Fredric E. Wondisford, and Anthony N. Hollenberg. Section of Endocrinology (R.N.C., F.E.W.) ...... Revision received March 29,. 2001. Accepted April 2, 2001.
The Specificity of Interactions between Nuclear Hormone Receptors and Corepressors Is Mediated by Distinct Amino Acid Sequences within the Interacting Domains
Ronald N. Cohen, Sabrina Brzostek, Brian Kim, Michael Chorev, Fredric E. Wondisford, and Anthony N. Hollenberg Section of Endocrinology (R.N.C., F.E.W.) Department of Medicine University of Chicago Chicago, Illinois 60637 Thyroid Unit (S.B., B.K., A.N.H.) Division of Endocrinology and Division of Bone and Mineral Metabolism (M.C.) Beth Israel Deaconess Medical Center Boston, Massachusetts 02215
The thyroid hormone receptor (TR) and retinoic acid receptor (RAR) isoforms interact with the nuclear corepressors [NCoR (nuclear corepressor protein) and SMRT (silencing mediator for retinoid and thyroid hormone receptors)] in the absence of ligand to silence transcription. NCoR and SMRT contain C-terminal nuclear hormone receptor (NHR) interacting domains that each contain variations of the consensus sequence I/L-x-x-I/V-I (CoRNR box). We have previously demonstrated that TR1 preferentially interacts with NCoR, whereas RAR␣ prefers SMRT. Here, we demonstrate that this is due, in part, to the presence of a novel NCoR interacting domain, termed N3, upstream of the previously described domains. An analogous domain is not present in SMRT. This domain is specific for TR and interacts poorly with RAR. Our data suggest that the presence of two corepressor interacting domains are necessary for full interactions with nuclear receptors in cells. Interestingly, mutation of N3 alone specifically decreases binding of NCoR to TR in cells but does not decrease NCoR-RAR interactions. In addition, while the exact CoRNR box sequence of a SMRT interacting domain is critical for recruitment of SMRT by RAR, the CoRNR box sequences themselves do not explain the strong interaction of the N2 domain with TR1. Additional regions distal to 0888-8809/01/$3.00/0 Molecular Endocrinology 15(7): 1049–1061 Copyright © 2001 by The Endocrine Society Printed in U.S.A.
the CoRNR box sequence are needed for optimal binding. Thus, through sequence differences in known interacting domains and the presence of a newly identified interacting domain, NCoR is able to preferentially bind TR1. These preferences are likely to be important in corepressor action in vivo. (Molecular Endocrinology 15: 1049–1061, 2001)
INTRODUCTION The thyroid hormone receptor (TR) and retinoic acid receptor (RAR) are members of the nuclear hormone receptor (NHR) superfamily (1). In the absence of their respective ligands, TR and RAR repress gene transcription by binding to nuclear corepressor proteins (2–8). These proteins, such as NCoR (nuclear corepressor protein) and SMRT (silencing mediator for retinoid and thyroid hormone receptors) mediate ligandindependent repression by binding a complex with histone deacetylase activity (9–11). NCoR and SMRT share a similar structure (Fig. 1), containing C-terminal nuclear receptor-interacting domains (IDs) and at least three N-terminal repressing domains. The IDs share limited homology with each other, suggesting that there may be specificity in terms of NHR recruitment of corepressors. In fact, data from a number of groups suggest that such specificity does exist. Previously we have demonstrated that TR1 prefers to interact with NCoR, and RAR␣ prefers to interact with SMRT, on DNA response elements (12). These preferences are mediated 1049
MOL ENDO · 2001 1050
Fig. 1. Nuclear Corepressor Interacting Domains Amino acid sequences of mNCoR and hSMRT are indicated. The NCoR IDs are outlined and numbered, and corresponding regions of SMRT are identified. The CoRNR box sequences of each interacting domain are indicated and numbered.
through the proximal IDs of NCoR and SMRT, N2 and S2, respectively (see Fig. 1). Additionally, Zamir et al. (13) have shown that RevErb, an orphan nuclear receptor, interacts with NCoR, but does not interact with SMRT. Wong and Privalsky (14) have demonstrated that the various RAR isoforms recruit SMRT with different affinities, such that RAR is deficient in interactions. In fact, knock-out of NCoR is embryonic lethal, possibly due to a key role of NCoR in TR-mediated erythrocyte development; therefore, SMRT cannot compensate for a lack of NCoR in vivo (15). Recent work has established that I/L-x-x-I/V-I motifs (or CoRNR boxes) and adjacent helical structure within the nuclear corepressor IDs are critical for mediating interactions with NHRs (16–18). While these domains exhibit similarity to LXDs (L-x-x-L-L motifs) found in coactivator IDs (19), their structure results in a conformation that favors release on ligand binding. In fact, changing the LXD of a coactivator to a CoRNR box results in a alteration in ligand dependence, such that the coactivator is bound in the absence of ligand (16). Data from a number of groups suggest that a single corepressor molecule binds a NHR dimer complex (homodimer or heterodimer) (12, 17, 18). These data suggest that each interacting domain may contact a single NHR. However, the specific roles of the different IDs in mediating this process have not been well defined. To identify the functions of the distinct interacting domains, we first searched for additional I/L-x-xI/V-I motifs in NCoR and SMRT. In fact, we found that NCoR contains a novel ID, proximal to its other two known IDs. This region (termed N3) binds TR well. This domain corresponds to the recently described interacting domain described by Webb et al. (20). N3 is not present in SMRT, suggesting a potential mechanism for the preferential interaction of TR1 complexes with NCoR over SMRT. Whether the distinct sequences within the CoRNR box domains mediate the specificity of interactions
Vol. 15 No. 7
with NHRs is unknown. In fact, each of the NCoR or SMRT IDs contains a unique CoRNR box (Fig. 1), suggesting the possibility that differences in corepressor binding by NHRs may depend on the distinct CoRNR box sequences within the IDs. For example, the distal CoRNR box in NCoR and SMRT appears to be important for retinoid X receptor (RXR) binding, whereas RXR binds the proximal CoRNR box sequences poorly (16). This is likely to be important in the binding of NCoR and SMRT to NHR heterodimers. It has previously been shown that regions of coactivators adjacent to LXDs dictate specificity in terms of binding to NHRs (21–23). We therefore studied the regions within the corepressor IDs to determine what portions contribute to the specific preferences of TR1 for NCoR, and of RAR␣ for SMRT. We focused on N2 and S2, as these domains appear to be important in mediating preferences, and share some degree of homology (12). Our data suggest that specificity of corepressor recruitment by TR and RAR may depend on distinct mechanisms.
