Abstract
Thousands of outer-arm dyneins (OADs) are arrayed in the axoneme to drive a rhythmic ciliary beat. Using electron microscopy, we determined the structure of OAD array bound to microtubule doublets (MTDs) in near-atomic details and illuminate how OADs coordinate with each other to move one step forward. OAD prefers a specific pattern of MTD protofilaments for its distinct microtubule-binding domains. Upon MTD binding, free OADs are induced to adopt a stable parallel conformation, primed for array formation. Extensive tail-to-head (TTH) interactions between OADs are observed, which need to be broken for ATP turnover by the dynein motor. ATP-hydrolysis in turn relaxes the TTH interfaces to sequentially effectuate free nucleotide cycle of downstream OADs. These findings lead to a model for how conformational changes of OADs produce coordinated actions.
Motile cilia are evolutionarily conserved organelles that drive the movement of individual cells or transport extracellular fluids. Outer-arm dynein (OAD) is the key motor protein that generates the majority of mechanical forces required for ciliary beating (1, 2). In a typical ‘9 + 2’ cilium (Fig. 1A), thousands of OAD molecules are assembled onto the axonemal microtubule doublets (MTDs) as ordered arrays. Defects of OAD arrays in human lead to diseases, particularly the primary ciliary dyskinesia (PCD) (3–5). To accomplish a rhythmic and energy-efficient beat, it is thought to be essential that OADs in the axoneme coordinate with each other. The regulation of ciliary beat involves various factors, including many different axonemal components, extracellular signals, and local geometry changes (2). Nevertheless, lacking a deep understanding of the structural basis for OAD array formation and conformational coordination between OADs has been a major barrier to unveiling the mechanism of ciliary beat and its implications in human diseases.
We purified native OAD and MTD from T. thermophila and reconstituted the isolated OAD onto MTD (OAD-MTD) (6) for biochemical and cryo-EM analysis (figs. S1 and S2, and table S1). Microtubule-gliding assays indicated that the isolated OADs are active in vitro (Fig. 1B). The gliding velocity is positively correlated with the OAD concentration as well as microtubule length (fig. S1, B and C) (7). Evidence from a previous study showed that the microtubule sliding is strikingly fast if OADs are aligned in bundles (8). Besides, our reconstitution assay reveals that ordered OAD arrays can be spontaneously formed in the presence of MTD, in line with a previous finding (6). These together suggest that multiple OADs tend to unify their forces to slide microtubules. To elucidate how OADs coordinate their actions, we performed cryo-EM on both free OAD and reconstituted OAD arrays. Free OADs are extremely flexible and do not have a stable conformation (fig. S1, D and E). We managed to obtain a structure of free OAD in a notable pre-parallel state at 5-12 Å resolutions in different regions (fig. S1, F to H). The OAD arrays adopt two distinct microtubule-binding states (MTBS-1 and MTBS-2) (fig. S2, D and E), which agrees with our observation by cryo-electron tomography (fig. S2, F and G). OAD unit together with the four binding protofilaments (OAD-PF) were locally refined to 2.8-3.8 Å resolutions for most regions in MTBS-1 and 3.8-6 Å resolutions in MTBS-2 (Fig. 1, C and D, figs. S3 to S6). Together with mass spectrometry and a genome-wide pattern search, eighteen unique subunits of OAD were unambiguously identified, built and refined (Fig. 1E, and tables S2 and S3).
