Direct observation of interdependent and hierarchical kinetochore assembly on individual centromeres

Yeast strains and genetic modifications

The S. cerevisiae strains are described in Supplemental Table S1 and are derivatives of SBY3 (W303). Standard genetic crosses, media preparation, and microbial techniques were used. Cse4 was internally tagged with EGFP at residue 80, flanked by linkers (pSB1617), and expressed under its native promoter at the endogenous locus (SBY22195). Other kinetochore genes were modified to include endogenously tagged GFP (pSB3156) and auxin-inducible degrons (-IAA7) using standard polymerase chain reaction (PCR)-based integration methods (Longtine et al., 1998), with modifications confirmed via PCR. The mif2 mutants were generated by CRISPR-based integration using vectors carrying Cas9 (pSB3218) and the donor sequences mif2-ΔN (pSB3354) and mif2^AME1N (pSB3237) [21]. The primers and plasmids used to construct the strains are listed in Supplemental Table S2.

Culture conditions

All liquid cultures were maintained in yeast peptone dextrose (YPD) medium. To synchronize cells in mitosis, 30 μg/ml benomyl was added to log-phase cultures for 2.5 h, ensuring at least 90% of cells exhibited large-budded morphology. For strains with auxin-inducible degron (AID) alleles (dsn1-AID), cultures were treated with 500 μM indole-3-acetic acid (IAA) dissolved in dimethyl sulfoxide (DMSO) with 30 μg/ml benomyl, following methods described [22–24]. Strains with galactose-inducible alleles were initially cultured in 2% raffinose before being cultured in 2% galactose until they were harvested at the specified time points. To achieve G1 arrest, yeast cells were treated with 1 μg/ml alpha factor for 3 h.

Cell fitness serial dilution assay

Yeast strains were cultured overnight in YPD medium. Cell density was determined using a spectrophotometer (Bio-Rad) and adjusted to an OD₆₀₀ of 1.0. Serial 1:5 dilutions were then prepared in water using a 96-well plate, and aliquots from each well were spotted onto YPD or YPD plates supplemented with benomyl (Sigma–Aldrich). The plates were incubated at the specified temperatures for two days before imaging.

Cell extract

The procedure for preparing cell lysate for kinetochore assembly assays was previously described [24]. Briefly, cells were cultured in liquid YPD medium until reaching logarithmic growth and arrested in mitosis with benomyl (30 μg/ml) in a total volume of 1 l, followed by harvesting via centrifugation at 4°C. All subsequent steps were carried out on ice using buffers maintained at 4°C. The cells were washed once with dH₂O containing 0.2 mM PMSF and then with Buffer L (25 mM HEPES, pH 7.6, 2 mM MgCl₂, 0.1 mM ethylenediaminetetraacetic acid, pH 7.6, 0.5 mM ethylene glycol tetraacetic acid (EGTA), pH 7.6, 0.1% NP-40, 175 mM K-Glutamate, and 15% glycerol) supplemented with protease inhibitors (10 μg/ml each of leupeptin, pepstatin, and chymostatin, 0.2 mM phenylmethylsulfonyl fluoride (PMSF)) and 2 mM dithiothreitol (DTT). The cell pellets were snap-frozen in liquid nitrogen and lysed using a Freezer/Mill (SPEX SamplePrep). The lysis involved 10 cycles, each consisting of 2 min of grinding at 10 times per second, followed by a 2-min cooling period. The resulting powder was weighed and resuspended in Buffer L using the formula: pellet weight (g) × 0.8 = volume (ml) of Buffer L. The resuspended lysate was thawed on ice, clarified by centrifugation at 16,100 × g for 30 min at 4°C, and the protein-rich supernatant was extracted using a syringe. The lysate was then aliquoted and snap-frozen in liquid nitrogen. The resulting soluble whole-cell extracts typically had a protein concentration of 70–80 mg/ml. Pellets, powder, and lysates were stored at −80°C.

DNA template preparation

As previously detailed [22], plasmid pSB963 was utilized to produce the wild-type CEN3 DNA templates, containing the centromere and surrounding pericentromeric sequences (277 bp total) flanked by additional linker DNA (212 bp upstream of CDEI and 302 bp downstream of CDEIII). Plasmid pSB972 was used to create the CEN3 mutant template for this study. PCR products were purified using the Qiagen PCR Purification Kit. For single-molecule TIRFM assays, the purified CEN DNA was diluted in tris(hydroxymethyl)aminomethane and ethylenediaminetetraacetic acid (TE) buffer to a final concentration of ∼1 ng/μl.

