Design of a Labile RNase A Using Protein Language Models

Protein language models (PLMs) have emerged as powerful
tools for
the generation of functional protein sequences. However, most efforts
focus on enhancing protein stability for industrial applications,
whereas there is untapped potential in designing proteins with reduced
stability, which can be advantageous in specific contexts. For instance,
in molecular biology workflows, enzymatic reagents such as DNase I
are commonly used and subsequently inactivated to prevent residual
activity from compromising downstream processes. Proteins that are
easily inactivated offer a streamlined alternative to physical removal,
simplifying protocols and reducing experimental complexity. In this
study, we leverage RNase A, a paradigmatically stable enzyme, as a
model for exploring the engineering of functional, yet less stable
proteins. By sampling sequences from the PLM embedding space near
the wild-type RNase A sequence, we engineered a variant, TempRNase,
with reduced stability while retaining its RNA degradation activity.
Using a fluorometric RNA degradation assay under varying conditions
of heat and reducing treatment, we benchmark TempRNase against its
wild-type counterparts and show that moderate heat and reducing treatment,
with marginal effect on the wild-type, permanently inactivates TempRNase.
Sequence and structural analyses of TempRNase reveal critical insights
into the stability modulation and protein dynamics. Our findings establish
the concept of engineering “worst of the best” enzymes
that are functional but less stable. Furthermore, we highlight RNase
A as a powerful model system for tuning protein stability with a
quantitative assay.