Visualisation of a kinesin-13 motor on microtubule end mimics

Summary

An expanding collection of proteins localises to microtubule ends to regulate cytoskeletal dynamics and architecture by unknown molecular mechanism(s). Electron microscopy is invaluable for studying microtubule structure, but because microtubule ends are heterogeneous, their structures are difficult to determine. We therefore investigated whether tubulin oligomers induced by the drug dolastatin could mimic microtubule ends. The microtubule end-dependent ATPase of kinesin-13 motors is coupled to microtubule depolymerisation. Significantly, kinesin-13 motor ATPase activity is stimulated by dolastatin-tubulin oligomers suggesting, first, that these oligomers share properties with microtubule ends and, second, that the physical presence of an end is less important than terminal tubulin flexibility for microtubule end recognition by the kinesin-13 motor. Using electron microscopy, we visualised the kinesin-13 motor-dolastatin-tubulin oligomer interaction in nucleotide states mimicking steps in the ATPase cycle. This enabled us to detect conformational changes that the motor undergoes during depolymerisation. Our data suggest that such tubulin oligomers can be used to examine other microtubule end-binding proteins.

Microtubules serve as tracks for molecular motors and are essential for many aspects of cellular function. They are dynamic polymers and the control of growth and shrinkage at their ends is crucial for cytoskeletal regulation. A variety of proteins localise and act at microtubule ends^1,2 so that a key question about microtubule dynamics is by what mechanism(s) do these proteins recognise and regulate microtubule ends? Such mechanistic questions can be usefully addressed by understanding the structures of the components and the ways in which they interact.

Among the proteins that control microtubule dynamics are the kinesin-13 family of microtubule depolymerising motors. Members of this family are important components of mitotic and meiotic spindles, and they also regulate interphase microtubules³. How do kinesin-13 motors depolymerise microtubules at their ends? The fundamental step of microtubule depolymerisation involves an ATP-induced deformation of the terminal tubulin dimer by the kinesin-13 motor domain, pulling the dimer away from its lattice position^4,5. The minimal kinesin-13 motor core is an inefficient depolymeriser compared to full-length, dimeric kinesin-13^6-8, but it shares many properties with longer, more efficient kinesin-13 constructs, making it a simple but useful model for the entire family. An important property shared by the motor core and larger constructs is the selective stimulation of the motor ATPase by microtubule ends compared to the microtubule lattice, thereby coupling depolymerisation to ATP turnover.

Because of the size and complexity of microtubules, electron microscopy (EM) is the method of choice for their structural study, and computational analysis of noisy EM images has enabled elucidation of many aspects of microtubule biology⁹. Medium- to high-resolution structure determination by EM relies on averaging data from many nominally identical entities¹⁰. However, the ends of microtubules are heterogeneous¹¹, so that their images cannot be averaged and their structures cannot be calculated using these methods. Tubulin dimers at microtubule ends probably have greater flexibility than dimers constrained within the microtubule lattice and recognition of this property is postulated to be one mechanism by which microtubule ends are discriminated. Furthermore, the crystal structures of kinesin-13 motor domains suggest that the motor surface is more complementary to the predicted curvature of microtubule ends than to the straight tubulin conformation found in the microtubule lattice^12,13. Strikingly, dimeric tubulin, which is presumably very flexible, stimulates the kinesin-13 ATPase activity in all constructs examined^7,8,14. However, for structural purposes, individual tubulin dimers are too small to visualise by EM.

A number of chemically distinct small molecules are known to perturb the microtubule cytoskeleton by disrupting polymerisation, a property that has made these compounds successful chemotherapeutic agents^15,16. Several of these molecules stabilise tubulin oligomers and we reasoned that such oligomers might facilitate both our experimental goals of a) mimicking flexibility at microtubule ends while b) being large and regular enough to allow analysis of EM images. Dolastatin-10, a peptide derived from the Indian Ocean sea hare Dolabella auricularia (Figure 1a), generates curved tubulin oligomers with an average radial curvature of ~220Å^17,18 that closely matches that adopted by unliganded tubulin^19,20. A variety of biochemical and modelling experiments^21,22 suggest that the dolastatin binding site is located at the so-called peptide site of the tubulin dimer, on the longitudinal inter-dimer interface and close to the nucleotide binding site on β-tubulin (Figure 1b). This site overlaps, but is not identical to, the vinblastine binding site²³.

Figure 1. Dolastatin-10 induces tubulin oligomers that mimic the flexible properties of microtubule ends.

