Differentially accessible Cdc4 phospho-degrons regulate Ctf19CCAN kinetochore subunit stability in mitosis

Kinetochores are multi-subunit protein assemblies that link chromosomes to microtubules of the mitotic and meiotic spindle. How effective, yet strictly centromere-dependent kinetochore assembly is coupled to cell cycle progression is incompletely understood. Here, by combining comprehensive phosphorylation analysis of native Ctf19CCAN subunits with biochemical and functional assays in the model system budding yeast, we demonstrate that Cdk1 phosphorylation activates phospho-degrons on the essential subunit Ame1CENP-U which are recognized by the E3 ubiquitin ligase complex SCF-Cdc4. Gradual phosphorylation of degron motifs culminates in M-Phase and targets the protein for degradation. Binding of the Mtw1 complex shields the proximal phospho-degron, protecting kinetochore-bound Ame1 from the degradation machinery. Artificially increasing degron strength partially suppresses the temperature-sensitivity of a cdc4 mutant, while overexpression of Ame1-Okp1 is toxic to cells, demonstrating the physiological importance of this mechanism. We propose that phospho-regulated clearance of excess CCAN subunits protects against ectopic kinetochore assembly and contributes to mitotic checkpoint silencing. Our results suggest a novel strategy for how phospho-degrons can be used to regulate the assembly of multi-subunit complexes.

phosphorylation of Okp1 was promoted by the hydrophobic patch, a known substrate 147 docking region in Clb5 and Clb2 cyclins (Supplementary Figure 1). The mapped 148 phosphorylation sites closely corresponded to the sites detected on native Ame1, in 149 particular phosphorylation of the residues Thr31, Ser41, Ser45, and Ser52/Ser53, was 150 both detected in vivo and in vitro ( Figure 1C). In case of Ser52/Ser53 either one, or 151 both adjacent sites may be phosphorylated. For the subsequent analysis, we focused 152 on seven N-terminal phosphorylation sites, as the two C-terminal sites Ser277 and 153 Ser323 were not phosphorylated by Cdc28-Clb2 in vitro. To analyze the functional role 154 of Ame1 phosphorylation we mutated the cluster to either alanine (Ame1-7A) or 155 glutamic acid (Ame1-7E) to eliminate or mimic phosphorylation, respectively. We 156 integrated Flag-tagged Ame1 constructs under their endogenous promoter with these 157 mutations into yeast and analyzed cell extracts by western blotting. Analysis of log 158 phase extracts showed that wild-type Ame1 displayed multiple slowly migrating forms 159 which were eliminated in the 7A mutant ( Figure 1D). By contrast, Ame1-7E migrated 160 much more slowly than wild-type, its position in SDS-PAGE corresponding to the most 161 slowly migrating forms of Ame1-WT. Ame1-7A and -7E mutants were viable when 162 expressed as the sole source of Ame1 in the cell. In an anchor-away approach, in 163 which endogenous Ame1 is removed from the nucleus upon addition of rapamycin, 164 both Ame1-7A and -7E variants supported viability with little difference in growth rate 165 on rich media compared to wild-type Ame1 ( Figure 1E). Analysis of internal 166 truncations, which maintained the essential Mtw1c-binding N-terminus (residues 1-15) 167 showed that deleting the region harboring the entire phospho-cluster (D31-116) 168 yielded a slow growth phenotype, while a more extensive deletion was inviable ( Figure  169 1F). We conclude that Ame1 phosphorylation is not required for viability, but the N-170 terminus contributes to an important aspect of Ame1 function, even when the Mtw1-171 binding domain is retained. 172 173 Non-phosphorylatable Ame1 mutants accumulate to increased protein levels 174 During our cellular characterization experiments for Ame1 phospho-mutants we 175 expressed wild-type or mutant versions of Ame1, along with its binding partner Okp1 176 from a two-micron plasmid under control of a Galactose-inducible promoter ( Figure  177 2A). Western Blot analysis showed that wild-type Ame1 gradually accumulated over 178 the course of five hours after switching the cells to Galactose. Strikingly, the non-179 phosphorylatable Ame1-7A mutant accumulated to much higher protein levels in the 180 same time span, leading to a roughly four-fold increase in steady state level compared 181 to wild-type after five hours in Galactose (Figure 2B and C). In this experiment, the 182 Ame1-7E mutant behaved similar to the -7A mutant, suggesting that it may constitute 183 a phospho-preventing rather than a phospho-mimetic mutation (Figure 2B and C). By 184 contrast, Okp1 expressed from the same plasmid showed no change in protein level 185 in the different Ame1 mutants, arguing that differences in plasmid stability or mitotic 186 retention cannot be the cause for the observed effect on the Ame1 protein level. 