A Novel Mechanism of Thyroid Hormone-dependent Negative Regulation by Thyroid Hormone Receptor, Nuclear Receptor Corepressor (NCoR), and GAGA-binding Factor on the Rat CD44 Promoter*

CD44 is an adhesion molecule in the extracellular matrix that shows various functions, including tumor genesis and metastasis. A recent study showed that CD44 expression level was strongly correlated with the generation of papillary thyroid carcinomas, the most prevalent malignancy of the thyroid gland. We report here that CD44 is negatively regulated by thyroid hormone (T3) through a novel mechanism. We demonstrate that nuclear receptor corepressor (NCoR) enhances thyroid hormone receptor (TR)-mediated basal transactivation by a weak TR·DNA interaction in the absence of T3, which is repressed by T3 through a transient TR ·DNA interaction. Initially, we identified that CD44 was negatively directly transcriptionally T3 -responsive. Deletion and mutation analysis indicated that both a weak TR and a GAGA-binding factor (GAF) binding sites on the CD44 promoter were required for negative regulation by T3. The weak TR·DNA interaction was further confirmed by electrophoretic gel mobility shift assay, chromatin immunoprecipitation, and transfection assays using a non-DNA-binding TRα1 mutant. More interestingly, NCoR acted as a co-activator to enhance TR-mediated basal transactivation in the absence of T3. This effect was eliminated by removal of TR or NCoR binding. Most strikingly, T3 induced a remarkable increase in TR·DNA binding at 40–60 min after T3 exposure that rapidly returned to basal levels, suggesting a T3-induced remodeling of chromatin structure at the early stage of T3 stimulation resulting in repression. Therefore, we propose a mechanism by which NCoR, GAF, and TR interact with the CD44 negative T3-responsive element to enhance basal transactivation, whereas T3 induces the remodeling of chromatin structure for repression.

Thyroid hormone affects important biological functions such as growth, development, and metabolism. The biologically active thyroid hormone triiodothyronine (T 3 ) 1 regulates gene ex-pression by interacting with two isoforms of the thyroid hormone receptor (TR), TR␣ and TR␤, each of which consists of a DNA binding domain (DBD), a ligand-binding domain, co-activator/co-repressor binding domains, and a transcription activation domain. The P-box of the zinc fingers within the DBD of TR recognizes the AGGTCA sequence of the thyroid hormoneresponsive element (TRE), which is diversely arranged in direct repeat (DRϩ4), palindrome, and inverted palindrome forms (1,2). When unliganded TR binds to TRE, it recruits co-repressors (NCoR/SMRT) and histone deacetylases for repression. On the other hand, T 3 binding to TR induces conformational changes in TR to release co-repressors and recruit co-activators such as CBP/p300, pCAF, and SRC-1 for transactivation (3)(4)(5)(6)(7)(8). Thus, although the mechanism of transcriptional regulation for the positive TRE (pTRE) is relatively well studied, T 3 -dependent negative gene regulation is still poorly understood and is limited to a few genes. It has not been possible to identify a consensus sequence of negative TRE (nTRE).
Recently, Wondisford and colleagues (9) provided evidence that direct binding of hTR␤2 to nTRE is required for T 3 -dependent negative regulation. This group generated a non-TRE binding hTR␤2 mutant (TR␤2-GS125) in which two amino acids (EG at 125 and 126 in the zinc finger region of DBD of TR␤2) were mutated to GS, preventing recognition of the TRE sequence. When TR␤2-GS125 was transfected into 293T cells, T 3 -dependent negative regulation of the hTSH␤ gene was abolished (9). Subsequently, Wondisford et al. showed that TR␤2 bound to the TRE sequence in the hTSH␤ fragment elicited a negative response to T 3 . The physiological significance of TR⅐DNA binding was further confirmed by the finding that TR␤2-GS125 knock-in homozygous mutant (TR␤ GS/GS ) mice displayed abnormal T 3 regulation of the hypothalamic-pituitary thyroid axis and retina, identical to abnormalities previously observed in TR␤ KO (TR␤ Ϫ/Ϫ ) mice, although TR␤ GS/GS mutant mice showed much less severe defects in hearing and outer hair cell loss than do TR␤ Ϫ/Ϫ mice (9). These results indicate that TR⅐DNA binding is necessary for negative transcriptional regulation of the TSH␤ gene.
Other studies suggest that TR binds directly to TREs or TRElike sequences for negative regulation by T 3 (9,10) and that various types of nTRE sequences resembling TREs are located in the promoters (trkB and amyloid precursor protein) (11)(12)(13), the first exons (c-myc) (14), and the 3Ј-untranslated regions (rat growth hormone and Clone 144) (15). For example, the TSH␣ promoter contains palindromic TREs necessary for negative regulation by T 3 (16), whereas mouse TSH␤ has a DRϩ2 TRE that includes a TRE near the TSS (9,17,18). RSV-LTR, Cyp7A1, and SCD1 promoters have different structures (10,19,20). However, no nTRE consensus sequence distinct from pTRE was clearly identified, raising the possibility of a novel nTRE different in sequence from these positively regulated genes.
On the other hand, there are also data that T 3 -dependent negative regulation can occur without direct TR⅐DNA interaction. Kushner and colleagues (21,22) showed that TR interacted with jun/fos through the AP1 site by a protein-protein interaction that was subsequently regulated by T 3 . Another recent study from our group indicated that JEG-3 cell-specific transcription factor-DNA binding induced T 3 -dependent negative regulation without direct TR⅐DNA binding (23). Jameson and colleagues indicated that TSH␣ gene expression might be negatively regulated by T 3 through a squelching mechanism (24,25). This model proposed that, in the absence of T 3 , unliganded TR bound to co-repressors, forming a complex that was subsequently sequestered from the promoter region, resulting in increased histone acetylation and transactivation through the recruitment of more co-activators to the promoter. However, in the presence of T 3 , T 3 -TR formed a complex with co-activators such as CBP/p300 that was segregated from the TSH␤ promoter, giving co-repressors NCoR/SMRT and histone deacetylases easy access to the promoter region for histone deacetylation and repression (25). When NCoR or SMRT was overexpressed in the presence of TRH and TSH␣ and -␤ promoter-Luc, TR-mediated basal transactivation of these promoters was enhanced by TR⅐NCoR interaction (9). These results suggest that negative regulation by T 3 could be induced without direct TR⅐DNA interaction (25). However, this model did not explain how the specificity of negative regulation is conferred. Thus, the question of whether TR⅐DNA interaction with the negatively regulated genes is essential for T 3 -dependent negative regulation is still not resolved.
