Unconventional kinetochore kinases KKT2 and KKT3 have unique centromere localization domains

Chromosome segregation in eukaryotes is driven by the kinetochore, the macromolecular protein complex that assembles onto centromeric DNA and binds spindle microtubules. Cells must tightly control the number and position of kinetochores so that all chromosomes assemble a single kinetochore. A central player in this process is the centromere-specific histone H3 variant CENP-A, which localizes constitutively at centromeres and promotes kinetochore assembly. However, CENP-A is absent from several eukaryotic lineages including kinetoplastids, a group of evolutionarily divergent eukaryotes that have an unconventional set of kinetochore proteins. There are six proteins that localize constitutively at centromeres in the kinetoplastid parasite Trypanosoma brucei, among which two homologous protein kinases (KKT2 and KKT3) have limited similarity to polo-like kinases. In addition to the N-terminal kinase domain and the C-terminal divergent polo boxes, KKT2 and KKT3 have a central domain of unknown function as well as putative DNA-binding motifs. Here we show that KKT2 and KKT3 are important for the localization of several kinetochore proteins and that their central domains are sufficient for centromere localization in T. brucei. Crystal structures of the KKT2 central domain from two divergent kinetoplastids reveal a unique zinc-binding domain (termed the CL domain for centromere localization), which promotes its kinetochore localization in T. brucei. Mutations in the equivalent domain in KKT3 abolish its kinetochore localization and function. Our work shows that the unique central domains play a critical role in mediating the centromere localization of KKT2 and KKT3.


Introduction
The kinetochore is the macromolecular protein complex that drives chromosome segregation during mitosis and meiosis in eukaryotes. Its fundamental functions are to bind DNA and spindle microtubules (Musacchio and Desai, 2017). In most eukaryotes, kinetochores assemble within a single chromosomal region called the centromere. While components of spindle microtubules are highly conserved across eukaryotes (Wickstead and Gull, 2011;Findeisen et al., 2014), centromere DNA is known to evolve rapidly (Henikoff et al., 2001). Nevertheless, it is critical that a single kinetochore is assembled per chromosome and its position is maintained between successive cell divisions. A key player involved in this kinetochore specification process is the centromere-specific histone H3 variant CENP-A, which is found in most sequenced eukaryotic genomes (Talbert et al., 2009). CENP-A localizes specifically at centromeres throughout the cell cycle and recruits HJURP, a specific chaperone that incorporates CENP-A onto centromeres (Black and Cleveland, 2011;McKinley and Cheeseman, 2016;Stankovic and Jansen, 2017). Besides CENP-A, components of the constitutive centromere-associated network (CCAN) also localize at centromeres throughout the cell cycle. CENP-A-containing nucleosomes are recognized by CCAN components, which in turn recruit the KNL1-Mis12-Ndc80 (KMN) network that has microtubule-binding activities. In addition to these structural kinetochore proteins, several protein kinases are known to localize at mitotic kinetochores, including Cdk1, Aurora B, Bub1, Mps1, and Plk1 (Cheeseman and Desai, 2008). These protein kinases regulate various aspects of mitosis, including kinetochore assembly, error correction, and the spindle checkpoint (Carmena et al., 2012;London and Biggins, 2014;Hara and Fukagawa, 2018).
Kinetoplastids are evolutionarily divergent eukaryotes that are defined by the presence of a unique organelle called the kinetoplast that contains a cluster of mitochondrial DNA (d' Avila-Levy et al., 2015). Centromere positions have been mapped in three kinetoplastids: 20-120 kb regions that have AT-rich repetitive sequences in Trypanosoma brucei (Obado et al., 2007), ~16 kb GC-rich unique sequences in Trypanosoma cruzi (Obado et al., 2005), and ~4 kb regions in Leishmania major (Garcia-Silva et al., 2017). Although some DNA elements and motifs are enriched, there is no specific DNA sequence that is common to all centromeres in each organism, suggesting that kinetoplastids likely determine their kinetochore positions in a largely sequence-independent manner. However, none of CENP-A or any other canonical structural kinetochore protein has been identified in kinetoplastids (Lowell and Cross, 2004;Berriman et al., 2005;Aslett et al., 2010). They instead have unique kinetochore proteins, such as KKT1-25 (Akiyoshi and Gull, 2014;Nerusheva and Akiyoshi, 2016;Nerusheva et al., 2019) and KKIP1-12 (D'Archivio and Wickstead, 2017;Brusini et al., 2019) in T. brucei. It remains unknown which of these proteins form the base of trypanosome kinetochores that recruits other proteins. There are six proteins that localize at centromeres throughout the cell cycle (KKT2, KKT3, KKT4, KKT20, KKT22, and KKT23), implying their close association with centromeric DNA. Indeed, we previously showed that KKT4 has DNA-binding activity in addition to microtubule-binding activity (Llauró et al., 2018;Ludzia et al., 2020). However, RNAi-mediated knockdown of KKT4 affected the localization of KKT20 but not other kinetochore proteins, suggesting that KKT4 is largely dispensable for kinetochore assembly.
In this study, we focused on KKT2 and KKT3, which are homologous to each other and have three domains conserved among kinetoplastids: a protein kinase domain that is classified as unique among known eukaryotic kinase subfamilies (Parsons et al., 2005), a central domain of unknown function, and divergent polo boxes (DPB). Presence of an N-terminal kinase domain and a C-terminal DPB suggests that KKT2 and KKT3 likely share common ancestry with polo-like kinases (Nerusheva and Akiyoshi, 2016). Interestingly, a protein kinase domain is not present in any constitutivelylocalized kinetochore protein in other eukaryotes, making these protein kinases a unique feature of kinetoplastid kinetochores. In addition to the three domains that are highly conserved among kinetoplastids, AT-hook and SPKK DNA-binding motifs are found in some species, suggesting that these proteins are located close to DNA (Akiyoshi and Gull, 2014). Although RNAi-mediated knockdown of KKT2 or KKT3 leads to growth defects (Akiyoshi and Gull, 2014;Jones et al., 2014), little is known about their molecular function. In this report, we have revealed a unique zinc-binding domain in the KKT2 and KKT3 central domain, which is important for their kinetochore localization and function in T. brucei. the effect of double knockdown. Although the growth defect of the KKT2/3 double RNAi was not dramatically different from that of individual knockdowns ( Figure 2G), which could be explained by residual KKT2/3 signals ( Figure 2I), defective kinetochore localization was found for KKT1 and KKT4 at 1 day post-induction ( Figure 2H, I). Taken together, these results show that KKT2 and KKT3 play important roles in recruiting multiple kinetochore proteins. It is possible that more efficient or rapid inactivation methods could reveal additional kinetochore proteins whose localizations depend on KKT2/3.

