To find the MATα1 protein sequence and the α-domains target sequences for homology modeling, the full-length MATα1 from Saccharomyces cereviseae was retrieved from the National Center for Biotechnology Information (NCBI) database. This sequence was then used as a template in a BLAST (Basic-Local-Alignment-Search-Tool) search using NCBI-BLASTP suite [32]. The homologous domain between residues 81 and 146 was identified as the α-domain and used. The α-domains of organisms Ogataea angusta (CAE84418), Kluyveromyces lactis (Q08398), Lachancea waltii (CAO02575), Debaryomyces hansenii CBS767 (CAG88405), Zygosaccharomyces rouxii (CAR29078), Candida albicans SC5314 (EAK 95705) were determined by the same procedure and used for the sequence alignment.

To find the optimum HMG-box template for homology modeling, a search of the PDB (Protein Data Bank) was done for all structures containing an HMG-box domain. Initially, 42 PDB records were obtained however we omitted point mutants and highly homologous HMG-box structures to yield 27 representatives for comparisons (see Table A1). The HMG domain sequence files were taken directly from the PDB. Upon analyzing pairwise alignments using DIALIGN-TX [33], we obtained the best templates for homology modeling as PDB-ID: 2E6O (21.1%), 2LEF (14.2%). The final sequence alignment file of the target, template and the atomic coordinate file of the template structure was used to build the model using the SWISS-MODEL workspace. The corresponding model generated by the SWISS-MODEL workspace was subjected to multiple rounds of sidechain and loop adjustments and energy minimization procedures using the SWISS-PDB viewer [34] and the program GROMOS. For models containing the C-terminal domain we took the amino acids 81 - 175 and submitted it to the ROBETTA server [35,36]. After 5 - 6 weeks, the server found 5 possible structures for the domain. Each ROBETTA model and SWISS-MODEL was assessed as described below.

The quality of the generated model was assessed by

Table 1. Structural models of the α-domain from Saccharomyces cereviseae.

checking the stereochemical parameters using PROCHECK [37], and ERRAT [38]. For each model we also checked its absolute quality by analyzing the QMEAN score, QMEAN Z-score, and the GA341 score [39-43]. The QMEAN is a scoring function consisting of a linear combination of six structural descriptors: 1) C-β interaction potential, 2) Solvation potential, 3) All-Atom interaction potential, 4) Tosional potential, 5) secondarystructure matching agreement (SSE-agree) and 6) solvent accessibility agreement (ACC-agree). The QMEAN score is a range from 0 to 1 with the most ideal value should be around 0.6 to 0.8. The QMEAN Z-score provides an estimate of the “degree of nativeness” of the structural model compared to experimental structures. High quality models will have a QMEAN Z-score less than 1 standard deviation from a similar sized high quality experimentally derived protein structure. The |QMEAN Z-scores| between 1 and 3 are acceptable quality and |QMEAN Z-scores| > 3 are considered “bad quality structures”. The GA341 is a score for the reliability of a model [40,41]. A model is predicted to be reliable when the model score is higher than a prespecified cutoff of 0.7. A reliable model has a probability of the correct fold that is larger than 95%.

Superposition were done using the SWISS-PDB viewer [34] and the program SUPERPOSE [44]. RMSD values were computed in the program SUPERPOSE. Models containing bound DNA were prepared by superpositioning the model onto the structure of LEF-1/DNA using the Swiss-PDB viewer and omitting the LEF-1 structure from the picture.

