Toward more potent imidazopyridine inhibitors of Candida albicans Bdf1: Modeling the role of structural waters in selective ligand binding

Abstract

Novel agents to treat invasive fungal infections are urgently needed because the small number of established targets in pathogenic fungi makes the existing drug repertoire particularly vulnerable to the emergence of resistant strains. Recently, we reported that Candida albicans Bdf1, a bromodomain and extra‐terminal domain (BET) bromodomain with paired acetyl‐lysine (AcK) binding sites (BD1 and BD2) is essential for fungal cell growth and that an imidazopyridine (1) binds to BD2 with selectivity versus both BD1 and human BET bromodomains. Bromodomain binding pockets contain a conserved array of structural waters. Molecular dynamics simulations now reveal that one water molecule is less tightly bound to BD2 than to BD1, explaining the site selectivity of 1. This insight is useful in the performance of ligand docking studies to guide design of more effective Bdf1 inhibitors, as illustrated by the design of 10 new imidazopyridine BD2 ligands 1a–j, for which experimental binding and site selectivity data are presented.

The co‐crystal structure of an imidazopyridine ligand (1) complex with Candida albicans Bdf1 BD2, a novel anti‐fungal drug target, reveals that one (W1) among six structural water molecules has been displaced. Molecular dynamic simulations provided insight into the mechanistic role of the displacement process. This informed more accurate molecular docking studies of a series of 1 analogues, leading to identification of novel, more potent selective inhibitors of Bdf1. Compound 1 bound in CaBD2 shows W1 of the 6 structural waters W0–W5, which are present in the apo structure, being displaced by the —OH group of the ligand.

1. INTRODUCTION

Invasive fungal infections (IFIs) have a high morbidity and mortality rate (overall 3‐month mortality rate of 51% in one US study) ¹ in immunocompromised lupus erythematosus, solid organ transplant or AIDS patients, and are estimated to cause more than 1.5 million deaths annually. ² These systemic fungal infections are becoming more common, and often lack effective drug therapies. ³ , ⁴ Candida bloodstream infections are among the most significant IFIs among hospitalized patients, with a 30‐day mortality rate of 41% according to a study conducted in Scotland in 2012–2013. ⁵ Candida albicans is the most common cause of Candida bloodstream infections, while Candida glabrata ranks second in isolation frequency. ⁶ Candida auris, a new multidrug‐resistant species that causes nosocomial transmission has been identified as an emerging urgent global health threat by the Centers for Disease Control and Prevention (CDC). ⁷ Legacy antifungal strategies target cell membrane integrity (polyene drugs, e.g., amphotericin B), the synthesis of ergosterol (azole drugs such as fluconazole), ⁸ nucleic acid polymerases (pyrimidine analogues like flucytosine) ⁹ or the cell wall (echinocandins) ¹⁰ with newer inhibitors targeting fungal dihydroorotate dehydrogenase (olorofim) and a Gwt1 (fosmanogepix). ³ However, the relatively limited number of drug classes available coupled with the increasing frequency of drug‐resistant fungal strains ¹¹ , ¹² have created an urgent need for new approaches to antifungal drug design. ³ , ⁴ , ¹³

Targeting of chromatin signaling pathways via small‐molecule inhibition of bromodomain and extra‐terminal domain (BET) bromodomains has been introduced as a therapeutic strategy against cancer and other non‐infectious diseases ¹⁴ , ¹⁵ but had not been explored as an approach to antifungal drug discovery prior to our initial report on the structure, function and inhibition of C. albicans BET bromodomains. ¹⁶

BET proteins are chromatin‐associated factors that regulate gene transcription and chromatin organization. ¹⁷ BET proteins contain two bromodomains (BD1 and BD2), small helical domains that specifically recognize short peptides acetylated on lysine residues and that mediate the association of BET proteins with acetylated histones in chromatin. ¹⁵ Whereas humans have four BET proteins (Brdt, Brd2, Brd3, and Brd4), only a single BET protein (Bdf1) is present in C. albicans. We previously showed that mutations inactivating the ligand binding functionality of both Bdf1 bromodomains suppress the viability and virulence of C. albicans. ¹⁶ Using a homogeneous time‐resolved FRET (HTRF) acetylpeptide ligand binding assay for Bdf1 bromodomains, we carried out a high‐throughput chemical screen to identify Bdf1 bromodomain inhibitors. One promising hit identified was an imidazopyridine compound (compound 1). This compound inhibited Bdf1 BD2 with a low micromolar IC₅₀ value and showed selectivity against the orthologous BD2 domain from the human BET protein Brd4. ¹⁶

Many successful drugs have emerged from high‐throughput screen augmented by computer‐aided rational design in recent years. ¹⁸ , ¹⁹ , ²⁰ , ²¹ Computer‐aided rational design can greatly accelerate the hit‐to‐lead phase of drug development and reduce the associated costs. ¹⁸ Despite recent advances in both computational capabilities ²² , ²³ and efficient screening strategies, ²⁴ drug discovery can still be complicated by complexities in ligand/binding site interactions. In this study of computer modeling‐based Bdf1 bromodomain inhibitor optimization, we apply ligand binding site docking and molecular dynamics (MD) to analyze the ligand binding pocket interactions of C. albicans Bdf1 BD2 (hereafter CaBD2), including the role of structural waters, with a series of designed analogues of 1.

