INTRODUCTION

Top Abstract Introduction Materials and methods Results Discussion References

Homeodomain proteins have been found in a wide range of eukaryotic organisms, spanning the spectrum from yeast to humans. These proteins have in common a conserved 60-residue DNA-binding domain and form a large family of transcription factors that play important roles in cell development (15). Although analysis of homeodomain proteins has shown that many of these proteins bind DNA with relatively low sequence specificity in vitro, they often confer highly specific regulatory activities in vivo (7, 10, 20, 35). One mechanism that homeodomain proteins use to achieve their biological specificity in vivo is through interactions with additional factors (4, 33, 41, 46). These protein-protein interactions function to increase the homeodomain DNA-binding affinity and specificity. One example of this type of interaction involves the 2 protein, which determines cell mating type in the yeast Saccharomyces cerevisiae (22). Although the 2 homeodomain protein binds DNA on its own in vitro, it must interact with one of two other proteins to regulate the cell-type-specific gene expression in vivo. In haploid  cells, 2 protein acts in combination with Mcm1, a MADS box protein, to bind DNA as a heterotetramer and repress transcription of a-specific genes (asg) (23, 31). In diploid a/ cells, 2 protein interacts with a1 protein, another homeodomain protein, to bind DNA as a heterodimer to repress transcription of haploid-specific genes (hsg) (9, 16-18). The interaction of 2 with these cofactors helps increase the affinity and specificity of 2 binding to its target sites (9, 18, 24, 38).

The three-dimensional structures of the 2 homeodomain bound to DNA alone and in complex with a1 and Mcm1 have been determined by X-ray crystallography (27, 40, 45). The 2 homeodomain adopts a fold similar to those of other homeodomains, and many of the DNA contacts are also highly conserved with the structures of other homeodomains bound to DNA (19, 25, 26, 30, 44). Although the DNA contacts made by 2 are almost identical in the three crystal structures, there are some minor differences. One difference among the structures is the position of the N-terminal arm, which makes several additional contacts with the DNA in the a1-2-DNA and 2-Mcm1-DNA ternary complexes that were not apparent in the 2-DNA cocrystal structure (27, 40, 45). There are also several water molecules at the protein-DNA interface in the a1-2-DNA ternary complex that were not visible in the 2 cocrystal structure (27). The a1 homeodomain folds in a conformation similar to the 2 homeodomain and makes an extensive set of base-specific contacts in the major groove of its own half site. The N-terminal arm of the a1 homeodomain is disordered in the crystal structure.

A comparison of the 2 binding sites in both asg and hsg operators yields the same consensus sequence, 5'-CATGTA-3'. These findings suggest that 2 has the same DNA-binding sequence specificities for both sites. However, it has previously been shown that an 2 homeodomain mutant, H3-3A, with alanine substitutions at residues Ser50, Asn51, and Arg54, which make base-specific contacts in the major groove, affects the ability of 2 to bind DNA and repress transcription in complex with Mcm1 but not with a1 (43). It has therefore been proposed that a1 provides the majority of the DNA-binding specificity and affinity for the a1-2 heterodimer and that contributions by the 2 homeodomain in complex with a1 are relaxed in comparison to those when 2 is in complex with Mcm1.

To test this model and to determine the contribution of each homeodomain to the DNA-binding specificity and affinity, we constructed a series of base pair substitutions in the a1-2 DNA-binding site as well as alanine substitutions in the a1 homeodomain. We examined their effects on a1-2-mediated repression in vivo and DNA-binding affinity in vitro. In general, our results correlate well with the structural analysis of the a1-2-DNA complex (27). Interestingly, we show that an 2 mutant, which is lacking all of the base-specific contacts in the major groove, has sequence specificity similar to that of the wild-type protein. This result indicates that the phosphate backbone and minor groove contacts play an important role in sequence-specific recognition by the 2 homeodomain. Finally, we show that a1 contributes to the DNA-binding affinity to the complex, but it appears to have relaxed specificity in comparison with 2.

