CM1-driven assembly and activation of yeast γ-tubulin small complex underlies microtubule nucleation

Microtubule (MT) nucleation is regulated by the γ-tubulin ring complex (γTuRC), conserved from yeast to humans. In Saccharomyces cerevisiae, γTuRC is composed of seven identical γ-tubulin small complex (γTuSC) sub-assemblies, which associate helically to template MT growth. γTuRC assembly provides a key point of regulation for the MT cytoskeleton. Here, we combine crosslinking mass spectrometry, X-ray crystallography, and cryo-EM structures of both monomeric and dimeric γTuSCs, and open and closed helical γTuRC assemblies in complex with Spc110p to elucidate the mechanisms of γTuRC assembly. γTuRC assembly is substantially aided by the evolutionarily conserved CM1 motif in Spc110p spanning a pair of adjacent γTuSCs. By providing the highest resolution and most complete views of any γTuSC assembly, our structures allow phosphorylation sites to be mapped, surprisingly suggesting that they are mostly inhibitory. A comparison of our structures with the CM1 binding site in the human γTuRC structure at the interface between GCP2 and GCP6 allows for the interpretation of significant structural changes arising from CM1 helix binding to metazoan γTuRC.


Introduction
The microtubule (MT) cytoskeleton plays an essential role in the spatio-temporal control of eukaryotic cellular organization, cytoplasmic transport, and chromosome segregation during mitosis (Desai and Mitchison, 1997). The organization and function of the cytoskeletal network is tightly controlled by regulating the rate and location of nucleation, as well as MT polymerization kinetics and stability (Akhmanova and Steinmetz, 2015;Howard and Hyman, 2009;Teixidó -Travesa et al., 2012).
In most cells, MT nucleation occurs primarily at MT organizing centers such as centrosomes or spindle pole bodies and is dependent on the universally conserved g-tubulin ring complex (gTuRC) (Lüders and Stearns, 2007). In budding yeast, homologues of the grip-containing proteins (GCPs) GCP2 and GCP3 (Spc97p and Spc98p) and two copies of g-tubulin (Tub4p) form a 300 kDa complex (gTuSC) (Vinh et al., 2002). Spc110p, a distant pericentrin homologue, recruits this complex to the nuclear face of the spindle pole body (SPB), while Spc72p recruits it to the cytoplasmic face (Knop and Schiebel, 1997;Knop and Schiebel, 1998;Nguyen et al., 1998). Both of the g-tubulin complex receptors contain the highly conserved centrosomin motif 1 (CM1) (Lin et al., 2014;Zhang and Megraw, 2007). In yeast, the CM1 motif is required for seven identical gTuSCs to helically assemble into a gTuRC at the SPB (Knop and Schiebel, 1998;Kollman et al., 2015;Lyon et al., 2016;Nguyen et al., 1998).
In metazoans and plants, the gTuRC is recruited to MT nucleation sites as a large, pre-formed ring-shaped 2.2 MDa complex (Teixidó -Travesa et al., 2012). The metazoan gTuRC is composed of 14 g-tubulins, and a smaller number of the g-tubulin binding proteins, GCP2-6, as well as other accessory proteins. While sharing only~15% homology and varying in size from 70 kDa to 210 kDa, GCP2-6 share a conserved core of two grip domains (Guillet et al., 2011). Structural and biochemical studies have shown that the N-terminal grip1 domain drives lateral association between GCPs, while the grip2 domain binds to g-tubulin (Choy et al., 2009;Farache et al., 2016;Greenberg et al., 2016;Guillet et al., 2011;Kollman et al., 2015). Recent cryo-EM structures revealed that five copies of the GCP2/3 gTuSC are integrated into the metazoan gTuRC along with a GCP4/5 and a GCP4/6 pseudo-gTuSC as well as other accessory proteins (Consolati et al., 2020;Liu et al., 2020;Murphy et al., 2001;Oegema et al., 1999;Wieczorek et al., 2020b). In the human gTuRC, two copies of the CM1-containing CDK5RAP2 gTuRC nucleation activator (gTuNA) were found at the GCP2-6 interface (Wieczorek et al., 2020a).
Previous moderate-resolution cryo-EM structural studies (8 Å ) had shown that wild-type (WT) yeast gTuSCs complexed with the N-terminal domain of Spc110p self-assemble into helical filaments (hereafter gTuRC) having 6.5 gTuSCs/turn, thereby presenting 13 g-tubulins to template 13protofilament MTs (Kollman et al., 2010;Kollman et al., 2015). Although close to MT symmetry, the g-tubulins within each gTuSC were too far apart to correctly match the MT lattice, adopting an open conformation. The relevant in vivo conformation was determined by cryo-tomography and sub-volume averaging, clearly showing a MT-matching geometry at the yeast SPB, suggesting that gTuRC closure might be an important regulatory step . To validate this hypothesis, g-tubulin was engineered with disulfides to stabilize a closed MT-like conformation (gTuRC SS ), resulting in significantly enhanced MT nucleation . This also had the benefit of improving the cryo-EM map (6.5 Å ) such that an initial pseudoatomic model Kollman et al., 2015) could be built based on the crystal structure of human g-tubulin (Aldaz et al., 2005;Rice et al., 2008) and the distant and much smaller (75 kDa vs. 97 or 98 kDa) human GCP4 (Guillet et al., 2011;Kollman et al., 2015). These structures suggest a hierarchical model of gTuSC activation, with gTuSC assembling at the SPB in an Spc110p-dependent manner into an open slightly active conformation of the gTuRC prior to gTuRC closure.
Biochemical studies on the role of Spc110p in gTuRC assembly revealed that higher-order oligomerization of Spc110p and its binding to gTuSCs was required to overcome the intrinsically weak lateral association of gTuSCs at physiologically relevant gTuSC concentrations (Kollman et al., 2010;Kollman et al., 2015;Lyon et al., 2016). Deletion studies identified that independent of oligomerization removal of Spc110p residues 1-111 (Spc110p 1-111 ) was lethal in vivo but only slightly compromised gTuRC assembly in vitro, perhaps suggesting an essential regulatory function. By contrast, deletion of the subsequent centrosomin motif 1 (CM1, Spc110p 117-146 ) additionally abolished gTuRC assembly in vitro ( Figure 1A). Supporting the need for precise regulation of gTuRC assembly and function, all the components of the gTuSC, as well as Spc110p and Spc72p, are phosphorylated at multiple sites in a cell-cycle-dependent manner (Fong et al., 2018;Keck et al., 2011). Mutations at several of these phosphorylation sites have been shown to impact cellular viability, spindle morphology, or shown to affect gTuRC assembly (Fong et al., 2018;Huisman et al., 2007;Keck et al., 2011;Lin et al., 2011;Lin et al., 2014;Lyon et al., 2016;Peng et al., 2015;Vogel et al., 2001). Together these data suggest a hierarchical model of gTuSC activation; with gTuSC assembling at the SPB into an open gTuRC that would be further activated by closure ( Figure 1B). However, owing to the lack of structural data, the molecular mechanisms by which Spc110p facilitates gTuRC assembly and activation have remained unclear.
Here, we use crosslinking mass spectrometry (XL-MS) and X-ray crystallography to identify and determine the structure of the N-terminal coiled-coil (NCC) of Spc110p-bound in gTuSC filaments previously observed in cryo-EM reconstructions. The combined data show that a unique pose of the coiled-coil satisfies most of the XL-MS restraints. Furthermore, integrative modeling indicates that only residues N-terminal to the coiled-coil from a single protomer were required to satisfy the majority of the crosslink restraints, suggesting an asymmetric mode of Spc110p binding to gTuRC.
We present cryo-EM structures of monomeric and dimeric gTuSCs at near-atomic resolution, as well as higher-resolution (~3.0-4.0 Å ) cryo-EM structures obtained from gTuRC filaments in the open and closed conformations (Kollman et al., 2010;Kollman et al., 2015). These have allowed de novo model building of unknown regions and reinterpretation of significant portions of gTuSC structure. Our atomic models of gTuSCs in different assembly and conformational states provide insights into the mechanisms of gTuRC assembly and activation mechanisms required for MT nucleation and reveal how N-terminal regions of Spc110p, notably CM1, facilitate gTuRC assembly. Many of the annotated phosphorylation sites had fallen in regions of gTuSC not present in GCP4, and hence not previously modeled. Thus, the new structure provides a powerful atomic framework for understanding the importance and mechanism of regulatory modifications.

