Human African trypanosomiasis threatens 50 million people in sub-Saharan Africa.(2
) As current therapies are either too dangerous or too limited, novel drugs are urgently needed. UDP-galactose 4′-epimerase (Tb
GalE), a protein critical to T. brucei
survival, is one potential drug target. We here use computer-aided drug design to identify 14 low-micromolar inhibitors of Tb
Computer Docking and Protein Dynamics
Traditional computer-docking methodologies often fail to identify true-positive inhibitors because the static protein structures typically used do not capture the highly dynamic reality of small-molecule binding.24,25
When a ligand approaches its receptor in vitro or in vivo, it encounters not a static protein structure but rather an ensemble of structures as the protein “breathes” in solution. Upon ligand binding, the population of receptor active-site configurations sampled by the protein may shift to better accommodate the ligand. Additionally, under the influence of a bound ligand, the protein may assume novel active-site configurations not sampled by the apo
protein at all.(26
To better understand TbGalE dynamics, we performed five 20 ns molecular dynamics (MD) simulations. Five short simulations were chosen, as opposed to one long simulation, in order to increase the diversity of protein conformations sampled and to ensure that the conformations sampled were geometrically similar to the known crystal structure. As TbGalE is a homodimer, each simulation provided two monomer trajectories, further increasing the diversity of conformations sampled. Clustering was subsequently used to identify 24 protein structures with active sites representative of the many active-site configurations sampled during the MD simulation. These 24 representative protein structures are said to constitute an ensemble.
The protein conformations of the ensemble proved useful in subsequent computer-docking studies.(27
) Recently, a new docking protocol called the relaxed complex scheme (RCS)(25
) has been developed that takes into account full protein flexibility. Rather than docking candidate compounds into a static protein receptor, compounds are docked into an ensemble of protein conformations typically extracted from an MD simulation. The compounds are then ranked by an ensemble-based score that accounts for active-site dynamics. The RCS has already been used to successfully identify inhibitors of FKBP,(28
) HIV integrase,(29
) and T. brucei
RNA editing ligase 1.(30
In the current work, we used AutoDock Vina (Vina)(31
) to perform a RCS screen of the NCI Diversity Set II into the 24 ensemble conformations extracted from the MD simulation. Like previous versions of AutoDock, Vina is freely available to the academic community. Additionally, it is 2 orders of magnitude faster than AutoDock 4.0 (AutoDock),(32
) the previous version. Vina performs well relative to AutoDock; while AutoDock is slightly better at predicting the energy of binding (standard error of 2.2 kcal mol−1
versus 2.8 kcal mol−1
), Vina more accurately reproduces cocrystallized ligand poses.31,32
To our knowledge, Vina has never been used in a RCS screen.
Compounds were docked into both the UDP-glucose and NAD+
binding pockets and were ranked by both an ensemble-average and an ensemble-best scoring scheme (Supporting Information
). Twenty-six high-scoring compounds were subsequently tested experimentally.
Experimental Validation Confirms Multiple Hits from the Primary Screen
Of the 26 compounds of the primary screen, 10 showed >50% average inhibition at 100 μM. Interestingly, at this same concentration, six compounds showed greater than 2-fold stimulation, suggesting allosteric cooperativity between the two monomers of the Tb
GalE homodimer, in harmony with previous studies that demonstrated GalE allostery in Kluyveromyces fragilis
and Saccharomyces fragilis
As we do not expect computer docking to be able to distinguish between an agonist and an antagonist, the effective hit rate of the primary screen was therefore 62% at 100 μM.
The 10 100-μM inhibitors were subsequently tested at 10 μM. Three showed >~50% average inhibition; compounds 1
, and 3
(clorobiocin) had IC50
values of 3.6 ± 0.8, 5.6 ± 0.8, and 5.0 ± 1.2 μM, respectively, and Hill slopes of 1.8 ± 0.6, 1.2 ± 0.3, and 2.1 ± 0.5, respectively (Tables and S1 in Supporting Information
). Interestingly, compounds 1
share a 2′-(phenylcarbamoyl)-[1,1′-biphenyl]-2-carboxylic acid core scaffold.
