Differential trafficking of Kif5c on tyrosinated and detyrosinated microtubules in live cells

The kinesin superfamily is a structurally diverse group of ATP- and microtubule (MT)-dependent motors (Lawrence et al., 2004). Kinesin-1 (conventional kinesin) (Brady, 1985; Vale et al., 1985a) moves towards the plus ends of MTs in vitro (Vale et al., 1985b). It is a tetramer consisting of two heavy chains and two light chains. The globular N-terminal domain of the heavy chain contains the motor, which interacts with MTs in an ATP-dependent manner. In the absence of MTs, ADP release is rate limiting (Hackney, 1988) and this release step is accelerated by the interaction of kinesin-1 with MTs.

Folding of kinesin-1 regulates its activity (Hackney et al., 1992). In the folded state, the tail of kinesin-1 binds to the head and ATPase activity is low. Binding of kinesin-1 to cargo unfolds the molecule, and allows it to bind to MTs and move along them in an ATP-dependent manner (Coy et al., 1999a; Friedman and Vale, 1999). A 65 amino-acid-long region at the end of the C-terminal tail is responsible for this cargo binding (Coy et al., 1999a). Kinesin-1 heavy chain can bind to its cargo either directly through adaptor proteins, or indirectly through the light chains associated with this region of the protein (for reviews, see Hirokawa and Takemura, 2005; Schnapp, 2003; Vale, 2003). The amount of kinesin-1 expressed in cells varies from 0.1-1 μM in tissues (Hollenbeck, 1989), but it is mostly found in a cytoplasmic pool where it is likely to be in a folded inactive form. A smaller amount of kinesin-1 is bound to organelles.

Kinesin-1 is a processive motor. In vitro, it moves along MTs for several μm, at a maximum velocity of ∼0.8 μm second^–1 (Coy et al., 1999b; Svoboda et al., 1993). Less is known about the motile behaviour of kinesin-1 in vivo. Most of our knowledge has been obtained indirectly, through the observation of putative cargoes. Velocity measurements from those experiments ranged from 0.034 μm second^–1 for RNA transport granules (Kanai et al., 2004) to 0.33 μm second^–1 for dense-core vesicles in clonal β-cells (Varadi et al., 2002). A recent study used a modified form of total internal reflection fluorescence microscopy to measure the dynamic behaviour of a motor domain construct in live cells, and found that average speeds were 0.78 μm second^–1 (Cai et al., 2007). The processive nature of kinesin-1 enables it to function as a trafficker in cells, transporting vesicles or other cargo over long distances.

It is still unclear how the cargo trafficking by kinesin-1 is spatially regulated in vivo. As it is a plus-end-directed MT motor, and the plus ends of MTs are oriented towards the cell periphery, kinesin-1 will broadly direct trafficking towards the plasma membrane. However, kinesin-1 can transport cargoes as diverse as membrane- and signalling-molecules to specific intracellular regions, and this is likely to be crucial in processes such as cell migration and neuronal differentiation. One possibility is that post-translational modifications of tubulin are important in directing trafficking to specific regions at the cell periphery (reviewed in Lakamper and Meyhofer, 2006). Microtubules can be post-translationally modified in a variety of ways, including through acetylation of α-tubulin at lysine at position 40, and through detyrosination by removing the C-terminal tyrosine residue of α-tubulin, which reveals a charged glutamate on the so-called `E-hook' of α-tubulin (reviewed in Luduena et al., 1992). A recent study has shown that the localisation of JNK-interacting protein 1 (JIP1), a protein trafficked by kinesin-1, depends on MT acetylation (Reed et al., 2006). Trafficking of intermediate filament proteins by kinesin-1 has been shown to depend on detyrosination (Gyoeva and Gelfand, 1991; Gurland and Gundersen, 1995).

Here, we have investigated the movement of kinesin-1 in live cells using a construct in which enhanced green fluorescent protein (eGFP) was fused to the N-terminal of the kinesin-1 isoform Kif5c (GFP-Kif5c). We then determined whether movement of GFP-Kif5c depends on post-translational modification of tubulin. Vertebrates contain three conventional kinesin-1 genes: Kif5a, Kif5b and Kif5c. Whereas Kif5b is ubiquitously expressed, Kif5a and Kif5c are enriched in neurons. The three proteins are highly homologous and functionally redundant (Kanai et al., 2000). We investigated the properties of GFP-Kif5c in neuronal and non-neuronal cells, and confirmed that it moves at a speed similar in vivo to that measured for the same motor in vitro. Moreover, we found that GFP-Kif5c preferentially moves along a subset of MTs, these MTs are post-translationally modified by detyrosination, and detyrosination affects velocity in vitro. Our results suggest that the spatial control of kinesin-1-based trafficking can be effected by the tyrosination-detyrosination cycle of MTs.

