Polarity and Specific Sequence Requirements of Peroxisome Proliferator-activated Receptor (PPAR)/Retinoid X Receptor Heterodimer Binding to DNA

The malic enzyme (ME) gene is a target for both thyroid hormone receptors and peroxisome proliferator-activated receptors (PPAR). Within the ME promoter, two direct repeat (DR)-1-like elements, MEp and MEd, have been identified as putative PPAR response elements (PPRE). We demonstrate that only MEp and not MEd is able to bind PPAR/retinoid X receptor (RXR) heterodimers and mediate peroxisome proliferator signaling. Taking advantage of the close sequence resemblance of MEp and MEd, we have identified crucial determinants of a PPRE. Using reciprocal mutation analyses of these two elements, we show the preference for adenine as the spacing nucleotide between the two half-sites of the PPRE and demonstrate the importance of the two first bases flanking the core DR1 in 5′. This latter feature of the PPRE lead us to consider the polarity of the PPAR/RXR heterodimer bound to its cognate element. We demonstrate that, in contrast to the polarity of RXR/TR and RXR/RAR bound to DR4 and DR5 elements respectively, PPAR binds to the 5′ extended half-site of the response element, while RXR occupies the 3′ half-site. Consistent with this polarity is our finding that formation and binding of the PPAR/RXR heterodimer requires an intact hinge T region in RXR while its integrity is not required for binding of the RXR/TR heterodimer to a DR4.

The peroxisome proliferator-activated receptors (PPAR) 1 form a group of lipid-activated transcription factors that belong to the nuclear receptor superfamily. Members of this superfamily are characterized by a structural organization in functional modules, comprising a N-terminal domain, a DNA binding domain, a hinge region, and a ligand binding domain that also contains a potent ligand-dependent transactivation domain and offers several interfaces for dimerization and protein-protein interaction. Although some of these receptors may bind to DNA as monomers, the majority binds as dimers to specific DNA sequences formed of two consensus half-sites. PPAR, thyroid hormone receptor (TR), vitamin D receptor (VDR), and all-trans-retinoic acid receptor (RAR) form a subgroup within the superfamily which heterodimerize with the 9-cis-retinoic acid receptor (RXR) and bind to response elements composed of two AGGTCA half-sites predominantly organized in a direct repeat. All the natural PPAR response elements (PPREs) described so far indeed consist of a direct repeat of two more or less conserved AGGTCA hexamers separated by a single base pair and are thus referred to as DR1 elements (1,2). This organization as direct repeat imposes a head to tail polarity to the bound heterodimer complex. Recent work has shown that in the case of TR/RXR and RAR/RXR bound to a DR4 and DR5, respectively, RXR occupies the 5Ј half-site (3)(4)(5). However, the polarity of RAR/RXR bound to a DR1 is opposite and results in a silencing of the ligand-dependent transactivation properties of RAR (6 -8). So far, the polarity of the PPAR/RXR heterodimer has not been determined.
Three different PPAR subtypes have been identified (␣, ␤/␦, and ␥). Each of them displays a distinct expression pattern in adult amphibians and rodents. PPAR␣ is predominantly expressed in hepatocytes, cardiomyocytes, proximal tubule cells of the kidney, and enterocytes. PPAR␤ (also called PPAR␦ or FAAR in rodents and NUCI in man) is more widely and often more abundantly expressed than PPAR␣ and ␥, whereas PPAR␥ is mainly restricted to the adipose tissue with some expression in spleen, retina, and hematopoietic cells (reviewed in Refs. 9 and 10). PPARs were named by virtue of their ability to be activated by peroxisome proliferators. However, new specific activators and ligands for the different PPAR subtypes are now emerging. Recent studies have shown that various fatty acids, eicosanoids, and hypolipidemic compounds, such as fibrates, directly bind to PPAR␣, ␤, and ␥. In addition, PPAR␥ binds antidiabetic thiazolidinediones, which lower the plasma levels of glucose, triglycerides, and insulin (reviewed in Ref. 11). Interestingly, not only can PPAR␥ direct adipocyte differentiation but all the PPAR target genes identified so far are involved in major steps of fatty acid metabolism, including fatty acid transport, intracellular binding,and ␤-oxidation, as well as fatty acid synthesis and storage as triglycerides (reviewed in Ref. 10). Thus, the nature of the PPAR ligands together with the observation that all PPAR␣ and ␥ target genes discovered so far are directly involved in lipid metabolism reinforce the novel concept of fatty acids and their derivatives acting as hormones and controlling their own fate through these specific nuclear receptors (reviewed in Refs. 12-14).
