A structural mechanism for directing corepressor-selective inverse agonism of PPARγ

Small chemical modifications can have significant effects on ligand efficacy and receptor activity, but the underlying structural mechanisms can be difficult to predict from static crystal structures alone. Here we show how a simple phenyl-to-pyridyl substitution between two common covalent orthosteric ligands targeting peroxisome proliferator-activated receptor (PPAR) gamma converts a transcriptionally neutral antagonist (GW9662) into a repressive inverse agonist (T0070907) relative to basal cellular activity. X-ray crystallography, molecular dynamics simulations, and mutagenesis coupled to activity assays reveal a water-mediated hydrogen bond network linking the T0070907 pyridyl group to Arg288 that is essential for corepressor-selective inverse agonism. NMR spectroscopy reveals that PPARγ exchanges between two long-lived conformations when bound to T0070907 but not GW9662, including a conformation that prepopulates a corepressor-bound state, priming PPARγ for high affinity corepressor binding. Our findings demonstrate that ligand engagement of Arg288 may provide routes for developing corepressor-selective repressive PPARγ ligands.

We thank the Reviewers for their time and constructive comments regarding our manuscript. Our revised manuscript includes new data in response to Reviewers 2 and 3 that together with other revisions recommended by all reviewers address the major points raised by the reviewers.
Our original manuscript contained 5 figures, and in this revised manuscript we split the previous version of Figure 5 into two figures ( Figure 5, which also contains some new data requested by Reviewer 3; and Figure 6). We also corrected the radar plot for the R288K mutant in Figure 3E; in the original version, we mistakenly used the wild-type (R288) plot for the R288K mutant. Our original supporting information document contained 4 figures and 1 table; the revised version contains 11 figures and 2 tables.
Below, our point-by-point response is formatted as a blue text indented paragraph with a left vertical line. We also included a version of the revised manuscript with Microsoft Word's Track Changes enabled to enable easy visualization of the changes made to the original manuscript.
Reviewer #1 (Remarks to the Author): The Authors compared the crystal structures of PPARg complexed with two covalent ligands only differing by a methine (CH) to nitrogen substitution and found that, despite no major structural differences, the two ligands show completely different agonist properties (one, GW9662, is a neutral antagonist, the other, T0070907, an inverse agonist). They found that the inverse agonist T0070907 (with the nitrogen substitution) realizes a water-mediated H-bond network that uniquely links R288 of PPARγ to the ligand pyridyl group. The Authors confirmed the stability of this network by MD simulations. Then, they used mutagenesis, activity assays and fluorescence polarization assays, with and without coregulators, to demonstrate that the unique H-bond network realized by T0070907 is responsible for the inverse agonism of the ligand. Moreover, the Authors discovered by NMR spectroscopy that PPARγ exchanges between two long-lived conformations when bound to T0070907, but not GW9662. Focusing on the 15N-PPARγ peaks of the residue G399, located near the coregulator hydrophobic cleft, they found that, in the presence of T0070907, one of the two conformations of PPARγ prepopulates a corepressorbound state, priming PPARg for high affinity corepressor binding. I found the manuscript very well written and the experiments very well designed and performed. The work is convincing and the conclusions original. The authors elegantly demonstrated their hypothesis. Particularly, the NMR studies, in the presence of increasing amounts of coregulators, proved to be very effective and powerful in understanding the ligand mechanism of action and open new routes to design and characterize PPARγ inverse agonists.

Authors' response:
We thank the reviewer for their comments and for pointing out the minor points listed below, which we have addressed in this revised manuscript.
Minor points to be addressed: Line 33: The authors affirm that "partial or graded agonists do not hydrogen bond to Y473, but mildly stabilize helix 12 via interactions with other regions of the ligand-binding pocket, resulting less pronounced changes in coregulatory affinity and transcriptional activation". Actually, there are also ligands, not cited by the Authors, that bind to Y473 but show partial agonism character because of unfavorable interactions with residues of helix 3 (Q286) or interactions with different regions of the LBD (H3, H11 and loop before H12) (Pochetti G. et al., JBC (2007) 282:17314-17324 andMontanari R. et al., JMC (2008) 51:7768:7776). For this reason I would modify the previous sentence in : "partial or graded agonists, generally, do not hydrogen bond to Y473…" Authors' response: These studies are indeed interesting and adds to the known structural mechanisms for directing PPARγ partial agonism. We modified our introduction to include a discussion of this mechanism and cited the above references.
