NuMA Targets Dynein to Microtubule Minus-Ends at Mitosis

To build the spindle at mitosis, motors exert spatially regulated forces on microtubules. We know that dynein pulls on mammalian spindle microtubule minus-ends, and this localized activity at ends is predicted to allow dynein to cluster microtubules into poles. How dynein becomes enriched at minus-ends is not known. Here, we use quantitative imaging and laser ablation to show that NuMA targets dynactin to minus-ends, localizing dynein activity there. NuMA is recruited to new minus-ends independently of dynein and more quickly than dynactin, and both NuMA and dynactin display specific, steady-state binding at minus-ends. NuMA localization to minus-ends requires a C-terminal region outside NuMA’s canonical microtubule binding domain, and it is independent of direct minus-end binders γ-TuRC, CAMSAP1, and KANSL1/3. Both NuMA’s minus-end-binding and dynein-dynactin-binding modules are required to rescue focused, bipolar spindle organization. Thus, NuMA may serve as a mitosis-specific minus-end cargo adaptor, targeting dynein activity to minus-ends to cluster spindle microtubules into poles.


INTRODUCTION
). Thus, it remains unclear whether dynein-dynactin and NuMA have specific binding sites at minus-ends, and if so, whether they are recruited by known minus-end binders. Finally, knowing how dynein is targeted to minus-ends would make it possible to test the in vivo role of minus-end-localized -compared to indiscriminately-localized -forces in spindle organization.
Here, we use laser ablation to create new, isolated minus-ends in the mammalian spindle in vivo, and quantitative imaging to map protein recruitment to these ends and its mechanistic basis. We demonstrate that NuMA binds at minus-ends independently of dynein, and that NuMA targets dynactin -and thereby dynein activity -to spindle minus-ends. This challenges the prevailing model that dynein delivers NuMA to spindle minus-ends and poles (Merdes et al., 2000;Radulescu & Cleveland, 2010). NuMA localization to minus-ends is independent of known direct minus-end binders g-TuRC, CAMSAP1, and KANSL1/3, and it requires both NuMA's canonical microtubule-binding domain and an additional region of its Cterminus. Thus, NuMA -which is sequestered in the nucleus at interphase (Lydersen & Pettijohn, 1980) -may serve as a mitosis-specific minus-end cargo adaptor, recruiting dynein activity to spindle minus-ends. Both NuMA's minus-end-binding domain and dynactin-binding domain are required for correct spindle architecture, supporting long-standing in silico predictions that localizing dynein to minus-ends enables effective clustering of parallel microtubules into poles. These findings identify a mechanism for mitosis-specific recruitment of dynein to microtubule minus-ends and, more broadly, illustrate how spatial regulation of local forces may give rise to larger-scale cytoskeletal architectures.

Dynactin and NuMA display mitosis-specific, steady-state binding at microtubule minus-ends.
To visualize the spatial targeting of the dynein-dynactin complex to microtubule minusends -which are normally buried in dense mammalian spindles -we used nocodazole washout and laser ablation to create resolvable minus-ends in mitotic cells. First, to determine whether dynein-dynactin and NuMA localize to individual microtubule minus-ends, we treated mammalian PtK2 and RPE1 cells with the microtubule-depolymerizing drug nocodazole and fixed cells 6-8 min after drug washout to capture acentrosomal microtubules with clearly visible plus-and minus-ends ( Figure 1A). p150 Glued (p150, a dynactin subunit) and NuMA strongly colocalized at the minus-ends of these individual microtubules, with a clear binding preference for minus-ends over the microtubule lattice or the plus-end, which is marked by EB1 ( Figure 1B). Interestingly, in prophase cells before nuclear envelope breakdown, p150 localized predominantly to plus-ends rather than minus-ends ( Figure 1B; Figure 1-figure supplement 1), consistent with dynactin's interphase localization (Vaughan et al., 1999). Thus, nuclear envelope breakdown (NEB) confers dynactin's preference for minus-ends, suggesting regulated, mitosis-specific spatial targeting.
Second, we sought to test whether dynactin and NuMA have finite binding sites at minus-ends by measuring the kinetics of their recruitment. To do so, we used laser ablation of k-fibers in PtK2 cells to create new minus-ends within the spindle body ( Figure 1C-E). By spatially and temporally synchronizing the creation of a bundle of minus-ends, laser ablation allowed for dynamic measurements of the recruitment to minus-ends of GFP-tagged dynactin (Arp1A) and NuMA and comparison to a direct minus-end binding protein, CAMSAP1. Dynactin, NuMA, and CAMSAP1 robustly recognized new microtubule minus-ends within the spindle ( Figure 1C-E). The kinetics of dynactin and NuMA recruitment to minus-ends were distinct from CAMSAP1, which reached a max intensity after approximately 5-10 s, but then decreased in intensity as if its binding sites at the minus-end were gradually obstructed ( Figure 1E-G). Unlike CAMSAP1, dynactin and NuMA intensities reached a stable plateau, suggesting that binding saturates, reaching a steady-state. This finite binding is consistent with a truly end-specific localization (rather than localization along the unbounded lattice near the minus-end). In addition, the rate of NuMA or dynactin accumulation at new minus-ends did not correlate with the length of k-fiber stubs created by ablation ( Figure 1H), which could indicate that recruitment rate is set by the number of individual microtubule minus-ends (which is similar across k-fibers (McEwen et al., 1997)) rather than k-fiber length.
Interestingly, NuMA intensity at new minus-ends increased at a faster rate and saturated sooner than dynactin ( Figure 1F,G). This observation hints at a model in which NuMA targets dynein-dynactin to minus-ends, and it is less easily consistent with the idea that dynein delivers NuMA to minus-ends. If true, this hypothesis would explain why dynactin's minus-end preference arises after NEB, when NuMA is released from the nucleus into the mitotic cytoplasm.