RESULTS NCoR Contains a Novel Interacting Domain, Not Present in SMRT Analysis of the NCoR amino acid sequence suggests the presence of an additional CoRNR box motif I-DV-I-I at amino acids 1949–1953 of murine (m)NCoR. Interestingly, although this sequence contains an I/Lx-x-I/V-I sequence (16, 18), it does not have a full extended helical motif as described by Perissi et al. (17), L-x-x-I/H-I-x-x-x-I/L. Therefore, to identify whether a region containing this sequence could function as an independent interacting domain in vitro, we analyzed interactions of NCoR amino acids 1939–2021 (termed N3) with TR1 in electrophoretic mobility shift assay (EMSA) (Fig. 2A). Glutathione-Stransferase (GST)-N3 was designed to be similar in size to GST-N2. As shown in lanes 2–6, GST-N3 binds strongly to TR1 on a DR⫹4 element. In isolation, GST-N3 and GST-N2 bind TR1 with high affinities (compare lanes 2–6 with 8–12), although N2 binding is stronger (compare particularly lanes 2 and 3 with 8 and 9). Both N3 and N2 are dissociated from TR1 in the presence of its ligand, T3 (data not shown). To determine whether this interaction is specifically dependent on the I-D-V-I-I CoRNR box sequence, N3mut was made. This protein contains NCoR amino acids 1,939– 2,021, but with a mutated CoRNR box sequence, A-DV-I-I. In fact, N3mut does not bind TR1 on a DR⫹4 element (Fig. 2B, lanes 4 and 5). Therefore, the CoRNR box sequence of N3 is vital for its interactions with the TR. To determine whether N3 is important for interactions of NCoR with TR1 in cells, we employed a mammalian two-hybrid assay (Fig. 2C). Portions of NCoR with and without N3 were placed downstream
Specificity of Corepressor Interactions
1051
Fig. 2. NCoR Contains a Third Interacting Domain A, Gel mobility shift assay with 4 l in vitro translated (IVT) TR1; 0.02 g (lanes 2, 8), 0.1 g (lanes 3, 9), 0.5 g (lanes 4, 10), 1 g (lanes 5, 11), or 1.5 g (lanes 6, 12) of indicated GST-CoR construct; and DR⫹4 radiolabeled probe. B, Gel mobility shift assay with 4 l IVT TR1; 0.02 g (lanes 2, 4) or 0.1 g (lanes 3, 5) of indicated GST-CoR construct; and DR⫹4 radiolabeled probe. C, CV-1 cells were cotransfected with 1.7 g UAS reporter; 80 ng of the indicated Gal 4 construct; and 80 ng of VP16- TR1 (or empty vector-VP16). Cells were also cotransfected with 20 ng of a CMV--galactosidase vector to control for transfection efficiency. The experiment was performed in triplicate, and repeated three times. Data are expressed as fold luciferase activity, in the presence vs. absence of cotransfected VP16-TR1 (mean ⫾ SE). D, Western blot of Gal4 constructs. CV-1 cells were transfected with 10 g of indicated Gal4 constructs. Protein nuclear extracts were analyzed by SDS-PAGE, transferred to nitrocellulose, and blotted with an anti-Gal4 antibody. Panel A, NCoR (1,939–2,142); panel B, NCoR (1,959–2,142); panel C, NCoR (2063–2142).
MOL ENDO · 2001 1052
of the Gal4 DNA-binding domain and tested for the ability to interact with VP16-TR1. In particular, Gal4NCoR 1,579–2,454, which contains N3, N2, and N1, interacts more than 20-fold with TR1. Gal4-NCoR 1,941–2,142, which contains N3 and N2 (but not N1), also interacts strongly with TR1. In contrast, deletion of just 20 further amino acids (Gal4-NCoR 1,961–2,142), which contains N2 but not N3, interacts minimally with TR1, as does a smaller N2-containing region (Gal4-NCoR 2,063–2,142). Interestingly, this region (containing only N2) binds TR1 strongly in EMSA (12). These data suggest that a single NCoR ID does not optimally bind TR1 in cells. Furthermore, N3, in the context of N2, allows for strong interaction with TR1. A Western blot of the Gal4 constructs (Fig. 2D), probed with an antibody to the Gal4 DNA-binding domain, shows that the Gal4 chimeric proteins are produced in CV-1 cells and that the decreased interactions with NCoR 1,941–2,142 and NCoR 2,063–2,142 are not due to decreased protein production. The Presence of Two CoRNR Box Motifs Are Necessary for Full Structural and Functional Interactions of NCoR with the TR To more fully define the role of individual IDs in the context of a larger portion of NCoR, a variety of CoRNR box mutations were introduced into NCoR amino acids 1,579–2,454. This fragment of NCoR contains the IDs of NCoR, but deletes the major repressing domains. It reverses basal repression when transfected into cells, by competing with endogenous NCoR for TR binding. It has therefore been termed NCoR inhibitor, or NCoRI (24). Mutations were made in each CoRNR box of NCoRI, either alone or in combination. These mutations each substitute an alanine for the initial amino acid of the I/L-x-x-I/V-I motif, such that N3m contains NCoR amino acids 1,579–2,454, but with an A-D-V-I-I in place of I-D-V-I-I. Similar mutations were made to create N2m and N1m. Combinations of these mutations were made to create N3–2 m, N3–1 m, and N2–1 m. Finally, the N3–2-1 m construct has similar mutations in all three CoRNR box motifs. To determine whether these mutations decreased interactions with TR1, GST interaction assays were performed. In these studies, 35S-labeled NCoRI constructs were tested for their ability to interact with GST-TR1. As shown in Fig. 3A, wild-type NCoR AA 1,579–2,454 interacted strongly with GST-TR1. This interaction was decreased (although not entirely eliminated) in the presence of T3 (lane 1). Both N3m (lanes 2) and N2m (lanes 3) have significantly decreased interactions with TR1. In contrast, N1m maintains strong interactions with TR1, consistent with the decreased ability of N1 in isolation to interact with TR1 in EMSA (12). Mutation of any two CoRNR box sequences (but in particularly N3–2 m and N3–1 m) results in a dramatic decrease in TR1 binding, suggest-
Vol. 15 No. 7
