A cohesion interaction network was created by collating information from two proteome databases and the literature (Figure 3). In Figure 3, lines are drawn between proteins to indicate known or potential interactions. The data from which the interactions are derived are indicated in a detailed key that differentiates between the two proteomic databases (and between the different sources of data within each database) and the literature. Four proteins (Esp1, Trf4, Prp11 and Tid3) interact directly with SMC or SCC proteins in S. cerevisiae. The interaction of Esp1 and Scc1 is currently known at a functional level [13], and its importance has already been discussed. This interaction is time-dependent and has not been identified in the yeast two-hybrid screen, and this information is not currently recorded in the YPD.

Trf4 is a protein involved in both mitotic chromosome condensation [38] and sister-chromatid cohesion [39]. In X. laevis Trf4 interacts with Smc1 and Smc2 [38], and in S. cerevisiae Trp4 interacts with Smc1 and Trf5 [40], another member of the TRF family. Trf4 homologs have been identified in S. pombe, C. elegans, Drosophila, human and Arabidopsis (Table 2). Trf4 has very recently been identified as a DNA polymerase with β-polymerase-like properties and is now designated DNA polymerase κ (the fourth class of nuclear DNA polymerases) [41]. Remote homologs of S. cerevisiae Trf4 include the caffeine-induced cell death protein I (Cid1) in S. pombe [42] (13.4% sequence identity) and the polynucleotide adenyltransferase enzyme from a number of organisms including S. pombe and humans (10.2% and 9.7% sequence identity respectively). Cid1 is of particular interest as it thought to play a part in the S-M checkpoint pathway in S. pombe [42]. As a homolog of Trf4, Cid1 could be the link between sister-chromatid cohesion and this checkpoint pathway.

Prp11 is a yeast splicing factor involved in the early stages of the spliceosomal assembly pathway [43]. Prp11 is a 266 amino-acid protein that includes a zinc-finger domain common to RNA-binding proteins [44]. This splicing factor forms a complex with two others, Prp9 and Prp21, which together with Prp5 are required for the binding of U2 snRNP to pre-mRNA [45,46]. There are homologs of this splicing factor in S. pombe, C. elegans, Drosophila, Arabidopsis, mouse and human (Table 2) and all include the RNA-binding motif. In mouse and humans, the homolog is SAP62 (spliceosome-associated protein), a spliceosomal protein that binds to pre-mRNA in the prespliceosomal complex [47].

Tid3 (NCD80) is a spindle pole body protein that has homologs in a number of eukaryotes (Table 2). Tid3 is predicted to interact with Smc1 and Smc2, and has been shown experimentally to interact with Spc24, another component of the spindle pole body. Interactions between the human homolog of Tid3, Hec1, and human Smc1 and Smc2 homologs have also been observed [48]. The interactions of Tid3 with subunits from both the cohesin and condensin macromolecules, places it alongside Trf4 and Scc1, as a protein integrally involved in both mechanisms. It is also proposed that Hec1 may be involved in chromatin assembly in the centromere and regulation of the kinetochore [48]. Spc24, one interaction partner of Tid3, also interacts with Prp11, the yeast splicing factor which is linked to the cohesin loading factors through its interaction with Scc2 (Figure 3).

The upstream regions of the genes encoding 17 proteins in the cohesin network (Figure 3) were searched for shared motifs using AlignACE. Three consensus motifs were identified that were common to subsets of the 17 genes. Only one motif was found to be relatively specific, however, matching upstream sequences of only 29 genes in the SGD [49,50] (see the Materials and methods). This motif has the consensus sequence A6[X]ACGCGTH2[X]RXAAX and includes the MluI cell-cycle box (MCB) element (consensus sequence ACGCGT) [51,52]. The extended consensus motif found in the current work was present in upstream regions of the genes encoding Scc1, Scc3, Smc3, Pds1, Eco1 and Spc24. This motif was located between 123-299 base pairs (bp) upstream of the genes encoding these six proteins. A search of the SGD revealed 23 additional genes containing this upstream motif. Eight of these additional genes encoded hypothetical proteins of unknown function. However, these additional genes also included those encoding chaperones (JEM1 and PDI1n), transcription factor components (TFA1, RFA2, RNA polymerase II, SPT20 and PRT1), and a YC component of the proteasome. When the search was extended to 2,000 bp upstream of the 5' untranslated regions of the yeast genome, the gene encoding Trf4 was also found to contain this consensus motif (1,560 bp upstream).

