Mechanism of how augmin directly targets the γ-tubulin ring complex to microtubules

Microtubules (MTs) originate from specific locations in the cell, which are broadly defined as MT organizing centers (MTOCs; Brinkley, 1985; Lüders and Stearns, 2007). MT nucleation at MTOCs requires the γ-tubulin ring complex (γ-TuRC), which provides a ring-shaped template to assemble α, β-tubulin heterodimers into a MT (Moritz et al., 1995; Zheng et al., 1995; Kollman et al., 2011). Therefore, it is essential to target γ-TuRC to these nucleation sites. Several factors have been identified in vivo to play a role in recruiting γ-TuRC to MTOCs, such as AKAP450 at the Golgi apparatus, CDK5RAP2 and NEDD1 at centrosomes, and augmin and NEDD1 at spindle MTs (Haren et al., 2006; Lüders et al., 2006; Fong et al., 2008; Rivero et al., 2009). However, whether they perform this function independent of other factors and how they recruit γ-TuRC to MTOCs on a molecular level are not known.

The eight-subunit protein complex augmin mediates MT nucleation from preexisting MTs (Petry et al., 2013). Augmin targets γ-TuRC to spindle MTs based on the findings that knockdown of augmin subunits diminishes both MT density and γ-tubulin signals in the mitotic spindle (Goshima et al., 2007, 2008; Lawo et al., 2009; Uehara et al., 2009; Ho et al., 2011). To perform this function, augmin needs to bind to MTs and γ-TuRC. Out of the eight augmin subunits (denoted HAUS1–8 or H1–8), the augmin subunit HAUS8/Dgt4 is primarily responsible for binding to MTs (Wu et al., 2008; Hsia et al., 2014). In contrast, the C-terminal half of the augmin subunit HAUS6/Dgt6 binds to the adapter protein NEDD1, which in turn binds to γ-TuRC (Uehara et al., 2009). In addition, the N termini of HAUS3/Dgt3 and HAUS5/Dgt5 were identified as NEDD1-binding sites (Chen et al., 2017). Human augmin was recently shown to have a Y-shaped structure (Hsia et al., 2014), yet where the subunits and functional sites are located within the Y-shaped augmin complex and whether additional ones exist are not known. Besides augmin and γ-TuRC, the protein TPX2 is required for MT nucleation from a preexisting MT (Petry et al., 2013). TPX2 is a downstream target of RanGTP (Gruss et al., 2001) and has been suggested to activate γ-TuRC for branching MT nucleation (Alfaro-Aco et al., 2017).

Although augmin has been implicated in localizing γ-TuRC to spindle MTs, its molecular basis remains to be determined. Here, we show that augmin is a direct targeting factor for γ-TuRC to spindle MTs by reconstituting this activity in vitro. Furthermore, we dissect augmin’s functional architecture, which explains how augmin performs this function.

To study augmin’s function and mechanism, we reconstituted the Xenopus laevis augmin holocomplex by coexpressing all eight subunits in SF9 insect cells and purifying the complex. Size-exclusion chromatography revealed that Xenopus augmin contains all eight subunits in equal stoichiometry, as previously reported for human augmin (Hsia et al., 2014; Fig. 1 A). To evaluate the functionality of recombinant augmin in physiological conditions, we established an activity assay consisting of several steps. First, we immunodepleted endogenous augmin from Xenopus egg extracts using custom-made antibodies against augmin subunits. Upon addition of constitutively active Ran (RanQ69L), branched MT networks formed in the control extract, whereas branching MT nucleation was no longer observed in the absence of augmin, as previously reported (Petry et al., 2013; Fig. 1 A). We selected the anti-HAUS1 (H1) antibody for these experiments because it completely depleted augmin while barely codepleting γ-TuRC. To test if our recombinant augmin was indeed active, we added it back to the augmin-depleted extract and assessed its activity via the branching reaction. Notably, recombinant augmin restored branching MT nucleation in a concentration-dependent manner and bound to MTs (Fig. 1 B and Video 1). Thus, reconstituted Xenopus augmin is active. We also note that the MT nucleation activity is reduced compared to the control reaction (Fig. S1), which could be due to a lack of posttranslational modifications and codepleted binding partners. In sum, this add-back system allowed us to directly study, for the first time, augmin’s function while resolving individual MT nucleation events.

Knowing that our recombinant augmin is active, we tested the key question whether augmin directly localizes γ-TuRC to MTs. We purified endogenous γ-TuRC from Xenopus egg extracts and verified its molecular content, size, and structure by mass spectrometry, sucrose gradient centrifugation, and negative-stain EM, respectively (Fig. S2, A and B). To first test whether augmin and γ-TuRC interact in solution, we incubated both purified complexes with each other and performed sucrose gradient sedimentation. As a result, augmin and γ-TuRC are marginally shifted and their fractions share some overlap compared with their individual sedimentation profiles (Fig. S2 A), consistent with previous findings in Drosophila melanogaster S2 extracts (Goshima et al., 2008) and our own investigations in Xenopus egg extracts (Fig. S2 C).

