DNA-independent and DNA-dependent mechanisms regulate the differential heterodimerization of the isoforms of the thyroid hormone receptor with retinoid X receptor.

Thyroid hormone receptors (TRs) require heterodimerization with retinoid X receptor (RXR) for maximum DNA binding affinity. Interaction with RXR occurs via two dimerization interfaces, one in the DNA-binding domain and one in the C-terminal “ninth heptad” of the receptors. We studied the relative importance of these two dimerization domains in naturally occurring C-terminal TR variants. TRα1 has a conserved ninth heptad and formed stable heterodimers with RXR in solution. TRα1·RXR heterodimers bound similarly to direct repeat 4 (DR4) sites with different 5′-flanking and spacer sequences. In contrast, TRα2, which contains a highly divergent ninth heptad, did not interact with RXR in solution and bound as a heterodimer with RXR only to specific DR4 sequences in which the downstream half-site was the preferred octameric binding site of TR (TNAGGTCA). Although the ninth heptad of TRα2 was insufficient for interaction with RXR off DNA, this region was required for DNA-dependent heterodimerization with RXR. TRα3, another naturally occurring TRα isoform whose ninth heptad differs from those of both TRα1 and TRα2, displayed intermediate behavior in heterodimerization with RXR. Thus, in the absence of a strong ninth heptad interaction an octameric downstream half-site allosterically promotes RXR heterodimerization with TRα2. Differential dependence upon DNA-binding for heterodimerization with RXR may influence transcriptional regulation by TRα isoforms.

RXR⅐RAR heterodimer, ligand binding by RXR has a lesser effect on transcription potential (8,9). In all cases, however, a major role of heterodimerization with RXR is to regulate the affinity and specificity of DNA binding (2).
RXR as well as the nuclear receptors with which it heterodimerizes have inherent affinity for the hexameric sequence AGGTCA. This sequence is recognized by amino acids collectively referred to as the P box within the classical DNA-binding domain (DBD) (10 -12). Both N- (13,14) and C-terminal (15) regions of the nuclear receptors have additional, important effects upon half-site affinity and specificity. In particular, amino acids immediately adjacent to the carboxyl end of the DBD also play a role in DNA recognition. These T and A boxes (16), also called the C-terminal extension of the DBD (17), increase specificity by making minor groove contacts with DNA sequences 5Ј to the hexameric core. This is particularly important for orphan receptors that have high affinity for monomeric sites, such as NGFI-B (16), SF1 (18), Rev-Erb (19) and ROR (20). It is also important for monomeric binding by TR, which greatly prefers the dinucleotide TA preceding the AGGTCA motif (21).
RXR heterodimers have selectively increased affinity for direct repeats of the AGGTCA motif (DRs), such that RXR⅐TR heterodimers have the highest affinity for DRs spaced by 4 base pairs (DR4), while RXR⅐PPAR, RXR⅐VDR, and RXR⅐RAR prefer DRs spaced by 1, 3, and 5 base pairs, respectively (22). Thus, heterodimerization with RXR is one of the most important determinants of the target gene specificity of receptor action. Furthermore, since DRs are inherently asymmetric, RXR heterodimers bind with nonrandom polarity. In the case of TR and RAR, the RXR binds to the upstream half-site, while the receptor partner binds preferentially downstream (23)(24)(25)(26)(27). At least two domains within the nuclear receptors contribute to heterodimerization with RXR. One domain is located within the C terminus of the receptors, in a region that has been referred to as the ninth heptad (28 -30), and is now recognized as helix 11 in the structure of the TR ligand-binding domain (31,32). This domain within TR and RAR is sufficient for RXR interaction irrespective of DNA binding. A second domain is located within the DBD of the receptors (25,27,33). This DBD interaction is relatively weak, but it serves an important function as the molecular determinant of the spacer requirements for selective binding to direct repeats.