Line 145: In Results the Authors affirm to have performed the MD simulations ranging from 4-26 microseconds in length, but in Methods regarding MD I didn't find a sentence about this length. Could you confirm that the MD simulation lasted 4-26 microseconds?
Authors' response: We have updated our Methods section to include this information as follows: "Production simulations were run in triplicate or quadruplicate for the following durations-T0070907 (modeled conformation; 3 total): 24.3, 26.4, and 13.4 µs; T0070907 (crystallized conformation; 4 total): all were 4 µs; GW9662 (crystallized conformation; 4 total): all were 4 µs." Lines 209 and 211: The Authors refer to Figures 3F and 3G, but in Figure 3 they are not present. Maybe should be in the text Figure 3D and 3E?
Authors' response: We corrected this typographical error, which should have referred to Figure  3D and 3E.
Line 317: D313 on helix 5, and not D311 ! Authors' response: We corrected this to read D313.
Line 384: maybe "chemical shit" is to be updated… Authors' response: We appreciate that the reviewer caught this typographical error and corrected it to "chemical shift". Line 414: "deconvoluted" and not "deconvoluated".
Authors' response: We corrected this typographical error.
Dr. Giorgio Pochetti Istituto di Cristallografia Area della Ricerca Roma1 -CNR Italy Reviewer #2 (Remarks to the Author): Brust et al. present a broad analysis of the interactions between a novel PPARγ ligand (T0070907) and the receptor using a broad range of techniques that include cell-based assays, in vitro binding assay, MD simulations, CD & NMR spectroscopy and crystallography. This molecule, T0070907, is a derivative of another PPARγ ligand, GW9662. The modification from the precursor to T0070907 is subtle -a methine to nitrogen substitution in T0070907. The authors also compare the mode of binding of these two PPARγ ligands. The authors conclude that unlike GW9662, T0070907 functions as an 'inverse agonist', a property that arises principally from a water-mediated H-bond network that links R288 of PPARγ to the T0070907 pyridyl group. Furthermore, they report that T0070907-bound PPARγ exists in two conformations, one similar to GW9662-bound PPARγ and a second that is unique. They conclude that this study has identified a 'novel structural mechanism.' Authors' response: We appreciate the reviewer's comments below, which have given us the opportunity to better clarify these points for the general readership in our revised manuscript.
Specific concerns: 1. Figure  Authors' response: In our revised manuscript, we include a new supplementary figure ( Figure S2; see the image to the right), which clearly shows that full-length PPARγ shows a transcriptional response (i.e., basal activity) in the absence of an exogenously added ligand (i.e., the vehicle/solvent DMSO control condition) relative to transfection of an empty control plasmid. Also, in Figure 1C we now show new qPCR gene expression data from 3T3-L1 preadipocyte cells treated with the same ligands used in the cellbased transactivation and biochemical profiling assays. The ligands show the same profiles in the expression of two known PPARγ target genes, FABP4/aP2 and CD36; and in both cases T0070907 represses the expression of these PPARγ target genes relative to DMSO control treated cells. These observations are consistent with our previous work published in Nature Communications, which showed that the direct antagonist ligands (SR2595 and SR10221) that we also studied in this report decrease PPARγ transactivation (transcriptional repression) relatively to vehicle (DMSO) control, which we similarly show in Figure 1B of this current manuscript. The only way transcriptional repression can be observed in these assays (reporter transactivation assay and the qPCR assay) is because PPARγ has significant observable basal transactivation in the absence of exogenous ligand. One reason that PPARγ has basal activity in the vehicle (DMSO) control condition is cells contain endogenous PPARγ ligands (lipids and fatty acids), which act as PPARγ agonists by enhancing coactivator binding and decreasing corepressor binding-as we show in Figure 1D for nonanoic acid.
