Reconstitution reveals two paths of force transmission through the kinetochore

Partitioning duplicated chromosomes equally between daughter cells is a microtubule-mediated process essential to eukaryotic life. A multi-protein machine, the kinetochore, drives chromosome segregation by coupling the chromosomes to dynamic microtubule tips, even as the tips grow and shrink through the gain and loss of subunits. The kinetochore must harness, transmit, and sense mitotic forces, as a lack of tension signals incorrect chromosome-microtubule attachment and precipitates error correction mechanisms. But though the field has arrived at a ‘parts list’ of dozens of kinetochore proteins organized into subcomplexes, the path of force transmission through these components has remained unclear. Here we report reconstitution of functional Saccharomyces cerevisiae kinetochore assemblies from recombinantly expressed proteins. The reconstituted kinetochores are capable of self-assembling in vitro, coupling centromeric nucleosomes to dynamic microtubules, and withstanding mitotically relevant forces. They reveal two distinct pathways of force transmission and Ndc80c recruitment.


Introduction
Chromosome segregation is a complex imperative faced by all eukaryotes, as failure to accurately distribute genetic material to daughter cells in mitosis and meiosis causes potentially lethal aneuploidy (Cimini et al., 2001;Cimini, 2008;Santaguida and Amon, 2015). Eukaryotic cells rely on a network of protein complexes ( Figure 1A and Table 1) to tether centromeres to the dynamic spindle microtubules that pull them to opposite poles of the dividing cell (Musacchio and Desai, 2017).
This network, the kinetochore, mediates the vital process of chromosome segregation by coupling microtubule dynamics to chromosome movement. The inner kinetochore binds the centromere; the outer kinetochore binds microtubules. A massive molecular machine, the kinetochore must harness the force of depolymerizing microtubules, transmit this force to the centromere, and sense aberrant attachments that fail to generate force. The kinetochore also serves as a regulatory hub for detection of unattached kinetochores and destabilization of improper kinetochore-microtubule attachments (Musacchio, 2011). Tension is the signal of correct attachment (i.e., biorientation) (Nicklas and Ward, 1994), and it directly stabilizes kinetochore-microtubule connections (Akiyoshi et al., 2010). But which of the myriad kinetochore proteins actually take part in force transmission and how much tension the kinetochore is under are unresolved questions.
Estimates of mitotic forces experienced by kinetochores in vivo span orders of magnitude (Nicklas, 1988;Fisher et al., 2009;Ye et al., 2016). In the case of budding yeast, each kinetochore contacts a single microtubule and is estimated to experience forces between 0.2 and 16 pN during metaphase (Fisher et al., 2009;Chacó n et al., 2014;Suzuki et al., 2016). Purified kinetochore particles isolated from budding yeast couple to microtubule plus ends with an average rupture strength in this range: 9.1 pN (Akiyoshi et al., 2010;Miller et al., 2016). Previous studies using purified proteins have concluded that (1) Ndc80c and Dam1c are both capable of coupling to microtubules independently, although Dam1c improves microtubule coupling by Ndc80c; (2) this process is regulated by Aurora B kinase (Ipl1 in budding yeast) (Cheeseman et al., 2002;Cheeseman et al., 2006;Caldas et al., 2013;Sarangapani et al., 2013;Umbreit et al., 2014;Kim et al., 2017); (3) Ndc80c transmits force to MIND (Kudalkar et al., 2015;Dimitrova et al., 2016), and (4) the kinetochore as a whole behaves as a tension-sensitive catch bond (Akiyoshi et al., 2010). Load-bearing tip-couplers can form in vitro by self-assembly of recombinant Ndc80c (Powers et al., 2009), MIND (Kudalkar et al., 2015), and Dam1c (Asbury et al., 2006), demonstrating that these components are sufficient to recapitulate a fundamental function of the outer kinetochore. Relatively little is known about how force is transmitted to the centromeric DNA. It is not even certain which inner kinetochore proteins are involved.
Here, using recombinantly purified S. cerevisiae proteins, we reconstitute kinetochore assemblies capable of tethering centromeric nucleosomes to dynamic microtubule plus-ends. We show that the assemblies are load-bearing, in the sense that they sustain tensile forces in the mitotically relevant piconewton range. In doing so, we demonstrate that there are at least two possible paths of force transmission through the inner kinetochore: the Mif2 and OA protein complexes. Each of these components can independently form load-bearing interactions with both MIND and the centromeric nucleosome.

