David Rideout, Associate Research Scientist at UC San Diego, in collaboration with Trey Ideker. Previous work in the Ideker Lab developed an approach called CliXO to automatically construct a gene ontology from pairwise association data. The core idea is to consider a subgraph of the complete graph on all genes, which includes only edges whose pairwise association score is above a given threshold. The maximal cliques of this subgraph form the terms of the computed ontology. As the threshold is decreased, one can form a complete gene ontology, whose terms are the maximal cliques arising from different values of the threshold. One of the challenges though with CliXO is the computation of maximal cliques in an arbitrary, large graph. Because the graph is an arbitrary subgraph of the complete graph, the best algorithms are exponential in the number of nodes. This makes CliXO challenging to use in practice, due to its extremely long runtime.

This pilot project sought to implement an improved version of this approach, which makes use of the fact that, if we know the set of all cliques in a graph, then it is much easier to compute the new cliques which form due to the addition of a single edge. This approach has the added benefit that it considers every possible threshold value, to that the output is not subject to an arbitrary schedule for increasing the threshold, as is done in CliXO. The pilot project team has written three algorithms implementing the above scheme and tested them on the DNA Repair subnetwork of the gene ontology (GO), which contains 504 genes and 126,756 gene pairs, using the Resnik semantic similarity of GO for the edge weights. Two of these ended up consuming an enormous amount of memory but the third, most efficient, algorithm runs with an extremely small memory footprint and can process the DNA repair subnetwork in 47 seconds. His group is now starting to test this algorithm on pairwise gene associativity scores which arise from experimental data, the method is expected to tell us more about the interrelationships among the genes than is currently coded into GO.

Natalia Jura, Assistant Professor at UCSF, in collaboration with Nevan Krogan. Recent studies in the Krogan and Grandis Labs led to the discovery that a subset of PI3KCA mutant variants in head and neck cancer cells preferentially associate with the HER3 receptor. Moreover, these PI3KCA mutants are dependent on HER3 binding to elicit its oncogenic signaling. These exciting findings point to a unique mechanism by which HER3 activates PI3KCA, whose understanding will help develop personalized therapies for the affected patients. At present, the nature of this selectivity is unknown due to the lack of direct insights into the mechanism of PI3KCA/HER3 binding. This pilot project aimed to address this gap through the biophysical characterization of the PI3KCA/HER3 interaction and its modulation by the cancer mutations, to provide a platform for the development of specific inhibitors that disrupt the signaling junction between HER3 and PI3K in disease. The pilot project team developed a baculovirus-based insect cell expression system for purification of the HIS6-tagged intracellular HER3 receptor fragment (residues 665-1342), containing a full C-terminal tail with all 6 YxxM sites (HER3-ICD), and a TEV cleavage site allowing for removal of the HIS6-tag. By reconstituting HER3-ICD on Ni2 -lipid containing vesicles in the presence of the purified EGFR kinase domain, HER3 tail becomes robustly phosphorylated. In collaboration with Dr. Bradley Webb’s Lab (West Virginia U.), the capability was developed to purify the PI3KCA holoenzyme composing of the full-length p110a domain and the p85DN domain, which contains all relevant subdomains (nSH2, iSH2 and cSH2) that form direct contact with the p110 domain. The PI3K construct (PI3KDN) was also shown to be enzymatically active. This preliminary work demonstrates the ability to obtain functional HER3 and PI3KCA constructs for further studies.