Results
Defining Spc110p:gTuSC interactions by XL-MS Our previous 6.9 Å cryo-EM reconstruction, derived from helical filaments of Spc110p bound to an engineered closed conformation of gTuRC SS , revealed an~40-residue-long segment of coiled-coil density contacting the N-terminal region of Spc97p. The limited resolution prevented rigorous assignment of this density to any particular portion of Spc110p. Given its importance for gTuRC assembly, the coiled-coil seemed likely to correspond to either the conserved Spc110p CM1 or the 45-residue segment (Spc110p 164-208 ) predicted with high probability to be a coiled-coil (the NCC, or Spc110p NCC ; see Figure 2A). Beyond this ambiguity, previous maps also lacked any density for the non-coiled-coil regions of Spc110p 1-220 known to be biochemically important and absolutely required for viability (Lyon et al., 2016).
To define the important interaction interfaces between Spc110p and gTuSC, we utilized XL-MS with the same Spc110 1-220 construct used for the cryo-EM as well as a longer Spc110 1-401 construct ( Figure 2-figure supplement 1). We observed a significant number of crosslinks between the N-terminal portions of Spc97p and Spc110p NCC (Figure 2-figure supplement 1). Thus, this region and not CM1 was responsible for the coiled-coil-gTuSC interaction apparent in the cryo-EM map. As shown below, CM1 instead binds at a cleft that spans two adjacent gTuSCs.  Lyon et al., 2016. Assembly data was generated using Spc110p-GCN4 tetramer fusion constructions, while in vivo data used full-length proteins in a red-white plasmid shuffle assay. (B) Schematic overview of gTuRC assembly: monomeric gTuSCs bind to Spc110p 1-220 and assemble into an open gTuRC, which undergoes closure prior to or concurrent with microtubule nucleation.
The Spc110p NCC  binds to gTuSC at the N-terminal regions of Spc97p Due to the limited resolution of the previous cryo-EM reconstruction, the derived atomic model of the coiled-coil contained only the peptide backbone Kollman et al., 2015). Motivated by the crosslinks observed between the Spc110p NCC and gTuSC, we sought a higher-resolution structure of the NCC region via X-ray crystallography. Previous work indicated that  Greenberg et al., 2016). The majority of crosslinking mass spectrometry (XL-MS) distance restraints are satisfied by this model. Satisfied and violated disuccinimidyl suberate (DSS) crosslinks are shown in cyan and purple, respectively. Satisfied and violated 1-ethyl-3-(3-dimethylaminopropyl) carbodiimide (EDC) crosslinks are shown in blue and red, respectively. Crosslinks that are satisfied by either Spc110p monomer are shown twice, one for each monomer. (D) Localization density map for the ensemble of integrative models consisting of two adjacent gTuSCs, each bound to an Spc110p 1-220 dimer. The map shows the positions of different parts of the complex in the ensemble of models from the top cluster; maps for all components are contoured at 2.5% of their respective maximum voxel values. The modeling results shown are based on the gTuSC-Spc110p 1-220 -GCN4 crosslinks; similar results were obtained using gTuSC-Spc110p 1-401 -GST crosslinks (see Appendix 1). The online version of this article includes the following figure supplement(s) for figure 2:    Spc110 1-220 is only weakly dimeric (Lyon et al., 2016). Using the proven strategy of fusing weakly interacting coiled-coils with stabilizing domains (Andreas et al., 2017;Frye et al., 2010;Klenchin et al., 2011), we found that an N-terminal fusion of Spc110p 164-207 with a domain from Xrcc4 produced high yields of soluble protein. The Xrcc4-Spc110p 164-207 construct crystallized in a variety of conditions, diffracted to 2.1 Å , and enabled phases to be obtained by molecular replacement using Xrcc4 as a search model. As expected, the electron density map was consistent with a coiled-coil, with interpretable density for Spc110p 164-203 residues ( Figure 2B, Supplementary file 1). When the coiled-coil was docked into the 6.9 Å cryo-EM map using cross-correlation in Chimera (Pettersen et al., 2004), the X-ray model occupied most of the alpha-helical cryo-EM density. Importantly in the docked conformation the majority of the unique crosslinks ( Figure 2C, Figure Greenberg et al., 2016;Kollman et al., 2015;Rout and Sali, 2019;Russel et al., 2012).
We first considered a single gTuSC bound to an Spc110p 1-220 dimer ( Figure 2-figure supplement 2A). Using a combination of the previous cryo-EM-based gTuSC pseudoatomic model Kollman et al., 2015), the X-ray structure of Spc110p 164-207 , and representing the rest of gTuSC and Spc110p 1-220 by flexible strings of beads representing the amino acid chain, approximately 3000 good-scoring models were obtained satisfying the crosslinks and stereochemistry (excluded volume and sequence connectivity). These models were clustered based on structural similarity (Figure 2-figure supplement 4), and~98% of the models were well represented by a single cluster that satisfied >90% of the crosslinks (see Appendix 1).
Consistent with visual inspection of the crosslinks, the localization probability density map from the most occupied cluster (Figure 2-figure supplement 2A) indicated that Spc110p 1-163 extended from the Spc110p NCC along the Spc97p-Spc98p interface towards g-tubulin and the C-termini of Spc97p/98p. The precision of the model was insufficient to distinguish separate paths for the noncoiled-coil regions of each protomer within the Spc110p dimer. Consequently, we also considered a model containing Spc110p 1-163 from a single protomer, which almost equally satisfied the crosslink restraints and indicated a similar path (Figure 2-figure supplement 2B).
As the localization probability map suggested that the two Spc110p protomers might follow different paths, with one path extending towards the adjacent gTuSC, we also modeled two adjacent gTuSCs, each bound to an Spc110p 1-220 dimer ( Figure 2D). By considering adjacent gTuSCs, the predicted path spans from the N-terminus of Spc97p of one gTuSC before proceeding towards the Spc98p from the adjacent gTuSC and binding in the space between the two gTuSCs. There is also a component that extends towards the Spc97p C-terminus and g-tubulin ( Figure 2D). Together these results suggest a complex path interacting with multiple gTuSCs taken by at least one of the two Spc110p N-termini.

