Sequence Requirements for Estrogen Receptor Binding to Estrogen Response Elements*

The estrogen receptor (ER) is a transcription factor that binds to a specific DNA sequence found in the regulatory regions of estrogen-responsive genes, called the estrogen response element (ERE). Many genes that contain EREs have been identified, and most of these EREs contain one or more changes from the core consensus sequence, a 13-nucleotide segment with 10 nucleotides forming an inverted repeat. A number of genes have multiple copies of these imperfect EREs. In order to understand why natural EREs have developed in this manner, we have attempted to define the basic sequence requirements for ER binding. To this end, we measured the binding of homodimeric ER to a variety of nonconsensus EREs. We discovered that an ERE containing even a single change from the consensus may be unable to bind ER. However, an ERE with two changes from the consensus may be capable of binding avidly to ER in the context of certain flanking sequences. We found that changes in the sequences flanking a nonconsensus ERE can greatly alter ER-ERE affinity, either positively or negatively. Careful study of sequences flanking a series of EREs made it possible to develop rules that predict whether ER binds to a given natural ERE and also to predict the relative amounts of binding when comparing two EREs.

The estrogen receptor (ER) is a transcription factor that binds to a specific DNA sequence found in the regulatory regions of estrogen-responsive genes, called the estrogen response element (ERE). Many genes that contain EREs have been identified, and most of these EREs contain one or more changes from the core consensus sequence, a 13-nucleotide segment with 10 nucleotides forming an inverted repeat. A number of genes have multiple copies of these imperfect EREs. In order to understand why natural EREs have developed in this manner, we have attempted to define the basic sequence requirements for ER binding. To this end, we measured the binding of homodimeric ER to a variety of nonconsensus EREs. We discovered that an ERE containing even a single change from the consensus may be unable to bind ER. However, an ERE with two changes from the consensus may be capable of binding avidly to ER in the context of certain flanking sequences. We found that changes in the sequences flanking a nonconsensus ERE can greatly alter ER-ERE affinity, either positively or negatively. Careful study of sequences flanking a series of EREs made it possible to develop rules that predict whether ER binds to a given natural ERE and also to predict the relative amounts of binding when comparing two EREs.
The estrogen receptor (ER) 1 is a transcription factor that regulates the expression of estrogen-responsive genes by binding to a specific DNA sequence found in their regulatory regions. This sequence is called the estrogen response element (ERE).
Many reports analyzing estrogen-responsive genes revealed a common sequence motif that was anticipated to be the ERE (1)(2)(3)(4)(5)(6). Experiments that placed this sequence upstream of a reporter gene confirmed that a 13-bp inverted repeat was an essential element for estrogen responsiveness (3). The essential ERE was determined to have the consensus sequence 5Ј-GGT-CAnnnTGACC-3Ј (4,7). The symmetry of the sequence was found to facilitate the binding of ER as a homodimer (8,9). However, only a handful of the most highly estrogen-responsive genes contain perfect consensus EREs. Many genes have been found to contain sequences that appear to be EREs, but most of these vary from the consensus by one or more nucleotides. Studies of ER binding showed that one or more changes from the consensus sequence resulted in lower ER-ERE affinity (10 -12) and that sequences immediately flanking the ERE impact ER-ERE binding (13). However, none of these reports delineated the requirements for ER binding to nonconsensus EREs. The results presented here define characteristics of nonconsensus EREs that determine whether or not productive ER-ERE binding occurs.
Highly Purified Human ER␣-Recombinant human ER␣ from baculovirus-infected insect cells was purchased from Panvera Corp. (Madison, WI) at a concentration of 5000 or 3400 nM for use in gel shift-based determinations of ER-ERE affinity and ERE competition assays.
ERE-containing Oligomers-Both strands of all oligomers were synthesized and polyacrylamide gel electrophoresis-purified by Genosys Biotechnologies (The Woodlands, TX). The regions flanking all sequences tested were identical (Fig. 1A). The oligomers were annealed, and 0.5 g was 32 P-end-labeled by Escherichia coli DNA polymerase I Klenow fragment incorporation of [␣-32 P]dATP and [␣-32 P]TTP (3000 Ci/mmol, NEN Life Science Products).
Gel Shift ERE Binding Assays-32 P-End-labeled DNA (60,000 cpm) and 0, 10, 20, 30, 40, or 50 l of ER were incubated with poly(dI-dC) (0.45 g, Midland Certified Reagents, Midland, TX) and an ER-supershifting antibody (either c-311 from Santa Cruz Biotechnologies, Inc., Santa Cruz, CA or H222, a gift from Abbott) in a total reaction volume of 60 l. Reactions were incubated at 0°C for 60 min and loaded onto a 5% nondenaturing polyacrylamide gel. Electrophoresis was performed at 150 V for 1.5 h at room temperature in 1ϫ TBE (100 mM Tris base, 0.831 mM boric acid, 1 mM EDTA). The gel was dried at 80°C under vacuum and exposed to film overnight.
ERE Competition Assays-6 l of 3400 nM recombinant human ER was diluted with 1040 l of TDPEK 100, 20% glycerol, 0.1% Nonidet P-40 containing 8.5 g of poly(dI-dC). 15 l of the ER-poly(dI-dC) mixture was added to tubes containing 6 g of 32 P-labeled p17 ERE that had been premixed with increasing concentrations of the appropriate competitor. Total reaction volume was 18 l. The mixture was incu-bated on ice for 1.5 h. 10 l was loaded onto a 5% native polyacrylamide gel and separated by electrophoresis at 150 V for 1.5 h. The gels were dried at 80°C under vacuum and exposed to a PhosphorImager screen (Molecular Dynamics, Inc., Sunnyvale, CA) overnight. DNA counts in the appropriate bands were quantitated and graphed.
