, the parameters α, k1, and k2 were shared between enzyme concentrations, and v′ allowed to float. Thus, each fit of the product accumulation data returned a single value of α, k1, and k2 constrained by all enzyme concentrations but multiple values of v′ (one for each concentration of enzyme). The reported value for v′ is the average ± SEM of values obtained. To fit the graphs of dP/dt (rate of product formation; Fig. 5A, center), the data were truncated at 20% substrate conversion and fit globally to Eq. 20 with T0 constrained for each data set. In these fits, α, k1, k2, and v′ were shared between data sets. We attribute the value of α to progress of the reaction before collection of the first data point (the dead time of the measurement). Our model does not allow us to assign an order to the rate constants k1 and k2 without a direct probe of the populations of E or E′, and therefore the numerical values of the rate constants k1 and k2 may be either of the values determined here.

View this table:

To confirm the unexpected three-state intramolecular mechanism for Aurora-A autophosphorylation, we analyzed Aurora-A activity with a DELFIA assay. Normalized curves for phosphorylated enzyme accumulation overlaid and showed an initial lag phase followed by an approximately linear reaction (fig. S5A, compare with Fig. 3, center). The time resolution was too low to permit meaningful differentiation. We did not detect any phosphorylation of inactive, unphosphorylated kinase [D274N mutant (5)] in the presence of a small quantity of wild-type, phosphorylated kinase, consistent with an intramolecular autophosphorylation mechanism (fig. S5B).

TPX2 enhances the intramolecular autoactivation reaction of Aurora A

In addition to phosphorylation, regulation of Aurora-A also occurs by the binding of protein partners, most notably TPX2, which stimulates the autophosphorylation of Aurora-A (11). We measured the autoactivation of Aurora-A in the presence of increasing concentrations of a fragment of TPX2 comprising residues 1 to 43 (Fig. 6A). The duration of lag phase appeared to decrease with increasing concentrations of TPX2, but further data processing was not informative because of the complexity of the mixture, in which three active species contribute to substrate turnover (7). To simplify analysis and enable normalization by enzyme concentration, we used a saturating concentration of TPX2 and varied the concentration of Aurora-A. This eliminated the lag phase entirely, indicating that saturating TPX2 increased the rate of the autoactivation reaction at least 20-fold (Fig. 6B). Autoactivation in the presence of TPX2 still proceeded by an intramolecular mechanism as indicated by the overlaid normalized product data (Fig. 6C).

Fig. 6 Protein binding partners stimulate autophosphorylation.

Reactions were measured by MSA. (A) Titrating TPX2 into a fixed concentration of Aurora-A (125 nM). TPX2 starting at 1.25 μM and decreasing through threefold serial dilutions dark blue–light blue. Open gray circles show Aurora-A alone. (B) Titrating Aurora-A into a fixed concentration of TPX2 (5 μM). Each curve is a twofold dilution of previous [enzyme] starting at 500 nM and progressing red-orange-yellow-green. (C) Curves in (B) normalized by enzyme concentration. Data shown in (A) to (C) represent two replicate experiments. (D) Titrating Nek9-CTD into a fixed concentration of Nek7 (2.5 μM). Nek9-CTD starting at 20 μM, proceeding with twofold serial dilutions from black, through dark blue to light blue. Open gray circles show Nek7 alone. Error bars are omitted for the sake of clarity. (E) Time taken to reach 20% substrate conversion, extracted from the data in (D). Individual data points from two replicate experiments are shown. (F) Representative progress curves from a single experiment for substrate phosphorylation by phosphorylated Nek7 starting at 325 nM and decreasing through twofold serial dilutions red-orange-yellow-green.

