Katanin spiral and ring structures shed light on power stroke for microtubule severing

Abstract

Microtubule-severing enzymes katanin, spastin and fidgetin are AAA ATPases important for the biogenesis and maintenance of complex microtubule arrays in axons, spindles and cilia. Because of a lack of 3D structures, their mechanism has remained poorly understood. Here we report the first X-ray structure of the monomeric AAA katanin module and cryo-EM reconstructions of the hexamer in two conformations. These reveal an unexpected asymmetric arrangement of the AAA domains mediated by structural elements unique to severing enzymes and critical for their function. Our reconstructions show that katanin cycles between open spiral and closed ring conformations, depending on the ATP occupancy of a gating protomer that tenses or relaxes inter-protomer interfaces. Cycling of the hexamer between these conformations would provide the power stroke for microtubule severing.

INTRODUCTION

Cells constantly assemble and disassemble their microtubule cytoskeleton through the concerted action of microtubule polymerases, depolymerases, crosslinkers and severing enzymes. Microtubule severing enzymes spastin, katanin and fidgetin generate internal breaks in microtubules thus modulating their dynamics and organization¹. The mechanism used by these enzymes to destabilize the microtubule, a polymer with a flexural rigidity comparable to Plexiglas®² is still poorly understood. Microtubule severing enzymes are critical in a wide range of cell biological processes including biogenesis of neuronal and non-centrosomal microtubule arrays^3–7, phototropism^5,7, spindle scaling^7–11, chromosome segregation^10,12, and control of centriole and cilia numbers^13–15. They are found throughout the animal kingdom, plants and protozoa, and mutations in microtubule severing enzymes cause severe neurodegenerative and neurodevelopmental disorders^1,14–16. All known microtubule-severing enzymes belong to the family of ATPases associated with various cellular activities (AAA ATPases).

Katanin was the first microtubule-severing enzyme discovered^17,18. Fittingly, it was named after the Japanese for sword. Katanin is composed of a catalytic (p60) and a regulatory (p80) subunit. The catalytic subunit contains the AAA ATPase motor and is sufficient for microtubule severing^10,19. The p80 subunit regulates association with the centrosome and enhances microtubule binding^18–21. ATP hydrolysis is required for severing and katanin p60 ATPase is stimulated by the microtubule¹⁹. The p60 AAA ATPase domain is connected to an N-terminal microtubule interacting and trafficking (MIT) domain through a poorly conserved linker, a sequence arrangement shared with the microtubule-severing enzymes spastin and fidgetin¹. In large part due to a lack of 3D structures, the mechanism of microtubule severing by this class of enzymes has remained poorly understood 25 years after the discovery of the first microtubule enzyme^17,18. Here we report the first X-ray structure of the monomeric AAA katanin module and cryo-EM reconstructions of the hexamer in two conformations. These reveal an unexpected asymmetric arrangement of the AAA domains mediated by structural elements unique to microtubule severing enzymes and that we show are critical for their function. Cryo-EM reconstructions at 4.4Å and 6Å resolution of the katanin hexamer reveal an open spiral and a closed ring conformations of the AAA core, depending on the nucleotide occupancy of a gating protomer that closes a ~ 40Å wide gate in the katanin hexamer. Together with solution small-angle X-ray scattering (SAXS) reconstructions, our integrated structural study supports a model whereby katanin makes multivalent interactions with the microtubule through its AAA core, flexible MIT domains and a newly defined linker element that crowns the AAA ring, and engages the C-terminal tails of tubulin through conserved pore loops that gradually pull tubulin dimers out of the microtubule lattice by cycling between open spiral and closed AAA ring conformations. Due to the high sequence homology, a similar mechanism is likely shared by all microtubule-severing enzymes. Lastly, our integrated study also provides insight into the many katanin mutants identified from classic genetic studies on meiosis where the katanin gene (also known as mei-1) was first identified as well as disease mutations found in microtubule severing enzymes.

RESULTS

Solution structure of the katanin hexamer

Both microtubule severing enzymes katanin and spastin assemble into a hexamer in their active, ATP-bound state^1,22. This hexamer is labile and falls apart at lower protein concentrations²². Consequently, the high-resolution architecture of an assembled microtubule-severing enzyme has remained elusive. Analytical ultracentrifugation (AUC) shows that the catalytic subunit of Caenorhabditis elegans katanin (MEI-1) populates a mixture of oligomerization states as a function of concentration and nucleotide. At the highest concentration (24 μM) and in the presence of ATP, the dominant species is a hexamer that dissociates at lower concentrations (Supplementary Figs. 1a, b). Introduction of a commonly used mutation in the Walker B element that retains ATP binding, but prevents hydrolysis, stabilizes the hexamer²² to concentrations as low as 3 μM (Supplementary Fig. 1c). Solution SAXS measurements of this species (hereafter referred to as katanin) reveal an elongated molecule with a maximal diameter (D_max) of ~ 250Å (Fig. 1a–c). SAXS ab initio reconstructions generated by imposing six-fold symmetry reveal a disc structure ~130Å in diameter with slender arms that fan out at variable angles (Fig. 1a and Supplementary Fig. 1d, e, f, h and Methods) indicating a high-degree of flexibility. This architecture is reminiscent of the one reported earlier for spastin²³. The central ring dimensions are compatible with a hexameric assembly of AAA ATPase domains^23,24. The MIT domains reside on average at the tips of the extended arms, as a deletion construct that misses this domain (hereafter ΔMIT katanin, Fig. 1d) has a D_max of ~180Å and shorter arms (Fig. 1b, c and Supplementary Fig. 1g, h).

Figure 1. Architecture of monomeric and assembled katanin from X-ray diffraction, solution SAXS and cryo-EM structures.