Teiresias, a pattern discovery algorithm [53,54], was used to search for common motifs between two or more sequences in the 17 proteins of the cohesion network. The highest number of proteins sharing a common motif was three, and these were the three SMC proteins, which have a high sequence identity and share known Prosite motifs (Table 3). More interesting were 24 pattern matches found between pairs of proteins in the network. A number of proteins share more than one sequence motif with the same protein. All shared motifs were either specific to the two proteins in the cohesion network, or in the case of three motifs, shared by one other protein sequence.

In the mouse, the sequence of SMC3 (SMCD) is 100% identical to two proteins with distinctly different functions, bamacan and Mmip1. The possibility that SMCs can 'moonlight' outside the cell in the basement membrane in the guise of the proteoglycan bamacan has recently been proposed [18]. Proteoglycans are glycoproteins present in connective tissues and on the cell surface, and modified proteoglycans are found in human tumors of epithelial origin (for example breast, colon and lung). In the current work we found that Mmip1 was 100% identical to the hinge, second coiled-coil and carboxy-terminal domain of SMCD. It is possible that Mmip1 is a protein resulting from alternative gene splicing or post-transcriptional modification. Mmip1 is thought to compete with the dimerization of Max and Mad family proteins, indirectly upregulating the transcriptional function of c-Myc, which plays a central role in cell proliferation, differentiation and apoptosis [26]. Hence, a single gene appears to encode three products, each with a different function.

The evolutionary tree presented here identifies five SMC families: Smc1, Smc2, Smc3, Smc4 and an ancestral SMC family that includes the SMC sequences from the Eubacteria and the Archaea. However, some eukaryotic SMC sequences, namely Rad18 and its homologs, are also clustered within the ancestral SMC family. Humans, Drosophila, C. elegans, S. cerevisiae and S. pombe have SMCs from all five families. A number of these eukaryotes have more than one sequence clustered in the ancestral family, which leads us to suggest that these proteins might form heterodimeric structures, similar to those observed in other SMC proteins (see, for example [8]). This evolutionary tree is similar in some respects to that recently reported by Cobbe and Heck [18], in that five families are identified. However, in our tree the Rad18 homologs are more closely related to the ancestral eubacterial and archaeal sequences. Rad18 is involved in DNA repair in response to damage caused by UV irradiation. Rad21 in S. pombe (homologous to Scc1 in the cohesin complex of S. cerevisiae) also functions in the repair of DNA damaged by ionizing radiation. From our analysis it is evident that the SCC proteins are not present in eubacteria or archea. Hence, it is possible that the SMCs of the ancestral family cluster with the eukaryotic Rad18 homologs because the eubacterial and archaeal proteins function both as DNA repair proteins and as structural proteins holding sister chromatids together during cell division.

The phylogenetic tree we have constructed is similar to that of Cobbe and Heck [18], but three new important results are revealed in our tree. First, Rad18 homologs cluster with the ancestral SMC proteins, as opposed to clustering as a separate family; second, two SMC homologs are identified in two species of eubacteria, as opposed to only single SMC homologs identified in these species [18]; and third, Mmip1 protein is identified as a new homolog of the SMC mouse protein, a protein not present in the tree previously constructed [18].

The interaction network is based on a combination of data sources, and includes data from two large-scale yeast two-hybrid screens [40]. It is known that this technique can lead to false-positive results [57,58] and false negatives (due to protein instability, for example [59]). However, it is still the only high-throughput technique that can be applied effectively to an entire genome. To screen more than 6,000 S. cerevisiae ORFs, some errors are inevitable and these have to be accepted as part of the data. We have combined yeast two-hybrid data with that available in the YPD and in the literature.

The cohesin macromolecular complex is shown to be at the center of a complex interaction network that connects the mechanism of cohesion to proteins in the condensation and spliceosomal pathways and in the spindle pole body (Figure 3). These interactions also show the potential overlap of proteins involved in sister-chromatid cohesion during mitosis and meiosis. Trf4 and Tid3 are both proteins that interact with Smc1 from the cohesin complex and Smc2 from the condensin complex. Scc2, one of the cohesin loading factors, interacts with Prp11, a splicing factor, which in turn interacts with Spc24, a protein of the spindle pole body that interacts with Tid3. Tid3 makes an important interaction with Dmc1, a meiosis-specific protein, which is required for recombination, synaptomal complex formation and cell-cycle progression [60]. Dmc1 interacts with large number of other proteins, including Apc2 which is a component of the anaphase-promoting complex (APC). The macromolecular APC is a ubiquitin protein ligase that is essential for mitotic cyclin proteolysis and sister chromatid separation [61].