The malic enzyme (ME) gene is one of the PPAR target genes whose product is involved in lipogenesis. ME catalyzes the oxidative decarboxylation of malate to pyruvate, which results in the production of NADPH for fatty acid synthesis. Thyroid hormone (T 3 ) leads to a pronounced stimulation of ME gene expression in the liver (15) and is involved in adipose differentiation for which ME is a late marker (16,17). Interestingly, thyromimetic effects of hypolipidemic fibrates on ME expression in rat liver have been observed (18,19). Since these compounds are PPAR ligands, these thyromimetic effects may be mediated by PPAR/RXR heterodimers binding to the ME promoter. We 2 and others (20,21) have identified two DR1-like elements in the ME promoter, hereafter referred to as MEp (proximal, located at bp Ϫ340/Ϫ328 with respect to the transcription initiation site) and MEd (distal, located at bp Ϫ463/ Ϫ451). However, conflicting results have been reported on the respective role of these two elements in PPAR-mediated regulation of ME gene expression; Castelein et al. (20) have identified MEp and Hertz et al. (21) MEd as the functional PPRE. Our interest in the hormonal cross-talk in adipocytes between the T 3 and fatty acids signaling pathways prompted us to pursue the characterization of the ME promoter with respect to PPAR-mediated gene regulation. In particular, we studied its regulation by PPAR␥, the subtype abundantly expressed in adipose tissue. In this work, the demonstration that only MEp and not MEd is a functional PPRE gave us a tool to identify the determinants of a functional PPRE, which go beyond the characteristics of the core DR1 element. Furthermore, we unveiled the polarity of the heterodimer PPAR/RXR bound to a DR1, which explains the nature of the PPRE-specific requirements.

EXPERIMENTAL PROCEDURES
Plasmids-The Xenopus PPAR␥ (1), human RXR␣ (22) and rat TR␣1 were subcloned into a modified pSG5, pSG5PL (gift of Dr. Hélène Richard-Foy). To optimize in vitro translation of the PPAR␥ and RXR␣ receptors, a Kozak (23) consensus sequence was introduced by recombinant polymerase chain reaction at the translational start site of the receptor.
The P box mutants of PPAR␥ and RXR␣ were obtained in two sequential mutagenic steps using the Kunkel et al. (24) method. The primers 5Ј-GCGTCCATGCATGTGGATCTTGCAAGGGGTTCTT-3Ј and 5Ј-GTGGATCTTGCAAGGTGTTCTTTAGAAGAAC-3Ј were used to create PPARpgr; the primers 5Ј-GAGTGTACAGCTGCGGGTCGTG-CAAGGGCTTCTT-3Ј and 5Ј-GCGGGTCGTGCAAGGTCTTCTTCAA-GCGGAC-3Ј were used to create RXRpgr. The primer 5Ј-GGCATGAA-GCGGGAATTCGAGGGGGAGGAGCGGCAGCG-3Ј was used to create the RXR(T) mutant, using the same approach. pME775 was created by cloning the 809-bp XbaI-BamHI fragment (bp Ϫ775/ϩ34) of the pME882 plasmid (25) into pBLCAT3 (26). The mutant reporter plasmids pME775pko and pME775dko, in which, respectively, the proximal or the distal putative PPRE has been mutated, were generated by recombinant polymerase chain reaction using the Expand High Fidelity kit (Boehringer Mannheim). The mutated sequences are as follow: within pME775dko from bp Ϫ469 to Ϫ445: TGCACTAGATCTGTCCGGTCTAACA; within pME775pko from bp Ϫ346 to Ϫ322: CATTCTAAGCTTGAGTTGATCCCCT.
The reporter constructs containing the wild-type or mutated PPREs upstream of the heterologous thymidine kinase promoter were created by cloning a single copy of the response element into the BamHI/ HindIII sites of the pBL-CAT8ϩ plasmid. All the constructs were verified by DNA sequencing.
