Quantitative proteomics of the thyroid hormone receptor-coregulator interactions.

The thyroid hormone receptor regulates a diverse set of genes that control processes from embryonic development to adult homeostasis. Upon binding of thyroid hormone, the thyroid receptor releases corepressor proteins and undergoes a conformational change that allows for the interaction of coactivating proteins necessary for gene transcription. This interaction is mediated by a conserved motif, termed the NR box, found in many coregulators. Recent work has demonstrated that differentially assembled coregulator complexes can elicit specific biological responses. However, the mechanism for the selective assembly of these coregulator complexes has yet to be elucidated. To further understand the principles underlying thyroid receptor-coregulator selectivity, we designed a high-throughput in vitro binding assay to measure the equilibrium affinity of thyroid receptor to a library of potential coregulators in the presence of different ligands including the endogenous thyroid hormone T3, synthetic thyroid receptor beta-selective agonist GC-1, and antagonist NH-3. Using this homogenous method several coregulator NR boxes capable of associating with thyroid receptor at physiologically relevant concentrations were identified including ones found in traditional coactivating proteins such as SRC1, SRC2, TRAP220, TRBP, p300, and ARA70; and those in coregulators known to repress gene activation including RIP140 and DAX-1. In addition, it was discovered that the thyroid receptor-coregulator binding patterns vary with ligand and that this differential binding can be used to predict biological responses. Finally, it is demonstrated that this is a general method that can be applied to other nuclear receptors and can be used to establish rules for nuclear receptor-coregulator selectivity.

Thyroid hormone (3,5,3Ј-triido-L-thyronine, T3) 1 regulates multiple physiologic processes in development, growth, and metabolism (1,2). Most T3 actions are mediated through the thyroid hormone receptors (TR), which regulate transcription of target genes either positively or negatively in response to hormone binding. There are two different genes that express different TR subtypes, TR␣ and TR␤. Each transcript can be alternatively spliced generating different isoforms (TR␣ 1 , TR␣ 2 , TR␤ 1 , TR␤ 2 ,) (3,4). While most of these isoforms are widely expressed, there are distinct patterns of expression that vary with tissue and developmental stage. In particular, TR␤ 2 is found almost exclusively in the hypothalamus, anterior pituitary, and developing ear. In addition, mice deficient in either TR␣ or TR␤ display unique phenotypes, suggesting that the different TR isoforms have unique regulatory roles (5)(6)(7)(8)(9)(10).
The thyroid hormone receptors belong to a superfamily of proteins known as the nuclear hormone receptors (NR). Like other members of the NR superfamily, TR has three functional domains: an N-terminal transactivation domain (NT), a central DNA binding domain (DBD), and a C-terminal ligand binding domain (LBD) (11). The DBD of TR recognizes short, repeated sequences of DNA found in T3-responsive genes, termed the thyroid hormone response elements (TREs). TR can bind to a TRE as a monomer, homodimer, or heterodimer with the retinoid X receptor (RXR) (12). However, receptor activation from the heterodimer complex is the best characterized to date. Both liganded and unliganded TR bind to TREs. In the absence of T3, TR associates with corepressor proteins at the TRE maintaining the chromatin in a compact state and therefore repressing gene activation. Upon binding of T3, TR undergoes a conformational change releasing corepressor proteins and allowing for the interaction with coactivator proteins that enhance TRE-driven gene transcription.
There is a large pool of coregulators available for interaction with TR. Although there appears to be some functional redundancy within the SRCs, there is also evidence that SRCs have distinct biological regulatory roles. While mice deficient in SRC1 exhibit resistance to thyroid hormone (RTH), the phenotypes for mice deficient in SRC2 and SRC3 are distinct with no evidence of RTH (30 -32). These studies, along with recent work with the progesterone and glucocorticoid receptors, have demonstrated that interaction with specific coregulators can elicit specific biological responses (33). However, it remains unclear how NRs discriminate between different coregulators. In this study we sought to define rules that govern TR-coregulator selectivity.