More importantly, can augmin directly target γ-TuRC to preexisting MTs? To test this, we performed MT cosedimentation assays. As expected from its minus-end–binding ability, γ-TuRC itself copelleted with Taxol-stabilized MTs at higher concentrations of 2 µM and 400 nM, but not at a lower concentration of 80 nM (Fig. 2, A and B). In contrast, in the presence of augmin, γ-TuRC was able to bind to MTs at this lower MT concentration of 80 nM and in an increased amount at 400 nM (Fig. 2, A and B), suggesting that γ-TuRC is recruited to additional binding sites on the MT via augmin. To assess where augmin targets γ-TuRC along the MT, we performed immunofluorescence (IF) of γ-TB on Cy5-labeled MTs incubated with either augmin or γ-TuRC or both (Fig. 2 C). In the absence of augmin, little γ-TuRC was detected on MTs (Fig. 2 C). However, when added together, both augmin and γ-TuRC specifically localized along the whole length of MTs (Fig. 2 C). Quantification of the γ-TuRC signal (normalized to the MT signal to account for MT bundling by augmin) revealed that addition of augmin significantly increases γ-TuRC localization to MTs (Fig. 2 D). Although augmin induced MT bundling, single augmin-bound MTs onto which γ-TuRC was recruited were also detected (Fig. S2 D). Thus, we reconstituted γ-TuRC localization along the length of MTs via augmin in vitro and thereby uncovered augmin as a direct targeting factor for γ-TuRC.

Because addition of the other essential branching factor TPX2 to Xenopus egg extracts induces the formation of branched fan-like structures (Petry et al., 2013), we tested whether recombinant augmin can promote MT nucleation from γ-TuRC in addition to being its localizer. When augmin is added to Xenopus egg extracts, it does not increase the number of MTs formed in extract (Fig. S3 A). This is in contrast to TPX2, which stimulates branching MT nucleation in extract (Petry et al., 2013; Alfaro-Aco et al., 2017). Moreover, augmin does not enhance the nucleation capacity of purified γ-TuRC in vitro (Fig. S3 B). These results further support that augmin’s function is to localize γ-TuRC along the length of MTs.

To study how augmin targets γ-TuRC to MTs, we sought to reveal how the eight subunits are arranged within the augmin holocomplex and performed two complementary interaction analyses. First, augmin antibodies were used to identify interacting augmin subunits by immunoprecipitation (IP) followed by Western blot (Fig. 3 A). Furthermore, these IP samples were labeled with different isobaric tandem mass tags and analyzed by quantitative liquid chromatography tandem mass spectrometry (Fig. 3 B), followed by hierarchical clustering to reveal how augmin is assembled (Fig. 3 C). The subunit H8 was shown to directly interact with H6 (Hsia et al., 2014), and this interaction was taken into account for clustering. Based on the resulting dendrogram, we propose two subcomplexes: one subcomplex includes H6, H7, and H8, whereas another one contains H1, H3, H4, and H5 (Fig. 3 D). H2 could have a role in connecting the two subcomplexes, or it could be peripheral (Fig. 3 D). Independently, we used our recombinant augmin expression system to probe direct interactions between two augmin subunits by in vitro pull-down assays (Fig. S4). Each subunit was tested to bind to each of the other subunits (Fig. S4) and unidirectional and reciprocal interactions are depicted in a summary table (Fig. 3 E). These in vitro interaction data were combined with the IP-based hierarchical clustering data to derive the following model for augmin assembly (Fig. 3 F). Reciprocal interactions exist between three subunit pairs, namely, H1/H4, H3/H5, and H6/H8. H6 appears to be the core of the complex because it interacts with five subunits (H1, H2, H4, H7, and H8). As a result, our data suggest that augmin might be assembled via several subcomplexes.

To test the predicted augmin assembly model, we attempted to coexpress and purify different combinations of subunits. Besides the augmin holocomplex, which contains all subunits, we generated an augmin octamer(H6ΔC) that contains all eight subunits, only that its H6 lacks the C-terminal half, which is known to interact with NEDD1 (Uehara et al., 2009; Hsia et al., 2014; Fig. 4, A and B). We were unable to reconstitute the trimer consisting of H6, H7 and H8 subunits, even when using the more soluble form of H6 lacking its C-terminal half (H6ΔC). Instead, the subunit H2 was required to reconstitute H6ΔC, H8, and H7 into a soluble tetramer (tetramer II [T-II]), consisting of the same subunits as the previously denoted T-II made from human augmin (Hsia et al., 2014). The remaining four subunits (H1/4/3/5) associated as a stable complex, which we denote as T-III (Fig. 4, A and B). The previously reported T-I of human augmin, consisting of subunits H6ΔC, H8, H1, and H4, could not be formed with our Xenopus subunits (Hsia et al., 2014). Each tetramer could not be broken down into smaller complexes, except for one dimer consisting of H1 and H4 that was soluble. Also, adding more subunits to each tetramer for coexpression and purification did not result in any monodisperse complexes. This suggests that both tetramers are unique and stable subcomplexes that make up the augmin holocomplex.

