Recently, several 3D images of kinesin-family motor domains interacting with microtubules have been obtained by analysis of electron microscope images of frozen hydrated complexes at much higher resolutions (9-12 Å) than in previous reports (15-30 Å). The high-resolution maps show a complex interaction interface between kinesin and tubulin, in which kinesin's switch II helix α4 is a central feature. Differences due to the presence of ADP, as compared with ATP analogues, support previously determined crystal structures of kinesins alone in suggesting that α4 is part of a pathway linking the nucleotide-binding site and the neck that connects to cargo. A 3D structure of the microtubule-bound Kar3 motor domain in a nucleotide-free state has revealed dramatic changes not yet reported for any crystal structure, including melting of the switch II helix, that may be part of the mechanism by which information is transmitted. A nucleotide-dependent movement of helix α6, first seen in crystal structures of Kif1a, appears to bring it into contact with tubulin and may provide another communication link. A microtubule-induced movement of loop L7 and a related distortion of the central β-sheet, detected only in the empty state, may also send a signal to the region of the motor core that interacts with the neck. Earlier images of a kinesin-1 dimer in the empty state, showing a close interaction between the two motor heads, can now be interpreted in terms of a communication route from the active site of the directly bound head via its central β-sheet to the tethered head.
Kinesin motors use energy supplied by ATP hydrolysis to move cargo along microtubules (MTs) or, in some cases, to depolymerise these cytoskeletal filaments. All have very similar motor domains and appear to follow the same cycle of strong binding to tubulin when the nucleotide-binding pocket is empty or contains ATP and weak binding, leading to detachment, when ADP is bound (Fig. 1). Researchers in this field are interested in the conformational changes that accompany these chemical changes and the mechanisms that control the cycle. The three main functional regions of the motor domain that need to exchange information are the microtubule-binding interface, the nucleotide-binding site and the neck region that connects the motor domain to the cargo-binding domain of a full-length protein (Fig. 2). Mutagenesis and biochemical studies first indicated that a fairly extensive part of the surface is involved in tubulin binding (Fig. 2A,B,D,E show examples of motor domains as viewed from the MT). The nucleotide-binding pocket, on an adjacent part of the surface, is surrounded by several loops: the loop known as switch I lies close to the α- and β-phosphates of the bound nucleotide, whereas the P-loop and switch II loop are close to the position of the γ-phosphate that is lost when ATP is hydrolysed to ADP. The neck associates with another part of the surface of the motor domain, on the opposite side from the nucleotide-binding site. Plus-end-directed kinesins and minus-end-directed kinesins differ in having their neck regions connected to the C-terminal and N-terminal ends, respectively, of the motor domain. However, the necks emerge from a similar point in both cases (see Fig. 2C,F) and may be controlled in similar ways. Nevertheless, it is unclear in either case how binding to a MT stimulates the release of ADP, how ATP binding promotes neck movement or how neck movement leads to detachment of the motor domain from tubulin.
The database of protein crystal structures currently contains over 40 near-atomic models of motor domains from the kinesin superfamily. Although the list includes a variety of proteins from distant subgroups and some mutants, most structures are of motors in which ADP is bound to the nucleotide-binding pocket, since this is the default state for a free motor. In this state, the neck is usually disordered (`undocked') in the case of plus-end-directed motors (see Fig. 2D), whereas the N-terminal neck of a minus-end-directed motor is usually docked on to the surface of the motor domain (Fig. 2A). A few structures of plus-end-directed kinesin motors include bound analogues of ATP (Nitta et al., 2004; Kikkawa et al., 2001; Ogawa et al., 2004) and show interesting differences from the majority, including changes in the positions of the necks. A minority of the ADP-bound proteins (Sack et al., 1997; Kozielski et al., 1997; Turner et al., 2001; Yun et al., 2003) also appear to have crystallised with their necks in the `wrong' conformations (Fig. 2B,E) but, overall, the structures seem consistent with the general idea that ADP and ATP binding each make a particular conformation of the neck more probable (Rice et al., 1999; Sindelar et al., 2002).
