7
construction of the connecting linkers and facilitate rapid diversification of input functions. Notably, many intracellular signal
transducers are organized in a modular fashion that facilitates
rewiring the response functions of bacterial [47–49] and eukaryotic [50–55] signal transducers or the construction of genetically
encoded [56–61] and semisynthetic protein sensors [62–67].
Visual inspections of protein structures are, however, relatively
crude design strategies that are nonquantitative, rely on manual
assessment, and frequently need to be optimized empirically
through experimental screening. Ideally, the function of a synthetic protein switch can be engineered computationally in an
automated fashion based on quantitative parameters, which also
reflects on our fundamental understanding how protein sequence
relates to protein structure and function.
Toward this goal, a number of computational strategies have been
pursued to analyze and engineer structural and functional properties of a protein a priori by means of computational design
[68–70]. In its most elementary form, molecular dynamic simulations compute the behavior of an ensemble of molecules based on
the physical forces that every single atom is subject to. Such highresolution models are however computationally expensive, and in
practice take prolonged periods of time to model the structure or
the conformational dynamics of proteins. As a result, molecular
dynamics simulations are primarily restricted to analytical studies
and thus not suited to iterate through large numbers of protein
mutants as necessitated in rational protein design.
Instead, increasing grades of abstraction and simplification are
introduced aiming to limit the conformational search space and
accelerate computation times [68–70]. This usually requires identifying, approximating, and weighing the key parameters that
underlie a structural, biophysical, or functional property. Specific
simplifications include restricting the dihedral angles of the polypeptide backbone and amino acid sidechains to the most frequently
occurring rotamers (in the same way structural biologists match
the tertiary structure of a protein to its electron density map) or
approximating secondary structure propensities, solvation terms,
electrostatic energies, and hydrogen bond potentials. This is
increasingly complemented by bioinformatic approaches mining
protein structures for functional motifs that can be grafted onto a
desired binding or enzyme catalyzed reaction.
In this way, a number of new protein structures and functions
could be computationally engineered including new folds [71, 72],
new ligand and substrate specificities [73, 74], as well as new catalytic functions [75, 76]. In contrast, predictably engineering the
conformational transitions that underlie the switch-like behavior of
synthetic protein switches has proven more difficult and primarily
relied on redesigning individual properties. In one recent example,
2.2 Design
by Molecular Modeling
Engineering Synthetic Protein Switches
construction of the connecting linkers and facilitate rapid diversification of input functions. Notably, many intracellular signal
transducers are organized in a modular fashion that facilitates
rewiring the response functions of bacterial [47–49] and eukaryotic [50–55] signal transducers or the construction of genetically
encoded [56–61] and semisynthetic protein sensors [62–67].
Visual inspections of protein structures are, however, relatively
crude design strategies that are nonquantitative, rely on manual
assessment, and frequently need to be optimized empirically
through experimental screening. Ideally, the function of a synthetic protein switch can be engineered computationally in an
automated fashion based on quantitative parameters, which also
reflects on our fundamental understanding how protein sequence
relates to protein structure and function.
Toward this goal, a number of computational strategies have been
pursued to analyze and engineer structural and functional properties of a protein a priori by means of computational design
[68–70]. In its most elementary form, molecular dynamic simulations compute the behavior of an ensemble of molecules based on
the physical forces that every single atom is subject to. Such highresolution models are however computationally expensive, and in
practice take prolonged periods of time to model the structure or
the conformational dynamics of proteins. As a result, molecular
dynamics simulations are primarily restricted to analytical studies
and thus not suited to iterate through large numbers of protein
mutants as necessitated in rational protein design.
Instead, increasing grades of abstraction and simplification are
introduced aiming to limit the conformational search space and
accelerate computation times [68–70]. This usually requires identifying, approximating, and weighing the key parameters that
underlie a structural, biophysical, or functional property. Specific
simplifications include restricting the dihedral angles of the polypeptide backbone and amino acid sidechains to the most frequently
occurring rotamers (in the same way structural biologists match
the tertiary structure of a protein to its electron density map) or
approximating secondary structure propensities, solvation terms,
electrostatic energies, and hydrogen bond potentials. This is
increasingly complemented by bioinformatic approaches mining
protein structures for functional motifs that can be grafted onto a
desired binding or enzyme catalyzed reaction.
In this way, a number of new protein structures and functions
could be computationally engineered including new folds [71, 72],
new ligand and substrate specificities [73, 74], as well as new catalytic functions [75, 76]. In contrast, predictably engineering the
conformational transitions that underlie the switch-like behavior of
synthetic protein switches has proven more difficult and primarily
relied on redesigning individual properties. In one recent example,
2.2 Design
by Molecular Modeling
Engineering Synthetic Protein Switches