Gel Shift-based Determinations of K d -Oligomers were dissolved at concentrations above 1000 ng/l, as determined by fluorimetry (Hoefer DynaQuant 200, Hoefer Instruments), and 32 P-end-labeled using E. coli DNA polymerase I Klenow fragment (3Ј-5Ј exo-) (New England Biolabs, Beverly, MA) incorporation of [␣-32 P]dGTP (3000 Ci/mmol, NEN Life Science Products). Labeling was stopped by heating the mixture to 70°C for 10 min, as per the manufacturer's instructions. The specific activity of labeled DNA was determined as follows. One-fortieth volume of the crude labeling mixture was spotted onto a TLC plate (polyethyleneimine-cellulose, EM Science, Gibbstown, NJ), and the plate was developed in an aqueous solution of 0.37 M HCl with 0.1 M sodium pyrophosphate. The unincorporated counts migrated with the solvent front, whereas the labeled DNA remained at the origin. The plate was divided into thirds, and each third was counted separately in scintillation vials containing EcoScint A (National Diagnostics, Atlanta, GA). The incorporated radioactivity at the origin and the quantity of DNA in the reaction were used to calculate the specific activity of each ERE. Increasing concentrations of each [ 32 P]DNA sample (1-l volume) were mixed with a constant amount (30 l) of diluted ER (6 l of 5000 nM Panvera ER␣, 1294 l of TDPEK 100, 20% glycerol, 0.1% Nonidet P-40 containing 3.0 g of poly(dI-dC)). Reactions were incubated for 1.5 h at 0°C, loaded onto a 5% native gel, and separated by electrophoresis at 150 V for 1.5 h. The gels were dried at 80°C under vacuum and exposed to a PhosphorImager screen overnight. DNA counts in the appropriate bands were quantitated and graphed. A semilog plot of the saturation curves was used to determine the ERE concentration at the midpoint of the maximal amount of ER⅐ERE complex. Actual K d values were calculated using the equation, where [ERE midpoint ] is the concentration of labeled ERE that results in a gel shift of half-maximal intensity (15,19,20). The concentration of ER had to be determined independently, because a significant proportion of the recombinant Panvera ER was incapable of binding ERE. The concentration of active ER was measured by titrating known ERE concentrations against a constant but unknown concentration of ER and subjecting the reactions to gel shift. In the shifted complex, the amount of ERE bound equaled the amount of ER bound. Since both the bound and free ERE were quantitated using the PhosphorImager, and the total amount of ERE was known, the molar quantity of bound and free ERE could be determined. At a saturating concentration of ERE, the maximal amount of ERE bound equaled the concentration of ER. This value was used to calculate K d .

RESULTS
DNA Substrates Used for ER Binding-We designed a series of ERE oligomers varying in sequence in our effort to define the requirements for stable binding to ER. Single-stranded oligomers containing variants of the ERE core consensus sequence were synthesized, gel-purified, and annealed, to create double-stranded DNA. Each substrate consisted of a test segment, having a sequence resembling the core consensus ERE, embedded in a larger oligomer. We call the additional length of oligomer the "background sequence." The background nucleotides were arranged to avoid any sequences related to the core consensus ERE. Therefore, the background is not able to stabilize ER-ERE interaction, except by providing a natural DNA extension on either side of the ERE. The background sequence is shown in Fig. 1A, and the test sequences are shown in Fig.  1B. The rationale for choice of the test sequences will be explained below. Binding was measured by gel shift, under conditions previously found to allow efficient ER-ERE binding (15).
ER Binding and Competition Assays-We performed binding assays using partially purified ER obtained from calf uteri, at concentrations of ER under 10 nM. We did this for two reasons: first, to have our in vitro assays approximate the levels of ER found inside the cell (21) and, second, to have the ER produced in, modified in, and obtained from a mammalian source in a mixture of proteins that naturally occur with ER. The binding assays then served to test whether the results obtained using the highly purified recombinant baculovirus-expressed ER apply to ER-purified directly from calf. Antibody was used to supershift the ER-containing complexes in the binding assays, FIG. 1. A-D, sequences of ERE constructs. A, the base oligomer sequence that surrounds the ERE test sequences. Braces enclose the region into which the test sequence is inserted. B, the test sequences, from 5Ј to 3Ј, surrounded by braces, that were inserted into the base oligomer. Only the upper strand of the test sequences is shown. The 15and 17-bp consensus sequences were based on the sequence of the highly estrogen-responsive Xenopus vitellogenin A2 (vit A2) gene. A change in one base was denoted as d1, two changes as d2, and three as d3. Underlining emphasizes changes from the core consensus. Boldface type indicates bases that have been altered from the vitellogenin A2 sequence in the flanking region. Lowercase type highlights nucleotides in the spacer region. Except for 13d1, EREs with a single change from the consensus were numbered using the base designations in D. C, the vitellogenin B1 ERE oligomers are shown in their entirety. Changes from the core 13-bp consensus ERE are underlined. The sequence flanking the ERE is that which is found in the vitellogenin A2 gene (2). The complementary strand (not shown) contained a 5Ј overhang sequence 5Ј-TAA-3Ј. D, the extended consensus inverted repeat. The 13-bp core consensus is underlined. The central 3-bp spacer is shown in lowercase. The flanking sequences are not underlined. The bases in the ERE are numbered according to their distance from the nucleotide at the center of the inverted repeat, with negative numbers on the left of the center, and positive numbers on the right. Labeled arrows below the sequence delineate the sizes of the inverted repeats referred to in the text. which resulted in a clear separation of the ER-containing shifts from those that were the result of non-ER proteins. Because of concerns that antibodies could affect the amounts of shifted complex, two different antibodies were used, in separate experiments. Both antibodies, H222 and c311 (see "Experimental Procedures"), gave identical results. One set of results is shown. The results of the experiments containing antibody were identical to those performed without antibody, except that the supershift allowed clear visualization of low levels of ER-ERE binding. For ERE competition experiments and for determinations of ER-ERE affinity, pure recombinant baculovirusexpressed human ER from Panvera was used because it was critical to have pure ER at high concentration. The correlation of results between ER-ERE binding assays, ER-ERE affinities, and ERE competition assays was absolute, showing that gel shift is a very robust and sensitive method for comparing ER binding to consensus and nonconsensus EREs. In addition, the results of experiments using human ER mirrored those seen with calf ER. Both estradiol-liganded and -unliganded ER behaved in an identical manner in gel shift assays using the consensus ERE (Ref. 15 and data not shown). All assays shown were performed with unliganded ER. Nucleotides within the test sequences are defined by the numbering system shown in Fig. 1D.
