John Karanicolas

Unlimited genetic and epigenetic diversity across the patient population, and within the tumor and among metastatic lesions of an individual patient, presents a challenge in developing targeted therapies to treat epithelial ovarian cancer (EOC). To find the Achilles heel of ovarian cancers, we designed an RNAi-based screen (>6,000 genes) to identify points of molecular vulnerability. These screens identified KIF11 as essential to maintain tumor cell viability. KIF11 encodes kinesin Eg5 (KIF11), a motor protein required for microtubule antiparallel sliding during mitosis that has been targeted clinically. Although KIF11 inhibitors are well tolerated by EOC patients, clinical response rates were disappointing in heavily treated patients with advanced disease. We hypothesize one mechanism through which KIF11 inhibitors’ efficacy is muted is via functional compensation of mitotic spindle assembly by KIF15, a second motor kinesin. Using a synthetic lethal approach, we have shown that RNAi-mediated silencing of KIF15 has no effect on cell viability, but significantly sensitized cells by >10- to 100-fold to a class of KIF11 inhibitors (KIF11is). TPX2-targeted siRNAs reduce cell viability by 5-fold alone, and sensitize cells by 100-fold in the presence of KIF11i. Based on these observations, we designed a high-throughput screening assay using Alpha Technology©. Informed by previous predictions of KIF15-TPX2 binding domains, we designed peptides corresponding to the appropriate domains of KIF15 (aa1149-1388, GST-tagged) and TPX2 (aa346-747, His-tagged), for use in the AlphaScreen^© assay. Following several optimization steps (e.g., buffer selection, protein titration, TruHits^© assay validation), we demonstrate that KIF15 binds TPX2, producing a signal >30-fold greater than background signal. Further, peptides designed to mimic either KIF15 or TPX2 binding domains, but not scrambled controls, exhibited concentration-dependent disruption of the KIF15-TPX2 interaction. Next, we screened a FDA library (3,474 compounds) and bioactive (1,902 compounds) library and identified 122 hits; fresh compound stocks were tested at 4 concentrations (20, 10, 5, and 2.5 μM), resulting in 53 confirmed hits (28 FDA, 25 bioactive). To identify false positives that interfere with assay chemistry, we repeated the screen at 3 concentrations, replacing the His-TPX2 and GST-KIF15 proteins with a His-GST fusion. As an orthogonal screen, we performed a Homogenous Time Resolved Fluorescence© (HTRF)-based screen to reconfirm hit compounds at the aforementioned concentrations. From these combined data, we have identified 4 hit compounds (KU-P001 to KU-P004) that disrupt KIF15/TPX2 protein-protein interactions. To further prioritize our hits, we developed in vitro (EOC cell lines were exposed to increasing concentrations, up to 1 µM, of KIF11i for 4 months) and in vivo (patient-derived ascites xenograft models of EOC were treated with 3 cycles of either 5 mg/kg or 10 mg/kg of a KIF11i and were serially passaged and treated for 3 additional generations) models of KIF11i-resistance to evaluate their activities. In resistant EOC cells we observed a 3-fold increase in KIF15 protein expression as compared with parental cells; TPX2 protein expression remained unchanged. In vivo, both KIF15 (68% increase) and TPX2 (2-fold) protein expression was increased in resistant versus untreated PDX tumor cells. Combination studies of KU-P004 and ispinesib (KIF11i) in vitro revealed a 2- to 64-fold increase in cell death for both responsive and resistant EOC cells compared with KIF11i treatment alone. In summary, we have developed a high-throughput screening assay platform to identify inhibitors of the KIF15-TPX2 interaction and are using iterative medicinal chemistry to guide optimization and functional evaluation of hits and lead compounds to develop new combination therapies for women with ovarian cancer. Citation Format: Rebecca J. Wates, Anuradha Roy, Frank Schoenen, John Karanicolas, Scott Weir, Andrew Godwin. Targeting the KIF11/KIF15/TPX2 axis to develop new therapies for ovarian cancer. [abstract]. In: Proceedings of the AACR Conference: Addressing Critical Questions in Ovarian Cancer Research and Treatment; Oct 1-4, 2017; Pittsburgh, PA. Philadelphia (PA): AACR; Clin Cancer Res 2018;24(15_Suppl):Abstract nr B03.