Next, we wanted to assess the functions of both tetramers compared with the augmin holocomplex and octamer(H6ΔC). For this purpose, each complex was added to the augmin-depleted extract and tested for their ability to promote branching MT nucleation. Both the augmin holocomplex and octamer(H6ΔC) restored branching MT nucleation, although the octamer(H6ΔC) displayed a reduced level of activity caused by the absence of H6’s C terminus (Fig. 4 C and Video 2). However, neither T-II nor T-III elicited branching nucleation after the depletion of endogenous augmin (Fig. 4 C and Video 2).

To assess why T-II and T-III independently cannot replace endogenous augmin and to determine their function, we tested the ability of each complex to bind MTs. Augmin T-II, similar to holocomplex and octamer(H6ΔC), binds to and bundles guanylyl-(α,β)-methylene-diphosphonate (GMPCPP)–stabilized MTs (Fig. 5 A). In contrast, T-III was not able to bind to MTs. These results are consistent with the finding that H8 within T-II of human augmin is mainly responsible for MT binding (Hsia et al., 2014). To act as a recruiting factor, augmin must also bind to γ-TuRC. To test this, we performed IP experiments using anti-GFP antibodies specific for GFP-tagged recombinant augmin complexes that were added back to augmin-depleted Xenopus egg extracts. Our results demonstrate that γ-TuRC interacts with the augmin holocomplex and, unexpectedly, with the octamer(H6ΔC) in the extract (Fig. S5 A). Thus, in contrast to previous work (Uehara et al., 2009), H6’s C-terminal lacking in the octamer(H6ΔC) is not required for γ-TuRC binding (Fig. S5 A), implying other γ-TuRC binding sites must exist. Interestingly, an interaction between γ-TuRC and T-III (H1/4/3/5) was detected (Fig. S5 A), thus uncovering T-III as a novel γ-TuRC–binding subcomplex of augmin. Importantly, this interaction pattern in the extract is in agreement with pull-down results performed in vitro with purified components (Fig. 5 B). Finally, we analyzed which components of T-III are responsible for γ-TuRC binding. Neither the individual subunits H3 and H5 nor the dimer H1/4 were able to bind to γ-TuRC (Fig. S5 B). Only the complete T-III binds to γ-TuRC, suggesting that the folded tetramer provides the functional binding site. In light of these findings, we tested whether T-II is sufficient to recruit γ-TuRC along the MT length with purified components. As expected, T-II was not able to perform this function, as it lacks the γ-TuRC–binding modules (Fig. S5 C). In summary, we identified two functionally distinct subcomplexes: T-III (H1/4/3/5) mediates γ-TuRC binding, and T-II (H6ΔC/8/2/7) binds to MTs. Both complementary activities (MT and γ-TuRC binding) need to be present in augmin for branching MT nucleation, suggesting that augmin directly bridges MTs and γ-TuRC.

To investigate the structure of the newly identified functional parts, we visualized augmin complexes by negative-stain EM. 2D class averages reveal that Xenopus augmin has a Y shape similar to human augmin (Hsia et al., 2014; Fig. 5 C and Fig. S6 A). Interestingly, our newly identified T-III has a rod shape, which roughly matches the stalk of augmin’s Y shape in dimensions and appearance (Fig. 5 C and Fig. S6 B). We were unable to obtain 2D class averages of the T-II subcomplex with a defined shape, suggesting that T-II is relatively flexible by itself and will only adopt a V-shape conformation when bound to the other tetramer (Fig. 5 C). Finally, we visualized augmin bound to MTs by negative-stain EM, in which the T-III stalk points away from the MT in a perpendicular manner (Figs. 5 D and S5 D). Altogether, this shows the functional architecture of augmin, in which T-II interacts with the MT lattice, whereas the complete T-III protrudes out from the MT to interact with γ-TuRC (Fig. 5 E).

In summary, our data lead to a model (Fig. 5 E) in which augmin binds to MTs via T-II and recruits γ-TuRC predominantly via T-III. Our data suggest H6’s C terminus is dispensable for γ-TuRC binding in vitro. To compare the exact contribution to γ-TuRC binding between the originally proposed C terminus of H6 and the newly identified T-III, it will be necessary to quantitatively compare their binding affinities to γ-TuRC. Although the complete T-III is sufficient to bind γ-TuRC, H6 may be located in the vicinity of T-III where it can also bind to γ-TuRC via NEDD1 (Fig. 5 E). This is in agreement with recent work in D. melanogaster where Dgt3 (H3), Dgt5 (H5), and Dgt6 (H6) were proposed to interact with Dgp71WD (NEDD1; Chen et al., 2017). It will be interesting to see whether the stalk-like T-III that directly binds to γ-TuRC could be a common structural feature of other γ-TuRC targeting factors.

In this study, we provide an interaction model of augmin’s subunits and place them into functional tetramers. Because each tetramer cannot be broken down into smaller units (except for the H1–H4 dimer), the tetramers assume an elongated structure, and each subunit seems to consist mostly of α-helices by secondary structure prediction (Drozdetskiy et al., 2015), it is likely that each tetramer’s four subunits fold into a helical coiled-coil structure (Fig. 5 C). An important next step will be to determine the location of each subunit within the holocomplex and ideally to obtain a high-resolution structure at the atomic level of augmin.