A Single Change in the Consensus ERE Abolishes ER-ERE Binding- Fig. 2, lanes 1-11, contains gel shift results comparing the binding of ER to two EREs that differ by a one-base pair substitution. Oligomer p13, shown in lanes 1-6, contains the consensus 5Ј-GGTCAnnnTGACC-3Ј inverted repeat. ER-ERE binding produced a prominent supershifted band. The sequence 13d1 in lanes 7-11 ( Fig. 2) contained a T to G base substitution in the Ϫ4-position (Fig. 1D). No supershifted band was evident, indicating that ER was not able to bind to this oligomer. Therefore, we concluded that ER-ERE binding can be disrupted by a single base change from the consensus. We extended these findings by testing ER binding to other EREs with single base substitutions at each position in one arm of the inverted repeat (Fig. 1B). In no case was any supershifted ER⅐ERE complex visible (data not shown).
Because ER binds ERE sequences as a homodimer (8,9,(22)(23)(24)(25)(26) and each ER molecule contacts a different arm of the ERE inverted repeat (27), any change made to an arm of the ERE should increase or decrease ER-ERE affinity in a manner identical to a change in the same position of the other arm of the inverted repeat. This was tested by comparing 13d1, which contains a T to G change at the Ϫ4-position of the ERE, to 13G1, where the ϩ4-position of the 13-bp ERE was changed from an A to a G. Both changes resulted in a loss of ER-ERE binding. Also, the Ϫ4-position in 13d1 was changed from a pyrimidine to a purine, whereas the ϩ4-position was changed from a purine to another purine. It was possible that ER-ERE affinity could be dependent on the general structure of the nucleotide, but this did not appear to be the case. We concluded that any change made in the left arm of the inverted repeat can be considered to have an effect identical to the mirror change made in the right arm of the inverted repeat. Although all possible base substitutions were not tested, these results are fully consistent with the hypothesis that all changes in the core consensus are equivalent in their disruption of binding.
Appropriate Flanking Sequences Allow ER Binding to Variants of the Core Consensus ERE-We considered whether sequences flanking the core 13-bp ERE could alter ER-ERE binding. The first experiment (Fig. 2, lanes 7-11) involved the labeled 13d1 oligomer (Fig. 1B), which has a single base change from the consensus 13-bp ERE. No supershifted ER was visible, indicating that ER was unable to bind this element. Lanes 12-16 ( Fig. 2) contained the 15d1 ERE, which has the same change from the consensus as 13d1. However, the nucleotides on either side of the 13-bp ERE have been substituted with nucleotides identical to those that immediately flank the core inverted repeat of the potent vitellogenin A2 ERE. ER binding to this new ERE produced an intense supershifted band. Lanes 17-21 show the binding of ER to 17d1, a 17-bp inverted repeat containing a single change from the consensus, with two nucleotides on either side of the core ERE converted to those of the vitellogenin A2 ERE. The appearance of a dark supershifted band confirmed that ER was also able to bind well to this ERE sequence. Although a single change from the consensus was enough to abolish ER binding in the presence of 13d1 background sequences, the sequences adjacent to the core in 15d1 and 17d1 rescued ER-ERE binding. Fig. 3 shows gel shift assays that contain substrates designed to explore the contributions of 5Ј A residues immediately adjacent to the core 13base pair ERE. As was seen previously, there was no binding of ER to 13d1 (Fig. 3, lanes 1-6). The addition of a single A residue adjacent to the 13d1 sequence (substrate 13d1 ϩ A, Fig.  3, lanes 7-12) resulted in the recovery of ER binding. Lanes 13-18 show ER binding to 15d1, which has a 5Ј A residue FIG. 2. ER binding to EREs with a single change from the consensus. The ability of ER to bind to a consensus or nonconsensus ERE in the context of the same flanking sequences was tested. The sequence of each ERE is shown, and the 13-bp core ERE sequence is boxed. The position of supershifted ER⅐ERE complex and free DNA is indicated to the left. The position of an ERE complex that does not contain ER is also indicated. Oligomer p13 contained the 13-base pair consensus ERE, and 13d1 contained the minimal consensus with a single base change (Fig.  1B), 15d1 had a 15-bp inverted repeat with the same base change, and 17d1 had a 17-bp inverted repeat also with the same base change. 32 P-End-labeled ERE was incubated with 0, 10, 20, 30, 40, or 50 l of Yellow 86 purified ER, 0.45 g of poly(dI-dC), and c-3ll antibody. The mixture was incubated at 0°C for 1 h and loaded onto a 5% nondenaturing polyacrylamide gel. Following electrophoresis, the gel was dried, and the bands were detected by autoradiography. present on both strands of the oligomer. The sensitive nature of this assay allowed the determination that there was clearly more ER binding to 15d1 than to 13d1 ϩ A. This result demonstrates that ER-ERE binding is influenced by the sequence at both of these sites.

5Ј-Flanking A Residues Rescue ER Binding to Nonconsensus EREs and Contribute to ER Affinity-
A Single 5Ј-Flanking G Residue Reconstitutes ER Binding to a Nonconsensus ERE, Showing Purines Are Preferred Immediately Flanking the 13-bp ERE-ER binding to an ERE with a single base change depends on the sequence immediately flanking the 13-base pair inverted repeat. The base oligomer around the 13d1 ERE included T and C at the Ϫ7-position of the upper and lower strand, respectively, but ER was unable to bind. This indicated that pyrimidines at this position inhibited ER-ERE binding. The addition of an A at the Ϫ7-position of either strand, however, supported ER binding (Fig. 3). In order to test whether the sequence preference for optimal ER-ERE binding included only A at the Ϫ7 position, ER-ERE binding to 13d1 ϩ G was tested (Fig. 4). In fact, the addition of a G residue at the Ϫ7-position reconstituted approximately the same level of ER binding as an A residue. For both 13d1 ϩ A and 13d1 ϩ G, the level of ER bound was significant but was below the level seen for ER binding to 15d1 (Figs. 3 and 4). This result suggests that a 15-bp extended ERE consensus sequence includes either G or A at the Ϫ7-position of each strand.
