Coupling of ATPase activity, microtubule binding, and mechanics in the dynein motor domain

The movement of a molecular motor protein along a cytoskeletal track requires communication between enzymatic, polymer‐binding, and mechanical elements. Such communication is particularly complex and not well understood in the dynein motor, an ATPase that is comprised of a ring of six AAA domains, a large mechanical element (linker) spanning over the ring, and a microtubule‐binding domain (MTBD) that is separated from the AAA ring by a ~ 135 Å coiled‐coil stalk. We identified mutations in the stalk that disrupt directional motion, have microtubule‐independent hyperactive ATPase activity, and nucleotide‐independent low affinity for microtubules. Cryo‐electron microscopy structures of a mutant that uncouples ATPase activity from directional movement reveal that nucleotide‐dependent conformational changes occur normally in one‐half of the AAA ring, but are disrupted in the other half. The large‐scale linker conformational change observed in the wild‐type protein is also inhibited, revealing that this conformational change is not required for ATP hydrolysis. These results demonstrate an essential role of the stalk in regulating motor activity and coupling conformational changes across the two halves of the AAA ring.


Introduction
Dyneins are minus-end directed, microtubule-based molecular motors that belong to the AAA+ (ATPases associated with diverse cellular activities) superfamily of proteins. Cytoplasmic dynein is responsible for the transport of numerous cargoes along microtubules (MTs), such as organelles, vesicles, viruses, and mRNAs (Vale, 2003;Vallee et al, 2004). In addition, cytoplasmic dynein plays key roles in facilitating basic cell biological processes such as spindle positioning during mitosis (Kiyomitsu & Cheeseman, 2013). Mutations and defects in cytoplasmic dyneins are associated with many diseases such as neurodegenerative diseases and cancers (Roberts et al, 2013).
The cytoplasmic dynein holoenzyme is composed of two identi-cal~500 kDa heavy chains and multiple associated polypeptide chains that primarily bind to the N-terminal tail of dynein (Pfister et al, 2006). Regulatory proteins such as Lis1 and NudE bind to some dyneins and can modify its motility properties (Kardon & Vale, 2009;Vallee et al, 2012). To initiate processive motility for cargo transport, human cytoplasmic dynein also requires dynactin as well as cargo-adaptor proteins such as BicD and Hook3 (McKenney et al, 2014;Schlager et al, 2014). However, the core element for motility of all dyneins lies in the conserved motor domain of the heavy chain, which consists of six different AAA domains that are linked together as an asymmetric hexameric ring (AAA1-AAA6). Only AAA1-AAA4 can bind nucleotides (Burgess et al, 2003;Carter et al, 2011;Schmidt et al, 2015;Kon et al, 2012;Schmidt et al, 2012;Cho et al, 2008;Kon et al, 2004;Fig 1A); ATP hydrolysis in AAA1 is microtubules (MTs). After ATP hydrolysis and phosphate release, the motor binds MTs with stronger affinity (K d < 1 lM; Kon et al, 2009). In this manner, the AAA ring controls the affinity of the MTBD for MTs. Conversely, interaction of the MTBD with MTs regulates the ATPase activity in the AAA ring (Kon et al, 2009). How this allosteric communication occurs is still poorly understood. In the case of kinesin and myosin, the ATPase-and track-binding sites are located relatively close (within~25 Å) to each other in the same domain (Vale & Milligan, 2000). In dynein, however, the very small~10 kDa microtubule-binding domain is spatially separated from the AAA ring by the~135 Å long coiled-coil stalk (Gibbons et al, 2005;Imamula et al, 2007;Carter et al, 2008;Kon et al, 2009;Redwine et al, 2012). Furthermore, the stalk is positioned between AAA4 and AAA5, which is on the opposite side of the ring from AAA1, resulting in a~240 Å separation between the main catalytic site and the MTBD.
To enable two-way communication between the MTBD and AAA ring, it has been suggested that the stalk undergoes conformational changes (Oiwa & Sakakibara, 2005;Shima et al, 2006;Kon et al, 2009). One hypothesis is that sliding between the two antiparallel helices of the stalk coiled-coil leads to changes in their register with respect to each other, with each registry corresponding to different microtubule affinities; the stalk in the b+ registry results in a low MT affinity state, and the a registry results in high MT affinity (Gibbons et al, 2005;Kon et al, 2009). This is further supported by structural work which has shown that when ADP-vanadate (ADPvi) is bound to AAA1, the coiled-coil 2 (CC2) of the stalk is kinked and slides together with the buttress relative to coiled-coil 1 (CC1; Schmidt et al, 2015). Another study speculates that local melting of the coiled-coil between different states of the hydrolysis cycle plays a major role in the communication (Gee & Vallee, 1998;Nishikawa et al, 2016). However, how relative length changes in the stalk either via sliding or local melting drive the communication between the ring and the MTBD is not well understood.
To gain better insights into the allosteric communication between the AAA ring and the MTBD, we have identified mutants in the dynein stalk that block communication between the ATPase-and microtubule-binding sites. These mutants show diffusive movement along MTs and also hydrolyze ATP at maximal rates in a microtubule-independent manner. Structural characterization by cryoelectron microscopy (cryo-EM) of one of these mutants reveals a stabilization of a previously uncharacterized open conformation of the AAA ring in the presence of the non-hydrolyzable ATP analogue AMPPNP. In the presence of ADP-vanadate (ADP-vi), mimicking the post-hydrolysis state of dynein, we observed that this mutant is primed for hydrolysis, but with the linker in an extended conformation, which differs from the bent conformation of the linker in wildtype dynein Schmidt et al, 2015). This result reveals that linker bending is not essential for ATP hydrolysis. Moreover, we gained new insights into domain movements in the AAA ring. The cryo-EM structure of the mutant in AMPPNP and A B D C Figure 1. Single-molecule motility properties of dynein stalk mutants reveal mutants with nucleotide-independent diffusive motility.
A Structure and domain organization of the motor domain of cytoplasmic dynein [PDB 4RH7 (Schmidt et al, 2015)]. Inset shows zoom of MTBD (white) and coiled-coil stalk (gray), which consist of two helices, CC1 and CC2. Well conserved residues that were used as anchor points to define CC1 and CC2 in this study are depicted as red spheres. Numbering is based on yeast cytoplasmic dynein. B Table showing