Interestingly, branching of actin filaments is generated with a fixed angle of 70 degrees because of the stable structure of the Arp2/3 complex (Mullins et al., 1998; Robinson et al., 2001). In contrast, branching MT nucleation occurs mostly parallel and with a range of shallow angles (Petry et al., 2013). Our structural data can rationalize this behavior, because augmin mostly binds to MTs in a perpendicular manner but still adopts various conformations around that angle (Figs. 5 D and S5 D). In the future, a 2D class average model of augmin bound to a MT can be obtained by collecting a myriad of particles.

We quantified MT nucleation kinetics in IgG and holocomplex add-back conditions (Fig. 1 B) by tracking EB1 comets (Fig. S1), which revealed two key differences. First, MT nucleation in the augmin add-back condition was delayed compared with the IgG control. Second, the total number of MTs was less upon augmin add back compared with the IgG control. Several explanations could account for these differences. It is possible that recombinant augmin might need to be activated in extracts, and extra time for posttranslational modifications or binding to endogenous partners like γ-TuRC is required after add back. During immunodepletion (ID), augmin-binding partners could get codepleted, which is common in this assay. In fact, coimmunoprecipitated proteins were identified using quantitative mass spectrometry (Fig. 3 C), including γ-TuRC components, and the absence of those proteins could prevent total recovery of MT nucleation after add back. It will be necessary to thoroughly investigate the function of augmin binding partners and their role in making branching MT nucleation more efficient.

It has been reported that recombinant human augmin independently produces asters and enhances Ran aster number in a dose-dependent manner in Xenopus egg extracts (Hsia et al., 2014). Instead, by using branching MT nucleation and total internal reflection fluorescence (TIRF) microscopy, we resolve individual, augmin-dependent MT nucleation events and show that Xenopus augmin does not change the number of MTs. Furthermore, we show that purified augmin does not activate purified γ-TuRC in vitro.

Interestingly, augmin seems to entirely decorate MTs within the branched fan-like structures (Fig. 1 B), yet new MTs only originate from discrete locations and with a certain delay along the MT lattice. Combined with our observation that augmin addition to γ-TuRC does not cause increased MT nucleation (Fig. S3 B), this suggests that the augmin–γ-TuRC interaction is not sufficient to nucleate a new MT from a preexisting MT. Additional factors may be required for MT nucleation from a preexisting MT. The missing piece could be TPX2, which is also essential for branching MT nucleation (Petry et al., 2013) and which was recently proposed to act as a γ-TuRC activator (Petry et al., 2013; Alfaro-Aco et al., 2017). Furthermore, γ-TuRC requires XMAP215 to synergistically nucleate MTs (Thawani et al., 2018). It will be critical to fully reconstitute branching MT nucleation in vitro to identify all factors involved along with their role and to establish the molecular mechanism of branching MT nucleation.

Xenopus augmin genes were cloned into pFastBac vectors for the use of Bac-to-Bac baculovirus expression system (Invitrogen), individually with and without N-terminal Streptag conjugated with GFP. Additionally, H3, H6ΔC (amino acids 1–430), and H6 (amino acids 1–978) genes were cloned with N-terminal ZZ tag (ZZ-H3, ZZ-H6, and ZZ-H6dC; Nilsson et al., 1987; Reck-Peterson et al., 2006) and H1 and H2 genes with C-terminal GFP-Histag (H1CG-His and H2CG-His) for the sequential affinity purifications.

Recombinant baculoviruses were amplified to reachviral titers of 10⁹ pfu/ml. For the expression of augmin complexes, 1–2 liters of SF9 cells (1.5–2.0 × 10⁶/ml) were infected with each baculovirus at a MOI of 1–3, which were collected after 72 h. The following baculovirus combinations were used for each augmin complex: holocomplex with ZZ-H6, H2CG-His, and the remaining six subunits without tag; octamer(H6ΔC) with ZZ-H6ΔC, H2CG-His, and the rest of six subunits without tag; T-II with ZZ-H6ΔC, H2CG-His, H7, and H8; and T-III with ZZ-H3, H1CG-His, H4, and H5.

