Nuclear Receptor-Coregulator Interaction Profiling Identifies TRIP3 as a Novel Peroxisome Proliferator-activated Receptor γ Cofactor*

Nuclear receptors (NRs) are major targets for drug discovery and have key roles in development and homeostasis as well as in many diseases such as obesity, diabetes, and cancer. NRs are ligand-dependent transcription factors that need to work in concert with so-called transcriptional coregulators, including corepressors and coactivators, to regulate transcription. Upon ligand binding, NRs undergo a conformational change, which alters their binding preference for coregulators. Short α-helical sequences in the coregulator proteins, LXXLL (in coactivators) or LXXXIXXXL (in corepressors), are essential for the NR-coregulator interactions. However, little is known on how specificity is dictated. To obtain a comprehensive overview of NR-coregulator interactions, we used a microarray approach based on interactions between NRs and peptides derived from known coregulators. Using the peroxisome proliferator-activated receptor γ (PPARγ) as a model NR, we were able to generate ligand-specific interaction profiles (agonist rosiglitazone versus antagonist GW9662 versus selective PPARγ modulator telmisartan) and characterize NR mutants and isotypes (PPARα, -β/δ, and -γ). Importantly, based on the NR-coregulator interaction profile, we were able to identify TRIP3 as a novel regulator of PPARγ-mediated adipocyte differentiation. These findings indicate that NR-coregulator interaction profiling may be a useful tool for drug development and biological discovery.

Nuclear receptors (NRs) 1 are ligand-inducible transcription factors involved in development and homeostasis that play key roles in many diseases, including diabetes, cancer, and obesity (1). NRs consist of several functional domains, which exhibit varying degrees of conservation among members of the receptor family. The poorly conserved N terminus contains the activation function 1 (AF-1) domain, the activity of which is often regulated by post-translational modifications. Centrally located is the DNA binding domain, which is highly conserved among species and between nuclear receptors. The ligand binding domain (LBD), which is also relatively well conserved in terms of primary amino acid sequence, mediates ligand binding, and contains the powerful ligand-dependent activation function (AF-2). LBD crystal structures have revealed a canonical fold consisting of 13 ␣-helices and a small four-stranded ␤-sheet (2). Upon ligand binding, the AF-2 helix (also referred to as helix 12) is stabilized in an active state (3). Depending on the conformation of the LBD and its modulation by ligand, NRs can recruit or release transcriptional coregulator proteins that perform all of the subsequent reactions needed to induce or repress transcription of target genes (4). Coregulators are often components of large multiprotein complexes that act in a sequential and/or combinatorial fashion to modify chromatin and to recruit basal transcription factors and RNA polymerase II (5). In general, the transcriptional coregulator family consists of coactivators, which associate with active, liganded receptors and corepressors, which interact with inactive, unliganded (or antagonistbound) receptors. Short peptide motifs within coactivator and corepressor proteins are responsible for their overlapping but non-identical binding to the LBD surface. LXXLL motifs (where L is leucine and X is any amino acid) are found in many coactivator proteins (6), whereas LXXXIXXXL motifs (where I is isoleucine) are present in most corepressor molecules (7). Both motifs probably form an amphipathic ␣-helix upon binding to the hydrophobic cleft on the surface of the LBD (8).
The NR superfamily includes the closely related peroxisome proliferator-activated receptors (PPARs) ␣, ␤/␦, and ␥. Although all three PPARs participate in lipid and glucose me-tabolism, the three isotypes exhibit different physiological roles due to (i) distinct expression patterns, (ii) specific activation by different ligands, and (iii) intrinsic functional differences between the different receptor proteins (9 -11). PPAR␥ is the key regulator of adipocyte differentiation, maintenance, and function (12,13) as exemplified by human familial partial lipodystrophy type 3 (FPLD3) patients. FPLD3 patients harbor heterozygous mutations in the PPARG gene and are characterized by aberrant fat distribution and metabolic disturbances, including insulin resistance and dyslipidemia (11). Synthetic PPAR␥ agonists include the thiazolidinediones (TZDs), which ameliorate insulin resistance and lower blood glucose levels in patients with type 2 diabetes. Treatment of diabetic patients with synthetic PPAR␥ ligands of the TZD class, however, has been linked to adverse side effects like undesired weight gain, fluid retention, peripheral edema, and potential increased risk of cardiac failure (14,15). These adverse side effects may be due to the use of high doses of full PPAR␥ agonists, suggesting that "activation in moderation" may be a more sensible approach (16). This may be achieved through the use of compounds displaying partial agonism, so-called selective PPAR␥ modulators (SPPAR␥Ms) (17).
