Advancing Luciferase Activity and Stability beyond Directed Evolution and Rational Design through Expert Guided Deep Learning.

Engineered luciferases have transformed biological imaging and sensing, yet optimizing NanoLuc luciferase (NLuc) remains challenging due to the inherent stability-activity trade-off and its limited sequence homology with characterized proteins. We report a hybrid approach that synergistically integrates computational deep learning with structure-guided rational design to develop enhanced NLuc variants that improve thermostability and thereby activity at elevated temperatures. By systematically analyzing libraries of engineered variants, we established that modifications to termini and loops distal from the catalytic center, combined with preservation of allosterically coupled networks, effectively enhance thermal resilience while maintaining enzymatic function. Our optimized variants - notably B.07 and B.09 - exhibit substantial thermostability enhancements (increases of 4.2 °C and 5.2 °C at 50 % solubility), leading to increased activity at elevated temperatures (320 % and 370 % of wild-type at 55 °C). These variants maintain NLuc's pH tolerance and retain improved activity with the alternative substrate coelenterazine. Molecular dynamics simulations and protein folding studies elucidate how these mutations favorably modulate conformational landscapes without perturbing substrate binding architecture. Beyond providing superior tools for bioluminescence applications, our integrated methodology establishes a broadly applicable framework for engineering enzymes where traditional homology-based approaches fail, and stability-activity constraints present formidable barriers to improvement.