GFP-Kif5c was transfected into mammalian cells and its distribution examined using live cell imaging. The N-terminal GFP tag was not expected to interfere with the interaction of the motor with its normal binding partners through the C-terminal tail (Coy et al., 1999a). Two main cell types were used; COS-7 fibroblasts and, because Kif5c was originally identified as a neuronal motor, the neuronal cell line Hdh^+/+. Only cells that expressed very low levels of GFP-Kif5c were used in this study to minimise any potential artifacts caused by expression of GFP-Kif5c at high levels. The morphology of these cells, as well as the distribution of intracellular organelles, appeared normal when transfected cells were compared with adjacent untransfected cells after fixation and immunostaining (data not shown). Organelles, which appear white in an inverted fluorescence image because they exclude GFP-Kif5c, are distributed throughout the cell. We confirmed that the majority of the larger organelles were mitochondria by using a mitochondrion-specific dye (mitotracker). We did not observe any differences in mitochondrial dynamics between transfected and untransfected cells (data not shown).

In cells whose expression levels of GFP-Kif5c were low, we observed several distinct distributions of GFP-Kif5c. GFP-Kif5c localised to fluorescent puncta along MTs and clustered at the ends of these MTs in specific cortical regions of the cell (Fig. 1). We also found a diffuse cytoplasmic distribution of GFP-Kif5c, which probably represents folded inactive GFP-Kif5c not bound to MTs. In addition, there was often a low level of diffuse labelling by GFP-Kif5c along the MTs, which allowed us to see their outline. This distribution was observed in both COS-7 (Fig. 1A,B,D, supplementary material Movies 1-3) and neuronal cells (Fig. 1C, supplementary material Movie 4). In cells that expressed high levels of GFP-Kif5c, this distribution along MTs was much brighter, puncta of GFP-Kif5c were able to crosslink MTs, and we were not able to observe any movement of the puncta (data not shown). However, in cells expressing low levels of GFP-Kif5c, only a subset of MTs was decorated with GFP-Kif5c and it was at the ends of these MTs that clusters of GFP-Kif5c puncta were localised. It seems reasonable to assume that these puncta represent GFP-Kif5c associated with a variety of cargo within the transfected cells. The size of these puncta varied. Judging from the size and the intensity of the puncta, the majority are likely to contain more than one molecule of GFP-Kif5c.

Co-imaging of GFP-Kif5c and fluorescently labelled MTs confirmed (by direct observation) that GFP-Kif5c moved along MTs (Fig. 2; supplementary material Movies 5 and 6). Single fluorescent puncta of GFP-Kif5c could be followed through sequential frames (Fig. 2A,C) as they moved along a single MT that had been fluorescently labelled using mCherry–α-tubulin (Shaner et al., 2004). The velocity of single puncta varied with time (Fig. 2B,D), suggesting that obstacles to movement exist in vivo that affect the velocity of Kif5c as it moves along the MT. We also found that some puncta stop moving for a short period of time, but remained attached to the MT, before resuming movement.

We determined the velocity of GFP-Kif5c movement along MTs by tracking GFP-Kif5c puncta in time-lapse images of COS-7 or neuronal (Hdh^+/+) cells. To ensure accurate tracking of fast-moving puncta, we used a frame capture rate of two per second. We found that the average velocities for individual puncta, including any stationary phases (Fig. 3A), were similar in Hdh^+/+ cells (0.32±0.13 μm second^–1, mean ± s.d., n=58 puncta) and COS-7 cells (0.29±0.13 μm second^–1, n=73 puncta). These average velocities are similar to those obtained by Varadi et al., who had imaged the movement of dense core vesicles by Kif5b in β-cells (Varadi et al., 2002). Excluding any stationary phases in calculating the average velocity (i.e. velocity=0 μm second^–1) slightly increased the average velocities to 0.39±0.14 μm second^–1 for Hdh^+/+ cells and to 0.34± 0.11 μm second^–1 for COS-7. The average maximum velocities measured for each of the puncta were higher in Hdh^+/+ cells (0.78±0.31 μm second^–1) than in COS-7 cells (0.62±0.24 μm second^–1). These maximum values are close to those reported for in vitro motility studies of kinesin-1, and to the recently reported velocities for movement of a fluorescently tagged motor domain in COS cells (Cai et al., 2007). We also measured the run lengths (distance moved per individual GFP-Kif5c fluorescent puncta) (Fig. 3B). GFP-Kif5c puncta could be tracked for up to 8 μm and we did not find any difference in run length between COS-7 and neuronal cells.

