Nonbiased Identification of DNA Sequences That Bind Thyroid Hormone Receptor a1 with High Affinity*

Thyroid hormone receptors are transcription factors that bind to specific DNA sequences and regulate gene expression in a ligand-dependent manner. Although thyroid hormone receptors are known to bind to the hexamer 5’-AGGTCA, it is not known if this represents the optimal binding site. Therefore, a nonbiased strat- egy was used to identify DNA sequences which bind thyroid hormone receptor a1 with high affinity. Such DNA sequences were isolated from a pool of random sequences using a strategy combining an electrophoretic mobility shift assay with the polymerase chain reaction. It was found that thyroid hormone receptor a1 binds with highest affinity to the octamer 5’- TAAGGTCA. Mutation of the two 5’-nucleotides decreased the affinity of thyroid hormone receptor a l for this DNA sequence approximately &fold, and the importance of those nucleotides in receptor binding was confirmed by DNA footprinting. A single copy of the octamer sequence (but not the hexamer AGGTCA) could impart T3 responsiveness to a heterologous pro- moter in a transient transfection assay. The results indicate that the optimal binding site for thyroid hor- mone receptor a1 is 2 base pairs larger than previously thought, and that a single binding site can function as a response element. In addition, we speculate that the optimal binding sites for thyroid hormone, vitamin D, and retinoic acid receptors may not be identical, as had previously been thought. PCR Labeling of by kinasing oligonucleotide primer autoradiography and The were treated with M for 30 min at 90 in loading and electrophoretically separated on a 10% sequencing gel. In


Nonbiased Identification of DNA Sequences That Bind Thyroid
Hormone Receptor a1 with High Affinity* (Received for publication, December 12, 1992, and in revised form, May 27, 1993)

Ronald W. Katz and Ronald J. KoenigS
From the Endocrinology Division, University of Michigan Medical Center, Ann Arbor, Michigan 48109-0678 Thyroid hormone receptors are transcription factors that bind to specific DNA sequences and regulate gene expression in a ligand-dependent manner. Although thyroid hormone receptors are known to bind to the hexamer 5'-AGGTCA, it is not known if this represents the optimal binding site. Therefore, a nonbiased strategy was used to identify DNA sequences which bind thyroid hormone receptor a1 with high affinity. Such DNA sequences were isolated from a pool of random sequences using a strategy combining an electrophoretic mobility shift assay with the polymerase chain reaction. It was found that thyroid hormone receptor a1 binds with highest affinity to the octamer 5'-TAAGGTCA. Mutation of the two 5'-nucleotides decreased the affinity of thyroid hormone receptor a l for this DNA sequence approximately &fold, and the importance of those nucleotides in receptor binding was confirmed by DNA footprinting. A single copy of the octamer sequence (but not the hexamer AGGTCA) could impart T3 responsiveness to a heterologous promoter in a transient transfection assay. The results indicate that the optimal binding site for thyroid hormone receptor a1 is 2 base pairs larger than previously thought, and that a single binding site can function as a response element. In addition, we speculate that the optimal binding sites for thyroid hormone, vitamin D, and retinoic acid receptors may not be identical, as had previously been thought.
