Self-assembly of pericentriolar material in interphase cells lacking centrioles

The major microtubule-organizing center (MTOC) in animal cells, the centrosome, comprises a pair of centrioles surrounded by pericentriolar material (PCM), which nucleates and anchors microtubules. Centrosome assembly depends on PCM binding to centrioles, PCM self-association and dynein-mediated PCM transport, but the self-assembly properties of PCM components in interphase cells are poorly understood. Here, we used experiments and modeling to study centriole-independent features of interphase PCM assembly. We showed that when centrioles are lost due to PLK4 depletion or inhibition, dynein-based transport and self-clustering of PCM proteins are sufficient to form a single compact MTOC, which generates a dense radial microtubule array. Interphase self-assembly of PCM components depends on γ-tubulin, pericentrin, CDK5RAP2 and ninein, but not NEDD1, CEP152, or CEP192. Formation of a compact acentriolar MTOC is inhibited by AKAP450-dependent PCM recruitment to the Golgi or by randomly organized CAMSAP2-stabilized microtubules, which keep PCM mobile and prevent its coalescence. Linking of CAMSAP2 to a minus-end-directed motor leads to the formation of an MTOC, but MTOC compaction requires cooperation with pericentrin-containing self-clustering PCM. Our data reveal that interphase PCM contains a set of components that can self-assemble into a compact structure and organize microtubules, but PCM self-organization is sensitive to motor- and microtubule-based rearrangement.


Introduction 29
The centrosome is the major microtubule organizing center (MTOC) in animal cells. It consists of two 30 centrioles surrounded by pericentriolar material (PCM) (reviewed in (Conduit et al., 2015;Paz and 31 Luders, 2017)). Major PCM components are microtubule-nucleating and anchoring proteins, which 32 can associate with centrioles and with each other. For a long time it was thought that the PCM is 33 amorphous, but super-resolution microscopy studies have shown that it has a distinct organization, 34 with some proteins likely attached to the centriole wall and others organized around them (Fu and  ninein, γ-tubulin and dynein is thus sufficient to form a radial microtubule array, similar to a 240 centrosome, and PCM clustering promotes dense microtubule organization. 241 242

Dynamics of caMTOC disassembly 243
To test whether the formation and maintenance of caMTOCs depends on microtubules, we 244 depolymerized them by treating cells with nocodazole at 37°C and found that caMTOCs fragmented 245 into small clusters upon nocodazole treatment and reassembled into a single structure after 246 nocodazole washout ( Figure 4A-C). Because we found that caMTOC formation is dynein-dependent, 247 we also included the dynein inhibitor dynapyrazole A in these experiments (Steinman et al., 2017). We 248 confirmed that treatment with dynapyrazole A for 3 hrs had no effect on dynein expression ( Figure  249 4D) and found that the addition of this drug before nocodazole treatment prevented the disassembly 250 of caMTOCs, whereas the treatment of cells with dynapyrazole A during nocodazole washout strongly 251 inhibited caMTOC re-assembly ( Figure 4A-C). These data indicate that both microtubule-dependent 252 dispersal and coalescence of PCM clusters into caMTOCs are driven by dynein activity. 253 We next studied PCM dynamics using stably expressed GFP-CDK5RAP2 as a marker in live cells where 254 microtubules were labeled with SiR-tubulin. GFP-CDK5RAP2 was mostly immobile within caMTOCs 255 before nocodazole treatment (Figure 1 -figure supplement 3B, Video 2). After a few minutes of 256 nocodazole treatment, when the microtubule density was significantly reduced, small PCM clusters 257 started to move out of the caMTOC and undergo rapid directional motility with speeds of up to 2 258 µm/sec, which is within the range characteristic for cytoplasmic dynein (Schlager et   the movement of GFP-CDK5RAP2-positive clusters stopped, indicating that it is microtubule-261 dependent but occurs only when the microtubule network is partially depolymerized. Since cluster 262 dispersal towards the cell periphery could be blocked by a dynein inhibitor, and since cytoplasmic 263 dynein is a minus-end-directed motor, these data indicate that during microtubule disassembly by 264 nocodazole at 37°C, there is a transient stage when PCM clusters interact with only a