We approached this problem by screening the rat genome for other genes whose expression is suppressed by T 3 . We determined that CD44 is negatively responsive to T 3 at a directly transcriptional level in various tissues and cell types. CD44 is a cell adhesion molecule involved in diverse biological processes, including angiogenesis, lymphogenesis, wound healing, inflammation, and cancer metastasis (26). Recent reports indicate that CD44 plays an important role in generation of papillary thyroid carcinomas (PTCs), the most prevalent malignancy of the thyroid gland. Some PTCs are thought to be initiated by the RET/PTC chimeric oncogene, which is generated by rearrangements of the RET receptor tyrosine kinase (26). RET/PTC signaling induces up-regulation of osteopontin and CD44, resulting in proliferation and invasion of transformed PC Cl 3 thyrocytes (26). Another interesting and relevant finding is that Alzheimer's disease-associated ␥-secretase cleaves CD44, generating intracellular domains for nuclear signaling and CD44 ␤-peptides, whose functions are unknown. This cleavage pattern is similar to that of ␤-amyloid precursor protein by the ␥-secretase, which produces an amyloid precursor protein intracellular domain for nuclear signaling and an amyloid ␤-peptide for ␤-amyloid plaque formation (27). Notch, low density lipoprotein receptor-related protein, E-cadherin, and ErbB-4 are also the same family proteins (28 -36). All of these results suggest a correlation between regulation of CD44 expression and progress of these diseases.
Here we provide evidence that, during negative regulation by T 3 , TR binds weakly but directly to a novel CD44 nTRE different from the pTRE. In addition, we identified two proteins (GAF and NCoR) required for T 3 -dependent negative regulation of CD44 gene expression. Of special note, cooperative interaction among TR, NCoR, and GAF was essential for enhanced unliganded TR-mediated basal expression, whereas repression was induced by T 3 through a transient, but strong in vivo TR⅐DNA interaction, probably also bringing about remodeling of the chromatin structure. These results suggest a novel mechanism for T 3 -dependent negative regulation.

MATERIALS AND METHODS
Cell Culture-Cells were cultured as described previously (23,37). Briefly, COS-7, JEG-3, and HeLa cells were maintained in Ham's F-10 media containing 14 mM sodium bicarbonate (pH 7.4) and 2 mM Lglutamine supplemented with 10% fetal bovine serum (Invitrogen) until cells were 70 -80% confluent (29). For stimulation with T 3 , culture medium was removed, the cells were rinsed once with phosphatebuffered saline, and Ham's F-10 medium containing charcoal-stripped 10% fetal bovine serum was added. The medium was changed after 1 day, and incubation continued for another 2 days. T 3 (50 nM) was diluted in Ham's F-10 medium, and charcoal-stripped 10% fetal bovine serum was added to cells for the times indicated in figure legends.
Plasmid Constructions and Mutagenesis-Standard techniques were used for all plasmid constructions, as described previously. A 5-kb 5Ј-FR genomic fragment, PCR-amplified 0.1-to 1.8-kb DNA fragments of 5Ј-FR, or synthesized double-stranded oligonucleotides were subcloned into KpnI and XhoI restriction sites of the pGL3-basic luciferase-expressing reporter vector (Promega, Madison, WI). A 177-bp CD44 promoter in pGL3BS-Luc (177-bp CD44-Luc) was mutated using the QuikChange site-directed mutagenesis kit (Stratagene, La Jolla, CA) as indicated in figure legends. To generate a non-TRE binding GS71 sequence as already described in JEG-3 cells (23), mTR␣1 cDNA was inserted in a CDM expression vector in which two bases from TSS of the mTRa1 cDNA were mutated: 614 (A to G) and at 616 (G to A). All constructs were confirmed by automated sequencing.
Transfection and Luciferase (Luc) Assay-Transfections were performed as previously described (23) with FuGENE 6 transfection reagent (Roche Applied Science) in COS-7 cells. Each plate was transfected with 2 g of Luciferase reporter plasmids and 0.5 g of TKGH, which constitutively expresses human growth hormone (hGH), as a control for transfection efficiency. 0.1 g of CDM8 vector expressing mouse TR␣1 (wild-type or mutant) or pCDNA3 vector expressing NCoR were transfected with and without 0.1-0.3 g of pSilencer 2.1-U6 neo (Ambion) expressing siRNAs of NCoR and GAF, as described in the figure legends. CDM8 and pCDNA3 empty vectors or GFP and random oligonucleotide siRNA-expressing vectors were used as controls. After transfection, cells were incubated in 60-mm culture dishes in the absence of thyroid hormone for 2 days and then exposed to 50 nM T 3 for 0 -24 h at 37°C. A luciferase assay was performed according to the manufacturer's protocol (Promega) and was normalized to hGH expression in the absence and presence of T 3 to calculate T 3 responsiveness. Transfection data are mean Ϯ S.D. of a minimum of triplicate samples.
Electrophoretic Mobility Shift Assay-EMSA was performed as described previously (23). Double-stranded oligonucleotides of wild-type and mutant CD44 nTRE were radiolabeled with [ 32 P]dCTP (PerkinElmer Life Sciences) by fill-in reaction using Klenow DNA polymerase, which was subsequently gel-purified (23). Chicken TR␣1 and human retinoid X receptor ␣ were overexpressed in Escherichia coli and purified as described previously (23). Radiolabeled probes were reacted with bacterially expressed cTR␣ and/or human retinoid X receptor and resolved by non-denaturing PAGE.