Nek9 enhances the intermolecular autoactivation reaction of Nek7

The noncatalytic C-terminal domain of Nek9 (Nek9-CTD) stimulates the activity of unphosphorylated Nek7 (8). Titrating Nek9-CTD into a fixed concentration of Nek7 (2.5 μM) initially reduced the duration of the lag phase (Fig. 6D), but as the concentration of Nek9-CTD exceeded 2.5 μM, this effect was reversed and the lag phase was prolonged (Fig. 6E). We also carried out the reciprocal experiment, using an excess of Nek9-CTD (20 μM) and varying the concentration of Nek7. Normalization of the progress curves showed that the activation mechanism of Nek7 was unchanged in the presence of Nek9-CTD and still occurred through a substrate-like intermolecular mechanism (fig. S6). Unlike TPX2 stimulation of phosphorylated Aurora-A (7), Nek9-CTD did not affect the activity of phosphorylated Nek7 (Fig. 6F and fig. S7).

Chk2 autophosphorylation is intramolecular, whereas Plk4 follows an intermolecular (substrate-like) mechanism

We also analyzed the autophosphorylation kinetics of two additional kinases, Chk2 and Plk4, which function in cell cycle arrest upon DNA damage and centriole duplication, respectively. Normalized time courses for Plk4 did not overlay, and the presence of a clear lag phase in the production of phosphorylated kinase (singly differentiated data, Fig. 7, A and B, center) indicated that this kinase follows an intermolecular, substrate-like autophosphorylation mechanism similar to that of Nek7. Conversely, normalized time courses for Chk2 indicated that autophosphorylation proceeded by an intramolecular mechanism; however, the wide spacing of time points with this particular substrate precluded further detailed analysis (Fig. 7C).

Fig. 7 Experimental progress curves for the Plk4 and Chk2 reactions measured by MSA.

(A) Experimental progress curves for the Plk4 reaction. (B) MSA data normalized by total amount of enzyme (T0). Left: product formation; center: first differential of left, dP/dt ∝ [ET*]; right: first differential of middle, d2P/dt2 ∝ dET*/dt. Each curve is a twofold dilution of previous [enzyme] starting at 600 pmol enzyme (20 μM) and progressing red–orange–yellow–dark green–mid green–light green. Data are truncated at 20% substrate conversion to avoid artifacts from non–Michaelis-Menten kinetics as substrate becomes limiting. Data shown represent two replicate experiments. (C) Experimental progress curves for the full-length Chk2 reaction. Left: measured curves of product formation; right: measured curves of product formation normalized by total amount of enzyme (T0). Each curve is a twofold dilution of previous [enzyme] starting at 0.48 pmol enzyme (16 nM) and progressing red–orange–yellow–dark green–mid green–light green. Data shown represent two replicate experiments.

A structural model of Aurora-A undergoing autophosphorylation shows that intramolecular phosphorylation is sterically possible

We are not aware of any crystal structures of an intramolecular phosphorylation state, and so we constructed a structural model of Aurora-A as it undergoes autophosphorylation to show that an intramolecular autophosphorylation conformation was sterically possible within a molecule displaying the structural features of an active kinase (27). We based our analysis on published crystal structures of Aurora-A in three of the four steps in the activation pathway and in which the activation loop is fully modeled [Protein Data Bank (PDB) codes 1OL5, 1OL7, and 4DEE] (Fig. 8, A to C) (5, 26). All three structures have similar overall conformations of the protein, including in the positions of active-site residues, such as Asp256 and Asp274, consistent with the observed catalytic activity in all three states (7). During autophosphorylation, Aurora-A must position Thr288 within the active site, in proximity to Asp256, for phospho-transfer to occur without disrupting any of the other conserved key active-site residues (27). It follows that the activation loop is the main, perhaps only, region of Aurora-A that changes conformation during the intramolecular autophosphorylation reaction.

Fig. 8 Models of autoactivation mechanisms for Aurora-A and Nek7.