The arrayed OADs adopt a fully parallel conformation (Fig. 1C) and fit well into in-situ cryo-ET maps (fig. S7, A to C) (9, 11–13), indicating that our reconstituted arrays reflect the native structures in cilia. Compared to the dynamic OAD in the free form, the parallel OAD has a characteristic conformation with the motor domains stacked together. The stacking is mediated by interactions between dynein linkers (Linkers) and AAA+ rings (Rings) [referred to as Linker-Ring (LR) interactions] (fig. S7, D and E). Our cryo-EM reconstruction reveals there are four subunits that form the core structure of OAD (core-OAD), including two heavy chains (α- and β-HC) and two intermediate chains (IC2 and IC3). The core-OAD is conserved across species and all the four subunits are critical for OAD assembly (1). They together serve as a scaffold for binding all other chains and mediate nearly all interactions between two adjacent OADs via a ‘tail-to-head’ (TTH) manner (Fig. 1, C and D). Apart from the core, the complete OAD from T. thermophila contains a special γ-HC and thirteen LCs (Fig. 1E, figs. S8 and S9) for regulatory roles (14–19). γ-HC extends out from the core-OAD array with its N-terminal kelch domain (γ-kelch) tightly bound to the helical bundle 6 (HB6) of β-tail (fig. S8, A and B). Following the γ-kelch are two consecutive immunoglobulin folds, Ig-PT and Ig-Fln (fig. S8C) (1, 20). In contrast to the classical unprimed cytoplasmic dyneins [post-powerstroke state 1 (Post-1)] (21–23) as well as OAD α- and β-HC in MTBS-1, the γ-HC adopts a distinct state [post-powerstroke state 2 (Post-2)] with its Linker upraised toward AAA3-4 (fig. S8D). Among all the thirteen light chains (LCs), ten (LC7, LC8, and Tctex families) are clustered by the IC2/3 N-terminal extensions (NTEs). They together form a tower-like structure (IC/LC-tower) (Fig. 1E and fig. S9A) and attach to α-tail en bloc, facing toward the inner side of axoneme. Each position of the tower is occupied by a unique LC for its specific surrounding interfaces (fig. S9B). Attributed to a special helical bar and a beta-hairpin of IC2, the Tctex dimer is folded back and pinned to LC8s, linking the IC/LC-tower to either N-DRC or IADs in the axoneme (fig. S9D). With more than ten different proteins interwoven, this local region is tightly attached to the HB8 of α-neck and is likely to mediate the communication between inner axonemal components and OAD array. The other three LCs include an LC3B-like subunit (LC3BL, thioredoxin) which contacts Ig-PT and links the γ-tail to β-Linker (fig. S8C), LC4A (calmodulin) which binds to α-HB6 and links α-tail to IC/LC-tower (fig. S9E), and LC1 bound to microtubule-binding domain of α-HC (α-MTBD) (Fig. 1, D and E).
Different from cytoplasmic dynein-1 which requires dynactin to align the two motor domains (7, 24), the parallel OAD conformation is spontaneously induced by microtubule binding (figs. S1C and S2B), which triggers formation of an ordered OAD array. Cryo-EM classification indicated that OAD units along each array are locally synchronized and have the same microtubule-binding state (fig. S10A). However, the relative rotations between each two adjacent protofilaments [inter-PF angles (25)] vary significantly around the MTD surface (fig. S10, B and C). To preserve a stable parallel architecture, the OAD stalks need to rotate with respect to the MTBD regions to compensate for the inter-PF angles. This is enabled by the hinges between stalk coiled-coil helix 1 and MTBD helix 1 (CC1-H1) (Fig. 2A, and fig. S10, B and C). We compared the inter-PF angles from our OAD-PF reconstruction with that of the native B5-B8 PFs. The overall patterns strikingly match each other (a drop of inter-PF angle in B6-B7 and an increase in B7-B8) (Fig. 2B) (25, 26), implying a preferred inter-PF pattern. Our cryo-EM reconstruction reveals that the pattern preference is attributed to the three distinctive MTBDs. The α-MTBD specifically binds the conserved LC1 with its helix H5 (Fig. 2A, and fig. S10D) (18). The α-MTBD/LC1 complex requires a wider inter-PF space (similar to that between B7 and B8) to properly interact with adjacent tubulins (Fig. 2B). This is mediated by a cluster of positively charged residues (27) on the LC1 surface that attracts the C-terminal tail of β-tubulin (β-CTT) from an adjacent PF (fig. S10D). Meanwhile, LC1-binding forces the flap of α-MTBD to fit a conformation without contacting β-CTT. By contrast, the γ-flap directly contacts β-CTT (fig. S10E). The β-MTBD has a much shorter flap (fig. S10F) and lacks density connection with β-CTT, similar to that in dynein-1 (28). These features collectively facilitate the landing of OADs onto MTD with a local inter-PF pattern similar to that in the native B5-B8 PFs. Such preference of inter-PF pattern may help with the landing of OADs onto MTD B-tubules during ciliogenesis and the effect could potentially be reinforced in the presence of docking complexes.
In the parallel OAD, the two ICs of core-OAD directly contact and cross over each other at multiple sites (fig. S9C), in addition to their interaction with HB3-5 of α/β-HCs. This facilitates core-OAD to locally form a tight hetero-tetrameric architecture near the WD domains (Fig. 2C). The two ICs and N-terminal dimerization domain (NDD) of α/β-HCs together enclose and position the tail region, which is primed to interact with the motor domain of another OAD. More OADs are subsequently induced to associate with each other in a TTH manner (Fig. 1B). We re-analyzed previously reported cryo-ET maps of axonemes and found all OAD arrays are assembled in the same manner in apo state across species (fig. S11) (9, 11, 29–31). The TTH interfaces involve IC3, NDD and α/β-HB1-3 of OAD0 tail region, and α/β-Linkers, β-AAA2 small subunit (β-AAA2S), β-AAA6S and β-CT-cap of OAD+1 head (Fig. 2, C and D, and fig. S12, A and B). Notably, the tail0 calipers the β+1-motor and potentially hinders its allosteric response to nucleotides unless the TTH interface is disrupted. We therefore sought to elucidate the array formation mechanism. In the absence of MTD, most free OADs are too flexible to support a stable inter-OAD interaction for array formation (fig. S1C). They transiently adopt a pre-parallel state, which is likely to facilitate a proper landing of OADs onto microtubules, but not yet ready for binding another OAD (fig. S13). It is the MTD binding that finally aligns the three motor domains, induces the LR interaction to stabilize a fully parallel architecture, and triggers multiple OADs to cooperatively associate with each other (Fig. 2E, and fig. S13, A and B). Even though the docking complex is not required for OAD array formation in vitro, it plays an important role in anchoring OADs to the right location in axoneme (32).
