Human RXR alpha and human RAR alpha was amplified by the polymerase chain reaction (PCR) from Uterus cDNA (PCR Ready-cDNA, Maxim Biotech, Inc., USA) and pituitary Cap site cDNA (Wako Pure Chemical Industries, Ltd., Japan.), respectively. These cDNA fragments were cloned into pGADT7 (CLONTECH, USA) and reamplified by PCR with primers (Table 1, RXR-5', RXR-3', RAR-5) to generate the restriction sites for sub-cloning. The Hind III fragment including nuclear localization signal (NLS) and GAL4 activation domain (GAL4 AD) was removed from pGADT7andthe cDNAs was inserted into the plasmid, respectively. The resulting plasmid was named pADH1_RXR and pADH1_RAR. The fragment including NLS and GAL4AD was amplified from pGADT7 by PCR and cloned. The fragment was inserted Bgl II sites at N-terminal end of RAR in pADH1_RAR and confirmed by sequencing. Resulting plasmid was named pADH1_NLS_GAL4AD_RAR. Foot-and-mouth disease virus (FMDV) 2A peptide sequence was amplified by annealing synthetic oligonucleotides (Table 1, FMDV_2A-F and FMDV_2A-R) and reamplified by PCR with primers (Table 1, FMDV_2A-5' and FMDV_2A-3') to generate the restriction sites. Then FMDV 2A fragment and NLS_GAL4AD_RAR was tandemly inserted at C-terminal end of RXR in pADH1_RXR. The resulting plasmid was named pADH1_RXR_2A_NLS_GAL4AD_RAR.pSUR (Gene Bank AB425277) was constructed previously and used as a reporter in the modified yeast one-hybrid system [8] . The positive control of a modified one-hybrid assay were constructed by inserting eight copies of the mouse cellular retinoic acid binding protein II (mCRABPII) RARE (Table 1, mCRABP2_F and mCRABP2_R) [9] at the upstream of the SPO13 promoter of pSUR. To generate the restriction sites and Kozak sequence for sub-cloning, N-terminal end side of RXR was amplified by PCR. The resulting PCR product was digested with Hind III and Sma I. C-terminal end side of RXR was obtainedby digesting with Sma I and XhoI. Both fragments were inserted into Hind III and XhoI sites pSP64 Poly (A) (Promega, USA) for in vitro transcription/translation. The cDNA for human RAR alpha was also inserted into the Sal I and the Xba I sites pSP64 Poly (A).

The human genomic library for a modified yeast one-hybrid assay was generated as previously described [8] .

The expression vector by placing FMDV 2A peptide sequence between RXR and RAR (pADH1_RXR_2A_ NLS-GAL4_RAR) was transformed into the 5FOA-selected yeast cells. The obtained transformants were grown on synthetic complete media lacking leucine, tryptophan, and uracil but containing 25 µg/ml 6-azauracil, 10 µM LG100754 (Wako Pure Chemical Industries, Ltd., Japan.), and 1 µM TTNPB (Funakoshi, Tokyo, Japan) for 3 weeks. Human genomic fragments were recovered from the positive colonies by colony-direct PCR with primers corresponding to the vector sequences (Table 1, pSRU-K1-F and pSRU-K1-R). The PCR fragments were directly sequenced with the forward and reverse primers (Table 1, pSRU-K2-F and pSRU-K2-R).

The RARE of rat rat cellular retinol binding protein type II (rCRBPII) were labeled with Cy5 and used as a probe (Table 1, rCRBPII-F and rCRBPII-R) [10] . Synthesis of human RAR and RXR proteins was performed with the TNT SP6 High Yield System (Promega, USA). In vitro synthesized RAR and RXR proteins were mixed with 500 ng of Calf thymus DNA (Invitrogen, USA) and 0.25 pmol of labeled oligonucleotide at 4˚C. In competition experiments, a 25- and 200-fold molar excess of unlabeled oligonucleotide was added to the reaction mixture. Used competitor sequences were listed in Table 2 and Table 3. The RARE of human RAR beta (Table 1, RARE-F and RARE-R) [11] and random sequences (Table 1, random_F and random_R) were used as positive and negative controls, respectively. The binding reaction was carried out in the EMSA binding buffer

containing 12 mM HEPES (pH 7.9), 60 mM KCl, 4 mM MgCl₂, 1 mM EDTA, 12% glycerol, and 0.5% Nonidet P-40. The reaction mixtures were directly loaded onto a 4% nondenaturing polyacrylamide gels made in 0.5xTBE. After electrophoresis was performed at 4˚C, the gels were analyzed by using a bio-imaging analyzer (FLA-7000 FUJIFILM).