The crystal structures have led to a dominant model for how cargo may be moved relative to a plus-end-directed motor domain when a molecule of ATP enters the nucleotide-binding site (Figs 1, 2). Rice et al. (Rice et al., 1999) used a range of techniques, including low-resolution 3D electron microscopy (EM) of nanogold-labelled proteins, to show that the neck linker (which connects the neck with the rest of the motor domain) of kinesin is free to move when the motor domain is empty or bound to ADP but is bound to a site on the surface of the motor domain when ATP or an ATP analogue is present. Crystal structures of both kinesin-1 and Kif1a show that, when the neck linker is docked in position, there are also changes in the positions of the switch II helix α4 and nearby helix α5 (see Fig. 2D,E). It was proposed that these latter changes are caused by ATP binding and directly promote neck-linker docking, which, in turn, provides a force to move cargo towards the MT plus end (Rice et al., 1999). Although it has become clear that that neck-linker docking cannot provide enough energy to move the maximum load that dimeric kinesin has been observed to pull from one binding site to another, it is still thought likely that ATP-dependent neck linker docking provides some force as well as a directional bias. An equivalent model for Ncd (Endres et al., 2006) or Kar3 is even more problematical because any force towards the minus end would need to be generated by neck detachment (Fig. 1A and Fig. 2B).
Since there are no crystal structures of any kinesin family motor in the empty state, it is completely unclear how ADP is released and replaced by ATP. Nor is it known how neck movement or nucleotide hydrolysis within the motor domain control its binding to and detachment from tubulin. As yet, no crystallisation conditions have been found to induce crystals of a kinesin-tubulin complex. Electron microscopists have therefore been attempting to obtain equivalent information from high-resolution images of motor domains bound to MTs, which is possible when rapidly frozen unstained specimens are studied under cryo-EM conditions. The difficulty lies in collecting enough good images to average out the noise that masks the fine details. Recent achievements are two 10 Å maps of a monomeric motor domain (from mammalian Kif1a, a kinesin-3 motor) bound to β-adenylyl-imidodiphosphate (AMPPNP) or ADP (Kikkawa and Hirokawa, 2006), three 10-12 Å maps of a C-terminal motor domain (from yeast Kar3, a kinesin-14 motor), including a strongly bound empty state (Hirose et al., 2006), and a 9 Å map of a single motor domain of human conventional kinesin (kinesin-1, a processive dimeric motor) (Sindelar and Downing, 2007). At these levels of resolution, it is possible to detect features such as α-helices and compare their positions with those in crystal structures by superimposing the latter as closely as possible on to the protein-density maps obtained by EM (as in Figs 3 and 4).
Images of kinesin-family motor domains bound to microtubules
The general orientation of motor domains on MTs has been clear for many years from mutagenesis and biochemical studies (Woehlke et al., 1997; Alonso et al., 1998). The recent high-resolution EM maps of different kinesin motors all show basically the same orientation (Kikkawa and Hirokawa, 2006; Hirose et al., 2006; Sindelar and Downing, 2007): there is general agreement that the switch-II-helix α4 is situated in the groove between α- and β-tubulin (see Fig. 5), whereas loop L8 and the loops at either end of α4 (L11 and L12) interact directly with tubulin (Fig. 3A,B and Fig. 4A). There is less agreement about structural changes that take place throughout the cycle. The currently available 3D data that show nucleotide-dependent structural changes are summarised in Table 1 and are described below in more detail. Tightly bound ATP-like states are most easily studied, by use of non-hydrolysable analogues such as AMPPNP to obtain MTs fully decorated with motor domains. Empty states are also usually tightly bound and occur when all free nucleotides are removed. It is most difficult to obtain full decoration in the weakly attached ADP-bound state but this may be achieved by a high motor protein concentration and low ionic strength.