187 Protein level differences between Ame1 phospho-mutants were also observed when 188 overexpression was performed in a mad1D strain background (Supplementary 189 Figure 2A). As the steady state protein level is determined by the rate of protein 190 translation versus degradation, and the rate of production should be unaffected in 191 these experiments, we reasoned that non-phosphorylatable Ame1 mutants may 192 accumulate due to impaired protein degradation. The levels of Cse4, part of the 193 centromeric nucleosome and a direct binding partner of AO at the inner kinetochore, 194 have been shown to be regulated by ubiquitin-dependent proteolysis via the E3 195 ubiquitin ligase Psh1 (Ranjitkar et al., 2010). Levels of Gal-expressed Ame1- WT,196 however, remained low in a psh1D strain background, while Ame1-7A and - 7E 197 accumulated as in the wild-type background (Supplementary Figure 2B). This 198 suggests that Psh1 is not involved in Ame1 level regulation under these conditions. 199 Another E3 ubiquitin ligase complex, Ubr2/Mub1 has been shown to regulate Dsn1,200 which is a subunit of the Ame1 binding partner Mtw1c (Akiyoshi et al., 2013). Similar 201 to psh1D, however, Ame1 protein levels were unaffected by the mub1 deletion and we 202 conclude that Ubr2/Mub1 is not involved in Ame1 level regulation under these 203 conditions either (Supplementary Figure 2C). 204 205

Identification of two phospho-degron motifs in the Ame1 N-terminus 206
To delineate the contribution of individual phosphorylation sites to Ame1 protein level 207 regulation in the overexpression setting, we constructed mutants in which we 208 prevented phosphorylation at selected sites individually or in combination ( Figure 2D). 209 Analysis of Ame1 protein levels after five hours of expression in the presence of 210 Galactose showed that preventing phosphorylation on Thr31 had relatively little effect 211 on Ame1 level when compared to the wild-type. By contrast, preventing 212 phosphorylation at Ser41, Ser45 or Ser52/53 led to accumulation of the protein, 213 roughly similar to preventing phosphorylation altogether in the 7A mutant ( Figure 2E). 214 We prepared the analogous Ame1 mutants as recombinant Ame1-Okp1 (AO) 215 complexes for in vitro kinase assays to evaluate the contribution of these individual 216 sites to overall Ame1 phosphorylation. Autoradiographs showed that in addition to 217 Ame1, also Okp1 can be phosphorylated by Cdc28-Clb2 ( Figure 2F, see also 218 Supplementary Figure 1 and 3). Ame1-WT appeared as two separated 219 phosphorylated forms after in vitro phosphorylation. The Ame1-T31A mutant displayed 220 a similar phosphorylation pattern, while the phosphorylation of Ame1-4A was clearly 221 decreased, with only the fast migrating Ame1 form remaining. The Ame1-7A mutant 222 completely eliminated Cdc28-Clb2 phosphorylation in vitro. Preventing 223 phosphorylation on either Ser41/Ser45 or Ser52/Ser53 allowed some residual 224 phosphorylation, but clearly decreased phosphorylation compared to wild-type. We 225 conclude that the residues responsible for Ame1 level regulation in vivo are major 226 targets for Cdc28 phosphorylation in vitro. Further analysis confirmed that mutating 227 the candidate Cdk1 site Ser26 in Okp1 to alanine prevented Cdc28 phosphorylation 228 and that the Ame1-7A/Okp1-1A complex was completely refractory to Cdc28 229 phosphorylation (Supplementary Figure 3). 230 Posttranslational modification via phosphorylation can be mechanistically linked to the 231 control of protein stability via the generation of so called phospho-degrons (Skowyra 232 et al., 1997). The best studied example for this mechanism is the controlled 233 ubiquitination and degradation of key cell cycle regulators by modular SCF complexes, 234 using F-box proteins as readers of phosphorylated substrates (Feldman et  interactions with Arg467 of Cdc4 ( Figure 3A). Further, phospho-Ser45 is engaged in 255 additional interactions with Arg443, Ser464, and Thr465 of Cdc4 ( Figure 3B). The 256 peptide residues Pro42, Ile43, Glu39, and Asn40 are also involved in interactions with 257 the protein ( Figure 3B). Overall, the doubly phosphorylated peptide showed a strong 258 hydrogen bond network at the binding pocket of Cdc4, highlighting the potential of this 259 peptide as a Cdc4 binder. 