Multiple domains of KKT2 are able to localize at centromeres in T. brucei
To understand how KKT2 localizes at centromeres, we determined which domain was responsible for its centromere localization by expressing a series of truncated versions of KKT2, fused with a GFPtagged nuclear localization signal peptide (GFP-NLS) in T. brucei ( Figure 3A). We previously showed that ectopically-expressed KKT2 DPB (residues 1024-1260) localized at kinetochores (Nerusheva and Akiyoshi, 2016). The present study confirmed this result and also identified two other regions (residues 562-677 and 672-1030) that localized at kinetochores from S phase to anaphase ( Figure 3A, B and Figure S1).
Based on our previous finding that KKT2 co-immunoprecipitated with a number of kinetochore proteins (Akiyoshi and Gull, 2014), we reasoned that these fragments might localize at kinetochores by interacting with other kinetochore proteins. To test this possibility, we immunoprecipitated KKT2 fragments and performed mass spectrometry to identify co-purifying proteins. Although the central domain co-purified only with limited amounts of KKT3 and KKT5 ( Figure 3C and Table S1), KKT2 672-1030 co-purified with several kinetochore proteins, with KKIP1 being the top hit ( Figure 3D and Table S1). Similarly, KKT2 DPB co-purified with several kinetochore proteins, which was abolished in the W1048A mutant that did not localize at kinetochores ( Figure 3E and Table S1) (Nerusheva and Akiyoshi, 2016). These results support the possibility that ectopically-expressed KKT2 672-1030 and DPB are able to localize at kinetochores from S phase to anaphase by interacting with non-constitutive kinetochore proteins (e.g. KKT1, KKT6, KKT7, KKT8, KKIP1). A corollary is that, in wild-type cells, the constitutively-localized KKT2 protein recruits these transient kinetochore proteins onto kinetochores using KKT2 672-1030 and DPB domains.