(A) Crystal structures of unphosphorylated and phosphorylated Aurora-A (colored blue and yellow, respectively). (B) Crystal structures of unphosphorylated Aurora-A (blue) and phosphorylated Aurora-A–TPX2 complex (red). The superposition of Ser283 in unphosphorylated Aurora-A and Arg286 in phosphorylated Aurora-A–TPX2 complex is indicated with a dashed outline. (C) Model of Aurora-A undergoing autophosphorylation, in which the activation loop adopts the leftward position around Ser283 and the inward position around Thr288. The side chain of Thr288 is restrained within H-bond distance (black dashed line) of Asp256, the catalytic base. In (A) to (C), the arrows show the relative position of the activation loop around Ser283 (leftward or rightward) and Thr288 (inward, shown with an arrow pointing up, or outward, shown with an arrow pointing down). Substrate peptide is modeled in gray. (D) Cartoon representation of Nek7 autophosphorylation proceeding through an intermolecular substrate-like mechanism, which may be enhanced by increased enzyme concentration, dimerization, or oligomerization. This mechanism results in propagation of kinase activation by amplification. (E) Cartoon representation of Aurora-A autophosphorylation proceeding through a three-state intermolecular mechanism, which may be enhanced by protein partner binding (green TPX2 molecule). This mechanism results in control of kinase activation at the level of individual molecules. (F) Schematic illustrations of the activation loop conformations of Aurora-A in the four states defined in (E).

There is little change in the overall conformation of the activation loop between the structures of the unphosphorylated and fully phosphorylated kinase (Fig. 8A). However, there are conformational changes between the phosphorylated TPX2–Aurora-A complex and the phosphorylated kinase alone. In the presence of TPX2, the activation loop moves inward (toward the active site) in the vicinity of Thr288 and rightward (away from the active site) in the vicinity of Ser283 (Fig. 8B). In this conformation, Gly291 is close to Asp256, and both Ser283 and Arg286 have moved by ~9 Å (equivalent to shifting the register of the loop by three residues; gray dashed oval in Fig. 8B). We used this conformational change as a precedent that the activation loop of Aurora-A can undergo remodeling within an active kinase. If the same shift in loop register occurred in the opposite direction, Thr288 would be positioned very close to Asp256, poised for phosphorylation.

To build a structural model that captures Aurora-A at the point of autophosphorylation, we restrained Thr288 to be within H-bond distance of the catalytic base Asp256 and remodeled the activation loop between His280 and Thr292 by applying energy minimization within standard crystallography software (Fig. 8C). This model of autophosphorylation showed the activation loop adopting a conformation that combines features of both the TPX2-bound (red) and TPX2-free (blue and yellow) structures. It adopts the leftward position in the vicinity of Ser283 (similar to TPX2-free conformation) and the inward position around Arg286-Thr288-Gly291 (similar to TPX2-bound structure). This model has good geometry and, therefore, shows a scenario that is physically realistic. Both Arg255 and Arg286 found potential hydrogen bonding partners in the activation loop (to the backbone oxygen of Pro280 and to the side chain hydroxyl of Ser284).


Our kinetic test showed that Nek7 and Plk4 autoactivate through intermolecular mechanisms and that Aurora-A and Chk2 autoactivate through intramolecular mechanisms. These results raise the question of why the prevailing view has emerged that kinase self-activation is an intermolecular process. Intramolecular reactions are arguably more efficient and are common in enzyme self-activation (for example, in autoprocessing of proteases). Intramolecular reactions can be tightly controlled at the level of individual molecules, whereas intermolecular reactions propagate through amplification (Fig. 8, D and E). Parallels could be drawn with the activation mechanisms of zymogens in which the inhibitory pro-domain is cleaved either intramolecularly (for example, subtilisins) or by an upstream protease in a cascade (for example, blood clotting and apoptosis).

Studies using biochemistry, chemical biology, and structural modeling have shown that protein kinase A undergoes an intramolecular phosphorylation event within its C-terminal region (28). Examples of activation loop autophosphorylation events following an intramolecular mechanism had previously been limited to a group of CMGC family kinases that have an activation loop sequence that does not match the usual substrate consensus sequence of the kinase (2, 3). The two kinases that we showed have intramolecular autophosphorylation mechanisms that do not fit into this category: Aurora-A is related to AGC family kinases and has an activation loop sequence that does match its consensus; Chk2 is a CAMK family kinase and has an activation loop sequence that does not match its consensus. Intramolecular autophosphorylation may be more widespread than previously thought.