260 The simulations also indicated that the singly phosphorylated variants establish less 261 interactions with Cdc4 with respect to the doubly phosphorylated Ser41/Ser45 peptide 262 (Supplementary Figure 4B, Supplementary Movies 1,2 and 3). Furthermore, the 263 protein-peptide complex involving the doubly phosphorylated peptide 264 (Supplementary Figure 4B, Supplementary Movies 1,2 and 3 Western blotting showed that Ame1-WT expressed from the GAL promoter strongly 279 accumulated in the skp1-3 mutant relative to a wild-type background ( Figure 4A). The 280 phospho-forms of Ame1 were preserved under these conditions, showing that the 281 skp1-3 mutant uncouples phosphorylation of Ame1 from its degradation. Interestingly, 282 under these conditions, also an accumulation of Okp1, expressed from the same 283 plasmid, was apparent. Okp1 appeared in two distinctly migrating forms, possibly 284 corresponding to phosphorylation. Combining the Ame1-7A mutation with the skp1-3 285 background revealed that the Ame1-7A protein was further enriched in the skp1-3 286 background compared to the wild-type strain, demonstrating that Ame1-7A can be 287 further accumulated by compromising the SCF machinery in addition to preventing 288 phosphorylation of Ame1 itself. We extended this analysis to mutant alleles in other 289 SCF subunits, in particular to identify which F-box protein is responsible for Ame1 290 regulation. Similar to the skp1-3 mutant, overexpressed Ame1 accumulated in mutant 291 alleles of the Cullin subunit Cdc53, the E2 enzyme Cdc34 and the F-Box protein Cdc4. 292 By contrast Ame1 levels remained low (or were even decreased relative to wild-type) 293 in a deletion mutant of the cytoplasmic F-box protein Grr1 ( Figure 4B). Interestingly, 294 the SCF mutants also had a pronounced effect on the level of overexpressed Okp1, 295 with particularly strong accumulation (30-fold increase) observable in the cdc4-1 296 mutant. In the background of the cdc34-2 allele, Okp1 accumulated only slightly when 297 Ame1 was wild-type, but more strongly when phosphorylation of Ame1 was prevented 298 ( Figure 4B). This indicates that in the context of the Ame1-Okp1 complex, level 299 regulation by phosphorylation may occur both in cis (only affecting the subunit itself) 300 or in trans (affecting also an interaction partner). 301 We tested the effect of Ame1-Okp1 expression from a GAL promoter in a serial dilution 302 assay. In a wild-type strain background, AO overexpression was tolerated well. In a 303 skp1-3 mutant background however, overexpression of AO, either in wild-type form or 304 with Cdk1 sites mutated to alanine, compromised growth at 30 °C and 34 °C ( Figure  305 4C). Similar results were obtained for the cdc34-2 mutant background, in which 306 overexpression of AO already greatly impaired growth at 30 °C. These effects are 307 consistent with AO being physiological substrates of the SCF machinery and they 308 show that accumulation of AO can negatively impact cell growth. Further, Leu56 of the peptide interacts with Arg443 and Thr465 of Cdc4, while the 326 peptide residues Val47, Pro53, and Pro54 also interact with protein residues 327 (Supplementary Figure 4D). Additionally, several van der Waals contacts are 328 established between the peptide (through Pro49, Leu51, Lys55, and Ile50) and Cdc4 329 (involving Leu637, Thr677, Ile676, Trp717, Ser464, Tyr574, and Gly636), indicating 330 an optimal fit of the peptide at the protein binding site. Overall, this strong network of 331 peptide-protein interactions indicates that the VQPILTPPKL peptide is predicted to be 332 a potent Cdc4 binder, even more than the doubly phosphorylated Ser41/Ser45 333 peptide. 334 Next, we evaluated the Ame1-CPD ILTPP mutation in the GAL overexpression setting. 335 Strikingly, neither Ame1 nor Okp1 protein was detectable by western blotting upon 336 overexpression under these conditions and preventing phosphorylation on the five 337 remaining sites (Ame1-5A-CPD ILTPP ) did not stabilize the protein ( Figure 5B). If the 338 strong CPD indeed exerts its effect via Cdc4-dependent recognition, then the protein 339 levels of Ame1-CPD ILTPP should be restored to wild-type in an SCF mutant. Combining 340 the Ame1-CPD ILTPP allele with the cdc4-1 mutant demonstrated that this is indeed the 341 case: The Ame1-CPD ILTPP mutant and also Okp1 were detectable and displayed 342 similar levels as the wild-type proteins in a cdc4-1 background ( Figure 5B). These 343 experiments provide evidence that protein levels of Ame1 and Okp1 are regulated by 344 activation of phospho-degrons in an SCF-Cdc4 dependent manner. We also tested 345 the effect of changing motif 1 into a strong CPD in the context of endogenous Ame1. 