Combinatorial peptide libraries have been used to define NR-coregulator specificity, and have revealed that the sequences immediately flanking the NR box are critical for specificity (34,35). However the peptides in these studies were generated from random libraries that do not represent the true NR box sequences. Other investigations focused on defining SRC NR box selectivity using a subset of coregulator NR boxes from the SRCs. This work has shown that ligands can allosterically modulate the coregulator binding pocket and therefore differentially alter specific SRC recruitment and NR box usage (36 -38). However, to date there has been no comprehensive study of the interactions of TR and natural coregulator NR boxes. To address this issue, we designed an in vitro binding assay to measure the equilibrium binding of TR␤ to a library of potential coregulators in a high-throughput manner using fluorescence polarization. With this method, binding constants for TR␤ to coregulator NR boxes were determined in a consistent format, including NR boxes from SRCs and nine other known coregulators. In addition the TR␤-coregulator binding patterns for three different ligands including T3, the synthetic TR␤selective agonist GC-1, and the T3 antagonist NH-3 were defined. This quantitative information can be used to establish rules for TR␤ coregulator selectivity, and these rules can be used for predicting biological responses.
Direct Binding Assay-Using a BiomekFX in the Center for Advanced Technology (CAT), hTR␤-LBD was serially diluted from 70 -0.002 M in binding buffer (50 mM sodium phosphate, 150 mM NaCl, pH 7.2, 1 mM dithiothreitol, 1 mM EDTA, 0.01% Nonidet P-40, 10% glycerol) containing 140 M ligand (T3, GC-1, or NH-3) in 96-well plates. Then 10 l of diluted protein was added to 10 l of fluorescent coregulator peptide (20 nM) in 384-well plates yielding final protein concentrations of 35-0.001 M and 10 nM fluorescent peptide concentration. The samples were allowed to equilibrate for 30 min. Binding was then measured using fluorescence polarization (excitation 485 nm, emission 530 nm) on an Analyst AD (Molecular Devices). Two independent experiments were assayed for each state in quadruplicate. Data were analyzed using SigmaPlot 8.0 (SPSS, Chicago, Il), and the K d values were obtained by fitting data to the following equation (y ϭ min ϩ (max Ϫ min)/1 ϩ (x/K d ) Hill slope).

RESULTS
Coregulator Peptide Library-A library of known coregulator peptides consisting of the LXXLL sequence plus 7-8 additional flanking residues at each terminus was synthesized (Fig. 1B). Previous screens with coactivator peptides established amino acid residues at ϩ6 to ϩ12 as critical for binding (38,42). To capture these specificity determinants, peptides of 20-amino acid length were generated. Additionally, negative control peptides were made for each coregulator NR box by replacing Lϩ4 and Lϩ5 with alanine (LXXAA), as this substitution has been shown to abolish interactions with NR (38). A thiol reactive fluorescent probe was covalently attached to each peptide via a cysteine positioned at the N terminus of each peptide. The coregulator peptides are listed in the far left column of Fig. 1B, where SRC1-1, SRC1-2, SRC1-3 represent the first, second, and third NR boxes in SRC1, respectively. This nomenclature, first proposed by O'Malley, is applied to all of the coregulator peptides studied throughout this report (30). All peptide probes were synthesized in parallel using the Fmoc strategy and purified by RP-HPLC. Identity and purity were confirmed using HPLC and MALDI-TOF or LCMS (Supplemental Information). Seven targeted coregulator NR boxes, TRAP100-1, TRAP100-5, RIP140-2, RIP140-4, TRAP220-1(Ϫ), TRBP-1(Ϫ), and TRAP100-6(Ϫ), are omitted as they could not be synthesized or purified after several attempts.