	MATERIALS AND METHODS

Top Abstract Introduction Materials and methods Results Discussion References

Plasmids and strains. The construction of derivatives of pYJ103, a CYC1-lacZ reporter plasmid containing the different hsg operators, and pAV115, a yeast CEN LEU2 plasmid containing a 4.3-kb MAT locus with the wild-type or mutant 2 gene, has been described (43). Plasmid pYJ195, a P_T7, His-tagged 2 C-terminal expression vector, was constructed by cloning a PCR-generated NdeI-XhoI fragment which contains a sequence encoding six histidine residues followed by 2 residues 123 to 210 into pET21a(+). pAV123, a yeast plasmid similar to pAV115 but containing the MATa locus, was modified to generate plasmid pYJ210. In pYJ210, a silent PstI site was engineered into the a1 gene at codons for homeodomain residues 43 and 44, and the second intron in a1 was deleted. Derivatives of pYJ210 containing mutant a1 genes were constructed by cloning synthetic PstI-XhoI fragments containing the desired mutations into pYJ210. Plasmid pYJ241, a P_T7, His-tagged a1 C-terminal expression vector, was constructed by cloning a PCR-generated NdeI-XhoI fragment that contains DNA encoding a1 residues 66 to 126 followed by six histidine residues into pET21a(+). The haploid MATa and diploid a/ and a/2-H3-3A strains used in the experiments were described previously (43).

-Galactosidase assays. -Galactosidase assays were performed as described by Keleher et al. (23). -Galactosidase activity was measured for three independent transformants for each mutant, and the values were averaged. The standard deviations for all values were less than 10%.

Protein purification. The 2 proteins used in the DNA-binding assays are C-terminal fragments containing the residues 123 to 210 with six histidine residues fused to the N terminus. The 2 proteins were expressed from plasmid pYJ195 in the BL21(DE3) pLysS strain. The a1 protein used in the experiments presented in Fig. 3 and 4 is the full-length protein with six histidine residues fused to the C terminus and expressed from plasmid pYJ173 in the BL21(DE3) strain. The a1 proteins used in the experiments presented in Fig. 5 are C-terminal fragments containing residues 66 to 126 with six histidine residues fused to the C terminus. Both 2 and a1 proteins were purified to greater than 90% homogeneity on nickel resin columns according to the manufacturer's protocol (Novagen).

EMSAs. DNA probes used in the electrophoretic mobility shift assays (EMSAs) were synthesized by PCR as described previously (21). EMSAs were performed in a buffer containing 20 mM Tris (pH 8.0), 0.1 mM EDTA, 5 mM MgCl₂, 10 mg of bovine serum albumin per ml (fraction V), 5% glycerol, 0.1% Nonidet P-40, and 10 µg of sheared salmon sperm DNA per ml. Protein dilutions were made in 50 mM Tris (pH 8.0), 1 mM EDTA, 500 mM NaCl, 10 mM 2-mercaptoethanol, and 10 mg of bovine serum albumin per ml. Five microliters of the 2 dilution and 5 µl of the a1 dilution were added to 40 µl of end-labeled operator fragment diluted in assay buffer, so that the final NaCl concentration was 100 mM. In the protein-free control, 10 µl of protein dilution buffer was added instead of the 2 and a1 proteins. Reaction mixtures were incubated at room temperature for at least 1 h, and then one half of the reaction mixture was loaded onto a 0.5× Tris-borate-EDTA native 6% polyacrylamide gel and electrophoresed at 200 V for 2 h. Dried gels were exposed to phosphor screens, and the images were scanned on a Molecular Dynamics model 425 phosphorimager.

	RESULTS

Top Abstract Introduction Materials and methods Results Discussion References

A consensus a1-2 site mediates repression as well as a wild-type site. We have designed a consensus hsg operator based on the sequence alignment of 17 potential a1-2 binding sites found in the promoters of hsg (6, 12, 14, 28, 29) (Fig. 1A). This site is very similar to the one used in determining the crystal structure of the a1-2-DNA ternary complex and differs from it only at bp 2 and 12, positions in which there are no apparent base-specific contacts in the ternary crystal structure (27). To assay whether the consensus site functions as an a1-2 repressor site in vivo, a reporter plasmid, pYJ103, was constructed by inserting oligonucleotides containing the site between the UAS and TATA sequences of the CYC1-lacZ promoter fusion in pAV73 (42). The presence of an hsg site in this promoter confers repression of lacZ expression that is dependent on both the a1 and 2 proteins (16). pYJ103 and derivatives containing a natural hsg site from the MAT1 promoter and the site used in the ternary crystal complex were individually transformed into an a/ diploid yeast strain and assayed for -galactosidase activity (Fig. 1B). The consensus hsg operator conferred 80-fold repression of lacZ expression, while the natural site found in the MAT1 promoter and the site used in the crystal structure conferred 50-fold and 70-fold repression, respectively. We conclude that the consensus operator functions in vivo at least as well as or better than the natural a1-2 site from the MAT1 promoter. We have therefore used this site as the wild-type standard in examining the binding characteristics of the a1-2 complex.

