C.3. Modeling Ligand Interactions with Signaling Molecules

C.3.1. Investigators/areas of scientific expertise: J.S. Lazo (PI), G. Romero, & A. Vogt (Pharmacology, Pitt), B. Day (Pharmaceutical Sci, Pitt); M. Cascio (co-PI) & T.E. Smithgall, (Mol Genetics & Biochem, Pitt); J. R. Stiles (PSC/CMU); I. Bahar & H. Meirovitch (Comp Biology & Bioinformatics, Pitt); P. Wipf (Combinatorial Chemistry, Pitt), R. Coalson (Chemistry, Pitt), J. Madura (Chem & Biochem, and Comp Sci, Duquesne); Ermentrout (Math, Pitt)

C.3.2. Specific Aims

The Specific Aims of this Development Project (DP3) are to design, develop, and implement tools for (1) modeling the interactions of selected kinases and phophatases with their cognate ligands to define new binding sites that can be targeted with small molecules, and (2) investigating the consequences of partial or complete disruption of the intracellular movements of these selected signaling targets after exposure to small molecules.

C.3.3. Background and Significance

C.3.3.1. Cancer is a disease of aberrant signaling. There is an increasing amount of experimental evidence indicating that cancer is fundamentally a disease associated with aberrant signaling processes including mutated growth factor receptors, modified signaling transducers, and disrupted cell cycle checkpoints (224;225). Consequently, cancer is an ideal disease model to probe for scalable computational methods for detecting new small molecules and modeling signaling interactions.

C.3.3.2. Characterization of the molecular interactions involved in cellular signaling is essential for defining molecular targets. Reversible chemical modification and protein-protein interactions dominate as methods used by mammalian cells for intracellular signaling. Among chemical modification methods, phosphorylation and dephosphorylation are most commonly used. There is an emerging understanding of the dynamic aspects of kinase and phosphatase interactions derived primarily from genetic and molecular biological studies. It is hypothesized that partial or transient disruption of specific kinases or phosphatases could be useful in probing the nature of these complex networks, as well as providing putative targets for novel therapeutic agents. Low-molecular weight, serine/threonine phosphatase inhibitors that penetrate living cells, such as okadaic acid and calyculin A, illustrate the utility of such reagents for functional studies. Unfortunately, selective, potent and cell penetrating small molecule inhibitors of most signaling pathways are absent. One important challenge is to develop computational methods that permit characterization of the interactions involved in cellular signaling, to define potential binding sites or pharmacophores that can be targeted with small molecules, and most importantly to correlate the molecular behavior (e.g. receptor-substrate kinetics) to microphysiological effects (e.g. spatial and temporal concentration of downstream effectors) and to cellular fates.

C.3.3.3. Reversible binding and translocation are two essential mechanisms of signal transduction. In many cases, the activation of signaling proteins involves binding interactions with adaptor complexes, cytoskeletal structures, and subcellular membranes or with the targets or activators themselves. These interactions can lead to altered subcellular localization that are highly regulated and reversible, leading to a dynamic signaling model that diverges from the historical “hardwired signaling concept”, in which the signaling protein stayed largely in place and the rapid diffusion of second messengers permitted signal transmission (224). The alternative “softwired-signaling concept” posits that the signaling proteins translocate and undergo reversible binding interactions as essential steps in the signaling pathways. Cancer-relevant examples include the plasma membrane recruitment of SH2 domain containing cytosolic proteins (224), or the cytoplasmic sequestration of Cdc25C or p53 after binding to protein kinase C inhibitor 14-3-3 (226), both to be examined in DP3. Mathematical models also suggest that prime considerations to develop a minimally accurate picture of cell signaling are compartmentalization and translocation of signaling components, which affect the effective concentrations and interactions (227).

C.3.3.4. Subcellular localization or translocation of signaling proteins can be probed by fluorescence microscopy. Until recently, the translocation of cancer relevant target molecules could only be followed microscopically by slow and tedious manual counting that was unsuitable for high-throughput or compound library analyses. Translocation studies in live cells were limited to selected cases where the signaling protein could be fluorescently labeled in vitro and then microinjected. The discovery of green fluorescent proteins (GFP) and its variants, which can be used as expressed fluorescence tags, now permits us to probe spatio-temporal dynamics proteins. The emergence of several imaging cytometers now allows both fixed and live cell analyses of ligand libraries.

