We have used an in vitro DNA binding site selection procedure to define the binding site preferences for six of the seven mouse DMRT proteins. The main conclusions of the work presented here are that the DMRT proteins select similar DNA binding consensus sites, which resemble that of DSX, and that some DMRT proteins can heterodimerize on DNA. The sites selected by the different proteins were not identical, however, and we observed qualitative and quantitative differences in DNA binding behavior in gel mobility shift assays. These differences may have implications for DNA binding in vivo, as discussed below.
The DM domain is bipartite, comprised of an amino-terminal zinc-binding module and a more variable carboxy-terminal domain that has been proposed to act as a recognition helix, contributing to DNA binding affinity and specificity [23
]. Mutagenesis studies have identified a number of residues in both portions of the DSX DM domain that are critical for binding to its consensus site [38
]. The DMRT proteins have recognition helices that differ from that of DSX at three of these critical positions. Despite these differences, however, most of the DMRT proteins select a site similar to that of DSX, and indeed several can efficiently bind a DSX site in a gel shift assay (data not shown). The similarity in DNA binding is surprising, given that all of the DMRT proteins have an amino acid change that eliminates efficient DNA binding by DSX (D78E in DSX) [38
]. The ability of the DSX and DMRT DM domains, despite their different recognition helices, to bind similar sequences underscores the need for a DM domain/DNA co-structure to help define how specific DM domain residues contribute to DNA affinity and to DNA sequence specificity. It also will be interesting to determine the preferred sites of more highly divergent DM domains such as those found in C. elegans
and see whether they also bind sites similar to that of DSX.
The sites selected by the DMRT proteins were very similar and most of the DMRT proteins were able to shift a DMRT1 site in the gel shift assay. However there were distinct sequence preferences at some positions. We tested the significance of the differences in sites selected by DMRT1 and DMRT5, and confirmed that several differentially selected nucleotides do have differential effects on binding by the two proteins in a gel shift assay (Figures and ). Based on these comparisons, it therefore appears that there are distinct binding preferences among the DMRT proteins, at least in vitro. The extent to which such differences in sequence preference also exist in vivo will require the future identification and characterization of DMRT target genes.
As described earlier, we were able to select bound DNA in solution using all of the DMRT proteins except DMRT6, but there were differences in the behavior of several proteins in the gel mobility shift assay that may be significant. DMRT2 and DMRT7 selected sites resembling those of the other DMRT proteins and DSX, indicating similar specificity, but exhibited less efficient DNA binding in the gel mobility shift assay, suggesting lower affinity for DNA. Although similar amounts of each protein were tested, we cannot exclude the alternative possibility that some of the proteins folded inefficiently and thus were less active. In contrast to the other proteins tested, DMRT4 formed a protein/DNA complex on a DMRT1 site that was excluded from the gel. There are several possible explanations for these differences in DNA binding behavior, which are not mutually exclusive. First, some of the proteins may form structurally distinct protein/DNA complexes, and this may be reflected in their behavior in the gel shift assay. For example, DMRT4 may form higher-order multimers on DNA. Second, efficient binding of proteins like DMRT2 and DMRT7 may require covalent modifications that were missing from the in vitro translated proteins. Third, some of the proteins may normally bind DNA in conjunction with one or more partner proteins (DM domain or otherwise) that were missing from the in vitro assay. A possible example is the efficient binding of DMRT3 as a heterodimer with DMRT1 (see below).
The DMRT protein binding sites, like that of DSX, comprise a punctuated palindrome flanking a single A/T base pair, and we found that DMRT1 and DMRT5 bind to this site as dimers. The site is not fully symmetrical, however, and there was stronger selection of DNA sequence on one side of most sites. Extensive sequence substitutions indicated that the side with the stronger selection is required for efficient DNA binding of both monomers and dimers, whereas the other side is primarily needed to allow dimer binding. Our data suggest a model in which the left side of the binding site is recognized first by a DMRT monomer. Once the first DMRT molecule has bound to the DNA, it presents a surface that in combination with the right side of the DNA site creates a high affinity binding site, which, allows a second DMRT molecule to efficiently bind. This model is supported by the base substitution experiments in Figure .
We found that DMRT1 can heterodimerize on DNA with DMRT3 and DMRT5, raising the possibility of combinatorial gene regulation by DMRT proteins. It is particularly striking that DMRT3 preferentially binds as a heterodimer in vitro. Comprehensive expression analysis has not been reported for most Dmrt genes, but it is clear that at least some overlapping expression occurs. For example, Dmrt1 and Dmrt3 are coexpressed in spermatogonia and Sertoli cells in the adult testis (unpublished data and the Mammalian Reproductive Genetics Database). Because the DMRT proteins vary in sequence preference outside the core consensus (Figure ), heterodimer formation has the potential to significantly increase selectivity for specific target sites. In addition, because the carboxy termini of the DMRT proteins are highly variable, it is possible that heterodimers can interact with a different spectrum of transcriptional regulatory partner proteins than do homodimers. It will be of interest in the future to better define the overlap of DM domain protein expression in mammals and other animals and to test the effect of heterodimerization on DNA binding specificity.
In contrast with the other DMRT proteins, DMRT6 did not select binding sites, and we were unable to detect binding of DMRT6 to a DMRT1 site (not shown). The other DMRT proteins are identical across 14 positions in the recognition helix (13 in DMRT7), whereas DMRT6 differs at six of these 14 positions. One of these variant residues is an arginine in the other DMRT proteins but a threonine in DMRT6, and is the site of a known intersex mutation in DSX (R91Q). Additionally, and perhaps more significantly, within the zinc binding module, a conserved lysine residue that in DSX has been proposed to form an important salt bridge to the DNA backbone (K60), instead is substituted with an alanine in DMRT6. We therefore suggest that DMRT6 may not directly bind DNA. We cannot, however, exclude the possibility that our in vitro conditions are incompatible with DNA binding by DMRT6.
An important question is how closely the DNA binding sites we have determined in vitro resemble those in vivo. Particularly for the proteins that did not exhibit efficient binding as homodimers, the sites occupied in vivo may only partially resemble those determined in vitro. Ultimately this question must be answered by identifying the in vivo targets of the DMRT proteins. In flies and worms DM domain proteins have been shown to use in vivo sites that closely mirror their in vitro binding preferences: DSX regulates transcription of the yolk protein gene Yp1
in flies and MAB-3 regulates transcription of the yolk protein gene vit-2
and the antineural bHLH gene ref-1
in worms and this requires sites closely matching those preferred in vitro [26
]. These results suggest that at least some of the DMRT proteins are likely to regulate their in vivo targets via sites similar to the ones we have defined.