QSAR models reveal new EPAC-selective allosteric modulators

Abstract

Exchange proteins directly activated by cAMP (EPAC) are guanine nucleotide exchange factors for the small GTPases, Rap1 and Rap2. They regulate several physiological functions and mitigation of their activity has been suggested as a possible treatment for multiple diseases such as cardiomyopathy, diabetes, chronic pain, and cancer. Several EPAC-specific modulators have been developed, however studies that quantify their structure-activity relationships are still lacking. Here we propose a quantitative structure-activity relationship (QSAR) model for a series of EPAC-specific compounds. The model demonstrated high reproducibility and predictivity and the predictive ability of the model was tested against a series of compounds that were unknown to the model. The compound with the highest predicted affinity was validated experimentally through fluorescence-based competition assays and NMR experiments revealed its mode of binding and mechanism of action as a partial agonist. The proposed QSAR model can, therefore, serve as an effective screening tool to identify promising EPAC-selective drug leads with enhanced potency.

Fig. 1. (A) The cAMP-bound EPAC2-CNBD (holo) state (blue, PDB 3CF6) is overlayed on the unbound (apo) state (yellow, PDB 1O7F). The phosphate binding cassette (PBC), the base binding region (BBR) and the hinge region are marked by black circles. The PBC and the hinge are in an ‘out’ conformation in the apo state and adopt an ‘in’ conformation in the holo state. cAMP, illustrated in sticks, forms key interactions between its phosphate group and residues in the PBC whereby equivalent residues in EPAC1 are written in brackets. The base of the nucleotide is oriented towards the BBR. (B) Map of residues (cyan surfaces) in the EPAC1-CNBD that showed NOE peaks to I942 protons. NOE peaks to protons in the phenyl group of I942 (2 and 3) originated from residues at the PBC, whereas the NOE peak towards the naphthalene proton (9) originated from the BBR residue, T261. (C) A scheme for the cAMP vs. I942 mimicry whereby regions of the molecular structures that are highlighted by the same shape are proposed to interact with similar regions of EPAC1-CNBD.

Fig. 2. Molecular structures of the I942 analogues in the (A) training and (B) test sets. The training and test sets include 45 and 11 molecules, respectively. The modifications, relative to I942 (3), are marked by blue for the training and red for the test set. The shaded circles highlight the positions that lack substituents originally found in I942. (C) Box plot representation for the distribution of the RFI values in the training (blue) and test (red) sets.

Fig. 3. Predicted vs. measured relative fluorescence intensities (RFI) correlation plots, with zero intercepts, for the (A) training and (B) test sets of I942 analogues shown in Fig. 2. Representative molecules are marked with black arrows and labeled in their assigned compound names as in Fig. 2.

Fig. 4. Independent validation of the QSAR model. (A) Molecular structures of new synthesized I942 analogues that are ‘unknown’ to the QSAR model. MLGM-2013, which was predicted to have the highest affinity towards EPAC1-CNBD (Table 2), is highlighted in blue and MLGM-2017, the compound with the lowest predicted affinity (Table 2) is shown in red. (B) EPAC1-CNBD binding isotherm for I942 (grey), MLGM-2013 (blue), and MLGM-2017 (red) measured through the 8-NBD-cAMP fluorescence-based competition assay. The resulting measured dissociation constants are included in the top right corner. The percentage of 8-NBD-cAMP bound to EPAC1 is represented by 〈v〉 on the y-axis. (C) The chemical shift differences between 50 μM apo EPAC1-CNBD (green) and 50 μM EPAC1-CNBD bound to 350 μM of MLGM-2013 (blue) were monitored by ¹⁵N-¹HSQC spectra. (D) The compounded chemical shift variations between 50 μM apo EPAC1-CNBD and 50 μM EPAC1-CNBD bound to MLGM-2013 (350 μM), and 100 μM apo EPAC1-CNBD and 100 μM EPAC1-CNBD bound to cAMP (1 mM) are plotted as blue and green bars, respectively. The secondary structure is shown on the top of the plot in boxes and key regions are highlighted in grey. (E) Compounded chemical shift differences of 50 μM EPAC1-CNBD in the presence of MLGM-2013 (350 μM) are plotted against the compounded chemical shift differences of 50 μM EPAC1-CNBD in the presence of I942 (350 μM). (F) 1D saturation-transfer reference (STR, blue) spectrum of EPAC1-CNBD: MLGM-2013 overlayed with the scaled saturation transfer difference (STD, red) spectrum. The assigned protons are marked in green and represented as circles on the structure of MLGM-2013 where the size of the circles reflects the relative STD/STR ratios (normalized to proton j with the highest STD/STR ratio). The structure of I942 with the previously determined STD/STR ratios are shown for comparison whereby the STD/STR ratios with the most significant differences are reported near the corresponding proton.

Fig. 5. (A) Residue specific compounded chemical shift variations between MLGM-2013-bound (green) or I942-bound (blue) and MLGM-2014-bound EPAC1 CNBD. Structural differences between the ligands are highlighted with corresponding color codes. (B) Vector representation of the CHESPA analysis. (C) Fractional activation X and (D) cos θ values of MLGM-2013-bound EPAC1 relative to cAMP-bound EPAC1 where values greater than 1 or less than −1 are not within the scale of the plot. (E) Fractional activation X and F) cos θ values of I942-bound EPAC1 measured under the same conditions as that of MLGM-2013-bound EPAC1. The asterisks correspond to the residues in the PBC which are more negative in the MLGM-2013-bound structure and the red asterisk marks N275, which exhibits the greatest change. The secondary structure of EPAC1-CNBD is shown in the same way as Fig. 4D.