The four-a-helix bundle, a common structural motif in globular proteins, provides an excellent forum for the examination of predictive constraints for protein backbone topology. An exhaustive examination of the Brookhaven Crystallographic Protein Data Bank and other literature sources has lead to the discovery of 20 putative four-a-helix bundles. Application of an analytical method that examines the difference between solvent-accessible surface areas in packed and partially unpacked bundles reduced the number of structures to 16. Angular requirements further reduced the list of bundles to 13. In 12 of these bundles, all pairs of neighboring helices were oriented in an anti-parallel fashion. This distribution is in accordance with structure types expected if the helix macro dipole effect makes a substantial contribution to the stability of the native structure. The characterizations and classifications made in this study prompt a reevaluation of constraints used in structure prediction efforts.Specification of the code that translates primary to tertiary structure remains unresolved. It has proven difficult to predict which features in an amino acid sequence will provide the basis for the three-dimensional structure or function of a given protein. However, proteins with only 10% identity at comparable positions in a polypeptide can have notably similar structures (1). Indeed, an individual tertiary structure will often fall into one of a limited number of structural classes (2). Tertiary-structure prediction systems incorporating combinatorial methods (3-5) and template methods (6, 7) target known structural classes in an attempt to reduce the number of structures created and examined. Therefore, to effectively predict tertiary structures, we must determine as many constraints describing the individual structural classes as possible.Among the structural class containing those proteins constructed predominantly from a-helical structures, a highly recurrent-motif is the collection of (anti)parallel a-helices known as the four-a-helix bundle. Four-a-helix bundles are found in proteins covering a wide range of structure and function, but they display some common characteristics. Weber and Salemme (8) were among the first to examine and characterize four-a-helix bundles as a class of supersecondary structure. Initially, their work suggested a common, right-handed, connective topology for all four-a-helix bundles. This characterization, derived from a limited data base of bundle topologies, proved too restrictive to be correct. The "handedness" constraint obscured further attempts to characterize the topology of previously known and newly discovered structures (9, 10). Incorporation of all the presently known four-a-helix bundle structures requires a different categorization scheme.The purpose of this study is to describe and present a thorough examination of the literature for four-a-helix bundles using generalized determination criteria. These criteria suggest a set of bundle structures beyond those initially...