Cell Culture and Transfections-NIH3T3 cells were maintained in culture and transfected as described previously (27). Briefly, each cuvette for electroporation contained 4 ϫ 10 6 cells at a density of 12.5 ϫ 10 6 cells/ml and a total of 70 g of plasmid DNA (20 g of chloramphenicol acetyl transferase (CAT)-reporter plasmid, 12 g of each expression vector as indicated in the figure legends, 1 g of pCMV␤gal as internal control for transfection efficiency, and pUC19 or salmon sperm DNA to complete to 70 g). After electroporation, cells were resus-pended and equally distributed in four 60-mm dishes containing the transfection medium supplemented with the appropriate activator. After 48 h, cell extracts were prepared and ␤-galactosidase and CAT activities were determined as described previously (27).
Electrophoretic Mobility Shift Assays-Proteins for electrophoretic mobility shift assays (EMSA) were obtained by in vitro transcription and translation using the TNT ® coupled reticulocyte lysate system (Promega). Parallel translations using [ 35 S]methionine (Amersham Corp.) followed by SDS-PAGE analysis and exposure in the phosphoranalyst (Bio-Rad) allowed standardization of the different protein preparations. Alternatively, nuclear extracts from Sf9 cells infected with a recombinant baculovirus overexpressing the mouse RXR␤ were used.
The probes and competitors corresponded to the double-stranded oligonucleotides indicated in the figures flanked on their 5Ј side and 3Ј side by the BamHI and HindIII overhang sequence, respectively. The ACO(A) and ME TRE oligonucleotides are (5Ј-GATCCCGAACGTGAC-CTTTGTCCTGGTCCCGATC-3Ј) and (5Ј-GATCAGGACGTTGGGGT-TAGGGGAGGACAGATC-3Ј), respectively.
In vitro translated proteins were preincubated for 15 min at room temperature, in a buffer containing 25 mM HEPES, pH 7.5, 5 mM MgCl 2 , 1 mM EDTA, 10% glycerol, 40 mM KCl, 1 mM dithiothreitol, and 8 g poly(dI-dC). After a further 20-min incubation period at room temperature in the presence of 20,000 cpm of labeled probe, the complexes were separated on a 6% native polyacrylamide gel with 0.25 ϫ TBE running buffer at 500 V, 25 mA, 4°C. For DNA binding competition experiments, a 10 -100-fold molar excess (as indicated) of the unlabeled double-stranded competitor oligonucleotide was added to the preincubation reaction. Gels were dried and exposed at Ϫ80°C to a Kodak X-Omat AR film with an intensifying screen. Scanning and treatment of the images were performed using the Cirrus 1.2 software. When appropriate, gels were analyzed by phosphor-analyst or densitometry.

RESULTS
The ME Promoter Contains an Element Responsive to PPAR␥-The possibility of cross-talk between PPAR and TR signaling pathways at the level of ME gene expression might be important in cells, such as adipocytes, in which all three proteins, PPAR, TR, and ME, are expressed. We thus tested if the ME promoter is responsive to PPAR␥, the PPAR subtype which is crucial for adipogenesis and is present at high levels in mature adipocytes (28,29). As seen in the schematic representation of the ME promoter in Fig. 1A, two DR1 elements, MEp (proximal, at bp Ϫ340/Ϫ328) and MEd (distal, at bp Ϫ463/ Ϫ451), are located upstream of the thyroid hormone response element ME TRE (27). Conflicting reports from Castelein et al. (20) and Hertz et al. (21) indicated mediation of the ME peroxisome proliferator response through either of these elements.
To clarify this point, we first tested whether the ME promoter, from Ϫ775 to ϩ34 base pairs relative to the initiation site was responsive to PPAR␥ and, as a positive control, to TR. Transfection analyses using a CAT reporter gene in NIH3T3 cells confirm that TR␣1 can control the ME promoter, as a T 3 -dependent 7-fold stimulation of the reporter gene expression was observed. Cotransfection of the reporter gene with a vector expressing PPAR␥ in the absence of PPAR activator moderately but reproducibly induced expression of the reporter gene (2.5-fold). Whether this induction was due to a constitutive activity of PPAR␥ or to endogenous PPAR activators was not further analyzed. Addition of the PPAR activator Wy 14,643 to the culture medium resulted in a 5.5-fold stimulation over the basal expression level of the reporter construct. A similar induction was obtained using the thiazolidinedione BRL49653, a PPAR␥ specific ligand (data not shown). These effects are receptor-specific, since T 3 and Wy 14,643 stimulated the ME promoter activity only in the presence of TR and PPAR, respectively (Fig. 1B).
