Identification of novel indole derivatives acting as inhibitors of the Keap1–Nrf2 interaction

Abstract Nine indole derivatives (9a-i) were tested as potential inhibitors of the Keap1–Nrf2 interaction. This class of compounds increases the intracellular levels of the transcription factor Nrf2 and the consequent expression of enzymes encoded by genes containing the antioxidant response element (ARE). In the ARE-luciferase reporter assay only 9e-g revealed to be remarkably more active than t-butylhydroxyquinone (t-BHQ), with 9g standing out as the best performing compound. While 9e and 9f are weak acids, 9g is an ampholyte prevailing as a zwitterion in neutral aqueous solutions. The ability of 9e-g to significantly increase levels of Nrf2, NADPH:quinone oxidoreductase 1, and transketolase (TKT) gave further support to the hypothesis that these compounds act as inhibitors of the Keap1–Nrf2 interaction. Docking simulations allowed us to elucidate the nature of the putative interactions between 9g and Keap1.


Introduction
The interaction between Kelch-like ECH-associated protein 1 (Keap1) 1 and the Nuclear Factor Erythroid 2-Related Factor 2 (Nrf2) 2 plays a crucial role in the homeostasis of cellular oxidative stress 3 . Under physiological conditions, the activity of Nrf2 as a transcription factor is negatively regulated by Keap1 through proteasomal degradation mechanisms 4 . An increase of the intracellular levels of reactive oxygen species (ROS) interferes with the stability of the Keap1-Nrf2 complexsituated in the cytoplasmby oxidation of cysteine residues located in a specific domain of Keap1. Disruption of the stability of the Keap1-Nrf2 complex triggers the release of Nrf2, allowing this protein to reach the nucleus where it behaves as a transcriptional activator of genes that contain the enhancer sequence antioxidant response element (ARE) 5 . As a result, several enzymes, such as NADPH:quinone oxidoreductase 1 (NQO1), heme oxygenase-1 (HO-1), glutathione S-transferase (GST), superoxide dismutase (SOD), catalase (CAT), and transketolase (TKT) are overexpressed [6][7][8] . These antioxidant enzymes reduce oxidative stress and, consequently, limit cellular damages. Thus, the Keap1-Nrf2-ARE system plays a pivotal role in cellular metabolism and redox balance. For the above reasons, inhibitors of the Keap1-Nrf2 interaction are currently being investigated as potential drugs to treat diseases involving chronic oxidative stress, such as diabetes, cancer and neurodegenerative disorders 9 .
The first non-peptide small molecules disrupting the Keap1-Nrf2 interaction by binding to Keap1 with micromolar affinities were reported by Hu et al. 10 and Marcotte et al. 11 using high throughput screening methods. This latter group also solved the 3D structures of two ligand-Keap1 complexes by X-ray crystallography 11 . The above quoted works gave impetus to additional X-ray diffraction studies and researches aimed at identifying novel ligands of Keap1 12 .
According to Jiang et al. 12 , the Keap1 binding cavity hosting inhibitors of the Keap1-Nrf2 interaction can be divided into six subpockets (P1-P6). P1 and P2 contain protonated arginine residues (R483, R415, R380) which give rise to strong electrostatic interactions with electron-rich parts of ligands; specifically, salt bridges with carboxylate groups, H-bonds with nitro oxygens or azole nitrogens, cation-p contacts with aromatic rings. Additional polar as well as hydrophobic interactions are established between ligands and the complementary Keap1 binding cavity. Some inhibitors of the Keap1-Nrf2 interaction, representative of the variety of a huge number of those reported in literature [10][11][12][13][14][15][16] are reported in Chart 1. They typically contain in their structures a planar or quasi-planar scaffold bearing at least one aromatic ring involved in cation-p interactions.
Our research group has been working for several years on different series of indole derivatives each of them binding with high potency and selectivity to the benzodiazepine receptor 17 , the A 2B adenosine receptor 18 and the translocator protein 19 . Given the commonly ascertained utility of indole as a scaffold for drug-like molecules (see compound 4 in Chart 1), we tried to identify novel indole derivatives which could act as inhibitors of the Keap1-Nrf2 interaction with the help of molecular modelling and substructure search methods. Specifically, a few indole-bearing models were superimposed on available 3D structures of inhibitors of the Keap1-Nrf2 interaction co-crystallised with Keap1. Starting from some designed indole derivatives we identified commercially available compounds using the SciFinder 20 substructure search routine. Examples of the above approaches are given in the Supporting Information. The present paper describes the results of our studies.