Our cryo-EM structures reveal that the interaction network of OAD array is dramatically remodeled when β-HC moves one step ahead from MTBS-1 to MTBS-2. Alteration of MTBS is coupled with four major groups of structural changes throughout the entire OAD. First, an 18-degree rotation of between the β-stalk (fig. S14A) and β-MTBD is required for the MTBS alteration, whereas β-MTBD remains in the high microtubule-affinity in MTBS-2 (fig. S14B). Second, a gear-shift-like switch of the LR interactions among the three motor domains is involved. In MTBS-1, the β-HC constantly docks its Linker in Post-1 (Fig. 3A), while in MTBS-2, the β-Linker is precisely switched to that in Post-2 along with the β-MTBD stepping forward (Fig. 3B). On the other hand, the α-Linker fits well into a groove (Groove-1) formed by AAA3S, AAA5 extension (AAA5E) and β-CT-cap in MTBS-1, while the β-Linker matches another groove (Groove-2) between AAA2S and AAA3S of γ-Ring (fig. S14, C and D). From MTBS-1 to MTBS-2, the α-Linker docking groove on β-Ring is also switched from Groove-1 to Groove-2 along with a slight rotation in the β/γ-LR contact site (fig. S14E). Third, the local network among β-tail, γ-tail and LC3BL is remodeled. In brief, γ-Ig-PT contacts β-AAA4L PS-I with the help of γ-Kelch in MTBS-1 (Fig. 3C), while the contact site for γ-Ig-PT on the β-motor domain is switched to the H2-β3 loop of β-AAA3L in MTBS-2 (Fig. 3D). From MTBS-1 to MTBS-2, LC3BL takes over γ-Ig-PT to contact β-AAA4L at the same site. Finally, the tail is rotated downward with respect to the head from MTBS-1 to MTBS-2.
Comparing with previously reported dynein structures reveals that the ADP-bound state of cytoplasmic dynein (33) has a most similar Linker docking mode to that in the Post-2 conformation of OAD β-HC. The ADP-bound state is thought to represent the rebinding of dynein MTBD to microtubule after moving one step ahead (34). Therefore, the β-HC in Post-2 is likely to mimic the state after one complete nucleotide cycle. After ATP treatment (35) on the native T. thermophila axonemes and reverting to ATP-free solution, we could observe both Post-1 and Post-2 states on the same cryo-ET data (fig. S1, F and G). Despite the remodeling of OAD array from MTBS-1 to MTBS-2, the pattern of TTH interaction remains nearly unchanged (fig. S14, F and G). In either MTBS, the core-OAD conformations are synchronized along the same array (fig. S14, F and G). However, the conformations in the two states are not compatible. Sterical clashes are unavoidable by substitution of OAD units between the two states (fig. S14, H and I). Therefore, the TTH interfaces need to be temporarily disrupted to complete the MTBS alteration. This is unlikely to occur within a locally synchronized OAD array except the ends or the transition points where the TTH interaction is temporarily relaxed.
What is the key factor that potentially affects the TTH interfaces? We propose two possibilities, nucleotide- and geometry-control, which can co-exist. First, we re-analyzed all previously reported cryo-ET maps of core-OADs using our atomic coordinates. An interesting finding is that the pre-powerstroke state shows a completely different interaction, in which the β-motor+1 is thoroughly released from IC-NDD0 region along with a clockwise and downward rotation, consistent across species (fig. S15A) (12, 29, 30). We then tested how the reconstituted OAD-MTD arrays respond to different nucleotides. Briefly, arrays could still be observed after incubation with either AMP-PNP or ADP but completely fell off the MTDs in the presence ATP, ADP·Vi or ATPγS (fig. S15B). We speculate it is the ATP hydrolysis that releases the MTBDs from MTD and subsequently breaks the array interfaces. Using gradient concentrations of ATPγS, we were able to gradually shorten OAD arrays in vitro owing to the low hydrolysis rate. Increasing the ATPγS does not cut the array into many shorter segments randomly, but rather sequentially shortens the arrays (fig. S15C), suggesting ATP hydrolysis is more likely to take place at the ends of an ordered OAD array.