Given the broad range of diseases in which NR-based drugs are currently being used (18 -20), high throughput profiling of interactions between a given NR and coregulator-derived peptides could be a very useful tool in drug development. Here we used peptide microarrays, as frequently used in enzymatic studies on kinases and proteases (21), to generate NR-coregulator interaction profiles. We generated ligand-specific interaction profiles (agonist, antagonist, or SPPAR␥M) and characterized NR mutants and isotypes (PPAR␣, -␤/␦, and -␥). In addition, we identified a novel biologically relevant interaction between PPAR␥ and thyroid hormone receptor-interacting protein 3 (TRIP3). These findings indicate that NR-coregulator interaction profiling may be a useful tool for drug development and biological discovery.

EXPERIMENTAL PROCEDURES
NR-Coregulator Interaction Profiling-Assay mixtures were prepared on ice in a master 96-well plate with a 5 nM concentration of either commercial preparations of GST-PPAR␣-LBD, GST-PPAR␤/ ␦-LBD, and GST-PPAR␥-LBD (PV4692, PV4694, and PV4546; Invitrogen) or equivalent amounts of purified PPAR-␥ LBD-GST wild type and mutants (see below), time-resolved fluorescence resonance energy transfer coregulator buffers J (for PPAR␣ and -␤/␦) and F (for PPAR␥) (Invitrogen), 25 nM Alexa Fluor 488-conjugated anti-GST antibody (A11131, Invitrogen), 5 mM DTT, 2% DMSO, and ligand at the indicated concentration. All assays were performed in a PamStationா-96 controlled by EvolveHT software (PamGene International B. V., 's-Hertogenbosch, The Netherlands) at 20°C at a rate of 2 cycles/min. Nuclear Receptor PamChipா Arrays (PamGene International B.V.) contained 48 peptides (first generation array used in Fig. 1) or 53 peptides (second generation array used in Figs. [2][3][4][5]. The arrays are made of a porous metal oxide carrier to which the peptides are spotted by means of piezo technology. Each spot has a diameter of 100 m, and because of the porous structure the surface area is ϳ500 times larger than that calculated based on spot diameter. A spot contains ϳ106 pores, each with a diameter of 0.2 m and a length of 60 m. In addition, pores are branched and interconnected. Arrays were incubated for 20 pump cycles with 25 l of blocking buffer (1% BSA, 0.01% Tween 20 in Tris-buffered saline) and then aspirated. Per array, 25 l of assay mixture was transferred from the master plate to the chip using a multichannel pipette. During the period of ligand incubation (ϳ40 min), a solution of GST-PPAR␥-LBD, fluorescent anti-GST antibody, and ligand was pumped through the porous peptide-containing membrane for 81 cycles at a rate of 2 cycles/min. Assay mixtures were incubated in the arrays for 80 cycles, and a .tiff format image of every array was obtained at cycles 21, 41, 61, and 81 by a charge-coupled device camera-based optical system integrated in the Pam-Station-96 instrument.
Peptide Microarray Data Analysis-Image analysis, consisting of automated spot finding and quantitation, followed by calculation of binding velocities was performed by Bionavigator software (PamGene International B.V.). In short, the boundaries of a spot were determined, and the median signal within the spot (signal) as well as that in a defined area surrounding it (background) were quantified. The signal-minus-background values were subsequently used for the calculation of the binding velocity. For each array and for each individual coregulator peptide, a binding curve of the NR to that coregulator motif was constructed from the binding at five consecutive time points using y ϭ y 0 ϩ y span ϫ (1 Ϫ e (Ϫk ϫ (x Ϫ xoffset)) ) (an exponential rise from level y 0 (at x ϭ x offset ) to y 0 ϩ y span (at x 3 ϱ) with rate constant k (k Ͼ 0)). The derived slope (binding velocity) of the fitted curve (cutoff, R 2 Ͼ 0.7) at time point 31 was used as the parameter for interpretation. Ligand dose responses were analyzed with Prism for Windows 4.02 (GraphPad Software Inc., San Diego, CA) using nonlinear regression, sigmoidal dose response (variable slope). Fitted dose-response curves delivered values for potency (half-maximal effective concentration (EC 50 )) and induction (difference in signal between the bottom and top value of the curve). EC 50 values were presented only when a curve met the criterion of R 2 Ͼ 0.5. Induction values were presented only when the 95% confidence intervals of the bottom levels of the curve did not overlap with the 95% confidence interval of the top of the curve.
