How augmin establishes the angle of the microtubule branch site

Abstract

How microtubules (MTs) are generated in the proper orientation is essential to understanding how the cytoskeleton organizes a cell and MT-dependent events such as cell division. In the spindle, most MTs are generated through the branching MT nucleation pathway. In this pathway, new MTs are nucleated from the side of existing MTs and oriented at a shallow angle by the branching factor augmin, ensuring that both MTs have the same polarity. Yet, how augmin binds MTs and sets the branch angle has remained unclear. Here, we report the cryo-electron microscopy structure of an augmin subcomplex on the MT. This structure resembles that of NDC80 bound to the MT, with the conserved CH domain of augmin’s Haus6 subunit directly proximal to the MT lattice. We find that the Haus6 CH domain is a bona fide MT binding site that increases augmin’s affinity for the MT and helps establish branch angle. A second binding site, located in the disordered N-terminus of Haus8, also establishes branch angle,. Thus, we find that augmin regulates MT branching using two domains, each tuned to modulate MT affinity and MT branch angle. This work expands our mechanistic understanding of branching MT nucleation and thus spindle formation.

Here, authors report the structure of the augmin’s CH domain bound to the microtubule lattice and show that its Haus8 subunit binds tubulin tails. Both domains are critical for microtubule binding and for microtubules to branch at shallow angles.

Introduction

Microtubules (MTs) are a primary component of the eukaryotic cytoskeleton, where they play critical roles in organellar and vesicular trafficking, cellular motility, and, during cell division, chromosome segregation. For each of these roles, new MTs must be generated on demand in real-time through MT nucleation. Moreover, the polarity of new MTs must be strictly controlled to ensure that cargos and chromosomes are correctly trafficked. MT nucleation proceeds through multiple biochemical pathways, and one of the most important during spindle assembly is branching MT nucleation^1–4. New branched daughter MTs emanate at a shallow angle, generally between 0° and 30°, and share the same polarity of the pre-existing mother MTs^5,6. Thus, on the micron scale, branching MT nucleation preserves spindle MT polarity.

Several MT-associated proteins (MAPs) have been shown to promote branching MT nucleation^6–8. However, in its simplest form, branching MT nucleation requires only two protein complexes: the universal MT nucleator γ-TuRC and the branching-specific hetero-octameric augmin complex^9–11. Augmin’s primary function in branching MT nucleation is to recruit γ-TuRC to pre-existing MTs^12,13. Moreover, augmin appears to establish the overall architecture of the MT branch site, as reconstitution of MT branching with only augmin and γ-TuRC recapitulates the shallow branch angles observed in endogenous branching^9–11,14. Yet, how augmin orients γ-TuRC and the daughter MT of the resulting branch has remained unclear.

Recent experimental models of both human and Xenopus laevis augmin, determined using single-particle cryo-electron microscopy (EM), have provided the groundwork to answer this question^14–16. These models demonstrated that augmin is a relatively rigid complex, composed of a γ-TuRC-binding stalk (T-III) and an arch (T-II) that binds the pre-existing MT^12,13,17. While augmin’s relatively rigid architecture thus made the complex seem a good candidate to enforce a defined range of angles between γ-TuRC and the pre-existing MT, one major issue remained. Prior biochemical data had implicated the N-terminus of T-II subunit Haus8, a low-complexity, highly electropositive 150-residue sequence, as the primary MT binding region (MTBR) within the complex^12,18. Yet, no electron density corresponding to this region could be found in any of the experimental datasets, presumably due to inherent flexibility^14–16. This begs the question: if augmin connects to the pre-existing MT only via a flexible, unstructured region, how can augmin possibly impose global structure on the branch site and define the range of branch angles.

To answer this question, in this work, we report the cryo-electron microscopy structure of an augmin subcomplex bound to the MT lattice. Our prior structures had found that augmin contains two CH domains, which are often MT binding regions. The structure of augmin on the MT reveals that the CH domain of Haus6 is, in fact, a bona fide MT binding region. Using structure-based biochemistry, we characterize the role of the MT binding regions and uncover how augmin regulates MT binding and branching angle, which is important for its function in the spindle during cell division.

Results

Engineering augmin T-II_bonsai

To understand how augmin interacts with the lattice of the mother MT, we needed to obtain the structure of augmin bound to the MT. Prior work from our lab and others^12,13 established that full-length augmin can be visualized sparsely decorating MTs. However, it was not possible to produce full-length augmin above the dissociation constant of 1 μM previously reported¹², making it infeasible to saturate the MT lattice with full-length augmin, as would be required for single particle cryo-EM. Thus, we explored truncations of the augmin complex, as well as subcomplexes, to identify a more tractable structural target.

Our prior structure of X. laevis augmin¹⁴ suggested that both the MTBR of Haus8, as well as the putative MT-binding CH domains of Haus6 and Haus7 were clustered at the N-terminus of tetramer T-II (Fig. 1a). Using the augmin structure as a guide, we located an internal hinge at the apex of T-II (Fig. 1a, Supplementary Fig. 1a, b). We truncated T-II at this hinge and were able to produce a shortened T-II that we termed augmin T-II_bonsai (Fig. 1a, Supplementary Fig. 1a, b). Surprisingly, Haus2 was excluded from the complex upon purification (Supplementary Fig. 1c). This is consistent with the more transient association of Haus2 with T-II that we had previously reported¹³. Nonetheless, the Haus6/Haus7/Haus8 heterotrimeric subcomplex was predicted to still form a folded, albeit remodeled, heterotrimer that, crucially, maintains the same architecture of the MT-binding regions as the full-length complex (Supplementary Fig. 1d). Furthermore, the complex was highly soluble and could be produced at a concentration of 10 μM. Using MT pelleting assays, we determined the affinity of T-II_bonsai for the MT lattice as 200 ± 80 nM (Supplementary Fig. 5a, b), in line with the previously reported affinity of the full length complex¹². Thus, we established that T-II_bonsai was a promising target for structural studies on the MT lattice.

Fig. 1. Structure of augmin T-II_bonsai on the MT.

a Above, the octameric augmin complex (PDB accession 8FCK) is biochemically composed of two functionally distinct tetramers, MT-binding T-II and γ-TuRC-binding T-III. T-II_bonsai—consisting of the N-terminal half of T-II—was generated from truncating the complex at a flexible hinge (dashed line). Below, the domain architecture of the four T-II subunits, Haus2 (cyan), Haus6 (green), Haus7 (red), and Haus8 (yellow), is shown with domain boundaries indicated. Regions of the subunits included in T-II_bonsai are shown in solid colors, and regions excluded from the T-II_bonsai construct are shown as outlines. Both the CH domains and MT binding region (MTBR) are included in T-II_bonsai. b Density reconstruction of a T-II_bonsai decorated MT (C1 reconstruction) is shown. Density is color-coded by polypeptide chain identity, where augmin subunits are color-coded as in (a), α-tubulin is light gray, and β-tubulin is dark gray. In all panels, the minus end of the MT is oriented to the left and the plus end to the right. The boxed area indicates the region covered by the symmetry-expanded map (sym) shown in (c). c Density reconstruction of a single T-II_bonsai molecule on the tubulin lattice, following symmetry expansion, is shown in top view. Density was segmented in Chimera segger and color-coded by polypeptide chain identity, where augmin subunits are color-coded as in (a), α-tubulin is light gray, and β-tubulin is dark gray. d Map-to-model fit of a single T-II_bonsai molecule on the tubulin lattice, following symmetry expansion, is shown in side view, where augmin subunits are color-coded as in (c) and electron density map is shown as a gray, partially-transparent surface. e All-atom cartoon model of T-II_bonsai on the MT is shown docked in the same orientation in side view. Two copies of α-tubulin are present within the symmetry-expanded map; however, T-II_bonsai makes contacts only within a single αβ-tubulin dimer, indicated above. The CH domain of subunit Haus6 is proximal to the MT lattice, whereas the tetramerization domains (tetramer) present in Haus6, -7, and -8 of T-II_bonsai project distally away from the lattice.

Structure of augmin T-II_bonsai on the MT

To determine the structure of augmin T-II_bonsai bound to the MT lattice, we prepared grids of GMPCPP-stabilized MTs, then flowed in the augmin T-II_bonsai construct before grid blotting and freezing. After screening for coated MTs (Supplementary Fig. 6), data collection (Supplementary Table 1), and data processing (Supplementary Fig. 6), we obtained a reconstruction of the 14-protofilament GMPCPP-stabilized MT decorated with density corresponding to T-II_bonsai at 3.8 Å resolution (Fig. 1b, Supplementary Fig. 6). Symmetry expansion followed by further refinement led to a reconstruction of a single tubulin dimer bound to a single copy of T-II_bonsai at 3.1 Å resolution (Fig. 1c). The highest resolution region of the map corresponded to the MT lattice (Supplementary Fig. 7a, b), however, the resolution of the T-II_bonsai density, ranging from 3 to 5 Å, was sufficient to interpret the structure of T-II_bonsai bound to the MT.

Although in the full augmin structure, T-II_bonsai forms a rigid unit, in our structure, only a portion of the T-II_bonsai is visible. This portion consists of a single CH domain, closest to the MT lattice, and an adjacent stalk region, composed of three α-helices, that points away from the lattice (Fig. 1d). This stalk is the precise length of the α-helical trimer at the N-terminus of T-II prior to incorporation of the Haus2 α-helix into a tetrameric bundle (Supplementary Fig. 8a). Thus, we may observe ordered density for only a portion of the complex because, in the absence of Haus2, the remainder fails to fold. Rigid-body docking of the N-terminus of T-II into the density was most consistent with Haus6 contributing the ordered CH domain, rather than Haus7, as this allowed for correct positioning of the α-helical stalk (Supplementary Fig. 8a, b). Furthermore, docking of the predicted structure of the Haus6 CH domain appropriately positions many bulky aromatic residues within the density (Supplementary Fig. 8c, d). These bulky residues are generally not conserved between Haus6 and Haus7 (Supplementary Fig. 8e), so this provides further agreement with our assignment of the Haus6 CH domain as closest to the MT lattice.