The central domain of KKT3 is able to localize at centromeres constitutively
We next expressed KKT3 fragments in trypanosomes ( Figure 4A, B). Similarly to KKT2, the Nterminal protein kinase domain of KKT3 did not localize at centromeres. KKT3 DPB had robust kinetochore localization only during anaphase ( Figure 4C), which differs from KKT2 DPB that localized from S phase to anaphase. Immunoprecipitation of KKT3 DPB identified a number of copurifying kinetochore proteins ( Figure 4D), raising a possibility that KKT3, like KKT2, recruits other kinetochore proteins by its DPB.

The Bodo saltans KKT2 central domain adopts a unique structure
To gain insights into how the central domains of KKT2 and KKT3 localize at centromeres, we expressed and purified recombinant proteins for their structure determination by X-ray crystallography. Our attempts to purify the T. brucei KKT3 (TbKKT3 hereafter) central domain were unsuccessful, but we managed to express and purify from E. coli the central domain of KKT2 from several kinetoplastids, including Bodo saltans (a free-living kinetoplastid (Jackson et al., 2016)) and Perkinsela (endosymbiotic kinetoplastids (Tanifuji et al., 2017)) ( Figure S2). We obtained crystals of BsKKT2 572-668 (which corresponds to residues 569-664 in T. brucei KKT2) and determined its structure to 1.8 Å resolution by zinc single-wavelength anomalous dispersion (Zn-SAD) phasing ( Figure 5 and Table 1). Our analysis revealed the presence of two distinct zinc-binding domains: the N-terminal one (referred to as the "CL" domain for its key role in centromere localization: see below) consists of 2 b-sheets (where b-strands 1, 4 and 5 comprise the first b-sheet, and b-strands 2 and 3 comprise the second b-sheet) and one a-helix, while the C-terminal one consists of one b-sheet (comprising b-strands 6 and 7) and one a-helix ( Figure 5A, B). The CL domain coordinates two zinc ions and the C-terminal domain coordinates one zinc ion.
A structural homology search using the DALI server (Holm and Laakso, 2016) indicated that the CL domain has weak structural similarity to proteins that have C1 domains (Table S2). C1 domains were originally discovered as lipid-binding modules in protein kinase Cs (PKCs) and are characterized by the HX12CX2CXnCX2CX4HX2CX7C motif (Colón-González and Kazanietz, 2006;Das and Rahman, 2014). C1 domains are classified into a typical C1 domain, which binds diacylglycerol or phorbol esters, and an atypical C1 domain, which is not known to bind ligands. The closest structural homolog of the BsKKT2 CL domain was the atypical C1 domain of the Vav1 protein (RMSD 2.7 Å across 52 C ). Although the CL domain and the C1 domain share some structural similarity, their superposition revealed fundamental differences ( Figure 6). Coordination of one zinc ion in the BsKKT2 CL domain occurs via the N-terminal residues Cys 580 and His 584, while that in the Vav1 C1 domain occurs via the N-terminal His 516 and C-terminal Cys 564. More importantly, the CL domain does not have the HX12CX2CXnCX2CX4HX2CX7C motif that is present in all C1 domains. Therefore, the structural similarity of these two distinct domains is likely a product of convergent evolution.
Structural analysis of the C-terminal zinc-binding domain of BsKKT2 revealed a classical C2H2-type zinc finger (Table S3). C2H2 zinc fingers are known to bind DNA, RNA, or protein (Krishna et al., 2003;Brayer and Segal, 2008). In most known cases, two or more C2H2 zinc fingers are used to recognize specific DNA sequences, which is typically achieved by specific interactions between the side chain of residues in positions -1, 2, 3, and 6 in the recognition a-helix (where -1 is the residue immediately preceding the a-helix) and DNA bases (Wolfe et al., 2000). Notably, some proteins with a single zinc finger can recognize specific DNA sequences (Omichinski et al., 1997;Dathan et al., 2002). The C-terminal zinc-binding domain of BsKKT2 consists of one C2H2 domain (-1: Ser 653, 2: Thr 655, 3: Lys 656, 6: Tyr 659). The sequence alignment of the BsKKT2 C2H2 zinc finger shows that residues at the position -1 and 3 are highly conserved in trypanosomatids, while those at position 2 and 6 are not ( Figure 8A).