Implications for cellular regulation of Nek7 and Aurora-A

An intermolecular mechanism for self-activation implies that the activation of Nek7 could be regulated through concentration-dependent effects, such as localization to an intracellular structure or through dimerization or oligomerization (Fig. 8D). Indeed, we showed that the noncatalytic, dimeric, coiled-coil region of Nek9 stimulated the activity of unphosphorylated Nek7; however, at high concentrations of Nek9-CTD, when each dimer would be associated with one molecule of Nek7, the activation of Nek7 was retarded. Nek7 autophosphorylation appeared rather slow for a mechanism that should respond rapidly to cellular events, even at the relatively high concentrations of enzyme that we studied. This might indicate that another factor facilitates Nek7 autophosphorylation or that the physiological activation mechanism of Nek7 requires transphosphorylation by an upstream kinase, as has been shown for the related kinase Nek6 (16). Transphosphorylation could be an initiating event that triggers Nek7 self-activation. Future work on Nek7 will define the contribution of autophosphorylation to its cellular activity.

Aurora-A was previously assumed to follow a substrate-like autoactivation mechanism based, in part, on the observation that the sequence of its activation loop matches its substrate consensus sequence. Our analysis showed that autoactivation of Aurora-A followed a three-state intramolecular mechanism and that the rate of this reaction was increased by the binding of TPX2 (Fig. 8E). This intramolecular mechanism is obligatory for Aurora-A, and we were not able to detect any intermolecular phosphorylation between an active, phosphorylated kinase and folded, inactive, unphosphorylated Aurora-A molecules.

We detected a mismatch between the rates of substrate phosphorylation by Aurora-A phosphorylated upon expression in E. coli and Aurora-A autophosphorylated in vitro [(7) and Table 1]. This suggests that there may be a fourth state of Aurora-A (an inactive conformation incapable of autophosphorylation) that interconverts with the first three on a time scale too long to be measured in our experiments (>6 hours). In a cell, it follows that although a proportion of Aurora-A may rephosphorylate rapidly after the activity of cellular phosphatases (12), rephosphorylation of the entire pool of Aurora-A may be much slower.

The rates of autophosphorylation of Aurora-A and Nek7 were both increased by the binding of protein partners, apparently without any change in mechanism. This showed that even among the class of autophosphorylating kinases, the activity of some (for example, Nek7) will self-amplify, whereas the activity of others (for example, Aurora-A) cannot. The intramolecular mechanism of Aurora-A autophosphorylation may explain why self-association of Aurora-A at the mitotic spindle is insufficient for activation. The intrinsic autophosphorylation rate of Aurora-A is slow; otherwise, it would spontaneously activate, and activation is therefore under the control of binding of protein binding partners, such as TPX2, Ajuba, or Cep192 (11, 13, 29). It is likely that other mechanisms, such as phosphatase-mediated dephosphorylation, also prevent uncontrolled activation of Aurora-A (12). How these properties relate to the cellular function of the kinase and the balance of regulation between different protein binding partners and protein localization remain to be determined.

Structural models of protein kinase autoactivation

Although there are many crystal structures of protein kinases, our understanding of kinase autophosphorylation mechanisms at the structural level is poor. Crystal structures of several kinases show domain-swapped dimer conformations, in which the activation loop is unfolded from one molecule toward the active site of another molecule, providing a potential structural mechanism for reciprocal intermolecular autophosphorylation (1, 4). However, the observation of a domain-swapped dimer does not allow unambiguous assignment of the mechanism because Chk2 has been captured in crystal structures that suggest either substrate-like or domain-swapped dimer conformations, whereas our kinetic data indicated an intramolecular mechanism (1, 30). Moreover, Aurora-A adopts a domain-swapped dimer conformation in several drug-bound crystal structures (PDB codes 2BMC, 3DJ5, and 3DJ6) despite having an intramolecular activation mechanism. The conformations of the dimeric Aurora-A in these structures suggest that they represent inactive states of the kinase. We therefore advise caution in using crystal structures or activation loop sequences to assign autophosphorylation mechanisms and recommend using a kinetic test, such as the one we have described.