Differences between ADP- and ATP-bound states
Images at 10 Å resolution obtained by Kikkawa and Hirokawa (Kikkawa and Hirokawa, 2006) show changes between ADP-bound and AMPPNP-bound Kif1a motor domains attached to tubulin that mostly resemble those in the equivalent crystal structures of Kif1a alone. Apart from the movement of the neck linker, the most striking difference between the ADP and ATP-like states, a 20° rotation of the core domain relative to α4, is also evident in the EM maps (Fig. 3B). A much smaller shift seen in different crystal structures of conventional kinesin-1 alone (Fig. 2D,E) also correlates with the position of the neck linker (Sindelar et al., 2002) but no nucleotide-dependent rotation was evident in EM maps of kinesin-1 motor domains bound to MTs (Rice et al., 1999). Moreover, the bound motor domains of ADP-filled dimeric kinesin molecules attached to MTs were, apparently, able to rotate freely through a large angle (30-40°) in the opposite direction, to relieve a clash between the coiled-coil stalk and the MT (Hirose et al., 1999) (see also Fig. 6C). No clear rotation, nucleotide-dependent or otherwise, has yet been detected in other plus-end-directed motors, such as Unc104 (Al-Bassam et al., 2003) and Cenp-E [reconstructed to 15 Å resolution (Neumann et al., 2006)], or in minus-end-directed (kinesin-14 family) motors, such as Kar3 (Hirose et al., 2006) or Ncd dimers [for which it would have been apparent from the position of the tethered head (Hirose et al., 1998; Wendt et al., 2002; Endres et al., 2006)]. There is thus no evidence that a large-scale rotation of the motor core is a fundamental feature of communication between the enzymatic site and the cargo-binding domain. Instead the rotation of the Kif1a motor core may be a side-effect of more fundamental changes in α4 during nucleotide exchange (see below).
An unexpected feature of the EM images of Kif1a is that α4 and its attendant loops, L11 and L12, appear as a longer rod of density in the map of the AMPPNP state, even though α4 is shorter in the AMPPNP crystal structure (see Fig. 3B). Also, the switch I loop of the superimposed crystal structure lies outside the strong density of the EM map of the ADP state, which suggests that this loop moves closer to the MT. However, the crystal structures and EM maps agree in showing the nucleotide-binding pocket of Kif1a is more open in the ADP state than in the AMPPNP state. The same is true for Kar3. The combined results from Kif1a and Kar3 clearly indicate that there is an important communication route extending via α4 and loops L11 and L12, from the nucleotide-binding pocket to the base of either the C-terminal neck-linker of a plus-end-directed kinesin (see Fig. 2D-F) or the N-terminal neck of a minus-end-directed motor (see Fig. 2A-C). As mentioned above, it has been proposed to carry the signal, during nucleotide exchange, from the active site to the neck. Alternatively or additionally, it may transmit a message in the opposite direction, from the cargo-binding region to the MT-binding region, to signal that the neck linker is docked, that there is currently no rearwards force and that the motor domain can safely detach from the MT (Rob Cross, personal communication). In myosin, the equivalent switch II helix is also thought to be a major route of communication (Vale and Milligan, 2000). However, it connects the nucleotide-binding site to the converter domain that controls the motion of the lever arm directly, rather than via the track-binding site as in kinesin.
Another very significant difference between the two states is in the position of helix α6 (shown green in Fig. 3C), which moves closer to tubulin when an ATP analogue is bound. The maps of the minus-end-directed Kar3 suggest similar changes may occur there (Fig. 4E). Since the P-loop is situated directly below α6 (Fig. 2), a movement of α6 in either plus-end-directed or minus-end-directed motor domains could transmit a signal directly from the nucleotide-binding pocket to the neck-binding side of the motor domain and thus contribute to the docking or undocking of the neck. Such a signal might operate in addition to that proposed to act via the switch II region.
Differences between nucleotide-bound states and the empty state
Both ends of the switch-II-helix α4 have been observed to vary in length in different crystal structures of Kif1a and it has been proposed by Nitta et al. (Nitta et al., 2004) that binding to tubulin is initiated by loop L12 at the C-terminal end of the helix, whereas nucleotide exchange causes loop L11 at the other end to bind more strongly and L12 more weakly. It was unclear how changes at one end of the helix could affect the other end but now the 3D image of the MT-bound empty state of Kar3 (Hirose et al., 2006) suggests that most of the helix melts during nucleotide exchange (Fig. 4C); its C-terminal half and loop L12 are no longer distinguishable from tubulin density. This dramatic loss of secondary structure is compensated by the intimate interaction with tubulin and it is probably significant that, in the absence of MTs, kinesin proteins without nucleotide tend to denature. The melting of α4 coincides with ADP loss but its secondary structure is restored when ATP binds. We suggest that this refolding represents the beginning of the process of detachment, which is completed after ATP hydrolysis and Pi release.
