PTPN21 and Hook3 relieve KIF1C autoinhibition and activate intracellular transport

The kinesin-3 KIF1C is a fast organelle transporter implicated in the transport of dense core vesicles in neurons and the delivery of integrins to cell adhesions. Here we report the mechanisms of autoinhibition and release that control the activity of KIF1C. We show that the microtubule binding surface of KIF1C motor domain interacts with its stalk and that these autoinhibitory interactions are released upon binding of protein tyrosine phosphatase PTPN21. The FERM domain of PTPN21 stimulates dense core vesicle transport in primary hippocampal neurons and rescues integrin trafficking in KIF1C-depleted cells. In vitro, human full-length KIF1C is a processive, plus-end directed motor. Its landing rate onto microtubules increases in the presence of either PTPN21 FERM domain or the cargo adapter Hook3 that binds the same region of KIF1C tail. This autoinhibition release mechanism allows cargo-activated transport and might enable motors to participate in bidirectional cargo transport without undertaking a tug-of-war.

I ntracellular transport is essential for cell polarity and function. Long-distance transport of cellular cargo is mediated by microtubule-based motors, dynein and kinesin. While dynein is the main transporter towards the minus end of microtubules, most kinesins walk towards the microtubule plus end 1 . However, many cargoes carry motors of both directionality in order to allow directional switching when they encounter a roadblock and the relative activity of the opposite directionality motors determines the net progress of the cargo towards the cell periphery (where usually most plus ends are located) or towards the cell centre (where microtubule minus ends are abundant) 1,2 . We have previously identified the kinesin-3 KIF1C as the motor responsible for the transport of α5β1-integrins. The delivery of integrins into cellular processes such as the tails of migrating cells allows the maturation of focal adhesion sites 3 . KIF1C is also required for the formation and microtubule-induced turnover of podosomes-protrusive actin-based adhesion structures-in both macrophages and vascular smooth muscle cells [4][5][6] . Moreover, KIF1C contributes to MHC class II antigen presentation, Golgi organisation and transport of Rab6-positive secretory vesicles 7,8 . In neurons, KIF1C transports dense-core vesicles both into axons and dendrites and appears to be the fastest human cargo transporter 9,10 . Consistently, human patients with missense mutations in KIF1C resulting in the absence of the protein suffer from spastic paraplegia and cerebellar dysfunction 11 . Mutations that result in reduced motor function also cause a form of hereditary spastic paraplegia 12 .
KIF1C-dependent cargo moves bidirectionally even in highly polarised microtubule networks and depletion of KIF1C results in the reduction of transport in both directions 3 , suggesting that KIF1C cooperates with dynein as had been suggested for other kinesin-3 motors 13 . To begin to unravel how KIF1C contributes to bidirectional cargo transport, we need to identify the mechanisms that switch KIF1C transport on and off. Most kinesin-3 motors are thought to be activated by a monomer-dimer switch, whereby cargo binding releases inhibitory intramolecular interactions of neck and stalk regions by facilitating the dimerisation of the neck coil and other coiled-coils regions in the tail [14][15][16][17] . The motor thereby transitions from an inactive, diffusive monomer to a processive dimer. In the alternative tail-block model, the motors are stable dimers, but regions of the tail interact with the motor or neck domains and interfere with motor activity until cargo binding occupies the tail region and releases the motor [18][19][20] .
Here we show that KIF1C is a stable dimer that is autoinhibited by intramolecular interactions of the stalk domain with the microtubule binding interface of the motor domain. We demonstrate that protein tyrosine phosphatase N21 (PTPN21) activates KIF1C by binding to the stalk region. This function does not require catalytic activity of the phosphatase, and its N-terminal FERM domain alone is sufficient to stimulate the transport of KIF1C cargoes in cells as well as increasing the landing rate of KIF1C on microtubules in vitro. The cargo adapter Hook3 binds KIF1C in the same region and activates KIF1C in a similar way, suggesting that both cargo binding and regulatory proteins might contribute to the directional switching of KIF1C-dynein transport complexes.

Results
KIF1C is an autoinhibited dimer. To determine the mechanism of KIF1C regulation, we first aimed to determine its mode of autoinhibition. The monomer-dimer-switch and tail-block models can be distinguished by determining the oligomeric state of the motor. Thus we performed classic hydrodynamic analysis using glycerol gradients and size exclusion chromatography on both purified recombinant full-length human KIF1C-GFP from insect cells and KIF1C-Flag in human cell lysate ( Fig. 1a-c, Supplementary Fig. 1). The sedimentation coefficient and Stokes radius were determined in comparison to standard proteins. We find that the apparent molecular weight determined both at physiological levels of salt (150 mM) and in high (500 mM) salt buffer is consistent with KIF1C being a dimer (KIF1C-GFP: apparent MW = 272 ± 57 kDa, expected dimer MW = 308 kDa; KIF1C-Flag: apparent MW = 178 ± 45 kDa, expected dimer MW = 251 kDa, all errors are SEM based on 40  errors of calibration curve fit parameters and estimated precision of peak location as 1/2 of fraction size). Consistent with this, the majority of KIF1C-GFP molecules bound to microtubules in the presence of non-hydrolysable AMPPNP show two bleach steps and the distribution of bleach steps found fits best to a simulation of 88% dimers and 12% tetramers (Fig. 1d, e), assuming that about 80% of GFPs are active, which is realistic based on previous findings 21 . Interestingly, KIF1C elongates at increasing salt concentrations from a moderately elongated conformation (frictional coefficient of 1.5) at physiological salt to highly elongated (frictional coefficient = 1.9) at 500 mM (Fig. 1c), suggesting that intramolecular electrostatic interactions might hold the KIF1C dimer in a folded, autoinhibited state.