Our structural model of intramolecular autophosphorylation indicated that even relatively short activation loops, such as the one in Aurora-A, have sufficient plasticity to gain access to the active site and undergo intramolecular activation. Our structural model does not, however, explain why other kinases do not undergo intramolecular autophosphorylation. We speculate that these kinases might have very low activity in their unphosphorylated states compared to the basal level of activity observed in unphosphorylated Aurora-A (7). This notion is supported by published crystal structures of Aurora-A and Nek7 in their unphosphorylated states: Aurora-A closely resembles an active kinase [PDB code 4DEE (26)], whereas Nek7 has several features incompatible with kinase activity, such as a tyrosine residue that blocks the active site [PDB code 2WQM (8)]. Further kinetic and structural studies are required to answer this question fully.

Crystal structures of Aurora-A in different activation states suggested that the regulation of its catalytic activity by TPX2 and phosphorylation are due to lever arm–like motions of the activation loop (5). We have extended this model to include the intramolecular autophosphorylation mechanism of Aurora-A (Fig. 8F). The function of TPX2 in this model is to stabilize the inward motion of the lever arm, which promotes autophosphorylation when combined with a leftward motion of the lever arm (gray in Fig. 8F) and promotes substrate phosphorylation when combined with a rightward motion of the lever arm (red in Fig. 8F). Phosphorylation of Thr288 prevents a leftward motion of the activation loop, which is then locked in the inward, rightward position by TPX2, as shown using protein crystallography.

Automated enzyme assays as tools for mechanistic studies and drug discovery

Automated enzyme assay platforms generate data at rate of over 150 data points per minute, yielding at least 10,000 data points in a working day, and allow for increased time resolution (in our case the reaction was monitored at intervals of ~30 s), outstanding reproducibility, and excellent signal-to-noise ratios. Most published mechanistic assays are based on a limited exploration of the relevant variables, such as enzyme concentration and concentration of binding partners. However, the properties of automated analysis that enable collection of many data points with good time resolution and data quality also enable the mechanisms of kinase regulation to be explored in greater depth. Here, we outlined one application of this technology and devised an assay to explore an outstanding problem in kinase activation.

The autophosphorylation reaction regulates the activity of many protein kinases in the cell and provides a target reaction for distinct, selective, small-molecule inhibitors. Although the autoactivation reactions of Nek7 and Plk4 are analogous to a conventional bimolecular enzyme-substrate reaction, those for Aurora-A and Chk2 are a distinct reaction. We thus believe that it may be possible to design Aurora-A or Chk2 inhibitors that target this reaction specifically.

Materials and Methods

Protein expression

Unphosphorylated Aurora-A (residues 1 to 403 and 122 to 403), Nek7, and Plk4 (residues 1 to 269 with an N-terminal 6xHis tag) kinases were produced by coexpression with λ-phosphatase in E. coli as previously described for Nek2 (31). TPX2 (residues 1 to 43) and Aurora-A (residues 122 to 403) were purified as previously described (7). Nek7 was purified essentially as previously described, with the exception that we used a modified buffer for size exclusion chromatography (SEC), consisting of 50 mM Hepes (pH 7.5), 200 mM NaCl, 5 mM dithiothreitol (DTT), 5% glycerol.

The C-terminal domain of Nek9 (residues 732 to 979) was expressed in E. coli with an N-terminal glutathione S-transferase tag and initially purified on a GS-TRAP column (GE Healthcare) in a buffer consisting of 50 mM Hepes (pH 7.5), 150 mM NaCl, 1 mM DTT, 5% glycerol, supplemented with 10 mM reduced glutathione for elution. Protein was further purified with an S75 SEC column pre-equilibrated in 50 mM Hepes (pH 7.5), 20 mM NaCl, 5 mM DTT, 5% glycerol.