Chemical biology has long sought to build protein switches for use in molecular diagnostics, imaging, and synthetic biology. The overarching challenge for any type of engineered protein switch is the ability to respond in a selective and predictable manner that caters to the specific environments and time scales needed for the application at hand. We previously described a general method to design switchable proteins, called “chemical rescue of structure”, that builds de novo allosteric control sites directly into a protein’s functional domain. This approach entails first carving out a buried cavity in a protein via mutation, such that the protein’s structure is disrupted and activity is lost. An exogenous ligand is subsequently added to substitute for the atoms that were removed by mutation, restoring the protein’s structure and thus its activity. Here, we begin to ask what principles dictate such switches’ response to different activating ligands. Using a redesigned β-glycosidase enzyme as our model system, we find that the designed effector site is quite malleable and can accommodate both larger and smaller ligands, but that optimal rescue comes only from a ligand that perfectly replaces the deleted atoms. Guided by these principles, we then altered the shape of this cavity by using different cavity-forming mutations, and predicted different ligands that would better complement these new cavities. These findings demonstrate how the protein switch’s response can be tuned via small changes to the ligand with respect to the binding cavity, and ultimately enabled us to design an improved switch. We anticipate that these insights will help enable the design of future systems that tune other aspects of protein activity, whereby, like evolved protein receptors, remolding the effector site can also adjust additional outputs such as substrate selectivity and activation of downstream signaling pathways.

Protein–protein interactions play important roles in virtually all cellular processes, making them enticing targets for modulation by small-molecule therapeutics: specific examples have been well validated in diseases ranging from cancer and autoimmune disorders, to bacterial and viral infections. Despite several notable successes, however, overall these remain a very challenging target class. Protein interaction sites are especially challenging for computational approaches, because the target protein surface often undergoes a conformational change to enable ligand binding: this confounds traditional approaches for virtual screening. Through previous studies, we demonstrated that biased “pocket optimization” simulations could be used to build collections of low-energy pocket-containing conformations, starting from an unbound protein structure. Here, we demonstrate that these pockets can further be used to identify ligands that complement the protein surface. To do so, we first build from a given pocket its “exemplar”: a perfect, but nonphysical, pseudoligand that would optimally match the shape and chemical features of the pocket. In our previous studies, we used these exemplars to quantitatively compare protein surface pockets to one another. Here, we now introduce this exemplar as a template for pharmacophore-based screening of chemical libraries. Through a series of benchmark experiments, we demonstrate that this approach exhibits comparable performance as traditional docking methods for identifying known inhibitors acting at protein interaction sites. However, because this approach is predicated on ligand/exemplar overlays, and thus does not require explicit calculation of protein–ligand interactions, exemplar screening provides a tremendous speed advantage over docking: 6 million compounds can be screened in about 15 min on a single 16-core, dual-GPU computer. The extreme speed at which large compound libraries can be traversed easily enables screening against a “pocket-optimized” ensemble of protein conformations, which in turn facilitates identification of more diverse classes of active compounds for a given protein target.

Protein-protein interactions are among today’s most exciting and promising targets for therapeutic intervention. To date, identifying small-molecules that selectively disrupt these interactions has proven particularly challenging for virtual screening tools, since these have typically been optimized to perform well on more “traditional” drug discovery targets. Here, we test the performance of the Rosetta energy function for identifying compounds that inhibit protein interactions, when these active compounds have been hidden amongst pools of “decoys.” Through this virtual screening benchmark, we gauge the effect of two recent enhancements to the functional form of the Rosetta energy function: the new “Talaris” update and the “pwSHO” solvation model. Finally, we conclude by developing and validating a new weight set that maximizes Rosetta’s ability to pick out the active compounds in this test set. Looking collectively over the course of these enhancements, we find a marked improvement in Rosetta’s ability to identify small-molecule inhibitors of protein-protein interactions.