ing that the one remaining ID is not sufficient for strong interactions with the TR. Although an isolated, single interacting domain is capable of binding the TR in EMSA (see Fig. 2A), the experiments in Fig. 3A were done in the context of a larger portion of NCoR, as well as a smaller amount of protein. Thus, the presence of two IDs is necessary for full TR-NCoR interactions. Not surprisingly, mutation of all three CoRNR box motifs results in compete loss of binding. To examine these interactions in the context of an underlying thyroid hormone response element, EMSAs were employed, using a 32P-radiolabeled DR⫹4 probe, and in vitro translated TR1 and mutant or wild-type NCoRI constructs (Fig. 3B, upper panel). As shown in lane 2, wild-type NCoRI (NCoR amino acids 1,579–2,454) binds strongly to the TR1 homodimer. In contrast, N2m and N1m bind weakly (lanes 2–3), and N3m not at all (lane 1), suggesting that N3 plays a critical role in binding the TR. In addition, mutation of any two CoRNR box motifs results in loss of all binding in this assay (lanes 4–6), again suggesting that two intact IDs are necessary for strong binding to TR1 in the context of NCoR. All NCoRI constructs were in vitro translated similarly (Fig. 3B, lower panel), suggesting that differences seen in the EMSA reflect actual differences in binding. We used the ability of NCoRI to reverse basal repression to provide an assay of the function of the distinct NCoR IDs (24). The ability of the mutated NCoRI variants to reverse repression on a DR⫹4 element was tested in transient transfections in CV-1 cells. As shown in Fig. 3C, transfection of TR1 results in approximately 4-fold ligand-independent repression (from ⬃40,000 light units to 10,000). Cotransfection of pKCR2-NCoRI decreases this repression by more than half, i.e. luciferase activity was decreased, but was still greater than half of basal activity. In contrast, cotransfection of either pKCR2-N3mut or pKCR2N2mut results in an intermediate level of repression, whereas the level of repression with pKCR2-N1mut is no less than that of pKCR2-NCoRI itself. Mutation of two IDs results in further loss of NCoRI function, if N3 is one of the mutated domains, and mutation of all three IDs yields a construct with minimal, if any, function. These data again suggest that two IDs are necessary for full interactions of NCoR with TR1 in cells, although they do suggest the possibility of residual function in the presence of a single N3 ID, perhaps through an alternative mechanism. In sum, these data indicate that N3 is important for interactions with the TR in cells; the combination of N3 and N2 allows for full interactions. N3 Contributes to NCoR Specificity of Binding to Nuclear Hormone Receptors To determine whether N3 binds RAR as well as TR, an EMSA paradigm was again used. In this EMSA, performed on a DR⫹5 element, RAR␣ is present in all lanes; RXR␣ additionally is present in even-numbered
Specificity of Corepressor Interactions
1053
Fig. 3. Two Intact CoRNR Box Sequences Are Necessary for Interactions with TR1 A, GST interaction assay of NCoRI mutants. Four microliters of in vitro translated 35S-methionine-labeled NCoRI constructs were incubated with GST-TR1, washed extensively, and analyzed by SDS-PAGE. B, Gel mobility shift was carried out using 4 l in vitro translated TR1; 4 l in vitro translated mutant or wild-type NCoRI construct; and DR⫹4 radiolabeled probe. Shown below is an SDS-PAGE of the NCoRI constructs. Constructs were in vitro translated with 35S methionine; 2 l of the reaction mixture was analyzed by SDS-PAGE. C, CV-1 cells were cotransfected with 1.7 g DR⫹4-luciferase reporter, 80 ng of TR1, and 160 ng of pKCR2-NCoRI construct (or empty vector pKCR2). Data were performed in triplicate and repeated twice. Transfection efficiency was controlled using a -galactosidase expression vector. Data are expressed as relative luciferase activity (mean ⫾ SE).
MOL ENDO · 2001 1054
lanes. The presence of RAR-RXR heterodimers is noted in the even-numbered lanes. In fact, in contrast to its interactions with TR1, N3 does not bind RAR␣ well in EMSA (Fig. 4A, lanes 9–10). In contrast, S2 binds well to the RAR-RXR heterodimer (Fig. 4A, lane 14). Since a region homologous to N3 is not present in SMRT, these data suggest that N3 plays a role in determining specificity of interactions with TR and RAR. A mammalian two-hybrid assay was used to determine whether N3 dictated such specificity in cells. In this experiment, Gal4-NCoR increased interactions 140-fold with VP16-TR1, whereas interactions were only increased 40-fold for VP16-RAR␣ (Fig. 4B, upper panel), consistent with previous data (12). As shown in Fig. 4B (middle), Gal4-NCoRI interacts strongly with TR1-VP16. As expected, this interaction is dramatically decreased by mutation of the N3 CoRNR box (Gal4-N3mut). The baseline interaction of Gal4-NCoRI and RARVP16 is significantly less than with TR1-VP16 (Fig. 4B, upper panel; and Ref. 12). In contrast to TR, however, Gal4-N3mut interacts with RAR-VP16 as strongly as does wild-type NCoRI. However, Gal4-N2mut was markedly deficient in interactions with RAR-VP16. Therefore, N3 dictates the preference of TR for NCoR. NCoR binds RAR more weakly, and N2, not N3, is required. Specificity of Corepressor Recruitment by TR and RAR Depends on Distinct Mechanisms To examine the roles of the sequences within the individual IDs in modulating interactions, we focused on the N2 and S2 domains. These particular IDs were chosen because they contain some degree of sequence homology; however, their differences suggest that there may be sequence-specific mechanisms governing the specificity of corepressor recruitment. We have previously shown that TR1 binds N2 more strongly than S2, whereas RAR␣ prefers S2; these interactions help explain the preferences of TR1 for NCoR and RAR␣ for SMRT (12). N3 was not used in the following studies because there is no analogous domain in SMRT. We therefore next studied what portions of N2 are important for TR binding. As shown in Fig. 5, various deletion constructs of N2 were made as GST fusion proteins. These constructs were used in EMSA on a DR⫹4 element, to determine their interactions with TR1. When the C terminus of N2 is gradually deleted, interactions with TR1 are reduced dramatically (compare lanes 2–5). In particular, there is a significant decrease in binding when amino acids 2,106 to 2,119 are deleted (compare lanes 3 and 4). However, some binding is still clearly observed (lane 3). In contrast, deletion of even the proximal 11 amino acid residues of N2 abolished binding to TR1 (lane 7), suggesting that this region is necessary for binding. To determine whether the proximal portion of N2 was sufficient for binding as well, we made constructs
Vol. 15 No. 7