Plk4 was purified on a His-TRAP column (GE Healthcare) in a buffer consisting of 50 mM tris (pH 7.5), 300 mM NaCl, 5 mM MgCl2, 5% glycerol, 10 mM 2-mercaptoethanol, 30 mM imidazole and eluted on a gradient of 30 to 250 mM imidazole. The protein was further purified with an S200 SEC column pre-equilibrated in 20 mM tris (pH 7.5), 300 mM NaCl, 5 mM MgCl2, 5% glycerol, 1 mM EDTA, 5 mM 2-mercaptoethanol. Recombinant full-length Chk2 expressed in Spodoptera frugiperda insect cells was purchased from Calbiochem.

Kinase assays

Aurora-A MSA was performed with the indicated quantities of enzyme at room temperature in 50 mM tris (pH 7.5), 200 mM NaCl, 5 mM MgCl2, 10% glycerol, 10 mM 2-mercaptoethanol with 3 μM fluorescein-labeled kemptide (Flu-LRRASLG, Caliper peptide 21) and 1 mM ATP. The concentration of ATP was well above the Km for the enzyme, meaning that we could ignore the contribution of ATP binding to the general reaction scheme in Fig. 2A. Reactions were started by the addition of concentrated enzyme solution to a premix of the other reagents, and enzyme dilutions were obtained by serial dilution of the most concentrated mix. Reactions were followed in duplicate in 25-μl reaction volumes with a Caliper EZ Reader II system (Caliper Life Sciences,

Nek7 MSA was performed with the indicated concentrations of enzyme at room temperature in 100 mM Hepes (pH 7.5), 0.003% Brij-35, 0.004% Tween 20, 10 mM MgCl2 with 3 μM labeled peptide (5-FAM-FLAKSFGSPNRAYKK, Caliper peptide 32) and 1 mM ATP. Reactions were started as above and carried out in duplicate in 30-μl reaction volumes.

Plk4 MSA was performed with the indicated concentrations of enzyme at room temperature in 100 mM Hepes (pH 7.5), 1 mM DTT, 0.003% Brij-35, 0.004% Tween 20, 10 mM MgCl2 with 3 μM labeled peptide (5-FAM-FLAKSFGSPNRAYKK-CONH2, Caliper peptide 32) and 1 mM ATP. Reactions were as above in 30-μl reaction volumes.

Chk2 MSA was performed with the indicated concentrations of enzyme at room temperature in 40 mM Hepes (pH 7.5), 40 mM KCl, 10 mM DTT, 7 mM MgCl2, 0.02% Tween 20 with 3 μM labeled peptide (5-FAM-KKKVSRSGLYRSPSMPENLNRPR-COOH, Caliper peptide 10) and 1 mM ATP. Reactions were carried out as described above.

DELFIA was performed in a 400-μl master reaction for each measurement containing the stated amount of enzyme and 1 mM ATP [and, in reactions monitoring intermolecular autophosphorylation reactions, 4.4 μM Aurora-A with a mutation in the active site (Asp274 to Asn) to render it catalytically inactive] in a 1-ml deep-well block in 50 mM tris (pH 7.5), 200 mM NaCl, 5 mM MgCl2, 10% glycerol, 10 mM 2-mercaptoethanol. At the indicated time points, 30 μl of reaction was removed from the master reaction and quenched in an equal volume of 40 mM EDTA in an Immulon high binding plate. The completed reaction was sealed and incubated overnight at 4°C before being developed with an antibody that recognizes pThr288 Aurora-A (New England Biolabs) diluted 1:2000 in branded assay buffer (PerkinElmer), Europium-chelating 2° (PerkinElmer), and branded enhancement solution (PerkinElmer).

Data are presented as either individual points for each experiment or as the midpoint of the data values for two independent experiments performed in duplicate, with bars representing the range of the original values.