The CL domain structure is conserved in Perkinsela KKT2a
We next asked whether the central domain structure is conserved among kinetoplastids. Perkinsela is a highly divergent endosymbiotic kinetoplastid that lives inside Paramoeba (Tanifuji et al., 2017).
Our homology search identified three proteins that have similarity to KKT2 and KKT3. Interestingly, similarities among these Perkinsela proteins are higher than those between them and KKT2 or KKT3 in other kinetoplastids ( Figure S3). Because these Perkinsela proteins overall have higher sequence similarity to KKT2 than KKT3, we call them PkKKT2a (XU18_4017), PkKKT2b (XU18_0308), and PkKKT2c (XU18_4564). PkKKT2c does not have a kinase domain, like KKT20 in other kinetoplastids (Nerusheva and Akiyoshi, 2016). Our sequence alignment suggests that PkKKT2a and PkKKT2b have a CL-like domain but lack a C2H2 zinc finger ( Figure S3).
The CL-like domain of PkKKT2a overlaps closely with that of BsKKT2 (RMSD: 0.79 Å across 39 C ), with the exception of some differences being localized to the loop insertion and the absence of a C2H2 domain in PkKKT2a 551-679 ( Figure 7C), consistent with our sequence analysis ( Figure S3).
Taken together, our structures have revealed that the CL domain is conserved in BsKKT2 and PkKKT2a. Given the sequence similarity of KKT2 between Bodo saltans and other trypanosomatids ( Figure 8A), it is likely that the unique CL domain structure is conserved among kinetoplastids.

Perkinsela KKT2a has DNA-binding activity
Although Perkinsela KKT2a lacks a C2H2 zinc finger, our sequence analysis of the KKT2 central domain revealed a putative C2H2 zinc finger not only in trypanosomatids and bodonids but also in one of Prokinetoplastina's KKT2-like proteins, PhM_4_m.86555 ( Figure S3). Furthermore, DNA-binding SPKK motifs (Suzuki, 1989) are present right after the C2H2 zinc finger in many kinetoplastids, while an AT-hook motif is present within the CL-like domain of Perkinsela KKT2b ( Figure S3). These observations suggest that the central domain of KKT2 and KKT3 have DNAbinding activity, perhaps stabilizing their localization at centromeres. To test this hypothesis, we performed fluorescence polarization assays using fluorescently-labelled DNA probes. Unfortunately, we were unable to obtain reliable data for T. brucei and B. saltans KKT2 central domains due to fluorophore quenching. We therefore focused on PkKKT2a 551-679 that has a CL-like domain. Our fluorescence polarization assay showed that PkKKT2a 551-679 has DNA-binding activity with a Kd of ~500 nM on three DNA probes that have different GC contents (50 mer CEN is 50-bp DNA sequence from the CIR147 centromere repeat in T. brucei) ( Figure 7D). To assess the importance of the structural integrity for DNA binding, we next performed fluorescence polarization assay for PkKKT2a 551-679 that has mutations in Zn-coordinating residues (C646 and C649, which correspond to C616 and C619 in TbKKT2) and found that it has a similar DNA-binding affinity compared to wildtype PkKKT2a 551-679 ( Figure 7E). As a comparison, we used a well-characterized zinc finger (Designed zinc finger) that binds a specific DNA sequence (Designed DNA) (Jantz and Berg, 2010).
This protein bound its optimal DNA sequence with a Kd of 8 nM, while it had weaker affinity for 20bp and 50-bp probes from the CIR147 centromere sequence ( Figure 7F). These results show that PkKKT2a 551-679 has DNA-binding activity. It is possible that PkKKT2a 551-679 has higher DNA-binding activity for Perkinsela centromere DNA sequences (yet to be identified).