Based on our current results and previously reported OAD structures, we propose the following model for how arrayed OADs coordinate with each other to take one step (Fig. 4). In the apo state (Post-1), OADs are attached to B-tubule and form an ordered array. The TTH interactions between OAD0 and OAD+1 sterically restrain nucleotide-induced allosteric response of OAD+1 and its downstream neighbors (toward cilium tip). Curvature changes may temporally relax the array interfaces and allow free nucleotide cycle locally. After relaxation of the TTH interfaces, the OAD is free for its nucleotide cycle and energy from ATP-hydrolysis is consumed to release OAD from B-tubule, rotate the motor domain and move it one step forward. This in turn relaxes the TTH interfaces between currently active OAD and its downstream neighbors. The process is rapidly propagated to the plus end of MTD. Meanwhile, the release of phosphate groups and ADPs from upstream OADs will lead to a re-binding of OADs to B-tubule in Post-2, which subsequently imposes tension on two adjacent MTDs. The tension is converted to local bending of MTD (fig. S16), which is propagated towards the plus end along with the nucleotide cycle propagation.
Our current cryo-EM structures of OAD arrays and the proposed model potentially explain how a rhythmic ciliary beat is generated through phased propagation of OAD nucleotide states and microtubule-binding states. The results also raise many more interesting questions, such as how the three motor domains coordinate their nucleotide cycles and microtubule-binding states, how the MTD-bending curvatures quantitatively affect the binding affinity of OADs to MTD, and how γ-HC, LCs, IDAs, DRCs as well as extracellular signals collectively regulate the OAD array activities.
Materials and methods
Purification of OAD and MTD
Mucocyst-deficient strain T. thermophila SB715 (26) was purchased from Tetrahymena Stock Center (Cornell University, https://tetrahymena.vet.cornell.edu/). The axoneme was purified by using a modified dibucaine method (36) from 4 liters of culture for each sample preparation. In brief, the pellet from every one-liter fresh cell culture was de-ciliated with 3 mM dibucaine (Sigma-Aldrich) in 150 mL fresh SSP medium, and centrifuged at 2, 000 xg for 10 min to remove the cell body. The cilia were spun down from the supernatant at 12, 000 xg for 10 min, resuspended by axoneme buffer (20 mM HEPES pH 7.4, 100 mM NaCl, 2 mM MgCl2, 1 mM DTT), and further demembranated with 1.0% TritonX-100 in axoneme buffer. The axoneme was then pre-treated with buffer containing high potassium acetate (HPA buffer 50 mM HEPES pH 7.4, 600 mM CH3COOK, 5 mM MgSO4, 0.5 mM EGTA, 1 mM PMSF, 1 mM DTT) for 30 min. Subsequently, the purified axonemes were treated in different conditions for different assays. For OAD purification, the axoneme was treated with high-salt buffer (HSC buffer 50 mM HEPES pH 7.4, 600 mM NaCl, 5 mM MgSO4, 0.5 mM EGTA, 1 mM PMSF, 1 mM DTT), and incubated on ice for 30 min. MTD and the majority of axonemal dyneins were separated by centrifugation at 21,000 xg for 10 min for further purification. To obtain high-quality OAD complex from the supernatant, we used an OAD extraction protocol (37) and optimized the parameters in our own experiments. The supernatant was laid over 5-25% (w/v) linear sucrose gradients in OAD buffer (50 mM HEPES pH 7.4, 100 mM NaCl, 1 mM DTT) and centrifuged at 153, 000 xg for 16 h at 4°C. The gradient was fractionated into 0.2 mL aliquots. Fractions containing IDAs and OAD were determined by SDS-PAGE (4-20% Mini-PROTEAN TGX precast protein gels run in SDS buffer (Bio-Rad)), stained by SYPRO Ruby or Page-Blue staining solution. The pooled fractions containing the OAD were dialyzed (REPLIGEN dialysis membranes) against the OAD buffer for 6h at 4 °C to remove the sucrose and then loaded on an EnrichQ 5/50 column (Bio-Rad) equilibrated with the ion-exchange buffer A (50 mM HEPES pH 7.4, 50 mM KCl, 1 mM DTT). The OAD was eluted with a linear salt gradient: 0-50% buffer B (with 1 M KCl) in 8 mL. The fractions containing OAD were determined by SDS-PAGE (Bio-Rad), pooled together, and adjusted to a final concentration of 0.3 mg/mL for reconstitution assay. The MTD pellet was resuspended in high salt buffer and dialyzed against the low salt buffer (50 mM HEPES pH 7.4, 0.5 mM EDTA, 1 mM DTT) overnight at 4°C (35). The MTD was then pelleted and resuspended with the fresh low salt buffer. The final concentration was adjusted to 0.6 mg/mL for the subsequent reconstitution assay. Grafix method (38) was used to improve the quality of free OAD samples for EM analysis. Briefly, the OAD fractions from the linear sucrose gradient were collected, dialyzed against the OAD buffer, concentrated to 1 mL and applied to Grafix [0-0.0125% (v/v) linear glutaraldehyde (Sigma-Aldrich) gradients along with a 5-25% (w/v) linear sucrose gradient in the OAD buffer]. The gradient was fractionated into 0.2 mL aliquots. The crosslinking was quenched by adding 10 µL Tris-HCl (1 M, pH 8.0) to each aliquot. The fractions containing the cross-linked OAD were determined by SDS-PAGE (Bio-Rad) and evaluated by ns-EM. The fractions containing properly cross-linked OAD were dialyzed against the OAD buffer for 6 h, concentrated to 50 µL and loaded to the TSKgel G4000SWXL column (TOSOH Bioscience) equilibrated with the gel filtration buffer (20 mM HEPES pH 7.4, 100 mM KCl, 1 mM DTT, 2 mM MgSO4). The peak fractions were pooled for subsequent negative-stain analysis. Protein concentrations were measured by using BioSpectrometer (Eppendorf).