TIRFM slide preparation

Coverslips and microscope slides were ultrasonically cleaned and polyethylene Glycol (PEG)-passivated as described previously [19, 20, 25]. Briefly, ultrasonically cleaned slides were treated with 3-aminopropyl triethoxysilane, 99% (Sigma–Aldrich) before incubation with 1% (w/v) biotinylated PEG-SVA MW-5000K/mPEG-SVA MW-5000K (Laysan Bio) in flow chambers constructed using double-sided tape. Passivation was conducted overnight at 4°C. Following passivation, the flow chambers were washed with Buffer L and then incubated with 0.3 M bovine serum albumin/0.3 M Kappa Casein in Buffer L for 5 min. The chambers were subsequently washed with Buffer L, incubated with 0.3 M Avidin DN (Vector Laboratories) for 5 min, and washed again with Buffer L. Next, ∼100 pM CEN3 DNA template was added to the chambers for 5 min, followed by another wash with Buffer L.

For endpoint colocalization assays, slides were prepared as follows: flow chambers were filled with 100 μl of cell lysate containing the protein(s) of interest using pipetting and wicking with filter paper or Pasteur pipettes. After lysate addition, slides were incubated for 90 min at 25°C, then washed with Buffer L to remove the lysate. The chambers were subsequently filled with Buffer L containing an oxygen scavenger system (10 nM protocatechuate-3,4-Dioxygenase (PCD), 2.5 mM protocatechuic Acid (PCA), and 1 mM Trolox) for imaging [26].

TIRFM time series imaging and analysis

As previously detailed [19, 20], all images were captured using a Nikon TE-2000 inverted RING-TIRF microscope equipped with a 100× oil immersion objective (Nikon Instruments) and an Andor iXon X3 DU-897 EMCCD camera. Images were acquired at a resolution of 512 × 512 pixels with a pixel size of 0.11 μm/pixel or 0.16 μm/pixel at a readout speed of 10 MHz. Atto-647-labeled CEN DNAs were excited at 640 nm for 300 ms, GFP-tagged proteins at 488 nm for 200 ms. Images were analyzed with the CellProfiler (4.2.6) to assess colocalization and quantify signals between the DNA channel (647 nm) and GFP channel (488 nm). Results were processed and visualized using FIJI (https://imagej.net/software/fij). Adjustments to example images, such as contrast or false color, were made using FIJI and uniformly applied across the entire field of view for each image. The lysates were washed off before imaging. At least 3000 DNA molecules were imaged for each time point from each repeat. Error bars are standard deviation over three biological repeats.

Colocalization-vs-time data were fit (in Igor Pro 9) with one of the four equations given below, which represent simple kinetic reaction pathways having either one, two, three, or four irreversible transitions (irreversible steps). Such simple reaction pathways do not explicitly model the complexity of kinetochore assembly, but they closely approximate the kinetics of more complex reaction pathways, and they provide excellent fits to our data using only two adjustable parameters. For a simple one-step assembly pathway, with one irreversible binding event, the fraction of complete assemblies will vary across time according to equation:

where represents time (e.g. in min), represents the rate constant (min⁻¹) for the transition, and represents the maximum fraction achieved when . For a simple two-step assembly pathway, with two irreversible transitions both having identical rate constants, , the fraction of complete assemblies will vary according to equation:

For a three-step assembly pathway, with three irreversible transitions all having identical rate constants, , the fraction of complete assemblies will vary according to

For a four-step assembly pathway, with four irreversible transitions all having identical rate constants, , the fraction of complete assemblies will vary according to

In general, the curve for a pathway with more step transitions exhibits a longer initial lag phase followed by a steeper rise to . We fit every dataset with all four equations, for one, two, three, and four steps, and then selected the best-fitting curve, with the minimum chi-squared deviation from the data. The rate constant, , determines the timescale over which rises to , but the best-fit values for are unimportant here because they do not correspond to any specific step in kinetochore assembly. Therefore, to compare curves we used , the time at which each curve reached a threshold of 30% (arbitrarily chosen). The best-fit values and uncertainties for and , the numbers of step transitions in the best-fitting curves, and the resulting chi-squired deviations between the data and the fits are all included in Supplemental Table S3.

