Structure-based development of PDEδ inhibitors

Introduction

Ras proteins are a family of small GTPases which regulate cell growth, proliferation and differentiation (Brunsveld et al., 2006; Gelb et al., 2006). They act as molecular switches that cycle between a GDP-bound inactive and a GTP-bound active state. GTP-loaded Ras recruit effector proteins to the plasma membrane (PM) leading to their activation and propagation of their signaling pathways (Pylayeva-Gupta et al., 2011). Consequently, the level of enrichment at the PM defines the signal transduction of Ras (Rocks et al., 2005, 2010). Oncogenic Ras mutations impair intrinsic and GAP-mediated GTP hydrolysis, leading to an accumulation of active GTP- and membrane-bound Ras which results in aberrant signaling (Cox et al., 2014). KRas is the most frequently mutated isoform, and occurs in more than 90% of pancreatic, 45% of colorectal and 30% of lung tumors (Cox et al., 2014).

The delta subunit of rod specific cyclic GMP phosphodiesterase, PDE6δ (from here on called PDEδ), plays a key role to sustain the dynamic distribution of KRas in the cell (Chandra et al., 2012). PDEδ binds and solubilizes farnesylated KRas, thereby enhancing its diffusion in the cytoplasm. Once in the perinuclear area, the release factor Arl2 discharges farnesylated KRas from the PDEδ prenyl-binding pocket. This mechanism results in an enrichment of KRas in the perinuclear area and allows an effective trapping of farnesylated KRas by the recycling endosome, which restores KRas localization to the plasma membrane by vesicular transport (Schmick et al., 2014). Blocking of the PDEδ/KRas interaction by means of small molecules diminishes the concentration of KRas at the plasma membrane and thereby its oncogenic signaling (Zimmermann et al., 2013).

Recently, three different PDEδ-inhibitor chemotypes were developed that bind to the prenyl-binding pocket of PDEδ in vitro with nanomolar affinity (Zimmermann et al., 2014; Murarka et al., 2017). Here we depict the road followed from the hits to the final compounds, guided by X-ray structure determination, modeling studies, biophysical experiments and medicinal chemistry approaches. These developments should pave the way towards new small molecule inhibitors of PDEδ that show high in vivo activity and are suitable for further clinical development targeting cancers that harbor KRas mutations.

The source of inspiration

Post-translational modifications of KRas, its regulators and effectors, structural features and oncogenic mutations have been extensively studied for more than a decade (Vigil et al., 2010; Cox et al., 2014). While the binding of N- and H-Ras to endomembranes is reversible and controlled by a (de-)palmitoylation cycle, binding of KRas to the plasma membrane is mediated by interaction of a S-farnesylated C-terminal polybasic motif with the PM’s negatively charged inner leaflet. Because association of KRas with the plasma membrane is absolutely required for its function, targeting membrane association of KRas as a possible strategy for anti-cancer therapy appeared to be a more accessible alternative than direct targeting of ‘undruggable’ KRas. After the FTase inhibitors (FTIs) had not kept their promise (Berndt et al., 2011), a better understanding of KRas trafficking was necessary to identify a more effective way to restrict KRas membrane association. The finding that PDEδ acts as a cytoplasmic chaperone for KRas, thereby increasing its plasma membrane localization, signified an unprecedented opportunity for subsequent strategies to target KRas membrane association and signaling.

Bastiaens and Wittinghofer and co-workers characterized the key role of the PDEδ/Arl2 system in sustaining the correct spatial organization and signaling of Ras (Nancy et al., 2002; Ismail et al., 2011; Chandra et al., 2012). Among those findings, a randomized endomembrane distribution of mCit-KRas was found in cells treated with PDEδ-specific siRNA. In addition, knockdown of PDEδ reduced the phosphorylation levels of Erk1/2, a downstream target of Ras, in comparison to untreated cells under EGF stimulation. These results demonstrated that PDEδ modulates the signaling of oncogenic KRas by affecting its localization, and called for the development of inhibitors that compete for the farnesyl-binding pocket of PDEδ and thereby block the KRas/PDEδ interaction.

Benzimidazole derivatives as inhibitors of PDEδ

In the search for small molecules that bind to the farnesyl-binding pocket of PDEδ, initial efforts focused on the development of a high-throughput screen of a compound library. We designed an assay utilizing the chemiluminiscence-based Alpha Screen technology (Zimmermann et al., 2013). It relies on a donor bead that recognizes the biotinylated and farnesylated KRas C-terminal peptide and an acceptor bead that binds to His-tagged PDEδ (Figure 1). The screening of an in-house library of ca. 150000 compounds, assembled considering structural diversity of the underlying scaffolds, drug-likeness, coverage of different established drug target classes and inspiration by natural products, allowed the identification of the small benzimidazole hit 1 (Figure 2) as ligand of PDEδ. That inhibitor was further characterized by measuring its binding affinity for the PDEδ binding pocket (K_D=165±23 nm) by using a competitive fluorescence polarization assay based on a known PDEδ ligand.