Mass spectrometry
Mass spectrometry (MS) on the isolated OAD sample was performed at Keck Biotechnology Resource Laboratory, Yale University. The OAD subunits identified from the MS data are summarized in table S1.
Microtubule gliding assay and analysis
Dynein gliding assay was adapted from a previously published protocol (39). In brief, HMDE buffer (30 mM HEPES-KOH, 5 mM MgSO4, 1 mM EGTP, 1 mM DTT, pH 7.4) was first introduced into the flow channel, followed by a 5-min incubation of 10 µL, 0.1 mg/mL purified outer arm dynein at room temperature to allow adsorption of dynein to the coverglass surface. Unbound dynein was then washed with HMDE buffer, followed by a 5-min incubation of 0.4 mg/mL casein. The channel was again washed by HMDE buffer. GMPCPP-stabilized microtubules were prepared as previously described (40) using bovine brain tubulin purified in-house (41). 10 µL of microtubule solution (0.15 µM tubulin dimer in HMDE + 1 mM ADP) was perfused in to bind to the motors with a subsequent wash by HMDE+1 mM ADP. 10 µL motility solution (HMDE + 1 mM ATP + 1 mM ADP) was then flowed in to initiate the microtubule gliding, imaged by interference reflection microscopy (IRM) as previously described (42)(41) with a frame rate of 13.5 Hz. Lengths and positions along the gliding paths of individual microtubules were tracked with tracking software FIESTA (43) after background subtraction. Tracking results were manually inspected to exclude immobile filaments, surface dirt particles, tracks less than 1 second, and tracking errors due to filament collisions. Position of individual microtubule filament was averaged over three frames (0.22 s interval) to reduce the experimental noise. Time-weighted average velocity and displacement-weighted average velocity were calculated as previously described (44) with the bin width of 0.4 µm/s. The standard error of the mean (SEM) of the displacement-weighted average velocity is equal to the standard deviation (SD) divided by , where N is the number of microtubules in each condition (N=51, 75, 65, 78 MTs for 100, 50, 20, 10 µg/mL of ODA with wildtype GMPCPP-microtubule, and N=51 for 100 µg/mL ODA with subtilisin-treated GMPCPP microtubule). P-value was calculated using Welch’s t-test.
OAD-MTD array reconstitution and nucleotide treatment
The reconstitution condition was optimized from our cryo-EM analysis based on a previously published protocol (6).To assemble the OAD-MTD complex, the freshly purified native OAD (not cross-linked) and MTD samples were mixed at a series of molar ratios (tubulin dimer/ OAD, 10-100) in the reconstitution buffer (20 mM Hepes pH 7.4, 100 mM KCl, 5 mM MgSO4, 1 mM DTT) and incubated on ice for 45 min. The reconstituted samples were analyzed by SDS-PAGE and negative-stain electron microscopy to screen out the optimal molar ratio. The concentration of the OAD-MTD complex was optimized by centrifugation and resuspension for subsequent experiments and checked by cryo-EM. All protein samples were verified by 4-20% Mini-PROTEAN TGX precast protein gel stained with Page-Blue solution. After array formation (45 nM OAD), different nucleotides at a final concentration of 1 mM or ATPγS at gradient concentrations were added to the reconstitution tubes. 4 µL solution in each reaction condition was immediately transferred to a glow-discharged continuous carbon grid (Electron Microscopy Sciences) for 15 seconds before blotting and staining. The grid was stained with 2% uranyl acetate and air-dried before loading to a Talos L120C microscope (Thermo Fisher Scientific). The length of the array was directly measured in the ImageJ and converted to the equivalent number of OADs for subsequent quantitative analysis.