NuMA localizes to minus-ends independently of dynein.
The finding that NuMA localizes to minus-ends more quickly than dynactin could suggest that NuMA directly or indirectly recognizes the exposed a-tubulin structure of minus-ends and subsequently recruits dynein-dynactin ('Structural Recognition,' Figure 2A). Alternatively, dynein-dynactin could walk toward minus-ends, carrying NuMA, and pile up or dwell there ('Walk and Pile-up,' Figure 2B). To test the 'Walk and Pile-up' model, we inhibited dyneindynactin in PtK2 spindles by over-expressing p50 (dynamitin), resulting in fully unfocused kfiber arrays ( Figure 2C). The 'Walk and Pile-up' model predicts that in the absence of dyneindynactin transport, NuMA -which requires dynactin for its interaction with dynein (Merdes et al., 2000) -should not reach minus-ends. Instead, GFP-NuMA was robustly recruited to minusends created by k-fiber ablation ( Figure 2C), indicating that NuMA can localize to minus-ends independently of dynein. Similarly, NuMA was recruited to ablation-created minus-ends when dynein was inhibited by p150-CC1 over-expression (Figure 2-figure supplement 1A,B). Thus, NuMA localizes to minus-ends without dynein carrying it there -consistent with early observations in extract asters and spindles (Gaglio et al., 1996;Heald et al., 1997) but in contrast to the prevailing view that dynein delivers NuMA to minus-ends (Merdes et al., 2000;Radulescu & Cleveland, 2010). Slower NuMA accumulation kinetics after p50 overexpression suggest that dynein-dynactin-NuMA complex formation may aid rapid NuMA recruitment, but dynein 'Walk and Pile-up' alone cannot explain NuMA's minus-end affinity ( Figure 2D,E).
To confirm that NuMA localizes to minus-ends even after genetic dynein deletion, we used an inducible CRISPR/Cas9 HeLa cell line to knock out dynein's heavy chain (DHC) (McKinley & Cheeseman, 2017). After dynein knockout (Figure 2-figure supplement 1C), both NuMA and dynactin localized robustly to k-fiber minus-ends ( Figure 2F). Together, these data indicate that NuMA can bind at minus-ends in the absence of dynein and are consistent with NuMAmediated recognition of minus-end structure ('Structural Recognition,' Figure 2A). In addition, they suggest that NuMA may interact with and target dynactin to minus-ends even in the absence of dynein, when dynein-dynactin complexes cannot form and dynein motility cannot deliver dynactin close to minus-ends.