Thyroid hormone receptors (TRs)' are ligand-dependent transcription factors which, along with the receptors for steroids, retinoic acid, and vitamin D, are members of the erbA superfamily of transacting proteins (1). TRs bind to triiodothyronine (T3) response elements (TREs) in the promoter regions of many genes and confer ligand-responsive transcriptional regulation. The molecular mechanism of transcriptional regulation is at present unclear. One important component of this regulation is the binding of the receptor onto the TRE. A modest number of TS responsive genes have had their TREs localized (2)(3)(4)(5)(6)(7)(8). Brent et al. (9) performed an * This work was supported by National Institutes of Health Grants DK44155 and DE00301. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
$ T o whom correspondence should be addressed University of extensive mutational analysis of the TRE in the 5'-flanking region of the rat growth hormone (GH) gene, and based upon this study proposed a consensus TR binding hexamer 5'-AGGTC/*A (9). Most of the TREs characterized to date contain multiple copies of variations of this hexamer, although conservation is frequently weak. Recently, Kim et al. (10) reported data showing that sequences surrounding the hexamer influence TR binding and function. In order to understand how variations in DNA sequence affect TR binding and transcriptional activation, we believe that it is necessary to determine the optimal DNA binding sequence of TRs without the bias introduced by the use and alteration of known TREs. By using a selection and amplification technique previously described for the determination of the DNA binding characteristics of myc (ll), we have determined the optimal DNA binding sequence for TRa1 monomers. In the experiments described below we demonstrate that high affinity binding is conferred by an 8-base pair sequence, 5'-TAAGGTCA, and that nucleotide substitutions or chemical modifications of this sequence can be correlated with decreased TR binding and qualitative changes in the ligand responsiveness of reporter constructs in transient transfection assays.

MATERIALS AND METHODS
DNA Construction-Following the design of oligonucleotides described previously ( l l ) , oligonucleotide primers A (5"TCCGA-ATTCCTACAG) and B (5"AGACGGATCCATTGCA) were synthesized. Primer A contains an EcoRI site and primer B contains a BamHI site. In addition, a pool of oligonucleotides of 49 nucleotides in length was synthesized containing the primer A sequence at the 5'-end, an internal random sequence of 18 nucleotides, and the reverse complement of primer B at the 3'-end. This internally random sequence oligonucleotide pool was made double stranded by annealing with primer B and filling-in with a Klenow reaction. The double strandedpool was then purified by polyacrylamide gel electrophoresis.
Selection and Amplification Procedure-Using the electrophoretic mobility shift assay (EMSA) described below, 100 ng of the random pool of double stranded DNA was incubated with 20,000 trichloroacetic acid-precipitable cpm of mouse T R a l (12) reticulocyte lysate translation product and then briefly run into a polyacrylamide gel. The portion of the gel above the region of unbound DNA (as visualized by ethidium bromide staining of an adjacent lane lacking poly(d1. dC) was cut out and the small amount of DNA contained was eluted into 0.5 M NH,OAc, 1 mM EDTA, 0.1% SDS, ethanol precipitated, and amplified by the polymerase chain reaction (PCR) using primers A and B. The PCR protocol utilized 30 cycles of denaturation at 94 "C for 1 min, annealing at 56 "C for 1 min, and extension at 72 "C for 1 min. The selection process was repeated and in the third EMSA an [36S]TRal .DNA complex could be visualized. For subsequent selections, the specific complex as revealed by autoradiography was excised from the dried gel, and the DNA was eluted and amplified by PCR.
Cloning of TRal -binding DNA Pools-PCR-generated TRal-binding DNA pools were ligated into either of the plasmid vectors pCRlOOO (Invitrogen) or pBluescript (Stratagene). For the latter, the DNA was digested with EcoRI plus BamHI and gel purified prior to ligation. Individual bacterial colonies were used to generate plasmid DNA, and PCR with primers A and B was then used to generate the contained 49-bp sequences. These were assayed for T R a l binding activity by performing EMSAs with [%3]TRal (see below). Individual subclones were sequenced using the dideoxynucleotide method with vector primers.
In order to measure the relative in vitro affinity of T R a l for individual sequences, a competition assay was devised. A synthetic palindromic TRE (14) was self-annealed and labeled with "P by a Klenow fill-in reaction. The full sequence of the single stranded oligonucleotide was CTAGAGATCTCAGGTCATGACCTGAGAT-CT (the palindromic TRE hexamer is underlined). Approximately 10,000 cpm of this labeled DNA were incubated with unlabeled in vitro synthesized T R a l (-12,000 cpm eq by volume) for 40 min a t room temperature prior to electrophoresis. Competition was assessed by determining how much unlabeled competitor DNA was required to decrease the intensity of the TRal.TRE,.I complex by 50% (C50; determined by densitometry of autoradiograms). The rat GH TRE (bp -188 to -160) was used in a series of graded amounts to provide competition standardization between different electrophoretic runs.