Cryo-EM sample preparation and data collection
4 µL OAD or OAD-MTD samples were applied to each Quantifoil R2/2 or C-flat R1.2/1.3 gold grid (for free OAD, the grids were coated a carbon layer), incubated in a Vitrobot Mark IV (Thermo Fisher Scientific) for 4 seconds, blotted for 2 seconds at 4 °C and 100% humidity, and then plunged into liquid ethane near melting point. Three cryo-EM datasets of ODA-MTD arrays in the apo state were collected on a 300 keV Titan Krios microscope (Thermo Fisher Scientific) equipped with a Bioquantum Energy Filter and a K2 Summit direct electron detector (Gatan) at Yale CCMI Electron Microscopy Facility. Data collection was automated by Serial EM software (45) and all micrographs were recorded in a super-resolution mode. The first two data sets were employed with the following parameters: 0.822 Å/pixel, 50 µm C2 aperture, 32 frames, 53.3 e-/Å2, 8 s exposure, -0.8 to -2.0 µm defocus range. Based on the results of these two data sets, the third data acquisition was optimized with a reasonable parameter set as follows: 1.333 Å/pixel, 50 µm C2 aperture, 40 frames, 53.3 e-/Å2, 12 s exposure, -1.2 to -3.0 µm defocus range. Three non-overlapping micrographs were recorded per hole in all the three data sets. The motion correction, particle picking, and CTF estimation were streamlined to evaluate the micrograph quality in real time during the data collection using a modified pre-processing script (https://www2.mrc-lmb.cam.ac.uk/research/locally-developed-software/zhang-software).
Cryo-ET data collection and reconstruction
Purified axonemes were treated with ATP at a final concentration of 1 mM for 5 minutes and reverted to nucleotide-free solution before freezing for cryo-ET. In total, 50 tomographic datasets were collected on the 300kV Titan Krios equipped with a K2 detector. The software SerialEM (45) was used for automatic data collection under the bidirectional scheme at a 3° interval and tilt angles ranging from -51° and +51°. Each of the final tilt series contains 35 movies with a pixel size of 2.8 Å at an average defocus of 5 µm and a total dose of 70 e/Å2. Individual movies were aligned by MotionCor2 (46). Motion-corrected images of each tilt series were aligned by using the patch-alignment approach in the IMOD software (47, 48). Sub-volume average was performed using PEET (49).
Pre-processing of cryo-EM data
Beam-induced drift was corrected using MotionCor2 (46) for all images. CTF parameters for each motion-corrected micrograph were estimated using Gctf (50). All particles were automatically picked using Gautomatch, extracted in RELION v3.0 (51) and imported to cryoSPARC v2.12 (52) for all subsequent processing, if not explicitly stated otherwise.
Structure determination of MTD
To obtain and analyze the structure of MTD, we first manually picked a small dataset (100 micrographs) at 4-nm intervals from Dataset 1 (table S1). These particles were analyzed in Cryosparc v2.12 to generate 20 good MTD 2D averages and used by Gautomatch for template-based particle picking. This generated 444, 603 raw particles from Datasets 1 and 2, and 680, 495 raw particles from Dataset 3 using a 4-nm distance cutoff (if the distance between two particles is less than 4 nm, the one with lower cross-correlation coefficient is removed). All the particles were extracted with a box size of 512 × 512. The micrographs from Dataset 1 and Dataset 2 were both scaled to a pixel size of 1.333 Å to match the Dataset 3 during the particle extraction. After 3-5 cycles of 2D classification to remove those particles that generated bad 2D averages, the high-quality images we selected and filtered by a 6-nm distance cutoff. This reduced the sampling of MTD to 8 nm and yielded 358, 116 good particles for subsequent 3D analysis. A previous MTD map from T. thermophila (EMDB: 8532) (35) was low-passed to 100 Å as an initial model. The 8-nm repeats were successfully separated into two classes of 16-nm repeats with comparable particles numbers after 3D classification in cryoSPARC v2.12 (51). The two 16-nm repeating maps were essentially the same except that they were shifted 8 nm with respect to each other. 196,740 good particles with 16-nm periodicity were selected for subsequent analysis. By restricting the refinement to each local region with 3 × 4 tubulins, we were able to improve the local tubulins at an average resolution of 3.1Å. A de novo model of the tubulin dimer was built on the best region and then expanded to all regions for manual refinement in Coot (53) and automatic refinement by Refmac5 (54). The 16-nm MTD repeat was used to estimate inter-PF distribution of tubulin lattice. Other structural information from MTD reconstruction using different parameters was not used in this work.