Second, we determined the relative contribution of the two putative PPREs to PPAR␥ responsiveness. For that purpose, we first mutated either the MEd or the MEp sequence within the homologous promoter creating pME775dko and pME775pko, respectively. The mutation of the MEd sequence did not alter the response of the reporter gene to Wy 14,643, while mutation of MEp suppressed responsiveness, indicating that MEp is the responsive element ( Fig. 1B; see also Castelein et al. (20)). To test if this result was independent of the position or relative orientation of each of these elements within the ME promoter (see Fig. 1A), we inserted, in the same orientation, the two elements encompassing the DR1 motif plus their 6 bp flanking either side, upstream of the herpes simplex virus thymidine kinase heterologous promoter in a CAT reporter gene. pMEpfl, containing the proximal element, and pMEdfl, containing the distal element, were then transfected into To test if the above reported difference in PPAR responsiveness of MEp and MEd reflects their ability to bind PPAR/RXR heterodimers, we performed EMSAs. We used MEpfl as a probe, in vitro translated PPAR␥, and cellular extracts from Sf9 cells infected by a recombinant baculovirus expressing RXR␤. Fig. 1C shows that PPAR␥/RXR binding complex could form on MEpfl, whereas no PPAR binding was observed in absence of RXR (data not shown). Unlabeled double-stranded oligonucleotides encompassing either the PPRE of the acyl-CoA oxidase gene (ACO(A)) or MEpfl itself efficiently competed PPAR/RXR complex formation on the probe MEpfl, whereas MEdfl, which was unresponsive in the functional test, was a very inefficient competitor. As expected, ME TRE did not compete for the PPAR/RXR complex binding.
These data show that of the two DR1 elements present in the ME promoter, only the MEp is able to bind PPAR␥/RXR heterodimers in a sequence-specific manner and can mediate peroxisome proliferator signaling.
Conversion of MEd to a Functional PPRE; Role of the DR1 Spacing Nucleotide and of Flanking Nucleotides-The inability of MEd to act as PPRE was puzzling since its sequence is closer to the consensus DR1 than that of the functional element MEp. Analysis of the compilation of the natural PPREs characterized so far (Fig. 2) and recent reports suggested that two regions of the PPRE may be given particular attention: the spacing nucleotide between the two half-sites which is predominantly an A and the sequence immediately 5Ј upstream of the DR1 core element (Fig. 2) (13,31,32). 3 In contrast to MEp, MEd strikingly diverges from the consensus in these two regions suggesting that these differences might be responsible for the lack of responsiveness of MEd to PPAR␥ and Wy 14,643. The role of the spacing nucleotide was first analyzed by changing the spacing nucleotide A of MEpfl to either G, C, or T, resulting in the MEp(G), MEp(C), and MEp(T) elements, respectively (see Fig. 3A). The presence of an A as in MEpfl reproducibly resulted in the strongest binding, while a C at this position always resulted in the weakest interaction (Fig. 3A). Notably, the nonfunctional sequence MEd has a C at the corresponding spacing position. Thus, we introduced the converse mutation in MEd, changing its spacing nucleotide from a C to an A, and observed a partial restoration of PPAR␥/RXR binding (MEd(DR-A); Fig. 3B). The inability of MEdfl to function as a PPRE might also be determined by the 5Ј-flanking nucleotides: TTCT in MEp versus TTAG in MEd. Indeed, transversion of AG to CT, creating MEd(CT), enhances the formation of PPAR␥⅐RXR complexes. Finally, combination of the mutations in the two regions, the spacing and the 5Ј-flanking nucleotides, in MEd(CTA) had a synergistic effect on PPAR␥/RXR binding (Fig. 3B), which was comparable to that observed on MEpfl.