To distinguish T R a l monomer and dimer complexes, an EMSA was performed using various DNA probes.
These studies utilized T R a l binding sequences that had previously been ligated into the BamHI site of pUTKAT3 (15). The probes were generated by PCR with vector primers, digestion with EcoRI, and Klenow fill-in with [(u-~'P]~TTP. The empty vector yields a 66-bp EcoRI fragment which served as a negative control. Sequences ligated into the BarnHI site included (with the relevant T R a l binding regions underlined, as explained under "Results": top strand sequence only is shown): GATCCTAAGGTCAG (denoted TKMA); -GATCAGCAGGTCAG (TKM1); and the palindromic sequence GATCTGAGGTCAT-GACCTCA (TKPal). The probes containing T R a l binding sites were 80-86 bp in length. Incubations utilized 20,000 cpm of radiolabeled probe and 250,000 cpm eq by volume of nonradiolabeled T R a l (or mock programmed reticulocyte lysate; 1 p1 of lysate volume in either case). These experiments utilized a relatively large amount of T R a l to enhance formation of the dimer band, which is not visible under the conditions of the competition EMSA. Specificity of the TRa1-DNA interaction was tested by competitions with 20 ng of nonradiolabeled DNA (TRE,., or nonspecific DNA from chick p-actin exon Methylution Sensitivity Analysis of TRal Binding-Two forms of methylation sensitivity DNA footprinting were evaluated, using a single representative subclone for both procedures. The ability of guanine methylation to interfere with protein-DNA binding was assessed by a standardized protocol (16). PCR generated DNA was labeled in the "bottom" strand by digestion with EcoRI and incubation with Klenow in the presence of [cY-~'P]~ATP. Labeling of the "top" strand was performed by kinasing oligonucleotide primer A with [y3*P]ATP, using it in a polymerase chain reaction, and purifying the resulting fragment from a native polyacrylamide gel. DNA was methylated by incubation with dimethyl sulfate. The radiolabeled, methylated DNA species were then incubated with Escherichia coli generated T R a l (17) and the protein-bound and free DNAs were separated by EMSA. These DNA species were localized by autoradiography of the wet gel, eluted, and ethanol precipitated. The DNA forms were treated with 1 M piperidine for 30 min at 90 "C, lyophilized, resuspended in loading buffer, and electrophoretically separated on a 10% sequencing gel. In parallel, the radiolabeled DNAs were chemically cleaved at guanine and adenine residues using the method of Maxam and Gilbert to generate a G + A sequencing lane for orientation. The gels were dried and autoradiographed at -70 "C for 2-4 days with an intensifying screen.

3,2096-2125 bp).
T o assay for the importance of the 5-methyl groups of thymine bases for the protein-DNA interaction, a uracil interference procedure was utilized (18). Deoxyuracil incorporation into the DNA sequence was accomplished by spiking a PCR reaction with a small amount of dUTP. The deoxyuracil incorporated DNA was labeled in a manner identical to that used for the methylation interference studies. This material was then incubated with TRal and subjected to EMSA as described above. The deoxyuracil incorporated DNAs were eluted from the wet gel, digested with uracil DNA glycosylase in PCR buffer for 1 h at 37 "C, ethanol precipitated, and cleaved with piperidine. The samples were resolved on the same sequencing gels as used for the guanine methylation interference studies.