To further test whether the observed distribution of GFP-Kif5c was dependent on its motor function, we compared its distribution with that of a non-motile mutant (GFP-Kif5cmut) in living cells. GFP-Kif5cmut contains a point mutation (T93N) in the motor domain that inhibits ATP hydrolysis (Nakata and Hirokawa, 1995). This mutant can bind to MTs but cannot move along them. It has been used previously as a `roadblock', and has been shown to decorate the MT network in vivo but not to re-organize it (Krylyshkina et al., 2002; Crevel et al., 2004). In agreement with this earlier work, we found that GFP-Kif5cmut heavily decorates MTs in the central region of the cell but does not appear as puncta or accumulate in cortical clusters or in the cytoplasm (Fig. 4A). Neither wild-type nor mutant GFP-Kif5c affected MT organisation (Fig. 4B). Expression of GFP-Kif5cmut also inhibited movement of vesicles, such as mitochondria in live cells (data not shown), as previously shown for this mutant (Varadi et al., 2002).

In addition, we found that accumulations of puncta at the cell periphery that were identified by visualising GFP-Kif5c were not simply static aggregates of kinesin, but exhibited dynamic behaviour by fluorescence recovery after photobleaching (FRAP). Cells were imaged on a laser scanning (LSM) confocal microscope with a heated stage, and a small area either at the periphery or in the centre of the cell was bleached (Fig. 4C, supplementary material Movie 7). After bleaching, the fluorescence intensity in peripheral accumulations of GFP-Kif5c recovered (Fig. 4C) with a t_1/2 of 16 seconds (n=7). FRAP was slightly faster at the periphery than in the centre of the cell (Fig. 4D). By contrast, photobleaching experiments with GFP-Kif5cmut showed very little recovery of fluorescence after photobleaching (Fig. 4C,D). These data indicate that the distribution of GFP-Kif5c in live cells is dependent on the motor function of this protein.

As we described above, the live-cell imaging experiments showed that GFP-Kif5c preferentially associates with a subset of MTs in cells (Fig. 1). We suspected that these were stable MTs and used two approaches to investigate this further. First, stable MTs tend to be more resistant to the MT poison nocodazole (Kreis, 1987). Therefore, we treated cells expressing GFP-Kif5c with nocodazole (Fig. 5) and found that GFP-Kif5c-decorated MTs persist after this treatment (Fig. 5A). GFP-Kif5c was able to move along the few remaining nocodazole-resistant microtubules (Fig. 5B, supplementary material Movie 8). To further investigate this, we performed dual imaging of GFP-Kif5c and of microtubules labelled with mCherry–α-tubulin during and after the addition of nocodazole to the culture medium (Fig. 5C, and supplementary material Movie 9). We found that microtubules not decorated with GFP-Kif5c before treatment with nocodazole disappeared rapidly, whereas MTs that were decorated with GFP-Kif5c persisted for longer and, eventually, began to accumulate more GFP-Kif5c (Fig. 5C).

Stable MTs accumulate a number of post-translational modifications that allow them to be identified by immunostaining. These include detyrosination, which creates MTs whose C-terminal tyrosine residue has been removed from the α-tubulin subunit (also known as Glu-MTs) (Gundersen et al., 1984), and acetylation (Piperno et al., 1987), in which lysine residue at position 40 is acetylated.

To determine whether the nocodazole-resistant MTs decorated by GFP-Kif5c are post-translationally modified, untransfected cells and cells expressing GFP-Kif5c were fixed and stained for acetylated, detyrosinated and tyrosinated tubulin after 60 minutes of nocodazole treatment (Fig. 5D). In untransfected cells, triple-stained for all three types of tubulin, nocodazole-resistant MTs stained for both acetylated and detyrosinated tubulin, and weakly for tyrosinated tubulin. GFP-Kif5c-expressing cells were additionally stained for tyrosinated and either detyrosinated or acetylated tubulin. Nocodazole-resistant MTs were decorated by GFP-Kif5c, and stained positively for both acetylated and detyrosinated tubulin but only weakly for tyrosinated tubulin. This confirms that GFP-Kif5c binds to stable MTs in nocodazole-treated cells.