Structural modeling

The structure of Aurora-A undergoing autophosphorylation was initially modeled with Coot (32) to manually relocate and rebuild the activation loop. The template used was PDB code 1OL7, with the activation loop between Ser283 and Thr288 based on PDB code 1OL5, and guided by the structure of phosphorylase kinase bound to a substrate peptide (PDB code 2PHK). The loop between Thr288 and Thr292 was rebuilt manually, guided by structures of Aurora-A in inactive conformations, such as PDB codes 1MUO and 2WTV. The side chain orientations of His176 and Arg180 were changed to resemble those found in 1OL5 and to generate space for the side chain of Arg286, which lies in a similar position to the arginine residue in the P+2 position of the substrate peptide in 2PHK. The resultant model was subjected to iterative rounds of geometry and minimization with PHENIX (33) and rebuilding with Coot. With PHENIX, the kinase structure was fixed except for activation segment residues 277 to 293, and the distance between atom OG1 of Thr288 and atom OD2 of Asp256 was constrained to an ideal distance of 3.0 Å, with a σ of 0.05. Geometry checking was performed with MolProbity, revealing overall good geometry with no Ramachandran outliers generated by the modeling (Arg304 is an outlier in the template used).

Numerical differentiation

This was carried out with a moving two-point gradient. dP/dt was calculated from the product progress curves, d2P/dt2 from dP/dt. The value of the gradient was plotted at the average x value for the pair. Curve smoothing for the insets in Fig. 5, right, was generated with the zero-order polynomial, two-point average in-built function in Prism ( We were cautious about using this function because we discovered that changing the smoothing window could introduce an artifactual phase in the first differential of perfect simulated intramolecular data. However, the impact of the smoothing window on the second differential of simulated data was to affect the gradient of the decay curve rather than to introduce an artifactual peak, giving us confidence in our limited use of it in this work.

General definitions for kinetic schemes

For the scheme in Fig. 2A, the total concentration of enzyme is defined as follows:ET*=[E*]+[E*S](1)T0=ET*+ET=[E*]+[E*S]+[E]=E0*+E0(2)where ET* is the total concentration of phosphorylated enzyme at a given time, E0* is the initial concentration of phosphorylated enzyme, and T0 is the total concentration of enzyme present in the reaction (irrespective of phosphorylation state).

We expect E0* to be a proportion of the total amount of enzyme, not a fixed amount at all enzyme concentrations. In the derivations here, we used E0*; however, for data fitting and modeling, we expressed this as a fractional population by making the substitution α=E0*T0 throughout.

The Km of phosphorylated enzyme for substrate can be written and then rearranged using Eq. 1 as follows:Km=[E*][S][E*S][E*S]=ET*[S]Km+[S](3)[E*]=KmET*Km+[S](4)

Kinetic scheme: Two-state intramolecular mechanism

Ek0E* (Fig. 2B)

Until substrate becomes limiting, the rate of formation of phosphorylated enzyme can be expressed asdET*dt=k0[E]=k0(T0ET*)(5)

Integrating gives the total amount of phosphorylated enzyme present at any time during the reaction: ET*=T0(T0E0*)ek0t(6)

From (3) and (6), the rate of product (phosphorylated peptide) formation isdPdt=kcat[E*S]=kcat[S]ET*Km+[S]=kcat[S][T0(T0E0*)ek0t]Km+[S](7)

Integrating (7) gives the concentration of product at any time:P=kcat[S]Km+[S][T0t+1k0(T0E0*)(ek0t1)]=v[T0t+1k0(T0E0*)(ek0t1)](8)where v=kcat[S]Km+[S] and is effectively the rate constant for the phosphorylation of peptide substrate by phosphorylated enzyme.

Kinetic scheme: Two-state intermolecular mechanism (reciprocal autophosphorylation)

E+Ek0E*+E* (Fig. 2D)

Until substrate becomes limiting, the rate of formation of phosphorylated enzyme is dET*dt=2k0[E][E]=2k0(T0ET*)2(9)

By integration, the concentration of phosphorylated enzyme at any time during the reaction is ET*=T0T0E0*2k0t(T0E0*)+1(10)

By analogy to (7) and integrating, the concentration of product throughout the reaction is as follows:P=v(T0t12k0ln(2k0t(T0E0*)+1))(11)where v=kcat[S]Km+[S].

Kinetic scheme: Two-state intermolecular mechanism (substrate-like autophosphorylation)

E+E*k0E*+E* (Fig. 2E)

Using a similar approach to that of Wu and Wang (22) and substituting from Eq. 4, until substrate becomes limiting, the rate of formation of the phosphorylated enzyme can be expressed as follows: dET*dt=k0[E*][E]=k0KmKm+[S]ET*[E]=kobsET*(T0ET*)(12)where kobs=k0KmKm+[S] and is the rate constant for the appearance of phosphorylated enzyme.