TRs are responsible for tissue-specific and developmentally regulated gene expression in response to thyroid hormone (T3). There are two TR genes, ␣ and ␤, with multiple isoforms (34). The TR␣ gene encodes TR␣1, which is a bona fide TR, as well as TR␣2 and TR␣3, which are identical to TR␣1 for 370 amino acids, including the DBD, before diverging within the sequence of heptad 9 (34 -36). The 40 C-terminal amino acids of TR␣1 that are lacking in TR␣2 and TR␣3 include 1) the C-terminal component of heptad 9, which has been shown to be essential for strong interaction with RXR and is highly conserved among RXR-heterodimerizing receptors, 2) amino acids necessary for T3 binding, and 3) an autonomous activation domain at the very C terminus that is highly conserved in the receptor superfamily and mediates the ligand-dependent transcriptional activation and interaction with putative coactivator proteins (37). In place of this important region of TR␣1, TR␣2 contains 122 additional amino acids with a unique sequence. TR␣3 is the result of splicing to an internal acceptor and contains the last 83 amino acids of the TR␣2 fused to amino acid 370 (Fig. 1). As a result of their lack of the 40 C-terminal amino acids of TR␣1, both TR␣2 and TR␣3 do not bind T3 or activate transcription (38).
The variant isoforms TR␣2 and TR␣3 are highly abundant in brain and developmentally regulated (39) and function as inhibitors of T3 action in vitro (40). TR␣2 has been more carefully studied, and its dominant negative activity appears to be at least in part due to competition with TRs for binding sites in target genes (41,42). Although other mechanisms may contribute to the inhibitory effects of TR␣2 (43), point mutations in the DBD that abolish DNA binding by TR␣2 abrogate its dominant negative properties. 2,3 DNA binding by TR␣2 differs from that of TR␣1 in some important respects. First, phosphorylation of the C terminus of TR␣2 reduces its monomeric binding affinity (15). Also, TR␣1 heterodimerizes with RXR on a variety of binding sites, whereas the ability to demonstrate TR␣2 heterodimerization with RXR has been more variable. TR␣2 and RXR do not heterodimerize on inverted repeats of AGGTCA, but they have been reported to heterodimerize on some DR4 sites but not others (15,30,43,44).
We previously hypothesized that the differential heterodimerization of TR␣1 and TR␣2 is related to their different ninth heptad sequences ( Fig. 1) (44). We have now studied RXR heterodimerization of TR␣1 and TR␣2 in the context of different DR4-binding sites. TR␣2, in contrast to TR␣1, formed stable heterodimers with RXR on only a subset of DR4 sequences. TR␣2-RXR heterodimerization on DR4 sequences required a downstream half-site that is an optimized TR-binding site (TNAGGTCA). Although the DNA-binding site determined whether TR␣2 would heterodimerize with RXR, implicating the heterodimerization domain within the DBD, mutation analysis revealed that even in the case of TR␣2 the ninth heptad was required for selective heterodimerization. Interestingly, TR␣3 heterodimerized weakly on even nonoptimal DR4s, suggesting that its ninth heptad heterodimerization domain was intermediate in strength between that of TR␣1 and TR␣2. Together, the results demonstrate the roles of the DBD and ninth heptad in heterodimerization of TR␣ isoforms and indicate that RXR heterodimerization and DNA binding may be regulated both by strong, DNA-independent protein-protein interactions and weaker protein-protein interactions that are allosterically favored by specific DNA-binding sites.

MATERIALS AND METHODS
Plasmids-The rat TR␣1 and TR␣2 cDNAs in pBluescript have been previously described (35). cDNA corresponding to the C terminus of rat TR␣3 was cloned by reverse transcription-polymerase chain reaction of rat GH3 cell RNA using the following primers: 5Ј-ACATTGGCCAGT-CACC-3Ј and 5Ј-CAAATAACAAGGGAGCTTGG-3Ј.