346 Interestingly, Ame1-CPD ILTPP expressed as the sole copy of Ame1 yielded a viable 347 strain which showed slight temperature-sensitivity at 37 °C and increased sensitivity 348 to benomyl ( Figure 5C). Upon combination with a cdc4-1 mutant, however, Ame1-349 CPD ILTPP was able to partially suppress the growth defect of cdc4-1 at 34 °C and In the experiments described above, cells were challenged with increased levels of 358 Ame1 following expression from a GAL promoter. How does this relate to the 359 regulation of endogenous Ame1? To test this, we constructed Ame1 mutants 360 expressed from their endogenous promoter. To simplify the complex phosphorylation 361 pattern of wild-type Ame1, we generated alanine mutants that either allowed 362 phosphorylation of the motifs 1 and 2, but prevented phosphorylation of the remaining 363 sites (Ame1-3A, CPD only), or, conversely, prevented motif 1 and 2 phosphorylation, 364 but allowed the remaining sites to be phosphorylated (Ame1-4A, CPD null). Western 365 blotting showed that motif 1/2 phosphorylation of endogenous Ame1 was cell cycle 366 dependent ( Figure 6A). S-Phase arrested cells displayed a single slowly migrating 367 Ame1 form in addition to unmodified Ame1, while M-Phase arrested cells were 368 maximally phosphorylated with two slowly migrating forms becoming apparent. In the 369 Ame1-4A mutant (CPD null), all slowly migrating forms were eliminated ( Figure 6A). 370 We followed Ame1 phosphorylation over the course of the cell cycle after release from 371 alpha-factor. Consistent with the analysis of the arrests, we observed that motif 1/2 372 phosphorylation occurred in a step-wise manner with fully phosphorylated forms 373 appearing 30 minutes into the cell cycle.   This study reveals novel aspects of phospho-regulation at the budding yeast inner 446 kinetochore. We show that an important function of Cdk1 phosphorylation is to 447 generate phospho-degron motifs on selected inner kinetochore subunits, including the 448 essential COMA subunit Ame1, which are then recognized by the conserved ubiquitin 449 ligase complex SCF with its phospho-adapter Cdc4. We note that Ame1 and Mcm21 450 peptides were also identified in a large-scale proteomic study geared towards 451 enriching peptides simultaneously regulated by phosphorylation and ubiquitination 452 (Swaney et al., 2013). In this context, ubiquitination of COMA subunits was detected 453 for Okp1 (on residue Lys57) and Mcm21 (on residue Lys229). 454 While Cdk1 has been thought to promote kinetochore assembly in most contexts 455 investigated so far, our study indicates that it can also act as a negative regulator of 456 kinetochore assembly by targeting subunits for ubiquitination and subsequent 457 degradation by the proteasome. This seems counterintuitive at first, given that 458 kinetochores perform their essential role in segregating sister chromatids during 459 mitosis. There are, however, important aspects in which the inner kinetochore subunit 460 Ame1 differs from previously studied SCF substrates: While for example the Cdk1 461 inhibitor Sic1 is fully degraded at the G1-S transition to allow replication initiation, our 462 experiments indicate that only a subset of Ame1 is phosphorylated and subjected to 463 the SCF-dependent pathway. Our biochemical experiments furthermore suggest that 464 the pool of Ame1 regulated by this mechanism corresponds to molecules that are not 465 bound to their binding partner within the kinetochore, the Mtw1c. Such excess Ame1 466 subunits were also present in our GAL-induced overexpression setting, in which we 467 initially characterized the SCF-dependent regulation of Ame1. While this experiment 468 creates an artificial situation in which the cell is challenged with an increased level of 469 Ame1, this scenario likely also applies to the natural kinetochore assembly process. 470 To ensure effective assembly during S-Phase, free kinetochore subcomplexes must 471 be present in excess amounts, otherwise they would become limiting for assembly 472 and prevent the effective formation of a new kinetochore. On the other hand, excess 473 free subcomplexes could favor ectopic assembly which would lead to genetic 474 instability. As shown in Figure 7, Mtw1c binding of Ame1-Okp1 subcomplexes shields 475 degron phosphorylation. This Mtw1c binding sensitive phosphorylation could ensure 476 that only free, unused subcomplexes are removed by degradation ( Figure 9A). From 477 a structural standpoint, kinetochore subcomplexes typically combine relatively short, 478 structured segments (often coiled-coil domains) with large unstructured domains that 479 are the preferred targets of phosphorylation. In this context, phospho-degrons could 480 be ideally suited as assembly-sensors for kinetochores since they allow to distinguish 481 excess subunits from properly assembled ones. By placing individually weak degron 482 signals on separate subunits, the cell may allow COMA assembly from AO and CM 483 complexes, while phosphorylation of the assembled COMA then creates stronger 484 composite binding sites for Cdc4 ( Figure 9A). 