Unique opportunities. This Development Project permits the faculty to exploit the close proximity of the College of Arts and Sciences, the School of Pharmacy, and the School of Medicine at the University of Pittsburgh, and the PSC. The availability of unique instrumentation within these institutions permits a new collaboration that will focus on the computational investigation of ongoing peer-reviewed research problems. Notably, the cell cycle signaling and control mechanisms examined in this project (DP3) overlap with those involved in the initiation of DNA repair (DP1) or regulation of apoptotic machinery (DP2). It is clear that the tools developed within the scope of DP1 and DP2 will be equally useful and applicable for DP3, or vise versa. A close cooperation with the other teams is therefore anticipated. In DP3, Cascio, Smithgall, Bahar and Meirovitch will combine experimental and computational methods to initially model the interactions of selected kinases; simultaneously, Lazo and Vogt will develop and implement the automated microscopic methodology to evaluate selected phosphatases. Based on the interactions that already exist among these individuals, we anticipate that the kinase and phosphatase groups will share many reagents and methods. Romero will assist in cell-based GFP protein studies. Chemical libraries will be developed and provided by Wipf, and computational chemistry and toxicology support will be provided by B. Day. Finally, Stiles, in cooperation with Madura, Coalson and Ermentrout, will develop and implement the computational framework for the simulation and visualization of signaling events in the presence of ligand-induced disruptions.

C.3.4. Research Design and Methods.

C.3.4.1. Specific Aim 1: Model the interaction of selected kinases and phosphatases to define new binding sites that can be targeted with small molecules

A. General problem and model system: Complex protein-protein interactions are a shared feature of many signal transduction systems. The recent development of high throughput methodologies such SPR allows for rapid accumulation of protein-protein or receptor-ligand kinetics data that may be used to iteratively refine the computational model and help in designing improved pharmacophores. As an initial model system, we have chosen the interaction of a well-known cytoplasmic protein-tyrosine kinase (c-Src) and a signal transducer and activator of transcription (Stat3). Both c-Src and Stat3 are composed of multiple domains, and their direct interaction is well documented. Members of DP3 have considerable experience with this system, thus, facilitating the overall prototype project (228-230). The goal is to combine experimental and computational models and techniques to determine the molecular mechanisms of kinase activation, substrate phosphorylation, and substrate release. Such information should guide the development of novel inhibitors of the signaling pathway in which they participate, which has been implicated in several forms of cancer and other diseases. The tools developed for this complex will be folded into our long-range goals – multiscale modeling of signaling pathways to measure receptor-ligand kinetics

B. Approach: Members of the Project have substantial preliminary data implicating Stat3 as a direct target for multiple Src kinase family members and demonstrating that the interaction of Src with Stat3 induces Src activation in vivo (231-233). In addition, we also have preliminary data suggesting that the interaction may involve both the SH2 and SH3 domains of Src and as yet unidentified target sequences within Stat3. We have already started to model the SH2 domain of Stat3 (229) using structural homology methods, and we are simultaneously using phage display to identify consensus-binding motifs for this domain. We propose to continue using phage display to identify consensus binding sequences for specific domains, exploit tethered site-directed studies to identify ligands at directed surfaces (234;235), and use SPR technique to determine the rates of association and dissociation, and binding energies. In our preliminary SPR studies, we observed that the SH3 domain of c-Src interacts weakly with a small peptide having the sequence of the putative SH3 ligand of Stat3. The ability to quantify the on and off rates (in addition to equilibrium constants) may be critical in differentiating the determinants important for initial binding (i.e. sites that affect the on rate) from other regions important in avidity (i.e. secondary sites that interact when brought into proximity, and may be reflected in the off rate). These experiments will be performed in iterative cycles in conjunction with computations, which would suggest residues for mutagenesis and further tests to refine our models and allow for the design of specific pharmacophores. The same approach will apply to studying the interactions between Cdc25C and 14-3-3 within the scope of Specific Aim 2.