NuMA is required for dynein activity and dynactin localization at minus-ends.
If NuMA does target dynein-dynactin to minus-ends, localizing force there, we would expect NuMA to be required for dynein to transport minus-ends created by k-fiber ablation (Elting et al., 2014;Sikirzhytski et al., 2014). To test this hypothesis, we made inducible CRISPR/Cas9 RPE1 cell lines to knock out NuMA (Figure 3-figure supplement 1) (McKinley et al., 2015;T. Wang et al., 2015). Indeed, when NuMA was knocked out -causing elongated, heterogeneously disorganized spindles with detached centrosomes -ablation-created minusends were no longer transported toward poles by dynein ( Figure 3A-C). These data are consistent with a previous finding that NuMA antibody injection prevents distal k-fiber looping events in monopolar spindles, in which dynein likely pulls on free k-fiber minus-ends (Khodjakov et al., 2003). Thus, NuMA is required for dynein activity at spindle microtubule minus-ends.
Dynein activity at minus-ends could require NuMA because NuMA modulates dyneindynactin's ability to pull on microtubules, or because NuMA localizes dynein to minus-ends. Given the findings in Figures 1-2, we hypothesized that NuMA localizes dynein activity by recruiting dynactin to minus-ends. Indeed, after NuMA knockout, dynactin (GFP-Arp1A) was no longer detectable at k-fiber minus-ends created by laser ablation ( Figure 4A). Immunofluorescence experiments confirmed that in the absence of NuMA, dynactin remained at kinetochores but no longer localized to minus-ends ( Figure 4B). Interestingly, dynein's localization within the spindle (labeled using an antibody against dynein intermediate chain) was less minus-end-specific than dynactin's, both before and after NuMA deletion (Figure 4figure supplement 1; Figure 4B).
In addition, nocodazole washout experiments revealed that dynactin (p150) localization to individual microtubule minus-ends was lost upon NuMA deletion ( Figure 4C). Instead, dynactin frequently co-localized with EB1 at plus-ends, similar to its interphase and pre-NEB localization pattern ( Figure 4C; Figure 1B). Thus, the data indicate that NuMA is required for the transport of minus-ends by dynein because NuMA binds at minus-ends and recruits dynactin there.

NuMA recognizes minus-ends independently of known minus-end binders.
How is NuMA targeted to minus-ends, independently of dynein? In vitro, canonical microtubule-binding regions of NuMA have shown no preference for minus-ends relative to the lattice or plus-end of purified microtubules (Du et al., 2002;Forth et al., 2014;Haren & Merdes, 2002;Seldin et al., 2016). Given this lack of minus-end-specific binding in vitro, we hypothesized that NuMA is indirectly recruited to spindle minus-ends by one of three known direct minusend binders active at mitosis: g-TuRC (Zheng et al., 1995), CAMSAP1 (Hendershott & Vale, 2014;Jiang et al., 2014) (Figure 1; CAMSAP2 and 3 are phosphorylated at mitosis and no longer interact with microtubules (Jiang et al., 2014)), and KANSL1/3 (Meunier et al., 2015). To test this hypothesis, we treated RPE1 cells with 30 µM gatastatin to block g-TuRC binding (Chinen et al., 2015), resulting in short spindles with fewer astral microtubules ( Figure 5A; Figure 5-figure supplement 1A). We also made inducible CRISPR/Cas9 RPE1 cell lines to knock out CAMSAP1 or KANSL1 ( Figure 5-figure supplement 1B,C), as KANSL1 depletion has been shown to delocalize the entire KANSL complex (Meunier et al., 2015). CAMSAP1 knockout caused a small reduction in spindle length (Figure 5-figure supplement 1D), consistent with the spindle minus-end protecting function ascribed to its Drosophila homolog, Patronin (Goodwin & Vale, 2010). To our surprise, however, none of these perturbations qualitatively altered NuMA localization at spindle poles ( Figure 5A). To check for a subtler contribution of g-TuRC, CAMSAP1, or KANSL1 to NuMA localization at minus-ends, we performed k-fiber ablations after 30 µM gatastatin treatment, CAMSAP1 knockout, or KANSL1 knockout and quantified GFP-NuMA recruitment kinetics at new minus-ends. NuMA recruitment to new minus-ends remained robust, and recruitment timescales were statistically indistinguishable from control ( Figure 5B-C). Thus, the data indicate that the direct mitotic minus-end binders g-TuRC, CAMSAP1, and KANSL1/3 are not responsible for NuMA's localization to spindle microtubule minus-ends.
Given this lack of involvement of direct minus-end binding proteins, we sought to confirm using super-resolution microscopy that NuMA specifically localizes at individual spindle microtubule minus-ends. Three-dimensional stochastic optical reconstruction microscopy (3D STORM (Huang et al., 2008;Rust, Bates, & Zhuang, 2006)) of NuMA at k-fiber minus-ends created by ablation revealed organized clusters of NuMA puncta, rather than a lawn of molecules along the lattice near the minus-end ( Figure 5D). The spacing between puncta was consistent with the distance between individual microtubules within k-fibers ( Figure 5E), as measured by electron microscopy in the same cell type (PtK2) (McDonald et al., 1992), consistent with the idea that these NuMA puncta are built on individual microtubule minusends. A yet undiscovered minus-end binding protein may recruit NuMA; alternatively, NuMA may have direct minus-end-specific binding ability that has not been recapitulated in vitro.