In vivo microscopy and analysis

Fixed cell images were captured using a DeltaVision Ultra deconvolution high-resolution microscope (GE Healthcare) equipped with a 100×/1.42 PlanApo N oil immersion objective (Olympus) and a 16-bit sCMOS detector. Cells were imaged in Z-stacks through their entire volume, with 0.4 μm steps, using plane-polarized light, 488 and 568 nm illumination. All images were deconvolved using standard settings, and Z projections of the maximum signal from all channels were exported as TIFF files for analysis in FIJI.

Centromeric region puncta in cells were identified using 568 nm images of Ndc10-mCherry. Signal intensity within these regions was quantified for the 568 nm channel and the corresponding 488 nm channel on a per-cell basis. Total intensity in the 488 nm channel was normalized to the total 568 nm intensity for each cell, and the normalized values were averaged across ∼33 cells per biological replicate, with three biological replicates analyzed per strain. Representative images shown from these experiments are maximum intensity projections across all Z images. Adjustments to example cell images, such as contrast, were made uniformly using FIJI.

Immunoblotting

Cell lysates were made by beat-beat pulverization (Biospec Products) with glass beads in sodium dodecyl sulphate buffer [27]. Sodium dodecyl sulphate–polyacrylamide gel electrophoresis gels were transferred to 0.45 μm nitrocellulose membrane using a wet transfer apparatus (Bio-Rad). The following antibodies were used for immunoblotting: α-PGK1 (Invitrogen; 4592560; 1:10,000), α-Cse4 (9536L, 1:500) [28]. Secondary antibodies used were sheep anti-mouse antibody conjugated to horseradish peroxidase (HRP; GE Biosciences) at 1:10,000 or donkey antirabbit antibody conjugated to HRP (GE Biosciences) at 1:10,000. Antibodies were detected using SuperSignal West Dura Chemiluminescent Substrate (Thermo Fisher Scientific). Immunoblots were imaged with a ChemiDock MP system (Bio-Rad).

Flow-cytometric analysis of mini-chromosome missegregation

A single-cell-based mini-chromosome loss assay was performed as described [29]. Briefly, a minichromosome that contains the Matα locus was introduced into strains of interest that contain a Mata specific gene tagged with GFP (Mfa1-3XGFP). Mfa1-3XGFP is repressed by the expression of the α2 transcriptional repressor from the minichromosome, so Mfa1-3XGFP is only expressed when the minichromosome is lost. Fixed cells were sonicated, and GFP-positive cells were detected and quantified by flow cytometry.

A single molecule assay to monitor the kinetics of kinetochore assembly

We previously developed an approach using TIRFM to observe the de novo assembly of individual centromeric nucleosomes and whole kinetochore particles in cell lysates [19, 25]. These efforts revealed molecular requirements for centromeric nucleosome assembly and directional asymmetry in the kinetochore-microtubule interface [19, 20, 25]. However, the kinetics of kinetochore assembly beyond the centromeric nucleosome remained unexplored.

To examine kinetochore assembly kinetics more broadly, we adapted our TIRFM approach to monitor the recruitment of various kinetochore proteins onto individual centromeric DNAs via time-series imaging. Atto-647-labeled DNA containing the centromere and surrounding pericentromeric sequences from chromosome III (CEN3 DNA) as well as additional linker DNA (791 bp total) were sparsely tethered onto a coverslip surface to allow detection of individual CEN3 molecules (Fig. 1B). One-step photobleaching confirmed that each CEN3 focus represented a single molecule (Supplemental Fig. S1A–C) [30]. To initiate kinetochore assembly on the surface-tethered CEN3 DNAs, we introduced yeast extracts containing GFP-tagged kinetochore proteins. We epitope tagged a representative component of each kinetochore subcomplex at its endogenous gene locus with GFP (Table 1) and confirmed that the GFP tag did not affect cell fitness at various temperatures and in the presence of microtubule-destabilizing drug treatments (Supplemental Fig. S1D). Because many CCAN proteins are nonessential, we used a more sensitive assay to determine whether the GFP tag affected their function. The nonessential CCAN genes become essential in a dsn1-3A mutant background [24, 31], so we crossed the GFP-tagged genes to dsn1-3A to confirm they are functional (Supplemental Fig. S1D). Once we had established a functional GFP tagged protein for each kinetochore subcomplex, we made lysates from cells arrested in mitosis with benomyl, to ensure the lysates were maximally competent for assembling kinetochores [24]. To minimize possible variation in the rate of kinetochore assembly, we used a consistent lysate concentration (70 mg/ml) and incubation temperature range (21°C–23°C) for the assay.

