Vidya Prasad

The nuclear hormone receptor peroxisome proliferator-activated receptor γ (PPARγ) plays central roles in adipogenesis and glucose homeostasis and is the molecular target for the thiazolidinedione (TZD) class of antidiabetic drugs. Activation of PPARγ by TZDs improves insulin sensitivity; however, this is accompanied by the induction of several undesirable side effects. We have identified a novel synthetic PPARγ ligand, T2384, to explore the biological activities associated with occupying different regions of the receptor ligand-binding pocket. X-ray crystallography studies revealed that T2384 can adopt two distinct binding modes, which we have termed “U” and “S”, interacting with the ligand-binding pocket of PPARγ primarily via hydrophobic contacts that are distinct from full agonists. The different binding modes occupied by T2384 induced distinct patterns of coregulatory protein interaction with PPARγ in vitro and displayed unique receptor function in cell-based activity assays. We speculate that these unique biochemical and cellular activities may be responsible for the novel in vivo profile observed in animals treated systemically with T2384. When administered to diabetic KKAy mice, T2384 rapidly improved insulin sensitivity in the absence of weight gain, hemodilution, and anemia characteristics of treatment with rosiglitazone (a TZD). Moreover, upon coadministration with rosiglitazone, T2384 was able to antagonize the side effects induced by rosiglitazone treatment alone while retaining robust effects on glucose disposal. These results are consistent with the hypothesis that interactions between ligands and specific regions of the receptor ligand-binding pocket might selectively trigger a subset of receptor-mediated biological responses leading to the improvement of insulin sensitivity, without eliciting less desirable responses associated with full activation of the receptor. We suggest that T2384 may represent a prototype for a novel class of PPARγ ligand and, furthermore, that molecules sharing some of these properties would be useful for treatment of type 2 diabetes. The nuclear hormone receptor peroxisome proliferator-activated receptor γ (PPARγ) plays central roles in adipogenesis and glucose homeostasis and is the molecular target for the thiazolidinedione (TZD) class of antidiabetic drugs. Activation of PPARγ by TZDs improves insulin sensitivity; however, this is accompanied by the induction of several undesirable side effects. We have identified a novel synthetic PPARγ ligand, T2384, to explore the biological activities associated with occupying different regions of the receptor ligand-binding pocket. X-ray crystallography studies revealed that T2384 can adopt two distinct binding modes, which we have termed “U” and “S”, interacting with the ligand-binding pocket of PPARγ primarily via hydrophobic contacts that are distinct from full agonists. The different binding modes occupied by T2384 induced distinct patterns of coregulatory protein interaction with PPARγ in vitro and displayed unique receptor function in cell-based activity assays. We speculate that these unique biochemical and cellular activities may be responsible for the novel in vivo profile observed in animals treated systemically with T2384. When administered to diabetic KKAy mice, T2384 rapidly improved insulin sensitivity in the absence of weight gain, hemodilution, and anemia characteristics of treatment with rosiglitazone (a TZD). Moreover, upon coadministration with rosiglitazone, T2384 was able to antagonize the side effects induced by rosiglitazone treatment alone while retaining robust effects on glucose disposal. These results are consistent with the hypothesis that interactions between ligands and specific regions of the receptor ligand-binding pocket might selectively trigger a subset of receptor-mediated biological responses leading to the improvement of insulin sensitivity, without eliciting less desirable responses associated with full activation of the receptor. We suggest that T2384 may represent a prototype for a novel class of PPARγ ligand and, furthermore, that molecules sharing some of these properties would be useful for treatment of type 2 diabetes. Peroxisome proliferator-activated receptor γ (PPARγ (NR1C3)) 6The abbreviations used are: PPARperoxisome proliferator-activated receptorTZDthiazolidinedioneHTRFhomogeneous time-resolved fluorescenceGSTglutathione S-transferaseLBDligand-binding domainFmocN-(9-fluorenyl)methoxycarbonylRXRretinoid X receptorNCORnuclear receptor corepressor. 6The abbreviations used are: PPARperoxisome proliferator-activated receptorTZDthiazolidinedioneHTRFhomogeneous time-resolved fluorescenceGSTglutathione S-transferaseLBDligand-binding domainFmocN-(9-fluorenyl)methoxycarbonylRXRretinoid X receptorNCORnuclear receptor corepressor. is a member of the nuclear hormone receptor superfamily of ligand-activated transcription factors (1Rosen E.D. Spiegelman B.M. J. Biol. Chem. 2001; 276: 37731-37734Abstract Full Text Full Text PDF PubMed Scopus (1068) Google Scholar, 2Nuclear Receptors Nomenclature CommitteeCell. 1999; 97: 161-163Abstract Full Text Full Text PDF PubMed Scopus (932) Google Scholar). PPARγ, together with PPARα (NR1C1) and PPARδ (NR1C2), form a subfamily of “lipid sensing” receptors that regulate important aspects of lipid and glucose homeostasis (3Evans R.M. Barish G.D. Wang Y.X. Nat. Med. 2004; 10: 355-361Crossref PubMed Scopus (1257) Google Scholar, 4Semple R.K. Chatterjee V.K. O'Rahilly S. J. Clin. Investig. 