DNA Binding Specificity and Function of Retinoid X Receptor a*

Retinoid X receptors are members of the erbA super- family of ligand-inducible transcription factors. Similar to several other members of this gene family, retinoid X receptors are known to bind to the hexameric DNA sequence AGGTCA. After binding to a direct repeat of this hexamer with a one-base pair spacer, retinoid X receptor homodimers are able to activate transcription in the presence of the ligand 9-cis-retinoic acid. However, it is not known if AGGTCA represents the highest affinity binding site for retinoid X receptors. A combination of the electrophoretic mobility shift assay and polymerase chain reaction was used to isolate from a pool of random DNA those sequences that bind retinoid X receptors with highest affinity. This approach, combined with mutational analysis and DNA footprinting, led to the iden- tification of the seven-base pair sequence GGGGTCA as the highest affinity retinoid X receptor binding site. A direct repeat of this sequence is substantially more active than a direct repeat of AGGTCA as a retinoid X response element.

Retinoid X receptors are members of the erbA superfamily of ligand-inducible transcription factors. Similar to several other members of this gene family, retinoid X receptors are known to bind to the hexameric DNA sequence AGGTCA. After binding to a direct repeat of this hexamer with a one-base pair spacer, retinoid X receptor homodimers are able to activate transcription in the presence of the ligand 9-cis-retinoic acid. However, it is not known if AGGTCA represents the highest affinity binding site for retinoid X receptors. A combination of the electrophoretic mobility shift assay and polymerase chain reaction was used to isolate from a pool of random DNA those sequences that bind retinoid X receptors with highest affinity. This approach, combined with mutational analysis and DNA footprinting, led to the identification of the seven-base pair sequence GGGGTCA as the highest affinity retinoid X receptor binding site. A direct repeat of this sequence is substantially more active than a direct repeat of AGGTCA as a retinoid X response element.
Retinoid X receptors (RXRs)' are ligand-inducible transcription factors that, along with the receptors for vitamin D (VDR), thyroid hormone (TR), and retinoic acid (RAR), belong to the erbA superfamily. RXR was first described by Hamada et al. (1) when RXRP, referred to as HBRIIBP, was found to bind and activate transcription of the murine major histocompatibility complex class I genes. Presently three isoforms termed a, p, and y have been identified (2)(3)(4). Compared with mouse RXRa, the P and y isoforms share 92 and 95% amino acid homology in the DNAbinding domain, and 87 and 86% amino acid homology in the ligand binding domain. The N-terminal regions of these isoforms show no conservation.
RXRs can heterodimerize with TRs, RARs, and VDR and thereby enhance the DNA binding of these receptors to their respective response elements. This enhanced DNA binding leads to increased thyroid hormone (T3), retinoic acid (RA), and vitamin D-mediated transcriptional activation (5-9). In addition to the heterodimer properties described above, RXRs also * This work was supported by National Institutes of Health Grant DK44155. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "uduertisement" in accordance with 18  It has been demonstrated that RxRs, TRs, RARs, and VDR all can bind to direct repeats of the sequence AGGTCA (12)(13)(14). For these receptors, the specificity of DNA binding is determined by the spacing between the AGGTCA half-sites. Thus, a direct repeat of AGGTCA with a one-bp spacer (DR+l) is a response element for RXR (RXRE) (141, whereas DR+3, DR+4, DR+5 are response elements for VDR, TR, RAR, respectively (12, 13).
However, although RXRs are known to bind to the hexamer AGGTCA, it is not known if this sequence represents the optimal binding site. Currently, few RXR responsive genes have been characterized, and the determination of the optimal binding site from such a limited number of genes would be difficult.
Therefore the goal of the present work was to identify the optimal DNA sequence for RXR binding. A nonbiased approach was taken similar to that employed by Blackwell et al. (15) to study the DNA binding of c-Myc. By using random DNA pool selection, competition assays, DNA sequencing and footprinting, mutational analysis, and transient transfections, we were able to identify the optimal RXRa binding site as the heptamer 5"GGGGTCA.