To establish the time series assay, we initially analyzed two inner kinetochore proteins (Ndc10-GFP and Mif2-GFP). Ndc10 is a component of the CBF3 complex that is required for assembly of the entire yeast kinetochore, including the Cse4 histone variant, suggesting that CBF3 binding is the initiating event [32–34]. The Mif2 protein interacts directly with the centromeric nucleosome, suggesting it assembles after Cse4. We incubated each lysate with surface-tethered CEN3 DNAs for various durations (0, 5, 10, 30, 60, 90, 120, 150, and 180 min) and then washed the lysate away prior to imaging. Removing the lysate ensured stable protein binding and a high signal-to-noise ratio, free from background due to out-of-focus GFP or autofluorescence. Multicolor TIRFM images were then recorded, and colocalization single-molecule spectroscopy (CoSMoS) was performed (Fig. 1C) [19, 25, 35, 36]. The fraction of wild-type CEN3 DNAs decorated with GFP increased as the incubation time was extended, until reaching a maximum. The rates of increase and the maximum levels of colocalization were markedly different for the two kinetochore proteins (Fig. 1C and D and Supplemental Fig. S1E). Colocalization of Ndc10-GFP on wild-type CEN3 DNAs rose above 50% in just 10 min and reached a maximum near 90% in 30 min. In contrast, Mif2-GFP assembled much more slowly than Ndc10-GFP, requiring 30 min to achieve 10% colocalization and 120 min to reach a maximum around 45%, consistent with the known requirement for Ndc10 to localize Mif2 [32–34]. As a negative control, we also tested a mutant CEN3 DNA containing a 3-bp substitution that abolishes kinetochore assembly [24, 37–39]. The fraction of CEN3 mutant DNA molecules decorated with Ndc10-GFP never exceeded 1%, confirming the specificity of Ndc10 binding.

We fit the colocalization versus time data for each protein with kinetic curves representing simple reaction schemes (as detailed in ‘Materials and methods’ section). To compare the assembly kinetics between proteins, we used , the time at which each curve reached a threshold of 30%, and the maximum colocalization, , which occurred when the curve plateaued at a steady value. The Ndc10-GFP colocalization plateaued at = 87 ± 1% and reached 30% colocalization at = 4.0 ± 0.2 min, consistent with the essential, early role Ndc10 plays in initiating kinetochore assembly [33]. The Mif2-GFP data plateaued at = 47 ± 1% and reached 30% colocalization at = 70 ± 4 min, indicating that Mif2 assembles later than Ndc10, after a significant delay (Fig. 1D). Altogether, these data show that the assembly of GFP-tagged kinetochore components specifically onto individual CEN3 DNAs can be monitored in time-series TIRFM, and that the rates and efficiencies of assembly are different for various components.

Inner kinetochore subcomplex incorporation is consistent with hierarchical assembly

Prior work indicates that kinetochore proteins are often expressed as biochemically stable, obligate heteromeric subcomplexes, most with two to four protein members [5, 6, 40]. Obligate members of the same subcomplex are expected to assemble onto the kinetochore simultaneously; distinct (nonobligate) subcomplexes, however, are thought to assemble hierarchically, with DNA-contacting inner subcomplexes assembling first and outer subcomplexes incorporating later, only after their direct contacts in the inner kinetochore have already assembled [40]. Co-localization dependencies and biochemical co-immunoprecipitations are consistent with this view, but the relative timing of subcomplex incorporation has not been measured directly, owing to the high-speed and complexity of the assembly process and the difficulty of distinguishing single kinetochores in vivo. We therefore extended our approach to compare assembly kinetics for representative members of every core kinetochore subcomplex, beginning with Cse4 (centromeric nucleosome) and the other five subcomplexes that make up the CCAN (Table 1).