Methylation Interference Assay-Dimethyl sulfate methylation interference assay was performed using a standard protocol as described previously (23).
Chromatin Immunoprecipitation Assay-ChIP assay was performed according to manufacturer's protocol (Upstate, Lake Placid, NY) (38,39). Briefly, 177-bp wild-type (177-bp CD-Luc) and mutant (177-bp CD-Dbl-Luc) constructs were co-transfected with TR and/or NCoR expression vectors into COS-7 cells and incubated in the presence or absence of 50 nM T 3 for 0 -24 h as represented in Fig. 8. Cells were cross-linked by 1% formaldehyde for 10 min and harvested in the presence of protease inhibitor (EDTA-free Complete, Roche Applied Science). These cells were then lysed and sonicated to generate 200-to 500-bp DNA fragments, which were subsequently confirmed by agarose gel electrophoresis. TR and NCoR binding fragments were immunoprecipitated using rabbit polyclonal anti-TR and goat polyclonal anti-NCoR antibodies (Santa Cruz Biotechnology, Santa Cruz, CA). As a control, immunoglobulin G (IgG) was used to determine the background level. Nonspecific interaction was minimized by thorough washing. Specifically bound DNA fragments were purified and quantified by real-time PCR using specific primers. The sense primer was pGL3P-4952S1 (pGL3 Promoter-Luc vector, Ϫ4952 to Ϫ4971: 5Ј-CTAGCAAAATAG-GCTGTCCC-3Ј), and the antisense primer was 31CD described in Fig.  8 (CD44, Ϫ31 to Ϫ51: 5Ј-CCAGGCTTTGAAAGAGTGACC-3Ј).
Miscellaneous-All chemicals were of reagent grade and obtained from commercial sources. Rat tissues were obtained under a protocol approved by the Brigham and Women's Hospital animal care and use committee and in compliance with National Institutes of Health guidelines.

RESULTS
The CD44 Gene Is Negatively Responsive to T 3 at a Transcriptional Level-To study the general mechanism of regulation of negatively T 3 -responsive genes, we performed screens to identify genes directly and transcriptionally negatively regulated by T 3 . First, we identified a group of negatively T 3responsive genes using mRNA from T 3 -deprived pituitary tumor (GC) cells grown in medium containing thyroid hormonefree serum (hypothyroid) and serum supplemented with normal (euthyroid) or excessive quantities of T 3 (hyperthyroid). T 3 -dependent mRNA expression was examined for 7000 known genes using the Affymetrix GeneChip (Rat Genome U34A), in which the most abundantly expressed genes were primarily selected and confirmed by quantitative real-time PCR. Cyclophilin was used as an internal control for normalization of mRNA expression. Amplified DNA was quantitated using the standard curve generated with serially diluted DNA after triplicate PCR as described under "Materials and Methods." The eight genes identified as strongly negatively regulated by T 3 were tested to determine whether they are transcriptionally or post-transcriptionally regulated by T 3 . To identify which of the cDNAs were transcriptionally regulated by T 3 , 2 M actinomycin D was used to block transcription, and the half-life of the mRNA was measured in the presence and absence of T 3 . Genes that showed no changes in mRNA half-life by T 3 and actinomycin D exposure were categorized as transcriptionally regulated (data not shown). To identify those directly responsive to T 3 , 30 M cycloheximide was added to hypothyroid GC cells 1 h prior to T 3 treatment. Cells were taken for RNA isolation after 0, 3, and 6 h of exposure to T 3 , and the changes in the specific RNAs were compared with control cells not exposed to cycloheximide. Genes that showed no changes in mRNA level by T 3 and cycloheximide exposure were categorized as directly responsive to T 3 (data not shown). Four genes met both criteria.
To identify which of these four genes was negatively regulated by T 3 in a general, as opposed to a tissue-specific fashion we examined mRNA of nine tissues from hypothyroid, euthyroid, and hyperthyroid (4-h and 24-h T 3 -treated) rats. These included blood, lung, anterior pituitary (AP), large intestine (L. Int.), stomach (STM), kidney, liver, heart, and hypothalamus (HPT) (Fig. 1). Serum T 4 and T 3 measurements were also performed to confirm the thyroid status of these animals. Examination of T 3 -dependent mRNA expression of the four genes in these tissues by quantitative real-time PCR showed that two were generally negatively regulated, whereas the other two were negatively regulated in a tissue-specific manner (data not shown). Because the CD44 gene was negatively regulated in most tissues (Fig. 1), it was selected as a model gene with which to study the T 3 -dependent negative regulation mechanism.
