Biology Reference
In-Depth Information
optimize the energy function for a single state, and extending them to consider multiple
states is not trivial. Therefore, algorithms specifically developed for this purpose need to be
employed.
An elegant and very general approach to this problem was presented by Leaver-Fay et al.
18
In this approach, the input to the calculation is a set of backbone models, together with an
algebraic rule of how to sum the scores (as determined by the energy function) of a certain
sequence on each backbone model into one value. During the calculation, the sequence
search is done by a genetic algorithm, and the value that is being optimized is the user-
specified sum over the scores of all considered states instead of the score of a single state.
When calculating the sum, absolute as well as relative scores of states can be considered in
positive (for the unwanted states) or negative (for the desired states) fashion. In their work,
the authors used this general framework to design sequences that in silico have favorable
interaction scores for a set of desired states while having unfavorable scores for another set
of competing, unwanted states. As this approach is still very recent, no experimental data on
sequences designed with it exist yet, but it does represent the most general and
comprehensive theoretical approach to the problem.
Several illustrative demonstrations of successful specificity redesign have been reported in
recent years, some of them with direct applications in synthetic biology contexts. In a
textbook example of two-sided design, Kapp et al.
19
designed an orthogonal GTPase/GEF
pair, which could be used as a valuable tool to study cell signaling, as well as serve as a
component in a synthetic signaling pathway. Starting from crystal structures of the GTPase
Cdc42 and its activator GEF, intersectin, the authors first identified positions in Cdc42 at
which mutations would interfere with intersectin binding without disrupting any of the
known interfaces with other binding partners or the active site. After identifying one
position (Phe-56), a cognate position on intersectin was identified where a salt-bridge could
be introduced if Phe-56 was concurrently mutated to Arg. The mutated GEF stimulated
nucleotide exchange in the mutated GTPase, but not in the wild-type, while the mutated
GTPase could be activated by the mutated GEF but not the wild-type, demonstrating the
orthogonality of the new pair. The new pair retained (albeit lower than wild-type) signaling
activity in vivo. Grigoryan et al.
20
introduced a computational framework that explicitly
considers competing states. In this study, the authors set out to design proteins that interact
with individual members of a class of transcription factors known as bZIPs. This class
comprises about 20 families, which share extensive structural and sequence similarity,
making the design of inhibitors specific for only one subset very challenging. To address this
problem, the authors developed an algorithm that first designs a sequence with highest
possible affinity for the target, and then in a second step modifies the found sequence to
increase the gap between the target state and the nearest competing state. Forty-six designs
were characterized, 10 of which interacted more strongly with their intended target than
with any competitor. However, to make the used algorithm computationally tractable, a
scoring function specifically developed for this class of proteins had to be used, making this
approach not easily extensible to other problems. In another study, Yosef et al.
21
redesigned
the calcium-dependent second-messenger protein calmodulin towards preferably binding
only one of its two major interaction partners. One of the designs, containing six mutations,
had significantly reduced affinity towards the undesired partner, but retained affinity for the
desired partner, resulting in a 900-fold specificity switch. Several more examples of
successful computational redesign of protein
106
protein interactions have been described in a
recent review.
22
Computational Design of Novel Protein Interactions
Complementary to redesigning existing protein interfaces, CPD can be used for the de novo
design of protein interactions. Generally, the objective in a de novo interface design problem