MATERIALS AND METHODS
Production and Purification of RXRa-The mouse RXRa cDNA was ligated in frame into the Escherichia coli expression vector pMAL (New England Biolabs Inc.) to produce a fusion protein consisting of maltose binding protein (MBP) followed by the recognition sequence for the protease factor Xa, fused to m a . Expression and purification of recombinant RXRa were performed according to the vendor's protocol. Briefly, E. coli strain XL1 was transformed with the expression plasmid, MBP-RXRa expression was induced with isopropyl-1-thio-P-D-galactopyranoside, and the recombinant fusion protein was purified by amylose affinity chromatography Purity was confirmed by SDS-polyacrylamide gel electrophoresis (PAGE) and Coomassie Blue staining, which revealed only a single band of -88 kDa, consistent with the expected combined mass of MBP (42 kDa) plus RXRa (46 kDa) (data not shown). Incubation with factor Xa was then performed to release RXRa from MBP. The cleaved MBP was not removed, since control studies indicated it does not bind DNA. The final products were shown by SDS-PAGE to be of the appropriate size.
Random DNA Pool Construction-Oligonucleotide primers A (5'-TC-CGAATTCCTACAG) and B (5"AGACGGATCCATTGCA) were synthesized. A pool of oligonucleotides of 49 nucleotides in length was synthesized containing the primer A sequence at the 5'-end, an internal random sequence of 18 nucleotides, and the reverse complement of primer B at the 3'-end. This internally random sequence oligonucleotide pool was converted to double-stranded DNA by annealing with primer B and filling-in with a Klenow reaction. The double-stranded DNA pool was purified by PAGE.
Electrophoretic Mobility Shifthsays (EMSAkProtein-DNA binding reactions were performed in 35 1. 11 of 20 mM HEPES, pH 7.8, 20% glycerol, 1.4 pg of poly(d1-dC), 1 mM dithiothreitol, 50 mM KC1 (in the later steps of the selection process up to 400 mM KC1 was employed), 0.1% Nonidet P-40, 32P-labeled DNA, and E. coli-expressed RXRa. The amounts of RXRa and 32P-labeled DNA are specified in subsequent sections. Reactions were incubated at room temperature for 45 min prior to electrophoresis. Electrophoresis was carried out on 0.25 X TBE  RXR a-monomer binding The sequences shown contain the conserved portions of the selected random 18-mers. Conserved nucleotides are numbered 1 through 7, with surrounding nonconserved nucleotides numbered -3 through -1 and 8 through 10. The number of clones containing either A, C, G, or T a t each position is indicated, and the consensus sequence is shown. Consensus:

G G G G T C A
(22 mM Tris base, 22 mM boric acid, 0.5 mM EDTA), 6% polyacrylamide gels (29:l acrylamidehisacrylamide) a t room temperature. Wet gels were exposed to film for 24 h a t 4 "C during the selection process. For all other EMSAs, the gels were fixed in 30% methanol, 10% acetic acid; dried; and exposed to film with an intensifying screen for 3-24 h a t Selection Process-The random pool of double-stranded DNA was end-labeled with I y-3zPIATP by T4 polynucleotide kinase. An EMSA was performed with 40,000 cpm of '*P-labeled random DNA pool plus 40 ng of nonradiolabeled random DNA pool and 2 pg of mouse RXRa. In the initial round of selection, the RXRa.DNA complex was not expected to be visible. Therefore a 52-bp DNA fragment containing the hexamer AGGTCA was constructed as a marker. This DNA was end-labeled, incubated with RXRa, and electrophoresed in a lane adjacent to the random DNA incubation. The location of the marker DNA.RXRa complex was used as a reference to identify the portion of the gel to excise in the random DNA pool-RXRa lane. The DNA was eluted into 0.1% SDS, 0.5 M NH,OAc, 1 mM EDTA, ethanol precipitated; and amplified by the polymerase chain reaction (PCR) using primers A and B. Cycling conditions included 1 min each of denaturation a t 94 "C, annealing a t 56 "C, and extension a t 72 "C for 30 cycles. The products were purified by PAGE, and the selection process was repeated. For subsequent selections, 40,000 cpm of the :j2P-DNA pool without nonradiolabeled DNA were used in the protein.DNA incubations. Eight rounds of selection were performed in total.