To determine the interaction surfaces involved in a possible autoinhibited state, we performed crosslink mass spectrometry. Purified KIF1C was treated with the 11 Å crosslinker BS3 or with the zero length crosslinker EDC, digested with trypsin and then subjected to tandem mass spectrometry analysis. Crosslinked peptides were identified using StavroX 22 . Only crosslinked peptides whose identity could be verified by extensive fragmentation with none or few unexplained major peaks were retained (Supplementary Figs. 2 and 3). These high-confidence crosslinks  were between K273-K591, K273-K645, K464-K645, K640-K645,  K464-E606 and E644-K854 (Fig. 2a, Supplementary Figs. 2 and 3), showing that the end of the FHA domain and the third coiled-coil contact the motor domain near K273. This residue is within alpha-helix 4 at the centre of the microtubule interaction interface of the motor domain 23 (Fig. 2b). In the presence of 500 mM NaCl, the abundance of the stalk-to-motor domain crosslinks was dramatically reduced consistent with the observed elongation of KIF1C dimer in glycerol gradients ( ARTICLE compact conformation in physiological conditions whereby the stalk domain blocks the motor domain from interacting with microtubules ( Fig. 2c). To confirm this intramolecular interaction and the autoinhibition directly, we purified bacterially expressed KIF1C motor domain and KIF1C stalk domains (Fig. 2d, e). Using microscale thermophoresis, we measured the affinity of motor and stalk domains as 1 µM (Fig. 2f). Consistent with the idea that the stalk competes with microtubule binding, we found that microtubule decoration by the KIF1C motor domain was significantly reduced in the presence of 1 µM stalk domain, but not when 1 µM GST was added as control (Fig. 2g, h). Thus KIF1C is a stable dimer that blocks its microtubule-binding interface with an intramolecular interaction in its autoinhibited conformation.
PTPN21 stimulates transport of KIF1C cargoes. We next aimed to identify a regulator that could engage with the stalk domain and thereby activate the motor. Two proteins had previously been shown to interact with the KIF1C stalk domain, non-muscle Myosin IIA 4 and protein tyrosine phosphatase N21 (PTPN21, also known as PTPD1) 24 (Fig. 2a). We reasoned that a KIF1C activator would be required for KIF1C function in cells and thus would phenocopy KIF1C depletion. As a readout of diminished KIF1C activity, we used the reduced podosome number in vascular smooth muscle cells that we reported previously 5 . Myosin IIA inhibition did not phenocopy KIF1C depletion. Instead, inhibition of Myosin IIA with Blebbistatin or indirectly via inhibition of Rho kinase using Y27632 did not result in reduced podosome number, while remaining actin stress fibres were efficiently removed (Fig. 3a-d). As depletion of Myosin IIA did also have opposite effects on the stability of cell tails and directional persistence of cell migration we previously reported for KIF1C depletion 3 , we excluded Myosin IIA as being involved in KIF1C activation. In contrast, PTPN21 depletion using siRNA resulted in a dramatic reduction of podosome number (Fig. 3e-h). To confirm specificity of the RNAi we rescued the phenotype with HA-tagged PTPN21. Interestingly, the catalytically inactive mutant PTPN21 C1108S (ref. 25 ) could also fully rescue the PTPN21 depletion phenotype (Fig. 3e, f), suggesting that a scaffold function is required rather than phosphatase activity. Strikingly, expression of PTPN21 C1108S also efficiently compensated for partial depletion of KIF1C and fully rescued podosome formation (Fig. 4a-c). An N-terminal 378 amino acid fragment of PTPN21, which contains a FERM domain, was previously shown to be sufficient to interact with KIF1C 24 . Therefore we tested this construct, and found that the PTPN21 FERM domain alone was sufficient to rescue the KIF1C depletion phenotype (Fig. 4a-c). This suggests that PTPN21 scaffolding could activate the remaining pool of KIF1C, which is reduced to about 25-30% in cells treated with siKIF1C-2 compared to control cells 3,5 . Alternatively, PTPN21 could also activate another kinesin to compensate for KIF1C depletion. KIF16B, another kinesin-3, was an obvious candidate for this as its interaction with PTPN21 FERM domain had been shown already 26 . To discriminate between these two possibilities, we depleted KIF1C and KIF16B individually and simultaneously in PTPN21 FERM rescue experiments. KIF16B depletion on its own resulted only in a mild reduction in podosome numbers and expression of FERM domain significantly increased the number of podosomes formed in control cells and in those where either of the kinesins was depleted. However, PTPN21 FERM did not increase podosome numbers in cells that were depleted for both KIF1C and KIF16B (Fig. 4d, Supplementary Fig. 4). These findings suggest that PTPN21 rescue depends on the presence of a sizeable pool of either KIF1C or KIF16B. To explore this activity further, we investigated more directly whether PTPN21 stimulates KIF1Cdependent transport processes. We have shown previously that KIF1C is required for the bidirectional transport of integrincontaining vesicles in migrating RPE1 cells 3 . The depletion of KIF1C results in a reduction of both plus-end and minus-enddirected transport and an increase in stationary vesicles. Expression of either wild-type PTPN21, the catalytically inactive PTPN21 C1108S mutant or PTPN21 FERM reactivated integrin vesicle transport in KIF1C-depleted cells (Fig. 4e, f). To test whether the ability of PTPN21 FERM to activate KIF1C-dependent transport universally applies, we next isolated primary hippocampal neurons and observed the transport of dense-core vesicles labelled with NPY-RFP in the presence of either a control plasmid or pFERM. Consistent with a function to activate KIF1Cdependent transport, we observed a dramatically increased frequency of vesicles moving in anterograde direction through the neurites in cells expressing the FERM domain of PTPN21 ( Fig. 4g-i).