Computations will be performed at two different levels: full atomic for small molecules and functional domains, and coarse-grained for multiprotein complexes. Full atomic computations will employ docking algorithms that also incorporate domain flexibility. On the experimental side, information on protein flexibility, a property functional for catalysis and substrate recognition, is inferred either indirectly, form the comparison of static X-ray structures, or for relatively small soluble proteins that can be probed by NMR. Computations will complement these approaches in that they can predict local conformational dynamics. A statistical mechanics methodology has been developed by Meirovitch and coworkers (236), which has been successfully used for predict the flexible solution structures and populations of cyclic peptides. This methodology relies on (1) a novel method for optimizing atomic solvation parameters (ASPs), (2) an extensive conformational search with the local torsional deformations (LTD) method, and (3) MC simulations and free energy calculations with the local states (LS) method. This methodology has been recently extended to surface loops in proteins ribonuclease A (64), and the transferability to loops of different size and amino acid sequence is being studied. To this aim, the efficiencies of both LTD and the LS methods are being enhanced. These unique tools are applicable to a wide range of problems in structural biology and drug design; in this project it is suggested to extend them to free-energy based flexible docking of ligands to an active site.

The multiprotein dynamics, on the other hand, will be explored with the GNM-based methods developed by Bahar and co-workers (see § A.3.1), using the PDB structures of Stat3, Src, or close homologs (237-243). Some of these are structures of the order of 10³ residues. Conventional MD simulations cannot efficiently characterize their dynamics. The elastic network models recently developed by Bahar and coworkers (32;90) are ideally suited - and even become more exact (due to the central limit theorem) - for modeling such large structures as shown in recent studies (91;92). Fig C.3.1 illustrates our preliminary results for c-Src, obtained with a computational time cost of a few minutes (real time) using the GNM algorithm with an R10,000 SGI processor. The calculations indicate the high mobility of the SH2 domain and the linker connecting the SH2 and SH3 domains, consistent with the mechanism of inactivation of the enzyme (244). These computations can rapidly identify the hinge sites for different types of collective motions, as well as the most flexible, potentially substrate recognizing sites, which will be combined with phage display and SPR experiments, and the above described novel conformation sampling methods (237), to define the target sites for ligand binding or allosteric regulation.

C.3.4.2. Specific Aim 2: Model the consequences of partial or complete disruption of intracellular movements of selected signaling targets after exposure to small molecules.

A. General Problem: Appropriate temporal and spatial regulation of various biochemical reactions is critical for cellular physiology (245). Although considerable attention has been directed towards the kinetic properties of cancer-related targets, the potential therapeutic benefit of disrupting the spatial properties of cancer-causing targets has largely been ignored. Spatial regulation is exemplified by the placement of an effector molecule in close proximity to its cognate target to allow for an effective interaction. For example, transcription factors must reside within the nucleus to interact with target sites of the transcriptional machinery, or growth factor receptor kinases must be associated with the plasma membrane to initiate their subordinate cascades after contact with extracellular growth factors. The macromolecular movement within the cell are thought to be driven by: (a) generation of new, often protein-protein, interactions (b) second messenger-mediated interactions between proteins and other partners such as lipids, in subcellular structures, or (c) direct interaction of ligands with their cognate receptors or acceptor targets. The best-studied examples of category (a) include the GTPase proteins, such as ras, which participate in the diffusion-mediated recruitment of kinases and other enzymes to the plasma membrane or other subcellular structures. Likewise, tyrosine phosphorylation often triggers the recruitment of SH2 domain-containing proteins to the plasma membrane. A cancer relevant process influenced by protein compartmentalization is the phosphorylation (on Ser216) of the G2/M checkpoint controlling phosphatase Cdc25C by Chk1 in response to DNA damage (see Figure C.2.1). This phosphorylation induces the binding of 14-3-3z to Cdc25c. The resulting cytoplasmic sequestration of Cdc25C is an effective mechanism of cell cycle arrest (196;197).

B. Model System: As part of their ongoing research projects funded by the NIH, biomedical researchers participating in the DP3 are already examining the subcellular localization of a number of essential signaling molecules in human tumor cells, namely in HeLa, PC-3, DU-145 and Hep3B cells, using automated fluorescence microscopy (246;247). Extensive data will be generated from these studies, especially information concerning the simultaneous or interdependent intracellular movement of signaling molecules after cellular treatment with small molecules, which will be exploited for developing computational algorithms. Currently, Drs. Lazo and Vogt are evaluating two chemical libraries for their ability to alter Cdc25C movement within cells; and they will start examining 14-3-3z movement as the next target. The procedure will be extended in future activities (as a Center) to other selected targets such as Stat3, c-Src, Cdk1, Cdc25A, Cdc25B, Erk1, p38, p53, and topoisomerase IIa once the general principles are established. This approach could be of great interest to DP1 and DP2 because most of these targets also participate in DNA damage signaling or regulation of apoptosis.

