As a first step in determining relationships among kinetochores in different organisms, we searched fungal genomes for point CENs similar in structure to those in S. cerevisiae. Three such examples are already known, C. glabrata, E. gossypii and K. lactis 28, but a significant number of newly sequenced genomes have not yet been analyzed. Finding new CENs with a CDEI-CDEII-CDEIII structure is not trivial because the number of identical bases in CDEI and CDEIII is relatively small, even among chromosomes in S. cerevisiae. Moreover, CDEII is not conserved in sequence but, rather, is characterized by high AT content and alternating runs of poly-A and poly-T. To capture this information we constructed a tri-partite computational model based on profiles for CDEI and CDEIII, a hidden Markov model (HMM) for CDEII (Figure 1a), and S. cerevisiae CENs as a training set. When the model was tested on C. glabrata, E. gossypii and K. lactis, organisms whose genomes are fully annotated, 6/13 centromeres in C. glabrata, 6/7 centromeres in E. gossypii and 6/6 in K. lactis were identified correctly (Figure 1b). Conversely, no point-CEN sequences were found in S. pombe, C. albicans or A. nidulans, organisms known to have regional CENs (Figure 1b). With a success rate of >70% and a false positive rate of <5%, we conclude that our computer model is effective at finding point CENs.

When unannotated genomes were analyzed using the tri-partite computational model, 15 CDEI-II-III sequences were found in S. bayanus,14 in S. mikatae and 15 in S. paradoxus (Figure 1b) 29. S. bayanus, S. mikatae and S. paradoxus contigs have not yet been fully assembled, but sequence similarity and synteny suggest that all 3 have 16 chromosomes, close to the number of putative CEN sequences identified computationally in each organism. When these newly identified point CENs were combined with those in the literature, 85 CDEI-II-III sequences from 7 organisms became available. These yielded a clear consensus for CDEI and CDEIII and revealed that, within a single organism, CDEII can vary in sequence from one chromosome to the next but that length distributions are very narrow (± 3%; Figure 1c, d). Most fungi have 84 bp CDEII sequences but E. gossypii and K. lactis have 164 bp CDEIIs, suggesting the presence of two copies of an underlying approximately 84 bp CDEII module (Figure 1d). To a first approximation, the extent of conservation among CDEI and CDEIII sequences on different chromosomes within a single organism was not much greater than the extent of conservation among syntenic CENs in different organisms (Figure 1c). Together, these data strongly imply that all organisms with CDEI-II-III point CENs arose from a relatively recent common ancestor.

Does the existence of CENs with similar CDEI-II-III structures imply the existence of similar DNA-binding kinetochore proteins? In addressing this question, the CDEI-binding Cbf1 protein is not very useful because it functions not only as a kinetochore subunit but also as a transcription factor for a set of highly conserved biosynthetic genes 30, implying conservation of non-kinetochore function. We therefore concentrated on components of the CBF3 complex, three of whose subunits are thought to function only in CDEIII-binding (the fourth subunit, scSkp1, is also a component of the SCF ubiquitin ligase complex 31 and, like Cbf1, has conserved non-kinetochore functions). When PSI-BLAST was used to search predicated open reading frames in 17 fungal genomes for orthologs of scCtf13, scCep3 and scNdc10, all 3 CBF3 subunits were found in the organisms with point CENs (7 in total), but not in organisms with regional CENs (Figure 1e). As a positive control for the PSI-BLAST search, orthologs of scMis6^Ctf3 and scSpc105 could be found in all fungi examined (Figure 1e). Importantly, Mis6^Ctf3 and Spc105 have approximately the same degree of sequence divergence in point-CEN containing fungi (51% and 48% similarity, respectively) as Ndc10 (48% similarity; Table 1). We provisionally conclude that CBF3 proteins are present only in fungi with CDEI-II-III CEN DNA whereas other kinetochore proteins (such as Spc105 and Ctf3) are ubiquitous. Moreover, when organisms with point CENs and CBF3 subunits are mapped on a phylogenetic tree (constructed using the highly conserved reference proteins α-tubulin, the signal recognition particle subunit SRP54 and PCNA) they were found to cluster closely together (Figure 1f). While recognizing the possibility for false-negative findings in cross-species sequence searching, we conclude that CDEI-II-III CENs and CBF3 CEN-binding proteins are probably found only in a subset of closely related budding yeasts and, thus, may have co-evolved. Intriguingly, the apparent common ancestor of point-CEN and regional-CEN organisms appears to be a fungus containing regional CENs, implying that simple point CENs arose from complex regional CENs and not the other way round.