In the second round of EMSA, an RXRa.:'2PP-DNA complex could be visualized. In the first two rounds, the protein.DNA incubations used 50 mM KCI, and in the first three rounds, 2 pg of RXRa was employed. To select for higher affinity binding sites, rounds 3-5 and 6-8 were performed with 150 and DNAs, and protein.DNA complexes were analyzed by EMSA. The dose of competitor DNA that competes 50% of the RXRn-:'"P clone 29 complex (C,,) was determined by densitometry and used as a measure of relative affinity. A competition assay using both a high affinity clone (clone 48; C,,, 2.4 ng) and a low affinity clone (clone 38; C,,, 13 ng) is shown. The samples were analyzed within the same experiment but were electrophoresed on separate gels. 400 mar KCI, respectively, and the amount of RXRa was reduced to 500 ng for rounds 4-7 and to 50 ng in the last round of selection. The selected RXRa-binding DNApool was ligated into the BamHI site of the plasmid vector pUTKAT3 (16). Individual bacterial clones were isolated and used to generate plasmid DNA. The sequences of all clones were determined by the dideoxynucleotide method using vector primers. PCR with primers A and B was used to amplify the contained 49-bp sequences, which were purified by PAGE and used in the studies described below.
Competition Assays-Four clones were randomly chosen, end-labeled with "' P, and used in an EMSA to test for RXRa monomer binding. The clone showing the strongest monomer band (clone 29) was chosen as the standard for further studies. The sequence of the random 18-mer within clone 29 is CCGGGGTCAACCAATTTC. Next, the affinity of RXRa monomers for all clones was assessed with a competition EMSA. To accomplish this, 20,000 cpm of "YP-labeled clone 29 DNA was incubated with 5 ng of RXRa plus graded doses of nonradiolabeled DNA from each clone. The amount of this competitor DNA required to reduce the intensity of the standard :' 2P clone 29 DNA,RXRa complex by 50% (C,,, measured by densitometry of autoradiograms) was taken as a measure of relative affinity. Competition assays were performed at least twice for all clones.
DNA Footprinting Analysis-Footprinting was performed on one high affinity clone, using methylation interference (17) to identify critical guanine residues and uracil interference footprinting (18) to identify critical thymine residues. For these studies, the DNA was radiolabeled by performing a PCR in which one of the two primers had been labeled at its 5'-end with [y-R*P]ATP. The PCR reactions were spiked with a small amount of dUTP (18). and the products were purified by PAGE and then methylated with dimethyl sulfate (17). The probes were incubated with R X R a and subjected to EMSA, and the protein.DNA complexes were eluted. Cleavage a t methyl guanines and uracils was accomplished by standard protocols (17,18), and the products were analyzed on a 10% polyacrylamide sequencing gel.