Our structure shows a modest binding surface between augmin and the MT lattice (Fig. 2a), with a buried surface area of 664 Ų. A little under half of the binding surface (309 Ų) contacts α-tubulin, mediated by αB’ and αE loop of Haus6 (Fig. 2a, Supplementary Fig. 9a). In addition, the short αF helix of Haus6, although down weighted by the deep-learning driven sharpening algorithm deepEMhancer, is visible in close proximity to β-tubulin in traditionally, globally-sharpened map (Supplementary Fig. 9b, c), and contributes an additional 355 Ų of buried surface area.

Fig. 2. MT binding interactions of T-II_bonsai.

a Top and side views of T-II_bonsai bound to the MT lattice. All ordered components are shown as atomic cartoons, where α- and β-tubulin are shown in light and dark gray, respectively. Augmin is colored by subunit: Haus6 (green), Haus7 (red), and Haus8 (yellow). Disordered regions of augmin (MTBR) and αβ-tubulin (α- and β-tails) are schematized with dashed lines. Tubulin-interacting helices of augmin are labeled. b Bottom view of T-II_bonsai to visualize the MT-binding interface of the complex. At center, the complex is shown as a cartoon and colored as in (a). At left, the complex is shown as a surface and colored by relative side-chain conservation, where white indicates non-conservation and purple indicates conservation. Organisms used for sequence alignment were Xenopus laevis, Homo sapiens, Takifugu rubripes, Ciona intestinalis, and Drosophila melanogaster, identical to the partial sequence alignment shown in Supplementary Fig. 5e. At right, the complex is shown as a surface colored by relative Coulombic potential, where red is electronegative and blue is electropositive. c MT localization of wildtype (WT) augmin octamer (top row) and mutants (bottom three rows) after 10 min of binding. GMPCPP MT seeds are shown in magenta on left and GFP-tagged augmin constructs are shown in green at right. Scale bars indicate 5 μm. d Quantification of MT localization of wildtype augmin and mutants as measured by average normalized augmin fluorescence intensity. MT localization was assayed using n = 3 independent experiments for each augmin construct over n = 214 MTs for wildtype augmin, n = 183 MTs for Haus8 MTBR mutant, n = 95 MTs for Haus6 CH domain deletion, and n = 253 MTs for Haus7 CH domain deletion. Data is shown as a box and whisker plot where the central line represents the median, the box encloses values between the 25th and 75th percentile, and whiskers correspond to points between the 5th and 95th percentile. Significance was determined by two-tailed unpaired t-test. **** corresponds to p < 0.0001. p values were 6.5 × 10⁻⁴² for WT vs. Haus8∆MTBR, 8.0 × 10⁻⁶³ for WT vs. Haus6∆CH, and 0.27 for WT vs. Haus7∆CH. Source data are provided as a Source Data File.

In addition to the CH domain of Haus6, augmin has been proposed to contain two other MT binding sites: the CH domain of Haus7¹⁶ and the MTBR of Haus8^12,18 As discussed above, the CH domain of Haus7 is not visible in our structure; however, its position can be inferred by docking larger fragments of augmin into our map (Supplementary Fig. 8a). Thus, our structure implies that the Haus7 CH domain may not directly contact the MT lattice. Surprisingly, we observe no additional density in our map that could correspond to the Haus8 N-terminal MTBR (Supplementary Fig. 10a, b). As the highly electronegative tails of α- and β-tubulin are predicted to be adjacent to the location to the Haus8 MTBR (Fig. 2a), we speculate that Haus8’s N-terminal electropositive MTBR may contribute to electrostatic attraction to the tubulin tails, thus enhancing augmin’s MT binding affinity without displaying appreciable ordered binding. Previous studies have shown that augmin shows substantial affinity for the tubulin tails¹⁹, and similar phenomena have been previously reported for regions of many electropositive MAPs, such as tau²⁰ and NDC80^21,22.

Roles of the bipartite binding interface in MT binding affinity

From our structure, the Haus6 CH domain appears to be well-positioned to interact with the tubulin lattice (Fig. 2a). Moreover, the surface adjacent to the tubulin dimer is both conserved and electropositive (Fig. 2b). Thus, we hypothesized that the Haus6 CH domain could contribute to augmin’s MT binding affinity. We tested this hypothesis by assessing which regions of augmin are necessary for MT binding. We generated augmin mutants that were missing the Haus8 MTBR or the Haus6 CH domain, as well as a mutant missing the Haus7 CH domain. Next, we visualized the binding of augmin mutants to immobilized, fluorescently labeled GMPCPP MT seeds via total internal reflection (TIRF) microscopy. As expected, wildtype augmin displayed robust binding to the MT seeds (Fig. 2c). Previous work using human augmin suggested that deletion of the Haus8 MTBR was sufficient to abrogate MT binding of the entire complex¹². However, when we deleted the Haus8 MTBR from X. laevis augmin, we found that MT binding was weakened but not abolished (Fig. 2c, d). Similarly, purified augmin lacking the Haus6 CH domain displayed reduced but detectable MT binding (Fig. 2c, d). In fact, MT binding of Haus6∆CH augmin may be more compromised than MT binding of Haus8∆MTBR augmin (Fig. 2d). In contrast, deletion of the Haus7 CH domain had no impact on augmin’s ability to bind MTs (Fig. 2c, d). Thus, both the Haus6 CH domain and the Haus8 MTBR, but not the Haus7 CH domain, contribute to MT binding of X. laevis augmin, and the CH domain may be more important for the MT affinity than the canonical MTBR. This result is consistent with prior observations of the NDC80 complex, where both the disordered MTBR and the CH domain of the Hec1/Ndc80 subunit contribute to MT binding^23,24.

Previously, we had identified a distant structural similarity between augmin’s T-II and the kinetochore-localized NDC80 complex^14,16. As this similarity centered on the presence of MT-binding CH domains, we hypothesized that augmin might recognize the MT in a similar way to other CH-domain containing proteins. Now, with the structure of augmin on the MT in hand, we could directly evaluate any similarities between augmin and other MT-bound CH domains. The first CH-domain found to bind to the MT derives from the plus-end tracking end-binding proteins (EBs)^25–27. However, EB proteins and augmin recognize the MT in highly divergent manners. EB proteins bind at the interface of four αβ-tubulin dimers^27,28, whereas augmin binds to a single αβ-tubulin dimer (Supplementary Fig. 11a). In contrast, the CH domain of NDC80’s Ndc80/Hec1 subunit bound to the MT superimposes almost precisely upon the CH domain of Haus6 bound to the MT (Supplementary Fig. 11b), further underscoring the similarity between these two complexes.

Since we do not observe appreciable Haus8 MTBR density in our cryo-EM maps, we wanted to further investigate its role in MT binding in vitro. Since there is evidence that augmin interacts with the C-terminal tails of tubulin, we speculated that Haus8 MTBR interacts with the tails since interactions with the tubulin tails non-specific and low affinity which would be in line with the reduction of MT binding when the Haus8 MTBR is deleted. To test this, we cleaved the C-terminal tails of GMPCPP MT seeds using subtilisin, and then used in vitro MT binding assays to test the ability of the cleaved MTs to bind a purified fragment of the Haus8 MTBR (i.e., Haus8(1-150)), wildtype augmin octamer, as well as the augmin octamer without the N-terminus of Haus8. This data is summarized in Supplementary Fig. 9. Firstly, we observe that even at 50 nM concentration, the decoration of Haus8(1-150) on the MT lattice is completely abrogated when the C-terminal tails are cleaved (Supplementary Fig. 12a). Moreover, when we test the wildtype augmin octamer, we observe a significant reduction in MT binding upon cleavage of the tails (Supplementary Fig. 12b). In this case, some binding is maintained due to the presence of the Haus6 MT binding interactions in this construct. Finally, we hypothesized if Haus8 MTBR is the primary region of augmin responsible for interacting with the tubulin tails, then deleting the Haus8 MTBR should show no preference between the GMPCPP seeds and those cleaved with subtilisin. Indeed, when we tested the augmin octamer missing the Haus8 MTBR, we observe little difference between its ability to bind cleaved seeds (Supplementary Fig. 12c). This again underscores a similarity to Ndc80 where augmin binds to the MT through both its CH domain in Haus6 as well as through Haus8 MTBR which interacts with the C-terminal tubulin tails.

MT lattice preference of branching factors

As described above, T-II and NDC80 have marked structural and biochemical similarities in how the complexes bind to MTs. Yet, one major difference between the two complexes is their localization on the MT. The NDC80 complex, in its role of capturing MT plus ends at the kinetochore, prefers and tracks the lattice at the growing MT plus end²³. In contrast, augmin tends to localize along the entire MT and enable branching MT nucleation uniformly along its length. Moreover, augmin has not been reported to have any plus-end tracking activity, either in vitro, in X. laevis egg extract, or in vivo. Therefore, we investigated whether augmin’s localization could be explained by a preference for a specific nucleotide state of the MT lattice.