Integrating (12) gives an expression for the total amount of phosphorylated enzyme present:ET*=E0*T0E0*+(T0E0*)ekobsT0t(13)

Again, using a similar approach to that of Wu and Wang (22), the rate of product formation isdPdt=kcat[E*S]=k2[S]ET*Km+[S]=kcat[S]E0*T0(Km+[S]){E0*+(T0E0*)ekobsT0t}(14)

Integrating (14) gives P, the concentration of phosphorylated peptide product at any time during the reaction: P=vkobsln[1E0*T0(1ekobsT0t)](15)where v=kcat[S]Km+[S].

Kinetic scheme: Three-state intramolecular mechanism

Ek1Ek2E* (Fig. 2C)

Working from left to right,d[E]dt=k1[E][E]=(T0E0*)ek1t(16)d[E]dt=k1[E]k2[E][E]=k1(T0E0*)k2k1(ek1tek2t)(17)

using the initial condition that t = 0, [E′] = 0 in Eq. 17 [if EE′ progressed in the absence of ATP, then we would instead be monitoring E′→E* (a two-state reaction) in our experiments].dET*dt=k2[E]=k1k2(T0E0*)k2k1(ek1tek2t)(18)

Integrating (18) givesET*=T0E0*k2k1[k1(ek2t1)k2(ek1t1)]+E0*(19)

which is symmetric in k1 and k2.

As previously,dPdt=kcat[E*S]=k2[S]ET*Km+[S](20)

Integrating for a final time,P=v(T0E0*)k1k2(k2k1)[k22(ek1t1)k12(ek2t1)]+vT0t(21)again where v=kcat[S]Km+[S]. Like Eqs. 18 and 19, Eq. 21 is also symmetric in k1 and k2.

The case of a three-state intramolecular reaction in which an equilibrium exists between the first two states (when E and E′ are in equilibrium) is considered explicitly in text S3.

Supplementary Materials

Text S1. Assumptions for kinetic analysis.

Text S2. Multiple mechanisms.

Text S3. More complicated kinetic mechanisms.

Text S4. Other kinetic tests.

Fig. S1. Activation of Aurora-A in vitro with different preincubation times.

Fig. S2. x-intercepts and induction times of progress curves.

Fig. S3. Experimental progress curves for full-length Aurora-A reaction measured by MSA.

Fig. S4. Aurora-A autophosphorylation is incomplete in vitro.

Fig. S5. Activation of Aurora-A measured by DELFIA assay.

Fig. S6. Experimental progress curves for Nek7 reaction in the presence of excess Nek9-CTD measured by MSA.

Fig. S7. Experimental progress curves for phosphorylated Nek7 reaction in the presence of Nek9-CTD measured by MSA.

Table S1. Mass spectrometry indicating the degree of phosphorylation in the sequence region encompassing Thr288.

References and Notes

Acknowledgments: We thank our former colleagues at the Institute of Cancer Research (ICR)—A. Paul and A. Thompson (mass spectrometry); J. Schmitt, K. Boxall, W. Aherne, R. Burke, and R. van Montfort (Caliper EZ Reader). K. Kulkani (ICR) and J. Dodson (Element Six Ltd.) provided invaluable mathematical advice. Funding: This work was supported by grants to R.B. from Cancer Research UK (C24461/A8032 and C24461/A12772) and a Royal Society University Research Fellowship. Author contributions: C.A.D., S.Y., and R.B. designed the research and wrote the paper. C.A.D., S.Y., T.H., and R.B. purified proteins and contributed to editing of the manuscript. C.A.D. and S.Y. carried out MSA assays and analysis. C.A.D. derived kinetic equations and carried out DELFIA experiments. R.B. carried out structural modeling. Competing interests: The authors declare that they have no competing interests.
View Abstract

Stay Connected to Science Signaling

Navigate This Article