Stalk mutants show nucleotide-independent diffusion
Given the spatial separation between dynein's catalytic AAA ring and the MTBD, it is apparent that allosteric communication must be mediated in some way via the stalk ( Fig 1A). To understand what regions of the stalk may play a role in allosteric communication, we aligned and analyzed 534 sequences of dynein's motor domain. We found that the length of the stalk is very well conserved (99% of the sequences have the exact same stalk length) among species and types of dynein, such as cytoplasmic, axonemal, and IFT dynein, but the sequence is not (Appendix Note S1, Fig EV1). Based on the conserved length of the stalk and our sequence analyses, we decided to investigate how insertions and deletions in the stalk affect dynein's motility. We designed a panel of 18 insertion and deletion mutants in the yeast cytoplasmic dynein background, based on our sequence analysis (Fig 1B, Appendix Figs S1 and S2, Appendix Table S1). We expressed and purified GST-dimerized versions of each mutant ( Fig EV2A) with an N-terminal GFP (Reck-Peterson et al, 2006;DeWitt et al, 2012), assessed the quality of the protein using negative stain electron microscopy to ensure structural integrity, and used single-molecule total internal reflection fluorescence (TIRF) microscopy assays (Reck-Peterson et al, 2006;Yildiz & Vale, 2015) for initial characterization of single-molecule motility. Our panel of mutants displayed a wide variety of phenotypes (Fig 1B and C). Of the eighteen mutants, seven mutants (mutants 1, 3, 4, 6, 7, 11, and 12) showed single-molecule movement with velocities and processivity that were between~50 and 100% of the wild-type protein ( Fig EV2). Remarkably, some of these mutants had relatively large insertions of 6 (mutant 4) or 9 (mutant 11) residues or a deletion of five residues (mutant 12), but still moved in a similar way and direction as the wild-type motor. Of these seven mutants that showed wild-type phenotypes, six are in CC1 (mutants 1, 3, 4, 7, 11, and 12), suggesting that this helix is more tolerant of changes in length than CC2 (mutant 6; Fig EV1L, Appendix Figs S1-S3, Appendix Note S2). One region that is particularly sensitive to mutation is at the interface of the stalk and buttress (Fig EV1A). Most of the mutations that resulted in a dead (mutants 8, 9, 10, 17, and 18) or unstable (mutants 15 and 16) motor are clustered in the proximal region of the stalk, close to the AAA ring (Fig EV1L), and are in regions that are important for the stalk and buttress interaction ( Fig EV1A and L, Appendix Fig S2). This stalk and buttress interface has been shown previously to play a role in nucleotidedependent conformational change (Schmidt et al, 2015), and thus, our observations suggest that mutations in the stalk and buttress interface can severely compromise dynein motility (Appendix Note S2), consistent with the model that the stalk and buttress interface is critical for dynein motility. Interestingly, we observed one mutant (mutant 14) that contains two distinct populations of molecules: The major population (96%, population 1) transiently binds to and releases from microtubules, and the minor population (4%, population 2) appears to move in a similar fashion as the wild-type motor (Fig EV3A and B). Surprisingly, the site of mutation for mutant 14 overlaps with that of mutant 18, yet single-molecule properties observed for mutant 18 show a dead motor (Fig EV2C, Appendix Figs S1 and S2). Lastly, three mutants from our panel (mutants 2, 5, and 13) presented a diffusive-like behavior, with single molecules randomly moving back-and-forth along the microtubule (Fig 1B and C, Appendix Fig S4, Movies EV1-EV4). This observation suggests that these diffusive-like motors are weakly bound to microtubules but unable to undergo effective unidirectional motion. We further analyzed the movement of mutant 5 along microtubules by measuring the displacement distance and directionality per 1-s interval. The histogram of the displacements (Appendix Fig S5) reveals a uniform Gaussian distribution centered close to zero with an average displacement of À3.3 nm. This analysis supported the notion that the back-and-forth motion of mutant 5 reflects random thermal-driven motion along the microtubule.
We decided to further characterize the three interesting mutants (mutants 2, 5, and 13) that showed similar one-dimensional diffusion along the microtubule and the one mutant (mutant 14) that showed weak binding and occasional directional motion. To assess the nucleotide dependence of the diffusive phenotypes, we carried out single-molecule experiments in the absence of ATP. As expected, the wild-type control showed no movement and was rigor bound to microtubules (Appendix Fig S4, Movie EV5). Surprisingly, in the absence of ATP, all three mutants (mutants 2, 5, and 13) displayed diffusive behavior very similar to that observed in the presence of ATP ( Fig 1D, Appendix Fig S4). Diffusion, that we observed even in the absence of nucleotide, suggests that mutant 2, mutant 5, and mutant 13 have a weakened interaction with microtubules ( Fig 1B,  Appendix Fig S4, Movies EV5-EV8). Mutant 14 also seems to have weak affinity for microtubules in the apo state because we observed transient binding events in the absence of ATP ( Fig EV3C).
We also assessed the nucleotide-dependent binding affinity of dynein for microtubules using a cosedimentation assay. In wild-type dynein, the motor binds tightly to microtubules in the absence of ATP, but weakly in the presence of ATP (Fig 2A). In contrast to the nucleotide-dependent microtubule affinity of wild-type enzyme, the microtubule affinity of the diffusive mutants (mutants 2, 5, and 13) and the transient binding mutant (mutant 14) was low in the absence of nucleotide and in the presence of ATP or AMPPNP (Figs 2B-D and EV3E, Appendix Table S2), which is consistent with the single-molecule motility results. These results confirm that the diffusive mutants and mutant 14 have a weakened microtubule affinity which remained unchanged in different nucleotide states.
Since we did not observe any directional movement of these three mutants in single-molecule assays, we asked whether there is any net directionality in a microtubule gliding assay when there are many motors interacting with a microtubule. In this microtubule gliding assay, dimeric dyneins (wild-type or mutants) were attached to a glass coverslip (Appendix Fig S6A). Results from this assay show that the three mutants generated microtubule gliding across the glass surface, although their velocities were~10-fold lower than wild-type dynein (Appendix Fig S6B, Movies EV9-EV12). This ª 2019 The Authors The EMBO Journal 38: e101414 | 2019 phenotype is reminiscent of human cytoplasmic dynein (dynein 1) purified from rat brains (McKenney et al, 2014), which also shows diffusive motility in single-molecule assays, but shows robust directional movement in gliding assays. We also determined whether the microtubules were moving in the same direction as for wild-type dynein and assessed the direction of motion with single molecules of a human homodimeric kinesin 1 (K490; Tomishige et al, 2006), which moves processively toward the microtubule plus end (Appendix Fig S6A). By observing the direction of kinesin movement along the gliding microtubules, we could assess their polarity. Our results showed that the direction of mutant 2, 5, and 13 in microtubule gliding assays was the same as for wild-type dynein. In conclusion, mutants 2, 5, and 13 show nucleotide-independent diffusive movement as single molecules, while ensembles of these motors can produce extremely slow directional movement toward the MT minus end.