Results
Assay to test quantitatively for spontaneous self-assembly of purified kinetochore subcomplexes The proteins of the S. cerevisiae kinetochore are organized into subcomplexes ( Figure 1A). We have recombinantly expressed and purified seven of these subcomplexes in E. coli and isolated bovine brain tubulin to allow reconstitution and mechanical testing of functional kinetochore assemblies in vitro ( Figure 1B). Previously, we showed that the microtubule-binding Ndc80c and Dam1c complexes, when introduced free in solution, will associate with bead-bound MIND to form a load-bearing attachment to a dynamic microtubule tip (Kudalkar et al., 2015). We have now extended this approach to test longer protein chains, including both inner and outer kinetochore components. In order to test a protein chain containing multiple components, one subcomplex carrying a His 6 -tag was bound directly to polystyrene microbeads via anti-His antibodies ( Figure 1C). The remaining subcomplexes, which did not have His 6 -tags, were added free in solution and bound to the beads only indirectly, by assembling together with the directly tethered, His 6 -tagged subcomplex. Using a laser trap to manipulate individual beads, the kinetochore assemblies were tested for their ability to couple to an individual, assembling microtubule tip. If they attached, then their coupling strength was subsequently measured by increasing the force until the bead detached from the tip (Asbury et al., 2006;Powers et al., 2009;Akiyoshi et al., 2010;Franck et al., 2010; Figure 1D). Protein chains that withstood the 1-pN preload force were considered load-bearing, and distributions of rupture force were used to compare relative mechanical strength. Because bond strength is determined by both kinetic and thermodynamic properties and varies depending on the loading rate (Evans and Ritchie, 1997;Merkel et al., 1999;Bustamante et al., 2004), loading rates and protein concentrations were held constant in all experimental conditions to allow meaningful comparison between the mechanical strengths of different protein linkages. In negative controls, microtubule coupling was abrogated by omitting one of the subcomplexes. Because these experiments were not conducted under single-molecule conditions, many kinetochore protein assemblies on each bead were presumably contacting each individual microtubule tip ( Figure 1-figure supplement 1); this arrangement mimics the physiological situation in budding yeast, where multiple microtubule-binding elements of the outer kinetochore form a multivalent attachment to a single microtubule.

OA forms microtubule attachments through MIND and Ndc80c
Both of the two essential inner kinetochore protein complexes, Mif2 and OA, bind directly to MIND (Hornung et al., 2014;Dimitrova et al., 2016). Thus we began by asking whether OA could form a microtubule attachment through MIND and Ndc80c. With OA bound to the polystyrene beads, and MIND and Ndc80c added in solution (at 10 nM), the vast majority of beads coupled to assembling microtubule tips, indicating that functional, OA-based co-complexes spontaneously assembled in vitro ( Figure 2A). No microtubule attachments were observed if MIND was omitted from the assay, demonstrating that OA must interact specifically through MIND -and not with Ndc80c or the microtubule directly -in order for the microtubule-coupling assemblies to form. Likewise, very few beads attached (6%) when they were decorated with a truncated mutant OA complex (DN-OA), lacking the N-terminal 20 residues of Ame1 that bind the MIND complex (Hornung et al., 2014;Dimitrova et al., 2016;Figure 2A). Together with the prior observation that MIND itself has no microtubule-binding activity (Kudalkar et al., 2015), these results indicate that a three-component chain, OA/MIND/Ndc80c, spontaneously assembles to couple OA indirectly to the microtubule tip. This arrangement is consistent with localization dependencies seen in vivo (De Wulf et al., 2003), co-immunoprecipitation (co-IP) dependencies in budding yeast (De Wulf et al., 2003;Lang et al., 2018), and with previous in vitro reconstitutions of metazoan kinetochore proteins (Cheeseman et al., 2006;Weir et al., 2016). (Although it should be noted that human homologs of OA are reported to directly bind microtubules and not the Mis12c/MIND Pesenti et al., 2018, suggesting interspecific differences in how this conserved subcomplex is utilized.) Phosphorylation of two sites on Dsn1 has been shown to partially alleviate an intra-complex autoinhibition, thereby allowing MIND to interact more stably with both OA and Mif2 Petrovic et al., 2016). However, we found that both the wild-type MIND complex and a phosphomimetic mutant, 2D-MIND, carrying aspartic acid substitutions at the two target phosphorylation sites on Dsn1 (S240D, S250D), supported the attachment of a similarly high fraction of OAcoated beads to assembling microtubule tips ( Figure 2A). This observation suggests that the intracomplex autoinhibition within MIND does not completely abrogate its interaction with OA.

