The sequence and structural similarity between γ- and α/β-tubulin suggested that nucleation results from microtubule-like contacts between γ- and α/β-tubulin⁸. The microtubule lattice consists of lateral and longitudinal contacts^9,10; the stronger longitudinal contacts define microtubule polarity, with “plus” and “minus” ends (Supplementary Fig. 1). The template model for microtubule nucleation predicts that γ-tubulins interact laterally to form a ring which makes longitudinal contacts with α/β-tubulin^2–5 minus ends; alternative models that predict lateral interactions between γ- and α/β-tubulin¹¹ have not been definitively ruled out.

Our previous 25 Å structure of Saccharomyces γTuSC was determined in buffer conditions that yielded predominantly monomeric complexes^12,13. Here, we show that buffer conditions that promote microtubule growth (BRB80: low salt, pH 6.9) also promote spontaneous assembly of γTuSCs into rings similar to Drosophila γTuRCs^2,4 (Fig. 1a,b). Ring formation was sensitive to both salt and pH (Supplementary Fig. 2a–c). The γTuSC rings bound microtubules, and many of these microtubule ends are capped (Fig. 1c), similar to microtubules nucleated in vivo^14,15 or from γTuRCs in vitro^3,4,16. Spontaneous assembly suggests that ring formation is an intrinsic property of γTuSC, and not dependent on γTuRC-specific proteins.

The N-terminal 220 residues of Spc110p (Spc110p^1-220), which attaches γTuSC to the nuclear face of the yeast spindle pole body, dramatically increased the stability and length of γTuSC assemblies (Fig. 1d). Copurification with Spc110p^1-220 yielded a continuum of γTuSC oligomers ranging from dimers to long, well-ordered helical filaments, even under conditions where γTuSC alone fails to assemble (Supplementary Fig. 2d–f). We determined the three-dimensional structure of γTuSC filaments from cryo-electron micrographs (Fig. 2a), using iterative helical real space reconstruction, a single particle approach to helical structure determination¹⁷. The resolution of the reconstruction, which included about 25,000 γTuSC subunits, was estimated at 8 Å by the Fourier shell correlation 0.5 cutoff (Fig. 2b).

The filament is a single spiral of laterally associated γTuSCs, without contact between layers (Fig. 2c,d; Supplementary Movie 1). The helical symmetry (54.3° rotation and 22.2 Å rise per subunit) gives rise to just over six and a half γTuSCs – or thirteen γ-tubulins – per turn, with a half γTuSC overlap. Each turn of helix forms a lock-washer shape similar to γTuRC⁴ (Fig. 2e). The thirteen-fold γ-tubulin symmetry of the filament is dictated largely by the extensive lateral interactions between Spc97p and Spc98p of adjacent γTuSCs (Figure 2e, Supplementary Fig. 6a), locking in the lateral tubulin contacts which on their own are flexible enough to accommodate a range of different symmetries. We propose that a γTuSC assembly very similar to a single ring from the filament provides the constraint that limits microtubules to 13 protofilaments in all eukaryotes in vivo⁷.

γTuSC in the filament is remarkably similar to free γTuSC^13,18 (Fig. 2f, Supplementary Fig. 3), indicating that oligomerization does not induce large scale conformational changes. The 8 Å structure provides new insight into the domain architecture of Spc97p and Spc98p. Spc97p and Spc98p dimerize at their N-terminal ends nearest the helical axis, and have extended central domains connecting to C-terminal γ-tubulin binding domains. The central domain of Spc98p is kinked, at a position previously shown to be the site of limited hinge-like flexibility¹³. The masses of the domains determined from the cryo-EM map provide a rough estimate of their boundaries in each sequence, indicating the positions of the grip1 and grip2 motifs, conserved in all γ-tubulin complex proteins¹⁹ (Supplementary Fig. 4a,b). The grip2 motif covers nearly half of the C-terminal domains, strongly suggesting that it is important for γ-tubulin binding. The grip1 motif is in the central domain, near inter-γTuSC contacts and the kink in Spc98p. We tentatively assign Spc110p^1-220 to a ridge of density running along the exterior face of γTuSC in the filament, making contacts primarily with Spc98p (Supplementary Fig. 3). The resolution of the reconstruction appears to be nonuniform, as tubes of alpha helical density are clear in the N-terminal domains of Spc97p and Spc98p at the core of the structure, while secondary structure features are not well defined in the peripheral density where γ-tubulin is located (Supplementary Fig. 5). The lower effective resolution in the γ-tubulin regions may be due to limited flexibility in the weak connections between the central and C-terminal domains of Spc97p and Spc98p.

The human γ-tubulin crystal structure²⁰ was fit into the density in the regions previously assigned to γ-tubulin¹³ (Fig. 3a,b). The minus end longitudinal surface of γ-tubulin is completely buried in the interface with Spc97p/Spc98p. The lateral contacts between γ-tubulins of neighbouring γTuSCs are nearly identical to microtubule lateral contacts. The two γ-tubulins within each γTuSC are skewed slightly apart, in a configuration incompatible with the microtubule lattice (Fig 3b,c), as observed in the free γTuSC structure¹³. This symmetry gives rise to an alternating pattern of γ-tubulin pairs with microtubule-like lateral spacing separated by gaps, generating a staggered mismatch with the microtubule lattice (Fig. 3c–e). The only microtubule lattice surface of γ-tubulin fully exposed in the filament is the plus end face, favouring a model in which γ-tubulin makes longitudinal contacts with α/β-tubulin. This, combined with the thirteen-fold γ-tubulin symmetry, provides the strongest evidence to date to support a γ-tubulin template mechanism for microtubule nucleation.