NR-Coregulator Peptide Affinity Determination-On a single array, a concentration range of the coregulator peptide was immobilized. K d values were determined from the binding of the NR at different peptide densities using non-linear regression with Prism for Windows 4.02 (sigmoidal dose response, variable slope, constraint, shared top).
Proteins were concentrated using Vivaspin centrifugal concentrators (Sartorius, Epsom, UK), and protein concentrations were determined using SDS-PAGE followed by Coomassie Brilliant Blue staining.
GST Pulldown Assays-The full-length coding sequences of PPAR␤/␦, PPAR␥, TRIP3, and TRIP3 m in the pcDNA3 or pcDNA3.1 expression vector were transcribed and translated in vitro in reticulocyte lysate in the presence of [ 35 S]methionine (Amersham Biosciences) according to the manufacturer's protocol (TNT T7 Coupled Transcription/ Translation kit, Promega, Madison, WI). 35 S-Labeled proteins were incubated with GST fusion proteins in NETN buffer (20 mM Tris, pH 8.0, 100 mM NaCl, 1 mM EDTA, 0.5% Nonidet P-40) containing protease inhibitors (Complete, Roche Applied Bioscience). Samples were subsequently washed and subjected to SDS-PAGE. Gels were incubated with Amplify (GE Healthcare) to enhance detection efficiency of 35 S-labeled proteins.

PPAR␥ (Ant)agonists Induce Specific Coregulator
Interaction Profiles-The molecular determinants dictating NR-coregulator interactions are still largely unknown. To generate NR-coregulator interaction profiles, we used PamChip-96 peptide microarrays containing peptides of ϳ25 amino acid residues derived from known NR coregulator proteins (Table I) harboring either LXXLL (in coactivators) or LXXXIXXXL (in corepressors) motifs (see supplemental Table 1). The peptides are immobilized on a porous metal oxide carrier by piezo technology (Fig. 1a) as previously used in kinase assays (28 -30). The peptide microarrays were incubated with NR LBDs by pumping the sample up and down through the threedimensional metal oxide carrier in the absence or presence of ligand ( Fig. 1, b and c). During each experiment images were collected at different incubation cycles using a PamStation-96 station, and a time-dependent increase of binding was clearly visible (Fig. 1c, lower panel). The spot signal intensity was plotted against cycle number (Fig. 1c, upper panel), and as a measure for binding the slope of the curve at cycle 31 (V ini ) was used (Fig. 1c, upper graph). The use of V ini values enhances the reliability of the interaction data, because they are calculated with more data points at cycles where binding has not reached saturation, resulting in a broader dynamic range (supplemental Fig. 1). To analyze the kinetics of the interactions, a peptide array was generated containing different amounts of TRAP220 LXXLL604 2 peptide (Fig. 1d, SYPRO Ruby staining of a 2-fold dilution series). Dissociation constants (K d ) for the unliganded (DMSO) and liganded (rosiglitazone) form of GST-PPAR␥-LBD were determined (Fig. 1d, upper panel) as was the ratio of V ini of liganded and unliganded GST-PPAR␥-LBD (modulation index (MI); Fig. 1d, lower panel) at each peptide concentration (see supplemental Fig. 2 for further explanation of K d and MI). Curves were obtained with high quality fits (global R 2 Ͼ 0.994) with K d values of 3.8 (log M, no ligand) and 3.2 (log M, rosiglitazone). The ratio of these K d values represents the drug efficacy for this interaction (see supplemental Fig. 2), which in this case means that a saturating concentration of rosiglitazone induces a GST-PPAR␥-LBD conformation that results in a ϳ4-fold (10 (3.8 -3.2) -fold) increased affinity for the TRAP220 LXXLL604 motif. These findings illustrate that at peptide concentrations higher than 0.25 mM clearly detectable interactions with NR proteins (V ini ) as well as robust ligand effects (MI) could be observed. Therefore all subsequent assays were performed with peptide concentrations of 1 mM.