Localization of CD44 Sequences (Ϫ129 to Ϫ66 bp from TSS) Conferring a Negative Response to T 3 -To determine whether negatively T 3 -responsive element (nTRE) is localized within the 5Ј-flanking region (5Ј-FR) of the rCD44 gene, we amplified this fragment by long distance PCR using a rat genomic DNA as a template, and inserted this fragment into a promoter-less luciferase-reporter pGL3 vector (5-kb CD44-Luc). When the 5-kb CD44-Luc construct was co-transfected with mouse thyroid hormone receptor ␣1 (mTR␣1) expressing plasmid into COS-7 cells, luciferase expression was decreased 2-fold with T 3 treatment ( Fig. 2A). Similar results were obtained using other cell lines such as anterior pituitary (GC) and hypothalamic (HT22) cells (data not shown), confirming that this 5-kb 5Ј-FR fragment was negatively responsive to T 3 . Serial deletion mutation analysis showed that the negative response to T 3 was maintained until the gene fragment was reduced to 177-bp (Ϫ129 to ϩ48 bp from the transcriptional start site (TSS)) ( Fig. 2A). The effect of TR␤1 was similar to that of TR␣1 (data not shown). When the 5Ј-FR was further reduced to 113-bp (Ϫ65 to ϩ48 bp), the repression ratio (ϩT 3 Luc/ϪT 3 Luc) of the 113-bp fragment was significantly reduced to 0.71 ( Fig. 2A), and the transcriptional potency of the construct in the absence of T 3 was also substantially lowered to near-background level (Fig. 2B). Furthermore, when GAGA factor (GAF) binding site (TCTC at Ϫ1 to ϩ3 bp) within the 113-bp fragment was mutated to TTTT, the weak residual TR-response was completely abolished (Fig. 2B, (Ϫ65ϳϩ48)-GAF). These results indicate that the GAF binding site and the sequences located between Ϫ129 and Ϫ66 bp are critically important for negative regulation of CD44 gene expression. This suggests that the sequences between Ϫ129 and Ϫ66 bp include TR binding sites for a negative response to T 3 . ing sites within the 64-bp fragment (Ϫ129 to Ϫ66 bp) that conferred a negative response to T 3 , we performed an EMSA assay. First, we divided the 177-bp rCD44 DNA fragment into two fragments: 89-bp (Ϫ129 to Ϫ41 bp) and 88-bp (Ϫ40 to ϩ48 bp), which were 32 P-labeled and incubated with bacterially expressed cTR␣1. EMSA showed that TR␣1 strongly bound the 89-bp fragment (Ϫ129 to Ϫ41 bp) but not with the 88-bp fragment, as expected (data not shown). Subsequent dimethyl sulfate footprinting assay with the 89-bp fragment showed that G-residues at Ϫ80, Ϫ78, Ϫ52, and Ϫ51 of the top strand and at Ϫ76, Ϫ75, Ϫ49, and Ϫ47 of the bottom strand were specifically protected by cTR␣1 (Fig. 3A). Interestingly, the protected regions consisted of two positive TRE half-sites with "everted" palindromic orientation separated by 20 bp (Fig. 3B). Each protected region contained an additional protected Gly residue at the end of the TRE half-site that was different from the typical pTRE, which consists of a direct repeat of a 4-bp sequence separation (DRϩ4).
When we examined the TR binding affinity to the 45-bp CD44 fragment containing the two TREs (Ϫ87 to Ϫ43 bp), we found that it was similar to the hdio1 DRϩ4 TRE (Fig. 3C,  lanes 1-7). However, nTREs such as hTSH␤ and RSV, and the 170-bp fragment of rD2 conferring a negative response to T 3 showed much lower affinity to TR␣1 than did the 45-bp CD44 fragment (Fig. 3C, lanes 8 -10). This raises the question of whether the presence of high affinity TREs in CD44 is required for negative regulation by T 3 .
The High Affinity TR Binding of TREs within the 64-bp Fragment (Ϫ129 to Ϫ66 bp) Was Not Required for T 3 -dependent Negative Regulation-We sought to determine whether these high affinity TREs were directly involved in negative regulation by T 3 . First, we generated several mutations at or around the TREs, as depicted in Fig. 4A. EMSA showed that mutations of both TRE sites resulted in a complete loss (M1) or a significant decrease (M2) in TR binding (Fig. 4B). However, shortening of the spacer length without changing the TR binding sites did not decrease the binding affinity (M3-M5) but actually increased binding in M6 (Fig. 4B). This again confirmed that these two TREs are high affinity TR binding sites (Fig. 4B).
Subsequently we generated more selective TRE mutations in the 177-bp CD44-Luc construct, as shown in Fig. 4C. These were the 177-bp CD-Dist (mutation in the distal TRE), the 177-bp CD-Prox (mutation in the proximal TRE), and the 177-bp CD-Dbl (mutation in both TREs). When these constructs were co-transfected with mTR␣1-expressing construct into COS-7 cells, none of them showed disrupted T 3 -dependent negative regulation (Fig. 4D). The total luciferase expression by mutated constructs (in the absence and presence of T 3 ) was decreased 2-to 3-fold compared with the wild type, but the changes in the repression ratios were small compared with those noted with the wild type (Fig. 4D, WT). When empty vector lacking mTR␣1 was used as a control, it did not respond to T 3 (Fig. 4D, CNT). These results indicated that high affinity TR binding TREs are not necessary for T 3 -dependent negative regulation of CD44 gene expression. This raises the possibility that weak or transient TR binding sites are present within the 64-bp CD44 fragment conferring negative response to T 3 , as illustrated in hTSH␤ (9) and RSV nTRE (10, 40) fragments (Fig. 3C). Direct TR⅐DNA binding is required for T 3 -dependent negative regulation of the mutant 177-bp CD44-Dbl fragment that lost its capacity for high affinity binding to TR␣1.
To determine whether there is direct TR⅐DNA interaction in vivo for negative regulation by T 3 , we generated a non-DNA binding mutant GS71. The amino acids at 71 and 72 in the P-box of DNA binding zinc-finger domain of mTR␣1 were converted from EG (Glu and Gly) to GS (Gly and Ser) of mTR␣1 (Fig. 5, A and B), which eliminated DNA binding capability, as was the case in the hTR␤2-GS125 mutant (9).
When the TRE double mutant 177-bp CD44-Dbl-Luc was co-transfected with the mTR␣1 expression construct into COS-7 cells, Luc expression was decreased by T 3 as expected (Fig. 5C, TR␣1). However, transient expression of GS71 instead of TR␣1 eliminated any T 3 -dependent negative regulation (Fig. 5C, GS71). Similar results were obtained using the wild-type 177-bp CD44-Luc (data not shown) as well as the 170-bp rD2-Luc and 1.2-kb hTSH␤-Luc conferring a negative response to T 3 (Fig. 5, D and E). To demonstrate the effects of these two receptors on these constructs, we included a potent positively responding reporter, a TK-Luc plasmid containing three copies of the consensus DRϩ4 TRE (41) from the hdio1 gene were used. These plasmids, along with an empty vector control, were transfected into COS-7 cells. The results of luciferase expression relative to an internal transfection efficiency control (TKGH), with and without T 3 , are shown in Fig. 5F. These data indicated that the potent positive regulation by the TR␣ was markedly decreased in GS71. In sum, these data indicated that the direct weak or transient TR binding site for T 3 -responsive repression was located within the 177-bp rCD44-Dbl DNA fragment, even in the absence of high affinity TR binding sites.