The similarities between the α-domain and the HMG-box have been established based on sequence identity [13]. However, despite extensive structural information on the HMG-box structure, no structure of the α-domain has been determined. Our motivation to do this work was to understand unique evolutionary aspects that exist between MATα1, the α-domain and the HMG-box containing proteins that are involved in mating-type determination. An initial BLAST search using the full-length MATα1, or the homologous α-domain did not reveal any sequence containing a solved structure in the PDB. Further analysis of the full-length MATα1 reveals a naturally disordered region (aa. 1 - 44) that exists on the N-terminus (data not shown). The C-terminus of MATα1 appears to form a structured core of which only residues 81 - 146 have sequence conservation in other yeast and HMG-box proteins. The region from amino acids 147 to 175 had only modest conservation in α-domain proteins of closely related yeasts. Secondary structural analysis of the C-terminal region of MATα1 from S. cereviseae reveals the presence of 5 or more helices (see Figure 3). The fifth helix was predicted with low confidence when analyzed using PsiPRED [45]. Other prediction gave similar results and thus the confidence scores for helix was slightly higher than for β-sheet and random coils. We then sought to compare the conserved region considered as the α-domain to all experimentally determined structures of HMG-box proteins. We were able to find 27 known structures of HMG-box domains that have been deposited into the PDB (see Table A1). All known structures were analyzed by a sequence structural alignment (data not shown). We took HMG-box domain sequences from the Protein Data Bank and α-domain sequences from the SWISSPROT database and performed a sequence alignment (Figures 4(a) and (b)). The sequence alignment was performed by the program MUSCLE and interestingly we found no difference in the alignment among the HMG-box proteins when only a sequencestructural alignment was performed. The critical amino acids within the HMG-box core are listed in Table A2. The α-domain protein from various ascomycete yeast were obtained and aligned along with the HMG-box proteins and colored in CLUSTALX [46] in JALVIEW [47] (see Figure 4 and Table A1). By comparing multiple models of structurally determined HMGbox domains, we were able to look very closely at common core elements among the HMG-box domains and determine how an α-domain would fold assuming the core of MATα1 folded as an HMG-box domain. Taken together, the HMG-box proteins shared very little sequence homology to the α-domain proteins with sequence homologies ranging from 5% to 21% identity (compared to the S. cerevisiae sequence). However, among the conserved sequence elements, the positions within the HMG-box that is required for either DNA binding or for hydrophobic interactions within the core are highly retained. A summary table is provided (see Table A2). Interestingly, residues involved in DNA binding include basic amino acids of the N-terminal loop, a conserved Asn at position 6 between the N-terminal loop and the first helix, and conserved intercalation at position 9 are conserved in all the α-domain proteins shown. The conserved Asn at position 6 is found in all sequence specific HMG-box proteins. The second intercalating residue of the first helix (at position 12) is an Arg in the α-domain which in the position to intercalate between the stacked bases, however, depending on the rotamer form used in the modeling, it can form a hydrogen bond to a nearby pyrimidine. Because Arg is a diversion away from a normal hydrophobic amino acid, this clearly represents a fundamental difference in how the α-domain binds to DNA relative to its close relatives of the HMG-box proteins. Other HMG-box domains have an Arg at this position, however, their structures have yet to be determined.

Despite high conservation of residues on the N-terminal loop and first half of helix 1, the rest of the sequence conservation is low. This exists when compareing all HMG-box proteins as well as α-domain proteins. However a close analysis reveals that amino acids that form hydrophobic contacts in the interior packing between the three helices are conserved. Aromatic residues at positions 8, 11, 42 and 53 are highly conserved in all HMG-box domains and present in all α-domains. Other conserved hydrophobic amino acids that stack between helix 1 and 2 are at positions 23 and 34. A diagram of these amino acids is in Figures 5(d) and (e).

By using comparative homology modeling, we have developed homology models of the α-domain from a representative ascomycete yeast S. cerevisiae (see Table 1). The modeling of the α-domain was performed by two different homology modeling programs: Swiss-Model [34], and ROBETTA [35]. Using the SWISS-MODEL workspace, a manually aligned sequence/template with the α-domain (residues 81 - 146) and the HMG-box protein (PDB ID 2E6O) was used. The resulting model retained many of the features of the template but with few exceptions (see Figures 5(a)-(c)). First, the C-terminal end of helix 1 deviates significantly from the template model. The φ and ψ angles also deviate away from the α-helical structure. Second, helix 3 is only about twothirds the length of helix 3 in the template. The hydrophobic core of the models retained the packing arrangement of the aromatic residues at positions 8, 11, 42, and 53 (see Table A2, Figure 4(a), Figures 5(d) and (e)). When the models produced are superimposed onto LEF-1/DNA, the positions of the critical intercalating Met (position 9) are in very close proximity (see Figure 5(f)). The deviations from the template HMG-box structure is consistent among all known HMG-box proteins used as a comparison here (data not shown). These two deviated positions are also consistent with a previously determined homology model of the α-domain from pezizomycotan yeast using the PHYRE [13].