Second, our data do match published observations and the general knowledge in the field that PPARγ has transcriptional activity in cells without exogenously added ligand. There are numerous publications showing this for other PPARs and other nuclear receptors. We found several publications that similarly observed an increase in basal activity upon cellular transfection of full-length PPARγ: 1. Figure Figure 5B in Hou Y, Moreau F, Chadee K. PPARgamma is an E3 ligase that induces the degradation of NFkappaB/p65. Nat Commun. 2012;3:1300. 3. Figure  and/or cellular treatment of a repressive ligand: 1. Figure 5A in Huang C, Zhang Y, Gong Z, Sheng X, Li Z, Zhang W, Qin Y. Berberine inhibits 3T3-L1 adipocyte differentiation through the PPARgamma pathway. Biochem Biophys Res Commun. 2006;348 (2):571-8.
We carefully read the four studies cited by the reviewer. One article cited by the reviewer does not focus on PPARγ or a nuclear receptor in general (Scientific rep. 2015, 5, 8256); it is titled "Efficient generation of gene-modified pigs via injection of zygote with Cas9/sgRNA". For another citation provided by the reviewer, we could not find an article under the citation provided (JBC 2000, 275(3), 1883-7).
In the Cell (2004) 116, 417-29 paper, we could not find any cellular transactivation data for PPARγ, but data are shown for a related receptor, PPARα, in Figure 4D. Consistent with our findings for PPARγ in this manuscript, PPARα showed basal activity of ~200 RLU units in the Cell paper.
In the original report of T0070907 (JBC 2002, 277, 22:19649 -19657), a Gal4-PPARγ LBD chimeric fusion protein was used to in Figure 2A to show that T0070907 inhibits rosiglitazone's ability to activate the Gal4-PPARγ LBD chimeric fusion. A close inspection of the data in Figure 2A in this JBC paper shows that the bar graph for T0070907 treated cells is lower than basal expression, which is consistent with transcriptional repression. To the right, we have cropped and increased the size of Figure 2A Figure 2A excerpt to the right). It is important to note that in this JBC study, T0070907 was cotreated with rosiglitazone because the main point of the paper is T0070907 inhibits binding of other synthetic ligands-this provided the initial "antagonist" label. However, the ability of T0070907 to inhibit the binding and activity of another ligand is not a pharmacological description of T0070907; the use of the term "antagonist" was not descriptive of T0070907's transcriptional properties on its own.
2. The authors refer to the molecule T0070907 as an inverse agonist. However, in the original study on this molecule, (JBC (2002) 277(22) Authors' response: As mentioned above, GW9662 and T0070907 were referred to as antagonists in their original studies because they bind covalently and block binding of other synthetic ligands such as rosiglitazone. However, the ability of a ligand such as GW9662 or T0070907 to block other ligands from binding to the orthosteric pocket is not a pharmacological description of their transcriptional activities. Our manuscript explores the pharmacological properties of GW9662 and T0070907 in head-to-head cellular (transactivation reporter and QPCR gene expression) and coregulator interaction assays by comparing T0070907 to GW9662 and other activating and repressive PPARγ ligands, which clearly shows T0070907 is an inverse agonist by definition: an inverse agonist enhances corepressor binding and decreases coactivator binding. In contrast, an antagonist by definition is a ligand that blocks the activity of other ligands while lacking the ability to modulate the receptor's activity on their own. This would mean that the activity of PPARγ bound to an antagonist should be similar to ligand free/apo-PPARγ. We added text to the manuscript introduction, results, and discussion sections to better describe the difference between GW9662's and T0070907's ability to inhibit ligand binding, which led to their original "antagonist" label; and their ability to influence PPARγ transcription, target gene expression, and coregulator interaction on their own, which provides their true pharmacological labels. We included a new supporting information table ( Table S1) that provides pharmacological definitions of nuclear receptor agonists, neutral antagonists, direct antagonists, and inverse agonists.
The reviewer indicated the original JBC paper only showed T0070907 to be an inverse agonist within the context of the GST-tagged PPARγ LBD. However, the JBC paper also showed that T0070907 increased interaction of full-length NCoR to full-length PPARγ/RXRα heterodimer bound to DNA using an GMSA/EMSA assay ( Figure 5 in the JBC paper). Consistent with these results, we provide new data in our manuscript showing that T0070907 strengthens the affinity (Kd) of our NCoR peptide for full-length PPARγ using a fluorescence polarization interaction assay ( Figure S3) and enhances the binding of our NCoR peptide to the PPARγ LBD and the full-length PPARγ/RXRα heterodimer bound to a PPARγ response element (PPRE) DNA sequence ( Figure S4).