NCoR Enhances TR-mediated Basal Transactivation of the rCD44 Gene-The indication that direct but weak TR⅐DNA binding is necessary for negative regulation by T 3 raises the possibility that other co-repressors interact with TR on the nTRE. Therefore, we addressed whether NCoR is involved in this negative regulation mechanism. When the 177-bp CD44- Dbl-Luc plasmid was co-transfected with mTR␣1 or mTR␣1 plus human NCoR-expressing constructs, NCoR further enhanced mTR␣1-mediated basal transactivation in the absence of T 3 , which then returned to basal or lower levels by T 3 (Fig.  6A, lane 2 versus 5). Consistently, expression of the C-terminal region of NCoR was enough to generate the same effect on the enhanced mTR␣1-mediated basal transactivation as did by the full-length NCoR (data not shown), suggesting that transactivation domain is localized within the C-terminal region. This result was further supported by the knockdown of NCoR expression. When NCoR siRNA was co-expressed, NCoR-mediated basal transactivation was completely abolished (Fig. 6A,  lanes 4 and 7), indicating the abrogation of TR-mediated basal transactivation. Decrease of NCoR mRNA expression after siRNA treatment was confirmed by PCR (data not shown). Co-expression of green fluorescence protein siRNA as a control did not significantly affect T 3 -dependent negative regulation (Fig. 6A, lane 8). Notably, when NCoR siRNA was co-expressed FIG. 3. Two high affinity TR binding sites located ؊75 and ؊52 bp from TSS of the CD44 promoter. A, dimethyl sulfate in vitro footprinting of 89-bp CD44 5Ј-FR fragment (Ϫ129ϳϪ41 bp from TSS). 32 P-Labeled 89-bp probe methylated by dimethyl sulfate was interacted with TR␣1. Bound and free fragments were isolated, and the protected Gly residues were identified by cleaving methylated G-residues with piperidine. Protected Gly residues are marked with arrowheads. B, schematic of Gly residues protected by cTR␣1. Potential TR binding half-sites are marked by arrows. C, EMSA using the DNA fragment containing positive and negative TREs. Interaction of bacterially expressed cTRa with hD1 DRϩ4 TRE, rCD44 nTRE (Ϫ87ϳϪ43 bp), hTSH␤ (1ϳϩ40 bp), RSV-nTRE (Ϫ115ϳϪ79 bp), and rD2 nTRE (Ϫ115ϳϩ55 bp). 1 ng of 32 P-labeled probe was reacted with 0.2 g of bacterially expressed cTR␣. with mTR␣1 without NCoR, TR-mediated basal transactivation was completely abolished (Fig. 6A, lane 4), suggesting not only that basal transactivation by TR␣1 on this promoter is regulated by endogenously expressed NCoR, but also that NCoR is required for TR-mediated basal transactivation. Negative regulation by T 3 was not observed when a non-DNA binding mutant GS71 was used (Fig. 6A, lane 3 and 6), suggesting that enhanced basal transactivation by NCoR was mediated through TR bound to DNA. As a control, 3XhD1-Luc containing three copies of human D1 DRϩ4 TRE was co-transfected with the NCoR expression plasmid; TR-mediated transactivation was repressed by NCoR (Fig. 6B). This was consistent with published data on the positive TRE (25,42,43). Thus, these data support the concept that NCoR acts as a co-activator to enhance TR-mediated basal transactivation of this negatively T 3 -responsive promoter. Unlike the mechanism responsible for the squelching of NCoR and histone deacetylases by TR, T 3 -dependent negative regulation of the CD44 promoter requires direct TR⅐DNA binding for the NCoR effect.
Identification of Weak TR Binding Site within the 64-bp CD44 Fragment (Ϫ129 to Ϫ66 bp)-We performed EMSA to clarify whether TR can interact with the CD44 promoter during the enhancement of TR-mediated basal transactivation by NCoR. When three 32 P-labeled DNA fragments of 29 bp (Ϫ129ϳϪ101 bp), 29 bp (Ϫ100ϳϪ72 bp), and 30 bp (Ϫ65ϳϪ36 bp) were incubated with bacterially expressed cTR␣1, a specific retarded band appeared in the band of one fragment (Ϫ100ϳϪ72 bp) (Fig. 7B, lane 3; Fig. 7C, lane 3), even though its binding affinity was about 100-fold less than that of DRϩ4positive TRE (Fig. 7B, lane 1; note the difference in loading amount). The retarded band was significantly reduced either by addition of the 50ϫ unlabeled 29-bp (Ϫ100ϳϪ72bp) fragment (Fig. 7C, lane 2) or by a mutation of the 6 bases at Ϫ82ϳϪ87 (Fig. 7A, Dbl-M1; Fig. 7C, lane 4). These data sug-
GAF Binding Is Required for T 3 -dependent Negative Regulation of the CD44 Gene-We further examined whether TR binding to weak TRE at Ϫ82ϳϪ87 bp was sufficient to bring about negative regulation by T 3 . When either the weak TRE at Ϫ82ϳϪ87 alone (Fig. 8A, Dbl-M1) or a GAGA factor (GAF) binding site at Ϫ1ϳϩ3 bp alone (Fig. 8A, Dbl-M2) was mutated within the 177-bp CD44-Dbl-Luc plasmid, T 3 -dependent negative regulation was not significantly affected (Fig. 8B, Dbl-M1 and Dbl-M2). However, when both the GAF and the weak TR binding sites were mutated, transcriptional activity was almost abolished, showing a Ͼ2000-fold reduction compared with 177-bp CD44-Dbl (Fig. 8, A and B, Dbl-M3). In addition, when GAF expression was knocked down by transfecting 0.1 or 0.3 g of GAF siRNA-expressing plasmid in COS-7 cells, NCoRenhanced TR-mediated basal transactivation was abolished (Fig. 8D, siGAF/0.1 and siGAF/0.3). As a control, a plasmid that produces random oligonucleotide siRNA was transfected at the same time but elicited no significant response (Fig. 8D, siRDM/  0.3). All of these data indicate that binding of both GAF and weak TR to sites at Ϫ1ϳϩ3 bp and Ϫ82ϳϪ87 bp (respectively) is required for NCoR-dependent TR-mediated basal transactivation and T 3 -dependent repression of CD44 gene expression.
