Identification of the DNA Binding Specificity and Potential Target Genes for the Farnesoid X-activated Receptor*

, The farnesoid X-activated receptor (FXR; NR1H4) is a member of the nuclear hormone receptor superfamily and functions as a heterodimer with the 9- cis -retinoic acid receptor (RXR). In order to determine the optimal DNA binding sequence for the FXR/RXR heterodimer, we have utilized the selected and amplified binding se- quence imprinting technique. This technique identified a number of related sequences that interacted with FXR/ RXR in vitro . The consensus sequence contained an inverted repeat of the sequence AGGTCA with a 1-base pair spacing (IR-1). This sequence was shown to be a high affinity binding site for FXR/RXR in vitro and to confer ligand-dependent transcriptional activation by FXR/RXR to a heterologous promoter. Electrophoretic mobility shift assays and transient transfection assays were used to investigate the importance of the core half-site sequences, spacing nucleotide, flanking sequences, and orientation and spacing of the core half-sites on DNA binding and ligand-dependent transcriptional activation by FXR/RXR. These studies demonstrated that the FXR/RXR heterodimer binds to the consensus IR-1 sequence with the highest affinity, although FXR/RXR can bind to and activate through a variety of elements including IR-1 elements with changes in the core half-site sequence, spacing nucleotide, and flanking nucleotides. In addition, FXR/RXR can bind to and transactivate served as a marker for migration of the FXR z hRXR a z DNA complex. The FXR z hRXR a z DNA complex was excised based on migration of the posi- tive control. The SAAB-selected oligonucleotides (S.O.s) were eluted in 0.5 ml of 0.5 M ammonium acetate, 10 m M MgCl 2 , 1 m M EDTA, and 0.1% SDS at 37 °C for 3 h. Glycogen (10 m g) was added, and the S.O.s were extracted twice with phenol:chloroform:isoamyl alcohol (25:24:1) and precipitated with ethanol. The S.O.s were resuspended in H 2 O and then reprecipitated using sodium acetate and ethanol. Approximately one- fifth of the S.O. pool was amplified by PCR using standard conditions.

Nuclear hormone receptors are ligand-activated transcription factors that belong to a superfamily consisting of over 150 different members (reviewed in Refs. [1][2][3]. These receptors elicit their actions by binding to hormone response elements (HREs) 1 in the promoters of target genes and regulating transcription in response to lipophilic ligands. HREs are typically composed of two 6-base pair half-sites that may be arranged as direct, inverted, or everted repeats (2)(3)(4). The HRE for a given receptor is determined by the nucleotide sequence of the halfsites, the orientation and spacing of those half-sites, and the nucleotide sequence flanking the half-sites (2)(3)(4). The nuclear hormone receptor superfamily can be divided into the following two major classes: type I or classic steroid receptors (which include estrogen, progesterone, glucocorticoid, and mineralocorticoid receptors) and type II or non-steroid receptors (which include the retinoic acid, retinoid X, thyroid hormone, and vitamin D receptors) (1)(2)(3). Type I receptors typically bind as homodimers to HREs that are inverted repeats separated by three nucleotides (IR-3) (3,4). Type II receptors show a larger variety of modes of DNA binding. Most type II receptors bind DNA as a heterodimer with the 9-cis-retinoic acid receptor (RXR) (1)(2)(3)(4). HREs for these heterodimers are generally direct or inverted repeats with between 0 and 5 base pair spacing (0 to 5 rule) (5)(6)(7). Some type II receptors bind DNA as monomers (e.g. NGFI-B) (8,9) or homodimers (e.g. RXR) (1)(2)(3)(4). The HRE recognized by a monomer consists of a single half-site with additional sequence requirements upstream of the half-site called the 5Ј-extended half-site (8,9). HREs that are recognized by homodimers are typically inverted or direct repeats (10 -12).
When a nuclear hormone receptor is bound to an HRE, it may activate or repress transcription depending on the presence of ligand, cell type, promoter, response element, or other signals. Recent studies demonstrate that HREs contain "signaling information" that the receptor must "interpret" to determine the activity of a receptor bound to that site. For example, the glucocorticoid receptor (GR) can activate transcription when bound to certain glucocorticoid response elements (GREs) or repress transcription when bound to other GREs (Ref. 13 and references therein). Yamamoto and co-workers (13) iden-tified a mutant GR (K461A) that activated transcription when bound to all GREs tested, implying that this residue was critical for interpreting the signal from the GRE. Another example of context-dependent activation is the liver-X receptor (LXR). LXR activates transcription when bound as a heterodimer with RXR to a direct repeat of the sequence AGTTCA spaced by 4 base pairs (DR4T element or LXR element) (14,15). However, LXR/RXR is unable to activate through a DR4 element with the sequence AGGTCA (DR4G element), even though LXR/RXR binds to DR4T and DR4G with similar affinities (14,15). Thus, HREs represent not only a specific binding site for nuclear receptors but can also significantly affect the function of the receptor.