NuMA function requires both minus-end-recognition and dynactin-recruitment modules.
To define the NuMA domain required for spindle minus-end localization, we performed rescue experiments with different NuMA truncations in the NuMA-knockout background ( Figure 6A). This avoids the C-terminus-mediated oligomerization (Harborth et al., 1999) with endogenous protein that has complicated interpretations of previous work. In this NuMAknockout background, full-length NuMA ('FL') localized to spindle minus-ends, as did a bonsai construct ('N-C') with most of the central coiled-coil removed ( Figure 6B). To our surprise, 'N-C' rescued spindle architecture as effectively as full-length NuMA, indicating that the extraordinary length of NuMA protein is not essential to its function in spindle structure ( Figure  6C). Importantly, NuMA's C-terminus ('C') alone localized to minus-ends even in the absence of endogenous full-length protein ( Figure 6B). Its intensity at minus-ends was less striking than that of full-length or N-C protein at poles, likely because minus-ends are not as densely concentrated in disorganized spindles, and perhaps NuMA's N-terminus facilitates formation of higher order NuMA assemblies. However, the preference of NuMA 'C' for spindle ends was clear. Because dynein-dynactin interacts with NuMA's N-terminus (Kotak, Busso, & Gönczy, 2012), the minus-end localization of NuMA's C-terminus provides further support for dyneinindependent minus-end recognition.
Importantly, NuMA's C-terminus ('C') localized to minus-ends but could not rescue proper pole focusing or spindle architecture, unlike full-length protein or N-C ( Figure 6C). This suggests that NuMA's function in spindle organization requires both minus-end binding (via its C-terminus) and the ability to recruit dynactin to minus-ends (via its N-terminus) (Kotak et al., 2012). Consistent with this hypothesis, NuMA's C-terminus ('C') was unable to recruit dynactin to minus-ends, while its N-terminus and C-terminus fused ('N-C') did ( Figure 6E; Table 1).
To test whether recruiting dynein-dynactin to the sides of microtubules was sufficient for proper microtubule clustering into poles, we fused NuMA's N-terminus to the Tau microtubule-binding domain ('N-Tau'). N-Tau localized along the length of spindle microtubules and recruited dynactin there ( Figure 6E). Despite combining dynein-dynactin binding and microtubule lattice binding, N-Tau was not enriched at minus-ends and was unable to rescue spindle architecture ( Figure 6C; Table 1).
Like a traditional cargo adaptor, NuMA may target force to spindle minus-ends using a cargo (minus-end) binding module ('C') and a dynactin-recruitment module ('N'). Furthermore, the inability of N-Tau to rescue spindle architecture in the absence of endogenous NuMA suggests that specifically targeting dynein-dynactin to minus-ends, not just all along spindle microtubules as N-Tau does, is critical for organizing a focused, bipolar spindle.

NuMA targets dynactin to minus-ends, spatially regulating dynein activity at mitosis.
Our data indicate that NuMA localizes to spindle microtubule minus-ends independently of dynein and minus-end binding proteins g-TuRC, CAMSAP1, and KANSL1/3. NuMA then targets dynactin to minus-ends, localizing dynein motor activity there. In addition, targeting dynein to the end of its track could permit amplification by motor pile-up, as NuMA at minusends captures processive dynein complexes. Altogether, our findings are consistent with a model in which NuMA confers minus-end targeting of the dynein-dynactin complex upon nuclear envelope breakdown (NEB), when NuMA is released from the nucleus. Indeed, active microtubule clustering by dynein is first observed coincident with NEB (Rusan et al., 2002). Thus, NuMA may provide both spatial (minus-end-specific) and temporal (mitosis-specific) regulation of dynein-powered force ( Figure 7A).

Function of force at spindle microtubule minus-ends.
Models of spindle assembly in silico predict that enriching a minus-end-directed motor at microtubule ends can break gridlock between parallel microtubules and allow robust minusend clustering into poles (Burbank et al., 2007;Goshima et al., 2005;Surrey et al., 2001). However, a lack of mechanistic information and tools made it difficult to test such hypotheses in vivo. The present study reveals a mechanism for direct recruitment of the motor to minus-ends, via NuMA, and is consistent with the prediction that targeting dyneindynactin to the sides of microtubules is not sufficient for robust spindle organization. Both the minus-end-binding module ('C') and dynein-dynactin-binding module ('N') of NuMA are required for bipolar focusing, while the distance between them is not critical for function. Fusing the dynein-dynactin-binding module to the Tau microtubule binding domain is not sufficient, suggesting a requirement for minus-end-specific forces ( Figure 6; Table 1). However, we cannot formally exclude that features of the NuMA C-terminus (missing from 'N-Tau') other than minus-end binding enable rescue of pole focusing. Altogether, the work is consistent with a model for mammalian spindle organization in which targeting poleward force to microtubule minus-ends (by NuMA-mediated dynactin recruitment) is critical for organizing microtubules into a focused, bipolar metaphase spindle ( Figure 7B). More broadly, the emergence of pole architecture at mitosis illustrates how spatial regulation of molecular-scale activities (like NuMA localizing dynein activity at individual minus-ends) can give rise to complex and diverse cellularscale structures.

