Nucleic Acids Research, Vol. 20, No. 12 3223-3232

Characterization of a retinoic acid responsive elementisolated by whole genome PCR

M.Patricia Costa-Giomi, Marie-Pierre Gaubl, Pierre Chambon' and Patricio Abarzu'a*Department of Oncology, Roche Research Center, Hoffmann-LaRoche Inc., Nutley, NJ 07110, USAand 1Laboratoire de Genetique Moleculaire des Eucaryotes du CNRS, U184 de Biologie Moleculaireet de Genie Genetique de l'INSERM, 67085 Strasbourg, France

Received November 27, 1991; Revised and Accepted May 12, 1992

ABSTRACT

We have used whole genome PCR in an attempt toisolate novel retinoic acid (RA) responsive genes. Wecloned several small genomic fragments from totalhuman DNA containing putative retinoic acidresponsive elements (RAREs) selected by directbinding to the retinoic acid receptor a (RARc). Wereport here that an oligonucleotide containing asequence from one of the cloned human DNAfragments, and referred to as cxl, functions as anauthentic RARE. It is shown that both RARa and RARt3produced in Cos cells as well as in vitro translatedRARa bind directly and sequence-specifically to thecxl RARE. By mutational analysis it is demonstrated thatthe cxlRARE consists of an imperfect direct repeat ofthe estrogen- and thyroid hormone-related AGGTCAhalf-site motif separated by a 5 bp spacer. Theorientation and spacing of the half-site repeats areshown to play a critical role in RAR recognition. Whencloned upstream of a TK-Luc reporter, the alRARE isshown to confer responsiveness to RA in anorientation-independent fashion in F9 and CV-1 cells.The magnitude of the RA response mediated by theal RARE differed in these cell lines.

INTRODUCTION

Retinoids, which include natural and synthetic derivatives ofvitamin A (retinol), have been shown to have profound effectsin a variety of biological processes including cell differentiation,embryogenesis and neoplasia. Retinoic acid (RA), one of the mostpotent natural retinoids, can dramatically change the biologicaland biochemical properties of normal and malignant cells in tissueculture, sometimes in opposite ways, depending on concentrationand cell type (1). Retinoids are also potent inhibitors of chemicalcarcinogenesis in rodent models (2). During embryogenesis, RAhas been found to be a very important endogenous signal in thedevelopment and regeneration of the limb and other structuresin vertebrates (3). In rodents and human, RA is teratogeniccausing a variety of crano-facial and limb malformations as well

as defects in brain development (4-7). Target tissues showunique temporal patterns of RA sensitivity that coincide withorganogenesis (5).The pleiotropic effects of retinoids have been thought to be

mediated at the gene transcription level (8). Several cDNA clonesrepresenting transcripts that increase or decrease in abundancein cells treated with RA have been isolated (9-13), howevera direct link between transcriptional regulation and RA was

lacking until the recent identification of the nuclear RA receptors(RARs) (14,15). Molecular cloning and sequence analysis haveled to the identification of three RAR subtypes (RARca, RARI3and RARy) encoded by three independent genes (16-18). Inaddition, multiple mRNA isoforms have also been identified foreach receptor subtype (19-21). Recently, a fourth retinoidnuclear receptor (RXR-cx) has been cloned and shown to bedistinct from the RARs (22). All these retinoid receptors belongto the steroid/thyroid hormone nuclear receptor superfamily(23 -25). Members of this family function as hormone-inducibletranscription factors and they are thought to bind as dimers totarget sequences called hormone responsive elements (HRE).RARs possess a domain structure comprising six regions,designated A to F (24,25). The amino acid domains responsiblefor DNA and ligand binding are located in regions C and E,respectively. The C domain of the receptors includes two zinc-finger motifs believed to mediate DNA binding (26-28). Thehigh level of amino acid homology among the RARs in the Cregion suggests that all three receptors are likely to recognizeidentical target sequences and, thus, may regulate identical setsof genes.One of the 'early' response genes shown to be transcriptionally

activated by RA is the RAR-g gene itself (29,30). The promoterregion was cloned and the sequences responsible for RAinducibility identified as an imperfect direct repeat of the sequence(A/G)GTTCA (31,32). The RARE(3 half-site repeat thus differsby a single non-conservative nucleotide change from theconsensus TRE/ERE motif. However, the spacing and orientationof the half-sites appears to be unique. Recently, characterizationof RAREs from the mouse laminin B1 (lamBI, refs. 33 and 34),mouse complement factor H (cfh, ref 35), human alcohol

* To whom correspondence should be addressed

.-. 1992 Oxford University Press

3224 Nucleic Acids Research, Vol. 20, No. 12

dehydrogenase (adh3, ref 36) and phosphoenolpyruvatecarboxykinase (pepck, ref 37) genes have all indicated that directrather than inverted repeats are responsible for RAresponsiveness. Moreover, in every case, the hexameric repeatis closely related to the AGGTCA motif. Despite thesesimilarities, the spacing and number of half-sites vary among

these natural RAREs. Whether these structural differences affectthe affinity or the mechanism of binding of RARs to theseelements is not known. It is also possible that receptor specificitymay be determined by some of these structural variations.Identification and characterization of more RA-responsive genesshould help to define what constitutes a consensus RARE andhow this element differs from other HREs and it should alsoenhance our knowledge of how retinoids work.

Although cDNAs for genes up- or down-regulated by RA havebeen cloned (9- 13), 'primary' response genes [defined as thosewhose transcription rates are increased or decreased in a proteinsynthesis-independent fashion by RA, examples are the RAR3and ERA-1 (Hox-1.6) genes] are the least represented. Thispaucity in known 'primary' RA-responsive genes is due toinherent limitations of screening large representative cDNAlibraries by conventional hybridization techniques (38,39). Forexample the RAR3 gene itself was never detected by differentialscreening of cDNA libraries from RA-treated cells (9-13). Toovercome this problem, we have chosen a different approach toisolate RA-responsive genes based on cloning of discrete genomicfragments of DNA containing RAREs selected by direct bindingto RARs. Then these fragments can be used as probes to isolatethe RA-target genes. This technique, referred to as whole genomePCR (40), has several advantages compared to more conventionalprocedures. In principle, any cell or tissue could serve as a source

of DNA, independent of whether or not RA-responsive genes

are expressed in that particular cellular milieu. Furthermore, thisapproach should lead to the isolation of novel RA-regulated genesand also increase our understanding of RAREs. As a first steptoward this goal, we report here the isolation and characterizationof a novel human RARE from total genomic DNA. Our studiessuggest that this approach may prove useful to identify genes

regulated by retinoids.

MATERIALS AND METHODSWhole genome PCRHigh molecular weight DNA from human placenta (Clontech)was sonicated on ice to an average size of 150-250 bp and madeblunt-ended by incubation with T4 DNA polymerase (4 units,New England Biolabs) and all four dNTPs. DNA was modifiedto a form appropriate for PCR amplification by ligation to 'catch-linkers' as previously described (40,41). To select for DNAsequences which bind to RARa, linker-modified DNA (0.5 ,Lg)was incubated with nuclear extracts from Cos cells transfectedwith an RARa expression vector (see below) in 20 tl of DNAbinding buffer (50 mM Hepes [N-2-Hydroxymethylpiperazine-N'-2-ethanesulfonic acid], pH 7.5, 50 mM KCI, 5 mM MgCl2,10 yM ZnSO4, 1 mM dithiothreitol, 20% glycerol) containing10 nM RA and 2 Ag poly (dI-dC) (Pharmacia) for 30 min at room

temperature. After DNA binding, 200 y1 of immunoprecipitationbuffer (40) and 5 ,al of rabbit anti-RARa antisera SP171 (42)were added and incubation continued for 1 hr. Immunocomplexeswere bound to protein-A-Sepharose (Sigma) for 1 hr at room

temperature, recovered by centrifugation in a microfuge andwashed three times with 1.5 ml of immunoprecipitation buffer.

DNA was released from the complex, extracted withphenol:chloroform (1:1), and ethanol precipitated. RecoveredDNA was amplified by 25 cycles of PCR with 'catch-linkers'as primers, and then used for additional rounds of selection.