High-resolution filament structures reveal previously uninterpretable regions of gTuSC
The observed binding site between Spc110p NCC and gTuSC explains how Spc110p oligomerized at spindle poles can stimulate gTuRC assembly by increasing the local gTuSC concentration. However, this fails to explain the critical biochemical and in vivo functional importance of residues N-terminal to the Spc110p NCC , such as the Spc110p CM1 region, for gTuRC assembly and MT nucleation (Lyon et al., 2016). While the crosslinking and integrative modeling data suggested a physical basis for these functional roles, the actual path and interactions taken by Spc110p 1-163 were unknown. Realizing that this would require much higher resolution of Spc110p-gTuSC interactions, we focused on obtaining higher resolution structures of the 'open' (gTuRC WT ) and disulfide trapped 'closed' (gTuRC SS ) filaments containing Spc110p (Kollman et al., 2010;Kollman et al., 2015) by collecting new datasets on a direct electron detector and incorporating symmetry expansion and 3D classification into the data processing pipeline.
Filaments were initially processed in Relion2 (Kimanius et al., 2016), while allowing for the refinement of helical parameters, prior to further refinement of alignment parameters in FREALIGN (Grigorieff, 2016). As with previous studies, a combination of local helical and conformational inhomogeneities led to significantly worse resolution in the Spc97p/Spc98p C-terminus/g-tubulin region compared to the N-terminal and middle domains of Spc97p/Spc98p, particularly for the gTuRC WT filaments. To improve the resolution, we performed symmetry expansion followed by focused classification of segments containing three adjacent gTuSCs. The resulting reconstructions were at a resolution of 3.6 Å and 3.0 Å for the gTuRC WT and gTuRC SS filaments, respectively ( Figure 3A Previous high-resolution crystal structures of g-tubulin have shown that it adopts a bent-like state when not in complex with GCPs, independent of its nucleotide state (Aldaz et al., 2005;Rice et al., 2008). This raised the possibility that g-tubulin might change conformation upon assembly into gTuSC. While the changes are small, in our structures, g-tubulin adopts a conformation distinct from the previously observed bent human g-tubulin or the yeast tubulin straight conformations (Figure 3figure supplement 5A). In the assembled state, g-tubulin H6 adopts what appears to be an intermediate conformation between the bent and straight conformations, while the C-terminal portion of the g-tubulin H6-H7 loop that most defines the interface with the incoming a-tubulin adopts a conformation similar to a straight yeast b-tubulin, likely potentiating MT formation (Figure 3-figure supplement 5A).
In looking for a potential cause for the altered g-tubulin H6-H7 loop conformation, there was one notable difference in the Spc/g-tubulin interface. The g-tubulin T7 loop in assembled g-tubulin moves such that it now more closely resembles the b-tubulin T7 loop of an assembled b-tubulin (Figure 3figure supplement 5B). The g-tubulin T7 loop is pinned between a loop (Spc98p H15-16 /Spc97p H16-17 ) located at the N-terminus of a small domain in Spc97p and Spc98p and the adjacent C-terminal helical bundles (Spc98p H22-23 /Spc97p H26-27 ). These results suggest that although subtle, assembly of yeast g-tubulin into a gTuSC may help promote a more MT-like g-tubulin plus end conformation, facilitating nucleation.
Previous structures of yeast GCPs and their assemblies suggested that the interface between the GCPs was largely formed from the two N-terminal helical bundles. Our high-resolution structures allow us to resolve large divergent N-terminal sequences present in both Spc97p and Spc98p, but absent in the shorter GCP4 'core' structure, which contribute to the intra-and inter-TuSC interfaces.
The GCP intra-gTuSC interface extends the entire length of the two N-terminal helical bundles of Spc97p and Spc98p, and also features significant contacts by the newly resolved N-terminal regions ( Figure 3C). Of the residues newly modeled, Spc97p 1-54,81-89 and Spc98p 163-179 contribute an additional~3600 Å 2 of buried surface area to the N-terminal interface. In addition, a previously unmodeled 33-residue insertion in the middle of Spc98p (Spc98p 672-704 ), between helices Spc98p H23 and Spc98p H24 , folds into a pair of strands, contributing an additional~1900 Å 2 of surface area. In the closed state, there is a small contact patch between the N-terminal region of Spc98p H27 and Spc98p H19 . Thus, while the much shorter GCP4 structure, which formed the basis of previous modeling efforts, suggested well-conserved N-terminal interactions, it is clear that a very large part of intra-gTuSC stabilization (~5400 out of~8000 Å 2 , total interface) arises from sequences in Spc97p and Spc98p not present in GCP4, suggestive of very tight binding. This is consistent with gTuRC assembly, particularly of smaller GCPs, being stabilized using non-gTuSC components in metazoans (Liu et al., 2020;Wieczorek et al., 2020a;Wieczorek et al., 2020b).
In contrast, the inter-gTuSC interface is much more limited in scope (total surface area~2900 Å 2 ). It is mainly composed of two smaller, largely hydrophilic contact patches located at the three N-terminal helical bundles, and a small set of hydrophobic contacts. In addition, a small contact between Spc97p K790 and Spc98p Y510 ( Figure 3D) involves almost no hydrophobic residues. The limited inter-g TuSC interface explains why gTuSCs fail to assemble under physiological concentrations (Kd~2 mM), and thus must rely on a combination of CM1 interactions (see below) and avidity effects provided by Spc110p oligomerization (Lyon et al., 2016).

Spc110p CM1 facilitates gTuRC assembly by binding at the inter-gTuSC interface
As before ( Figure 2B, C), we observed coiled-coil density for the Spc110p NCC in our higher resolution maps. Given the observed pitch of the coiled-coil in the crystal structure, as well as density for   larger side chains, we were able to assign the register of the NCC (Figure 4-figure supplement 1).
To assess the path of Spc110p N-terminal to the Spc110p NCC , we generated a difference map between our experimental density maps and an atomic model for gTuRC, which did not include Spc110p atoms. This difference map should contain density for Spc110p and any regions not included in the atomic model. Indeed, the difference map revealed clear density extending from the NCC to a helical density that spans the inter-gTuSC interface and beyond ( Figure 4A). Based on the side-chain features, we were able to unambiguously assign CM1 117-141 to the helical inter-gTuSC density ( Figure 4B). While the density connecting the Spc110p CM1 helix with Spc110p NCC was at a lower resolution, we were able to model residues Spc110p 112-206 spanning the Spc110p CM1 and Spc110p NCC (Figure 3-figure supplement 3C, D).
Interestingly, a pair of helix-dipole/hydrogen bond interactions augment binding of the CM1 helix with Spc98p, with Spc98p D542 hydrogen bonding with the N-terminus of the Spc110p CM1 helix, and Spc110p K120 hydrogen bonding with the C-terminus of helix Spc98p H19 (Figure 4-figure supplement 2A). On Spc97p, the C-terminus of the Spc110p CM1 helix interacts with helices Spc97p H23 and Spc97p H28 and the loop C-terminal to Spc97p H21 , as well as the insertion between Spc97p H7 and Spc97p H9 at the N-terminus of Spc97p H8 ( Figure 4C, Figure 4-figure supplement 2B). While we were unable to trace residues Spc110p 1-111 in our structure, numerous crosslinks map to the region between Spc97p and Spc98p and g-tubulin at the intra-gTuSC interface ( Figure 2-figure supplement 2C). These residues may therefore be involved in contacts facilitating activation and closure.
Together, our data reveal that one protomer of Spc110p 112-206 within each Spc110p 1-220 dimer adopts a complex path across two gTuSCs, while the Spc110p 112-165 region of the second protomer is unresolved, a path that defines the molecular role of the conserved CM1 motif. Beginning with Spc110 NCC (Spc110p 164-208 ) bound to the N-terminus of Spc97p near the intra-gTuSC interface and moving towards the N-terminus, Spc110p next interacts with Spc98p and then weaves a path along the surface of Spc97p. From there, the CM1 helix binds across the inter-gTuSC interface to Spc98p on the adjacent gTuSC. After that, it continues along the surface of Spc98p, then turns towards the Spc97p C-terminus ending near g-tubulin ( Figure 2-figure supplement 2D, Figure 4A, C). Integrating these data generates a continuous path across two gTuSC subunits ( Figure 2-figure supplement 2D). This is in good agreement with modeling predictions.
To assess the generality of the observed CM1 binding mode, we mapped conservation of the CM1 motif and its binding sites on Spc97p and Spc98p (Figure 4-figure supplement 3A, B). Of note, the more C-terminal portion of CM1 that binds to Spc97p is better conserved than the N-terminal portion that binds to Spc98p (