NuMA as mitotic dynein adaptor or activator.
The data suggest that NuMA in the mammalian spindle body may function as a traditional cargo adaptor, with a cargo (minus-end)-binding module and a motor-binding module. In the spindle body context, microtubule minus-ends are the cargo, just as the cortex can be framed as cargo for cortical dynein -analogous to more canonical interphase cargo like membranous vesicles or organelles. This framework raises the question of whether NuMA additionally serves as a dynein activator, inducing highly processive motility like interphase dynein adaptors can (McKenney et al., 2014;Schlager et al., 2014).
Dynein's localization in mammalian mitosis is more ubiquitous than that of dynactin ( . These observations could be explained if NuMA and dynactin at minus-ends selectively activate dynein therethrough a conformational shift that renders the motor more processive, for example -rather than simply selectively localizing it there. In other words, dynein's localization within the spindle body may be less tightly regulated than its activity. The loss of dynein activity at minusends observed after NuMA knockout ( Figure 3C) may stem from a lack of dynein activation (without NuMA and dynactin present at minus-ends) rather than a lack of dynein enrichment.

Figure 7. Model: NuMA spatially targets dynein activity to minus-ends at mitosis to ensure minus-end clustering into poles.
(A) Our findings suggest that NuMA confers minus-end targeting to the dynein-dynactin complex upon NEB, when NuMA is released from the nucleus. NuMA binding at minus-ends requires a region of its C-terminus (amino acids 1701-1981, in red) that includes tail1 in addition to lattice-binding domain tail2A. NuMA targets dynactin to minus-ends, localizing dynein motor activity.
(B) Without minus-end targeting, dynein molecules acting on the red microtubule oppose each other, resulting in gridlock (left). NuMA-mediated targeting of force to minus-ends allows for productive clustering of the red microtubule into the pole (right). Altogether, our data support a model for mammalian spindle organization in which targeting poleward force to microtubule minus-ends specifically -by NuMA-mediated dynactin recruitment -provides robust clustering of microtubules into a focused, bipolar spindle.  Table S1. The plasmid used to express sgRNAs under control of the hU6 promoter (pLenti-sgRNA) was a gift from T. Wang, D. Sabatini, and E. Lander (Whitehead/Broad/MIT). An RPE1 cell line containing doxycycline-inducible human codon-optimized spCas9 was a gift from I. Cheeseman (Whitehead/MIT) and was generated as described in (McKinley et al., 2015) using a derivative of the transposon described in (G. Wang et al., 2014). We infected this inducible-spCas9 RPE1 cell line with each pLenti-sgRNA as described in (T. Wang et al., 2015) using virus expressed in HEK293T cells and 10 µg/mL polybrene and selected with 6 µg/mL puromycin. For each targeted gene, we tested 3 independent sgRNA sequences, each of which generated indistinguishable spindle phenotypes (data not shown), and picked one line for subsequent studies. 4 days before each experiment, spCas9 expression was induced with 1 μM doxycycline hyclate.

MATERIALS AND METHODS
Live imaging and laser ablation. For live imaging, cells were plated on glass-bottom 35mm dishes coated with poly-D-lysine (MatTek Corporation, Ashland, MA) and imaged in a stage-top humidified incubation chamber (Tokai Hit, Fujinomiya-shi, Japan) maintained at 30°C and 5% CO2. To visualize tubulin, 100 nM siR-Tubulin dye (Cytoskeleton, Inc., Denver, CO) was added 2 h prior to imaging, along with 10 µM verapamil (Cytoskeleton, Inc.). Under these conditions, there was no detected defect in spindle appearance or microtubule dynamics. As described elsewhere (Elting et al., 2014), cells were imaged using a spinning disk confocal inverted microscope (Eclipse Ti-E; Nikon Instruments, Melville, NY) with a 100X 1.45 Ph3 oil objective through a 1.5X lens, operated by MetaMorph (7.7.8.0; Molecular Devices, Sunnyvale, CA). Laser ablation (30 3-ns pulses at 20Hz) with 551 nm light was performed using the galvo-controlled MicroPoint Laser System (Andor, Belfast, UK). For laser ablation experiments, images were acquired more slowly prior to ablation and more rapidly after ablation (typically 7 s prior and 3.5 s after ablation).