To correlate PPAR binding affinity and transcriptional activity, MEp(C) and the MEd mutants were inserted upstream of the herpes simplex thymidine kinase promoter and used as reporter constructs in cotransfection experiments (Fig. 3C). Compared with MEpfl, MEp(C) lost most of its capacity to mediate PPAR␥ induction. While the mutants MEd(DR-A) and MEd(CT) did not confer a PPAR responsiveness to the reporter gene, the double mutant MEd(CTA) mediated a PPAR-dependent increase of the basal level of expression, which was further induced by Wy 14,643. This transactivation pattern of MEd(CTA) correlates well with its capacity to bind PPAR/RXR and closely resembled that observed with MEpfl, confirming the importance of both the central adenylate and the 5Ј flank in defining a PPRE.
The Relative Importance of the 5Ј-Flanking Sequences Depends on the Core DR1-Previous work on PPREs stressed their DR1-type structure, particularly since a synthetic perfect DR1 core sequence exhibits a high PPAR/RXR binding affinity. The newly discovered role of the 5Ј flank in native elements raises the question as to whether it mainly compensates for a weak interaction of the heterodimer with imperfect DR1 core sequences as found in native elements (see Fig. 2). In that respect, the poorly conserved 3Ј half-site (AGTTGA) of the MEp PPRE is an excellent example. To answer the above question, we assessed the importance of the flanking sequences for PPAR/RXR binding both in the context of the native MEp and that of a perfect synthetic DR1 element. EMSAs, as the one shown in Fig. 4, demonstrate that PPAR/RXR binding to MEp is 80% less efficient (mean of two independent experiments) when its specific 5Ј flank is mutated. In contrast, binding to the perfect DR1 is affected to a lesser extent by the sequence of the 5Ј flank, exhibiting a 26% loss of binding efficiency in presence of the mutated versus the wild-type 5Ј MEp flank (mean of three independent experiments). This result reinforces the hypothesis that the 5Ј-flanking sequence does play a role in the specific DNA recognition by PPAR, as PPAR/RXR heterodimer; together with an imperfect core DR1, it contributes to selective binding of PPAR (see "Discussion").
Polarity of the PPAR␥/RXR Complex on MEp-Based on the results described above, we hypothesized that PPAR may bind to the extended 5Ј half-site of a PPRE, while RXR binds to the 3Ј hexamer. To determine the PPAR/RXR binding polarity, we converted the P box of PPAR and of RXR, into that of the glucocorticoid receptor (GR), creating PPARpgr and RXRpgr, respectively (Fig. 5A). Such hybrid receptors will recognize the consensus hexamer -AGAACA-of the GR response element (GRE) (3, 4). Consequently, we tested in EMSA two hybrid MEp elements in which either the 5Ј hexamer or the 3Ј hexamer was replaced by a GRE half-site, giving MEpfl:GRE5Ј and MEpfl:GRE3Ј, respectively (Fig. 5A). The formation of PPAR/ RXR complex was barely detectable either on MEpfl:GRE5Ј or on MEpfl:GRE3Ј (Fig. 5A, compare lane 1 to lanes 5 and 9). In contrast, the combination of PPAR and RXRpgr led to the formation and binding of a complex to MEpfl:GRE3Ј but neither to MEpfl nor to MEpfl:GRE5Ј (Fig. 5A, compare lane 6 to  lanes 2 and 10). This result suggested that RXR indeed binds to the 3Ј half-site of the PPRE. As expected, the heterodimer PPARpgr/RXR did not bind to MEpfl:GRE3Ј (Fig. 5A, lane 7); however, it was not able to bind to MEpfl:GRE5Ј either (Fig.  5A, lane 11). As shown in the top panel of Fig. 5, that latter probe associates a GRE half-site -AGAACA-in 5Ј and the poorly conserved MEp 3Ј half-site -AGTTGA-. Thus, it is likely that the divergence of the overall sequence of MEpfl:GRE5Ј from a DR1-like element is too important to accommodate nuclear receptor binding.