Cell Culture and Transfections-JEG-3 cells were grown in 90% Eagle's medium plus 10% iron-enriched calf serum supplement (Life Technologies Inc.) and were transfected using standard calciumphosphate precipitation (13). T R a l was expressed from the vector pCDM (13). Transfections included 3 pg of pCDMTRal (or vector) plus an additional 3 pg pCDM as "filler" plasmid. Ligand responsive reporter plasmids were constructed by ligating putative TRE sequences into pUTKAT3 (15) a t a unique BarnHI site located just upstream of the minimal herpes simplex virus thymidine kinase promoter directing chloramphenicol acetyltransferase (CAT) expression (described above in the "EMSA methods). Reporter plasmids were transfected a t a dose of 4 pg/Petri dish. T o control for transfection efficiency, each transfection also included 0.5 pg of pRSVGH, in which the Rous sarcoma virus promoter directs expression of human GH. Cells in 60-mm Petri dishes were transfected in the presence of 10% charcoal stripped iron-enriched calf serum supplemented with 100 nM dexamethasone, and then cultured for 2 days & a receptor saturating dose of T3 (100 nM). Cell lysates were assayed for CAT activity and media for human GH as previously described (13). Ligand responsiveness is defined as CAT/human GH for cells cultured with ligand divided by CAT/human GH for cells cultured without ligand.
Results are presented as the mean f S.E. for a t least four transfections per construct.
Transfections also were performed in COS cells. The technique was similar to that described above, except the cells were grown in 90% Dulbecco's modified Eagle's medium plus 10% fetal bovine serum.

Nonbiased Selection Reveals a Highly Conserved 8-Base Pair
Sequence-Using the EMSA and PCR amplification of gel eluted DNA, an original pool of double stranded DNA sequences with an internal span of 18 random nucleotides was selected for those species capable of binding TRal.  To determine the relative affinities of specific DNA sequences for T R a l , a competition EMSA was devised using a reproducible T R~u~-[~* P ] D N A complex as a baseline. The labeled DNA sequence consisted of the well characterized synthetic palindrome TRE,.] (5"TCAGGTCATGACCTGA). T o assess T R a l affinity, varying amounts of unlabeled DNA of each of the subclones were added to binding reactions and the decrease in the intensity of the baseline TRal-TRE,,] complex was determined. For comparison, varying amounts of the rat GH gene promoter TRE were used to assess relative affinity and standardize results across several electrophoretic runs.
Using this assay, a Cso (50% competition) value for various TRal-binding DNA sequences was determined (Fig. 2). For the rat GH gene TRE, the CsO was 3.6 ng. Ten individual subclones were characterized with CS0 values clustered around 1.5 ng (1.5 k 0.2, mean f S.E.). A C50 analysis for representative subclone A3 is shown in Fig. 2. Sequencing of these 10 subclones demonstrated that they all contained the 8-bp sequence 5"TAAGGTCA. By performing competitions with a single dose of 5 ng of DNA, an additional 4 subclones were identified which gave -80% competition at that dose. Based upon the slope of the Cso curves for the other 10 subclones, these 4 subclones also would be predicted to have c 5 0 values of 1-2 ng. All 4 of these subclones contained the sequence 5'-TAAGGTCA. Thus, a total of 14 high affinity binding subclones were identified, and all contained this conserved octamer. However, among these 14 subclones, no sequence conservation was detected in the nucleotides surrounding this octamer. In addition, the position of the octamer within the random 18-mer was not conserved, nor was the orientation of the octamer relative to the primers. All other subclones had C50 values greater than 3 ng. Several of these were sequenced, and all contained variations of the octamer sequence (Table  I). Overall, the results suggest that the optimal binding sequence for T R a l is 2 bp larger than previously believed.
