Discovery by Virtual Screening of an Inhibitor of CDK5-Mediated PPARγ Phosphorylation

Abstract

Thiazolidinedione PPARγ agonists such as rosiglitazone and pioglitazone are effective antidiabetic drugs, but side effects have limited their use. It has been posited that their positive antidiabetic effects are mainly mediated by the inhibition of the CDK5-mediated Ser273 phosphorylation of PPARγ, whereas the side effects are linked to classical PPARγ agonism. Thus compounds that inhibit PPARγ Ser273 phosphorylation but lack classical PPARγ agonism have been sought as safer antidiabetic therapies. Herein we report the discovery by virtual screening of 10, which is a potent PPARγ binder and in vitro inhibitor of the CDK5-mediated phosphorylation of PPARγ Ser273 and displays negligible PPARγ agonism in a reporter gene assay. The pharmacokinetic properties of 10 are compatible with oral dosing, enabling preclinical in vivo testing, and a 7 day treatment demonstrated an improvement in insulin sensitivity in the ob/ob diabetic mouse model.

The peroxisome proliferator-activated receptors (PPARα, PPARβ/δ, and PPARγ), are nuclear receptors that function as ligand-dependent transcriptional regulators of multiple genes involved in adipogenesis, insulin sensitization, and lipid metabolism.¹ PPARγ is mostly expressed in adipose tissue and has been referred to as the master regulator of adipogenesis, and its activation is associated with beneficial metabolic effects. It has therefore been a prominent target for the pharmaceutical development of insulin-sensitizing and antidiabetic agents.²⁻⁴ The most important synthetic PPARγ agonists are the thiazolidinediones (TZDs), represented by the approved drugs rosiglitazone 1 and pioglitazone 2.⁵ These compounds are highly efficacious in treating insulin resistance and type 2 diabetes, but their use has been associated with a number of side effects, primarily edema and weight gain.⁶ Although 1 and 2 are structurally very similar, they exhibit different clinical safety profiles. Concerns regarding the cardiovascular safety of 1 led to its removal from the market in the EU,⁷ whereas 2 has been reported to reduce the risk of major cardiac adverse events in type 2 diabetic patients.⁸ Compound 2 has been associated with an increased risk of bladder cancer⁹ (although this is contested by recent reports¹⁰), which is absent for 1.¹¹ These findings have stimulated research to develop new PPARγ modulators that preserve the strong antidiabetic effect of TZDs but reduce or eliminate side effects.

In particular, selective PPARγ modulators (SPPARγMs) or PPARγ partial agonists have been investigated for this purpose.^2,3,12,13 This class is quite heterogeneous in terms of structure and pharmacological profile. It has been hoped that partial PPARγ agonists (defined as possessing <50% agonist efficacy compared with rosiglitazone in a reporter gene assay¹⁴) would possess fewer side effects than full agonists. Preclinical results have been variable with this class, and no PPARγ partial agonist is currently approved for clinical use in a diabetes indication.

The mechanism of PPARγ activation by full agonists such as TZDs 1 and 2 is described as a molecular switch composed of the distal carboxyl terminal helix 12 (H12) of the PPARγ ligand binding domain (LBD). Typically, a TZD, carboxylic acid, or other polar moiety of the ligand forms an extensive hydrogen-bond network with Ser289, His323, His449, and Tyr473 and thereby mediates the stabilization and positioning of H12 in a conformation that allows it to dock against H3 and H11, thus forming a key part of the coactivator-binding site, the so-called AF2 surface.¹⁵ On the contrary, partial PPARγ agonists (such as MRL24 3,¹⁶ nTZDpa 4,¹² INT-131 5,¹⁷ and BVT.13 6;¹⁸Figure 1) bind to a region of the PPARγ ligand binding pocket that is distant from H12 and activate the receptor allosterically, independently of a direct interaction with H12 (Figure 2). As a consequence, these ligands cause a lower degree of H12 stabilization, which has been proposed to lead to a differential recruitment of cofactors and consequently affect the transcriptional profile of the activated receptor–ligand complex.¹⁴

Figure 1.

Representative PPARγ ligands: full agonists (1 and 2) and partial agonists (3–6) and inhibitors of PPARγ phosphorylation (7–9).

Figure 2.

Space-filling models of full agonist 1 (red, panel A) and partial agonist 6 (green, panel B) in complex with the PPARγ LBD, exemplifying the lack of H12 interaction that is characteristic of PPARγ partial agonists (generated from PDB 2PRGand2Q6S).

Several PPARγ partial agonists (for example, GQ-16¹⁹) have been reported to maintain the beneficial insulin-sensitizing effects of full agonists but with fewer side effects, but this has not resulted in approved drugs exploiting this mode of action.