To delineate further which kinetochore proteins are specific to point CENs, and which are more widely distributed, we analyzed all known S. cerevisiae kinetochore proteins for sequence conservation. As a starting point we examined scMis12^Mtw1 and scNdc80^Hec1, kinetochore proteins first identified in yeast and subsequently shown to have human orthologs (hsMis12 and hsNdc80^Hec1) that localize to kinetochores and play a role in chromosome segregation 2025. Experimental and sequence data establish that yeast and higher cell Ndc80^Hec1 and Mis12^Mtw1 proteins represent true orthologs 20323334. Nonetheless, the overall degree of similarity among Ndc80^Hec1 and Mis12^Mtw1 proteins across eukaryotes was found to be relatively modest (approximately 15% to 30%) as compared to proteins involved in DNA replication (PCNA, approximately 75%) or protein translocation (SRP54, approximately 60%). Multiple protein sequence alignments of fungal, plant, and metazoan Ndc80^Hec1 and Mis12^Mtw1 showed that sequence similarity is confined to 30 to 100 residue blocks interspersed by stretches of non-homology, many of which correspond to coiled coils (Figure 2a, b). This pattern of block-by-block similarity was also observed with five other kinetochore proteins for which orthology has been established experimentally, and is consistent with previous proposals that kinetochore proteins have evolved rapidly 35 (Figure 2c). Importantly, for our purposes, data obtained from known kinetochore orthologs suggests that it is necessary to use conserved blocks, rather than complete sequences, when searching kinetochore proteins for patterns of sequence conservation.

When 55 S. cerevisiae kinetochore proteins (including the CBF3 subunits discussed above) were used in PSI-BLAST queries to search 14 fully annotated fungal genomes (Additional data file 1), 41 were found to have orthologs in organisms with both point and regional CENs (Figure 3). These proteins included kinetochore regulators such as the Mad1-3, Bub1, BubR1/Mad3 and Mps1 checkpoint proteins and the Ipl1-AuroraB kinase, as well as many structural components. In addition to the 41 proteins mentioned above, conservation was observed for proteins such as Skp1 31, Cbf1 3036 and some MAPs 37 that function at kinetochores as well as at other locations in the cell. As noted above, these proteins are likely to have been conserved for reasons other than their presence at kinetochores, and they cannot be used to infer overall similarity in kinetochore structure. In this respect, kinesin motor proteins are also difficult to analyze. Eukaryotic cells contain multiple kinesins, which are known to fall into 14 highly conserved protein families based on sequence, structure and function 38. Typically, each kinesin has more than one cellular function and kinetochores in different organisms recruit different kinesin family members, making it difficult to determine (in the absence of experimentation) which kinesins should be considered kinetochore associated.

Leaving these complications aside, among 55 fungal kinetochore components analyzed, 11 were found in the 7 organisms with point CENs and nowhere else, implying that they are specific to a CDEI-II-III CEN architecture (Figure 3). These 11 proteins include the CBF3 subunits scCtf13, scCep3 and scNdc10 described above, the non-essential CNN1 gene product, 1 subunit of the SPC105 complex (Ydr532c), two subunits of the COMA linker complex (scAme1 and scOkp1) and 4 proteins that require COMA for CEN-association (scMcm22, scMcm16, scNkp1 and scNkp2). Among organisms in which they are found, the 11 point CEN-specific proteins are as well or better conserved than ubiquitous kinetochore proteins, implying that failure to identify orthologs in more distant fungi is a consequence of their actual absence. We therefore propose that approximately 20% of the overall kinetochore in fungi containing CDEI-II-III CENs is specialized to their simple CENs. As expected, these specialized kinetochore subunits include proteins in direct contact with CEN DNA (Figure 3).