To investigate whether this array allows interaction profiling of agonist-and antagonist-occupied NRs, we incubated the GST-tagged LBD of PPAR␥ with rosiglitazone (31) or GW9662 (32), respectively. In both cases binding to coregulator-derived peptides was measured using the PamChip-96 and Pamstation-96. Importantly the PPAR␥-coregulator peptide interactions observed upon agonist treatment confirmed previous protein-protein interaction studies (Table I): rosiglitazone specifically induced binding to the LXXLL motifs derived from the coactivators CBP, p300, SRC1, TIF2, SRC3, PRIP, IKBB, DAX1, SHP, RIP140, and TRAP220, whereas interactions with the corepressor motifs NCOR LXXXIXXXL2263 and SMRT LXXXIXXXL2342 were reduced compared with vehicle treatment (Fig. 2, a and b, and supplemental Fig. 3). In contrast to rosiglitazone treatment, the antagonist GW9662 failed to induce binding to any of the coactivator peptides and even slightly reduced several ligand-independent interactions while preserving the interactions with the NCoR and SMRT corepressor peptides (Fig. 2, a and b).
To characterize the binding properties of each spotted peptide on the array in more detail, dose-response curves were generated using increasing concentrations of rosiglitazone and GW9662 (2 ϫ 10 Ϫ11 -2 ϫ 10 Ϫ4 M) (Fig. 2, c and d, and supplemental Fig. 3). For each peptide curve fitting was applied to determine EC 50 and the difference between minimum and maximum binding (induction; supplemental Fig. 2), whereas the accompanying R 2 was used as quality measure for the fit. In the case of rosiglitazone, 30 of the total 48 peptide binding curves showed R 2 values of Ͼ0.7 (see sup-plemental Fig. 4 and supplemental Table 2). Within this group, EC 50 values varied from 6 ϫ 10 Ϫ7 M (TIF2 LXXLL690 ) to 4 ϫ 10 Ϫ8 M (CBP LXXLL70 ). In the case of GW9662 only six peptides showed R 2 values of Ͼ0.7 (see supplemental Table 2) with most of the peptides showing the opposite effect of that obtained with rosiglitazone, reflecting the agonistic and antagonistic activity of rosiglitazone and GW9662, respectively. In conclusion, the peptide microarray approach allows dynamic studies on the interactions between a given NR and 48 coregulator motifs in a single experimental run.
The SPPAR␥M Telmisartan Induces a Specific PPAR␥-Coregulator Interaction Profile-SPPAR␥Ms are ligands that show partial modulatory effects on PPAR␥ activity compared with traditional fully agonistic TZDs like rosiglitazone or pioglitazone (17). To establish a broad SPPAR␥M-induced coregulator interaction profile, PPAR␥ was incubated with telmisartan, an angiotensin receptor blocking drug with SPPAR␥M characteristics (33). NR-coregulator interactions were analyzed with PamChips containing 53 peptides. Using telmisartan and rosiglitazone concentrations ranging from 10 Ϫ9 to 10 Ϫ4 M, dose-response curves were generated.
A large number of peptides followed the binding pattern of rosiglitazone-treated GST-PPAR␥-LBD only with reduced induction and/or increased EC 50 values (Fig. 3, a and b, and  supplemental Fig. 5). For instance, in the case of the PRIP LXXLL1491 peptide, telmisartan showed an induction similar to that of rosiglitazone but with an increased EC 50 (Fig. 3a). In the case of peptide PGC1␤ LXXLL343 , induction was reduced, and EC 50 increased (Fig. 3, a and b). Furthermore in the case of the P300 LXXLL81 and TRAP220 LXXLL645 peptides, telmisartan failed to induce binding as only a minimal induction value was observed, whereas rosiglitazone treatment resulted in increased interactions (Fig. 3, a and b). The rosiglitazone-induced interaction between GST-PPAR␥-LBD and the four SRC1 peptides on the array (SRC1 LXXLL633 , SRC1 LXXLL690 , SRC1 LXXLL749 , and SRC1 LXXLL1435 ) was more efficient and potent compared with the effect of telmisartan (Fig.  3b). To study this interaction by independent experimental means, we performed GST pulldown experiments. GST-SRC1 fusion protein harboring three LXXLL motifs (amino acids 570 -780) was incubated with 35 S-labeled PPAR␥ in the absence or presence of rosiglitazone or telmisartan (10 Ϫ8 -10 Ϫ4 M). In accordance with the PamChip data, rosiglitazone treatment resulted in a profound interaction between PPAR␥ and SRC1, whereas the effect of telmisartan was more modest (Fig. 3c). In conclusion, our data confirm and expand the qualification of telmisartan as a bona fide SPPAR␥M.