STORM.
PtK2 cells expressing GFP-a-tubulin (gift of A. Khodjakov, Wadsworth Center) were plated on photo-etched, gridded coverslips (G490; ProSciTech, Kirwan, Australia) coated with 1 mg/mL poly-L-lysine (P-1524; Sigma) and imaged at 29-30°C in a homemade heated aluminum coverslip holder using the confocal microscope and ablation system described above. 20-30 s after k-fiber ablation, imaging media was replaced with fixative (as above) chilled to -80°C, and the coverslip holder was placed on ice for 1 min. Cells were incubated with 3% BSA in PBS for 1 hr at RT, and then with primary antibodies overnight at 4°C. Secondary antibodies ((anti-mouse Cy3B; Jackson Immunoresearch, West Grove, PA); anti-rabbit AF647 (Life Tech, Carlsbad,CA)) were incubated for 30 min at RT. Antibody incubations were followed by 4 washes with 0.2% BSA in PBS. Samples were stored in PBS during confocal imaging, and coverslip grid was used to re-find the individual ablated cell. For 3D STORM imaging, samples were mounted on glass slides with a standard STORM imaging buffer consisting of 5% (w/v) glucose, 100 mM cysteamine, 0.8 mg/mL glucose oxidase, and 40 µg/mL catalase in 1M Tris-HCI (pH 7.5) (Huang et al., 2008;Rust et al., 2006). Coverslips were sealed using Cytoseal 60. STORM imaging was performed on a homebuilt setup based on a modified Nikon Eclipse Ti-E inverted fluorescence microscope using a Nikon CFI Plan Apo λ 100x oil immersion objective (NA 1.45). Dye molecules were photoswitched to the dark state and imaged using either 647-or 560-nm lasers (MPB Communications, Montreal, CAN); these lasers were passed through an acousto-optic tunable filter and introduced through an optical fiber into the back focal plane of the microscope and onto the sample at intensities of ~2 kW cm -2 . A translation stage was used to shift the laser beams towards the edge of the objective so that light reached the sample at incident angles slightly smaller than the critical angle of the glass-water interface. A 405-nm laser was used concurrently with either the 647-or 560-nm lasers to reactivate fluorophores into the emitting state. The power of the 405-nm laser (typical range 0-1 W cm -2 ) was adjusted during image acquisition so that at any given instant, only a small, optically resolvable fraction of the fluorophores in the sample were in the emitting state. Emission was recorded with an Andor iXon Ultra 897 EM-CCD camera at a framerate of 220 Hz, for a total of ~120,000 frames per image. For 3D STORM imaging, a cylindrical lens of focal length 1 m was inserted into the imaging path so that images of single molecules were elongated in opposite directions for molecules on the proximal and distal sides of the focal plane (Huang et al., 2008). Two-color imaging was performed via sequential imaging of targets labelled by AF647 and Cy3B. The raw STORM data was analyzed according to previously described methods (Huang et al., 2008;Rust et al., 2006).

Drug treatment and microtubule re-growth.
To inhibit g-tubulin, 30 µM gatastatin (gift of Takeo Usui and Ichiro Hayakawa, University of Tsukuba and Okayama University, respectively) (Chinen et al., 2015) was added 25 -60 min before imaging (from 30 mM stock). For microtubule re-growth (nocodazole washout) experiments, cells were treated with 5 µM nocodazole (M1404; Sigma) for 15 min at 37°C. After 3 washes, cells were incubated at room temperature for 6-8 min before fixation and immunofluorescence (as above).