After two and four rounds of binding and amplification,libraries were constructed as previously described (40,41).Fragments that were enriched by the selection procedure wereidentified by colony hybridization (43) with a probe preparedfrom the amplified and selected DNA (10-50 ng) by randomprimer labelling with [a-32P]dCTP. Highly repetitive sequenceswere removed by pre-annealing the probe to 1 mg/ml sheared,heat-denatured human placental DNA in 5 x SSC (1 x SSC is0.15 M NaCl, 0.015 M Na Citrate) for 15 min at 68°C.

Plasmid constructionsAll plasmid constructions were performed by standard cloningtechniques (44). The RARa (hRARaO) and RAR,3 (hRAR30)expression vectors were previously described (16,17). The RARyand VDR expression vectors were a gift of Dr. P.LeMotteand Dr. W.Hunzinker (F.Hoffmann-LaRoche Ltd., Basle)respectively. The RXRa expression vector was a gift ofDr. J.Grippo (Hoffmann-LaRoche Inc., Nutley).The reporter plasmid pTK-Luc was constructed by subcloning

a BamHlI fragment from plasmid ptkluc (45) into the BamHI siteof the bluescript KS(+) plasmid. This fragment contains the HSVthymidine kinase promoter, positions -109 to + 50, the fireflyluciferase gene and SV40 splicing and polyadenylation signals.The a 1 (1 13 bp) and a 13 (167 bp) fragments were isolated byEcoRI digestion, made blunt by fill in with the Klenow fragmentof E. coli DNA polymerase I, and cloned upstream of the TKpromoter into a unique SmaI site present in the multiple cloningsite of bluescript KS( +) to generate the a 1/TK-Luc and a 13/TK-Luc reporters, respectively. To construct the alRARE3/TK-Lucreporter, a unique BglII cloning site was created by inserting aBglII linker into the SmaI site of pTK-Luc. A 73 bp syntheticfragment containing three tandem copies of the a 1 sequence (Fig1) and flanked by BamHI and BglII recessive ends was thenligated to BglIf-digested pTK-Luc. The structure and orientationof the a lRARE3/TK-Luc reporter was verified by restrictionenzyme mapping and nucleotide sequencing. All plasmids weremaintained in E. coli DH5a and prepared by banding twice inCsCl gradients.

Synthetic oligonucleotidesAll oligonucleotides were synthesized on a Applied BiosystemDNA synthesizer, deblocked and purified by denaturatingpolyacrylamiide gel electrophoresis. Two 33-base oligonucleotideswere synthesized for each of the a 1, 01 and TREGH probes. Thesense and antisense strands of the cal probe consisted of thesequences 5'-AAGTGGATCCCAAGGGCAGGAGAAGGTCA-GAGA-3' and 5'-AGGAAGATCTTCTCTGACCTTCTCC-TGCCCTTG-3', respectively. When these oligonucleotides areannealed, a 10 base 5' overhang is generated at each end. Theoligonucleotides used to prepare the 01 and TREGH probes weresynthesized with the same overhangs and the a l sequence wasreplaced with information from the human RAR,B gene (-59 to-33) and the rat growth hormone gene (- 186 to -158),respectively (31,46). 32P-labelled probes were prepared byfilling in the 5' overhangs with dATP, dTTP, dGTP, and[a-32P]dCTP using the Klenow fragment of E. coli DNApolymerase I. Oligonucleotides used as competitors were preparedin a similar manner except that unlabelled dCTP was added.

Nucleic Acids Research, Vol. 20, No. 12 3225

Oligonucleotides containing mutations in the a 1 sequence werebased on a 43-base oligonucleotide with the sequence5'-GTGGATCCCAAGGGCA-GGAGAAGGTCAGAGAA-GATCTTCCTAA-3' containing the same 23 bases of a1information as above. A single oligonucleotide was synthesizedfor each mutant and annealed to a 14 base primer of the sequence5 '-TTAGGAAGATCTTC-3'. Full-length mutant oligo-nucleotides used as competitors were made by extending theprimer as described above.

Tissue culture, transfections, and luciferase assaysF9, CV l or HeLa cells were plated at a density of 5 x I05 cellsper 100 mm dish in Dulbecco's modified Eagle's medium(DMEM) supplemented with 10% fetal calf serum (FCS). F9cells were cultured on gelatin-coated dishes as describedpreviously (47). The following day, the medium was replacedwith DMEM containing 10% charcoal-treated FCS and cells weretransfected with 5 jig of luciferase reporter plasmid, 1 Ag ofreceptor expression vector, 10 ,ug of the ,3-galactosidaseexpression vector pCH110 (Pharmacia) as an internal control fortransfection efficiency and 14 itg of bluescript plasmid as carrier.Transfections were performed as described by Chen andOkayama (48) for 16-17 h at 370C in 5% Co2. Aftertransfection, cells were washed with PBS and refed with DMEMplus 10% charcoal-treated FCS. All-trans-RA (Hoffmann-LaRoche Inc.), dissolved in 100% ethanol, was administered 4h later to a final concentration of 1 1M. Untreated controlsreceived only the vehicle. Cells were harvested 24 h after RAaddition and luciferase (49) and ,B-galactosidase (50) activitieswere measured as previously described. Protein concentrationwas determined by the Bradford procedure (51). For eachexperiment, luciferase activity was normalized to that of (-galactosidase activity. All transfections were independentlyrepeated at least three times unless otherwise noted.

Preparation of nuclear extractsCos cells were transfected with 7 itg of RARa or RARi3expression vectors by the DEAE-dextran procedure (44). Afterincubating the transfected Cos cells for another 48 h, the cellswere harvested and nuclear extracts were prepared according toLee et al (52). Briefly, cells were resuspended in one packedcell volume of buffer A (53) and allowed to swell on ice for 15min. Cells were lysed by passing the suspension 13 times througha 1 ml syringe attached to a 26 1/2 gauge needle. The cellhomogenate was centrifuged for 20 seconds in a microfuge andthe nuclear pellet was resuspended in two thirds of the initialpacked cell volume of buffer C (53) containing 0.4 M NaCl.Nuclei were extracted with stirring for 30 min on ice. Nucleardebris was removed by centrifugation for 5 min in a microfugeand the soluble nuclear extract was aliquoted, quick-frozen andstored at -80°C.

In vitro transcription and translationRARa and RAR,B expression vectors were linearized by BamHIrestriction enzyme digestion and capped mRNA transcriptsprepared using T7 RNA polymerase (Stratagene). These mRNAswere then used to program translation in a rabbit reticulocytelysate system as recommended by the manufacturer (Promega).For gel retardation assays, 5 ul of the lysate was added to thebinding reaction.

Gel retardation assaysCos cells nuclear extracts (9-15IAg) were preincubated in thepresence of 1 jLg of poly (dI-dC) in 20 tl of binding buffer (47mM Hepes, pH 7.9, 50 mM KCL, 5 mM MgCl2, 1 mM DTT,0.05% NP40 and 10% glycerol) for 15 min at room temperature.Subsequently, 0.25 ng of the radiolabelled probe was added andthe incubation was continued for 45 min. For the competitionstudies, a 50-fold molar excess of unlabeled competitoroligonucleotide was added to the reaction after the preincubationstep and the mixture further incubated for 15 min before additionof the radiolabelled probe. When antibodies and peptides wereused in the assay, they were preincubated with nuclear extractsin binding buffer for 1 h at 4°C prior to the addition of the probe.Protein/DNA complexes were analyzed by electrophoresisthrough a nondenaturating 6% polyacrylamide gel (80:1acrylamide:bisacrylamide) prepared and electrophoresed in1 xTBE (0.1 M Tris, 0.1 M boric acid, and 1 mM EDTA, pH8.3) for 2 h at 200 volts. Gels were pre-run for 30 min at roomtemperature.

RESULTSIsolation of RARa binding sites by whole genome PCRTo isolate RARE sequences from total human DNA we followedthe whole genome PCR procedure described by Kinzler andVogelstein (40). In this technique, genomic DNA is first modifiedto a form that can be amplified by the polymerase chain reaction.Specific DNA sequences are then selected by binding to aregulatory protein followed by immunoprecipitation of theprotein/DNA complex. Because DNA fragments recovered fromthe immunoprecipitation step are amplified by PCR, selectioncan be performed under stringent conditions with little concernabout yield. Accordingly, sheared, linker-modified human DNAwas incubated with nuclear extracts from RAR-a transfected Coscells and receptor/DNA complexes were immunoprecipitated withrabbit polyclonal antibody SP171 (42). DNA was recovered, PCRamplified, and then used for further rounds of selection andamplification. After two and four of these cycles, DNA wasdigested with EcoRI and cloned. Colonies were screened forenriched sequences by hybridization to the same selected andamplified labelled DNA. Under these conditions, only coloniescontaining sequences whose abundance is 0.05% or higher ofthe total sequence complexity would be expected to score aspositive (38,39).