The identification of this extensive network of interacting proteins provided us with a data set of genes and proteins that we could search for common motifs. We identified a large number of protein sequence motifs that were shared by pairs of proteins in the network, and which could potentially represent shared binding sites. The most interesting discovery was the motif shared by Scc2 and Chk1, which comprised part of the serine/threonine protein kinase motif, including the active-site residue. Eukaryotic protein kinases are characterized by 12 subdomains and 12 conserved residues [62]. In a ClustalX alignment of Scc2 and Chk1 from S. cerevisiae, Scc2 (which comprises 1,493 residues compared to the 527 residues of Chk1) does not have the 12 subdomains, but it does have 5 of the 12 conserved residues, only two of which are in the shared catalytic-site motif identified using the Teiresias algorithm. The potential for Scc2 to function as a kinase needs to be validated experimentally. If Scc2 functioned as a kinase it would provide further evidence for the regulation of sister-chromatid cohesion by phosphorylation. It has been already been shown that phosphorylation of Scc1 is a requirement for its subsequent cleavage by the Esp1 separin, which occurs at the metaphase-to-anaphase transition of mitosis [13]. If Scc2 is a kinase, then the cohesin-loading function of the Scc2-Scc4 complex might simply be the phosphorylation of Scc1 or one of the other cohesin subunits to allow DNA binding.

The detection of regulatory motifs in gene promoter regions may provide evidence for the functional assignment of genes of unknown function [63]. We found genes encoding six (or seven if an extreme upstream region is considered) cohesion network proteins sharing a common upstream regulatory sequence that includes the MCB box element. The MCB element is found in upstream regions of genes that encode proteins preferentially synthesized at the G1/S boundary of the cell cycle. MCB or MCB-like elements are common upstream of genes for DNA replication, repair and recombination and of a number of cyclin genes [51]. An upstream DNA motif common to a number of genes could be indicative of DNA-binding preferences of transcription factors and hence coexpression or co-regulation of the genes. It would be possible to investigate coexpression levels experimentally using microarray technology. We found that three of the subunits of the cohesin complex - Scc1, Scc3, Smc3 - shared this common upstream motif, along with Eco1, Pds1 and Spc24 (a protein not directly associated with the cohesin complex). We searched for the expression levels of these six proteins in the data from the microarray hybridization experiment of cell-cycle-regulated genes in S. cerevisiae [64,65]. The data for these six proteins was incomplete, but we found that the mRNA levels of five of the six genes (Scc1, Scc3, Smc3, Pds1 and Eco1) peaked in G1 phase of the cell cycle, and that Spc24 appeared not to be cell-cycle regulated. We also found that the mRNA levels of Smc1 peaked in G1. Spellman [64] observed from the expression data that Smc1, Smc3, Scc1, Pds1 and Pds5 showed tight temporal regulation. This set of proteins includes three of those we found to contain the MCB upstream regulatory region. Further specific microarray data is needed to investigate the expression levels in more detail. However, the coexpression of Pds1 with subunits of the cohesin complex (Scc1, Scc3 and Smc3) would fit with the theory that Pds1 acts a protector of cohesin, inhibiting its breakdown which signifies the onset of anaphase.

The securin Pds1 is an anaphase inhibitor that is complexed to Esp1 for part of the cell cycle. The destruction of Pds1 by the ubiquitin ligase APC is the control point for the separation of sister chromatids [66,67]. APC is activated by Cdc20 during mitosis and the proteolytic function of Cdc20-APC depends on a destruction box motif [54] which has been identified in Pds1. We found three destruction box motifs in Smc3. It is possible that two of the potential destruction box motifs in Smc3 are linked to the periodicity of residues in the coiled-coil domain. But one destruction box motif is in the hinge region, and if this targets the proteins for destruction by APC, then proteolysis of Smc3 could also be connected with the metaphase-to-anaphase transition. APC is activated by Cdh1 during late mitosis/G1 and Cdh1-APC requires a KEN box recognition signal (KENXXXN) to identify its substrates [56]. We identified a KEN box in the sequence of Smc2. If this targets Smc2 as an APC substrate, then proteolysis of this protein could be an event that occurs in late mitosis/G1 of the cell cycle.