Over the past decade, protein-protein interactions have emerged as attractive but challenging targets for therapeutic intervention using small molecules. Due to the relatively flat surfaces that typify protein interaction sites, modern virtual screening tools developed for optimal performance against “traditional” protein targets perform less well when applied instead at protein interaction sites. Previously, we described a docking method specifically catered to the shallow binding modes characteristic of small-molecule inhibitors of protein interaction sites. This method, called DARC (Docking Approach using Ray Casting), operates by comparing the topography of the protein surface when “viewed” from a vantage point inside the protein against the topography of a bound ligand when “viewed” from the same vantage point. Here, we present five key enhancements to DARC. First, we use multiple vantage points to more accurately determine protein-ligand surface complementarity. Second, we describe a new scheme for rapidly determining optimal weights in the DARC scoring function. Third, we incorporate sampling of ligand conformers “on-the-fly” during docking. Fourth, we move beyond simple shape complementarity and introduce a term in the scoring function to capture electrostatic complementarity. Finally, we adjust the control flow in our GPU implementation of DARC to achieve greater speedup of these calculations. At each step of this study, we evaluate the performance of DARC in a “pose recapitulation” experiment: predicting the binding mode of 25 inhibitors each solved in complex with its distinct target protein (a protein interaction site). Whereas the previous version of DARC docked only one of these inhibitors to within 2 Å RMSD of its position in the crystal structure, the newer version achieves this level of accuracy for 12 of the 25 complexes, corresponding to a statistically significant performance improvement (p < 0.001). Collectively then, we find that the five enhancements described here – which together make up DARC 2.0 – lead to dramatically improved speed and performance relative to the original DARC method.

Protein–protein interactions represent an exciting and challenging target class for therapeutic intervention using small molecules. Protein interaction sites are often devoid of the deep surface pockets presented by “traditional” drug targets, and crystal structures reveal that inhibitors typically engage these sites using very shallow binding modes. As a consequence, modern virtual screening tools developed to identify inhibitors of traditional drug targets do not perform as well when they are instead deployed at protein interaction sites. To address the need for novel inhibitors of important protein interactions, here we introduce an alternate docking strategy specifically designed for this regime. Our method, termed DARC (Docking Approach using Ray-Casting), matches the topography of a surface pocket “observed” from within the protein to the topography “observed” when viewing a potential ligand from the same vantage point. We applied DARC to carry out a virtual screen against the protein interaction site of human antiapoptotic protein Mcl-1 and found that four of the top-scoring 21 compounds showed clear inhibition in a biochemical assay. The K_i values for these compounds ranged from 1.2 to 21 μM, and each had ligand efficiency comparable to promising small-molecule inhibitors of other protein–protein interactions. These hit compounds do not resemble the natural (protein) binding partner of Mcl-1, nor do they resemble any known inhibitors of Mcl-1. Our results thus demonstrate the utility of DARC for identifying novel inhibitors of protein–protein interactions.

Small-molecules that inhibit interactions between specific pairs of proteins have long represented a promising avenue for therapeutic intervention in a variety of settings. Structural studies have shown that in many cases, the inhibitor-bound protein adopts a conformation that is distinct from its unbound and its protein-bound conformations. This plasticity of the protein surface presents a major challenge in predicting which members of a protein family will be inhibited by a given ligand. Here, we use biased simulations of Bcl-2-family proteins to generate ensembles of low-energy conformations that contain surface pockets suitable for small molecule binding. We find that the resulting conformational ensembles include surface pockets that mimic those observed in inhibitor-bound crystal structures. Next, we find that the ensembles generated using different members of this protein family are overlapping but distinct, and that the activity of a given compound against a particular family member (ligand selectivity) can be predicted from whether the corresponding ensemble samples a complementary surface pocket. Finally, we find that each ensemble includes certain surface pockets that are not shared by any other family member: while no inhibitors have yet been identified to take advantage of these pockets, we expect that chemical scaffolds complementing these “distinct” pockets will prove highly selective for their targets. The opportunity to achieve target selectivity within a protein family by exploiting differences in surface fluctuations represents a new paradigm that may facilitate design of family-selective small-molecule inhibitors of protein-protein interactions. Despite intense interest and considerable effort, there are few examples of small molecules that directly inhibit protein-protein interactions. Crystal structures of early successes have highlighted the plasticity of the protein surface, as some inhibitor-bound proteins are captured in conformations that are distinct from both their unbound and their protein-bound conformations. The lack of a single well-defined structure presents a challenge for predicting the members of a protein family to which a given compound will show activity (ligand selectivity). Here we generate ensembles of conformations from simulation, and show that we can predict ligand selectivity based on which family members sample conformations complementary to the ligand. This approach may present a new avenue for designing highly-selective inhibitors of protein-protein interactions.

