Tobacco etch virus protease (TEVp) is a prominent and widely used biotechnological
tool for which no reversible protein inhibitors are currently known. Developing a
specific inhibitor would broaden its applications in synthetic biology and
bionanotechnology, notably in the design of molecular mechanosensors. In this master's
thesis, we utilized state-of-the-art deep learning pipelines (RFdiffusion, ProteinMPNN,
and AlphaFold2) to computationally design de novo protein binders that act as TEVp
inhibitors. The design strategy was based on structural data of the interaction between
TEVp and its own C-terminus, which naturally binds to the active site.
The viability of the computationally designed proteins was experimentally evaluated
through their expression in E. coli, purification via affinity and size-exclusion
chromatography, and fluorimetric enzyme kinetics measurements. The results
demonstrated that inhibitors mimicking the C-terminal binding mechanism of the
enzyme were the most effective at reducing proteolytic activity. Furthermore, SPR
analysis revealed low-nanomolar dissociation constants (K D) for the top candidates,
with A8 yielding a KD of 5.5 nM and A12 exhibiting the highest binding affinity with a
KD of 1.75 nM. By developing fusion constructs intended for mechanical force
transduction in biosensors, we observed that introducing specific linkers to prevent
unwanted proteolysis slightly reduced the binding affinity to the active site;
nevertheless, the constructs retained substantial inhibitory activity. These findings
validate the efficacy of computational de novo design in developing functional enzyme
inhibitors.
|
