Structural basis of human kinesin-8 function and inhibition

The complex machinery of the mitotic spindle is built from discrete populations of MTs. Each population is tightly regulated to perform specific functions and to collectively ensure the efficiency and accuracy of cell division. Kinesin-8s move processively along specific spindle MTs to reach their plus-ends and, on arrival, control MT plus-end dynamics; for example, kinetochore MTs are regulated by Kif18A to enable chromosome alignment before metaphase (11, 12). The dual activities of kinesin-8s—plus-end–directed stepping and ability to modulate MT dynamics—are encoded within the kinesin dimer, with the tail domain enhancing motor processivity, MT cross-linking activities, as well as MT end occupancy, thereby making significant contributions to motor function (13, 21, 24, 25, 47). However, our current structural and functional studies, together with previous cell biological work (43), also demonstrate that specific nucleotide-dependent conformational changes within the Kif18A monomeric motor domain are crucial for canonical plus-end stepping and for modulation of MT end dynamics.

We show that the major structural rearrangements within the Kif18A motor domain that drive plus-end–directed motility occur in the transition between the MT-bound NN and ATP-like states (Fig. 2). The closure of the nucleotide-binding site brings the surrounding conserved loops into a catalytically competent conformation, and induces extension of helix-α6 and reorientation of the neck linker toward the MT plus-end. Based on a previous, lower resolution Kif18A_MD reconstruction (29), we concluded that helix-α4 is bent in the Kif18A-MT complex. However, with the improved resolution of our current reconstructions, we can clearly see density corresponding to an extended helix-α4 in all nucleotide states; thus, the earlier conclusion arose owing to the limitations of the resolution. The conformational changes that we now see in our Kif18A_MDNL reconstructions are structurally conserved in multiple families of plus-end directed kinesins (31, 32, 39, 48). High-resolution structures of tubulin-bound kinesin-1 revealed that these changes arise from nucleotide-dependent subdomain rearrangements within the motor domain (31), and the models that we generated using available crystal structures are consistent with such conformational changes also occurring in Kif18A. Thus, in response to nucleotide binding, the main movements take place in the Switch I/II and P-loop subdomains with respect to the largely static tubulin-binding subdomain, which is centered on the continuous helix-α4 that binds the MT at the intradimer interface.

Our data show that the mechanism of MT depolymerization by the monomeric Kif18A motor domain is neither intrinsically limited to MT plus-ends nor dependent on processive stepping by dimers. Furthermore, our experiments reveal that Kif18A_MD is a weaker MT depolymerizer and forms rings much less readily than Kif18A_MDNL (Fig. 3 A and B and SI Appendix, Fig. S5 A and B). This difference in depolymerization activity implicates neck linker docking in conformational changes associated with modulation of MT end dynamics. However, surprisingly, the Kif18A_MD supports plus-end–directed motility in a gliding assay (Fig. 1B). We speculate that the ensemble nature of the gliding assay allows for the motile force generated by this truncated construct to be captured (49). The 2D profiles of the Kif18A_MDNL-MT-bound reconstruction and of Kif18A_MDNL within the tubulin rings are very similar (Fig. 3E), suggesting that ring formation involves similar subdomain rearrangements within the Kif18A motor domain in response to ATP— including neck linker docking—as seen on the MT lattice. This idea is also supported by the observation that both gliding and depolymerization are blocked by BTB-1 (Fig. 4). A very different structural mechanism for MT depolymerization was recently proposed for the kinesin-8 Kif19A (involved in length regulation of primary cilia) involving large-scale unfolding of the Kif19A motor domain (30); however, our data do not support such a mechanism in Kif18A.

Notably, comparison of the MT-bound Kif18A motor with the Kif18A crystal structure reveals a shift of loop8 and helix-α3 toward the MT surface (SI Appendix, Fig. S2). An equivalent conformational change has been reported previously (29) and in the motor domain of kinesin-8 Kif19A, which is also a dual-function motor (30), but not in kinesin-1, -3, or -5 (31, 32, 39, 50). Thus, this MT-induced conformational change may be specific for kinesin-8s, possibly as part of their adaptation to dual function on both the MT lattice and at MT ends. The work done by Kif18A_MDNL at the MT end (induction/stabilization of tubulin curvature) as opposed to the lattice (MT movement) is different because of the context; this is presumably achieved by responding to the relative deformability of the tubulin dimer to which the motor is attached. This suggests that the mechanochemistry of Kif18A is adaptable according to its molecular context.