OA forms a load-bearing attachment to MIND
Because tip-coupled kinetochores must support piconewton-scale loads in vivo, we asked whether the reconstituted OA/MIND/Ndc80c chains could support such loads in vitro. In order to quantify their load-bearing ability, we used a force-ramp assay . After attachment of a bead to a single growing microtubule tip and application of a weak preload force (~1 pN), the force was gradually increased (at 0.25 pN s À1 ) until the attachment ruptured. Rupture is a stochastic event, and therefore rupture forces measured for populations of identical protein linkages always exhibit significant variability (Evans and Ritchie, 1997;Merkel et al., 1999;Akiyoshi et al., 2010;Kudalkar et al., 2015;Helgeson et al., 2018). For this reason, dozens of ruptures were collected for every condition tested. The median rupture strength of tip-attachments formed by OA/MIND/ Ndc80c chains was 3.0 pN ( Figure 2B,C), indicating that both the OA/MIND and MIND/Ndc80c interfaces can bear load. When the phosphomimetic 2D-MIND was used instead of wild-type MIND, the median rupture strength of the OA/2D-MIND/Ndc80c chains increased slightly, to 3.9 pN. This observation suggests that the autoinhibition within the MIND complex slightly weakens the loadbearing capacity of the OA/MIND interface.
While rupture strengths by themselves do not provide information about which protein-protein interface breaks during each rupture, some inferences can be made from comparative analysis. For example, previously published work demonstrates that MIND/Ndc80c-mediated microtubule attachments do not differ in strength from Ndc80c-mediated attachments (Kudalkar et al., 2015), implying that the MIND/Ndc80c interface is mechanically stronger than the Ndc80c/microtubule interface. This conclusion was additionally supported by the prior observation that adding Dam1c, which specifically strengthens the Ndc80c/microtubule interface, significantly increases the strength of MIND/ Ndc80c-mediated attachments (Kudalkar et al., 2015). In order to determine if the OA/MIND interface is likewise mechanically strong relative to the Ndc80c/microtubule interface, we specifically strengthened the latter by adding Dam1c in solution to our rupture force assay. OA/MIND/Ndc80cmediated attachments nearly doubled in strength, to a median rupture force of 7.6 pN, with the addition of Dam1c ( Figure 2B,C). This observation suggests that the OA/MIND interface, like the MIND/Ndc80c interface, is mechanically strong relative to the Ndc80c/microtubule interface.

Mif2 forms microtubule attachments through MIND and Ndc80c
We then asked if Mif2 could form microtubule attachments through MIND and Ndc80c. Mif2 was bound to polystyrene beads, while wild-type MIND and Ndc80c were added in solution. A substantial fraction of the beads, 45%, coupled to microtubule tips, demonstrating that functional, threecomponent Mif2/MIND/Ndc80c chains spontaneously assembled in vitro ( Figure 3A). Hypothesizing that intra-complex autoinhibition within MIND might be responsible for the 55% of beads that failed to bind microtubules, we repeated the same experiment, but using the phosphomimetic mutant, 2D-MIND, instead of wild-type MIND. With this substitution, the fraction of Mif2-coated beads capable of coupling to microtubules increased to 81% ( Figure 3A), suggesting that intra-complex autoinhibition within MIND tightly regulates its interaction with Mif2. To confirm that MIND was essential for connecting Mif2 to Ndc80c, we omitted MIND from the assay. Beads coated in Mif2 did not couple to microtubules in the absence of MIND ( Figure 3A), indicating that Mif2, like OA (Figure 2A), must bind directly to MIND, and not to Ndc80c or to microtubules. Unlike OA, however, Mif2 requires alleviation of autoinhibition within the MIND complex in order to bind tightly. For this reason, 2D-MIND was used in all subsequent experiments.
The conserved N-terminus of Mif2 is sufficient for binding to the Mtw1/Nnf1 head of the MIND complex (Hornung et al., 2014;Dimitrova et al., 2016). In order to probe the specific Mif2 domains involved in assembling functional microtubule-coupling chains, we purified a truncated version lacking the first 40 N-terminal residues, DN-Mif2. Only 16% of beads coated with DN-Mif2 interacted with microtubules ( Figure 3A), suggesting that the N-terminal domain of Mif2, like the N-terminus of Ame1, is an important point of attachment between the inner kinetochore and MIND. But because a significant fraction of beads was still able to bind microtubules in the absence of the first 40 residues of Mif2, it seems likely that other amino acids also contribute to the Mif2/MIND interface. A region of looser conservation extends beyond the well-conserved first 40 residues at the N-terminus of Mif2 and its homologs (Screpanti et al., 2011). Source data 1. Raw rupture force data for Mif2-based linkers.

Mif2 forms a load-bearing attachment to MIND
To quantify the rupture strength of Mif2-based assemblies, we applied the force-ramp assay. Attachments mediated by the Mif2/2D-MIND/Ndc80c chains supported piconewton-scale loads, with a median rupture strength of 4.4 pN. Notably, their strength was not significantly different from that of our OA-based assemblies ( Figure 3B,C). As explained above, the addition of Dam1c significantly strengthened the OA-based linkers, indicating that the OA/MIND interface is mechanically strong relative to the Ndc80c/microtubule interface. To ask if the same is true of the Mif2/MIND interface, we tested whether Mif2/2D-MIND/Ndc80c chains could be strengthened by the addition of Dam1c. In contrast to our results with OA, adding Dam1c had no significant effect (p=0.67) on the strength of the Mif2-based chains ( Figure 3B,C). These observations suggest that the OA/MIND interface is  mechanically strong relative to the Ndc80c/microtubule attachment, but that the Mif2/MIND interface is not.