Substitutions That Are Two Nucleotides Distant from the 13-bp Core Consensus Influence ER-ERE Binding-We also needed to know whether sequences beyond those immediately flanking the core ERE affect ER binding affinity. The 17d1 oligomer was used to test whether increasing the inverted repeat beyond the A at Ϫ7 in 15d1 contributes to ER affinity. However, ER binding to 17d1 and 15d1 was not noticeably different. The high affinity binding of ER to either sequence resulted in a very stable ER⅐ERE complex. This made it difficult to determine whether additional vitellogenin A2 flanking sequences present in 17d1 resulted in more ER-ERE binding than was seen for 15d1 (data not shown). To overcome this difficulty, ER binding to 17d1AT was compared with ER bind-ing to 15d1AT. Reduced overall ER binding allowed a better assessment of possible low level binding enhancement to 17d1AT versus 15d1AT. Gel shift assays were performed using oligomer 15d1AT, in which the 5Ј A residue in position Ϫ7 on both strands, immediately flanking the imperfect core ERE, was exchanged for a T. This substitution conserved the length of the inverted repeat and the presence of an A-T base pair flanking the imperfect sequence on both sides but put the A-T base pair in an opposite orientation with respect to 15d1 (Fig.  1). This resulted in complete loss of ER binding, identical to the result seen for 13d1. Because ER was unable to bind 15d1AT, any ER binding to 17d1AT can be attributed to the extended flanking sequences. Subsequently, gel shifts were performed in parallel using 17d1AT, the same sequence as 15d1AT, except for the addition of a 5Ј C residue at position Ϫ8 on both strands (Fig. 1), similar to the sequence of the vitellogenin A2 ERE. The presence of the C in the Ϫ8 position of both strands resulted in a supershifted ER⅐ERE complex (data not shown). A single C on one strand of the ERE at Ϫ8 was not sufficient to reconstitute ER binding in 15d1AT. Recovery of ER-ERE binding required the correct sequence at position Ϫ8 on both strands of the 13d1 ERE. This shows that the identity of bases two residues distant from the core consensus is important for ER-ERE binding to nonconsensus EREs.
Flanking Sequences Can Compensate for One or Two, but Not Three Changes from the Consensus ERE-The gel shift in Fig.  5 demonstrates the decrease in binding of ER as the ERE sequence varies from the consensus. In lanes 1-6, where the ERE has only a single change from the consensus, a strong ER supershift was seen. In lanes 7-12, where the ERE has two changes from the consensus, there was a relative decrease in FIG. 3. A single flanking A residue rescues ER binding to a nonconsensus ERE. ER binding to EREs that contained no 5Ј-flanking A (13d1), one 5Ј-flanking A (13d1 ϩ A), or two 5Ј-flanking A residues (15d1), were compared. The sequence of each ERE is shown, and the 13-bp core ERE sequence is boxed. Oligomer 13d1 contained the minimal consensus with a single base change of T to G at position Ϫ4; 13d1 ϩ A had the same base change plus a 5Ј A residue added at position Ϫ7. 15d1 had a 15-bp inverted repeat with the same base change, with a 5Ј A residue on both strands of the ERE. The position of supershifted ER⅐ERE complex and free DNA is indicated to the left. The position of an ERE complex that does not contain ER is also indicated. ER/ 32 P-ERE mixtures were incubated with poly(dI-dC) and antibody and resolved as in Fig. 2.   FIG. 4. A flanking G residue rescues ER binding to a nonconsensus ERE. ER binding to EREs that contained a 5Ј-flanking G (13d1 ϩ G) or two 5Ј-flanking A residues (15d1), were compared. The sequence of each ERE is shown, and the 13-bp core ERE sequence is boxed. Oligomer 13d1 ϩ A contained the 13-bp consensus with a single base change T to G at position Ϫ4 plus a 5Ј G residue added at position Ϫ7. 15d1 had a 15-bp inverted repeat with the same base change, with a 5Ј A residue on both strands of the ERE (Fig. 1B). The position of supershifted ER⅐ERE complex and free DNA is indicated to the left. The position of an ERE complex that does not contain ER is also indicated. ER/ 32 P-ERE mixtures were incubated with poly(dI-dC) and antibody and resolved as in Fig. 2. ER-ERE binding. In lanes 13-18, where the sequence incorporates three changes from the consensus, there was no ER binding. Similar results were seen for 17d1, 17d2, and 17d3 (data not shown). These experiments reveal that, with the appropriate flanking sequences, up to two changes from the core 13-bp consensus can support ER-ERE binding. However, there was no case where we observed ER-ERE binding when the ERE contained more than two changes from the core consensus.
A Single A in the Immediately Adjacent 5Ј-Flanking Sequence Compensates for One, but Not Two, Changes in the Core Consensus-Can one flanking nucleotide compensate for two changes in the core sequence? The experiment in Fig. 6 was designed to test the binding of ER to a sequence that contained the same two changes from the consensus as the construct 15d2 but had only one compensatory A residue, at the Ϫ7-position of one strand. As seen in Fig. 5, lanes 7-12, 15d2 was able to bind ER, showing that the presence of A residues at Ϫ7 on both strands was sufficient to compensate for two changes to the sequence core. In Fig. 6, however, there was clearly no ER binding to 13d2 ϩ A, showing that a single flanking sequence cannot compensate for two changes to the core. This result is consistent with the pattern in which each single change in the 13-bp inverted repeat requires a compensatory change in the immediate flanking sequence.