(a) Top and side views of twenty superimposed ab initio bead models of full-length katanin p60 calculated from solution SAXS data (Supplementary Fig. 1 and Methods). (b) Top view of an overlay of representative models for full-length (grey) and ΔMIT katanin (magenta). Approximate dimensions shown. (c) Pair distance distribution functions for full-length and ΔMIT katanin. (d) Domain diagram of katanin p60; MIT, microtubule interacting and trafficking domain, yellow; flexible linker, pink; AAA domain, grey; fishhook-shaped linker element, light-blue; α1, dark-blue; α12 and 468–472 C-terminal residues, dark-orange. NBD and HBD are highlighted by light green and dark green hatches, respectively as in panel (e). Residue numbers for C. elegans katanin. (e) Cartoon representation of the katanin AAA core crystal structure; N-terminal helix α1, blue; NBD, light green; HBD, dark green; C-terminal helix α12, dark-orange; sulfate ion as stick representation; unresolved residues shown as spheres. (f) Views of the final sharpened 3D density map (13 σ) of the katanin hexamer filtered to 4.4 Å resolution showing a spiral architecture (rotation angles between the different views indicated with arrows). Top, microtubule-binding face²³. Approximate dimensions shown. Structural elements color-coded as in (d). Bracket indicates open gate between P1 and P6.

X-ray and cryo-EM structures reveal asymmetric architecture of the katanin hexamer

To obtain a higher-resolution view of katanin, we employed a hybrid structural approach combining X-ray crystallography of the monomer and cryo-EM of the hexamer. We solved the X-ray structure of the nucleotide-free monomer by X-ray crystallography at 3.3Å resolution (Fig. 1e and Table 1). This showed well-resolved density only for the AAA domain (residues 173–467). It is comprised of an enzymatic, α/β nucleotide binding domain (NBD) that contains the Walker A and B motifs involved in coordinating ATP, and a helix bundle domain (HBD). The katanin AAA domain is distinguished from that of other AAA ATPases like ClpX, N-ethylmaleimide sensitive factor (NSF) and Heat-shock protein (Hsp) 104 by two helices (N-terminal α1 and C-terminal α12) that augment the NBD and are characteristic of severing enzymes²³ (Fig. 1d and Supplementary Fig. 2).

Table 1.

Data collection and refinement statistics

	Katanin AAA apo (5WC1)
Data collection
Space group	P65
Cell dimensions
a, b, c (Å)	99.4,99.4,76.6
α, β, γ (°)	90, 90, 120
Resolution (Å)	50.0–3.2 (3.26–3.20)
Rsym (%)	9.8 (56.7)
I/σ(I)	36.7(3.4)
Completeness (%)	98.7(100.00)
Redundancy	9.8 (10.2)
Refinement
Resolution (Å)	46.0 – 3.3
No. reflections	7090
Rwork / Rfree	30.7/25.8
No. atoms
Protein	1793
Ion (SO4)	10
Water	0
B factors
Protein	107.6
Ligand/ion	177.2
Water
R.m.s. deviations
Bond lengths (Å)	0.001
Bond angles (°)	0.359

Data collected from one crystal

^aValues in parentheses are for highest-resolution shell.

We also obtained structures of the assembled full-length katanin hexamer by single particle cryo-EM (Table 2, Fig. 1f, Supplementary Fig. 3 and Methods). The data were refined and classified without imposing symmetry, yielding two reconstructions of distinct conformational states at 4.4Å and 6.0Å resolution (FSC=0.143 criterion), respectively (Supplementary Figs. 3, 4 and Methods). Well-resolved secondary structure elements in both reconstructions allowed derivation of atomic models by flexibly fitting the X-ray structure of the AAA monomer followed by real-space refinement (Table 2, Figs. 2, 3, and Methods). The 4.4Å reconstruction represents the dominant class (49% of particles sorted by 3D classification; Supplementary Fig. 4). It reveals a hexamer with a right-handed spiral arrangement of subunits with a ~ 5Å rise and ~60° twist per protomer (Fig. 2) that results in a ~30Å offset between the first and last protomers (hereafter P1 and P6), and a ~ 40Å wide gate between them (Fig. 1f, Fig. 2). Notably, the microtubule severing enzyme spastin and other AAA ATPases were proposed to form functional helical assemblies during their mechanochemical cycles based on their propensity to form helical arrangements in crystals²⁵. Both katanin and spastin form helices in crystals (Supplementary Fig. 5) that are left-handed and much more open (~13–16 Å rise and ~60° twist per crystallographic copy) than the right-handed spiral arrangement we see in our cryo-EM reconstruction thus resulting in a larger opening between the boundary subunits (~75Å), consistent with the more limited crystal packing interactions between the AAA domains. The second conformation identified by our cryo-EM analysis (24% of particles sorted by 3D classification) reveals a planar ring arrangement of the protomers that results in the closure of the P1–P6 gate (Fig. 3 and Supplementary Fig. 4).

Table 2.

Cryo-EM data collection, refinement and validation statistics

	#1 Spiral conformation (EMD-8794, PDB 5WC0)	#2 Ring Conformation (EMD-8796, 5WCB)
Data collection
Microscope	FEI Titan Krios	FEI Titan Krios
Camera	Gatan K2 Summit	Gatan K2 Summit
Magnification	22,500	22,500
Voltage (kV)	300	300
Electron dose (e−/Ǻ2)	51	51
Defocus range (μm)	−1.0 to −3.5	−1.0 to −3.5
Pixel size (Ǻ)	1.31	1.31
Reconstruction and Model Refinement
Particles	38,072	16,185
Map sharpening B factor (Ǻ2)	−98.36	−95.45
Resolution (Å)	4.4	6.0
FSCaverage	0.884	0.90
Model composition
Non-hydrogen atoms	11477	11422
Protein residues	1660	1662
Ligands (ATP)	6	5
R.m.s. deviations
Bond lengths (Å)	0.003	0.004
Bond angles (°)	0.74	0.83
Validation
MolProbity score	1.77	2.18
Clashscore	3.91	6.76
Poor rotamers (%)	0.00	0.00
Ramachandran plot
Favored (%)	92.57	87.86
Allowed (%)	7.37	12.14
Disallowed (%)	0.06	0.00

Figure 2. Cryo-EM maps and three-dimensional models for the katanin hexamer in the spiral conformation.

(a) Views of the katanin spiral conformation (rotation angles between the different views indicated with arrows). Protomer P1, green; P2, cyan; P3, blue; P4, orange; P5, purple; P6, red. NBD, HBD, light and dark hue, respectively. Black arrow indicates the open gate between protomers P1 and P6. (b) Top view of the spiral conformation with fitted atomic model. Cryo-EM map, shown as a transparent gray isosurface. The resolution of the map precluded de novo building of the fishhook linker element. Protomers colored as in (a). (c) Views of the atomic models for the katanin spiral conformation in the orientations shown above in panel (a).