To circumvent this experimental limitation, we repeated the same experiment using the synthetic element DR1fl and chimeric response elements in which we introduced a GRE halfsite in place of the 5Ј hexamer or of the 3Ј hexamer giving DR1fl:GRE5Ј and DR1fl:GRE3Ј, respectively (Fig. 5B). Very clearly, the PPAR/RXRpgr heterodimer bound to DR1fl:GRE3Ј but not to DR1fl:GRE5Ј (Fig. 5B, lanes 11 and 7, respectively), whereas PPARpgr/RXR heterodimer bound to DR1fl:GRE5Ј but not to DR1fl:GRE3Ј (Fig. 5B, lanes 8 and 12, respectively). In other words, the complex that contained the mutant PPARpgr could form only on the response element that bears the GRE sequence in the 5Ј position while the complex containing the mutant RXRpgr could form only on the element with the GRE sequence in the 3Ј position. Together these results provide evidence for a defined binding polarity of the PPAR/ RXR heterodimer onto its response element, with PPAR anchored to the 5Ј extended half-site and RXR to the 3Ј half-site. Interestingly, the role of the 5Ј-flanking sequences clearly appeared in the context of the mutant receptors and mutant PPREs, as exemplified by PPAR/RXR and PPARpgr/RXR which bound to DR1fl:GRE5Ј with a greater efficiency than to DR1:GRE5Ј (Fig. 5B, compare lane 6 to lane 18 and lane 8 to lane 20; see "Discussion").
In agreement with these binding studies, PPARpgr was unable to activate the expression of the MEpfl reporter construct while it could efficiently activate the reporter DR1fl:GRE5Ј (Fig. 5C). Cotransfection of RXRpgr with either of these reporter genes strongly inhibited the PPAR␥-induced expression, suggesting a dominant negative effect of this mutant receptor. In contrast, the reporter construct that contains the GRE at the 3Ј half-site was not activated by PPARpgr, was poorly activated by PPAR in presence of the endogenous RXR, but was significantly stimulated if RXRpgr is cotransfected with PPAR (Fig.  5C). Thus, these functional studies further confirmed the binding polarity of PPAR/RXR to PPRE.
Dimerization Interface in PPAR/RXR Interactions-The binding polarity of PPAR/RXR onto a PPRE should be reflected in the dimerization surface used by each partner. Structural and biochemical analyses of nuclear receptor heterodimers bound to direct repeat elements demonstrate that the second zinc finger of the receptor which binds to the 5Ј half-site contacts the T box and the first zinc finger of the receptor which binds to the 3Ј half-site (3)(4)(5)(32)(33)(34)(35)(36)(37). Hence we mutated 3 amino acids in the T box of RXR, creating RXR(T) (Fig. 6). These mutations should affect the function of RXR as the 3Ј-binding receptor and thus the formation and binding of the heterodimer PPAR/RXR to a PPRE such as MEp. In contrast, they should not alter the function of RXR as the 5Ј-binding receptor, and consequently still allow the formation and binding of the complex RXR/TR on a TRE such as the ME TRE . Indeed, as seen in Fig. 6, PPAR is not able to bind MEp when using RXR(T) as partner while a complex corresponding to the heterodimer TR/RXR(T) can form on ME TRE albeit with a weaker affinity than TR/RXR (Fig. 6, lanes 3 and 6, respectively). These results are clearly consistent with the polarity that we demonstrate for the PPAR/RXR complex bound to PPRE and further suggest that the T region of RXR is involved in the dimerization surface between RXR and PPAR.

DISCUSSION
The analysis of the regulation by peroxisome proliferators of the ME promoter led us to a better understanding of the molecular mechanisms governing PPAR-mediated gene regulation. First, the observation that in our experimental model, only the proximal DR1 sequence located at Ϫ340/Ϫ328 (MEp) can function as a PPRE, while the element Ϫ451/Ϫ463 (MEd) does not, prompted us to further analyze the structural definition of a PPRE. Our results add the following three main properties of native PPREs to the initial definition as DR1: (i) an extended 5Ј half-site, (ii) an imperfect core DR1, and (iii) an adenine as a spacing nucleotide between the two hexamers. Second, we provide evidence that this PPRE structure reflects the polarity with which PPAR/RXR binds to the element: PPAR recognizes the 5Ј extended half-site and RXR the 3Ј half-site.