In stabilizing curved tubulin oligomers, including rings, in the presence of AMPPNP, the depolymerization activity of Kif18A monomers appears similar to that of kinesin-13s (51–53). However, the depolymerization activity of Kif18A monomers is greater in the presence of AMPPNP than in the presence of ATP (Fig. 3A), clearly distinguishing it from the robust depolymerization capability of kinesin-13s in the presence of ATP (54). Thus, while the Kif18A monomeric motor has the ability to induce/stabilize tubulin in a curved conformation, multiple motors may need to be retained at MT ends for depolymerization to occur. Below this threshold, however, accumulating Kif18A motors may influence MT dynamics by stabilizing a curved tubulin conformation, thereby inhibiting MT growth. Other motile kinesins also influence MT dynamics, for example, kinesin-4s that act as pausing factors at MT plus-ends (55, 56). The difference in depolymerization capacity in kinesin-8s compared with kinesin-13s may reflect a tunable mechanochemistry that exists across the kinesin superfamily encoded in their motor domains (43). In vivo, other factors work together with Kif18A on the ends of kinetochore MTs (e.g., ref. 57) to define the complex and tightly regulated dynamics of kinetochore MTs.

A recent investigation of the molecular mechanism of the Kip3 depolymerase (58) showed a number of parallels to our findings regarding Kif18A, highlighting the common mechanisms likely shared by kinesin-8s. The monomeric motor domains of these proteins, 50% identical by sequence, depolymerize MTs from both ends, demonstrating that depolymerization activity is likely intrinsic to kinesin-8 motor domains. As with Kif18A, the Kip3 neck linker makes an important contribution to its depolymerase activity. Furthermore, ATP binding rather than hydrolysis is sufficient for Kip3 depolymerase activity, as also demonstrated by our observation of faster depolymerization by Kif18A in the presence of AMPPNP compared with ATP. This work demonstrates the critical role of Kip3 loop11 for sensing MT ends, suppressing Kip3 ATPase activity, and promoting depolymerization. It also highlights the role of a specific residue in α-tubulin (equivalent in our complex to α-tubulin D116) in this depolymerization activity. In the Kif18A_MDNL-MT structure, the distance between loop11 residues and D116 is too far for a direct interaction (∼14 Å), but curvature in the tubulin at MT ends—and as seen in our ring structures—likely would make interaction between these regions possible. While the structure of Kif18A in projection looks very similar in the rings and the lattice, a relatively small difference in the conformation of loop11 would not be detectable in our current 2D view of the rings. Thus, a similar role for Kif18A loop11 in modulating its depolymerization activity is not excluded by our data. It is possible that tubulin flexibility at MT ends supports the interaction between loop11 and this more distant region of α-tubulin; this in turn could lead to a subtle distortion of the catalytic site, suppressing motor ATPase activity while still allowing ATP binding. As a result, the subdomain rearrangements that are ordinarily coupled to ATPase hydrolysis on the MT lattice are instead used to facilitate tubulin’s distortion or removal from MT ends. Further work on both Kif18A and Kip3 is needed to address this crucial mechanistic question.

We identified a potential binding site for the inhibitor BTB-1 between helix-α2 and helix-α3 of the Kif18A motor domain that is revealed due to MT-dependent subdomain movements. This region contains Kif18A-specific residues, explaining the specificity of BTB-1 for Kif18A inhibition compared with other mitotic kinesins (38). To accommodate BTB-1 in this region, the conformational shift of helix-α3 on MT binding is essential; the equivalent position in the MT-free Kif18A crystal structure is too small to accommodate BTB-1, providing a structural explanation for the specificity of BTB-1 for MT-bound Kif18A. BTB-1 thereby locks Kif18A in the MT-bound state (Fig. 8A), a mechanism that has been described as “gain-of-function” inhibition, in the sense that inhibited Kif18A gains a novel function of tight, rather than dynamic, MT binding (37) Although this kinesin inhibition mechanism is not strictly analogous to classical genetics gain-of-function mutations, it offers a useful point of comparison with Kif11 inhibitors (discussed below). The putative BTB-1– binding site lies at the junction of the SwI/II and P-loop subdomains that move with respect to each other when ATP binds (Fig. 2 and SI Appendix, Fig. S4). Thus, while BTB-1 inhibition is ATP-competitive, it acts allosterically by blocking the conformational changes that accompany ATP binding and reorganization of the catalytic site that enables ATPase activity. Our work provides a fresh perspective on the mechanism of Kif18A inhibition by BTB-1 that will be important for the development of novel small molecules capable of specifically inhibiting Kif18A in the cellular context (59).