The FXR (NR1H4) (16) was originally isolated from a rat liver cDNA library by degenerate PCR, utilizing primers to the highly conserved DNA binding domain of nuclear hormone receptors (17). The FXR cDNA was found to be homologous (94% identity) to a previously identified mouse receptor called RXR-interacting protein 14 (RIP14) (17,18). These receptors are expressed in liver, kidney, gut, and adrenal cortex of adult rats and mice and in several embryonic tissues (17,18). Sequence comparison indicates that FXR is closely related to the ecdysone receptor (EcR) from Drosophila, particularly in the DNA-binding domain where they share 81% identity (17). The EcR is known to function as a heterodimer with ultraspiracle (USP), the Drosophila homolog of RXR (19 -21). The EcR/USP heterodimer binds to and transactivates through an EcR element (EcRE) that consists of an inverted repeat with 1-base pair spacing (IR-1) (19 -21). The Drosophila heat shock protein 27 (hsp27) promoter contains an IR-1 that functions as an EcRE (22). Based on homology between FXR (or RIP14) and EcR, it was shown that FXR/RXR or RIP14/RXR bound to the hsp27-EcRE (17,18). RIP14/RXR was also shown to bind to several other elements including the retinoic acid response element from the hRAR␤2 promoter (an imperfect direct repeat with 5-base pair spacing; DR-5), synthetic DR-2, DR-4, and DR-5 elements (18); however, it is not known whether these various motifs also function as positive or negative transcriptional elements involved in the regulation of gene expression by FXR. A detailed investigation of the DNA sequences necessary for binding and activation of transcription by FXR/RXR has not been reported. FXR/RXR was originally shown to activate reporter genes with multiple copies of the EcRE upstream of a heterologous promoter in response to the isoprenoid farnesol or the related insect isoprenoid Juvenille Hormone III. Based on these observations, the receptor was named the farnesoid X-activated receptor (17). Surprisingly, RIP14 was not activated by farnesol (23). In contrast, both FXR and RIP14 can be activated by all-trans-retinoic acid or a synthetic retinoid (TTNPB) (23). More recently, a number of bile acids, including chenodeoxycholic acid (CDCA), have been shown to bind to and potently transactivate FXR and RIP14 (24 -26). Neither farnesol, Juvenille Hormone III, nor all-trans-retinoic acid have been shown to bind directly to either FXR or RIP14, although TTNPB has been shown to displace CDCA from FXR (24). The FXR/RXR heterodimer can also be activated by RXR-specific ligands (including 9-cis-retinoic acid) through the RXR moiety (17,18,23).
Here we report the identification of high affinity binding sites for FXR using an unbiased in vitro selection technique. We identified an IR-1 element as a high affinity binding site and functional response element for FXR/RXR. In addition, a number of non-consensus IR-1 and DR sequences were shown to function as response elements for FXR/RXR when placed upstream of a minimal promoter. We identified naturally occurring IR-1 elements in the promoters of several genes that both bind FXR/RXR in vitro and confer transcriptional activation by FXR ligands in vivo. These results suggest that FXR may regulate expression of these genes. Consistent with this hypothesis, we demonstrate that FXR ligands increase the mRNA levels of at least one of these genes and that overexpression of FXR further increases these levels. These studies will assist in the identification of genes and pathways regulated by FXR.

EXPERIMENTAL PROCEDURES
Plasmid Constructs-Full-length cDNA for FXR (pKS-OR2) was kindly provided by Dr. Cary Weinberger (NIEHS, National Institutes of Health). FXR (amino acids 111-469) was amplified by PCR using pKS-OR2 as the template and primers incorporating a BamHI site upstream and an XhoI site downstream of the indicated coding region. This fragment was subcloned into BamHI/XhoI-digested pRSET B (Invitrogen, Carlsbad, CA) to generate pRSET-FXR⌬110. Mammalian expression vectors for FXR (pCMX-FXR) and hRXR␣ (pCMX-hRXR␣) were kindly provided by Dr. Ron Evans (Salk Institute, La Jolla, CA). Full-length hRXR␣ was amplified by PCR using pCMX-hRXR␣ as the template and subcloned into pRSET B, as described above, to generate pRSET-hRXR␣.
Selection and Amplification of Binding Sequences (SAAB)-The SAAB technique was performed essentially as described (27). Briefly, a randomized single-stranded oligonucleotide template (SAAB template, 5Ј-agtggaattcgcgaagataaggtcannnnnnnnnnnggagaccagtaggggtaccaccc-3Ј) (see Fig. 1A), a 5Ј SAAB primer (5Ј-agtggaattcgcgaagataag-3Ј), and a 3Ј SAAB primer (5Ј-gggtggtacccctactgg-3Ј) were purchased (Life Technologies, Inc.). Double-stranded randomized templates were generated by annealing the 3Ј SAAB primer to the SAAB template and extending with Klenow fragment of DNA polymerase (Life Technologies, Inc.) at 37°C for 1 h. The double-stranded SAAB templates were separated from single-stranded oligonucleotides by electrophoresis on a 15% polyacrylamide gel, excised, and