2006; 116: 581-589Crossref PubMed Scopus (667) Google Scholar). At least two PPARγ isoforms exist, γ1 and γ2, resulting from transcription from two different promoters upstream of the PPARG gene (5Fajas L. Auboeuf D. Raspe E. Schoonjans K. Lefebvre A.M. Saladin R. Najib J. Laville M. Fruchart J.C. Deeb S. Vidal-Puig A. Flier J. Briggs M.R. Staels B. Vidal H. Auwerx J. J. Biol. Chem. 1997; 272: 18779-18789Abstract Full Text Full Text PDF PubMed Scopus (1075) Google Scholar). PPARγ1 is expressed broadly in many tissues, whereas PPARγ2, which possesses an additional 30 amino acids at its N terminus, is expressed predominantly in adipose tissue. Both “gain of function” and “loss of function” studies have firmly established PPARγ as a master regulator of adipocyte differentiation and a link with the diseased state in type 2 diabetes (3Evans R.M. Barish G.D. Wang Y.X. Nat. Med. 2004; 10: 355-361Crossref PubMed Scopus (1257) Google Scholar). The important roles PPARγ plays in lipid metabolism, inflammatory cytokine production, and macrophage function may also have direct impacts on dyslipidemia, atherosclerosis, and cardiovascular diseases (6Ricote M. Valledor A.F. Glass C.K. Arterioscler. Thromb. Vasc. Biol. 2004; 24: 230-239Crossref PubMed Scopus (140) Google Scholar).The first insight into the link between PPARγ and diabetes came from the discovery of PPARγ as the biologic target for the thiazolidinedione class of antidiabetic drugs (TZDs) (7Forman B.M. Tontonoz P. Chen J. Brun R.P. Spiegelman B.M. Evans R.M. Cell. 1995; 83: 803-812Abstract Full Text PDF PubMed Scopus (2713) Google Scholar, 8Lehmann J.M. Moore L.B. Smith-Oliver T.A. Wilkison W.O. Willson T.M. Kliewer S.A. J. Biol. Chem. 1995; 270: 12953-12956Abstract Full Text Full Text PDF PubMed Scopus (3443) Google Scholar). Two such drugs, rosiglitazone (Avandia™; GlaxoSmithKline) and pioglitazone (Actos™; Takeda Pharmaceuticals), have been prescribed either as monotherapy or in combination with sulfonylureas, metformin, or insulin to achieve glycemic control in diabetic patients. Interestingly, although TZDs lower glucose and improve lipid and adipocytokine profiles in type 2 diabetics, as full agonists they also stimulate adipocyte differentiation in vitro and result in weight gain in vivo, which normally aggravates the diabetic state. Additional undesirable side effects associated with TZD treatment include edema/hemodilution, cardiomegaly, and anemia (9Nesto R.W. Bell D. Bonow R.O. Fonseca V. Grundy S.M. Horton E.S. Le Winter M. Porte D. Semenkovich C.F. Smith S. Young L.H. Kahn R. Diabetes Care. 2004; 27: 256-263Crossref PubMed Scopus (506) Google Scholar). These side effects most likely result directly from activation of the receptor, because other PPARγ agonists with unrelated chemical structures show similar profiles (10Picard F. Auwerx J. Annu. Rev. Nutrition. 2002; 22: 167-197Crossref PubMed Scopus (370) Google Scholar).Paradoxically, the reduction of PPARγ activity also results in improvement in insulin sensitivity. Human genetic studies have identified genetic variants in PPARγ associated with altered risks to development of type 2 diabetes. For example, one common polymorphism, Pro12Ala, correlates both with a reduction in the risk of type 2 diabetes and with a reduction in body mass index (11Knouff C. Auwerx J. Endocr. Rev. 2004; 25: 899-918Crossref PubMed Scopus (239) Google Scholar, 12Florez J.C. J. Clin. Endocrinol. Metab. 2004; 89: 4234-4237Crossref PubMed Scopus (11) Google Scholar). The in vitro characterization of this polymorphism suggests that the Ala allele causes a partial loss of PPARγ function as a result of decreased DNA binding affinity and transcription activity (13Deeb S.S. Fajas L. Nemoto M. Pihlajamaki J. Mykkanen L. Kuusisto J. Laakso M. Fujimoto W. Auwerx J. Nat. Genet. 1998; 20: 284-287Crossref PubMed Scopus (1190) Google Scholar, 14Masugi J. Tamori Y. Mori H. Koike T. Kasuga M. Biochem. Biophys. Res. Commun. 2000; 268: 178-182Crossref PubMed Scopus (181) Google Scholar). In rodents, the reduction of PPARγ expression in PPARγ+/- heterozygotes protects these animals from high fat diet- or aging-induced adipocyte hypertrophy, obesity, and insulin resistance. The heterozygous animals also show reduction in factors associated with insulin resistance, including free fatty acids and tumor necrosis factor α, an up-regulation of leptin and adiponectin, and a significantly increased rate of fatty acid β-oxidation. This latter finding may explain the decrease in triglyceride content of white adipose tissue, skeletal muscle, and liver seen in these animals (15Yamauchi T. Kamon J. Waki H. Murakami K. Motojima K. Komeda K. Ide T. Kubota N. Terauchi Y. Tobe K. Miki H. Tsuchida A. Akanuma Y. Nagai R. Kimura S. Kadowaki T. J. Biol. Chem. 2001; 276: 41245-41254Abstract Full Text Full Text PDF PubMed Scopus (560) Google Scholar). These results, in addition to the analysis of other PPARγ polymorphisms (11Knouff C. Auwerx J. Endocr. Rev. 2004; 25: 899-918Crossref PubMed Scopus (239) Google Scholar, 16Argmann C.A. Cock T.A. Auwerx J. Eur. J. Clin. Investig. 2005; 35: 80-92Crossref PubMed Scopus (54) Google Scholar), suggest that the receptor-mediated side effects associated with TZD treatment could potentially be decoupled from the insulin sensitization effect. In theory, a PPARγ ligand that could modulate receptor activity and mimic the effects observed with a P12A allele or PPARγ heterozygotes might abolish or at least reduce the unwanted side effects associated with full activation of the receptor while still triggering the physiologic responses that allow proper glycemic control.The PPARγ ligand-binding site is a large Y-shaped cavity with a total volume of ∼1,300 Å3, as seen in the apo-PPARγ crystal structure (17Nolte R.T. Wisely G.B. Westin S. Cobb J.E. Lambert M.H. Kurokawa R. Rosenfeld M.G. Willson T.M. Glass C.K. Milburn M.V. Nature. 1998; 395: 137-143Crossref PubMed Scopus (1658) Google Scholar). Rosiglitazone binds in this pocket in a “U-shaped” conformation, wrapping around helix 3 with its central benzene ring directly behind helix 3 and the TZD head group extending toward the AF2 helix to form a direct H-bond with the hydroxyl moiety of the Tyr473 side chain. This interaction is critically important for stabilizing the AF2 helix in a conformation that interacts with coactivator proteins. Because this large binding pocket underlies the surface important for receptor interactions with coactivator and corepressor proteins (17Nolte R.T. Wisely G.B. Westin S. Cobb J.E. Lambert M.H. Kurokawa R. Rosenfeld M.G. Willson T.M. Glass C.K. Milburn M.V. Nature. 