We next investigated the association of GFP-Kif5c with tyrosinated, detyrosinated or acetylated MTs in cells not treated with nocodazole to confirm that GFP-Kif5c-decorated MTs tended to be stable, post-translationally modified MTs. Untransfected cells were triple-stained for tyrosinated, detyrosinated and acetylated tubulin (Fig. 6A) to determine the relative amounts of modified and unmodified MTs. An analysis of 60 randomly selected MTs from ten cells showed that 95% of the MTs were predominantly tyrosinated (Fig. 6A), and the remaining 5% contained modified (acetylated and/or detyrosinated) tubulin. Of the modified MTs, 10% were predominantly detyrosinated, 27% were predominantly acetylated, and the remainder was both acetylated and detyrosinated.

In GFP-Kif5c-expressing cells, GFP-Kif5c associated with all three types of MT examined (Fig. 6B). However, further analysis and quantification (Fig. 6C) showed that a MT in which 50% or more of its visible length was decorated with GFP-Kif5c was much more likely to be detyrosinated (70% of MTs) or acetylated (20%) than tyrosinated (<2%). Thus, GFP-Kif5c appears to associate predominantly with the 5% of modified MTs in the cell and to show a strong bias towards the MTs that are predominantly detyrosinated.

To further investigate the interaction of kinesin-1 with modified MTs, we used in vitro motility assays to determine whether kinesin-1 moved modified and unmodified MTs differently. We used three different sources of tubulin; predominantly tyrosinated tubulin purified from HeLa cells (Cytoskeleton), tubulin from HeLa cells treated with carboxypeptidase A (CPA), which removes the C-terminal tyrosine and generates detyrosinated tubulin (Chapin and Bulinski, 1991), and highly modified tubulin isolated from pig brain. The latter source of tubulin is most commonly used in in-vitro motility assays. A dot-blot assay showed that tubulin from untreated HeLa cells contained at least 95% tyrosinated tubulin (Fig. 7A) as well as partially acetylated tubulin, whereas tubulin from CPA-treated cells was at least 95% detyrosinated only (Fig. 7A), and pig-brain tubulin was heavily modified containing high levels of both detyrosinated and acetylated tubulin (Fig. 7A). Tubulin from CPA-treated cells looked similar to that of untreated cells when analysed on SDS-PAGE (Fig. 7B) and, when polymerised, all three types of tubulin formed MTs that were indistinguishable using enhanced video differential interference microscope (Fig. 7C). The MTs were used in in-vitro motility assays (Fig. 7D) together with a purified Kif5c-motor-domain construct. Movement of MTs in these assays was tracked and their velocity calculated.

The assays showed that the velocity of MTs depended on the type of tubulin (Fig. 7E). The velocity for detyrosinated MTs was significantly lower (0.58±0.07 μm second^–1, mean ± s.d., n=206; P<0.05) than for unmodified (tyrosinated) MTs (0.66±0.06 μm second^–1, n=103). Interestingly, the velocity of highly modified MTs from pig brain was significantly slower than that on detyrosinated MTs (0.55±0.05 μm second^–1, n=207, P<0.5). It has been reported previously (Liao and Gundersen, 1998) that purified kinesin-1 from squid binds with a 2.8-fold greater affinity to tyrosinated tubulin from HeLa cells than to detyrosinated tubulin (K_d= 29 nM and K_d=81 nM, respectively). We performed a similar binding assay to confirm that the binding of the Kif5c motor domain to tyrosinated and detyrosinated MTs from HeLa cells showed the same trend (data not shown). Without ATP, the Kif5c motor domain was incubated with taxol-stabilised tyrosinated and detyrosinated MTs from HeLa cells or MTs from pig brain. Samples were then centrifuged, and pellet and supernatant analysed by western blotting or SDS-PAGE. We found that ∼1.5 times more Kif5c was bound to tubulin from pig brain and to detyrosinated tubulin than to tyrosinated tubulin, confirming that Kif5c has a higher steady-state occupancy on detyrosinated than on tyrosinated MTs (data not shown).

To determine whether a small difference in velocity and the reported 2.8-fold greater binding affinity of kinesin-1 to detyrosinated MTs compared with tyrosinated MTs (Liao and Gundersen, 1998) is sufficient to bias transport of Kif5c towards modified MTs, a simple mathematical model was used. The model is a `hopping' model for the dynamics of kinesin-1 behaviour on a track of N+1 sites (Fig. 8A). It was used to predict the equilibrium distribution (stationary number density distribution of kinesin-1) termed c̄_n (for 0<n<N) for kinesin-1 along MTs, the average number density of kinesin-1 in the steady-state (c̄), the equilibrium average velocity (v̄) and the equilibrium average flux (j̄). This is a very general model that uses only a small number of parameters. It results in a statistical representation of the kinesin-1 distribution on an ensemble of MTs. Full details of the model, including assumptions made and equations used, are given in Materials and Methods.