containing only the proximal portion of N2. In fact, constructs containing the proximal 27 amino acids of N2 bound to TR on the DR⫹4 element (data not shown). These data suggest that the proximal region of N2 is both necessary and sufficient for TR binding, consistent with the results of other groups in that it contains a CoRNR box motif (16–18). However, the marked decrease in binding of the C-terminal deletion constructs (e.g. Fig. 5, lanes 3 and 4) suggests that an additional (more distal) portion of N2 also contributes to TR binding. As discussed previously, the most proximal portion of N2 contains the I-C-Q-I-I CoRNR box sequence. It is known that mutations of this region abolish binding to TR (16–18). However, it is not clear if the differences in the distinct ID CoRNR box sequences contribute to the specificity of corepressor recruitment. In fact, an examination of the corepressor IDs shows that each contains a unique CoRNR box (Fig. 1A). For example, the N2 CoRNR box sequence is I-C-Q-I-I-, whereas the S2 CoRNR box sequence is I-S-E-V-I. To define the importance of the N2 and S2 CoRNR box sequences in mediating the specificity of interactions with NHRs, we made GST fusion proteins composed of constructs that swap these domains. Thus, N2* contains the S2 CoRNR box in the context of the full N2, and S2* contains the N2 CoRNR box in the context of the full S2 (Fig. 6A). As demonstrated in Fig. 6B, on a DR⫹5 element, RAR␣ binds S2 much more strongly than N2, both in the context of the supershift and the decrease in the remaining RAR/RXR heterodimer (compare lanes 6 and 14). Changing the S2 CoRNR box to that of N2 (S2*) markedly decreases these interactions (lanes 14 and 18). In contrast, changing the N2 CoRNR box to that of S2 (N2*) increases interactions with RAR (lanes 6 and 10). Therefore, the S2 CoRNR box plays a pivotal role in mediating the preference of RAR␣ for S2 (over N2). In contrast, TR1 binds N2 much more strongly than S2 on a DR⫹4 element (Fig. 6C, lanes 3 and 7). Mutating N2 to be N2* does not, however, decrease interactions with TR (lanes 3 and 5). Moreover, changing S2 to S2* decreases, not increases, binding to TR (lanes 7 and 9). Therefore, even though the N2 CoRNR box is vital in the binding of TR to NCoR, the specific N2 CoRNR box sequence (I-C-Q-I-I) does not mediate the specific preference of TR for N2 (over S2). In fact, the TR appears to bind the S2 CoRNR box sequence (I-S-E-V-I) as well as the N2 CoRNR box. Therefore, the region of N2 important for mediating this specificity must reside elsewhere in N2. Analysis of GST constructs by SDS-PAGE showed that all the constructs were produced appropriately (Fig. 6D); after SDSPAGE, a Bradford assay was performed, so that equivalent amounts of GST construct could be used in each experiment. As the motif just proximal to the N2 and S2 CoRNR boxes have also been shown to be important in binding these IDs to NHRs (17), we next wanted to determine whether these regions explained the phenome-
Specificity of Corepressor Interactions
1055
Fig. 4. N3 Contributes to NCoR Specificity A, Gel mobility shift assay with 4 l IVT RAR␣; 2 l IVT RXR in even numbered lanes (or unprogrammed reticulocyte lysate, in odd numbered lanes); 0.02 g (lanes 3–8) or 1.5 g (lanes 9–14) GST-CoR construct; and DR⫹5 radiolabeled probe. B, CV-1 cells were transiently transfected (upper panel) with 1.7 g UAS-luciferase reporter, 80 ng Gal4-NCoRI (or Gal4-empty vector); and 80 ng of VP16-TR1 or VP16-RAR␣. Transfection efficiency was controlled using a -galactosidase expression vector; data are expressed as relative luciferase activity. For middle and lower panels, CV-1 cells were transiently transfected with 1.7 g UAS-luciferase; reporter 80 ng of Gal4-NCoRI (or mutant); 80 ng of TR1-VP16 (upper panel) or RAR-VP16 (lower panel). Transfection efficiency was controlled using a -galactosidase expression vector. Data are expressed as relative fold activation. A fold activation of 1.0 is defined as the luciferase activity produced in the presence of VP16-TR1 and wild-type Gal4-NCoRI (middle panel); or VP16-RAR␣ and wild-type Gal4-NCoRI (lower panel).
MOL ENDO · 2001 1056
Vol. 15 No. 7
Fig. 5. Proximal and Distal Portions of N2 Play Distinct Roles in Binding TR Gel mobility shift assay was carried out using 4 l in vitro translated (IVT) TR1; 100 ng of the indicated GST-CoR construct; and DR⫹4 radiolabeled probe. A schematic illustration of the GST-CoR constructs is also shown; hatched marking indicates position of ICQII motif. The numbering shown is based on the mNCoR sequence. hNCoR numbering is as follows: mNCoR 2,063–2,142 correspond to hNCoR 2,045–2,128; mNCoR 2,063–2,105 correspond to hNCoR 2,045–2,091; mNCoR 2,063–2,119 correspond to hNCoR 2,045–2,105; mNCoR 2,063–2,132 correspond to hNCoR 2,045–2,118; mNCoR 2,074–2,142 correspond to hNCoR 2,056–2,128; mNCoR 2,084–2,142 correspond to hNCoR 2,066–2,128; mNCoR 2,093–2,142 correspond to hNCoR 2,179–2,128; mNCoR 2,107–2,142 correspond to hNCoR 2,193–2,128.
non of corepressor specificity. We therefore mutated N2 so that it would contain the proximal domain and CoRNR box of S2 (m1); and S2 so it would contain the proximal domain and CoRNR box of N2 (m2). As shown in Fig. 7A, however, these changes did not in themselves alter the specificity of TR for N2; in particular, TR binds m1 much stronger than m2 (see lanes 4–5). These data are in agreement with the hypothesis that regions of N2 C-terminal to the CoRNR box are also important in binding TR. To further define the region of N2 important for mediating specificity, additional chimeric proteins were made, containing regions common to N2 or S2. m3 contains the proximal 23 amino acids of N2, followed by the remainder of S2. m4 contains the proximal 23 amino acids of S2, followed by the distal portion of N2 (Fig. 6A). The first 23 amino acids of N2 and S2 contain not only the CoRNR box sequences, but also both regions immediately proximal and distal to this motif. As shown in Fig. 7B, N2 again binds TR1 more strongly than S2 on a DR⫹4 response element (compare lanes 2 and 3, or 6 and 7). When N2 is
changed to m3 (so that it contains distal S2), significant TR binding is lost (compare lanes 6 and 8). In contrast, when S2 is changed to m4 (so that it contains the N2 distal region), binding is regained (compare lane 7 with lane 9). Thus, a region in the distal domain of N2 is important in mediating the preference of TR1 for N2. However, both m3 and m4 bind TR more weakly than does N2, suggesting that both portions of N2 are important in specifying strong TR binding (compare lanes 4 and 5 with lane 2; or lanes 8 and 9 with lane 6).