PPAR␥ Mutants Display Altered Coregulator Interaction Profiles-To investigate the effects of PPAR␥ mutations on the coregulator interaction profile, we used the naturally occurring R425C mutant and the artificial L468A/E471A mutant. The FPLD3-associated R425C mutant in which a conserved arginine at position 425 is changed into a cysteine (34) displays reduced ligand binding and ligand-mediated coactivator Peptides derived from NR coregulators (red), containing either LXXLL or LXXXIXXXL motifs, are covalently attached to the porous carrier material (light gray). GST-NR fusion proteins (blue) are pumped through the porous material together with a fluorescently labeled antibody against GST (green). Depending on the presence of ligand, the GST-NR protein will bind to coregulator-derived peptides, which can be detected by the fluorescent antibodies. Background fluorescence signal is reduced because unbound protein-antibody complexes are collected behind an optic barrier (dark gray), taking these out of the focal plane of the optical detection system. c, time-dependent binding of GST-PPAR␥-LBD to immobilized TRAP220 LXXLL604 motif visualized by fluorescently labeled anti-GST antibodies (lower panel). During incubation, images were acquired at cycles 21, 31, 61, 81, and 101, and binding was quantified. Cycle number was plotted against signal, a recruitment (23). As a result, the transcriptional activity of this mutant is reduced (Fig. 4a), ultimately leading to impaired capacity to induce adipocyte differentiation (23). The artificial L468A/E471A double mutant contains two alanine residues in helix 12 of the LBD instead of the highly conserved hydrophobic leucine and charged glutamic acid residues. This mutant still binds ligand, but coactivator binding is ablated, whereas corepressors are recruited more avidly in the absence of ligand (22). In agreement with earlier reports (22,23), the ability of the PPAR␥ L468A/E471A to activate a 3xPPRE-tk-Luc reporter was completely abolished (Fig. 4a).
We then subjected bacterially expressed and purified wild type and mutant (R425C and L468A/E471A; Fig. 4b) GST-PPAR␥-LBD proteins to peptide microarray analysis in the absence or presence of rosiglitazone. Dose-response curves were determined for all coregulator-derived peptides. Depending on the quality of the curve fit (R 2 Ͼ 0.5) and the overlap of 95% confidential intervals of top and bottom values, potency (EC 50 ) and induction values were calculated ( Fig.  4c and supplemental Fig. 6). For wild type PPAR␥, proper dose-response curves (R 2 Ͼ 0.5) could be obtained for most peptides. However, mutation of Arg-425 dramatically altered the binding characteristics of the PPAR␥ protein because in this case only five peptides showed dose-response curves with R 2 Ͼ 0.5 (SRC1 LXXLL1435 , DAX1 LXXLL146 , NRIP1 LXXLL185/C177S , PGC1␣ LXXLL144(motif1) , and PGC1␣ LXXLL144(motif2) ) but with minimal induction values (Fig. 4, c and d). This profound effect on the overall interaction profile is in agreement with the reduced ligand binding affinity of the R425C mutant (23). The L468A/ E471A mutant protein showed strongly reduced coactivator recruitment upon ligand stimulation with no peptide displaying significant ligand-induced interaction. Ligand-independent binding to LXXXIXXXL repressor motifs, however, was stronger than in the case of wild type PPAR␥ (Fig. 4, c and d), which is in line with the stronger recruitment of corepressors by this mutant (22). Taken together, these data indicate that naturally occurring and artificial PPAR␥ mutants display distinct coregulator interaction profiles in line with their functional characterization in in vitro and cell-based assays.
