Regulation of kinesin-2 motility by its β-hairpin motif

Abstract

Members of the kinesin-2 family coordinate with other motors to power diverse physiological processes, but the structural mechanisms regulating kinesin-2 activity have been unknown. Distinctively, kinesin-2s canonically function as heterotrimers of two different motor subunits (for example Kif3A and Kif3B in humans) and Kap3, but the role of heterotrimerization has yet to fully emerge. Here, we combine structural, cell biological and single-molecule approaches to dissect kinesin-2 regulation as a heterodimer, heterotrimer and quaternary complex with a cargo adaptor (APC). We identify a conserved motif in the tail of kinesin-2s (the β-hairpin motif) that, in conjunction with the adjacent coiled coil, controls kinesin-2 motility by sequestering the motor domains away from their microtubule track. Our data reveal how Kap3 binds via a multipartite interface with Kif3A and Kif3B. Rather than activating motility directly, Kap3 provides a platform on which cargo adaptors can engage and occlude the β-hairpin motif. Together, these data articulate a structural framework for kinesin-2 activation, recycling by dynein and adaptation for different biological functions.

Fig. 1. A β-hairpin motif predicted in the tail of kinesin-2 in diverse eukaryotes.

a, Top, sequence diagrams of human kinesin-2 Kif3A and Kif3B. Bottom, AlphaFold3 model of the C-terminal coiled-coil region and β-hairpin motif, with the negatively charged residues at its apex shown in stick representation. The full AlphaFold3 model with pLDDT and PAE scores is shown in Extended Data Fig. 1. b, Structure-based sequence alignment of kinesin-2s from different eukaryotic supergroups. Negatively charged residues at the apex of the β-hairpin are shown in red. Conserved aromatic residues in the –3 and +4 positions relative to the apex are shown in bold. c, Aligned AlphaFold3 models of the β-hairpin in different kinesin-2s, colored as in b. d, AlphaFold3 models of homodimeric kinesin-2 Kif17 and OSM-3, which both feature an additional loop between the coiled coil and short α-helix preceding the β-hairpin. e, Size-exclusion chromatogram of reconstituted Kif3AB heterodimer with schematic of the construct inset. a.u., arbitrary units. f, SDS–PAGE of peak size-exclusion chromatography fraction after labeling SNAP-tagged Kif3A with the Alexa Fluor 647 fluorophore. g, Mass photometry of purified Kif3AB. The main peak (174 ± 15 kDa, mean ± s.d.) is consistent with the Kif3AB heterodimer mass (166 kDa).

Fig. 2. The β-hairpin motif mediates kinesin-2 autoinhibition.

a, Composite TIRF images of the 488-nm channel (MT) and 640-nm channel (Alexa-Fluor-647-labeled Kif3A), offset in y axis by 8 pixels (px). Top, with 1 mM ATP. Bottom, with 1 mM AMPPNP. MT; microtubule. b, Left, schematic of the MT gliding assay with biotinylated Kif3AB. Middle, example kymograph. Right, plot of gliding velocities from three technical replicates. Colored circles; individual data points. White circles; average from each separate replicate. Lines, mean ± s.e.m., n = 33, 25 and 28 microtubules analyzed per replicate. c, Left, single-molecule velocity, run length and microtubule landing rate of Alexa-Fluor-647-labeled Kif3AB constructs on microtubules. Constructs were progressively truncated from their C termini, as indicated in the schematics above the plots. Measurements were taken from three technical replicates from separate motility chambers, with three fields of view per replicate. For velocity and run length, the numbers of measurements per technical replicate are: Kif3A^Δ89B, n = 120, 100 and 120; Kif3A^Δ103B, n = 113, 123 and 106; Kif3A^Δ103B ^Δ39, n = 121, 105 and 102; Kif3A^Δ103B^Δ90, n = 144, 108 and 103; Kif3A^Δ103B^Δ114, n = 105, 102 and 116; Kif3A^Δ103B^Δ156, n = 124, 108 and 104. For landing rate, n = 5 microtubules per replicate. Gray circles, average from each replicate. Lines, mean ± s.e.m. For Kif3A–Kif3B, Kif3A^ΔC38–Kif3B, and Kif3A^ΔC59–Kif3B constructs, no motile events were observed across three replicates (depicted with a cross), suggestive of strong autoinhibition at the single-molecule level. For other constructs, velocity and run-length values were not significantly different than each other (P > 0.1 one-way analysis of variance (ANOVA) followed by Tukey’s multiple comparison test). Landing-rate values for Kif3A^Δ103B^Δ90, Kif3A^Δ103B^Δ114 and Kif3A^Δ103B^Δ156 were significantly higher than those of other constructs (P < 0.05). Exact P values are given in the Methods. Right, an example kymograph of Kif3A^ΔC103B^ΔC156. d, Example kymographs of Kif3AB β-hairpin mutants. Kif3A^KK mutant: E612K D613K. Kif3B^KKK mutant: E607K E608K E609K. Kif3A^Gly-mut: H609G W610G W617G Q618G. Motility parameters for all constructs are given in Extended Data Table 1.