The CL domain of KKT2 is important for long-term viability in T. brucei
To examine the functional relevance of the CL domain and C2H2 zinc finger, we tested their mutants in T. brucei. We first made various mutants in full-length TbKKT2 and found that all mutants localized at kinetochores ( Figure S4). Because TbKKT2 has multiple domains that can independently localize at kinetochores (Figure 3), we next expressed mutants in our ectopic expression of the central domain (TbKKT2 562-677 ). We found that mutations in Zn-coordinating residues of the CL domain (C576A, H580A, C597A, C600A, C616A, C619A) all abolished kinetochore localization ( Figure 8A, B). In contrast, similar mutations in the C2H2-type zinc finger (C640A C643A) did not affect the localization.
To gain insights into how the KKT2 CL domain may promote kinetochore localization, we analyzed conservation and electrostatic potential of the BsKKT2 CL domain surface residues to identify possible patches that may be involved in this process. Our analysis revealed a highly conserved acidic patch, centered around residue BsKKT2 D626 ( Figure 5C, D, and Figure S5).
Interestingly, this aspartic acid is strictly conserved in all KKT2 and KKT3 proteins ( Figure S3). To test the importance of this residue, we mutated the corresponding residue in T. brucei and found that TbKKT2 562-677 with either D622A or D622E failed to localize at kinetochores ( Figure 8B). Taken together, our results show that the CL domain, but not the C2H2 zinc finger, is important for the kinetochore localization of the TbKKT2 central domain.
To test the importance of the TbKKT2 central domain for cell viability, we next performed rescue experiments. We replaced one allele of TbKKT2 with an N-terminally YFP-tagged TbKKT2 construct that has either wild-type or mutant versions of the central domain, and performed RNAi against the 5′UTR of the TbKKT2 transcript to knockdown the untagged allele of TbKKT2 ( Figure   S6) (Ishii and Akiyoshi, 2020). As expected, mutants in the CL domain (C576A and D622A) and the C2H2 zinc finger (C640A C643A) both localized at kinetochores ( Figure 8C and Figure S7). Upon induction of RNAi, however, the CL domain mutants failed to support normal cell growth after day 4, while the C2H2 zinc finger mutants rescued the growth defects. These data confirm the importance of the CL domain for the function of TbKKT2 in vivo.

Localization of KKT3 depends on the central domain in T. brucei
The central domain of TbKKT3 can localize at kinetochores throughout the cell cycle ( Figure 4). The sequence similarity of the central domain between KKT2 and KKT3 suggested that TbKKT3 likely consists of two domains, which correspond to the CL domain and the C2H2 zinc finger present in KKT2 ( Figure S3). Consistent with this prediction, mutating TbKKT3 residues that align with Zncoordinating histidine or cysteine residues in the CL domain of KKT2 abolished the kinetochore localization of the ectopically-expressed full length TbKKT3 protein ( Figure 9A, B). We also found that the conserved aspartic acid D692 was essential for kinetochore localization. In contrast, mutations in the TbKKT3 C2H2 zinc finger (C707A C710A) did not affect the kinetochore localization.
We next performed rescue experiments by replacing one allele of TbKKT3 with a Cterminally YFP-tagged construct that has either wild-type or mutant versions of the central domain, and performed RNAi against the 3′UTR of TbKKT3 to knockdown the untagged allele of TbKKT3 ( Figure S6). We first confirmed that TbKKT3 CL domain mutants (C668A C671A and D692A) were unable to localize at kinetochores, while the TbKKT3 C2H2 zinc finger mutant (C707A C710A) localized normally ( Figure 9C and Figure S7). Upon induction of RNAi, CL mutants failed to rescue the growth defect, showing that kinetochore localization is essential for the TbKKT3 function ( Figure   9C). In contrast, the TbKKT3 C2H2 zinc finger mutant supported normal cell growth. These data show that the TbKKT3 CL domain is essential for the localization and function of TbKKT3.