Transient Interaction of TR and NCoR with CD44 nTRE in Vivo-It has recently been shown that the ChIP assay can be applied to the study of endogenous, integrated, or transiently transfected TR-regulated genes (9,38,39,44). Therefore, we decided to perform ChIP assays to determine whether we could detect TR⅐DNA interaction in vivo using a transient transfection system. First, we optimized a protocol to quantitate protein⅐DNA binding by a combination of ChIP assay and quantitative real-time PCR (quantitative real-time PCR) as described in detail under "Materials and Methods." First, as a control experiment, we determined TR⅐DNA interaction on the high affinity TRE of the wild-type 177-bp CD44 promoter. Wild-type 177-bp CD44-Luc was co-transfected with TR␣1 (or GS71) and NCoR expression constructs into COS-7 cells with and without 50 nM T 3 for 24 h, followed by crosslinking with 1% formaldehyde to fix the cells, and ChIP assay was performed using a specific anti-TR␣1 antibody. The TR-bound DNA was analyzed by quantitative real-time PCR using promoter-specific primers as depicted in Fig. 9A. Transfection efficiency was normalized to hGH by co-transfection of TKGH plasmid. As an immunoprecipitation control, immunoglobulin G (IgG) was used to establish a nonspecific background against which normalization of specific signals was measured. Ratios of signals to background by IgG are illustrated in Fig. 9B.
The results showed a strong TR binding to CD44 promoter in the absence of T 3 , which was more than 10-fold greater than the IgG background (Fig. 9B, lane 2). This interaction was decreased after 24-h T 3 exposure (Fig. 9B, lane 2). Similarly, EMSA showed that T 3 significantly decreased TR binding affinity to DNA, even though its interaction was not completely abrogated, probably due to differences between in vivo and in vitro experimental conditions (data not shown). On the other hand, transient expression of non-TRE binding GS71 showed that the interaction between GS71 and DNA was much less than that by TR␣1 (Fig. 9B, lanes 2 versus 4). This result was consistent with the expected high affinity TR⅐DNA interaction on the wild-type CD44 promoter, validating this protocol as a suitable method by which to analyze in vivo protein-DNA interaction.
We then examined whether TR⅐DNA interaction could be detected using a 177-bp CD-Dbl-Luc that contained no high affinity TREs using the same protocol. In the absence of T 3 , TR␣1-DNA interaction was compared with GS71 as a control. As expected, we detected a weak but direct TR␣1⅐DNA interaction that was significantly higher than that with GS71, even though it was just above the background nonspecific interaction level by IgG (Fig. 9C, lane 2, TR␣1 versus GS71). These results indicate that the difference in DNA binding affinity between TR␣1 and GS71 represents actual TR␣1-DNA binding, because nonspecific binding to DNA by IgG was relatively higher than the back ground level by GS71. This suggests that TR interacts weakly but specifically with the CD44 promoter. Thus, the in vivo result was consistent with the in vitro results using EMSA (Fig. 7, B and C). Similar results were also obtained with NCoR, suggesting that NCoR interacts with the TR⅐DNA complex (Fig. 9C, lane 3).
We extended this experiment to examine the effects of T 3 on TR⅐DNA interaction. When ChIP assay was performed with COS-7 cells that were co-transfected with 177-bp CD44-Dbl- Luc plasmid, and TR␣1-and NCoR-expressing constructs, and were subsequently exposed to T 3 for 0 min, 10 min, 20 min, 30 min, 40 min, 50 min, 60 min, and 24 h, T 3 surprisingly induced transient interaction of TR⅐DNA over 60 min with increases seen at 10, 30, 50, and 60 min, and decreases at 20 and 40 min (Fig. 9D). Interestingly, the peak of the strong TR⅐DNA inter-action appeared between 50 -60 min (Fig. 9D). When we repeated the same experiment with 20-min intervals of T 3 stimulation, TR⅐DNA binding was decreased at 20 min of T 3 exposure but remarkably increased at 40 min. TR⅐DNA interaction then decreased and maintained a steady lower level over 24 h (data not shown). In further experiments, the same pat- tern of transient TR⅐DNA interaction was reproduced, even though the peak of the strong TR⅐DNA interaction was slightly shifted (possibly due to small differences in experimental conditions). This clearly indicates that TR interacted with DNA strongly during the early stages of T 3 stimulation. Transfection efficiency was monitored by hGH expression, maintaining a steady level during all transfections, with 8% standard deviation. Amounts of DNA used for immunoprecipitation did not vary significantly between samples, as measured by ethidium bromide staining of the agarose gel. This transient interaction suggested that T 3 -TR recruited chromatin remodeling enzymes to generate repression at the early stages of T 3 exposure. Thus, all of these results suggest that TR interacts with a potential nTRE sequence at Ϫ82ϳϪ87 bp of the CD44 promoter, weakly in the absence of T 3 but transiently strongly at the early stage of T 3 stimulation. This again confirms that TR binds to CD44 nTRE in vivo in T 3 -dependent negative regulation.
Interestingly, NCoR showed a similar pattern of dynamic interaction, although much weaker than TR binding (Fig. 8D). This suggested that NCoR also interacted weakly with the TR⅐DNA complex.