Fig. 3. The β-hairpin motif is crucial for Kif3 recycling and spatial regulation in intraflagellar transport.

a, Left, representative IMCD-3 cilia in control cells and Kif3B KO cells stably expressing the indicated constructs. Cells were immunofluorescently labeled for acetylated tubulin (red) marking the ciliary axoneme, and gamma tubulin (yellow), marking the basal body. Right, quantification of cilia length from three technical replicates. Gray circles, individual data points. White circles, average from each replicate. Lines, mean ± s.e.m. Control, n = 86; Kif3B KO, n = 76; Kif3B^WT, n = 78; Kif3B^ΔC156, n = 76; Kif3B^KKK, n = 74 cilia. Cilia length is not significantly different for control, Kif3B^WT and Kif3B^KKK cells (P > 0.4, one-way ANOVA followed by Tukey’s multiple comparison test) or for Kif3B KO and Kif3B^ΔC156 cells (P > 0.9). Cilia length is significantly shorter in Kif3B KO and Kif3B^ΔC156 cells than in control, Kif3B^WT or Kif3B^KKK cells (P < 0.0001). Exact P values are given in the Methods. b, Top, representative images of mScarlet (mSc)-tagged Kif3B^WT and Kif3B^KKK constructs expressed in Kif3B KO cells. Bottom, plot of the average mSc-Kif3B fluorescent signal from line scans along the cilium length (bars in the top panel indicate the distance analyzed), aligned at the ciliary tip (peak in the Kif3B^WT trace). Values are normalized to Kif3B^WT peak value. Traces show mean intensity ± s.e.m.; n = 42 (Kif3B^WT), 51 (Kif3B^KKK) cilia measured in three separate experiments. An analysis of Kap3 localization is shown in Extended Data Fig. 2e.

Fig. 4. Kap3 makes a multipartite interaction with Kif3AB that permits autoinhibition.

a, Size-exclusion chromatogram of reconstituted Kif3AB–Kap3 complex (red); schematics are also shown. Normalized Kif3AB and Kap3 traces are shown for comparison (dashed orange and yellow lines respectively). b, SDS–PAGE of the peak size-exclusion chromatography fraction. c, Top, composite TIRF image of the 488-nm channel (MT) and 640-nm channel (Alexa-Fluor-647-labeled Kif3A), offset in the y axis by 8 pixels, and the 561-nm channel (TMR-labeled-Kap3), offset in the y axis by 16 pixels. The assay was conducted with 1.5 nM Kif3AB, 7.5 nM Kap3 and 1 mM ATP. The presence of Kap3 does not activate Kif3AB microtubule binding or motility. Bottom, the same experiment as above, but with 1 mM AMPPNP. The assay was performed with 1 nM Kif3AB and 1 nM Kap3. Kif3AB and Kap3 colocalize in a complex. d, Top, example class averages of the Kif3AB–Kap3 complex, with features labeled. Bottom, corresponding AlphaFold3 (AF) model projections. e, Ribbon representation of the Kif3AB–Kap3 AlphaFold3 model. Protruding coiled-coil regions of Kif3AB shown in white. Further analysis of coiled-coil regions in this model shown in Extended Data Fig. 5. f, Close-up of the composite binding interface between Kap3 and Kif3AB C-terminal region in the AlphaFold3 model. The three main interfaces are labeled. Kif3AB β-hairpins are not occluded by Kap3 binding. g, Plot of the fraction of TMR-labeled Kap3 colocalizing with Alexa-Fluor-647-labeled Kif3AB for each indicated construct: ΔInt3 (Kif3A–Kif3B^ΔC90), ΔInt2+3 (Kif3A–Kif3B^ΔC114), ΔInt1 (Kif3A^ΔC89–Kif3B) and ΔInt1–3 (Kif3A^ΔC103–Kif3B^ΔC156). Measurements were taken from three technical replicates. Gray circles, average from each replicate. Lines, mean ± s.e.m. WT, n = 49, 68 and 81; ΔInt3, n = 54, 75 and 85; ΔInt2+3, n = 46, 70 and 82; ΔInt1, n = 68, 44 and 56; ΔInt1–3, n = 59, 58 and 28 molecules analyzed per replicate. Mean values for WT, ΔInt2+3, ΔInt1 and ΔInt1–3 are significantly different from each other (P < 0.01, one-way ANOVA followed by Tukey’s multiple comparison test). Exact P values are given in the Methods.