View larger version (36K): [in this window] [in a new window]   FIG. 1.   A comparison of naturally occurring and synthetic a1-2 binding haploid-specific gene operators. (A) Sequence alignment of 17 naturally occurring a1-2 binding sites located upstream of haploid-specific genes and the synthetic a1-2 binding sites used in the ternary crystal complex (27) and in this study. DNA sequences are written from 5' to 3' (left to right). (B) Comparison of the repression by naturally occurring and synthetic a1-2 binding sites. Transcription reporter constructs that contain a1-2 binding sites in the promoter region of the CYC1-lacZ fusion were transformed into a diploid strain for -galactosidase assays. The fold (×) repression was calculated by comparing the -galactosidase (-Gal.) activities from strains that carry a plasmid containing the a1-2 site with the activity of a plasmid without the a1-2 site. The levels of repression are shown for a natural site found in the MAT1 promoter, the synthetic site used to determine the crystal structure of the ternary complex (27), and a partially symmetric synthetic consensus site that we have used as our standard.

Mutations in the a1-2 site show reduced a1-2-mediated repression in vivo. To determine the contribution of each base pair in the hsg operator to a1-2-mediated repression, sites with single base pair substitutions were cloned into the CYC1-lacZ reporter promoter and assayed for -galactosidase activity in wild-type diploid a/ cells (Fig. 2B). The effects of these substitutions were then compared with the DNA contacts made by 2 or a1 homeodomain observed in the crystal structure of the ternary complex (27). The predicted contacts to the consensus a1-2 site are summarized in Fig. 2A. In general, substitutions that show the largest reduction in repression occur at positions which are specifically contacted by multiple homeodomain residues in the crystal structure. Substitutions at other positions, in which there is only one base-specific contact in the crystal structure, do not have as large an effect on repression. Interestingly, substitutions at positions in which there are no base-specific contacts also have an effect on repression, suggesting that contacts to the phosphate backbone may have a role in sequence-specific recognition. How specific substitutions correlate with the structure of the ternary complex is summarized below.

View larger version (33K): [in this window] [in a new window]   FIG. 2.   Effects of base pair substitutions in a consensus a1-2 DNA-binding site. (A) Predicted base-specific contacts to a consensus a1-2 binding site. The number of each base pair in the site corresponds to the numbering system used in the analysis of the a1-2-DNA ternary complex (27). Predicted base-specific contacts in the major groove made by 2 (left half site) or a1 (right half site) are shown above the DNA sequence. Minor groove contacts made by the N-terminal arm of the 2 homeodomain are shown below the DNA sequence. Hydrogen bonds are indicated by arrows, and van der Waals interactions are indicated by lines that end in circles. Water-mediated contacts are indicated by the letter w in a circle. Positions on the DNA where the proteins make sugar-phosphate backbone contacts are indicated with the letter p. (B) Effects of substitutions in the a1-2 binding site on repression of a heterologous promoter. Values in the table are the repression ratios calculated by comparing the -galactosidase activities in the presence and absence of the a1-2 site.

Mutations in the hsg operator have reduced a1-2 DNA-binding affinity in vitro. The results shown above indicate that substitutions at positions in which there are base-specific contacts as well as sugar-phosphate backbone contacts affect a1-2-mediated repression in vivo. To correlate our data for in vivo repression with the effects of these substitutions on the DNA-binding affinity of the complex in vitro, we assayed the operator mutants for their binding affinity by EMSAs with purified fragments of the a1 and 2 proteins (Fig. 3). In general, the DNA-binding results agree with the in vivo repression data. For example, the T7A and T19G substitutions, which show less than 10% of wild-type repression in vivo, cause a 30- to 50-fold decrease in DNA-binding affinity compared to the wild-type site. Likewise, T5A and C20G, which have moderate effects on repression in vivo, show approximately fivefold decrease in DNA-binding affinity. Substitutions with no effect on repression in vivo (A4G, T15C, and C20T) have essentially wild-type levels of DNA binding. We conclude that the in vitro a1-2 DNA-binding affinities of these hsg operator mutants correlate well with their in vivo repression activities and that the decreases in the level of repression are a direct result of lower binding affinity for the mutant sites.