PTPN21 relieves KIF1C autoinhibition. Our experiments in cells identified PTPN21 as a potential activator of KIF1C. To confirm that the PTPN21 FERM domain can directly activate KIF1C, and to further interrogate the mechanism of activation, we reconstituted KIF1C motility in vitro. Hilyte647-and biotinlabelled microtubules were assembled in the presence of GTP, then stabilised with Taxol and attached to a PLL-PEG-biotincoated coverslip. Six hundred picomolar KIF1C was added in the presence of 1 mM ATP and an ATP-regenerating system. At 25°C, individual KIF1C-GFP dimers moved processively towards the plus end of the microtubule where they accumulated (Fig. 5a). The average speed and run length were 0.45 µm/s and 8.6 µm respectively (Fig. 5d). We then purified recombinant PTPN21 FERM -6His and for comparison Ezrin FERM -6His from Escherichia coli (Fig. 5c). Consistent with being an activator, the landing rate of KIF1C motors increased by about 40% in the presence of PTPN21 FERM (Fig. 5b-d). The frequency of observing running motors was also increased by 40%. Ezrin FERM , acting as a negative control, did not significantly affect the landing rate or the frequency of running motors ( Fig. 5b-d). These findings are consistent with the idea that PTPN21 opens the KIF1C motor by binding to its tail domain and thereby relieves autoinhibition and increases the binding rate of the motor.
If PTPN21 relieves the autoinhibition, the following two predictions should hold true: (i) PTPN21 FERM binding should remove the interactions between stalk and motor domain of KIF1C and (ii) deleting the region required for autoinhibition should result in a dominant-active motor that cannot be further activated by binding of PTPN21 FERM . To test the first prediction, we performed crosslinking mass spectrometry experiments with KIF1C and PTPN21 FERM . We found two new crosslinks, but could no longer find the intramolecular crosslinks in KIF1C (Fig. 5e, f). The new crosslinks show interactions of the FERM domain with lysines 591 and 645 in KIF1C ( Supplementary  Fig. 5). Because the crosslinks between motor domain and stalk were absent in the presence of PTPN21 FERM , the relative abundance of the un-crosslinked LKEGANINK peptide at the microtubule binding interface of KIF1C increased more than twofold in the KIF1C + PTPN21 FERM sample compared to KIF1C alone (Fig. 5e). These data confirm that the presence of the FERM domain is capable of freeing the motor domain, and relieving the autoinhibition of KIF1C. To test the second prediction, we generated two deletion mutants, KIF1CΔCC3 in which we removed coiled-coil 3, which makes the predominant contact with the motor domain in the autoinhibited conformation and KIF1CΔS where we removed coiled-coil 3 and the stalk region previously identified as minimal PTPN21-binding domain (Fig. 6a). First, we assessed whether either or both constructs are hyperactive by localisation of the construct in cells. We had described previously that KIF1C accumulates specifically in tails of migrating RPE1 cells 3 . Expression of KIF1CΔCC3-GFP and KIF1CΔS-GFP in RPE1 cells resulted in accumulation of the construct in cell tails and also elsewhere at the periphery of the cell (Fig. 6b). As tail accumulation is sensitive to tail state, i.e. whether it is forming or retracting 3 , we used KIF1C-mCherry as an internal control. In comparison to KIF1C-GFP, which was enriched in tails to a similar extent as KIF1C-mCherry (tail: cytoplasm ratio 6.1 ± 1.3 for KIF1C-mCherry and 7.4 ± 1.2 for KIF1C-GFP, n = 26 cells, errors are SEM), KIF1CΔCC3-GFP was almost threefold more enriched (tail:cytoplasm ratio 3.9 ± 0.6 for KIF1C-mCherry and 13.6 ± 2.1 for KIF1CΔCC3-GFP, n = 32 cells), and KIF1CΔS-GFP was fourfold more enriched (tail: cytoplasm ratio 1.8 ± 0.3 for KIF1C-mCherry and 7.7 ± 0.8 for KIF1CΔS-GFP, n = 39 cells) (Fig. 6c). Secondly, we purified recombinant KIF1CΔCC3-GFP and KIF1CΔS-GFP from insect cells (Fig. 6d) and performed single-molecule assays. KIF1CΔCC3-GFP had a threefold higher landing rate compared to wild-type KIF1C-GFP, and addition of PTPN21 FERM did not further increase the landing rate of KIF1CΔCC3-GFP ( Fig. 6e-g, k). KIF1CΔS-GFP had a 20-fold higher landing rate than KIF1C-GFP and also moved at about threefold higher average speed ( Fig. 6h-m). Therefore, these experiments confirm that binding of PTPN21 to the stalk domain of KIF1C relieves the autoinhibition of KIF1C and thereby enables the motor to engage with microtubules.
Hook3 binds to KIF1C stalk and also activates KIF1C. Finally, we wondered how universal this mechanism is and whether other proteins binding KIF1C stalk could activate KIF1C in a similar manner. To identify KIF1C stalk interactors we performed a BioID experiment with full-length KIF1C and KIF1CΔS (Fig. 7a). We identified 43 proteins or clusters of proteins isolated with streptavidin-beads upon biotinylation that were significantly enriched in either the KIF1C-BioID2 or KIF1CΔS-BioID2 samples. Eight of those were enriched more than 10-fold in the fulllength sample. The most significant hit was Hook3 (Fig. 7b   appeared as a possible KIF1C stalk interactor. We confirmed that both Hook3 and PTPN21 bind to KIF1C in a stalk-dependent manner using co-immunoprecipitation from HEK293 cells (Fig. 7c, Supplementary Fig. 6). Hook3 has previously been identified as an activator of dynein/dynactin-an activity requiring its N-terminal globular domain and all three coiled-coil domains [27][28][29][30] . In contrast, the C-terminus of Hook3 was shown to interact with KIF1C 31 . To test whether Hook3 could also activate KIF1C, we purified recombinant human Hook3 with a Cterminal SNAP-tag from insect cells (Fig. 7d), labelled it with Alexa647 and performed single-molecule assays with KIF1C.