The model predicts that an increase from 0% detyrosination to 5% detyrosination, similar to the levels we observed in COS-7 cells, increased the levels of bound kinesin-1 (stationary number density distribution of kinesin) by about 3% (Fig. 8B). The model also predicts that the density of kinesin-1 on MTs, the average flux of kinesin-1 along a MT and the average velocity depends on the degree of detyrosination (Fig. 8B). The change in stationary number density distribution of kinesin-1 as a result of detyrosination is small, and one possible reason for this is that the k_on value (or landing rate) in the model is independent of detyrosination. If k_on were to increase when tubulin is detyrosinated, the difference between detyrosinated and tyrosinated MTs in our model would increase further.

The results presented here show that GFP-Kif5c binds to and moves along MTs in live cells with binding affinities and velocities similar to those measured in vitro. Distribution of GFP-Kif5c is consistent with that of exogenously expressed Kif5b in fixed and immunostained COS-7 cells described by other investigators (Macioce et al., 2003), and also with the distribution of the kinesin-3 family member Kif1c expressed as a GFP fusion protein in macrophages (Kopp et al., 2006). The binding and movement of GFP-Kif5c along MTs is dependent on the presence of an active motor, because a rigor mutant GFP-Kif5cmut (T93N) was able to bind to, but not move along, MTs in the same system. Although GFP-Kif5c can bind to tyrosinated, detyrosinated and/or acetylated MTs in vivo, we found that it preferentially associated with detyrosinated MTs and, to a lesser extent, with acetylated MTs. Our in vitro work indicates that tubulin detyrosination directly influences the velocity of MTs moved by Kif5c.

Why does GFP-Kif5c preferentially traffic along detyrosinated MTs in vivo? A clue might lie in the results of our in vitro assays. We observed slower sliding of detyrosinated MTs (0.58 μm second^–1 vs 0.66 μm second^–1), indicating that detyrosination causes a slower stepping/ATP-turnover rate (72.5 second^–1 vs 82.5 second^–1, assuming 8-nm steps). This observation is consistent with a model in which detyrosination of MTs shifts the steady-state binding equilibrium of kinesin-1 heads towards strong binding by reducing the detachment rate of the weak binding kinesin.ADP intermediate. This would reduce the stepping rate and increase the steady-state occupancy of MTs. In current models for the kinesin-1 stepping mechanism, reducing the detachment rate in the weak-binding kinesin.ADP state will decrease the probability of detachment following a step, giving rise, therefore, to longer processive runs. Detyrosination would then result in a slower motor that made longer processive runs, and increase the steady-state occupancy on MTs, as we observed. Our mathematical modelling confirms this relationship between the detachment rate, the run length and the steady-state occupancy. An analogy might be that detyrosination gives the motor `magnetic boots', attaching it more reliably to the track, but making stepping correspondingly more difficult.

Detyrosination might also result in a slower but more consistent and sustained delivery of kinesin-1 cargo to specific intracellular sites. That GFP-Kif5c is capable of processive movement on heavily modified MTs in vivo was confirmed by our observations of motor movement along nocodazole-resistant MTs in drug-treated cells (supplementary material Movie 8).

In support of this idea, it has been shown previously that complete cleavage of the `E-hook' reduced the velocity of a fungal kinesin-1 from 1.6 μm second^–1 to 1.0 μm second^–1 (Lakamper and Meyhofer, 2005; Skiniotis et al., 2004). Cleavage of the E-hook also shifted the normally weak binding kinesin.ADP state towards strong binding. The E-hook is the glutamate-rich sequence at the C-terminus of both α and β tubulin that contributes to an electrostatic interaction between kinesin-1 and tubulin (Thorn et al., 2000). Both detyrosination and E-hook cleavage could produce their effects by strengthening the binding of the kinesin-ADP intermediate. If correct, this idea would also account for our observation of higher steady-state occupancy of detyrosinated MTs compared with tyrosinated MTs. Kinesin-1 is known to bind with increased affinity to detyrosinated MTs (Liao and Gundersen, 1998), and this would be expected if detyrosination tightens the binding of the kinesin.ADP intermediate.