Figure 3. Cryo-EM maps and three-dimensional models for the katanin hexamer in the ring conformation.

(a) Views of the katanin ring conformation (rotation angles between the different views indicated with arrows). Protomer P1, green; P2, cyan; P3, blue; P4, orange; P5, purple; P6, red. NBD, HBD, light and dark hue, respectively. Black arrow indicates the newly formed P1–P6 interface. (b) Top view of the ring conformation with fitted atomic model. Cryo-EM map, shown as a transparent gray isosurface. Protomers colored as in (a). (c) Views of the atomic models for the katanin ring conformation in the orientations shown above in panel (a).

The cryo-EM reconstructions lack density for the MIT domains and the N-terminal half of the linker (Fig. 1d), suggesting that these parts of katanin are flexible and thus averaged during image processing, consistent with our SAXS analyses that show a large distribution of angles for the arms protruding from the AAA ring (Fig. 1 and Supplementary Fig. 1). The dimensions of the AAA ring in the cryo-EM reconstructions are consistent with those obtained from our low-resolution SAXS envelopes (Fig. 1). The C-terminal ends of the linkers that immediately precede the α1 helices (shown in dark-blue in Figs. 1d, f) form fishhook-shaped structures (shown in light-blue in Figs. 1d, f). The fishhooks and the α1 helices crown the N-terminal entry to a 20 Å-wide axial pore that opens up into the 40Å wide gate between protomers P1 and P6 formed due to their spiral offset (Fig. 1f). Well-defined density reveals a spiral arrangement of the pore loops 1 and 2 which are essential for katanin and spastin severing activity^26,23. These loops are poorly ordered in the monomer katanin X-ray structure (Fig. 1e) indicating that they are stabilized upon hexamerization. Pore loop 1 contains a conserved aromatic residue characteristic to all AAA ATPase polypeptide unfoldases that is critical for direct interaction and translocation of the substrate, while pore loop 2 consists of positively charged residues that are highly conserved among severing enzymes (Supplementary Fig. 2 and ref.²³).

The NBD and HBD of the AAA core of each protomer form two lobes of a crescent. Protomers P2 through P5 make canonical convex to concave AAA interactions between successive protomers^27–29 (Fig. 2a). For a given protomer, the convex face of its NBD interacts with both the NBD and HBD of one neighboring protomer, while its concave face interacts with the NBD of the other neighbor. In the spiral arrangement, the gate is opened by the loss of convex and concave interactions for P1 and P6, respectively (Fig. 2). In the ring conformation, protomers P2 through P5 retain their helical arrangement (~ 5Å rise and ~60° twist per protomer as for P1 through P6 in the spiral conformation) with the canonical AAA ring interactions (Fig. 3). However, because of a 44° rotation, the P1 NBD makes near-canonical interactions with the HBD in P6 and non-canonical interactions on both its convex and concave interfaces with the NBD of P6 and P2, respectively, thus closing the P1–P6 gate (Fig. 3).

Katanin hexamer assembly is mediated by structural elements unique to microtubule severing enzymes

Katanin augments this mode of inter-protomer association with interactions mediated by structural elements characteristic to microtubule severing enzymes: the fishhook-shaped linker element, helix α1 and helix α12 (Fig. 1d, f, Supplementary Fig. 2). Helix α1 (shown in dark-blue in Fig. 1f) of the concave face of the NBD of one protomer makes contacts with the linker that crowns the NBD of the neighboring protomer (Fig. 1f). Additional contacts are mediated by the α1-α2 and α5-β4 loops that are positioned to sense conformational changes in the nucleotide-binding pocket and become ordered upon ATP-dependent hexamer formation (Supplementary Fig. 6a). Notably, mutation of conserved Ile 195 to lysine in the α1-α2 loop inactivates katanin in C. elegans^30,31. α12 helices form part of a belt around the C-terminal side of the axial channel (shown in orange in Fig. 1f). This arrangement is mediated by a tight interaction between the conserved α11-α12 linker of one protomer and the four C-terminal residues of the adjacent protomer that immediately follow helix α12 (Supplementary Fig. 6b). These four terminal residues are highly conserved among microtubule severing enzymes and are disordered in the monomeric crystal structures of both katanin (Fig. 1e) and spastin^23,25, consistent with their role in hexamer assembly. They include invariant Phe469 and Gly470 (Supplementary Figs. 2 and 6b). Mutation of Gly470 to Asp impairs katanin activity in C. elegans^30,31, while mutation of Phe469 impairs both ATPase and microtubule severing activity of katanin in vitro (Supplementary Table 1 and Methods). Mutation of the equivalent aromatic residue in spastin also has a drastic effect on microtubule severing activity²³.

Nucleotide state of a gating subunit mediates transition between katanin spiral and ring conformations

In the spiral conformation, all nucleotide-binding sites are occupied by ATP (Fig. 4), with clear density visible for all three phosphates as indicated by difference map analysis between the cryo-EM map and the nucleotide-free atomic model (Figs. 4c–e, Supplementary Fig. 7a and c). ATP binds at the NBD-HBD junction (Fig, 4a). Accordingly, the angle between the NBD and HBD is similar in all six protomers (Cα rmsd ~ 0.3Å; Fig. 4a). AAA ATPases contain arginine residues at the inter-protomer interface, termed “Arg fingers” that stimulate hydrolysis of ATP in trans around their rings³². In katanin, invariant Arg351 and Arg352 from protomers P2 through P6 are ~ 4Å away from the ATP β- and γ-phosphates of the neighboring protomer, yielding a catalytically competent arrangement of the nucleotide binding pockets in protomers P1 through P5 (Figs. 5a, b, Supplementary Fig. 6c). The exception is protomer P6 where the ATP is ~ 36Å from the arginine fingers in P1 (Fig. 5c). Mutation of both Arg351 and Arg352 drastically reduces katanin ATPase and inactivates microtubule severing (Supplementary Table 1).