Finally, using our fluorescence polarization coregulator interaction assay we found that delipidation of PPARγ LBD protein does not significantly affect coregulator affinity (see the figure to the right), likely because any co-purified bacterial lipids are bound substoichiometrically because of our extensive purification methods where any bound lipids would likely exchange off during column chromatography. This is supported by our coregulator profiling data we showed in Figure 1D,E, where addition of nonanoic acid-a human dietary PPARγ ligand that is also present in bacteria and can be co-purified with PPARγ ( acids are endogenous agonists that function by forming hydrogen bonds with Tyr473 and other residues near helix 12, which stabilizes the AF-2 surface similar to synthetic agonists. If our purified protein was saturated with bacterial lipid, we would not have observed such a dramatic effect upon addition of nonanoic acid in the coregulator profiling assay. Furthermore, our data show that the inverse agonist coregulator recruitment properties of T007 is enhanced, not unfounded as suggested by the reviewer, when compared to PPARγ bound to lipid (as we showed in Figure 1D) because lipids function as PPARγ agonists not inverse agonists or neutral antagonists.
4. Although the GW9662 ligand is defined as an antagonist, the structure of the GW9662-PPARg LBD complex (3B0R) is of PPARg in the agonist conformation (helix 12 conformation), suggesting the opposite. This is important because this study derives much of its conclusions in comparison to the GW9662-PPARg LBD complex (3B0R) structure.
Authors' response: As we stated in our manuscript and show in Figure S8 (previously Figure  S2; and we updated this figure to show nearby symmetry related molecules), the "active" or "agonist" conformation of helix 12 is a crystallization artifact. In many of the published ligandbound PPARγ crystal structures, helix 12 of chain B docks into the AF-2 surface of chain A, which artificially forces chain A helix 12 into the "agonist" conformation. We discussed this in our manuscript as part of the text that leads up to our NMR studies-and we would argue that it is our comparative NMR studies, not the comparison of the crystalized helix 12 conformations, that is important in deriving the conclusions in our study. In short, as we stated in the manuscript, the crystal structures were critical to identify the pyridyl-water-Arg288 interaction, but our NMR studies were essential to show what happens to the conformation of PPARγ in solution (not in the crystalline state) when bound to T0070907 or GW9662-that T0070907 populates a unique corepressor-like conformation in solution, which due to packing interactions in the crystal are not able to be observed in the crystalline/crystal structure state. Figures 3F and 3G.

Authors' response:
We mistakenly called out Figure 3F and 3G; these callouts should have been to Figure 3D and 3E. We thank the reviewer for pointing this out and we corrected this error.
5. The NMR study shows the chemical shifts in some distinguishable resonances. However, a significant portion of the peaks are unresolved. It is likely that many of the shifts are conjugated with other movements that are not observed in this study. Ironically, there is no R288, the residue about which this entire study hinges, is unobserved.

Authors' response:
PPARγ is a relatively large protein to study by NMR, and therefore there is a good amount of NMR chemical shift (peak) overlap in the center of the spectrum, which includes the backbone resonance for R288, making analysis difficult or impossible. In situations like these, it is common to analyze well resolved or "distinguishable" resonances-and to this point, we would also point out that in Figure S9 (previously Figure S4) we showed that 11 other wellresolved NMR peaks clearly show slow exchange properties on the NMR time scale. Furthermore, in the analysis of ZZ exchange data, the appearance of exchanging crosspeaks typically only occurs for a few resonances when very strict criteria are met: when the two conformations that are exchange have large separation in the 15N dimension. This is best stated in a well-accepted review on NMR experiments to study protein dynamics: "EXSY may be limited by spectral crowding and/or poor sensitivity because it functions by introducing additional, often weak, signals into the spectrum. Practically though, many EXSY studies only require a few structural probes to address the questions of interest (as opposed to the tens of structural probes typically required for other NMR-based techniques).
Finally, we note that in an NMR experiment we performed in response to Reviewer 3's critique, we show new NMR data where we assigned the R288 side-chain Nε/Ηε NMR peak (by mutagenesis), determined how it is affected when PPARγ is bound to T0070907 vs. GW9662, and consistent with our T0070907-bound crystal structure we also observed water interactions with the R288 side-chain Nε/Ηε group using NMR experiments.