Structure determination of OAD-PF
To eliminate the interference of microtubule in OAD structure determination, we linearly weakened the microtubule signals to improve the alignment of OAD. In brief, the coordinates of all good particles we selected during the MTD reconstruction were split and backtracked to their original micrographs. We manually checked all micrographs one by one to make sure they are centered and evenly spaced in each MTD. If not, we then manually adjusted the uncentered, added the missing particles, or deleted the undesirable ones. The MTD signal was weakened by removing the weighted average within a rectangle mask slightly wider than the MTD. The OAD particles from the MTD-weakened micrographs were picked by Gautomatch using 20 best templates generated from a negative-stain dataset of free OAD. After 2D classification, we selected the fifty best 2D averages for another cycle of automatic particle picking. Due to the severe orientation preference, we used a very low cross-correlation (CC) cutoff (0.08) and also a very small distance cutoff (150 Å) for automatic picking by Gautomatch. The purpose was to include as many views as possible at the beginning even if there were some false pickings. This generated 824, 659 particles from Dataset 1 and 2, and 2, 022, 385 particles from Dataset 3. Cycles of 2D and 3D classification (for screening purpose) were performed on the 8 times shrunk images to remove MTDs and low-quality particles. In total, 346, 320 good particles were selected for subsequent 2D and 3D analysis. All particles from the above processing were re-extracted with a box size of 510 × 510 at 1.333 Å pixel size (Dataset 1 and 2 were re-scaled to this pixel size) and merged for subsequent processing. To further remove particles that were less consistent with the major classes, we performed iterative 2D and 3D classification. Briefly, all the particles were separated into four subsets to accelerate the processing. Each cycle of 2D cycle was followed by two cycles of 3D classification. 58, 096 particles were further excluded via this 2D and 3D classification. All the subsets were merged again that yielded 288, 224 good particles for a final cycle of 3D classification. This generated nine good classes and one bad class. Six of the nine classes were categorized to microtubule-binding state 1 (MTBS-1), while the rest three were in MTBS-2. At this stage, we had 191, 776 particles in MTBS-1 and 76, 936 for subsequent local refinement. A multi-level masking scheme was applied to the local refinement. Briefly, we gradually decreased the size of the mask applied to a certain region to ensure a stable local refinement. We divided each of the OAD-PF class into five major parts: (a) MTBD-tubulin region, (b) α-motor domain, (c) β-motor domain, (d) γ-motor domain, (e) tail region. The α-motor domain in MTBS-1 was straightforward improved from 10.1 Å to 4.5 Å after one cycle of local refinement. At such resolution, we were able to build backbones, but ab-initio assignment of the side chains was very challenging. The map was further improved by optimizing the following aspects: (i) more cycle of local classification, (ii) refinement of the particle centers, (iii) manual optimization of the local mask, (iv) 2D classification based on 3D alignment parameters, (v) local CTF refinement, (vi) non-uniform refinement (55). By combining these approaches in an iterative way, the map of α-motor domain was finally improved to an average resolution of 3.19 Å resolution for final model building. We applied the same strategy to improve the β-motor domain which finally generated a map at 3.3 Å resolution overall. Focusing on the AAA2-4 subdomains allowed us to slightly improve this region which helped with the model building a bit. The γ-motor domain, MTBD region and tail region are much more complicated than the other two. We were able to overcome the issue by. To ensure that the maps from two adjacent regions can be smoothly combined, we applied a third mask that fully covered the boundary between each pair of adjacent masks. Finally, we integrated all 31 locally refined parts into an entire unit of OAD-PF array in Chimera (56). The structures of free OAD and OAD-PF in the AMP-PNP bound state were determined using the same approach.
Identification of the light chains
We built the atomic model of all the ten IC-binding light chains de novo in combination with our mass spectrometry data. First, each of the ten LCs was manually built as a poly-Ala model. All side chains were tentatively assigned to several groups: (1) large (Trp, Try, Arg, Phe, His), (2) middle (Leu, Gln, Asn, Ile, Met, Lys), (3) small (Pro, Val, Ser, Thr, Cys, Glu, Asp, Ala), and (5) Gly. Here we categorized Glu and Asp into the group ‘small’ because the side chain densities of negatively charged residues are typically weak in cryo-EM reconstruction. We then performed two parallel approaches to identify all the light chains: (a) pattern recognition and (b) penalty function. The approach (a) is based on regular expression match using the ‘gawk’ command on a CentOS 7.5 Linux system. In the second approach, we tried to fit all predicted homologs into a certain position, e.g. the LC8-2b position and assigned the residues. All the residues that did not match the side-chain density were manually counted. The counts were regarded as penalty scores for all LC homologs. We then compared the final scores and selected the best one for subsequent model building and refinement. A protein was regarded as ‘identified’ only if it meets the following requirements: (i) it exists as a significant hit from the mass spectrometry data; (ii) its side chains simultaneously match the cryo-EM density map; (iii) no other homologs have better results of (i) or (ii). We identified IC2, IC3, γ-Kelch, all the ten IC-binding light chains. The LC7-a/b is not the standard LC7A/B heterodimer, but a heterodimer comprising LC7B (LC7-b) and an unnamed LC7A homolog (TTHERM_00348650). The full-length protein is 159 residues long (XP_976918.2) while the truncated one is 103 residues long (XP_976918.1). We unambiguously assigned the residues from S58 to G152. The extra density that links γ-HB6 to LC7-b was tentatively assigned as the N-terminus of LC7-b. Despite the similar core structures, each of the six LC8-like proteins (LC8s) are clearly different from any other five by its characteristic side chain densities and loops, which allowed us to distinguish them unambiguously. The positions of 1a, 1b, and 2a are taken by LC10, DLC82 and LC8E, respectively. The rest three (2b, 3a, 3b) were simply predicted to be LC8 homologs without standard names in TGD (TTHERM_00023950 for LC8-2b, TTHERM_01079060 for LC8-3a and TTHERM_000442909 for LC8-3b). Neither TCT1A nor TCT1B matched the key features of our cryo-EM maps. The Tctex-a position was identified as a hypothetical homolog (TTHERM_00392979) while the best hit for Tctex-b is LC2A (57).