DISCUSSION The TRs and RARs repress gene transcription in the absence of ligand. This action is mediated by the nuclear corepressor proteins, NCoR and SMRT. NCoR and SMRT bind TR and RAR via their C-terminal interacting domains. Ligand binding causes a conformation change in the receptor, such that helix 12
Specificity of Corepressor Interactions
Fig. 6. The S2 CoRNR Box Mediates Specificity with RAR A, Amino acid sequences of the proximal 27 amino acids of N2 and S2 are indicated, along with sequences of the corresponding portions of mutant constructs. B, Gel mobility shift assay carried out with 4 l IVT RAR␣; 2 l IVT RXR in even numbered lanes (or unprogrammed reticulocyte lysate in odd numbered lanes); 0.02 g (lanes 3, 4, 7, 8, 11, 12, 15, 16) or 1 g (lanes 5, 6, 9, 10, 13, 14, 17, 18) GST-CoR construct; and DR⫹5 radiolabeled probe. C, Gel mobility shift carried out with 4 l IVT TR1; 2 l IVT RXR in even numbered lanes (or unprogrammed reticulocyte lysate in odd numbered lanes); 0.02 g indicated GST-CoR construct; and DR⫹4 radiolabeled probe. D, SDS-PAGE of GST-CoR constructs. After analysis by SDS-PAGE, protein quantification was performed by Bradford assay. Panel A, N2*; panel B, S2; panel C, S2*; panel D, N2. Equivalent amounts of protein were used in each EMSA, as indicated.
1057
Fig. 7. A Distal Domain in N2 Is Required for Specificity A, Gel mobility shift assay carried out with 4 l IVT TR1, 20 ng indicated GST-CoR construct, and DR⫹4 radiolabeled probe. B, Gel mobility shift assay carried out with 4 l IVT TR1, 20 ng (lanes 2, 3, 4, 5) or 1 g (lanes 6, 7, 8, 9) indicated GST-CoR construct, and DR⫹4 radiolabeled probe.
rotates. This repositioning of the NHR releases bound corepressor (8) and forms a hydrophobic coactivatorbinding surface (25). The L-x-x-L-L motifs present in coactivators mediate binding to this pocket formed in the NHR ligand-binding domain (LBD). Amino acids adjacent to this motif are important in mediating the specificity of interactions with distinct NHRs (21–23). In contrast, the amino acid sequences within the corepressor IDs that dictate specificity in terms of interactions with NHRs are less well defined. Corepressor IDs are characterized by CoRNR box motifs: regions containing the sequence I/L-x-x-I/V-I (16–18). It appears that the portion of the NHR that
MOL ENDO · 2001 1058
contacts the corepressors is similar to the portion that makes contact with coactivators, but that the position of helix 12 is important to distinguish which class of cofactor is able to bind. Thus TR and RAR bind corepressors in the absence of ligand, but upon ligand binding, corepressors are released and coactivators are recruited. In contrast, the estrogen receptor binds corepressors only in the presence of antagonists such as 4-OH tamoxifen (26–28). This antagonist binds the ligand-binding domain in such a way as to rotate helix 12 to stabilize corepressor binding. Finally, RevErb binds NCoR and has no known ligand. RevErb lacks a helix 12, thereby stabilizing NCoR binding; moreover, side chains fill the ligand-binding cavity, and it may therefore bind NCoR constitutively (29). Specificity in cofactor recruitment is likely to play an important role in NHR action. For example, distinct roles for coactivators are becoming better understood through recent gene knockout experiments. Knockout of steroid receptor coactivator 1 (SRC-1) causes a variety of defects, including impaired decidual stimulation (a progesterone-mediated response), impaired prostate growth in response to androgen, and decreased mammary gland ductal branching (30). In addition, these animals appear to have a degree of thyroid hormone resistance (31). Although the lack of SRC-1 (NCoA-1) may be partially replaced by an upregulation in glucocorticoid receptor-interacting protein 1 (GRIP-1) (TIF2/SRC-2), it is clear that not all SRC-1 action is replaced (30). Interestingly, knock out of p/CIP (SRC-3/TRAM-1/AIB1/ACTR) results in a more severe phenotype, including growth retardation and delayed puberty (32). Thus, although coactivators appear to have redundant roles in vitro, they play more specific roles in vivo. In terms of the corepressors, deletion of NCoR is embryonic lethal, suggesting that NCoR and SMRT play distinct roles in development (15). In turn, specificity in terms of corepressor recruitment by the distinct NHRs may play an important role in mediating these distinct functions. Whereas the region adjacent to the L-x-x-L-L motifs in coactivators appear to play an important role in mediating specificity in terms of interactions with NHRs, the domains important for specificity in terms of corepressor recruitment are less well defined. Hu and Lazar (16) have shown that RXR preferentially binds to N1 over N2 (16). We have shown that TR1 prefers to bind NCoR, and RAR␣ prefers to bind SMRT, and that this preference is mediated by the proximal IDs (12). In this report, we have examined the regions of N2 and S2 that mediate preferential interactions. We have found that the I-S-E-V-I CoRNR box motif in S2 is vital in dictating the preference of RAR␣ for S2. In contrast, the I-C-Q-I-I CoRNR box motif in N2, while important for binding, does not in itself dictate the preference of TR1 for N2. Regions downstream of the CoRNR box play a pivotal role in this respect. Moreover, deletion of this C-terminal region of N2 impairs binding to TR1. Therefore, the TR requires additional sequences in the
Vol. 15 No. 7
corepressor IDs, distal to the CoRNR box, for optimal binding. Our data also show that NCoR contains a third interacting domain, N-terminal to the other two domains. This domain, which we term N3, binds TR1 strongly; it has also recently been described by another group (20). Our data confirm the importance of N3 as an NCoR interaction domain and also extends previous findings to establish a role for N3 in mediating the specificity of corepressor interactions. Our data suggest that N3 contributes to the preferential recruitment of NCoR by TR. In isolation, N3 binds TR almost as well as N2, and more strongly than N1. In addition, mutation of the N3 CoRNR box impairs binding of NCoR to TR to a greater degree than does mutation of N2 or N1. Therefore, N3 is likely to be an important domain for interactions of NCoR with TR. Interestingly, a region analogous to N3 is not present in SMRT. Therefore, the presence of N3 provides an important mechanism to explain the TR1 