In the presence of Hook3, the landing rate of KIF1C increased twofold (Fig. 7e, f), thus confirming that Hook3 functions as an activator similarly to PTPN21 by binding to the KIF1C stalk region adjacent to the site engaged in intramolecular interactions of KIF1C during autoinhibition. As we could observe that Hook3 was co-transported with KIF1C (Fig. 7e), we asked whether PTPN21 also maintained contact with KIF1C after activating it. We purified and labelled FERM-SNAPf with Alexa647 similarly to our Hook3 construct and performed two-colour TIRF assays both with KIF1C-GFP and KIF1CΔS-GFP. Both activators were co-transported with full-length KIF1C, but not with KIF1CΔS ( Fig. 8a-d). This suggests that both PTPN21 and Hook3 act as scaffolds to activate KIF1C by engaging with its stalk region (Fig. 8e, f).

Discussion
Our findings support a model whereby KIF1C is a stable dimer that is held in an autoinhibited conformation by interaction of its stalk region including the third coiled-coil domain with the microtubule binding surface of the motor domain (Fig. 8e). Autoinhibition is relieved upon binding of PTPN21 FERM domain or Hook3, allowing the motor domain to engage with microtubules ( Fig. 8f). While the model agrees with the finding that KIF1C motors are dimeric in cells 32 , this is in contrast to the mode of autoinhibition described for other kinesin-3 motors, KIF1A, Unc104, KIF16B, KIF13A and KIF13B that all undergo a monomer-dimer transition [14][15][16][17] . However, interactions of the stalk or tail region with the motor domain have also been described for KIF13B and KIF16B 19,20 . These might act as a second layer of activity control once dimers are formed or stabilise the inhibited monomer conformation. Our data suggest that KIF1C autoinhibition acts by steric blockage of the microtubule binding. In contrast, kinesin-1, which is also autoregulated by a tail-block mechanism, is inhibited by crosslinking of the motor domains that prevent movement required for neck linker undocking 33,34 .
We identify the phosphatase PTPN21 as a KIF1C activator. The first 378 amino acids containing the FERM domain are sufficient to activate KIF1C in vitro and in cells and does neither require the catalytic activity nor the phosphatase domain of PTPN21. Even though a scaffold activity is sufficient, it is still possible that PTPN21-mediated dephosphorylation of KIF1C 24 can further modulate the activity of KIF1C, modify KIF1C cargo, cargo adapters or the activity of adjacent motors, a possibility that will be interesting to explore in a future study. PTPN21 has been shown to dynamically localise to focal adhesions 35 or EGF receptor recycling sites 36 . It is possible that KIF1Cmediated transport facilitates the efficient turnover of PTPN21 at these sites. Thus the phosphatase could be both a cargo and a regulator of KIF1C. We have implicated KIF1C previously in the transport of integrins required for the maturation of trailing focal adhesions 3 . It is possible that additionally, KIF1C transport facilitates PTPN21-mediated regulation of Src tyrosine kinase and FAK signalling that promote cell adhesion and migration 35 .
PTPN21 FERM was able to compensate for depletion of KIF1C by activating KIF16B, a highly processive early endosome transporter that has been shown to also interact with PTPN21 FERM 15,26,37 . The expression of the FERM domain efficiently stimulated dense-core vesicle transport in primary neurons. PTPN21 has been associated with schizophrenia in a genomewide association study 38 and shown to promote neuronal survival and growth 39,40 . While the latter is thought to occur via NRG3 or ERK1/2 signalling, neuronal function and morphology is compromised when microtubule transport is perturbed and the function of PTPN21 as neuronal transport regulator might contribute to the complications in schizophrenia patients with PTPN21 mutations. Further, cAMP/PKA pathway stimulation of kinesin-1 has been shown to reverse aging defects in Drosophila neurons, and a similar age-driven decrease in KIF1C transport of dense-core vesicles and other organelles may have similar effects in neurodegenerative diseases 41 .
The findings that Hook3 can activate both dynein/ dynactin 27,29,30 and KIF1C (this study), and that the binding sites for these opposite directionality motors are non-overlapping 29,31 , suggests that Hook3 could simultaneously bind to KIF1C and dynein/dynactin and provide a scaffold for bidirectional cargo transport. Evidence for the existence of a complex of dynein/ dynactin, KIF1C and Hook3 has recently been provided in a preprinted manuscript 42 . We note that this study did not report an activation of KIF1C upon binding of Hook3; however, this is based solely on the analysis of speed and run lengths, while we find that activation primarily increases KIF1C landing rates. How the directional switching would be orchestrated in such a KIF1C-DDH complex is an exciting question for the future. It is important to note that Hook3 is not the only dynein cargo adapter which binds KIF1C. BICDR1 has been shown to bind to the proline-rich C-terminal region of KIF1C 9 , and BICD2 appears to interact with KIF1C biochemically 43 . Whether BICDR1 or BICD2 are able to activate the motor is unclear, but it is possible that different adapters not only mediate linkage to a different set of cargoes, but also recruit opposite polarity motors in different conformations and thus relative activity. For dynein/ dynactin, such a difference is seen in BICD2 recruiting only one pair of dynein heavy chains while BICDR1 and Hook3 recruit two pairs and thus are able to exert higher forces 28 . BICDR1 also binds Rab6 and recruits both dynein/dynactin and KIF1C to participate in the transport of secretory vesicles 9 . Rab6 in turn has been shown to bind and inhibit the KIF1C motor domain 7 .