OA strengthens the Mif2/2D-MIND interface
Having demonstrated that both Mif2 and OA can independently transmit force to MIND, we then asked if these two subcomplexes are stronger in combination. We added free OA to the Mif2/2D-MIND/Ndc80c chain, thereby increasing its median rupture force from 4.4 pN to 5.4 pN ( Figure 4A,  B). This increase suggests that OA is able to strengthen the Mif2/MIND interface, which we believe is relatively weak (based on our experiments with Dam1c). We then tested whether the Mif2-based chains with the addition of free OA could be further strengthened by adding free Dam1c. Adding Dam1c in solution did not change the 5.4 pN rupture force ( Figure 4A,B), strongly suggesting that the Mif2/2D-MIND interface, although strengthened by the addition of OA, remains the primary site of rupture in Mif2/OA/2D-MIND/Ndc80c linkers.
In order to better understand how OA strengthens Mif2/2D-MIND/Ndc80c linkers, we replaced the free OA added in solutions with DN-OA, which is incapable of binding MIND (Figure 2A). In contrast to the increase in rupture force that we observed when free full-length OA was added to Mif2based linkers, free DN-OA did not increase the strength of these linkers ( Figure 4A,B). This observation shows that the MIND-binding activity of OA is essential for strengthening the Mif2/2D-MIND interface. There are two ways in which this could occur. Because '2D' phosphomimetic mutations on Dsn1 do not entirely alleviate autoinhibition within the MIND complex , it is possible that OA strengthens the Mif2/2D-MIND interface by further alleviating that autoinhibition. Alternatively, if the free OA were able to bind both 2D-MIND and Mif2, then it could theoretically recruit additional MIND/Ndc80c to the linker, thereby increasing the valency and strength of the Ndc80c-microtubule interaction. Either scenario is consistent with DN-OA being unable to strengthen Mif2/2D-MIND/Ndc80c linkers. We favor the former theory, because we have no evidence of a direct interaction between OA and Mif2 in the absence of post-translational modifications, and because experiments with Dam1c ( Figure 4A,B) suggest that the Mif2/2D-MIND interface is mechanically weak relative to the Ndc80c/microtubule interface.
In OA-based linkers, on the other hand, the OA/2D-MIND interface was not the weakest link, leading to the prediction that strengthening that interface should not increase the overall strength of OA-based chains. Indeed, the addition of Mif2 in solution did not strengthen OA/2D-MIND/ Ndc80c-mediated attachments (Figure 4-figure supplement 1), consistent with our earlier inference that the Ndc80c/microtubule interface is usually the site of rupture in our OA-based assemblies.
Both OA and Mif2 assemble with centromeric nucleosomes Previous work shows that Mif2 and OA can bind centromeric nucleosomes (Kato et al., 2013;Xiao et al., 2017;Anedchenko et al., 2019;Fischbö ck-Halwachs et al., 2019), which carry Cse4, a specialized variant of histone H3 (human CENP-A) (Palmer et al., 1991;Stoler et al., 1995). To ask if either Mif2 or OA can assemble in our assay onto these specialized nucleosomes, histone complexes containing H2A, H2B, H4, and Cse4, were wrapped in 601 DNA and the resulting nucleosome core particles (Cse4-NCPs) were bound to polystyrene microbeads. Mif2 or OA was then added in solution, along with 2D-MIND and Ndc80c. With OA in solution, 78% of beads bound microtubules; with Mif2 in solution, 24% bound ( Figure 5A). Possible reasons for the relatively low percentage of active Cse4-NCP-decorated beads in the experiments with added Mif2/2D-MIND/Ndc80c are discussed below. Nevertheless, these observations confirm that OA and Mif2 can individually link the outer kinetochore subcomplexes, 2D-MIND/Ndc80c, to centromeric nucleosomes, spontaneously forming the four-component microtubule-binding chains, Cse4-NCPs/OA/2D-MIND/Ndc80c and Cse4-NCPs/Mif2/2D-MIND/Ndc80c, respectively. As expected (Lang et al., 2018), the Cse4-NCPdecorated beads completely failed to bind microtubules if both OA and Mif2 were omitted ( Figure 5A), confirming that the NCPs themselves do not interact directly with MIND, Ndc80c, or microtubules. To test whether Mif2 and OA bind specifically to centromeric nucleosomes, we also wrapped canonical nucleosomes containing histone H3 (instead of Cse4) with 601 DNA and attached these H3-NCPs to beads. The H3-NCP-decorated beads did not bind microtubules with either Mif2 or OA in solution ( Figure 5A), indicating that both Mif2 and OA bind selectively to centromeric, Cse4-containing nucleosomes. Altogether these results show that Mif2 and OA can serve as molecular bases for two distinct paths connecting the outer kinetochore to the centromere.