Mutation  pairs TRal Binding-In order to confirm that the two nucleotides upstream of the classical TRE hexamer are important for T R a l binding, the following experiment was performed. A representative subclone (A3) was selected and a "mutant" version (A3M) was synthesized that contained substitutions in the two bases upstream of the TRE hexamer, altering the sequence to 5"EAGGTCA. When the original sequence and the mutant sequence were compared by EMSA for their abilities to compete away the TRa1-["*P]TREP.~ complex, it was observed that the substitution of the two nucleotides increased the C50 approximately 5-fold, from 1.5 to 7.7 ng (Fig. 2). Fig. 2 also demonstrates the lack of competition by two subclones (52 and 53) that do not bind TRal. TRal Binds as a Monomer to the Sequence TAAGGTCA-It was important to demonstrate that TRal binds only as a monomer to the sequence TAAGGTCA. Therefore, an EMSA was performed comparing the binding of T R a l t o various DNA sequences, including TAAGGTCA, in the context of a 66-bp EcoRI fragment derived from the vector pUTKAT3 (Fig. 3 lane 7), which again shows specific competition (lanes 8 and 9). When the T R a l binding site was altered to GCAGGTCA, which contains only the traditional single half-site hexamer, binding of TRa1 was weakened, but the mobility of the monomer complex was unaltered (TKM1, lane 10). This weakened binding is consistent with the competition data presented above (Fig. 2), which indicates that, in the context of the full 49-bp A3 clone, T R a l binds to the sequence TAAGGTCA with approximately 5-fold greater affinity that to the mutated sequence GCAGGTCA.
Methylation Sensitivity Confirms That the Binding Sequence Is a n Octamer-DNA footprinting studies were performed to confirm that TRal did indeed bind to the full octamer, and also that important contacts were not made with surrounding bases. Critical guanine residues were identified by methylation interference, and critical thymine residues by uracil interference. As can be observed in Fig. 4, T R a l exhibits markedly reduced binding when any of the guanine residues from either strand in the octamer sequence is methylated. Importantly, the loss of the thymine 5-methyl group by replacement with deoxyuracil markedly reduces the affinity of T R a l for the DNA only when the thymine affected is part of the upstream " T A of the octamer (Fig. 4). In addition it is clear that chemical modification of nucleotides outside the octameric sequence has no major effect on T R a l binding. These data reinforce the observation that the two nucleotides upstream of the classical TRE hexamer play an integral role in the formation of the protein-DNA complex, and that TRal makes critical base contacts only within the octamer sequence.
A Single Octamer, but Not a Hexamer, Can Confer T3 Responsiveness to a Reporter Construct-In order to determine whether the octamer sequence is able to function as a TRE that will increase transcription, the 8-bp sequence 5'-TAAGGTCA was inserted upstream of a minimal thymidine kinase promoter driving CAT expression (pTKMA). Transient transfection of JEG-3 cells with pTKMA plus T R a l led to a T3-dependent 4.2-fold induction of CAT activity (Fig. 5). This induction is quantitatively identical to that seen with the classical TRE in the wild type rat GH promoter (2,19). When the octamer sequence was inserted in the reverse orientation (pTKMB), ligand responsiveness decreased significantly. This suggests that the single octamer requires a specific orientation relative to the initiation site to function as a transcriptional activator.
When the mutant octamer 5"SAGGTCA was inserted into the same reporter construct to create pTKM1, T3 responsiveness was lost (Fig. 5). This mutant sequence retains the idealized T R E hexamer sequence and, as noted earlier, still binds T R a l (albeit with reduced affinity). The lack of T3 responsiveness with pTKMl is not surprising, since studies with the rat GH gene TRE indicate that TI responsiveness is absent if only one of the three hexamers is intact (9).
To confirm the above results, transfections also were carried out in COS cells. Following transfection with T R a l , T3 induced CAT expression from pTKMA 5.6 f 0.9-fold ( n = 4), whereas the induction with pTKM1 was only 2.2 f 0.2-fold ( n = 4). Furthermore, unliganded T R a l did not suppress CAT expression in either cell line. In both COS and JEG cells, the normalized CAT activity for cells transfected with pTKMA and TRal and cultured without T3 was 90% of that for cells transfected with empty vector in place of TRa1. Thus, the T3 induction of CAT activity from pTKMA represents a true increase in gene expression over the basal state, not just a relief of repression induced by unliganded receptor.