Chemistry
Nine indole derivatives fulfilling the pharmacophoric requirements to act as inhibitors of the Keap1-Nrf2 interaction 12 were selected for biological evaluation (9a-i reported in Chart 2). Compounds 9a-d were synthesised (see Supporting Information), whereas 9e-i were purchased from AKos (AKos GmbH, Steinen, Germany).
Based on their acid-basic properties, these compounds can be divided into three groups: (a) non-ionisable (9a-d); (b) acidic (9e, f); ampholytic (9g-i). We reasoned that the presence of methoxy group(s) or a methylendioxy moiety on the benzene rings in the structures of the selected compounds might allow them to interact with Keap1 through cation-p interactions and H-bonds. The thiophene ring featured by 9e-g confers more conformational rigidity to the above compounds and an electron-rich ring capable to establish cation-p interactions.

Biology
The biology experiments are detailed in the Supporting Information. To evaluate the capability of the indole derivatives 9a-i to inhibit the Keap1-Nrf2 interaction, we performed a cellbased luciferase reporter assay in which induction of ARE-driven luciferase activity is mediated by Nrf2. HeLa cells were transiently transfected with ARE-luciferase reporter plasmids and treated with all compounds at the concentration of 10 lM, except for 9g which was tested at the concentration of 5 lM owing to its limited solubility in phosphate buffer. t-Butylhydroxyquinone (t-BHQ), one of the canonical activators of Nrf2 12,21,22 , was employed as a positive Chart 1. Structures of representative inhibitors of the Keap1-Nrf2 interaction. For each compound a reference is reported in parentheses. Compounds 6 and 8 exhibit nanomolar affinities for Keap1, whereas the remaining compounds bind to Keap1 with micromolar affinities.
control at the concentration of 50 mM, a value which induced in our experiments the maximum luciferase activity.
Compounds 9e, 9f, and 9g increased luciferase activity by 152, 263, and 486%, respectively (Figure 1). Compared with the activity exhibited by t-BHQ (48% increase), the activities of 9e-g were higher by 3.2-, 5.5-, and respectively, 10.1-folds. The remaining compounds displayed poor activities in this assay, comparable or lower with respect to t-BHQ. The best performing compounds 9e-g share a common thiophene-carboxylic moiety. Among this subset, 9e and 9f are acidic, whereas 9gby far the most active compoundis ampholytic.
To assess the inhibition of the Keap1-Nrf2 interaction exerted by 9e-g, we evaluated the expression of Nrf2 and of two enzymes encoded by its downstream target ARE genes, namely NQO1 and TKT. Hela cells were treated with 9e and 9f at the concentration of 10 mM while 9g was employed at the concentration of 5 mM. We used t-BHQ as a reference compound at the concentration of 50 mM. As shown in Figure 2, expression of NQO1 and of TKT remarkably increased after treatment with 9e-g. In Figure 2, it can be appreciated a significant increase in Nrf2 levels determined by our indole derivatives, clearly much higher compared to that produced by t-BHQ. The original picture of the Western blot gel is reported in the Supporting Information.
Compounds 9e-g, selected as the most active indole derivatives in the luciferase assay, were evaluated for their cytotoxicity after 72 h of incubation in human peripheral blood lymphocytes (PBLs) ( Table S1). The compounds were tested either in quiescent and in proliferating PBLs (proliferation was induced by phytohematoagglutinin (PHA) as a mitogenic stimulus).
The obtained data indicate that the above compounds can be considered safe for human cells. In the absence of PHA, 9e, f exhibited GI 50 values greater than 100 mM, while 9g showed GI 50 values greater than 5 mM (higher concentrations of this compound could not be used due to its limited solubility in phosphate  buffer). Compounds 9e, f slightly increased their cytotoxicity in proliferating PBLs showing GI 50 values of 61.0 lM and, respectively, 84.9 mM. Compound 9g killed 6 and 10% of quiescent and, respectively, proliferating PBLs.