Discussion
A major open question concerning the biology of kinetoplastids is how these organisms assemble kinetochores specifically at centromeres using a unique set of kinetochore proteins. Studies in other eukaryotes have shown that constitutively localized kinetochore proteins (such as CENP-A and CENP-C) play crucial roles in kinetochore specification and assembly (French and Straight, 2017;Hamilton and Davis, 2020;Kixmoeller et al., 2020). Among the six proteins that constitutively localize at kinetochores in T. brucei (KKT2, KKT3, KKT4, KKT20, KKT22, and KKT23), we previously showed that KKT4 is important for recruiting KKT20 but not many other proteins including KKT2 and KKT3 (Llauró et al., 2018). In this study, we show that KKT2 and KKT3 are important for recruiting multiple kinetochore proteins (including KKT4, KKT22, and KKT23), while their localization is independent from various kinetochore proteins. Together with the fact that KKT2 and KKT3 have DNA-binding motifs, these results support a hypothesis that KKT2 and KKT3 locate at the base of kinetoplastid kinetochores and play crucial roles in recruiting other kinetochore proteins.
It will be important to identify which proteins directly interact with KKT2/3 to understand the mechanism of kinetochore assembly. KKT2/3 have sequence similarity with polo-like kinases (Nerusheva and Akiyoshi, 2016). In addition to an N-terminal protein kinase domain and a C-terminal divergent polo boxes, they have a central domain that is highly conserved among kinetoplastids. By ectopically expressing fragments of KKT2 and KKT3 in T. brucei, we established that their central domains can localize specifically at centromeres. The crystal structure of the Bodo saltans KKT2 central domain revealed a unique structure, which consists of two distinct zinc-binding domains, the CL domain and C2H2-type zinc finger. It is likely that the central domain of T. brucei KKT2 has a similar structure based on high sequence similarity between BsKKT2 and TbKKT2 proteins. Importantly, mutational analyses of TbKKT2 revealed that the CL domain is important for the localization of the central domain, while the C2H2 zinc finger is not. Furthermore, although full-length TbKKT2 CL mutants localized at kinetochores (likely due to interactions with other kinetochore proteins via other domains of TbKKT2), they were not fully functional. Taken together, these data have established that the CL domain is essential for the function of TbKKT2, which is consistent with the presence of CL, but not C2H2 zinc finger, in Perkinsela KKT2a. It remains unclear whether the structure of the central domain is conserved between KKT2 and KKT3. Nonetheless, our functional studies showed that the equivalent domain of CL in TbKKT3 was also essential for the kinetochore localization and function, while the equivalent domain of C2H2 zinc finger was not, showing that the functional importance of CL is conserved in KKT3. It will be important to obtain KKT3 central domain structures to reveal structural similarity or difference between KKT2 and KKT3. It is noteworthy that all identified KKT2/3 homologs in deep-branching Prokinetoplastina have higher similarity to KKT2 than KKT3.
We speculate that ancestral kinetoplastids had only KKT2-like proteins that carried out all necessary functions and that KKT3 in trypanosomatids and bodonids represents a product of gene duplication that became specialized in certain functions (such as more efficient centromere localization by its central domain compared to KKT2).
It remains unclear how kinetoplastids specify kinetochore positions. The fact that KKT2/3 central domains manage to localize at centromeres suggests that they are able to recognize something special at centromeres. What might be a unique feature at centromeres in kinetoplastids that lack CENP-A? Histone variants are one possibility. T. brucei has four histone variants, H2AZ, H2BV, H3V, and H4V. However, none of them is specifically enriched at centromeres (Lowell and Cross, 2004;Lowell et al., 2005;Siegel et al., 2009), and histone chaperones did not co-purify with any kinetochore protein (Akiyoshi and Gull, 2014). Alternatively, there might exist certain posttranslational modifications on histones or DNA specifically at centromeres (e.g. phosphorylation, methylation, acetylation, ubiquitination, or sumoylation). The KKT2 CL domain has a highly conserved acidic patch, which might act as a "reader" for such modifications. Although there is no known histone or DNA modification that occurs specifically at centromeres, KKT2/3 have a protein kinase domain and KKT23 has a Gcn5-related N-acetyltransferase (GNAT) domain (Nerusheva et al., 2019). It will be important to examine whether these enzymatic domains are important for proper recruitment of KKT2/3 central domains. Another unique feature at centromeres is the presence of kinetochore proteins, which could potentially recruit newly synthesized kinetochore components by direct protein-protein interactions. Finally, it is important to note that it remains unclear whether kinetoplastid kinetochores build upon nucleosomes. It is formally possible that the KKT2/3 central domains directly bind DNA and form a unique environment at centromeres. Understanding how the KKT2/3 central domains localize specifically at centromeres will be key to elucidating the mechanism of how kinetoplastids specify kinetochore positions in the absence of CENP-A.