Based on success in identifying fungal orthologs of S. cerevisiae kinetochore proteins, we expanded our set of target organisms to higher eukaryotes (see Figure 4 for a schematic of the approach). Alignments were created for 41 ubiquitous fungal proteins and conserved blocks determined. The non-redundant NCBI protein database was then searched for these conserved blocks using PSI- BLAST or Prosite pattern searching algorithms (see Materials and methods for details). Potential orthologs differing greatly in size from the fungal proteins and candidates with well-established non-kinetochore functions were eliminated from further consideration. The remaining proteins were then aligned to confirm the presence of conserved blocks. This search led to the identification, in a wide variety of organisms, of previously unreported orthologs of many S. cerevisiae kinetochore proteins (Additional data file 1), among which were four new human kinetochore proteins (Figure 4). Recent analysis of S. pombe kinetochore complexes by mass spectrometry revealed the presence of a set of proteins for which orthologs could not be found in S. cerevisiae 3940. When conserved sequence blocks from these S. pombe proteins were used to search the genomes of higher eukaryotes, two additional human proteins were flagged as likely kinetochore subunits (Figure 4). Regardless of which fungi contributed to the sequence blocks, the most highly conserved kinetochore subunits were invariably regulatory proteins such as the Mad and Bub checkpoint proteins and the Aurora B kinase. Structural proteins such as Ndc80^Hec1, Nuf2, CENP-C^Mif2 and Mis12^Mtw1 were considerably more diverged.

The four human proteins representing hitherto unrecognized orthologs of S. cerevisiae kinetochore subunits were provisionally named hsNnf1-Related (hsNnf1R; also known as PMF1 41; Figures 4 and 5), hsNsl1R (also known as DC8 or DC31), hsMcm21R and hsChl4-R. hsNnf1R shares with its fungal counterpart 2 conserved blocks of 30 to 35 residues with 47% and 67% similarity, hsNsl1R shares 1 conserved block of 35 residues with 43% similarity, hsMcm21R shares 3 conserved blocks of 15 to 30 residues with 46%, 87% and 33% similarity and hsChl4R shares 2 conserved blocks of 20 and 50 amino acids with 45% and 40% similarity (Figure 5). The potential human orthologs of S. pombe Fta1 and Sim4 were provisionally named hsFta1R and hsSim4R (also known as Solt 42). hsFta1R shares with its fungal counterpart three conserved sequence blocks of 40, 25 and 30 residues with 48%, 49% and 58% similarity and hsSim4R one block of 27 residues with 65% similarity (Figure 6). Elsewhere we will describe experimental data showing that hsChl4R, hsNsl1R, hsMcm21R, hsNnf1R, hsFta1R and hSim4R localize to kinetochores in human cells and are required for accurate chromosome segregation (AD McAinsh et al., submitted). Importantly, for the purposes of the current analysis, the identification of new human kinetochore proteins means that one or more subunits are present in metazoans for each of the four multi-protein linker complexes forming the core of the S. cerevisiae kinetochore. Thus, it appears that simple point CENs in budding yeast and complex regional CENs in human cells probably share fundamental architectural similarities.

S. cerevisiae DASH is a 10-protein MT-binding complex that has attracted considerable recent interest because it forms rings encircling MTs 4344. DASH subunits are conserved among fungi but we have found few if any potential orthologs in higher eukaryotes. The closest match to a DASH protein in humans, NYD-SP28 45, has an amino-terminal domain of about 30 amino acids 40% similar to S. cerevisiae Spc34 (Additional data file 2). The Chlamydomonas rheinhardtii ortholog of NYD-SP28 localizes to the flagellum 46, implying that NYD-SP28 might be involved in interactions with MTs. Our preliminary conclusion is that higher eukaryotes do not contain a protein complex closely related to fungal DASH, although further investigation of NYD-SP28 is warranted.

Several kinetochore proteins first identified in human cells have previously been shown to have fungal orthologs, including CENP-C (orthologous to scMif2p 47) and CenH3^CENP-A (orthologous to scCse4 48). We therefore wondered whether additional orthologs might be found in fungi for kinetochore proteins hitherto characterized only in higher eukaryotes, such as CENP-E, CENP-H, Rod, Zwint and Zwilch 4950515253. We found that, among fungal proteins, hsCENP-H is most similar to S. pombe spFta3 (Figure 7a), which was shown recently to be a fission yeast kinetochore protein 39. It has been suggested previously that S. cerevisiae scNnf1 is the budding yeast CENP-H ortholog 54 (Figure 7b) but we find that scNnf1 is actually much more similar to hsNnf1R^Pmf1 and spNnf1 than to CENP-H (Figure 7c). We therefore propose that CENP-H is orthologous to the fungal Fta3 family of proteins. Searches using PSI-BLAST revealed that the Fta3 protein, like the Sim4 and Fta1 proteins with which it interacts in S. pombe 39, has apparent orthologs only in organisms with regional CENs (Additional data file 1). The presence of Sim4 and Fta1 in the budding yeast Yarrowia lipolytica, which has regional CENs, but not in yeasts with point CENs, is striking, since Y. lipolytica is significantly closer in overall sequence to S. cerevisiae than to S. pombe. We therefore conclude that Fta3, Sim4 and Fta1 are members of a class of kinetochore proteins found specifically in fungi and metazoa with regional CENs and not in fungi with point CENs.