Competition with Selected EREs Confirms Results of Direct Binding Assays- Fig. 7A demonstrates how different unlabeled EREs compete with labeled p17 for binding to ER. This experiment was repeated at least three times for each oligomer, with similar results. The EREs with the highest affinity for ER, p15 and p17, effectively competed away the shifted complex. At a 40-fold excess of unlabeled over labeled ERE, less than 5% of the initial complex remained. The essentially indistinguishable curves show that the affinities of both of these EREs for ER are similar. The 15d1 ERE, which has the 15-bp inverted repeat with a single change from the consensus, did not compete as efficiently as the intact 15-and 17-bp inverted repeats. Inter-estingly, however, 15d1 competed more efficiently than the perfect 13-bp inverted repeat, showing that this imperfect ERE has higher affinity for ER than the perfect p13 sequence. The 15d2, with two changes from the p15 consensus, and the 17d1AT, which has only the outermost vitellogenin A2 flanking sequences intact at Ϯ8-positions, have much lower, but similar affinities for ER. 15d2 and 17d1AT were the constructs that had the lowest levels of detectable binding in the gel shift assays performed at physiological levels of ER, as above. Neither 15d3, which had three changes from the p15 consensus, nor 13d1, which had a single change from the p13 consensus, bound detectable levels of ER in the binding assays. The competition experiments revealed that they both have extremely low affinity for ER. After a 160-fold excess of 15d3 was added, 90% of the binding activity remained. We were unable to detect any competition by 13d1 for ER binding, even at a 320-fold excess of ERE. It was surprising that the effect of this single change was so profound.
ER-ERE Affinity with Selected Consensus and Nonconsensus EREs-Gel shift assays were used to calculate the binding affinity of p13, p15, and 15d2 for ER. Increasing concentrations of ERE were added to a constant amount of ER. The bound and free ERE bands were quantitated using the PhosphorImager. Semilogarithmic plots of ERE titrations resulted in three distinctly different S-shaped saturation curves. The experiment was repeated three times for each construct, and the average value Ϯ S.E. was graphed (Fig. 7B). Affinities were determined as outlined under "Experimental Procedures." The affinity of ER for the p17 inverted repeat (0.25 nM) has been previously determined by other methods (15). Because both the binding and competition assays showed little difference between p17 and p15 affinities for ER, we expected the calculated affinity of p15 to be close to 0.25 nM. In fact, the calculated affinity was 0.54 nM, very close to the expected value. Because 15d2 bound ER with low affinity, it was not possible to achieve saturation, even at maximal concentrations of 15d2. However, at saturation the amount of bound DNA equals the amount of active ER. Because the amount of ER used in each experiment was identical, the maximum possible amount of bound 15d2 equals the amount of active ER. Since we could use this anticipated saturation value, the points on the curve at high concentrations of FIG. 5. The effect of one, two, and three changes from the consensus ERE on ER binding. Lanes 1-6 show ER binding to 15d1, an ERE with one change from the consensus. ER binding to 15d2 is shown in lanes 7-12. ER binding to 15d3 is shown in lanes 13-18. The sequence of each ERE is shown, and the 13-bp core ERE sequence is boxed. The position of supershifted ER⅐ERE complex and free DNA is indicated to the left. The position of an ERE complex that does not contain ER is also indicated. 32 P-End-labeled ERE was incubated with 0, 10, 20, 30, 40, or 50 l of Yellow 86-purified ER, 0.45 g of poly(dI-dC), and H222 antibody. The mixture was incubated at 0°C for 1 h and resolved as in Fig. 2.   FIG. 6. A single flanking A residue does not compensate for two changes in the consensus ERE. Lanes 1-6 show ER binding to 13d2 ϩ A, an ERE with two changes from the consensus and a 5Ј A residue at Ϫ7. For comparison, ER binding to 15d1 is shown in lanes 7-12. The sequence of each ERE is shown, and the 13-bp core ERE sequence is boxed. The position of supershifted ER⅐ERE complex and free DNA is indicated to the left. The position of an ERE complex that does not contain ER is also indicated. ER/ 32 P-ERE mixtures were incubated with poly(dI-dC) and antibody and resolved as in Fig. 2. 15d2 were not necessary for an accurate determination of K d . The K d for 15d2 was calculated to be 20.0 nM. Given the previous results seen for the ER-ERE binding and competition experiments, we expected the binding curve for p13 to position between that of p15 and 15d2. The plot of the binding data showed that this was in fact the case, and the K d for p13 was determined to be 6.9 nM. DISCUSSION We have identified sequence requirements for ER to recognize EREs. It had been reported that the 13-bp inverted repeat 5Ј-GGTCAnnnTGACC-3Ј ERE is the minimum length sequence having two essential qualities, the ability to bind in a stable complex with ER and to mediate the induction of estrogen-responsive genes (3). This 13-bp palindromic structure was termed the core consensus ERE. However, most estrogen-responsive genes contain variations of the core ERE. Our experiments were designed to reveal how natural genes evolved to contain variant, yet functional, EREs, and to explore the contributions of flanking sequences to ER-ERE affinity. Our hy-pothesis is that sequences surrounding the variant core consensus EREs in natural genes allow them to retain ERE function despite the core sequence variations. We found that the core ERE placed in an inert sequence background must be perfect to retain function. However, within the appropriate background, the entire sequence would be operational despite changes in the core.
It is of interest to know whether a particular sequence is an ERE. We show here that sequences resembling EREs may not bind ER or function in hormone response. Sequences that vary from the core consensus by as little as a single nucleotide can lose the ability to bind ER. For example, even in huge excess, 13d1 did not bind ER and failed to compete for ER binding with the consensus ERE. Many ERE-like sequences may be the target of other transcription factors that bind "GGTCA" motifs. Because many genes contain ERE candidates and the exact flanking sequences that specify functional nonconsensus EREs were unknown, a number of studies were performed to measure the binding and transcriptional activity of various EREs (2, 12,   FIG. 7. A, consensus and nonconsensus ERE binding competition. 6 g of 32 P-labeled p17 ERE was mixed with increasing amounts of the indicated unlabeled competitor ERE, and each was added to 15 l of 19.2 nM ER containing 48 ng of poly(dI-dC). The mixtures were incubated on ice for 1.5 h. 10 l was subsequently loaded onto a 5% native polyacrylamide gel and separated by electrophoresis at 150 V for 1.5 h. The gels were dried at 80°C under vacuum, and exposed to a PhosphorImager screen for 12 h. Radioactivity in shifted DNA was quantitated and graphed as a percentage of initial binding, where initial binding was the intensity of the shifted band in the absence of competitor. B, saturation plot of ER-ERE binding. Increasing concentrations of the indicated 32 P-labeled ERE were added to a constant amount of recombinant human ER. The formation of the ER⅐ERE complex was followed by gel shift, and the amount of binding was quantitated using the PhosphorImager. The amount of bound ERE was graphed versus the log of the concentration of ERE. The experiment was done in triplicate, with the error bars indicating Ϯ S.E., as shown. The concentration of active ER was equal to the amount of ERE bound at saturation. K d values were determined as they were under "Experimental Procedures," by subtracting 1 ⁄2[ER] from the ERE concentration at half-maximal binding. 28 -33). It was found that a gene in one species may contain a functional ERE, whereas the ERE in the analogous gene of a second species is nonfunctional. A goal of our work is to predict functional ER-ERE binding on the basis of ERE sequence.