Tryp cells and plasmids, microscopy, immunoprecipitation, and mass spectrometry
All trypanosome cell lines, plasmids, primers, and synthetic DNA used in this study are listed in Table S4. All trypanosome cell lines used in this study were derived from T. brucei SmOxP927 procyclic form cells (TREU 927/4 expressing T7 RNA polymerase and the tetracycline repressor to allow inducible expression) (Poon et al., 2012). Cells were grown at 28˚C in SDM-79 medium supplemented with 10% (v/v) heat-inactivated fetal calf serum (Brun and Schönenberger, 1979).
Details of other plasmids are described in Table S4. Site-directed mutagenesis was performed using primers and template plasmids listed in Table S4. All constructs were sequence verified.
Plasmids linearized by NotI were transfected into trypanosomes by electroporation into an endogenous locus (pEnT5-Y derivatives and pBA148/pBA192/pBA892 derivatives) or 177 bp repeats on minichromosomes (pBA310 derivatives and p2T7-177 derivatives). Concentrations of drugs used were as follows: 5 µg/ml for phleomycin, 25 µg/ml for hygromycin, 10µg/ml for blasticidin, 30 µg/ml for G418, and 1 µg/ml for puromycin. To obtain endogenously-tagged clonal strains, transfected cells were selected by the addition of appropriate drugs and cloned by dispensing dilutions into 96-well plates. Clones that express mutant versions of KKT2 or KKT3 from the endogenous locus were screened by Sanger sequencing of genomic DNA. Expression of GFP-NLS fusion proteins (pBA310 derivatives) was induced by the addition of doxycycline (10 ng/ml). RNAi was induced by the addition of doxycycline (1 µg/ml).
Fluorescence microscopy, immunoprecipitation of GFP-fusion proteins, and mass spectrometry were performed essentially as described previously (Nerusheva and Akiyoshi, 2016;Ishii and Akiyoshi, 2020). Proteins identified with at least two peptides were considered as significant and shown in Table S1.

Multiple sequence alignment
Protein sequences and accession numbers for KKT2 and KKT3 homologs were retrieved from

Protein expression and purification
Multiple sequence alignment together with secondary structure predictions of the KKT2 central domain were used to design constructs in Bodo saltans and Perkinsela. To make pBA1660 (BsKKT2 572-668 with an N-terminal TEV-cleavable hexahistidine (His6) tag), the central domain of Bodo saltans KKT2 (accession number BSAL_50690) was amplified from BAG50 (a synthetic DNA that encodes Bodo saltans KKT2, codon optimized for expression in Sf9 insect cells (Table S4) coli cells at 20˚C using auto induction media (Formedium) (Studier, 2005).
Briefly, 500 mL of cells were grown at 37˚C in 2.5 L flasks at 300 rpm until OD600 of 0.2-0.3 and then cooled down to 20˚C overnight (2 L for BsKKT2, 6 L for PkKKT2a, and 2L for Designed zinc finger). Cells were harvested by centrifugation and resuspended in 50 ml per litre of culture of lysis buffer (25 mM Hepes pH 7.5, 150 mM NaCl, 1 mM TCEP, 10 mM imidazole, and 1.2 mM PMSF). Proteins were extracted by mechanical cell disruption using a French press (1 passage at 20,000 PSI) and the resulting lysate was centrifuged at 48,384 g for 30 min at 4˚C. Clarified lysate was incubated with 5 mL TALON beads (Takara Bio), washed with 150 mL lysis buffer and eluted in 22 mL of elution buffer (25 mM Hepes pH 7.5, 150 mM NaCl, 1 mM TCEP, and 250 mM imidazole) in a gravity column, followed by TEV treatment for the removal of the His6 tag. Salt concentration of the sample was subsequently reduced to 50 mM NaCl using buffer A (50 mM Hepes pH 7.5, and 1 mM TCEP) and the sample was loaded onto a 5 mL HiTrap Heparin HP affinity column (GE healthcare) pre-equilibrated with 5% buffer B (50 mM Hepes pH 7.5, 1 M NaCl, and 1 mM TCEP) on an ÄKTA pure 25 system. Protein was eluted by using a gradient from 0.05 to 1 M NaCl, and proteincontaining fractions were combined, concentrated with an Amicon stirred cell using an ultrafiltration disc with 10 kDa cut-off (Merck), and then loaded onto a HiPrep Superdex 75 16/60 size exclusion chromatography column (GE healthcare) pre-equilibrated with 25 mM Hepes pH 7.5, 150 mM NaCl, and 1 mM TCEP. Fractions containing the protein of interest were pooled together, concentrated with an Amicon stirred cell using an ultrafiltration disc with 10 kDa cut-off, and stored at -80˚C. Designed zinc finger was buffered exchanged into 50 mM Tris pH 7.5, 1 mM ZnCl2, 50 mM NaCl and 1 mM TCEP prior storage. Protein concentration was measured by Bradford assay.