eluted in 0.5 ml of 0.5 M ammonium acetate, 10 mM MgCl 2 , 1 mM EDTA, and 0.1% SDS at 37°C for 12 h. Templates were then precipitated with ethanol, resuspended, and then reprecipitated using sodium acetate and ethanol. For first round selection, random templates were end-labeled with T4 polynucleotide kinase (New England Biolabs) and [␥-32 P]ATP (NEN Life Science Products). Labeled double-stranded templates were mixed with approximately 50 ng of FXR⌬110 and 25 ng of hRXR␣ proteins (described above) and incubated in binding buffer (10 mM HEPES, pH 7.6, 0.5 mM DTT, 40 mM NaCl, 2.5 mM MgCl 2 , 0.05% (v/v) Nonidet P-40, 10% (v/v) glycerol, 50 ng/l poly(dI-dC) (Amersham Pharmacia Biotech), 2.5 mg/ml non-fat milk) for 1 h at room temperature. The ratio of FXR⌬110 to hRXR␣ was 2:1 in order to favor formation of FXR/hRXR␣ heterodimers rather than hRXR␣ homodimers. DNA-protein complexes were separated on a 6% non-denaturing polyacrylamide gel at 4°C for 2 h. The gel was dried and visualized by autoradiography. An oligonucleotide probe containing a single EcRE, a known binding site for FXR/hRXR␣, served as a marker for migration of the FXR⅐hRXR␣⅐DNA complex. The FXR⅐hRXR␣⅐DNA complex was excised based on migration of the positive control. The SAAB-selected oligonucleotides (S.O.s) were eluted in 0.5 ml of 0.5 M ammonium acetate, 10 mM MgCl 2 , 1 mM EDTA, and 0.1% SDS at 37°C for 3 h. Glycogen (10 g) was added, and the S.O.s were extracted twice with phenol:chloroform:isoamyl alcohol (25:24:1) and precipitated with ethanol. The S.O.s were resuspended in H 2 O and then reprecipitated using sodium acetate and ethanol. Approximately onefifth of the S.O. pool was amplified by PCR using standard conditions. PCR products were purified on a 15% polyacrylamide gel as described above. These purified amplified templates were labeled for subsequent rounds of selection by PCR; approximately 5 ng of templates were amplified by one cycle of PCR with 100 ng each SAAB 5Ј primer and SAAB 3Ј primer, 30 Ci of [␣-32 P]dCTP, 0.25 mM each dATP, dTTP, and dGTP in standard PCR buffer. Unincorporated [␣-32 P]dCTP was removed by gel filtration on BioSpin 6 columns (Bio-Rad), and approximately 100 pg of labeled template was used for the next round of selection (EMSA). After three rounds of selection, selected oligonucleotides were subcloned by TA cloning into the pCR2.1-TOPO cloning vector (Invitrogen). Twenty-six clones were picked and sequenced using the T7 Sequenase version 2.0 kit (Amersham Pharmacia Biotech).
Construction of Reporter Plasmids-Double-stranded oligonucleotides (described above) were ligated into BamHI/BglII digested pTK-Luc. Plasmids were selected in which two copies of the binding site were oriented in the same direction. For pTK-2x(IR-1) constructs, the binding sites were oriented such that the extended half-site (tttt in consensus) was downstream of the IR-1. For pTK-2x(EcRE), -(PLTP), -(CPT II), and -(PNMT), binding sites were oriented as found in their natural promoters.
HepG2 cells were transiently transfected with 2 g of pTK-Luciferase reporter plasmids (described above), 1 g of pCMV-␤-galactosidase, 50 ng each of pCMX-FXR and pCMX-hRXR␣, and 3 g of pTKCIII (carrier plasmid) per 60-mm dish using the MBS Mammalian Transfection Kit (Stratagene) with minor modifications. Following transfection, cells were grown in medium containing 10% charcoal resin-stripped (super-stripped) FBS  Production of Stable Cell Lines-Full-length FXR was subcloned into an MSCV-based retrovirus upstream of an internal ribosome entry site and neomycin resistance gene. This plasmid (MSCV-FXR-neo) or the empty vector (MSCV-neo) was transfected into HEK 293T cells along with a A helper virus, and the supernatant, containing the replicationincompetent retrovirus, was then used to infect HepG2 cells. Stable cells were selected using 800 g/ml G418 sulfate (Geneticin®, Life Technologies, Inc.). Neomycin-resistant cells (at least 1 ϫ 10 4 colonies) were pooled, expanded, and tested for expression of FXR by Northern blot and by transient transfection with the pTK-2x(PLTP)-Luciferase construct.
RNA Isolation and Northern Blot Analyses-FAO, HepG2-FXR-neo, or HepG2-neo cells were placed in medium containing super-stripped FBS for 24 h prior to the addition of ligands. Cells were then incubated in the presence of ligand for 24 h, and total RNA was isolated using TRIzol reagent (Life Technologies, Inc.). Ten g of total RNA (per lane) were separated on 1% agarose/formaldehyde gels, transferred to nylon membranes, and fixed by cross-linking with UV light. Random primed [␣-32 P]dCTP-radiolabeled probes were generated using the Oligonucleotide labeling kit (Amersham Pharmacia Biotech). Membranes were hybridized using the QuikHyb Hybridization Solution (Stratagene) according to the manufacturer's instructions. Blots were normalized for differences in RNA loading by hybridization to a glyceraldehyde-3phosphate dehydrogenase probe or to the ribosomal protein 36B4 probe and analyzed by PhosphorImaging or densitometry (Molecular Dynamics).