(a) Diagram of the pentapeptide dolastation-10. Previous studies have suggested that the consecutive valine, dolaisoleucine and dolaproine residues are the most important features for binding to tubulin²². This likely accounts for the reduced tubulin affinity of dolastatin-15, a closely related seven subunit peptide from the same organism that lacks these key residues³², and our observation of its failure to induce tubulin oligomers (data not shown). (b) i) View of β-tubulin from the microtubule plus-end; residues predicted to bind dolastatin-10 mainly via hydrophobic interaction are shown as space-filling representations in light blue²¹,²² and the bound GDP shown in space-filling representation in red/green/blue. The tubulin C-terminal α-helices H11 and H12, the principal binding site of kinesin, are shown in yellow. Coordinates from 1SAO.pdb were used³³; ii) View of the αβ-tubulin heterodimer from the side and relationship of the predicted dolastation-10 binding site with respect to the microtubule lattice; iii) Schematic of dolastatin-10 induced tubulin oligomers built through protofilament-like longitudinal tubulin contacts and viewed from the side. (c) Electron micrograph of dolastatin-10 induced tubulin oligomers, visualised by negative stain. Bar = 400Å. The inset shows the image of an individual ring (boxed) that has been band-pass filtered to reveal more clearly the single band of tubulin density from which the rings are built. Dolastatin-10, dissolved in DMSO, was incubated with tubulin (purchased from Cytoskeleton, Inc (Denver, CO)) at a molar ratio of 2:1 in polymerisation buffer (40mM Pipes, pH 6.8, 1.5mM MgCl₂, 12% DMSO for 1 hour at room temperature. Oligomers were diluted in BrB₂₀ (20mM Pipes, 2mM MgCl₂, 1mM EGTA, pH6.8) prior to use. 1 μM tubulin-dolastatin rings were applied to home-made, continuous carbon EM grids glow discharged in the presence of amylamine. The rings were negatively stained using 1% uranyl acetate. Images of the dolastatin rings were collected on SO-163 photographic film on a Phillips CM120 electron microscope operating at 100kV under low dose conditions and at a nominal magnification of 45,000x. Micrographs were digitised using a Zeiss SCAI microdensitometer with a final pixel size at the sample of 3.1126Å/pxl (calibrated using TMV). (d) Stimulation of pKinI ATPase by dolastatin-10 induced tubulin oligomers. V_max=0.88μM/s; K_m=0.31μM; n=2-4 for each data point; error bars = standard deviation. pKinI was prepared as previously described and dialysed against BrB₂₀/1mM DTT prior to use⁵. The ATPase activity of pKinI (0.25μM) at 25°C was measured using an NADH-coupled system as described previously⁷. (e) Representative image averages of dolastation-10 tubulin rings; each average contains, respectively, 93, 91, 111 and 114 individual images. Bar = 100Å. For image analysis, dolastatin-tubulin rings were selected manually using the MRC program Ximdisp³⁴, their defocus was calculated using the MRC program CTFFIND2 and phases were corrected for the contrast transfer function using SPIDER³⁵. Individual images were band-pass filtered, normalised and centred using SPIDER to a rotational average of the previously calculated pKinI-induced tubulin ring⁵. Centred images were then subjected to multivariate statistical analysis (MSA) and classification in IMAGIC³⁶. Selected class averages were used iteratively for further rounds of multi-reference alignment until the outcome of MSA had stabilised (~2 additional rounds of alignment).

In order to understand the kinesin-13 mechanism, it is essential to understand the motor’s interaction with microtubule ends. We therefore used the well-studied motor core of kinesin-13 from the malaria parasite (which we call pKinI^5,7,13) to investigate the ability of dolastatin-induced tubulin oligomers to mimic microtubule ends.

Flexible tubulin oligomers mimic microtubule ends

When tubulin is incubated with dolastatin-10, single protofilaments form curved oligomers such as rings and spirals (Figure 1c). Crucially for our purposes, dolastatin-10-induced oligomers stimulate the ATPase activity of the kinesin-13 minimal motor domain pKinI to the same extent as tubulin dimers and microtubule ends, and in contrast to the inhibition of the pKinI ATPase on binding to the microtubule lattice (Figure 1d, V_max for tubulin-dolastatin oligomers = 0.88μM/s compared to 0.78 μM/s for tubulin heterodimers and 0.63 μM/s for sheared microtubules⁷). This observation is noteworthy because it suggests that it is the relative flexibility of microtubule ends that is the property by which the kinesin-13 motor ATPase is activated, rather than by end-specific surfaces on the terminal tubulins. Tubulin surfaces buried by lateral lattice contacts are more likely to be accessible at microtubule ends and are also present in the dolastatin-rings. Such surfaces might contribute to end recognition mechanisms, but because we are working with a minimal kinesin motor which is well established to interact with the front face of the tubulin dimer, we favour flexibility as the most likely contributing factor to pKinI ATPase activation. Since the plus- and minus-end surfaces of microtubules are structurally very different, this is also consistent with the ability of kinesin-13 motors to depolymerise microtubules at both ends⁴.