A PSI-BLAST [68] search of the nonredundant protein sequence database (GenBank CDS translations, PDB, SWISSPROT, PIR and PRF) for sequence homologs of Smc1, Smc3, Scc1, Scc2, Scc3 and Scc4 from S. cerevisiae was conducted. PSI-BLAST was used for five iterative rounds of searching and a cut-off threshold of 0.001. Homologs from nine eukaryotic species are listed in Table 1. A PSI-BLAST search was also conducted for four proteins (Trf4, Prp11, Tid3 and Esp1) that interact with cohesin proteins (Table 2).

The sequence homologs to Smc3 from S. cerevisiae identified from the PSI-BLAST search were globally aligned in ClustalX (using default parameters). This alignment was used as input for the PHYLIP program [69,70]. PHYLIP (PHYLogeny Inference Package) includes a number of programs for the creation of evolutionary trees. PROTPARS was used to estimate the phylogeny from the ClustalX sequence alignment using the parsimony method. CONSENSE was used to create a consensus tree using the majority-rule consensus tree method, and DRAWTREE was used to plot an unrooted phylogeny (Figure 2).

Each cohesin protein sequence (Smc1, Smc3, Scc1, Scc2, Scc3 and Scc4) and the four interacting proteins (Trf4, Prp11, Tid3 and Esp1) from S. cerevisiae were searched for PROSITE motifs [71] and PFAM domains [72]. The PROSITE database was searched using ProfileScan [73]. PFAM was searched using the protein search facility at the PFAM website [74].

A large-scale yeast two-hybrid screen has been conducted to identify protein-protein interactions in S. cerevisiae [40]. This screen revealed 957 putative interactions involving 1,004 proteins, and the data is publicly available from the CuraGen Corporation [75]. The Yeast Proteome Database (YPD) [76] collates information from the literature, and includes data on protein-protein interactions. Some of these interaction data are derived from the yeast two-hybrid screen. Both the YPD and the yeast two-hybrid data were searched for each of the six cohesin proteins (Smc1, Smc3, Scc1, Scc2, Scc3 and Scc4) to find their interaction partners. This enabled the creation of an interaction network that surrounds the cohesin molecule (Figure 3).

Teiresias, a two-phase general purpose pattern discovery algorithm developed at IBM [53,54], was used to search for common motifs between two or more protein sequences in a network of 17 proteins (Figure 3). The sequence pattern discovery search engine was used for 'exact discovery' with parameter L = 4 (minimum number of literals, that is, non-wild characters in pattern) and W = 6 (maximum extent spanned by any L consecutive literals in the reported pattern). All other parameters were used with default settings. Significance values (calculated by assuming each pattern is used as a predicate to search a database with the size and composition of GenPept) were calculated, and only those with a probability value of = ≤ 30.0 were considered significant.

A search was then made for the presence of the significant motifs in SWISSPROT, TREMBL, and PIR, using the 'findpatterns' facility of GCG (Wisconsin Package Version 10.0). This facility in GCG was also used to search for destruction box (RXXLXXXXN) and KEN box (KENXXXN) motifs in the 17 proteins in the cohesion network.

AlignACE, a Gibbs sampling algorithm for identifying motifs over-represented in DNA sequences [77,78] was used to search for common upstream regions of the genes encoding 17 proteins in the cohesion network. The 'Full Analysis' option (which includes three procedures: Extraction, AlignAce and CompareACE) was selected to search the upstream regions of the selected S. cerevisiae genes. Extraction selects the upstream regions of each gene, AlignACE is the motif finding algorithm, and CompareACE compares the extracted motifs with a set of previously identified motifs from S. cerevisiae. Only motifs with a Map score of > 10.0 were considered significant.

The significant motifs were then used to search the upstream regions of the complete yeast genome using the PatMatch search facility of the SGD [49,50]. PatMatch was used to search 1,000 and 2,000 bp upstream of the 5' untranslated regions, searching both strands, with no mismatches, deletions or insertions.