View larger version (80K): [in this window] [in a new window]   FIG. 3.   EMSAs of mutant a1-2 operator sites. Labeled fragments containing either wild-type or mutant a1-2 sites were assayed for binding in the presence of a constant amount of a1 and dilutions of 2 (residues 123 to 210) ranging by fivefold increments from 3 × 1010 M (lanes 2, 7, 12, 17, 22, 27, 32, and 37) to 2.4 × 1012 M (lanes 5, 10, 15, 20, 25, 30, 35, and 40). Numbers below the substitutions are the fold (×) repression ratios as measured in the -galactosidase assay described in the legend for Fig. 2.

An 2 mutant, lacking base-specific contacts, has sequence specificity similar to that of the wild-type protein. Substitutions at base pairs which are contacted by 2 have large effects on a1-2-mediated repression and DNA binding. Although these results would normally be expected, they stand in contrast to our earlier finding that an 2 homeodomain mutant, called 2:H3-3A, with alanine substitutions of residues Ser50, Asn51, and Arg54, has little or no effect on a1-2-mediated repression and DNA-binding affinity in complex with a1 (43). Since these substitutions effectively remove the side chains that make base-specific contacts in the major groove, this raises the question of whether the H3-3A mutant is able to distinguish between the wild-type and mutant sites.

MAT2

View larger version (42K): [in this window] [in a new window]   FIG. 4.   (A) Effects of substitutions in the a1-2 binding site on repression by 2 homeodomain mutant 2:H3-3A. The values are the fold (×) repression of the reporter promoter measured by -galactosidase assay in the diploid wild-type and a/2:H3-3A strains and were calculated as described in the legend for Fig. 1. (B) EMSA of a1 and 2:H3-3A bound to the a1-2 operator sites. 2 contains residues 123 to 210. Labeled fragments containing either wild-type or mutant a1-2 sites were assayed for binding in the presence of a constant amount of a1 and dilutions of 2 ranging by fivefold increments from 6 × 1011 M (lanes 2, 7, 12, and 17) to 4.8 × 1013 M (lanes 5, 10, 15, and 20).

Mutations in the a1 homeodomain produce reduced repression in vivo. The results shown above indicate that the 2 protein binds in complex with the a1 protein with high specificity and affinity to the site even if it is missing side chains that contact the DNA. This result brings up the question of whether a1-2-mediated repression and DNA binding would also be unaffected by substitutions in the a1 homeodomain. In the crystal structure of the a1-2-DNA ternary complex the a1 homeodomain makes an extensive set of base-specific and sugar-phosphate backbone contacts similar to 2 and other homeodomains (27). Five residues in the a1 homeodomain make base-specific contacts in the major groove of DNA (Fig. 2A). To determine the contribution of a1 to the binding specificity and affinity of the a1-2 heterodimer, we constructed a1 homeodomain mutants with single alanine substitutions and assayed their effects on a1-2-mediated repression in vivo (Table 1). For comparison, we have made similar substitutions at the same positions in the 2 homeodomain and examined their effects on repression as well. As predicted by the crystal structure and in contrast to what is observed for similar mutations in 2, substitutions at the residues in a1 that contact DNA significantly reduce the level of repression. For example, the invariant Asn51 residues of both proteins make nearly identical hydrogen-bond contacts with an adenine in their half sites (Fig. 2A) (27). An alanine substitution of this residue in a1 has only 6% of wild-type activity, while the same substitution in 2 has 80% of wild-type activity. The Arg53 residue is conserved among almost all homeodomains and makes similar phosphate backbone contacts in all homeodomains in which the three-dimensional structures have been determined. Alanine substitution at this residue in a1 has a large effect on the repression, while the same mutation in 2 has virtually no effect on repression. These results show that residues in the a1 homeodomain make a larger contribution to the activity of the complex than residues at the same positions of the 2 homeodomain.