Dolastatin-10 tubulin rings imaged by negative stain EM (Figure 1c) were processed computationally to allow rings with similar appearances to be averaged together, allowing visualisation of reproducible structural details. The raw images show that the rings are flexible, and this variability of curvature is reflected in the calculated averages (Figure 1e). The class averages also show that there are 3 main geometries present in the ring population – 13x, 14x and 15x tubulin dimers, with 14x being the most prevalent as previously described^17,18. As expected from the resolution of these images, α- and β-tubulin cannot be discriminated, nor do the structural details around the rings show repetitive variation that would allow intra- and inter-dimer interfaces to be identified. This suggests that intra- and inter-dimer curvatures are not correlated (or that differences are undetectable) in these images. This is in contrast to the tubulin oligomers induced by the peptide cryptophycin²⁴, demonstrating the variety of tubulin oligomeric forms induced by peptides binding at very similar sites on β-tubulin. The similarity of dolastatin-tubulin and unliganded tubulin oligomer curvatures, as well as their ability to stimulate pKinI ATPase, encouraged us to examine the interaction between pKinI and dolastatin rings more closely. Using EM and image analysis of the pKinI-ring complexes, we dissected the nucleotide-dependent interaction of the motor with these microtubule end mimics.

Nucleotide-dependent interaction of pKinI with rings visualised by EM

Raw images of pKinI-dolastatin-10-tubulin rings show the presence of additional protein density on the inside of the tubulin rings (corresponding to the microtubule outside surface, Figure 1b, Figure 2a). Dolastatin-tubulin rings also appear less flexible in the presence of pKinI, suggesting that at least some of the ring flexibility is constrained by motor binding (compare Figure 1c with Figure 2a). Despite this, the pKinI-bound dolastatin-10-tubulin rings were not perfectly circular, suggesting the presence of some residual flexibility at the inter-dimer contacts where dolastatin-10 is presumed to bind. We examined pKinI-ring complexes in the presence of ADP, no nucleotide, AMPPNP (a non-hydrolysable ATP analogue) and ADP.AlF₄ (a transition state analogue, mimicking ADP.Pi), thereby capturing different stages of the pKinI ATPase cycle. Representative averages for each nucleotide state are shown in Figure 2b. In each, density corresponding to the pKinI motor associates with 2 tubulin densities, as is typical for kinesin motors. We observed good pKinI binding in all nucleotide states apart from ADP, where the motor density is more poorly defined. In all cases, 13x, 14x and 15x geometries were observed (Supplementary Figure 1a) but the 14x geometry provided the majority of the data for each nucleotide state (an average of 63% of the data compared to 14% and 23% for 13x and 15x geometries respectively).

Figure 2. Nucleotide-dependent interaction of pKinI with dolastation-10 induced tubulin rings.

(a) Electron micrograph of pKinI-AMPPNP bound to dolastatin-10 tubulin rings. Bar = 400Å. The inset shows the image of an individual ring (boxed) that has been band-pass filtered to reveal more clearly the double band of density from which the rings are built; the inner layer corresponds to pKinI which binds to the outer ring of tubulin. The background on this micrograph is high compared to Figure 1c due to the presence of unbound pKinI. 1 μM tubulin-dolastatin rings were incubated with 5μM pKinI and 1mM nucleotide or nucleotide analogue (ADP, AMPPNP, ADP.AlF₄, or in the absence of nucleotide) for 5 minutes at room temperature and these mixtures were prepared and imaged by EM as described in the legend to Figure 1. (b) Representative image averages of pKinI-dolastatin-10 tubulin rings with 14x geometry in different nucleotide states; each average contains, respectively, 102, 187, 193 and 84 individual images and was calculated as described in the legend to Figure 1. The ring averages show a marked handedness, demonstrated by the skew of the tubulin in the outside ring and direction in which the pKinI density faces on the inner ring; this handedness is clockwise for the rings in the absence of nucleotide and in the presence of AMPPNP, and counter-clockwise for ADP and ADP.AlF4 and is a reflection of the way in which the rings fall on the grid and of the fact that each nucleotide data set was processed independently. Bar = 200Å.