Another large change observed in the MT-bound empty state of Kar3 is the movement of the region that includes loop L7. This loop was originally thought to be a part of the MT-binding region (Woehlke et al., 1997), but superimposing the crystal structures on to previous EM maps showed some distance between L7 and tubulin. The new Kar3 map in the empty state shows movement of this loop toward helix H4 of β-tubulin (Fig. 4A). Since L7 interacts with the switch II loop in crystal structures, binding of the MT to L7 and the switch II helix may move the loops surrounding the nucleotide-binding pocket and trigger release of ADP.
A 3D map of conventional kinesin bound to a MT without added nucleotide (Sindelar and Downing, 2007) has the highest resolution obtained so far (see Fig. 3A) – a remarkable achievement given that these authors found the motors to be so weakly bound that they had difficulty in keeping them attached while preparing EM grids. In contrast to the nucleotide-free Kar3 map, the kinesin-1 map shows an elongated switch II helix and no movement of L7. This could be owing to a difference between the minus-end-directed and plus-end-directed motors but it is more likely that these two structures are in different states. The reason for the weak binding to MTs reported by Sindelar and Downing is unclear, because an empty motor is normally strongly bound. The only difference in the conditions was that, during apyrase treatment to remove ATP and ADP, they exposed the motor domains to GTP, which would have bound to the kinesin nucleotide pocket (Cohn et al., 1989) and have been hydrolysed to GDP. It is possible, though unexpected, that this bound GDP was not lost during the subsequent centrifugation through a nucleotide-free sucrose cushion and dilution with a nucleotide-free buffer. No nucleotide is detectable in the 3D map (Fig. 3A) but the resolution is too low to be sure of its absence.
When loop L7 moves in Kar3, the central β-sheet, which is associated with this loop, should also change structure and this appears to affect the `point' at the top of the sheet (Loop L10 in Fig. 2A, blue arrowheads in Fig. 4A). These changes reveal a further communication route: between the MT-binding loops and the neck. In the case of Kar3, changes in the β-sheet may lead to displacement of the neck from the site to which it docks in the ADP state. Similar changes in the β-sheet of kinesin-1 would be expected to have a different effect because of the difference in the necks (Fig. 2).
Lower-resolution images of dimeric motors
Although there is not yet any high-resolution image showing changes in the β-sheet of a plus-end-directed kinesin, there is some evidence from low-resolution images of kinesin dimers that the conformation of the core domain of kinesin-1 in the empty state is significantly different from that of the nucleotide-bound states. Whereas heads of a nucleotide-filled kinesin dimer appear either as separate entities or loosely associated (e.g. Fig. 6C), in conformations likely to be similar to the rat kinesin dimer crystal structure (Fig. 6A), in the nucleotide-free state the tethered head appears to be closely associated with the directly bound head (Fig. 6D). As detailed in Table 1, the empty state is the only one for which images of dimeric kinesin from different research groups have all shown density corresponding to a specifically bound tethered head1. Relative to the crystal structure of dimeric kinesin (Fig. 6A), the tethered head in Fig. 6D appears to tilt downwards (as modeled in Fig. 6B), which may be a consequence of structural changes associated with the proposed distortion of the central β-sheet.
The same second-head arrangement of dimeric kinesin decorating MTs in the absence of nucleotide was reported independently by Arnal and Wade (Arnal and Wade, 1998), and even an image published by Hoenger et al. (Hoenger et al., 1998), who reported in their text that they saw only one head, shows a similar, though slightly smaller, feature. The reduced apparent size is easily explained by blurring owing to the greater mobility of a domain not in direct contact with the MT. To see equal-sized heads, we selected images with the best-ordered second heads for inclusion in the averaged data. Proponents of the model in which the dwell time (see main text) is spent with both heads simultaneously bound to adjacent sites on the MT (Hoenger et al., 1998; Hoenger et al., 2000; Hackney, 2007) have suggested that the tethered heads seen in our reconstructed images belong to a non-specifically bound extra layer of molecules. This does appear to be a reasonable explanation for some reconstructed images (Hoenger et al., 2000) (K.H. and L.A.A., unpublished) obtained from poorly soluble motor-domain constructs, when the decorated MTs appear thickly encrusted with non-specifically bound molecules. In such cases, molecules in a second layer (and sometimes even a third layer) occupy the holes in the underlying layer so that each contacts 2-4 different heads in that layer. This is in marked contrast to specifically bound kinesin dimers, whose second heads each make contact with only one directly bound head.