1998; 395: 137-143Crossref PubMed Scopus (1658) Google Scholar, 18Xu H.E. Stanley T.B. Montana V.G. Lambert M.H. Shearer B.G. Cobb J.E. McKee D.D. Galardi C.M. Plunket K.D. Nolte R.T. Parks D.J. Moore J.T. Kliewer S.A. Willson T.M. Stimmel J.B. Nature. 2002; 415: 813-817Crossref PubMed Scopus (508) Google Scholar), we reason that ligands bind to different regions of this large pocket, may differentially affect receptor-coregulatory protein interactions, and selectively modulate aspects of receptor activity.Here we describe a novel, potent, and selective PPARγ ligand, T2384, which may represent this new class of molecule. Using a variety of biochemical and cell-based assays, we demonstrate that T2384 binds to PPARγ in multiple conformations, each with potentially distinct biological activities in vitro. Moreover, we suggest that the unique in vitro properties of the ligand we have described might be responsible for this distinctive in vivo pharmacology. In rodent models of diabetes, T2384 effectively lowered plasma glucose and insulin levels without the induction of body weight gain and anemia.EXPERIMENTAL PROCEDURESIn Vitro Reagents and Assay Protocols—Plasmids used, homogeneous time-resolved fluorescence (HTRF) assay, ligand-binding assay, and transient transfection assay were all described previously (19Lee G. Elwood F. McNally J. Weiszmann J. Lindstrom M. Amaral K. Nakamura M. Miao S. Cao P. Learned R.M. Chen J.L. Li Y. J. Biol. Chem. 2002; 277: 19649-19657Abstract Full Text Full Text PDF PubMed Scopus (240) Google Scholar). Briefly for HTRF assays, the reaction conditions were as follows: a 100-μl reaction volume contains 50 mm Tris, pH 7.9, 50 mm KCl, 1 mm EDTA, 0.5 mm 2-mercaptoethanol, 0.1 mg/ml bovine serum albumin, 800 ng/ml anti-GST-(Eu)K antibody (PerkinElmer), 1 ng/ml GST-PPARγ, 1.5 μg/ml streptavidin conjugated with allophycocyanin (Streptavidin-APC, PerkinElmer), 200 nm biotin-peptide, 5 μl of compound in Me2SO as indicated in figure legends. GST-PPARγ/anti-GST-(Eu)K and biotin-peptide/streptavidin were pair-wise preincubated separately in 20 μl each for 1 h at room temperature before being combined with the remaining components for additional 1 h at room temperature. The reactions were set up in 96-well plates (black polypropylene, Whatman Inc.), and fluorescence was measured on LJL Analyst (LJL Bio-Systems). The data were expressed as the ratio, multiplied by a factor of 1000, of the emission intensity at 665 nm to that at the 620 nm.Mutagenesis was performed using a QuikChange site-directed mutagenesis kit from Stratagene. All of the data in the manuscript are representative of at least two to three independent experiments, and the results are presented as the means of duplicate or triplicate determinations.Adipocyte Differentiation Assays—Isolated human preadipocytes were cultured and induced to differentiate as described by the manufacturer's (ZenBio, NC) protocol, except either rosiglitazone or T2384 was used in place of the PPARγ agonist in the supplied differentiation medium. The lipid content of the cells was measured with Nile Red staining (Molecular Probes #N1142).Protein Preparation—Human PPARγ LBD (residues 206–478) was PCR-cloned into a pET-30 vector (Novagen, WI) with an N-terminal His6 tag, and human RXRα LBD (residues 221–462) was in pET-15b vector (Novagen, WI) with an N-terminal His6 tag as well. Both proteins were expressed in Escherichia coli BL21(DE3) cells (Invitrogen) by growing in LB medium. The proteins were purified by a Ni+-nitrilotriacetic acid-agarose column (Qiagen) and an anion exchange column of Mono-Q (Pharmacia) and then further purified by a gel filtration column of Superdex 75 (Pharmacia). For the heterodimer complex of PPARγ LBD and RXRα LBD, the two proteins were mixed and purified by a cation exchange column of Source 30S (Pharmacia) to remove the excess RXRα. The purified PPARγ LBD protein was concentrated to 1 mg/ml in 20 mm Tris-HCl, pH 7.9, 100 mm NaCl, 2 mm EDTA, and 5 mm dithiothreitol before mixing with 5-fold excess of T2384. The protein-ligand mixture was further concentrated to 5–7 mg/ml for crystallization. The purified heterodimer complex of PPARγ and RXRα was concentrated to 1 mg/ml before mixing with 5-fold excess of T2384 as well as 9-cis-retinoic acid and 2–3-fold excess of a coactivator peptide containing the LXXLL motif derived from helical domain one (TSHKLVQLLTTT) of SRC-1. The mixture was then concentrated to 5–7 mg/ml for crystallization.Crystallization—Cocrystals of wild-type PPARγ LBD with T2384 were grown at 20 °C by either hanging drop or sitting drop with 2.5 μl of the protein solution and 2.5 μl of the well solution containing 22–26% (w/v) polyethylene glycol 4000, 0.1 m Hepes, pH 7.5, and 0.2 m sodium acetate. Cocrystals of two mutant PPARγ LBDs with T2384 were obtained with the heterodimer of PPARγ/RXRα LBDs in the presence of 9-cis-retinoic acid and the coactivator peptide in 0.1 m Tris, pH 9.0, 25–27% polyethylene glycol 2000 monomethyl ether, 0.8 m sodium formate. The crystals were transferred into a well solution that contained an additional 15–20% (w/v) of ethylene glycol and then flash frozen in liquid nitrogen.Data Collection, Structure Determination, and Refinement—X-ray diffraction data sets were collected on synchrotron radiation beam lines 5.0.1 and 5.0.2 at the Advanced Light Source (Berkeley, CA). The data were integrated using either DENZO/SCALEPACK (20Otwinowski Z. Minor W. Methods Enzymol. 1997; 276: 307-326Crossref Scopus (38355) Google Scholar) or MOSFLM (21Leslie A.G.W. Joint CCP4 + ESF-EAMCB Newsletter on Protein Crystallography. 