DISCUSSION
Despite great interest in the molecular mechanism of transcriptional regulation by erbA superfamily members, it is still unclear how these proteins interact with DNA and other truns factors to produce ligand specific regulation of gene expression. Although up to 5-10% of liver proteins may be regulated by TB (20, 21), only a small number of TREs have been extensively studied by techniques such as footprinting and mutational analysis (2,4,7,9). Perhaps the most extensively studied T R E is that derived from the 5'-flanking region of the rat GH gene. This TRE contains three binding sites (commonly called half-sites) for TR. Starting with the wild type sequence, Brent et al. (9) performed an extensive mutational analysis that demonstrated all three half-sites are important for full T3 responsiveness. Furthermore, these studies led to the proposal that the optimal TR binding sequence is 5'-AGGTC/AA. Indeed, most TREs identified to date contain one or more variations of this hexamer, although rarely is the exact hexamer found. However, even though it is clear this sequence binds TR, it is not clear it represents the optimal binding sequence. The sequence 5'-AGGTC/AA was derived with a pre-existing bias as to what a TRE sequence should look like (the rat GH TRE), and practical considerations limited the number of specific mutations that could be tested. The potential importance of sequences outside the hexamer was not investigated, possibly because of a preconceived bias derived from steroid hormone response elements (22) that binding half-sites are likely to be hexamers.
Not only do TREs vary in half-site sequence, but the halfsites can be oriented as direct repeats (4), palindromes (4,14), or inverted palindromes (7,23). Furthermore, the spacing between half-sites is not easily predicted. While directly repeated half-sites usually are separated by 4 bp (24,25), exceptions exist (26,27). Palindromic TREs can have spacing of 0-5 bp (4,14,27,28), and inverted palindromic TREs have been reported with spacings of 2 and 4 bp (7,23). Most of the TREs studied to date appear to contain 2 or 3 half-sites, but the a-glycoprotein subunit gene negative TRE appears to contain just one (29).
In addition, considerable uncertainty exists as to the nature of the actual protein that binds to each of the half-sites of any TRE in uiuo. Using the rat GH TRE as a model, it has been shown that TR can bind all three half-sites in vitro, and the effects of mutations in any half-site on TR binding in vitro correlate closely with the effects on T3-dependent reporter gene activation in transfected cells (3,9). These data suggest TR occupies all three half-sites in vivo. However, TR binds to previously characterized positive TREs (all of which contain at least two half-sites) in vitro with a higher affinity as a heterodimer with the retinoid X receptor (RXR) than as a monomer or homodimer (30)(31)(32)(33)(34), and mutations that disrupt heterodimerization also impair T3-dependent gene activation (35). Furthermore, cotransfecting cells with RXR enhances T3-dependent reporter gene expression (30)(31)(32)(33)(34)36). Thus, it is currently felt that TR-RXR heterodimerization may play a central role in T3-dependent gene activation, although it is unclear whether specific TRE half-sites bind TR or RXR in vivo.
Clearly, a better understanding of the TR-DNA interaction is required to understand the mechanism of T3-dependent gene activation. We wished to address this problem without any preconceived bias derived from studies of the few already characterized TREs. In addition, we felt that the great variation in number, orientation, and spacing of half-sites in known TREs, as well as the uncertainty as to which half-sites actually are occupied by TR, would seriously confuse the evaluation of TR binding sequences selected for homodimer, heterodimer, or multimer binding properties. Rather, we felt the initial nonbiased characterization of TR binding sites would be easiest to interpret if we focused on single half-sites that could bind TR monomers. This information could then be used to reassess what factors actually bind to naturally occurring TREs with multiple half-sites, and could be used as a frame of reference to interpret sequences selected in the future for binding TR homodimers, TR-RXR heterodimers, and possibly multimers.
Our data indicate that the highest affinity binding site for T R a l is the octamer 5"TAAGGTCA. It is interesting to ask whether the unexpected 5'-nucleotides (TA) are found in any of the already well characterized TREs. A review of the rat GH (4,9), malic enzyme (4,5), and a-myosin heavy chain (4, 6, 37) TREs indicates that these elements contain a total of 8 half-sites, with the initial Thd being present in 4 and the following Ado in but 1. The remaining nucleotides 3-8 of the octamer (AGGTCA) all are present in at least 4 of these 8 half-sites. Thus, although the initial TA may be less well conserved than the traditional hexamer in these TREs, it remains possible that half-sites which lack this initial Thd and/or Ado are the ones that prefer to bind RXR, not TR, in vivo.