Figure 1:

Design of a high-throughput screening assay for the detection of small molecule inhibitors of the KRas4B/PDEδ interaction.

Figure 2:

Chemical structure and binding affinities of compounds 1 (A), 2 and 3 (B) and 4 (C).

X-Ray structures of benzimidazole compounds 1 (D), PDB code: 4JV6), 2 (E), PDB code: 4JVB) and 4 (C), PDB code: 4JVF) in complex with PDEδ. Key residues are represented as sticks. Dotted lines represent H-bond interactions.

A crystal structure of compound 1 in complex with PDEδ (Figures 2 and 3), revealed that two molecules of the benzimidazole bind inside the prenyl binding pocket. The binding of the inhibitors is mediated by hydrophobic interactions and hydrogen bonding to the Tyr149 and the Arg61 side-chains. An obvious next step to obtain higher affinity compounds consisted of linking those two benzimidazole fragments (Zimmermann et al., 2014). Guided by molecular modeling (Schrödinger, Maestro suite), both moieties were connected while trying to introduce minimal disturbance in the benzimidazole orientations (Figure 3).

Figure 3:

Design of linked benzimidazoles.

(A) Surface representation of the PDEδ cavity in a crystal structure with two benzimidazole fragments 1 in complex with PDEδ (PDB 4JV6); the distance between linked atoms is highlighted. (B) Design of linked bis-benzimidazole ethers. (C) Best scoring docking pose obtained by docking a bis-benzimidazole (allyl group as R) into the crystal structure of benzimidazole fragment 1 in complex with PDEδ (PDB 4JV6).

The allyl-containing linked benzimidazole 2 bound to PDEδ in nearly the same manner as the two individual molecules, showing H-bonds to Y149 and R61 and an almost complete overlapping (Figure 2E and Zimmermann et al., 2013). This also validated the guiding docking studies, since the best scoring docking pose was highly similar to the structure observed in the crystal structure (Figure 3C). The fluorescence polarization assay confirmed that compound 2 (K_D=39±11 nm) was 4-fold more potent than compound 1.

The crystal structure of 2 in complex with the protein (Figure 2E) revealed that the allyl group is located in a previously non-filled hydrophobic cavity, pointing towards the carbonyl of the Cys56 backbone. It was thus planned to substitute that allyl moiety by different groups, ultimately arriving at a piperidine group, which would ideally form a previously unexploited H-bond interaction with the Cys56 carbonyl group resulting in a more potent PDEδ inhibitor. A series of compounds having different piperidine moieties was tested (Table 1), but only compound 3 proved to retain some affinity for PDEδ (K_D=87±35 nm) in comparison to the parent compound. The presence of the piperidine moiety greatly increased, however, the solubility of those small molecules in water, so the piperidine functionality was maintained while further improving the binding mode of the benzimidazole derivatives.

Table 1:

PDEδ affinities of ether linked bis-benzimidazoles measured by fluorescence polarization.

Together with the binding affinities of the piperidine containing compounds, docking studies suggested the backbone carbonyl of Cys56 was unreachable, due to the rigidity of the short ether linker. The exchange for a larger and more flexible ester linker led to compound 4, which had a K_D=10±3 nm and displayed a third H-bond to the Cys56 backbone, as determined by crystal structure analysis (Figure 2F).

It was eventually found that the (S)-enantiomers of both the ester and ether linked bisbenzimidazoles have higher affinity than the (R)-enantiomer. The (S)-enantiomer of compound 3, termed Deltarasin (Figure 2), was selected for further biological studies due to its highly stable ether linker, good solubility and membrane permeability and PDEδ specificity. Deltarasin was the first small molecule that proved to inhibit oncogenic RAS signaling in vivo by breaking the PDEδ/KRas interaction (Zimmermann et al., 2013).

Pyridazinone inhibitors targeting PDEδ

After the successful development of the small molecule inhibitor Deltarasin, efforts were focused on understanding and overcoming its deficiencies. While Deltarasin binds to the PDEδ prenyl binding pocket with low nanomolar affinity in vitro, micromolar concentrations of compound are needed to affect growth inhibition in cell-based assays (Zimmermann et al., 2013). In addition, Deltarasin displayed a ‘switch-like’ inhibition of proliferation, which could be indicative of general cytotoxicity or/and interactions with off-targets. This prompted us to develop a novel PDEδ inhibitor chemotype to address the question whether those limitations were chemotype-dependent or intrinsic to the PDEδ-inhibition approach.