preference for NCoR over SMRT. Not only does TR bind NCoR over SMRT, but the reverse is also true: NCoR binds TR more strongly than it binds RAR. This is, in part, due to the presence of the N3 domain. Although N3 binds TR1 well, it binds RAR␣ poorly. Mutating the N3 domain dramatically decreases interactions of NCoRI with TR in cells but has no effect on the binding of NCoR to RAR. Therefore, the N3 domain also plays an important role in mediating the preferences of corepressors for the distinct NHRs. Although N2 binds TR1 well in EMSA, the isolated domain interacts weakly with TR1 in cells, as measured by the mammalian two-hybrid system. However, constructs containing N3 and N2 interact strongly with TR1 in mammalian cells. These data suggest that two IDs are necessary for full interactions with TR1 in vivo. Prior studies showed that deletion of N2 or N1 (in the presence of N3) results in preserved functional activity of a dominant inhibitory form of NCoR (NCoRI) in cells (33). Our current study extends this finding and shows that mutation of any two ID CoRNR boxes blocks NCoR binding to TR in EMSA, and significantly impairs NCoRI function in transient transfections, particularly if N3 is one of the domains mutated. This is consistent with the N3-N2 combination being sufficient for functional interactions with the TR homodimer complex in vivo. In fact, data from our group and others suggest that the NHR complexes (homodimers and heterodimers) each bind a single NCoR with a stoichiometry of one receptor dimer to one NCoR corepressor (12, 17, 18), with each interacting domain contacting a member of the dimer pair. If two NCoR IDs are necessary for binding, why are there three such domains in NCoR? Our data suggest that the N3-N2 ID combination may be the optimal pair to bind the TR-TR homodimer. It has been suggested that RXR has a preference for the distal corepressor IDs (16). Therefore, the binding of NCoR to TR-RXR heterodimers or TR-TR homodimers may rely on dif-
Specificity of Corepressor Interactions
ferent sets of IDs (Fig. 8). In contrast, the preference of SMRT for receptor heterodimers may depend, in part, on the lack of a region homologous to N3 in SMRT. Thus these data support a model in which the specificity of interactions between corepressors and NHRs depends on sequences within the corepressor interacting domains, both within the CoRNR box motifs (for RAR) and outside that motif. Moreover, these data suggest the presence of a novel interacting domain (N3) that plays a key role in binding TR and mediating the preference of TR homodimers for NCoR. The differences between the NCoR and SMRT IDs may explain corepressor specificity, and thus the distinct physiological roles that corepressors have in vivo (15). A more detailed understanding of the mechanisms by which the corepressor IDs mediate this specificity will allow an opportunity to modulate these effects.
MATERIALS AND METHODS Plasmids All GST fusion proteins were cloned into the vector PGEX4T1 as EcoRI fragments or EcoRI-XhoI fragments. PCR was used to amplify the indicated amino acid sequences from either human (h)NCoR or hSMRT, which were then placed downstream of the GST moiety. A schematic illustration of
1059
these constructs is shown in Fig. 6A. GST-N2 includes amino acids (aa) from hNCoR corresponding to mNCoR aa 2,063– 2,142 (34). GST-S2 includes hSMRT aa 2,119–2,266. GSTN2* includes the same amino acid residues as GST-N2, but with aa 2,073–2,077 mutated to be ISEVI instead of ICQII. Similarly, GST-S2* contains the same amino acids as GSTS2, but mutated such that it contains ICQII instead of ISEVI (aa 2,029–2,033). m1 is similar to N2*, but additionally with Q in position 2071; similarly, m2 is equivalent to S2* but with D in position 2117. m3 contains the proximal 23 aa of N2 and the distal 125 aa of S2; similarly, m4 contains the proximal 23 aa of S2 and the distal portion of N2. PKCR2-NCoRI consists of a portion of hNCoR corresponding to amino acids 1579–2454 of mNCoR (23). Mutations of pKCR2-NCoRI were made using the QuikChange Site Directed Mutagenesis Kit (Stratagene, La Jolla, CA); the codon coding for the initial isoleucine of each CoRNR box was mutated to code for alanine, either separately or in combination. Gal4-NCoR was made by cloning aa 1,579–2,454 of NCoR downstream of the sequence encoding the GAL4 DNA-binding domain in the SV40-driven expression vector pECE. All other Gal4-N2 and Gal4-N3 constructs were made by PCR and placed downstream of the GAl4-DNA binding domain as either ECoRI or EcoRI-PstI fragments. Construct integrity was confirmed by restriction endonuclease digestion and dideoxy sequencing. GST Fusion Proteins GST fusion proteins were expressed in DH5␣ or BL21 Escherichia coli expressing thioredoxin by induction with 0.1 mM isopropylthio--D-galactosidase, as described previously (12). Proteins were isolated by lysis with lysozyme and purified on Sepharose beads. For experiments using EMSA, the bound proteins were eluted using a glutathione buffer. Verification of protein synthesis was obtained on SDS-PAGE. The amount of protein generated was quantified by Bradford assay, such that equivalent amounts of protein could be used in each EMSA. EMSA EMSAs were carried out as previously described (12, 35) with either a 32P-radiolabeled DR⫹4 or DR⫹5 probe. GST-corepressor proteins (GST-CoRs) were purified on Sepharose beads and eluted using a glutathione buffer. Nuclear receptors were in vitro translated in reticulocyte lysate (Promega Corp., Madison WI) using T7 polymerase. For each EMSA, 4 l of in vitro translated TR or RAR were used. For experiments with RXR, 2 l were used, or an equivalent amount of unprogrammed reticulocyte lysate as a control. When NCoRI mutant constructs were used, 4 l of in vitro translated protein were added. When GST-CoR constructs were used, the amount of GST protein used is indicated in each figure. Quantification was determined by Bradford assay. Incubations were carried out for 20 min, and complexes were resolved on a 5% nondenaturing gel, followed by autoradiography. Cell Culture and Transfection
Fig. 8. Binding of NCoR to Homodimers and Heterodimers Is Distinct Schematic model of the interactions between NCoR and TR-TR homodimers and TR-RXR heterodimers. The presence of N3 allows for strong binding of NCoR to the TR homodimer, whereas N1 may play a role specifically in binding the TR-RXR heterodimer.