1992; 26Google Scholar). The PPARγ crystals grew in space group C2 with unit cell dimensions a = 92, b = 63, c = 119 Å, β = 105°. The PPARγ/RXRα heterodimer crystals also grew in space group C2 with unit cells of a = 180, b = 54, c = 67 Å, β = 107°.The cocrystal structures of PPARγ LBD with T2384 were solved by molecular replacement using previously published PPARγ coordinates as a search model. The Protein Data Bank code of 2LBD was used for the PPARγ LBD cocrystal structure, and Protein Data Bank code of 1FM9 was used for the heterodimer cocrystal structures. Manual model building was carried out in Quanta (Accelrys), and refinement was carried out with both CNX (22Brunger A.T. Adams P.D. Clore G.M. DeLano W.L. Gros P. Grosse-Kunstleve R.W. Jiang J.S. Kuszewski J. Nilges M. Pannu N.S. Read R.J. Rice L.M. Simonson T. Warren G.L. Acta Crystallogr. Sect. D Biol. Crystallogr. 1998; 54: 905-921Crossref PubMed Scopus (16929) Google Scholar) and REFMAC5 in CCP4 (23Collaborative Computational Project N. Acta Crystallogr. Sect. D Biol. Crystallogr. 1994; 50: 760-763Crossref PubMed Scopus (19703) Google Scholar). The x-ray data collection and refinement statistics are shown in Table 1 (Table 1). Note that the overall heterodimer structures we obtained for the two mutants are similar to the one reported previously (24Gampe Jr., R.T. Montana V.G. Lambert M.H. Miller A.B. Bledsoe R.K. Milburn M.V. Kliewer S.A. Willson T.M. Xu H.E. Mol. Cell. 2000; 5: 545-555Abstract Full Text Full Text PDF PubMed Scopus (510) Google Scholar), especially the RXRα part that contains 9-cis-retinoic acid in the ligand-binding site and the coactivator peptide.TABLE 1Statistics of crystallographic data and refinement Rmsd is the root-mean-square deviation from ideal geometry. The numbers in parentheses are for the highest resolution shell. Rsym = Σ|Iavg – Ij|/ΣIj. R = Σ|Fo – Fc|/ΣFo, where Fo and Fc are the observed and calculated structure factors, respectively. Rfree was calculated from a randomly chosen 10% of reflections excluded form the refinement, and Rcryst was calculated from the remaining 90% of reflections.CrystalWild typeG284IL228W/A292W/L333WWavelength (Å)1.01.01.0Space groupC2C2C2Cell constants (Å)a91.6180.0180.4b62.454.053.5c118.967.166.9β102.6107.7107.3Resolution (Å)2.52.42.3Total reflections21406489826100297Unique reflections226672434527310Completeness (%)94.9 (88.4)100.0 (100.0)100.0 (99.7)I/σ25.6 (1.9)7.9 (1.8)9.5 (3.7)Rsym (%)4.0 (53.9)5.8 (37.5)4.6 (19.4)RefinementRcryst (%)22.626.224.2Rfree (%)27.335.429.4Rmsd bond lengths0.010.020.025Rmsd bond angles1.162.242.44Total non-hydrogen atoms995149014165 Open table in a new tab In Vivo Profiling PPARγ Ligands—KKAy male mice (5–7 weeks of age; Harlan) were prescreened for body weight and a range of clinical parameters (glucose, insulin, leptin, nonesterified free fatty acids, triglycerides, total cholesterol, and high density lipoprotein cholesterol). All blood sampling was performed from the retro-orbital plexus under ether anesthesia. The mice were singly housed and randomized into groups of n = 8/treatment group. Compounds were dissolved in ethanol and added to powdered diet (Purina 5053) and after drying overnight to allow for removal of ethanol, delivered to mice in preweighed food jars (day 1). On day 4 of the study, body weight change and food intake were determined, and blood was sampled and assayed for glucose and insulin. EDTA-treated blood samples were submitted for hematology and serum samples for clinical chemistry (Quality Clinical Labs, Mountain View, CA).RESULTST2384 Partially Activates PPARγ and Antagonizes Rosiglitazone Promoted Adipocyte Differentiation—T2384 is chemically distinct from the thiazolidinedione class of PPARγ agonist (Fig. 1A) but binds to the receptor with an affinity similar to that of rosiglitazone with a Ki of 200 nm (Fig. 1B) and shows specificity toward PPARγ (data not shown). Despite a similar affinity for PPARγ, these two ligands show distinct profiles in cell-based assays for receptor activation. Cell-based reporter gene assays in which HEK293 cells were transiently transfected with an expression construct containing the PPARγ LBD fused to the Gal4-DNA-binding domain, together with a luciferase reporter gene under the transcriptional control of the Gal4 upstream activating sequence (Gal4-UAS), were used to measure ligand-dependent effects on transcription mediated by PPARγ. In these cotransfection assays, rosiglitazone activated transcription by 12-fold, whereas in this same context, T2384 only partially activated the receptor (∼3-fold) with an EC50 value of 0.56 μm. Moreover, T2384 inhibited PPARγ transactivation in the presence of rosiglitazone (Fig. 1C), as expected for a partial agonist. Similar results have also been observed using a full-length receptor together with a luciferase reporter gene under the control of a DR1 sequence (data not shown).PPARγ agonists are known to promote the conversion of a variety of preadipocyte cell lines and primary cells into mature adipocytes. Therefore, we compared the ability of rosiglitazone and T2384 to stimulate triglyceride accumulation in preadipocytes (Fig. 1D). Incubation of isolated human preadipocytes with rosiglitazone resulted in their efficient conversion to adipocytes, as indicated by dramatically increased lipid accumulation in these cells. In contrast, T2384 treatment stimulated very little lipid accumulation in these cells (Fig. 1D). T2384 was able to efficiently antagonize the ability of rosiglitazone to convert the preadipocytes into adipocytes (Fig. 1D), consistent with its effects in the reporter gene assay (Fig. 1C).T2384 Can Adopt Multiple Conformations in the Ligand-binding Pocket of PPARγ—Binding of ligands to PPARγ triggers conformational changes that promote interactions between the receptor and an assortment of coregulatory proteins. These transcriptional regulatory proteins, including both