Both OA and Mif2 form load-bearing attachments to centromeric nucleosomes
Having established that both OA and Mif2 can individually and selectively link bead-bound Cse4-NCPs to outer kinetochore subcomplexes, we used our force-ramp assay to test whether these interactions are load-bearing. With Cse4-NCPs on the beads, attachments mediated by Cse4-NCP/Mif2/ 2D-MIND/Ndc80c chains ruptured at 3.3 pN ( Figure 5B,C). Likewise, attachments mediated by Cse4-NCP/OA/2D-MIND/Ndc80c chains also ruptured at 3.3 pN. The two survival curves are not Source data 1. Raw rupture force data for nucleosome core particle-based linkers.  significantly different from one another (p=0.55) ( Figure 5B,C). However, both are significantly weaker than the chains assembled with Mif2 or OA bound directly to the beads, without NCPs (p=0.00015 for the comparison with Mif2/2D-MIND/Ndc80c; p=0.00405 for OA/2D-MIND/Ndc80c).
Given our previous finding that OA strengthens Mif2-based chains ( Figure 4A,B), we were curious whether the simultaneous presence of both inner kinetochore complexes, OA and Mif2, would strengthen Cse4-NCP-based chains. It did. With a median rupture force of 4.0 pN, Cse4-NCP-based chains with both Mif2 and OA added in solution were significantly stronger than chains containing only Mif2 (p=0.015) or containing only OA (p=0.037) ( Figure 5B,C). This strengthening suggests that the Mif2/Cse4-NCP and OA/Cse4-NCP interfaces created in our assay are individually relatively weak but can mutually reinforce one another.
OA is reported to bind both DNA and the N-terminus of Cse4 (Hornung et al., 2014;Anedchenko et al., 2019;Fischbö ck-Halwachs et al., 2019). In order to assess the relative contributions of these two interactions to the OA/Cse4-NCP interface, we immobilized His 6 MBP-tagged Cse4 (residues 1-50) on the surface of the beads and added free OA, 2D-MIND, and Ndc80c in solution ( Figure 5-figure supplement 1). These Cse4 (1-50) -based chains exhibited a median rupture strength of 3.1 pN and were not significantly weaker than chains based on complete centromeric nucleosomes (p=0.91). This observation indicates that interactions of OA with the 601 DNA, the rest of Cse4, or histones other than Cse4 made no significant contribution to the strength of the OA/ Cse4-NCP interface in our experiments. Because Mif2 homologs bind the C-terminus of Cse4 homologs (Carroll et al., 2010;Guse et al., 2011), we expected that adding Mif2 in solution would not strengthen the Cse4 (1-50) -based assemblies. Indeed, the addition of free Mif2 did not strengthen the Cse4 (1-50) /OA/2D-MIND/Ndc80c chains (p=0.33)( Figure 5-figure supplement 1).
To further assess the relative strength of the OA/Cse4-NCP interface, we added free Dam1c to the Cse4-NCP/OA/2D-MIND/Ndc80c chains ( Figure 5-figure supplement 2). The addition of Dam1c, which specifically strengthens the Ndc80/microtubule interface, had no significant effect on the rupture strength of the Cse4-NCP/OA/2D-MIND/Ndc80c chains (p>0.5), indicating that the Cse4-NCP/OA interface is probably the weakest link in these chains. We attempted to strengthen that interface by adding Chl4/Iml3 (CI) in solution, but it also had no effect on overall chain strength (p>0.5) ( Figure 5-figure supplement 2). Altogether, these data indicate that both essential components of the inner kinetochore, OA and Mif2, are capable of independently transmitting piconewton-scale forces from the outer kinetochore to centromeric nucleosomes.