Conformational changes of Spc97p and Spc98p during assembly
To better resolve fundamental questions about the molecular basis for gTuRC assembly and activation, we determined the cryo-EM structure of unassembled gTuSC, without Spc110p or filament formation, from images of frozen-hydrated single particles. At the concentration of~1 mM used in data collection, micrographs and 2D classes show a mixture of gTuSC monomers and dimers, with a small number of larger oligomers (Figure 5-figure supplement 1). We were able to obtain a structure of the gTuSC monomer at~3.7 Å , and of a gTuSC dimer at~4.5 Å resolution ( Figure 5-figure supplements 2 and 3, Supplementary file 2). The gTuSC dimer is formed from two gTuSCs in lateral contact using the same interface as observed in the gTuRC:Spc110p filament structures, but as expected lacks density for both the Spc110p NCC and the Spc110p CM1 helix.
In order to assess the changes that occur during assembly of monomeric gTuSCs into the gTuRC and the subsequent closure, we aligned the N-terminal two helical bundles of Spc97p and Spc98p (residues Spc97p 52-276 and Spc98p 178-342 ). This alignment allows for a concise description of the joint conformational changes in both proteins that occur as gTuSCs assemble into gTuRCs and the subsequent closure required for MT nucleation (  During the transition from monomer to assembled open state (as seen in gTuRC WT ), the g-tubulins move in the same overall direction, approximately orthogonal to the plane of the Spc97p/Spc98p contact interface ( Figure 5-figure supplement 4A). The center of mass of the g-tubulins shifts~13.8 Å and~15.6 Å when bound to Spc97p and Spc98p, respectively, as a result of twisting the helical bundles in Spc97p and Spc98p. All of the conserved contacts in Spc97p and Spc98p observed in assembled gTuSC filaments occur in the N-terminal three helical bundles. Notably, much of the bottom three helical bundles show only minor changes when assembling to the open state. The dominant changes occur on loop Spc98p H10-S1 in the middle contact, which moves~4.1 Å , and at the N-terminus of helix Spc98p H11 , involved in the top contact, which moves~6.6 Å . The large conformational change in Spc98p that occurs to create these contacts in the gTuSC dimer is a major contributor to the even larger-scale changes during gTuRC assembly and activation. In addition, the Spc98p-bound g-tubulin undergoes a twisting motion of~5-6˚. During these conformational changes, the inter-gTuSC contacts make only minor alterations, mainly in the N-terminal three helical bundles of Spc97p and Spc98p, which undergo complex tilting and twisting motions. Overall, these conformational changes alter the pitch and twist of the gTuRC assemblies from~140 Å /turn and 54.5 o in the open state to~132 Å /turn and 55.1 o in the closed state ( Figure 5A, B, D, E).
Excising a full turn in our gTuRC filament structures containing seven gTuSC subunits provides a good model for an isolated gTuRC as it might bind at the SPB. This reveals that within each gTuRC there are only six complete CM1 binding sites, the last one being interrupted at the end of the ring. This in turn suggests that only six Spc110p molecules need to be bound to a gTuRC in vivo to stabilize the full ring. This helps explain the apparent symmetry mismatch between the underlying hexameric organization of Spc42p (Bullitt et al., 1997;Drennan et al., 2019;Muller et al., 2005) within the SPB and the heptameric gTuRC. The geometry is such that the Spc110p NCC binding site most proximal to the SPB would be empty ( Figure 5E, F).
Finally, by local 3D classification we observed that a closed state is populated in our gTuRC WT data (Figure 5-figure supplement 1, Supplementary file 2). This state had previously not been observed in the gTuRC WT structure as robust symmetry expansion and 3D classification techniques had not yet been developed for cryo-EM when the structure was published. While not identical to the disulfide crosslinked closed state, the differences are minimal, indicating that the conformational changes observed at highest resolution in gTuRC SS are representative of those occurring in the closed WT gTuRCs ( Figure 5-figure supplement 4C). The fact that the WT closed state can occur spontaneously and is sampled in our open population suggests that, in the presence of Spc110p, the energy differences between the open and closed states are not large.

Mapping phosphorylation sites on the gTuRC suggests largely inhibitory roles
The gTuSC is heavily phosphorylated in a cell-cycle-dependent manner, and perturbing phosphorylation has been shown to affect spindle morphology (Fong et al., 2018;Keck et al., 2011;Lin et al., 2011;Peng et al., 2015;Vogel et al., 2001). The role of many of these phosphorylation sites To better understand the potential role of phosphorylation in gTuRC assembly, regulation. and function, we mapped a recently determined set of phosphorylation sites, including a re-analysis of previously determined data and newly acquired data from SPBs (Fong et al., 2018), onto a dimer of our gTuRC SS structure ( Figure 6A). Here, we focus on the serine/threonine sites, given the minimal tyrosine kinase activity in yeast. Surprisingly, phosphorylation at the majority of the mapped sites would seem to destabilize the assembled gTuRC and thus may help keep unassembled or partially assembled components inactive. Phosphorylation at two sites would likely stabilize assembly, indicating the complex modulatory role played by phosphorylation.
Many of these phosphorylation sites map to potentially important interfaces: the Spc110p/ Spc97p interface (at the Spc110p NCC and at the Spc110p NCC-CM1 loop), the inter-gTuSC interface, the g-tubulin/a-tubulin interface, as well as a cluster of sites at the Spc97/98p:g-tubulin interface. There are also a large number of unmapped phosphorylation sites, the majority of which are located on low-resolution or unresolved regions in the N-termini of Spc98p and Spc110p.
Strikingly, our gTuRC SS structure reveals a cluster of phosphorylation sites, with many exhibiting in vivo phenotypes, that maps near the Spc110p:gTuSC interface in the Spc110p NCC region and near the loop connecting Spc110p NCC and Spc110p CM1 . Of particular note are a set of sites on Spc110p (Spc110p T182 , Spc110p T188 ) and the adjacent interface on Spc97p (Spc97p S84 , Spc97p T88 ). Together, these would add numerous negative charges in a portion of the Spc97p/Spc110p interface that is already highly negatively charged, especially the Spc110p NCC . Phosphorylation at two of these sites (Spc110p T182 and Spc97p S84 ) would likely negatively impact Spc110p binding, whereas Spc97p T88 is adjacent to a positively charged patch; phosphorylation at this site would likely promote Spc110p binding.
Three sites on Spc97p (Spc97p S130 , Spc97p S208 , Spc97p S209 ) and two sites on Spc110p (Spc110p S153 , Spc110p S156 ) map onto or near the loop connecting Spc110p CM1 with the Spc110p NCC and its interface with Spc97p. Mutation of Spc97p S130 exhibited a temperature-sensitive phenotype, and the Spc97p S208A/S209A , Spc97p S208D/S209D double mutants were lethal, consistent with phosphorylation of this region potentially having a regulatory role (Fong et al., 2018;Lin et al., 2011). While the loop has a lower resolution than other portions of the map (Figure 3-figure supplement  3C, D), the backbone approximately tracks with a long negatively charged patch along Spc97p and Spc98p ( Figure 6C). Furthermore, the Spc110p 150-161 loop has two negative charges and one positive charge. Although phosphorylation at Spc110p S153 and Spc110p S156 was not consistently observed (Fong et al., 2018), phosphorylation at these sites, as well as on the opposite Spc97p interface, would likely destabilize Spc110p binding.
One site on Spc97p maps near the inter-gTuSC dimer interface. The interface is rearranged only by a few Å ngstroms during activation, so any effect of phosphorylation would presumably only impact assembly. Spc97p S797 mutations produce a mild phenotype (Fong et al., 2018), and it is unresolved in all of our structures, but it is likely on a flexible loop near a positive patch in Spc110p and Spc97p and may thus favor assembly.  Figure supplement 2. WT g-tubulin small complex (gTuSC) processing and resolution. Figure supplement 3. Segmented single-particle reconstructions of g-tubulin small complex (gTuSC) monomer and dimer. Figure supplement 4. Conformational changes in g-tubulin small complex (gTuSC) during assembly and activation. Figure 5-video 1. Conformational changes of g-tubulin small complex (gTuSC) during assembly. https://elifesciences.org/articles/65168#fig5video1 Figure 5-video 2. Conformational changes of g-tubulin ring complex (gTuRC) during activation. https://elifesciences.org/articles/65168#fig5video2 Finally, g-tubulin S71 localizes near the g-tubulin:a-tubulin interface, likely decreasing binding affinity, and perhaps even interfering with GTP binding. g-tubulin S71 and g-tubulin S74 mutants (A or D) both exhibit phenotypes, likely reflecting the importance of proper hydrogen bonding near the gtubulin GTP binding site ( Figure 6D).