coactivators and corepressors, mediate contact between the PPARγ-RXR heterodimer, chromatin, and the basal transcriptional machinery to promote activation or repression of target gene expression.To study T2384-bound PPARγ conformations and their effects on receptor and transcriptional regulatory protein interactions, a homogeneous HTRF based assay was developed (19Lee G. Elwood F. McNally J. Weiszmann J. Lindstrom M. Amaral K. Nakamura M. Miao S. Cao P. Learned R.M. Chen J.L. Li Y. J. Biol. Chem. 2002; 277: 19649-19657Abstract Full Text Full Text PDF PubMed Scopus (240) Google Scholar) using a panel of peptides derived from these regulatory proteins. Rosiglitazone showed an interaction profile characteristic of full agonists in these assays (Fig. 2). In contrast, T2384 induced a profile that was distinct both from rosiglitazone and a previously described PPARγ antagonist T0070907 (data not shown and Ref. 19Lee G. Elwood F. McNally J. Weiszmann J. Lindstrom M. Amaral K. Nakamura M. Miao S. Cao P. Learned R.M. Chen J.L. Li Y. J. Biol. Chem. 2002; 277: 19649-19657Abstract Full Text Full Text PDF PubMed Scopus (240) Google Scholar). For example, when a peptide derived from coactivator protein DRIP205 was used, T2384 showed a partial agonist profile at concentrations less than 0.1 μm and an antagonist profile at higher concentrations (Fig. 2A). This biphasic profile is also observed when a peptide from the NCoR corepressor protein was used. The first partial agonist phase, where the NCoR peptide is displaced from the receptor, is at concentrations lower than 0.1 μm, and the second antagonist phase, where NCoR peptide is recruited to the receptor, occurred at higher concentrations (Fig. 2B). One possible explanation for these intriguing observations is that T2384 is structurally flexible, adopting multiple conformations with different intrinsic affinities to the ligand-binding pocket of PPARγ. We speculate that these different conformations of T2384 and their unique interactions with the receptor may be responsible for the different activities seen.FIGURE 2T2384 induced unique profile in PPARγ-LBD and coactivator/corepressor interactions in biochemical assays. A, dose response of rosiglitazone and T2384 in an HTRF assay with GST-PPARγ LBD and peptide derived from coactivator DRIP205 protein. B, dose response of rosiglitazone and T2384 in an HTRF assay with GST-PPARγ LBD and peptide derived from corepressor NCoR protein. T2384 displayed biphasic responses in both assays.View Large Image Figure ViewerDownload Hi-res image Download (PPT)More direct evidence for multiple binding conformations came from x-ray crystallographic analysis of T2384-bound PPARγ crystals. We solved the cocrystal structure of T2384 with the PPARγ ligand-binding domain to a resolution of 2.5 Å. Strikingly, the structure shows two crystallographically independent monomers, each with T2384 bound in a different conformation (Fig. 3, A and B). One monomer of PPARγ has one T2384 molecule in the ligand-binding pocket with T2384 wrapping around Cys285 from helix 3 in a U-shaped conformation (Fig. 3C), with B and C rings bind in a similar area as rosiglitazone (17Nolte R.T. Wisely G.B. Westin S. Cobb J.E. Lambert M.H. Kurokawa R. Rosenfeld M.G. Willson T.M. Glass C.K. Milburn M.V. Nature. 1998; 395: 137-143Crossref PubMed Scopus (1658) Google Scholar). In this conformation, the A ring occupies the space between helix 7 and helix 3, which rosiglitzone does not exploit, and makes aromatic stacking interactions with Phe363. Hydrogen bonds are formed between Lys367 in helix 7 and one of the sulfonamide oxygen atoms of T2384, whereas the B ring is sandwiched between Cys285 of helix 3 and Leu330 of helix 5. The region between helix 3 and the β-sheet forms a pocket into which the C ring binds, and this region is defined as the U pocket. The high affinity binding between T2384 and PPARγ results from the single hydrogen bond and numerous hydrophobic interactions between the receptor and the bound ligand. Unlike full agonists, such as the TZDs or tyrosine analogs (17Nolte R.T. Wisely G.B. Westin S. Cobb J.E. Lambert M.H. Kurokawa R. Rosenfeld M.G. Willson T.M. Glass C.K. Milburn M.V. Nature. 1998; 395: 137-143Crossref PubMed Scopus (1658) Google Scholar, 24Gampe Jr., R.T. Montana V.G. Lambert M.H. Miller A.B. Bledsoe R.K. Milburn M.V. Kliewer S.A. Willson T.M. Xu H.E. Mol. Cell. 2000; 5: 545-555Abstract Full Text Full Text PDF PubMed Scopus (510) Google Scholar), T2384 shows no direct hydrogen bond interactions with the AF2 helix. Because stabilization of the AF2 helix in the activated conformation is a key characteristic of full agonism, the lack of direct interaction between this domain of the receptor and T2384 may explain the partial agonist/modulator activities of T2384 observed in vitro.FIGURE 3Structure of PPARγ ligand-binding domain in complex with T2384. A, an overall view of the PPARγ LBD in a dimeric form, shown in ribbon diagram, in the presence of T2384 shown in a sphere model. The coloring scheme is red for oxygen atom, blue for nitrogen, orange for sulfur, green for chlorine, light blue for fluorine, and cyan or pale green for carbon. B, ribbon diagram of one PPARγ LBD monomer structure. The ligand-binding pocket is shown in a gray surface, and T2384 is shown in a stick model. C, stereo view of the ligand-binding pocket of one subunit of the PPARγ homodimer with bound T2384 shown in a stick model. The color scheme is the same as in A and B except the protein molecule has a wheat color for the carbon atom. D, stereo view of the ligand-binding pocket of the other subunit of the PPARγ homodimer bound with two T2384 molecules. The color scheme is the same as in A and B except the protein molecule has a magenta color for the carbon atom. The interactions between the protein and the ligand are highlighted with dashed lines with hydrogen bonds colored in magenta.View Large Image