Specificity and Promiscuousness of PPRE and DR1 Elements-Binding of PPAR/RXR to DR1-like elements raises the question of the selectivity of this interaction since these elements are binding sites for several members of the nuclear receptor superfamily. DR1 elements were shown to mediate 9-cis-retinoic acid responsiveness through binding of RXR homodimers, as first demonstrated on the cellular retinol binding protein II element (CRBP-II) (38). Moreover, an unbiased search for RXRE as well as analyses of synthetic elements revealed the importance of an A or a G as the base immediately 5Ј of perfect core hexanucleotides in a DR1 configuration (39 -42). Interestingly, however, no or very weak binding of RXR homodimer was observed in previous work describing naturally occurring PPREs, indicating that discriminating parameters must account for RXR homodimer and PPAR/RXR heterodimer selectivity. In addition, DR1 elements are also binding sites for at least three orphan members of the nuclear receptor superfamily: HNF-4, ARP-1, and ear-3 (or COUP-TF), the most promiscuous response element being a synthetic DR1-G element, in which the two consensus hexamer are flanked by a G in 5Ј (see Ref. 41 and references therein). Competition for binding and functional interference can indeed occur between PPAR/ RXR and HNF4 (43,44) as well as between PPAR/RXR and COUP-TF (45,46). However, the native HNF4 binding site characterized in the ␣ 1 -antitrypsine gene only poorly fits the DR1 consensus and binds neither COUP-TF nor ARP1 (47). Again this underscores that subtle differences in the natural DR1-type response elements must be important for their selectivity. The alignment in Fig. 2 of the sequence of 19 native PPREs previously characterized, with the number of occurrences for each nucleotide at each position (from 1 to 17) presented in the bottom panel, reveals specific characteristics of native PPREs. DR1 motifs clearly appear between position 5 and 17, with an obvious lack of nucleotide preference only at a single position (nucleotide 8), while the spacing nucleotide between the half-sites (position 11) is remarkably conserved. Our mutation analyses show that indeed an adenine as the spacing base results in the strongest heterodimer binding, whereas cytidine is the least desirable of the four possibilities at that position. A certain degree of conservation is also present in the 5Ј-flanking nucleotides (AACT, position 1-4), suggesting a potential role for this region. Herein, we demonstrate that the nucleotides in positions 3 and 4 extend the 5Ј half-site and are an integral part of the PPRE, in agreement with recent work done with the PPRE of the CYP4A6 gene (31). Importantly, the role of the 5Ј-flanking sequence is especially apparent when the DR1 sequence is poorly conserved with respect to a perfect DR1, as for MEp and MEpfl (Fig. 4) but also as in the chimeric elements DR1:GRE5Ј and DR1fl:GRE5Ј (Fig. 5B). The same applies when RXR itself binds poorly because it has been altered, as in RXRpgr, leading to a PPAR/RXRpgr complex that binds to DR1fl but not to DR1. Thus, it appears that a weakened interaction of PPAR/RXR with the core DR1 is tolerated as long as specific contacts in the 5Ј flank can stabilize it. These results clearly plead for the importance of the association of a specific 5Ј-flanking sequence, an imperfect core DR1, and a central adenine as structural characteristics allowing the discrimination of PPRE from other DR1 response elements by the PPAR/RXR heterodimer.