Coregulator Peptides Bind to hTR␤-LBD in Four Different
Binding Modes-Initial peptide binding studies were carried out with SRC2-2, and the ligand binding domain of the human thyroid hormone receptor ␤ (hTR␤-LBD) in the presence of T3. As shown in Fig. 2A, SRC2-2 binds to TR␤ in a saturable dose-dependent manner with a measured K d of 0.7 M, consistent with literature reports (0.8 M) (15). We also confirmed that this interaction was specific by carrying out binding studies with a mutated SRC2-2 peptide (LXXAA, SRC2-2(Ϫ)). The trace observed in Fig. 2A reveals that SRC2-2(Ϫ) does not interact with TR␤ in the presence of T3.
To further elucidate the coregulator binding pattern of TR␤ in the presence of T3, direct binding assays were carried out with the entire coregulator peptide library. This set of experiments was executed by maintaining a constant concentration of coregulator peptide (10 nM) and varying the TR␤ concentration from 0.001-35 M in the presence of saturating amounts of T3 (90 M). The results revealed that the 34 different coregulator peptides bound to liganded TR with varying degrees of affinity. Individual Klotz plots were constructed for each coregulator peptide. This analysis revealed four different binding modes. Example equilibrium affinity curves are summarized in Fig. 2 (A-D) and K d value ranges are reported in Fig. 3A. In no case did the negative peptide controls bind to liganded TR␤.
The next binding mode included peptides where binding FIG. 1. Coregulator peptides. A, block diagram of SRC1 identifying some of the functional domains. The nuclear interaction domain (NID) contains three nuclear receptor interaction domains (NR boxes) that are known to interact with NR (SRC1-1, SRC1-2, SRC1-3). The activation domains AD-1 and AD-2 interact with other known coregulators including p300/CBP and CARM-1. SRC1 has two isoforms, SRC1a and SRC1e. SRC1-a has an additional NR box at the C terminus designated SRC1-4 that has been shown to interact with some NR. B, sequence alignment of the coregulator peptides tested in this study. The conserved NR box, LXXLL, is enclosed in a box. Negative control peptides were also synthesized with leucine ϩ4 and ϩ5 replaced with alanines (LXXAA). * denotes omitted peptide sequences that were not recovered after synthesis. appeared to be reaching saturation but did not have a clear plateau, as defined by at least two points with indistinguishable y-ordinates (Fig. 2B). This assumption is based on previous binding studies conducted with these coregulator peptides where similar changes in polarization values were observed for saturating binding isotherms (data not shown). These peptides bound to TR␤ with a K d range of 10 -30 M and included SRC1-2, SRC1-3, SRC3-2, TRAP220-2, TRBP-2, and ARA70. To accurately obtain K d values, however, the binding studies would need to be carried out at protein concentrations varying from 1-300 M as this would give the widest range of polarization values. Working with TR␤ protein concentrations higher than 100 M is problematic due to protein aggregation and decreased protein stability. In order to reflect the inability to obtain an unambiguous K d value, we report a K d range,10 -30 M, for coregulator peptides that exhibit this binding mode.
The third binding mode is one in which polarization increases with protein concentration but does not seem to be reaching saturation. This mode is exemplified by SRC3-3 (Fig.  2C) where the polarization of SRC3-3 slowly increases with TR␤ concentration. Other coregulators included in this category are SRC1-1, SRC3-1, SRC3-3, RIP140-3, RIP140-8, and SHP. The binding isotherms for this group suggest that the coregulator peptides are binding non-specifically or that they bind with a K d value significantly above the working assay range, (e.g. Ͼ30 M). This class of coregulator peptides is distinguished from the final group of peptides where no binding is observed (Fig. 2D).