In 2010, a molecular mechanism was reported by which partial and full PPARγ agonists act to improve insulin sensitivity not by the classical recruitment and stabilization of H12²⁰ and activation of a large number of PPARγ target genes but rather via inhibition of the CDK5-mediated phosphorylation of PPARγ2 Ser273 (Ser245 in PPARγ1; PPARγ2 numbering is used throughout in the text and figures).

With the use of hydrogen/deuterium exchange–mass spectroscopy (HDX-MS), Ser273 phosphorylation inhibition was linked to increased ligand-induced stabilization of the region encompassing the Ser273-containing H2–H2′ loop. The potential combination of potentially fewer undesirable side effects of a partial agonist/modulator together with the claimed antidiabetic effect of Ser273 phosphorylation inhibition sparked interest in generating additional PPARγ ligands exhibiting this profile. This led to the discovery of SR1664 7 and later UHC-1 8(21−23) and GQ-16 9.¹⁹

During our target validation activities, we observed that 7 was not tolerated in ob/ob mice (data not shown), and we therefore sought to identify structurally unrelated inhibitors of PPARγ Ser273 phosphorylation with minimal PPARγ agonism. Herein we report the identification of compound 10 (Scheme 1), a potent PPARγ binder that inhibits Ser273 phosphorylation in vitro but displays essentially no PPARγ agonism. We also describe its subsequent in vivo characterization in the ob/ob diabetic mouse model.

Scheme 1. Synthesis of 10.

Reagents and conditions: (a) (i) KMnO₄, K₂CO₃, H₂O, reflux, overnight, (ii) hydrazine sulfate, H₂O, reflux, 2 h, 32% over two steps; (b) (i) conc. H₂SO₄, EtOH, reflux, overnight, (ii) 4-(bromomethyl)-1,2-dichlorobenzene, Cs₂CO₃, MeCN, 2 h, (iii) 2 M aq NaOH, THF rt, overnight, 68% over three steps.

Our hit finding approach was a targeted structure-based virtual screen (VS). Previous groups have shown PPARγ to be a suitable target for structure-based VS²⁴⁻²⁶ and importantly also shown the possibility of targeting the selection of compounds toward partial agonists in contrast with full agonists.²⁷ Our approach was based on the assumption that full agonists interact and stabilize the AF2 region, whereas partial agonists do not, resulting in diminished conformational stabilization of the AF2 surface as compared with full agonists.¹⁴

The VS workflow was applied, targeting the shape and volume of the area occupied by the partial PPARγ agonists 4 and 6 (as observed in their PPARγ LBD complex structures, PDB IDs 2Q5Sand2Q6S, respectively; see the Supporting Information). These two ligands fulfill the criterion of not interacting with the AF2 region but instead occupying the distal part of the binding pocket close to the β-sheet. In addition, by occupying nonoverlapping volumes of the binding pocket, they increase the volume of the pocket probed and thus increase the potential hit rate.

The VS was performed on the AstraZeneca corporate collection and consisted of two different sequentially applied docking techniques. First, the fast 3D shape-based docking software FRED²⁸ was used followed by a more accurate docking algorithm using the software Glide SP.²⁹ Ranking by Glide score and visual inspection resulted in 2943 compounds being selected for single replicate concentration–response testing in both a human PPARγ time-resolved fluorescence energy transfer (TR-FRET) binding assay and a PPARγ reporter gene agonist assay.³⁰

201 compounds were defined as active (PPARγ binding affinity <1 μM and PPARγ agonist efficacy <25%), leading to a hit rate of 7%. Further in silico profiling of physicochemical and DMPK (drug metabolism and pharmacokinetics) properties using in-house derived models³¹ (including predicted logD < 3, aqueous solubility >100 μM, and Hu mics CL_int < 10 μL/min/mg) led to the identification of 10 as the most promising compound, representing a novel class of PPARγ modulators. Compound 10 is a 24 nM binder to human PPARγ but displays no PPARγ agonism at concentrations below 10 μM (Table 1 and Figure 3) and significantly lower PPARγ agonist potency (EC₅₀ > 50 μM) compared with 7 (EC₅₀ 726 nM; see Table 1 and Figure S4).

Table 1. In Vitro PPARγ Profile of Compounds 1, 7, and 10.

compound	1	7	10
PPARγ binding IC50 (nM)a	90 ± 27	154 ± 12	24 ± 14
PPARγ GAL4 reporter gene EC50 (nM)/max agonist effectb (%)a	27 ± 16/86 ± 17	726c/28 ± 7	>50 000/27 ± 7
PPARγ pS273 IC50 (nM)a	4,400 ± 820	580 ± 44	160 ± 6

^aValues presented as the mean ± standard deviation (n ≥ 3).