View this table: [in this window] [in a new window]   TABLE 1.   The effects of similar mutations in the a1 and 2 proteins

View larger version (55K): [in this window] [in a new window]   FIG. 5.   Effects of alanine substitutions in the a1 homeodomain in combination with wild-type 2 and 2:H3-3A. (A) Plasmids that express the wild-type a1 protein or one of the indicated mutants were cotransformed with the hsg reporter plasmid into either a wild-type or 2:H3-3A strain. In the absence of a1, the reporter vector produces 270 U of -galactosidase activity. In the presence of wild-type a1 the reporter plasmid produces 8 U, giving 34-fold repression of the promoter by the wild-type protein. EMSAs of the a1 mutants with (B) wild-type 2 protein or (C) the 2:H3-3A mutant are shown. a1 contains residues 66 to 123, and 2 contains residues 123 to 210.

Mutations in the a1 homeodomain produce relaxed DNA-binding specificity. We have shown that the protein produced by the 2:H3-3A mutant has DNA-binding specificity similar to that of the wild-type protein (Fig. 4). This result raises the question of whether the proteins produced by a1 mutants also retain the same DNA-binding specificity as the wild-type protein. To address this question, we assayed for the ability of a1 mutants to repress transcription of reporter constructs with mutant a1-2 binding sites (Table 2). Our results show that alanine substitutions at residues which contact a specific base in the site still retain some sequence preferences at those base pairs. For example, in the crystal structure residue Ile50 of a1 makes a van der Waals contact with T15. An alanine substitution at this residue, which presumably removes this contact, also produces a decrease in repression with the T15G mutant to a level similar to that for the wild type protein. It is likely that this base-pair substitution sterically interferes with binding by the protein. Interestingly, the Ile50Ala and Met54Ala mutants show increased repression at the A16T and C17G sites. The effects of these mutations in the proteins are not the result of a general increase in the DNA-binding affinity since they do not suppress the effects of base-pair substitutions at other positions (T15G and A18G). It is likely that the amino acid substitutions at the smaller side chain remove the steric interference caused by the base-pair substitutions at positions 16 and 17. 

View this table: [in this window] [in a new window]   TABLE 2.   Repression by a1 mutations of transcription of reporter constructs with mutant a1-2 sites

In combination with a1 homeodomain mutants, the 2:H3-3A displays a mutant phenotype. In contrast to the effects observed for 2 homeodomain mutants, similar substitutions in the a1 homeodomain show reduced hsg repression. This result supports a model in which a1 provides the majority of the DNA-binding energy in the a1-2-DNA complex. If this model is correct, then the a1-2-DNA complex may be more sensitive to mutations in the 2 homeodomain if the DNA contacts by a1 are weakened. Under these conditions, substitutions in the 2 homeodomain may have an effect on hsg repression. To test this model, the a1 mutants were cotransformed with the hsg reporter plasmid into the haploid 2:H3-3A strain and assayed for -galactosidase activity (Fig. 5A). In the presence of wild-type a1 the same levels of hsg repression are observed in the wild-type 2 and 2:H3-3A strains. However, in the presence of the a1 mutants, there is a difference between the levels of hsg repression in the two strains. For example, while the a1 Arg53Ala and Arg55Ala mutants show fourfold and sixfold repression in combination with wild-type 2, these same mutants fail to show any repression in the 2:H3-3A strain. The repression activities by a1 mutants Met54Ala and Ile50Ala are also partially reduced in the 2:H3-3A strain.

	DISCUSSION

Top Abstract Introduction Materials and methods Results Discussion References

The yeast 2 homeodomain protein requires either the Mcm1 or a1 protein to bind with high affinity and specificity to its target sites and regulate the expression of different cell-type-specific genes in vivo. The interaction between 2 and its cofactors dictates which sets of target sites are bound and therefore which sets of genes are regulated. How does the interaction with these proteins influence the DNA-binding specificity of 2? In this study we examined the DNA-binding specificity of the 2 protein in complex with a1.