Mutational Analysis-The consensus sequence of the highest affinity clones was subjected to mutagenesis. A series of mutant oligonucleotides was made, and these were used in a C,, EMSA analysis with the clone 29 as a probe. The data were used in conjunction with the foot-  ZYansient ZYansfections-JEG-3 cells were grown in 90% Eagle's minimum essential medium plus 10% fetal bovine serum and were transfected using standard calcium phosphate precipitation (19). 'R-ansfections included a potentially RXR responsive chloramphenicol acetyltransferase-expressing plasmid, a human growth hormone (GH)-expressing plasmid (internal control), and an expression vector for mouse RXRa or P (3) (or the empty vector). Chloramphenicol acetyltransferase-expressing vectors were used a t a dose of 0.4 pg/60-mm Petri dish and were derived from pUTKAT3 (16) in which a basal thymidine kinase promoter drives reporter gene expression. Potential RXREs were ligated into a BanHI site 5' to the basal promoter. The sequences of these potential RXREs are as follows (top strands excluding BanHI compatible overhangs): R7, TGGGGTCAGGGGTCACGTGC; R8, TAG-GTCACAGGTCAC; and CRBP, CAGGTCACAGGTCACAGGTCACAG-TTCA. T W O forward oriented copies of R7 and R8 were ligated into pUTKAT3, and one forward oriented copy of CRBP was ligated, creating four potential RXR binding sites in each construct. The internal control vector pTKGH also uses a basal thymidine kinase promoter to express GH, and was transfected a t a dose of 1 pg/60-mm Petri dish. RXRa was expressed from the vector pSG5 (20) and RXRP from the vector pCDM (19) at 0.3 or 3 pg/60-mm Petri dish as noted. Empty vector pCDM was added to all transfections as "filler" plasmid to achieve a total of 6 pg of DNNtransfection. Cells in 60-mm Petri dishes were transfected in the presence of 10% charcoal-stripped fetal bovine serum and 100 nM dexamethasone. Cells were cultured f 1 p~ 9-cis-RA for 2 days prior to harvest. Chloramphenicol acetyltransferase and human GH were assayed as described previously (21). Ligand responsiveness was calculated as chloramphenicol acetyltransferasehuman GH for cells cultured with 94s-RA divided by chloramphenicol acetyltransferase/ human GH for cells cultured without 9-cis-RA. Transfection data are presented as the mean f S.E. for at least four independent paired transfections/condition.

Selection of High Affinity R X R a Monomer-binding DNA
Pool-The selection process utilized a pool of random doublestranded 18-bp oligonucleotides flanked by known sequences (total length, 49 bp). After 8 rounds of selection using EMSA and PCR amplification of gel eluted DNA, a DNA pool that bound RxRa with high affinity was obtained (Fig. 1).
The RxRa monomer-binding DNA pool was subcloned, and 19 different clones were studied. As shown in Tables I and 11,ll of these 19 clones contained the sequence GGGGTCA. However, neither the position nor orientation of this heptamer within the 18-bp random sequence was conserved nor was there conservation of sequence outside the heptamer. In EMSA competition assays, the clones containing the heptamer GGGGTCA showed the highest affinity for RXRa with a C,, of 3.4 r 0.3 ng (mean 2 S.E.); the other 8 clones including those containing the idealized hexamer AGGTCA had lower affinities with C,, values ranging from 7.8 to 12.7 ng (Table I1 and Fig. 2).
Mutational Analysis-To confirm the above results, a series of mutant oligonucleotides was synthesized, and their affinities were tested by EMSA competition. Changing any of the 4 guanines to thymines within the heptamer significantly decreased the affinity for R x R a (Fig. 3). Thus, the GGGGTCA-containing competitor oligonucleotide had a C,, of 0.6 ng (lanes 2-41, whereas the C,, values for oligonucleotides with mutations at the first, second, or third plus fourth guanines were 4, >8, and >8 ng, respectively (lanes 6- 10, 11-14, and 16-19). However, mutation of the two nucleotides immediately 3' to the heptamer did not affect the affinity of RxRa binding (data not shown). R7 is an oligonucleotide containing a DR+1 of GGGTCA(the first guanine of the 3"heptamer is considered to be the one-nucleotide spacer), and R8 contains a DR+1 of the hexamer AGGGTCA. Radiolabeled probes were incubated with 5 ng of RXRu plus the indicated amounts of competitor DNA(actin signifies a nonspe-M and D indicate the locations of R X R a cific competitor) and analyzed by EMSA.