The second component to assemble after Ndc10 was the centromeric histone Cse4 that reached 30% at = 12 ± 1 min, only 7 min after Ndc10, and reached a maximum of = 71 ± 1% by 90 min (Fig. 2A). Next was the essential CCAN subcomplex, CENP-QU, represented by Ame1, reaching a similar maximum ( = 79 ± 3%) but more slowly, with = 33 ± 3 min. The remaining four nonessential CCAN subcomplexes, assayed using GFP-tagged Ctf19 (CENP-OP) [37], Ctf3 (CENP-HIK) [41], Iml3 (CENP-LN) [42] and Wip1 (CENP-TW) [17, 40], all assembled later and with lower maximum efficiency, following kinetic curves roughly similar to each other and to that of Mif2, with = 38 ± 1% (Ctf19), 56 ± 1% (Ctf3), 37 ± 1% (Iml3), and 49 ± 2% (Wip1) and = 74 ± 10 min (Ctf19), 68 ± 2 min (Ctf3), 72 ± 3 min (Iml3), and 89 ± 7 min (Wip1), respectively (Supplemental Table S3). The similar timing of the nonessential CCAN components suggests they may be pre-assembled. We attempted to test more directly for the arrival of pre-assembled CCAN components using two-color, fast time-lapse imaging, in lysates where two different proteins were labeled with different fluorophores. Unfortunately, the relative fluorophore photostability and autofluorescence from the lysate precluded reliable detection of co-arrivals when obligate members of the same subcomplex were tagged (e.g. Okp1-GFP and Ame1-mKate).

Taken together, these results are consistent with a hierarchical inner-kinetochore assembly process, initiated by Ndc10 and followed by Cse4 histone incorporation. One of the essential CCAN subcomplexes, represented by Ame1, then binds to the centromere followed by the other essential CCAN protein, Mif2, and the remaining four nonessential CCAN subcomplexes (Fig. 2B). In general, the tended to decrease the further a subcomplex was from the CEN3, consistent with our prior observations of de novo kinetochore assembly in lysates assayed by immunoblotting, where centromere-distal components assembled with reduced efficiency compared to centromere-adjacent components [24].

Relieving autoinhibition of the Mtw1 subcomplex improves outer kinetochore assembly

We next sought to analyze outer kinetochore assembly using a representative GFP-tagged protein from three outer kinetochore subcomplexes, Dsn1 (Mtw1c) [28, 40, 43], Nuf2 (Ndc80c) [44, 45], Kre28 (KNL subcomplex) [43], as well as the Stu2 protein (human chTOG homolog) [23, 32]. Unfortunately, the overall colocalization for the outer kinetochore subcomplexes was too low (∼2–10%) for reliable kinetic analyses (Fig. 3A and Supplemental Fig. S3A). The inefficient assembly of the outer kinetochore is consistent with our previous study where de novo assembly was measured in a bulk assay by immunoblotting [24]. To overcome this limitation, we added phosphomimetic aspartic acid substitutions at two serine residues in the Dsn1 protein, S240D, S250D (dsn1-2D), which enhance outer kinetochore assembly by relieving autoinhibition of the Mtw1 subcomplex to promote its binding to Mif2 (Fig. 3B) [24, 31, 46]. Consistent with bulk kinetochore assembly assays [24], lysates from dsn1-2D strains produced much higher maximum colocalizations for GFP-tagged outer kinetochore proteins in the time-series TIRFM assay with = 54 ± 2% (Dsn1-2D), 57 ± 1% (Nuf2), 52 ± 2% (Kre28), and 48 ± 1% (Stu2) (Fig. 3C, Supplemental Fig. S3A, and Supplemental Table S3). Dsn1-2D assembly reached 30% at = 56 ± 5 min (Fig. 3). Nuf2 assembled more slowly than Dsn1-2D, reaching 30% at = 88 ± 4 min (Nuf2). Finally, Kre28 and Stu2 assembled last with = 146 ± 9 min (Kre28) and 134 ± 6 min (Stu2), respectively. Altogether, these data confirm that relieving autoinhibition of the Mtw1 subcomplex by phosphomimetic mutations significantly improves outer kinetochore assembly in the assay, and they suggest that the Mtw1 subcomplex likely assembles first, followed by the Ndc80 subcomplex, and then the KNL subcomplex and the Stu2 protein.