Harvested insect cells were resuspended with lysis buffer containing 20 mM Tris, pH 7.4, 150 mM NaCl, 3 mM β-mercaptoethanol, 0.05% (vol/vol) Tween-20, and Complete EDTA-free protease inhibitor (Roche) and lysed by EmulsiFlex (Avestin), followed by centrifugation at 80,000 g for 30 min at 4°C. Cleared lysates were mixed with 1–2 ml of IgG-Sepharose (GE Healthcare) slurry for 2 h using rotator at 4°C. IgG beads were collected via a chromatography column (Econo-Column; Bio-Rad) and then washed with 20 times of bed volume of TBS-T (Tris-buffered saline, pH 7.4, with 0.1% [vol/vol] Tween 20). Washed IgG beads were mixed with 3C protease cleavage buffer (20 mM Tris, pH 7.4, 150 mM NaCl, 3 mM β-mercaptoethanol, 10% [vol/vol] glycerol, and 100–200 µg of GST-HRV3C protease). Augmin complexes were eluted from IgG beads by protease cleavage overnight at 4°C. On the following day, 3C protease was removed by GSTrap column (GE Healthcare). 1 ml Ni-NTA agarose slurry (QIAGEN) was added to the flow-through fraction of GSTrap column and incubated for 1 h at 4°C. Ni-NTA beads were collected with Poly-Prep chromatography columns (Bio-Rad) and washed with 10 ml of Ni-NTA wash buffer (20 mM sodium phosphate, pH 7.4, 200 mM NaCl, and 20 mM imidazole). Augmin complexes were eluted by Ni-NTA elution buffer (Ni-NTA wash buffer with 300 mM imidazole), followed by gel filtration chromatography using Superdex 200 increase 10/300 GL (GE Healthcare) column equilibrated with TBS, pH 7.4, to confirm complex formations or by dialysis overnight at 4°C with CSF-XB buffer containing 10 mM Hepes, pH 7.7, 100 mM KCl, 1 mM MgCl₂, 5 mM EGTA, and 10% (wt/vol) sucrose. Augmin complexes in CSF-XB buffer were flash-frozen with liquid nitrogen and stored at −80°C for functional assays. The void volume of the gel filtration column (8.73 ml) was determined by running 2 mg/ml blue dextran 2000 (GE Healthcare). Protein concentrations were measured at UV 280 nm by NanoDrop (Thermo Fisher Scientific) using the extinction coefficient corresponding to each protein complex.

Rabbit polyclonal antisera against Xenopus H1 (1–861 bp), H2 (1–669 bp), H3 (1–1794 bp), H6 (1873–2973 bp), and H8 (1–1,118 bp) were generated by custom antibody production service (GenScript). For antibody purification from antisera, antigens were conjugated to Affi-gel 10 or 15 (Bio-Rad) overnight at 4°C in coupling buffers according to the product instruction. On the following day, the antigen-coupled beads were washed with PBS, pH 7.4, and 0.1 M glycine, pH 2.5, followed by equilibration with PBS. Then, antisera were diluted in PBS (1:4) and incubated with antigen-coupled beads for 2 h at 4°C, followed by washing with PBS with 0.5 M NaCl. The beads were washed again with PBS to remove salts. Antibodies were eluted with 0.1 M glycine, pH 2.5, and the buffer quickly neutralized with 1 M Hepes, pH 7.7 (1:10 ratio).

Cytostatic factor (CSF)–arrested Xenopus egg extracts were prepared as described (Hannak and Heald, 2006). For branching MT nucleation assay, 6.5 µl of the extract was mixed with 1.5 µl of 1 µM augmin complexes (or CSF-XB buffer), 0.5 µl of 10 mM vanadate, 0.5 µl of 200 µM RanQ69L, 0.5 µl of 1 µM mCherry-EB1, and 0.5 µl of 2 µM Cy5-labeled porcine tubulin in 10 µl reaction volume. The extract mixture was introduced into the flow cell as described previously (King and Petry, 2016). Images were collected on a Nikon TiE microscope using a 100× 1.49 NA oil-immersion TIRF objective and three laser channels(488, 561, and 641 nm). Images were acquired on Nikon elements software with an Andor Zyla sCMOS camera. Images were collected every 2 s for 20–30 min.

150 µl Dynabeads Protein A (Life Technologies) were equilibrated with TBS-T and coupled with 35 µg of each antibody overnight at 4°C. Antibody-conjugated beads were washed three times with 150 µl TBS-T and two times with 150 µl CSF-XB. The beads were retrieved by magnetic separation rack and split into two tubes, which were sequentially mixed with 65 µl CSF extracts and placed on ice for 45 min for each round. Immunodepleted extracts were retrieved and transferred to another tube on ice.

After carrying out ID, beads were retrieved from the immunodepleted extracts and washed three times with 150 µl TBS-T. Immunoprecipitated samples were eluted from beads with 0.1 M glycine, pH 2.5. Proteins in the eluates were identified and quantified by the Thermo Fisher Scientific Center for Multiplexed Proteomics at Harvard Medical School (Boston, MA) after sixplex isobaric tandem mass tag (Thermo Fisher Scientific) labeling. Quantified peptides were normalized to the summed signal across all samples, which were used to generate the heatmap and hierarchical clustering using the “heatmaply” package of R.

After running SDS-PAGE gels with immunodepleted extracts or IP samples, proteins were transferred to nitrocellulose membranes, which were incubated with blocking buffer containing TBS-T and 10% (wt/vol) fat-free dry milk for 30 min, followed by incubation with primary antibodies that were used for ID and IP in blocking buffer for 1 h at RT or 4°C overnight. Membranes were washed three times for 15 min each with blocking buffer and then incubated with HRP-conjugated secondary anti–mouse or anti–rabbit IgG antibodies (GE Healthcare) in blocking buffer (1:3,000 dilution) for 1 h at RT or 4°C overnight. Membranes were washed three times for 15 min each with TBS-T. ECL solutions were added to the membranes for chemiluminescence detection. For the detection of γ-TuRC components, the following commercial antibodies were used: anti-NEDD1 antibody (ab57336; Abcam), anti–γ-TB antibody (T6557; Sigma), anti-GCP4 antibody (sc-271876; Santa Cruz Biotechnology), and anti-GCP5 antibody (sc-365837; Santa Cruz Biotechnology).