Our work has focused on the importance of detyrosination, but other tubulin modifications probably also affect the interaction of kinesin-1 with MTs. Polyglutamination, another post-translational modification seen at the E-hook, influences the trafficking of Kif1a cargo in mouse brain (Ikegami et al., 2007). Acetylation also affects kinesin-1 trafficking. Drosophila kinesin-1 binds with a higher affinity and moves acetylated MTs from Tetrahymena faster (∼20%) than MTs isolated from mutants in which acetylation was prevented (Reed et al., 2006). This latter finding is intriguing, because acetylation occurs on the lysine residue at position 40, which is positioned on the internal face of α-tubulin and, therefore, within the lumen of the MT. This modification would not be expected to have a direct effect on binding affinity of kinesin-1, but could have an indirect effect by influencing the structure of the tubulin dimer. This might also mean that acetylation affects a different step in the kinesin ATPase cycle compared with detyrosination, and this might explain why detyrosination increases binding affinity and reduces velocity, whereas acetylation increases both binding affinity and velocity. Further work is needed to understand how these different tubulin modifications affect kinesin-1 binding and movement. However, it is possible that both modifications are important in directing kinesin-1 trafficking, which may be related to the large variety in MT acetylation and/or detyrosination between different cell types (Bulinski et al., 1988).

The results described here, in combination with work by previous investigators (Kreitzer et al., 1999; Liao and Gundersen, 1998), allow us to propose a biologically relevant explanation for the phenomenon of tubulin detyrosination. We propose that the targeting of MT-motor-dependent transport in vivo is the specific reason for this tubulin modification. Preferential trafficking along detyrosinated MTs could provide targeting cues for intracellular transport, such as those previously reported by others (Nakata and Hirokawa, 2003). Detyrosinated MTs in migrating cells penetrate into lamellipodia that extend in the direction of travel (Gundersen and Bulinski, 1988; Kupfer et al., 1982). Furthermore, signalling events that leading to the generation of detyrosinated MTs are mediated by the activity of a variety of effectors downstream of the small GTPase Rho (Cook et al., 1998; Palazzo et al., 2001). Recently, the final link in this pathway has been suggested to be dependent upon the APC tumour suppressor protein (Bienz, 2001). Consistent with this, co-immunostaining of transfected cells indicated that APC and GFP-Kif5c distributions extensively overlapped along MT distal segments as well as in clusters located at the cell periphery (S.D., M.P. and E.E.M., unpublished data). This observation supports the hypothesis that this signalling pathway – which terminates with APC-mediated MT stabilisation – provides directional cues for kinesin-1-dependent cargo transport.

In conclusion, our observations confirm by direct in vivo observation that the full-length heavy chain of kinesin-1 is associated with MTs where it forms part of a processive plus-end-directed motor complex that moves with a velocity comparable with that seen in vitro. Our results also define a mechanism that exploits post-translational tubulin modifications associated with MT stability to target kinesin-1 dependent transport to specific regions within cells. Finally, our data help to explain the 20-year old observation that vesicles in lobster axons are found only on a small proportion of the MTs, suggesting that distinct classes of MTs have distinct roles in either trafficking or maintaining structural integrity of the cell (Miller et al., 1987).

Rat anti-α-tubulin antibody (YL1/2) specific for the EEY epitope of tyrosinated MTs was obtained from Serotec. Rabbit anti-detyrosinated-tubulin antibody (T12) specific for the xEE epitope of α-tubulin was a kind gift from Holly Goodson, University of Notre Dame, IN (Kreis, 1987). Mouse anti-acetylated-tubulin antibody was obtained from Sigma-Aldrich. To amplify the GFP-Kif5c signal in fixed cells, anti-Kif5c (Abcam, Ltd) or anti-KHC (Chemicon International) were used. Secondary antibodies used for immunofluorescence were Alexa-Fluor-488, -594 and -647 conjugates (Molecular Probes). Nocodazole was obtained from Sigma-Aldrich and used at a final concentration of 2.5 μg ml^–1.

Two fluorescently tagged Kif5c fusion proteins were used in this study. GFP-Kif5c was constructed by cloning the full-length rat Kif5c cDNA (a kind gift from Scott Brady, University of Illinois, IL) into pEGFP-C3 (BD Biosciences), creating an N-terminally tagged GFP-Kif5c fusion protein. The rigor kinesin-1 mutant, GFP-Kif5cmut was created using an overlap PCR approach to introduce the T93N mutation (Nakata and Hirokawa, 1995). The mCherry-α-tubulin expression vector was a kind gift from Vic Small (Institute of Molecular Biotechnology, Vienna, Austria).