C. Approach: We use both fixed and live cell multiparameter fluorescence detection to assign the location of selected signaling proteins in a cell after exposure to novel chemical entities (246). In these studies, cells are treated for short times, namely <1 h, with lead compounds and the subcellular location of selected proteins are examined. The effect of a small targeted array library of 25-100 compounds, will be tested in each case, in collaboration with Pitt Combinatorial Chemistry Center, using the methods described in our recent work (141;248;249). We will examine the possible binding of the disruptive compound using recombinant Cdc25C or 14-3-3, and the SPR technique as described in Specific Aim 1. If an interaction is found, we will map the region on the protein by using deletion or point mutants. These experimental studies will be complemented by GNM-based computations for rapidly estimating the key residues that coordinate the global dynamics of the proteins. Moreover, the availability of targeted array libraries of 25-100 structurally similar compounds should allow us to model the ligand interactions with the target molecule quickly. We recently successfully completed a pilot project with Cdc25B and a small molecule inhibitor that has given us some experience with the overall approach we might use (250). Obviously, we are able to use the same approach for c-Src and Stat3 interactions (see Aim 1) as well as in cooperation with members of DP2. In the studies following Cdc25C movement, we are likely to identify compounds that do not directly interact with Cdc25C or 14-3-3; indeed these may be among the most interesting compounds.

Irrespective of the targeted signaling molecules, the computational modeling of the movement and interactions of these and other targets within the cell is of great interest. To simulate the simultaneous dynamics of multiple coupled molecular targets is a monumental task that goes beyond the scope of a developmental project and present capabilities. Nevertheless, the potential importance of such simulations is equally monumental and more than justifies a simplified proof-of-principle approach at this stage using Cdc25C and 14-3-3 as targets, and later ERK, c-Src and Stat3.

The first task will be to acquire the data for Cdc25C and 14-3-3 (and other signaling proteins) concentrations/activities with different concentrations of disruptor (or stimulus) and at different times of exposure. This will provide information on the stimulus response behavior for these targets, as well as the possible smooth (Michealian) or sigmoidal (ultrasensitive, bistable) behavior of the target response. Depending on how quickly we can obtain the chemical libraries and target information, we will begin constructing a model that includes a minimal number of spatial and chemical kinetic free parameters. Obviously, we will use the experimental data provided by SPR and multiparameter fluorescence as well as the results from combinatorial chemistry for building our models, although the data from isolated experiments with purified components may not reflect the effective concentrations and interaction rates in subcellular environment.

As described in § A.4., we will adopt a two-level approach: First, we will identify the positive or negative feedback modules, or the control motifs, for the signaling networks involving the investigated proteins and compare their dynamics predicted by stochastic simulations (MCell) and by mathematical models (see § A.4.4). Even simple signaling circuits of a few components can be bistable and can exhibit hysteresis (irreversible change) in the presence of strong positive feedback (117). Ferrell has also shown that a switch-like response results if a stimulus causes both a signaling molecule and its activator to translocate to the same subcellular compartment (251). The analysis performed by Ferrell for the simultaneous induced translocation of Cdc25c and Cdc2-cyclinB1 to the nucleus indeed provides a valuable guidance for the control motifs that are directly affected by Cdc25c translocation, which will be analyzed as described in § A.4.4. Likewise, the cell division control model and and corresponding phase-plane portraits and bifurcation analyses by Tyson and coworkers (see for example ref 122) will serve as excellent test modules for bridging between MC simulations and mathematical models, as described in see § A.4.4. Secondly, we will focus on the biochemical and biophysical characteristics of the elements (e.g. Cdc25c, Wee1, Cdc2-cylinB1, in different forms) of the same motif(s) to identify the minimal geometric and energetic features at the molecular level that need to be incorporated in the MC simulations of the control motifs (given computational time and memory limitations) to ensure that our models reproduce the experimentally observed (qualitative) behavior. See the approaches summarized in § A.4.5.

If even this much can be accomplished it will be a major achievement, but in reality the complexity of real cell structure and variability, together with biochemical network complexity, may complicate matters considerably. Regardless of outcome, however, the process of building such models for the first time will be invaluable to identification of required new modeling methods and multiscale design. Thus, this second aim of DP3 will be a major investment in modeling and simulation infrastructure that must be undertaken, in essence as a first step toward a real capability to “forecast cellular weather”. As the modeling and simulation infrastructure grows, the accuracy of our meteorological predictions will also grow, with enormous potential payoff.