Figure 4. Different nucleotide occupancies in the spiral and ring conformations of the katanin hexamer.

(a) Superposition of protomers P1 through P6 in the spiral conformation. (Cα overall r.m.s.d. ~ 0.3Å). Protomers colored as in Fig. 2a. (b) P1 structure in the spiral conformation. Bound ATP shown as stick representation. (c-e) Enlarged views of the nucleotide-binding pocket showing bound ATP for protomers P1, P3 and P6 in the spiral conformation with the difference map as transparent gray isosurface (Supplementary Fig. 7a, c and Methods). (f) Superposition of P1, P3 and P6 in the ring conformation showing the 44° rotation of the NBD in the P1 protomer. (g) P1 structure in the ring conformation. Oval highlights ATP-binding region. (h, i) Enlarged views of the nucleotide-binding pocket at two contour levels, 15 σ (h) and 10 σ (i) showing absence of nucleotide in P1. (j) Enlarged view of the nucleotide-binding pocket showing bound ATP for protomer P3 in the ring conformation with the difference map as transparent gray isosurface (Supplementary Fig. 7b, d).

Figure 5. Protomer-protomer interface rearrangements between the spiral and ring conformations: transition between tense and relaxed states of the P1 gating protomer.

(a, b) Canonical P3–P4 (a) and P1–P2 (b) interfaces in the spiral conformation highlighting the potential salt bridge between invariant Arg244 and Asp322 (Cα shown as spheres). (c) Lack of contacts between the P6 and P1 protomers in the spiral conformation. (d) Canonical P3–P4 interface in the ring conformation highlighting the potential salt bridge between invariant Arg244 and Asp322 (Cα shown as spheres) (e) The relaxed non-canonical P1–P2 interface in the ring conformation. (f) The P6–P1 interface in the ring conformation. Walker B region highlighted by a dashed-line box in all panels and arginine fingers are represented as gray ball and stick (a, b) or Cα spheres (c-f).

In the ring conformation, no nucleotide density is visible in the P1 protomer (Figs. 4h, i and Supplementary Figs. 7b). Protomers P2 through P6 have bound ATP (Fig. 4j and Supplementary Fig. 7b and d). Consistent with the absence of ATP at the P1–P2 interface, the Arg fingers that would be supplied in trans by P2 are ~ 12Å from the nucleotide-binding pocket (Figs. 5b, and e). The loss of ATP in P1 widens the angle between the NBD and HBD by ~18 ° (Fig. 4f). As a result, the P1–P2 interface relaxes. It loses NBD-NBD contacts and the interface is formed only by interactions between the P1 HBD and P2 NBD (Fig. 3a, b). As a result, the salt bridge between invariant Arg244 in P1 and invariant Asp322 in P2, present in the spiral conformation (Fig. 5b) is lost (Fig. 5e). Consistent with its importance, mutation of Asp322 reduces ATPase by 96% and inactivates katanin in in vitro microtubule severing assays (Supplementary Table 1). The rotation of the P1 NBD brings it into contact with P6, thereby closing the gate (Figs. 5f, 3 and Supplementary Video 1). In the canonical interface, α5 is involved in extensive inter-protomer interactions, notably with the Walker B motif that senses nucleotide identity while α1 makes contact with the fishhook-shaped linker element of the neighboring protomer (Figs. 5a, b, d and Fig. 1f). The loss of nucleotide in P1 displaces α5 so that it packs against α1 in P6 and makes more limited inter-protomer contacts (Fig. 5f). Notably, mutations of residues in helix α5 of spastin are found in hereditary spastic paraplegia cases and impair microtubule-severing activity (Supplementary Fig. 2 and refs.^23,33).

DISCUSSION

AAA proteins actively thread their substrate biopolymers through an axial channel with the help of conserved pore loops, leading to remodeling or disassembly of their substrates³⁴. Katanin and spastin are thought to sever the microtubule by engaging the C-terminal tails of tubulin with their essential pore loops and pulling on them to dislodge tubulin subunits out of the microtubule lattice^23,35. The lack of structural information on microtubule severing enzymes has however impeded the advancement of a mechanistic framework for the mechanochemical cycle for this class of enzymes. The spiral and ring conformations of the katanin AAA motor that we uncovered in our study suggest a power stroke that links ATP hydrolysis and nucleotide release to microtubule severing. The katanin hexamer binds the tubulin tails essential for severing^18,35–38. In the spiral ATP- bound conformation, the pore loops stack sequentially to form a spiral ladder optimal for substrate translocation, with pore loop 1 of protomer P1 positioned at the bottom of the spiral close to the microtubule-binding surface, and the open gate between P1 and P6 facilitating tubulin C-terminal tail access to the axial pore. Moreover, the NBD of P1 is exposed in this conformation and accessible to the tubulin tail, which may thus directly stimulate its ATPase upon binding. Our structure reveals that the novel linker element, pore loops 1 and 2, and α12 form a contiguous interface through the axial channel that is optimal for substrate binding and translocation (Fig. 1f and Supplementary Video 1). The location of the fishhook linker element indicates that it is likely one of the first structural elements to interact with the microtubule surface. This structural element is absent in the vps4 hexamer of the meiotic clade of AAA ATPases^39,40 and thus is a defining feature of katanin. Interestingly, linker residue Ser135, immediately proximal to the fishhook element, is phosphorylated by Aurora A kinase. This phosphorylation impairs katanin microtubule stimulated ATPase and severing, thus positively regulating spindle size in Xenopus laevis^9,41. The additional negative charges proximal to the channel entrance likely interfere with the productive engagement of the negatively charged microtubule substrate. In spastin, mutations of residues in this region or deletion of this linker element impair microtubule severing^23,35.

We propose that ATP hydrolysis and nucleotide release in P1 drives the transition to the ring state, thus displacing the linker and pore loop in P1 by ~ 20Å together with the bound tubulin substrate (Figs. 6a, b and Supplementary Video 1) from the bottom of the spiral and up and into the axial channel of the hexamer. The tubulin chain can be then further pulled into the axial channel by the engagement of loop 1 in protomer P2 for another cycle through a sequential mechanism of ATP hydrolysis in the AAA ring. This ~ 20Å displacement would correspond to a ~ 5 amino acid length translocation of the C-terminal tubulin tail from the microtubule surface towards the center of the axial pore.