All transient transfections were performed in CV-1 cells, which were maintained as previously described (33). Mammalian two-hybrid transient transfections were performed in six-well plates using the calcium phosphate technique, with each well receiving 1.7 g of upstream activating sequencethymidine kinase luciferase reporter; 80 ng of Gal4-corepressor construct; 80 ng of VP16-TR1 or VP16-RAR (or empty vector VP16 as a control); and 20 ng of a cytomegalovirus (CMV) -galactosidase construct. Fifteen to 18 h after trans-
MOL ENDO · 2001 1060
fection, cells were washed in PBS and refed with 10% steroid hormone-depleted FBS, as previously described. Forty to 44 h after transfection, cells were lysed and assayed for luciferase and -galactosidase activity. -Galactosidase was used to control for transfection efficiency. Experiments were performed in triplicate. Data are expressed as fold stimulation ⫾ SEM. For Western blot experiments of the Gal4 constructs, 10 g of the indicated construct were transiently transfected into CV-1 cells. Twenty-four hours after transfections, cells were washed in PBS and changed to fresh media; 24 h later, proteins from nuclear extracts were isolated, run on SDSPAGE, transferred to nitrocellulose, blotted using a Gal4 polyclonal antibody (Santa Cruz Biotechnology, Santa Cruz, CA), and visualized by ECL⫹ (Amersham Pharmacia Biotech, Arlington Heights, IL). For transient transfections analyzing the ability of NCoRI or NCoRI mutants to reverse basal repression, 1.7 g DR⫹4pA3Luc, 80 ng of pKCR2-TR1, 160 ng of pKCR2-NCoRI construct (or empty vector pKCR2), and 20 ng of a CMV -galactosidase construct were transfected. After transfection, the procedure was the same as outlined above for the mammalian two-hybrid assay. Data are expressed as relative luciferase activity, after correcting for -galactosidase activity, ⫾ SEM.
Vol. 15 No. 7
4. 5.
6.
7.
8. 9.
10. GST Protein Interaction Assay GST-TR was expressed in BL21 E. coli expressing thioredoxin by induction with 0.1 mM isopropylthio--D-galactosidase. The proteins were isolated with lysis by lysozyme and purified on sepharose beads. Verification of protein synthesis was obtained on SDS-PAGE. GST-TR was incubated with 4 l 35S methionine-labeled in vitro translated NCoRI construct. The concentration of T3 used was 10⫺6 M. After extensive washing, the bound proteins were eluted by boiling in loading buffer and run on SDS-PAGE.
11.
12. 13.
Acknowledgments We would like to thank R. Evans, C. Glass, and N. Moghal for plasmids, and A. Takeshita for thioredoxin-expressing BL21 E. coli.
14.
15. Received February 7, 2001. Revision received March 29, 2001. Accepted April 2, 2001. Address requests for reprints to: Dr. Anthony N. Hollenberg, Thyroid Unit, Department of Medicine, Beth IsraelDeaconess Medical Center, 330 Brookline Avenue Research North 325, Boston, Massachusetts 02215. E-mail: THollenb@ caregroup.harvard.edu. This work was supported by NIH Grants to R. Cohen (DK-02581), F. Wondisford (DK-49126 and DK-53036), and A. Hollenberg (DK-56123).
16. 17.
18.
REFERENCES 1. Mangelsdorf DJ, Thummel C, Beato M, Herrlich P, Schutz G, Umeseno K, Blumberg B, Kastner P, Mark M, Chambon P, Evans RM 1995 Overview: the nuclear receptor superfamily: the second decade. Cell 83:835–840 2. Seol W, Choi HS, Moore DD 1995 Isolation of proteins that interact specifically with the retinoid X receptor: two novel orphan nuclear receptors. Mol Endocrinol 9:72–85 3. Horlein AJ, Naar AM, Heinzel T, Torchia J, Gloss B, Kurokawa R, Ryan A, Kamei Y, Soderstrom M, Glass CK, Rosenfeld MG 1995 Ligand-independent repression by
19. 20.
21.