2. METHODS

2.1. Chemistry

All reagents and solvents were purchased from Sigma Aldrich, TCI or Oakwood Chemical and were used directly as received. Reaction products were purified by flash chromatography using the ISCO CombiFlash® Rf+ Lumen. ¹H, ¹³C, and ¹⁹F NMR spectra were obtained on Varian 400‐MR, VNMRS‐500, or VNMRS‐600 spectrometers. NMR spectra were processed with MestReNova 9.0.0 or 11.0.2. Multiplicities are quoted as singlet (s), doublet (d), triplet (t), unresolved multiplet (m), doublet of doublets (dd), doublet of doublet of doublets (ddd), doublet of triplets (dt), triplet of doublets (td), and broad (b). All chemical shifts (δ) are reported in parts per million (ppm) relative to residual CD₂HOD in CD₃OD (δ 3.34, ¹H NMR), CHCl₃ in CDCl₃ (δ 7.26, ¹H NMR) and CFCl₃ (δ 0.00, ¹⁹F NMR). Mass spectrometry was performed on a Finnigan LCQ Deca XP Max. Mass spectral data were calculated using ChemDraw 19.1.21. Isotopic mass for compounds containing a bromine atom was calculated for ⁸¹Br. Compound IUPAC names were assigned by MarvinSketch 20.3.

2.2. General method for synthesis of 1a–j

The 1a–j series was prepared via a two‐step reaction (Scheme 1) as previously reported. ¹⁶ The bromoacetophenone 3 (1 eq) and corresponding toluidine 4 (1.1 eq) were triturated with Na₂CO₃ (0.47 eq) for 2 min to form the α‐aminoketone intermediates 5 (TLC). The product was washed 3× with water and dried under vacuum, then was dissolved in aqueous iPrOH (0.16 M). 3‐Methylpyridin‐2‐amine (1 eq) was added dropwise, followed by ZnI₂ (0.3 eq), and 4 Å molecular sieves. The reaction mixture was heated to 80°C for 12 h, and the product 1 was taken up into EtOAc and purified via flash chromatography (Hex/EtOAc, 0%–90% gradient) (5%–20%).

SCHEME 1.

Synthesis of 1a–j.

2‐Fluoro‐4‐(8‐methyl‐3‐(p‐tolylamino)imidazo[1,2‐a]pyridin‐2‐yl)phenol (1a). ¹H NMR (400 MHz, CD₃OD): δ 7.80 (dq, J = 6.8, 0.8 Hz, 1H), 7.68 (dd, J = 12.8, 2.0 Hz, 1H), 7.61 (ddd, J = 8.4, 2.1, 1.0 Hz, 1H), 7.12 (dt, J = 6.8, 1.1 Hz, 1H), 6.98–6.92 (m, 2H), 6.90 (dd, J = 9.1, 8.4 Hz, 1H), 6.79 (t, J = 6.8 Hz, 1H), 6.44–6.40 (m, 2H), 2.62 (s, 3H), 2.20 (s, 3H); ¹³C NMR (101 MHz, CD₃OD): δ 152.58, 150.20, 144.61, 144.47, 143.12, 142.54, 129.64, 127.96, 126.34, 124.48, 123.29 (d, J = 3.2 Hz), 120.69, 117.19 (d, J = 3.3 Hz), 114.60, 114.40, 112.84, 112.17, 19.09, 15.27; ¹⁹F NMR (376 MHz, CD₃OD): δ 139.62 (dd, J = 12.9, 9.3 Hz); LRMS (ESI, m/z): [M+H]⁺ calcd for C₂₁H₁₉FN₃O, 348.15; found, 348.60 (Figures S3–S7).

3‐Fluoro‐4‐(8‐methyl‐3‐(p‐tolylamino)imidazo[1,2‐a]pyridin‐2‐yl)phenol (1b). ¹H NMR (400 MHz, CD₃OD, δ): 7.79 (dq, J = 6.7, 0.7 Hz, 1H), 7.42 (t, J = 8.5 Hz, 1H), 7.15 (dt, J = 6.8, 1.2 Hz, 1H), 6.94–6.88 (m, 2H), 6.83 (t, J = 6.8 Hz, 1H), 6.63 (dd, J = 8.5, 2.4 Hz, 1H), 6.56 (dd, J = 12.0, 2.4 Hz, 1H), 6.39–6.32 (m, 2H), 2.61 (s, 3H), 2.18 (s, 3H); ¹³C NMR (101 MHz, CD₃OD, δ): 163.40, 160.93, 160.41, 160.30, 144.44, 143.57, 132.85 (d, J = 5.4 Hz), 130.53, 128.88, 127.56, 125.38, 122.12, 114.23, 113.31, 112.17 (d, J = 3.0 Hz), 103.77, 103.52, 20.28, 16.49; ¹⁹F NMR (376 MHz, CD₃OD): −114.06 (dd, J = 12.0, 8.5 Hz); LRMS (ESI, m/z): [M+H]⁺ calcd for C₂₁H₁₉FN₃O, 348.15; found, 348.50 (Figures S8–S12).

