Analysis of crystal structures of EF with various substrate analogues was the first step in our design process. EF can be allosterically activated by the presence of other proteins, such as calmodulin, which is a Ca²⁺ ion sensor present in host cells. Inhibitors targeting sites for such allosteric activators have recently been identified [15]. Our studies focused on the active site (circled in the structure of EF bound to calmodulin, shown in Figure 1Top). Comparison of the active site conformation in various crystal structures in the Protein database (PDB) (which differed in the number and types of bound metal ions and substrates [16]) revealed important information about how the active site of the toxin differed from the mammalian adenyl cyclase enzymes. These crystal structures, with or without the bound metal ions, were used for docking potential inhibitors identified by our fragment based pharmacophore.

A fragment library was built that contained small molecules with at most one rotatable bond. The HINT program was used to select those fragments that bound to areas in the active site of EF. The Hydropathic INTeractions, or HINT, program [18,19,20] uses experimental solvent partitioning data as a basis for calculating free energy scores of binding. Interaction energy calculations used to score fragment binding included terms for hydrophobic, ionic, and hydrogen bond interactions (Figure 2 and Figure 3). Initially, a smaller library, from the NCI, was screened with the pharmacophore and 8 compounds selected from this list that had particularly good scores with the FlexX docking program. Then these compounds were used to identify larger fragments that were used to screen the ZINC library for compounds.

Molecular docking played an important role in our identification of a novel inhibitor of EF, compound 1 [23] (Figure 3B top), when used in combination with pharmacophore-based compound selection, and experimental screening with a cell-based bioassay [16,24]. Autodock was used for this work, after we compared several different docking programs for how well they were able to reproduce the position within the active site of analogues of ATP compared to that of the crystal structure. This work revealed that that the protonation state of inhibitors had a pronounced effect on docking, consistent with work by others [25]. From about 5000 compounds chosen in our initial screening of the NCI and ZINC libraries, about 20 purchasable compounds were assayed as described below. Of these, three compounds were active at concentrations below 10 µM in the cell assay. One of these, (compound 1) proved to be an effective lead compound, as its activity was reproducible; it had reasonable (but not ideal) aqueous solubility, and showed no toxicity in our initial tests.

All compounds and synthesized derivatives were dissolved in DMSO and then diluted at least 100 fold into cell culture medium before assay. Their ability to reduce total extracellular cAMP production in mammalian cells treated with EF/PA was determined. Assays were done in triplicate and samples with IC₅₀ values < 20 μM were assayed at least twice on different days, and re-assayed in 2-fold dilution steps to reduce errors at lower doses.

Our standard assay for the activity of the selected inhibitors was inhibition of cAMP released to the medium of cells treated with edema toxin (EF + PA). An example of the assay, for two of our inhibitors, is shown in Figure 5.

Our reliance on this assay was for several reasons. First, experience has shown that direct inhibitors of enzyme activity can fail at the level of the whole cell, due to inability to enter the cell, or problems with toxicity that may be direct effects or due to off target effects on other cellular enzymes. Further, the cell assay was exquisitely sensitive, able to measure the effects of very small amounts of EF (administered with PA). This assay gives more variability than using isolated enzymes, but it is biologically more relevant, as the concentrations needed to inhibit diarrhea and intestinal damage in the ETEC murine model (7.5–15 μg/mouse) were similar to those indicated by the cell culture IC₅₀ values for compound 1 [26], as discussed below.

Due to the cost of testing the inhibitors against B. anthracis infection, assays for which must be done in BSL-3 conditions, a BSL-2 experiment was conducted to determine whether our inhibitors could prevent intestinal edema and diarrhea during entertoxigenic E. coli (ETEC) infection in mice. This murine model of bacterial infection was used as ETEC produce an adenylyl cyclase toxin that has a high degree of identity to EF, known as heat-labile enterotoxin (LT) [27]. ETEC is a leading cause of traveler’s diarrhea [28,29]. Periodic outbreaks occur in the developing world [30] and with increasing frequency in the US [31,32]. A murine model was developed to test the effect of our inhibitors on the progress of the infection, and particularly development of diarrhea, using a gavage method to infect the animals, with the inhibitor supplied intraperitoneally both before and after the inoculation of the mice. In this minimally invasive model, the flow in the intestine was not interrupted, and it thus approximated that of a natural infection. Our lead compound significantly decreased intestinal colonization of ETEC in this model. However, the toxin inhibitor did nothing to inhibit the growth of several different pathogenic bacteria in flask culture [26]. This example illustrates the need for testing toxin inhibitors in animal models, as in this case the cAMP secretion induced by the toxin might have served to inhibit cell attachment by the bacteria [27] or as a quorum sensor that stimulates bacterial replication [33]. This sort of activity would not be clear from its effects on bacteria growing rapidly in normal culture medium, where the toxin would not be required as a growth stimulator [34].

Compound 1 inhibited cAMP production induced by PA and EF (IC₅₀ 2–9 μM in cells) [35] and reduced bacterial colonization and diarrhea caused by ETEC in mice with an oral gavage dose of 15 μg/mouse. While there was no obvious toxic effect in the cell cultures or mouse studies, and the Ames II™ Mutagenicity Assay determined the compound 1 to be non-mutagenic in all conditions tested, the compound gave a positive response for genotoxicity in a Green Screen HC [GADD45α (growth arrest and DNA damage gene)-Green Fluorescent Protein (GFP)] assay test on mammalian cells at 10–20× its active dose (62.5 µg/mL vs. active dose in mice of about 3 µg/mL, assuming 5 mL serum/mouse).