The 14 Low-Micromolar TbGalE inhibitors Indentifieda
In one recent study, 95% of the inhibitors identified in a high-throughput screen acted through a nonspecific aggregation-based mechanism. This same study suggested that aggregation-based inhibition typically produces steep Hill slopes that are much greater than unity, with average values around 2.2.(35
) As the Hill slopes of compounds 2
(clorobiocin) were significantly greater than unity (Table S1, Supporting Information
GalE inhibition was measured at 30 μM inhibitor concentration (~5 × IC50
) in the presence and absence of a detergent that disrupts colloidal aggregates (0.06% n
-octylglucopyranoside). No significant differences in inhibition were noted, suggesting that inhibition is specific rather than aggregation-based.(30
) Given that the possibility of aggregation was eliminated, the steep Hill slopes of these two compounds provide further evidence for allostery between the two monomers of the Tb
Clorobiocin: An Interesting Inhibitor
One of the hits from the primary virtual screen, compound 3
(clorobiocin), an aminocoumarin derived from several Streptomyces
species, has previously been shown to inhibit the growth of Trypanosoma cruzi
), a close relative of T. brucei
) As clorobiocin is a known bacterial topoisomerase II inhibitor, some have hypothesized that topoisomerase II may be the T. cruzi
protein target as well,(39
) although other targets could not be ruled out.(38
) The current work suggests that UDP-galactose 4′-epimerase may also be among the proteins targeted by this apparently polypharmacophoric compound.
We note with interest that novobiocin, a compound structurally similar to clorobiocin that likewise inhibits the growth of T. cruzi
) did not show greater than 50% Tb
GalE inhibition at 10 μM despite the fact that our computational model predicted high binding affinity. The crude scoring function employed by Vina, optimized not only for accuracy but also for speed, seems unable to differentiate between the apparently subtle differences in the protein−ligand interactions of clorobiocin and novobiocin.
Predicted Binding Poses of Top Inhibitors
Computer docking suggests that compound 3 (clorobiocin) occupies the NAD+-binding pocket. To further characterize the clorobiocin binding pose, we examined the Vina scores of clorobiocin docked into each of the protein configurations of the ensemble and selected the binding pose/protein configuration associated with the best score for further analysis. The predicted protein−ligand interactions are represented schematically in Figure .
The predicted ligand−protein interactions of clorobiocin bound to TbGalE, shown schematically.
Clorobiocin is predicted to participate in multiple hydrogen bonds with the backbone atoms of amino acids lining the NAD+ binding pocket. One ligand hydroxyl group is predicted to form two hydrogen bonds with the backbone amines of N202 and A203; a second ligand hydroxyl group is predicted to form hydrogen bonds with the backbone amines of V35, G36, and S33. Finally, a ligand secondary amine is predicted to form a hydrogen bond with the backbone carbonyl oxygen atom of A100. We note again, however, that many of these same protein−ligand interactions characterize the predicted binding mode of novobiocin as well, despite the fact that novobiocin binding to TbGalE is weak.
The binding modes of compounds 1 and 2, both predicted to bind in the UDP pocket, were not so paradoxical. These compounds contain similar 2′-(phenylcarbamoyl)-[1,1′-biphenyl]-2-carboxylic acid core scaffolds (Table ) and similar predicted binding modes (Figure ). For each of these two ligands, we again examined the Vina scores of the ligand docked into the various protein configurations of the ensemble and selected the binding pose/protein configuration associated with the best score for further analysis (Figure ). The core scaffold common to compounds 1 and 2 participates in many of the same hydrogen bonds that characterize UDP-glucose binding. R335, N202, and R268 all form hydrogen bonds with the diphosphate moiety of UDP-glucose; they likewise form hydrogen bonds with the carboxylate group and the carbonyl oxygen atom of the core scaffold. Additionally, H221 may also form a hydrogen bond with the carboxylate group of the core scaffold (Figure ).
The predicted binding modes of compounds 1 and 2. Some portions of the protein have been removed to facilitate visualization. Hydrogen bonds between the protein and ligand are shown as black lines.
Cation−π interactions seem to play a critical role in the binding of compounds 1
. In these interactions, the positive charge of the cation is electrostatically attracted to the quadrupole moment of an aromatic group.(40
) If the predicted binding of the core scaffold of compounds 1
is correct, both R335 and R268 participate in cation−π interactions with the ligand.
Compounds 1 and 2 have polyaromatic moieties that extend into the pocket normally occupied by the uridine and ribose moieties of UDP-glucose. Much work can yet be done to optimize these fused ring systems in order to improve binding, as hinted at by the hydrogen bond predicted to form between compound 1 and the backbone carbonyl oxygen atom of T220 (Figure a). Other potential interacting groups in the uridine and ribose portions of the UDP-glucose binding pocket include the backbone carbonyl oxygen atom of P253 and the backbone amine of F255. It may also be possible to add moieties to the fused ring system that exploit the F255 aromatic side chain, which can participate in π−π and cation−π interactions.