Model Building and Refinement
We used different model building approaches for different regions. Most of the regions were refined at better than 3.5 Å resolution, which allowed us to build them in Coot (53, 58) with side chains assigned and refined ab initio. For the regions that were slightly worse, we were able to build backbone models with the residues assigned based on the relative positions among the large resides (such as Try and Arg) of each domain. For the regions that show clear backbone density with low-quality sidechain density, we coarsely assigned the residues using previously published homologous structures as references or predicted model from Phyre2 web server (59). For those regions that were solved at a resolution with helices clearly separated, we fitted the predicted model into the density as rigid bodies in Chimera (56). If the predicted model contained more than one subdomain, we then refined the fitting of each subdomain as a rigid body in Coot (58). All models at better than 4 Å resolution were automatically refined by Refmac5 (59), followed by manual check in Coot (58). The process was repeated until all parameters were reasonably refined.
Inter-PF rotation angle measurement
The inter-PF angle is defined as the lateral rotation angles between a pair of adjacent microtubule protofilaments, as described in a previous publication (25). To calculate the inter-PF angle of MTD, we fitted individual tubulin dimer model built from the 16-nm MTD reconstruction into the 48-nm MTD map as rigid bodies. We calculated the inter-PF angle between each pair of tubulin dimers throughout each protofilament. The averaged value and standard deviation were estimated from the six measurements along each protofilament. All the inter-PF angles were calculated in PyMOL (https://pymol.org/2/).
MTD curvature analysis
After manual checking all the micrographs, we selected 15, 541 MTDs in total. Using the manually verified coordinates, the curve of each MTD was approximated with a polynomial regression of degree 3. All curves were resampled with a sampling rate of 4 nm. The axial direction of MTD at each point was estimated as the tangent line. The curvature of the fitted curve at each resampled point was calculated to approximate the local MTD bending. The maximum orientation difference among all the tangent lines of each MTD was regarded as the bending angle.
Visualization and representation
The figures and movies were created by Chimera (56), ChimeraX (60) and PyMOL (https://pymol.org/2/). Other tools used in this research include FIJI (61), EMAN2 (62) and ESPript (63). This manuscript for BioRxiv was formatted using a modified LaTeX template from the Henriques Lab (https://www.overleaf.com/latex/templates/henriqueslab-biorxiv-template/nyprsybwffws.Wp8hF1Cnx-E)
Funding
This work was supported by start-up funds from Yale University and Rudolf J. Anderson Fellowship awards to L.H. and Y.W.
Author contributions
Q.R. prepared all samples and performed biochemical characterization; Q.R., K.Z., P.C., Y.W., L.H., R.Y., and Y.Y. determined the apo ODA-PF structures; P.C., K.Z. and Q.R. determined all other structures in this work; Q.R. and Y-W.K. performed motility assays; all involved in analyzing the results; K.Z. and Q.R. prepared the manuscript with the help from all other co-authors.
Competing interests
None.
Data availability
Cryo-EM maps and atomic coordinates have been deposited in the Electron Microscopy Data Bank under accession codes EMD-22677, EMD-22679, EMD-22840, EMD-XXXX and in the Protein Data Bank in under accession 7K58, 7K5B, 7KEK, XXXX.
ACKNOWLEDGEMENTS
We thank S. Wu, K. Zhou, M. Llaguno and K. Li for technical support on microscopy, J. Kanyo for mass spectrometry support, Y. Xiong, F. Sig-worth, J. Liu, and S. Baserga for their valuable research advice, S.M. King, A. Yildiz, A.P. Carter and Y. Xiong for valuable feedback on the manuscript, and P. Sung, W. Konigsberg, A. Garen, Y. Xiong as well as many others for their generous support during K.Z.’s lab setup.