RXRa:
Competitor: [32P]DNA: ' V DNA Footprinting-DNA footprinting analysis was performed using a high affinity clone containing the conserved sequence GGGGTCA. Methylation interference was used to identify critical guanine residues, and uracil interference was used to identify critical thymine residues. As depicted in Fig. 4, when any of the guanine residues from either strand in the heptamer sequence was methylated, RXRa binding was impaired. In a similar way, loss of a thymine 5-methyl group by changing the bottom strand thymine to uracil within the heptamer interfered with RXRa binding. No footprinting was noted outside the conserved heptamer. These data, combined with the random selection results and mutational analysis, indicate that the heptamer GGGGTCA represents the highest affinity RXRa binding sequence.
fiunsient fiansfections-To test the function of the optimal RXR binding site, transient transfections were performed in JEG-3 cells. Potential response elements were ligated upstream of a basal thymidine kinase promoter driving expression of chloramphenicol acetyltransferase. In particular, we wished to determine whether the heptamer GGGGTCA is a stronger RXRE than the idealized hexamer AGGTCA. Initially we studied both GGGGTCA and AGGTCA a s potential singlesite response elements with co-transfected RXRa or RXRP.
However, neither sequence supported 9-cis-RA induction of chloramphenicol acetyltransferase (data not shown), indicating that a single RXR binding site is unable to function a s a response element under these conditions. Therefore we studied multiple copies of RXR binding sites. Since it had previously been shown that the optimal spacing for an RXRE is a DR+1 (14), we constructed oligonucleotides that contained a DR+1 of AGGTCA (R8) or of GGGGTCA (R7) (in the latter case, the first G of the 3"half-site functions as the spacer, since the sequence is a heptamer rather than a hexamer). EMSAs show RXRa monomers and homodimers bind with higher affinity to R7 than to R8 (Fig. 5). Two copies of either the AGGTCA DR+1 (R8AA) or the GGGGTCA DR+1 (R7AA) were ligated into PUT-KAT3 and tested for 9-cis-RA induction. Transient transfection of J E G S cells with RXRa demonstrated that R7AA is -6-fold more potent an RXRE than is R8AA (Fig. 6A). Similar results were obtained when RXRP was used instead of RXRa; the 9-cis-RA induction of chloramphenicol acetyltransferase activity was -4-fold greater for R7AA than for R8AA (Fig. 6B). To confirm these data, a separate series of transfections was performed comparing R7AAto the RXRE of the rat cellular retinol binding protein I1 (CRBPII) gene (14). This natural RXRE contains four AGGTCA-like half-sites arranged as a series of DR+ls (Table   111). Consistent with the previous data, transfection with RXRa demonstrated that 9-cis-RA induction of chloramphenicol acetyltransferase was 3.7-fold greater for R7AA than for CRBP (Fig. 6C). Thus these data demonstrate that not only does RXRa bind to GGGGTCA better than to AGGTCA, but also the heptamer is more active as an RXRE half-site. DISCUSSION RXRs are able to heterodimerize with TRs, RARs, and VDR, increasing both DNA binding and transcriptional activation (5)(6)(7)(8)(9) and suggesting a diversity of RXR-dependent signaling pathways. These heterodimers are present in solution and are further stabilized by binding to their DNA response elements. In addition to their role in heterodimeric complexes, RXRs can bind as monomers and homodimers to DNA. The natural ligand for RXR is 9-cis-RA (10,11). Thus, 9-cis-RA induction using RXR homodimers defines a second retinoid signaling pathway distinct from that using all-trans-RA and -M s .