We tested the capacity of γTuSC oligomers to nucleate microtubules (Fig. 4). At pH 6.9 both γTuSC alone and γTuSC filaments provided modest levels of nucleation, slightly higher for γTuSC filaments than γTuSC alone. At pH 7.5 γTuSC alone does not nucleate microtubules, whereas the filaments retain a low level of nucleation. As γTuSC rings do not form at pH 7.5, but γTuSC filaments remain intact (Supplementary Fig. 2b,c), these results suggest that γTuSC nucleation activity is assembly-dependent. The levels of nucleation observed are consistent with previous measurements for γTuSC^2,21, but less robust than seen with γTuRC^2,5, suggesting that assembly alone is insufficient to fully activate γTuSC nucleating activity. The arrangement of γ-tubulin in γTuSC oligomers provides a structural explanation for their relatively modest nucleating activity. Nucleation likely arises from the inter-γTuSC γ-tubulin pairs, which have the correct microtubule lattice spacing. Simulations indicate that a γTuSC assembly in which all of the γ-tubulins make lateral microtubule-like contacts would provide greatly enhanced nucleation (L. Rice, personal communication).

The structure provides a clear hypothesis for how nucleation could be fully activated. We previously predicted that bending at the flexible kink in Spc98p is required to bring the intra-γTuSC γ-tubulins to the microtubule spacing^13,18. We speculated that γTuSC assembly might drive this change, but that clearly is not the case – a similar conformational change is still required in γTuSC rings. In the lower resolution γTuSC structure we predicted that the movement would be a closure of the gap between γ-tubulins; here we see that the movement must be more perpendicular to the edge of the ring, bringing γ-tubulin in toward the helical axis. A rotation of 23° about the kink in Spc98p would reposition γ-tubulin by 26 Å, bringing it to the microtubule lattice spacing (Supplementary Fig. 7, Movies 2 and 3). The staggered γ-tubulin arrangement likely serves a regulatory function, maintaining γTuSC oligomers in a low activity state until a signal (protein binding, post-translational modification, etc.) directs the rearrangements in Spc98p necessary to form a template with exact microtubule lattice geometry. Although less likely, rearrangement of the γ-tubulins could be induced by binding of α/β-tubulin. In such a model γTuSC would function primarily as a cap for stabilizing and localizing microtubule minus ends, rather than as a strong nucleator.

The dramatic enhancement of γTuSC oligomer stability by Spc110p, combined with its role in γTuSC localization, likely serves to ensure that microtubule template assembly in yeast occurs only at the spindle pole body. We propose a general model for microtubule nucleation in which Spc110p or its functional equivalent directly attaches γTuSC to microtubule organizing centres, promoting template assembly. A subsequent activation step then fully activates nucleation by rearranging the γ-tubulin network (Fig. 5a). In a template with seven γTuSCs, the location of the half γTuSC overlap defines the position of the 13 protofilament microtubule seam; a single lateral contact between γ-tubulin and α-tubulin would be made at the overlap, as well. It is unclear how many γTuSCs are required to nucleate a microtubule – an incomplete ring may be sufficient to initiate growth.

The γTuSC filament structure provides unique insight into the roles of γTuRC-specific proteins. Our results clearly show that γTuSC assembly alone establishes thirteen-fold γ-tubulin symmetry, indicating that the γTuRC-specific proteins are not required as a scaffold. This is consistent with the observation that all of the γTuRC-specific proteins can be depleted without affecting centrosomal microtubule nucleation of thirteen protofilament microtubules ^22,23. While not required as a scaffold, the γTuRC-specific proteins may serve to stabilize the ring structure or fully activate nucleation activity, and they are essential for γTuRC localization at non-centrosomal sites, as in augmin-dependent binding within the mitotic spindle²⁴ (Fig. 5b).

We suggested above that the grip1 and grip2 motifs, conserved in all the γTuRC-specific proteins, are involved in ring assembly contacts and γ-tubulin binding, respectively. This raises the intriguing possibility that the γTuRC-specific proteins may each bind γ-tubulin and substitute for Spc97p or Spc98p in the ring itself. To do this, they might form hybrid γTuSCs with one of the γTuRC-specific proteins plus Spc97p or Spc98p, alternative γTuSCs with two different γTuRC-specific proteins, or unique half γTuSCs (Supplementary Fig. 8). Such alternative γTuSCs might serve to initiate or terminate γTuSC oligomerization, or to stabilize the ring at the overlapping ends, while providing unique attachment sites in the structure of the ring itself.

1

Supplementary Movie 1

Supplementary Movie 2

Supplementary Movie 3

Supplementary Movie 4

supplement 1