few 265 microtubules, some of which have their minus-ends facing outwards, and these microtubules serve as 266 tracks for PCM transport. To support this idea, we used motor-PAINT, a technique that employs 267 nanometric tracking of purified kinesin motors on the extracted cytoskeleton of fixed cells to super-268 resolve microtubules and determine their orientation (Tas et al., 2017). Using this approach, we 269 11 determined microtubule orientations in centrinone-treated AKAP450/CAMSAP2 knockout cells and in 270 cells that were also treated with nocodazole for 15 min to induce partial microtubule disassembly 271 ( Figure 4G, Figure 4 -figure supplement 1B). We found that the cells contained a significant number 272 of minus-end-out microtubules, and their proportion increased during early stages of nocodazole 273 treatment, possibly because minus-end-out microtubules are more stable ( Figure 4G, e.g., minus-end-274 out microtubules constituted ~23% of the total microtubule length determined from kinesin-1 275 trajectories in the untreated cell and ~46% in the nocodazole-treated cell). These microtubules could 276 serve as tracks for outward movement of PCM, causing the disassembly of caMTOC when the overall 277 microtubule density around the caMTOC was strongly reduced ( Figure 4H). These data suggest that 278 the dense network of PCM-anchored microtubule minus-ends around a caMTOC allows for its 279 compaction via dynein-mediated forces, but that dynein can pull the caMTOC apart when 280 microtubules are disorganized. 281 To further confirm that caMTOC disassembly is an active microtubule-dependent process, we also 282 depolymerized microtubules by a combination of cold (4°C) and nocodazole treatment. When all 283 microtubules were depolymerized, caMTOCs did not fall apart, even when the cells were subsequently 284 warmed to 37°C in the presence of nocodazole, so that microtubules could not re-grow ( Figure 4I). 285 However, we noticed that in these conditions, the continuity and cylindrical organization of the PCM 286 cluster were often perturbed. This raised the possibility that the elongated arrangement of PCM 287 components within caMTOCs is microtubule-driven. Indeed, when cells were subjected to cold (4°C) 288 treatment in the absence of nocodazole, most of microtubules depolymerized, but some short cold-289 stable microtubules remained associated with the caMTOC (Figure 4 -figure supplement 1C). These 290 data indicate that PCM self-assembly in the absence of centrioles is microtubule-dependent, and 291 microtubules are involved in shaping the PCM cluster. Once assembled, the PCM cluster is quite stable, 292 unless microtubule organization is altered and dynein-driven microtubule-based transport pulls it 293 apart. 294 295

Dynamics of caMTOC assembly 296
When nocodazole-mediated microtubule disassembly was carried out at 4°C, the caMTOC remained 297 intact and after nocodazole washout it served as the major microtubule nucleation site, similar to the 298 centrosome in untreated wild-type cells ( Figure 5 -figure supplement 1A). However, when 299 nocodazole-mediated disassembly of the caMTOC was carried out at 37°C, the cluster fell apart and 300 reassembled upon nocodazole washout ( Figure 4A, B), providing a way to study the dynamics of PCM 301 self-assembly and the roles of different PCM components during this process. Small PCM clusters 302 positive for pericentrin, CDK5RAP2, γ-tubulin and the centriolar satellite protein PCM1 that co-303 localized with the plus-ends of microtubules (labeled with EB1) could be detected 30 s after 304 nocodazole washout; these PCM clusters and nascent microtubules did not colocalize with the Golgi 305 membranes ( Figure 5A, Figure 5 -figure supplement 1B). Ninein was not detected within the clusters 306 at this early stage of microtubule regrowth but could be found 2 min after nocodazole washout. In 307 contrast, no clusters of CEP192 or NEDD1 were observed even 10 min after nocodazole washout 308 stably expressing GFP-CDK5RAP2 showed that when pericentrin was depleted, CDK5RAP2 clusters 321 were not detectable, and the microtubule network, both before nocodazole treatment and after 322 nocodazole washout, was disorganized ( Figure 5 -figure supplement 2, Video 3). Taken together, our 323 data show that pericentrin and γ-tubulin form microtubule-nucleating and anchoring units, which are 324 clustered by the self-association of pericentrin and assembled into larger structures by dynein-based 325 transport. CDK5RAP2 contributes to microtubule nucleation efficiency, whereas ninein appears to act 326 somewhat later and contributes to the formation of a compact PCM cluster and a radial microtubule 327 network. Importantly, all these proteins can cluster in the absence of centrioles, and together they 328 can efficiently nucleate and anchor microtubules. 329