To directly investigate augmin’s lattice preference, we employed two variations on a dual-lattice TIRF microscopy binding assay, inspired by Mauer et al. and Zhang et al. ^28,29. In the first variation, we initially nucleated Atto 647N-labeled GMPCPP MT seeds, then performed another round of polymerization with Alexa568 tubulin and GTPγS. The resulting MTs had regions that mimic the GTP MT lattice (GMPCPP), at the plus end, and the GDP + Pi transition state (GTPγS), further from the plus end, each labeled with a distinct fluorophore. As a proof of concept, we incubated seeds with EB1 and found, as previously reported, that the protein preferred the GTPγS lattice over GMPCPP (Fig. 3a, b, Supplementary Fig. 13)^26,29. The branching factor TPX2 has been previously reported to prefer GMPCPP lattice³⁰, and we also recapitulated this preference for the GMPCPP lattice (Fig. 3a, b, Supplementary Fig. 13). Having shown that proteins can preferentially localize to both lattice types in our system, we next assayed augmin’s lattice preference. We found that augmin, like EB1, strongly prefers the GTPγS lattice over GMPCPP (Fig. 3a, b), suggesting that, consistent with prior observations, augmin does not preferentially bind lattice found at the plus-end^11,31. In addition, augmin lacking the MTBR of Haus8 maintains its preference for GTPγS lattice over GMPCPP, suggesting that lattice preference is encoded in the Haus6 CH domain (Fig. 3a, b).

Fig. 3. MT nucleotide preferences of augmin and TPX2.

a Representative cropped fields of view of MAPs—80 nM EB1-GFP (green), 2 nM BFP-TPX2 (white), 100 nM wildtype augmin octamer (green), and 100 nM Haus8∆MTBR augmin octamer (green) —bound to dual-color, dual-nucleotide MT seeds, where GMPCPP lattice is magenta and GTPγS cyan. Scale bars are 10 μm. b Quantification of MT localization of MAPs from (a). Mean MAP fluorescence intensity is plotted segmented according to the GMPCPP (magenta) or GTPγS (cyan) signal. The number of MTs averaged were: EB1 on GMPCPP (188) and GTPγS (169), TPX2 on GMPCPP (273) and GTPγS (308), wildtype (WT) augmin octamer on GMPCPP (162) and GTPγS (217), Haus8∆MTBR augmin octamer on GMPCPP (208) and GTPγS (128). Data is shown as a box and whisker plot where the central line represents the median, the box encloses values between the 25th and 75th percentile, and whiskers correspond to points between the 5th and 95th percentile. Unpaired two-tailed t-tests were performed. **** indicates p < 0.0001. p values were 0 for EB1, 0 for TPX2, 5 × 10⁻³⁰¹ for augmin wildtype, and 4 × 10⁻²⁵⁸ for augmin Haus8∆MTBR. Source data for (b, d) are provided as a Source Data File. c Representative cropped fields of view of various MAPs as in (a) bound to a mixture of dual-color, dual-nucleotide MT seeds, where GTPγS lattice is in cyan and GDP in red. Scale bars are 10 μm. d Quantification of MT localization of MAPs from (c). Mean MAP fluorescence intensity is plotted segmented according to the GTPγS (cyan) or GDP (red) signal. The number of MTs averaged were: EB1 on GTPγS (251) and GDP (59), TPX2 on GTPγS (472) and GDP (101), wildtype augmin octamer on GTPγS (103) and GDP (19). Data is shown as a box and whisker plot where the central line represents the median, the box encloses values between the 25th and 75th percentile, and whiskers correspond to points between the 5th and 95th percentile. Unpaired two-tailed t-tests were performed. **** indicates p < 0.0001. p values were 0 for EB1, 4 × 10⁻³⁰⁸ for TPX2, and 5 × 10⁻⁵⁴ for augmin wildtype.

Prior experiments using dynamic MTs found preferential accumulation of augmin and other branching factors on older lattice. This could be due to slow kinetics of augmin accumulation or, alternatively, a preference for GDP lattice. Thus, we next asked whether augmin displayed any preference for GDP lattice over the other types. To address this question, we modified our lattice preference conditions by mixing the above-described GMPCPP-GTPγS MTs with orthogonally-colored dynamic MTs—presumed to be fully converted to GDP lattice—capped by GMPCPP for stability. Using this modified set-up, we find that, as previously reported, EB1 preferentially binds the GTPγS lattice over the GDP lattice (Fig. 3c, d). Under our assay conditions, TPX2 displays a minor but significant preference for GTPγS over GDP. In addition, we observe that augmin displays a minor but statistically significant preferential localization to the GTPγS over GDP lattice (Fig. 3c, d). Taking all this data together, we find that augmin’s primary preference is for more native, compacted lattice instead of the expanded GMPCPP lattice. Additionally, augmin displays a slight preference for GTPγS lattice over GDP lattice, although this preference is less pronounced than the preference for GTPγS over GMPCPP.

To confirm that lattice mismatch did not cause distortions to the structure of augmin on the MT, we solved a moderate-resolution structure of T-II_bonsai bound to the GTPγS lattice (Supplementary Figs. 14–16, Supplementary Table 1). At this resolution, approximately 8 Å around augmin (Supplementary Figs. 12d, 16a), we can resolve secondary structure corresponding to T-II_bonsai. We find that Haus6’s CH domain directly contacts the GTPγS lattice in a similar manner to the GMPCPP lattice (Supplementary Fig. 16b). Compaction of the GTPγS lattice by comparison to GMPCPP, which comprises both interdimer and intradimer compaction as previously reported²⁸, causes a slight reorientation of the Haus6 CH domain. This reorientation, imposed by a minor reorientation of the α11’ helix of α-tubulin, may underlie augmin’s preference for the more biological compacted lattice state, although at the moderate resolution of the GTPγS structure, this cannot be proved conclusively.

In vitro branching reconstitution provides functional implications for the bipartite MT binding region of augmin

We and others have previously shown that branching MT nucleation can be reconstituted in vitro using only augmin and γ-TuRC, while still recapitulating the conserved polarity and shallow branch angles observed in X. laevis meiotic egg extract^9–11,14. Therefore, we hypothesized that the MT binding sites of augmin contribute to setting the branch angle. To test this, we performed a series of in vitro branching assays from stabilized MT seeds using recombinant augmin. We expressed and purified wild type augmin as well as octameric augmin constructs lacking either of its MT binding sites, i.e., lacking either the CH domain of Haus6 (Haus6∆CH) or the N-terminus of Haus8 (Haus8∆MTBR). To increase the dwell time of augmin on the MT seeds for long enough to observe branching events, we first bound BFP-tagged full-length TPX2 at a concentration below its phase boundary³². As discussed above, TPX2 strongly binds to the GMPCPP MT lattice. In addition, it is resistant to washout and helps retain augmin on the lattice (Supplementary Fig. 17) long enough to observe branching events for all constructs tested, even the mutants with reduced MT binding. Once augmin was bound to MTs, we incubated the MT seeds with γ-TuRC purified from X. laevis egg extract. Finally, fluorescently labeled soluble tubulin and GTP were added to initiate branching from the MT seed (Fig. 4, Supplementary Fig. 18, Supplementary Movie 1).

Fig. 4. In vitro reconstitution of branching MT nucleation reveals that both of augmin’s MT binding domains control the branch angle.

a Schematic of workflow for in vitro reconstitution of branching MT nucleation from stabilized MT seeds by TIRF microscopy. Schematic created in BioRender. Kraus, J. (2025) https://BioRender.com/9k1ncvc. b Still images of the soluble tubulin channel from time-lapse movies of branching MT nucleation using wildtype (WT) augmin octamer (magenta) and Haus8∆MTBR (cyan) and Haus6∆CH (gray) mutants. All constructs can initiate branched MTs from a template. For each augmin construct, three independent live movies were collected. Scale bars are 5 μm for (b, c). c Representative still images of the soluble tubulin channel taken at later time points for each augmin construct. For each construct, 5–10 still images were collected after the last time point of the live movies to obtain more micrographs for quantification of branching MT nucleation. d Quantification of number of branches per seed over the TIRF microscopy data sets. Branch angles were manually counted using criteria described in the methods section using the independent replicates. Number of branches were 38 for the wildtype (WT) augmin, 26 for the Haus8∆MTBR mutant, and 29 for the Haus6∆CH mutant. Individual data points are shown as black points, the colored bars represent the mean, and measurement standard deviations are shown. Significance was determined by two-tailed unpaired t-test. ** corresponds to p < 0.01. p values were 0.0075 for Haus8∆MTBR vs. WT, 0.0049 for Haus6∆CH vs. WT, and 0.54 for Haus8∆MTBR vs. Haus6∆CH. Source data for (d, e) are provided as a Source Data File. e Histograms for quantification of branch angles for wildtype augmin (magenta), augmin Haus8∆MTBR mutant (cyan), and augmin Haus6∆CH mutant (gray). Branches analyzed are identical to those quantified in (d). The branch angles are binned by 10° from 0 to 180°, shown as individual bars in the plot. The percentage of MTs for a given construct contained into each branch angle category is annotated by the percentages listed on the y-axis.

As we have shown previously, augmin octamer and γ-TuRC are sufficient to reconstitute branching from MT seeds¹⁴. This minimal reaction recapitulates key features of branching, including a shallow branch angle and conservation of overall MT polarity. However, we previously observed branch angles focused around 15–30° for branching performed with augmin octamer¹⁴, despite the fact that the most abundant branch angle observed in Xenopus egg extract is 0 to 15 degrees^6,31. When we more closely examined the in vitro branching reaction, we found that the most frequent branch angle is in fact between 0 and 15°, due to the presence of previously-overlooked daughter MTs that form parallel (0–5 degrees) to the MT seed. These parallel MTs were observed both through live imaging as well as by an increase in tubulin intensity at the polymerizing MT plus end in later time points. Parallel MTs are easily distinguishable in our live movies of the wildtype augmin octamer, where kymographs of the plus end shows that a new branched MT nucleates at less than 5 degrees alongside the plus end, while the plus end is polymerizing (Supplementary Fig. 19 and Supplementary Movie 7). Additionally, when analyzing still images from different fields of view at later time points, quantification of several MTs at the plus end revealed a mean intensity approximately 2-fold higher than other branched MTs, indicating that a parallel MT existed at the plus end in addition to the one polymerized from the MT seed (Supplementary Fig. 20). This is likely due to the bundling properties of the Haus8 MTBR, as discussed below.