As depicted in Fig. 5b, a striking PPAR isotype difference was observed with respect to the interactions with the two PGC1␣ peptides (which differ only in length; see supplemental Table 1): PPAR␤/␦ displayed clear ligand-dependent binding, whereas the interactions with PPAR␥ were ligand-independent (Fig. 5b). Because NR-coregulator peptide interactions may be influenced by the concentration of spotted peptide (supplemental Fig. 2), we analyzed the interaction between the PPAR␤/␦ and -␥ isotypes and the PGC1␣ LXXLL144 peptide using an array containing different peptide concentrations (Fig. 5c). PPAR␥ displayed only modest ligand-dependent binding, which decreased with increasing peptide concentration, PPAR␤/␦, however, displayed considerable ligand effects even at the highest peptide concentration (Fig. 5c), indicating that the difference between PPAR␤/␦ and -␥ observed in the standard assay ( Fig. 5b; 1 mM peptide) is independent of peptide concentration. GST pulldown assays were performed to verify these findings (Fig. 5d). Indeed ligand treatment clearly increased the interaction between PGC1␣ and PPAR␤/␦ as observed earlier (26,37). The interaction between PGC1␣ and PPAR␥ was highly ligand-independent with no ligand-induced binding, which was also reported before (38). These data kinetics curve was fitted, and its derivative (V ini ) at cycle 31 was used as the parameter for binding (upper graph). d, a 2-fold dilution series of the TRAP220 LXXLL604 peptide was prepared,  Table I, TRIP3 had not been implicated in PPAR␥ signaling before. The 155-amino acid TRIP3 pro-NCOR1 LXXLL2263 (white circle) and the coactivator motif TRAP220 LXXLL645 (gray circle) are highlighted. b, coregulator peptide binding profiles for GST-PPAR␥-LBD treated with DMSO, rosiglitazone (Rosi), and GW9662. Images of all arrays were quantified, and for each coregulator peptide V ini values (in arbitrary units (A.U.)) are plotted. The experiment was performed in duplicate. c, dose-response curves for GST-PPAR␥-LBD treated with rosiglitazone. GST-PPAR␥-LBD was treated with a concentration series of rosiglitazone, and images of coregulator binding per array, each with a different ligand concentration, were obtained (upper panel). For peptides NCOR1 LXXII2263 and TRAP220 LXXLL645 data were quantified and plotted in a graph with the x axis representing the 10  U2OS cells were transfected with expression vector encoding PPAR␥2 wild type (wt), PPAR␥ L468A/E471A , or PPAR␥ R425C and a 3xPPRE-tk-Luc reporter. Activation of the luciferase reporter, in the absence or presence of 1 M rosiglitazone, is expressed as -fold induction over that with tein harbors a HIT type 3 zinc finger and a well conserved LXXLL motif (Fig. 6a) and is widely expressed (39,40). TRIP3 has been shown to interact with multiple NRs: the protein was originally found to interact with thyroid hormone receptor ␤ only in the presence of thyroid hormone (39). It also showed a ligand-dependent interaction with 9-cis-retinoic acid receptor ␣ but did not interact with the glucocorticoid receptor under any condition (39). Subsequent studies showed an interaction with the orphan receptor hepatocyte nuclear factor-4␣ (40). The interaction profile showed a significant ligand-dependent interaction between the TRIP3 peptide and PPAR␥ (Figs. 4d and 6b; see also Figs. 2b and 5b and supplemental Figs. 3, 5, 6, and 7) that was abolished in the R425C and L468A/E471A mutants (Fig.  4d). To test the full-length TRIP3 protein for its ability to interact with PPAR␥, we performed GST pulldown experiments in the presence or absence of rosiglitazone. A rosiglitazone-dependent interaction between full-length TRIP3 and GST-PPAR␥-LBD was detected (Fig. 6c, upper  panel). Importantly mutation of the LXXLL motif (TRIP3 m ) completely abolished this interaction (Fig. 6c, lower panel). Because of the key role of PPAR␥ in adipogenesis, we next investigated whether TRIP3 would also play a role in this process. As a first experiment, protein expression of TRIP3 was studied during the differentiation of 3T3-L1 cells into adipocytes (Fig. 6d). Induction of differentiation, as monitored by the expression of the differentiation marker FABP4, resulted in increased expression of the TRIP3 protein after 2-3 days. To address the relevance of TRIP3 in adipogenesis, the expression of this protein was reduced using RNA interference-mediated knockdown. Using a pool of siRNA oligonucleotides, TRIP3 expression was efficiently reduced in 3T3-L1 cells (Fig. 6e). Knockdown of TRIP3 resulted in a partial inhibition of adipocyte formation as illustrated by staining of triglycerides with oil red O at day 6 of differentiation ( Fig. 6e). To confirm this reduction in adipocyte differentiation independently, protein expression levels of FABP4 and PPAR␥ were determined. Although the FABP4 and PPAR␥ proteins were clearly induced upon differentiation, treatment of cells with TRIP3 siRNA oligonucleotides blunted this response (Fig.  6e). Knockdown of PPAR␥ itself resulted in a complete inhibition of adipogenesis (Fig. 6e). These results therefore implicate the TRIP3 protein in PPAR␥-mediated adipogenesis and indicate that novel, biologically relevant interactions can be identified based on the NR-coregulator interaction profiling method used here.