330
The role of CAMSAP2-stabilized minus ends in defining microtubule network geometry 331 The results of nocodazole treatment and washout suggested that PCM can self-assemble into a 332 caMTOC which nucleates and anchors microtubules, but this structure is sensitive to microtubule 333 organization. This observation prompted us to investigate in more detail how the microtubules that 334 are not anchored at PCM clusters affect PCM organization in steady state conditions. An abundant 335 13 population of stable minus ends that do not attach to PCM is decorated by CAMSAP2. In centrinone-336   treated wild type cells, CAMSAP2-bound microtubule minus ends were anchored at the Golgi (Wu et  337 al., 2016)( Figure 1D), whereas in centrinone-treated AKAP450 knockout cells they were distributed 338 randomly ( Figure 1D, 6A). Live imaging showed that CAMSAP2-decorated minus ends displayed only 339 limited motility on the scale of hours and thus formed a relatively stationary, disorganized microtubule 340 network (Video 4). Live imaging of GFP-CDK5RAP2 together with SiR-tubulin in these cells 341 demonstrated that small PCM clusters were distributed throughout the cytoplasm ( Figure 6B). These 342 clusters moved along microtubules and encountered each other, but the direction of the movements 343 was random and the clusters did not coalesce into a single structure ( Figure 6B, Figure 6 - figure  344 supplement 1A, Video 5). Treatment with nocodazole and subsequent nocodazole washout confirmed 345 that the motility of GFP-CDK5RAP2 clusters in centrinone-treated AKAP450 knockout cells was 346 microtubule-dependent, and that these clusters could nucleate microtubules and move together with 347 microtubule minus ends, but did not converge into a single caMTOC ( clustering of GFP-CDK5RAP2 in AKAP450 knockout cells is pericentrin-dependent. Based on these data, 353 we conclude that in AKAP450 knockout cells, pericentrin still forms PCM clusters that can nucleate 354 microtubules and can be moved by dynein along other microtubules, similar to what occurs in wild-355 type cells. However, in the absence of AKAP450, CAMSAP2-stabilized microtubules form a 356 disorganized network, which imposes a random motility pattern on pericentrin-dependent PCM 357 clusters and thus prevents their assembly into a single caMTOC, likely because PCM interactions are 358 not sufficient to trigger their stable association ( Figure 6C). 359 If the geometry of the CAMSAP2-stabilized microtubule network determines PCM distribution, 360 focusing CAMSAP2-bound minus ends is expected to bring PCM together. To test this idea, we linked 361 CAMSAP2-stabilized minus ends to a minus-end-directed motor. In order to avoid potential cell 362 toxicity associated with manipulating cytoplasmic dynein, we used the motor-containing part of a 363 moss kinesin-14, type VI kinesin-14b from the spreading earthmoss Physcomitrella patens (termed 364 here ppKin14). The C-terminal motor-containing part of this protein can efficiently induce minus-end-365 directed motility of different cargoes in mammalian cells when it is tetramerized through a fusion with 366 the leucine zipper domain of GCN4 (GCN4-ppKin14-VIb (861-1321)) and recruited to cargoes using 367 inducible protein heterodimerization (Jonsson et al., 2015;Nijenhuis et al., 2020). We employed a 368 chemical heterodimerization system that is based on inducible binding of two protein domains, FRB 369 14 and FKBP, upon the addition of a rapamycin analog (rapalog) (Pollock et al., 2000). To ensure that all  370   CAMSAP2-decorated microtubule minus ends were linked to kinesin-14, we rescued centrinone-371   treated AKAP450/CAMSAP2 knockout cells by expressing CAMSAP2 fused to a tandemly repeated  372 FKBP domain (2FKBP-mCherry-CAMSAP2) ( Figure 6D-F). This construct was co-expressed with the FRB-373 GCN4-tagBFP-ppKin14 fusion, which by itself localized quite diffusely, with only a weak enrichment 374 along microtubules, as described previously (Nijenhuis et al., 2020) (Figure 6D,F). In