We next tested the effect of the augmin Haus8∆MTBR and Haus6∆CH constructs on the branching reaction. Despite their decreased affinity for the MT, both constructs are competent to support the nucleation of daughter MTs from the pre-existing seed, (Fig. 4b, Supplementary Movies 2–5). Indeed, by quantifying the average number of branches per seed, we observe an approximately 50% decrease in branches per seed for both mutants compared to the wildtype octamer (Fig. 4d). This is consistent with the approximate 50% reduction in binding to the MT when either domain is deleted (Fig. 2), suggesting that reduced MT binding is responsible for this effect.

Strikingly, when comparing the branch angles of the mutants to those of the wildtype, we observed a widening of the branch angle for both the Haus8∆MTBR and Haus6∆CH constructs (Fig. 4e). For the Haus8∆MTBR construct, the mean branch angle widened to 70°, while for the Haus6∆CH construct, the branch angle widened to 41°. Therefore, both MT binding sites significantly contribute to setting the architecture of the branch site in augmin. Additionally, branches emerged at a relatively broad distribution of angles for all three conditions, with wildtype augmin showing the smallest standard deviation in angles (24°), deletion of the Haus6 CH domain showing a similar standard deviation of 28°, and deletion of the Haus8 MTBR showing a slightly larger standard deviation of 39°.

To understand the source of the variability in branch angle, we further examined our live movies of MT branching and found that, at early timepoints, some MT branch angles are not fixed but rather fluctuate (Supplementary Fig. 21). This effect can also be observed in prior reconstitutions of branching MT nucleation^9,11 but has not been previously remarked upon. Follow-up studies will be necessary to fully characterize these fluctuations and determine whether they are confined to the minimal reconstituted system or also relevant in the context of the endogenous complement of factors, including motors and other crosslinkers. Nonetheless, our preliminary characterization finds that, at early timepoints, some individual branches sample a broad range of branch angles, similar in breadth to the range of branch angles found in our bulk sample (Fig. 4e). Some branches are less mobile than others, especially long branches and highly acute branches (with angles <5°). This last is consistent with the idea that acute branches are bundled with the mother MT (Supplementary Fig. 21).

To further characterize the role of MT bundling in determining branch angle, we investigated the ability of different augmin constructs to bundle MTs, which had been speculated upon previously¹⁸. To test this, we performed an in vitro bundling assay using TIRF microscopy, where we immobilized an Alexa568-MT seed on the glass coverslip and flowed in a solution of augmin and Cy5-MT seeds. Co-localization of augmin with both fluorescently-labeled seeds would indicate augmin is able to bundle the two MTs together, similar to a 0 degree branch angle. We performed this assay with purified Haus8(1–150), wildtype augmin, and augmin octamer Haus8∆MTBR construct. In our data, we observe that wildtype augmin can clearly bundle MTs together, and this happens through the MTBR of Haus8 because, when it is deleted, augmin loses its bundling ability (Supplementary Fig. 22).

Discussion

In this work, we have reported the cryo-EM structure of an augmin subcomplex, T-II_bonsai, bound to the MT lattice. Surprisingly, this structure displayed no ordered density for the known MT binding N-terminal region of Haus8. Instead, it strongly resembled the structure of NDC80 bound to the MT and revealed a second MT binding site composed of the CH domain of Haus6 in complex with an αβ-tubulin dimer. This interaction between CH domain and MT is critical for augmin to bind the MT with high affinity, as is, to a lesser extent, the Haus8 MTBR. We further found that the augmin complex displays a strong preference for GTPγS-bound MTs over GMPCPP-bound MTs. This spatial preference is the opposite of the strong GMPCPP-preference of the branching factor TPX2. Despite this opposing spatial preference on endogenous-like MTs, we could reconstitute MT branching in vitro using both TPX2 and augmin, along with the MT nucleator γ-TuRC. This allowed us to probe the function of augmin’s two MT binding sites during MT branching. We concluded that both the ordered binding of the Haus6 CH domain, as well as the disordered binding of the Haus8 MTBR, play critical roles in establishing branching geometry and thus in regulating spindle polarity.

We were only able to obtain the structure of a relatively small augmin subcomplex on the MT, as the full-length complex proved a poor structural target. However, by combining our prior knowledge of the structure of the full augmin octamer with the structure of T-II_bonsai on the MT, we can arrive at a model of full-length augmin bound to the MT (Fig. 5a). From this model, we observe that the tip of the augmin complex projects approximately 40 nm from the surface of the MT lattice. The disordered Haus8 MTBR, comprising 154 residues, would, at its full extension, be able to extend 54 nm from the MT lattice surface, although AlphaFold3 models (Supplementary Fig. 3) suggest that it will most commonly sample a region extending 19 nm. Depending where on augmin γ-TuRC binds^12,13,17,33, augmin thus may be able to capture both soluble γ-TuRC and growing MT branches at a distance of more than a MT diameter, aiding the complex in its role as a γ-TuRC tether. By modeling multiple augmin complexes bound to the MT, we also note that the MT would quickly become surrounded by a meshwork of augmin complexes. These complexes appear spaced together too tightly to allow a large molecule such as γ-TuRC—or another MT—to reach down to the MT lattice. This suggests that there may be a tension at the branch site between, on the one hand, recruiting enough augmin to retain weakly-binding soluble γ-TuRC versus, on the other, leaving enough physical space to accommodate other complexes required for branching.

Fig. 5. Model of augmin bound to the MT at the branch site.

a Two views of a 14-protofilament GMPCPP MT coated with full-length augmin (model derived from fitting full-length augmin into the GMPCPP C1 map) with three full-length X. laevis augmin complexes modeled bound to adjacent protofilaments (derived from PDB accession code 8FCK). At left, the MT is shown in profile as in Fig. 1b and, at right, the MT is shown looking down the MT axis from the plus-end. Tubulin subunits are colored light gray for α-tubulin and dark gray for β-tubulin, and the plus and minus ends of the MT are indicated. Augmin is colored by subunit as in Fig. 1a. The distance between the tip of the augmin complex and the MT lattice is measured as 40 nm and indicated by a black line. A black dashed line indicates the volume occupied by the Haus8 MTBR across 5 independent AlphaFold2-Multimer structural positions, extending 19 nm from the side of the MT. b Biochemical model of MT branching showing the contribution of augmin to different steps in the reaction. At left, augmin binds to the MT lattice, excluding the GTP cap, using predominantly its Haus6 CH domain (green circle), as well as interactions between the Haus8 MTBR (yellow dashed line) and tubulin tails (black lines). At right, a MT branch capped by γ-TuRC (black triangle) is formed at an acute angle, which is controlled primarily by the Haus8 MTBR and secondarily by the Haus6 CH domain. The Haus8 MTBR may also stabilize the branch through MT bundling, for example by interacting with disordered tubulin tails on both the pre-existing and branched MTs.

Prior biochemical work¹² had suggested that the N-terminal MTBR of Haus8 was insufficient to explain the affinity of augmin for the MT lattice. Our work has borne out this hypothesis, finding that the CH domain of Haus6 substantially enhances the affinity of X. laevis augmin for the MT (Fig. 5b). However, the function of the dual MT binding regions of augmin is not simply to increase augmin’s ability to bind MTs. The bipartite binding site also functions downstream in organizing the MT branch site and controlling the angle at which the new MT emerges. It is relatively straightforward to explain how the Haus6 CH domain accomplishes this. Our structure suggests that the CH domain binds in an ordered manner to the MT lattice. Thus, deletion of an ordered binding site should cause the augmin complex to become more flexibly attached, and would result in the broader distribution of branching angles that we observe (Fig. 5b). However, the role of the Haus8 MTBR is harder to explain. We observe that this region, although disordered in our structure, is critical to keep MT branch angles shallow. Similarly, divergent interphase and mitotic Haus8 MTBRs have been hypothesized to establish cell-cycle-dependent branch angles in vivo in Arabidopsis³⁴. One possible hypothesis for how Haus8 is acting comes from the ability of the Haus8 MTBR to bundle MTs (Supplementary Fig. 12)¹⁸. As discussed above, the Haus8 MTBR might bundle MTs either directly, by interacting with tubulin tails on both preexisting and nascent MTs, or else through indirect means, for example reducing electrostatic repulsion by neutralizing the tubulin tails^35,36. Regardless of the mechanism, MT bundling, acting on the nascent MT, could lead to a preference for more acute angles; thus, removal of the MT bundler would lead to wider angles, which is in fact what we observe.

While our work answers some outstanding questions about the structure and mechanism of MT branching, it also raises others. In particular, the only major component missing from our model of the branch site is now γ-TuRC. Thus, determining where on the augmin complex γ-TuRC binds will complete our structure of the MT branch. Recent work from two groups has provided clues as to augmin’s interaction with γ-TuRC via the NEDD1 adaptor protein. Recent work using A. thaliana augmin and available in preprint from the Al-Bassam group³⁷ shows that the N-terminal WD40 domain of NEDD1 binds to the top of the T-II arch. This binding location is perhaps surprising, as the NEDD1 WD40 domain will, first of all, be located too far from the MT surface when docked onto T-II to provide additional MT binding affinity as has been previously hypothesized¹¹. Secondly, as the WD40 domain is connected to the C-terminus of NEDD1, and thus the remainder of γ-TuRC via a flexible linker, this binding interaction will lead to a flexible, not fixed, linkage between augmin and γ-TuRC that seems unlikely to support branching at a fixed angle. A solution to this conundrum is suggested by a second publication from the Pfeffer and Schiebel groups describing the structure of the C-terminus of NEDD1 bound to the base of the γ-TuRC cone³⁸. Using molecular modeling, the authors predict a second, fixed-orientation interaction between the extreme C-terminus of Haus1, representing the farthest tip of T-III, and the C-terminal 4-helix bundle of NEDD1. Thus, it appears that, just as augmin uses both ordered and flexible interactions to bind the MT, it may also use both ordered and flexible interactions to capture and orient γ-TuRC. Prior work¹¹ has also suggested that multiple augmin complexes may be required at a single branch site. In addition to determining how these augmin complexes could recruit one another and their relative orientations, this also begs the question of why multiple copies of augmin are needed. We have previously observed that binding between a single augmin complex and γ-TuRC in solution is relatively weak¹³. Therefore, it is possible that the presence of multiple augmin complexes on the MT would strengthen the interaction with γ-TuRC through the effects of avidity.