Figure 6. Pore loop displacement of the gating protomer suggests power stroke for microtubule severing.

(a) Superposition of the katanin spiral and ring conformations shows the large movement of the gating protomer P1 (green arrow) and the smaller-scale movement of P6 (red arrow). Protomers are colored as in Fig. 2 in the spiral conformation and are gray in the ring conformation. (b) Cartoon depicting the movement of protomer P1 between the spiral and ring conformations. Protomers colored as in Fig. 2. Arrow highlights the translocation of loop 1 of the P1 gating protomer. (c) Cartoon illustrating proposed power stroke that extracts a tubulin dimer and initiates microtubule lattice breakdown and severing. Left panel, Katanin (blue) assembles as a hexamer with a spiral configuration of the AAA domains and the MIT domains emanating from the AAA motor core and making multivalent interactions with the microtubule (green). The flexible tubulin tail is engaged in the axial pore of the katanin hexamer. Right panel, ATP hydrolysis and release in the gating protomer P1 leads to closure of the AAA ring and a ~20Å displacement in the P1 loop that translocates with it the bound C-terminal tail of a tubulin subunit. The cycle is repeated until lattice contacts unravel and the microtubule severs.

The MIT domains bind the microtubule autonomously with low affinity⁴². Our SAXS analyses demonstrate they are located at the tip of flexible tethers that connect them to the AAA core. This flexibility is likely important to the ability of the MIT domains and the AAA ring to simultaneously engage the microtubule substrate through multivalent interactions and thus help the enzyme stay anchored to the microtubule as it remodels its AAA ring during the mechanochemical cycle and pulls on the tubulin tails with its pore loops (Fig. 6 c). The avidity-based interaction with the microtubule through the MIT domains could also help retain the assembled enzyme on the polymer for subsequent engagement of neighboring tubulin tails if the extraction of one tubulin subunit is not enough to initiate catastrophic breakdown of the microtubule lattice. Complete unfolding and axial translocation of the entire polypeptide chain as in the case of ClpX or Vps4^43,44 is not absolutely necessary as the microtubule could be destabilized by repeated “tugging” on the tubulin to apply a distorting force to the protofilament structure by cycling through the spiral and closed ring states in a mechanism more analogous to the one recently proposed for NSF⁴⁵.

In addition to our results here on microtubule severing, structural studies of AAA ATPases that remodel protein complexes or aggregates such as Hsp104⁴⁶, ClpX²⁹ the archaeal homolog of p97 of unknown function, VAT⁴⁷, the proteasome^48,49 and Vps4^39,40 have identified asymmetric spiral architectures of AAA rings with pore loops stacked in a spiral suggestive of a relay of the substrate through their central pores. Recent cryo-EM studies identified both open and closed spiral configurations for VAT and Vps4 in the presence of substrate and as a function of nucleotide state^40,47 suggesting a mechanism analogous to katanin, although there are differences in nucleotide occupancies between these structures, consistent with their different domain organization and function. The substrate-bound structures of Vps4⁴⁰ and VAT⁴⁷ are asymmetric with five of the pore loops bound to the unfolded substrate polypeptide and arranged in a tight spiral and the sixth subunit in a different conformation and disengaged from the substrate. Thus, a common theme is emerging for some AAA ATPases whereby the power stroke for substrate translocation is generated by breaking the helical symmetry of a spiral arrangement of subunits by a boundary subunit in response to nucleotide hydrolysis. Such a mechanism for substrate translocation was originally advanced for nucleic acid AAA ATPases translocases such as the DNA helicase E1⁵⁰ and the RNA translocase Rho^51,52. In contrast to these, the recent cryo-EM reconstructions of p97 show six-fold symmetric arrangements indicative of a concerted mechanism of substrate engagement and remodeling⁵³, and highlight the mechanistic richness used by AAA ATPases, consistent with their large representation in the genome and diverse cellular substrates and functions. Thus, details of nucleotide occupancies and nucleotide hydrolysis mechanism are likely different between AAA ATPases and might even differ between substrates for the same enzyme as studies on ClpB and Hsp104 have shown different ATP hydrolysis mechanisms depending on the type of substrate they engage^54,55. This is an interesting aspect to consider when comparing microtubule severing enzymes and other closely related AAA ATPases, as the former are able to disassemble one of the stiffest non-covalent polymer structures found in eukaryotic cells².

The mechanistic complexity of AAA ATPases highlights the importance of obtaining high-resolution structural and kinetic information for each enzyme along its entire mechanochemical cycle and with its physiological substrates. Future structures of katanin and spastin in different nucleotide-bound states and with the microtubule substrate will be critical for elucidating how these enzymes generate the force needed to disrupt lattice interfaces in the microtubule polymer that ultimately lead to its catastrophic disassembly. Our structural study is an important first step towards establishing a mechanistic framework for these fascinating molecular machines.

ONLINE METHODS

Protein expression and purification

Caenorhabditis elegans katanin p60 was expressed in Escherichia coli as a maltose binding-fusion protein. Cultures were grown at 37°C to an OD₆₀₀ of ~ 1.0 and katanin expression was induced with 0.5 mM IPTG at 16°C and for fourteen hours. Harvested cells were resuspended in 50 mM HEPES pH 7.5, 500 mM KCl, 10 mM MgCl₂, 1 mM DTT, 1 mM PMSF, and lysed with a microfluidizer in the presence of a protease inhibitor cocktail (Roche). The supernatant was collected by centrifugation at 31,000 x g for 40 min and loaded onto amylose resin (New England Biolabs), equilibrated in 50 mM HEPES pH 7.5, 500 mM KCl, 10 mM MgCl₂ and 1 mM DTT. Katanin was released from the resin by cleavage with Tobacco Etch Virus protease. Katanin was further purified by anion exchange chromatography. The sample was exchanged into buffer containing 20 mM HEPES pH 7.5, 300 mM KCl, 10 mM MgCl₂, 1 mM TCEP and 15 % glycerol and flash-frozen in liquid nitrogen for storage at −80°C. Prior to conducting all analytical ultracentrifugation (AUC), solution small-angle x-ray scattering (SAXS) and electron microscopy (EM) analyses katanin was further purified by size exclusion chromatography on a Superose 6 Increase 10/300 GL column (GE Healthcare). In order to determine the stoichiometry of ATP to the katanin hexamer we chromatographed full-length and ΔMIT katanin p60 (100 μL of 100 μM) on a Superose 6 10/300 GL Increase column in the buffer containing excess ATP (100 μM). As the protein migrates through the gel filtration matrix, it depletes ATP from the buffer resulting in a negative peak. Bound ATP was determined from the area of the negative ATP depletion peak. The determined molar ratio of ATP per katanin hexamer was 5.6±0.3 (n=3). Mutants were generated using Quickchange (Stratagene).