Selection of High Affinity Binding
Sites for FXR-In order to determine the DNA sequences that represent high affinity binding sites for FXR, we utilized the SAAB technique (27). We synthesized a 60-base pair double-stranded oligonucleotide, which contained the canonical nuclear receptor half-site, AG-GTCA, upstream of 11 randomized base pairs and is referred to as the "SAAB template" (Fig. 1A). This oligonucleotide should allow for selection of either direct or inverted repeats with 0 -5-base pair spacing between half-sites. The SAAB template (Fig. 1A) was end-labeled and used in EMSA with recombinant, purified FXR and RXR proteins. An oligonucleotide containing the hsp27 EcRE sequence (EcRE-WT) was used as a positive control in the EMSA and allowed for identification of the FXR⅐RXR⅐DNA complex. Shifted DNA complexes were recovered from the gel, and the DNA was amplified by PCR. Following each round of selection, selected oligonucleotides were endlabeled with [ 32 P]ATP and used in EMSAs without or with FXR and/or RXR to test for enrichment of high affinity binding sites. The formation of specific complexes was largely dependent on the presence of both FXR and RXR, suggesting that FXR/RXR heterodimers were favored over FXR/FXR homodimers, RXR/ RXR homodimers, or monomers (data not shown). However, a minor shifted complex, presumably containing RXR/RXR homodimers, was formed in the absence of FXR (data not shown). We utilized an amino-terminal truncated FXR protein (Ϸ41 kDa) along with full-length RXR (Ϸ54 kDa), which allowed us to distinguish between FXR/RXR heterodimers, RXR/RXR or FXR/FXR homodimers, or monomers. The migration of the major protein-SAAB template complex was consistent with the formation of FXR/RXR heterodimers since the complex migrated in the gel to a position approximately equal to that of the positive control (FXR⅐RXR⅐EcRE-WT complex) (data not shown). Following three rounds of selection, selected oligonucleotides were subcloned, and 26 independent colonies were chosen and sequenced. The cloned sequences were aligned, and the frequency which given nucleotides were selected in the randomized 11 nucleotide region is shown in Fig. 1B. Analysis of the data indicates that the selected sequences contain a consensus that consists of an inverted repeat with a downstream half-site sequence AGGTCA or AGTTCA (reading 5Ј to 3Ј on the bottom strand) with 1-base pair spacing between half-sites (an IR-1 element) (Fig. 1B). In addition to selecting a consensus sequence in the downstream half-site (positions 2-7), the selected sequences show a preference for a T or A as the spacer nucleotide (position 1) and for a stretch of T nucleotides flanking the downstream half-site (positions 8 -11), referred to as the "extended half-site" (Fig. 1B). Investigation of the individual cloned sequences revealed that 23 of the 26 sequences were identical to or highly related to the consensus IR-1 element (Fig. 1C). One cloned sequence contained a direct repeat with 1-base pair spacing (DR-1 element) and one contained a direct repeat with 3-base pair spacing (DR-3 element) (Fig. 1C). One of the clones contained sequences with no obvious FXR/RXR-binding site (data not shown).
The IR-1 Consensus Sequence Functions as a High Affinity Binding Site for FXR/RXR-To determine if the IR-1 consensus sequence (IR-1-CON) selected by the SAAB technique is a high affinity binding site for FXR/RXR, we performed EMSAs with purified FXR and RXR proteins. Both FXR and RXR proteins were required to shift the IR-1-CON oligonucleotide, and migration of the FXR⅐RXR⅐IR-1-CON complex was identical to the migration of a FXR⅐RXR⅐EcRE-WT complex (Fig. 2D,  lanes 1 and 5, data not shown). In direct binding experiments, a higher percentage of the IR-1-CON labeled probe was shifted by FXR/RXR compared with the EcRE-WT (Fig. 2D, lane 1  versus 5). To determine the relative affinity of FXR/RXR for the IR-1-CON versus the EcRE-WT, competitive EMSAs were performed. Radiolabeled IR-1-CON was incubated with FXR and RXR proteins in the absence (Fig. 2A, lane 1) or presence of increasing amounts of unlabeled IR-1-CON oligonucleotide ( Fig. 2A, lanes 2-7) or EcRE-WT oligonucleotide ( Fig. 2A, lanes  8 -13). The IR-1-CON was an effective competitor, competing approximately 95% of the shifted probe at 250-fold molar excess, whereas the EcRE-WT was considerably less effective, competing only about 20% at the same concentration (Fig. 2, A  and B). When the EcRE-WT was used as the 32 P-labeled probe, unlabeled IR-1-CON was a much better competitor than unlabeled EcRE-WT (data not shown). Scatchard plot analysis of the data indicates that the IR-1-CON had a 10-fold higher affinity for FXR/RXR than the EcRE-WT (data not shown). These results demonstrate that the FXR/RXR heterodimer can bind to the IR-1-CON with high affinity.
Since the EcRE-WT oligonucleotide differs from the IR-1-CON oligonucleotide in both half-sites (Fig. 2C), as well as in flanking nucleotides, we tested whether the difference in affinity of FXR/RXR for the IR-1-CON versus the EcRE-WT was due to differences in the half-site sequences or in flanking nucleotides. Oligonucleotides were generated where either the 5Ј, 3Ј or both half-sites of the IR-1-CON oligonucleotide were mutated to the sequence of the EcRE-WT half-sites, respectively (Fig. 2C, IR-1-M7, -M8, and -M27). Conversely, half-sites of the EcRE-WT were mutated to consensus half-sites (Fig. 2C,  EcRE-M1, -M2, and -M3). These oligonucleotides were endlabeled with 32 P and used in EMSAs with purified FXR and RXR proteins. Mutation of either half-site of the IR-1-CON to half-site sequences from the EcRE-WT resulted in decreased formation of a shifted complex (Fig. 2D, lane 1 versus lanes 2  and 3). Mutation of both half-sites of the IR-1-CON to those of EcRE (IR-1-M27) resulted in a significant decrease in complex formation (Fig. 2D, lane 4 versus lane 1), to a level that was approximately equal to that obtained with the EcRE-WT (Fig.  2D, lane 4 versus lane 5). Conversely, mutation of either binding site of the EcRE-WT to the consensus sequence of the IR-1-CON resulted in an increase in the formation of a shifted complex, whereas mutation of both half-sites increased complex formation to near that of IR-1-CON (Fig. 2D, lane 5 versus  lanes 6 -8). These results demonstrate that the core half-site nucleotides, not the flanking regions, represent the primary determinant for complex formation.