5‐((2‐(4‐Hydroxyphenyl)‐8‐methylimidazo[1,2‐a]pyridin‐3‐yl)amino)‐2‐methylphenol (1c). ¹H NMR (600 MHz, CD₃OD): δ 7.82–7.80 (m, 2H), 7.80 (d, J = 2.1 Hz, 1H), 7.11 (dt, J = 6.9, 1.2 Hz, 1H), 6.85 (d, J = 8.1 Hz, 1H), 6.80 (dd, J = 4.4, 2.4 Hz, 2H), 6.79 (d, J = 1.9 Hz, 1H), 6.07 (dd, J = 8.1, 2.3 Hz, 1H), 5.91 (d, J = 2.3 Hz, 1H), 2.61 (s, 3H), 2.06 (s, 3H); LRMS (ESI, m/z): [M+H]⁺ calcd for C₂₁H₂₀N₃O₂, 346.16 m/z; found, 346.30 (Figures S13–S15).

2‐((2‐(4‐Hydroxyphenyl)‐8‐methylimidazo[1,2‐a]pyridin‐3‐yl)amino)‐5‐methylphenol (1d). ¹H NMR (400 MHz, CD₃OD): δ 7.85–7.78 (m, 3H), 7.11 (dt, J = 6.8, 1.2 Hz, 1H), 6.82–6.75 (m, 3H), 6.68 (dd, J = 1.8, 0.7 Hz, 1H), 6.38–6.30 (m, 1H), 5.86 (d, J = 7.9 Hz, 1H), 2.62 (s, 3H), 2.16 (s, 3H); ¹³C NMR (101 MHz, CD₃OD): δ 157.02, 144.52, 142.42, 138.40, 131.23, 128.41, 128.35, 126.05, 124.67, 124.25, 120.70, 120.04, 115.16, 114.75, 111.95, 111.52, 19.33, 15.28; LRMS (ESI, m/z): [M+H]⁺ calcd for C₂₁H₂₀N₃O₂, 346.16; found, 346.40 (Figures S16–S19).

4‐(8‐Chloro‐3‐(p‐tolylamino)imidazo[1,2‐a]pyridin‐2‐yl)phenol (1e). ¹H NMR (400 MHz, CD₃OD, δ): 7.89 (dt, J = 6.8, 1.0 Hz, 1H), 7.88–7.75 (m, 2H), 7.39 (dd, J = 7.4, 1.0 Hz, 1H), 7.02–6.89 (m, 2H), 6.89–6.71 (m, 3H), 6.48–6.36 (m, 2H), 2.20 (s, 3H); LRMS (ESI, m/z): [M+H]⁺ calcd for C₂₀H₁₇ClN₃O, 350.11; found, 350.36 (Figures S20–S22).

4‐(8‐Bromo‐3‐(p‐tolylamino)imidazo[1,2‐a]pyridin‐2‐yl)phenol (1f). ¹H NMR (600 MHz, CD₃OD) δ: 7.94 (dd, J = 6.7, 1.0 Hz, 1H), 7.85–7.79 (m, 2H), 7.58 (dd, J = 7.3, 1.0 Hz, 1H), 6.98–6.94 (m, 2H), 6.80–6.76 (m, 3H), 6.44–6.39 (m, 2H), 2.20 (s, 3H); ¹³C NMR (151 MHz, CD₃OD: δ 158.80, 144.32, 141.31, 141.25, 131.10, 130.05, 129.53, 129.05, 125.54, 123.76, 121.72, 116.23, 114.33, 113.44, 111.46, 20.51); LRMS (ESI, m/z): [M+H]⁺ calcd for C₂₀H₁₇BrN₃O, 396.05; found, 396.35 (Figures S23–S26).

4‐(8‐Iodo‐3‐(p‐tolylamino)imidazo[1,2‐a]pyridin‐2‐yl)phenol (1g). ¹H NMR (600 MHz, CD₃OD): δ 7.93 (dd, J = 6.7, 1.1 Hz, 1H), 7.84–7.81 (m, 2H), 7.79 (dd, J = 7.1, 1.1 Hz, 1H), 6.98–6.94 (m, 2H), 6.80–6.75 (m, 2H), 6.63 (t, J = 6.9 Hz, 1H), 6.45–6.37 (m, 2H), 2.20 (s, 3H); ¹³C NMR (151 MHz, CD₃OD): δ 157.29, 142.94, 141.51, 139.80, 134.51, 129.65, 128.61, 128.02, 124.24, 123.00, 120.30, 114.76, 112.88, 112.57, 82.56, 19.08; LRMS (ESI, m/z): [M+H]⁺ calcd for C₂₀H₁₇IN₃O, 442.04; found, 442.22 (Figures S27–S30).