Analytical ultracentrifugation

AUC experiments were conducted at 24, 12, 6 and 3 μM of katanin p60 concentrations in 20 mM HEPES pH 7.5, 300 mM KCl, 10 mM MgCl₂, 1 mM TCEP with or without 100 μM ATP. Measurements were collected in a ProteomeLab XL-I analytical ultracentrifuge (Beckman Coulter, Indianapolis, IN) using absorption optics. To obtain a final sample optical density in the 0.25 to 1.00 range, cells of 3-mm or 12-mm optical path length were used. Loaded cells, placed in the four-hole AN-Ti rotor, were thermally equilibrated in the ultracentrifuge. After thermal equilibrium was reached at rest at 10°C or 20°C, the rotor was accelerated to 45,000 rpm to start the collection of intensity scans at 280 nm. The data were collected until no further sedimentation boundary movement was observed. Sedimentation velocity data analyses were performed with the program SEDFIT using a continuous c(s) model⁵⁶. All accepted fits have root-mean-square deviations (RMSD) less than 0.008. Partial specific volume and buffer parameters were calculated with SEDNTERP (http://sednterp.unh.edu/). Sedimentation coefficient distribution values were corrected to standard conditions of water at 20°C, s_20,w.

Crystallization and X-ray structure determination

For crystallization, katanin p60 (E293Q) eluted from the size execution column (see above) was supplemented with 1 mM ATP and concentrated to 10 mg/ml. Crystals grew by hanging drop vapor diffusion at room temperature from protein mixed 2:1 (v/v) with the crystallization solution containing 1.0 M ammonium sulfate and 0.1 M Tris pH 7.8. Crystals were cryoprotected with 25% glycerol before flash freezing in liquid nitrogen. X-ray diffraction data were collected at the Advanced Light Source (ALS), beamline 5.0.2. Crystals grew with the symmetry of space group P6₅ with one katanin protomer per asymmetric unit (Unit Cell: a = 99.4 Å, b = 99.4 Å, c = 76.6 Å). Most crystals diffracted poorly and were not suitable for structural determination. Diffraction data were reduced with HKL2000⁵⁷. Details of the experimental data collection and processing statistics are in Table 1. The structure was solved by molecular replacement using PHASER⁵⁸ with the previously published structure of spastin as the search model (PDB entry 3B9P²³). Multiple rounds of model building and refinement were performed with COOT⁵⁹ and PHENIX⁶⁰. The current refinement model for katanin p60 consists of 262 residues and one sulfate ion. Despite the presence of ATP in the crystallization conditions, no well-ordered nucleotide is visible in the maps. Six regions of the polypeptide chain (residues 1–172; 186–190, 263–269, 297–306, 324–332, 467–472) are not well-resolved and presumed disordered. Residues 173–183, 270–284 and 310–322 were built as poly-Ala as the sequence register is based on the sequence alignment of the spastin X-ray structure. The fishhook element visible in the cryo-EM map is also visible in the X-ray model, although less well defined in the latter because of the lack of stabilization by protomer-protomer contacts. The quality of the map in this region was not sufficient to allow confident de novo building in either the X-ray or cryo-EM map. This structural feature comprises ~ 25 residues N-terminal to helix α1. The current crystallographic model of the katanin protomer at 3.3 Å resolution has R_work and R_free of 30.7 % and 25.8 %, respectively (Table 1) with 95.1% of residues in favored regions and 0.8% in disallowed regions. Coordinate error is 0.53Å. Figures were prepared with UCSF Chimera⁶¹.

Solution small-angle X-ray scattering

Solution SAXS experiments of full-length and ΔMIT katanin p60 (E293Q) were performed at beamline 12-ID-B of the Advanced Photon Source (APS) at Argonne National Laboratory using an in-line FPLC AKTA micro setup with a Superose 6 Increase 10/300 GL size exclusion column. The wavelength, λ, of X-ray radiation was set to 0.886 Å. Scattered X-ray intensities were measured using a Pilatus 2M detector. The sample-to-detector distance was set such that the detecting range of momentum transfer q [=4π sinθ/λ, where 2θ is the scattering angle] was 0.004–0.55 Å⁻¹. The sample passed through the FPLC column and was fed to a flow cell for SAXS measurements. The flow cell is a cylindrical quartz capillary 1.5 mm in diameter and 10 μm wall thickness, and is equipped with a Peltier cooling device that keeps the flow cell at 10°C. SAXS data were collected at the elution peak while background data were collected before and after the peak. The exposure time was set to <1 second to reduce radiation damage. More than forty images were acquired for each sample (around the elution peak) and background in order to obtain good signal-to-noise ratio values. The 2-D scattering images were converted to 1-D SAXS (I(q) vs q) curves through azimuthally averaging after solid angle correction and then normalizing with the intensity of the transmitted X-ray beam flux, using custom scripts. The sample concentration was ~2.0 mg/ml – 4.0 mg/ml at the elution peak. The SAXS profile of the protein was obtained by subtracting background from the sample data. The radius of gyration (R_g) was calculated using the Guinier equation⁶², i.e., $I\left(q\right)=Io\text{exp}\left(-R_g^2{q}^2/3\right)$, where I_o is the forward scattering, an indicator of molecular weight⁶³. The pair distance distribution function P(r) that is the inverse Fourier transform of X-ray scattering data and roughly a weighted histogram of atomic-pair distances in the molecule, was calculated using GNOM v4.6⁶⁴. R_g were calculated from the second momentum of the P(r) function and compared favorably to the initial value derived from the Guinier plots (fig. S1E), 67.7±0.4 Å versus 63 ± 5 Å for full-length katanin p60 and 51.4 ± 0.3 Å versus 51 ± 1 Å for ΔMIT katanin p60. Molecular weights were estimated directly from the background-subtracted SAXS data using the program SCATTER v2.3h⁶⁵ and were 360 kDa and 260 kDa for katanin p60 and ΔMIT katanin p60, respectively. The expected molecular weights of the katanin p60 and ΔMIT katanin hexamers are 319 kDa and 252 kDa, respectively. Three-dimensional molecular envelopes were calculated from SAXS data up to q of 0.30 Å⁻¹, using DAMMIF v1.1.2, a fast version of DAMMIN⁶⁶. Twenty and twenty-four runs were performed for katanin p60 full-length and the ΔMIT construct, respectively. The reconstructions were aligned, averaged and filtered on the basis of occupancy using DAMAVER v5.0⁶⁷. P6 symmetry was applied in the DAMMIF and DAMAVER calculations. The overall size of the DAMAVER averaged bead model was significantly reduced comparing to the individuals, indicating that the regions beyond the central ring have a high degree of flexibility (Fig. 1a and Supplementary Figs. 1f, d, h). Therefore, both the individual and averaged models are reported, with the former being a better representation of molecular size.

