Using the new version of a Protein Structural Database, BIPED, β-tums have been extracted from 58 non-identical proteins (resolution ≤ 2 Å) using the standard criteria that the distance between Cαiand Cαi+3is < 7 Å and that the central residues are not helical. It has been shown that 42% of these do not fit into the eight conventional turn types (I, I′, II, II′, VIa, VIb and VIII) as defined by φ, ψ limits for residues i+1 and i+2. Most of the unclassified turns are of the standard turn types, lying just outside the specified limits. Eleven per cent of the turns are not related to the standard types, and have completely different φ, ψ combinations. Therefore, the conformational space available to a trans tetrapeptide has been explored to find all two-residue φ, ψ combinations which satisfy the criteria for β-turn formation. The conformations generated by the search are the conformations observed in the data. On the basis of these observations a new nomenclature is suggested for β-turns based on a shorthand descriptor of populated regions of the φ, ψ plot. For example, it is proposed that a type I turn should now be described as an αα turn, the descriptor indicating the φ, ψ regions of the i+1 and i+2 positions of the turn. The use of descriptors, which convey information about the turn type conformation, should aid protein structural workers in turn classification and visualization. The sequence preferences for the αβ turn have been elucidated: Pro, Asp and Ser at i+1; Asn and His at i+2; Pro at i+3. These preferences have been explained in terms of specific interactions involving the side-chains. The β-turn prediction program, BTURN 1.0, has been further developed by alterations to the calculation of parameters, the removal of incompatible multiple turn predictions and the inclusion of the distorted types αα and βpγdata in the respective sets of parameters, to yield BTURN 2.0. A variation of this β-turn prediction program, called GORBTURN 1.0, has been developed, which uses the directional parameters of Gibrat et al. in combination with equivalent parameters produced from work by Garrett et al., to eliminate potential helix and strand-forming residues from the β-turn prediction. This has led to improved turn prediction in protein sequences, although it is not suitable for predictions on short peptide sequences. BTURN 2.0 is the program of choice for these sequences. The incorporation of αβ parameters into GORBTURN 1.0 led to a small decrease in predictive accuracy, and therefore these parameters are not yet refined enough to give improved prediction.
Bibliographical noteFunding Information:
We thank Steve Gardner, David Smith and the Protein Engineering Club for the use and help with BIPED; Malcolm McGregor for use of a program to produce number density 0,i/' plots; Drs Suhail Islam and Mansoor Saqi for help with the torsional driver programs; Mark Swindells for the use of a program to calculate correlation coefficients; and Jean-Francois Gibrat and Jean Gamier for discussions on the GOR program. C.W. was supported by a SERC studentship.
- Protein secondary structure prediction