Using the SWISS-MODEL workspace, we could success-

fully model the conserved region of the α-domain; however, analysis of the C-terminus indicated that this core region of the protein is not likely an independent domain. When modeled independently, several regions, particularly on helix 3, could not be repositioned without high residue error (data not shown). The amino acids at the C-terminus of helix 3 are concentrated with many hydrophobic groups which are not likely found exposed to solvent. We then used ROBETTA [35] to find a better structure that included the region of the C-terminus from amino acids 146 to 175. ROBETTA uses the ROSETTA [36] de novo and comparative modeling methods simultaneously to find full chain structural models. The de novo models are built through fragment insertion and simulated annealing. The whole 94 amino acid C-terminus (aa. 81 - 175) was determined by ROBETTA. ROBETTA returned five possible solutions to the structure of the C-terminal domain (see Table 1 and Figures 6(a)-(e)). Interestingly, couldn’t find any solutions for the 28 amino acid non-homologous C-terminal region (aa. 147 - 175) without the presence of the homologous HMGcore region (aa. 81 - 146) of the α-domain. The solutions to the HMG core that ROBETTA determined compared well with our HMG core using the SWISS-MODEL server and Swiss PDB viewer (see RMSD values in Table 1 and Figure 5(c)). The superposition of the modeled HMG core regions (81 - 146) from ROBETTA with the template 2E6O had a higher RMSD values than the superposition of the HMG core determined using the SWISSMODEL workspace and Swiss-PDB viewer (value of 1.71 Ǻ for Model 1_SWM verses an average of 2.46 Ǻ for Model 1 ROB through Model 5 ROB). The quality of the models was analyzed by the QMEAN, QMEAN Z-score [41-43] and the GA341 score from ModEval [39,40]. The QMEAN, QMEAN Z-score and GA341 are briefly discussed in the methods section. All models were analyzed by other structural analysis servers as well. The stereochemical analysis by PROCHECK [37] can be found in the Table A3. The results of the evaluation show that Model 3_ROB had the best quality scores when you compare the QMEAN score and the GA341 score of 1.000. The QMEAN score for Model 3_ROB is slightly lower than ideal range. We believe that this is due to its small size (94 amino acids) as compared to most of the test proteins analyzed. Analysis by PROCHECK reveals that Model 3_ROB has 85.4% of residues dihedral angles are in the most favored region (see Table A3). This is below the 90% for an ideal protein but much higher than the template used (2E6O) for modeling the HMG core region of Model 3_ROB. We also considered that a good model should be able to bind DNA stably and thus the C-terminus which has very little sequence conservation is likely not to participate in DNA binding. Each model was analyzed as to whether the C-terminus would interfere with DNA binding (see Table 1). The models were superimposed and merged onto the structure of LEF-1/DNA. The LEF-1 structure was then omitted and regions that clashed with DNA were highlighted. We did consider the possibility that minor conformational changes could occur upon DNA binding. Only Model 3_ROB, and Model 4_ROB could bind DNA reversibly without major conformational changes and all the amino acids that participate in DNA binding were available in these models. A picture of Model 3_ROB bound to DNA is in Figure 6(h). The C-terminus of Model 5_ROB does not interfere with DNA binding however; ROBETTA modeled a β-sheet in the N-terminal arm which would interfere with DNA binding.

When the C-terminal region of the α-domain (147 - 175 of S. cerevisiae) is analyzed among the ascomycete yeast species, only little conservation is observed (Figure 4(b)). A fourth helix appears with some conserved residues (aa. 69 - 73) “FVEWL”. A close look at Model 3_ROB reveals some elegant interactions between these conserved residues in the fourth helix. In S. cerevisiae, Phe 69, Trp 72 and Leu 73 of Helix IV form a hydrophobic pocket with Tyr 60 of Helix III (Figures 6(f) and 6(g)). Val 70 interacts with Helix I and the conserved acidic residue Glu 71 makes a contact to the basic N-terminal loop. These interactions are unique to the α-domain.