The ability to selectively activate function of particular proteins via pharmacological agents is a longstanding goal in chemical biology. Recently, we reported an approach for designing a de novo allosteric effector site directly into the catalytic domain of an enzyme. This approach is distinct from traditional chemical rescue of enzymes in that it relies on disruption and restoration of structure, rather than active site chemistry, as a means to achieve modulate function. However, rationally identifying analogous de novo binding sites in other enzymes represents a key challenge for extending this approach to introduce allosteric control into other enzymes. Here we show that mutation sites leading to protein inactivation via tryptophan-to-glycine substitution and allowing (partial) reactivation by the subsequent addition of indole are remarkably frequent. Through a suite of methods including a cell-based reporter assay, computational structure prediction and energetic analysis, fluorescence studies, enzymology, pulse proteolysis, X-ray crystallography, and hydrogen–deuterium mass spectrometry, we find that these switchable proteins are most commonly modulated indirectly, through control of protein stability. Addition of indole in these cases rescues activity not by reverting a discrete conformational change, as we had observed in the sole previously reported example, but rather rescues activity by restoring protein stability. This important finding will dramatically impact the design of future switches and sensors built by this approach, since evaluating stability differences associated with cavity-forming mutations is a far more tractable task than predicting allosteric conformational changes. By analogy to natural signaling systems, the insights from this study further raise the exciting prospect of modulating stability to design optimal recognition properties into future de novo switches and sensors built through chemical rescue of structure.

Docking Approach using Ray Casting (DARC) is structure-based computational method for carrying out virtual screening by docking small-molecules into protein surface pockets. In a complementary study we find that DARC can be used to identify known inhibitors from large sets of decoy compounds, and can identify new compounds that are active in biochemical assays. Here, we describe our adaptation of DARC for use on Graphics Processing Units (GPUs), leading to a speedup of approximately 27-fold in typical-use cases over the corresponding calculations carried out using a CPU alone. This dramatic speedup of DARC will enable screening larger compound libraries, screening with more conformations of each compound, and including multiple receptor conformations when screening. We anticipate that all three of these enhanced approaches, which now become tractable, will lead to improved screening results.

Traditional drug targets have historically included signaling proteins that respond to small molecules and enzymes that use small molecules as substrates. Increasing attention is now being directed toward other types of protein targets, in particular those that exert their function by interacting with nucleic acids or other proteins rather than small-molecule ligands. Here, we systematically compare existing examples of inhibitors of protein–protein interactions to inhibitors of traditional drug targets. While both sets of inhibitors bind with similar potency, we find that the inhibitors of protein–protein interactions typically bury a smaller fraction of their surface area upon binding to their protein targets. The fact that an average atom is less buried suggests that more atoms are needed to achieve a given potency, explaining the observation that ligand efficiency is typically poor for inhibitors of protein–protein interactions. We then carried out a series of docking experiments and found a further consequence of these relatively exposed binding modes is that structure-based virtual screening may be more difficult: such binding modes do not provide sufficient clues to pick out active compounds from decoy compounds. Collectively, these results suggest that the challenges associated with such non-traditional drug targets may not lie with identifying compounds that potently bind to the target protein surface, but rather with identifying compounds that bind in a sufficiently buried manner to achieve good ligand efficiency and, thus, good oral bioavailability. While the number of available crystal structures of distinct protein interaction sites bound to small-molecule inhibitors is relatively small at present (only 21 such complexes were included in this study), these are sufficient to draw conclusions based on the current state of the field; as additional data accumulate it will be exciting to refine the viewpoint presented here. Even with this limited perspective however, we anticipate that these insights, together with new methods for exploring protein conformational fluctuations, may prove useful for identifying the “low-hanging fruit” among non-traditional targets for therapeutic intervention.