december 11 2025, 14:00, salle Herpin, in French

Statistical engineering of specificity of S1A serine proteases

Enzymatic specificity conversion, particularly within the serine protease S1A family, represents a major challenge in protein engineering. Despite decades of research, rational reprogramming of catalytic function remains difficult to achieve, often limited by an incomplete understanding of allosteric effects and evolutionary constraints. This thesis aims to decipher the sequence–function relationship in these enzymes by combining two complementary approaches : (1) high-throughput experimental screening using microfluidic droplet sorting (FADS) and (2) directed mutagenesis guided by computational analysis of evolutionary data.

First, we build upon the work of Valentin Senlis on the deep mutational scan (DMS) of rat trypsin [186]. The analysis, supported by a biophysical model of the selection experiment, confirmed a bimodal fitness landscape, where the majority of mutations are either neutral or highly deleterious.

Next, we interpret Senlis’ results using a two-dimensional selection method that allowed identification of mutations, including distal ones, modulating the arginine/lysine specificity ratio. We provide quantitative measurements of this ratio for the selected mutants.

Next, we explored the ability of generative models (DCA : Direct Coupling Analysis), trained solely on sequence data, to design functional synthetic proteases. This proof had already been demonstrated experimentally for enzymes called chorismate mutases; we show here that it also extends to enzymes with far more complex catalysis, such as serine proteases. Finally, in collaboration with Marion Chauveau, a rational design approach based on DCA (COMBINE) was applied to specificity conversion [29]. A combination of only three mutations in the binding pocket was sufficient to achieve a partial but successful conversion of trypsin into chymotrypsin. In contrast, the conversion of chymotrypsin into trypsin remained unsuccessful, confirming an asymmetry already observed in the literature.

In conclusion, this work highlights the complexity of evolutionary trajectories controlling enzymatic specificity. Our results show that detectable specificity conversions can be achieved by modifying only the active site, without requiring distal mutations. However, to reach activity comparable to the natural sequence acquisition of permissive distal mutations appears necessary. We hypothesize that the development of robust predictive models will require the acquisition of massively annotated functional datasets, despite the experimental challenges this entails, in order to capture the subtle constraints governing the sequence–function relationship in these enzymes.