To identify alternative chemotypes, further analysis of the previously developed high-throughput Alpha Screen hit set pointed towards pyrrolopyridazinone 14 and pyrazolopyridazinone 15 as inhibitors of the PDEδ/KRas interaction (Figure 4) (Papke et al., 2016; Murarka et al., 2017).

Figure 4:

Novel inhibitor chemotypes pyrrolopyridazinone 14 and pyrazolopyridazinone 15 identified through Alpha Screen.

Pyrrolopyridazinone derivatives

With the first hit 14 in hand (Figure 4), a series of anilide derivatives and their corresponding bioisosteric ester analogs was prepared (Table 2). Some of these compounds, containing both electron-withdrawing or -donating substituents in the aryl ring, retained the potency of the parent compound, but none of them displayed K_D values closer to those of the bisbenzimidazole chemotype.

Table 2:

PDEδ K_D values for pyrrolopyridazinone-based anilides and esters as determined by fluorescence polarization experiments.

Modeling studies suggested that the new pyrrolopyridazinone derivatives bind the PDEδ pocket similarly to the bisbenzimidazoles, displaying hydrogen bonds to Y149 and R61 (Figure 5). This binding mode would leave some room inside the pocket to accommodate larger amide substituents. In an attempt to move away from the metabolically labile anilides, and guided by docking studies, a derivative having a benzyl amide substituent was also synthesized and evaluated (Table 2, compound 26). Compound 26 showed an improved binding affinity in comparison to the first hit, so the benzyl amide scaffold was maintained while looking for further optimization.

Figure 5:

Comparison between the binding mode of inhibitors 26 and 2.

Left: Chemical structure, PDEδ affinity and best docking pose of compound 26 inside the PDEδ binding pocket (Schrödinger, Maestro suite). Right: Chemical structure, PDEδ affinity and crystal structure of inhibitor 2 in complex with PDEδ.

We characterized a collection of pyrrolopyridazinone derivatives with different linker lengths and amide structures (Table 3). As anticipated by molecular modeling, a C3 linker length was found to be optimal for binding, since it allows the inhibitor to precisely reach both H-bonds to Tyr149 and Arg61. Different aryl substitutions were well tolerated (Table 3, compounds 27–30), while the introduction of different heterocycles resulted in a diminished affinity for PDEδ (compounds 33–35). Interestingly, several cycloalkyl-amides, the larger the better, proved to bind effectively the PDEδ binding pocket (compounds 36–41).

Table 3:

PDEδ K_D values for pyrrolopyridazinones with different linker lengths and amide structure as determined by fluorescence polarization measurements.

While a series of 1,2,4- and 1,3,4-oxadiazoles failed to work as PDEδ inhibitors with improved drug-like properties, compound 42 was the only member of the pyrrolopyridazinone family that co-crystallized in complex with PDEδ (Figure 6). As expected, the binding is not only mediated by key hydrophobic interactions but also by two hydrogen bonds between the carbonyl group of the pyridazinone to Arg61, and the oxadiazole to Tyr 149 (Figure 6). This crystal structure validated the docking experiments and provided enough information for future optimization, if required, of this compound class.

Figure 6:

Structure, binding affinity and crystal structure of inhibitor 42 in complex with PDEδ at 2.10 Å resolution.

Pyrazolopyridazinone derivatives

In this case, the first hit 3 was co-crystalized in complex with PDEδ (Figure 7). This pyrazolopyridazinone inhibitor is located inside PDEδ in an identical manner to that of the previous chemotype, involving hydrogen bonds to Tyr149 and Arg61. In analogy to the pyrrolopyridazinone compound class optimization, a library of compounds bearing different arylamide substituents was synthesized (Table 4). The binding efficacy of these analogs was evaluated by means of the established fluorescence polarization assay. These results showed that some benzyl and homobenzyl amides maintain a high affinity for PDEδ (Table 4, compounds 44 and 60). Directed by molecular modeling, a methyl substituent was attached to the N-phenyl moiety that points deep into the farnesyl binding site. By satisfying more hydrophobic contacts, these analogs were more potent binders of PDEδ than the non-methylated compounds (Table 4, compounds 43–45 and 48–50 vs. 51, 55, 56, 57, 59, 60). With more than 30 potent inhibitors in hand (some not shown), it was assumed that the extra methyl group at the benzylic position of compound 63 would diminish its metabolic degradation. That inhibitor, named Deltazinone 1, was selected for further biological evaluation in order to compare the in vivo behavior of a different chemotype with that of Deltarasin.