A number of potential a1-2 binding sites have been identified in the promoters of haploid-specific genes, and many show strong sequence similarity at particular positions in the sites (Fig. 1A). The conclusions from our mutagenesis data about the relative importance of each position in the site correlate well with the sequence conservation among the natural sites. In general, positions which are not conserved among the natural sites, such as positions 11, 13, and 14, can accommodate substitutions without large changes in the level of repression. Positions that we have shown are critical for a1-2 binding and repression, such as bp 6, 7, and 19, are almost invariant among the known or predicted natural binding sites. The sequence conservation at other positions, however, is not as easy to explain. Among the predicted natural sites, positions 16, 17, 20, and 21 are strongly conserved, but we have shown that many of the substitutions at these positions have little or no effect on repression or DNA binding. One possible explanation for this result is that we have made these substitutions in the context of a strong binding site. It is possible that in the context of weaker binding sites, these positions may make a larger relative contribution to a1-2 binding and would therefore be conserved.

Only a few of the natural sites have been shown by genetic and biochemical studies to be functional a1-2 repressor sites (6, 16, 28). The other sites were identified based on their homology with the known sites and therefore may not be functional repressor sites or may only function weakly. For example, the HO(1) and HO(8) sites vary at several positions that we have shown are important for a1-2 binding and repression. We have tested these sites for repression in the context of the heterologous CYC1-lacZ reporter promoter and found that they only weakly (two- to threefold) repress the promoter (data not shown). Although these sites function only weakly on their own, they may work synergistically with other sites in the promoter to increase the local concentration of the repressor complex at the HO promoter to repress transcription.

The effects of mutations in the 2-Mcm1 binding site on repression and DNA binding have also been examined (39, 47). A comparison of substitutions in the 2-Mcm1 and a1-2 sites shows that many of the mutations have the same effects on both sites. This result suggests that the binding specificity of 2 in complex with Mcm1 is similar to that in complex with a1. However, we did observe several significant differences between the sites. Substitutions at positions T10 and T11 in the a1-2 site have relatively little or no effect on repression or DNA binding by the a1-2 complex. In contrast, substitutions at the analogous positions in the 2-Mcm1 binding site have a large effect (less than 5% activity) on 2-Mcm1-mediated repression in vivo (47). One possible explanation for this difference is that in the 2 and 2-Mcm1 structures the Arg7 side chain makes a contact to the base on the top strand at position 10, while in the a1-2 complex it makes a contact to the base on the bottom strand (27, 40, 45). These different contacts may have different sequence preferences. However, in the 2-Mcm1 complex these positions are also contacted by the Mcm1 protein and we have shown that substitutions of these bases have a large effect on DNA binding and transcription regulation of the Mcm1 protein on its own (1). It is most likely that this difference in the specificity of DNA binding to the two sites is primarily due to binding by Mcm1 and not to differences in recognition by 2.

A second difference between the two sites is the effects of substitutions at position A8. A substitution of T at this position in the a1-2 site produces almost wild-type activity (75%), while the analogous substitution in the 2-Mcm1 site produces only 14% of the wild-type activity (39, 47). In the a1-2-DNA crystal structure, there are base-specific contacts in the minor groove at this position by residues Arg4 and Gly5 in the N-terminal arm of 2 (27). In contrast, only Gly5 makes a base-specific contact at this position in the 2-Mcm1-DNA structure (40). Although one would assume that there are fewer sequence specific requirements at this position in the 2-Mcm1 site than in the a1-2 site, we found the opposite. It is not clear from the analysis of the structures why there is this difference between the sites. Although there are these subtle differences, our results show that 2 appears to bind to its half site with similar specificities in complex with Mcm1 and in complex with a1.

The analysis of a1-2 binding to the hsg operator suggests two general properties for homeodomain DNA recognition. First, the base-pair specificity of DNA recognition seems to extend beyond the positions that are contacted by homeodomain residues in the crystal structure. Substitutions at almost every position in the operator have at least a moderate effect on repression and DNA-binding affinity. Although there are not direct contacts to the bases at some positions, there are contacts to the sugar-phosphate backbone. One possible explanation for these results is that these backbone contacts may also be sequence dependent. Substitutions at these positions may affect the precise configuration of the backbone atoms or the overall DNA structure and therefore have an effect on the DNA-binding affinity of the complex. Although substitutions at these positions have a relatively small effect on the DNA-binding activity in vitro, they have a significant effect on repression in vivo. These results may suggest why sites for many homeodomain proteins from higher eukaryotes that appear to have similar DNA-binding affinities in vitro may have different activities in vivo (5, 8, 11, 32, 35).