After two cycles of selection and amplification less than 0.1 %of the colonies were detectable following hybridization to theprobe, and this required prolonged exposure to X-ray film. Afterfour cycles, close to 100% of the colonies hybridized to the probe.Restriction enzyme analysis of plasmid DNA isolated frompositive colonies selected at random showed about 30% to containdistinctive fragment inserts. The remaining contained only linkersequences, suggesting a PCR artifact.Ten independently-isolated positive clones containing inserts

(average size 144 bp, range 76 to 291 bp) were sequenced bythe dideoxy procedure and found to represent nine uniquesequences. Extensive sequence analysis revealed most of theclones to contain only small stretches of homology amongthemselves and to previously described HREs (31,46,54). Themost prevalent region of homology contained variants of theAGGTCA motif found in EREs and TREs. Preliminary evidencesuggesting that at least some of the fragments contained RAR

3226 Nucleic Acids Research, Vol. 20, No. 12

CA AGGGCA GGAGA AGGTCA GAGATG GAATCA ACTTG AGTACA TGGAG AA A GCC A AT - A A A TAG CA CC T TAT GTTGCA TCCTG AGTTCA CTTCCGGAAGCAGAACA AAGGCA GTTA

Figure 1. Sequence of a series of oligonucleotides containing putative RARabinding sites. Human DNA fragments isolated by whole genome PCR weresequenced and the regions showing significant homology were directly aligned.The sequences with the highest identity were then selected and double-strandedoligonucleotides synthesized. All oligonucleotides contained identical 10 bpoverhangs for labelling and they are not shown. A 1 bp gap was introduced intoc18 for optimal alignment. The AGGTCA-related motifs are shown within theboxes. Arrows indicate the orientation of the motifs.

--, w

ELI I.:: .

i

Figure 2. In vitro binding of RARa and RAR,8 to oligonucleotide al. 10tg ofnuclear extracts from mock-transfected Cos cells (M: lanes 1, 4, 7 and 10), orCos cells transiently transfected with RARa (a: lanes 2, 5, 8 and 11) or RAR/3(,B: lanes 3,6,9 and 12) were incubated with [32P]-labelled al oligonucleotide.For competition experiments, a 50-fold molar excess of unlabeled a 1 (lanes 4,5 and 6), 01 (RARE,B) (lanes 7, 8 or 9) or TREGH (lanes 10, 11 and 12)oligonucleotides were added to the binding reactions 15 min prior to the additionof the labelled probe. Protein/DNA complexes were resolved by polyacrylamidegel electrophoresis and analyzed by autoradiography. The arrows indicate thespecific complexes Cl and C2. Addition of RA to the binding reaction did notchange the size or intensity of the complexes (data not shown).

binding sites was obtained by immunoprecipitation experimentsperformed as described for the selection procedure. When thesefragments were tested for binding by gel retardation assays usingnuclear extracts from transfected Cos cells, several protein/DNAcomplexes were detectable. Initially, we could not observesignificant differences in the intensity or the number of bandswhen nuclear extracts from mock- or RARa-transfected cellswere used in the binding reaction (data not shown). Because ofthe presence of several protein/DNA complexes, some of themvery strong, we wondered whether multiple binding sites fornuclear factors present in the full length fragments wereinterfering with detection of the specific RAR/DNA complex.To overcome this problem, we synthesized oligonucleotidescontaining putative RAR binding sites as deduced from sequenceanalysis and test them for RAR binding by gel retardation assays.Fig 1 shows the best computer-assisted alignment for five such

Figure 3. Characterization of the protein/DNA complexes formed with the a I

probe with anti-RAR polyclonal antibodies. Binding of nuclear extracts to thea 1 probe was performed in the absence of antibody (-: panel A, lanes I and2; panel B, lanes I and 2) or in the presence of non-immune rabbit serum (C:panel A, lanes 3 and 4; panel B, lanes 3 and 4), anti-RARa antibody SPI71(a: panel A, lanes 5 to 8) or anti-RAR/3 antibody SP172 (3: panel B, lanes 5to 10). Nuclear extracts from mock-transfected Cos cells (panel A: lanes 1, 3,5 and 7; panel B: lanes 1, 3 , 5 , 7 and 9) or Cos cells transfected with RARca(a: panel A, lanes 2, 4 6 and 8) or RARI3 (/3: panel B, lanes 2, 4, 6, 8 and10) were pre-incubated with antibodies for 60 min at 4°C prior to the additionof the radiolabelled probe. To show the specificity of the antibodies, a RARapeptide (panel A: lanes 7 and 8; panel B: lanes 9 and 10) or a RAR( peptide(panel B: lanes 7 and 8) containing the SP171 and SP172 epitopes respectivelywere incubated for 60 min at 4°C with the respective anti-serum before additionto the binding reaction. The position of the specific complexes C and C2, andthat of the shifted complex are indicated by arrows.

putative sites as found in five of the cloned fragments. The spatialorganization and sequence composition showed a degree ofresemblance to the RARE found in the RAR-3 gene (31,32). Weshow below by direct binding and functional assays that the a 1

oligonucleotide functions as an authentic RARE.

Binding of RAR-a and RAR-f to al oligonucleotideTo test if RARs would bind to any of the sequences shown inFig 1, double stranded oligonucleotides were synthesized, end-labelled with [oa-32P]dCTP, and incubated with nuclear extractsprepared from Cos cells transiently transfected with RARexpression vectors. Protein/DNA complexes were resolved bygel retardation assays and detected by autoradiography. Underthese experimental conditions only the a 1 oligonucleotide yieldedspecific bands. Two major protein/DNA complexes could bedetected by this assay with the ac1 probe. Complex 1 (Cl, Fig2) was only present when nuclear extracts from either RARa-or RARf3-transfected Cos cells were used in the binding reaction(Fig 2, lane 2 and 3). In contrast, complex 2 (C2, Fig 2) was

detected when both mock- and RAR-transfected nuclear extractswere incubated with the a1 probe.

Sequence-specific binding of the RARs to the al probe was

demonstrated in competition assays (Fig 2 lanes 4-12). Additionof a 50-fold molar excess of unlabelled a 1 or 01 oligonucleotide(RAREf3, ref 31) almost completely abolished the formation ofCl, whereas an oligonucleotide containing a TRE from the ratgrowth hormone gene failed to compete under these conditions.On the other hand, formation of C2 was only competed by a 1

(lanes 4-6, Fig 2), suggesting that RARs are not involved inthe formation of this complex. Identical results were obtained

a - 1a -17a -18a -24a -26

0-

a

:4,-:::x.4 "a

Nucleic Acids Research, Vol. 20, No. 12 3227

Antibody - C (I ARNA 4+ - + -

_

B

C

DR1C A A G G G C A G G A G

. , .. .T .

. . . . . . .

.T---------

-CT.CT--- --

DR2A A GG T C A G A

- G T- -TTCG T

T C

C T

G A Wild type- ml

- - m2- - m3- - m4- - m5- - m6- - m7- - m8

m.9- - m13

DR1 > DR2 >

C A A G G G C A G G A G A A G G T C A G A G A Wild typeI--

--GG GA m10G A A - - - - - - ml1

--- --G A .

----- - - - - m12!~ JG G TA G A

-- - - - m15

DR1 )P g DR2

LA G G G C Al T G A C C Ti- - - - m14

1 2 3 4 5 6

Figure 4. In vitro translated RARa binds directly to the al oligonucleotide. 5

Al of rabbit reticulocyte lysate programmed with no RNA (-: lanes 1, 3 and5) or with in vitro transcribed RARa mRNA (+ : lanes 2, 4 and 6) were pre-incubated in the absence of antibody (-: lanes 1 and 2), or in the presence ofnon-immune serum (C: lanes 3 and 4), or in the presence of anti-RARa polyclonalantibody (ca: lanes 5 and 6) before addition of the radiolabelled cal probe.Protein/DNA complexes were analyzed by gel retardation assay. Arrows indicatedthe specific (left) and shifted (right) complexes respectively.

when the fragment was used as the probe. However, theintensity of the RAR-specific complex was reduced (data notshown).