Figure 7:

Comparison between the binding mode of inhibitors 3 and 63.

Upper row: Structure and binding affinities of pyrazolopyridazinone derivative 3 and 63 (Deltazinone 1). Lower row: Crystal structure of inhibitor 3 in complex with PDEδ at 2.60 Å resolution (PDB code: 5E80) and predicted binding mode of Deltazinone 1 (best docking pose; Schrödinger, Maestro suite). Besides Y149 and R61, Gln78 is also shown as stick to highlight a probable third H-bonding interaction.

Table 4:

PDEδ K_D values for pyrazolopyridazinones with different amide structures and N-phenyl substituents, as determined by fluorescence polarization measurements.

To determine their selectivity profile and detect pharmacologically relevant off-targets, Deltazinone 1 and Deltarasin were subjected to a panel of ca. 100 binding, enzymatic and cell-based assays. With the exception of one assay (59% inhibition of specific control binding to adenosine A3 receptor), Deltazinone 1 did not show an inhibition or stimulation higher than 50% at a compound concentration of 5 μm, which is considered to represent a significant effect of the test compound (Figure 8). Deltarasin, however, was found to bind to different GPCRs, ion channels and transporters. This suggests that Deltarasin eventually modulates several off-targets and so the ‘switch-like’ behavior in growth inhibition assays may be due to general cytotoxicity, rather than dependent on the approach to block the PDEδ/KRas interaction. Indeed, Deltazinone 1 inhibited the cell growth of KRas dependent cell lines in a clear dose-dependent manner, and did not show unspecific cytotoxicity even at 24 μm concentrations (Papke et al., 2016). Because Deltazinone 1 needed at least 10 μm concentrations to show an effect in cell-based assays, this limitation may not be chemotype- but system-dependent. An adequate explanation points towards displacement of the small molecule inhibitors of PDEδ by GTP-bound Arl2. In that scenario, only the development of compounds that bind tighter to PDEδ (i.e. using a high number of non-covalent interactions) or PDEδ covalent inhibitors could overcome this limitation.

Figure 8:

Characterization of Deltarasin and Deltazinone 1 in a panel of biochemical and pharmacological assays.

Both compounds were tested at a concentration of 5 μm in the CEREP diversity panel (Eurofins Pharma Discovery Services) including a comprehensive set of approximately 100 receptor binding, enzymatic and cell-based assays.

Even though Deltazinone 1 revealed an acceptable early ADME profile and showed an improvement in cell-based assays in comparison to Deltarasin, this compound is rapidly metabolized in mouse liver microsomes, making it non-suitable for in vivo experiments. Efforts thus focused on increasing the metabolic stability of Deltazinone 1 to obtain a compound that can be used in mouse models. Using non-quantitative in silico predictions (SMARTCyp) and LCMS/MS techniques (after incubation of Deltazinone 1 in the presence of mouse liver microsomes), four metabolic hot spots affected by CYP450-mediated metabolism were identified (Figure 9). Removal of the benzylic methyl functionality and blocking of the para-phenyl position with a fluorine atom led to 69 (Table 5). This compound retained the potency of Deltazinone 1 (K_D=8±4 nm), while showing a substantial increase in metabolic stability (Table 5). An extended plasma exposure and terminal half-life was found for compound 69 in comparison to Deltazinone 1 (63) after intravenous dosing in mice (Figure 9) (Murarka et al., 2017). These results proved compound 69 to be a valid candidate for further pharmacological development using mouse models.

Figure 9:

Metabolic soft spots of 63 and PK profiling of 63 and 69.

(A) In silico prediction of metabolic soft spots in Deltazinone 1 using SMARTCyp. (Yellow, most likely liable site; red, second most likely liable site; green, third most likely liable site). (B) PK profiling of pyrazolo-pyridazinones 63 and 69 were administered to NMRI mice IV at a dose of 3 mg/kg each. Blood samples were taken over a period of 24 h and analyzed for the total compound concentrations in plasma. Left panel: plasma concentration-time-profile over 24 h for 63 and 69. Right panel: comparison of PK parameters of 63 and 69 calculated from the total plasma concentrations measured.

Table 5:

PDEδ K_D and clearance values for pyrazolopyridazinones.

Conclusions and further perspectives

While some preliminary results during the bisbenzimidazole inhibitor development raised several questions, it has been proven that different PDEδ inhibitor chemotypes can overcome specific limitations. The finding that Deltazinone 1 causes cell death in a set of KRas dependent hPDAC cell lines with high specificity and little off-target effects further validates PDEδ as a relevant target for anti-cancer drug discovery.

Future directions should focus on the development of more potent compounds or covalent inhibitors that can better resist the allosteric displacement by Arl2.