Diffusive mutants show microtubule-independent hyperactive ATP hydrolysis
Since single-molecule analysis of mutants 2, 5, and 13 showed diffusive movement uncoupled from the nucleotide state, we asked whether these mutants were capable of hydrolyzing ATP. One possible hypothesis was that the mutants could no longer bind or hydrolyze ATP, while another possibility is that ATP hydrolysis was uncoupled from directional movement. We measured the ATPase activity of mutant 2, mutant 5, and mutant 13 at varying concentrations of microtubules. In wild-type dynein, ATPase activity is stimulated in the presence of microtubules, resulting in a characteristic increase in ATPase activity as microtubule concentration increased, until maximal ATPase activity is reached. For wild-type dynein, we measured a basal ATPase turnover of 0.75 AE 0.34 ATP/motor/s, which increased with increasing concentrations of microtubules to a k cat of 15.18 AE 1.18 ATP/motor/s and K M of 0.50 AE 0.17 lM for tubulin dimer ( Fig 2E). These ATPase values are similar to those previously reported (Carter et al, 2008;Cho et al, 2008;Toropova et al, 2014; Appendix Table S3). Surprisingly, and in contrast to wild-type dynein, the three diffusive mutants (mutants 2, 5, and 13) showed high basal ATPase activity that did not significantly increase upon the addition of microtubules. Interestingly, the basal ATPase activities of mutants 5 and 13 were very similar to the maximal microtubule-stimulated ATPase activity of the wild-type protein ( Fig 2E and F, Appendix Table S4). In addition to the diffusive mutants (mutants 2, 5, and 13), mutant 14 also showed high basal ATPase activity that is independent of microtubule concentration ( Fig EV3D, Appendix Note S2). Together, these results indicate that the four weak binding mutants (2, 5, 13, and 14) all showed high ATPase activity and loss of microtubule regulation of the ATPase activity.
Structural basis for hyperactivity of mutant 5 Our functional and biochemical assays showed that insertions and deletions in mutants 2, 5, and 13 result in (i) diffusive movement of single dynein molecules on microtubules, (ii) constitutively hyperactive ATPase, and (iii) constitutively weak microtubule binding that is not modulated by nucleotide. Taken together, these results suggest that these mutations disrupt the two-way communication between the MTBD and AAA ring in the dynein motor domain. Next, we sought to understand the structural basis underlying the uncoupling between the microtubule and ATPase sites in the diffusive mutants. Because of its high basal ATPase activity, we decided to focus on mutant 5. We first collected a cryo-EM dataset of mutant 5 in the presence of 2 mM AMPPNP to mimic one of the post-force-generating states at AAA1 and AAA3. After 3D classification and refinement, we identified two distinct classes, with reconstructions at~7.5-8 Å resolution (Fig EV4A and B, Appendix Fig S7). This resolution allowed us to establish conformational changes at the subdomain level and model helices in some parts of the structure (Fig EV4C, Appendix Fig S7). Each AAA domain consists of a large subdomain (AAAL) and a small subdomain (AAAs), which can be considered as rigid bodies in the context of our resolution. Each AAAL and AAAs subdomain is fit independently as rigid bodies into each density map to generate a model corresponding to each map.  Table S2 shows fit equation and rate quantification for microtubule affinity data. E Microtubule-stimulated ATPase activity of wild-type (gray), mutant 2 (red), mutant 5 (orange), and mutant 13 (purple). Appendix Table S4 shows fit equation and rate quantification for ATPase data. F Bar plot of basal ATPase activity of wild-type (gray), mutant 2 (red), mutant 5 (orange), and mutant 13 (purple).
Data information: Error bars show standard deviation of three repetitions of different dynein preparations.

