[Thesis]. Manchester, UK: The University of Manchester; 2018.
Protein - protein interactions govern every aspect of the cellular life cycle. Despite
the pivotal role of interprotein association, many of its aspects remain poorly understood.
This pertains particularly to the specificity determinants in interactions between
large families of proteins and in intrafamily interactions. To elucidate the origins
of affinity and specificity in paralogous inter- and intrafamily interactions, a series
of in silico techniques of increasing theoretical sophistication and computational
cost were employed on several datasets from key physiological pathways, under the
initial assumption that interactions are mediated through a common interface on a
conserved steric scaffold.
A large-scale bioinformatics study on all combinations of potential interactors within
the examined systems was carried out first, performing side chain replacement on X-ray-
and NMR-derived templates to produce up to thousands of models of the various binary
interactions within the examined systems. Simultaneously, polar and nonpolar areas,
buried upon complexation, and the energy of electrostatic interaction between the
binding partners were computed. Comparison of surfaces and energies between interacting
and non-interacting pairs, identified from literature, reveals that all three parameters
are significantly different between interactors and non-interactors, with electrostatics
being most discriminatory of the three interfacial descriptors.
Despite the statistical significance of the separation between binders and non-binders,
considerable overlap remains, making any predictions solely based on buried surface
and charge interactions unreliable. To probe deeper into the binding process, extensive
molecular mechanics - Poisson-Boltzmann surface area calculations were then performed
on a medium-sized set of 60 protein - peptide complexes from the Bcl-2-family of proteins
- key regulators of the intrinsic apoptotic pathway. Per-residue decomposition of
the enthalpy of interaction between the different protein - peptide pairs provides
much finer detail on the binding process than the large-scale surface and charge calculations
previously performed. This allowed pinpointing where affinity and specificity within
the system originate, identification of key interactions, determination of how affinity
is dependent on peptide properties, and provided a quantitative estimate of the energetics
of binding. Crucially, this work demonstrates that the proteins' per-residue energies
can be viewed as an energy fingerprint.
Finally, this point was further developed by performing free energy calculations
at a higher level of theory - thermodynamic integration - on eight large, drug and
drug-like compounds bound to the Bcl-xL and Bcl-2 proteins. Comparison of the information
content provided by energetic fingerprinting with a traditional two-dimensional quantitative
structure-activity relationship study demonstrates the added value of free energy
calculations. Crucially, this method affords a more comprehensive description of the
binding process and every individual protein - ligand/peptide/protein complex, and
extends the framework of four-dimensional molecular dynamics - quantitative structure-activity
relationships (4D-MD/QSAR). Finally, directions for future work aiming to derive and
validate hyperpredictive 4D-MD/QSAR models incorporating ligand- and receptor-based
descriptors are set out.