A highly interdependent relationship between Cnn1 and Mtw1 assembly

The Ndc80c has two distinct kinetochore receptors in yeast, the Mtw1 and Cnn1 subcomplexes, that are thought to compete for binding to Ndc80c [18, 47]. Although the autoinhibited Mtw1c in wild-type lysates did not assemble well (∼10% colocalization for Dsn1-GFP after 300-min incubation), Wip1-GFP, representing the Cnn1/Wip1 yeast homolog of the CENP-TW subcomplex, assembled relatively efficiently ( = 49 ± 2%) (Fig. 3D). It was therefore surprising that Ndc80c did not assemble well (∼8% for Nuf2-GFP), which indicates that most of the centromere-bound Cnn1 in wild-type lysates failed to recruit Ndc80c (Fig. 3D). For another test of the contribution of Cnn1 to Ndc80c recruitment, we measured Nuf2-GFP assembly in lysates from mutant cnn1Δ cells. Nuf2-GFP assembly dropped from ∼8% in wild type to ∼3% in the cnn1Δ lysates (Fig. 3D and Supplemental Fig. S3B). However, Dsn1-GFP assembly also dropped by a similar amount, from ∼10% to ∼3%, suggesting that Cnn1/Wip1 somehow facilitates Mtw1c recruitment but might not directly recruit Ndc80c in the TIRF assay (Fig. 3D and Supplemental Fig. S3C). To further test this hypothesis, we analyzed Mtw1c and Ndc80c recruitment in the presence or absence of Cnn1 and with Mtw1c autoinhibition relieved by including the dsn1-2D substitutions. Similar to wild-type Dsn1-GFP, we found that Dsn1-2D-GFP assembly dropped when Cnn1 was deleted, from 54 ± 2% to 37 ± 1% (Fig. 3D). Nuf2-GFP assembly also dropped, by a similar amount (from 57 ± 1% to 31 ± 2%) (Fig. 3D). These close correspondences between Ndc80c and Mtw1c recruitment, and the poor correspondence of Ndc80c recruitment with Cnn1/Wip1 recruitment, across all four strain backgrounds (± cnn1Δ, ± dsn1-2D), suggest that Mtw1c is the major recruiter of Ndc80c in the assay. The consistent drop in Mtw1c recruitment (by about half) upon deletion of Cnn1 suggests that the Cnn1/Wip1 subcomplex contributes to assembly of the Mtw1 subcomplex via an unknown mechanism, thereby facilitating assembly of Ndc80c indirectly.

Relieving Mtw1 autoinhibition promotes CCAN and Mif2 assembly

In the simplest models of hierarchical assembly, stable incorporation of inner DNA-proximal subcomplexes would occur independently of the outer subcomplexes, and altering the outer Mtw1 subcomplex (e.g. using dsn1-2D) would not be expected to affect the assembly of inner subcomplexes. To test this hypothesis, we measured assembly kinetics in dsn1-2D lysates for the same inner-kinetochore subcomplexes that we had measured earlier in DSN1 cells. As expected, the kinetics for GFP-tagged Ndc10 and Cse4 were similar to those measured in wild-type lysates ( = 88 ± 1% and 76 ± 1%, respectively; and = 2.6 ± 0.2 min, 12 ± 1 min, respectively) (Fig. 4A and B). However, Ame1-GFP showed slightly lower colocalization and slower assembly kinetics the in dsn1-2D background ( = 77 ± 3% and = 49 ± 4 min) for reasons that are unclear (Fig. 4C). The GFP-tagged nonessential CCAN components, Ctf19, Iml3, and Ctf3 assembled in dsn1-2D lysates with similar timing relative to wild-type lysates ( = 53 ± 8, 52 ± 2, and 69 ± 23 min). However, they reached higher maximum efficiencies ( = 80 ± 5%, 60 ± 1%, and 69 ± 1%) (Fig. 4D–F). Wip1-GFP showed similar kinetics in wild type and dsn1-2D (Supplemental Fig. S4A). The protein that exhibited the biggest change was Mif2-GFP, which assembled much earlier in dsn1-2D lysates ( = 26 ± 1 min), now preceding Ame1 instead of following it and reaching higher maximum colocalization than in wild-type lysates ( = 70 ± 1%) (Figs 4G and H, and 5A, and Supplemental Fig. S4B). These observations show that the timing and efficiency of inner subcomplex assembly can be strongly affected by modifications of outer subcomplexes, further indicating that kinetochore assembly occurs through a network of interdependent steps, rather than a simple sequence of independent events.

Relieving Mtw1c autoinhibition via dsn1-2D also relieves Mif2 autoinhibition

Adding phosphomimetic dsn1-2D substitutions to the outer Mtw1 subcomplex improved assembly of the inner Mif2 protein, so we tested whether the improved assembly required Mtw1c. To do this, we depleted Mtw1c using a dsn1-AID strain. For reasons that are unclear, Mif2 assembly in the dsn1-AID lysates that were not treated with auxin was higher than the wild-type strain. However, auxin treatment led to a reduction in Mif2 incorporation, especially at earlier time points (Fig. 5B). Together, these data confirm that Mif2 assembly kinetics are regulated by Mtw1c.