Details of the interactions between the two heads have been clarified by recent studies of kinesin-tubulin complexes. Alonso et al. have shown that the `gate' allowing ADP to be released from kinesin, which is closed in the absence of tubulin, can be opened by soluble tubulin and not just by assembled MTs (Alonso et al., 2007). An interaction with soluble tubulin had previously been observed in the case of the specialised members of the kinesin family that act as MT depolymerisers (Moores and Milligan, 2006) but now it is clear that conventional kinesin-1 behaves similarly. Moreover, the interaction with tubulin, like that with MTs, releases ADP from only one of the heads of a kinesin dimer and the other must wait for the first to bind ATP before it can follow suit. As long as the first head is empty, it seems to shield the second head from interacting with tubulin and releasing its ADP. These results undermine models in which the so-called waiting state relies on a signal transmitted by tension produced by having both heads bound simultaneously to the MT lattice. Instead, they strongly support the idea of a direct interaction between the heads and indicate that the nucleotide-free state observed by cryo-EM corresponds to an actual stage in the cycle of processive movement (Fig. 1B; bound head at stage 2 and tethered head at stage 5). Combining these observations with the new results obtained by high-resolution EM thus leads to the idea that changes in the β-sheet of kinesin-1 may provide a parking site for the tethered head on the top of the currently active head during the time spent waiting for ATP to bind (the `dwell time') and then dislodge it when ATP binds, so that it is free to bind to the next site on the MT. Parking of the tethered head on top of the directly bound head could also send a signal in the opposite direction to allow ATP to bind.
High-resolution images of kinesin-family motor domains bound to MTs under conditions that mimic specific stages in the nucleotide hydrolysis cycle clearly support the idea that the switch II helix is the central structure involved in MT binding and is in close communication with the nucleotide-binding pocket. For both plus-end- and minus-end-directed motors, there are changes at each end of this helix, between the ADP-bound and ATP-analogue-bound states. The vital role of α4 has been further emphasised by a 3D image of Kar3 in the tightly bound empty state, where it appears to melt on to the MT surface. In this state, loop L7 also interacts closely with tubulin, distorting the central β-sheet, which may be important in controlling the binding site on the motor domain for the neck that connects to the cargo-binding region. Both of these changes were a surprise because no equivalent state has yet been seen by X-ray crystallography and this tightly bound state has not yet been observed at high resolution for another type of motor domain.
In addition to the need for supporting images of the tightly-bound empty state, there remain many questions about the sequence of changes in tubulin-bound motor domains when ATP binds and is hydrolysed. The movement observed for helix α6, that also seems to connect the nucleotide-binding and neck regions, is intriguing and needs to be compared in more detail for plus-end-directed and minus-end-directed motors. Ideally, it would be desirable to crystallise kinesin-tubulin complexes and view the interfaces between the proteins in different states at near-atomic resolution. Failing this, there remains the hope of pushing the resolution of the EM images much higher, perhaps to 4 Å or better. The specimens used in the work reviewed here did not appear to be limiting the resolution obtainable. The number of images required to see detail at higher resolution will grow steeply but, fortunately, the human effort involved will be reduced by recently developed computer programs. In particular, being able to analyse images of 13- or 14-protofilament MTs (Li et al., 2001; Kikkawa, 2004; Sindelar and Downing, 2007) will help in data collection. Previous use of the rare 15-protofilament tubes, which are perfectly helical and therefore easier to analyse, meant that only a small fraction of the images recorded were used. In future it should be possible to use the majority of images.