In addition to validating CRACLe using the ZDock Benchmark 4.0, CRACLe has been applied to two additional projects. The first is a collaboration with the Cance laboratory from Roswell Park Cancer Institute in Buffalo, New York to develop a small molecule inhibitor for the interaction between the Focal Adhesion Kinase (FAK) and Human Epidermal growth factor Receptor 2 (HER2). The second is a collaboration with the Asokan laboratory at the University of North Carolina at Chapel Hill to predict antigenic epitopes for the viral envelope protein gp120.
3.3.1 Predicting the FAK-HER2 Interaction
FAK is a tyrosine kinase that plays an important role in a number of cellular functions, including integrin-mediated signaling, cellular motility, and protection against apoptosis, and HER2 has been used as a marker to evaluate the aggressiveness of a particular cancer [98]. Several studies have shown that FAK plays an important role in upregulation of the HER2 signaling pathway [98], and the Cance laboratory has produced experimental evidence that FAK not only plays a role in the signaling pathway but actually binds HER2. Based on this knowledge, their laboratory is attempting to design a small molecule inhibitor to disrupt this interaction.
We used CRACLe to suggest possible binding sites on the surface of the FAK FERM domain. CRACLe predicted two binding sites that have been previously validated in the literature (Figure 3.9). The first site consists of residues Y180 and V196 on the FERM F2 domain. These residues form part of a hydrophobic pocket that binds and inhibits the FAK kinase domain [99]. The second predicted binding site consists of residues K218 and K222, both of which were found to be critical for an interaction with the proto-oncogene c-Met [100]. CRALCe also predicted two additional binding sites on the FERM domain that are un- dergoing experimental validation for binding activity with HER2. Our collaborators have designed small molecules inhibitors to bind at each of these predicted binding sites and are
Figure 3.9: CRACLe predicted two binding sites on the FAK FERM domain (PDB code 2AL6) that correspond to experimentally validated hot spots. (Red) Predicted binding site with residues Y180 and V196; this site corresponds to a hydrophobic pocket formed by residues Y180, M183, V196, and L197 that binds the FAK kinase domain. M183 is a part of the surface triplet, and L197 is not on the surface in the tessellation. (Purple) Predicted binding site with residues L281 and L222; both of these residues are experimentally validated hot spots for the interaction with c-Met.
currently testing to see the affect each small molecule has on the FAK-HER2 interaction. In- hibitors for the first predicted site have been found to reduce cell viability in cancer cell lines; unfortunately, the results suggest that this binding site does not interact with HER2. However, inhibitors for the second binding site do appear to be disrupting the FAK-HER2 interaction. We are unable to provide additional data at this time as the results are not yet published.
3.3.2 Predicting Antigenic Epitopes
In order to fight infection, B cells produce antibodies that are able to identify and neutral- ize foreign proteins called antigens. Antibodies bind these antigens at a unique, and usually conserved, part of the protein, referred to as antigenic epitopes ([101]). Prediction of antigenic epitopes could lead to the development of specialized antibody drugs and a better understand- ing of host-pathogen interactions [101, 102]. Ofttimes, antigenic epitopes are continuous, i.e., formed by a strand of sequentially adjacent peptides, but many known epitopes are discon-
tinuous. The former are much easier to predict, and many various methodologies have been developed for that purpose; the latter much less so [103, 101]. Based in part on the ZDock Unbound results for the Ab-Ag complexes, we hypothesize that CRACLe could be used to predict antigenic epitopes on pathogenic proteins. To this end, we decided to predict the binding sites for the well-known HIV-I envelope protein gp120.
We have compiled a dataset of 18 gp120 and gp160 precursor proteins from the Immune Epitope Database (IEDB) based on sequence similarity for known antigenic epitopes. Our initial results have not correlated well with the known antigenic epitopes; however, visual in- spection of the crystal structure shows that the binding sites predicted by CRACLe may be correct. For example, we analyzed the structure of gp120 co-crystallized with CD4 and an antibody (PDB code 2QAD) and found that the antigenic epitopes given in the IEDB do not occur where the antibody is bound (Figure 3.10A), but instead in an internal part of the pro- tein, at the interface with a gp120 dimer, and on anα-helix on the opposite side of the protein from where the antibody is bound. These differences may be explained by problems inter- preting the crystal structure or existence of multiple antibodies that bind in different locations. Interestingly, CRACLe predicted three binding sites, one at the interface with the antibody in the crystal structure, another at the gp120 dimer interface, and a third structurally adjacent to the antigenic epitope found on theα-helix. Further analysis is required before our predic- tions may be validated; however, these initial results suggest that CRACLe may be useful for prediction of new and validation of existing antigenic epitopes.