Computational prediction and design for creating iteratively larger heterospecific coiled coil sets

A major biochemical goal is the ability to mimic nature in engineering highly specific protein-protein interactions. We previously devised a computational interactome screen to identify eight peptides that form four heterospecific dimers despite 32 potential off-targets. To expand the speed and utility of our approach and the PPI toolkit, we have developed new software to derive much larger heterospecific sets (≥ 24 peptides) while directing against antiparallel off-targets. It works by predicting Tm values for every dimer based on core, electrostatic, and helical propensity components. These guide interaction specificity, allowing heterospecific coiled coil sets to be incrementally assembled. Prediction accuracy is experimentally validated using circular dichroism and size exclusion chromatography. Thermal denaturation data from a 22 coiled coil (CC) training-set was used to improve software prediction accuracy, and verified using a 136 CC test-set consisting of 8 heterospecidic dimers and 128 off-targets. The resulting software, qCIPA, individually now weighs core a-a’ (II/NN/NI) and electrostatic g-e’+1 (EE/EK/KK) components. The expanded dataset has resulted in emerging sequence context rules for otherwise energetically equivalent CCs; for example, introducing intra-helical electrostatic charge-blocks generated increased stability for designed CCs while concomitantly decreasing the stability of off-target CCs. Coupled with increased prediction accuracy and speed, the approach can be applied to a wide range of downstream chemical and synthetic biology applications, in addition to more generally to impose specificity in structurally unrelated PPIs.