In contrast to CenH3^CENP-A, CENP-C and CENP-H, potential orthologs of the human CENP-E, Rod, Zwint and Zwilch proteins were not found in any of the fungi examined. The apparent absence of a fungal Rod or Zwilch is particularly interesting, since their binding partner at human kinetochores, Zw10, has a potential ortholog in S. cerevisiae, Dsl1 55. Both hsZw10 and scDsl1 play a role in membrane trafficking during interphase 56, but scDsl1 is not known to localize to kinetochores. Thus, whereas human hsZw10 functions in vesicle-MT and chromosome-MT interaction, scDsl1 appears to have only the former function, presumably because Rod and Zwilch are not present. The absence of Zw10 from fungal kinetochores is also sufficient to explain the absence of Dynein: the Rod/Zw10/Zwilch (RZZ) complex is needed for the association of Dynein with human and Drosophila kinetochores 57. Considering these data together, we conclude that animal cell kinetochores contain proteins, currently comprising perhaps 25% of the total (and likely to increase), that are absent in fungi with either regional or point CENs.

Thus far, we have distinguished only between point and regional CENs but a more nuanced view can be obtained from phylogenetic analysis of kinetochore structural proteins. As a reference for these comparisons, a tree was constructed by combining data on three well-conserved eukaryotic proteins: α-tubulin, PCNA and SRP54 (Figure 8a; this reference tree closely matches reference trees constructed by others 5859). The reference tree exhibited prototypical clustering of fungi in one branch and metazoa in another so that Drosophila and C. elegans were much closer to humans than S. pombe or S. cerevisiae. However, the phylogenetic trees for Ndc80^Hec1 and Nuf2 were remarkably different: overall sequence divergence was much greater and Drosophila and C. elegans Ndc80^Hec1 (or Nuf2) proteins were not significantly more similar to their human than their fungal counterparts (Figure 8b, c). Drosophila Ndc80 and Nuf2 were particularly striking in occupying a branch of the phylogenetic tree distant from all other animals. This great divergence in Drosophila kinetochore protein sequence is also illustrated by the fact that, apart from regulatory components, such as the Mad-Bub proteins and a few MAPs, only a limited number of structural kinetochore proteins have been identified in flies (for example, CENP-C 60, CenH3^CID 61, the RZZ complex 62, Ndc80^Hec1, Nuf2, and Mis12^Mtw1 ;Figure 9).

Encephalitozoon cuniculi is a microsporidium and intracellular parasite that has been subjected to considerable evolutionary pressure to reduce its genome to the smallest possible size. As a consequence, E. cuniculi and related microsporidia have the smallest known eukaryotic proteome (1,997 potential open reading frames) and many cellular structures in E. cuniculi lack redundant and non-essential genes 63. Using our HMM for CDEI-II-III, no sequences similar to point CENs were found on any of the 11 E. cuniculi chromosomes, nor were CBF3 proteins found by PSI-BLAST (Figure 10a). We therefore speculate that E. cuniculi contains a regional CEN of some sort. Orthologs of CenH3 and CENP-C^Mif2 are present in E. cuniculi, as are all four components of the NDC80 linker complex, three components of MIND and SPC105 (Figure 10b, Additional data file 3). No subunits of COMA, the fourth S. cerevisiae linker, were found. Among regulatory proteins, E. cuniculi Ipl1/Aurora B and Survivin^Bir1 orthologs were present as were Mps1 and Bub3, but not other proteins required for the spindle assembly checkpoint in yeast or human cells (Figure 10b). When Cdc20, an essential activator of the anaphase promoting complex (APC/C) was examined for sequence motifs, further evidence was obtained that E. cuniculi lacks a spindle checkpoint. APC/C is an E3 ligase required for the ubiquitination of proteins whose destruction is necessary at the metaphase-anaphase transition 64. In all eukaryotes examined to date, an activated form of the Mad2 checkpoint protein binds to Cdc20 via a short conserved peptide so as to block Cdc20 from activating APC/C, thereby arresting cells at the metaphase-to-anaphase transition 6566 (Figure 10c). E. cuniculi Cdc20 contains the WD-domain implicated in APC/C interaction but lacks any sequence similar to a Mad2 binding domain (Figure 10c), implying that it is not subject to checkpoint control. From these data we conclude that E. cuniculi probably contains a very simple kinetochore, based on a regional CEN that contains about one-half the proteins found in S. cerevisiae. In contrast, other large multi-protein structures in E. cuniculi are only slightly less complex than their higher eukaryotic counterparts. For example, E. cuniculi ribosomes are composed of 77 subunits as compared to 84 subunits in S. cerevisiae. Symptomatic of the simplicity of the E. cuniculi kinetochore is the absence of the vast majority of potential MAPs. Nonetheless, it is significant that the E. cuniculi kinetochore contains three of the four linker complexes that appear to form the core of budding yeast and human kinetochores.