Comparison of yeast gTuRC with metazoan TuRC structures
Recent efforts by several labs have been successful in providing the first models for the more complex metazoan gTuRCs (Consolati et al., 2020;Liu et al., 2020;Wieczorek et al., 2020a;Wieczorek et al., 2020b). These new structures provide much needed clarity on the stoichiometry of the five different GCPs (GCP2-6) and how they are organized within the gTuRC ring. They also reveal unexpected structural roles for numerous accessory components. Of interest to us was the role CM1-containing accessory proteins might have in metazoan gTuRC assembly and conformation.
Further support for a conserved role for CM1 is apparent in the recently published structure of the human gTuRC purified by affinity with gTuNA, an N-terminal truncation of CDK5RAP2, which includes its CM1 motif. The authors assign CM1 to the helical density at the interface between GCP2 and GCP6 (Wieczorek et al., 2020a;Wieczorek et al., 2020b). This is precisely equivalent to our assigned yeast CM1 helix at the Spc98p-Spc97p interface (Figure 7-figure supplement 1A).
Notably, separate structural studies of human and Xenopus gTuRC, where the gTuRC was purified by affinity against GCP2 and g-tubulin, respectively, showed no density at the same interface (Consolati et al., 2020;Liu et al., 2020). Furthermore, when the human gTuNA-bound map is filtered to low resolution, density similar to that observed in our yeast gTuRC ss difference map continues from the N-terminus of the CM1 helix along the surface of GCP6 towards GCP4 (Figure 7figure supplement 1B, C). Taken together, these results stress the broad conservation and importance of CM1 binding.
These compositional differences led us to wonder whether CM1 binding might also drive conformational rearrangements in the metazoan gTuRCs, analogous to the changes we observed during yeast gTuRC assembly. Due to the lower resolution and lack of deposited atomic coordinates for the GCP2 affinity-purified human structures, we limited our comparison to the gTuNA-bound human and Xenopus (gTuNA-unbound) structures. Perhaps surprisingly, both metazoan gTuRC structures show a very poor match to MT symmetry and would require substantial g-tubulin motions to match the MT (Figure 7). The g-tubulins in the metazoan gTuRC structures are displaced up to~46 Å from their ideal MT-like positions, as opposed to the 9 Å observed in our closed yeast gTuRC ss structure, suggesting that the metazoan gTuRCs may be even more strongly dependent upon additional factors or post-translational modifications (PTMs) to achieve an active conformation than the yeast gTuRCs.
While the human and Xenopus GCPs overlay very well at g-tubulin positions 1-10, the terminal four positions show a different twist and pitch. We suggest here that these differences arise from CM1 binding at GCP2:GCP6 interface in the human gTuRC. The relative position of the g-tubulins bound to GCP2 and GCP6 changes upon CM1 binding to much more closely match what we observe. That is, during the 'transition' from a CM1-absent gTuRC (Xenopus) to a CM1-present (human) gTuRC, the GCP6-bound g-tubulin moves by~10 Å to better match the position observed in our closed TuRC ss structure (Figure 7-figure supplement 2A, B). From this we speculate that binding of CM1-containing accessory proteins at other sites within the gTuRC would further optimize their conformation and MT nucleating ability. Within the centrosome, CM1-containing proteins are expected to be in very high local concentrations due to the highly colligative/phase condensation behavior of the pericentriolar material (Feng et al., 2017;Woodruff et al., 2015), further promoting gTuRC activation.
In contrast to the yeast structures, having CM1 bound in the human gTuRC seems to correlate with breaking the GCP2/6 N-terminal interface (Figure 7-figure supplement 2C, D). The dissociation of this N-terminal interface may be due to an intrinsically weaker GCP2/6 interaction, enhancing the role that additional factors that bind at CM1 or the inter-GCP interface could play in regulating MT nucleation.

Discussion
Using a combination of single-particle and filament cryo-EM data, we have determined structures for monomeric and dimeric gTuSCs, along with assembled open and closed state gTuRCs at near-atomic ! (!) 7!/! /8 -9'7 : *2!/! ;'" 9-&< ;'" 9-&= ;'" ;'"> ;'" '  resolutions. Our structures complement existing structural and biochemical data with high-resolution snapshots of the yeast gTuSC and gTuRC. Together with previous work, these provide a framework for understanding the molecular basis for MT nucleation and regulatory processes likely necessary to ensure that MTs are only nucleated at the SPB. We provide the first molecular understanding for the critical role of the conserved Spc110p CM1 region in yeast gTuRC assembly. The structures suggest that nucleation is positively controlled in at least three ways: (1) assembly of gTuSCs into an open ring mediated by Spc110p oligomers and Spc110p CM1 , (2) closure of each gTuSC from an open state to a closed state to fully align the g-tubulins to the MT lattice, and (3) phosphorylation at Spc110p T88 can support Spc110p binding and directly impact gTuRC assembly. In addition, a number of phosphorylation sites on Spc110p and g-tubulin would have a negative impact on assembly or MT nucleation, either inhibiting gTuSC binding to Spc110p or ab-tubulin binding ( Figure 6).
Although minor, we also observe conformational changes in g-tubulin upon assembly into gTuSCs that mimic aspects of the bent-to-straight transition in ab-tubulin and would thus be expected to enhance MT nucleation. Unresolved is to what extent these differences arise from differences in the protein sequence from yeast to metazoans or represent an assembly-driven enhancement in g-tubulin conformation. There is at least some role for sequence as we know that there is a strong species barrier such that yeast gTuRC is hundreds-fold more potent at stimulating yeast tubulin polymerization than mammalian tubulin .
As initially observed in negative stain EM (Choy et al., 2009), our new cryo-EM structures of monomeric and dimeric gTuSCs show that Spc97p and Spc98p intrinsically adopt an open conformation at the intra-gTuSC interface such that the attached g-tubulins fail to make MT-like lateral contacts. Our structures of open and closed assembled gTuRCs show that Spc97p and Spc98p undergo large conformational changes during assembly into rings, bringing them much closer to MT geometry. Only smaller conformational changes occur as they transition from the open to the closed state during activation. The observation that a population of gTuRCs in the WT filaments adopts a locally closed conformation indicates a small energetic barrier to a single gTuSC closing, although simultaneous closing of an entire ring is unlikely. Thus, the addition/removal of PTMs or the binding of other factors could allosterically drive a more ideal template state. Indeed, we know that the yeast CK1d kinase, Hrr25, is needed for proper spindle formation in vivo and that it binds to gTuRCs and stimulates MT assembly in vitro (Peng et al., 2015), indicating that it is one such activator. gTuRC closure may also be stabilized by the process of MT assembly.
Our structures also resolve a long-standing mystery: how the sixfold symmetric Spc42p layer at the SPB (Bullitt et al., 1997) could facilitate the formation of a gTuRC containing specifically seven gTuSCs (Figure 8). This is resolved by recognizing that Spc110p CM1 within each dimer extends from one gTuSC to another, contributing to cooperative assembly, and cannot bridge across the large gap between the last and first gTuSCs in the ring. Thus, we suggest that six Spc110p dimers are symmetrically bound to the Spc42p lattice at the SPB. These would thus present the six CM1 motifs required to bind at the six complete CM1 binding sites formed within gTuRC heptamer. Given the observed pattern of connectivity where Spc110p CM1 extends across the interface in the same direction as the helical rise ( Figure 4A), this would leave the terminal NCC site nearest to the SPB unoccupied.
Our high-resolution structures are further poised to help inform on the mechanism of activation of the stable metazoan gTuRC complexes. Both of the published gTuRC structures would require large conformational changes in pitch and rise to match MT symmetry (Figure 7). In the human gTuRC, the gTuNA CM1 helix is bound at the GCP2:GCP6 interface, and the distance between the GCPs at this interface closely matches that observed in our gTuRC structures, indicating a conserved and more optimal spacing upon CM1 binding (Figure 7-figure supplements 1 and 2B). This suggests that the mode of interaction of the Spc110p CM1 helix with GCPs is broadly conserved. The fact that both structures have low-resolution density extending past the CM1 N-terminus towards the adjacent GCP suggests that there may also be a conserved functional role for the residues N-terminal to CM1 (Figure 7-figure supplement 1BC).
Simple addition of a CDK5RAP2 homologue during purification did not yield observable CM1 density in the Xenopus gTuRC complexes (Liu et al., 2020), suggesting either a lower affinity for the other sites or that other factors could be important. Combined with our additional observation of a change in the local twist and pitch of GCP:g-tubulin conformation near GCP2:GCP6, the data suggests that binding of a CM1 helix at the five GCP2:GCP3 inter-gTuSC interfaces could cooperatively rearrange the gTuRC to much better match the MT pitch and spacing, leading to activation of MT nucleation.
Recent work on MT nucleation from single purified metazoan gTuRCs has suggested that MT nucleation remains a highly cooperative process, requiring~4-7 ab-tubulin dimers (Consolati et al., 2020;Thawani et al., 2020). An optimal nucleator that perfectly matches the MT symmetry would be expected to exhibit non-cooperative behavior, as is observed in elongating MTs. In metazoans, factors that allosterically drive a more ideal template state could reduce the energetic barrier to nucleation and stimulate MT nucleation.
Consistent with this interpretation, CDK5RAP2 has been found to stimulate MT nucleation in vitro (Choi et al., 2010). This further suggests a functional role for increasing the local concentration of CM1-containing proteins through either an ordered oligomerization processes, as with Spc110p (Lyon et al., 2016), or through a more colligative phase-condensate mechanism. For example, if the additional CM1s were to come from proteins tightly bound within the PCM (such as pericentrin, centrosomin, or CDK5RAP2), the effect would be to couple gTuRC activation to PCM localization, similar to Spc110p confining yeast gTuRC function to the SPB. Despite these major advances, significant gaps remain in our understanding of how the binding of regulatory proteins and PTMs acts to modulate activation of yeast and metazoan gTuRCs. XL-MS experiments and data analysis are described in the Materials and methods section. All raw and processed data, along with complete information required to repeat the current analyses, can be found at https://proxl.yeastrc.org/proxl/p/cm1-tusc as described in the Materials and methods . # () (/ 0. 1. Figure 8. Model of g-tubulin ring complex (gTuRC) assembly and activation. g-Tubulin small complex (gTuSC) monomers bound to Spc110p display an improved binding for gTuSC due to the presence of the overhanging Spc110p CM1 binding surface. This leads to cooperative assembly of further gTuSCs to form an open gTuRC. The open gTuRC then transitions into a closed structure either prior to or concurrent with microtubule nucleation. Only six full Spc110p CM1 binding sites exist in a fully formed gTuRC, matching the hexagonal Spc42p symmetry at the spindle pole body (Bullitt et al., 1997;Drennan et al., 2019;Muller et al., 2005). section. In addition, the complete crosslinking dataset and analysis presented in this paper can be viewed, downloaded, examined, and visualized using our web-based interface, ProXL, at the URL above.