485 The observation that even placing the strong CPD ILTPP degron into Ame1 does not 486 eliminate the endogenous protein fully, is consistent with the idea that at least some 487 budding yeast kinetochores are always assembled and that therefore a significant 488 fraction of Ame1 is continuously protected from phosphorylation and subsequent 489 degradation. Elimination of those COMA molecules that are not bound to Mtw1c might 490 be important for multiple reasons: Increased levels of COMA may be dangerous 491 because they could facilitate the formation of ectopic kinetochores. Indeed, we show 492 that GAL-based overexpression of all four subunits of the COMA complex is toxic for 493 cells. COMA might be an especially important target for regulation in this regard, 494 because it is near the top of the assembly hierarchy and contacts multiple other inner 495 kinetochore subunits (Yan et al., 2019). It also contains DNA-binding elements, such 496 as AT hooks, that may allow it to bind to chromosome arms when not properly targeted 497 to centromeres. 498 In summary, we propose the following model for COMA phospho-regulation by 499 Cdc28 Cdk1 (Figure 9B): Initial phosphorylation on Ame1 motifs 1 and 2 starts in S-500 Phase but is not complete before M-Phase. This ensures that sufficient free 501 subcomplexes are available for kinetochore assembly. In parallel, the observation that 502 individual degrons on AO or CM are weak, permits COMA assembly from its 503 subcomplexes. In M-Phase, full degron phosphorylation destabilizes assembled 504 COMA complexes, unless they are bound to the Mtw1 complex, which shields the 505 degrons. The proper timing of phosphorylation is critical in this model. Premature 506 phosphorylation of COMA would target it for destruction too early, likely compromising 507 kinetochore assembly in S-Phase. This could be the reason, why the Ame1 phospho-508 sites are only gradually phosphorylated in vivo. Conversely, placing the stronger 509 CPD ILTPP phospho-degron into Ame1 compromised growth in otherwise wild-type 510 cells, but partially rescued the growth phenotypes of cdc4-1 mutants. This argues that 511 indeed key mitotic SCF substrates reside at the inner kinetochore. Notably, the 512 Histone H3 variant Cse4 has recently been reported to be an SCF substrate (Au et al., 513 2020). 514 While our experiments strongly implicate Ame1 as a Cdk1 and SCF substrate, 515 they also show that the Ame1-7A mutant effectively prevents phosphorylation in vivo 516 and in vitro, but does not induce a strong mitotic delay such as the skp1-3 mutant. We 517 speculate that the Ame1-7A mutant is not sufficient to fully prevent SCF regulation of 518 the inner kinetochore, and additional phospho-targets must exist. The molecular 519 requirements for the recognition of phosphorylated substrates may be complex in the 520 case of multi-protein complexes such as COMA. Here, degron sequences located on 521 multiple subunits and phosphorylated in different combinations might be required for 522 effective binding by Cdc4. This could involve both non-consensus sites 523 phosphorylated by Cdk1, and also kinases in addition to Cdk1, as is the case for 524 Previous work has defined the role of SCF-Cdc4 at the G1-S transition, but has 534 also shown that Cdc4, as well as Cdc34 or Cdc53, remain essential genes in sic1D 535 mutants (Schwob et al., 1994), demonstrating that there must be additional key 536 substrates. Furthermore, SCF mutants that complete replication, arrest at G2-M with 537 a short spindle and an activated mitotic checkpoint (Goh and Surana, 1999 Complex, kinetochores are also molecularly linked to SCF, the other major RING-type 546 E3 ligase complex that regulates the eukaryotic cell cycle. 547 548

Clb2) 552
Expression constructs for kinetochore proteins AO, Mtw1c and Mtw1-Nnf1 (MN) and 553 the kinase complex Cdc28-Clb2 used in this study were created by amplification of the 554 DNA for the respective genes from yeast genomic DNA and cloning into pETDuet-1, 555 pET3aTr/pST39, pST44 plasmids (bacterial expression) following the protocol for 556 restriction free cloning or pESC two-micron plasmids (yeast expression) using 557 classical cloning methods. Restriction free cloning was also used to produce vectors 558 encoding phospho-eliminating mutants in Ame1. Site directed mutagenesis (Agilent 559 technologies) was applied for introduction of amino acid substitutions. A list of all 560 vectors used for protein production and purification in bacteria or insect cells can be 561 found in Table 3.