In recent years, visual analytics (VA) has shown promise in alleviating the challenges of interpreting black-box deep learning (DL) models. While the focus of VA for explainable DL has been mainly on classification problems, DL is gaining popularity in high-dimensional-to-high-dimensional (H-H) problems such as image-to-image translation. In contrast to classification, H-H problems have no explicit instance groups or classes to study. Each output is continuous, high-dimensional, and changes in an unknown non-linear manner with changes in the input. These unknown relations between the input, model and output necessitate the user to analyze them in conjunction, leveraging symmetries between them. Since classification tasks do not exhibit some of these challenges, most existing VA systems and frameworks allow limited control of the components required to analyze models beyond classification. Hence, we identify the need for and present a unified conceptual framework, the Transform-and-Perform framework (T&P), to facilitate the design of VA systems for DL model analysis focusing on H-H problems. T&P provides a checklist to structure and identify workflows and analysis strategies to design new VA systems, and understand existing ones to uncover potential gaps for improvements. The goal is to aid the creation of effective VA systems that support the structuring of model understanding and identifying actionable insights for model improvements. We highlight the growing need for new frameworks like T&P with a real-world image-to-image translation application. We illustrate how T&P effectively supports the understanding and identification of potential gaps in existing VA systems.

GS-9148 (2′-Fd4AP, 4) has been identified as a nucleoside phosphonate reverse transcriptase (RT) inhibitor with activity against wild-type HIV (EC50 = 12 μM). Unlike many clinical RT inhibitors, relevant reverse transcriptase mutants (M184V, K65R, 6-TAMs) maintain a susceptibility to 2′-Fd4AP that is similar to wild-type virus. The 2′-fluorine group was rationally designed into the molecule to improve the selectivity profile and in preliminary studies using HepG2 cells, compound 4 showed no measurable effect on mitochondrial DNA content indicating a low potential for mitochondrial toxicity.