Further evidence for the involvement of RARs in the formationof C 1, but not C2, was obtained from retardation assays in thepresence of RAR-specific antibodies (Fig 3A and 3B). Incubationof binding reactions with anti-serum SP171, but not non-immuneserum, led to the appearance of a higher molecular weightcomplex with a concomitant decrease in the intensity of C whenRARax-transfected Cos cell nuclear extracts were the source ofreceptor. Moreover, an RARa-specific peptide containing theepitope recognized by the SP171 anti-serum blocked theformation of this large complex (compare lanes 6 and 8 in Fig3A). Similarly, incubation of RARI3-transfected Cos cell nuclearextracts with rabbit polyclonal antibody SP172 raised against a

RAR,B peptide (42) also led to the formation of a slowly migratingcomplex with an accompanying decrease in the intensity of C1(Fig 3B lane 6). Incubation with the RAR,8 peptide, but not an

RARca peptide, blocked the shift to the high molecular weightform of the complex (compare lanes 8 and 10 with lane 6 in Fig3B). In contrast, neither antibody affected the intensity or

migration of C2. These results demonstrate that both RARca andRAR,B bind with high affinity to the al oligonucleotide.Furthermore, the presence of a second specific protein/DNAcomplex also generated by the mock-transfected Cos cell nuclearextracts suggests the existence of other nuclear factor(s) that bindto the same element as the RARs. It is unlikely that endogenousmonkey (Cos cells) RARs are responsible for C2 since the 01oligonucleotide failed to compete, even though it is a highlyefficient RARE activated by endogenous RARs in human,monkey and mouse cells (31,32).To investigate whether the formation of the complex between

RARs and the cal probe requires additional factor(s) present innuclear extracts, in vitro transcribed RARat mRNA was used toprogram a rabbit reticulocyte lysate and the in vitro translatedreceptor was then tested in gel retardation assays for binding to

Figure 5. Sequence of a 1 oligonucleotide mutants. Oligonucleotides containingthe same 23 bp of information as al (Fig 1) were prepared containing: (A) aseries of single or double point mutations within the direct repeats (DR) or flankingsequences; (B) deletions of I to 3 bp or addition of I bp within the spacer region;and (C) an inversion of the second direct repeat (DR2). The region of the wild-type al oligonucleotide that was mutated and the specific changes that were madeare shown. A deletion is indicated by a dot. The AGGTCA-related sequencesare shown within boxes. Arrows indicate orientation of the half-site repeats.

the al oligonucleotide. As shown in Fig 4, a protein/DNAcomplex could be detected only when the acl probe was incubatedwith reticulocyte lysate programmed with the RARca mRNA.Addition of polyclonal antibody SP171 to the binding reactioninduced a shift in this band toward a high molecular weight form,demonstrating the presence of the receptor in this complex (Fig4, lane 6). This antibody-associated shift was not observed whennon-immune serum was added to the binding reaction (Fig 4,lane 4). Similar results were obtained for RAR,B (data not shown).These data demonstrate that RARs can bind directly to the calprobe. Note that the addition of the SP171 antiserum appearsto increase the intensity of the complex formed between in vitrotranslated RARca and the cal probe (compare lane 6 with lanes2 and 4 in Fig 4). Likewise, we have observed that addition ofnuclear extracts from untransfected Cos cells, but not bovineserum albumin, increases the amount of specific complex at least5-fold, suggesting that accessory nuclear factor(s), although notrequired for binding, may either increase the stability of theRAR/DNA complex or facilitate dimer formation. These nuclearfactor(s) may be limiting since a 5-fold dilution of the nuclearextracts leads to a similar decrease in the amount of specificcomplex (55). This is in agreement with published reportsdemonstrating the existence of multiple cell type-specific factorsthat increase the binding affinity of the RARs for responseelements (56). Based on the high affinity binding of the RARsto the cal oligonucleotide we conditionally designated it asa 1RARE.We failed to observe specific RAR/DNA complexes with the

other oligonucleotides shown in Fig 1, however, such a resultcan occur if the gel retardation assay is not sensitive enough todetect weaker, but significant interactions between RARs andsome DNA sequences. In particular, the a24 oligonucleotideshares 62% overall sequence homology with the RARE from theRAR,3 gene (see Fig 9A). Moreover, its 3'-end matches perfectlyat 9 out of 12 positions to a shorter but still active version ofthe RARE3 (32), yet we could not detect a RAR-specificcomplex. Consequently, we tested these oligonucleotides for RAR

3228 Nucleic Acids Research, Vol. 20, No. 12

U m

Figure 6. Sequence and structural requirements for RAR specific binding to thea RARE. Double-stranded mutant oligonucleotides shown in Fig 6 were usedas competitors at a 50-fold molar excess in binding reactions containing nuclearextracts from RARa-transfected Cos cells and labelleda(l probe. (A) Importanceof the nucleotide sequence and involvement of the half-site repeats in RARarecognition of the ceIRARE. (B) Effects of spacer sequence (lanes 4 and 5), spacerlength (lanes 2, 3, 6, 7 and 8), and orientation of the half-sites (lane 9) for bindingof RARa to the acRARE. Arrows indicate specific complex Cl and C2.

binding in competition assays against the a 1 probe to increasethe probability of detecting weak but sequence-specificinteractions (data not shown). As would be expected fromsequence similarity, under these conditions the a24oligonucleotide was able to compete for binding to the RARaalmost as efficiently as the ca1 sequence itself, demonstrating thatat least two of the nine immunoselected human DNA fragmentscontain authentic RAR binding sites.

Mutational analysis of the alRARETo help define the sequences and structural determinants criticalfor RAR binding to the caiRARE, we synthesized several mutantoligonucleotides (ml to m1S, shown in Fig 5) and assayed themfor their ability to compete for binding to the RARs present innuclear extracts from RAR-transfected Cos cells (Fig 6). Mutants1 and 2 both contain a single nucleotide change converting themto a perfect direct repeat of motif 2 (DR2) or 1 (DR 1)respectively. In gel retardation assays, ml competed even more

effectively than the wild-type a 1 oligonucleotide for binding tothe RARa-specific complex (Fig 6A, compare lane 2 and 3).In contrast, m2 was a very poor competitor even at the 50-foldmolar excess used in these experiments (Fig 6A, lane 4). Similarquantitative results were obtained when we used a 25- or 100-foldmolar excess of competitor or when RAR,8 extracts were used(data not shown). These dramatic changes induced by single basesubstitutions suggest that high affinity binding of RARs to thecalRARE is highly sequence-specific and it also reveals theimportance of the direct repeats in RAR recognition.The aiRARE contains two imperfect direct repeats, each

consisting of 9 nucleotides (Fig 1). HRE for other members ofthe steroid/thyroid receptor superfamily have been shown tocontain hexameric half-sites arranged as inverted repeats(26-28,46,54). To investigate the possible hexameric nature ofthe half-sites in the alRARE, mutant oligonucleotides were

synthesized containing either single or double base substitutionswithin DR2 (Fig 5). Since we have already shown above thatthe DRI sequence AGGGCA is not optimal for binding, we

expected that any mutation critical for RAR recognition in DR2

would have a more dramatic effect upon the ability of the mutantoligonucleotide to compete for the formation of the specificcomplex, and indeed that was the case. Any mutation affectingthe motif AGGTCA (e.g. m2, m3, m4 and mS) had profoundeffects upon the ability of the mutant DNAs to compete againstthe al probe for RAR binding (Fig 6A). By contrast, either a

single (m8) or a double (m7) base change immediately followingthis hexameric element did not appreciably affect competition,suggesting that these nucleotide positions are not crucial forsequence-specific binding by the RARs. Similarly, m5 and m6were equally ineffective in competing for complex 1, suggestingthat the adenosine residue immediately preceding the hexamericmotif is not involved in recognition. Mutations within the regionseparating the hexameric half-sites either have no effect (ml3)or slightly decreased the ability of the oligonucleotide to compete(m9). Altogether, these results delineate the two AGGTCA-related motifs as critical for sequence-specific RAR binding. Notethat mutant competition data for complex 2 follow the same