Fig. 5. Structural basis for Kif3 autoinhibition.

a, Close-up of the AlphaFold3 model of the autoinhibitory Kif3AB interface in cartoon representation, colored as in previous figures. Key features are labeled. The negative-stain EM class average from Fig. 4d is shown in the inset. b, Open-book representation of binding the interface between the Kif3A motor domain and Kif3A β-hairpin and coiled coil in the AlphaFold3 model. The model is shown in surface representation and colored by electrostatic potential with key residues annotated. Note the complementary charges between motor domain (positively charged; blue) and β-hairpin and coiled coil (negatively charged; red). c, Open-book representation of the binding interface between Kif3B motor domain and Kif3B β-hairpin and coiled coil. d, Schematic of kinesin-2 autoinhibition. The Kif3AB motor domains fold back to interact with β-hairpin motifs and the C-terminal coiled coil, occluding the tubulin-binding surface of each motor. Kap3 binds C-terminal regions of Kif3AB and does not relieve autoinhibition, remaining available to bind other factors that might activate Kif3. e, Example kymographs of Kif3AB^WT and Kif3AB^CC-mut (Kif3A-E559K D562K E556K; Kif3B-D554K D557K E561K). The gels show purified Kif3AB^CC-mut. Motility parameters are provided in Extended Data Table 1.

Fig. 6. Kap3-dependent activation of Kif3 by adaptor binding to the β-hairpin motif.

a, Top, size-exclusion chromatogram of reconstituted Kif3AB–Kap3 complex in the presence of either APC^ARM (green) or APC^ARM-Mut (L395Y K399E K441E L437Y R440E, gray). Bottom, SDS–PAGE of size-exclusion chromatography fractions. Traces for APC^ARM and APC^ARM-Mut alone shown in Extended Data Fig. 4a. b, Example kymographs of Kif3AB-Kap (Alexa-Fluor-647-labeled Kif3A) in the presence of microtubules and APC^ARM or APC^ARM-Mut. c, AlphaFold3 model of APC^ARM in complex with Kif3AB–Kap3. Conserved APC^ARM residues at the interface are shown as black spheres and annotated (mutated in APC^ARM-Mut). APC^ARM occludes the Kif3A β-hairpin. d, Equivalent view to c, with the autoinhibited Kif3A motor domain binding site overlaid, showing steric clash with APC^ARM. The APC^ARM and Kif3A motor domain are shown in surface representation. e, Schematic model of kinesin-2 activation. Isolated Kif3 favors a compact autoinhibited conformation in which the β-hairpin motifs and C-terminal coiled coil interact with the motor domains, sequestering the motor domains from the microtubule. Binding of a cargo adaptor, exemplified here by APC^ARM, occludes the β-hairpin motif, promoting an extended activated conformation in which the motor domains are free to drive processive motility along the microtubule.

Extended Data Fig. 1. AlphaFold3 analysis of Kif3AB.

Left, Kif3AB AlphaFold3 model showing β-hairpin motif, colored by pLDDT (predicted local distance difference test) according to the key. Right, predicted aligned error (PAE) plot of Kif3AB, full-length proteins.

Extended Data Fig. 2. Purification of Kif3AB constructs, CRISPR knockout of Kif3B, and Kap3 imaging in cilia.

a, SDS-PAGE of purified Kif3AB constructs (post size-exclusion chromatography) with indicated C-terminal truncations. b, Genotype for homozygous Kif3B KO cell line, with insertions highlighted by alignment with the reference sequence. Clones were extensively Sanger sequenced to determine genotype (representative traces shown). c, Immunofluorescence images of cilia in Kif3B KO cell lines and cell lines stably expressing Kif3B^WT, Kif3B^ΔC156 and Kif3B^KKK. Cells were stained for gamma tubulin (yellow), acetylated tubulin (red) and DAPI (blue). d, Western blot showing expression of FLAG-tagged Kif3B^WT, Kif3B^ΔC156 or Kif3B^KKK in Kif3B KO cells, detected using anti-FLAG. GAPDH used as loading control. e. Top, representative images of mNeonGreen (mNG) tagged Kap3 in background of unlabelled Kif3B^WT and Kif3B^KKK constructs expressed in Kif3B KO cells. Bottom, plot of average mNG-Kap3 fluorescent signal from line scans along the cilium length (bars in top panel indicate distance analyzed). Kif3B^WT background, black. Kif3B^KKK background, green. Values are normalized relative to Kif3B^WT peak value. Traces show mean intensity ± s.e.m.; n = 39 (Kif3B^WT), 31 (Kif3B^KKK) cilia measured from three technical replicates.