To map the obtained sequences on the human genomic assembly (GRCh37), these sequences were analyzed using NCBI’s BLAST and were searched for AGGTCA motifs. For the stringency of the search, we allowed up to 2-bp mismatches. For each predicted RAR/RXR binding site, the nearest gene and the distance from the center of the binding site to the transcriptional start site of the gene within 1000 kb was identified with GREAT (http://bejerano.stanford.edu/great/public/html/).

To identify RAR/RXR heterodimer―but not RXR homodimer-binding sites by a modified yeast one-hybrid system, the simultaneous expression of two transcription factors in yeast was required. Internal ribosome entry site (IRES) has been widely used for this purpose, but it has a major limitation. Namely, translation efficiency of a gene placed after IRES is much lower than that of a gene located before IRES [12] . The limitation can be over

come by a 2A peptide, a “self-cleaving” small peptide, which was identified in the FMDV [13] . As 2A-me-di- ated cleavage is a universal phenomenon in eukaryotic cells [14] [15] , we constructed a single expression vector by placing FMDV 2A segment between RXR and RAR. To minimize the effect of RXR homodimer on the reporter activity, RXR is expressed as the native protein, although RAR is expressed as a fusion to the GAL4 AD (Figure 1). To evaluate the function of RAR/RXR heterodimer, yeast cells were transformed with these effectors and the indicated reporters (Figure 2). The transformants were grown on synthetic complete media includ-

ing RAR-specific ligand TTNPB [16] and antagonist of RXR homodimers LG100754 [17] [18] . The transformants expressing either RAR or RXR alone could not grow, whereas the transformants expressing both RAR and RXR could grow in a RARE-dependent manner. These results indicated that RAR/RXR heterodimer could activate the reporter gene via the RARE.

The human genomic library after 5FOA selection [8] was transformed with the human RAR/RXR expression vector (Figure 1). More than 1 × 10⁷ of the library was selected for uracil prototroph. Human genomic fragments were recovered from the validated colonies by colony-direct PCR and sequenced. Two hundred and eleven unique sequences were obtained from 364 clones.

RAREs are typically composed of two directrepeats of a core motif, RGKTCA [19] . The classical RARE is a 5 bp-spaced direct repeat (DR5). Furthermore, the heterodimers also bind to direct repeats separated by 1 bp (DR1) or 2 bp (DR2) [20] . To analyze the obtained genomic sequences, AGGTCA motifs were computationally searched. For the stringency of the search, we allowed up to 2-bp mismatches. As an initial test, we examined direct interaction in vitro synthesized RAR/RXR and the known RARE (rCRBPII RARE) (Figure 3). Incubation of the the cy5-labeled rCRBPII oligonucleotides with the combination of RAR and RXR retarded complexes, but no shifted band was observed with either receptor alone (Figure 3, lanes 2-4). The complexes repre- sented a sequence-specific interaction between the rCRBPII probe and the RAR/RXR heterodimer, since the formation of this complex was specifically reduced with molar excess of unlabeled competitors (Figure 3, lanes 5-7). Moreover, the addition of anti-RXR or anti-RAR antibody created a slower-migrating complex (Figure 3(a), lanes 7 and 8). No super shifted bands were observed with anti-HNF4 antibodies (Figure 3, lane 9). These results indicated that the sequence-specific binding complex contained both RAR and RXR, presumably as a heterodimer.

As a next step, we examined whether the 193 predicted RARE should interact with RAR/RXR heterodimer. At least 160 predicted RAREs in the obtained genomic sequences could directly interact with RAR/RXR heterodimer (Table 2). These RAR/RXR-binding sites were located around or in the genes with various functions,

such as cytoskeleton (TUBB6, KIF16B, KRT18 and CTNNA1), extracellular matrix (LARGE), signal transduction (CREB5 and STAT3), transcription (TAF12, POLQ and TRERF1), translation (EIF2B3 and EIF5B) and development (INHBA, ISX, KIAA1715, LHX1, and PBX3). Remarkably, several genes previously known to be regulated by ATRA including FBLN1 [21] and RAR beta [11] [22] were also included. Although the RARE of human RAR betagene was already reported, the exon 9 was a novel binding site. Furthermore, we confirmed that RAR/RXR could directly interact with non-canonical motifs, such as inverted repeats (IRs) and everted repeats (ERs) (Table 2).

Functional rSNPs in transcription factor-binding sites may predictably lead to differences in gene expression and associate with disease susceptibility. Then, we identify 23SNPs on each RAR/RXR-binding sequence by using the NCBI’s SNP database and examined the effect on the RAR/RXR-binding affinity. As a result, 19 functional rSNPs were identified by analyzing the difference in the DNA-binding affinities (Table 3).