[9, 21]. Implicit solvent models have been actively developed for
many years, both for medium and high-throughput applications.
They usually involve a continuum electrostatic component, such as
a Poisson–Boltzmann (PB) or Generalized Born (GB) energy term.
For high-throughput CPD, additional approximations are usually
necessary, outlined below.
1.2 PDZ–Peptide
Issues
PDZ–peptide binding presents specific difficulties. First, the
unbound peptide is quite flexible, and it is challenging to explore
its motions and quantify the effect of mutations on its flexibility and
entropy. Second, the binding interface is large and the affinity arises
from many small contributions, which should be accurately captured. Third, with such a large interface, many residues undergo
extensive burial upon binding, changing from a solvent-rich to a
solvent-poor environment. This change often leads to electronic
polarization, which is still a difficulty for molecular mechanics
models. Fourth, peptide phosphorylation regulates binding in
some cases, and phosphate–protein binding is also challenging to
model. Fifth, the use of ncAAs to enhance binding or peptide
stability means they must be part of the molecular mechanics
model. This often means a specific extension of the model is
needed.
1.3 Chapter
Overview
We first outline our high-throughput CPD approach. We describe
only briefly the technical details, which are available in published
articles [16, 22]. We describe a protocol that allows us to select
peptide variants based on their relative binding free energies, which
is of great interest. The approach is based on an adaptive Monte
Carlo method [22]. We present illustrative results for the Tiam1
PDZ domain. Next, we describe our medium-throughput
approach [23], which is readily applied to dozens of peptide or
protein variants (or a few hundred with more resources; roughly
one variant per day and per GPU card). The complex is simulated
with MD and explicit solvent. Then the binding free energy is
computed with a free energy function that combines a Poisson–
Boltzmann description of solutes and solvent, along with additional
nonpolar SA and vdW free energy terms. We include selected results
for the Tiam1 PDZ domain binding to a collection of peptides
[23], some of which were taken from CPD predictions. Figure 1
shows Tiam1 bound to the Syndecan-1 peptide (Sdc1), which
corresponds to the C-terminus of its natural target protein.
2 High-Throughput Design of PDZ–Peptide Binding
Many successful CPD examples have been reported in recent years
[3, 24–30]. Many were obtained with energy functions that
included knowledge-based terms, such as those in the Rosetta
energy function. Proteus, on the other hand, relies on “physicsComputational Design of Binding
239
many years, both for medium and high-throughput applications.
They usually involve a continuum electrostatic component, such as
a Poisson–Boltzmann (PB) or Generalized Born (GB) energy term.
For high-throughput CPD, additional approximations are usually
necessary, outlined below.
1.2 PDZ–Peptide
Issues
PDZ–peptide binding presents specific difficulties. First, the
unbound peptide is quite flexible, and it is challenging to explore
its motions and quantify the effect of mutations on its flexibility and
entropy. Second, the binding interface is large and the affinity arises
from many small contributions, which should be accurately captured. Third, with such a large interface, many residues undergo
extensive burial upon binding, changing from a solvent-rich to a
solvent-poor environment. This change often leads to electronic
polarization, which is still a difficulty for molecular mechanics
models. Fourth, peptide phosphorylation regulates binding in
some cases, and phosphate–protein binding is also challenging to
model. Fifth, the use of ncAAs to enhance binding or peptide
stability means they must be part of the molecular mechanics
model. This often means a specific extension of the model is
needed.
1.3 Chapter
Overview
We first outline our high-throughput CPD approach. We describe
only briefly the technical details, which are available in published
articles [16, 22]. We describe a protocol that allows us to select
peptide variants based on their relative binding free energies, which
is of great interest. The approach is based on an adaptive Monte
Carlo method [22]. We present illustrative results for the Tiam1
PDZ domain. Next, we describe our medium-throughput
approach [23], which is readily applied to dozens of peptide or
protein variants (or a few hundred with more resources; roughly
one variant per day and per GPU card). The complex is simulated
with MD and explicit solvent. Then the binding free energy is
computed with a free energy function that combines a Poisson–
Boltzmann description of solutes and solvent, along with additional
nonpolar SA and vdW free energy terms. We include selected results
for the Tiam1 PDZ domain binding to a collection of peptides
[23], some of which were taken from CPD predictions. Figure 1
shows Tiam1 bound to the Syndecan-1 peptide (Sdc1), which
corresponds to the C-terminus of its natural target protein.
2 High-Throughput Design of PDZ–Peptide Binding
Many successful CPD examples have been reported in recent years
[3, 24–30]. Many were obtained with energy functions that
included knowledge-based terms, such as those in the Rosetta
energy function. Proteus, on the other hand, relies on “physicsComputational Design of Binding
239