Data Analysis.
To determine the percentage of p150 at plus-ends vs. minus-ends ( Figure 1B, Figure 4C), we used single microtubules where both ends were clearly visible, determined p150 localization relative to the EB1-labeled plus-end, and calculated the percentage of p150 at each location within each cell. Percentages for multiple cells were averaged for Figure 1B and 4C. Pre-NEB cells were distinguished from post-NEB cells by the exclusion of microtubules from the nucleus, circle-shaped chromosome packing in the nucleus, and, when possible, NuMA localization within the nucleus. Kymographs of GFP-Arp1A, GFP-NuMA, and GFP-CAMSAP1 puncta and pole position over time ( Figure 1C-E, Figure 2C, Figure 6D) were generated in ImageJ (Version 2.0.0/1.51h). To measure GFP intensity at ablation-created minus-ends over time ( Figure 1F, Figure 2D, Figure 5B), we used a home-written Matlab (R2012a Version 7.4) program to integrate GFP intensity within a 1.4 µm-diameter circle centered on the manually-tracked k-fiber minus-end, and to measure local background intensity within a surrounding 2.7 µm-diameter 'donut'. After background subtraction, the intensity measured at the cut site during the three frames before ablation (k-fiber intensity) was averaged and set to zero. For NuMA and Arp1A, we then fit a sigmoid function ( = # $%& '()'*) , , where y = intensity and x = time) to each trace, normalized to plateau height (a = 1), and solved for the time b at which y = 0.5*a to determine time to halfmaximum intensity ( Figure 1G, Figure 2E, Figure 5C). For CAMSAP1, we normalized each trace to peak height (mean intensity from t = 5 s to t = 20 s) and found the first point at which intensity passed 0.5 to determine time to half-maximum intensity. Finally, to generate mean intensity traces, data from all traces were collected into 5 s wide bins in time and their intensities were averaged. Stub length (distance between k-fiber plus-and minus-ends, Figure  1H) was measured in ImageJ at the first frame following ablation. Minus-end position data ( Figure 3C) were generated by manual tracking of ablationcreated k-fiber minus-ends (marked by GFP-CAMSAP1) and spindle poles in time-lapse videos, using a second home-written Matlab program. Nearest neighbor distances between NuMA puncta in STORM imaging ( Figure 5H) were measured as the center-to-center distance from each NuMA puncta to its nearest neighboring puncta. NuMA truncation rescue capability ( Figure 6C) reports the percentage of bipolar spindles with two focused poles compared to disorganized spindle architecture characteristic of NuMA knockout (detached centrosomes, loss of k-fiber focusing into two poles). Percentage was calculated for each experiment (n = 3-5 experiments) and then averaged.
Statistics. All data are expressed as average ± standard error of the mean (SEM). Calculations of correlation coefficients (Pearson's r) and p-values were performed in Matlab. All other p-values were calculated using two-tailed unpaired t-tests with GraphPad Software. Quoted n's are described in more detail in Figure Legends, but in general refer to individual measurements (biological replicates, e.g., individual spindle lengths, intensity measurements over time after an individual k-fiber ablation, etc.). Video preparation. Videos show a single spinning disk confocal Z-slice imaged over time and were formatted for publication using ImageJ and set to play at 35x relative to real time. Videos were corrected to play at a constant frame rate, even when the acquisition rate was not constant.

Figure 1 -figure supplement 1. Dynactin localization shifts from plus-ends to minus-ends upon nuclear envelope breakdown.
Immunofluorescence showing localization of p150 (dynactin subunit; magenta) and NuMA (red) within asters formed after washout of 5 µM nocodazole in RPE1 cells. Before nuclear envelope breakdown (pre-NEB), NuMA is sequestered in the nucleus and p150 is visible at aster plus-ends (arrowheads), facing outward. After nuclear envelope breakdown (post-NEB), both NuMA and p150 concentrate at aster centers (arrowhead) where minus-ends are. (Similar to Figure 1B, where individual microtubules in the cell periphery with resolvable plus-and minus-ends were analyzed instead of the asters we show here.) Scale bar, 5 µm. (A) Live images of GFP-α-tubulin immediately before and after k-fiber ablation at targeted sites (red 'X's) in a PtK2 cell in which dynein cargo-binding is inhibited by transfection of the dominant negative p150-CC1 fragment (Quintyne & Schroer, 2002). Scale bar, 5 µm. (B) Immunofluorescence image of NuMA (magenta) and α-tubulin (yellow) in cell from (A), fixed after ablation. NuMA (arrowheads) localizes to new minus ends. Scale bar, 5 µm.    (A) Mean spindle length ± SEM in control RPE1 cells compared to cells treated with 30 µM gatastatin (g-tubulin inhibitor) for 25 min. The mean length of gatastatin-treated spindles was reduced by 23%, very similar to previously reported measurements after g-tubulin inhibition (Chinen et al., 2015). n = 24 control cells; n = 25 gatastatin-treated cells from 3 separate experiments. ***P < 0.0001. (B-C) Western blot showing > 85% depletion (normalized to actin) of (B) CAMSAP1 and (C) KANSL1 protein at the cell population level after knockout (KO). Whenever possible, complete protein loss within individual cells analyzed was verified by immunofluorescence. (D) Mean spindle length ± SEM in control RPE1 cells (no doxycycline added) and CAMSAP1 knockout cells (after Cas9 induction by doxycycline). Complete CAMSAP1 loss was verified by immunofluorescence for all cells analyzed. n = 11 control cells; n = 12 CAMSAP1 knockout cells from one experiment. ***P < 0.001.