Polarity of the DNA-bound PPAR/RXR Heterodimer-Consistent with the discriminatory role of the PPRE 5Ј flank, we demonstrated that PPAR binds to the extended 5Ј half-site of the PPRE, while RXR binds to the 3Ј hexamer. This is in contrast to the RXR/VDR, RXR/TR, and RXR/RAR heterodimers which bind with the reverse polarity to DR3, DR4, and DR5, respectively (3)(4)(5), but is in register with the RAR/ RXR complex bound to a DR1 (6 -8). Asymmetric contacts operating between the two partners of a heterodimer bound to a direct repeat occur between their respective DNA binding domain (DBD) and carboxyl-terminal extension (CTE) that comprises the T box and A box. The hinge region, which also includes the CTE, would allow adequate rotation of the interacting ligand binding domains with respect to the DBD (reviewed in Refs. 49 and 50). While the three-dimensional structure of the DBD and CTE region of PPAR has not yet been solved, some of its properties can be inferred from the polarity to which PPAR binds to PPRE and from detailed biochemical studies and structural analyses of RAR/RXR and TR/RXR heterodimers bound to direct repeat sequences (3-5, 32-37). In the configuration of an RAR/RXR heterodimer bound to a DR1 FIG. 6. An intact RXR T box is required for the interaction of RXR with PPAR. Binding analysis of wild-type PPAR or TR with either RXR or RXR(T) to MEpfl and ME TRE . Radiolabeled probes, as indicated, were incubated with standardized amounts of rabbit reticulocyte translated wild-type or mutant receptors. A schematic representation of the DNA binding domain indicating the position of the T box is shown at the top.  (51). Peptide sequences of the cognate receptor region involved in the specific recognition of these additional bases, are aligned starting at the end of helix 2, as defined for the glucocorticoid and estrogen receptors (77,78). Identical residues are boxed. For comparison the T/A region of RXR is shown; the three amino acids that are mutated in RXR(T) are underlined. element, the crucial amino acids for heterodimerization of the 5Ј-positioned receptor (RAR) are located in the second zinc finger outside its first knuckle or D box while the 3Ј-positioned receptor (RXR) contributes to the dimerization interface via its T box. Exclusion from the dimerization interface of the D box of the 5Ј-positioned receptor could explain why PPARs have a D box of 3 amino acids instead of 5 in other members of the superfamily. Consistently, exchanging this D box with that of RXR did not alter PPAR/RXR binding to a PPRE. 4 Because of the 5Ј location of the PPAR molecule on a PPRE, its T/A region may be involved in the recognition of the 5Ј extension of the PPRE half-site. This has been described for receptors which can bind extended half-sites as monomers such as FTZ-F1, NGFI-B/Nurr1, ROR/RZR, Rev-erb␣, and TR␣1. Their recognition of the base pairs extending the consensus hexamer in 5Ј involves critical amino acids in their respective CTE region (48, 50 -55). In PPARs the corresponding region is highly conserved between the different subtypes but differs from the other above mentioned receptors. The closest similarity to PPAR within this region is found in ROR/RZR and Rev-erbA␣, which correlates with the closest 5Ј-extended sequence similarity of their respective response elements (see Fig. 7). However, in contrast to these latter receptors, PPAR is unable to bind as a monomer.
The polarity of PPAR/RXR onto the direct repeat may also explain the spacing of 1 nucleotide between the two half-sites of the PPRE. Indeed, VDR, TR, and RAR by occupying the 3Ј half-site of DR3, DR4, and DR5 respectively, dictate the spacing that provides the specificity of their respective response element, since all of them have the same partner RXR. Accordingly, RXR on the 3Ј half-site of RXRE and PPRE elements dictates the common spacing of 1 nucleotide, likely through physical constraints residing in the role of its T region as dimerization surface. One consequence of this reasoning is that if RXR binds to the 3Ј half-site, as it does in the context of the RAR/RXR, PPAR/RXR and RXR/RXR dimers, selectivity cannot be conferred by spacing. Instead, like PPAR, the heterodimerization partner might select a specific 5Ј extended half-site, allowing discrimination between DR1 elements.
Elucidating the polarity with which nuclear receptors bind to their cognate response elements is not only important for understanding the molecular mechanism of DNA-protein interaction and of dimerization properties, but it may give insight into some functional aspects of receptor activation by the ligand and of transactivation. Along this line of thought, Kurokawa et al. (7) recently demonstrated that in the context of the DR1-bound RAR/RXR complex, RAR fails to release NCoR in the presence of ligand, suggesting a molecular mechanism for the repression caused by RAR/RXR complexed to a DR1. How do co-activators and co-repressors interact with each receptor within a PPAR/ RXR heterodimer and how do the respective ligands influence these interactions remain to be solved. Interestingly, cooperativity in PPAR and RXR signaling pathways is also mediated by PPRE, as demonstrated by the additive and synergistic transcriptional effect of their respective ligands in cell culture (56,57) and in vivo (58). In that respect, our detailed characterization of the ME PPRE unveils some critical mechanisms by which PPAR can achieve functional specificity and as such is a first step toward the understanding of the molecular mechanisms underlying cooperativity and hormonal cross-talk via the ME promoter.