4‐(3‐([4‐Aminophenyl]amino)‐8‐methylimidazo[1,2‐a]pyridin‐2‐yl)phenol (1h). ¹H NMR (400 MHz, CD₃OD): δ 7.83–7.77 (m, 3H), 7.09 (dt, J = 6.9, 1.2 Hz, 1H), 6.80–6.75 (m, 3H), 6.64 (d, J = 8.6 Hz, 2H), 6.36 (d, J = 8.6 Hz, 2H), 2.61 (s, 3H); LRMS (ESI, m/z): [M+H]⁺ calcd for C₂₀H₁₉N₄O, 331.16; found, 331.37 (Figures S31–S33).

4‐(3‐([4‐Bromophenyl]amino)‐8‐methylimidazo[1,2‐a]pyridin‐2‐yl)phenol (1i). ¹H NMR (600 MHz, CD₃OD): δ 7.80 (dd, J = 6.8, 1.0 Hz, 1H), 7.79–7.75 (m, 2H), 7.26–7.23 (m, 2H), 7.13 (dt, J = 6.9, 1.4 Hz, 1H), 6.83 (d, J = 6.8 Hz, 1H), 6.81–6.79 (m, 2H), 6.48–6.42 (m, 2H), 2.62 (s, 3H); ¹³C NMR (151 MHz, CD₃OD): δ 157.21, 145.14, 142.63, 138.65, 131.91, 128.45, 126.34, 124.44, 124.38, 120.45, 118.01, 114.84, 114.60, 112.27, 110.13, 15.28; LRMS (ESI, m/z): [M+H]⁺ calcd for C₂₀H₁₇BrN₃O₃, 396.05; found, 396.20 (Figures S34–S37).

4‐(8‐Methyl‐3‐([4‐nitrophenyl]amino)imidazo[1,2‐a]pyridin‐2‐yl)phenol (1j). ¹H NMR (400 MHz, CD₃OD): δ 8.14–8.03 (m, 2H), 7.85 (ddd, J = 6.8, 1.3, 0.7 Hz, 1H), 7.78–7.70 (m, 2H), 7.20 (d, J = 6.9 Hz, 0H), 6.88 (t, J = 6.8 Hz, 1H), 6.85–6.78 (m, 2H), 6.63 (d, J = 8.7 Hz, 2H), 2.65 (s, 3H); ¹³C NMR (151 MHz, CD₃OD): δ 171.57, 157.44, 152.14, 142.97, 139.71, 139.03, 128.44, 126.60, 125.84, 124.70, 124.04, 120.28, 116.32, 114.94, 112.64, 15.29; LRMS (ESI, m/z): [M+H]⁺ calcd for C₂₀H₁₇N₄O₃, 361.13; found, 361.43 (Figures S38–S41).

2.3. Computational studies

2.3.1. Dscore calculation

All chains in the published PDB structures were extracted and analyzed with inclusion of the five conserved structural waters located inside the binding pocket. ¹⁷ Structural waters were manually added to chains lacking all five conserved waters from complete chains. The chains were prepared and minimized using the Schrödinger Suite (release 2017‐3) protein preparation wizard to ensure consistency using the force field OPLS3e. ²⁵ The minimized chains were then analyzed using Schrödinger Suite SiteMap, to obtain the druggability score (Dscore) with default parameters. ²⁶

2.3.2. Molecular docking

The atomic coordinates of CaBD2 including those of the structural water molecules (W0–W5) were taken from the structure of CaBD2 bound to imidazopyridine 1 (PDB code: 5N18). Ligand molecules were simulated as flexible, while the protein structure remained fixed, except for induced fit docking. The pH was set to 7 and all relevant structures at this pH were examined. For each high scoring molecule, docking was first performed in ICM‐Pro Molecular Modeling Software (Molsoft) version 3.9‐2b, with default settings. The molecule was then re‐docked to the protein using Schrödinger Suite SP‐Glide docking, with default settings. Finally, docking was redone using the Schrödinger Suite Induced Fit module, with residues near the binding pocket set to flexible.