of 13
The EMBO Journal 38: e101414 | 2019 ª 2019 The Authors The most evident change in the motor domain of the majority (~71% of all particles-class 1, 7.7 Å resolution) of mutant 5 particles was a substantial opening between the small and large domains of AAA5 (Fig 3A and B), which was previously only observed as a minor conformation for the wild-type motor (Appendix Note S3, Appendix Fig S8). In addition, density for most of the distal stalk as well as the buttress is missing, suggesting that these regions are flexible. For the minor conformation (~29% of all particles-class 2, 7.6 Å resolution), the cryo-EM map shows a closed ring with no gap between the small and large domains of AAA5 and the helices of the initial part of the stalk and for the buttress are well defined (Figs 3A and EV4C). In this class, we can identify a conformation that has previously been referred to as the high microtubule affinity state in which the coiled-coil 2 of the stalk is not kinked (Fig EV4C;Schmidt et al, 2015). An additional and more subtle difference between the class 1 density and class 2 density is found in the N-terminal GFP tag at the end of the linker. In contrast to class 2, for class 1 (major class with "open" ring), the density for the N-terminal GFP tag is not well defined (Fig EV4D and E), which may indicate that the N-terminus of the linker is more flexible and potentially undocked from the ring at AAA5. Looking at domain movements in both class 1 and class 2 (Fig 3C and D), we also found that AAA2L is positioned away from the active site of AAA1 (Fig EV4F-R). Since the gap between AAA1 and AAA2 must close for productive ATP hydrolysis, we concluded that the ring of mutant 5 in the AMPPNP state is not primed for hydrolysis, as is true for wild-type dynein.
We next examined mutant 5 in the ADP-vanadate (ADP-vi) state, which mimics the post-hydrolysis state of dynein (Schmidt et al, 2015). In this state, the AAA domains in wild-type dynein adopt a more compact conformation in which the gap between AAA1 and AAA2 closes, which primes AAA1 for nucleotide hydrolysis ( Fig EV4R). In addition, in the wild-type protein, the linker changes from a "straight" conformation (extended linker spanning from AAA1 to AAA5) to a "bent" conformation (the N-terminus of the linker making contacts with AAA 3/2). Our cryo-EM data for mutant 5 in the presence of 2 mM ATP and 2 mM vanadate resulted in ã 9 Å reconstruction, for which subdomain movements could be mapped with confidence (Figs 4A and EV5A). Based on fitting AAAs and AAAL domains into our density as described above, our data show that the gap between AAA1L and AAA2L for mutant 5 closes when transitioning from the AMPPNP to the ADP-vi state (Fig 4B,  Movie EV13), similar to what was observed for wild-type dynein. The AAA2L domain of mutant 5 undergoes a rotation between the AMPPNP and ADP-vi state of~21°which is similar to the~20°d omain rotation in wild-type dynein (Fig EV5C and D). The approximate distance between the arginine finger and Walker-A motif of mutant 5 and wild-type decreases from~22 to~20 Å in the AMPPNP state, respectively to~17 and~14 Å in the ADP-vi state, respectively (Fig EV5E), highlighting that the gap between AAA1L and AAA2L for mutant 5 and wild-type dynein indeed closes in a similar manner.
Unlike the bent linker observed for wild-type dynein in the presence of ADP-vi (Schmidt et al, 2015), the mutant 5 linker is not bent at the hinge-point ( Fig 4A). However, the N-terminal region of the linker is undefined, suggesting increased flexibility at its N-terminal region (Figs 4A and C and EV5B). We confirmed the binding of vanadate to the AAA1 nucleotide-binding pocket of mutant 5 by demonstrating vanadate-mediated UV photo-cleavage and vanadate inhibition of the ATPase activity (Fig EV5F-H). Thus, our structural data for mutant 5 in the presence of ATP and vanadate indicate that the motor is primed for hydrolysis, but does not undergo the large conformational change in the linker that is believed to be essential for motility (Burgess et al, 2003;Bhabha et al, 2016).
To better understand how mutant 5 can be primed for hydrolysis while the linker remains in a straight conformation, we analyzed the   (Figs 5B and EV5I). In contrast to the pronounced nucleotide-dependent motions in the AAA2s-AAA5L half of the ring for wild-type dynein, very little motion is observed for these domains in mutant 5 and the mode of movement is different (Fig 5B). Thus, between the AMPPNP and ADP-vi states, mutant 5 exhibits normal AAA domain movements in one-half of the ring (AAA5s-AAA2L), but shows a considerable lack of motion in the other half (AAA2s-AAA5L; Fig 5A and B). This result reveals that this stalk mutation uncouples nucleotide-dependent conformational changes in the two halves of the ring. Moreover, these results provide new insight into domain movements of the AAA ring and could explain why mutant 5 shows high ATPase activity but little motility, as will be described in the discussion.

Discussion
We have identified mutations in the dynein stalk that show nucleotide-independent weak binding to microtubules and diffusional motion along the microtubule surface. A microtubulestimulated ATPase assay revealed that these mutants hydrolyze ATP independently of microtubule concentration; two of these mutants (mutants 5 and 13) are hyperactive and have a basal ATPase activity that is as high as the maximal microtubule-stimulated turnover rate in the wild-type protein. Performing structural analysis on one of these mutants (mutant 5) using cryo-electron microscopy, we found that nucleotide-dependent "straight-tobent" conformational change in the linker domain is inhibited. Moreover, we observed that AAA domain movements in one part of the ring are altered, while the other part of the ring becomes primed for hydrolysis very similarly as in wild-type dynein. These data provide new information on how the microtubulebinding domain (MTBD), stalk, linker, and AAA ring A C B Figure 4. Cryo-EM structure of mutant 5 in the presence of ADP-vanadate shows priming for hydrolysis with unbent linker.
A Cryo-EM reconstructions and fitted models from 3D classification of the data. The cryo-EM density map is shown as a semi-transparent surface with a fitted model (fit as described in Materials and Methods) as cartoon. The black arrow indicates the position of the gap between AAA5L and AAA5s of mutant 5 in the presence of ADP-vi. This gap is smaller than for mutant 5 in the presence of AMPPNP but larger than in wild-type ADP-vi ( Fig EV5J). Color coding of domains is the same as for were used. For mutant 5 AMPPNP, we used the class 1 structure. D Model for domain movements in the AAA ring of wild-type dynein during ATP hydrolysis at AAA1. The AAA ring can be divided into two halves that are connected by two springs. Upon ATP binding/hydrolysis, the gap between AAA1/AAA2 closes and moves AAA6/AAA5 which in turn pulls on AAA4 so that the gap between AAA4/AAA5 closes as well. In addition, this conformational change will pull the buttress and therewith change the stalk registry. E In the stalk mutant, the spring between AAA4/AAA5 does not close upon ATP binding/hydrolysis presumably due to a disruption of the stalk-buttress interface.
Moreover, the gap at the AAA4/AAA5 interface is larger for the stalk mutant in both states, AMPPNP and ADP-vi, than for wild-type in the AMPPNP state. This "loose spring" at AAA4/AAA5 uncouples these domains from the closure of the AAA1/AAA2 interface, and this accounts for microtubule-independent hydrolysis.
ª 2019 The Authors The EMBO Journal 38: e101414 | 2019 communicate with one another during the ATPase cycle, as discussed below.