the absence of 375 rapalog, CAMSAP2-decorated microtubule minus ends were distributed randomly, similar to 376 endogenous CAMSAP2 in AKAP450 knockout cells ( Figure 6F). However, upon rapalog addition, 377 ppKin14 was rapidly recruited to CAMSAP2-decorated microtubule ends, and after 2 hrs, more than 378 90% of cells acquired a radial microtubule organization ( Figure 6E-G). In rapalog-treated cells, 379 CAMSAP2-bound microtubule minus ends formed either a tight cluster or a "whirlpool-like" ring in the 380 cell center ( Figure 6D Presence of PCM in the caMTOC induced by minus-end-directed transport of CAMSAP2 might be a 389 passive consequence of microtubule reorganization, but might also play an active role in forming this 390 caMTOC. To distinguish between these possibilities, we attached CAMSAP2 to ppKin14 in centrinone 391 -treated cells where both AKAP450 and pericentrin were knocked out 392 (AKAP450/CAMSAP2/p53/pericentrin knockout). In the absence of rapalog, CAMSAP2-stabilized 393 minus ends and the whole microtubule network were disorganized, as expected, and the same was 394 true when the two constructs were expressed separately, with or without rapalog ( Figure 7A,D, Figure  395 7-figure supplement 1). After rapalog addition, microtubules in cells expressing both constructs 396 acquired a radial organization, but their minus ends usually did not converge in a single spot but rather 397 accumulated in a ~30-70 µm-large ring-like structure ( Figure 7A Video 8). Staining for PCM markers showed that CDK5RAP2 and γ-tubulin were enriched in the vicinity 399 of CAMSAP2-positive microtubule minus ends, whereas ninein appeared rather diffuse ( Figure 7A). To 400 determine the nature of the structure "corralled" by the ring of CAMSAP2-decorated minus ends in 401 rapalog-treated cells, we stained for different membrane organelles and found that whereas there 402 was no strong correlation with the nucleus, Golgi membranes or lysosomes, the majority of 403 mitochondria were found within the CAMSAP2 ring, and the endoplasmic reticulum (ER) displayed 404 increased density overlapping with the CAMSAP2 ring ( Figure  To support this notion further, we also generated cells that were knockout for AKAP450, CAMSAP2, 414 CDK5RAP2, myomegalin (MMG, homologue of CDK5RAP2), p53 and pericentrin. To achieve this, we 415 used the previously described RPE1 cell line knockout for AKAP450, CAMSAP2, CDK5RAP2 and MMG 416 (Wu et al., 2016), in which we sequentially knocked out p53 and pericentrin (Figure 7 Figure 6D), we observed that CAMSAP2 clustering was even less efficient than in 424 AKAP450/CAMSAP2/p53/pericentrin knockout cells ( Figure  7D,E). 49% of 425 AKAP450/CAMSAP2/CDK5RAP2/MMG/p53/pericentrin knockout cells had small bundles of CAMSAP2 426 stretches dispersed throughout the cytoplasm, and only 37% of these cells formed a ring of CAMSAP2-427 decorated minus ends, whereas 80% of AKAP450/CAMSAP2/p53/pericentrin knockout cells formed 428 such a ring. We examined the ER and mitochondria in these cells and found that in cells that did form 429 a CAMSAP2 ring, the ER displayed an overlapping ring-like density, although the mitochondria inside 430 the CAMSAP2 ring were more scattered compared to those of AKAP450/CAMSAP2/p53/pericentrin 431 knockout cells (Figure 7 -figure supplement 2B). In cells with dispersed CAMSAP2-positive bundles, 432 no increased ER density or central accumulation of mitochondria were observed (Figure 7 -figure  433 supplement 2B). These data further support the notion that minus-end-directed transport of stable 434 minus ends alone is insufficient to generate a caMTOC, and that synergy with PCM is required. 435 436

Recapitulation of PCM self-organization by computer simulations 437
To rationalize the appearance of the different microtubule arrangements and to find a minimal set of 438 interactions between filaments and motors that would lead to self-organization into structures that 439 we observed experimentally, we set up agent-based computer simulations with Cytosim (Nedelec and 440 Foethke, 2007). First, we sought to recapitulate the formation of a single PCM cluster. The numerical 441 values for the parameters of our simulations have been taken from literature or reasonably chosen 442 otherwise (Supplemental Table S1). For simplicity, we considered a two-dimensional circular cell with 443 a radius of 10 μm. We described a mobile PCM complex as a bead with a radius of 