2.3.3. Molecular dynamics

The atomic coordinates of ligand‐free CaBD1 (PDB 5N15) and of the unbound (PDB 5N13) and 1‐bound (PDB 5N18) forms of CaBD2 were prepared using the Schrödinger Suite protein preparation function at pH = 7. For each bromodomain, the conserved water molecules W0–W4 (as well as the non‐conserved water W5) within the CaBD2 binding pocket were kept, while all other water molecules were deleted. We replaced the glycerol molecule in the complexed CaBD2 structure (PDB 5N15) by a water molecule, using the position of the corresponding W1 water molecule in unbound CaBD1 as a reference. Then each protein with its bound water molecules was minimized using the protein preparation wizard. All the prepared proteins or protein‐ligand systems were then uploaded to http://charmm-gui.org to generate the required GROMACS files, with Force Field CHARMM36m applied. ²⁷ The water boxes were set to the default size, with Na⁺ and Cl⁻ added to create 0.15 mM NaCl solution environments. The temperature was changed to 310 K and a 100 ns MD simulation was performed for each system using Gromacs 5.1.

2.3.4. MD analysis

The trajectories were analyzed using the MD analysis tool WORDOM. ²⁸ , ²⁹ Occupancy of each conserved water molecule position was determined using the three atoms from the binding site residues list in Tables S1 and S2. The area of the intersection of three spheres was calculated using a Monte‐Carlo method, with iteration set to 1e6.

2.4. HTRF assay

The biotinylated tetra‐acetylated histone H4 peptide (H4Ac₄, N‐terminal H4 peptide, residues 1–20, with acetylation of K5, K8, K12 and K16) was purchased from Covalab (Villeurbanne, France). HTRF reagents and buffers were purchased from Cisbio. HTRF assays were performed as previously described. ¹⁶ Briefly, GST‐tagged bromodomains (human Brd4 BD1 and BD2, CaBdf1 BD1 and BD2) in 25 mM Hepes pH 7.5, 150 mM NaCl, 0.5 mM DTT were assayed at 5 nM final concentration. The corresponding final concentrations of biotinylated H4Ac₄ peptide used were 50, 600, 300 and 400 nM, respectively. The antibody‐conjugated donor (Mab anti‐GST‐Tb; Cisbio) was used at 0.5 nM and the streptavidin‐conjugated acceptor (streptavidin‐d2; Cisbio) was used at 1/8 of the H4ac₄ peptide concentration. Inhibitors were tested by performing a nine‐point dilution series with a final concentration ranging between 0.013 and 20 μM. Experiments were performed in 384‐well white plates (Greiner 781080) in a final volume of 16 μl. Plates were incubated at 4°C for 4 h (BD1) or 24 h (BD2) and subsequently analyzed in a ClarioStar plate reader (BMG LABTECH). Excitation was at 330 nm and emission intensities were measured at 620 and 665 nm with an integration delay of 60 μs and an integration time of 400 μs.

3. RESULTS

3.1. MD analysis of water molecules in the binding pocket of Bdf1 bromodomains

A highly conserved structural feature of bromodomains is an array of five water molecules (designated W0–W4) that form hydrogen bonds with protein main chain and side chain atoms within the ligand binding pocket, ¹⁵ as observed in the unliganded structure of CaBD1 (Figure 1A). These water molecules, which reduce the volume and alter the shape of the binding pocket, ³⁰ play an important role in ligand recognition since they can bridge hydrogen bonds between the ligand and protein residues. ³¹ , ³² They also affect the druggability of the pocket when present, as reflected in the change in Dscore (Figure S2 and Table S3; a higher Dscore corresponds to a more druggable pocket). Huang et al. have reported that some of these water molecules can be displaced by other solvents. ³³ Interestingly, in the structure of unbound CaBD2 (PDB 5N13), which was determined using crystals soaked in a glycerol‐containing solution, ⁴ water position W1 is occupied by the hydroxyl group of a glycerol molecule (Figure 1B). Similarly, in the crystal structure of CaBD2 bound to inhibitor 1 (PDB 5N18), position W1 is occupied by the hydroxyl group of the ligand (Figure 1C). In this structure, another (non‐conserved) water molecule (W5) serves as a water bridge between the receptor and the ligand (Figure 1C).

FIGURE 1.

Comparison of the conserved water molecules (red number labels) of (A) CaBD1 (PDB 5N15), (B) CaBD2 (PDB 5N13) and (C) CaBD2 bound to compound 1 (PDB 5N18)

We hypothesized that conserved water W1 in CaBD2 is selectively displaced by ligands because it is relatively more weakly bound to the CaBD2 ligand binding pocket. To verify this hypothesis, we applied the analysis method of Huang et al. ³³ by monitoring the behavior of water molecules W0–W5 in MD simulations of CaBD1 and CaBD2 (Figure 2).

FIGURE 2.

Overlap of the binding pose observed for 1 in the co‐crystal structure with CaBD2 (PDB 5N18) (yellow) with that predicted by molecular docking (pink)

Three 100 ns MD analyses with identical parameters were separately conducted on CaBD1 (PDB 5N15) and on the unbound (PDB 5N13) and 1‐bound (PDB 5N18) forms of CaBD2 (Figure 3) (for unbound CaBD2, the glycerol ligand was replaced by a water molecule at the W1 position). The position of each monitored water molecule was determined by three nearby protein atoms, with a preference for oxygen and nitrogen atoms that could form hydrogen bonds. Water occupancy was calculated using threshold values of distances between the water molecules and corresponding protein atoms (Tables S1 and S2). The threshold values were adjusted to keep a constant volume (22 ± 2 Å³) at each water position.

