Marcia E. Newcomer is at Vanderbilt University, School of Medicine, Department of Biochemistry, Nashville, Tennessee 37232, USA.newcomer@lhmrba.hh.vanderbilt.edu
Protein polymerization via 3D domain swapping has been suggested to promote amyloid plaque formation. Two recent papers provide support for a role for domain swapping in the formation of highly ordered aggregates such as the fibrils characteristic of amyloid plaques.
A number of neurodegenerative diseases are associated with the accumulation of amyloid fibrils, insoluble protein aggregates of otherwise soluble proteins. In familial amyloid polyneuropathy, the fibrils are composed of transthyretin, while for Creutzfeldt-Jakob disease, the fibrils contain prion protein PrP. X-ray fiber diffraction studies indicate that amyloid fibrils share a common structural motif (an extended -sheet in which the strands are oriented perpendicular to the long axis of the fiber) despite the fact that the proteins that assemble into these structures do not themselves have structural homology. Furthermore, the ability to assemble into fibrils is not unique to the disease-associated proteins, as in vitro fibril assembly has been demonstrated for a variety of proteins. Nor is there a sequence motif associated with the propensity for fibril assembly. 3D domain swapping has been proposed as a mechanism to promote the polymerization of soluble proteins into fibrils1,
2. Now on page 316 of this issue of Nature Structural Biology3 and in a paper published in the previous issue4 two groups present crystallographic structures that support a role for domain swapping in fibril formation.
3D domain swapping At the heart of these structures is 3D domain swapping, an oligomerization mechanism which has been described for a variety of native as well as non-native structures. Eisenberg defined 3D domain swapping as the replacement of a portion of the tertiary structure of a protein with an identical piece from a second polypeptide chain2 (Fig 1). When the exchange is reciprocated, domain-swapped dimers embrace with the exchange of elements of secondary structure or domains. However, if the exchange is not reciprocated but propagated along multiple polypeptide chains higher order assemblies, such as the fibrils characteristic of amyloid plaques, may form.
Retinol binding protein (top) is a monomeric lipocalin, while the homologous odorant binding protein (bottom) is a dimer in which the helices have been swapped.
A variety of domain-swapped structures have been described. The extent to which the proteins are intertwined may be limited, as it is for domain-swapped staphylococcal nuclease5, or extensive, as it is for the dimeric interferon- (ref. 6). Both swapped and unswapped versions of a protein fold shared by homologous sequences have been described, as well as swapped and unswapped versions of a single protein (ref. 2). In either case, in order to 'swap' structural elements there must be a hinge, or linker region, which permits the protein to recapitulate the native fold from two polypeptide chains, to form essentially what one might term a 'hybrid' fold. Eisenberg has defined the structure that the polypeptide adopts when monomeric as the 'closed' monomer and the conformation of the polypeptide in the domain- swapped oligomer is known as the 'open' monomer2. The 'closed' interface is the intramolecular interface found in the monomer structure and recreated by two polypeptide chains in the domain-swapped structure. The 'open' interface is the interface unique to the oligomeric form of the protein and consequently any favorable interactions found here provide a gain in free energy for the oligomeric fold.
The RNase A dimer In principal, any protein is capable of oligomerization by 3D domain swapping, given that certain criteria are met. The protein fold must contain an exchangeable element appended to a hinge region and this hinge region should allow the placement of the swapped region in both an intra- and intermolecular fashion. Furthermore, the energy barrier between the monomeric fold and domain swapped folds must be low to allow transition between the two conformations. Given a destabilizing milieu (or mutation) and a high local concentration, monomeric proteins could transiently unfold and refold, aggregating via 3D domain swapping as the polypeptides intertwine to reconstruct the native fold from multiple polypeptide chains. However, the domain-swapped structures described to date are for closed oligomers, that is, dimers or trimers that cannot form higher order polymers because there are no more interacting surfaces available for propagation of the assembly. These closed oligomers are therefore 'dead end' structures on what should be the path to polymerization. The significance of two domain-swapped bovine pancreatic ribonuclease A (RNase A) structures described by Liu et al.4,
7 is that they elegantly illustrate a mechanism for polymerization by the swapping of multiple domains.
RNase A forms two types of dimers with different biophysical and biochemical properties. In one, the dimer forms by swapping its N-terminal -helix with that of an identical molecule. In the second domain swapped dimer structure for RNase A described by Liu et al.4 the C-terminal -strand of one monomer is swapped with that of an identical molecule. The existence of two modes of domain swapping available to a single protein suggests a model for how higher order assemblies might occur. In the context of the RNase A fold, there are no topological constraints to restrict the formation of both 'swaps' in a single molecule. Each molecule can trade both termini with two partners and thus higher order polymers can readily form (Fig. 2).