COS-7 cells were cultured as described previously (Morrison et al., 1998). The neuronal cell line Hdh^+/+ was cultured as described previously (Zuccato et al., 2003) Cells were transiently transfected with pGFP-Kif5c using GeneJuice (Novagen) according to the manufacturer's protocol.

For immunofluorescence studies cells were cultured on acid-washed glass coverslips (VWR International) coated with 0.02% gelatine (Sigma). Cells were either fixed in cold methanol (Morrison et al., 1998; Musa et al., 2003) or paraformaldehyde (Swailes et al., 2006), immunostaining was performed as described previously. Fluorescence imaging was performed using a deconvolution microscope (Deltavision) or confocal laser scanning microscope (LSM) 510 Meta (Zeiss).

Cells were grown in 35-mm glass-bottomed culture dishes (Iwaki brand; Asahi Techno Glass Corporation, Japan; supplied by Bibby Sterilin). 18-48 hours post-transfection the growth medium was exchanged for CO₂-independent medium (Invitrogen) supplemented with 10% foetal calf serum (FCS), 100 μg ml^–1 penicillin-streptomycin and 4 mM glutamine for imaging.

Imaging at the slower frame rate of one frame every 5 seconds was performed essentially as described previously (Langford et al., 2006). Cells were transferred to a Zeiss Axiovert 200 inverted microscope with the stage enclosed by a heated chamber (Solent Scientific, UK), and the temperature maintained at 37°C. Cells were examined by fluorescence microscopy using a Zeiss Plan Apochromat 63×/1.4NA oil-immersion lens. An excitation-emission filterset optimised for eGFP imaging was used (Chroma Technology Corp., Brattleboro, VT; filterset ID 86007). Time-lapse images were obtained using Ludl shutters and a Hamamatsu Orca II ER camera. Microscope, camera, filter wheels and shutters were controlled by Kinetic Imaging AQM 6 software.

To perform imaging at higher frame rates (two frames per second) or to perform multicolour time-lapse live cell imaging, cells were transferred to an Olympus microscope enclosed in a heated chamber (Solent Scientific, as above). Cells were imaged using an Olympus 60×/1.4NA oil-immersion lens, and camera, shutters and software were controlled using Softworx software package (Deltavision, USA). The image size was set to 256×256 pixels, using 1×1 binning to obtain the highest resolution of puncta. Only sequences captured at this faster frame rate and at this higher spatial resolution were used in the tracking analysis to calculate GFP-Kif5c velocities. Particle tracking analyses were conducted on original datasets using Kinetic Imaging Motion Analysis software, or using GMview (kindly provided by Gregory Mashanov, NIMR, London, UK; (Mashanov et al., 2004). GFP-Kif5c puncta were tracked from frame to frame. When puncta temporarily stopped moving, velocity dropped to zero; these phases were counted as `stationary'. The average velocity for each punctum was calculated either by including these stationary phases or by excluding them. As expected, excluding them increased velocities slightly. Time-lapse image series were cropped and converted into Quicktime or AVI movies using Image J, Adobe ImageReady 7, Photoshop CS3 or Fast Movie Processor. Some movies are presented using an inverted greyscale colour look-up table to increase perceived contrast.

Tubulin purified from HeLa cells (Cytoskeleton) was modified to produce pure populations of detyrosinated and tyrosinated subunits as described (Chapin and Bulinski, 1991). Dot-blot assays were used to assess the purity of the tyrosinated and detyrosinated MTs, and to determine the extent of post-translational modifications of tubulin purified from pig brain. All tubulin was polymerised according to the methods of Lockhart and Cross (Lockhart and Cross, 1996) and MTs were resuspended in BRB 80 (80 mM PIPES-KOH pH 6.9, 1 mM MgCl₂, 1 mM EGTA) containing 20 μM taxol. The kinesin-1-motor-domain construct used was K430 [430aa; as described in Crevel et al. (Crevel et al., 1997)]. Flow cells were constructed by sealing together two ethanol-washed glass coverslips using two parallel strips of silicone grease (Dow Corning). An aliquot of 5 μM Kif5c-motor-domain diluted in BRB 80 was washed into the flow cell. The kinesin solution also contained 100 μg ml^–1 casein as surface block. The motor was allowed to adsorb to the surface for 5 minutes at room temperature. Non-adsorbed Kif5c-motor-domain was washed out with 2-3 flow cell volumes of BRB 80 supplemented with 5 mM DTT and 20 μM taxol. Two flow cell volumes of MTs (∼2 μM) in BRB 80 with 1 mM DTT, 20 μM taxol and 1 mM ATP were then allowed to flow through. The slide was immediately transferred to the microscope and MT gliding was observed by video-enhanced differential interference microscopy (VEDIC). The measurement of MT velocities was performed using Nick Carter's (MCRI, Oxted, Surry, UK) freeware RETRAC program (http://mc11.mcri.ac.uk/retrac.html) for image capture and tracking.