^bAgonist effect at the highest assay concentration (50 μM).

^cEstimated EC₅₀. (See the Supporting Information.)

Figure 3.

Concentration–response data for 10 in the human PPARγ reporter gene agonist (green circles, n = 4, no curve fitted to data) and binding (blue squares, n = 15) assays. Error bars indicate the SD from the mean of individual assay occasions for the reporter gene assay and the SD from the mean of aggregated data from all screening occasions for the binding assay.

Next, 10 was evaluated for its ability to inhibit the CDK5-mediated Ser273 phosphorylation of full-length human PPARγ.³⁰ Compound 1 inhibited CDK5-mediated PPARγ Ser273 phosphorylation in a concentration-dependent manner (Table 1). Importantly, 10 was shown to be an inhibitor of PPARγ Ser273 phosphorylation, with a potency greater than that of 1 or 7. In addition, 10 was shown not to inhibit CDK5 kinase activity in two separate in vitro CDK5 activity assays. (See the Supporting Information.)

To enable to perform further characterization, we resynthesized 10, as shown in Scheme 1. The methyl groups of compound 11(32) were oxidized using potassium permanganate to afford the corresponding dicarboxylic acid, and subsequent cyclization with hydrazine afforded the corresponding phthalazine derivative 12 in 32% yield. A sequence of esterification, N-alkylation, and ester hydrolysis resulted in the final product 10 in 68% yield over three steps. The physicochemical and DMPK properties of 10 are shown in Table 2.

Table 2. Physicochemical and in Vitro DMPK Profile of 10.

clogP/LogD7.4	4.9/2.3 (n = 3)
aqueous solubility (from dried DMSO solution)	324 μM (n = 4)
mouse/hu ppb (% unbound)	0.022/<0.1 (n = 2 mouse, n = 4 hu)
Hu Mics/heps CLint	<4 μL/min/mg (n = 4)/15.2 μL/min/1 × 10–6 (n = 1)
rat/mouse heps CLint (μL/min/1 × 10–6)	8.8 (n = 3)/10.0 (n = 2)
Caco2 permeability (1 × 10–6 cm/s)	13 (n = 2)
GSH conjugates (Hu mics)	not detected
CYP inhibition profile	2C9: IC50 = 4.6 μM; inactive on 1A2, 2C19, 2C8, 2D6, 3A4

The in vitro PPARγ profile of 10, with minimal agonism and potent inhibition of phosphorylation at Ser273, was in good agreement with the binding mode probed by the VS. To confirm that 10 bound in a similar fashion to other PPARγ partial agonists and did not make agonism-favoring interactions with H12, we generated a 2.2 Å resolution X-ray crystal structure of 10 bound to the PPARγ ligand binding domain. (See the Supporting Information.)

Gratifyingly for the chosen VS strategy, the overall structure of the PPARγ LBD-10 complex is identical to the docked pose (Figure S2) and reminiscent of the PPARγ-6 and 7 structures, and 10 recapitulates their key pharmacophoric elements with good shape complementarity (Figure 4). As seen in Figure 4, 10 binds remotely from H12, in agreement with its very low level of agonist activity. It also accepts a H-bond from the backbone NH of Ser370, an interaction exploited by many PPARγ partial agonists.¹⁴

Figure 4.

(A) X-ray crystal structure of the PPARγ LBD in complex with 10 (orange, PDB7QB1), showing a lack of interaction with PPARγ H12 and proximity to the H2–H2′ loop containing Ser273 (highlighted in red). (B) Close-up of the binding mode of 10 showing the H-bond interaction (dotted purple line) between the carboxylate of 10 and the Ser370 backbone NH.

In agreement with both the X-ray crystal structure and the negligible PPARγ agonism observed for 10, no ligand-induced stabilization of H12 of the PPARγ LBD was observed in a differential HDX-MS experiment (Figure 5). (See the Supporting Information.) Stabilization of regions directly lining the binding site of 10 (H3, the C-terminus of H5 and the β-sheet) was observed, and longer range stabilization of H2, H7, and the Ser273-containing H2–H2′ loop was also observed.

Figure 5.

Differential HDX-MS data of the 10-induced stabilization of the PPARγ LBD (mapped onto PDB 2PRG). Deeper shades of blue indicate greater ligand-induced stabilization, and regions of the protein for which no peptides were detected in MS are colored in tan. Compound 10 is displayed as a green space-filling model.