50 nm from which 444 one microtubule plus end could grow. Microtubule growth and shrinkage were simulated with the 445 classical microtubule model from Cytosim. When the microtubule reached a maximal length of 7.5 446 μm, its growth was stopped. In this way, we limited the microtubule length to avoid long microtubules 447 that push their minus end to the periphery of the cell. Additionally, one dynein molecule was attached 448 to a PCM complex with its cargo-binding domain. With this configuration, dynein molecules could 449 transport PCM complexes along microtubules growing from other PCM complexes. Once they were 450 bound to a microtubule, they walked towards the minus end in a force-dependent manner. When a 451 dynein motor reached a minus end, it detached. Furthermore, we implemented a reversible binding 452 interaction between PCM complexes to make them adhere to each other. Because we were interested 453 in the spatial arrangement of the PCM complexes, we introduced steric interactions between them. 454 However, we neglected steric interactions between microtubules to effectively account for the three- In our experiments, we saw that the presence of CAMSAP2-stabilized microtubule network prevented 467 PCM clustering ( Figure 1D,E, Figure 6A-C), and that CAMSAP2-bound minus ends appeared almost 468 stationary on the scale of minutes (Video 4) compared to the rapid dynein-driven movement of small 469 PCM clusters (Video 5). To simulate such a situation, we assumed that 300 microtubules could grow 470 and shrink with random orientations from CAMSAP2-stabilized microtubule ends that were randomly 471 distributed and stationary within the cell. Furthermore, we assumed that some of the moving PCM 472 complexes were not efficient in nucleating microtubules, because the presence of a PCM-independent 473 microtubule population would reduce the concentration of soluble tubulin and thus nucleation 474 efficiency. To effectively account for such an effect, we allowed only 50 of the 300 PCM complexes to 475 nucleate microtubules. In this system, the PCM cluster formation was disrupted, and PCM complexes 476 randomly moved around the cell ( Figure 8C to each other, provided that the interaction between PCM complexes is not too tight. 484 Next, we examined how CAMSAP2-stabilized microtubules would organize when minus-end-directed 485 kinesin-14 motors could attach to them and carry the minus ends. CAMSAP2-stabilized microtubule 486 ends bound to kinesin-14 motors (termed "CAMSAP-kin14 complexes" in the text below) were 487 modelled as a bead with a radius of 50 nm from which a microtubule could grow and to which 6 488 kinesin-14 molecules were attached. Because the biophysical properties of single plant kinesin-14 489 molecules used in our assays are poorly understood, we assumed typical kinesin-1 values with an 490 opposite directionality. CAMSAP-kin14 complexes only interacted sterically but did not adhere to each 491 other. In a simulation of 300 CAMSAP-kin14 complexes, a loose ring-like arrangement appeared 492 ( Figure 8G), similar to the one observed in our simulations of PCM complexes that could not bind to 493 each other ( Figure 8B). Introducing adhesive interactions between CAMSAP-kin14 complexes was 494 sufficient to promote compact cluster formation ( Figure 8H). 495 Finally, to investigate if a compact cluster of CAMSAP-kin14 complexes would emerge in the presence 496 of self-associating PCM complexes, we set up simulations with 150 CAMSAP-kin14 complexes together 497 with 150 adherent PCM complexes. The steady state of such a system displayed a compact central 498 cluster, in which both types of complexes were mixed ( Figure 8I,K,L, Video 9I), while making PCM 499 complexes non-adhesive prevented cluster compaction ( Figure 8J,K,L, Video 9J). Taken together, our 500 simulations suggest that the PCM complexes provide enough adhesive interactions to compact the 501 cluster of non-interacting CAMSAP-kin14 and PCM complexes, very similar to our experimental 502 observations. 503

Discussion 504