Details of the model and the equations used are given below. To solve the equations detailed there, a C code was written that solves equations 1, 2, 3 by the Euler method. Given the solution of the number distribution, the code then computes the flux and the velocity distributions, as well as the average values of the number density, the flux and the velocity at any given time (equations 7, 8, 9). The code has been run for long enough to obtain the steady-state solution (equations 10, 11, 12). Given the steady number density distribution, the code computes the steady-state average number-density, average flux and average velocity as a function of the tyrosination f (equations 13, 14, 15).

Kinesin 'walks' along the MT – taking unit steps – in a stochastic manner starting from the minus end of the filament (site n=0; Fig. 8A) towards the plus-end of the MT (site n=N; Fig. 8A). The starting point of our model is that detyrosination of the tubulin dimers results in a stronger binding of kinesin as the reported K_d decreases from 81 nM to 29 nM (Liao and Gundersen, 1998), and in a slower rate of motion along the MT by kinesin molecules (as described here).

Our experimental data show that, the fraction of MT-binding sites that are tyrosinated is 95%, kinesin moves along MTs with a velocity of 0.66 μm second^–1 on tyrosinated MTs and at 0.58 μm second^–1 on detyrosinated MTs. Furthermore it has been shown previously that the K_d for detyrosinated MTs (29 nM) is about 3 times lower than that for tyrosinated MTs (81 nM) (Liao and Gundersen 1998). Finally, the on-rate is diffusion limited and was described as 2×10⁷ M^–1 second^–1, the detachment rate as 20 second^–1 (Johnson and Gilbert, 1995, and Table 1 within), and the forward rate of motion for a kinesin molecule on tyrosinated MTs is ∼82.5 second^–1, given the size of the steps (8 nm) (Svoboda et al., 1993) and its velocity (0.66 μm second^–1).

Given these experimental data, a mean field model (Edelstein-Keshet, 2005) is an appropriate description of the kinetics of kinesin walking on a MT, where a fraction f of the sites are tyrosinated and (1–f) are detyrosinated. According to our experiments f is 95%. We introduce k^t₊ and k^g₊, the forward rates for fully tyrosinated and fully detyrosinated MTs, respectively. We have a=k^g₊/k^t₊=0.58/0.66=1/1.379 (from the two measured velocities of kinesin movement on tyrosinated and detyrosinated MTs). We have introduced the parameter a=1/1.379 that implies slower forward walking along detyrosinated MTs. For a partially detyrosinated track, the forward rate is given by k₊=f k^t₊+(1–f)k^g₊=k^t₊[f+a(1–f)] with k^t₊=82.5 second^–1. The on-rate is given by k_onϕ, with k_on=2×10⁷ M^–1 second^–1 and ϕ=10^–7 M. The detachment rate, k_off depends on f as follows. k_off is approximately three times lower for fully detyrosinated than for fully tyrosinated MTs (Liao and Gundersen, 1998). We can therefore write this in the mean field description k_off=fk^t_off+(1–f), where k^t_off is the detachment rate for fully tyrosinated MTs and is equal to k^t_off =20 second^–1. Finally, the backward rate of motion can be parametrised in terms of the bias b=k₊/k_– so that k_–=b^–1k₊, which we have set to 100. The probability that kinesin motors take a backwards step is very low, but increasing the bias to 1000 or even larger only has very small effects on the results presented here.

We also considered whether we can model the effects of acetylation. It has been recently reported (Reed et al., 2006) that acetylation affects the binding and velocity of conventional kinesin. Lack of acetylation reduces binding substantially, to ∼10% of wild-type levels, and velocity to ∼20% of wild-type levels. However, because the extent of detyrosination of the MTs used in those experiments was not measured, it is unclear how this may have affected the results.

This work was funded by the Wellcome Trust, Cancer Research UK and the Marie Curie Foundation. S.D. was an MRC-funded PhD student. We thank Holly Goodson for antibodies used in this work, and Nikolai Belyaev for providing us with the neuronal cell line.