T3 Recruitment of Coregulators-We report here the first comprehensive investigation of coregulator recruitment to liganded TR␤ with 12 different coregulators and 32 unique NR boxes as well as appropriate negative controls (LXXAA). Of the 32 coregulator peptides tested, 20 appear to interact with TR␤ in the presence of T3 with varying degrees of affinity. The strongest recruitment observed was with SRC2-2 which exhib- The coregulator peptides that clearly did not interact with T3 liganded TR␤ included ARA55, all of the TRAP100 peptides, and some of the RIP140 and DAX1 NR box peptides. The remaining coregulator peptides bound to TR␤ weakly with K d values ranging from 10 to 30 M or Ͼ 30 M as discussed above.
The SRCs are a family of coregulators whose interaction with TR␤ have been extensively studied using non-quantitative methods (15,38,43,44). Our data are consistent with previously published work where it was determined that TR␤ has a strong preference for SRC2 NR box peptides with the overall observed preference being SRC2-2 Ͼ SRC2-3 Ͼ SRC2-1 (15). The next family member, SRC1, bound to liganded TR␤ to a lesser extent with SRC1-2 ϳ SRC1-3 Ͼ SRC1-1. Only weak interactions were observed for SRC3 where affinity has been reported as SRC3-2 Ͼ SRC3-1 ϭ SRC3-3.
The TRAP coactivator complex has been shown to associate with NRs and help initiate transcription (19,20). Two single subunits TRAP220 and TRAP100 were investigated for their ability to interact with liganded TR␤. TRAP220 has two NR boxes and in this report it was determined that the first NR box, TRAP220-1, interacted more strongly than the second NR box, TRAP220-2. This is the opposite of what has been reported for TR␣, supporting the notion that TR isoforms can differentially recruit coactivator NR boxes (20,45). The TRAP100 protein contains 7 NR boxes, 5 of which were studied here. As previously reported, none of these NR boxes interact with liganded TR␤ (19).
Another coregulator that has been shown to interact with TR␤ is TRBP. This coactivator is ubiquitously expressed and appears to be a general coactivator that can associate with NR including TR, ER, PPAR, as well as other transcriptional pro- teins such as AP-1, CRE, and NFB-response element (25). There are two LXXLL motifs in TRBP and both can interact with TR␤. Our studies indicate that TRBP-1 is preferentially recruited to TR␤ in the presence of T3.
RIP140 is a coregulator that contains 9 LXXLL motifs and has been shown to interact with many NRs including TR␤ (23). It has been suggested that RIP140 directly competes with other coregulators (23). Unlike traditional coregulators, however, RIP140 represses transcription upon binding to NR (46 -49).
Here we show liganded TR␤ has a clear preference for three of the NR boxes in RIP140, RIP140-3, RIP140-5, and RIP140-8. One NR box peptide in particular, RIP140-5, bound fairly tightly with a K d of 2.5 M Ϯ 0.4.
Three additional coactivators, p300, ARA70, and DAX1, were also shown in this report to associate with TR␤ with varying degrees. ARA70 and DAX1 had not previously been investigated for their interaction with TR␤.
Ligands Alter Coregulator Recruitment-To investigate ligand effects on coregulator recruitment, binding studies were performed in the presence of the TR␤ selective agonist, GC-1. GC-1 is a halogen-free thyromimetic that is ϳ10-fold selective for binding to TR␤ versus TR␣ (39). It has been shown that the oxyacetic acid at the carbon-1 position (Fig. 3A) is responsible for the selective TR␤ binding of GC-1 (50). Additionally, crys-tallographic studies have confirmed that the oxyacetic acid group participates in a hydrogen bonding network in the TR␤ LBD polar pocket (51). We sought to determine how these interactions might alter coregulator specificity.