DISCUSSION
For NR-mediated gene transcription to occur, these proteins need to work in concert with transcriptional coregulators, including corepressors and coactivators. The specific set of NR coregulators in complex with an NR is determined by the cellular context, the type of NR, and the type of ligand.
Here we used an in vitro microarray approach to functionally characterize ligand-induced NR-coregulator recruitment based on interactions between NRs and peptides derived from known NR coregulators.
A number of peptide-based methods have been described in which NR-coregulator interaction profiles can be generated, including homogeneous time-resolved fluorescence resonance energy transfer (41)(42)(43)(44) and bead-based multiplexed measurements coupled to flow cytometry (45,46). In addition, phage display library screening has been used to identify artificial interacting peptides (46 -50). Methods in which full-length coregulator proteins rather than LXXLL/ LXXXIXXXL peptides were used include high throughput GST pulldown assays (51), two-hybrid-based methods (52), and protein microarrays in which either NR or coregulators were immobilized (53). The peptide microarray approach we used has several advantages over these methods. First, it combines sensitive high throughput interaction profiling with real time kinetics, enhancing the reliability of the interaction data (V ini values; Fig. 1 and supplemental Fig. 1). Second, the peptide microarray approach avoids time-consuming cloning procedures. Third, only small amounts of NR protein (Ͻ10 ng of protein/array) are required with a relatively low degree of purity (data not shown).
An obvious application of high throughput NR-coregulator interaction profiling is the initial screening of novel NR (ant)agonists that may have clinical applications. Although NRs may be regarded as natural drug targets, NR-based drugs are often associated with (tissue-specific) adverse side effects. For example, synthetic PPAR␥ ligands of the TZD class, which are used in the treatment of insulin resistance associated with type 2 diabetes, have been associated with serious side effects like undesired weight gain, fluid empty vector (pCDNA3) in the absence of ligand after normalization for Renilla luciferase activity. Results are averages of at least three independent experiments assayed in duplicate ϮS.E. b, protein isolation of GST-PPAR␥-LBD, GST-PPAR␥-LBD L468A/E471A , and GST-PPAR␥-LBD R425C . GST-PPAR␥-LBD fusion proteins were bacterially expressed and purified. SDS-PAGE followed by Coomassie Brilliant Blue staining was performed to quantify protein amounts using BSA standards. c, coregulator binding profiles for GST-PPAR␥-LBD wild type and mutants. PamChip arrays were used to analyze the ligand-dependent coregulator binding of GST-PPAR␥-LBD, GST-PPAR␥-LBD R425C , and GST-PPAR␥-LBD L468A/E471A . Dose-response curves were determined upon rosiglitazone stimulation. EC 50 values and induction values were determined as described in Fig. 3b. d, Fig. 3b. For each PPAR isotype efficacies were normalized to the highest induction value obtained in the experiment. All dose-response curves were performed in duplicate. c, the effect of increasing amounts of PGC1␣ LXXLL144 peptide on the ligand-induced interaction of PPAR␤/␦ and PPAR␥ with PGC1␣ LXXLL144 . Increasing amounts of PGC1␣ LXXLL144 were spotted on a PamChip array, and at each peptide concentration V ini values were determined either in the presence or absence of ligand. d, in vitro interaction between GST-PGC1␣ and PPAR␤/␦ or PPAR␥. GST pulldown experiments were performed to analyze PGC1␣ binding by PPAR␤/␦ or PPAR␥ upon ligand stimulation. In vitro translated PPAR␤/␦ and PPAR␥ were treated with 440 nM GW501516 and 10 M rosiglitazone, respectively, and subjected to a pulldown experiment with GST-PGC1␣. Experiments were performed with DMSO as control. A.U., arbitrary units. retention, peripheral edema, and potential increased risk of cardiac failure. These effects may be due to the use of high doses of full PPAR␥ agonists (14,15). Activation in moderation through selective NR modulators (SNRMs), which elicit a more restricted, tissue-specific transcriptional response, may therefore be a more sensible approach (16,17). The validity of the SNRM concept is clearly underscored by tamoxifen, which is widely used for the treatment and prevention of breast cancer: this drug functions as an estrogen receptor agonist in some tissues (e.g. bone) but as an agonist in others (e.g. breast) (54,55). Besides SNRMs for PPAR␥ and