general pattem as for the RAR specific complex except for mutant2. In m2, the T to G transversion at DR2 only marginally affectedcompetition for complex 2 while still impairing its ability tocompete for complex 1. This observation suggests that the proteincomponent(s) involved in formation of complex 2 may recognizerelated but not identical sequences.The half-site repeats in the a lRARE as defined above are

spaced by a five nucleotide long 'linker' sequence. In thisarrangement, each position in one half-site is separated by 11bp from its homolog. This locates both half-sites on the sameside of the DNA helix as it has also been suggested for the/3RARE and the lamB1 RARE (32-34). By analogy to othermembers of the nuclear receptor family, RARs are believed tobind as dimers to RAREs and the stereo-alignment of the half-sites may serve to orient the two monomers along the DNA.Therefore, alterations of the spacing between the repeats mightbe expected to be deleterious for RAR binding. Indeed, insertionor deletion of a single nucleotide into the spacer of the a 1 RAREdid abolish the ability of these mutant oligonucleotides to competefor binding to the RAR-specific complex (mI5 and mlO, Fig 6B).Unexpectedly however, deletion of two or three bases from thespacer region was not detrimental for binding, since ml 1 andm12 competed as well as the wild-type oligonucleotide for bindingto RARa (Fig 6B, lanes 6 and 7). Our results suggest that atleast for sequence-specific DNA binding, positioning of the half-sites on the same side of the DNA helix might not be a criticalparameter. Most likely, deletion or insertion of a single bp inthe spacer region as in mlO and ml5 disrupts the spatial geometrybetween the receptor dimer and the half-sites, since eachnucleotide added or removed would shift the half-sites positionsby 3.4 A and impose a 360 rotation around the DNA helix (57).This detrimental effect on receptor binding may be partially (e.g.ml 1) or totally (e.g. m12) compensated by rotating and movingthe half-sites closer together to within a distance where functionalinteraction with the receptor dimer may take place. Note thatthe size of the spacer does not impair the ability of the mutantsto compete for complex 2, demonstrating that formation ofcomplex 1 and 2 have different structural requirements.RARs have been shown to activate both direct and inverted

repeat responsive elements (31,32,34,35,37,58-60). Thispromiscuity displayed by the RARs suggested that the orientationof the half-sites might be functionally equivalent between thedirect and inverted form. To test this possibility, we synthesizeda mutant a(loligonucleotide (Fig SC, m14) containing the second

..... WW*

A..-A:A.AAk 'k

Nucleic Acids Research, Vol. 20, No. 12 3229

A

L.

.1

.,

ez

2

3

.,

7

6

5

4-

3-I2 nI

RA -+-+ -+-+ -+-+RARa --++ --++ --++

TK-Luc al1fTX-Luc al3fflX-Luc

10

9

8-

7

6

5-

4-

3-

2-

1

R - + + - + +

RARRa - + + - + +

TK-Luc a1RARE3/YK-Luc (F) alRARE3/KX-Luc (R)

Figure 7. Transactivation of the aIRARE by RA and RARcx in F9 cells. (A)The reporter plasmid al/TK-Luc contains the original human al genomic fragmentcloned upstream of the TK promoter. The reporter plasmid a 13/TK-Luc containsa human genomic fragment that is not bound by the RARs and it was used as

a negative control. The reporter plasmid, TK-Luc, contains only the TK promoterdriving the luciferase gene. (B) Reporter plasmids cilRARE3/TK-Luc (F) andacRARE3/TK-Luc (R) each contain three copies of the aiRARE (Fig 1) in theforward and reverse orientation, respectively, cloned upstream of the TK promoter.Luciferase reporter plasmids (5 ug) and ,B-galactosidase expression vector pCHl 10

(10 jig) were cotransfected into F9 cells with (+) or without (-) an RARcaexpression plasmid (1 jig). After transfection, 1 j4M RA (+) or vehicle alone(-) were added to the culture media. Cell extracts were prepared 24 h later andassayed for luciferase and ,B-galactosidase activity. Relative luciferase activityhas been corrected for protein concentration and normalized for (-galactosidaseactivity.

direct repeat in the opposite orientation relative to the wild-typesequence while preserving the half-site spacing, thus convertingit to the inverted repeat form. This oligonucleotide was testedfor its ability to compete against the alI probe for binding to theRARs. As shown in Fig 6B, lane 9, m14 was completely inactiveas a competitor for the specific RAR complex, demonstratingthat the direct repeat arrangement of the half-sites in the caIRAREis highly specific for RAR recognition. Similarly, m14 also failedto compete for complex 2, suggesting that the protein(s) involvedin formation of this complex also recognize direct rather thaninverted repeats.

Functional analysis of the alRARETo investigate whether the putative a lRARE could confer RAresponsiveness to a heterologous promoter, the ca1 fragment(al1/TK-Luc) as well as a double stranded oligonucleotidecontaining three tandem copies of the a 1 sequence(ac1RARE3/TK-Luc) were cloned upstream of the HerpesSimplex Virus Thymidine Kinase (TK) promoter driving the

Table I. Activation of various TK-Luc reporter plasmids by RA and RARa inCV-1 cells.

Relative Luciferase Activity(a)Transfected Receptor

(-) RARa

Reporter plasmid -RA +RA -RA +RATK-Luc 1.0 1.4 1.8 1.7aIRARE3/TK-Luc (F)(b) 1.0 1.4 1.6 2.7crIRARE3/TK-Luc (R)(c) 1.0 0.8 1.2 2.9cx26RARE3/TK-Luc(d) 1.0 1.2 1.3 1.2

(a) Relative luciferase activity was determined by dividing the normalizedluciferase activity from each experiment by the normalized luciferase activity foruntreated, non-receptor transfected cells.(b) Forward orientation of the calRARE (as shown in Fig 1) relative to the TKpromoter.(C) Reverse orientation of the calRARE relative to the TK promoter.(d) Plasmid ci26RARE3/TK-Luc contains three copies of oligonucleotide a26(Fig 1).

firefly luciferase reporter gene. These constructs were thentransiently cotransfected with or without an RARa expressionvector into F9 embryonal carcinoma cells and tested for RAinducibility.Under the experimental conditions used, the TK promoter itself

was unresponsive to RA (Fig 7A and B). In the absence ofcotransfected RARca, the c1l/TK-Luc reporter showed basalluciferase activity comparable to the TK-Luc reporter, with aslight increase upon addition of RA. In contrast, when RARawas cotransfected, a modest but reproducible 2- to 3-fold increasein luciferase activity was observed following addition of RA(Fig 7A). As a control we constructed a reporter (a 13/TK-Luc)containing a genomic human fragment (c 13) isolated by the sameprocedure. The a13 fragment neither binds nor competes forbinding to the RARs by gel retardation assays (data not shown).As expected, the a13/TK-Luc reporter failed to show anyincrease in luciferase activity upon addition of RA with or withoutcotransfected RARca (Fig 7A).

Fig 7B summarizes the results obtained with theceIRARE3/TK-Luc reporter. The reporter with the responseelement cloned in either orientation showed basal luciferaseactivity comparable to the TK-Luc reporter, with slight increaseupon addition of RA. In contrast, when RARa was cotransfected,a significant and reproducible 6- to 8-fold increase in luciferaseactivity was observed following addition of RA. This RA- andRARa-associated increase in luciferase activity was orientation-independent, since similar levels of activation was observed witheither the forward or reverse construct. In addition,transactivation was copy number-dependent since thealRARE3/TK-Luc reporter induced a 3-fold higher responsethan the al/TK-Luc construct, which contains a single copy ofthe a 1 sequence. Similar results were obtained when thealRARE3/TK-Luc reporter was cotransfected with RARa intoCV1 (Table I) and Hela cells (data not shown), although the foldincrease in luciferase activity was lower than in F9 cells. Again,transcriptional activation observed in CVI and Hela cells wasorientation-independent. These results demonstrate that thea 1RARE can function as an RA-inducible enhancer element indifferent cell types in the context of the TK promoter.