the thyroid hormone receptor mediated by a nuclear receptor co-repressor. Nature 377:397–404 Chen JD, Evans RM 1995 A transcriptional co-repressor that interacts with nuclear hormone receptors. Nature 377:454–457 Sande S, Privalsky ML 1996 Identification of TRACs (T3 receptor-associating cofactors), a family of cofactors that associate with and modulate the activity of, nuclear hormone receptors. Mol Endocrinol 10:813–825 Ordentlich P, Downes M, Xie W, Genin A, Spinner NB, Evans RM 1999 Unique forms of human and mouse nuclear receptor corepressor SMRT. Proc Natl Acad Sci USA 96:2639–2644 Park EJ, Schroen DJ, Yang M, Li H, Li L, Chen JD 1999 SMRTe, a silencing mediator for retinoid and thyroid hormone receptors-extended isoform that is more related to the nuclear receptor corepressor. Proc Natl Acad Sci USA 96:3519–3524 Glass CK, Rosenfeld MG 2000 The coregulatory exchange in transcriptional functions of nuclear receptors. Genes Dev 14:121–141 Alland L, Muhle R, Hou HJ, Potes J, Chin L, SchreiberAgus N, DePinho RA 1997 Role for N-CoR and histone deacetylase in Sin3-mediated transcriptional repression. Nature 387:49–55 Heinzel T, Lavinsky RM, Mullen TM, Soderstrom M, Laherty CD, Torchia J, Yang WM, Brard G, Ngo SD, Davie JR, Seto E, Eisenman RN, Rose DW, Glass CK, Rosenfeld MG 1997 A complex containing N-CoR, mSin3 and histone deacetylase mediates transcriptional repression. Nature 387:43–48 Nagy L, Kao H-K, Chakravarti D, Lin RJ, Hassig CA, Ayer DE, Schreiber SL, Evans RM 1997 Nuclear receptor repression mediated by a complex containing SMRT, mSin3A, and histone deacetylase. Cell 89:373–380 Cohen RN, Putney A, Wondisford FE, Hollenberg AN 2000 The nuclear corepressors recognize distinct nuclear receptor complexes. Mol Endocrinol 14:900–914 Zamir I, Zhang J, Lazar MA 1997 Stochiometric and steric principles governing repression by nuclear hormone receptors. Genes Dev 11:835–846 Wong CW, Privalsky ML 1998 Transcriptional silencing is defined by isoform- and heterodimer-specific interactions between nuclear hormone receptors and corepressors. Mol Cell Biol 18:5724–5733 Jepsen K, Hermanson O, Onami TM, Gleiberman AS, Lunyak V, McEvilly RJ, Kurokawa R, Kumar V, Liu F, Seto E, Hedrick SM, Mandel G, Glass CK, Rose DW, Rosenfeld MG 2000 Combinatorial roles of the nuclear receptor corepressor in transcription and development. Cell 102: 753–763 Hu X, Lazar MA 1999 The CoRNR motif controls the recruitment of corepressors by nuclear hormone receptors. Nature 402:93–96 Perissi V, Staszewski LM, McInerney EM, Kurokawa R, Krones A, Rose DW, Lambert MH, Milburn MV, Glass CK, Rosenfeld MG 1999 Molecular determinants of nuclear receptor-corepressor interaction. Genes Dev 13: 3198–3208 Nagy L, Kao HY, Love JD, Li C, Banayo E, Gooch JT, Krishna V, Chatterjee K, Evans RM, Schwabe JW 1999 Mechanism of corepressor binding and release from nuclear hormone receptors. Genes Dev 13:3209–3216 Heery DM, Kalkhoven E, Hoare S, Parker MG 1997 A signature motif in transcriptional co-activators mediates binding to nuclear receptors. Nature 387:733–736 Webb P, Anderson CM, Valentine C, Nguyen P, Marimuthu A, West BL, Baxter JD, Kushner PJ 2000 The nuclear receptor corepressor (N-CoR) contains three isoleucine motifs (I/LXXII) that serve as receptor interaction domains (IDs). Mol Endocrinol 14:1976–1985 Darimont BD, Wagner RL, Apriletti JW, Stallcup MR, Kushner PJ, Baxter JD, Fletterick RJ, Yamamoto KR
Specificity of Corepressor Interactions
22.
23.
24.
25.
26.
27.
28.
1998 Structure and specificity of nuclear receptor-coactivator interactions. Genes Dev 12:3343–3356 McInerney EM, Rose DW, Flynn SE, Westin S, Mullen TM, Krones A, Inostroza J, Torchia J, Nolte RT, AssaMunt N, Milburn MV, Glass CK, Rosenfeld MG 1998 Determinants of coactivator LXXLL motif specificity in nuclear receptor transcriptional activation. Genes Dev 12:3357–3368 Chang C, Norris JD, Gron H, Paige LA, Hamilton PT, Kenan DJ, Fowlkes D, McDonnell DP 1999 Dissection of the LXXLL nuclear receptor-coactivator interaction motif using combinatorial peptide libraries: discovery of peptide antagonists of estrogen receptors ␣ and . Mol Cell Biol 19:8226–8239 Hollenberg AN, Monden T, Madura JP, Lee K, Wondisford FE 1996 Function of nuclear co-repressor protein on thyroid hormone response elements is regulated by the receptor A/B domain. J Biol Chem 271:28516–28520 Feng W, Ribeiro RCJ, Wagner RL, Nguyen H, Apriletti JW, Fletterick RJ, Baxter JD, Kushner PJ, West BL 1998 Hormone-dependent coactivator binding to a hydrophobic cleft on nuclear receptors. Science 280:1747–1749 Smith CL, Nawaz Z, O’Malley BW 1997 Coactivator and corepressor regulation of the agonist/antagonist activity of the mixed antiestrogen, 4-hydroxytamoxifen. Mol Endocrinol 11:657–666 Wagner BL, Norris JD, Knotts TA, Weigel NL, McDonnell DP 1998 The nuclear corepressors NCoR and SMRT are key regulators of both ligand- and 8-bromo-cyclic AMPdependent transcriptional activity of the human progesterone receptor. Mol Cell Biol 18:1369–1378 Jackson TA, Richer JK, Bain DL, Takimoto GS, Tung L, Horwitz KB 1997 The partial agonist activity of antagonist-occupied steroid receptors is controlled by a novel
1061
29.
30.
31.
32.
33.
34.
35.
hinge domain-binding coactivator L7/SPA and the corepressors N-CoR or SMRT. Mol Endocrinol 11:693–705 Renaud JP, Harris JM, Downes M, Burke LJ, Muscat GE 2000 Structure-function analysis of the Rev-erbA and RVR ligand-binding domains reveals a large hydrophobic surface that mediates corepressor binding and a ligand cavity occupied by side chains. Mol Endocrinol 14: 700–717 Xu J, Qiu Y, DeMayo FJ, Tsai SY, Tsai MJ, O’Malley BW 1998 Partial hormone resistance in mice with disruption of the steroid receptor coactivator-1 (SRC-1) gene. Science 279:1922–1925 Weiss RE, Xu J, Ning G, Pohlenz J, O’Malley BW, Refetoff S 1999 Mice deficient in the steroid receptor coactivator 1 (SRC-1) are resistant to thyroid hormone. EMBO J 18:1900–1904 Xu J, Liao L, Ning G, Yoshida-Komiya H, Deng C, O’Malley BW 2000 The steroid receptor coactivator SRC-3 (p/CIP/RAC3/AIB1/ACTR/TRAM-1) is required for normal growth, puberty, female reproductive function, and mammary gland development. Proc Natl Acad Sci USA 97:6379–6384 Cohen RN, Wondisford FE, Hollenberg AN 1998 Two separate NCoR interacting domains mediate corepressor action on thyroid hormone response elements. Mol Endocrinol 12:1567–1581 Seol W, Mahon MJ, Lee Y-K, Moore DD 1996 Two receptor interacting domains in the nuclear hormone receptor corepressor RIP13/N-CoR. Mol Endocrinol 10: 1646–1655 Hollenberg AN, Monden T, Flynn TR, Boers M-E, Cohen O, Wondisford FE 1995 The human thyrotropin-releasing hormone gene is regulated by thyroid hormone through two distinct classes of negative thyroid hormone response elements. Mol Endocrinol 10:540–550