The polymerase chain reaction product was cut with EcoNI and ligated into the EcoNI site of the TR␣2 cDNA to produce full-length TR␣3 cDNA, which was completely sequenced. 4 The nonphosphorylatable form of TR␣2 (TR␣2-SA) in pBluescript has been described previously (15). For creation of C-terminal deleted proteins, TR␣1 and TR␣2 were subcloned into the EcoRI and the XhoI/BamHI site of pCMX, respectively. The cDNAs were then digested with a convenient C-terminal restriction enzyme (as indicated below) and NheI, which cuts immediately upstream of multiple stop codons of pCMX. The vector was then blunt-ended using Klenow enzyme and ligated to produce the following truncated proteins: TR␣2⌬405 (cut with NsiI), TR␣2⌬387 (cut with Xma3), TR␣⌬347 (cut with BspMI), and TR␣⌬250 (cut with EcoNI). HA epitope-tagged-TR␣1 was created by inserting the EcoRI fragment of TR␣1 into the MunI site of a T7 promoter-driven expression vector (pGEM Z) encoding the HA epitope (45). HA⅐TR␣2 was made by subcloning the unique C-terminal fragment from TR␣2 pBluescript into the AccI and BamHI of HA⅐TR␣1 ⅐ pGEM Z. RXR␣ in pBluescript was kindly provided by R. Evans. Gal4 DBD (amino acids 1-147) and the Gal4⅐TR␣1-(121-410) in pECE have been previously described (41). To construct the Gal4⅐TR␣2-(121-492), the following polymerase chain reaction primers were made to isolate TR␣2 in pBluescript from amino acids 121-492 (missing the DNA-binding domain): 5Ј-AAAGGATC-CAAGCCATGGACCTGGTTCTAGA-3Ј and 5Ј-GTAAAACGACGGCC-AGT-3Ј.
Following the polymerase chain reaction, the TR␣2 fragment was cloned in frame into the BamHI site of pECE containing the Gal4 DBD-(1-147). VP16-RXR fusion protein was kindly provided by K. Chatterjee. The (Gal4 UAS ϫ 5)-SV40-luciferase reporter contained five copies of the Gal4 17-mer binding site and has been previously described (46). Proteins were translated in vitro and labeled with [ 35 S]methionine, using the T3 polymerase TNT reticulocyte lysate transcription and translation system (Promega).
Electrophoretic Mobility Shift Assay (EMSA)-EMSA was performed by incubating 5 l of each protein at room temperature for 20 min in a 30-l binding reaction mixture containing 10 mM HEPES, pH 7.9, 80 mM KCl, 5% glycerol, 30 ng of poly(dI-dC), 75 ng of denatured salmon sperm DNA, 10 mM dithiothreitol, 30 ng of bromphenol blue, and 100,000 cpm of 32 P-labeled double-stranded DNA probe. Following incubation, reaction mixtures were loaded on a 5% polyacrylamide gel, dried, and subjected to autoradiography. For off-rate experiments, 5% gels were prerun for 2 h prior to loading. Proteins were preincubated for 1 h with labeled probe at room temperature, and then the first lane was loaded into a running gel. A 500-fold molar excess of unlabeled probe was added to the remaining reaction mix, and equal aliquots were loaded onto the remaining lanes at times indicated in the appropriate figure. For competition experiments, proteins were preincubated for 1 h at room temperature with unlabeled probe at specified -fold excess. Labeled probe was then added and incubated for 20 min at room temperature.
Complementary single-stranded oligonucleotides containing a single copy of a TR-binding site were annealed and labeled with [␣-32 P]dCTP and used as probes. The sequences of probes used are as follows (hexameric half-sites underlined): Hex-Hex, CCAGGTCAAAGGAGGTCA; Hex-Oct, CCAGGTCAAATAAGGTCA; Oct-Hex, TAAGGTCAAAAAAG-GTCA; Oct-Oct, TAAGGTCAAATAAGGTCA. Also, Oct-Oct probes with different spacers were made by varying the third base pair of the spacer.
Cell Culture and Transfection-293T cells were maintained in Dulbecco's modified Eagle's medium supplemented with 10% bovine calf serum and switched to Dulbecco's modified Eagle's medium plus anion exchange resin and charcoal-treated serum 12 h prior to transfection. 60-mm dishes were transfected by the calcium phosphate precipitation method using 1-2 g of receptor expression vectors, 1 g of luciferase reporter, and 0.5 mg of ␤-galactosidase expression vector. Cells transfected were lysed in Triton X-100, and cell lysates were subject to luciferase and ␤-galactosidase assays as described previously (46). Results were expressed as -fold activation relative to the control level and normalized to ␤-galactosidase activity, which served as internal control for transfection efficiency.
Coimmunoprecipitation-In vitro translated proteins were incubated in 75 l of buffer H (20 mM HEPES, pH 7.9, 50 mM KCl, 2 mM EDTA, 0.1% Nonidet P-40, 10% glycerol, 0.5% nonfat dry milk, 5 mM dithiothreitol) in the absence or presence of DNA (2.5 pmol of probe to 45 fmol of protein) for 20 min at room temperature. Samples were then precleared with 50 l of protein A-Sepharose twice in buffer H for 15 min at room temperature on a rotator. Precleared supernatants were then incubated with 50 l of protein A-Sepharose prebound to 4 l of HA antibody for 1 h on a rotator. Reactions were then microcentrifuged for 10 s and then washed four times with ice-cold buffer, boiled in 1 ϫ SDS loading buffer, and separated by electrophoresis on a 10% polyacrylamide gel. Gels were fixed, incubated with En 3 Hance, dried, and subjected to autoradiography.

TR␣1 but Not TR␣2 Interacts with RXR in Solution-
We first studied the ability of TR␣1 and TR␣2 to heterodimerize with RXR in the absence of TR-binding sites. We have previously shown biochemically that Gal4⅐TR␣1 could interact with RXR in a DNA-independent manner (41). This was confirmed using a mammalian two-hybrid assay. All experiments were performed in the absence of T3. Fig. 2A shows that a chimeric protein consisting of the Gal4 DBD fused to the C terminus (amino acids 120 -410) of TR␣1 did not activate transcription from a reporter gene containing five Gal4-binding sites under these conditions (lane 2). Co-transfection of VP16⅐RXR activated Gal4⅐TR␣1 (lane 3), indicating an in vivo interaction between TR␣1 and RXR. Thus, the C terminus of TR␣1 was sufficient for this interaction, since in this fusion protein the TR␣1 DBD is replaced by the Gal4 DBD. In contrast VP16⅐RXR was unable to stimulate transcription by Gal4⅐TR␣2-(120 -492) ( Fig. 2A, lane 7), indicating that TR␣2 and RXR did not interact in vivo. In addition, cotransfection of wild-type TR␣1 was able to inhibit activation of Gal4⅐TR␣1 by RXR⅐VP16 (lane 4), probably by competition for RXR; wild-type TR␣2 did not inhibit the interaction of Gal4⅐TR␣1 and RXR⅐VP16 (lane 5), again consistent with its inability to heterodimerize with RXR. The inherent ability of full-length TR␣1, but not full-length TR␣2, to heterodimerize with RXR was confirmed biochemically using a coimmunoprecipitation assay. In vitro translated 35 S-labeled RXR was incubated with HA epitope-tagged TR␣1 or TR␣2 prior to immunoprecipitation with HA antibody. Fig. 2B shows that the labeled RXR coprecipitated with HA⅐TR␣1 in the presence of HA antibody, whereas HA⅐TR␣2 did not (compare lanes 3 and 5). These results showed that TR␣1 but not TR␣2 interacts with RXR in solution due to the strong heterodimerization domain in the C terminus of TR␣1.
TR␣2 and RXR Heterodimerize on DR4 Sequences That Contain an Optimal TR Monomer-binding Site as the Downstream Half-site-The above results were independent of DNA binding by TR⅐RXR heterodimers. Since TR␣2 has been shown to heterodimerize with RXR on some but not all TR-binding sites (15,42), we systematically studied the role of preferred TR monomer-binding sites as the upstream and downstream half-sites of DR4 elements. For these experiments we created DR4 sites with hexameric half-sites (CCAGGTCA, referred to as the "Hex" sequence) or optimal TR monomeric half-sites (TAAG-GTCA (21), referred to as the "Oct" sequence) as one or both half-sites (referred to as Hex-Hex, Hex-Oct, Oct-Hex, and Oct-Oct). Fig. 3 shows that TR␣1 bound weakly as a monomer to all four sites and bound much more strongly as a heterodimer with RXR to all four of the DR4s with little dependence upon the sequences flanking the AGGTCA hexamer in either of the half-sites. TR␣2 also bound as a monomer to all four of the DR4s, even more weakly than TR␣1 monomers, in keeping with our earlier observations that TR␣2 is phosphorylated in reticulocyte lysate and that phosphorylated TR␣2 has less DNA binding affinity than nonphosphorylated TR␣2 (15). RXR alone displayed virtually no binding to these sites.
The striking result of the experiments shown in Fig. 3 was that TR␣2 and RXR heterodimerized on a specific subset of the DR4s. The TR␣2⅐RXR heterodimer bound strongly to the DR4 in which both half-sites were optimized for TR monomer binding (Oct-Oct) but did not bind to the DR4 in which both halfsites were nonoptimal (Hex-Hex). We hypothesized that since TR binds to the downstream half-site of the DR4 (24,25,27,33), it might be the downstream octamer half-site that was of most importance. Indeed, converting the downstream half-site of Oct-Oct to the hexameric sequence (Oct-Hex) abolished TR␣2⅐RXR heterodimer binding. In contrast, the TR␣2⅐RXR heterodimer bound strongly to the Hex-Oct site, in which only the downstream half-site is optimized for TR. The TR⅐RXR interaction was clearly cooperative, since TR␣2 itself had much lower affinity for either half-site, and RXR alone did not bind at all. Thus, the ability of TR␣2 to cooperatively bind DNA as heterodimer with RXR was allosterically regulated by the DNA-binding site. The simple rule governing this interaction was the requirement for an optimal TR monomer site as the downstream half-site in the context of the DR4.
DNA-dependent Interaction of TR␣2 and RXR-The role of the DNA-binding site was confirmed in the coimmunoprecipitation assay. Fig. 4 shows that, as shown earlier, RXR coimmunoprecipitated with HA⅐TR␣1 in the absence of DNA, whereas it did not coprecipitate with HA⅐TR␣2 under the same conditions. TR␣1-RXR was mildly stimulated by both the nonoptimized DR4 (Hex-Hex) as well as a DR4 with both half-sites optimized for TR binding (Oct-Oct; compare lanes 5 and 7 with lane 3), consistent with previous reports (47). The effect of including DNA in the coimmunoprecipitation was much more striking for TR␣2-RXR heterodimerization. Inclusion of the Oct-Oct site increased the affinity of TR␣2 for RXR such that under these conditions RXR was efficiently coimmunoprecipitated with HA⅐TR␣2, almost to the level of TR␣1 (compare lanes 7 and 13). In contrast, the Hex-Hex oligonucleotide had little effect on the interaction between HA⅐TR␣2 and RXR, consistent with the DNA binding studies shown earlier.  5 shows that a T as the third nucleotide in the spacer (i.e. 2 base pairs 5Ј to the downstream AGGTCA) was sufficient to promote heterodimerization between TR␣2 and RXR. Although any nucleotide immediately adjacent to the downstream halfsite (i.e. the fourth nucleotide in the spacer) was permissive for heterodimerization, Fig. 5 also shows that a G in this position resulted in the strongest binding of the TR␣2⅐RXR. In contrast, monomer binding by TR␣1 or TR␣2 and heterodimer binding by TR␣1⅐RXR were not nearly as favored by the TG dinucleotide (e.g. compare TR⅐RXR heterodimer binding in lanes 2 and 4, or TR␣2 monomer binding in lanes 9 and 11). Thus, the presence of a G adjacent to the AGGTCA specifically favored TR␣2⅐RXR heterodimer formation.
DNA-dependent RXR Heterodimerization of Nonphosphorylated TR␣2-Phosphorylation of serines 474 and 475 in the C terminus of TR␣2 greatly reduces its monomeric DNA binding affinity (15). We tested the role of phosphorylation in the mechanism of TR␣2⅐RXR heterodimerization by comparing wild type TR␣2 with a nonphosphorylatable form of TR␣2 (TR␣2-SA). Fig. 6, panel A (as well as panel B and Fig. 7) shows that the binding of TR␣2-SA⅐RXR heterodimers to permissive DNAbinding sites was greater than that of the wild type TR␣2 heterodimers. Two potential explanations for this difference in binding were considered: 1) RXR heterodimerized with phosphorylated TR␣2 had reduced affinity for the Oct-Oct site or 2) RXR heterodimerized with phosphorylated TR␣2 had very little affinity for the Oct-Oct site, and the binding seen was due to a fraction of nonphosphorylated TR␣2 in the in vitro translated preparation. To distinguish between these two possibilities, off-rates were determined for the wild type and TR␣2-SA preparations bound to the Oct-Oct site with RXR. As shown in Fig.  6B, the half-lives were nearly the same in each case, although the initial binding was again greater for the TR␣2-SA. These results suggest that phosphorylated TR␣2 bound very weakly as a heterodimer with RXR even to the Oct-Oct site, and the observed heterodimers actually represented a fraction of the in vitro translated preparation that was nonphosphorylated. This is consistent with our observations of variable degrees of DNA binding by TR␣2 (compared with TR␣2-SA) related to different batches of reticulocyte lysate (data not shown). In any case, Fig.  6C shows that, as shown earlier for wild type TR␣2, nonphosphorylated TR␣2-SA only bound cooperatively as a heterodimer with RXR to the DR4s with the optimized TR monomer sequence as the downstream site, indicating that phosphoryla- tion is not the key determinant of the DNA site specificity of TR␣2 heterodimerization with RXR.
The Ninth Heptad of TR␣2 Is Required for DNA-dependent Heterodimerization with RXR-The results thus far indicate that TR␣1 interacts with RXR in solution, and this preformed heterodimer binds similarly to the Hex-Hex and Oct-Oct sequences. In contrast, TR␣2 and RXR only heterodimerized on DNA but did so cooperatively in a manner that was highly dependent on the composition of the spacer between the two half-sites. These data were consistent with a major role of the strong ninth heptad dimerization domain in TR␣1, with the weaker domain located in the DBD of both TR␣1 and TR␣2 being more important in TR␣2. In order to assess whether there was any role of the ninth heptad of TR␣2, we created a series of C-terminal deletions of TR␣2 (Fig. 7A) and assayed their ability to heterodimerize on the Oct-Oct binding site. Fig.  7B confirms that TR␣2-SA⅐RXR heterodimers bound more strongly than TR␣2⅐RXR heterodimers (compare lanes 3 and 5). C-terminal deletion to amino acid 405, which removes the phosphoserines that reduce the DNA binding affinity of TR␣2, restored the binding of the TR␣2⅐RXR heterodimer to the level seen with TR␣2-SA (compare lanes 5 and 7). Continued Cterminal deletion to amino acid 387, which still contained the ninth heptad region of TR␣2, did not affect the DNA-dependent heterodimerization on this site (compare lane 9 with lane 5). Note, however, that further deletion into heptad 9 abolished RXR heterodimerization. Parenthetically, deletion to amino acid 347 enhanced TR␣ homodimer formation (compare lane 12 with other even-numbered lanes in which RXR is not present), consistent with the work of others (48) and suggesting an inhibitory role of the C terminus on homodimerization. Nevertheless, both of the C-terminal deletion mutants that lacked heptad 9 (⌬347 and ⌬250) did not heterodimerize with RXR. Thus, the DBD dimerization domain alone was insufficient for even the DNA-dependent RXR heterodimerization of TR␣2 (or TR␣1, since all of the C-terminal deletions past amino acid 370 are identical in TR␣1 and TR␣2). It is noteworthy in this context that the identification of the DBD dimerization interface has only been demonstrated, both functionally and structurally, at high concentrations of TR and RXR DNA-binding domains. The lack of the C terminus would obliterate potential cis-acting effects of the C terminus on monomer (e.g. by phosphorylation (15)) and dimer (lane 12, and see Ref. 48) binding. The present data suggest that for full-length receptors at lower (and closer to physiological) protein concentrations, the unique ninth heptad of TR␣2 creates an increased dependence upon the DNA-dependent heterodimerization domain in the DBD but still is required for heterodimerization between TR␣2 and full-length RXR.
The DNA Dependence of RXR Heterodimerization with TR␣3 Is Intermediate between That of TR␣1 and TR␣2-Although less well studied, TR␣3 is also a developmentally regulated C-terminal splice product of the TR␣ gene (36,39,49). Given the above results suggesting a less dominant but definite role for the ninth heptad of TR␣2 in permitting the DNA-dependent heterodimerization with RXR, we were interested in comparing the DNA binding and heterodimerization of TR␣3, which has yet a third ninth heptad sequence. Fig. 8 shows the DNAdependence of RXR heterodimerization of TR␣3. Like both TR␣1 and TR␣2, TR␣3 bound strongly with RXR to DR4 sites containing optimal TR monomer-binding sites as downstream half-sites. However, TR␣3 weakly heterodimerized with RXR on DR4s that lacked optimal TR half-sites downstream. Thus, the characteristics of RXR heterodimerization by TR␣3 were intermediate between those of TR␣1 and TR␣2 in that it was stimulated but not absolutely dependent upon optimization of the spacer.

DISCUSSION
Heterodimerization with RXR is a major determinant of DNA binding specificity for TR as well as receptors for retinoids, vitamin D, and eicosanoids. In the simplest model, RXR heterodimerization targets TR to DR4 sites, RAR to DR2 and DR5 sites, VDR to DR3 sites, and PPAR to DR1 (2). However, there are clearly additional layers of complexity in heterodimer binding to DRs. For example, the 5Ј-flanking sequence of the upstream half-site in DR1s is critical for PPAR⅐RXR binding (50). The preferred flanking sequence converts the half-site into a Rev-Erb monomer-binding site (19), consistent with the similarities in the T and A boxes of PPAR and Rev-Erb. We have now shown that a similar preference rule relating to the 5Ј-flanking sequence of the downstream half-site governs TR␣2 heterodimerization with RXR.
Our understanding of the mechanisms regulating heterodimerization with RXR comes from biochemical and structural studies. Gel shift analyses using full-length receptors have implicated the ninth heptad of the receptors as being necessary for RXR interaction (29,51,52). Evidence for this domain is based on mutational analysis, and the regions of direct protein-protein contact are not definitive because the C termini of the receptors have not been co-crystallized with RXR. In contrast, co-crystallization of the TR and RXR DBDs on a DR4 element revealed the importance of the DBD dimerization interfaces (27). This interaction domain can be demonstrated in gel shift assays at high protein concentration (25,33) but is weak and generally insufficient for heterodimerization in the absence of the ninth heptad in the context of the full-length receptors (44,51).
Against this background, the comparison of RXR heterodimerization with TR␣1, TR␣2, and TR␣3 is revealing for a variety of reasons. First, these are naturally occurring forms of TR␣, which are developmentally and tissue-specifically expressed. Second, these proteins have identical DBDs and divergent ninth heptads. The ninth heptad of TR␣1 is highly similar to those of TR␤ as well as RAR, VDR, and PPAR (28), whereas the C-terminal portion of the ninth heptad regions of TR␣2 and TR␣3 are unique. Thus, our results are of biological as well as mechanistic importance. It is not clear why TR␣3 is intermediate in its RXR heterodimerization, although the ninth heptad sequence comparison shown in Fig. 1 reveals that the major difference between TR␣2 and TR␣3 is that TR␣3 contains an arginine at position 375 that is conserved in TR␣1 and other RXR-heterodimerizing receptors but is a glutamine in TR␣2 (53).
The ability of TR␣1 to heterodimerize with RXR in solution in the absence of DNA is likely to require the ninth heptad, since TR␣2 does not form heterodimers under the same conditions. Furthermore, the C termini alone were sufficient for interaction in vivo, indicating that the DBD dimerization interface is not required for the heterodimerization between TR␣1 and RXR. The existence of preformed TR␣1⅐RXR heterodimers in solution probably explains in part the lack of strong requirement for specific 5Ј-flanking or spacer sequences. In addition, TR␣1⅐RXR heterodimers bind relatively stably to other DR sites such as DR5 (54), as well as to non-DR sites such as inverted and everted repeats (42,52,(55)(56)(57).
In contrast to the DNA-independent heterodimerization of TR␣1 and RXR, TR␣2 heterodimerization with RXR was not demonstrable in the absence of DNA and in fact only occurred on a specific subset of DR4 sequences. Specifically, the presence of a T two base pairs upstream from the AGGTCA (i.e. in the third position of the DR4 spacer) was required for TR␣2 heterodimerization with RXR. Significantly, this creates a high affinity TR monomer binding sequence at the downstream site (21). TR has been shown to bind to the downstream half-site in the context of binding with RXR to a DR4 (24,27), and this polarity of binding probably explains why the same flanking sequences in the context of the upstream half-site were ineffective at binding TR␣2⅐RXR.
The heterodimeric DNA binding of TR␣2 and RXR was clearly cooperative because TR␣2 bound weakly as a monomer, and monomeric RXR binding was undetectable on the same sites under identical conditions. The receptors must also directly interact, because no heterodimers were observed when the positions of the Hex and Oct half-sites were simply reversed. This indicates that the specific DNA-binding site allosterically favored the interaction of TR␣2 and RXR. Since the TR monomer site was necessary for heterodimerization, we suggest that DNA binding altered the conformation of the TR DBD interaction domain to favor interaction with RXR bound to an upstream half-site. It is also possible that the conformations of both TR and RXR are altered by binding to the specific DR4. Such an allosteric change in receptor structure upon DNA binding would be consistent with the nonvectorial changes in the TR⅐RXR⅐DR4 complex that we observed in circular permutation analysis yet could not be attributed to DNA bending (54). The ability of DNA to allosterically regulate the conformation of the TR would be formally analagous to the ability of T3 ligand to allosterically alter TR conformation (9). DNA binding has also been shown to allosterically regulate interaction with transcriptional cofactors (58). Indeed, DBD-dependent proteinprotein interactions have already been documented in the crystal structure of the TR⅐RXR bound to DR4 (27). The present results suggest that in the absence of a strong C-terminal interaction, the inter-DBD interactions are regulated in part by the sequence bound by the TR. We propose a model in which TR␣1 and RXR exist as a heterodimer in solution due to their strong C-terminal interaction. The DBD interaction determines the spacing (DR4) and polarity (TR downstream) requirements of binding but does not depend upon specific DNA sequence of the spacer. In contrast, TR␣2 and RXR do not form a stable heterodimer in solution due to the altered ninth heptad of TR␣ and thus do not bind as heterodimers to all DR4 sites. Since RXR has little monomeric DNA binding affinity (59), we speculate that the first step in TR␣2⅐RXR heterodimerization on selected binding sites is binding of TR␣2 to the TNAGGTCA downstream site. The affinity of this interaction is reduced by phosphorylation of TR␣2. We suggest that binding of TR␣2 causes an allosteric change in conformation, which promotes the interaction with RXR, which binds weakly to the upstream site and stabilizes the TR␣2⅐RXR⅐DNA complex. The interaction between TR␣2 and RXR requires both the DBD and C-terminal interactions as well as a permissive downstream half-site. Such differential heterodimerization may allow TR␣2 to inhibit T3 action at selected target genes. Furthermore, allosteric regulation of protein conformation by specific DNA-binding sites may be a general mechanism of regulating DNA binding specificity and gene regulation of nuclear hormone receptors. In addition to regulating heterodimerization with RXR, DNA-dependent conformational changes in receptor structure may also regulate interactions with other proteins. For example, DNA-dependent interactions with transcriptional coactivators such as SRC-1 (60), RIP-140 (61), Trip1 (62), or TIF-1 (63) could explain why receptors such as TR activate transcription only on a subset of high affinity binding sites (64).