Reviewer #3 (Remarks to the Author): This manuscript addresses structural mechanism for the difference in effect between the highly similar compounds, GW9662 and T0070907, on peroxisome proliferator-activated receptor gamma (PPARγ) activity. Despite simple methane to nitrogen substitution, GW9662 is a neutral antagonist of PPARγ, while T0070907 is an inverse agonist. Using crystallography, molecular dynamics (MD) simulations and mutagenesis, they showed that water-mediated hydrogen bond network linking the T0070907 pyridyl group to Arg288 is essential for the inverse agonism. They also performed NMR experiments, and showed that PPARγ adapts two long-lived conformations when bound to T0070907 but not GW9662, one of which is resembles a GW9662-bound state and the other is unique state that is similar to the corepressor-bound state. This manuscript is well-written and provided insights into the mechanism for inverse agonism of PPARγ. However, I have some concerns as follows.

Authors' response:
We thank the reviewer for their comments, suggested experiments, and alternate data interpretations, which we have addressed in this revised manuscript and enhances our work.
1. The relationship between the water-mediated hydrogen bond network and the prepopulated corepressor-bound conformation in the T0070907-bound state is not clear. I wonder weather the hydrogen bond network is necessary for adapting the corepressor-bound conformation. This point should be addressed, for example, by NMR measurements of a T0070907-bound Arg288 mutant.

Authors' response:
This was an excellent experiment to suggest. In Figure 5B and Figure S9C, we show new 2D NMR data for the R288K and R288A mutants. Interestingly, and consistent with our biochemical and cellular mutagenesis findings, the R288K mutant significantly populates the corepressor-like "unique" conformation, whereas the R288A mutant significantly populates the coactivator-like "mutual" conformation that is shared with GW9662. In these studies, we also found that the GW9662-like/mutual/coactivator-like conformation observed for T0070907-bound PPARγ is less populated in the R288K mutant. In the discussion, we indicate that this may explain why the R288K mutant shows weakened affinity for the TRAP220 coactivator compared to T0070907-bound PPARγ. Furthermore, we found that the R288A mutant lowly populates the unique/corepressor-like conformation. As we discuss in the results section, this indicates that: (1) an extended pyridyl-water hydrogen bond network to residues other than R288, which is observed in the T0070907-bound crystal structure and would be present in wild-type PPARγ and the R288 mutants, may be involved in lowly populating the unique conformation but it is not sufficient for directing inverse agonism; and (2) the pyridyl-water interaction with R288 (or a positively charged residue) is necessary for significant population of the unique (corepressor-like) conformation and directing inverse agonism.
2. If bound water is actually located as shown in Figures 2B and 2C, a NOE signal will be observed between Arg288 Hε and water when PPARγ is bound to T0070907 but not GW9662. The NOE signal will be a strong evidence for the existence of the water-mediated hydrogen bond network.
Authors' response: To assign the R288 side-chain Nε/Ηε NMR peak, we compared 2D [ 1 H, 15 N]-HSQC of T0070907-bound wild-type PPARγ and the R288K mutant. One peak in the Arg sidechain Nε/Ηε spectral region disappeared in the R288K mutant ( Figure S5). Interestingly, this peak is also missing in GW9662-bound wild-type PPARγ, indicating that the pyridyl-water-R288 network is involved in stabilizing µs-ms time scale motions (intermediate exchange on the NMR time scale) of the R288 side-chain. To determine if there is a water NOE signal to the R288 sidechain Nε/Ηε group, we performed a 3D 15 N-NOESY-HSQC experiment on T0070907-bound PPARγ ( Figure 2D). In the R288 side-chain Nε/Ηε NOESY strip, a NOE peak is observed at 4.77 ppm, which is consistent with a water NOE. We also confirmed this result using a Phase-Modulated CLEAN chemical EXchange (CLEANEX-PM) NMR experiment (Figure S6), which detects water-protein interactions. Because the R288 side-chain Nε/Ηε NMR peak is not visible when bound to GW9662 due to intermediate exchange, we did not collect the 3D 15 N-NOESY-HSQC experiment on GW9662-bound PPARγ. We provide some additional insight into these data in the discussion section.