Crystallization
Both BsKKT2 572-668 and PkKKT2a 551-679 crystals were optimized at 4˚C in sitting drop vapour diffusion experiments in 48-well plates, using drops of overall volume 400 nL, mixing protein and mother liquor in a 3:1 protein:mother liquor ratio. BsKKT2 central domain crystals grew from the protein at 26 mg/mL and mother liquor 40% PEG 400, 0.2 mM (NH4)2SO4, and 100 mM Tris-HCl pH 8. The 40% PEG400 in the mother liquor served as the cryoprotectant when flash-cooling the crystals by plunging into liquid nitrogen. PkKKT2a 551-679 crystals grew from the protein at 13 mg/mL and mother liquor 19% MPD, 50 mM Hepes pH 7.5, and 10 mM MgCl2. The crystals were briefly transferred into a cryoprotecting solution of 30% MPD, 50 mM Hepes pH 7.5, 10 mM MgCl2 prior to flash-cooling.

Data collection and structure determination
X-ray diffraction data from a BsKKT2 central domain crystal were collected at the I04 beamline at the Diamond Light Source (Harwell, UK) at the Zinc K-edge wavelength (l=1.28297 Å). A set of 1441 images were processed in space group I222 using the Xia2 pipeline (Winter, 2010), with DIALS for indexing and integration (Winter et al., 2018) and AIMLESS for scaling (Evans and Murshudov, 2013) to 1.8 Å resolution. Initial 3 Zn atoms were localized by interpreting the anomalous difference Patterson, single anomalous dispersion (SAD) phases were estimated using Crank2 (Skubák and Pannu, 2013), and an initial model was built with BUCCANEER (Cowtan, 2006). The structure was completed by several cycles of alternating model building in Coot (Emsley et al., 2010) and refinement in autoBUSTER (Blanc et al., 2004;Bricogne et al., 2017).
PkKKT2a 551-679 X-ray diffraction data were collected at the I03 beamline at Diamond Light Source (Harwell, UK) also at the zinc K-edge (l=1.28272 Å) and processed using the autoPROC pipeline (Vonrhein et al., 2011) using XDS (Kabsch, 2010 for indexing/integration and AIMLESS (Evans and Murshudov, 2013) for scaling, to a resolution of 3.8 Å. Two initial Zn positions were determined by interpreting the anomalous difference Patterson, and single anomalous dispersion (SAD) phases were estimated using Crank2 (Skubák and Pannu, 2013) and SHARP (Vonrhein et al., 2007) in space group P64. An initial model was manually built in Coot and refined once with RosettaMR (Terwilliger et al., 2012). The structure was completed by several cycles of alternating model building in Coot (Emsley et al., 2010) and refinement in autoBUSTER (Blanc et al., 2004;Bricogne et al., 2017).
A higher resolution dataset was collected from a PkKKT2a 551-679 crystal at the I24 beamline at Diamond Light Source (Harwell, UK), at a wavelength of l=0.9686 Å. Data were processed using Xia2 pipeline (Winter, 2010), DIALS (Winter et al., 2018), and AIMLESS (Evans and Murshudov, 2013) in space group P64 to a resolution of 2.9 Å. The model obtained from the 3.8 Å dataset was used for further model building and refinement with autoBuster (Bricogne et al., 2017) and Coot (Emsley et al., 2010).

Fluorescent anisotropy DNA-binding assay
All experiments were performed in binding buffer (25 mM Hepes pH 7.5, 50 mM NaCl and 1 mM TCEP) using 1 nM FAM-labelled dsDNA sequences purchased from IDT (Table S4). Prior to the assay, proteins were buffer exchanged into binding buffer using a Zeba spin desalting column (Thermo Fisher), serially diluted at 2:3 v/v ratio and then incubated with DNA for 20 min at room temperature. Fluorescence anisotropy was measured at 25°C using a PHERAstar FS next-generation microplate reader (BMG-Labtech). Each data point is an average of 3 independent experiments. Data were fitted with SigmaPlot using a standard four-parameter logistic equation to calculate Kd.  In parentheses the values relative to the highest resolution shell. In parentheses the values relative to the highest resolution shell.   showing that KKT1 and KKT4 failed to form normal kinetochore-like dots but instead formed bright blobs upon KKT2/3 depletion. Cells were fixed at day 1 post-induction. Scale bars, 5 µm. (E) TbKKT2 DPB WT, not W1048A, co-purifies with several kinetochore proteins. Cell lines, BAP517, BAP535. Inducible GFP-NLS fusion proteins were expressed with 10 ng/mL doxycycline, and immunoprecipitation was performed using anti-GFP antibodies. See Table S1 for all proteins identified by mass spectrometry. (D) TbKKT3 DPB co-purifies with several kinetochore proteins. Cell line, BAP520.