An essential feature of an ERE is the ability to bind ER. We measured the ability of ER to retard an ERE-containing DNA segment in gel electrophoresis and also the capacity of various EREs to compete for binding to ER. Affinities were calculated for consensus and nonconsensus EREs. To determine whether the core consensus ERE represents the minimal element for ER binding, we designed an oligomer called the "background sequence," which lacked any series of nucleotides present in the core consensus ERE and was unable to stabilize ER-ERE interaction (Fig. 1A). Test EREs were embedded in this background. Results showed that variation of a single nucleotide in the palindromic 13-bp core ERE prevented ER-ERE interaction. Furthermore, the single change abolished its ability to compete, even at a 300-fold excess, with a functional ERE for binding ER. We concluded that ER binding to nonconsensus EREs depended on nucleotides flanking the core.
Flanking nucleotides would most likely contribute to ERE function by direct interaction with the ER. Candidate nucleotides were chosen from the ERE of the vitellogenin A2 gene, a highly inducible estrogen-regulated gene (2)(3)(4). Its ERE is an extended 17-bp palindrome, containing an A residue at the Ϫ7-position of both strands and a C residue at the Ϫ8-position of both strands. We asked whether the vitellogenin A2 ERE contains the optimal flanking sequence or whether outer nucleotides of this ERE represent a subset of sequences that can compensate for changes in the core sequence. For example, the presence of an A residue at Ϫ7, as in vitellogenin A2, was able to compensate for a core variation, rescuing ER binding. Is the 5Ј-flanking A of vitellogenin A2 ERE necessary for ER binding to EREs that have a single change from the consensus, or can other bases substitute? We knew that 13d1, like some other 13-bp constructs that did not bind ER, had a 5Ј-flanking T residue on the upper strand and a 5Ј-flanking C residue on the lower strand. This suggested that pyrimidines could not substitute for the A residue at position Ϫ7. However, when ER 13d1 ϩ G was tested, ER binding was restored at about the level seen for 13d1 ϩ A. Apparently, purines are preferred for the 5Ј-position immediately flanking the core. Therefore, while the vitellogenin A2 sequence served as a useful guide for testing nucleotides, other base substitutions also restored ERE binding ability.
We considered the possibility that the symmetry of the inverted repeat sequence contributed to ER affinity. If so, flanking nucleotides that extend the core inverted repeat would promote binding, and the identity of the nucleotides would then be of lesser importance. Construct 15d1A-T exchanged the positions of the flanking A and T residues found in 15d1, while maintaining the inverted repeat. This sequence was identical to the 13d1 construct, except that there was an A residue placed 3Ј of the core inverted repeat. No ER binding was observed. This demonstrated that the identity of individual nucleotides is important and that a simple expansion of the inverted repeat does not increase ER binding.
It is important to know the effective distance of flanking base sequences from the core sequence. It was clear that the Ϫ7 sequence immediately flanking the 13-bp core had a large effect on ER-ERE affinity. To test the influence of nucleotides further out, the nucleotides at Ϫ8 were tested in the context of an ERE-like sequence that did not bind ER, allowing minor contributions to be detected. Although a correct nucleotide at Ϫ7 on either strand could compensate for a single change in the core, correct nucleotides at Ϫ8 on both strands were required to fully recover ER binding to an ERE having the same core defect. A single C at Ϫ8 was not sufficient to recover ER binding to 15d1AT. Detectable ER-ERE binding required a C at Ϫ8 on both strands, as in 17d1AT. We suspect that both residues at the Ϫ8-position exert an equal small positive effect on ER-ERE binding affinity and that both residues are needed to overcome the significant negative effect of the changed core consensus on binding to EREs at physiological concentrations of ER. In ranking the relative contributions of flanking sequences to ER-ERE affinity, we conclude that bases at the Ϫ7-position exert more influence than bases at the Ϫ8-position.
Stabilization of ER binding by flanking sequences at position Ϫ8 of the inverted repeat is supported by additional evidence. The DNase I footprint of ER spans 22 nucleotides (15), and the crystal structure of the ER-DNA binding domain complexed with DNA shows contacts with 17 base pairs of core and flanking sequence (27). The crystal structure shows that four side chains on the surface of the ER-DNA recognition helix form hydrogen bonds with the GGTC bases in the ERE sequence. Further contacts are made with the backbone of the DNA, suggesting that these are not sequence-specific. However, comparisons of crystal structures of ER and glucocorticoid receptor DNA binding domain complexed with DNA revealed that even subtle changes in the DNA-protein interface can substantially alter key contact points. Single DNA base changes are likely to result in the rearrangement of side chains and multiple stabilizing bonds (27). As a result, specific contacts made by ER could vary extensively depending on the DNA sequence presented. The ER would not be unique in its ability to rearrange stabilizing contacts for recognition of nonconsensus sequences. Other nuclear hormone receptors can bind remarkably disparate DNA sequences; e.g. not only will COUP-TF bind to inverted or direct repeats of 5Ј-GGTCA-3Ј with spacing up to 12 nucleotides (34), but it can also tolerate changes from the consensus (35,36).