A series of nucleoside phosphonate reverse transcriptase (RT) inhibitors have been synthesized and their anti-HIV activity and resistance profiles evaluated. The most potent analog [5-(6-amino-purin-9-yl)-2,5-dihydro-furan-2-yloxymethyl]-phosphonic acid (d4AP) demonstrated a HIV EC50 = 2.1 μM, and the most favorable resistance profile against HIV-1 variants with K65R, M184V or multiple thymidine analog mutations in RT.

Phosphonomethoxy nucleoside analogs of the thymine containing nucleoside reverse transcriptase inhibitors (NRTIs), 3′-azido-2′,3′-dideoxythymidine (AZT), 2′,3′-didehydro-2′,3′-dideoxythymidine (d4T), and 2′,3′-dideoxythymidine (ddT), were synthesized. The anti-HIV activity against wild-type and several major nucleoside-resistant strains of HIV-1 was evaluated together with the inhibition of wild-type HIV reverse transcriptase (RT). Phosphonomethoxy analog of d4T, 8 (d4TP), demonstrated antiviral activity with an EC50 value of 26 μM, whereas, phosphonomethoxy analogs of ddT, 7 (ddTP), and AZT, 6 (AZTP), were both inactive at concentrations up to 200 μM. Bis-isopropyloxymethylcarbonyl (bisPOC) prodrugs improved the anti-HIV activity of 7 and 8 by >150-fold and 29-fold, respectively, allowing for antiviral resistance to be determined. The K65R RT mutant virus was more resistant to the bisPOC prodrugs of 7 and 8 than bisPOC PMPA (tenofovir DF) 1. However, bisPOC prodrug of 7 demonstrated superior resistance toward the RT virus containing multiple thymidine analog mutations (6TAMs) indicating that new phosphonate nucleoside analogs may be suitable for targeting clinically relevant nucleoside resistant HIV-1 strains.

Modified purine analogs of GS-9148 [5-(6-amino-purin-9-yl)-4-fluoro-2,5-dihydro-furan-2-yloxymethyl]-phosphonic acid (2'-Fd4AP) were synthesized and their anti-HIV potency evaluated. The antiviral activity of guanosine analog (2'-Fd4GP) was comparable that of to 2'-Fd4AP in MT-2 cells, but selectivity was reduced.

Deep learning (DL) models have shown performance benefits across many applications, from classification to image-to-image translation. However, low interpretability often leads to unexpected model behavior once deployed in the real world. Usually, this unexpected behavior is because the training data domain does not reflect the deployment data domain. Identifying a model's breaking points under input conditions and domain shifts, i.e., input transformations, is essential to improve models. Although visual analytics (VA) has shown promise in studying the behavior of model outputs under continually varying inputs, existing methods mainly focus on per-class or instance-level analysis. We aim to generalize beyond classification where classes do not exist and provide a global view of model behavior under co-occurring input transformations. We present a DL model-agnostic VA method (ProactiV) to help model developers proactively study output behavior under input transformations to identify and verify breaking points. ProactiV relies on a proposed input optimization method to determine the changes to a given transformed input to achieve the desired output. The data from this optimization process allows the study of global and local model behavior under input transformations at scale. Additionally, the optimization method provides insights into the input characteristics that result in desired outputs and helps recognize model biases. We highlight how ProactiV effectively supports studying model behavior with example classification and image-to-image translation tasks.

Cyclic phosphonomethoxy pyrimidine nucleosides that are bioisosteres of the monophosphate metabolites of HIV reverse transcriptase (RT) inhibitors AZT, d4T, and ddC have been synthesized. The RT inhibitory activities of the phosphonates were reduced for both dideoxy (dd) and dideoxydidehydro (d4) analogs compared to the nucleosides. Bis-isopropyloxymethylcarbonyl (BisPOC) prodrugs were prepared on selected compounds and provided > 150-fold improvements in antiviral activity.