In the presence of GC-1 the coregulator peptides bound to TR␤ with varying degrees of affinity and all four binding modes were observed. Overall the coregulator binding patterns for GC-1 and T3 were similar in terms of which coregulator peptides were recruited. However, the degree to which they bound varied. In most cases, the coregulator peptides bound with similar or slightly lower affinity to TR␤⅐GC-1 than to TR␤⅐T3. Several NR boxes, particularly those of the SRC family, exhibited significant differences in affinity to TR␤⅐GC-1 relative to TR␤⅐T3. All of the SRC2 NR boxes bound TR␤⅐T3 in a measurable K d range, whereas in the presence of GC-1 a saturated binding curve was only observed for SRC2-2. Additionally, SRC1-2 appears to be much more strongly recruited by TR␤⅐T3, whereas the opposite is true for SRC3-3. Other notable differences between T3 and GC-1 were seen with TRAP220 and TRBP where recruitment decreased in the presence of GC-1.
In addition to studying agonist induced coregulator recruitment, we wanted to explore how antagonists may affect coregulator binding. The recently reported T3 antagonist NH-3 (Fig.  3A) was tested against the entire coregulator peptide library   FIG. 3. Coregulator binding patterns and specificity determinants. A, the structure of the TR␤ ligands tested; endogenous thyroid hormone T3, synthetic TR␤-selective agonist GC-1, and T3 antagonist NH-3. The equilibrium binding constants for TR␤-coregulator NR boxes are reported for each ligand. The coregulator peptides are listed in the left column where SRC1-1, SRC1-2, SRC1-3, SRC1-4 represent the first, second, third, and fourth NR box in SRC1, respectively. Each color represents a unique K d range as defined by the legend on the left. Significant differences between TR␤⅐T3 and TR␤⅐GC-1 are boxed in black. B, representative coregulator peptide is listed for TR␤⅐T3, TR␤⅐GC-1, ER␣⅐E2 with amino acids highlighted in green representing amino acids that convey specificity for each NR⅐ligand state. The information for ER␣⅐E2 was extracted from two sources for comparative purposes: 1) a time-resolved fluorescence assay conducted with SRCs (36) and 2) a phage peptide library screen with ER␣⅐E2 (35). ⌽ denotes hydrophobic amino acids, and represents hydrophilic amino acids. (40). In the presence of saturating concentrations of NH-3, no coregulators from the library were recruited to the TR␤⅐NH-3 complex (Fig. 3A). DISCUSSION The NRs show a clear preference for particular NR boxes. One factor that drives this specificity is the amino acid residues immediately flanking the NR box (15, 34, 36 -38). Most prior work focused on individual coregulators or the p160/SRC coregulator family (15,19,23,25,38,(43)(44)(45)52). To expand our understanding of coregulator recruitment, we investigated the ability of TR␤ to bind to a library of known coregulator NR boxes using a homogenous equilibrium binding assay. The results from this screen demonstrate that the coregulator binding pattern for TR␤ is distinct from other NR (ER␣, ER␤, AR, data not shown) and new TR␤-coregulator peptide interactions, including RIP140-5 (K d ϭ 2.5 M), ARA70 (K d ϭ 10 -30 M), and DAX1-3 (K d ϭ 3.6 M) were identified. The NR box peptides that interacted with TR␤ revealed specificity elements including a propensity for hydrophobic groups at the Ϫ1 position and for proline at the Ϫ2 position as seen in TRAP220 and TRBP-1 (20,45). Additional trends included histidine at Ϫ3, glutamine at ϩ6, and the presence of a serine in the C-terminal positions ϩ7 to ϩ10 (Fig. 3B). Previous phage display studies isolated non-natural peptides with similar specificity determinants (35).
Ligands are another factor that can impact the recruitment of coregulators to NR (36 -38,53). In this study, the ability of both agonists and antagonists to modulate the coactivator binding pocket of TR␤ was investigated. As predicted, the NR box binding patterns for TR␤⅐T3 and TR␤⅐GC-1 were different, and no coregulators were recruited in the presence of the T3 antagonist NH-3. Based on the differential NR box recruitment observed for the GC-1 ligand, specificity determinants that are distinct from T3 can be defined. While there is still a high propensity for hydrophobic amino acids at the Ϫ1 position and for serine at positions ϩ7 to ϩ10, proline at Ϫ2 and histidine at Ϫ3 do not seem to be important for specificity in the presence of GC-1 (Fig. 3B). In addition, the specificity determinants for TR␤⅐T3 and TR␤⅐GC-1 are distinct from those observed for the estrogen receptor with its cognate ligand (ER␣⅐E2) (Chang et al. Ref. 35).