Nucleotide-dependent conformational changes by pKinI bound to tubulin rings

To further probe the nucleotide-dependent conformational state of pKinI on the tubulin rings, we wanted to combine data of the motor-tubulin complex from the individual 14x averages and around each ring. Since the rings are built from nominally identical pKinI-tubulin dimer repeats, further averaging of these individual building blocks should enable additional structural details to be extracted. However, simple rotational averaging results in degradation of the average because the rings are not perfectly circular and such rotational averaging was never imposed (Supplementary Figure 1). Instead, we selected the pKinI-tubulin dimer units from the best 14x class averages for each nucleotide state (treating each motor-tubulin dimer repeat in the ring averages as a single particle) and aligned and averaged these together to create a pKinI-tubulin dimer image for each nucleotide state. These averages demonstrate that there are only very small changes in the tubulin in the different nucleotide states, but that pKinI passes through a series of distinct conformational states during its ATPase cycle (Figure 3a, b). To understand these pKinI states in more detail, we used the Student’s t-test to calculate the statistically significant differences between nucleotide states of pKinI (Figure 3c²⁵). As data shown in these averages are moderate resolution 2D projections of the 3D structure of the motor-tubulin complex, we have interpreted the nucleotide-dependent differences we observe conservatively. To facilitate this interpretation and provide an overview of the changes occurring in the motor domain during its ATPase cycle at microtubule ends, we generated an atomic model of the pKinI-tubulin interaction based on available structural information (Figure 3d). The pKinI interaction with tubulin is centred on helix-α4, the nucleotide sensing, switch II relay helix (shown in yellow²⁶), which sits above the tubulin intra-dimer interface. Towards the plus-end, residues in loop12/α5 and loop 8 of the motor lie close to the β-tubulin surface, while at the minus-end, helix α6 and loop 2, containing residues essential for depolymerisation are close to α-tubulin (shown in red^12,13).

Figure 3. Conformational changes in pKinI during its ATPase cycle.

(a) Averages of the pKinI-tubulin dimer repeat for each nucleotide state. These were calculated as follows: following calculation of the pKinI-dolastatin-tubulin-ring class averages, the best classes were evaluated for homogeneity by MSA and also by visual inspection of their constituent members prior to further processing. At no point was 14-fold rotational symmetry imposed. From the best 14X averages for each nucleotide state, individual pKinI-tubulin dimer complexes were picked manually and were aligned. These steps are illustrated in Supplementary Figure 1c. For the ADP, no nucleotide, AMPPNP and ADP.AlF4 states, 4, 5, 7 and 7 class averages respectively were used, generating respectively 40, 70, 98 and 98 individual pKinI-tubulin dimer units. ADP rings were incompletely decorated so only segments with an obvious motor density were selected. An average and variance map for each nucleotide state was calculated and the statistically significant differences between states, representing nucleotide-dependent conformational changes, were evaluated pixel by pixel using Student’s t tests²⁵. To confirm the validity of these observations, the statistically significant differences between half data-sets of the same nucleotide state were examined and in all cases were found to be minimal in comparison to differences between nucleotide states. Differences between the nucleotide states calculated using the t-test analysis were confirmed using MSA in IMAGIC. (b) Schematic of the kinesin-13 ATPase cycle at the microtubule end; the steps are numbered according to the corresponding panels in Figure3C. (c) Evaluation of the statistically significant differences between consecutive nucleotide states in the ATPase cycle (indicated beneath) representing conformational changes in the motor. The right hand column shows the output of the Student t-test, contoured at a probability of <0.005. The positions of these peaks of statistically significant difference are indicated on the relevant image averages with red circles. (d) Model of the pKinI-tubulin interaction obtained by docking pKinI (1RY6.pdb¹³) onto GDP-tubulin (1SA0.pdb³³). This docking was performed using Kif1a-tubulin coordinates (2HXH.pdb²⁸) as a template in Pymol (www.pymol.org) and is consistent with other kinesin-tubulin structures⁷,¹²,³⁷. The pKinI-GDP-tubulin atomic model is presented so that it matches the EM projection maps as closely as possible. pKinI is shown in blue and the GDP-tubulin dimer is shown in green; a space filling model of ADP indicates the kinesin nucleotide binding site, although no nucleotide was found in the pKinI crystal structure. The pKinI relay helix, α4, is coloured yellow and sits at the tubulin intra-dimer interface; this region of pKinI appears to be invariant during ATP binding and hydrolysis steps. Other portions of pKinI likely to be involved in conformational changes during the motor’s ATPase cycle are depicted in red and include loop 2 (L2) which contains kinesin-13 depolymerisation-specific residues; helix 6 (α6); loop 12/helix 5 (L12/α5) and loop 8 (L8).