All currently known TREs that induce gene expression contain at least two half-sites. In addition, mutation of any two of the three rat GH gene TRE half-sites totally abolishes T g induction (3,9). Thus, it has been thought that positive TREs require at least two half-sites. Given this, we were interested to test the T3 responsiveness of a single octamer half-site. Surprisingly, these studies indicate this octamer can confer a modest 4-5-fold T3 induction upon a neutral heterologous basal promoter. This response is abolished by mutation of the 5"TA within the octamer. It is important to emphasize that the wild type rat GH TRE, which has three imperfect half-sites, is no more active than this single octamer (2,19). Indeed, our EMSA data indicate TRal binds a single octamer with a higher affinity than it binds the intact rat GH TRE with all three half-sites. Thus, although wild type TREs can be mutated and polymerized to create constructs that show 25 or more fold TS induction (9), the single octamer TRE yields an hormonal response similar to that of a naturally occurring TRE. The ability of the octamer to function as a TRE was orientation dependent, suggesting that when T R a l occupies the element in reverse orientation it cannot interact favorably with other transcription factors. This is not surprising considering that the octamer is not a palindrome.
The conclusion that a single octamer is a functional TRE would be compromised if a cryptic TRE half-site had been introduced into pTKMA during construction of this reporter vector. However, this remote possibility is excluded by the EMSA presented in Fig. 3, which shows that only a TR monomer-DNA complex forms when T R a l is incubated with an 80-bp EcoRI fragment from pTKMA that includes a minimum of 23 bp of sequence flanking each side of the TRE octamer.
These data lead to the prediction that certain naturally occurring positive TREs may consist simply of a single octamer half-site. Since TR would presumably bind as a monomer to these TREs, the TS response probably would not be dependent on the presence of RXR. Thus, genes with this class of TRE might be regulated in a very different manner than the currently known genes with positive TREs.
Since TRs, retinoic acid receptors, vitamin D receptors, and RXRs all bind to the sequence AGGTCA, it is likely all will bind to the octamer TAAGGTCA. However, we would speculate that only T R a l would bind this octamer with the highest affinity, and that alterations in the first two octamer nucleotides may create optimal binding sites for these other erbA superfamily members. Thus, although it has been proposed that half-site spacing determines whether a response element is specific for TRs, retinoic acid receptors, or vitamin D receptors (24, 25), it is possible that the full half-site sequence (octamer, not hexamer) also helps determine receptor (and hormone) specificity.
Recent studies by Kim et al. (10) bear relevance to our work. These authors studied a sequence near the TATA box of the rat GH promoter that can function as a negative TRE. By performing a mutational analysis of the TR binding site in this region, they proposed that an optimal half-site is 10 bp long, with 2 extra bp both 5' and 3' to the traditional hexamer. Our nonbiased selection strategy supports the assignment of one of their four novel bases, the most 5' Thd. However, we find the second base to be an Ado, whereas they proposed Guo, and we find no conservation of sequence (or footprinting) 3' to the classical hexamer. Indeed, their data in support of assignment of the 9th and 10th nucleotides appear weaker than that which supports the first two nucleotides. In addition, they did not test adenine in the second position, illustrating the practical limitations inherent in a nonrandom mutational analysis. However, it should be noted that their transfection data were performed in the context of the basal rat GH promoter, and it is likely that both context and sequence together determine hormonal responsiveness.
In summary, our data show that the optimal TRa1 binding sequence is an octamer, and that a single such sequence is as potent a TRE as is the wild type rat GH gene TRE. Expansion of these studies to identify optimal binding sites for RXRs and other erbA superfamily members as monomers, dimers, and multimers should help clarify what constitutes an hormone response element and what trans factors occupy each half-site in the element.