Cryo-EM specimen preparation, data acquisition and analysis

Katanin p60 (E293Q) was diluted to 0.6 mg/ml in 20 mM HEPES-KOH pH 7.5, 300 mM KCl, 10 mM MgCl₂, 1mM TCEP and 1mM ATP and spun at 69,800 x g for 15 min at 4°C. 5 μl was applied to a glow-discharged C-flat holey carbon grid with 1.2 μm hole, 1.3 μm space (C-flat™, Electron Microscopy Sciences). The grids were blotted for 5 seconds at 90% humidity and plunge-frozen in liquid ethane cooled by liquid nitrogen using Leica EM GP (Leica Microsystems, Germany). To assess sample stability and monodispersity, cryo-EM data were first collected on a Tecnai TF20 electron microscope (FEI) operated at 200 kV, equipped with a K2 Summit direct electron detector camera (Gatan) operated in electron counting mode. To obtain a preliminary reconstruction, ~800 images were recorded at a nominal magnification of 29,000x, corresponding to a physical pixel size of 1.27 Å/pixel. Images were recorded with a defocus range from −1.5 μm to −3.5 μm. With no specimen present, the rate of exposure of the detector was 6 e⁻/Ų/s. Exposure-fractionated movie stacks of 40 frames were recorded for 10 s with a total electron dose of 60 e⁻/Ų.

To obtain a preliminary reconstruction of katanin p60 (E293Q), images were aligned with each other using Unblur⁶⁸. Defocus parameters were estimated from aligned sums of 5 frames using CTFFINF4⁶⁹. 90,000 particles were manually picked from movie sums with the total dose of 60 e/Ų and extracted in Relion using a 300 pixel box size⁷⁰. Images were downsampled to a pixel size of 2.54 Å and classified in 2D using Relion. An initial 3D reconstruction with C6 symmetry was generated using a subset of particles (40,000) in FREALIGN⁷¹, following a protocol described previously⁷². To obtain the initial reconstruction, images were downsalmpled to 5.08 Å/pixel and frequencies only up to 1/40 Å⁻¹ in the initial rounds and up to 1/15 Å⁻¹ in the final rounds of refinement were used. The most populated class was low-pass filtered to 60 Å and used as an initial reference for 3D classification of the full data set. For 2D and 3D classification in Relion, particles were re-extracted from the movie sums with the total dose of 35 e⁻/Ų. 3D classification was carried out without application of symmetry operators. Only spatial frequencies up to 1/10 Å⁻¹ were used to avoid overfitting. After 50 rounds of 3D classification with 4 classes, particles in 2 classes that produced similar 3D reconstructions were combined for 3D auto-refinement. The 3D reconstruction from ~27,000 particles was resolved to 14.1 Å (FSC=0.143 criterion).

To obtain the final high-resolution 3D reconstructions, a dataset of ~2,100 images was collected on a Titan Krios microscope (FEI) operated at 300 kV, equipped with a K2 Summit direct electron detector camera, operated in electron counting mode. Micrographs indicated monodisperse distribution of particles in ice (Supplementary Fig. 3a). Movie stacks were recorded at a nominal magnification of 22,500x, corresponding to a physical pixel size of 1.31 Å/pixel at a defocus range from −1.0 μm to −3.5 μm. The rate of exposure was 2.91 e⁻/Ų/s. 50-frame movie stacks were recorded for 17.5 s with an exposure of 1.02 e⁻/Ų per frame and a total electron dose of 51 e⁻/Ų. Data collection was automated with SerialEM⁷³. Data collection statistics are in Table 2.

All image frames (51 e⁻/Ų total dose) were aligned and summed with Unblur through the Relion pipeline^68,70. Defocus parameters were estimated from aligned sums of 5 frames using CTFFINF4⁶⁹. Images were downsampled to a pixel size of 2.62 Å/pix and ~403,023 manually picked particles were classified in 2D using Relion. Reference-free 2D class averages showed fine molecular features and presented the complex in different orientations, suggestive of its structural order and the lack of preferential orientation in ice (Supplementary Fig. 3b). After 3 rounds of the 2D classification ~197,779 particles were 3D classified into 4 classes without application of symmetry operators (Supplementary Fig. 4). The reconstruction from the TF20 dataset was low-pass filtered to 60 Å and used as the initial reference for the 3D classification. 100 rounds of 3D classification with no applied symmetry produced two similar classes with a spiral conformation (~95,468 particles total, Classes 2 and 4 in Supplementary Fig. 4), one class with a ring conformation (~48,565 particles, Class 3 in Supplementary Fig. 4) with the rest of the particles classifying into a low-resolution class (~53,746 particles, Class 1 in Supplementary Fig. 4). The low-resolution class consists of broken or misaligned particles and looked structurally different from the three major classes. Further classification of particles in this class did not identify any dominant conformation and did not improve the overall quality of the map. The particles from the two classes with the spiral conformation were combined and reclassified into 3 classes. All three classes looked alike and contained a similar number of particles, indicating the structural homogeneity of this particle set. Consequently, the particle set was refined in 3D as a single class. Similarly, the particle subset that produced the reconstruction with the ring conformation was further classified into 2 classes. Both new classes looked similar and contained a similar number of particles indicating the structural homogeneity of this particle set. Consequently, the original particle subset was refined in 3D as a single class.