Secondly, our unexpected finding that the augmin complex prefers the GTPγS lattice, over both the expanded GMPCPP lattice as well as, to a lesser extent, the GDP lattice, is consistent with augmin’s homology to the plus-tip tracker NDC80, but adds a layer of complexity to another open question, the spatiotemporal control of branching. Prior work has suggested that, on dynamic MTs, augmin does not plus track and in fact prefers older MT lattice¹¹. Therefore, although augmin may thermodynamically prefer the structure of “new” MT lattice, it appears that, in practice, binding of augmin to MTs is driven by kinetic effects to prefer the older, although structurally non-ideal, GDP lattice. In this, augmin is similar to TPX2, which, although under some conditions recognizes the GTP or transition state lattice³⁹, in general accumulates on older lattice through a slow (or biphasic) process^1,11,30. In line with this, MT branch sites preferentially form on older MT lattice^11,31. Thus, within the spindle, it appears that not only MTs themselves exist within a kinetic, not thermodynamic regime, but similarly the factors that control new MT nucleation, at least via the branching process, are themselves controlled by kinetic considerations.

In summary, our structural and biochemical work have established the basis for how augmin binds the MT and how augmin binding to the MT helps to orient new MT branches. This work thus advances our understanding of the MT branch site, which is key to generating spindle MTs and to establishing the structure of the spindle itself.

Methods

All work in this manuscript was carried out in accordance with the NIH guidelines for Biosafety in Microbiological and Biomedical Laboratories and the approved Princeton University Institutional Biosafety Committee (IBC) protocol 1082. Xenopus work was carried out in accordance with the NIH Guide for the Use and Care of Laboratory Animals and the approved Princeton University Institutional Animal Care and Use Committee (IACUC) protocol 1941. Either 3 or 4 Xenopus laevis animals (female adults, sourced from Nasco and/or Xenopus1) were used for the production of each meiotic egg extract.

Cloning

X. laevis T-II_bonsai was cloned into the Duet multisubunit expression system (Novagen) for expression in E. coli. DNA fragments of Haus6.L_1-269 and Haus8.L_1-268 codon-optimized for E. coli expression were amplified using the following oligonucleotide pairs—Haus6 (TATAGGATCCATGCAATCTGGCTCAC/TATACTCGAGTTACTGATCAATGTTGCCACGAACC) and Haus8 (TATACTCGAGTTACGGCATGATTTCTTCCAGCAG/TATACTCGAGTTAGGTGCTATCCAGAGCGGTTG)— digested with BamHI-HF/XhoI for Haus6 and BamHI-HF/XhoI for Haus8, and cloned into a modified pETDuet-1⁴⁰ digested with BamHI-HF/SalI-HF for Haus6 and BglII/XhoI for Haus8, placing a TEV-cleavable His₆ tag on Haus6. Haus2.L_1-116 and Haus7.L_1-264 codon-optimized for E. coli expression were likewise amplified using the following oligonucleotide pairs—Haus2 (TATACCATGGATGATGGCTGCTAACCCGTG/TATACTCGAGTCAAACCGGCTGTGCCAGAC) and Haus7 (TATAGGATCCCATGACGGGTGGCAAAG/TATACTCGAGTTATTGAAATTCGTTGGCATAAATCTGGTTGAAC—digested with NcoI-HF/XhoI for Haus2 and BamHI-HF/XhoI for Haus7, and cloned into pRSFDuet-1 digested with NcoI-HF/SalI-HF for Haus2 and BglII/XhoI for Haus7. CH domain deletions were introduced into previously-described pFASTBAC plasmids for full-length augmin constructs^13,14 via subcloning⁴¹ using the following oligonucleotide pairs—Haus6 (TATAGAATTCATACAAGATCCCCAAAAGGCACTG/CGACAAGCTTTTAGTGGCAAAAGACG) and Haus7 (TATAGGATCCATGGAATCATTGTCTCACAGCACAGAGAAAAATG/TATAGTCGACTCAGTCCTGCAATGAACTTTGTTGGAATG)—and restriction enzyme digest with EcoRI-HF/HindIII-HF for Haus6 and BamHI-HF/SalI-HF for Haus7. Plasmids were verified by sequencing. Recombinant bacmids were produced by transformation of pFASTBAC plasmids into DH10Bac competent cells (NEB), screening for recombination on XGAL plates, followed by bacmid isolation⁴² and confirmation through diagnostic PCR.

Recombinant protein expression and purification

T-II_bonsai was produced by co-transformation of pETDuet and pRSFDuet plasmids into Rosetta2 (DE3) pLysS cells (Millipore Sigma, catalog number 71397-3). Dually-transformed cells were grown to mid-log in modified Terrific broth, inducted with 0.5 mM isopropyl-β-D-thiogalactopyranoside, and protein was expressed at 18 °C for 16–18 h. Cells were harvested and resuspended in lysis buffer (50 mM Tris pH 7.5, 200 mM NaCl, 20 mM imidazole, 5 mM β-mercaptoethanol (BME)) supplemented with 1 mM phenylmethylsulfonyl fluoride (PMSF) and 10 μg/mL DNase I from bovine pancreas (Roche). The cellular resuspension was then lysed using an Avestin Emulsiflex C3 and clarified by ultracentrifugation at 105,000 x g at 4 °C for 30 min. The supernatant was then loaded onto a 10 mL Ni-NTA agarose (Qiagen), washed with 100 mL lysis buffer, 50 mL lysis buffer supplemented with 40 mM imidazole, then eluted with 30 mL lysis buffer supplemented with 300 mM imidazole. Solid ammonium sulfate was added to a final concentration of 1 M then, following 0.22 μm syringe filtration, eluate was loaded onto a 1 mL Butyl Sepharose HP column (Cytiva) pre-equilibrated with HIC buffer (50 mM Tris pH 7.5, 1 M ammonium sulfate, 1 mM DTT). T-II_bonsai was eluted using an ammonium sulfate gradient from 1 to 0 M, and complex eluted at a concentration of 330 mM ammonium sulfate. Eluate was concentrated in a 100 kDa molecular weight cut-off (MWCO) concentrator (Cytiva), then loaded onto a Superdex 200 Increase 10/300 gel filtration column (Cytiva) pre-equilibrated with CSF-XB (10 mM HEPES pH 7.7, 100 mM KCl, 5 mM ethylene glycol tetraacetic acid (EGTA), 1 mM MgCl₂, 1 mM dithiothreitol (DTT)). Peak fraction purity was assessed via MES SDS-PAGE, and peak fractions were pooled. Pool was concentrated to 12 μM, as determined via Bradford assay calibrated to a bovine serum albumin (BSA) standard curve, followed by snap freezing in liquid nitrogen.

Octameric augmin lacking residues 431-978 of Haus6 was prepared similarly to what was previously reported¹⁴. 2 L of Sf9 cells (Gibco, 12659017) at 1 × 10⁶ cells/mL were infected with eight separate recombinant baculoviruses, each containing a single augmin subunit. After 68 h, cells were harvested, lysed, and recombinant augmin was purified via IgG Sepharose affinity chromatography, proteolytic release by PreScission protease, followed by Superose 6 gel filtration. Wild type augmin contained a C-terminal GFP tag on Haus2 and N-terminal GFP tag on Haus1, whereas augmin lacking the N-terminal MTBR of Haus8 (residues 1-154) was tagged only with a C-terminal GFP on Haus2 (as previously described in ref. ⁴⁰), and the protein concentration was quantified by Bradford assay as approximately 300 nM. Augmin lacking the N-terminal CH domain of Haus6 (residues 1-160), due to issues with protein stability, had three GFP tags—on the C-terminus of Haus2, the C-terminus of Haus4, and the N-terminus of Haus8—and the protein concentration was quantified by Bradford assay as approximately 100 nM. Augmin lacking the N-terminal CH domain of Haus7 (residues 1-131) was soluble but prone to aggregation, and was prepared at approximately 300 nM with GFP tags on the C-terminus of Haus2, the C-terminus of Haus1, and the N-terminus of Haus8.

The purification strategy for BFP-TPX2³² has been described previously. Briefly, X. laevis TXP2 was cloned as an N-terminally tagged Strep-6xHis-BFP-TEV-TPX2 fusion, then transformed into Rosetta2 E. coli cells, grown at 37 °C in high-salt Luria broth to mid-log, then cooled to 27 °C and induced with 0.75 mM IPTG for another 7 h. Cells were lysed using an Emulsiflex (Avestin) in lysis buffer (50 mM Tris-HCl, pH 7.75, 15 mM imidazole, 750 mM NaCl) containing 2 mM PMSF, 6 mM BME, cOmplete EDTA-free Protease Inhibitor tablet (Sigma 5056489001), and 10 μg/mL DNase I from bovine pancreas. Lysate was clarified by ultracentrifugation at 105,000 × g at 4 °C for 30 min. Clarified lysate was bound to pre-washed Ni-NTA agarose beads (Qiagen 1018236) in a gravity column and beads were washed with 10 column volumes (CVs) of lysis buffer. Protein was eluted in lysis buffer containing 200 mM imidazole and then further purified via gel filtration (Superdex 200 HiLoad 16/600, GE Healthcare—28-9893-35) in CSF-XB buffer (10 mM HEPES, pH 7.75, 0.5 M KCl, 2 mM MgCl₂, 1 mM CaCl₂, 4 mM EGTA, 10% w/v sucrose). Peak fractions were pooled, concentrated (Amicon Ultra, ThermoFisher—30,000 Da MWCO), flash frozen in liquid nitrogen, and stored at −80 °C.