Data deposition
Integrative modeling scripts and final models and densities are available at https://salilab.org/ gtuscSpc110 and have been deposited to the Protein Data Bank archive for integrative structures (https://pdb-dev.wwpdb.org/) with depositions codes PDBDEV_00000077, PDBDEV_00000078, and PDBDEV_00000079. gTuSC purification gTuSC was prepared essentially as described (Vinh et al., 2002;Lyon et al., 2016).

Crosslinking and mass spectrometry (XL-MS)
XL-MS was carried out as described by Zelter et al., 2015. All gTuSC-Spc110p reactions were in 40 mM HEPES pH 7.0, 150 mM KCl, and contained a final concentration 0.4 mM gTuSC and 0.8 mM Spc110. DSS reactions were carried out at room temperature (RT) for 3 min using 0.44 mM DSS prior to quenching with 100 mM ammonium bicarbonate. EDC reactions were carried out at RT for 30 min using 5.4 mM EDC plus 2.7 mM Sulfo-NHS prior to quenching with 100 mM ammonium bicarbonate plus 20 mM 2-mercaptoethanol. After quenching, reactions were reduced for 30 min at 37˚C with 10 mM dithiothreitol (DTT) and alkylated for 30 min at RT with 15 mM iodoacetamide. Trypsin digestion was performed at 37˚C for 4 or 6 hr with shaking at a substrate to enzyme ratio of 17:1 or 30:1 for EDC and DSS reactions, respectively, prior to acidification with 5 M HCl. Digested samples were stored at À80˚C until analysis. Mass spectrometry and data analysis were performed as described . In brief, 0.25 mg of sample was loaded onto a fused-silica capillary tip column (75 mm i.d.) packed with 30 cm of Reprosil-Pur C18-AQ (3 mm bead diameter, Dr. Maisch) and eluted at 0.25 mL/min using an acetonitrile gradient. Mass spectrometry was performed on a QExactive HF (Thermo Fisher Scientific) in a data-dependent mode and spectra converted to mzML using msconvert from ProteoWizard (Chambers et al., 2012).
Proteins present in the sample were identified using Comet (Eng et al., 2013). Crosslinked peptides were identified within those proteins using Kojak version 1.4.1 or 1.4.3  available at http://www.kojak-ms.org. Percolator version 2.08 (Käll et al., 2007) was used to assign a statistically meaningful q value to each peptide spectrum match (PSM) through analysis of the target and decoy PSM distributions. Target databases consisted of all proteins identified in the sample analyzed. Decoy databases consisted of the corresponding set of reversed protein sequences. Data were filtered to show hits to the target proteins that had a Percolator assigned peptide level q value 0.01 and a minimum of two PSMs. The complete list of all PSMs and their Percolator assigned q values is available on the ProXL web application (Riffle et al., 2016) at https://proxl. yeastrc.org/proxl/p/cm1-tusc along with the raw MS spectra and search parameters used.
Xrcc4-Spc110 164-207 purification and X-ray crystallography DNA encoding residues 2-132 of Homo sapiens Xrcc4 (UniProt ID Q13426) fused in frame with residues 164-207 of Spc110p were synthesized by GeneArt (Thermo Fisher Scientific) and cloned into pET28a expression vector with N-terminal 6His tag, 3C protease cleavage site, and six-residue linker with sequence GSGGSG. Xrcc4-Spc110 164-207 was expressed in Escherichia coli BL21-CodonPlus-RIL (Agilent). Cells were harvested by centrifugation, then resuspended in lysis buffer (50 mM potassium phosphate pH 8, 300 mM NaCl, 5 mM EDTA, 1 mM DTT, 0.3% Tween-20, 1x cOmplete protease inhibitor, EDTA-free [Roche]). Cells were lysed by Emulsiflex C3 (Avestin). Lysate was cleared by ultracentrifugation at 40,000 x g for 30 min in a Type 45Ti rotor (Beckman-Coulter). Xrcc4-Spc110 164-207 was then purified by NiNTA affinity chromatography followed by addition of 3C protease overnight at 4˚C to cleave the 6His tag. Xrcc4-Spc110 164-207 was further purified by size exclusion chromatography (Superdex 75; GE Healthcare Life Sciences), anion exchange chromatography (MonoQ; GE Healthcare Life Sciences), with a final size exclusion polishing and buffer exchange step (Superdex 75). Crystals of Xrcc4-Spc110 164-207 were obtained by hanging drop vapor diffusion with 8 mg/ mL protein and a well solution containing 13% PEG3350 and 0.2 M magnesium formate. Crystals were cryo-protected by rapid transfer to well solution with 30% PEG3350. Diffraction data was collected under cryogenic conditions at Advanced Light Source beamline 8.3.1. Diffraction data was processed with XDS (Kabsch, 2010) and indexed in space group P1. Phases were obtained by molecular replacement using Phaser within the Phenix package (Adams et al., 2010;McCoy et al., 2007). The search model was the PDB ID 1FU1 residues 1-150, with the coiled-coil residues 133-150 mutated to alanine. The S-(dimethylarsenic)cysteine at position 130 in 1FU1 was modified to cysteine. The majority of the structure was built with phenix.autobuild (Terwilliger et al., 2008) with the remainder built manually in Coot (Emsley et al., 2010) and refined with phenix.refine (Afonine et al., 2012). The final structure contains Spc110 residues 164-203, along with the Xrcc4 fusion domain.
The buffer used during purification was modified to contain 40 mM HEPES pH 7.5, 100 mM KCl, 1 mM EGTA, 2 mM MgCl2, 10% glycerol, and 1 mM DTT. Samples were concentrated and buffer exchanged to obtain a final glycerol concentration of 2.5% glycerol.
Oxidation of gTuSC SS filaments was performed overnight at 4˚C by dialysis into 1 mM oxidized glutathione, removing DTT.