Integration/replacement constructs) 645
Yeast strains were constructed in the S288C or W303 (SCF strains) background. A 646 list of all yeast strains used in this study can be found in Table 4, a list of all vectors 647 used for generation of novel yeast strains can be found in Table 3. Yeast strain 648 generation and methods were performed by standard procedures (Daniel et al., 2006). 649 The anchor-away approach for characterization of Ame1 in SCF mutants or wild-type 650 strain was performed as described (Haruki et al., 2008), using the ribosomal RPL13-651 FKBP12 anchor. Final rapamycin concentration in plates or liquid media was 1 μg/ml. 652 Serial two-fold dilutions of overnight cultures were prepared on 96-well plates in 653 minimal medium starting from OD600 of 0.4 for anchor-away approach or 0.5 for 654 overexpression approach. The dilutions were spotted on YPD medium with and 655 without rapamycin or on minimal medium with either S-glucose or S-raffinose + 656 galactose (2 % each) and grown at 30 °C for 2-3 days. To confirm phenotypes 657 observed in the serial dilution assays for Ame1 mutants, Ame1 hemizygous deletion 658 strains were used to introduce Ame1-wild-type or phospho-mutants at an exogenous 659 locus before haploid spores were produced. Pds1-13xMyc was integrated 660 exogenously into haploid strains for cell cycle experiments. For overexpression of 661 proteins, pESC two-micron plasmids were integrated in haploid wild-type or SCF 662 mutant strains without integration into the genome and clone pools were used for 663 further analyses. Selective pressure was used for maintenance of the plasmids. 664 Expression of integrated proteins was checked for all created yeast strains by protein 665 extraction from yeast (Kushnirov, 2000) and western blotting against the respective 666 tags of individual proteins (see Table 2). 667

GAL overexpression 669
For overexpression studies strains containing two-micron plasmids were grown 670 overnight in YEPD. Next morning, cells were washed twice in YEP-raffinose (2 %) and 671 incubated in YEP+R for 3 hours at 30 °C. Overexpression was induced with the 672 addition of 2 % galactose and timepoints were taken after 0, 3 and 5 hours in YEP+RG. 673 Protein extracts (Kushnirov, 2000) for western blotting analysis were prepared for 674 Mesh Ewald (PME) for computing the long-range electrostatic interactions (Cheatham 740 et al., 1995). The systems were minimized in two steps (using 10,000 conjugate 741 gradient and 10,000 steepest descent cycles for both steps). In step1, the protein-742 peptide complex was restrained using a force constant of 50 kcal/mol A −2 , and only 743 the ions and solvent molecules were allowed to relax. In step 2, the restraints were 744 removed, and the whole assembly was allowed to relax. The minimized systems were 745 heated to room temperature (40 ps) and underwent equilibration (5 ns). Finally, GaMD 746 simulations were performed for 100 ns (3 replicas, NPT). The trajectories were 747 processed using the Cpptraj code (Roe and Cheatham, 2013), and the downloadable 748 version of the Ligplot tool was used to create 2D interaction plots (Wallace et al., 749 1995 CENP-T 90 17 2 (T42, S192) 2 (T21, S177) Mhf1 CENP-S 93 3 1 (T34) -Mhf2 CENP-X 95 1 1 (S60) -978