The IR-1-CON Serves as a Functional Response Element for FXR/RXR-To determine if the IR-1-CON can function as a response element for FXR/RXR, two copies of the IR-1-CON oligonucleotide were cloned upstream of the TK minimal promoter and luciferase reporter gene (pTK-luciferase vector). Two copies of the binding sites were chosen to allow for a relatively high level of expression of the reporter and to limit the synergistic activation that can occur when multiple binding sites are used. As a positive control, two copies of the EcRE-WT oligonucleotide were also cloned into the pTK-luciferase vector. The reporter plasmids were cotransfected into HepG2 cells along with expression vectors for FXR and RXR. As expected, the EcRE-WT reporter was activated by ligands for FXR (CDCA or TTNPB) or RXR (LG10153) (Fig. 2E). Addition of both FXR-and RXR-specific activators resulted in additive levels of expression of the EcRE-WT reporter (Fig. 2E) in agreement with a previous report (19). The IR-1-CON reporter con-

FIG. 1. Selection of high affinity binding sites for the FXR/RXR heterodimer.
A, schematic representation of the SAAB template. High affinity binding sites for FXR/RXR were selected using the SAAB technique as described under "Experimental Procedures." The SAAB template consisted of a double-stranded oligonucleotide containing a consensus nuclear receptor half-site, AGGTCA, upstream of 11 randomized base pairs (designated positions 1-11). The SAAB template was 32 Pend-labeled and incubated with purified, recombinant, histidine-tagged FXR and RXR proteins as described under "Experimental Procedures." Protein-bound DNA complexes were separated from unbound DNA by non-denaturing gel electrophoresis. The selected (shifted) oligonucleotides were recovered from the gel, amplified by PCR, and used for subsequent rounds of selection. After 3 rounds of selection, selected oligonucleotides were subcloned. B, summary of SAAB-selected sequences. Twenty six independent clones were isolated and sequenced. The number of times a given nucleotide was selected and the consensus sequence at each position in the 11-base pair randomized region are shown. C, sequences of individual binding sites selected. The individual binding sites selected by SAAB are shown arranged according to the orientation and spacing of the half-sites. The core half-site sequences are shown in capital letters, and the nucleotides that differ from the consensus half-site sequence (AGGTCA) are boxed. Nucleotides outside of the half-sites are indicated by n. IR-1 represents inverted repeat with 1-base pair spacing between half-sites; DR-1 and DR-3 represent direct repeats with 1-or 3-base pair spacing, respectively. struct was also activated by ligands for either FXR or RXR (Fig.  2E). In the presence of both FXR-and RXR-specific ligands, additive levels of expression were observed (Fig. 2E). The IR-1-CON reporter construct was activated to a higher level than the EcRE-WT reporter (Fig. 2E), consistent with the higher affinity of FXR/RXR for the IR-1-CON versus the EcRE-WT ( Fig. 2A). Coexpression of FXR, but not RXR, was required for activation of either the IR-1-CON or the EcRE-WT reporter  CON (lanes 2-7) or EcRE-WT (lanes 8 -13) competitor oligonucleotide. Oligonucleotides used in this experiment were IR-1-CON, the IR-1-consensus sequence selected by SAAB technique (see Fig. 1), and EcRE-WT, the ecdysone response element from the hsp27 gene. B, graphical representation of the EMSA competition experiment shown in A. The radioactivity present in shifted protein-DNA complexes and free DNA probe (A) were quantitated by PhosphorImaging, and the values obtained for IR-1-CON (q) or EcRE (f) as competitor are shown. C, comparison of core half-site sequences for variants of IR-1-CON and EcRE-WT. The name and core half-site sequence for IR-1-CON, EcRE-WT, and mutant oligonucleotides are shown. For the EcRE-WT, the bottom strand is shown. Nucleotides that differ from the IR-1-CON are boxed. D, direct binding of FXR/RXR to the IR-1-CON, EcRE-WT, or the indicated mutant oligonucleotides. EMSA of purified FXR and RXR proteins incubated with 2.5 fmol of the indicated 32 P-labeled oligonucleotide under high stringency conditions, as described under "Experimental Procedures." E, comparison of transactivation by FXR/RXR conferred by IR-1-CON or EcRE. Duplicate plates HepG2 cells were transiently transfected with expression vectors for FXR and RXR, CMV-␤-galactosidase, and pTK-luciferase reporter constructs with two copies of either IR-1-CON (white bars) or EcRE-WT (black bars). Following transfection, cells were incubated with the indicated ligands or Me 2 SO vehicle for 24 h, harvested, and luciferase and ␤-galactosidase activities were determined. The results are representative of three independent experiments. ␤-Galactosidase activity was used to normalize for transfection efficiency. genes in HepG2 (data not shown). However, coexpression of RXR, together with FXR, further increased the activity of both reporter genes (data not shown). These results demonstrate that the IR-1-CON can function as a high affinity binding site and a functional response element for FXR/RXR.