Based on the simple structure of their CENs, it is widely assumed that S. cerevisiae kinetochores represent an ancestral structure from which complex regional kinetochores evolved. Several findings in the current work suggest, however, that CDEI-II-III CENs arose in combination with a set of 11 proteins as a specialization of a regional CEN. First, all annotated organisms containing point CENs (S. cerevisiae, C. glabrata, K. lactis, and E. gossypii) have a common origin in one relatively shallow branch of the fungal phylogenetic tree. Were CDEI-II-III sequences an ancestral CEN, the current distribution of regional CENs would require loss of point CENs from multiple independent evolutionary branches. Second, we could obtain no evidence for CDEI-II-III CEN DNA or CBF3 proteins in the microsporidium E. cuniculi, which is thought to have arisen through an ancient divergence in the fungal kingdom 68.

If the speculation that CDEI-II-III point-CENs evolved from regional CENs is correct, we must consider the possible existence of other short CENs that are also based on sequence-specific DNA binding interactions just not CBF3. By way of precedent, the emergence of CDEI-II-III CENs is coincident with large-scale chromosomal changes that gave rise to the HMR, HML and MAT loci, thereby changing the sexual potential of S. cerevisiae and related yeasts 29. S. pombe and its close relatives undergo mating type switching analogous to that in S. cerevisiae, but the molecular mechanisms of switching are completely different 69. Functional analysis of fungi with short uncharacterized CENs will be needed to test the speculation that just as different forms of mating-type switching have developed based on distinct biochemistry, point CENs with structures other than CDEI-II-III might exist.

Sequence comparison reveals that conservation among orthologous kinetochore proteins is invariably restricted to relatively short sequence blocks embedded in longer regions of low sequence similarity. The restriction of sequence similarity to small blocks explains the relative difficulty in finding orthologs and the widespread assumption that yeast and human kinetochores are very different. Henikoff and colleagues 67 have studied the evolutionary divergence of CenH3 and CENP-C^Mif2 in some detail and propose that kinetochore proteins are under positive selection in plants and animals as a consequence of meiotic drive by CEN DNA during female meiosis. Rapid evolution in protein sequence is most apparent in worms and flies, and in this study we have added only dmNdc80, dmNuf2 and dmMis12 to the list of likely structural Drosophila kinetochore proteins. Why the rate of kinetochore protein evolution is so much greater in flies and worms as compared to mammals, plants and fungi remains a mystery but it is reminiscent of data on other key regulators of chromosome segregation. Securin and its protease separase are also highly diverged in D. melanogaster: Drosophila securin, unlike the human and yeast proteins, consists of two separate gene products, called three rows and pimples, that interact with an unusually short separase 70. Moreover, unlike the majority of eukaryotes that utilize an RNA-templated reverse transcriptase to replicate telomeres, D. melanogaster uses an alternative mechanism based on transposition of the HeT-A and TART retrotransposable elements 71. It seems very likely that several distinct classes of kinetochore arose early in evolution. Perhaps surprisingly, fungal kinetochores appear to be as good a model for their human counterparts as kinetochores in organisms such as worms and flies.