A previous study showed that, like Mtw1c, Mif2 is also autoinhibited [Fig. 5C(i)] [21]. The Mif2 N-terminal region, which directly recruits Mtw1c, can instead form autoinhibitory, intramolecular bonds to other regions of Mif2 that normally anchor it to the centromeric nucleosome. When Mif2 interacts with Cse4 nucleosomes, its autoinhibition is relieved, allowing its N-terminal region to bind Mtw1c [21]. Our discovery that dsn1-2D improves Mif2 assembly suggested that Mtw1c binding may also help stabilize Mif2’s “open” status, promoting its binding to Cse4 [see right side of Fig. 5C(i)]. To further test this idea, we analyzed two Mif2 mutants lacking autoinhibition, Mif2-ΔN and Mif2^Ame1N [Figs 5C (ii) and (iii)]. We first deleted residues 2 through 35 from the N-terminus of Mif2, creating Mif2-ΔN, and thereby eliminating its ability to bind Mtw1c as well as its autoinhibition [21, 48]. Compared to wild type, the GFP-tagged truncation mutant Mif2-ΔN assembled onto centromeric DNAs much more quickly ( = 38 ± 1 min for truncated Mif2-ΔN versus 70 ± 1 min for wild type) and to a higher final colocalization percentage ( = 84 ± 1% for truncated Mif2-ΔN, versus 47 ± 1% for wild type) (Fig. 5D), consistent with autoinhibition in the wild-type Mif2 normally limiting its assembly at the centromere. Moreover, assembly kinetics for the Mif2-ΔN truncation mutant were indistinguishable regardless of whether the lysate contained wild-type DSN1 or phosphomimetic dsn1-2D. These observations indicate that the truncation mutant Mif2-ΔN is already uninhibited and therefore does not require the relief of autoinhibition that wild-type Mif2 normally receives by associating with Mtw1c.

We next tested another Mif2 mutant, Mif2^Ame1N, that lacks autoinhibition but retains its interaction with Mtw1c [21]. The Mtw1 subcomplex interacts with the N-terminus of Ame1 in a manner that is not expected to be strongly regulated by dsn1-2D [46], so we made a Mif2 swap mutant where the N-terminus of Mif2 was replaced with the N-terminus of Ame1 [creating Mif2^Ame1N; Fig. 5C(iii)] [21]. Similar to Mif2-ΔN, the Mif2^Ame1N swap mutant assembled significantly earlier than wild-type Mif2 ( = 25 ± 1 min for swap mutant Mif2^Ame1N versus 70 ± 4 min for wild type) (Fig. 5E). Assembly kinetics for the swap mutant Mif2^Ame1N were indistinguishable regardless of whether the lysate contained wild-type DSN1 or phosphomimetic dsn1-2D, consistent with the Ame1 interaction with Mtw1c being independent of regulation by dsn1-2D (Fig. 5E). Together, these data indicate that Mtw1c binding relieves the autoinhibition of wild-type Mif2, allowing it to assemble more quickly and efficiently onto the centromere.

Interactions between Mif2 and Mtw1c promote CCAN incorporation

We next asked whether the accelerated arrival of Mif2 affects the assembly of other CCAN proteins. To do this, we analyzed assembly kinetics for GFP-tagged CCAN components in lysates carrying the mif2-ΔN truncation mutant, which lacks the N-terminus and cannot bind to Mtw1c. Ctf19 assembly was normal ( = 40 ± 2%, = 94 ± 15 min), but other nonessential CCAN components showed lower efficiency in mif2-ΔN lysates ( = 29 ± 1% for Iml3, 41 ± 1% for Ctf3, and 22 ± 1% Wip1), and slower kinetics ( = 174 ± 33 min for Iml3, 107 ± 3 min for Ctf3, and > 180 min for Wip1, which never reached 30%) relative to lysates with wild-type Mif2 (Fig. 6A–D). These data indicated that the N-terminus of Mif2 enhances CCAN assembly. To test whether this enhancement was due to Mtw1 binding to Mif2, we used the Mif2^Ame1N swap mutant. In lysates carrying the Mif2^Ame1N swap mutant, the Ctf19 subcomplex again assembled normally ( = 43 ± 3%, = 84 ± 17 min), but the other nonessential CCAN proteins, Iml3, Ctf3 and Wip1, now showed higher colocalization efficiency ( = 68 ± 1% for Iml3, 80 ± 1% for Ctf3, and 67 ± 1% for Wip1) compared to lysates with wild-type Mif2. Two CCAN proteins, Iml3 and Wip1, also exhibited faster kinetics ( = 51 ± 2 min for Iml3 and 60 ± 1 min for Wip1) (Fig. 6A–D). These observations indicate that interactions between Mtw1c and Mif2 promote assembly of many components of the inner kinetochore, a result that is inconsistent with the simplest models, where inner components would assemble completely independently of outer components.