In this study, we explored the mechanisms of interphase PCM self-assembly in the absence of 505 centrioles. Our experiments and simulations support the idea that PCM complexes can form a single 506 cluster through a positive feedback mechanism, whereby dynein motors carry microtubule minus-507 ends to other microtubule minus ends (Cytrynbaum et al., 2004). However, for a compact cluster to 508 emerge, the minus ends must not only be able to move towards each other but also to bind to each 509 other, and this idea is fully supported by our simulations. Interphase PCM is thus capable both of 510 dynein binding and self-association sufficient to organize a compact microtubule-nucleating and 511 anchoring structure in the absence of centrioles. However, the formation of a compact MTOC in 512 acentriolar cells is slow and less robust than in centriole-containing cells, and the resulting structure 513 is sensitive to the overall organization of microtubules and to dynein function. This means that 514 interphase PCM self-association is by itself reversible and not sufficiently tight to resist dynein-driven 515 forces. Centrioles can thus be regarded as catalysts of PCM assembly and stabilizers of interphase 516 centrosomes, preventing PCM movement on microtubules oriented with their minus ends away from 517 the centrosome. 518 Our experimental system allowed us to examine which PCM components are capable of associating 519 with each other and with dynein to promote microtubule nucleation and anchoring independently of 520 centrioles. The major scaffold for interphase acentriolar PCM assembly is pericentrin, which can self- Our simple cellular system allowed us to dissect the molecular details of acentriolar PCM assemblies. 564 We found that caMTOCs recruited multiple MAPs and a subset of centriolar proteins. For example, 565 caMTOCs accumulated several +TIPs, such as CLASPs, CLIP170 and chTOG, though the depletion of 566 these proteins had no effect on caMTOC formation or function. In contrast, the core components of 567 the +TIP complexes, EB1 and EB3, were neither enriched in caMTOCs nor required for their assembly. 568 While the negative results on other +TIPs might be due to their incomplete depletion, EB1 and EB3 569 function was tested using genetic knockouts, indicating that interphase PCM function is EB-570 independent. Among the centriolar proteins, we detected CPAP, CP110 and CEP120 in caMTOCs, and 571 20 it will be interesting to test whether any of these microtubule-binding factors contribute to 572 microtubule organization independently of their participation in centriole and centrosome assembly. 573 Furthermore, our work provided insight into the self-assembly properties of interphase PCM. It seems 574 that while oligomerization and clustering of pericentrin molecules is important for MTOC formation 575 both in interphase and in mitosis, during mitotic entry, pericentrin forms condensates ( . We found that CAMSAP2-592 stabilized minus ends can exert a highly dominant effect on PCM organization, and this property likely 593 explains how the Golgi, which anchors CAMSAP2-stabilized microtubules, recruits PCM and becomes 594 the major MTOC in cells lacking centrosomes. Importantly, our experiments and simulations showed 595 that coupling stable minus-ends to a minus-end directed motor is by itself insufficient to form a 596 compact MTOC, but self-clustering PCM can contribute to this process. Altogether, self-association of 597 interphase PCM appears to be strong enough to promote its clustering but is sufficiently dynamic to 598 allow PCM reorganization dependent on other microtubule regulators present in the cell. 599 An interesting question that remains unanswered by our work is the inhibitory role of PLK4 in 600 interphase caMTOC formation. We did observe caMTOCs in cells depleted of PLK4, indicating that, 601 unlike cells lacking TRIM37, which form PCM clusters containing catalytically inactive PLK4 (Meitinger 602 et al., 2020; Yeow et al., 2020), interphase cells studied here do not rely on enzymatically inactive PLK4 603 for PCM assembly. PLK4 is known to phosphorylate NEDD1 (Chi et al., 2021), and it is possible that the 604 lack of phosphorylation prevents this γ-TuRC activator and its partners, such as CEP192, from 605 participating in interphase caMTOC assembly. It is, of course, also possible that PLK4 phosphorylation 606 inhibits the interactions or activities of some of the players driving caMTOC formation. The easy-to-607 manipulate cellular model that we have described here will allow these questions to be addressed and 608 facilitate detailed studies of the interactions and functions of PCM components in nucleating and 609 stabilizing interphase microtubule minus ends. 610

DNA constructs and protein purification 612