Secondly, in general, homeodomain proteins seem to have relatively low sequence specificity and display a significant degree of tolerance for different DNA sequences. Our results show that the a1 and 2 homeodomains are similar in this regard, since even substitutions at positions in which there are base-specific contacts (4, 8, 10, 20) have only moderate effects on DNA binding and repression. One possible explanation for this observation is that the proteins may be able to form favorable contacts with the substituted base pairs, and therefore changes at those positions do not dramatically affect the binding affinity. The low sequence specificity may be advantageous for the function of homeodomain proteins. Since homeodomain proteins are often involved in many different combinatorial regulatory circuits, the relaxed DNA specificity may allow these proteins to interact with a number of other cofactors to bind different DNA sites and regulate the expression of different genes (22).

The results from the mutational analysis of the a1 homeodomain also correlate well with the crystal structure of the a1-2-DNA ternary complex. Mutations at residues that make hydrogen-bond contacts with the DNA have a stronger effect on repression and DNA binding than mutations at the residues that make van der Waals contacts. These results were expected since van der Waals interactions are long range and are likely to contribute less energy than hydrogen bonds. Compared with the effects of the same mutations in the 2 homeodomain, mutations in a1 have a larger effect on repression. For example, the Asn51Ala and Arg53Ala substitutions in 2 do not appear to have an effect on a1-2-mediated repression, while the same mutations in the a1 homeodomain show only about 6% activity of the wild-type protein. These results indicate that despite their similar structures and many similar contacts to the DNA, the two homeodomain proteins do not make equivalent contributions to the DNA-binding affinity of the a1-2 heterodimer.

The result that a1 homeodomain mutants show stronger effects in the 2:H3-3A strain than in the wild-type 2 strain also supports the concept that in the a1-2-DNA complex, a1 provides a large portion of the DNA-binding energy. Although the 2:H3-3A mutant shows the wild-type level of repression in complex with wild-type a1, when in complex with a1 mutants, the 2:H3-3A mutant displays a more pronounced mutant phenotype. This result suggests that residues in 2 which make base-specific contacts do make a contribution to the DNA-binding affinity of the complex but that it is relatively minor compared to the contribution by the analogous side chains in a1.

Our result for the 2:H3-3A mutant created a paradox, i.e., mutations at side chains that make contacts in the major groove of the DNA have little effect on repression and DNA binding, whereas substitutions of the base pairs that are contacted by these residues have a strong effect on repression and DNA-binding affinity. One explanation for this discrepancy is that in the a1-2-DNA complex, these side chains may not make significant contributions to the binding affinity. Although the substitutions with alanine at these positions remove the 2 base-specific contacts in the major groove, the sequence specificity and binding energy provided by the sugar-phosphate backbone contacts and the base-specific contacts by the N-terminal arm in the minor groove may be sufficient for 2 to bind DNA with a1 at close to the wild-type level. A similar phenomenon has been observed for the GCN4 bZIP protein, in which alanine substitutions at residues that contact DNA in the cocrystal structure do not affect the protein's function in vivo (13, 34). However, our finding that base pair substitutions have a strong effect on repression and DNA-binding affinity is rather surprising. There are two possible explanations. First, although the wild type 2 protein may not require these specific contacts to bind DNA, the Ser50, Asn51, and Arg54 side chains may fit into the DNA only when certain base pairs are at these positions. Changes of these base pairs might cause steric interference with the 2 residues which would result in the reduced binding affinity and repression. If this model is correct, one would then expect that mutant proteins, in which the side chains that cause steric interference are effectively removed by the alanine substitutions, would bind with wild-type affinity to the mutant operator sites. We have shown that proteins containing alanine substituted at a1 residues involved in base-specific contacts bind with relaxed specificity to mutant DNA sites. However, our result that the 2:H3-3A protein is able to discriminate among the mutant operators to the same degree as the wild-type protein argues against this possibility for 2. It is likely that the contacts by the N-terminal arm contribute to some of the sequence specificity of binding by the 2:H3-3A mutant. Another explanation is that changes of these base pairs may have altered the DNA conformation and therefore affected other protein-DNA contacts. For example, it is possible that many of the protein contacts to the sugar-phosphate backbone are sequence specific. Therefore, substitutions in the DNA will not only interfere with contacts to the bases but also alter the positions of the atoms in the backbone and thus weaken the protein-DNA contacts. In this case, the crystal structure of the a1-2:H3-3A dimer bound to the hsg operators is needed to solve the paradox.