This could provide a potential mechanism for a second layer of regulatory control of KIF1C activity to facilitate its minus enddirected transport with dynein-dynactin-Hook3.
Taken together, we provide mechanistic insight into the regulation of KIF1C, a fast long-distance neuronal transporter. We show that KIF1C is activated by a scaffold function of PTPN21 and the dynein cargo adapter Hook3, but the mechanism of autoinhibition release described here is likely to be universal and we expect cargoes and further scaffold proteins binding to the stalk region around the third coiled-coil in KIF1C to also activate the motor and initiate transport along microtubules. This opens
RNAi-protected PTPN21 was made using pHA-PTPN21-WT as a template in a three-step mutagenesis PCR using AS397, AS380 and AS264. The fragment containing the mutation was replaced in pHA-PTPN21-WT using NheI to generate pHA-PTPN21 RIP . RNAi-protected inactive PTPN21, pHA-PTPN21 RIP C1108S was generated by replacing the fragment containing mutation C1108S in pHA-PTPN21 RIP using BsgI.
pFERM expressing HA-tagged PTPN21 1-377 was generated by digesting pHA-PTPN21 RIP and pEGFP-N1 (Clontech) with HindIII and inserting the fragment in pEGFP-N1. As the GFP is not in frame, the FERM domain is expressed from this plasmid with an N-terminal HA-tag only.
To express and purify FERM domains from E. coli, pET22b-HA-PTPN21 1-381 -6His was generated by amplifying pHA-PTPN21 RIP by PCR with primers AS558 and AS592, digesting with NdeI and NotI and replacing EB1 in pET22b-mEB1-6His 47 . For labelling FERM domain, SNAPf-26 was codon optimised for E. coli, synthesised, and ligated into a PCR cloning vector generating pCAPs-SNAPf-26 with the addition of 5′ and 3′ NdeI and KpnI sites. The HA tag in pET22b-HA-PTPN21 1-381 -6His was exchanged for SNAPf-26 in a three-fragment ligation where the backbone was digested with PvuI and NdeI or PvuI and KpnI, and the SNAPf-26 insert was digested with NdeI and KpnI.
Hook3 was cloned by PCR from random-primed cDNA reverse transcribed from RPE1 RNA using primers AS690 and AS691, which was incorporated between the EcoRI and BamHI site of pKIF1C RIP2 GFP, replacing the KIF1C and generating pKan-CMV-Hook3-GFP. A multi-tag version of pFastBacM13 (Invitrogen) was created by ligating an insect-cell codon optimised synthesised cDNA for 8His-ZZ-LTLT-BICDR1-SNAPf between the RsrII and NotI sites creating pFastBacM13-8His-ZZ-LTLT-BICDR1-SNAPf. Hook3 was cloned into this by PCR from random-primed cDNA reverse transcribed from RPE1 RNA using Anti-Hook3 K IF 1 C -F la g K IF 1 C -F la g G F P -F la g G F P -F la g K IF 1 C Δ S -F la g K IF 1 C Δ S -F la g  Recombinant protein purification. For baculovirus expression, pFastBac-M13-6His-KIF1CGFP, pFastBac-M13-6His-KIF1CΔCC3-GFP, pFastBac-M13-6His-KIF1CΔS-GFP and pFastBacM13-8His-ZZ-LTLT-Hook3-SNAPf plasmids were transformed into DH10BacYFP competent cells 48 and plated on LB-Agar supplemented with 30 µg/ml kanamycin (#K4000, Sigma), 7 µg/ml gentamycin (#G1372, Sigma), 10 µg/ml tetracycline (#T3258, Sigma, 40 µg/ml Isopropyl β-D-1thiogalactopyranoside (IPTG, #MB1008, Melford) and 100 µg/ml X-Gal (#MB1001, Melford). Positive transformants (white colonies) were screened by PCR using M13 forward and reverse primers for the integration into the viral genome. The bacmid DNA was isolated from the positive transformants by the alkaline lysis method and transfected into SF9 cells (Invitrogen) with Escort IV (#L-3287, Sigma) according to the manufacturer's protocols. After 5-7 days, the virus (passage 1, P1) is harvested by centrifugation at 300 × g for 5 min in a swing out 5804 S-4-72 rotor (Eppendorf). Baculovirus infected insect cell (BIIC) stocks were made by infecting SF9 cells with P1 virus and freezing cells before lysis (typically around 36 h) in a 1°c ooling rate rack (#NU200 Nalgene) at −80°C. P1 virus was propagated to passage 2 (P2) by infecting 50 ml of SF9 (VWR, #EM71104-3) culture and harvesting after 5-7 days as described above 49 . For large-scale expression, 500 ml of SF9 cells at a density of 1-1.5 × 10 6 cells/ml were infected with one vial of BIIC or P2 virus. Cells  were equilibrated with the lysis buffer and the cleared lysate obtained is mixed with the beads and batch bound for 1 h. Next, the beads were loaded onto a 5 ml disposable polypropylene gravity column (#29922, Thermo Scientific) and washed with at least 10 CV SP wash buffer (50 mM sodium phosphate pH 7.5, 150 mM NaCl) and eluted with SP elution buffer (50 mM sodium phosphate pH 7.5, 300 mM NaCl). The peak fractions obtained were pooled and diluted with Ni-NTA lysis buffer (50 mM sodium phosphate pH 7.5, 150 mM NaCl, 20 mM Imidazole, 10% glycerol) and batch bound to Ni-NTA beads (#30230, Qiagen) for 1 h. The beads were loaded onto a gravity column and washed with at least 10 CV of Ni-NTA wash buffer (50 mM sodium phosphate pH 7.5, 150 mM NaCl, 50 mM Imidazole and 10% glycerol) and eluted with Ni-NTA elution buffer (50 mM sodium phosphate pH 7.5, 150 mM NaCl, 150 mM Imidazole, 0.1 mM ATP and 10% glycerol). The peak fractions were run on a SDS-PAGE gel for visualisation and protein was aliquoted, flash frozen and stored in liquid nitrogen. Hook3-647 was purified and labelled in a two-step process utilising both the His and ZZ affinity tags at 4°C. A pellet corresponding to 500 ml insect culture was resuspended in 40 ml Adaptor Lysis Buffer (50 mM HEPES pH 7.2, 150 mM NaCl, 20 mM imidazole, 1× cOmplete