It was assumed that the above receptors recognize the hexamer AGGTCA as an optimal binding sequence since they have identical P boxes. The P box is a region located at the base of the first zinc finger that is critical for DNA half-site recognition specificity (22). However, by using a nonbiased approach, we found that the optimal DNA binding site for R X R a is the heptamer GGGGTCA. This sequence binds RXRa with higher affinity and also gives significantly higher transcriptional activation in transient transfection studies than does the hexamer AGGTCA. With a similar approach, it was shown that the optimal TRa monomer binding site is the octamer TAAGGTCA (23). These results would indicate that regions in the receptors in addition to the P box are important for half-site specific recognition. Indeed it has been shown that two other regions, denoted the T and A boxes and located 3' to the second zinc finger, play an important role in the protein.DNA interaction. The A box is important for recognition of the sequence AA at the 5'-end of the core motif for the NGFI-B/nur77 orphan receptor (24). Recently by using a (NG)GGTCA motif as a half-site, it was noted that the T box for RXR and RAR and the Abox for TR are important for binding to monomer sites (25).
A single copy of the optimal R X R a monomer binding sequence was unable to function as an RXRE when placed 5' to a basal thymidine kinase promoter. This is in contrast to the optimal TRa monomer binding sequence, which is a functional T3 response element, when placed a s a single copy 5' to the same promoter (23). Furthermore, certain other members of the erbA superfamily, such a s NGFI-B (26) and Rev-erhA (27) appear to activate transcription from monomer response elements. The reason RXRa is unable to accomplish this is not known, but it may relate to structural differences in transactivation domains or other conformational differences.
Only a few RXREs have been characterized, including those found in region I1 of the mouse major histocompatibility class I gene locus (l), the rat CRBPII gene (14), section A of the apolipoprotein AI gene (28), and the hepatitis B virus enhancer I (HBVGB element) (29). Analysis of these 4 RXREs has revealed 10 half-sites (Table 111). The perfect GGGGTCA heptamer sequence is found in only one of these 10 half-sites, located in the apolipoprotein AI RXRE. This is not contradictory to our data since half-site sequence should not be considered in isolation. for at lease four independent transfections. b, similar to a, except FZRP was used in place of m a . C, similar to a, except R7AA was tested versus the RXRE from the rat CRBPII gene, which contains four AGGTCA-like half-sites arranged as a series of DR+ls.
The total number of half-sites, their arrangement within the response element, the similarity of each half-site to the optimal sequence, and the overall DNA context of the response element all will contribute to the inducibility and function of the response element. For example the CRBPII RXRE lacks a perfect half-site, but it contains four imperfect sites arranged as a series of DR+1 elements. Nevertheless, in a transfection analysis, the CRBPII RXRE was less potent than were two copies of the GGGGTCA DR+1. In summary, our results indicate that the optimal monomer binding site for RXRa is GGGGTCA. It is not known if this heptamer sequence also is the optimal binding site for the RXR isoforms p and y. Functional differences have been described among these isoforms; for example, RXRa and y are more potent than p in activating transcription of the CRBPII gene (4).
Therefore it is possible that the isoforms may have slightly different optimal binding sequences.
Our data indicate that the optimal DNA half-site for RXRa binding differs from that for TRa binding (TAAGGTCA) (23). In addition, the optimal binding element for VDR appears to be GGTTGCA (30) or (MGIGGTGA (311, and evidence suggests that the optimal element for R A R a is AGAGGTCA (32). Thus, each of these receptors may have its own unique high affinity binding sequence. These observations suggest that heterodimers of RXR with TR, VDR, or RAR may prefer to bind to imperfect direct repeats with the appropriate spacing. For example, RXR.TR heterodimers may bind best to the sequence GGGGTCANNTAAGGTCA, and this sequence may be activated preferentially by T3-occupied RXR.TR heterodimers as opposed to TR homodimers. In contrast, we speculate that a perfect direct repeat of a receptor's optimal monomer binding site (with appropriate spacing) might be activated preferentially by the receptor homodimer (e.g. TAAGGTCANNTAAG-GTCA for TR homodimers). In addition, other erbA superfamily members may form functional heterodimers with TR, VDR, or RAR on response elements that contain half-sites of appropriate sequence and spacing. The details of the importance of half-site sequence in receptor function are just beginning to be explored.