Each augmin subunit with Strep-GFP tag was expressed alone and coexpressed with each of the other seven subunits by baculovirus coinfection to 50 ml of insect cells (1.5–2.0 × 10⁶). Cells were harvested by centrifugation at 3,000 g for 20 min, followed by resuspension with the lysis buffer described above. Cells were lysed by sonication, and lysates were cleared by centrifugation at 17,000 g for 30 min. 100 µl Strep-Tactin (IBA) slurry was added to each supernatant and mixed with rotator for 2 h at 4°C. Beads were collected by centrifugation at 4,000 g for 3 min, followed by three washes with 10 ml TBS. Samples were eluted by boiling with SDS-PAGE sample buffer and loaded to a SDS-PAGE gel (4–12% polyacrylamide gradient).

Augmin complexes (150 nM) were added back to the immunodepleted extract, followed by branching MT nucleation assays as described above. For IP experiments, anti-GFP antibody (ab290; Abcam)–conjugated Dynabeads were prepared as for ID experiments. Augmin complexes (150 nM) were added to the ID extract (15 µl) to make up 20 µl of samples and incubated for 30 min on ice. Anti-GFP antibody conjugated Dynabeads (20 µl) were retrieved and mixed with the extract samples. The extract sample and beads were mixed for 1 h by rotation at 4°C. The beads were collected from the extract sample and washed three times with 150 µl TBS-T. Samples were eluted by boiling with SDS-PAGE sample buffer, followed by Western blot.

20 µM bovine tubulin was mixed with 2 µM biotinylated porcine tubulin, 2 µM Cy5-tubulin, and 1 mM GMPCPP in BRB80 buffer (80 mM K-Pipes, pH 6.8, 1 mM EGTA, and 1 mM MgCl₂) on ice. The tubulin mixture was incubated at 37°C for 1 h. The polymerized GMPCPP MT seeds were diluted with BRB80 (1:400) and mixed with 150 nM augmin complexes. Dimethyldichlorosilane-treated coverslips were used to make flow cells. 0.1 mg/ml anti-biotin antibody was applied to the coverslip and unattached antibodies were washed out with 50 µl BRB80, followed by blocking with 1 mg/ml kappa-casein. After washing with 50 µl BRB80, MT/augmin mixtures were attached to the coverslip via anti-biotin antibodies. After a final wash with BRB80 supplemented with oxygen scavenging system (Trolox, PCA, and PCD; Aitken et al., 2008), the sample was imaged.

10 ml of Xenopus egg extract was diluted with 5 vol of CSF-XB buffer supplemented with 1 mM GTP and Complete EDTA-free protease inhibitor (Roche), followed by spinning at 3,000 g for 10 min at 4°C to pellet any large particles. The supernatant was further diluted twofold with CSF-XB buffer containing 1 mM GTP and filtered through a 22-µM filter. γ-TuRC was precipitated from the filtered extract by addition of 6.5% (wt/vol) polyethylene glycol 8000 and incubated on ice for 30 min. After centrifugation for 20 min at 17,500 g at 4°C, the polyethylene glycol pellet was resuspended in 20 ml CSF-XB buffer with 0.05% NP-40. The resuspended pellet was centrifuged at 136,000 g at 4°C for 7 min. The supernatant was then precleared with recombinant protein A Sepharose beads (GE Healthcare) for 20 min at 4°C. The beads were removed, 1 ml γ-tubulin antibody (1 mg/ml) was added to the sample, and the sample was incubated at 4°C for 2 h with rotation. After this, 1 ml of washed recombinant Protein A Sepharose beads was incubated with the sample on the rotator for 2 h. The sample was then centrifuged at 500 g for 5 min and the flow-through removed. The beads containing bound γ-TuRC were washed twice with 10 ml CSF-XB buffer with 0.05% NP-40 and poured onto a column. The beads were then washed with CSF-XB buffer containing 250 mM KCl (30 ml), with CSF-XB buffer with 1 mM ATP to 10 ml remove heat-shock proteins, and finally with 10 ml of CSF-XB buffer. 1 ml γ-tubulin peptide (residues 412–451) at 0.4 mg/ml in CSF-XB buffer was applied to the column and allowed to incubate overnight. The following day, additional CSF-XB buffer was added to the column and fractions 1–5 (0.5 ml each) were collected. All fractions contained γ-TuRC, with fractions 1–3 having the highest concentration of γ-TuRC.

Sucrose was dissolved in CSF-XB buffer (without sucrose) to prepare light and heavy solutions (10–50% or 15–45% [wt/vol]). Continuous sucrose gradient was made by mixing the two solutions using Gradient Master (Biocomp). After loading samples, differential centrifugation was performed using TLS-55 rotor (Beckman Coulter) at 200,000 g for 4 h at 4°C. Samples were fractionated and loaded on SDS-PAGE gels, followed by Western blot analyses to determine γ-TuRC containing fractions.