A variety of ERE-containing oligomers were tested for their ability to compete with the p17 consensus for ER binding. These experiments provided an independent and corroborative assessment of relative affinities. Based on both binding and competition assays, relative binding affinities were ranked in the following order: p15 ϳ p17 Ͼ 15d1 ϳ 17d1 Ͼ p13 Ͼ 17d1AT ϳ 15d2 ϳ 17d2 Ͼ 15d3 ϳ 13d1.
The dissociation constant (K d ) for p17 was reported as 0.25 nM. Affinity measurements for p15 yielded a K d of 0.54 nM, comparable with p17 within experimental variance. Therefore, K d determinations using gel shift analyses could accurately estimate affinities of different EREs for ER. The affinities of 15d1 and 17d1 for ER were higher than that of p13, which was calculated to be 6.9 nM. The K d for 17d2 was 20 nM, and the affinities of 15d2 and 17d1AT should be similar, based on binding and competition experiments.
Could we define the limit of base substitutions in the core sequence that could be rescued by flanking sequences? The answer is determined by the position of the substitutions. Our results indicate that two, but not three, substitutions in the same arm of the inverted repeat will function in the appropriate flanking sequence context. Can all double substitutions be rescued? Reports by others provide evidence that ER cannot bind to a subset of EREs with two changes from the consensus, although they have flanking sequences expected to rescue ER-ERE binding. The murine c-fos oncogene contains an ERE with two changes from the consensus, one on each arm of the inverted repeat, and does not bind ER (37). This sequence, 5Ј-AAGGTCTaggAGACCCC-3Ј (38), is incapable of binding despite a stimulatory sequence at the Ϫ7-position. The Ϫ373 ERE found in the 5Ј-flanking region of c-jun is similar. It has two changes from the consensus sequence and a G residue at the Ϫ7-position, i.e. 5Ј-CGGGTCCgctGGACCTT-3Ј, and is also unable to bind ER (13). ER would not bind an ERE-like sequence in the 5Ј regulatory region of the rat prolactin gene, 5Ј-TTT-GTCActaTGTCCTA-3Ј (39,40). The rat calbindin D-9K gene contains a functional ERE with a single change from the consensus, which when mutated to 5Ј-AGATCActgTGATCT-3Ј, with a change in both arms of the inverted repeat, did not bind ER (41). Together, these reports indicate that, although two changes from the consensus may be tolerated in a single arm of the inverted repeat in the presence of appropriate flanking sequences, a change in each arm of the inverted repeat is not.
It appears that ER binding to EREs can be described by a general set of rules. For ER to bind, at least 10 of the 12 nucleotides located between two and seven nucleotides from the center of the inverted repeat have to be of the correct sequence. The furthest flanking positions eight nucleotides from the center also contribute to ER binding affinity, but both correct bases are required to reconstitute binding when the rest of the ERE has only nine of the required bases. Any ERE with two changes from the core consensus is nonfunctional when the changes are on different arms of the inverted repeat. EREs with three or more changes from the core consensus do not bind ER. Also, between Ϫ7 and ϩ7, progressively more changes from the optimal sequences result in lower affinity binding. Conversely, more bases in agreement with the optimal sequences will result in tighter binding.
From these rules, one can predict accurately whether a putative ERE sequence will bind to ER and also determine the relative affinity of ER to different EREs. For example, the vitellogenin B1 gene contains two EREs in the regulatory region. We designated them vitellogenin 1B1 (5Ј-CCAGTCActgT-GACCCA-3Ј) and vitellogenin 2B1 (5Ј-CAAGTTAtcaTGAC-CTC-3Ј) (Fig. 1C). Vitellogenin 1B1 has a single change from the consensus, and 2B1 has two changes from the consensus. The core sequence is underlined, and changes from the consensus are in boldface type. Applying our rules for ER-ERE binding, we would predict that vitellogenin 1B1 will bind ER because it contains a 5Ј G residue at Ϫ7 on the lower strand that compensates for the single base substitution in the core consensus sequence. Vitellogenin 1B1 also has a 5Ј C residue at position Ϫ8 in the upper strand, which contributes slightly to binding affinity. Vitellogenin 2B1 has a 5Ј flanking A residue at the Ϫ7-position on both strands, which compensates for the two changes in the core consensus, allowing ER-ERE binding. As with vitellogenin 1B1, the 5Ј C residue in the Ϫ8-position has a small positive effect on binding. Because vitellogenin 2B1 has two changes in the core consensus and vitellogenin 1B1 has one, vitellogenin 1B1 should bind more ER than vitellogenin 2B1. Gel shifts with these EREs (not shown) confirmed these predictions.
The ERE sequence in the hsp70-related gene has a single nucleotide change from the consensus, 5Ј-CTGGTCActcCGAC-CAG-3Ј (42). Krawczyk et al. (42) predicted that this ERE would confer estrogen responsiveness to a reporter gene. This prediction was based on its similarity to the EREs found in the rabbit uteroglobin gene (5Ј-CAGGTCAccaTGCCCTC-3Ј) (43), the human pS2 ERE (5Ј-CAGGTCActgTGGCCCT-3Ј) (12), and other sequences that had a single core substitution and mediated an estrogen response. Based on flanking sequence requirements for ER-ERE binding, we would predict that the rabbit uteroglobin ERE and the human pS2 ERE would bind ER with approximately the same affinity as the 13-bp core ERE, while the hsp70-related gene ERE would bind barely detectable amounts of ER. In fact, the authors reported that the hsp70related gene ERE did not confer hormone responsiveness to reporter genes. They concluded that the T nucleotide in the right half inverted repeat was the essential element for estrogen responsiveness. While that explanation is consistent with their results, we would predict that the combination of a single core substitution with inappropriate flanking sequences makes this apparent ERE inactive. One value of our findings is that the same modified core sequence might be active in another gene, having the correct flanking sequence. Expectation of an essential role of the T nucleotide would then predict the wrong result.