Since we have shown above that RAR3 also bound to the a 1probe with high affinity in gel retardation assays, we tested otherretinoid receptors (RAR,B, RARy and RXRca) for transcriptionalactivation of the a IRARE3/TK-Luc reporter in F9 cells. As a

3230 Nucleic Acids Research, Vol. 20, No. 12

3

2

0

RA -+ -+ -+ + +

C RARa RAR#

RARfyRXRa

Figure 8. Transcriptional activation of the caIRARE by different retinoid receptors.S Ag reporter plasmid aIRARE3/TK-Luc (R) was cotransfected with

1 jg ofeither a control plasmid vector containing the RARa cDNA in the antisenseorientation (C), or a RARa, RARI,, RAR-y, or RXRa expression vector intoF9 cells. Four hours after transfection, 1 ltM RA (+) or vehicle alone (-) wereadded to the culture medium and cell extracts prepared 24 h later and assayedfor luciferase and ,B-galactosidase activity. Relative luciferase activity wasdetermined as in Fig 7.

negative control, we cotransfected an antisense RARa expressionvector. As shown in Fig 8, addition of RA induced an increasein luciferase with all four receptors tested, but not with theantisense control. The strongest response was obtained with theRARfl expression vector. We do not know the reasons for thispreferential transactivation by RAR,3. Our binding studies withRARa and RAR,B did not reveal any significant differences inbinding affinities to the a I probe. The only reproducibledifference so far observed between nuclear extracts from RARaand RAR,B transfected Cos cells is the greater abundance ofcomplex 2 with RAR,8 (compare lanes 1 and 3 in Fig 2). Acomplex with similar properties has also been detected in nuclearextracts from F9 and Hela cells (data not shown), suggesting thatcomplex 2 is not a particular attribute of Cos cells. Whether thiscomplex is physiologically relevant in transactivation of thealRARE remains to be investigated.RXRa was identified by molecular cloning and sequencing

analysis as a new member of the steroid/thyroid receptorsuperfamily. Transcriptional activation studies revealed thatRXRa responds specifically to Vitamin A metabolites, includingRA, but it was considered unlikely that RA itself was the ligandsince RXRa only shares 27% amino acid homology with theligand binding domain of RARa (22). Additionally, recent studieshave shown that the natural ligand that activates RXRa appearsto be 9-cis RA, a new stereoisomer of RA (61,62). As shownin Fig 8, RXRa was able to induce luciferase activity from thea IRARE3/TK-Luc reporter upon addition of RA. The levels ofinduction were slightly lower than for RARa. One possibleexplanation of these results is that F9 cells can isomerize someall-trans RA to 9-cis RA, which it acts as a ligand for RXRa.

DISCUSSIONWe have shown above that both RARca and RARf3 produced inCos cells bind in a sequence-specific manner to the a 1oligonucleotide in gel retardation assays. Moreover, RARstranslated in vitro also bound the a 1 probe, demonstrating thatno additional nuclear protein is required for this specificinteraction. A second retarded complex detected by the a 1 probewas also present in nuclear extracts from mock-transfected Cos

A

R

RARE-a1RA;E-a24RAFE-U-B2RAFIE-r-a2RARE-dh

RAJEqnhcqRAIRE--6

Cons

CAAGGGCAGGAGAAGGTCAGAATGTTGCATCCTGAGTTCACTAGGGTTCACCGAAAGTTCACTCGAGTTCAGCAAGAGTTCAGCGCAGGTCA CTGACAGGGCATAAGGGGTCA TTCAGAGTTCA GTGGAGGTGAACAGGAGGTCAGCTCAGGTCACCAGGAGGTCAGA

NNAGGTCANNNNNAGGT CANNT

ThE-pW TCAGGTCA- TGACCT AVVAE-od CTAGGTGA --C-- TCACCG G

EE-vlt TCAGGTCAC-A-GTGACCTGA

RARE-crbp- GTIAG TCA AA GTCAGA

RW-Iwv61 G !GGCT 0 AAGCCCTTAGAAAAA GO AACC

RARE-ppdk AGGTGTCACTCCCA CAIA3GGTCA]TGAGAA

Figure 9. Sequence comparison of retinoic acid responsive elements. (A) Naturaland synthetic direct repeat RAREs. The sequence of the a 1 and (24 RAREsis from Fig 1; RARE-rar-32, the human and mouse RAR-,B2 isoform genepromoters (-55 to -35, refs 31 and 32); RARE-rar-a2, the mouse RAR-a2isoform gene promoter (-57 to -37, ref 73); RARE-cth, the mouse complementfactor H gene promoter (- 144 to - 125, ref 35); RARE-adh, the human alcoholdehydrogenase gene promoter (-302 to -282, the antisense strand is shown,ref 36); RARE-mhc-r, synthetic RARE derived from the rat cardiac myosin heavychain a subunit gene promoter (- 151 to - 132, ref 62); RARE-dr-5, syntheticRARE described in ref 63. (B) Palindromic response elements. TRE-pal, syntheticthyroid hormone responsive element derived from the rat growth hormone genepromoter (46); VDRE-ost, the human osteocalcin gene promoter (-512 to -497,ref 60); ERE-vit, the Xenopus vitellogenin A2 gene promoter (-334 to -316,ref 54). (C) Atypical RAREs. RARE-crbp-I, the mouse cellular retinol bindingprotein I gene promoter (- 1013 to -996, ref 71); RARE-lamB 1, the mouselaminin B1 gene promoter (-470 to -432, the antisense strand is shown, ref33); RARE-pepck, the rat phosphoenolpyruvate carboxykinase gene promoter(-457 to -425, the antisense strand is shown, ref 37). The AGGTCA-relatedmotifs are shown within the boxes. Arrows indicate the spacial orientation ofthe half-sites.

cells. The strongest evidence against a role for RARs in formationof complex 2 is the failure of the ,3RARE oligonucleotide (01,Fig 2) to compete for this complex and the inability of both anti-RAR antibodies to induce a shift in the mobility of complex 2.In addition, the structural requirements and, to lesser degree,the sequence requirements for binding displayed by complex 2were different from those exhibited by the RARs. We presentlydo not know the identity of the protein(s) involved in formationof complex 2. Interestingly, complex 2 is always more abundantin nuclear extracts prepared from RAR3-transfected Cos cells(compare lane 3 with lanes 1 and 2 in Fig 2). Since we cultureCos cells in the absence of RA, induction of this activity byretinoids is unlikely. Further studies will be required to identifythe factor(s) involved in this complex.The a( 1RARE contains nine-bp-long imperfect direct repeats,

however, our mutational analysis indicates that the functional half-site unit is the motif AGGTCA, also shown to be important forERE and TRE function. Nucleotide positions outside thehexameric half-sites seem to contribute little to sequence-specificbinding. We have also shown that the spacing and orientationof the hexameric repeats play a critical role in RAR binding.However, our competition data does not support the simplehypothesis that the stereo-alignment of the half-sites on the sameside of the DNA helix determines receptor binding as proposedfor the RARE: and the lamBI RARE (32,34). Instead, certain

Nucleic Acids Research, Vol. 20, No. 12 3231

spacer length are incompatible with RAR binding. These spacervariants could encode different receptor specificities as proposedrecently for a subset of the nuclear receptor family that includesthe RARs (63). Our data cannot exclude, however, that phasingof the half-sites on the same side of the DNA helix may promotemore efficient interaction of the receptor dimer with thetranscriptional machinery of the cell, resulting in strongeractivation. As shown in Fig 9A, in the majority of the naturaldirect repeat RAREs the half-sites are spaced by 5 nucleotides,suggesting a strong selection for this spacing option.

