We solved the X-ray structure of the nucleotide-free, monomeric AAA domain of Drosophila spastin (residues 464–758) at 2.7Å resolution (R_free = 28.7%; Supplementary Information). Similar to other AAA proteins, the enzymatic core of spastin contains a central α/β nucleotide-binding domain (NBD) and a smaller four-helix bundle domain (HBD). A marked feature of the spastin structure is its open nucleotide pocket, which explains the absence of a bound nucleotide, despite the presence of 0.5 mM adenosine 5’-O-(3-thiotriphosphate) (ATPγS) in the crystallization solution. Comparison of our nucleotide-free spastin structure with the ATP-bound structure of N-ethylmaleimide-sensitive fusion protein (NSF) (an AAA protein involved in membrane fusion¹⁵) reveals that an extended loop involved in nucleotide contact and protomer–protomer interactions in NSF (Supplementary Fig. 2) is pulled away from the nucleotide pocket in spastin by the packing of the linchpin Trp 482 in a conserved hydrophobic pocket (Fig. 1g). The pocket for Trp 482 is sub-optimal, compatible with movement of the tryptophan and rearrangement of the flap on ATP-induced hexamerization or/and substrate binding.

Uniquely among known AAA structures, spastin has two helices (N-terminal α1 and C-terminal α11) that embrace the NBD (Fig. 1c and Supplementary Fig. 3). The amphipathic N-terminal helix is anchored to the body of the NBD by interdigitating hydrophobic residues (Leu 470/Ile 473/Val 474; Fig. 1d). Mutation of these residues to alanine reduced ATPase activity by ~90% and abolished microtubule-severing, while preserving microtubule binding (Fig. 1f and Supplementary Fig. 4). Mutation of invariant Leu 567 located at the helix α1–NBD interface causes hereditary spastic paraplegias (HSP)¹⁶. Solvent-exposed residues on the N-terminal helix also have important roles; L465F and the triple mutant L465A/D471A/E472A markedly decreased microtubule-severing (Fig. 1f) without significantly affecting the ATPase. The C-terminal helix, which is also present in the closely related enzyme VPS4 (vacuolar sorting protein 4)¹⁷, and part of the preceding conserved linker wrap around the phosphate-binding loop (P loop) of the NBD (Fig. 1e). Mutation of the highly conserved Tyr 753 at the end of the C-terminal helix to alanine effectively inactivated the enzyme (Fig. 1f), whereas a Y753F mutation still showed severing activity in vivo (Supplementary Fig. 4). Thus, our structural and mutational analyses indicate that helices α1 and α11 of spastin have important roles in allosteric control of the ATP-binding site and possibly in substrate binding (discussed below).

We next obtained structural information on a hexameric spastin construct using small-angle X-ray scattering (SAXS). A Caenorhabditis elegans MIT–AAA construct was used because it is monodisperse at the concentrations (> 5 mg ml⁻¹) required for collecting high-quality SAXS data. Compared to Drosophila spastin, C.elegans spastin lacks an N-terminal domain and has a shorter linker between the MIT and AAA domains; nonetheless, it displays microtubule-severing activity¹⁸ (Supplementary Fig. 5a). We first examined the oligomeric state of C. elegans spastin in its nucleotide-free (apo) and ATP-bound states by static multi-angle light scattering. To create a stable, noncycling ATP-bound state, we prepared a well-described AAA mutation that blocks nucleotide hydrolysis (E278Q; E583Q in Drosophila spastin). Static light scattering revealed that the apoenzyme exists in equilibrium between a monomeric and a weak dimeric state, whereas ATP-bound spastin is a hexamer, a quaternary structure adopted by many AAA proteins¹⁹ (Supplementary Fig. 5). Unlike many AAA ATPases, but similar to katanin²⁰, spastin exists mostly as a monomer at submicromolar concentrations, even in the presence of ATP (data not shown).

SAXS data were used to generate low-resolution ab initio²¹ models of three-dimensional arrangements of scattering centres that provide the shape of the molecular envelope of the hexamer (Methods; Fig. 2 and Supplementary Fig. 6). The models from seven independent ab initio simulations were aligned, averaged and filtered on the basis of occupancy to obtain a most probable model. The close agreement between the total volume enclosed by the superposition of the individual runs (the composite structure) and the most probable density map (the filtered structure) indicates the robustness of the ab initio reconstructions. The filtered structure shows a central ring with a double trapezoid cross-section (130Å × 65Å), a ~20-Å-diameter central pore, and slender arms radiating ~50Å outward and extending towards one face of the ring (Fig. 2). The clear reconstruction of the arms also indicates that the linker, although unlikely to be rigid, adopts some defined structure and is not completely disordered. Shortening the linker to <40 residues disables microtubule-severing (but not microtubule-binding, Supplementary Fig. 1), suggesting some length and/or sequence requirement for this region. The asymmetric position of the arms defines a polarity to the overall structure (two faces, herein termed face A and B). We generated an atomic model for the AAA hexameric core of spastin by superimposing our nucleotide-free spastin monomer X-ray structure onto the crystal structure of the NSF hexamer. This model was docked into the SAXS reconstruction with the N- and C-terminal helices on faces A and B, respectively (Figs 2b,c; for details about the fit, see Supplementary Fig. 6 and Supplementary Information).