protease inhibitor cocktail, 1 mM DTT) and lysed with 20 strokes in a douncer. Lysates were cleared of insoluble material by centrifugation at 50,000 × g for 40 min in a Sorvall T-865 rotor. The cleared lysate was bound to 2 ml Ni-NTA beads for 2 h, and subsequently washed with 200 CV of lysis buffer followed by 200 CV lysis buffer supplemented with 60 mM imidazole. Protein was eluted in 5 CV of lysis buffer containing 350 mM Imidazole, and the eluate was batch bound to 1 ml IgG Sepharose (#17-0969-01, GE Healthcare Life Sciences) for 2 h. Beads were briefly washed with TEV cleavage buffer (50 mM Tris-HCl pH 7.4, 148 mM KAc, 2 mM MgAc, 1 mM EGTA, 10% (v/v) glycerol, 1 mM DTT) before being collected in a 1.5 ml Eppendorf tube and incubated with 3.5 µM SNAP-Surface Alexa Fluor 647 substrate (#S9136S, New England Biolabs) for 2 h. Beads were washed with 200 CV TEV cleavage buffer and resuspended in a 1.5 ml Eppendorf tube containing 50 µg/ml TEV protease, which was allowed to digest the protein off of beads overnight at 4°C. Finally, eluate was collected from the beads, concentrated to between 5 and 10 µM in a Centrisart I 20 kDa ultrafiltration device (#13249-E, Sartorius), flash frozen and stored in liquid nitrogen. Labelling ratio was 1.05 for Hook3-647.
Hydrodynamic analysis. Size exclusion chromatography was carried out using Superdex 16/60 200 pg (#28989335, GE Healthcare) column on an AKTApurifier10 FPLC controlled by UNICORN 5.11 (GE Healthcare). The column was equilibrated with the SEC Buffer (35 mM sodium phosphate pH 7.5, 150 mM NaCl, 1.5 mM MgCl 2 ). Two hundred microlitres of~0.5 mg/ml KIF1C-GFP or KIF1C-Flag was injected into the column and 0.5 ml fractions were collected using the fraction collector. GFP fluorescence in the fractions was measured using Nanodrop 3300 Fluorospectrometer (Thermo Scientific) to determine peak positions. One hundred microlitres of 5 mg/ml standard proteins were injected individually. The Stokes radius for the standard proteins used were as follows thyroglobulin (#T9145, Sigma) 8.5 nm 50 , apoferritin (#A3660, Sigma) 6.1 nm 51 , catalase (#C9322, Sigma) 5.2 nm 52 , bovine serum albumin (#A7906, Sigma) 3.48 nm 53 . Log R s vs. (elution volume-void volume) for standard proteins was plotted and stokes radius R s for KIF1C was determined from the slope of the line, y = (−0.014 ± 0.002) ×x + (0.97 ± 0.04). Three independent repeats for KIF1C-GFP were carried out at 500 mM NaCl and two at 150 mM NaCl. KIF1C-Flag was carried out once at 150 mM NaCl.
The molecular weight was calculated from stokes radius and sedimentation coefficient using equation M r = 4205 ×R S × S as described in ref. 56 . The frictional coefficient was calculated from the following formula, ƒ/ƒ min = S max /S with S max = 0.00361·M 2/3 (ref. 56 ) using M = 308,000 Da for KIF1C-GFP dimer and S is the sedimentation coefficient determined from glycerol gradient centrifugation.
Crosslinking mass spectrometry. Two cross-linkers, BS3 (bis (sulpho-succinimidyl) suberate) (#21580, Thermo Scientific) and EDC (1-ethyl-3-(3-dimethylaminopropyl) carbodiimide) (#22980, Thermo Scientific) were used to analyse protein interactions using crosslink mass spectrometry. Five millimolar of crosslinker was freshly prepared in MilliQ water and was mixed in the ratio of 1:2 with the protein(s) of interest by pipetting. KIF1C-GFP concentration was 1 µM when crosslinked alone. For crosslinking in the presence of PTPN21-FERM, both proteins were at 0.5 µM. Final concentration NaCl in the crosslinking reaction was 150 mM or increased to 500 mM for high salt sample. For EDC crosslinking, 3 nM Nhydroxysulfosuccinimide (#24510, Thermo Scientific) was added to the reaction with EDC to improve efficiency of the crosslinking. The reaction was incubated shaking at 400 r.p.m. for an hour at room temperature and then quenched with 50 mM Tris-HCl pH 7.5. Next, the protein was diluted in equal volume of 50 mM ammonium bicarbonate (#A6141, Sigma) and reduced using 1 mM DTT (#MB1015, Melford) for 60 min at room temperature. The sample was then alkylated with 5.5 mM iodoacetamide (#I6709, Sigma) for 20 min in the dark at room temperature and digested using 1 µg trypsin (sequencing grade; #V5111, Promega) per 100 µg of protein overnight at 37°C. The crosslinked peptides were desalted using C18 stage tips. Twenty microlitre samples were then analysed by nano LC-ESI-MS/MS using UltiMate ® 3000 HPLC series using Nano Series™ Standard Columns for separation. A linear gradient from 4% to 35% solvent B (0.1% formic acid in acetonitrile) was applied over 30 min, followed by a step change 35-70% solvent B for 20 min and 70 to 90% solvent B for 30 min. Peptides were directly eluted (~300 nl/min) via a Triversa Nanomate nanospray source into a Orbitrap Fusion mass spectrometer (Thermo Scientific). Positive ion survey scans of peptide precursors from 375 to 1500m/z were performed at 120 K resolution (at 200m/z) with automatic gain control 4 × 10 5 . Precursor ions with charge state 2-7 were isolated and subjected to HCD fragmentation in the Orbitrap at 30 K resolution or the Ion trap at 120 K. MS/MS analysis was performed using collision energy of 33%, automatic gain control 1 × 10 4 and max injection time of 200 ms. The dynamic exclusion duration was set to 40 s with a 10 ppm tolerance for the selected precursor and its isotopes. Monoisotopic precursor selection was turned on. The instrument was run in top speed mode with 2 s cycles.
