Recovering Prebiotic Rossmann-like Folds Using AI-guided Sequence Design

The first small peptides and functional proteins were likely composed of only the ten prebiotic amino acids available on the early Earth: alanine, aspartic acid, glutamic acid, glycine, isoleucine, leucine, proline, serine, threonine, and valine. Among the proposed earliest protein domains is the ~90-residue Rossmann-like fold. Many modern Rossmann-fold proteins bind FeS clusters or nucleotides, suggesting that this fold may represent a transitional link between FeS-driven and nucleotide-dependent electron transfer required for biological energy production. This study aims to use AI deep learning algorithms in combination with experimental validation to reconstruct functional, prebiotic Rossmann-like folds that might have existed at the origin of cellular life. ProteinMPNN was used to generate candidate prebiotic sequences, followed by validation of Rossmann-like folds using AlphaFold3. Sequences with high probability of correct folds were then analyzed to select both consensus-based and diversity-based prebiotic sequences for experimental testing. Plasmids containing these prebiotic sequences were transformed into E. coli, and prebiotic Rossmann-like folds were expressed and characterized, including from both soluble and inclusion body fractions. We hypothesize that recovering prebiotic Rossmann-like folds will shed light on the evolutionary transition that allowed primitive peptides to give rise to complex functional proteins.