For the majority of kinetochore proteins we have little knowledge of their biochemical functions or their structure. It is tempting to speculate that conserved sequence blocks represent protein-protein interaction domains or interaction surfaces under tight evolutionary pressure. However, with very few exceptions (for example, the kinase domains of checkpoint and regulatory proteins and motor domains of kinesins), blocks of conserved sequence do not correspond to recognizable functional domains. This stands in contrast to the situation in nuclear pore complexes, in which highly conserved and recognizable domains correspond to key functional units 72. The most abundant structural elements in kinetochore proteins are coiled coils, which are known to function in protein-protein association 73 and act as springs and levers 74. Coiled coils in the budding yeast spindle pole body protein scSpc42 also create a crystalline core involved in spindle pole body duplication 75. Biochemical and electron microscopy experiments have shown that the heptad repeat domains in all four subunits of the S. cerevisiae NDC80 complex associate to form an extended stalk that is linked to two globular heads 3233. Whether the stalk is simply a spacer or some sort of mechanical element remains unexplored. Only detailed structural and biochemical experiments, backed up by analysis in vivo, will reveal the logic of sequence conservation among kinetochore subunits.

A key conclusion in this paper is that four multi-protein linkers that form the core of the S. cerevisiae kinetochore, MIND, the SPC105 complex, the NDC80 complex and COMA are also likely to be present in a wide variety of species (Figure 11). Along with CenH3 and CENP-C, SPC105, MIND and NDC80 complexes are ubiquitous. In budding yeast, linker complexes are thought to form a bridge between proteins in direct contact with DNA and those that bind MTs 51415, and it will be important to show that this is also true in other organisms. Prior to the current work, biochemical experiments had led to the identification of SPC105, MIND and NDC80 complexes in S. pombe, C. elegans and human cells 1819 but our systematic sequence analysis extends these observations to a greater variety of organisms, including E. cuniculi, a microsporidium with a remarkably small proteome. The presence of the structural kinetochore proteins listed above appears to be more fundamental for chromosome segregation than a Mad2-dependent spindle assembly checkpoint, which does not seem to exist in E. cuniculi. Thus, ascertaining the precise molecular functions of the MIND complex, NDC80 complex, Spc105, CenH3 and CENP-C and of the macromolecular assemblies in which they participate is a key task in the study of kinetochore biology.

Budding yeast with CDEI-II-III point CENs contain a set of 11 proteins that are not present in fungi such as S. pombe or C. albicans (Figure 11). Three of the eleven point-CEN specific proteins are involved in sequence-specific binding to CDEIII while six are part of the COMA complex or of a COMA-dependent assembly pathway. Only three of the eleven components of the COMA pathway in S. cerevisiae (Mcm21^Mal2, Chl4^Mis15 and Ctf3^Mis6/CENP-I) are conserved among fungi and mammalian kinetochores (Figure 11). In S. pombe, an alternative set of eight proteins, including spSim4 and spFta1-7, are bound to the COMA components spMcm21^Mal2, spChl4^Mis15 and spMis6^Ctf3. At least three of these proteins (CENP-H^Fta3, Fta1 and Sim4) are members of a class of proteins found in fungal and metazoan organisms with regional CENs whereas the other four proteins have no obvious orthologs (Figure 11a). Overall, these data point to COMA and COMA-associated proteins as kinetochore components with a particularly high degree of sequence divergence through evolution. It seems reasonable to speculate that COMA helps to accommodate kinetochore subunits that are highly conserved among regional and point CENs, such as the NDC80 complex, to diverged components, such as CBF3. By analogy, it seems likely that specialized proteins have evolved to meet the special structural demands of holocentric CENs; ceKNL-3, a kinetochore protein bound to the C. elegans MIND and NDC80 complexes 18 but absent from other kinetochores, may be an early example of a holocentric adaptor.

The MT binding components of kinetochores are unlike kinetochore structural components in that almost all are involved in multiple MT-based processes (Figure 11). In humans for example, EB1^Bim1 and APC^Kar9 are found not only at kinetochores, but also at sites of MT association with the cell cortex; CLIP-170^Bik1 and Dynein play important roles in vesicle trafficking and ch-Tog1^Stu2 is required for spindle assembly. From yeast to humans, only one or two of the six to ten kinetochore MAPs and motors are specific to kinetochores. CENP-A functions in most organisms to determine CEN location without recognizing CEN-specific sequences; similarly, the NDC80-MIND-SPC105-COMA complexes must determine the specialized biochemistry of MT-kinetochore linkages without resort to many kinetochore-specific MAPs.