To generate the lentiviral vector pLVX-GFP-CDK5RAP2-IRES-Puro, pLVX-IRES-Puro plasmid (Clontech) 613 was digested with AgeI and NotI (FastDigest, Thermo Fisher), and then Gibson Assembly (NEB) was 614 performed with gel-purified PCR product of GFP-CDK5RAP2 ( To remove centrioles, RPE1 cells were treated with 125 nM centrinone B containing complete medium 666 for ~10 days, and drug-containing medium was refreshed every 24 hrs; cell confluence was maintained 667 around ~50-80% during the treatment. 668 For the microtubule disassembly and regrowth assay, the acentriolar RPE1 cells were seeded onto 669 coverslips in 24-well plates and incubated for 24 hrs, then cells were treated with 10 μM nocodazole 670 for 1 hr in an incubator (37°C, 5% CO2) and followed by another 1 hr treatment at 4°C to achieve 671 complete disassembly of stable microtubule fragments. Nocodazole washout was then carried out by 672 at least six washes on ice with ice-cold complete medium; subsequently, plates were moved to a 37°C 673 water bath and pre-warmed medium was added to each well to allow microtubule regrowth. 674 For cell cycle synchronization, centrinone-treated AKAP450/CAMSAP2/P53 knockout cells were 675 treated with 5 mM Thymidine (Sigma-Aldrich) overnight, released in centrinone containing medium 676 for 4 hrs and subsequently treated with 5μM proTAME (Boston Biochem, I-440) for 2 hrs before being 677 released in centrinone containing medium for 1-4 hrs followed by live imaging and fixation. 678 For the inducible ppKin14-CAMSAP2 heterodimerization experiment, acentriolar cells were seeded 679 onto coverslips in 24-well plates, cultured with centrinone B containing medium and co-transfected 680 with 2FKBP-mCherry-CAMSAP2 and FRB- TagBFP

Generation of CRIPSR/Cas9 knockout cell lines 704
The CRISPR/Cas9-mediated knockout of p53-, pericentrin-, AKAP450-and CAMSAP2-encoding genes 705 was performed as described previously (Ran et al., 2013). In brief, AKAP450/CAMSAP2 knockout RPE1 706 25 cells (Wu et al., 2016) were transfected with the vectors bearing the appropriate targeting sequences 707 using FuGENE 6. One day after transfection, the transfected AKAP450/CAMSAP2 knockout RPE1 cells 708 were subjected to selection with 15 µg/ml puromycin for up to 3 days. After selection, cells were 709 allowed to recover in normal medium for ∼7 days, and knockout efficiency was checked by 710 immunofluorescence staining. Depending on the efficiency, 50-500 individual clones were isolated 711 and confirmed by immunofluorescence staining, and the resulted single colonies were characterized 712 by Western blotting, immunostaining and genome sequencing. AKAP450/CAMSAP2/p53 and 713 AKAP450/CAMSAP2/MMG/CDK5RAP2/p53 knockout cell lines were generated first and subsequently, 714 each of them was used to knock out the gene encoding pericentrin. The mutated portions of the p53-715 and pericentrin-encoding genes were sequenced using gel-purified PCR products obtained with 716 primers located in the vicinity of the corresponding sgRNA targeting sites. Detected particles were linked into tracks again using DoM, which performs a quicker variant of a 866 nearest neighbor search, with a maximum distance of 5 pixels (~320nm) between consecutive frames 867 30 and no permitted frame gap. Tracks were later filtered to remove those shorter than 4 frames or 868 longer than 200 frames, those in which an angle between parts of the trajectory exceeded 90 degrees, 869 and those in which the speed of the motor was less than 100nm/s or more than 1500nm/s. 870 The particle table was then split into four particle tables corresponding to the four quadrants of the 871 image with tracks sorted based on their net displacement (i.e., Δx>0 ∧ Δy>0; Δx>0 ∧ Δy<0; Δ 872 x<0 ∧ Δy>0; Δx<0 ∧ Δy<0), as described previously (Tas et al., 2017). These directionality- To quantify the percentage of minus-end-out microtubule length to total microtubule length before 881 and after nocodazole treatment, the length of each microtubule (determined from kinesin-1 882 trajectories) in the cell was measured by calculating the Euclidean distance between all subsequent 883 pairs of points along the microtubule and summed. The ratio was calculated as the total minus-end-884 out microtubule length divided by the total microtubule length. 885 886

Analysis of PCM cluster dynamics 887