Discussion
We have reconstituted kinetochore assemblies capable of transmitting piconewton-scale forces from dynamic microtubules to Cse4-containing nucleosomes ( Figure 5). To our knowledge, these are the first kinetochore assemblies made from individually purified, recombinant S. cerevisiae components that reconstitute what is arguably the kinetochore's most fundamental activity: coupling centromeric nucleosomes to dynamic microtubule tips under physiologically relevant loads.
The self-assembly of full-length kinetochore subcomplexes at nanomolar concentrations suggests that they might have greater affinities for one another than suggested by the micromolar dissociation constants obtained using short peptides (Hornung et al., 2014;Dimitrova et al., 2016;Petrovic et al., 2016), possibly because of additional points of contact, although the immobilization of one subcomplex on beads might also affect binding dynamics. Both OA and Mif2 are individually capable of forming load-bearing interactions with MIND, demonstrating that there are two distinct paths of force transmission through the inner kinetochore. However, these paths are not equivalent. They are differentially sensitive to regulation by the major mitotic kinase, Ipl1 (Aurora B in humans), with the Mif2 path depending on the presence of phosphomimetic mutations on the MIND complex, whereas the OA pathway is only slightly strengthened by them. Also, while both the Mif2/2D-MIND and OA/2D-MIND interfaces are load-bearing, the former is mechanically weak relative to the Ndc80c/microtubule interface, but the latter is strong (Figure 2 and Figure 3).
We also found that OA can strengthen Mif2/2D-MIND/Ndc80c linkers, whereas Mif2 slightly weakens OA/2D-MIND/Ndc80c linkers. Considering that '2D' phosphomimetic mutations on MIND partially alleviate autoinhibition but do not abrogate it completely , we speculate that OA binding to 2D-MIND might alleviate this residual autoinhibition, promoting MIND's interaction with Mif2 and strengthening the Mif2-based chain. In contrast, for the OA-based linker, we hypothesize that OA disinhibition of 2D-MIND might permit Mif2 to outcompete OA for binding to 2D-MIND ( Figure 4A,B). In other words, Mif2 might decrease the rupture force of the OA/2D-MIND/Ndc80c linkers by 'stealing' the disinhibited 2D-MIND from bead-bound OA, thus detaching MIND from the bead. Competition for binding to MIND is plausible, because previous work has shown that OA and Mif2 bind to adjacent sites on the same head of MIND .
When NCPs are on the beads, adding both OA and Mif2 in solution strengthens their attachment to microtubules ( Figure 5B,C). OA and Mif2 probably do not compete with each other for binding to a centromeric nucleosome, because OA binds the N-terminus of Cse4, whereas Mif2 likely binds the C-terminus of Cse4 (Carroll et al., 2010;Guse et al., 2011;Anedchenko et al., 2019;Fischbö ck-Halwachs et al., 2019). If OA and Mif2 each recruit one MIND complex to the nucleosome, then together they would increase the total number of Ndc80c complexes in the assembly. Both OA and Mif2 self-assemble exclusively on centromeric nucleosomes ( Figure 5A), a kinetochore function essential for genome stability. Yet we consistently find that Cse4-NCP-based chains are weak compared to shorter chains without NCPs. We hypothesize that this relative weakness occurs because the Cse4-NCPs were wrapped in 601 DNA rather than centromeric DNA. Budding yeast Mif2 is reported to have a 40-fold greater affinity for Cse4-containing nucleosomes if they are wrapped in centromeric DNA (Xiao et al., 2017). We were unable to test the effect of this increased affinity on rupture force because wrapping nucleosomes in centromeric DNA proved intractable. Nonetheless, the rupture strengths of our reconstituted assemblies fall within the range of mitotic forces estimated for budding yeast kinetochores in vivo.
In addition to the absence of centromeric DNA, there are other possible reasons why the strongest chains studied here still had a median force below the 9.1 pN strength measured previously for native purified kinetochore particles (Akiyoshi et al., 2010). First, studies of native kinetochore particles have not tested force transmission through the inner kinetochore; force has been exerted only through outer kinetochore components via MIND-based bead linkages (Akiyoshi et al., 2010). Thus, the inner kinetochore connections might also be weaker in the native particles. Second, post-translational modifications retained on the native proteins but absent from our recombinant proteins might strengthen interactions within the inner kinetochore. Given that methylation of Cse4 is known to weaken the OA-Cse4 interface (Anedchenko et al., 2019), it seems plausible that additional modifications that strengthen the inner kinetochore might also be found.
Although both OA and Mif2 can transmit force from Cse4-NCPs to MIND, they are neither interchangeable nor redundant. There are several reasons that having multiple paths of force transmission might be advantageous. First, each path can be differentially regulated, which our data suggest is the case. Second, OA and Mif2 together increase the total number of outer kinetochore components that can be recruited. Estimates of the number of Ndc80 complexes present in each kinetochore vary from six to thirty copies (Joglekar et al., 2006;Lawrimore et al., 2011;Dhatchinamoorthy et al., 2017), and biophysical experiments demonstrate that multiple Ndc80 complexes are necessary for effective microtubule coupling (Powers et al., 2009;Volkov et al., 2018). But MIND and Ndc80c interact in a 1:1 ratio , implying that the inner kinetochore must serve as an oligomerization platform, assembling on a single centromeric nucleosome and recruiting multiple copies of MIND and Ndc80c. If one OA complex and one Mif2 dimer assembled onto each of the two faces of a centromeric nucleosome, they could collectively recruit up to 6 MIND/Ndc80c, with additional copies of Ndc80c potentially being recruited by the Cnn1 pathway (Lang et al., 2018).
Our findings also raise intriguing questions about how these multiple possible paths of force transmission and Ndc80c recruitment may be differentially utilized at different stages of the cell cycle or by different organisms. For instance, the human homolog of OA binds microtubules (Pesenti et al., 2018), but budding yeast OA clearly does not (Figure 2A). For some organisms, OA is essential (Pot et al., 2005;Kagawa et al., 2014); other species appear to lack OA homologs van Hooff et al., 2017). A smaller subset of organisms has no known homolog of Mif2 (Drinnenberg et al., 2014;van Hooff et al., 2017). Evidently, the relative importance of the OA and Mif2 pathways varies between organisms. In budding yeast, the extreme N-terminus of Ame1 is essential, while the N-terminus of Mif2 is not (Hornung et al., 2014), suggesting the primacy of the OA-based path in this organism. There is also strong evidence for a third pathway of Ndc80c recruitment, through Cnn1 (Bock et al., 2012;Huis In 't Veld et al., 2016;Lang et al., 2018). Although Cnn1 is not essential in yeast (Giaever et al., 2002;De Wulf et al., 2003), an additive growth defect is observed when deletion of Cnn1 is combined with deletion of the N-terminus of Mif2 (Hornung et al., 2014). Homologs of Cnn1 also play important roles in several other organisms (Gascoigne et al., 2011;Nishino et al., 2013;Huis In 't Veld et al., 2016;Cortes-Silva et al., 2020). Thus, while the branched architecture of the inner kinetochore is a feature conserved across many species, the relative importance of particular branches apparently varies between species.
Taken together, our work represents a major advance towards the goal of reconstituting a complete, functional kinetochore from purely recombinant proteins. Going forward, the reconstituted 'skeleton' kinetochore presented here can be used as a scaffold to study the contributions of other kinetochore proteins to force transmission. Of particular interest are Cnn1, which could increase rupture force by recruiting additional copies of Ndc80c (Schleiffer et al., 2012;Lang et al., 2018), and Stu2, which is known to contribute significantly to kinetochore-microtubule coupling (Miller et al., 2016).