Mutational Analysis of the IR-1-CON-Several non-consensus IR-1 elements, which differ from the IR-1-CON by only a single nucleotide in the downstream half-site, were selected by the SAAB technique (Figs. 1C and 3A). To determine if the IR-1-CON represents the highest affinity binding site for FXR/ RXR, we introduced single nucleotide mutations in the downstream half-site of the IR-1-CON. Some of these mutations were made to investigate non-consensus IR-1 sequences selected by SAAB (IR-1-M1, -M2, -M25, and -M26) (Figs. 1C and  3A). Other mutations were introduced to test whether specific sequences, not selected by SAAB, could also function as binding sites for FXR/RXR heterodimers (IR-1-M3, -M4, -M5, and -M6) (Fig. 3A). Each of the oligonucleotides shown in Fig. 3A were end-labeled and used in EMSAs with purified FXR and RXR proteins. Fig. 3B shows that FXR/RXR binds with high affinity to number of oligonucleotides containing non-consensus inverted repeats. Certain single point mutations resulted in a significant decrease in the formation of a shifted complex (Fig.  3B, lanes 6 and 9 versus lane 2). These results demonstrate that the SAAB technique resulted in isolation of sequences that are bound by FXR/RXR with a broad affinity range. Most of the sequences (22 of 26) identified by SAAB are high affinity FXR/ RXR-binding sites, although some were of lower affinity (Fig.  3B, IR-1-M25). In addition, some high affinity binding sites (IR-1-M3 and -M5, Fig. 3B) were not selected presumably because only 26 clones were analyzed (Fig. 1B). Nonetheless, these results demonstrate that the SAAB technique identified sequences that, in general, were of higher affinity than previously identified FXR/RXR-binding sites.
To test how mutations in the downstream half-site sequence of the IR-1-CON affect function as a response element for FXR/RXR, we generated luciferase constructs, each containing two copies of the indicated response element (IR-1-M1 through M6). The resulting reporter constructs were cotransfected into HepG2 cells along with expression vectors for FXR and RXR. The results demonstrate that the induced level of reporter gene activity, in the presence of the FXR ligand TTNPB, was M3 Ͼ M1 ϭ M5 Ͼ M2 Ͼ CON Ͼ M4 Ͼ M6 (Fig. 3C). The rank order of activation was similar to the order of DNA binding affinity (Fig. 3B).
Importance of the Spacer Nucleotide and Extended Half-site Sequence-In addition to selecting a specific sequence for the downstream half-site, the SAAB technique also selected specific nucleotides in the spacer nucleotide (T or A) and "extended" half-site (TTTT) (Fig. 1B). To investigate the importance of these regions on DNA binding by FXR/RXR, we generated oligonucleotides with mutations in these regions (Fig. 4A). EM-SAs showed a small decrease in binding affinity of FXR/RXR when the spacer nucleotide is changed from T to G (IR-1-M9) (Fig. 4B, lane 2 versus lane 1) or when the extended half-site was changed from TTTT to GGGG (IR-1-M14, -M19, or -M24) (Fig. 4B, lanes 1, 4 and 6 versus lanes 3, 5, and 7). These studies indicate that the spacer nucleotide and flanking nucleotides have a minor effect on the DNA binding affinity of FXR/RXR and together with previous results (Fig. 2D) demonstrate that the core half-sites are the critical determinants of FXR/RXR binding.
FXR/RXR Binds to and Activates through Direct Repeats-The SAAB technique selected two direct repeat elements (DR-1 and DR-3), in addition to the many inverted repeat elements (Fig. 1C). To test if other direct repeat elements could function as high affinity binding sites for FXR/RXR, EMSAs were performed using oligonucleotides with consensus DR-1 through DR-5 elements. All five DR elements were able to form a specific complex with FXR/RXR (Fig. 5A, lane 2-6). Competitive EMSAs indicated that the affinity of FXR/RXR for DR-4 and DR-5 was similar to that for EcRE (data not shown). Fig.   FIG. 3. Effect of changes in one half-site of the IR-1-CON on binding and transactivation by FXR/RXR. A, comparison of core half-site sequences of IR-1s used in this study. The name, core half-site sequence, and number of times in which that core half-site sequence was selected by the SAAB technique (see Fig. 1) is shown. NS refers to not selected by the SAAB technique. Nucleotides that differ from the IR-1-CON sequence are boxed. B, direct binding of FXR/RXR to various IR-1s. High stringency EMSAs were performed as described under "Experimental Procedures." Each oligonucleotide was labeled to approximately equal specific activity. C, transactivation by FXR/RXR through various IR-1 sequences. Reporter plasmids, containing two copies of the indicated IR-1 binding site (see A) subcloned into pTKluciferase, were transiently transfected into HepG2 cells in duplicate dishes together with expression vectors for FXR, RXR, and ␤-galactosidase. Following transfection, cells were incubated for 24 h with either 5B shows that reporter genes containing two copies of DR-4 or DR-5 were activated by ligands TTNPB or CDCA in the presence of coexpressed FXR and RXR. Activation required coexpression of FXR, but maximal activation was dependent on the coexpression of both FXR and RXR (Fig. 5C). The slight activation of the DR-5 element by TTNPB in the absence of coexpressed FXR (Fig. 5C, panels 5 and 7) likely results from the activation of endogenous RAR/RXR. The latter heterodimer (RAR/RXR) is known to bind to a DR-5 element and to be activated by TTNPB (28,29). Taken together, these results demonstrate that FXR/RXR can bind to and activate expression through direct repeat elements.