Domain movements in the AAA ring
Dynein is a large and complex allosteric protein that must coordinate the conformations of four independent domains: (i) the AAA ring (consisting of 6 AAA domains), (ii) the largely helical linker (which spans over the ring and serves as a mechanical element), (iii) the small, globular microtubule-binding domain, and (iv) the stalk-buttress apparatus (a pair of antiparallel coiled coils that extend from the AAA ring and connect via the stalk to the microtubule-binding domain). Current structural data suggest that ATP binding to AAA1, with ADP bound at AAA3, drives full AAA ring closure (Kon et al, 2012;Bhabha et al, 2014;Schmidt et al, 2015), which is associated with a large-scale conformational change in the linker and a shift in registry of the two antiparallel coiled coils that affects the affinity of the distal microtubule-binding domain. However, the manner in which these different domains communicate with one another is incompletely understood. Previous models for conformational changes in the AAA ring upon ATP binding suggest a rigid body movement of AAA2-AAA4, which propagates as a rotational motion to AAA5s and AAA6L which in turn pull the buttress relative to the stalk Schmidt et al, 2015). ATP hydrolysis and/or product release then straightens the linker (thought to be the "power-stroke") and relaxes the ring back to its original conformation. Another model by Kon et al (2012) suggests an important role of the C-terminal domain (C-sequence), located on the surface of the ring opposite to the linker, in connecting AAA1L and AAA5s and triggering a movement of the buttress and, in turn, conformational changes in the stalk and MTBD. Even though the full C-terminal domain is not found in every dynein (Schmidt et al, 2015), the H1 alpha helix of the C-sequence that staples the AAA1L/AAA6s and AAA6L/AAA5s blocks together (Kon et al, 2012) appears to be present in virtually all dyneins.
Comparing the conformational states sampled by mutant 5 with the conformational states previously reported for the wild-type protein, we can understand how the stalk mutation in mutant 5 perturbs normal conformational changes in the dynein motor protein, thus provide insight into stalk-mediated conformational changes in the AAA ring.
Our cryo-EM data of mutant 5 revealed wild-type-like nucleotidedependent AAA domain conformational rearrangements in one part of the ring (AAA5s-AAA6-AAA1-AAA2L), but absence/alteration of these conformational changes in the other part of the ring (AAA2s-AAA3-AAA4-AAA5L; Fig 5A and B, Movie EV13). Specifically, movements in AAA5L and AAA4s are much smaller in magnitude, and different in their vector of movement (Fig 5B, Movie EV13). This result indicates that the domain movements in the AAA5s-AAA2L block are insulated, at least to some extent, from the rest of the ring and from disruptive mutations in the stalk (Fig 5C, Movie  EV14). Thus, we hypothesize that the two halves of the ring can undergo two independent modes of conformational change, which require the stalk-buttress apparatus to be properly coupled.
Based on the conformational changes seen in the cryo-EM data for mutant 5, we speculate that ATP binding does not primarily propagate in a clockwise (viewed from the linker side), domino-like manner from AAA1 to AAA6 (Carter, 2013;Bhabha et al, 2014;Schmidt et al, 2015). Instead, the data for mutant 5 show bidirectional domain movement around the ATP-bound pocket of AAA1, with a block of AAA5s-AAA6Ls-AAA1L moving toward the AAA1s-AAA2L block upon ATP binding to AAA1 (Movie EV14). The Cterminal domain might provide the underlying bridging support between AAA5s/AAA6L and AAA6s/AAA1L that allows this block of AAA domains to move in a unified manner, consistent with the proposal of Kon et al (2012).
The structural data also emphasize the important role that the stalk-buttress plays in coupling the conformational changes in the two halves of the ring. Although we cannot see the precise lesion in the stalk caused by mutant 5 due to flexibility in this region, the downstream effect is an enlarged gap between AAA5s and AAA5L, which we speculate is the underlying cause in the disruption in the allosteric communication within the ring (Fig 5D and E). Specifically, our data for mutant 5 in the presence of ADP-vanadate suggest that movement of AAA5s/AAA6L toward the nucleotide-binding pocket of AAA1 is unable to pull AAA5L/AAA4s with it (and the other half of the ring; Movie EV14, Fig 5C). The interaction between the buttress (emerging from AAA5s) and stalk (emerging from AAA4s) is likely needed for this coordination between the two halves of the ring. Interestingly, the failure to connect AAA5L/ AAA4s to the movement of the AAA5s-AAA6Ls-AAA1Ls-AAA2L block severely impacts the nucleotide-dependent movements of the half of the ring from AAA2s-AAA5L. This result suggests that domain movements in this part of the ring are dependent upon the integrity of the stalk-buttress apparatus and its connection to the autonomous nucleotide-driven motions of the AAA5s-AAA6Ls-AAA1Ls-AAA2L block.
This new model suggests why the stalk mutants hydrolyze ATP independently of microtubule concentration. Specifically, the disruption of the stalk-buttress interface allows the buttress and AAA5s/AAA6/AAA1L to undergo open-closed transitions accompanying ATP binding, hydrolysis, and product release, without any regulation by microtubules through the stalk-buttress apparatus (Fig 5D and E). Interestingly, buttress mutations that presumably disrupt the stalk-buttress interaction (Kon et al, 2012) also show high ATPase activity independent of microtubules, similar to our mutants 5 and 13. Thus, we suggest that the stalk-buttress interaction is a key regulator for dynein's ATPase activity by controlling the coupling between AAA4s/AAA5L and AAA5s/AAA6L and thus the coordination of domain movements in the two halves of the AAA ring.

