In order to design a novel protein to self-assemble into a complex but well-defined architecture, the protein molecule must contain multiple self-associating interfaces. It turns out, somewhat surprisingly, that two distinct self-associating interfaces is sufficient to create a wide range of outcomes, from cages to extended three-dimensional materials. The designed assembly problem therefore boils down to desiging a protein molecule having two geometrically precise interaction interfaces. A design idea we put forward in 2001 was to create protein molecules of this type by fusing together two simpler oligomeric proteins, e.g. a dimer and a trimer. The resulting fusion protein inherits the self-associating interfaces, and thereore the symmetry properties, of the two daughter components. The key problem is how to make such a fusion in a way that the two oligomeric components are held in a specific geometry required to make some desired architecture. One solution is to genetically fuse an oligomer that ends in an alpha helix to an oligomer that begins in an alpha helix, using helix-preferring amino acids for any extra linking residues. If the two oligomeric components are connected by a continuous helix running between them, then their relative orientation can be anticipated in advance. In initial proof of concept experiments, two students, Jennifer Padilla and Chris Colovos, identified a dimer and trimer that could be fused in a way that was predicted to give the right geometry for self-assembly of a 12-subunit tetrahedral cage, about 160A in diameter and a half-megadalton in mass. Those experiments gave rise to assemblies that were roughly consistent with the design, but which were too polymorphic to crystallize and confirm the results.

In 2012, a new student (Yen-Ting Lai) re-inspected the originally designed protein sequence and identified a few amino acids that could be causing steric classes and preventing correct assembly. A new construct was obtained, differening from the original design by just two amino acids, which assembled into homogeneous 12-subunit complexes. The protein was crystallized and shown to match the designed tetrahedral cage structure overall, but with substantial deviations arising from alternative packing modes of the dimeric component, combined with helix flexibility. At essentially the same time, former student Neil King, as a postdoc in David Baker's laboratory, was able to design two novel protein assemblies by using computational interface design. In that case, a natural oligomer (a trimer) provided one self-associating interface, and a second geometrically specific (dimeric) interface was designed computationally using the Rosetta-Design program; several computationally designed constructs had to be screened to find successful designs. These design strategies and others being developed in various laboratories are opening the way to a broad range of engineered protein materials with biomedical and nanotechnology applications.