Database searches were performed on NCBI non-redundant and EST databases using PSI-BLAST and BLAST (protein-protein BLAST (blastp) and genomic BLAST (tblastn)) 76. Pattern searches were performed using ScanProsite 77. Multiple sequence alignments were built with ClustalW, MUSCLE and T-Coffee and edited by hand 787980. Coiled coil predictions were based on the COILS program using a window size of 28 81. Human Nnf1R and Fta1R were identified in PSI-BLAST searches using the full-length S. pombe Fta1 or S. cerevisiae Nnf1 protein sequences as the queries. To identify human Mcm21R, the S. cerevisiae Mcm21 protein sequence was first used as a query in PSI-BLAST searches that yielded fungal Mcm21 related proteins. These proteins were assembled in a multiple sequence alignment from which the motif [HYF]- [KRHDENQ]- [VLI]-x- [HYF]- [ST]- [IVL]- [P]-x-x- [IL]-x- [ILV] was derived, and then used in a pattern search to identify metazoan orthologs. To identify orthologs of S. cerevisiae Nsl1 and Chl4, S. pombe Sim4 or H. sapiens CENP-H conserved blocks were first identified. PSI-BLAST searches were carried out using S. pombe Sim4, S. cerevisiae Nsl1, S. cerevisiae Chl4 or H. sapiens CENP-H as query sequences. This approach identified a set of fungal Sim4, Nsl1 or Chl4 related proteins and a set of metazoan CENP-H related proteins. Each set of proteins was then assembled into multiple sequence alignments and conserved blocks identified (amino acids 1 to 143 for U. maydis Nsl1, 1 to 114 for S. cerevisiae Chl4, 341 to 373 for S. pombe Sim4 and 224 to 269 for H. sapiens CENP-H). The sequences present in these conserved blocks were then used in PSI-BLAST searches to identify new fungal (for CENP-H) or metazoan (for Sim4, Chl4 or Nsl1) proteins.

Phylogenetic alignments were generated with MUSCLE using GBlocks to identify conserved blocks 82. Conserved blocks were selected only if single positions were conserved in at least 50% of the sequences, with higher stringency at flanking positions (80%). A maximum of eight contiguous non-conserved positions were allowed. The minimum block length was five amino acids. Positions with gaps were allowed only if their number did not exceed 50%. Conserved blocks and the number of positions used for each protein family are described in Additional data file 5. To calculate the distances between sequences we took a maximum likelihood approach using TREE-PUZZLE 83 with the 'Pairwise distance calculation only' option, the Jones-Taylor-Thornton substitution matrix 84 and gamma-distributed rates (eight categories) to account for rate heterogeneity (parameters were estimated from the dataset). A neighbor-joining tree was constructed from the distance matrix with the NEIGHBOR program from the PHYLIP package 85. Reliability of the dataset was assessed by bootstrap. We generated 100 permutation datasets using the SEQBOOT program from the PHYLIP package. From these 100 datasets we calculated distance matrices and constructed neighbor-joining trees using the parameters described above. TREE-PUZZLE was then used with the 'Consensus of user defined trees' option to generate a consensus tree from all neighbor-joining trees (nodes with support less than 50% were collapsed) 86. Trees were visualized using the SPLITS TREE tool 87. Amino acid similarity percentages used in multiple sequence alignments are given in Additional data file 4.

The point CEN model was constructed from three different sub-models based on the known structure of point CENs. The first sub-model searched for CDEI-like regions in the query sequence using the [T|G]CA[C|G|T][A|C|G]TG motif. The second sub-model then searched for adjacent CDEII-like AT rich regions. The CDEII region was modeled with a HMM using CDEII from S. cerevisiae 88. For the negative model, S. cerevisiae genomic DNA was used (the effect of including CENs in the genomic DNA was disregarded). For both datasets, base transition frequencies were determined and the transition matrix for the HMM was calculated. The quality of the HMM was evaluated by screening annotated budding yeast genomes and assessment with a bit score:

Given the identification of CDEI and CDEII sequence elements, a third sub-model searched for an adjacent CDEIII motif using an expression based on the highly conserved CCGGAA motif. Positive hits were evaluated with the bit score calculated from the CDEII HMM, length distribution, AT length, AT runs and synteny.