Plasmids and constructs
All expression vectors are listed in Figure 1-source data 1.
Briefly, His 6 -tagged protein subcomplexes were expressed from polycistronic pST39 vectors in BL21 cells and induced with 0.4 mM isopropyl b-D-1-thiogalactopyranoside (IPTG) for 16 hr at 18˚C. Cells were lysed with a French press and the subcomplex was immobilized on a Ni-charged IMAC resin column (Bio-Rad) in either MIND buffer (50 mM sodium phosphate buffer, pH 7.0, 200 mM NaCl), Ndc80c buffer (50 mM HEPES buffer, pH 7.6, 300 mM NaCl), or Dam1c buffer (50 mM sodium phosphate buffer, pH 6.9, 500 mM NaCl) supplemented with protease inhibitors (Roche), 5 mM imidazole, and 1 mM PMSF. The resin was then washed, and subcomplex eluted with buffer containing 300 mM imidazole before being further purified using a Superdex 200 size exclusion column (GE Healthcare) in either MIND buffer, Ndc80c buffer with 200 mM NaCl, or Dam1c buffer.

Okp1/Ame (OA)
Wild-type (WT) and truncated His 6 -tagged OA was expressed in a polycistronic pST39 vector in BL21 cells and induced with 0.4 mM IPTG for 16 hr at 18˚C. Cells were lysed with a French press and OA was immobilized on a Ni-charged IMAC resin column (Bio-Rad) in 50 mM sodium phosphate buffer, pH 7.0 supplemented with 200 mM NaCl, protease inhibitors (Roche), 5 mM imidazole, and 1 mM PMSF, washed, and eluted in the same buffer with 300 mM imidazole. OA was further purified using a Superdex 200 size exclusion column (GE Healthcare) in SEC-OA buffer (50 mM sodium phosphate buffer, pH 7.0 supplemented with 200 mM NaCl.) WT and truncated FLAG-tagged OA was expressed similarly to His 6 -tagged OA except that it was immobilized on an anti-FLAG M2 affinity gel (A2220; Sigma). The affinity gel was washed, and protein was eluted with 0.1 mgÁmL À1 3X FLAG Peptide (F4799; Sigma). OA was further purified using a Superdex 200 size exclusion column (GE Healthcare) in SEC-OA buffer.
After SEC, OA was concentrated with Amicon Ultra centrifugal filters (Millipore), stored in SEC-OA buffer with 5% glycerol, and immediately stored at À80˚C.