Uncoupling of linker bending from robust ATP hydrolysis
Our structural analysis revealed that mutant 5 has uncoupled nucleotide-dependent changes in one-half of the AAA ring and ATP hydrolysis from the large conformational change in the linker domain. Previous findings DeWitt et al, 2015) suggested that in order for ATP hydrolysis to proceed, the linker must be undocked from the ring, to allow full closure of AAA2L. Thus far, structures in which the AAA ring is primed for hydrolysis (i.e., the gap between AAA2 and AAA1L is closed) have been accompanied by linker bending and docking onto AAA3/2 Schmidt et al, 2015). However, our cryo-EM reconstruction of mutant 5 in the presence of ADP-vanadate shows a motor 8 of 13 The EMBO Journal 38: e101414 | 2019 ª 2019 The Authors that is primed for hydrolysis with an unbent linker (Fig 4A and C). Weak/missing density for the N-terminus of the linker suggests that it is flexible and might be undocked; however, it is clearly not in a bent conformation (Fig EV5B). These data therefore suggest that linker bending is not a prerequisite to prime the motor for hydrolysis. However, the linker bending is not only necessary for efficient directional motion Cleary et al, 2014; since single mutant 5 motors show random bidirectional motion) but also important to set the directionality of the motor (Can et al, 2019).
Together, these data further support our model that the hydrolysis cycle arises from autonomous conformational changes within the AAA5s-AAA6Ls-AAA1Ls-AAA2L block and neither require linker bending nor an intact stalk-buttress interface.
Our structural data also allow us to speculate on how linker bending is initiated and why it fails to occur in the mutant 5. In the nucleotide-free and ADP state, the linker forms contacts with AAA5L. Movement of AAA5L upon ATP binding in AAA1 may induce a steric clash with the linker [perhaps with contributions from AAA4L (Schmidt et al, 2015)], and linker bending may ensue to minimize such clashes. In mutant 5 however, AAA5L movement is minimal and thus may be unable to induce the linker steric clash (Fig EV5K), which is highlighted in the "gap" between AAA5L and AAA5s. We therefore speculate that the interface between AAA5s and AAA5L (AAA5s connecting to the AAA1 ATPase site via the AAA5s-AAA6Ls-AAA1Ls-AAA2L block and AAA5L interacting with the linker and AAA4s/ stalk/MTBD) may be a critical region for coordinating ATPase activity (AAA1), microtubule binding (stalk/MTBD), and mechanics (linker) in the dynein motor domain (Movie EV14).

Bioinformatic analysis of dynein sequences
The detailed process is described in Appendix Note S1. Briefly, we used 677 unique axonemal and cytoplasmic dynein heavy chain sequences from 229 fully sequenced eukaryotic genomes, which we received from Christian Zmasek, Godzik Lab, Burnham. This dataset was pruned based on well-defined criteria as listed in Appendix Note S1 and analyzed using Jalview (Waterhouse et al, 2009). Remaining sequences were aligned using MAFFT (Katoh et al, 2002) in the Bioinformatic Toolkit (Alva et al, 2016), and mutations in the stalk were manually identified by comparing sequences in Jalview. More information on the sequence alignments can be found in the Appendix Note S1.

Yeast strains used in this study
Recombinant Saccharomyces cerevisiae cytoplasmic dynein (Dyn1) truncated at the N-terminus (1219-4093 aa) was used in this study. All constructs used in this study are listed in Appendix Table S1. Dimeric constructs are based on VY208 and were created by artificially dimerization through an N-terminal GST-tag (Reck-Peterson et al, 2006) and tagged with a HaloTag (Promega) at the C-terminus as well as a GFP at the very N-terminus. Monomeric constructs (VY137) are GFP tagged at the N-terminus. Stalk mutations were inserted by homologous recombination as previously described (Reck-Peterson et al, 2006).

Microtubule preparation
Tubulin was purified and polymerized as previously described (McKenney et al, 2014). For single-molecule motility assays, unlabeled tubulin, biotinylated tubulin, and fluorescent tubulin were mixed at an approximate ratio of 20:2:1 in BRB80 (80 mM Pipes (pH 6.8), 1 mM EGTA, and 1 mM MgCl 2 ). For the gliding assay, unlabeled tubulin and fluorescent tubulin were mixed at an approximate ratio of 20:1 in BRB80. For tubulin that was used in the ATPase assay as well as the microtubule affinity assay, only unlabeled tubulin was used. We added 1 mM GTP to all polymerization reactions. Then, the mixtures were incubated for 15 min in a 37°C water bath. 20 lM of Taxol (Sigma, T1912) was added afterward, and the mixture was incubated for 2 more hours at 37°C. Before usage, microtubules were spun over a 25% sucrose cushion in BRB80 at~160,000 g for 10 min in a tabletop centrifuge.