<p>Deep learning (DL) models have shown performance benefits across many applications, from classification to, more recently, image-to-image translation. However, low interpretability often leads to unexpected model behavior when deployed in the real world. Usually, this unexpected behavior is because the training data domain does not reflect the deployment data domain. Identifying model breaking points under input conditions and domain shifts, i.e., input transformations, is essential to improve models. Although visual analytics (VA) has shown promise in studying the behavior of model outputs under continually varying inputs, existing methods mainly focus on per class or instance-level analysis. We aim to generalize beyond classification where classes do not exist and provide a global holistic model behavior overview under co-occurring input transformations. We present a DL model-agnostic VA method (ProactiV ) to help model developers proactively study output behavior under input transformations to identify and verify breakpoints. ProactiV relies on a proposed input optimization method to determine how far a given transformed input has to move in the input data manifold to achieve the desired output. The data from this optimization process allows the study of global and local model output behavior under input transformations at scale. Additionally, the optimization method provides insight into the input characteristics that result in desired outputs and helps recognize model biases. We highlight how ProactiV effectively supports studying model behavior under input transformations to identify and verify model breakpoints with example classification and image-to-image translation tasks.</p>

<p>Citation Information: V. Prasad, R. J. G. v. Sloun, A. Vilanova and N. Pezzotti, "ProactiV: Studying Deep Learning Model Behavior under Input Transformations," in IEEE Transactions on Visualization and Computer Graphics, doi: 10.1109/TVCG.2023.3301722.</p> <p><br></p> <p>© 2023 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.</p> <p><br></p> <p>Code is available here: https://github.com/vidyaprsd/proactiv</p>

<p>Citation Information: V. Prasad, R. J. G. v. Sloun, A. Vilanova and N. Pezzotti, "ProactiV: Studying Deep Learning Model Behavior under Input Transformations," in IEEE Transactions on Visualization and Computer Graphics, doi: 10.1109/TVCG.2023.3301722.</p> <p><br></p> <p>© 2023 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.</p> <p><br></p> <p>Code is available here: https://github.com/vidyaprsd/proactiv</p>

Diffusion models have garnered significant attention since they can effectively learn complex multivariate Gaussian distributions, resulting in diverse, high-quality outcomes. They introduce Gaussian noise into training data and reconstruct the original data iteratively. Central to this iterative process is a single Unet, adapting across time steps to facilitate generation. Recent work revealed the presence of composition and denoising phases in this generation process, raising questions about the Unets' varying roles. Our study dives into the dynamic behavior of Unets within denoising diffusion probabilistic models (DDPM), focusing on (de)convolutional blocks and skip connections across time steps. We propose an analytical method to systematically assess the impact of time steps and core Unet components on the final output. This method eliminates components to study causal relations and investigate their influence on output changes. The main purpose is to understand the temporal dynamics and identify potential shortcuts during inference. Our findings provide valuable insights into the various generation phases during inference and shed light on the Unets' usage patterns across these phases. Leveraging these insights, we identify redundancies in GLIDE (an improved DDPM) and improve inference time by ~27% with minimal degradation in output quality. Our ultimate goal is to guide more informed optimization strategies for inference and influence new model designs.

<p>In recent years, visual analytics (VA) has shown promise in alleviating the challenges of interpreting black-box deep learning (DL) models. While the focus of VA for explainable DL has been mainly on classification problems, DL is gaining popularity in high-dimensional-to-high-dimensional (H-H) problems such as image-to-image translation. In contrast to classification, H-H problems have no explicit instance groups or classes to study. Each output is continuous, high dimensional, and changes in an unknown non-linear manner with changes in the input. These unknown relations between the input, model and output necessitate the user to analyze them in conjunction, leveraging symmetries between them. Since classification tasks do not exhibit some of these challenges, most existing VA systems and frameworks allow limited control of the components required to analyze models beyond classification. Hence, we identify the need for and present a unified conceptual framework, the Transform-and-Perform framework (T&P), to facilitate the design of VA systems for DL model analysis focusing on H-H problems. T&P provides guidelines to structure and identify workflows and analysis strategies to design new VA systems, and understand existing ones to uncover potential gaps for improvements. The goal is to aid the creation of effective VA systems that support the structuring of model understanding and identifying actionable insights for model improvements. We highlight the growing need for new frameworks like T&P with a real-world image-to-image translation application. We also illustrate how T&P effectively supports the understanding and identifying potential gaps in existing VA systems.</p>

<p>Citation Information: V. Prasad, R. van Sloun, S. van den Elzen, A. Vilanova, and N. Pezzotti, “The Transform-and-Perform framework: Explainable deep learning beyond classification,” IEEE Transactions on Visualization and Computer Graphics, 2022. https://doi.org/10.1109/TVCG.2022.3219248</p> <p><br></p> <p>© 2022 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.</p>

The stereospecific interaction of a neutral probe molecule, acetonide, with immobilized α1-acid glycoprotein (AGP) was investigated. Enantioselectivity was found to be influenced by the choice of organic modifier, temperature, and pH. These parameters could be varied to the extent that a reversal of elution order could be induced. Our studies found that the role of hydrogen bonding in the chiral discrimination of acetonide was minimal. An inclusion mechanism is proposed with the investigated parameters affecting the access to the binding sites either through induced conformational changes or steric hindrance.

Abstract ChemInform is a weekly Abstracting Service, delivering concise information at a glance that was extracted from about 200 leading journals. To access a ChemInform Abstract of an article which was published elsewhere, please select a “Full Text” option. The original article is trackable via the “References” option.

<p>Citation Information: V. Prasad, R. van Sloun, S. van den Elzen, A. Vilanova, and N. Pezzotti, “The Transform-and-Perform framework: Explainable deep learning beyond classification,” IEEE Transactions on Visualization and Computer Graphics, 2022. https://doi.org/10.1109/TVCG.2022.3219248</p> <p><br></p> <p>© 2022 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.</p>