Mif2
Mif2 expression plasmids contain a C-terminal (Ct) linker and maltose-binding protein (MBP) tag in order to improve solubility and expression of Mif2. The location of the tag was rationally selected, as the N-terminus of Mif2 homologs is reported to mediate binding to MIND Petrovic et al., 2016), while a central domain mediates binding to the centromeric nucleosome (Carroll et al., 2010;Kato et al., 2013).
The flowthrough was discarded, beads washed with HSB, and histones eluted with 400 mM imidazole and 50 mM EDTA. The eluate was concentrated, and histone octamers were further purified on a Superdex 200 size exclusion column (GE Healthcare) in SEC-H1 buffer containing 2M NaCl, 50 mM HEPES buffer, pH 7.5, 1 mM TCEP.
Histone octamers were wrapped in the 147-basepair double-stranded Widom DNA (henceforth '601 DNA') (Lowary and Widom, 1998) as described (Migl et al., 2020). 601 DNA was produced by PCR and final purification performed using HiTrapQ column. 601 DNA and histone octomers were combined in a 1.1:1 molar ratio in a hydrated 7K MWCO Slide-A-Lyzer MINI Dialysis Device (69560; ThermoFisher), which was placed in 0.5 L of high-salt dialysis buffer (2 M NaCl, 30 mM HEPES buffer, pH 7.5, 1 mM TCEP, and 5 mM EDTA). The nucleosome wrapping was achieved through dialysis against low salt dialysis buffer (200 mM NaCl, 30 mM HEPES buffer, pH 7.5, 1 mM TCEP, and 5 mM EDTA) over the course of 60 hr at 4˚C. Purification of excess DNA was done by using a Superdex 200 or Superose six size exclusion columns (GE Healthcare) with SEC-H2 buffer 150 mM NaCl, 30 mM HEPES buffer, pH 7.5, 1 mM TCEP. NCPs were stored on ice, and all experiments using NCPs were performed within one week of wrapping.

Optical trap rupture force assay
Streptavidin-coated 0.56 mm polystyrene beads (Spherotech) were coated with biotinylated anti-His5 antibodies (Qiagen), and 3.5 pM beads were incubated with 10 nM His 6 -tagged protein complex as described Umbreit et al., 2014), such that each bead was decorated with 3000 protein complexes. Flow cells were prepared using double-sided tape and plasma-cleaned coverslips and incubated with 25 mL of 1 mgÁmL À1 biotinylated BSA (Vector Laboratories) for 15 min at room temperature, then washed with BRB80 (80 mM PIPES, pH 6.9, 1 mM MgCl 2 , 1 mM EGTA). Flow cells were then incubated with 30 mL of 1 mgÁmL À1 avidin DN (Vector Laboratories) for 5 min at room temperature, followed by another BRB80 wash. GMPCPP biotinylated tubulin seeds in BRB80 were bound for 5 min and washed twice with 42˚C growth buffer (BRB80 plus 1 mM GTP and 105 mM k-casein). Protein-decorated beads were introduced into the flow chamber in growth buffer with 1.4 mgÁmL À1 tubulin, 313 mgÁmL À1 glucose oxidase, 37.5 mgÁmL À1 catalase, 30 mM glucose, 1 mM DTT, and 10 nM each of the free, non-His-tagged protein complexes.
For the negative controls (e.g., OA-decorated beads with Ndc80c but not MIND in solution), all beads and soluble proteins were maintained at the concentrations used in the experimental reactions, except for the subcomplex being omitted.
For each condition, a bead was manipulated to contact a single microtubule (Figure 1-figure  supplement 1). If the bead-microtubule contact did not persist after the laser was switched off, that bead was considered to be a nonbinder. Optical trap assays were performed at room temperature using custom instrumentation to capture and manipulate beads as described . Rupture force assays were performed as described (Akiyoshi et al., 2010;Tien et al., 2010;Sarangapani et al., 2013). Once beads were bound to microtubule tips, a test force of 1 pN was applied, and only beads that tracked with~100 nm of tip growth were subjected to ramping force of 0.25 pNÁs À1 until detachment. All attachments that withstood the 1 pN preload force were included in our analysis. Records of force vs. time were collected, and maximum rupture force was determined using custom Igor Pro software (Wavemetrics) (Source data 1). Protein chains that withstood the 1-pN preload force were considered load-bearing, and distributions of rupture force were used to compare relative mechanical strength.

Data analysis and figure preparation
Data from optical trap assays were analyzed in R (Wickham, 2009;R Development Core Team, 2013;Kassambra and Kosiniski, 2018) and Igor Pro (Wavemetrics). Figures were produced using R and Adobe Illustrator. Charles L Asbury

Genentech
Cameron L Noland Yoana N Dimitrova The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication. . Transparent reporting form

Data availability
All data analyses during this study are included in the manuscript and supporting files. Source data files have been provided for Figures 1-5 and their supplements.