The RNase A dimer and the polar zipper model There is also an aspect of the RNase A structure that has direct bearing on the 'polar zipper' model8 for fibril assembly. In individuals afflicted with Huntington's disease, the protein huntingtin assembles into amyloid fibrils, and the capacity to do so has been linked to the presence of an expanded polyglutamine sequence in the protein. Perutz proposed that stretches of glutamine fold into antiparallel -sheets or hairpins. In addition to the hydrogen bonds characteristic of -sheets, hydrogen bonds can form between the amides of the glutamine side chains (Fig. 3). Three layers of hydrogen bonds then stabilize the polar zipper: side chains that participate in hydrogen bonds flank a central layer of peptide backbone hydrogen bonds. In the C-terminal domain-swapped RNase A structure, the hinge region participates in an antiparallel -sheet with the same region from its dimer mate. A consequence is the juxtaposition of Asn 113 from both polypeptide chains. The structure clearly reveals a hydrogen bond between the two asparagines and thus gives us a glimpse of a polar zipper.
The ground is now fertile for more speculation: could the presence of polyglutamine repeats promote domain swapping? Here we can see how polyglutamine expansions might indeed exacerbate the inherent ability of proteins to swap domains. In order to accommodate the insertion into the tertiary structure, some adaptation to the native fold will be necessary and so the insertion itself will be destabilizing. In addition, the insertion itself serves as the requisite hinge that permits the domains to make intermolecular contacts that recapitulate the intramolecular contacts of the monomer: the insertion provides a flexible spacer between the N- and C-terminal regions. Furthermore, the hinge region can form a polar zipper with like regions of its domain-swap partner to form an 'open interface' rich in stabilizing hydrogen bonds.
Cystatin dimer In a separate paper Janowski et al.3 describe the 3D domain swapped structure of cystatin C, a protein with amyloidogenic properties and a potent inhibitor of cysteine proteases. The cysteine protease inhibitor is found in amyloid deposits of elderly patients with cerebral amyloid angiopathy. The substitution of Leu with Gln at position 68 (L68Q) in the 120-amino acid polypeptide is associated with hereditary cystatin C amyloid angiopathy. The description of domain swapping in this protein, which has clearly been established as a component of amyloid fibrils, provides additional support for a role for domain swapping in fibril assembly. Furthermore, it is apparent from this structure that the disease causing mutation might promote polymerization by destabilization of the monomeric fold, making the transition to the domain-swapped fold accessible.
However attractive the hypothesis that the mutation destabilizes the native fold and allows for polymerization via domain swapping may be, there are still some findings to be reconciled. The reported structure does not indicate how fibril formation may occur since as described it is a 'dead end' dimer with no means to polymerize. The authors propose a model for how such an assembly might be constructed. In the proposed model, the domain swapping is not reciprocated but continues in an open-ended fashion such that molecule A donates to molecule B, which donates to molecule C and so on (Fig. 4). Molecules in the polymer form a growing spiral rather the dyad found in the crystal structure.
Figure 4. Schematic representations for the reciprocated and propagated swap models for Cystatin A.
In principal there is no reason to discount such a model, but for this particular protein in the observed swap mode it is not clear what might stabilize this polymerization. Since monomeric and domain swapped versions of a protein have in common their 'closed' interfaces, the added stabilization energy for the polymeric fold is a result of the additional contacts in the 'open' interface. The L68Q mutation is at the core of the closed interface of cystatin C. This mutation would then destabilize the native fold, and the free energy required to unfold and adopt a domain-swapped conformation is no longer prohibitive. However, the 'closed' interface is equally destabilized in the domain-swapped version. The open interface, or what is gained by dimerization, described in the reciprocated domain swap cannot form in the proposed fibril model and so it is not apparent what stabilization energy is gained by polymerization. Nonetheless, the structure clearly demonstrates that given a high local concentration and a destabilizing element, strands can be swapped between protomers and in such a fashion higher order assemblies may form from interlaced subunits.
Conclusions These two reports combine to provide tremendous support for a role for 3D-domain swapping in amyloid fibril formation. Liu et al.4 beautifully illustrate how any protein, given the appropriate conditions, might assemble itself into an ordered aggregate by the intertwining of domains and secondary structural elements from multiple polypeptide chains. And Janowski et al.3 present us with what may be an intermediate in this process for a protein that actually forms fibrils in vivo.
-sheet in which the strands are oriented perpendicular to the long axis of the fiber) despite the fact that the proteins that assemble into these structures do not themselves have structural homology. Furthermore, the ability to assemble into fibrils is not unique to the disease-associated proteins, as in vitro fibril assembly has been demonstrated for a variety of proteins. Nor is there a sequence motif associated with the propensity for fibril assembly. 3D domain swapping has been proposed as a mechanism to promote the polymerization of soluble proteins into fibrils
(ref. 
-helix with that of an identical molecule. In the second domain swapped dimer structure for RNase A described by Liu et al.
