Biology Reference
In-Depth Information
compromise
. Further, while the biophysical principles governing binding are generally
understood, and significant insight into the principles underlying enzymes
'
catalytic power
has been obtained,
33
there are still open questions about the relative importance of certain
phenomena for catalysis, such as the trade-off between active-site dynamics and
preorganization.
34
However, developing an algorithm to approach a certain CPD problem
necessitates understanding of the biophysical factors playing a role in the problem, and if
this understanding is incomplete, the resulting algorithms might be lacking critical features.
Thus, while impressive early results have been achieved in computational enzyme design,
the procedure is still less reliable than binding or thermostability design.
'
Another computational challenge unique to enzyme design is that to accurately model
chemical reactivity, a quantum-level treatment of the system is required, for example to
assess the effect of the active site electronic environment on the energy levels of the
substrate
s molecular orbitals. All CPD energy functions employ a classical physical
representation of the system however, and since billions of different sequences are usually
considered by the side-chain placement algorithm, using a more accurate quantum-level
model would be prohibitively slow. Thus, the energy function is essentially blind with
respect to the catalytic competence of the designed active site. The most-often-used
workaround for this problem is based on restraining the identity and allowed geometry for
a handful of the active site residues.
35
In this section, we will introduce methods used and
successful examples for each of the three cases.
'
Computational Redesign of Enzyme Specificity
The objective of specificity redesign usually is to take an existing enzyme and change it such
that it transforms a different substrate. Generally, the catalytic residues of the enzyme active
site (i.e. the residues that mediate the chemical steps of the reaction), and consequentially
the type of chemistry and mechanism that the enzyme performs, remain unchanged. Only
those active site residues that play a role in binding the substrate are modified by the design
algorithm. Thus, the new substrate needs to be similar to the enzyme
111
s natural substrate, in
that it must feature the same reactive moieties that are acted upon by the enzyme. The new
substrate is only allowed to differ in the nonreacting parts of the molecule.
'
As for most other CPD projects, a structural model, optimally a crystal structure, of the to-be-
designed system is required as the input. For specificity redesign, the optimal starting point is
a crystal structure of the enzyme of interest in complex with its native substrate, or with a
substrate or transition state analogue. Additionally, a basic understanding of the catalytic
mechanism and knowledge of the most important active site residues is required in order to
prevent the design algorithm from mutating away from these critical residues. The first step is
to generate a model of the new desired substrate. If this substrate features rotatable bonds, an
ensemble of possible conformations also needs to be generated. Usually, this ensemble has
restricted diversity in the moiety of the target substrate that resembles the wild-type substrate,
but full diversity in the differing regions. The target substrate ensemble is then superimposed
onto the wild-type substrate, such that the shared moiety is in the same region of the active
site and making identical contacts to the catalytic residues. Next, the side-chain placement
algorithm is used to design a sequence that accommodates the new substrate. In this step,
besides sampling the identity and conformational diversity of the active site residues, the
substrates conformational ensemble is also sampled. Usually, only the subset of residues that
contact the differing moiety of the target substrate is allowed to mutate. After running the
side-chain placement algorithm, the resulting structure is usually refined through gradient-
based minimization, and these two steps are iterated several times. If a stochastic side-chain
placement algorithm is used, the calculation is carried out several hundreds or thousands of
times, leading to a large number of models. The best scoring models are selected, and