On binding of pKinI to tubulin, ADP is released from the motor and according to our data, this is coupled to conformational changes in pKinI that are spread across most of the motor domain. The overall shape of the t-map in Figure 3ci resembles that of the motor domain, reflecting the incomplete occupancy of pKinI-ADP binding and suggesting that in the absence of nucleotide, pKinI sits more firmly on the terminal tubulin following ADP dissociation. This is in contrast to the localised conformational change we identified in pKinI when it releases ADP on interaction with the microtubule lattice⁷. The more widespread changes we observe when pKinI binds to dolastatin-tubulin rings likely reflect engagement of the entire kinesin-13 motor domain with its tubulin substrate in preparation for ATP binding and hydrolysis, neither of which can occur on the non-flexible microtubule wall⁷.

We previously described induction of curled protofilaments by pKinI-AMPPNP from the ends of stabilised microtubules that likely represent a key step in microtubule depolymerisation⁵. In our current experiments, we are observing pKinI on curved tubulin that is longitudinally constrained by the ring, so we do not expect to see additional curving in the tubulin when AMPPNP (mimicking ATP) binds to pKinI. Nevertheless, on AMPPNP binding, pKinI apparently undergoes concerted conformational changes, with alterations at the pKinI-tubulin interface dominating the t-map (Figure 3cii). One cluster of statistically significant differences is located towards the motor’s plus-end in regions that correspond to important elements of kinesin’s energy transduction system including helix-α5/loop 12 and loop 8. Loop 12 is conserved in all kinesins and is thought to be important in kinesin-microtubule interactions²⁷ while nearby loop 8 has been proposed to act as a sensor for regulation of ATP hydrolysis in depolymerising kinesins¹². At the motor’s minus-end, there are also conformational changes in pKinI close to the surface of tubulin – these probably involve α6 and the kinesin-13-specific loop 2 region. It seems likely that the extended conformation of loop 2 seen in the crystal structure will alter when pKinI is bound to its tubulin substrate. These clusters of conformational changes occur at either end of the motor on either side of an apparently static anchor point that is likely to be formed, at least in part, by helix-α4. It is interesting to note that this mirrors the behaviour of the functionally very different kinesin KIF1A, in which the kinesin motor domain was observed to rearrange itself around its fixed relay helix during its ATPase cycle²⁸.

Similar regions of the motor appear to alter their conformations when ATP is hydrolysed by pKinI (Figure 3ciii). However, while major changes near the tubulin surface were observed on ATP binding, changes in the body of the motor itself are more significant on ATP hydrolysis. It has previously been postulated that the hydrolysis step of the kinesin-13 ATPase cycle is important for release of the terminal tubulin from the microtubule end⁴. In the context of the dolastatin rings, such release is probably prevented because the tubulin to which pKinI is attached is constrained at both ends. Our data is consistent with this idea and is further supported by the wide-spread differences in pKinI when comparing the ADP.Pi and ADP states, suggesting that release of pKinI from its tubulin substrate is coupled to phosphate release (Figure 3civ).

Our data have revealed the dynamic response of the kinesin-13 motor domain that allows coupling of ATP turnover to depolymerisation at microtubule ends. What might we predict for larger kinesin-13 constructs including the native dimer? The kinesin-13 neck sequence, not present in our pKinI construct, is essential for efficient motor activity, although the precise mechanism by which it acts is unknown⁶. In a recently solved crystal structure (2gry.pdb; http://www.rcsb.org/pdb), the proximal part of the kinesin-13 neck helix (the only part of the neck that is visible) lies directly above loop 2 (its position is indicated by the asterisk in Figure 3d). In the light of our data, one possible mechanism by which the neck sequence could be involved in motor efficiency is by direct coupling of the neck to conformational changes in the motor core during the ATP-binding and –hydrolysis steps, perhaps enabling amplification of these conformational changes, and transmission to adjacent protofilaments of the microtubule²⁹. The contributions of the rest of kinesin-13 protein to and the significance of dimerisation for the depolymerisation mechanism are less clear. Further experiments will aim to probe these questions and to establish the role of the actual microtubule end in depolymerisation efficiency.

Cell biology studies have told us a great deal about the location and colocalisation of a large number of proteins at the dynamic ends of microtubules², but we know very little about the detailed molecular mechanisms by which microtubule ends are identified. Our ability to observe the activity and conformational changes of an end-specific motor using dolastatin-induced rings has provided insight into important properties of microtubule ends that distinguish them from the bulk microtubule cytoskeleton. These rings could be extremely useful in discriminating between different binding mechanisms of the many proteins that are active at microtubule ends, and to visualise the variety of protein-protein interactions that are found there^30,31.