Purified bovine brain α, β-tubulin was purchased (PurSolutions). γ-TuRC obtained after sucrose gradient fractionation was concentrated and exchanged into BRB80 buffer using an Amicon Ultra-4 ml 10,000 NMWL centrifugal filter units. Before use in the assay, unlabeled α, β-tubulin + 10% Alexa Fluor 568–labeled porcine brain tubulin, BRB80 buffer and 1.5 mM GTP was combined and clarified by centrifugation at 80,000 rpm (TLA100.1 rotor; Beckman Coulter) for 20 min at 2°C. A sample of the clarified tubulin mixture was reserved for confirmation of final tubulin concentration in the reaction. The final concentration used in the nucleation reactions was similar for each experiment and ranged from 10 µM to 12 µM. The clarified tubulin mixture was combined with γ-TuRC with or without recombinant augmin at described concentrations to a final volume of 20 µl. The solution was mixed on ice, allowed to rest for 2 min, and then incubated at 37°C for 5 min. The nucleation mixture was then diluted with 80 µl warm BRB80, terminated with the addition of 100 µl of 2% glutaraldehyde in BRB80 buffer (for final glutaraldehyde concentration of 1%), and left at RT for 5 min. The samples were then diluted 10-fold in BRB80 buffer, layered on top of 5 ml cushion of 20% (vol/vol) glycerol in BRB80 buffer prepared in a 15 ml Corex tube fitted with a custom insert to support a round poly-lysine–coated coverslip. The sample was centrifuged for 45 min at 25,000 g in an HB-6 rotor at 4°C. After centrifugation, the cushion was removed and ice-cold methanol was added to the tube. The coverslip was mounted in Prolong Diamond and imaged with TIRF microscopy on Nikon TiE microscope using 100× 1.49 NA oil immersion TIRF objective. The images were acquired with Andor Zyla sCMOS camera. All images from each experiment were taken with the same TIRF angle and the same laser exposure during the same imaging session. At least 20 random fields of equal size were imaged for each sample, and representative images are shown.

10–50 nM augmin complexes (5 µl) were introduced onto carbon EM grids and stained with 2% uranyl acetate. The negative-stain data for holocomplex, T-II, and T-III were collected using a FEI Tecnai F20 Twin transmission electron microscope operating at 120 keV at a nominal magnification of 50,000 with a defocus ranging from −2 to −3 µm. A total of ∼200 images for each sample were automatically recorded on a Gatan 4k × 4k-pixel charge-coupled-device camera using the LEGINON data collection software (Suloway et al., 2005). The calibrated pixel size of the recorded image is 2.1 Å.

The contrast transfer function was estimated using CTFFIND3 program within the APPION processing environment (Mindell and Grigorieff, 2003; Lander et al., 2009). The particles were automatically picked from the raw micrographs using a template-based method in Gautomatch program (http://www.mrc-lmb.cam.ac.uk/kzhang/Gautomatch/). Particle images stacks were generated by Relion-2 program featuring GPU acceleration (Kimanius et al., 2016). The reference-free 2D classification of the particle image stack was also performed in Relion-2 using 50 classes and a regularization parameter T = 2.

GMPCPP MT seeds were made as described above, spun down at 17,000 g for 20 min at RT, and resuspended with 100 µl of BRB80 mixed with 20 nM augmin holocomplex or octamer(H6ΔC). The MT/augmin mixture (5 µl) was applied to EM grids and stained likewise. The negative-stain data were collected with CM100 TEM (Philips) at 80 keV at a magnification of 64,000 and 130,000, respectively, for octamer(H6ΔC) and holocomplex with a defocus ranging from −2 to −4 µm. Images were recorded using ORCA camera with a calibrated pixel size of 7.0 Å.

Anti-GFP antibody–conjugated Dynabeads were preparedas in the ID experiments. 100 nM of each augmin complex was mixed with 10 nM γ-TuRC and the beads (20 µl) using a rotator for 2 h at 4°C. After incubation, beads were retrieved from solution and washed with three times of 150 µl TBS-T. Samples were eluted by boiling with SDS-PAGE sample buffer, followed by Western blot.

For MT cosedimentation assays, 40 µM bovine tubulin was mixed with 1 mM DTT and 1 mM GTP in BRB80 and incubated at 37°C for 30 min. 50 µM Taxol was then added to the mixture, which was incubated for another 30 min. MTs were pelleted at 17,000 g for 20 min at RT and resuspended with BRB80Tx buffer (BRB80 with 50 µM Taxol). γ-TuRC and augmin were spun at 100,000 g for 15 min in the presence of 45% (wt/vol) sucrose cushion, and supernatants were collected for the following assay. MTs (2, 0.4, and 0.08 µM) were mixed respectively with 5 nM γ-TuRC and 5 nM γ-TuRC/50 nM augmin. 20-µl samples were loaded onto 80 µl of 45% (wt/vol) sucrose/BRB80Tx cushion and spun down at 100,000 g for 15 min at 22°C using a TLA-100 (Beckman Coulter) rotor. Supernatants were 30 µl from the top and pellets were 30 µl from the bottom of the tubes. Samples were loaded onto SDS-PAGE gels, followed by Western blot analyses.

