Catching a common fold

As the first proteins were sequenced in the 1950s, it was evident that they belonged to families. The determination of protein three-dimensional structures during the late 1960s and early 1970s (e.g., insulins, globins, and serine proteinases) confirmed that related proteins from different species adopt similar tertiary structures characteristic of each family. The sequence variations within a family reflected the restraints of the tertiary structures: apart from the catalytic or binding residues, invariant amino acids were most often in the protein core, inaccessible to solvent and with a key role in the protein architecture. The fascination with families of proteins was deepened with the realization that many proteins, with quite unrelated sequences, could adopt a common fold. Rossmann, Matthews, Branden, Richardson, and many others recognized similarities between the tertiary structures or domains that occur in many quite different proteins (Richardson, 1981); these included ap-nucleotide binding motifs (Rossmann fold), &-barrels (TIM barrel), &jelly rolls, four ahelix bundles, and immunoglobulin domains (0-Ig fold). These protein topologies underlined the fact that tertiary structures could be considered as simple combinations of secondary structural elements packed together in a limited number of ways: apa/3aO, mas, P@PP, and so on. It seemed that protein structures could be predicted from sequences by combinatorial assembly of the basic elements of secondary structure, following various rules about handedness of the loops connecting them and the avoidance of strands that were “cross-overs.” However, such combinatorial approaches to the protein folding problem depend on correct assignment of a-helices, 0strands, and coils, and this remains a formidable challenge.