Raw data files were converted to.mgf format using the ProteoWizard msconvert toolkit 57 . Sequences are visualised using Scaffold (Proteome software) for percentage coverage and purity followed by analysis using StavroX 22 . Crosslinked peptides were identified using the StavroX software with appropriately defined parameters for the crosslinker used. A crosslinked peptide was considered as valid if it passed the 5% false discovery rate (FDR) cut-off along with a StavroX score of at least 100, which is based on: (i) the presence of ion series fragmentation for both peptides, specifically those fragment ions that include the crosslinker and the attached second peptide, (ii) the proximity of observed fragmentation ion mass to expected fragment ion mass, (iii) number of ion fragments for b or y ions in the a peptide, (iv) number of unidentified high intensity signals in the spectra. The MS/MS spectra were also manually inspected and only those crosslinks were accepted for which fragmentation ions were observed for both peptides and three or more fragments for b or y ions in the alpha peptide were required to match. Any crosslinks of continuous peptides, which could indicate intradimer interactions, were rejected as these could not be distinguished from partially cleaved peptides that have been modified by the crosslinker. Once a significant crosslink was identified in at least one sample, we used the information on retention time and mass of the precursor ion to verify the presence of the crosslinked peptide in other samples.
Label-free quantification of crosslinked precursor ions and unmodified peptides was done using MaxQuant (V1.5.5.1). The LFQ module calculates the integrated peak area of the precursor ion based on a retention time window of 4 min and ppm error window of 10 ppm to quantify the presence and absence of a precursor ion of interest in different mass spectrometry output files. LFQ intensity data for each sample were normalised to the sum of unmodified and modified peptides containing LKEGANINK. BS3 experiments were repeated 3-5 times and EDC experiments 2-3 times for each sample.
Microscale thermophoresis. GST-stalk was rebuffered into MST buffer (50 mM Tris pH 7.4, 150 mM NaCl, 10 mM MgCl2, 0.05 % Tween-20) using Zeba Spin Desalting Columns (Thermo Scientific). The resulting 37 µM stock solution was used to prepare a twofold dilution series. Motor-GFP was diluted in MST buffer to a final concentration of 250 nM, 2 µl of which was mixed with 8 µl of each GSTstalk dilution. Capillaries were filled with 10 µl protein mix and thermophoresis performed in a Monolith NT.115 (Nanotemper) using 30% Nano Blue detection and 40% MST power at 25°C. The experiment was repeated three times. Traces were analysed in 0.5 and 1.5 s heating window and the Kd model was fitted using MO.Affinity Analysis software (Nanotemper).
BioID protein interaction analysis. For each BioID experiment, three 14.5 cm dishes of RPE1 cells were grown, and each was transfected at 90% confluency with 20 µg DNA diluted in 1 ml Optimem (#31985062, Fisher) with 60 µg polyethylenimine (PEI, #408727, Sigma). After 24 h, media was replaced and supplemented with 50 µM biotin (#8.51209, Sigma). After a further 12 h, cells were harvested by trypsinisation, washed with PBS and lysed in RIPA buffer (#9806S, New England Biolabs) supplemented with 250 U of benzonase (#E1014, Sigma). Lysate was cleared by centrifugation in a tabletop centrifuge at 20,238 ×g for 30 min at 4°C (#5417R, Eppendorf), and bound to 50 µl streptavidin-coated Dynabeads (#65601, Invitrogen) for 1.5 h, washing away unbound proteins with 3 × 1 ml of PBS. Beads were resuspended in 45 µl 50 mM ammonium bicarbonate, reduced by adjusting to a final concentration of 10 mM TCEP, and subsequently alkylated by adjusting to 40 mM chloroacetamide (#C0267, Sigma), incubating at 70°C for 5 min. Proteins were digested off of the beads with 0.5 µg Trypsin (#V5111, Promega) at 37°C overnight. Peptides were separated from beads by filtration through a 0.22 µm tube filter (#CLS8169, Sigma) and desalted using C18 stage tips. Twenty microlitres of peptides were then analysed by nano LC-ESI-MS/MS in an Ultimate 3000/Orbitrap Fusion. Hits were automatically identified, thresholded and viewed in Scaffolds software. We performed three independent experiments of which one had three technical replicated. Statistical analysis was performed in Scaffolds using a Fisher's exact test with Benjamini-Hochberg correction, which resulted in a significance threshold of 0.0019 for a 5% FDR.
Cells were imaged using a Deltavision Elite Wide-field microscope and Z-stacks of individual cells were acquired using a ×40 NA 1.49 objective and a Z-spacing of 0.2 µm.