Since the RARs have been shown to recognize both direct andinverted repeat binding sites, it was surprising to find thatchanging the relative orientation of DR2 in the aiRARE as inm14 would have such a drastic down-effect in competition assaysfor RAR binding. A similar observation has been reported fora consensus GRE. In this case changing the half-sites frominverted to direct repeats resulted in complete loss of GR binding(64). The 1800 rotational inversion of the half-sites from invertedto direct repeats (or vice versa) implies that the relative orientationof the DNA binding domain of the receptor dimer must also bealtered for binding to occur. Our results with m14, and thoseobtained with the GR suggest that the same molecular form ofthe receptor, whether homodiner or heterodimer, cannot undergosuch a rotational inversion of the DNA binding domain. Thus,how the RARs are able to recognize both inverted and directrepeat binding sites is unclear. It is possible that differentmechanisms may exist for selective recognition of palindromicand direct repeat responsive elements. Recently, multiple cell-type specific nuclear factors that modulate DNA binding activityof the RARs to inverted and direct repeat elements have beendescribed (56). Therefore, heterodimer formation among receptorproteins themselves and/or similar interactions with nuclearfactors may play an active role in selective recognition ofpalindromic versus direct repeat RAREs by the RARs. After thiswork was completed several groups have shown that RXRs actas coregulators that heterodimerize to RARs to enhance bindingto target sequences (65 -70).Both the a 1 and ca24 sequences described here compare well

with previously characterized natural and synthetic direct repeatRAREs (Fig 9A). As expected, the highest identity among all8 sequences coincide with the AGGTCA-related motifs. Flankingnucleotides, as well as nucleotides located between the half-siterepeats, are less conserved. Linear alignment of the sequencessuggests a consensus direct repeat RARE consisting of half-sitesof the sequence AG(G/T)TCA spaced by 5 bp. Three deviationsfrom this consensus are the crbpI-, lamBl-, and pepck-RAREs(Fig 9C). The crbpl-RARE solely differs from the consensus inthe spacing of the half-sites (71) and is similar to m12. Theexistence of a natural 2 bp spaced RARE is consistent with ourmutational analysis. By contrast, the lamBI-RARE contains threenon-uniformly spaced AGGTCA-related motifs and all three havebeen shown to be required for maximal activation. In addition,binding of the RARs to the lamBI-RARE was weak (34).Similarly, for maximal activation of the pepck-RARE, sequencesdirectly upstream of the 1 bp spaced imperfect direct repeats havebeen shown to be required (37). These sequences are not relatedto the AGGTCA motif and they are not essential for RARbinding, suggesting that additional protein(s) may be involvedin in vivo recognition of this RARE. The existence of theseatypical response elements suggests that RARs may regulate geneexpression in more subtle ways.The magnitude of the RA response by the a1RARE3/TK-Luc

reporter was 5- to 20-fold lower than that observed with a similarreporter plasmid containing three copies of the (3RARE in allthree cell lines we transfected (data not shown). In addition,receptor overexpression was required in F9 cells for full activationof the alIRARE3, whereas endogenous RARs could activate thej3RARE3 reporter. Our mutational analysis demonstrated thatconverting the aIRARE to a perfect direct repeat of the AGGT-CA motif resulted in increased efficiency of competition forbinding to the RARs, suggesting that DRI is not optimal forbinding. This may partially explain the requirement for receptoroverexpression observed in F9 cells. Our studies have alsodemonstrated that binding of RARs produced in Cos cells to theaIRARE probe is strong. Still, the induction levels of thealRARE3 were not only dependent upon transfected receptorin F9 cells, but also unexpectedly low in CV-1 and Hela cells.Thus, this lack of correlation between binding and transcriptionalactivation makes it unlikely that DNA binding alone can explainthese disparate observations. One interpretation of these resultsis that cell-type specific factor(s) may exist that modulate the RAresponse by binding directly to the RAR-RA complex or bybinding to factors required for transcription. Alternatively,overlapping binding sites for the RARs and complex 2 in theclIRARE may antagonize receptor binding and repress activation.Overlapping binding sites at the ca-fetoprotein and humanosteocalcin gene gene promoters have been implicated in mutualrepression by c-jun/c-fos and members of the nuclear receptorsuperfamily (60,72). Whether complex 2 plays any role inrepression and/or activation of the aiRARE remains to be shown.However, we find it unlikely that binding antagonism occurs atthe cIlRARE between RARs and the factors involved in formationof complex 2, because we have observed that complex 2 or acomplex 2-like activity seems more abundant in untransfectedF9 cells than in Cos and Hela cells (data not shown).

Regardless of the mechanism that may be used to achieve thedifferential activation of the calRARE in a cell-type specificmanner, our results show that F9 cells seem more permissiblefor RA-induced transcriptional activation mediated by thisenhancer element. The embryonic nature of these cells and thefact that RA is potent inducer of their differentiation suggest thatthe a lRARE may regulate a gene involved in development. Asearch in the sequence data bank (GenBank release 65) did notuncover significant homology between the al fragment andknown gene sequences. Southern blot analysis of human genomicDNA revealed discrete bands consistent with a low copy numbersequence (data not shown). We are currently screening a genomiclibrary to isolate the gene(s) regulated by the calRARE.

In conclusion, our work has shown that it is possible to isolateauthentic RAREs from a highly complex DNA mixture by usingwhole genome PCR. This approach may not only be useful toisolate RA-responsive genes, but it may also serve to identifynovel genes regulated by other sequence-specific DNA bindingproteins, provided that a stringent selection procedure can bedeveloped. One possible drawback of our selection procedurebased on coimmunoprecipitation of RAR binding sites seems tobe that the polyclonal antibody we used may stabilize weakinteractions between RARa and DNA. Since low-affinity sitesshould be more abundant in a random sequence, for this approachto be more efficient a different or modified selection strategy maybe needed (41). When the promoter region of a gene has beencloned, whole genome PCR could provide a quick alternativeto end-point deletion analysis to identify a regulatory sequencebound by a known protein factor.

3232 Nucleic Acids Research, Vol. 20, No. 12

ACKNOWLEDGMENTS

We thank J.Grippo for the RXRa expression vector and forcritical reading of the manuscript; P.LeMotte for RAR-yexpression vector; M.L.DePamphilis for plasmid ptkluc;R.Spathis for help with nucleotide sequencing; W.McComas andR.Motyka for oligonucleotide synthesis; W.Danho for peptidesynthesis; and D.Schoenhaut for critical comments duringpreparation of the manuscript. We are especially grateful toM.I.Sherman for support and encouragement during the earlystages of this work.

REFERENCES1. Roberts, A.B. and Spom, M.B. (1984) In Spom, M.B., Roberts, A.B. and

Goodman, D.S. (eds.), The Retinoids. Academic Press, Orlando, Vol. 2,pp.209-286.

2. Moon, R.C. and Itri, L. (1984) In Spom, M.B., Roberts, A.B. and Goodman,D.S. (eds.), The Retinoids. Academic Press, Orlando, Vol. 2, pp.327-371.

3. Tabin, C.J. (1991) Cell, 66:199-217.4. Alles, A. J. and Sulik, K.K. (1989) Teratology 40:163-171.5. Kistler, A. (1981) Teratology, 23:25-31.6. Lammer, E.J., Chen, D.T., Hoar, R.M., Agnish, N.D., Benke, P.J., Braun,

J.T., Curry, C.J., Fernhoff, P.M., Grix, A.W., Lott, I.T., Richard, J.M.and Sun, S.C. (1985) New Engl. J. Med., 313:837-841.

7. Sulik, K.K. and Dehart, D.B. (1988) Teratology, 37:527-537.8. Sherman, M.I. (1986) In Sherman, M.I. (ed.), Retinoids and cell

differentiation. CRC Press, Boca Raton, pp161 - 186.9. Astigiano, S., Barkai, U., Abarzda, P., Tan, S., Harper, M.I. and Sherman,

M.I. (1991) Differentiation, 46:61-67.10. LaRosa, G.J. and Gudas, L.J. (1988) Proc. Natl. Acad. Sci. USA,

85:329-333.11. Levine, R.A., LaRosa, G.J. and Gudas, L.J. (1984) Mol Cell. Biol.,

4:2142-2150.12. Smits, H.L., Floyd, E.E. and Jetten, A.M. (1987) Mol. Cell. Biol.,

7:4017-4023.13. Wang, S.Y., LaRosa, G.J. and Gudas, L.J. (1985) Dev. Biol., 107:75-86.14. Petkovich, M., Brand, N.J., Krust, A. and Chambon, P. (1987) Nature,

330:444-450.15. Giguere, V., Ong, E.S., Segui, P. and Evans, R.M. (1987) Nature,

330:624-629.16. Brand, N., Petkovich, M., Krust, A., Chambon, P., de ThM, H., Marchio,

A., Tiollais, P. and Dejean, A. (1988) Nature, 332:850-853.17. Krust, A., Kastner, P., Petkovich, M., Zelent, A. and Chambon, P. (1989)