The differential binding patterns for TR␤⅐T3, TR␤⅐GC-1, and TR␤⅐NH-3 may be used to predict biological responses. In hypothyroid and euthyroid hypercholesterolemic mice, GC-1 behaves like T3 to potently reduce serum cholesterol (54). Additionally, while T3 potently induces positive inotropic and chronotropic cardiac effects, GC-1 is devoid of significant cardiac effects through a wide dose range. Although these observations may be partially explained by the selective binding of GC-1 to TR␤ as well as preferential liver versus heart uptake (54), coregulator selectivity may also play a role. In the presence of T3 there is a stronger preference for the recruitment of SRC1 and SRC2 coregulator peptides to TR␤, whereas SRC2 and SRC3 NR boxes are more strongly recruited to TR␤⅐GC-1. Recent investigations focusing on the SRC1 role in regulating T3-responsive genes have revealed that SRC1 is important for the pituitary-hypothalamus-thyroid axis and for T3 affects in the heart but not for regulation of hepatic genes that regulate cholesterol levels (55)(56)(57). This suggests that SRC2 and SRC3 regulate cholesterol-modulating genes in the liver. In support of this argument, studies have shown that SRC2 and SRC3 liver expression is increased in hypothyroid mice while there is a slight decrease in SRC1 expression (58).
Studies utilizing cDNA microarrays have revealed that thyroid hormone can both positively and negatively regulate genes (59). One mechanism of T3-driven repression may involve the recruitment of coregulators that repress transcription, such as RIP140 and DAX1. In our studies we find that both TR␤⅐T3 and TR␤⅐GC-1 strongly interact with RIP140-5 and DAX1-3, but TR␤⅐NH-3 fails to recruit these coregulators. From these observations, it can be predicted that T3 and GC-1 can repress gene transcription but NH-3 lacks this ability. Thus NH-3 treatment may result in partial activation of genes that are normally repressed by TR␤⅐T3. If this is the case, then NH-3 would display unique pharmacology by blocking ligand activation of positively regulated T3-responsive genes and causing derepression of negatively regulated T3-responsive genes.
Presumably there are additional factors that influence NR recruitment of coregulators such as post-translational modifications, structural determinants arising from specific DNA response elements, cooperativity, cellular environment, and additional interaction surfaces on the NR and coregulator proteins. To fully dissect NR-coregulator interactions, more complex models will need to be developed. The use of full-length molecules for determining NR-coregulator binding affinities has been employed with the estrogen receptors and members of the SRC family (60). Although this work demonstrated that the binding affinities are 3-5 fold higher than predicted with coregulator peptides and NR-LBD, the overall selectivity of ER isoforms for SRC members was consistent with previous investigations. This emphasizes the utility of a simple affinity model as a first step for establishing rules of NR-coregulator selectivity.
NR signaling is a multivariant complex process that utilizes differences in NR, NR isoforms, a diverse set of coregulators, ligands, tissue variability, and unique DNA response elements. It remains unclear how this protein network can potentiate signals for specific biological responses. However one point of regulation may derive from specific NR-coregulator interactions. Using an equilibrium binding assay, the binding affinities of TR␤ for a large set of NR boxes in the presence of multiple ligands were quantitatively determined and some rules were defined that account for the specificity of these interactions. Additionally, it was shown that these binding patterns could be used to predict biological responses. Finally, we believe that this method may be generalized to other nuclear receptors to establish patterns of NR-coregulator selectivity.