estrogen receptor, such compounds are being FIG. 6. TRIP3 regulates PPAR␥-mediated adipocyte differentiation. a, schematic representation of the human TRIP3 protein. Indicated are the HIT type zinc finger, the LXXLL motif, and an alignment of the LXXLL amino acid sequence in human TRIP3 (NCBI accession number NP_004764) with mouse (CAI25491), Xenopus (AAI29775), and salmon TRIP3 (ACI66212). b, the effect of TRIP3 LXXLL101 peptide concentration on ligand-induced interaction with GST-PPAR␥-LBD was analyzed using a PamChip array with increasing amounts of TRIP3 LXXLL101 peptide. c, in vitro interaction between GST-PPAR␥-LBD and TRIP3. In vitro translated 35 S-labeled wild type TRIP3 and TRIP3 containing a mutated LXXLL motif (TRIP3 m ) were subjected to GST pulldown experiments with GST-PPAR␥-LBD in the absence or presence of rosiglitazone. d, TRIP3 protein expression is induced during adipocyte differentiation. The expression of TRIP3 and FAPB4 during 3T3-L1 differentiation was monitored using Western blot analysis. e, knockdown of TRIP3 expression impairs adipocyte differentiation. During differentiation 3T3-L1 cells were treated with siRNA oligonucleotides against TRIP3 or PPAR␥ or with control oligonucleotides. Protein levels of TRIP3, PPAR␥, and FABP4 were analyzed by Western blotting (left panel), and differentiated cells were stained with oil red O (right panel). A.U., arbitrary units; diff., differentiated; undiff., undifferentiated. developed by pharmaceutical companies for the progesterone receptor, androgen receptor, and glucocorticoid receptor (56). NR-coregulator interaction profiling allows characterization of compounds, including SNRMs, as illustrated by the interaction profile induced by the SPPAR␥M telmisartan (Fig. 3). Together with experiments assessing the selectivity of a compound for a given NR (Fig. 5 and supplemental Fig.  8), such analyses could aid the selection of effective compounds in the early stages of drug development to help minimize the costs of expensive gene expression analyses and animal studies.
Another interesting application of NR-coregulator interaction profiling is the functional characterization of NR mutants. A number of diseases are associated with mutations in NR genes, frequently affecting the LBD (57,58). In some cases, the functional defect can be (partially) rescued by specific ligands as exemplified by the effects of synthetic tyrosinebased agonists on the FPLD3-associated PPAR␥ V318M and P495L mutants (59). Characterization of NR mutants using NR-coregulator interaction profiling can therefore provide the rationale for therapy in such cases.
Several novel ligand-dependent interactions were observed in our interaction profiling experiments. Previously unknown interactions ( Table I) include those of PPAR␥ with IKBB, GCN5, TRIP3, and TRIP8 ( Fig. 2 and supplemental Figs. 3, 5, 6, and 7). Our data further implicate TRIP3 in PPAR␥-mediated adipocyte differentiation. In yeast, TRIP3 lacks intrinsic transcriptional activity when tethered to DNA (39), suggesting that its function as a transcriptional coregulator depends on other mammalian proteins. Interestingly the TRIP3 protein harbors a HIT type zinc finger, named after the yeast HIT1 protein, which is mainly found in nuclear proteins involved in gene regulation and chromatin remodeling (60). Recent studies on the closely related ZNHIT2 protein revealed that this type of zinc finger is not involved in DNA binding but probably serves as a protein-protein interaction surface (61). Future studies are therefore required to reveal the exact role of TRIP3 and its HIT type zinc finger in PPAR␥-mediated transcription processes.
Although best known for their application in enzymatic studies on kinases and proteases (21), our studies indicate that peptide microarrays may have wider applications in protein-protein interaction studies. Of particular interest in this respect are docking interactions of kinases and phosphatases with their substrates, which critically depend on short peptide motifs within these substrate proteins (62). In addition, this technology may be extended toward the interactions between N-terminal histone tails and their binding partners in which the interactions are dictated by specific posttranslational modifications (63). Because NRs, kinases/phosphatases, and histone-interacting proteins like lysine and arginine methylases are involved in many diseases and often qualify as potential drug targets, peptide microarray platforms may present an important generic tool in drug development.