FIGURE 3.

Occupancy of structural water positions in the binding pocket of CaBD1 and BD2. (A–C) A data point indicates that a water molecule is present at a given time. (A) The dynamics of water molecules W0–W4 for unbound CaBD1 (PDB 5N15). (B) The dynamics of water molecules W0–W5 for unbound CaBD2 (PDB 5N13). (C) The dynamics of water molecules W0 and W2–W5 for 1‐bound CaBD2 (PDB 5N18; 8HZ = 1). (D) Histogram comparing the water occupancy of positions W0–W5

The results reveal that the six water molecules in unbound CaBD2 have very different occupancies (Figure 3). The position of water W1, which was displaced by the ligand in CaBD2 (Figure 3C, 3D), had a relatively low water occupancy (~15%) compared to the other conserved water positions W0 and W2–W4 (~60%) (Figure 3B, 3D). In contrast, the W1 position in CaBD1 had a high water occupancy (~80%), demonstrating that the presence of this water was energetically favorable (Figure 3A, 3D). Water W5, which is not a canonical conserved water in bromodomain structures, displayed a low water occupancy in CaBD2 only (~15%) (Figure 3). In addition, comparing the water occupancies of the bound and unbound forms of CaBD2 showed that the presence of the imidazopyridine ligand stabilized all five water molecules common to both forms (W0 and W2–W5), increasing the water occupancy by 5%–10% (Figure 3BD).

The difference in stability of water molecule W1 in CaBD1 vs CaBD2 contributes to the difference in water shell and pocket size between these two bromodomains. For ligand docking studies, it therefore is preferable to retain this W1 in the binding pocket of CaBD1 and omit it in CaBD2, although this presents a potential challenge in attempting to design single compounds that potently inhibit both BDs.

3.2. Ligand docking studies

We next performed ligand docking studies using Schrödinger Suite SP‐Glide docking to better understand how 1 interacts with the CaBD2 binding pocket and aid identification of more effective ligands. First, we re‐docked 1 to CaBD2, retaining the conserved water molecules observed in the crystal structure (W0, W2–W5). The predicted binding pose with the non‐protonated form of the ligand was essentially identical to that observed experimentally (Figure 2), enhancing confidence in the reliability of the docking procedure.

Because the occupancy of water W1 is significantly lower than that of the other conserved waters, we excluded it from the docking model. To confirm this choice and further validate the docking protocol, we docked the same ligand 1 to CaBD2 with all six water molecules including W1. The resulting model gave a docking score of −7.25 kJ/mol, significantly worse than the score obtained when W1 was omitted (−9.90 kJ/mol). The interaction graph (Figure S1) now showed a poor fit for the ligand, which was forced partially outside the pocket by the additional water.

3.2.1. The effect of pK _a on inhibitor potency

Our initial high‐throughput chemical screen yielded a hit compound, 2, that inhibited CaBD2 much more potently than 1. Using MarvinSketch 20.3 to estimate pK _a values for 2 and 1 (Figure 4), we found that the pK _a corresponding to the protonated imidazole nitrogen atom differed significantly between the two compounds: 3.10 in 2 versus 6.24 in 1, owing to destabilization of the protonated species by the meta‐pyridyl nitrogen (Figure S43).

FIGURE 4.

Structures of inhibitors 2 and 1

To better understand the role of the ligand pK _a in binding, we docked 1 in both the non‐protonated and protonated forms to CaBD2 using Schrödinger Suite SP‐GLIDE docking (Figure 5).

FIGURE 5.

Binding interaction diagram of unprotonated (left) and protonated (right) 1 (PDB 5N18)

The modeling results revealed that non‐protonated 1 had a score of −9.90 kJ/mol, while that of the protonated form was −8.86 kJ/mol. The ligand interaction diagram reveals that this difference may originate from a change in the bridging role of water molecule W5, which serves as a hydrogen donor instead of a hydrogen acceptor. The standard pK _a for all of the compounds are calculated to be within the range of 3.10–6.24 suggesting they would all be ionized at pH 7, promoting hydrogen bonding.

3.3. Potential for additional polar interactions

The interaction diagram in Figure 5 shows that the methyl of the methylphenyl group of 1 is very close to leucine 420. The distance between the methyl group carbon of 1 and the backbone carbonyl oxygen of leucine 420 is 4.3 Å. One might be able to introduce additional polar interactions that enhance inhibitor potency by replacing the methyl group with a hydrogen donor or by adding hydrogen donor groups to the phenyl ring. Induced fit docking using the Schrödinger Suite supported this hypothesis. Adding a hydroxyl group in the ortho or meta position (1c and 1d) with respect to the methyl group was predicted to yield an additional polar interaction with the Leu420 carbonyl group. Replacement of —CH₃ with —NH₂ did not show a notable improvement in the docking score (1h), but distance between the —H from —NH₂ to the Leu420 oxygen is within 3 Å, suggesting the feasible formation of an additional hydrogen bond (Figure 6).