Identification of Rat CD44 Gene That Is Negatively Responsive to T 3 at a Transcriptional Level-
We successfully isolated a model gene, CD44, to study the general mechanism of T 3 -dependent negative regulation. Using the gene expression profiles with mRNAs from hypothyroid-and hyperthyroid-conditioned pituitary tumor cell line (GC) cells, we identified a group of abundantly expressed genes. Eight were confirmed by quantitative real-time PCR. Further screening showed that four of these were directly regulated at a transcriptional level. In vivo experiments with animal tissues showed that two were generally regulated, and the other two were tissue-specifically regulated. Finally, we selected the CD44 gene as an example of a generally negatively T 3 -responsive gene containing all necessary characteristics for studying the T 3 -dependent negative regulation mechanism. This indicated that our approach constituted a reasonably efficient method to identify directly ligand-responsive genes that are negatively regulated in a general and a tissues-specific manner.
Localization of the Negative T 3 -responsive Fragment-Interestingly, the effect of TR␣1 on T 3 -dependent negative regulation of luciferase expression from the core CD44 fragment was higher or equivalent to that of TR␤ (data not shown). Similar results were also obtained using 1.2-kb hTSH␤-Luc and 0.65-kb rD2-Luc reporter constructs that conferred a negative response to T 3 (data not shown). This indicates that TR␣1 acts as a potent negative regulator of nTRE. This contradicts a previous report that TR␣1 was a more potent activator on pTRE than TR␤ that has a preference for negative regulation by T 3 (45). It is likely that negative regulation by TR␤2 is rather tissue-specific, especially to anterior pituitary and hypothalamus (46). Therefore, we chose TR␣1 to study the general mechanism of T 3 -dependent negative regulation.
After demonstrating that a negative T 3 response and thus an nTRE is located within the 5-kb 5Ј-FR, we performed deletion analysis and identified a minimal fragment for negative regulation by T 3 . We initially obtained a 177-bp DNA fragment (Ϫ129ϳϩ48 from the TSS) that conferred a negative response to T 3 ( Fig. 2A). Further 5Ј deletion to the 113-bp fragment (Ϫ65ϳϩ48 bp) reduced but did not completely eliminate the negative response to T 3 ( Fig. 2A). The transcriptional activity of the 113-bp fragment was largely abolished to just above background levels (Fig. 2B). This suggests that TR still interacted (although weakly) with the 113-bp fragment, which was almost non-functional. We found that this residual response to T 3 was caused by the GAF binding to a site Ϫ1ϳϩ3 bp from the TSS (Fig. 8B). When the GAF binding site of the 113-bp fragment was mutated to be non-functional, the residual T 3 response was completely abolished (Fig. 8C). This suggests that the GAF bound to the 113-bp fragment interacts weakly with TR and is responsible for the residual weak negative response to T 3 . Thus, these results support the notion that a major TR binding site for negative regulation by T 3 is located within a 64-bp (Ϫ129ϳϪ65 bp) fragment.
Weak TR and GAF Binding Sites Are Required for Negative Regulation by T 3 -Initially, we evaluated whether an nTRE could consist of the same sequences as a pTRE, because two high affinity TREs were identified within a Ϫ129ϳϪ65 bp fragment conferring negative response to T 3 . Two half-sites of TRE consensus sequences were located at Ϫ79ϳϪ74 bp (AG-GACA) and Ϫ53ϳϪ48 bp (AGGTCA) with a 20-bp separation. However, mutational analysis indicated that these TREs were not necessary for T 3 -dependent negative regulation (Fig. 4D). This result contradicts published results suggesting the existence of positive TRE-like sequences within negative T 3 -responsive promoters (10, 13, 16 -18, 23, 40, 47-50). For example, hTSH␤ contains a TRE sequence and RSV-nTRE two everted pTRE-like sequences with an 8-bp separation (10). This suggests that TR might not interact with pTRE for T 3 -dependent negative regulation, but rather that these high affinity TREs are involved in the increase of total transcriptional activity of the 177-bp CD44 promoter (Fig. 4D). In fact, EMSA showed very weak TR binding to the other nTRE fragments chosen for study (Fig. 3C), even though the hTSH␤ and RSV promoters contain pTRE sequences in the core negative fragment (9,10).
This result led us to re-examine whether there was any weak or transient TR binding site on the 177-bp CD44-Dbl that was mutated at the high affinity TR binding site. Functional analysis of the non-DNA binding mutant GS71 showed that TR binding was required for T 3 -dependent negative regulation of the 177-bp CD44-Dbl-Luc. Similar results were obtained using 1.2-kb hTSH␤ and 170-bp rD2, which conferred a negative response to T 3 (Fig. 5, D and E). Consistently, EMSA also showed that TR bound weakly to the specific sequence TTT-GGG at Ϫ82ϳϪ87 bp. In the same context, sequences of hTSH␤ and RSV and 170-bp rD2 conferring a negative response to T 3 also showed a weak interaction with TR␣1 (Fig.  3C). This interaction was further supported by ChIP assay of the 177-bp CD44-Dbl fragment (Fig. 9). Thus, it is likely that weak TR⅐DNA interaction is a general characteristic of the nTREs.
However, the weak TR⅐DNA interaction alone was not sufficient to elicit negative regulation by T 3 as shown in Fig. 8B. This suggested that interaction of unknown TR-associated proteins to TR bound to the nTRE was required for the cooperative repression. We identified one of them, GAF, a DNA binding transcription factor. Mutation of both the TR binding sequence TTTGGG and the GAF binding site at Ϫ1ϳϩ3 bp almost completely abolished transcriptional activity, whereas mutation of either one alone had no significant effect. This suggests that there is a putative interaction among nTRE, TR, and GAF to confer negative regulation by T 3 . Interestingly, the same sequence is also present in the core nTRE of hTSH␤, TTT-GGGTCA (Ϫ4ϳϩ5 bp from the TSS). When GGG was mutated to AAA in the nTRE of hTSH␤, EMSA and ChIP assays resulted in a loss of not only TR binding but also functional repression by T 3 (9). This again supports the contention that TTTGGG might be a novel nTRE consensus sequence.
