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.