Molecular modelling
The potential interactions of 9g with Keap1 were modelled in silico by docking simulations. Details of these calculations are available in the Supporting Information. Compound 9g was treated as a zwitterion, as this species prevails to a significant extent either in neutral aqueous solutions or in the essentially polar environment of the Keap1 binding cavity. As represented in Figures 3  and 4, in our docking model the ligand occupies the central cavity of Keap1 with the carboxylate group pointing outwards and the rest of the molecule extending into the central channel of the protein. Based on the partition of the Keap1 binding cavity proposed by Jiang et al. 12 into subpockets P1-P6, 9g seemingly occupies mainly P1 and part of P2 and P3, taking into account that arginine 415 is located at the border of P1 and P2.
Several strong electrostatic ligand-protein interactions take place: the carboxylate of 9g forms a salt bridge with the arginine 483 side chain and a charge-reinforced H-bond with the P1 serine 508 side chain; the thiophene ring of the ligand establishes a cation-p interaction with the arginine 415 side chain (subpockets P1 and P2). Weaker hydrophobic interactions contribute to the ligand-protein affinity through favorable contacts between the indole nucleus of the ligand with the P3 alanine 556 methyl group and the arginine 415 dimethylene fragment.
The (m-methoxy)benzylaminomethyl substituent, in its protonated state, points deep down into the central Keap1 channel where it establishes hydrophobic interactions with valine 512 and leucine 472 side chains and a H-bond between the m-methoxy oxygen and the leucine 472 backbone NH.
The results of our theoretical calculations suggest that the thiophene ring featured by 9e-g, engaged in a strong cation-p interaction, is responsible for their considerable activities as inducers of antioxidant enzymes. This hypothesis is consistent with the much lower activities exhibited by 9h and 9i in which a dimethylene and, respectively, a trimethylene chainin place of the thiophene moietybear a carboxyl group. However, the entropic advantage offered by the thiophene ring in reducing the conformational freedom of 9e-g with respect to 9h, i cannot be ruled out.

Discussion
The results of the biological experiments and the consistency of our model of the 9g-Keap1 complex with SARs suggest that 9e-g act as inhibitors of the Keap1-Nrf2 interaction. These three  compounds share a thiophene-carboxylate moiety which gives rise to putative strong electrostatic interactions with arginine 483 and serine 508 of Keap1 and, additionally, limits conformational freedom.
Acidic inhibitors of the Keap1-Nrf2 interaction bearing carboxylic groups exist in aqueous neutral solution mainly as anionic species. This have been considered an obstacle to translocation into cells 15,23 . To circumvent such a problem, bioisosteric replacements of a carboxylic group with a tetrazole ring 24 or a nitro group 16 have been attempted, yielding compounds which retained high affinity for Keap1 and exhibited improved activity in cell-based experiments. Compound 9g stands out as the most active indole derivatives among those investigated. To our knowledge, 9g represents the first inhibitor of the Keap1-Nrf2 interaction with ampholytic properties.
Several physicochemical and pharmacokinetic properties of 9ai were calculated using the Maestro QikProp tool 25 (Table S1). The drug-likeness of the compounds is indeed confirmed by these data that show a negligible number of Lipinski Rule of 5 and Jorgensen Rule of 3 violations, good oral absorption and a prevalently lipophilic profile for each of them.
The data reported in the present paper, together with the docking model of 9g-Keap1 complex, will be exploited for continuing the design and the synthesis of novel indole derivatives as inhibitors of the Keap1-Nrf2 interaction.