Both the GAGA-binding factor (GAF) and the GAF-responsive element have been extensively investigated, including demonstrations that GAF functions in transcriptional regula-tion and chromatin remodeling (51). Our functional analysis of the CD44 promoter showed that GAF binding was required for negative regulation by T 3 . It is also possible that GAF is generally involved in the T 3 -dependent negative regulation mechanism, not only with the CD44 gene. Sequence data indicated that other potential GAF binding sites were located at Ϫ207ϳϪ204 bp and at ϩ52ϳϩ55 bp of the hTSH␤ gene. It will be interesting to determine whether TR bound to TRE at Ϫ1ϳϩ5 bp of hTSH␤ interacts with GAF at one of these binding sites.
NCoR-mediated Basal Transactivation-Different from the "squelching" model, which proposed that NCoR enhanced unliganded TR-mediated basal transactivation without TR⅐DNA interaction, our results indicate that T 3 -dependent negative regulation requires TR⅐nTRE interaction, even though its interaction was weak in the absence of T 3 (demonstrated by EMSA, ChIP, and transfection assays with GS71). In addition, both GAF and NCoR were required for T 3 -dependent negative regulation, possibly through NCoR, GAF-DNA binding, and TR⅐nTRE interaction. These results support a novel mechanism underlying T 3 -dependent repression.
It would be especially valuable to characterize the dual functions of NCoR, which appear to be dependent on the promoter context. Thus, on the positive TRE, NCoR acts as co-repressor, while on the negative TRE, as a co-activator, possibly by interacting with GAF and TR. When the C-terminal region of NCoR, which contains CoRNR boxes that interact with nuclear receptors for repression, was transiently expressed in COS-7 cells, the effect was the same as that generated by the full-length NCoR (data not shown). This raises the interesting question of whether the C-terminal domain of NCoR also contains TR and co-activator interaction domains for TR-mediated basal transactivation of T 3 -dependently negatively regulated genes.
Alternatively, the varying function of NCoR depending on the promoter context may be explained by its acting as a bridging molecule, rather than as a co-activator or co-repressor. It is possible that the TR⅐nTRE complex has a high affinity to certain domains in the C-terminal NCoR to recruit co-activators such as GAF for enhanced transactivation, whereas TR⅐pTRE complex binds to CoRNR boxes of NCoR to induce repression.
Transient Interaction of TR with DNA during T 3 Exposure-TR⅐DNA interaction on the CD44 promoter was also confirmed in vivo by ChIP assay. In the absence of T 3 , TR weakly but significantly interacted with DNA. Surprisingly, in the presence of T 3 , TR bound to DNA in a pattern of transient interac-tion at early stages of T 3 stimulation. TR⅐DNA interaction peaked 50 -60 min after the addition of T 3 in vivo (Fig. 9D). TR-associated nuclear proteins are probably recruited in vivo onto the nTRE, enhancing TR⅐DNA interaction, with a peak 50 -60 min after T 3 exposure. If true, this suggests that the most important result of TR⅐DNA interaction was to initiate a repressed state in the chromatin structure for repressed function. This again implies that there was transient but strong direct TR⅐DNA interactions in vivo on the CD44 promoter, despite its lack of high affinity TRE binding sites. The generation of peak interaction at such an early stage of T 3 stimulation is a novel phenomenon.
Other interesting questions remain to be answered in broadening our understanding of the T 3 -dependent repression mechanism. What is the basal mechanism underlying T 3 -dependent repression? How are histone molecules modified by acetylation, methylation, and phosphorylation? Which other chromatin remodeling proteins, such as FACTs (facilitates chromatin transcription) and NURFs (the ATP-dependent nucleosome remodeling factors), are recruited?
The working model we propose is depicted in Fig. 10. In the absence of T 3 , TR interacts weakly with nTRE (Fig. 10, part a). Without NCoR or GAF, TR⅐nTRE interaction allows basal expression. When NCoR and GAF are present, GAF binds to its recognition site at Ϫ1ϳϩ3 bp and subsequently forms the TR⅐NCoR⅐GAF complex, containing the DNA loop structure that enhances transactivation (part b). At this stage, basal transactivation is further enhanced through the TR⅐nTRE interaction (part b). Addition of T 3 causes repression through a transient TR⅐DNA interaction (parts c and d). After 50 -60 min of T 3 stimulation, TR⅐nTRE interaction is maximized, possibly recruiting chromatin remodeling proteins such as NCoR/ SMRT, histone deacetylases, and NURFs (part c). As time goes on, chromatin structure is remodeled to its repressed state, and TR dissociates from DNA (part d). This state is maintained until the T 3 level is reduced (part e). When the T 3 level falls to hypothyroid concentrations, TR⅐nTRE interaction resumes, supporting basal transactivation (part a).
Summary-We have identified a new model gene, CD44, with which to study negative regulation by T 3 . Investigation of the CD44 promoter revealed several novel aspects of the T 3 -dependent negative regulation mechanism: 1) the core sequence TTTGGG of the nTRE was different from that in pTREs; 2) the nTRE interacted directly, but weakly, with TR in vitro; 3) TR-mediated basal transactivation is enhanced by NCoR during TR⅐DNA interaction; 4) both TR⅐DNA and GAF⅐DNA FIG. 10. A proposed model for NCoR-mediated transactivation and repression by T 3 . In the absence of T 3 , nTRE at Ϫ82ϳϪ87 bp is exposed to TR for basal expression of CD44 mRNA. Increased expression of NCoR and/or GAF enhances TR-mediated basal transactivation. T 3 induces remodeling of chromatin structure dynamically by repeated TR⅐DNA interactions at the early stage for 60 min; repressed state is maintained until T 3 is lowered.
binding are required for negative regulation by T 3 ; 5) T 3 induces transient TR⅐DNA interaction in vivo, resulting in a maximal interaction during 50 -60 min of T 3 stimulation; and 6) NCoR is recruited to DNA, presumably during the TR⅐NCoR interaction. This is the first demonstration of a T 3 -dependent negative regulation mechanism involving NCoR, GAF, and TR on the novel nTRE of the CD44 gene.