To represent the motion of PCM clusters during nocodazole treatment, ImageJ plugin 888 KymoResliceWide v.0.4 (https://github.com/ekatrukha/KymoResliceWide) was used for generating 889 kymographs from the time lapse images. The velocity of PCM clusters was measured manually using 890 kymographs starting from the time point when a small PCM cluster moved out of a caMTOC. 891 Microtubule density around each PCM cluster was determined by measuring the mean fluorescence 892 intensity of SiR-tubulin in a circular area with a 2 μm radius centered on the PCM cluster and 893 normalizing it to the mean fluorescence intensity of 20 images prior to nocodazole addition (set as 894 100%). The moment when a PCM cluster started to move out of the caMTOC was set as the initial time 895 point (0 min), and the subsequent PCM cluster motion velocity and the relative local microtubule 896 density at 43 time points were calculated and averaged. 897 The movement trajectories of PCM clusters were generated using ImageJ plugin TrackMate (version 898 is 6.0.2). The parameters and the settings used were as following: LoG (Laplacian of Gaussian) detector 899 31 with estimated blob diameter: 14.9 µm; thresholding value 12.25; sub-pixel localization was selected. 900 HyperStack Displayer was selected to overlay the spots and tracks on the current hyperstack window. 901 Simple LAP tracker was selected to track the distance and time with the linking max distance: 32.0 µm, 902 gap-closing max distance: 55.0 µm and gap-closing max frame gap: 2. All other parameters and 903 settings were used as the default.  Table S1. The configuration file is provided as Supplemental file 1. 910 We defined the following components in the simulation: 911 Cell shape: We considered a two-dimensional system with a circular cell with a radius of 10 µm. As 912 commonly used in Cytosim simulations, we set the intracellular viscosity to 1 pN s/ µm 2 . 913

Molecular motors:
The binding process of a molecular motor to a microtubule was described by a 914 binding rate kon and the unbinding from the microtubule by a force-dependent unbinding rate koff = 915 k 0 off exp(F/Fd). When a motor was engaged with the microtubule it moved along the microtubule with 916 a linear force-dependent velocity, characterized by v(F) = v0 (1 -F/Fs). Dynein, as well as kinesin-14 917 motors moved to the minus end of microtubules. 918 Microtubule filaments: We used a classical model for microtubule dynamics which is described by a 919 catastrophe rate, a growth speed, and a shrinkage speed. The growth speed is force-dependent with 920 a characteristic growing force. For simplicity, we ignored rescue events. The catastrophe rate was set 921 as kcat = vg/LMT, in which the mean microtubule length LMT was 5 µm. To further restrict the microtubule 922 length, we set a maximum of 7.5 µm. This limitation was necessary to avoid that long microtubules 923 were pushing the minus ends to the periphery. 924 PCM complexes: We described a PCM complex as a bead with a radius of 50 nm. We randomly placed 925 one microtubule nucleation site and one dynein on the bead. To effectively account for an unspecific 926 adhesive interaction between PCM complexes, we introduced two molecules that can bind to each 927 other. One was implemented as a 10 nm Cytosim fiber and the binding partner as a Cytosim hand with 928 a binding range of 100 nm, binding rate of 10 s -1 , force-free unbinding rate of 0.01 s -1 , and 929 characteristic unbinding force of 3 pN. We randomly placed one of each molecule on a PCM complex. 930 In the simulations with strong adhesive interactions, we increased the number of adhesive binding 931 32 molecules on the beads and kept all the other parameters the same. We defined five random 932 attachment points on the beads and placed to each point five molecules. In this setup each PCM 933 complex was covered with 25 Cytosim binding filaments and 25 Cytosim binding hands. Therefore, 934 when two PCM complexes were close to each they formed multiple bonds between each other. 935 CAMSAP-kin14 complexes: We described a complex consisting of a CAMSAP-stabilized microtubule 936 end with kinesin 14 motors attached as a Cytosim bead with a radius of 50 nm. We attached five kin14 937 motors and one microtubule nucleation site randomly on the bead. When we implemented adhesive 938 interaction between CAMSAP-kin14 complexes, we used exactly the same binding molecules and 939 arrangements as used for the PCM complexes. Microtubule release from the centrosome in migrating cells.                            anti Ku80