Neurodegenerative disorders are frequently associated with beta-sheet-rich amyloid deposits. Amyloid-forming proteins can aggregate under different structural conformations known as strains, which can exhibit a prion-like behavior and distinct pathophenotypes. Precise molecular determinants defining strain specificity and cross-strain interactions (cross-seeding) are currently unknown. The HET-s prion protein from the fungus Podospora anserina represents a model system to study the fundamental properties of prion amyloids. Here, we report the amyloid prion structure of HELLF, a distant homolog of the model prion HET-s. We find that these two amyloids, sharing only 17% sequence identity, have nearly identical beta-solenoid folds but lack cross-seeding ability in vivo, indicating that prion specificity can differ in extremely similar amyloid folds. We engineer the HELLF sequence to explore the limits of the sequence-to-fold conservation and to pinpoint determinants of cross-seeding and prion specificity. We find that amyloid fold conservation occurs even at an exceedingly low level of identity to HET-s (5%). Next, we derive a HELLF-based sequence, termed HEC, able to breach the cross-seeding barrier in vivo between HELLF and HET-s, unveiling determinants controlling cross-seeding at residue level. These findings show that virtually identical amyloid backbone structures might not be sufficient for cross-seeding and that critical side-chain positions could determine the seeding specificity of an amyloid fold. Our work redefines the conceptual boundaries of prion strain and sheds light on key molecular features concerning an important class of pathogenic agents.< Réduire