Grid preparation: gTuSC
Prior to grid preparation, gTuSC aliquots were centrifuged in a benchtop centrifuge (Eppendorf 5415D) at 16,000 g for 15 min and transferred to a new tube. The sample concentration was assessed on a nanodrop and diluted to a final concentration of~1 mM (O.D. at 280 nm wavelength of 0.28-0.35) such that the final buffer conditions were 40 mM HEPES pH 7.5, 2 mM MgCl 2 , 1 mM EGTA, 1 mM GDP, 100 mM KCl, and 2.5% v/v glycerol.
Data used for initial model generation and refinement had final buffer conditions of 40 mM HEPES pH 7.5, 1 mM MgCl 2 , 1 mM EGTA, and 100 mM KCl.
C-flat 1.2-1.3 4C grids were used for sample freezing and glow discharged for~30 s at À20 mA immediately prior to plunge-freezing. Grids were frozen on a Vitrobot Mark II or Mark IV, with the humidity set to 100%, and using Whatman 1 55-mm filter papers.
Grid preparation: gTuRC filaments Quantifoil 1.2-1.3 400-mesh grids were used for sample freezing and glow discharged for~30 s at À20 mA immediately prior to plunge-freezing. Grids were frozen on a Vitrobot Mark IV, with the humidity set to 100%, and using Whatman 1 55-mm filter papers.
Electron microscopy: gTuSC single-particle data Micrographs used in gTuSC initial model generation were collected using an FEI Tecnai F20 operated at 200 kV at a nominal magnification of 29,000Â (40,322Â at the detector). The data was collected with a 20 mm C2 aperture and a 100 mm objective aperture with a target underfocus of~1-2.5 mm. UCSF Image4  was used to operate the microscope. Dose-fractionated micrographs were collected on a Gatan K2 Summit camera in super-resolution mode at a dose rate of~8.5-9.5 electrons per physical pixel per second for 12 s, with the dose fractionated into 40 frames.
Micrographs included in the final model were collected using an FEI Tecnai Polara operated at 300 kV at a nominal magnification of 31,000Â (39,891Â at the detector). Data was collected with a 30 mm C2 aperture and a 100 mm objective aperture inserted with a target underfocus of~1-3 mm. Leginon (Suloway et al., 2005) or SerialEM (Mastronarde, 2005) were used to operate the microscope. Dose-fractionated micrographs were collected on a Gatan K2 Summit camera in super-resolution mode at a dose rate of approximately six electrons per physical pixel per second for 20 s, with the dose fractionated into 100 frames.
Electron microscopy: gTuRC WT filament data Data was collected in two sessions on a Titan Krios operated at 300 kV at a nominal magnification of 22,500Â (47,214Â at the detector). The data was collected with a 70 mm C2 aperture and a 100 mm objective aperture with a target underfocus of~0.9-2.0 mm. Dose-fractionated micrographs were collected on a Gatan K2 Summit camera in super-resolution mode at a dose rate of six electrons per physical pixel per second for 15 s, with the dose fractionated into 75 frames.
Electron microscopy: gTuRC SS filament data Data was collected in two sessions on a Titan Krios operated at 300 kV at a nominal magnification of 22,500Â (47,214Â at the detector). The data was collected with a 70 mm C2 aperture and a 100 mm objective aperture with a target underfocus of~0.6-2.0 mm. Dose-fractionated micrographs were collected on a Gatan K2 Summit camera in super-resolution mode at a dose rate of 6.7 electrons per physical pixel per second for 12 s, with the dose fractionated into 120 frames.
Image processing: gTuSC single-particle initial model generation Dose-fractionated image stacks were corrected for drift and beam-induced motion as well as binned twofold from the super-resolution images using MotionCorr (Li et al., 2013). CTF estimation was performed using CTFFIND4 (Rohou and Grigorieff, 2015). Particle coordinates were semi-automatically picked from filtered and binned images using the e2boxer 'swarm' tool (Tang et al., 2007). Particles were extracted using Relion (Scheres, 2012) with a box size of 384 physical pixels resampled to 96 pixels for initial processing. A dataset of~50,000,000 particles from 217 micrographs was used to generate 300 2D classes using Relion 1.3. 23 classes were selected and used in the generation of a gTuSC monomer initial model using the e2initialmodel.py function in EMAN2. This model was then used as a reference in Relion 1.3 for 3D classification into four classes of a 115,701 particle dataset from 507 micrographs with a 384 pixel box. Particles from the best gTuSC monomer class were then used for further processing and classification into four classes in FREALIGN (Grigorieff, 2016). The best class, with a resolution of~9 Å , was then used as a 3D reference for processing of the Polara data.
Image processing: gTuSC single-particle Polara data Images were drift-corrected and dose-weighted using MotionCor2 (Zheng et al., 2017). Initial processing to generate monomer and dimer reconstructions was performed with CTFFIND4 (Rohou and Grigorieff, 2015), Relion (Scheres, 2012), and FREALIGN (Grigorieff, 2016). Processing leading to the final reconstructions was performed in cisTEM (Grant et al., 2018). Particles were automatically picked from 7381 images in cisTEM, yielding 3,210,917 initial particle coordinates. 2D classification was performed to eliminate junk and ice particles, with 1,187,292 particles being included in the initial 3D classification. During 3D classification, particles were extracted from unbinned superresolution micrographs with a box size of 376.02 Å (600 pixels).
Classification and alignment were performed using the cisTEM 'Manual Refine' tool, as delineated in Figure S13.

Image processing: gTuRC WT filaments
Images were drift-corrected, dose-weighted, and binned twofold using MotionCor2 (Zheng et al., 2017). Filaments were manually picked using e2helixboxer (Tang et al., 2007) from 2204 micrographs. Filaments were extracted in Relion2 (Kimanius et al., 2016) and boxed approximately every three asymmetric units, using a rise of 21 Å with a box size of 635.4 Å (600 pixels on the micrographs, rescaled to 448 pixels), yielding 28,753 boxed filament images. 2D classification was performed to eliminate junk particles and filament ends, with 28,648 filament images remaining after culling. These images were initially aligned in Relion2 (Kimanius et al., 2016, p. 2), while allowing for the refinement of helical parameters. Particle alignments were exported into FREALIGN (Grigorieff, 2016) for additional helical refinement. FREALIGN alignments were used for helical symmetry expansion as implemented in Relion2. Symmetry-expanded alignment parameters were then imported into cisTEM (Grant et al., 2018) for local alignment and classification. A user-generated mask was supplied for these refinements, with the final mask containing approximately three gTuSC subunits with a total molecular weight of approximately 900 kDa. Prior to classification, the defocus was refined in cisTEM. Focused classification was performed in cisTEM, as delineated in Figure 3figure supplement 1.

Image processing: gTuRC SS filaments
Images were drift-corrected, dose-weighted, and binned twofold using MotionCor2 (Zheng et al., 2017). Filaments were manually picked using e2helixboxer from 3024 micrographs. Filaments were extracted in Relion2 (Kimanius et al., 2016) and boxed approximately every three asymmetric units, using a rise of 21 Å with a box size of 635.4 Å (600 pixels), yielding 175,500 boxed filament images. 2D classification was performed to eliminate junk particles and filament ends, with 152,798 filament images remaining after culling. These images were initially aligned in Relion2, while allowing for the refinement of helical parameters. Particle alignments were exported into FREALIGN (Grigorieff, 2016) for additional helical refinement. FREALIGN alignments were used for helical symmetry expansion as implemented in Relion2. Symmetry-expanded alignment parameters were then imported into cisTEM (Grant et al., 2018) for local alignment and classification. A user-generated mask was supplied for these refinements, with the final mask containing approximately three gTuSC subunits with a total molecular weight of approximately 900 kDa. Prior to classification, the defocus was refined in cisTEM. Focused classification was performed in cisTEM, as delineated in Figure 3figure supplement 2.