VIDEO LEGENDS
Video 1. Dynactin is robustly recruited to new spindle microtubule minus-ends. See also Figure 1C. Live confocal imaging of a PtK2 cell expressing GFP-Arp1A, with spindle microtubules labeled by siR-tubulin. GFP-Arp1A is recruited to k-fiber minus-ends (arrowhead) created by ablation ('X') and moves with them as dynein pulls them poleward (Elting et al., 2014). Time is in min:sec, with the frame captured immediately following ablation set to 00:00. Plus-end of ablated kfiber is marked by '*'. Scale bar, 5 µm.
Video 2. NuMA is robustly recruited to new spindle microtubule minus-ends. See also Figure  1D. Live confocal imaging of a PtK2 cell expressing GFP-NuMA, with spindle microtubules labeled by siR-tubulin. GFP-NuMA is quickly recruited to k-fiber minus-ends (arrowhead) created by ablation ('X') and moves with them as dynein pulls them poleward (Elting et al., 2014). Time is in min:sec, with the frame captured immediately following ablation set to 00:00. Plus-end of ablated k-fiber is marked by '*'. Scale bar, 5 µm.
Video 3. CAMSAP1 is robustly recruited to new spindle microtubule minus-ends. See also Figure 1E. Live confocal imaging of a PtK2 cell expressing GFP-CAMSAP1, with spindle microtubules labeled by siR-tubulin. GFP-CAMSAP1 is immediately recruited to k-fiber minus-ends (arrowhead) created by ablation ('X') and moves with them as dynein pulls them poleward (Elting et al., 2014). Time is in min:sec, with the frame captured immediately following ablation set to 00:00. Plus-end of ablated k-fiber is marked by '*'. Scale bar, 5 µm.
Video 4. NuMA quickly localizes to k-fiber minus-ends despite dynein-dynactin inhibition. See also Figure 2C. Live confocal imaging of a PtK2 cell expressing GFP-NuMA, in which dynein-dynactin is inhibited by overexpression of p50 (dynamitin) (Shrum et al., 2009). K-fibers are unfocused and splayed due to dynein-dynactin inhibition, but NuMA is still robustly recruited to k-fiber minus-ends (arrowheads) created by ablation ('X'). Time is in min:sec, with frame captured immediately following ablation set to 00:00. Plus-end of ablated k-fiber is marked by '*'. Scale bar, 5 µm.
Video 5. NuMA is required for transport of minus-ends by dynein. See also Figure 3A. Live confocal imaging of a RPE1 cell in which NuMA has been knocked out using an inducible Cas9 system, and GFP-CAMSAP1 is expressed to label minus-ends. After ablation ('X'), the kfiber minus-end (arrowhead) is not transported poleward by dynein and remains detached from the disorganized spindle. Time is in min:sec, with the frame captured immediately following ablation set to 00:00 s. Plus-end of ablated k-fiber is marked by '*'. Scale bar, 5 µm.
Video 6. Dynactin is not recruited to minus-ends in the absence of NuMA. See also Figure 4A. Live confocal imaging of a RPE1 cell in which NuMA has been knocked out using an inducible Cas9 system. In the absence of NuMA, 2xGFP-Arp1A does not localize to k-fiber minus-ends, and its recruitment is not detectable at new minus-ends (arrowheads) created by ablation ('X').
Note that mislocalized 2xGFP-Arp1A puncta diffuse randomly within the spindle, including within microns of the new minus-end, but do not remain there. Time is in min:sec, with frame captured immediately following ablation set to 00:00. Plus-end of ablated k-fiber is marked by '*'. Scale bar, 5 µm.
Video 7. The 'tail1+2A' region of NuMA's C-terminus is sufficient for minus-end localization. See also Figure 6D. Live confocal imaging of a RPE1 cell expressing GFP-C-tail1+2A in a NuMA knockout background. After ablation ('X'), GFP-C-tail1+2A is recruited to k-fiber minus-ends (arrowhead). The k-fiber stub slowly polymerizes, but its minus-end is not transported by dynein and remains detached from the spindle. Note that C-tail1+2A does not rescue spindle architecture; video shows one confocal slice of a three-dimensionally disorganized spindle. Also note that bright GFP signal from a neighboring interphase cell is noticeable on the left side of the video. Time is in min:sec, with the frame captured immediately following ablation set to 00:00 s. Plus-end of ablated k-fiber is marked by '*'. Scale bar, 5 µm.