Proc. Natl. Acad. Sci. USA, 86:5310-5314.18. Zelent, A., Krust, A., Petkovich, M., Kastner, P. and Chambon, P. (1989)

Nature, 339:714-717.19. Kastner, P., Krust, A., Mendelsohn, C., Gamier, J.M., Zelent, A., Staub,

A. and Chambon, P. (1990) Proc. Natl. Acad. Sci. USA, 87:2700-2704.20. Leroy, P., Krust, A., Zelent, A., Mendelsohn, C., Gamier, J.-M., Kastner,

P., Dierich, A. and Chambon, P. (1991) EMBO J., 10:59 -69.21. Zelent, A., Mendelsohn, C., Kastner, P., Krust, A., Gamier, J.-M.,

Ruffenach, F., Leroy, P. and Chambon, P. (1991) EMBO J., 10:71-81.22. Mangelsdorf, D.J., Ong, E.S., Dyck, J.A. and Evans, R.M. (1990) Nature,

345:224-229.23. Green, S. and Chambon, P. (1988) Trends Genet., 4:309-314.24. Evans, R.M. (1988) Science, 240:889-895.25. Beato, M. (1989) Cell, 56:335-344.26. Green, S., Kumar, V., Theulaz, I., Wahli, W. and Chambon, P. (1988)

EMBO J., 7:3037-3044.27. Mader, S., Kumar, V., de Verneuil, H. and Chambon, P. (1989) Nature,

338:271-274.28. Umesono, K. and Evans, R.M. (1989) Cell, 57:1139-1146.29. de The, H., Marchio, A., Tiollais, P. and Dejean, A. (1989) EMBO J.,

8:429-433.30. Hu, L. and Gudas, L.J. (1990) Mol. Cell. Biol., 10:391 -396.31. de The, H., Vivanco-Ruiz, M., Tiollais, P., Stunnenberg, H. and Dejean,

A. (1990) Nature, 343:177-180.32. Sukov, H.M., Murakami, K.K. and Evans, R.M. (1990) Proc. Natl. Acad.

Sci. USA, 87:5392-5396.33. Vasios, G.W., Gold, J.D., Petkovich, M., Chambon, P. and Gudas, L.J.

(1989) Proc. Natl. Acad. Sci. USA, 86:9099-9103.34. Vasios, G.W., Mader, S., Gold, J.D., Leid, M., Lutz, Y., Gaub, M.-P.,

Chambon, P. and Gudas, L.J. (1991) EMBO J., 10:1149-1158.

35. Mufnoz-Canoves, P., Vik, D.P. and Tack, B.F. (1990) J. Biol. Chem.,265:20065 -20068.

36. Duester, G., Shean, M.L., McBride, M.S. and Stewart, M.J. (1991) Mol.Cell. Biol., 11:1638-1646.

37. Lucas, P.C., O'Brien, R.M., Mitchell, J.A., Davis, C.M., Imai, E., Forman,B.M., Samuels, H.H. and Granner, D.K. (1991) Proc. Natl. Acad. Sci. USA,88:2184-2188.

38. Cochran, B.H., Reffel, A.C. and Stiles, C.D. (1983) Cell, 33:939-947.39. Dworkin, M.B. and Dawid, I.B. (1980) Dev. Biol., 76:449-464.40. Kinzler, K.W. and Vogelstein, B. (1989) Nucleic Acids Res., 17:3645 -3653.41. Kinzler, K.W. and Vogelstein, B. (1990) Mol. Cell. Biol., 10:634-642.42. Gaub, M.P., Ruperte, E., Petkovich, M., Brand, N. and Chambon, P. (1989)

Proc. Natl. Acad. Sci. USA, 86:3089-3093.43. Buluwela, L., Forster, A., Boehm, T. and Rabbitts, T. H. (1989) Nucleic

Acids Res., 17:452.44. Sambrook, J., Fritsch, E.F. and Maniatis, T. (1989) Molecular Cloning:

A Laboratory Manual. Cold Spring Harbor Laboratory Press, Cold SpringHarbor.

45. Martfnez-Salas, E., Linney, E., Hassell, J. and DePamphilis, M.L. (1989)Genes & Development, 3:1493-1506.

46. Glass, C.K., Holloway, J.M., Devary, O.V. and Rosenfeld, M.G. (1988)Cell, 54:313-323.

47. Abarzua, P. and Sherman, M.I. (1990) Methods Enzymol., 189:337-348.48. Chen, C. and Okayama, H. (1987) Mol. Cell. Biol., 7:2745-2752.49. Brasier, A. R., Tate, J.E. and Habener, J.F. (1989) BioTechniques,

7:1116-1121.50. Herbomel, P., Bourachot, B. and Yaniv, M. (1984) Cell, 39:653-662.51. Bradford, M. (1976) Anal. Biochem., 72:248-254.52. Lee, K.W., Bindereif, A. and Green, M.R. (1988) Gene Anal. Techn.,

5:22 -31.53. Dignam, J.D., Lebovitz, R.M. and Roeder, R.G. (1983) Nucleic Acids Res.,

11:1475-1489.54. Klein-HiipaB, L., Schorpp, M., Wagner, U. and Ryffel, G.U. (1986) Cell,

46:1053- 1061.55. Costa-Giomi, P. and Abarzua, P. (Unpublished data)56. Glass, C.K., Devary, O.V. and Rosenfeld, M.G. (1990) Cell, 63:729-738.57. Kornberg, A. (1980) DNA Replication. W. H. Freeman and Company, San

Francisco, C.A.58. Umesono, K., Giguere, V., Glass, C.K., Rosenfeld, M.G. and Evans, R.M.

(1988) Nature, 336:262-264.59. Glass, C.K., Lipkin, S.M., Devary, O.V. and Rosenfeld, M.G. (1989) Cell,

59:697-708.60. Schule, R., Umesono, K., Mangelsdorf, D.J., Bolado, J., Pike, J.W. and

Evans, R.M. (1990) Cell, 61:497-504.61. Levin, A. A., Sturzenbecker, L. J., Kazmer, S., Bosakowski, T., Huselton,

C., Allenby, G., Speck, J., Kratzeisen, C., Rosenberger, M., Lovey, A.,and Grippo, J. F. (1992) Nature, 355:359-361.

62. Heyman, R. A., Mangelsdorf, D. J., Dyck, J. A., Stein, R. B., Eichele,G., Evans, R. M., and Thaller, C. (1992) Cell, 68:397-406.

63. Umesono, K., Murakami, K.K., Thompson, C.C. and Evans, R.M. (1991)Cell, 65:1255 -1266.

64. Dahlman-Wright, K., Wright, A., Gustafsson, J.-A. and Carlstedt-Duke,J. (1991) J. Biol. Chem., 266:3107-3112.

65. Yu, V. C., Delsert, C., Andersen, B., Holloway, J. M., Devary, 0. V.,Naar, A. M., Kim, S. Y., Boutin, J. -M., Glass, C. K., and Rosenfeld,M. G. (1991) Cell, 67:1251-1266.

66. Leid, M., Kastner, P., Lyons, R., Nakshatri, H., Saunders, M., Zacharewski,T., Chen, J.-Y., Staub, A., Gamier, J. -M., Mader, S., and Chambon, P.(1992) Cell, 68:377-395.

67. Zhang, X. -K., Hoffmann, B., Tran, P. B. -V., Graupner, G., and Pfahl,M. (1992) Nature, 355:441-446.

68. Kliewer, S. A., Umesono, K., Mangelsdorf, D. J., and Evans, R. M. (1992)Nature, 355:446-449.

69. Bugge, T. H., Pohl, J., Lonnoy, O., and Stunnenberg, H. G. (1992) EMBOJ., 11:1409-1418.

70. Marks, M. S., Hallenbeck, P. L., Nagata, T., Segars, J. H., Appella, E.,Nikodem, V. M., and Ozato, K. (1992) EMBO J., 11:1419-1435.

71. Smith, W.C., Nakshatri, H., Leroy, P., Rees, J. and Chambon, P. (1991)EMBO J., 10:2223-2230.

72. Zhang, X-.K., Dong, J.-M. and Chiu, J.-F. (1991) J. Biol. Chem.,266:8248-8254.

73. Leroy, P., Nakshatri, H. and Chambon, P. (1991) Proc. Natl. Acad. Sci.USA, 88:10138-10142.