Identification of Natural Response Elements for FXR/ RXR-We performed a data base search using the IR-1-CON sequence and identified several candidate genes that contain sequences in their proximal promoters that correspond closely to the consensus IR-1 sequence. Three of these genes, phospholipid transfer protein (PLTP) (30), phenylethanolamine N-methyltransferase (PNMT) (31, 32), and carnitine palmitoyltransferase II (CPT II) (33), were chosen for further investigation. These genes are expressed in liver, kidney, adrenals, and/or gut (30 -37), tissues where FXR expression has been observed (17,18). The sequences from the promoters of these genes that correspond closely to the consensus IR-1 element are shown in Fig. 6A. To determine if FXR/RXR could bind to these elements, oligonucleotides were generated which correspond to the IR-1 elements from the PLTP, PNMT, and CPT II genes. Each of these elements were specifically bound by FXR/RXR with affinities that are similar to that of IR-1-CON (Fig. 6B,  lanes 1-4). These elements also function as response elements for FXR/RXR when placed upstream of a minimal promoter and luciferase reporter gene (Fig. 6C). Interestingly, each of these reporters was activated by either TTNPB or CDCA to higher levels than the IR-1-CON reporter (Fig. 6C). The PLTP reporter was consistently activated to the highest level by TT-NPB (35-fold) or CDCA (60-fold) (Fig. 6C). The PNMT reporter was also highly activated by TTNPB (15-fold) or CDCA (31fold) (Fig. 6C). The IR-1 element from the PLTP promoter has a sequence of GGGTCA in both half-sites, whereas the PNMT element has this sequence in one half-site and the consensus AGGTCA in the other half-site (Fig. 6A). We noted earlier that the IR-1-M3 reporter, which has this same downstream halfsite sequence (GGGTCA), was the most highly activated reporter (Fig. 3C). Taken together, these findings may indicate a preference for a G nucleotide in the 1st position of the half-site for high level activation of IR-1 elements by FXR/RXR. These results demonstrate that FXR/RXR can bind to and activate expression through naturally occurring IR-1 elements contained within the promoters of the PLTP, PNMT, and CPT II genes.
Regulation of the Endogenous PLTP mRNA Levels by FXR/ RXR-Based on its role in lipid transport and lipoprotein metabolism, PLTP appeared to be the most promising candidate gene and thus was chosen for a more detailed investigation. To determine if FXR/RXR could regulate expression of the endogenous PLTP gene, Northern blots were performed on RNA derived from two different cell types that express FXR. Northern blot analysis demonstrated that FAO cells, a hepatocytederived cell line, express endogenous FXR (data not shown). When FAO cells were treated with either of two different FXR ligands, the PLTP mRNA level was increased (1.7-2.5-fold) (Fig. 7A). To demonstrate that FXR mediates activation of PLTP, we generated cells overexpressing FXR by infecting HepG2 cells with a virus encoding FXR and the neomycin resistance gene (Fig. 7B, FXR) or the neomycin resistance gene only (Fig. 7B, neo). Neomycin-resistant cells were selected and then treated with CDCA or Me 2 SO for 24 h. The PLTP mRNA levels showed a dramatic increase in cells infected with FXR when compared with cells infected with the empty vector (Fig.  7B). Furthermore, PLTP mRNA levels were induced by treatment with the FXR ligand CDCA (Fig. 7B). These results demonstrate that FXR/RXR regulates the level of expression of the endogenous PLTP mRNA. DISCUSSION No detailed study of the DNA binding specificity of FXR has been reported to date, even though such information is crucial for definitive analyses of FXR target genes. In this study, we utilized an unbiased, PCR-based selection technique (the SAAB technique) (27) to identify high affinity binding sites for the FXR/RXR heterodimer. The majority of the selected sequences contained elements that were identical to or highly related to the IR-1 consensus sequence. These findings are consistent with previous reports that demonstrate that FXR/ RXR (or RIP14/RXR) can bind to an imperfect IR-1 (the EcRE-WT) or to an idealized IR-1 element (17,18). The nucleotide sequences selected for high affinity binding by FXR/RXR (IR-1 consensus) in this study are similar to sequences selected (by FIG. 4. The effect of the spacing nucleotide and the nucleotides that comprise the extended half-site on binding of FXR/ RXR. A, comparison of spacing nucleotide, downstream half-site sequence, and extended half-site sequence for the indicated IR-1s used in these experiments. Each binding site is an IR-1 that differs from the IR-1-CON oligonucleotide only where boxed. S and NS indicate selected or not selected by the SAAB technique ( Fig. 1), respectively. B, direct binding of FXR/RXR to IR-1s that differ from the IR-1-CON sequence in the spacing nucleotide or the extended half-site as shown by the boxed nucleotides. High stringency EMSAs were performed as described under "Experimental Procedures." the SAAB technique) by the related EcR⅐USP complex from Drosophila (38) and sequences shown to bind Mosquito EcR/ USP with high affinity (39,40).
By using direct binding and competitive EMSAs, we demonstrate that the consensus IR-1 represents the highest affinity FXR/RXR-binding site known to date. Our studies show that FXR/RXR binds to the consensus IR-1 with greater than 10-fold higher affinity than to the previously identified EcRE-WT sequence (Fig. 2, A and B) or to DR elements (Fig. 5A). More significantly, the consensus IR-1 was shown to function as a response element for FXR/RXR and to be more effective than the EcRE-WT in conferring responsiveness to ligands for FXR (CDCA or TTNPB) or RXR (LG10153) (Fig. 2E). The level of transactivation by FXR/RXR through various IR-1 elements was shown to correlate with DNA binding affinity, determined from in vitro EMSAs (Fig. 3, B and C, data not shown).