Gliding and single-molecule motility assay
We made custom flow chambers using laser-cut double-sided adhesive sheets (Soles2dance, 9474-08x12-3M 9474LE 300LSE). We used glass slides (Thermo Fisher Scientific, 12-550-123) and coverslips (Zeiss, 474030-9000-000). We cleaned the coverslips in a 5% v/v solution of Hellmanex III (Sigma, Z805939-1EA) at 50°C overnight and then washed them extensively with Milli-Q water. The flow cells were assembled in a way that each chamber holds approximately 10 ll.
Every data collection was carried out at room temperature (~23°C) using a total internal reflection fluorescence (TIRF) inverted microscope (Nikon Eclipse Ti microscope) equipped with a 100 × (1.45 NA) oil objective (Nikon, Plan Apo ƛ). We used an Andor iXon 512 × 512 pixel EM camera, DU-897E and a pixel size of 159 nm. Dynein [always as dimer and either labeled with GFP only or with GFP and a Halo488 dye (Promega, G1001)] was excited with a 488 nm laser (Coherent Sapphire 488 LP, 150 mW), kinesin with a 561 nm laser (Coherent Sapphire 561 LP, 150 mW), and microtubules with a 640 nm laser (Coherent CUBE 640-100C, The EMBO Journal 38: e101414 | 2019 100 mW). For the gliding assay, images were recorded with 100-ms exposure time and a 2-s frame rate for MTs and a 100-ms frame rate for kinesin. For the single-molecule assay of dynein, we used 100ms exposures and a 2-s frame rate and a 100-ms frame rate for kinesin. The acquisition software was lManager (Edelstein et al, 2010) 2.0, and data were analyzed in ImageJ (Schneider et al, 2012). For the gliding assay, we first added 10 ll of GFP antibody (Abcam, ab1218) and incubated for 5 min. Then, we washed with 20 ll of DAB with 2 mg/ml b-casein and 0.4 mg/ml j-casein. We then added 10 ll of dimeric dynein and incubated for another 5 min which was followed by an additional wash with 20 ll of DAB with 2 mg/ml b-casein and 0.4 mg/ml j-casein. Next, we added 10 ll of polymerized microtubules and incubated for 5 min. Then, we washed with 30 ll of DAB with 2 mg/ml b-casein and 0.4 mg/ml jcasein. Finally, 10 ll of DAB with kinesin, 0.4 mg/ml j-casein, 10 lM Taxol, 1 mM Mg-ATP, and the PCA/PCD/Trolox oxygen scavenging system (Aitken et al, 2008) were added.
Prior to the single-molecule motility assays, dynein was labeled with Halo488 dye (Promega, G1001) as previously described . Briefly, dynein constructs were mixed with 20 lM Halo Alexa488 dye and incubated on ice for 10 min and a PD MiniTrap G-25 column (GE Healthcare) equilibrated with dynein gel filtration buffer was used to remove excess dye afterward.
The flow chambers for the single-molecule motility assay were prepared as previously described (Yildiz & Vale, 2015). Briefly, we first added 10 ll of 5 mg/ml Biotin-BSA in BRB80 and incubated for 2 min. Then, we washed with 20 ll of BRB80 with 2 mg/ml bcasein (Sigma, C6905) and 0.4 mg/ml j-casein (Sigma, C0406). Afterward, we added 10 ll of 0.5 mg/ml streptavidin in PBS for a 2-min incubation. Next, we again washed with 20 ll of BRB80 with 2 mg/ml b-casein and 0.4 mg/ml j-casein. This was followed by the addition of 10 ll of polymerized microtubules and a 5-min incubation. Then, we washed with 30 ll of DAB (50 mM K-Ac, 30 mM HEPES, pH 7.4, 2 mM Mg(Ac) 2 , 1 mM EGTA) with 2 mg/ ml b-casein, 0.4 mg/ml j-casein, and 10 lM Taxol. Finally, we added 10 ll of dynein and kinesin in DAB with 0.4 mg/ml jcasein, 10 lM Taxol, 1 mM Mg-ATP, and the PCA/PCD/Trolox oxygen scavenging system (Aitken et al, 2008). In the single-molecule assay where ATP was omitted, the final solution contained 10 ll of dynein in DAB with 0.4 mg/ml j-casein, 10 lM Taxol, and the PCA/PCD/Trolox oxygen scavenging system (Aitken et al, 2008). The acquisition software was lManager (Edelstein et al, 2010) 2.0.

Diffusion analysis of single-molecule movements
The imaging was performed as described for the single-molecule motility assay. Subsequently, tracks of wild-type dynein and mutant 5 along microtubules in the presence of 1 mM ATP were obtained. We tracked single molecules using the "localization microscopy" plug-in from lManager 2.0 (Edelstein et al, 2010) by fitting emitters with a Gaussian-based maximum-likelihood estimation (Mortensen et al, 2010) as previously described (Niekamp et al, 2019) and extracted tracks based on a nearest neighbor search. We then straightened these single-molecule traces along the main axis of motion along the microtubule using a principal component analysis implemented in the "localization microscopy" plug-in from lManager 2.0 (Edelstein et al, 2010). Following this, the displacement of wild-type and mutant 5 was binned into 1-s intervals and the polarity of microtubules was determined by analyzing the directionality of human homodimeric kinesin 1 (K490; Tomishige et al, 2006), which moves processively toward the plus end of microtubules.

ATPase assay
The ATPase assays were carried out in DAB (50 mM K-Ac, 30 mM HEPES, pH 7.4, 2 mM Mg(Ac) 2 , 1 mM EGTA) as follows. We mixed dynein (monomeric for all constructs) to a final concentration of 10-20 nM with 2 mM Mg-ATP (Sigma), 0.2 mM NADH (Sigma), 1 mM phosphoenolpyruvate (Sigma), 0.01 U pyruvate kinase (Sigma), 0.03 U lactate dehydrogenase (Sigma), 10 lM Taxol, 1 mM DTT, and 0-5 lM microtubules in DAB. Absorbance at 340 nm was continuously measured in an Eppendorf Spectrophotometer (UV-Vis BioSpectrometer), and the data were fit to the following equation  using an excel curve fitting routine: The vanadate inhibition of dynein ATPase activity was performed as previously described (Höök et al, 2005). Briefly, we mixed dynein [monomeric of wild-type and mutant 5 (from the same batch that was used to solve the structure of mutant 5 in the presence of ADP-vi)] to a final concentration of 20 nM with 1 mM Mg-ATP (Sigma), 0.2 mM NADH (Sigma), 1 mM phosphoenolpyruvate (Sigma), 0.01 U pyruvate kinase (Sigma), 0.03 U lactate dehydrogenase (Sigma), 10 lM Taxol, 1 mM DTT, 6 lM microtubules, and 0-100 lM vanadate (Sigma) in DAB. The vanadate was boiled for 10 min before usage. Absorbance at 340 nm was continuously measured in an Eppendorf Spectrophotometer (UV-Vis BioSpectrometer), and the turnover rate was calculated as described above.

Vanadate-mediated UV photo-cleavage
Protein from the same batch that was used to solve the structure of mutant 5 in the presence of ADP-Vi was used in the vanadatemediated UV photo-cleavage. The assay was performed in a similar way as previously described (Schmidt et al, 2015). Briefly, mutant 5 monomer was mixed with 2 mM Mg-ATP (Sigma) and 2 mM vanadate (Sigma) and either exposed to UV-light (365 nm) or kept in the dark for 90 min. The vanadate was boiled for 10 min before usage. Afterward, the samples were analyzed by SDS-PAGE.

