Thermodynamics and selection of the plasminogen activator inhibitor-1 latency transition

Plasminogen activator inhibitor-1 (PAI-1), like other serine protease inhibitors, exists in a functionally active metastable conformation. Unlike other serine protease inhibitors, PAI-1 undergoes a relatively rapid rearrangement to a lower energy latent conformation. We employ deep mutational scanning (DMS) to simultaneously probe the effects of nearly all single amino acid substitutions (92%) in PAI-1 on its latency transition as well as its ability to inhibit its canonical protease target, urokinase-type plasminogen activator (uPA). The DMS results are interpreted in the context of variant effect predictors (evolutionary model of variant effect), AlphaMissense, and cross-protein transfer model and protein stability predictors (EvoEF and FoldX), as well as the evolutionary conservation of the PAI-1 sequence space across extant vertebrate species. We demonstrate that while variant effect predictors can generally partition PAI-1 as functional or nonfunctional with respect to uPA inhibition, they perform poorly when attempting to discriminate the effects of PAI-1 variants on its latency transition. However, by employing protein stability predictors, we demonstrate that the PAI-1 latency transition is most likely driven by changes in the energy barrier to the latency transition. Finally, we show that PAI-1's ability to inhibit uPA, as well as its thermodynamic instability in its active conformation, are both under purifying selection that limits the sequence space available to PAI-1. These findings suggest that DMS of collateral fitness effects, including PAI-1's latency transition, may be better interpreted in contexts other than variant effect predictors, including protein thermodynamics and evolution.