Additional studies reveal that FXR/RXR can bind to and transactivate through various IR-1 elements. The highest level of transactivation by FXR/RXR was mediated by an IR-1 where one half-site had the sequence GGGTCA (IR-1-M3) (Fig. 3C). Interestingly, the optimal monomer binding sequence for RXR has been shown to be GGGGTCA (41), and the optimal RXR homodimer binding site has been shown to be a DR-1 element with one half-site of the sequence GGGTCA (42). This may suggest that the FXR/RXR heterodimer favors an orientation on the IR-1-M3 element where the RXR moiety is bound to the GGGTCA half-site, and this orientation may result in a greater level of activation of the FXR/RXR heterodimer.
We also demonstrate that FXR/RXR binds to DR elements. These results are consistent with earlier reports that utilized RIP14, the mouse homolog of FXR, and RXR (18) or EcR/USP (38 -40). However, it was not known if DR elements could function as response elements for RIP14/RXR or FXR/RXR. Importantly, we now demonstrate that FXR/RXR can activate expression of reporter genes through DR-4 and DR-5 elements (Fig. 5). DR-4 and DR-5 elements have been described as response elements for several nuclear receptor heterodimers, including LXR/RXR and TR/RXR (DR-4) and RAR/RXR (DR-5) (5,6,15,16,43,44). Therefore, activation of genes through DR-4 or DR-5 elements by FXR/RXR will likely depend, in part, on the relative concentration of these different receptors in the cell and their relative binding affinities for these elements.
The identification of DNA sequences that both are recognized by FXR/RXR in vitro and serve as response elements in vivo has allowed us to identify three putative target genes that may be regulated by FXR. Each of these genes contained an IR-1 element in their proximal promoter region that was bound by FXR/RXR and when placed upstream of a heterologous promoter was able to confer ligand-dependent regulation by FXR/RXR. One of these potential targets was in the proximal promoter of the human phospholipid transfer protein (PLTP) gene (30). Reporter genes containing two copies of the PLTP IR-1 element (GGGTCAgTGACCC) are the most responsive to induction by FXR ligands tested to date (Fig. 6). Importantly, we demonstrate that in FAO and HepG2 cells, endogenous PLTP mRNA levels are increased in response to FXR ligands (Fig. 7). In addition, in HepG2 cells, PLTP mRNA levels are further increased in response to overexpression of FXR (Fig.  7B). Consistent with these observations, preliminary studies have shown that the natural PLTP promoter can be regulated by FXR/RXR or RIP14/RXR. 2 Thus, PLTP represents only the second known target gene for activation by FXR (25). PLTP mediates transfer of phospholipids and cholesterol from triglyceride-rich lipoproteins to high density lipoproteins (reviewed in Refs. 45 and 46) (47). Taken together, these findings suggest that FXR may play a role in the regulation of lipoprotein metabolism by mediating expression of PLTP.
We identified other functional FXR/RXR-binding sites in the proximal promoters of the CPT II gene (33) and the PNMT gene (31). CPT II is important in the entry of acyl groups into the mitochondrion for ␤-oxidation of fatty acids (reviewed in Ref. 48). PNMT is primarily expressed in the adrenal medulla and encodes the final enzyme in the synthesis of epinephrine. Additional studies are required to demonstrate that these target genes are regulated by FXR in vivo. Our studies, defining the DNA binding specificity for FXR/RXR, should both assist in the identification of additional genes and pathways regulated by FXR, bile acids, and retinoids. FIG. 6. Identification of genes that contain inverted repeats in their proximal promoter that can serve as functional response elements for FXR/RXR. A, comparison of sequences of IR-1 elements found in four natural promoters with IR-1-CON. Nucleotides that differ from the IR-1-CON sequence are boxed. B, direct binding of FXR/RXR to IR-1 elements found in the proximal promoters of the CPT II, PLTP, and PNMT genes. EMSAs were performed as described under "Experimental Procedures" using 2.5 fmol of the indicated 32 P-labeled oligonucleotide and purified FXR and RXR. Shifted protein-DNA complexes and free DNA probe are indicated. C, IR-1 elements from natural promoters confer transactivation by FXR/RXR. Reporter plasmids containing two copies of the IR-1 element taken from the indicated gene (see A) were transiently transfected into duplicate dishes of HepG2 cells along with CMV-␤-galactosidase and expression vectors for FXR and RXR. Cells were then treated with the indicated ligands for 24 h and harvested, and luciferase and ␤-galactosidase activities were determined. The normalized luciferase units, obtained as described under "Experimental Procedures," are shown. The results are representative of three independent experiments.  1 and 2) in duplicate. Northern blots were performed using a 1.3-kilobase pair SacI fragment of the PLTP cDNA or glyceraldehyde-3-phosphate dehydrogenase (GAPDH) as a loading control. The blots were quantitated by PhosphorImaging or densitometry. Relative expression values are the averages of duplicate lanes following normalization for differences in RNA loading. B, PLTP mRNA levels are dramatically increased by overexpression of FXR. Stably infected HepG2 cells overexpressing either FXR and the neomycin resistance gene (FXR, lanes 3 and 4) or the neomycin resistance gene only (neo, lanes 1 and 2) were treated with 100 M CDCA (lanes 2 and 4) or Me 2 SO vehicle only (lanes 1 and 3). Northern blots were probed with a 1.3kilobase pair SacI fragment of the PLTP cDNA or 36B4 as a loading control. Quantitation was performed by PhosphorImaging, and relative expression values are normalized for RNA loading.