FIGURE 6.

Binding mode of 1 analogues in induced fit docking (left: addition of —OH [1c]; right: replacement of —CH₃ with —NH₃ [1h])

3.4. Inhibitor design

As discussed above, when performing molecular docking we only retained the conserved waters observed in the ligand‐bound CaBD2 crystal structure (PDB 5N18). The docking was performed with ICM‐Pro Molecular Modeling Software (Molsoft) version 3.9‐2b using a total position scan. Analogues with relatively high docking score were selected and re‐docked on CaBD2 using Schrödinger Suite SP‐Glide, via rigid and flexible pathways. The output compounds were then selected based on our analysis of the pocket property (Figure 7).

FIGURE 7.

Designed analogues of 1

Compounds 1a–1b were designed to enhance the hydrophobic interactions inside the pocket (residues F411, V415, Y425, C464, F467, V474), by introduction of an F substituent. In 1c–1d and 1h, we sought to enable additional hydrogen bonds with Leu420, as discussed above. Compounds 1e–1g were designed to lower the pK _a of the target ligand by introducing electron‐withdrawing substituents. The original imidazopyridine methyl group was replaced by halogen atoms (Cl, Br, or I) to mimic the electron density of compound 2 while preserving its selectivity. F was not considered in this role, because of its small size. In compounds 1i–1j we replaced the toluidine methyl group with Br and NO₂ substituents, as prototypes for conjugation with a second drug.

3.5. Synthesis

The 1a–1j series of compounds were prepared via a two‐step synthetic strategy (Scheme 1) originally developed for 1. ¹⁶ Thus, the α‐aminoketones 5 were formed by reaction of bromide 3 with toluidine 4. Formation of the final imidazo[1,2‐a]pyridines 1a–1j was realized by Lewis acid‐catalyzed condensation of the appropriate 5 derivative with 3‐methylpyridin‐2‐amine (6) in the presence of oxygen. All compounds were characterized by their NMR and MS spectra.

3.6. Inhibitory activity of compounds against CaBD2

HTRF assays showed that compounds 1c, 1e, 1f, 1g, and 1h have relatively higher potency (lower IC₅₀ values) compared to 1, consistent with the predictions made from molecular docking (Table 1, Figure S42). However, the addition of a fluorine atom in compounds 1a and 1b did not decrease the IC₅₀ value. Compounds 1c and 1d with a hydroxyl group at R4 or R5, respectively, were not predicted to differ in potency, but in fact 1c showed significantly increased potency compared to 1 and 1d. The introduction of halogens in position X (1e, 1f, and 1g) showed no significant difference in potency between the three, as expected with the docking results, but compared to 1 increased potency. This is likely due to the decrease in pK _a of the nitrogen hydrogen bonding with W5 (Figure, Figure S43). 1h showed increased hydrogen bonding with Leu420 based on docking results, and was shown to increase inhibition compared to 1. When the methyl group in position R3 was replaced with a bromo or nitro group as in 1i and 1j, there was no significant decrease in activity as predicted by our docking results.

TABLE 1.

IC₅₀ of ligand analogues 1a–j

Comp	R1	R2	R3	R4	R5	X	ICM‐pro docking scores (CaBD2, unitless)	IC50 (μM)				Selectivity (CaBD2 vs. hBrd4 BD2)
								CaBD1	CaBD2	hBrd4 BD1	hBrd4 BD2
1							−42.04	>20	3.90 ± 0.35	>20	>20	>5
1a	—F	—H	—Me	—H	—H	—Me	−41.31	>20	6.30 ± 2.78	14.8 ± 6.9	8.7 ± 1.8	1.4
1b	—H	—F	—Me	—H	—H	—Me	−41.73	>20	5.22 ± 2.12	>20	>20	>4
1c	—H	—H	—Me	—OH	—H	—Me	−41.47	>20	0.78 ± 0.15	>20	>20	>26
1d	—H	—H	—Me	—H	—OH	—Me	−41.89	>20	9.74 ± 0.19	>20	>20	>2
1e	—H	—H	—Me	—H	—H	—Cl	−43.21	>20	1.50 ± 0.42	>20	16.0 ± 1.2	10.6
1f	—H	—H	—Me	—H	—H	—Br	−43.69	>20	0.75 ± 0.25	>20	>20	>26
1g	—H	—H	—Me	—H	—H	—I	−44.09	>20	1.01 ± 0.19	>20	>20	>20
1h	—H	—H	—NH2	—H	—H	—Me	−41.20	>20	0.37 ± 0.09	>20	2.24 ± 0.65	6.1
1i	—H	—H	—Br	—H	—H	—Me	−41.68	>20	4.35 ± 0.63	>20	12.8 ± 3.2	2.9
1j	—H	—H	—NO2	—H	—H	—Me	−41.25	>20	4.39 ± 1.04	>20	>20	>4.5