For the 3D refinement particles were re-extracted from unbinned image sums that showed tone rings to 5 Å or higher spatial frequencies. The 3D refinement was carried out using a created mask with a soft edge of 6 pixels to mask out flexible protein parts around the AAA ATPase hexameric core. Prior to application of the mask, no other well-defined features were visible outside the hexameric core. The 3D reconstruction of the spiral conformation was refined to 5.2 Å and the ring conformation to 7.3 Å (FSC=0.143 criterion). To improve the quality of the 3D maps further, particles were sorted based on their differences with their aligned, CTF-corrected references as described in refs.^46,74 A subset of particles with z-scores ≤ 0.7 and ≤ 0.8 for the spiral and the ring conformations, respectively, were re-refined in 3D. This procedure improved the quality of the final 3D maps with the resolution of the spiral reconstruction calculated to 4.4 Å and that of the planar reconstruction to 6.0 Å after implementation of the PostProcessing procedure with automatic masking and negative Bfactor application. Both reconstructions were sharpened with the automatically determined negative B-factors of ~100 Ų for both structures. 38,072 and 16,185 particles contributed to the final maps in the spiral and ring conformations, respectively. Data collection statistics and image-processing summary are in Table 2.

To generate the difference maps in Fig. 4 and Supplementary Fig. 7, the atomic models for the protein were converted to mrc format using the bgex routine in Bsoft⁷⁵ and low-pass filtered using the proc3d routine in EMAN⁷⁶. The difference maps for the spiral conformation between the cryo-EM reconstruction and the atomic model, low-pass filtered to either 5.5 Å for P1 and P6 or to 4.4 Å for P2 through P5, were generated using diffmap.exe (http://grigoriefflab.janelia.org/software). The resolution cut-off values for each protomer were determined by the blocres program in Bsoft⁷⁵. Atomic models for the ring conformation were low-pass filtered to 8 Å for P1 and P6 and to 6 Å for P2 through P5. The difference maps were generated following the procedure described for the spiral conformation.

Model building and refinement.

Consistent with the estimated 4.4 Å resolution, we observe grooves in α-helixes, separated β-strands within β-sheets and densities for some bulky side chains. The X-ray structure of the katanin protomer was docked into the sharpened map using UCSF Chimera⁶¹ and its fit was improved with the collage option in Situs⁷⁷. The atomic model was fit flexibly into the cryo-EM reconstructions using NAMD with Molecular Dynamics Flexible Fitting (MDFF)⁷⁸. Secondary structure, chirality and cis-peptide restraints were applied to the model, and MDFF simulations were performed in implicit solvent at temperature T=300 K to enhance the protein conformational sampling. Energy minimization was performed for 200 steps. MDFF simulations were carried out for 150 ps with a coupling constant (gscale) = 0.3. MDFF simulation convergence was monitored by plotting the root mean square deviation over time. Additional adjustments to the backbone and side chains for this model were performed manually in COOT⁵⁹, residue by residue. Densities for ATP were clearly visible in all six protomers of the spiral conformation (Fig. 4c–e and Supplementary Fig. 7a and c). The α11 – α12 linker, the C-terminus and pore loop1 were built de novo as they are disordered in the monomer X-ray structure. Weak residual density features are present near pore loop1 that were not interpreted at the current resolution limit. None of the side chains of the pore loop residues were built. The resulting model was subjected to real space refinement in PHENIX⁶⁰. The final atomic model has an overall correlation to the map of 0.884. For the ring conformation, the X-ray structure was first rigidly docked using Situs and then flexibly fitted using MDFF and COOT into the cryo-EM map as described for the spiral conformation. Additional adjustments to the backbone for this model were performed manually in COOT. Densities for ATP were clearly visible in five of the six AAA protomers (Fig. 4j and Supplementary Fig. 7b and d). The model was subjected to real space refinement in PHENIX. The final atomic model has an overall correlation to the map of 0.90. MolProbity was used to evaluate model geometry⁷⁹. Model statistics for both conformations are listed in Table 2.

ATPase and microtubule severing assays

ATPase assays were performed at room temperature in BRB80, 50 mM KCl, 2.5 mM MgCl₂, 1 mM ATP, 10 μM taxol at 50 nM katanin concentration. ATPase activities were measured using an EnzChek Phosphate Assay (Life Technologies). Initial rates were calculated from the linear portion of the reaction profiles after addition of 1 mM ATP. ATPase rates were adjusted by subtraction of the measured release of phosphate in the absence of ATP. Rates reported are means from four separate reactions.

Chambers for TIRF microscopy were assembled as previously described⁸⁰. GMPCPP brain microtubules containing 1% biotinylated tubulin and 20% HiLyte647-labeled tubulin were immobilized in the chamber with 2 mg/ml Neutravidin (Sigma) and imaged by TIRF or DIC microscopy in severing buffer (BRB80 buffer with 2 mg/ml casein, 14.3 mM 2-mercaptoethanol, 2.5% glycerol, 50 mM KCl, 2.5 mM MgCl₂, 1 mM ATP, 1% Pluronic F127 (Life Technologies) and oxygen scavengers prepared as described in⁸⁰). The chamber was then perfused with 20 nM katanin in severing buffer. Severing rates were calculated by manual counting of severing events as a function of time as described previously³⁷.

A Life Sciences Reporting Summary accompanies this paper. All data and constructs used in this study are available upon request from the corresponding author. Protein Data Bank accession number for atomic coordinates and structure factor amplitudes for the X-ray structure is 5WC1. The EMDB accession codes for the spiral and ring maps are EMD-8794 and EMD-8796, respectively. The atomic coordinates for the cryo-EM models were deposited with PDBIDs 5WC0 and 5WCB for the spiral and ring conformation, respectively.