Extended Data Fig. 3. Characterization of the Kif3AB-Kap3 and Kif3AC-Kap3 heterotrimer.

a, Mass photometry of purified Kif3AB-Kap3. Main peak (291 ± 23 kDa, mean ± s.d.) is consistent with the theoretical mass of heterotrimer (278 kDa). b, Electron micrograph of negatively stained Kif3AB-Kap3. c, Left, AlphaFold3 models of Kif3AB-Kap3 and Kif3AC-Kap3 colored by pLDDT according to the key. PAE plots alongside. Interface 3 highlighted with black arrowhead for Kif3B and lack of equivalent interface with white arrowhead for Kif3C. Right, Close-up views of Kap3 for each model. Note interface 2 with Kap3 common to Kif3B and Kif3C and lack of interface 3 with Kif3C (black versus white arrowheads). d, Left, SDS-PAGE of purified Kif3AC-Kap3 heterotrimer following size-exclusion chromatography, peak fraction shown. Right, Kap3 co-localization assay with full-length Kif3AC (top) and with Kif3AC lacking C-terminal regions (Kif3A^ΔC103C^ΔC155; bottom). Composite TIRF images are shown of 488 nm channel (MT), 640 nm channel (Alexa-Fluor-647-labeled Kif3A) offset in y-axis by 8 pixels, and 561 nm channel (TMR-labeled Kap3) offset in y-axis by 16-pixels.

Extended Data Fig. 4. Characterization of APC^ARM and interaction with Kif3AB-Kap3.

a. Size-exclusion chromatogram of purified APC^ARM and APC^ARM-Mut proteins. SDS-PAGE of input shown left and of fractions beneath. b, Size-exclusion chromatogram of binding reaction between Kif3AB and APC^ARM with SDS-PAGE of fractions beneath. Note lack of co-elution of APC^ARM with Kif3AB compared to when Kap3 is present (Fig. 6a). c, Examples kymographs of Kif3AB (Alexa-Fluor-647-labeled Kif3A) in presence of microtubules and APC^ARM (unlabeled). As there is non-specific binding of Kif3AB to the coverslip surface in this experiment, kymographs are shown for regions on and off the microtubule. APC^ARM does not activate the motility of Kif3AB in the absence of Kap3. d, Electron micrograph of negatively stained Kif3AB-Kap3 in presence of APC^ARM. Particles with an elongated appearance indicated with arrowheads. e. AlphaFold3 model of Kif3AB-Kap3-APC^ARM colored by pLDDT according to the key. Right, PAE plot.

Extended Data Fig. 5. Analysis of the coiled-coil regions in Kif3AB.

a, Kif3AB-Kap3 AlphaFold3 model with predicted coiled coil regions colored in rainbow from N-terminus (blue) to C-terminus (red). Note the predicted separation of the strands of the N-terminal coiled coil in the autoinhibited state, resulting in a fine structure protruding from the motor domain (black arrowhead). This model is consistent with EM class averages showing equivalent fine protruding structure (white arrowhead, inset). b, Sequence diagrams of Kif3A and Kif3B, with predicted coiled coil regions colored in rainbow as in panel a. The location of the putative hinge in the coiled coil (Kif3A- G485 G486; Kif3B-G477 G478) is indicated. c, AlphaFold3 model of the Kif3AB coiled coil region lacking the motor domains (thereby preventing the formation of the autoinhibited conformation). This model shows an extended coiled coil, with a hinge at the predicted location, and may represent conformation of the coiled coil in active Kif3AB. Note that transition between the autoinhibited conformation in panel a and the extended conformation in panel c is not well described by simple folding at putative hinge. d, Mutation of the α-helix-breaking glycine residues at the putative hinge (Kif3A-G485 G486; Kif3B-G477 G478) to a residue with higher α-helical propensity (glutamic acid; E) is known to relieve autoinhibition in kinesin-2 (Brunnbauer et al. PNAS 107, 10460–10465, 2010). An AlphaFold3 model of the Kif3AB coiled coil region including the equivalent substitutions (Kif3A-G485E G486E; Kif3B-G477E G478E – GG > EE) shows a continuous coiled coil rather than a hinge, which could prevent access of the motor domains to the β-hairpin motif, consistent with the observed activation.