To determine the number of podosomes formed in each cell, images of the cortactin channel were first transformed in a Z-projection. Z-projection images were then segmented using the ImageProAnalyzer 7 software by applying a threshold and a minimal size filter of 16 pixels (2.58 µm 2 ). Individual objects identified in the cortactin Z-projection were then visually compared to the actin channel to confirm they are podosomes, removed if the cortactin staining did not coincide with an actin spot and podosome clusters were split into individual podosomes. All experiments were repeated three times with 30 cells being analysed for each condition in each experiment.
Live cell imaging. Live cells were imaged using a ×60 oil NA 1.4 objective on an Olympus Deltavision microscope (Applied Precision, LLC) equipped with eGFP, mCherry filter sets and a CoolSNAP HQ2 camera (Roper Scientific) under the control of SoftWorx (Applied Precision). The environment was maintained at 37°C and 5% CO 2 using a stage-top incubator (Tokai Hit) and a weather station (Precision control).
RPE1 α5-integrin-GFP cells were seeded 24 h before imaging into quadrant glass-bottom dishes (#627975, Greiner) coated with 10 µg/ml Fibronectin. Cells with tails were selected and the mid region of the tail was bleached using the 488 nm laser. Vesicle movement was imaged for 200 time points at 0.7-1.0 s per frame. α5-integrin-GFP trafficking was analysed using ImageJ. Kymographs were generated from parallel 21 pixel wide lines covering the area of the tail. Vesicle movement was extracted manually from all kymographs using the segmented line tool. Vesicles moving less than a total of 1.5 µm over the period of imaging were classified as stationary.
Primary mouse hippocampal neurons transfected with pNPY-RFP and either pFlag (as control) or pFERM and imaged at DIV5 or DIV6. Images were acquired with 500 ms exposure every 1.5 s for 160 s. The frequency of NPY-positive vesicles passing through a location in the neurite in anterograde and retrograde direction was determined manually from kymographs. RPE1 cells co-transfected with pKIF1C-mCherry and either pKIF1C-GFP, pKIF1CΔCC3-GFP or pKIF1CΔS-GFP were imaged 36 h post transfection. Images were acquired with 500 ms exposure in the eGFP channel and 1 s exposure in the mCherry channel. To determine the ratio of enrichment at the tail, a region of interest was drawn manually surrounding the accumulation observed and the mean intensity was measured at the tail, in the cytoplasm near the nucleus and the image background outside the cell in both channels. The ratio of KIF1C enrichment at the tail was calculated as I tail −I background /I cytoplasm −I background for GFP vs. mCherry channel. diluted into imaging buffer. Images were taken once every 275 ms using 50 ms exposure with a 647 nm laser line followed by 150 ms exposure with a 488 nm laser line, using laser powers of 18% and 21%, respectively. Images were analysed by generating kymographs from each microtubule using the ImageJ plugin by Arne Seitz (https://biop.epfl.ch/TOOL_KYMOGRAPH.html) and manually identifying runs by scoring the kymographs for moving and static motors. The speed and run length was calculated from the length and slope of each run using a custom ImageJ macro. Blind analysis was carried out to determine landing rates and frequency of running motors in all data sets.
For TIRF-based microtubule binding assays, 80 nM motor-GFP was mixed with either 20 µM GST-stalk or the same amount of GST as a control. These mixtures were diluted 1:20 into motility mix as described for KIF1C-Hook3 single-molecule experiments, yielding a final protein concentration of 4 nM motor domain to 1 µM tail or GST control. The motility mixture was flown into chambers prepared with unlabelled GDP-taxol microtubules, and microtubule intensities in ten random fields-of-view were captured with an exposure time of 400 ms and 21.4% 488 laser power. Control and experimental chambers were prepared in parallel, and we alternated which chamber was imaged first to account for time effects. For analysis, microtubules were traced by hand with a line ROI in ImageJ, and the mean intensity was taken. A region close to the microtubule was selected as the local background, and subtracted from the mean to generate a local-background corrected intensity.
Bleach step analysis. A flow chamber was prepared as described above with biotin-647-microtubules immobilised on the coverslip. Six hundred picomolar KIF1C-GFP was flown into a chamber with MRB80, 1 mM AMP-PNP, 100 µM taxol and 0.2 mg/ml κ-casein. Images were acquired every 500 ms for 600 s at an exposure of 400 ms using the 488 nm laser line at 50% laser power. Images were analysed using the Plot Profile function in ImageJ for each individual spot and manually counting bleach steps. A mixed binomial distribution for dimers and tetramers of KIF1C was fitted to the data using Eq. (1), where P(k) is the probability to find k bleach steps, p is the fraction of active GFP and x is the fraction of tetramers.
Statistical analysis and figure preparation. Statistical data analyses and graphs were generated using Origin Pro 8.5 (OriginLab), python's Matplotlib and Scipy packages or R. Box plots show quartiles with whiskers indicating 10% and 90% of data as well as the mean indicated with a circle. All statistical significance analyses were carried out using two-tailed two-sample t-tests assuming equal variance (referred to a t-test is legends), Mann-Whitney U-test, ANOVA with post hoc Tukey, or Kolmogorov-Smirnov test. Where necessary, p-values were adjusted for multiple comparisons using Bonferroni or Holm-Šídák corrections. Figures were prepared by adjusting min/max and applying and inverting look up tables using ImageJ and assembled using Adobe Illustrator.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Data supporting the findings of this manuscript are available from the corresponding author upon reasonable request. A reporting summary for this Article is available as a Supplementary Information file. The source data underlying Figs. 1a, 1e, 3c, 3d, 3f, 3g, 3h, 4c, 4d, 4f, 5d, 5e, 6c, 6j, 6k and 7c, 7f and Supplementary Figs 2e and 6 are provided as a Source Data file. Our BioID mass spectrometry data are provided in Supplementary Data 1. The crosslinking mass spectrometry data have been deposited to the ProteomeXchange Consortium via the PRIDE 59 partner repository with the dataset identifier PXD013939.