MTs were prepared in the same ways as previously mentioned. All reactions and buffers were at RT unless otherwise specified, and all incubations were performed in a humidity chamber. 60 nM MTs were mixed with 75 nM augmin holocomplex and 5 nM γ-TuRC in a 10-µl reaction at RT for 10 min. The mixture was then flown into a flow channel and allowed to adhere for 5 min. Unattached MTs were washed out by 20 µl BRB80, followed by fixation with −20°C methanol for ∼1 min and then by another wash in 20 µl of BRB80. The coverslips were washed with 50 µl blocking buffer (5% normal goat serum S1000 in BRB80; Vector Labs) and incubated for 1 h at 4°C. Reactions were incubated overnight at 4°C in blocking buffer with custom rabbit polyclonal anti–γ-TB at 1:100 dilution. The following day, reactions were subjected to three rounds of washing with 50 µl BRB80 each followed by a 15-min incubation. Reactions were incubated for 1 h in blocking buffer containing goat anti–rat IgG (H+L) secondary antibody and Alexa Fluor 568 conjugate (Thermo Fisher Scientific) at 1:2,500 dilution. Again, reactions were washed with 50 µl BRB80 three times, and each wash was followed by a 15-min incubation. Finally, reactions were mounted with ProLong Diamond Antifade Mountant (Life Technologies) and stored until imaging.

The images were acquired on Nikon TiE TIRF microscope using 1.49 NA oil-immersion objective and captured with an Andor Zyla sCMOS camera. To avoid bleaching in the red and infrared channels, random fields were located in the green channel and then imaged in all three channels using a multichannel acquisition function in the software. All adjustable imaging parameters (exposure time, laser intensity, and TIRF angle) were kept the same between reactions and experiments. Images from all experiments were collected in a single session.

Images were analyzed using MATLAB. To segment MTs, tubulin signal was first thresholded via Otsu method and further dilated to create a MT mask. MTs were isolated from the mask by finding connected components in the binary image, whereas small aggregates and false positives were eliminated by setting the minimum perimeter as 100 pixels. Mean γ-tubulin signal on each MT per pixel was calculated. The mean intensity from reverse mask of the entire field of view was subtracted from each MT and its corresponding γ-tubulin signal as background. Each experiment was repeated thrice, and at least 30 MTs were analyzed for each condition. Data from all three experiments were pooled and reported. Boxplots were generated using notBoxPlot function in MATLAB Central File Exchange (https://www.mathworks.com/matlabcentral/fileexchange/26508-notboxplot). P-values were calculated with the Wilcoxon rank-sum test in MATLAB.

The online supplemental material contains the quantification of MT nucleation kinetics in IgG and augmin add-back conditions (Fig. S1), which belong to Fig. 1 B. In addition, more experiments depict the purification of γ-TuRC, the weak interaction between augmin and γ-TuRC in solution, and the fact that augmin can target γ-TuRC to single MTs in vitro (Fig. S2 for Fig. 2). We show that augmin alone does not induce MT nucleation when added to extract or purified γ-TuRC (Fig. S3). All SDS-PAGE gels of the in vitro pull-down assays are depicted (Fig. S4 for Fig. 3). Finally, IP and IF experiments show that holocomplex, octamer, and T-III, but not T-II, bind to g-TuRC and recruit it along preexisting MTs (Fig. S5 for Fig. 5). Fig. S6 shows negative-stain analysis of augmin. Video 1 depicts that recombinant augmin can replace endogenous augmin in Xenopus egg extracts (for Fig. 1 B). In Video 2, different augmin complexes were added back to extracts (for Fig. 4 C).

We would like to thank Raymundo Alfaro-Aco for help in purifying γ-TuRC and Elbegduuren Erdenee for help in testing augmin subunit interactions. We are grateful to Dr. Martin Wuehr for support with the quantitative mass spectrometry experiments and Dr. Eva Nogales for the use of the University of California Berkeley EM facility. We also thank Dr. Gohta Goshima for critical reading of the manuscript and helpful comments.

This work was supported by a National Institutes of Health DP2 New Innovator Award (National Institute of General Medical Sciences,1DP2GM123493-01), the Pew Scholars Program in the Biomedical Sciences, and the David and Lucile Packard Foundation (all to S. Petry). R.S. Kadzik was supported by an National Institutes of Health postdoctoral fellowship (1F32GM119195-01) and A. Thawani was supported by an American Heart Association Graduate Research Fellowship (17PRE33660328).

The authors declare no competing financial interests.

Author contributions: J.-G. Song designed and conducted the experiments. M.R. King purified augmin antibodies and carried out IF experiments and quantification of the data (Fig. 2, C and D). R. Zhang performed negative staining EM and 2D class averaging (Fig. 5 C). R.S. King purified endogenous γ-TuRC and performed in vitro MT nucleation assays (Figs. S2 B and S3 B). A. Thawani analyzed MT nucleation kinetics (Fig. S1) and contributed to the quantification of IF data (Fig. 2 D). S. Petry contributed to the experimental design and oversaw the work. J.-G. Song and S. Petry wrote the manuscript. All authors reviewed the data and the manuscript.