Microtubule affinity assay
The microtubule affinity assays were carried out in DAB (50 mM K-Ac, 30 mM HEPES, pH 7.4, 2 mM Mg(Ac) 2 , 1 mM EGTA) as follows. We mixed dynein (monomeric for all constructs) to a final concentration of approx. 50 nM with 10 lM Taxol, 1 mM DTT, and 0-8 lM microtubules in DAB. For the measurements with ATP, we added 5 mM Mg-ATP (Sigma), and for the experiment with AMPPNP, we added 5 mM Mg-AMPPNP (Sigma). After a 3-min incubation at room temperature, the samples were spun over a 25%

of 13
The EMBO Journal 38: e101414 | 2019 ª 2019 The Authors sucrose cushion in DAB at~160,000 g for 10 min in a tabletop centrifuge. The concentration of dynein in the supernatant (unbound) and in the pellet (bound) was determined by measuring the intensity of the N-terminal GFP on a Typhoon laser scanner (GE Healthcare). The data were fit to the following equation using an excel curve fitting routine which B M is the maximum binding, K d is the dissociation constant, k basal is the basal "binding" fraction and accounts for the pelleting of dynein without microtubules present, and k obs is the observed fraction of dynein bound (pelleted) over the total amount of dynein. We could have also used the simplified equa- constant) but we wanted to account for potential pelleting of dynein without microtubules (k basal ). However, since k basal is very low, using the simplified equation gives almost identical results for B M and K d .

Electron microscopy data collection
For negative stain, data for mutants 5 (monomer) were collected on a Tecnai F20 microscope with a Tietz F416 CMOS detector at the New York Structural Biology Center (NYSBC). Leginon software (Suloway et al, 2005) was used for the semi-automated collection of 825 images at a magnification of ×62,000 and a pixel size of 3 Å per pixel. For cryo-EM data collection, 1,200 movies of mutant 5 (monomer) mixed with 2 mM AMPPNP were recorded with Seri-alEM (Mastronarde, 2005) at 300 kV on a Titan Krios (FEI) equipped with a K2 summit camera (Gatan) at 0.655 Å per pixel in super-resolution mode at Janelia Research Campus. Another 664 movies of the same mutant (mutant 5-monomer) mixed with 2 mM ATP and 2 mM vanadate were recorded with SerialEM at 200 kV on a Arctica (FEI) equipped with a K2 summit camera (Gatan) at 0.578 Å per pixel in super-resolution mode at New York University.

Electron microscopy data processing and analysis
For the images of the negatively stained sample, particles were selected using DoG picker  in APPION (Lander et al, 2009) and then extracted in Relion 2.1.0 (Scheres, 2012) into boxes of 180 × 180 pixels, leading to 156,199 boxes for mutant 5. A round of 2D classification was performed to remove junk and noisy particles, leading to 54,913 particles selected. Subsequent image processing steps were carried out using CryoSPARC (Punjani et al, 2017). After having generated an ab-initio model, those particles were used to generate eight 3D classes. Because of the similarity between all those classes, a final round of 3D refinement was completed using all of the particles. For the cryo-EM images (see also Appendix Table S5), the movies of mutant 5 with 2 mM AMPPNP were first aligned and binned to 1.31 Å per pixel with MotionCor2 v1.0.5 (Zheng et al, 2017), and then, the contrast transfer function parameters were estimated with GTCF 1.06 (Zhang, 2016). The particles were picked automatically in Relion 2.1.0 (Scheres, 2012) using a Gaussian blob as a reference, and further processing was done in CryoSPARC (Punjani et al, 2017). Out of the 310,085 regions automatically picked, 136,056 were kept after evaluation of 2D classes. Two ab-initio models were first generated in CryoSPARC, and the best one was used in a 4-class 3D heterogeneous refinement. Then, two 3D homogeneous refinements were completed: one with class 3 (here referred to as class 2-with 29% of remaining particles and with a resolution of 7.6 Å) and another one (here referred to as class 1-with 71% of remaining particles and with a resolution of 7.7 Å) with the three other classes which looked very similar and were therefore combined before refinement. Note that the overall and local resolutions we report are the average resolution after refinement in CryoSPARC v2.5.0 (Fig EV4A and B, and Appendix Fig S7). The final maps were then filtered for display using a B-factor of À400. For modeling, we used PDB 4W8F as a reference. The PDB file was split into 13 domains (small and large subdomains for each AAA domain, and the linker), and for each of those domains, we simultaneously fit all 13 subdomains into the map using UCSF Chimera (Pettersen et al, 2004). We noticed that the rigid body of the buttress region in class 2 map did not perfectly fit the densities (Fig EV4C). This model was therefore subjected to the real_space_refine algorithm in PHENIX (Adams et al, 2010) using two cycles and 100 iterations to optimize the fit. Figures and movies were generated with the UCSF Chimera package or the Pymol Molecular Graphics System (version 2.0, Schröodinger, LLC).
For the images of mutant 5 with 2 mM ATP and 2 mM vanadate acquired on the Arctica, a similar process was followed. First aligned and binned to 1.31 Å per pixel with MotionCor2 v1.0.5 (Zheng et al, 2017), and the contrast transfer function parameters estimated with GCTF 1.06 (Zhang, 2016). A first round of autopicking was conducted in Relion 2.1.0 (Scheres, 2012) using a Gaussian blob as a reference. Two of the resulting classes were then used as template for a round of reference-based auto-picking. Further processing was also conducted in CryoSPARC (Punjani et al, 2017). Out of 35,565 picked particles, 32,442 particles were kept for the generation of ab-initio models and a 4 and a 5-class heterogeneous refinement were tried. One class with 8,653 particles leads to a clear dynein 3D model that we refined to 9.2 Å ( Fig EV5A) and finally filtered for display using a B-factor of À400. Another class at 17 Å seemed to show only the AAA domains while the linker could not be seen. The model for the 9.2 Å map was constructed as for mutant 5 in the AMPPNP state, using rigid body docking of domains from PDB 4W8F, but was not further refined in PHENIX due to its lower resolution.

Statistics
For each result obtained, the inherent uncertainty due to random or systematic errors and their validation are discussed in the relevant sections of the manuscript. Details about the sample size, number of

Data availability
Density maps for the structures were deposited in the Electron Microscopy Data Bank under accession codes EMD-7829 (mutant 5 in the presence of AMPPNP, class 1), EMD-7830 (mutant 5 in the presence of AMPPNP, class 2), and EMD-9386 (mutant 5 in the presence of ADP-vanadate). The sequence alignment files used to create the mutants are available as Appendix. All other data are available from the corresponding author upon request.
Expanded View for this article is available online.