Finally, the nucleophilic side-chain thiol of C266 is attractive from a drug-design perspective; fragments with electronegative moieties could be added to the fused ring systems of compounds 1
in order to facilitate the formation of a covalent adduct with the C266 thiol group, a strategy similar to that employed by Kerr et al. in designing potent vinyl-sulfone inhibitors of parasitic cysteine proteases, including K777, which, pending FDA approval, will soon enter phase I clinical trials.(41
) Additionally, as this cysteine is a glycine in Hs
GalE, compounds that target C266 may be selective for the trypanosomal form of the enzyme, although caution is advised as nonspecific reactions with protein thiols could lead to toxic side effects.
The Secondary Virtual Screen: A Similarity Search
Encouraged by these initial results, we next searched the entire NCI database for compounds similar to the three low-micromolar inhibitors verified experimentally. Eighty additional compounds were subjected to the same RCS protocol used in the primary virtual screen, and 14 novel compounds were thus identified as potential binders. One of these compounds, kedarcidin, a structural orthologue of clorobiocin, was unavailable from the NCI, and another failed the purity criteria. Thus, 12 compounds were subsequently tested experimentally.
Eleven of the 12 compounds showed >50% average inhibition at 10 μM. Subsequent experimental analysis confirmed that these compounds had IC50 values between 0.9 and 6.8 μM (Table ). As previously, aggregation effects were excluded, and compound identity and purity were confirmed by accurate mass and LC-MS.
Taken together with the compounds of the initial screen, these inhibitors constitute a novel hit series based on a 2′-(phenylcarbamoyl)-[1,1′-biphenyl]-2-carboxylic acid scaffold. Additionally, a singleton hit (clorobiocin) was also found to be potent, although novobiocin, a related compound that might have otherwise been considered a member of the same hit series, was not.
All low-micromolar Tb
GalE inhibitors identified were tested for their ability to inhibit the growth of cultured T. brucei
and human liver MRC5 cells using the established Alamar Blue protocol.42,43
Two compounds containing the 2′-(phenylcarbamoyl)-[1,1′-biphenyl]-2-carboxylic acid core scaffold, compounds 12
, had EC50
values of 24.4 and 28.5 μM against whole-cell T. brucei
, respectively. Additionally, the natural product clorobiocin (compound 3
) had an EC50
value of 4.4 μM. Only compound 14
, a compound with no activity against whole-cell T. brucei
, demonstrated inhibition of MRC5 growth (Table ).
To better understand why most of the Tb
GalE inhibitors failed to inhibit whole-cell T. brucei
, we calculated the LogP value of each (Table , Supporting Information
). With only one exception, the logP values were all high, either near the upper bound for what is considered “druglike” or, in several cases, well beyond that bound.44,45
As these compounds are very hydrophobic, we postulate that they are retained in the cellular membrane, explaining the reduced efficacy against whole-cell T. brucei
. The one compound with a druglike LogP value was compound 3
(clorobiocin), which has a measured whole-cell EC50
of 4.4 μM, suggesting inhibition of intracellular targets including Tb
GalE and, potentially, topoisomerase II.(39
) As an alternate explanation for poor membrane traversal, we note also that the charged carboxylate moiety characteristic of the main hit series, absent in clorobiocin, may likewise block access to intracellular targets.
Designing compounds similar to those based on the 2′-(phenylcarbamoyl)-[1,1′-biphenyl]-2-carboxylic acid core scaffold but with reduced hydrophobicity should not be difficult. We note that these inhibitors have roughly the same IC50 values regardless of the fused ring system that is attached to the core scaffold. This suggests that the fused ring systems likely contribute to the overall binding affinity principally through nonspecific interactions, if at all. Hydrophilic functional groups could likely be added to the fused ring systems without compromising the overall binding affinity, thereby increasing the propensity of these molecules to traverse cellular membranes.
Several of the fused ring systems have functional groups that may be amenable to chemical modification. Hydrophilic moieties could be added to the fused ring systems of compounds 1 and 6 via substitution at the hydroxyl group, for example. Several of the TbGalE inhibitors identified also have aromatic halides that could facilitate fragment addition.
As these compounds are already quite large, however, a better approach may be to entirely substitute the fused ring systems of the compounds tested here with other more hydrophilic molecular fragments. Reactions between the readily available compound [1,1′-biphenyl]-2,2′-dicarboxylic acid (NSC1966) and varied aromatic and even nonaromatic amines could expand the chemical series explored here to include compounds with more favorable partition coefficients.
If the above modifications fail to produce high-affinity inhibitors of cellular growth, the charged carboxylate of the main-series compounds could be methylated to potentially facilitate membrane traversal. Once in the cell, this prodrug could perhaps be demethylated by cellular esterases.(46