The purification strategy for EB1-GFP was adapted from the protocol described previously³¹. Briefly, H. sapiens EB1 was cloned as a C-terminally tagged GFP-6xHis fusion in pET21, then transformed into Rosetta2 E. coli, grown at 37 °C in Terrific broth to mid-log, then cooled to 18 °C and induced with 0.5 mM IPTG for 16 h. Cells were lysed using an Emulsiflex in lysis buffer (50 mM Tris, pH 7.5, 20 mM imidazole, 200 mM NaCl) containing 2 mM PMSF, 6 mM BME, and 10 μg/mL DNase I from bovine pancrease. As for the TPX2 prep above, lysate was clarified, loaded onto Ni-NTA agarose beads, washed with 10 CVs lysis buffer, followed by 10 CVs lysis buffer supplemented with 40 mM imidazole, then eluted in lysis buffer containing 300 mM imidazole. Protein was concentrated using a 30 kDa MWCO concentrator, then further purified by gel filtration (Superdex 200 HiLoad 16/600) in CSF-XB buffer. Peak fractions were pooled, flash frozen in liquid nitrogen, and stored at −80 °C.

To purify endogenous y-TuRC, we expressed and purified a 51 amino acid fragment of human CDK5RAP2 (51-100) containing the CM1 motif/γ-TuNA domain, as previously described⁴³, as a Strep-His-Halo-tag fusion protein construct. Yield and purity were assessed via SDS-PAGE gel and Coomassie stain. Concentration was assessed via Bradford assay. γ-TuNA construct yielded between 40 and 60 mg of protein (per 2 L TB culture) at >98% purity.

Endogenous pulldown of native X. laevis γ-TuRC

To purify native γ-TuRC from X. laevis egg extract, we employed a strategy as previously described in Rale et al.⁴³. In short, we coupled 15 mg human Halo-γ-TuNA directly to beads via the affinity of the Halo-tag for its substrate. These beads were then used to pull down γ-TuRC from X. laevis egg extract. X. laevis egg extracts were prepared as described previously⁴⁴. In brief, 3-4 adult female X. laevis frogs (sourced from Nasco/Xenopus1) were induced to ovulate in Marc’s Modified Ringer solution at 16 °C overnight. Meiotic egg extract was isolated at 4 °C following a low-speed spin, flash-frozen, and stored at −80 °C until use.

Halo-human γ-TuNA beads were incubated with the extract for 2 h at 4 °C, while rotating. The beads were then removed from the extract, washed with 4.5 CV of CSF-XB, and then resuspended with 2 mL of PreScission Elution buffer (600 µg PreScission protease diluted in CSF-XB, 2% w/v sucrose). Proteins were eluted from the beads overnight at 4 °C, 15–18 h, with rotation.

The elution containing cleaved γ-TuNA, γ-TuRC, and other factors, was concentrated to ~300 µL in a 100 kDa MWCO Amicon 4 mL spin concentrator, and then spun through a 10–50% w/w sucrose gradient (in CSF-XB buffer) using a TLS55 rotor in a Beckman Coulter Optima MAX-XP ultracentrifuge at 200,000 x g, 4 °C for 3 h. The gradient was manually fractionated such that the first fraction was the same size as the input (~300 µL), and each subsequent fraction was 140 µL in size. Samples of each fraction were run on a 4–12% Bis-TRIS SDS-PAGE gel. The SNAP i.d. 2.0 rapid Western blotting system (cat. #SNAP2MM, EMD-Millipore, Burlington, MA) was used to determine the peak γ-TuRC fraction (usually fraction 7; γ-tubulin, GTU-88 mouse antibody, cat. # T6557, Sigma, St. Louis, MO, 1:500 dilution; sheep anti-mouse HRP-linked antibody, cat # NA931, Amersham, 1:500 dilution). The peak γ-TuRC fraction was stored on ice and used within 24 h or snap-frozen and stored at −80 °C. This prep yielded an average peak concentration between 150–200 nM γ-TuRC (per 140 µL fraction), as we recently estimated⁴³.

Cryo-EM sample preparation and data collection

Guanylyl-α,β-methelynediphosphate (GMPCPP)-stabilized bovine MT seeds were prepared via standard procedures⁹. In brief, 20 μM bovine tubulin (PurSolutions) was incubated with 1 mM GMPCPP (Jena Biosciences) in 1x BRB80 (80 mM PIPES, pH 6.8, 1 mM EGTA, 1 mM MgCl₂) at 37 °C for 1 h. Guanosine 5’-O-[γ-thio]-triphosphate (GTPγS)-stabilized bovine MT seeds were prepared using a protocol adapted from ref. ⁴⁵. In brief, 36 μM bovine tubulin was incubated with 1 mM GTPγS (Sigma Aldrich) in 1x BRB80 lacking MgCl₂ and supplemented with 25% v/v glycerol at 37 °C for 1 h. Seeds were sonicated for 1 min, then 10 μL seeds were added to a fresh 100 μL polymerization reaction containing 33 μM bovine tubulin, 1 mM GTPγS, and 1x BRB80 lacking MgCl₂ at 37 °C for 1 h. After the second round of polymerization, seeds were pelleted at room temperature at 17,000 x g for 15 min and resuspended in 1 mM GTPγS and 1x BRB80 lacking MgCl₂ at a final tubulin concentration of 66 μM. T-II_bonsai was prepared for grid freezing by buffer exchange into cryo-EM buffer (80 mM PIPES pH 6.8, 1 mM EGTA, 1 mM MgCl₂, 1 mM DTT, 0.05% v/v Tween-20) at 4 °C via an 0.5 mL 7 kDa MWCO ZebaSpin desalting column (Thermo Scientific). After buffer exchange, T-II_bonsai protein was clarified by ultracentrifugation at 279,000 x g at 4 °C for 10 min. Estimation of protein concentration by NanoDrop absorbance at 280 nm suggested a final T-II_bonsai concentration of 10 μM.

Quantifoil 1.2/1.3 holey carbon grids on a 400-mesh copper support were glow-discharged for 60 s at 10 mA using a Ted Pella easiGlow then loaded onto a Leica GP2 in a humidity chamber pre-incubated at 25 °C and 85% humidity. 3 μL undiluted GMPCPP or GTPγS MT seeds were incubated on grids for 30 s, removed by manual blotting, then 3 μL T-II_bonsai were added and incubated on grids for 30 s, manually blotted, followed by a second application of T-II_bonsai and 30 s incubation. Finally, grids were blotted for 5–7 s by the GP2 and plunge-frozen in liquid ethane at −182 °C.

Data were collected at the Princeton University Imaging and Analysis Center using ThermoFisher EPU version 2.12.1.2782REL and a G3 Titan Krios equipped with a Gatan BioQuantum energy filter and Gatan K2 Summit direct electron detector (Supplementary Figs. 6a and 14a). Data collection parameters are summarized in Supplementary Table 1.

Cryo-EM data processing

Cryo-EM data processing was performed as previously described in ref. ⁸ using cryoSPARC version 4.4.0⁴⁶. Motion correction was performed using the patch motion correction algorithm, followed by CTF correction using CTFFind version 4⁴⁷. For the GMPCPP structure, approximately 100 micrographs were manually picked, and templates were generated by 2D classification. Templates were then used for autopicking using the filament tracer utility in cryoSPARC. For the GTPγS structure, microtubules were manually picked as filaments in Relion 3.1.0⁴⁸, and particles were extracted along the filament length at a separation of 82.5 Å. Particles were sorted using two rounds of 2D classification (Supplementary Figs. 6b, 14b), then 13- and 14-protofilament MTs were sorted by heterogeneous refinement from undecorated references. Alignments for the predominant 14-protofilament class were then further refined using the helical refinement and local refinement routines in cryoSPARC (Supplementary Figs. 6c, 14c). Particles from local refinement were then transferred to Frealign version 9^49,50 for two rounds of seam searching using the protocol previously outlined^8,51. Correctly aligned particles were reimported into cryoSPARC, individually CTF-corrected, and locally refined to produce the C1 map (Supplementary Figs. 6d, 14d). For GMPCPP, symmetry expansion was performed using a helical rise of 8.988 Å, a helical twist of −25.75°, and 11-fold symmetrization (to avoid mis-averaging of α, β-tubulin across the microtubule seam). For GTPγS, symmetry expansion was performed using a helical rise of 9.455 Å, a helical twist of −27.69°, and 11-fold symmetrization The symmetry-expanded data set was further refined using a mask around a tubulin dimer (located at the opposite side of the seam) plus augmin density (Supplementary Figs. 6e, 14e). The GMPCPP map was sharpened using both cryoSPARC “Sharpening Tools” with a B-factor of 60 Ų and using DeepEMhancer with the tightT setting⁵². The GTPγS map was filtered using cryoSPARC’s “Local Filter” job.