The N-terminus of Mif2 becomes important in vivo when Cse4 is overexpressed

Although the Mif2 mutants lacking the N-terminus are viable, it was previously shown that overexpression of the Mif2^Ame1N mutant led to chromosome missegregation [21]. We wondered whether the N-terminus of Mif2 might also become critical in cells where kinetochore assembly is stressed, such as by Cse4 overexpression. To test this, we overexpressed Cse4 in MIF2-GFP, mif2-ΔN-GFP, mif2^AME1N, and dsn1-2D strains using a galactose-inducible promoter. Consistent with previous studies, Cse4 overexpression by itself did not alter cell fitness [21, 49, 50] (Fig. 7A and Supplemental Fig. S5A). We also did not detect growth defects when Cse4 was overexpressed in mif2^AME1N, dsn1-2D, or mif2^AME1N dsn1-2D double mutant strains, all of which maintain the interaction between Mif2 and the Mtw1c (Supplemental Fig. S5B and C). However, there was a strong growth defect when Cse4 was overexpressed in mif2-ΔN mutant cells, consistent with the Mif2 N-terminus becoming important when the histone variant is misregulated (Fig. 7A and Supplemental Fig. S5A). To determine whether Mif2 localization was altered, we analyzed GFP-tagged Mif2 or Mif2-ΔN in cells containing Ndc10-mCherry to mark the endogenous kinetochores. Cells were released from G1 into galactose media to induce Cse4 overexpression, and we performed microscopy on mitotic cells. In wild-type large budded cells, we nearly always observed two kinetochore foci, consistent with yeast kinetochores clustering into two foci when kinetochores biorient and come under tension [51]. However, when Cse4 was overexpressed in MIF2-ΔN-GFP cells, we often observed more than two Mif2-GFP and Ndc10-mCherry foci (Fig. 7B and C). The Mif2-GFP and Mif2-ΔN-GFP foci were colocalized with Ndc10-mCherry in >99% of the cells, although the mCherry intensity was much lower than the GFP signal. Because Ndc10 specifically binds to CDEIII, the additional foci appear to represent endogenous, not ectopic kinetochores, and we hypothesized that they likely have impaired kinetochore-microtubule attachments. To test if there is chromosome mis-segregation under these conditions, we used a quantitative chromosome transmission fidelity assay [29]. Consistent with this hypothesis, pGAL-CSE4 mif2-ΔN shows a 3.8-fold increase in mini-chromosome loss rate relative to wild type (Supplemental Fig. S7). Defects in kinetochore-microtubule attachments trigger the spindle assembly checkpoint [52–54], so we asked whether the mif2-ΔN cells overexpressing Cse4 activate the checkpoint. We released cells from G1 in the presence of galactose to induce Cse4 overexpression and analyzed the percentage of cells that entered anaphase 120 and 150 min after release. By 150 min, >60% of MIF2-GFP, mif2-ΔN-GFP, and MIF2-GFP pGAL-CSE4 cells had entered anaphase. In contrast, <20% of the pGAL-CSE4 mif2-ΔN cells had progressed into anaphase. To determine whether this slow progression was due to spindle assembly checkpoint activation, we deleted the MAD3 checkpoint gene. Deletion of MAD3 rescued anaphase entry, indicating that the cells overexpressing Cse4 with Mif2-ΔN triggered the checkpoint, strongly suggesting they have defective kinetochore-microtubule attachments (Fig. 7D). Given that we did not find evidence for ectopic kinetochores, it is likely that Mif2-ΔN overexpression affects endogenous kinetochore assembly. Regardless of the mechanism, these data suggest that the N-terminus of Mif2 becomes important for kinetochore function when there is excess Cse4.