Several AAA proteins (for example, the bacterial proteins ClpX, ClpA and ClpB) remodel their substrates by threading the end of the polypeptide chain through a central pore in their rings^19,22,23. The microtubule-severing activities of spastin and katanin depend on the ~20-residue disordered and negatively charged C-terminal tails of tubulin^3,6, suggesting an analogous mechanism for spastin and katanin. In support of this model, we found that a 23-mer peptide corresponding to the C-terminal tail of β-tubulin inhibited microtubule-severing by ~70% at 0.5 mM, whereas a randomized (scrambled) peptide of identical amino acid composition or an α-tubulin peptide that contains the C-terminal tyrosine (α-Tyr peptide) did not show detectable effects (Fig. 3a). The large concentration of peptide needed to observe inhibition is not surprising given the high local concentration of tubulin tails encountered by microtubule-bound spastin. Involvement of the β-tubulin tail is consistent with genetic data showing that a charge-reversal mutation in this region suppresses the lethality of ectopic katanin activity²⁴. We also found that an antibody that recognizes exposed glutamate residues on the C-terminal tails of tubulin (detyrosinated α-tubulin with a final C-terminal glutamate as well as β-tubulin and polyglutamylated tubulin) completely inhibited spastin-mediated severing. In contrast, a ‘Tyr’ antibody that recognizes β-tubulin with a C-terminal tyrosine²⁵ (~50% of brain tubulin^26,27) did not inhibit severing, even though the antibody binds to microtubules (Fig. 3b and Supplementary Fig. 7b). Although we did not detect a robust inhibitory effect of a detyrosinated α-tubulin peptide, an antibody that recognizes the tail of Glu– α-tubulin²⁷ partially inhibited severing (Supplementary Fig. 7c). Collectively, these in vitro data support a model in which spastin interacts with the acidic tubulin C-terminal peptide during the severing reaction and may recognize specific features of the C-terminal peptide.

To explore this model further, we examined the roles of three solvent-exposed loops within the pore that are highly conserved among spastins and katanins (Fig. 3c). Mutations in pore loop 1 of Drosophila spastin, which has been shown to be important for the substrate-remodelling activity of several other AAA proteins^22,23,28, abolished severing (Figs 3c, d) but preserved microtubule binding (Supplementary Table 2 and Supplementary Fig. 8). After submission of this work, similar results were obtained in ref. ¹⁴. Mutations of solvent-exposed residues in pore loops 2 and 3 also completely inhibited or severely crippled the enzyme (Figs 3c, d). However, the disease mutant S589Y retains some activity, suggesting neurons are susceptible to disease with partial spastin activity. Mutations of surface residues leading to the pore (Fig. 3d) also markedly affected the activity of spastin (for example, L465F and L465A/D471A/E472A in Fig. 1f, and K562A and K562R in Supplementary Fig. 4).

In conclusion, the combination of X-ray crystallography, SAXS ab initio reconstructions and structure-guided mutagenesis provides the first structural information on microtubule-severing proteins and allows us to propose a molecular model for spastin-mediated severing (Fig. 4a). Owing to their similar domain organization and high sequence similarity, this model probably pertains to katanin as well. We propose that face A of the spastin AAA ring docks onto the microtubule, placing the positively charged N-terminal pore entrance in contact with the negatively charged C terminus of tubulin. The translocation from face A to face B would correspond to the direction of substrate translocation proposed for the distantly related AAA ATPases ClpX, ClpA and ClpB^22,28,29. The linker and MIT domains extending from the ring would make additional contacts with the microtubule, thus increasing microtubule avidity and potentially stabilizing the hexamer on the microtubule²⁰. On the basis of our affinity measurements, only a subset of the six arms is likely to make strong binding interactions with the microtubule (Fig. 4a).

We propose that the tubulin polypeptide is threaded through the pore, perhaps driven by nucleotide-driven conformational changes of the pore loops. However, spastin may not need to completely translocate the tubulin polypeptide substrate, but instead just grip the C-terminal tubulin tail and exert mechanical ‘tugs’ that might partially unfold tubulin or locally destabilize protomer–protomer interactions, leading to catastrophic breakdown of the microtubule lattice. It also remains possible that the MIT domains could participate in this nucleotide-driven process by binding and ‘feeding’ the C-terminal tails to the pore. Further biophysical characterization will be needed to decipher the structural details of substrate recognition and mechanical force production. Our data also suggest that spastin may selectively recognize post-translationally modified tubulins (‘Glu’ tubulins) that are part of stable microtubules. Consistent with this idea, loss of spastin in Drosophila results in the accumulation of stabilized polyglutamylated tubulin in neurons⁸ and spastin knockout mice show axonal swellings enriched in detyrosinated, stable microtubules³⁰. Our structure also provides the first glimpse into how spastin disease mutations contribute to spastin dysfunction and disease, most of which we suggest are involved in destabilizing protomer–protomer interactions, microtubule- or ATP-binding (Fig. 4b and Supplementary Fig. 4); in such cases, spastin-linked HSP is probably caused by haploinsufficiency and not a dominant negative effect. Further elucidation of the mechanistic details of how spastin interacts with particular tubulin isoforms and post-translational modifications and leads to microtubule destabilization may provide insight into the origin of spastin paraplegias and potential treatments for this disease.

Supplementary Figures 1-8 and Legends, Supplementary Methods and Supplementary Tables 1-2

01