Model building, refinement, and validation

6DPU, a structure of the porcine 14-protofilament GMPCPP MT lattice²⁸, and chains E-H of 8FCK, a structure of the X. laevis augmin complex¹⁴, were used as initial models and rigid-body fit into the symmetry-expanded, sharpened map using UCSF Chimera version 1.13.1⁵³. To improve precision of pixel size determination, we empirically determined GMPCPP structure pixel size by calculating the best fit to the tubulin atomic model, slightly rescaling the resulting map from 1.426 to 1.405 Å/pix⁵⁴. This recalibrated pixel size was used throughout for the GTPγS structure. For the GMPCPP structure, the sequence of 6DPU β-tubulin was mutated to match that of bovine tubulin, and disordered, non-visible regions of augmin were removed from the model in COOT version 0.9.8.93⁵⁵. Refinement was performed using Phenix version 1.17.1-3660 real_space_refine⁵⁶. For the GTPγS structure, the GMPCPP structure was fit as four rigid bodies—chains A, B, C, and the heterotrimer composed of chains F/G/H—in Chimera, and the GTPγS nucleotide was manually added in COOT. The GMPCPP C1 map is available from the EMDB using accession code EMD-46488, the GMPCPP symmetry-expanded map is available from the EMDB using accession code EMD-46489, and the GTPγS symmetry-expanded map is available from the EMDB using accession code EMD-70593. The GMPCPP model coordinates are available from the PDB using accession code 9D2B and the GTPγS coordinates using accession code 9OLH. Predicted structures of X. laevis augmin T-II_bonsai (tetramer), T-II full-length, and T-II_bonsai (trimer) were generated using AlphaFold3⁵⁷ and are accessible from ModelArchive using the respective accession codes 10.5452/ma-xqslg [https://www.modelarchive.org/doi/10.5452/ma-xqslg], 10.5452/ma-k2itn [https://www.modelarchive.org/doi/10.5452/ma-k2itn], and 10.5452/ma-tkb40 [https://www.modelarchive.org/doi/10.5452/ma-tkb40]. Theoretical atomic-coordinate-based maps were calculated using the UCSF Chimera utility molMap at a resolution cut-off of 3.1 Å, the nominal resolution of our structure. Difference maps were calculated between the theoretical map and the sharpened maps using the UCSF Chimera utility vop subtract, followed by a Gaussian volume filter of 1.0. Buried surface area was calculated using the PISA server⁵⁸. Map and model figures were generated using UCSF Chimera version 1.13.1⁵³, using the deepEMhancer sharpened map except where indicated, and protein sequence alignments were generated using Clustal Omega⁵⁹.

MT sedimentation assays

20 μM bovine brain tubulin was resuspended in BRB80 supplemented with 1 mM GTP and 0.1 mM DTT and allowed to polymerize at 37 °C for 20 min. Paclitaxel solubilized in isopropanol was added to a final concentration of 50 μM and polymerization was allowed to continue for an additional 10 min at 37 °C. Wildtype or mutant augmin T-II_bonsai was added to a final concentration of 1 μM in BRB80 supplemented with 50 μM paclitaxel and allowed to bind at room temperature for 10 min. Reactions were spun through a 140 μL 40% v/v glycerol cushion in BRB80/paclitaxel at 279,000 x g for 10 min at room temperature. 20 μL from the top of the tube was removed as the supernatant fraction, then, after removal of the glycerol cushion, the pellet was resuspended in cold BRB80 supplemented with 10 mM CaCl₂. Soluble and pellet fractions from each condition were run on 4–12% MES-buffered SDS-PAGE, then silver stained using the SilverQuest kit. Experiments were performed in technical quadruplicate, and band intensities were quantified in FIJI/ImageJ version 2.14.0/1.54f⁶⁰. Band intensities were averaged across all augmin subunits in each lane. Fraction bound (f_b) was computed as Eq. (1):

$$f_{\mathrm{b}}=\frac{{\mathrm{Augmin}}_{\mathrm{pellet}}}{{\mathrm{Augmin}}_{\mathrm{soluble}}+{\mathrm{Augmin}}_{\mathrm{pellet}}}$$ 1

and plotted as a function of total MT concentration. K_D was calculated for each MT concentration and each technical quadruplicate using Eq. (2):

$$K_{\mathrm{D}}=\frac{\left(\left[{\mathrm{Augmin}}_{\mathrm{total}}\right]-f_{\mathrm{b}}\times \left[{\mathrm{Augmin}}_{\mathrm{total}}\right]\right)\times \left(\left[\mathrm{MT}\right]-f_{\mathrm{b}}\times \left[{\mathrm{Augmin}}_{\mathrm{total}}\right]\right)}{f_{\mathrm{b}}\times \left[{\mathrm{Augmin}}_{\mathrm{total}}\right]}$$ 2

Data points for 5 and 10 μM tubulin were excluded from the K_D calculation due to a systematic underestimate of the augmin signal caused by increased background inherent to the high concentration of tubulin in those lanes. The K_D was reported as the mean of all calculated K_Ds, along with the standard error.

Preparation of functionalized coverslips and imaging chambers

We used a previously published method to generate functionalized glass coverslips for TIRF imaging^43,61,62. Briefly, coverslips were sonicated in 3 M NaOH followed by Piranha solution (2:3 ratio of 30% w/w H₂O₂ to sulfuric acid). After washing with MilliQ water, we dried and then treated the coverslips with 3-glycidyloxypropyl trimethoxysilane (GOPTS, cat. # 440167, Sigma) at 75 °C for 30 min. Unreacted GOPTS was removed with two sequential acetone washes, and the coverslips were then dried with nitrogen gas. Between sandwiched coverslips, we melted a powder mix of 9 parts HO-PEG-NH₂ and 1 part biotin-CONH-PEG-NH₂ by weight (cat. #103000–20 and #133000-25-20, Rapp Polymere, Tübingen, Germany) at 75 °C. We pressed out air bubbles and repeated cycles of 75 °C incubation and pressing. After an overnight incubation at 75 °C, the sandwiches were separated and washed in MilliQ water. These were then spun dry in air and stored at 4 °C for up to 1 month.

We made imaging chambers by first making a channel on a glass slide with double-sided tape. To this channel, we added 2 mg/mL PLL-γ-PEG (SuSOS AG, Dübendorf, Switzerland) in MilliQ water. After a 20 min incubation, the channel was rinsed thoroughly with MilliQ water and dried with nitrogen gas. Using a diamond pen, functionalized coverslips were cut into quarters. Each quarter piece was then added to the double sided tape on the slide with the functionalized surface facing down. Chambers were made fresh on the day of each experiment.

Tubulin labeling and preparation of double cycled GMPCPP-stabilized MTs

Bovine brain tubulin was labeled following prior methods⁹. Using Alexa568-NHS ester (Invitrogen; A20003) yielded 36–40% labeling efficiency. Double cycled GMPCPP-stabilized MTs were made as previously described (42, 63). Briefly, 18 μM unlabeled bovine tubulin supplemented with 1 μM Alexa568 labeled tubulin and 1 μM biotinylated tubulin was polymerized in BRB80 buffer (80 mM PIPES, pH 6.8, 1 mM EGTA, and 1 mM MgCl₂) in the presence of 1 mM GMPCPP for 1 h at 37 °C. After 1 h, the MT seed mixture was centrifuged at 17,000 x g at room temperature for 15 min. The remaining pellet was then resuspended in cold BRB80 and left on ice to depolymerize for 20 min, followed by a second round of polymerization with 1 mM GMPCPP for 1 h at 37 °C. After 1 h, the mixture was centrifuged again at 17,000 x g at room temperature for 15 min. The supernatant was removed, and the pellet was resuspended in warm BRB80 and snap frozen in 1 μL aliquots and stored at −80 °C for later use.

Preparation of subtilisin-cleaved, GMPCPP MT seeds

Single cycled, GMPCPP-stabilized MT seeds were first prepared. Briefly, 18 μM unlabeled bovine tubulin supplemented with 1 μM Alexa568 labeled tubulin and 1 μM biotinylated tubulin was polymerized in BRB80 buffer (80 mM PIPES, pH 6.8, 1 mM EGTA, and 1 mM MgCl₂) in the presence of 1 mM GMPCPP for 1 h at 37 °C. After 1 h, the MT seed mixture was centrifuged at 17,000 x g at room temperature for 15 min. The supernatant was removed and the pellet was resuspended in warm BRB80 supplemented with 1 mM GMPCPP. The MTs were then incubated with 100 ug/ml Subtilisin A, and incubated at 37 °C for 1 h. 2 mM PMSF was then added to quench the reaction, and the seeds were spun down at 17,000 × g at room temperature for 15 min. The supernatant was removed, and the pellet was washed one time with BRB80 to remove any additional Subtilisin A. The final pellet was resuspended in warm BRB80, and used immediately for TIRF microscopy experiments.

Preparation of dual-labeled MT seeds

To prepare dual-labeled GMPCPP-GTPγS MT seeds, we modified a previously established protocol for growing GTPγS MTs from GMPCPP templates²⁸. 20 μM bovine brain tubulin with 10% Atto 647 N (ATTO-TEC Gmbh) labeled tubulin and 10% biotinylated tubulin were incubated on ice for 5 min then polymerized in the presence of 1 mM GMPCPP at 37 °C for 1 h in BRB80. The seeds were pelleted by centrifugation at 17,000 ×g at room temperature for 15 min, then resuspended in 10% Alexa568 labeled 50 μM tubulin in modified BRB80 buffer containing 0.2 mM MgCl₂ and 1 mM guanosine 5’-O-[γ-thio]triphosphate (GTPγS) (pre-incubated on ice for 5 min). The mixture was then polymerized for 1 h at 37 °C. These seeds were pelleted again by the same means and resuspended in 0.2 mM MgCl₂ BRB80 supplemented with 1 mM DTT.