SR1664 7 was run as a comparator under identical HDX-MS conditions. 7 induced less stabilization of the H2–H2′ loop compared with 10 (see Figure S3), which is consistent with its lower pS273 inhibition potency (Table 1).

Confident that 10 possessed the in vitro profile we sought (i.e., potent PPARγ binding, minimal PPARγ agonism, and inhibition of PPARγ Ser273 phosphorylation), we continued toward profiling the compound in the ob/ob diabetic mouse model (Figure 6). A comprehensive secondary pharmacology profile of 10 (see the Supporting Information) showed no hits considered stoppers for continued preclinical profiling. In vivo pharmacokinetic (PK) profiling (100 μmol/kg by oral gavage) in lean C57/BL6 mice was carried out, with the compound exhibiting excellent oral bioavailability (109%) and total C_max (167 μM) and reasonable t_1/2 (5 h). The ob/ob diabetic mouse study used two dose groups of 10 and positive control 1 (10 and 100 μmol/kg/day, dosing by oral gavage) and vehicle and lean controls.

Figure 6.

In vivo pharmacodynamic (PD) effects of 10 in ob/ob mouse model of diabetes. Compounds 1 and 10 were dosed by oral gavage at 10 and 100 μmol/kg/day for 7 days. Effects on (A) terminal fasting plasma glucose, (B) insulin, and (C) body weight. 100 μmol/kg/day of 10 resulted in reduced plasma insulin, corresponding to an increase in insulin sensitivity. Data are presented as the mean ± SEM. Data normality was assessed using the D’Agostino and Pearson test. Statistical analysis was performed using one-way ANOVA followed by Dunnett’s multiple comparison’s test between the ob/ob vehicle group and the substance-treated groups. For data sets that did not meet equality of variances, statistical analysis was performed using the Brown–Forsythe test and Welch’s one-way ANOVA followed by Dunnett’s T3 multiple comparison test. A value of P < 0.05 was considered statistically significant.

After 7 days of dosing, both 1-treated groups showed the expected normalization of terminal fasting plasma glucose and insulin levels matching those of the lean counterparts (Figure 6A,B). In contrast, the effects of 10 on glucose and insulin were more modest, with only the 100 μmol/kg dose group showing a significant reduction in plasma insulin and a trend toward a reduction in plasma glucose, which indicates an overall increase in insulin sensitivity. Under these conditions, both 1 and 10 were well tolerated and did not impact the body weight (Figure 6C).

In summary, we have described the identification by VS of PPARγ modulator 10, a novel, high-affinity binder to PPARγ, which possesses minimal PPARγ agonism and exhibits potent in vitro inhibition of the CDK5-mediated phosphorylation of isolated full-length PPARγ. Compound 10 has a PK profile amenable to in vivo administration, and dosing in ob/ob mice demonstrated a modest improvement of insulin sensitivity, which can be used as a starting point for the further development of novel insulin sensitizers for the treatment of metabolic disorders.

Experimental Procedures

See the Supporting Information for the details of all experimental procedures, including the synthesis and characterization of 10.

Glossary

Abbreviations

AF-2

activation function-2

Cl

clearance

C_max

maximum concentration

CYP

cytochrome P₄₅₀

DMPK

distribution, metabolism, and pharmacokinetics

GSH

glutathione

HDX-MS

hydrogen/deuterium exchange–mass spectroscopy

heps

hepatocytes

HOMA-IR

Homeostatic Model Assessment for Insulin Resistance

Hu

human

IC₅₀

half-maximal inhibitory concentration

LBD

ligand binding domain

mics

liver microsomes

PK

pharmacokinetic

PPAR

peroxisome proliferator-activated receptor

ppb

plasma protein binding

TR-FRET

time-resolved fluorescence energy transfer

TZD

thiazolidinedione

VS

virtual screen

V_ss

steady-state volume of distribution

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsmedchemlett.1c00715.

• Experimental methods including the synthesis and characterization of 10, in vivo ob/ob study methods and data, crystallographic information, VS procedure, and secondary pharmacology profile of 10 (PDF)

• Deuterium uptake data for 7 and 10 (XLSX)

• HDX data table for 7 and 10 (XLSX)

Author Present Address

^△ D. E. Shaw Research, 120W 45th St., New York, NY 10036, United States

The authors declare the following competing financial interest(s): All authors were employees of AstraZeneca and may have held AZ stock at the time the work was performed.

Notes

The atomic coordinates for the crystal structure of compound 10 in complex with human PPARγ LBD have been deposited in the Protein Data Bank (PDB) (www.rcsb.org/pdb/home/home.do) with accession code 7QB1.