How an amino acid sequence encodes the information necessary for a protein to adopt a unique tertiary structure remains unresolved. We are addressing this problem by designing "from scratch" protein molecules that will adopt predetermined three-dimensional structures. Based on this strategy, two identical four-stranded 3-sheets were designed to dimerize and form a 1sandwich protein, called betadoublet. A synthetic gene encoding half the 13-sandwich protein was expressed in Escherichia colt, and the protein was purified to homogeneity. Biophysical characterization of betadoublet in aqueous solution demonstrated that the disulfide formed between the two sheets and that the dimer was a compact unaggregated globular protein, consing predominantly of 1-sheet and stable to thermal denaturation. It has some backbone amide protons whose exchange is slow enough to be measured by NMR but binds more of the dye 1-anilinonaphthalene-8-sulfonate than a well-folded protein.Our understanding of the relationship between a protein's tertiary structure and its amino acid sequence directly effects our comprehension of protein folding and the connection between protein structure and function and our ability to design macromolecules with new functions. One approach to expanding this knowledge involves de novo design of protein molecules of specific three-dimensional structure (1-10). Recent advances in molecular biology and computer-aided design tools have increased the potential for rational design of proteins. We have undertaken the design, synthesis, and structural characterization of a (3-sandwich protein, betadoublet.Betadoublet was designed to form a pair of identical four-stranded (3-sheets that, facing one another, produce a sandwich. This motif is found in many naturally occurring proteins, such as immunoglobulins (11), concanavalin A (12), exotoxin A (13), and superoxide dismutase (14). A disulfide bond connecting the sheets was added to mimic the disulfide in the immunoglobulin fold and as a design diagnostic tool. The challenges of designing a (-sandwich molecule include accounting for the nonlocal interactions inherent in (3-structure (vs. a-structure) and production of a soluble small ,(protein using only naturally occurring amino acids in a nonrepetitive native-like distribution. Our goal was to produce a stable three-dimensional (-sandwich molecule based on a specific design model.Our approach has been one of a continuing cycle of design and evaluation beginning with the betabellins, a series of proteins based on this (3-sandwich motif that have been prepared by solid-phase peptide synthesis (15, 16). Successive versions of betabellin have improved in total amount of (3-structure and solubility. These molecules have also provided the opportunity to test the resolution of tight-turn geometry using D-amino acids (17)(18)(19). Most recently, betabellin 12 has been shown to have fully assignable NMR spectra in dimethyl sulfoxide, with many local but no longrange nuclear Overhauser effects observable (19). The beta...