Difference map generation
The gTuRC SS reconstruction was resampled to 400 pixels using resample.exe included in the cisTEM package. A molecular map of a trimer of gTuSCs from the gTuRC SS model, but not including Spc110p, was generated in Chimera using the molmap command with a resolution of 3.3 Å . A difference map was generated using the diffmap.exe software obtained from the Grigorieff lab website (https://grigoriefflab.umassmed.edu/diffmap). The difference map was sharpened with a b-factor of À40 Å 2 and filtered to 5.5 Å with a five-pixel fall-off using the bfactor software obtained from the Grigorieff lab website (https://grigoriefflab.umassmed.edu/bfactor).

Local resolution estimation
Local resolutions were estimated in blocres (Heymann, 2001) using a box size of 20 and a step of either 1 (gTuRC SS , gTuRC WT ) or 2 (monomer and dimer). The local resolution estimate was applied to the gTuRC SS reconstruction using SPOC in Figure 3-figure supplement 3C (Beckers and Sachse, 2020).

Initial atomic model generation: gTuSC monomers
To generate an initial atomic model, the crystal structure of human GCP4 and a previously generated pseudo-atomic model were used as templates. Prior to fitting, the GCP4 structure was threaded with the Spc97p and Spc98p sequence, and the human g-tubulin was threaded with the Tub4p sequence. These initial models were fitted into preliminary structures into segmented density using Rosetta's relax function. Missing residues were built using RosettaCM density-guided model building (DiMaio et al., 2015), with the human GCP4, g-tubulin threaded models and the pseudo-atomic model being sampled separately. Well-scoring structures were then compared to the density, assessing the quality of the fit to determine the register. In cases where the register was poorly fit and the correct register was clear, the register was manually adjusted to fit map details. Certain regions were built using the RosettaES algorithm (Frenz et al., 2017). This procedure was iterated, with occasional manual modification of the structure in Coot (Emsley et al., 2010).
As a final step, final half-maps were used in the refinement, with the best preliminary models relaxed and refined through iterative backbone rebuilding (Wang et al., 2016) into one half-map reconstruction, and iteratively refined using Rosetta. This model was used as a starting point for atomic model building into the higher resolution gTuSC SS filament structure.

Atomic model generation: gTuRC SS
The initial model from monomer fitting was relaxed into the gTuRC SS structure using Rosetta's relax function and refined using iterative backbone rebuilding as previously described. Poorly fitting and missing regions were either built in Coot or using the RosettaES algorithm. Residues Spc110p 112-150 were manually built in Coot. Finally, the models were iteratively refined using a procedure that involved using Rosetta to relax the models into one half-map and iterative backbone rebuilding, with the best models as assessed using the FSC to the second half-map being combined using the phenix combine_models function, followed by Phenix (Adams et al., 2010) real-space refinement (Afonine et al., 2018) and manual modification. This model was used as the basis for the single-particle monomer model, and the gTuRC WT models. Models were further iteratively refined using Rosetta, Coot, and Phenix. Finally, the Spc110p 164-208 crystal structure was relaxed into gTuRC SS density, with the residues 151-164 built manually in Coot. Spc110p was iteratively relaxed into density using Rosetta to relax the models into one half-map and iterative backbone rebuilding, with the best models being visually inspected and manually modified in Coot. A final round of manual refinement of Spc110p was performed in ISOLDE (Croll, 2018) using a density-modified map generated in Phenix (Terwilliger et al., 2020).

Atomic model generation: gTuRC WT open state
The initial model from gTuSC SS fitting was relaxed into the gTuSC WT structure using Rosetta's relax function and refined using iterative backbone rebuilding, with the best models as assessed using the FSC to the second half-map being combined using the phenix combine_models function. Models were further iteratively refined using Rosetta, Coot, and Phenix.

Atomic model generation: gTuRC WT closed state
The initial model from gTuSC SS fitting was relaxed into the closed gTuSC WT structure using Rosetta's relax function and refined using iterative backbone rebuilding, with the best models as assessed using the FSC to the second half-map being combined using the phenix combine_models function, followed by Phenix real-space refinement and manual modification. Models were further iteratively refined using Rosetta, Coot, and Phenix.

Model generation: gTuSC monomer
The initial model from gTuSC SS fitting was relaxed into the gTuSC monomer structure using Rosetta's relax function and refined using iterative backbone rebuilding, with the best models as assessed using the FSC to the second half-map being combined using the phenix combine_models function, followed by Phenix real-space refinement and manual modification. Models were further iteratively refined using Rosetta, Coot, and Phenix. The final round of Phenix real-space refinement was performed against the full map.

Model generation: gTuSC dimer
The dimer model was generated by using Rosetta's relax function to fit two gTuSC SS models generated as above into dimer density. The nucleotide was subsequently modified to GDP, and poorly fitting regions were deleted. This model was used solely for the segmentation shown in

Surface area calculations
Surface area calculations were performed using NACCESS (Hubbard and Thornton, 1993).
2D classification: Figure S4-figure supplement 1 Monomer and dimer stacks (384 pixel stacks used in final reconstruction generation) were separately classified using cisTEM (Grant et al., 2018). Classes showing high-resolution features were extracted for figure generation using IMOD (Kremer et al., 1996).

Wiring diagrams
Wiring diagrams were generated using the PDBSum online portal (Laskowski, 2009).

Sequence alignments for conservation surfaces
Sequences for Spc97p and Spc98p homologues from H. sapiens, Mus musculus, Danio rerio, Xenopus laevis, Drosophila melanogaster, Arabidopsis thaliana, Glycine max, Dictyostelium discoideum, and Saccharomyces pombe were aligned to the sequence from Saccharomyces cerevisiae using the MAFFT algorithm (Katoh and Standley, 2013) implemented on the MPI bioinformatics website (Zimmermann et al., 2018). The Spc110p sequence from S. cerevisiae was similarly aligned to orthologs from H. sapiens, M. musculus, D. rerio, X. laevis, D. melanogaster, D. discoideum, and S. pombe. Sequence alignments were imported using the Multalign Viewer in Chimera (Pettersen et al., 2004), which was subsequently used to color the surfaces and ribbons by conservation.

Figure generation
Structural figures were generated in UCSF Chimera (Pettersen et al., 2004) or ChimeraX (Goddard et al., 2018). FSC plots were generated in Excel from Part_FSC estimates in cisTEM (Grant et al., 2018). Map-to-model FSCs were generated in Phenix (Adams et al., 2010), with default parameters. Figures panels were compiled into figures in Affinity Designer. The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication. Additional files

Supplementary files
. Supplementary file 1. X-ray data collection and refinement statistics.
. Transparent reporting form Data availability Structure factors and model coordinates for the Xrcc4-Spc110p164-207 fusion X-ray crystal structure have been uploaded to the RCSB protein data bank with PDB ID 7M3P. Cryo-EM reconstructions and model coordinates have been deposited to the EMDB and PDB for the gTuSC monomer (EMDB ID: EMD-23638 PDB ID: 7M2Z), gTuRCSS (EMDB ID: EMD-23635 PDB ID: 7M2W), gTuRCWT open (EMDB ID: EMD-23636 PDB ID: 7M2X) and gTuRCWT closed (EMDB ID: EMD-23637 PDB ID: 7M2Y) states. The cryo-EM reconstruction for the gTuSC dimer (EMDB ID: EMD-23639) has been deposited to the EMDB. Accession codes are also available in Tables S1 and S2. XL-MS experiments and data analysis are described in the Methods section. All raw and processed data, along with complete information required to repeat the current analyses, can be found at https://proxl.yeastrc.org/proxl/ p/cm1-tusc as described in the Methods section. In addition, the complete crosslinking dataset and analysis presented in this paper can be viewed, downloaded, examined and visualized using our web-based interface, ProXL, at the URL above. Integrative modeling scripts and final models and densities are available at https://salilab.org/gtuscSpc110 and have been deposited to the Protein Data Bank archive for integrative structures (https://pdb-dev.wwpdb.org/) with depositions codes PDBDEV_00000077 PDBDEV_00000078 PDBDEV_00000079.
The following datasets were generated: