Plasticity of tetramer formation by retinoid X receptors. An alternative paradigm for DNA recognition.

Retinoid X receptors (RXRs) are transcription factors that traditionally have been thought to bind DNA as protein dimers. Recently, however, it has been recognized that RXRs can also bind to DNA as protein tetramers. Receptor tetramers form cooperatively on response elements containing suitably reiterated half-sites, and play an important role in determining the specificity of DNA recognition by different nuclear receptors. We report here that RXR tetramers exhibit significant functional plasticity, and form on response elements possessing diverse half-site orientations and spacings. This ability of RXRs to form tetramers and related oligomers appears to contribute to the synergistic transcriptional activation observed when multiple, spatially separated response elements are introduced into a single promoter. Oligomerization may therefore be a common paradigm for DNA recognition and combinatorial regulation by several different classes of transcription factors.

Retinoid X receptors (RXRs) are transcription factors that traditionally have been thought to bind DNA as protein dimers. Recently, however, it has been recognized that RXRs can also bind to DNA as protein tetramers. Receptor tetramers form cooperatively on response elements containing suitably reiterated half-sites, and play an important role in determining the specificity of DNA recognition by different nuclear receptors. We report here that RXR tetramers exhibit significant functional plasticity, and form on response elements possessing diverse half-site orientations and spacings. This ability of RXRs to form tetramers and related oligomers appears to contribute to the synergistic transcriptional activation observed when multiple, spatially separated response elements are introduced into a single promoter. Oligomerization may therefore be a common paradigm for DNA recognition and combinatorial regulation by several different classes of transcription factors.
It is now clear, however, that receptor dimers are not the only paradigm for DNA recognition, and that certain members of the nuclear receptor family can also bind to DNA as protein complexes larger than dimer in size (27). These high order oligomers include receptor trimers, tetramers, and pentamers, and bind with high cooperativity to response elements containing suitably reiterated half-sites (27). RXR tetramers have also been observed in solution (28 -30). Oligomer formation by RXRs permits recognition of DNA sequences that are not recognized by receptor dimers, and contributes to isoform-specific promoter utilization (27).
We wished to investigate the nature of these newly elucidated receptor oligomers. Here, we report that RXR tetramer formation is relatively plastic in character, and that tetramers form on response elements possessing a wide variety of halfsite orientations and spacings. In fact, receptor oligomers can form cooperatively on spatially separated response elements and may contribute to the synergistic gene activation observed when multiple response elements are introduced into a single promoter. Intriguingly, the half-site spacing and orientation required for tetramer formation are non-equivalent at different positions in the response element, suggesting a corresponding anisotropism in the RXR tetramer. High order oligomer formation may be a common means of generating DNA specificity and combinatorial regulation for a variety of transcription factors.

EXPERIMENTAL PROCEDURES
Proteins and Oligonucleotide Probes-The wild-type, ⌬N terminus, and the chimeric mouse RXR proteins were obtained as nuclear extracts from Sf9 cells infected with the appropriate recombinant baculovirus (27). The construction of the ⌬N terminus RXR␤, ⌬N terminus RXR␥, and the ␥⅐␤ RXR chimera was described previously (27). The TR⅐RXR chimera represents a fusion of codons 1-159 of the avian thyroid hormone receptor ␣-1 sequence to codons 228 -464 of mouse RXR␥. The ⌬N⌬C RXR␥, representing codons 139 -228, was isolated as glutathione S-transferase fusion protein from Escherichia coli bearing a suitable pGEX plasmid vector (31). No significant functional differences were detected when the same receptor was isolated from the two different expression systems (27). Oligonucleotides were obtained as complementary, single-stranded DNAs (Operon Incorporated) and were annealed to create double-stranded DNAs with 4-base overhangs. For use as probes in electrophoretic mobility shift experiments, the overhangs were filled in with radiolabeled nucleotides and Klenow fragment of DNA polymerase I (31,32). For use as response elements, the doublestranded DNAs were introduced into the SalI site of pD33-CAT (31). An adjacent, cryptic half-site present in the pD33-CAT vector itself (AG-GTCG) was removed from these constructs by PstI and HindIII cleavage, converting the overhanging ends to blunt with T4 DNA polymerase, and religation.
Transient Transfections-Drosophila SL-2 cells were maintained at room temperature in Schneider's medium supplemented with 14% fetal bovine serum. Transient transfections were performed using a calcium phosphate co-precipitation protocol and typically employing 1 g of pD33-CAT reporter, 0.15 g of pHSP82-lacZ (used as an internal standard), and 0.1 g of either pA 5 C-RXR␤ or pA 5 C-RXR␥ expression plasmid/35-mm plate (31). After incubation in the absence or presence of 100 nM 9-cis-retinoic acid (Ligand Research), the transfected cells were harvested, and the chloroamphenicol acetyltransferase activity was determined relative to that of the ␤-galactosidase control (31).

RESULTS
RXR␣ and ␥ Bind Cooperatively to Reiterated Response Elements as Protein Tetramers-Both RXR␣ and ␥ can efficiently bind to reiterated elements containing four half-sites oriented as direct repeats ( Fig. 2A). These protein-DNA complexes migrate at a position characteristic of 4-fold occupancy (4R), and form with extremely high cooperativity and no evidence of prior dimer formation (27). We have therefore defined these RXR␣ and RXR␥ complexes as protein tetramers (27). In contrast, RXR␤ primarily forms dimers (2R) and neither binds these reiterated elements as tetramers in vitro nor efficiently medi-ates gene activation through them in vivo ( Fig. 2A and Ref. 27). It should be noted that at high protein concentrations, RXR␤ can occupy all four half-sites on a reiterated element by the non-cooperative binding of two independent RXR␤ dimers ( Fig.  2A); this is, however, in clear contrast to the highly cooperative binding of tetramers observed with RXR␣ and ␥ ( Fig. 2A).
The various complexes observed in Fig. 2A are mediated by the corresponding RXR isoforms, as demonstrated by the ability of RXR-directed antisera to supershift the complex, and by the absence of complex formation by equivalent protein preparations isolated from uninfected Sf9 cells, or from cells in- with the locations of the receptor domains thought to be involved in DNA recognition, hormone binding, receptor-receptor dimerization, and transcription (Tx) regulation indicated above. The locations of the Nand C-terminal deletions, and the junctions used to create the RXR ␥⅐␤ chimera and the TR⅐RXR chimera, are shown below. The core sequences of the different response elements employed in these studies are also presented (panel B); flanking sequences employed to standardize the length of the oligonucleotides and to permit cloning and radiolabeling are not shown. Arrows indicate the locations and orientation of the AGGTCA half-sites, which are numbered as 1, 2, 3, and 4 for convenience. The locations at which 2-, 5-, or 10-base spacers were inserted, as described under "Results," are indicated above the top sequence.
FIG. 2. Tetramer formation by RXR␣ and RXR␥ on reiterated DNA elements. The ability of wild-type RXR␣, ␤, or ␥ to bind to DNA was tested (panel A), using a reiterated response element composed of four half-sites oriented as direct repeats with a 1-base spacer element (top, see Fig. 1B). Increasing amounts of receptor (estimated at 0, 1, 3, or 5 ng, respectively) were added to a fixed amount of radiolabeled response element, and the resulting complexes were resolved by electrophoretic shift assay and autoradiography. The position of protein-DNA complexes representing 2-fold occupancy (2R) or 4-fold occupancy (4R) of the response element by receptor are indicated on the right, along with the location of free probe. The identity of the RXR⅐DNA complexes were confirmed by supershift experiments, comparing the mobilities of the complexes formed in the absence of specific antisera (panel B, left two lanes) to those of the complexes formed in the presence of anti-RXR antisera (panel B, right two lanes). Combinatorial experiments were also performed to confirm our assignment of stoichiometry to these complexes, utilizing an N-terminal deletion (⌬N␥) of RXR␥, full-length RXR␥ (␥), or mixtures of the two (panels C and D). DNA elements containing two (panel C) or four (panel D) half-sites were employed, using the same form of electrophoretic analysis as in panel A. The proposed combinatorial products, expressed as mixtures of fulllength (R) and deleted (r) RXR molecules, are indicated on the right of panels C and D. Small quantities of an RXR␥ degradation product could be detected as a minor band flanking the major complex in the right lane of panel C. The free probe was electrophoresed off the gel in panel D to improve the resolution of the different complexes. fected by non-recombinant baculovirus (Fig. 2, A and B). 2 Furthermore, analogous complexes were formed by purified glutathione S-transferase-RXR␥ fusion proteins isolated from E. coli, suggesting that RXR is likely the only protein necessary for complex formation (27).
Our assignment of stoichiometry to the various complexes in Fig. 2A was based initially on titration experiments with elements containing differing numbers of half-sites; a stepwise decrease in the mobility of the complex occurred corresponding to the number of half-sites in the element (27). A more absolute assignment of stoichiometry was obtained by a combinatorial experiment. We compared the migration pattern of DNA complexes generated by an N-terminal deletion of RXR␥ (which produces a rapidly migrating DNA-protein complex) with that of the full-length RXR␥ (which produces a slower migrating complex) (Fig. 2, C and D). Mixing these two different receptor derivatives on a two-half site element produced a single intermediate complex, the combinatorial result predicted if this element is indeed bound by receptor dimers (i.e. RR, Rr, rr). In contrast, mixing these same two receptor derivatives with our four half-site element produced three intermediate complexes ( Fig. 2D) precisely the number predicted for combinatorial mixtures of a receptor tetramer (i.e. RRRR, RRRr, RRrr, Rrrr, rrrr). We therefore conclude that the RXR␣ and RXR␥ complexes observed on DNA elements containing four half-sites do indeed represent 4-fold occupancy by receptor tetramers.
The RXR Hormone-binding Domain Is Necessary, but Not Sufficient, to Confer Tetramer Formation-What RXR domains are involved in tetramer formation? We have previously shown that the N terminus of RXR confers the differing oligomerization properties of the ␣, ␤, and ␥ isoforms (27). Exchanging the N termini of the RXR ␤ and ␥ isoforms exchanges the oligomerization phenotype (Fig. 3A). However, the N terminus plays a negative role in oligomerization, not a positive one; deletion of the N terminus does not affect oligomer formation by RXR␥, and actually enhances oligomerization by RXR␤ (⌬N-␤ and ⌬N-␥; Fig. 3A). Thus, the actual oligomerization interface(s) must lie outside the N-terminal domain. Indeed, deletion of both the N terminus and the C-terminal hormone-binding domain disrupted cooperative tetramer formation without abolishing dimerization (⌬N⌬C␥; Fig. 3A). We conclude that the zinc-finger domain is sufficient for dimer formation, but that additional, C-terminal sequences are necessary for tetramer formation.
We next tested if the oligomerization properties associated with the C terminus of RXR were transferable to another nuclear receptor. Thyroid hormone receptor-␣-1 (TR␣-1) binds DNA as a dimer, but does not appear to form tetramers on 4-fold reiterated TR response elements (Refs. 1-7, and data not shown). We therefore replaced the C-terminal hormone-binding domain of TR␣-1 with that of RXR␥ (Fig. 1). Although this chimera readily bound to a reiterated TR response element as a homodimer, it failed to form the cooperative tetramers characteristic of the RXR␥ parent (Fig. 3B). It was conceivable that the TR N terminus might disrupt tetramer formation by the TR⅐RXR␥ chimeras, analogous to the negative effects of the N terminus of RXR␤. However, deletion of the N terminus failed to confer tetramer formation on either the TR⅐RXR chimera, or on the native TR ( Fig. 3B and data not shown). We conclude that the C terminus of RXR is necessary for oligomerization, but is not sufficient under the conditions employed here.

A Wide Variety of Reiterated Elements Can Accommodate Tetramer Formation by RXR␥, but the Individual Receptor
Molecules in the Tetramer Are Non-equivalent-Nuclear hormone receptor dimers can recognize a variety of half-site orientations and spacings (15)(16)(17)(23)(24)(25). RXR homodimers, for example, bind efficiently to direct, convergent, and divergent repeats of half-sites spaced by 1, 0, or 2 bases, respectively. To dissect the topological constraints operative on the RXR tetramer, we first explored the effects of varying the spacing of the half-sites within the 4-fold reiterated element. Introduction of a 2-or 10-base spacer at the center of the element (i.e. between half-sites 2 and 3) had little or no detectable effect on tetramer formation by RXR␥ (Fig. 4A), indicating that the oligomer could readily accommodate changes in spacing at this location. Formation of tetramers by RXR␥ was slightly destabilized, though not abolished, by introduction of 5 bases at the same location (Fig. 4A) presumably reflecting the non-integral rotation about the DNA imposed by this spacer. None of these changes in spacing conferred tetramer binding on the RXR␤ isoform (Fig. 4A).
In contrast to its relative lack of effect at the center of the element, the 10-base insertion abolished tetramer formation by RXR␥ when introduced between half-sites 1 and 2, or between half-sites 3 and 4 (Fig. 4B). Instead of tetramers, RXR␥ bound to these elements relatively weakly, and as a broad complex migrating in a position characteristic of 3-fold occupancy (denoted 3R; Fig. 4B). We attribute this broadening or smearing phenomenon to an instability of this trimeric complex during the electrophoretic separation, a feature that was observed with several other non-optimal elements. Apparently the 10base insertion at these flanking sites interferes with recruitment of a fourth receptor molecule to the DNA, resulting in formation of a receptor trimer. We conclude that the receptor interface between half-sites 2 and 3 is relatively insensitive to spacing, whereas the receptor interfaces between half-sites 1 and 2, or 3 and 4, are spacing-dependent.
We next varied the relative orientations of the half-sites in the reiterated element. Direct (ϾϾϾϾ), convergent (ϾϾϽϽ), and divergent (ϽϽϾϾ or ϾϽϾϽ) orientations between the central half-sites 2 and 3 were all compatible with tetramer formation ( Figs. 2A and 5). Similarly, both direct (ϾϾϾϾ) and convergent (ϾϽϾϽ) repeat orientations were acceptable between half-sites 1 and 2 or between 3 and 4 ( Figs. 2A and 5). However, inversion of just a single half-site at position 3 (ϾϾϽϾ) strongly inhibited tetramer formation and produced an unstable trimer, presumably reflecting a non-optimal orien-tation of this half-site relative to half-site 4 (Fig. 5). Similar elements lacking half-site 4 (ϾϾϽ) or containing a dysfunctional half-site at position 3 (ϾϾ8Ͼ) also failed to form tetrameric complexes (Fig. 5). Thus, most, but not all orientations of individual half-sites could function to recruit an RXR␥ tetramer.
It should be noted that identical receptor preparations were employed in these experiments, and that the overall length and base composition of the different probes were kept as similar as possible; we therefore attribute the changes in mobility of the various protein-DNA complexes to alterations in tetramer formation, rather than arising from changes in protein modification or DNA conformation.

RXR␤ Can Bind Cooperatively to a Topologically Restricted Subset of Reiterated Half-sites-
The N terminus of the RXR␤ isoform prevents tetramer formation on the consensus reiterated element composed exclusively of direct repeats ( Fig. 2A). If this interference with tetramer formation is due to a steric hindrance mediated by the RXR␤ N terminus, it might be partially relieved on DNA elements with a different half-site topology. Indeed, although tetramers of RXR␤ also failed to form on most of the elements tested (e.g. Fig. 4A), reiterated elements with a divergent orientation between half-sites 2 and 3 bound four molecules of RXR␤ with detectable cooperativity, i.e. with some degree of tetramer formation (compare the binding of RXR␤ to the ϽϽϾϾ or ϾϽϾϽ orientations versus the ϾϾϽϽ orientation; Fig. 6). Nonetheless, the extent of tetramer formation by RXR␤ on these divergent elements was signifi-

FIG. 4. The effects of varying the half-site spacing on tetramer formation by RXR␤ and RXR␥.
A series of reiterated response elements was created containing four-half-sites, all oriented as direct repeats as in Fig. 2, but differing in the spacing between half-sites. The elements are presented schematically above the relevant panels, with each individual half-site depicted as an arrow. The ability of wild-type RXR␤ and ␥ to bind these elements was examined by electrophoretic shift assay as in Fig. 2. The effect of introducing different length spacers between half-sites 2 and 3 (panel A) and of introducing the same length spacer, but between different half-sites (panel B), were tested. Approximately 1, 3, or 6 ng of receptor were used in panel A, and 0, 2, 4, 6, or 8 ng of receptor in panel B. Lanes labeled "Dimer" and "Tetramer" represent the complexes resulting when saturating amounts of RXR␥ were bound to elements containing two or four half-sites, respectively, and were used as markers. cantly less than observed for RXR␥ (compare Fig. 6 to Fig. 5).
RXR Tetramer Formation in Vitro Correlated with Reporter Gene Activation in Vivo-We tested the ability of the various response elements to mediate transcriptional activation in transient transfections (Fig. 7). Generally, the reiterated elements that permitted tetramer formation by RXR␥ in vitro also conferred efficient RXR␥-mediated gene activation in transient transfection assay, whereas elements with half-site spacings or orientations that disrupted tetramer formation failed to activate reporter gene expression. This correlation was observed over a range of receptor DNA concentrations (data not shown). Similarly, the tetramer-deficient RXR␤ isoform exhibited a much lower activity on all of the reiterated elements tested (Fig. 7), despite RXR␤ and ␥ possessing near equal transcriptional activities on elements composed of only two half-sites (Ref. 27 and data not shown) and despite the use of identical expression vectors and transfection conditions. One dramatic exception to this general correlation was noted, however; our element possessing an inversion of the third half-site (ϾϾϽϾ) conferred extremely strong transcriptional activation by RXR␥, despite failing to allow efficient tetramer formation in vitro (Fig. 7B).
Hormone has been reported to enhance dimer formation by RXRs on two-half-site elements (13). We therefore asked if the presence of hormone could influence RXR tetramer formation, perhaps accounting for the apparent discrepancy between our DNA binding studies (performed in the absence of hormone) and the transfection studies (measured in its presence). Indeed, inclusion of 9-cis-retinoic acid in the DNA binding assay significantly enhanced RXR␥ tetramer formation on several of the reiterated elements tested. This effect was particularly strong for the prototype direct repeat element (ϾϾϾϾ; Fig. 8A) and for the element containing an inverted third half-site (ϾϾϽϾ; Fig. 8B), accounting at least partly for the strong activity of the latter element in transient transfections. Notably, hormone did not enhance RXR␥ tetramer formation on the elements that were inactive in the transient transfection assays, nor did hormone confer tetramer binding by RXR␤ on any of the elements tested (data not shown). Under these conditions, inclusion of 9-cis-retinoic acid had only modest effects on dimer formation by RXR␥, as determined on elements containing only two half-sites ( Fig. 8A and data not shown).

Multiple, Spatially Distinct HREs May Function in an Analogous Manner as Reiterated Elements by Recruiting Receptor
Oligomers-It was intriguing that insertion of a 10-base spacer in the center of a reiterated element was fully compatible with tetramer formation and with reporter gene activation. This modified element is, in essence, two dimeric response elements (ϾϾ) separated by one turn of the DNA. It is known that multiple copies of a dimeric response element, when introduced into a promoter, can stimulate reporter gene expression to levels much greater than that seen with one copy (e.g. [33][34][35][36]. To test if this synergy on separated dimeric response elements is related to the ability of receptors to form high order oligomers, we compared activation by RXR␤ versus RXR␥ on reporter genes containing multiple dimeric response elements (each separated by 18 bases). Indeed, increasing the number of dimeric response elements resulted in a synergistic increase in reporter gene activation by RXR␥, but had only a much weaker effect on RXR␤ (Fig. 9, compare B to A). Thus the ability of a receptor to form oligomers on a single reiterated half-site element closely parallels the ability of the same receptor to mount a synergistic response on multiple, "separate" dimeric elements. This suggests that formation of high order receptor-DNA complexes may underlie both phenomena. DISCUSSION

RXR Tetramers Form on Reiterated Response Elements with a Broad Topological Plasticity-
The specificity of nuclear hormone receptors for their cognate response elements has traditionally been believed to operate at two levels: (a) recognition of the nucleotide sequence of the individual half-sites by each receptor monomer, mediated by contacts between the receptor FIG. 6. The effects of varying the half-site orientation on tetramer formation by RXR␤. The same form of experiment as in Fig. 5 was performed, but using the RXR␤ isoform. Marker lanes are as described in Fig. 4. were introduced individually into a pD33-CAT reporter. The reporter constructs were transfected into SL-2 cells together with either an RXR␤ or RXR␥ expression vector, as indicated in each panel. The cells were incubated in the absence (hatched bars) or presence (filled bars) of 9-cis-retinoic acid, harvested, and the chloramphenicol acetyltransferase activity was determined relative to a ␤-galactosidase internal standard, as detailed under "Experimental Procedures." zinc-finger domain and bases in the DNA, and (b) recognition of the spacing and orientation of the half-sites in an element, determined by the nature of the protein-protein interface in the receptor dimer (17)(18)(19)(20)(21)(22)(23)(24)(25). More recently, however, we have suggested that there is a third level of DNA recognition, conferred by the ability of certain nuclear receptors to bind cooperatively as high order oligomers to response elements containing highly reiterated half-sites (27). The ultimate affinity of a receptor for a given response element appears to be the combined consequence of all three components, and the destabilizing effects of a non-optimal half-site sequence can be counteracted if the half-sites are highly reiterated (27).
Many nuclear hormone receptors display a wide diversity in their ability to recognize dimeric response elements containing a variety of half-site orientations and spacings (1-7, 18 -25). In common with these dimers, the RXR tetramer also exhibits a broad plasticity in its ability to accommodate a variety of halfsite orientations and spacings in the reiterated elements examined here. Not all orientations were acceptable between all half-sites (see below), but a surprising mixture of direct, convergent, and divergent orientations could be efficiently bound by receptor tetramers in vitro and mediate reporter gene activation in vivo.
Tetramers of RXR have also been observed in solution (28 -30). Although probably reflecting a similar phenomena, it is unclear if these free tetramers are the immediate precursors of the DNA-bound complexes we observe here. The tetramers in solution are destabilized by hormone (28 -30), in contrast to the hormone-mediated enhancement of tetramer formation we observe on DNA. In addition, for the RXR tetramers in solution to be precursors to DNA-associated tetramers, either they must pre-exist as a mixed population of differing monomer orientations or they must rearrange on DNA binding so as to allow recognition of a wide range of different half-site orientations and spacings.
At Least Two Distinct Protein-Protein Interfaces Are Involved in Receptor Oligomerization-Dimers of nuclear hormone receptors are stabilized by protein-protein interfaces that map to both zinc-finger and hormone-binding domains (Fig. 10A, and Refs. 18 -26). The protein-protein interface in the zinc-finger domain is thought to dictate recognition of half-site spacing and orientation; small changes in the spacing or orientation of half-sites can destabilize dimer binding by preventing formation of the proper receptor-receptor contacts in this zinc-finger domain (15,17,19,20,24,25). In contrast, the dimerization interface in the hormone-binding domain exhibits significant FIG. 9. Effect of multiple copies of response elements on RXRmediated reporter gene regulation. A dimeric response element (composed of two copies of an AGGTCA half-site) was introduced into the pD33-CAT reporter as either one, two, three, or four copies. The reporter (1 g) was introduced into SL-2 cells together with 1 g of RXR␤ (panel A) or RXR␥ (panel B) expression vector, the cells were incubated in the absence (hatched bars) or presence (filled bars) of 9-cis-retinoic acid and harvested, and the chloramphenicol acetyltransferase activity was determined relative to a ␤-galactosidase internal standard, as detailed under "Experimental Procedures." CAT activity in the absence of hormone was near zero and is not readily visible at this scale. Each RXR molecule is presented as a dumbbell, with the zinc-finger (Zn) and hormone-binding domains (HBD) displayed as distinct regions. RXR is believed to dimerize through the interactions of at least two separate dimerization interfaces, one in the zinc-finger domain and one in the hormone-binding domain (panel A). Our results suggest two possible models of tetramer formation by RXRs; the tetramer may consist of two dimers held together through the previously elucidated hormone domain interface (panel B) or through a novel "tetramerization" interface, also mapping to within the hormone domain (panel C). In either case, each dimer in the tetramer is stabilized through the zinc-finger interface. functional plasticity, and is able to stabilize receptor dimer formation on response elements displaying a variety of half-site orientations and spacings (see Refs. 24 -26, and references therein). This contrast between the tight topological constraints imposed by the zinc-finger dimerization domain, compared with the more permissive nature of the hormone-binding domain interface, implies either the presence of a flexible "swivel" between the zinc-finger and hormone-binding domain, or the existence of multiple, alternative dimerization interfaces that are differentially invoked with the different half-site orientations (24 -26).
In light of this work on dimers, it is notable that the topological requirements for tetramer formation are non-equivalent at the different half-sites in the reiterated elements. Most striking, the spacer between half-sites 2 and 3 can vary significantly in length without effect on RXR tetramer formation, whereas similar changes in the spacing between half-sites 1 and 2, or between half-sites 3 and 4, strongly impair tetramer formation. These requirements suggest that the interactions between RXRs at half-sites 1 and 2 (or between 3 and 4) may be mediated through the zinc-finger domain interfaces previously described for two-half-site elements. In contrast, the ability of half-sites 2 and 3 to accommodate a variety of spacers is consistent with the properties observed for the hormone-binding domain interface.
Thus the RXR tetramer may be thought of as a "dimer of dimers," with one dimer bound to half-sites 1 and 2, and a second, interacting dimer cooperatively bound to half-sites 3 and 4. Two possible conceptualizations, both consistent with the data presented here, are shown in Fig. 10. In the first model (Fig. 10B), the interface holding the two dimers together is the same hormone-binding domain interface as that previously described for receptor dimers, but now oriented away from the zinc-finger domain interface. In the second model (Fig. 10C), the two receptor dimers are held together by a novel interface, also located within the hormone-binding domain, but now distinct from the domain previously characterized in stabilization of receptor dimers. In either model, additional determinants mapping outside of the hormone-binding domain would function in certain contexts to restrict tetramer formation (as observed for the N-terminal deletions and the TR⅐RXR chimera).
The Ability of Receptors to Bind Cooperatively to Reiterated Half-sites May Contribute to the Synergistic Transcriptional Activation Observed with Multiple "Dimeric" Response Elements-Naturally occurring, hormone-responsive promoters often possess multiple copies of the relevant response element (e.g. Refs. [33][34][35][36][37]. These multiple response elements appear to act synergistically to enhance the magnitude of the hormone response (33)(34)(35)(36)(37)(38)(39). We demonstrate here a correlation between the ability of different RXRs to bind to a single, tetrameric response element, and the ability of these same receptors to synergistically activate transcription from multiple, dimeric response elements. We suggest that multiple, dimeric response elements are functionally equivalent to a single reiterated element containing a large central insertion. Thus, the synergistic effects observed with multiple HREs may reflect, in part, a cooperative binding of high order receptor oligomers to these repeated elements (36).

Recognition of Reiterated Elements May Extend to Other Members of the Nuclear Hormone Receptor Family, and Might Provide a Potential Nexus for Combinatorial Transcriptional
Regulation-It is notable that many naturally occurring response elements comprise three or more half-sites (e.g. Refs. 38 -40). It is tempting to suggest that these multiple half-sites are recognized by either homo-or hetero-oligomeric forms of receptors. There are several attractive consequences of such a hypothesis. (a) The same naturally occurring elements often contain non-optimal half-site sequences that, in dimeric elements, are destabilizing for receptor binding. In the reiterated elements, the destabilizing effects of the non-optimal half-sites would be compensated by the multiple DNA-protein contacts provided by the cooperative binding of a receptor oligomer. (b) Reiterated sites also provide a potent nexus for combinatorial regulation, perhaps recruiting mixed oligomers of different receptors. The constituents of the receptor oligomer might vary from cell to cell, or from response element to response element, permitting the transcriptional response to be precisely tailored in a cell-type and promoter-specific manner.
It is intriguing that the STAT family of transcription factors also bind cooperatively to reiterated DNA sites, a phenomenon that plays an important role in DNA recognition by, and functional interactions between different STAT family members (41). We suggest that oligomer formation, as observed here for RXR, may be a common means of generating DNA specificity and combinatorial regulation for a variety of transcription factors.