Computer simulations to engineer PDZ-peptide recognition

Informations générales
Détails de la thèse/HDR
Jean-Philip PIQUEMAL
Directeur (pour les thèses)
Résumé en anglais
Protein-protein interactions (PPIs) mediate complex signaling networks in cells. Engineering them can provide understanding
and is a recognized strategy for drug design. PDZ domains are among the most widespread domains mediating PPIs.
They recognize the 4-10 C-terminal amino acids of their target proteins. They can also bind the corresponding peptides, which
can thus serve as inhibitors or model systems.

We have developed and tested two computational approaches to characterize and engineer PDZ:peptide recognition.
We applied them to the PDZ domain of the Tiam1 protein, a Rac GTP exchange factor involved in neuronal protrusion
and axon guidance. Its natural target protein is the Syndecan1 protein. Experimental affinities are available for the C-terminal
Syndecan1 peptide, denoted Sdc1, and several mutants; this data was used for testing and benchmarking.

We first developed a novel high throughput strategy for protein and peptide design. A Monte Carlo simulation was done
where protein and peptide side chains could change conformations and selected positions could mutate. Solvent was modeled implicitly.
The energy landscape was adaptively flattened, following the Wang-Landau method, so that a very diverse set of sequences
was sampled. By performing separate simulations for the PDZ:peptide complex and the unbound peptide, we could recover the
relative binding free energies of around 100,000 peptide variants using just one hour of CPU time on a desktop machine.
The computed affinities were consistent with (sparse) available experimental data.

Next, we developed a much more detailed and accurate simulation model. The solute and solvent were described in full atomic detail,
using molecular dynamics simulations and a state-of-the-art force field that explicitly accounts for electronic polarization: the Charmm Drude force field.
Polarizability is expected to be important for PDZ:peptide binding, since residues that form the binding interface are transferred
from a solvent-rich to a solvent-poor environment upon binding. We performed alchemical free energy simulations to compare the binding
free energies of four peptide variants, which all differ in one or two ionic side chains. The calculations gave good agreement with experiment.
In contrast, the Charmm and Amber “additive” force fields, which treat electronic polarization implicitly, gave poorer agreement.
These calculations represent the first example of relative protein:ligand binding free energies computed with alchemical free energy simulations
and a polarizable force field.

Finally, to allow future modeling of phosphorylated peptide variants, we extended the Drude force field to include both methyl phosphate
and phosphotyrosine. The force field gave excellent agreement with experimental phosphate--magnesium binding free energies.