To prepare a 1:1 mixture of labeled GTPγS and GDP seeds, we modified the above protocol to produce GMPCPP-Atto-647N GTPγS seeds by omitting fluorescently labeled tubulin from the first GMPCPP polymerization step and substituting 10% Atto-647N-labeled tubulin for Alexa568-labeled tubulin in the second GTPγS polymerization step. We produced GMPCPP-capped Alexa568-labeled GDP seeds in the following way. 20 μM bovine brain tubulin with 10% Alexa568-labeled tubulin and 10% biotinylated tubulin in BRB80 supplemented with 15% v/v glycerol were incubated on ice for 5 min then polymerized in the presence of 1 mM GTP at 37 °C for 45 min. The seeds were pelleted by centrifugation at 17,000 × g at room temperature for 15 min, then resuspended in 1 μM tubulin in BRB80 buffer containing 1 mM GMPCPP and 15% v/v glycerol, which had been preincubated on ice for 5 min. The second polymerization step was then performed for 45 min at 37 °C. Seeds were again pelleted as above and resuspended in room temperature 0.2 mM MgCl₂ BRB80. For attachment, GMPCPP-GTPγS and GDP-GMPCPP seeds were diluted and mixed using a final ratio of 1 part GMPCPP-GTPγS, 1 part GDP-GMPCPP, and 300 parts 0.2 mM MgCl₂ BRB80.

Attachment of stabilized MTs to PEG-functionalized coverslips

Flow chambers were incubated with 5% v/v Pluronic F-127 in water (Invitrogen; catalog no.: P6866) for 5 min at room temperature and then washed with assay buffer (BRB80; 5 mM BME, 0.075% w/v methylcellulose, 1% w/v glucose, 0.02% v/v Brij-35 [Thermo Scientific; catalog no.: 20150]) supplemented with 50 μg/mL κ-casein. Flow chambers were then incubated with an assay buffer containing 50 μg/mL NeutrAvidin (Invitrogen; catalog no.: A2666) for 2 min on a metal block on ice and subsequently washed with BRB80. Next, flow chambers were incubated for 5 min at room temperature with Alexa568-labeled biotinylated GMPCPP-stabilized MTs diluted 1:2000 in BRB80. Unbound MTs were removed by additional BRB80 washes.

Binding of proteins to GMPCPP MT seeds

Before initiating branching MT nucleation, TPX2 (10 nM), augmin constructs (~75 nM), and y-TuRC (~75 nM) were sequentially flowed into the flow chambers containing attached GMPCPP-stabilized MT seeds attached to the surface. TPX2 was added at a concentration of 10 nM, below the phase boundary, to prevent formation of condensates. All proteins were diluted into CSF-XB buffer with 8% w/v sucrose. Proteins were incubated for 5 min at room temperature. Unbound proteins were removed with additional BRB80 washes prior to incubation of the next protein. Additionally, prior to each protein incubation, the flow chamber was washed with a blocking buffer containing 1 mg/mL κ-casein. For experiments testing the binding of augmin constructs to GMPCPP seeds without TPX2 present (as in Fig. 2), all augmin constructs were diluted directly into a CSF-XB-based imaging buffer containing oxygen scavengers and imaged directly after a 10 min incubation without washing the channels.

MT bundling experiments

To determine if a MAP was capable of bundling two MTs, biotinylated, Alexa647 GMPCPP-stabilized MT seeds were first bound to a glass coverslip as described in the previous section. Next, augmin constructs were mixed with Alexa568 GMPCPP-stabilized MT seeds and immediately flowed into the channel. This mixture was incubated for 10 min before washing with BRB80 and imaging via TIRF. The augmin construct was considered able to bundle if the two colors of MT seeds (Alexa568, Alexa647) co-localized with one another along with the GFP-labeled augmin.

Binding of proteins to dual-labeled MT seeds and analysis

Dual-labeled seeds were attached to PEG-functionalized coverslips (as above), and EB1, TPX2, and augmin octamer imaging mixtures were prepared in a 5% w/v sucrose CSF-XB-based imaging buffer as described above. All reactions were imaged within 10 min. 3 × 3 arrays were imaged using a Nikon Ti-E inverted stand (RRID:SCR_021242) with an Apo TIRF 100 x oil objective (NA = 1.49) and an Andor iXon DU-897 EM-CCD camera with EM gain set to 300. Microscopy data were processed using a custom macro in FIJI/ImageJ. Background subtraction was performed using the Subtract Background command on the MAP channel. MT seeds were segmented into regions of interest by automatically thresholding the GMPCPP, GTPγS, and/or GDP channels as applicable using the Otsu method⁶³ and selecting particles the “Analyze Particles” function with minimum 1.0 μm² area and a circularity range of 0.0–0.5. Average MAP fluorescent signals per pixel were calculated for each MT.

MT nucleation assays on PEG-functionalized surfaces

For branching MT nucleation reactions in vitro, BFP-TPX2 (10 nM), GFP-augmin ( ~ 75 nM) and γ-TuRC ( ~ 75 nM) were added sequentially to the chamber containing attached GMPCPP-stabilized MTs and incubated for 5 min at room temperature. Unbound proteins were removed by additional BRB80 washes, followed by washes with 1 mg/mL κ-casein in BRB80. The final assay mixture was flowed into the chambers: 80 mM PIPES, pH 6.8, 30 mM KCl, 1 mM EGTA, 1 mM MgCl₂, 1 mM GTP, 5 mM BME, 0.075% w/v methylcellulose (4000 cP), 1% w/v glucose, 0.02% v/v Brij-35, 250 nM glucose oxidase, 64 nM catalase, 1 mg/mL BSA, 19 μM unlabeled bovine tubulin and 1 μM Cy5-labeled bovine tubulin. Before use, all proteins were pre-cleared of aggregates via ultracentrifugation at 346,000 × g for 15 min in a TLA100 rotor in an Optima MAX-XP ultracentrifuge at 4 °C.

TIRF imaging and image analysis

Total internal reflection fluorescence (TIRF) microscopy was performed with a Nikon TiE microscope using a 100 × 1.49 NA objective. Andor Zyla sCMOS camera was used for acquisition, with a field of view of 165.1 × 139.3 µm. 2 × 2 binned, multi-color images were acquired using NIS-Elements version 5.02.01-Build 1270 (Nikon). All adjustable imaging parameters (exposure time, laser intensity, and TIRF angle) were kept the same within experiments. For MT nucleation assays in vitro, the TIRF objective was warmed to 33 °C using an objective heater (Bioptechs, 150819–13). For all time-lapse imaging, one multi-color image was taken at the beginning of the experiment followed by collection of time-lapse images on the 647 nm channel every 2 s for 10 min. After the time lapse was collected, another multi-color image was taken. For still images of augmin binding and branched MT nucleation experiments, an additional 15–30 multi-color images were collected. Brightness and contrast were optimized individually for display.

Images used for the quantification of MT binding were analyzed using FIJI/ImageJ. Background subtraction was performed using the Subtract Background command on the MAP channel. To segment MTs, the tubulin signal was first thresholded via the Otsu method. MTs were isolated from the mask by setting the minimum particle area as 1 or 2 μm². Average fluorescent signals per pixel, for the MT or for bound proteins, were calculated for each MT. The average intensity from the reverse mask of the entire field of view was subtracted from the average intensity on each MT. For branching MT nucleation experiments in vitro, MTs were counted manually using time-lapse experiments, and multiple aspects of the data were taken into account to exclude MTs of ambiguous origin. In the case where branched MTs were quantified from still images, the MTs were counted manually and only included in the analysis if there was a clear distinction between the plus and minus end of the MT. Additionally, branched MTs were excluded from analysis if two microtubule seeds were close to each other on the glass coverslip, and the origin of the daughter MT was ambiguous in a still image. Branched MTs were also excluded if there was a possibility of what we term a crossing event, where a spontaneously nucleated MT could be above the immobilized seed. In the case that the origin of the branched MT was still ambiguous, kymographs were prepared from the live movies to detect which MTs were true branches vs. spontaneously nucleated (Supplementary Fig. 22). Branch angles were measured manually using FIJI/ImageJ. For nearly parallel branch angles of <5 degrees, kymographs were made of the MT plus ends to show that a second MT was growing in addition to the polymerizing MT end (Supplementary Fig. 19, Supplementary Movie 6). To quantify parallel branch angles in the still images, the mean intensity of the plus end was quantified to show that the intensity was twice of that expected if there was only a single MT polymerizing from the plus end (Supplementary Fig. 20). The branch angle distributions were plotted as histograms in Prism.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

S.M.T.: Conceptualization, Methodology, Formal Analysis, Investigation, Resources, Data Curation, Writing—Original Draft and Editing, Visualization. J.K.: Conceptualization, Methodology, Formal Analysis, Investigation, Resources, Writing—Original Draft, Reviewing, and Editing, Visualization. C.T.M.: Methodology, Formal Analysis, Investigation, Visualization, Writing—Reviewing and Editing. K.G.: Investigation, Resources. R.Z.: Investigation, Resources, Data Analysis, Writing—Reviewing and Editing, Supervision. S.P.: Conceptualization, Writing—Reviewing and Editing, Supervision, Project Administration, Funding Acquisition.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Data availability

The reconstructed electron density maps generated in this study have been deposited in the EMDB under accession codes EMD-46488 (GMPCPP C1), EMD-46489 (GMPCPP symmetry-expanded), and EMD-70593 (GTPγS). The rigid-body fit structural model associated with these map and generated in this study have been deposited in the PDB under accession codes 9D2B (GMPCPP) and 9OLH (GTPγS). Rigid-body fit structural models were adapted from the following published models: 6DPU (GMPCPP-bound MT) and 8FCK (X. laevis augmin complex). Predicted structures of X. laevis augmin T-II_bonsai (tetramer), T-II full-length, and T-II_bonsai (trimer) have been deposited in ModelArchive under accession codes 10.5452/ma-xqslg [https://www.modelarchive.org/doi/10.5452/ma-xqslg] (T-II_bonsai tetramer), 10.5452/ma-k2itn [https://www.modelarchive.org/doi/10.5452/ma-k2itn] (T-II full-length), and 10.5452/ma-tkb40 [https://www.modelarchive.org/doi/10.5452/ma-tkb40] (T-II_bonsai trimer). Source data for Figs. 2b, 3b, 3d, 4d, and 4e and for Supplementary Figs. 5b, 12a–c, 20b, and 21 are provided as a Source Data file. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-025-64650-1.