Predictions based on principles determined here extend to other examples in the literature. ERE sequences of some genes differ between species, and it is important to know whether ER binds to these EREs. The ERE sequences of the mouse and human lactoferrin gene are different. The mouse lactoferrin gene contains an ERE with one change from the consensus, 5Ј-CAGGTCAAGGTAACCCA-3Ј (32). We predict that the 5Ј A residue compensates for that change, allowing ER-ERE binding. Teng (32) reported that ER bound to this ERE and that the ERE was functional in transfection experiments. The human lactoferrin gene ERE sequence is 5Ј-CAGGTCAAGGC-GATCTT-3Ј. Because this ERE contains a 5Ј A residue on both strands, we expect this ERE to bind ER although there are two changes from the consensus. Teng (32) reported that this ERE also binds ER and functions in transfection experiments. Although no direct comparison was made between the two sequences, we expect that the mouse ERE binds more ER than the human ERE, based on the number of changes from the consensus.
There are numerous reported ERE sequences to which the binding of ER is predictable by our rules. These include rat and bovine oxytocin genes (30); cathepsin D ERE (33); calbindin D-9K gene ERE (41); vitellogenin Bl, vitellogenin A2, and chicken apo-very low density lipoprotein gene EREs (2); rabbit uteroglobin ERE (43); hsp70-related gene ERE (42); and Hageman factor XII gene ERE (44). The 5Ј-flanking region of the tst-1/oct6 gene has a single change from the consensus combined with an A residue at the Ϫ7-position; it bound ER and generated an estrogen response in reporter genes. The authors also synthesized two EREs which had three changes from the consensus, neither of which bound ER (45). Other EREs to which the rules successfully apply are the Ϫ334 Xenopus vitellogenin ERE (7), Xenopus ER gene ERE (46), and the salmon gonadotropin II ␤-subunit gene (47).
Although we report here that purified, homodimeric ER does not bind EREs with more than two changes from the consensus and others have seen the same result (28,30,45,48), genes are known where estrogen response is facilitated through ERE elements that contain three or more changes from the consensus. These are c-Fos and c-Jun, transcription factors that bind to the AP-1 site and rat creatine kinase B. It has been suggested that ER binding to EREs that vary by three or more nucleotides is dependent on heterodimerization or other novel mechanisms. For example, mouse proto-oncogene c-jun contains 5Ј-AAGCAGAgcaTGACCTT-3Ј (13) that binds ER. Mutation of this ERE so that the left arm of the inverted repeat contains only a single base in common with the consensus (5Ј-AATTAGAgcaTGACCTT-3Ј) did not diminish ER binding. Given these unexpected results, the authors concluded that ER does not bind to these sequences as a homodimer, as was seen with the consensus ERE. They suggested that ER interacts with an unidentified protein factor to regulate estrogen induction of c-jun expression in a cell-specific pattern. The sequence of the rat creatine kinase B ERE is 5Ј-AGGTCAgaaCACCCT-3Ј, which also binds ER (49). Estrogen regulation of the rat creatine kinase B gene is complex, depending on Sp1 binding and a second regulatory region that has no ERE. The mouse c-fos proto-oncogene was reported to have an ERE, capable of ER binding, similar to the rat creatine kinase B ERE. The sequence of the c-fos ERE is 5Ј-CAGGTCAccaCAGCCCA-3Ј, containing three changes from the consensus (50,51). The authors report that gel shift experiments show a difference in mobility for ER binding to the c-fos ERE compared with the consensus ERE (51). This is a significant finding, because gel shifts for homodimeric ER⅐ERE complexes show consistent migration behavior. The authors note that the complex they observe may contain a heterodimer, where ER associates with an unknown protein, similar to their findings for c-jun. In fact, ER binding to degenerate ERE/Sp1 sequences was reported to involve heterodimerization (52). Heterodimerization of ER may be a prerequisite for ER binding to nonconsensus EREs with three or more changes. Adding another level of complexity to ER regulation, ER has been shown to interact with Fos-Jun complexes directly at the protein-protein level, activating transcription through DNA sequences that contain GGTCA-like elements, without binding to the DNA itself (53) We would expect the DNA sequence requirements for heterodimeric ER binding or for ER binding as part of a multiprotein complex to be different from those for homodimeric ER. This hypothesis places ER action into two models. In the first, ER binds as a homodimer directly to EREs that contain two or fewer changes from the consensus. Here, ER acts alone to up-regulate a gene. In the second model, ER binds as a heterodimer or as part of a multiprotein complex. These two models can be distinguished by the ERE sequence. With ERE sequences containing more than two changes from the consensus, ER regulation would depend on the presence of other transcription factors, allowing for a more complex response. Furthermore, the more mutations one arm of an ERE contains, the more it resembles an ERE half-sequence. Half-ERE sequences have been observed to bind ER and regulate the expression of certain genes (54 -58). Although we could not detect ER binding to a half-ERE (1/2 ERE p17, Fig. 1) (data not shown), the conditions under which ER binds to half-EREs seem to require high ER concentrations and multiple half-sites in tandem (59). In other reports, ER regulates gene expression through half ERE sequences without binding directly to these sequences (58). Experiments exploring the nature of ER-ERE binding to ERE-like inverted repeats that have three or more changes from the consensus are currently being conducted.
In summary, our study emphasizes that the sequence requirements for a "minimal" consensus ERE are more complex than previously thought. The minimal sequence required for ER binding encompasses a wide variety of possible combinations, which display stable ER binding. The optimal binding sequence is 5Ј-C(A/G)GGTCAnnnTGACC(T/C)G-3Ј. Any change from this optimal consensus diminishes ER-ERE affinity. The underlined core consensus sequence represents the minimum length for binding activity. However, at that length, no variation is allowed. Our experimental results are consistent with the hypothesis that ER can bind as a homodimer to any ERE with up to two changes in one arm of the core consensus if appropriate flanking sequences are present. One change in the core requires a corresponding single correct nucleotide in the immediate flanking sequence at the Ϫ7-position or two changes in the far flanking sequence at the Ϫ8-position. Two changes in the core require two correct nucleotides in the immediate Ϫ7 flanking sequence. If an ERE contains a change in each arm of the inverted repeat, ER cannot bind. Because these rules apply to EREs with one or two changes from the consensus but do not apply for EREs with three or more changes, we suggest that ER does not bind alone as a homodimer to EREs with more than two changes from the core consensus.