4. DISCUSSION

Similarly to human BDs, CaBDs contain a flexible binding pocket associated with structurally conserved waters that are necessary to take into account when performing computational modeling. A MD study of CaBD1 and CaBD2 reveals that the stability of the five conserved waters varies between binding sites. Structural water W1 in CaBD2 was shown to be less stable than the other bound water molecules, allowing it to be displaced by potent inhibitors. This finding was crucial for accurate molecular modeling of the co‐crystal structure, which showed inhibitor 1 displacing W1 in CaBD2. Specifically, the conserved water W1 must be removed before performing ligand docking on CaBD2 to achieve accurate binding predictions.

The experimental results obtained from the HTRF inhibition assays are generally consistent with the relative binding interaction energies predicted for our compound SAR suite 1a–1j in the ligand docking analysis. A key role in binding is played by formation of a hydrogen bond between the hydrogen of the OH of the ligand and the W2 oxygen (Figure 2). The relatively lower activity of compounds 1a, 1b, and 1 may be due to the decreased pK _a of the phenol OH caused by introduction of the F substituent (7.86, 8.10, and 9.27 for 1a, 1b and 1, respectively; calculated by MarvinSketch 20.3) (Figure S43).

Addition of a hydroxyl group ortho to the methyl group in position R3 in 1c resulted in increased potency compared to substitution at the meta position (1d). This could be due to an increased distance from Leu420, weakening or preventing hydrogen bonding. Lowering the pK _a by replacing the methyl group with halogen atoms (—Cl, —Br, —I; compounds 1e–1g) enhanced the potency up to 3‐fold (Table 1 and Figure S42). The docking results predicted a slight difference in activities between these compounds (I > Br > Cl in order of decreasing potency), which was too small to detect in the experimental inhibition assays. The deprotonated forms of these compounds are predicted to form a hydrogen bond with the W5 water bridge. A pK _a calculation indicates that the —OH in all these halo‐substituted ligands is deprotonated at pH 7, rationalizing the similar potencies observed. Whereas the potencies of compounds 1i and 1j, in which the methyl group on the phenyl ring was replaced by a nitro‐ or bromo‐ substituent, were similar to that of the original imidazopyridine 1, an amino group at this position in compound 1h yielded a 10‐fold enhancement of potency, likely due to hydrogen bonding to Leu420 as predicted in the docking calculations.

5. CONCLUSIONS

The discovery of inhibitor 1 demonstrated the feasibility of selectively inhibiting CaBD2 vs human BET BD2. ¹⁷ The present study demonstrates the use and advantages of MD studies to facilitate the design of more potent and selective inhibitors. MD simulations provide a more realistic understanding of the binding mode and were here explored as a potential computational tool for design of drugs targeting fungal BDs. The co‐crystal structure of 1 bound to CaBD2 reveals that W1, a conserved water molecule close to the binding site, is missing when the bound ligand is present. Displacement of this water molecule by 1 is predicted by the MD analysis, in terms of lowered occupancy compared to the other conserved waters. This insight informed creation of a series of compounds predicted to have higher BD2 binding affinity than 1.

Compounds chosen to have decreased pK _a (1e–1g) relative to 1 showed a 3‐ to 5‐fold decrease in IC₅₀ while still retaining good selectivity against human Brd4 BD2 (>10 fold). This confirms a role of the imidazole nitrogen in binding. The methyl group on the imidazopyridine ring also was found to contribute to selectivity. Compounds 1c and 1h, designed to form additional hydrogen bonds with Leu420 (Figure 6), showed a 5‐fold (1c) or 10‐fold (1h) decrease in IC₅₀. In conclusion, we identified and rationalized displacement of a structural water of CaBD2, W1, using MD as a key mechanism in selective ligand binding to this site. This allowed development of a more accurate model for docking calculations and in silico SAR predictions, which were validated by experimental IC₅₀ data. Apart from application to CaBD2 as in the present study, this approach could be usefully extended to similar ligand‐receptor systems to achieve more active compound predictions.

Supporting information

Appendix S1 (1) water position parameters, pKa calculation data, binding mode comparison details, druggability analyses (Schrodinger SiteMap); (2) chemical characterization data for 1a‐1j (¹H, ¹³C, ¹⁹F NMR spectra and MS data)

Zhou Y., Overhulse J. M., Dupper N. J., Guo Y., Kashemirov B. A., Wei K., Govin J., Petosa C., McKenna C. E., J. Comput. Chem. 2022, 43(32), 2121. 10.1002/jcc.26997

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.