Kif11 was the first kinesin for which a specific inhibitor —monastrol—was identified (35). Since then, numerous other allosteric inhibitors that bind to Kif11 loop5 have been developed and incorporated into cancer clinical trials (34). These compounds prevent tight MT binding (60), thereby blocking the force generation function of Kif11 within the spindle; that is, they work as “loss-of-function” inhibitors (37) (Fig. 8B). A similar mechanism also might operate in inhibitors for other mitotic kinesins—for example, the kinesin-14 HSET/KifC1, which is involved in MT clustering at spindle poles. A set of specific inhibitors have been designed to target HSET loop5 to block its activity in cancer cells and lead to apoptosis (61); however, redundancy within the spindle machinery can mean that such loss-of-function inhibition is prone to be overridden by other spindle components. Indeed, Kif15, a kinesin-12 with overlapping roles with Kif11 in mitotic MT organization, can functionally substitute for Kif11 in Kif11-independent dividing cells, thereby allowing the completion of mitosis in the absence of Kif11 (62, 63). The discovery of Kif11 inhibitors represented a paradigm shift in both the ability to functionally dissect the roles of kinesins in cell division and in establishing Kif11—and kinesins in general—as chemotherapeutic targets. Subsequent work has also highlighted the need to consider modes of kinesin inhibition that enhance MT binding to produce a mitotic block that is less readily overridden.

Such inhibition may be particularly advantageous for drugging the mitotic spindle. By trapping the target motor on its MT track and preventing its dynamic turnover, knock-on blockage of much of the dynamic spindle machinery can be achieved, thereby amplifying the antimitotic effect. In contrast to most Kif11 inhibitors and apart from the data presented here, to date no structural information is available about kinesin inhibitors that trap the motors on MTs, being by definition essentially largely inaccessible for study by X-ray crystallography. In addition to Kif18A inhibition by BTB-1, there are several reports of inhibitors of Kif11 that act allosterically and increase Kif11 MT binding within the spindle (37, 64). Similarly, the allosteric inhibitor (GSK923295) of CENP-E, a motor involved in metaphase chromosome alignment, blocks the release of inorganic phosphate from MT-bound CENP-E, and as demonstrated by mutagenesis and photolabeling, also binds in the vicinity of helix-α2, helix-α3, and loop5 (65). A series of selective HSET inhibitors that are ATP/ADP-competitive, specific for the motor’s MT-bound state, and targeted to loop5 also have been described (66, 67). It seems likely that these inhibitors will work in a similar way and recognize specific sequences in this region of the particular motor. A recent study of inhibition of kinesin-1 by the general anesthetic propofol showed that kinesins active in interphase are also susceptible to perturbation of their ATPase cycles by small-molecule binding; this could be explained through mechanisms related to disruption of subdomain rearrangements (68). Thus, our study of BTB-1 inhibition of Kif18A reinforces a common mechanism of allosteric kinesin inhibition through binding at the dynamic interface of subdomains within the kinesin motor domain, either on or off the MT (69).

Mitotic kinesin inhibitors continue to be valuable tools for dissection of the mechanisms and roles of different motors within the complex spindle machinery, and may have clinical applications as well. The emergence of resistance to monotherapies by accumulation of point mutations presents a challenge to effective drug development and highlights the need to investigate multiple modes of mitotic motor inhibition to overcome drug resistance. The mechanistic redundancy intrinsic to mitosis highlights another problem with monotherapeutic approaches. Thus, the development of combinatorial drug therapies—in particular, inhibitors that trap motors tightly on MTs—in which several mitotic molecules are inhibited simultaneously is an important approach. This is critical to specifically inhibit collaborating or synergistic mitotic partners. In this context, work on Kif18A is relevant because it collaborates with CENP-E in chromosome alignment and is up-regulated in a subset of cancers (19, 70). Thus, inhibitors like BTB-1 may be important when considered synergistically with CENP-E inhibition, particularly when considering the induction of CIN as a strategy for inducing tumor death (71). With the evolution of personalized treatments of heterogeneous cancers, the possibility of accurately targeting different components of the spindle in different ways is becoming increasingly important.

We thank Anthony Roberts for help in designing the TIRF assays, access to the TIRF microscope, and very useful discussions. A.P.J. and M.T. also thank Joseph Newcombe, Martyn Winn, Ardan Patwardhan, and Ingvar Lagerstedt for useful technical discussions. J.L., A.P. and C.A.M. were supported by funding from Cancer Research UK (C33336/A13177) and Worldwide Cancer Research (16-0037), and thank Dr. Charles Sindelar (Yale University) for reconstruction algorithms. A.P.J. and M.T. are supported by the Medical Research Council, UK (MR/M019292/1 and MR/N009614/1). T.U.M. and M.M.M. were supported by funding from the Deutsch Forschungsgemeinschaft (CRC 969) and the Konstanz Research School of Chemical Biology.