1. Introduction

The notion of protein structure classification emerged from several studies conducted during the late 1970s and early 1980s that aimed to elucidate the basic principles of protein folding and protein structure evolution. The early work of Chothia and coworkers pioneered the division of protein structures into four major classes based on their secondary-structure composition and demonstrated that simple geometrical features of secondary-structural elements govern their mutual arrangement in distinct architectures (Chothia, 1984; Chothia et al., 1977; Levitt & Chothia, 1976). Later, Jane Richardson provided a more detailed classification deduced from the topological details of less than 200 structures (Richardson, 1977, 1981). The progress made in the field in the 1980s was reviewed by Chothia & Finkelstein (1990). By that time, the term `fold' was already established and it was intended to outline three major aspects of protein three-dimensional structure: the secondary structures of which the protein is composed, their relative arrangement and the path taken through the structure by the polypeptide chain. Thus, the fold of a protein was defined through its composition, architecture and topology.

It also became apparent that homologous proteins of similar sequence adopt the same fold. It was also noted that some folds are populated by proteins with dissimilar sequences. These observations raised the question of whether the structural similarities between different proteins indicate distant homology or simply result from the basic principles of physics and chemistry. The Structural Classification of Proteins (SCOP) database, established in 1994, circumvented this argument by introducing a new category: the superfamily. This level of classification aimed to group together proteins that have probably descended from a common ancestor but whose sequences have diverged beyond detectable similarity (Murzin et al., 1995). The notion of probable distant homology `relaxed' the definition of protein fold by transforming it into a `consensus' fold common to a set of evolutionarily related proteins. Dissecting the protein structure into an evolutionarily conserved core and a variable periphery kept the discrete classification, thus avoiding the continuous folding space and Russian-doll problems that arise from partial structural similarities.

It was thought at that time that the number of architectural types was limited. Moreover, although some structural variations were observed amongst evolutionarily related proteins, none of these affected the common structural core. Therefore, it was assumed that the protein fold is evolutionarily stable in that it retains its characteristic features, although some structural changes could be anticipated. Similarly, it was thought that every protein folds into a single three-dimensional structure and that its structural core is insensitive to large conformational changes related to function or formation of quaternary structure.

At the advent of structural genomics (SG) and the Protein Structure Initiative (PSI; https://www.nigms.nih.gov/Initiatives/PSI/ ), analysis of the trends of newly discovered folds seemed to indicate that most of protein fold space had been explored. The growth in the number of new folds in SCOP had almost stalled and the rate of discovery of new superfamilies and families obviously slowed down. After the PSI launch the number of new folds, superfamilies and families rose again, mainly because the PSI SG centres targeted proteins with no significant sequence similarity to known structures. Currently, owing to the joint efforts of SG centres and independent structural biology groups there is more than one structural representative for most of the characterized families (Andreeva et al., 2008). Recent analysis of the distribution of protein families characterized by structural genomics has confirmed the dominant role of the largest known superfamilies, which have grown further in their number of constituent families (Andreeva et al., 2008). In addition, other superfamilies have grown large rather unexpectedly. The evolutionary success of these `new rich' superfamilies is probably a consequence of the presence of unusual conserved and presumably functionally important features in their folds. One of these `new rich' superfamilies, for example, is the dimeric α+β barrel superfamily in SCOP, several new members of which have come from the first structures of metagenomic sequences (Yooseph et al., 2007).

Initially, it was anticipated that a large number of new folds would be discovered owing to the breath of coverage of fold space targeted by the PSI. Interestingly, this has not turned out to be the case as a substantial portion of the structures coming from SG revealed significant structural similarities to already known folds and in fact represent variations of existing protein architectures and topologies. However, there were several unexpected findings of previously unseen topologies and architectures. For instance, a number of SG structures revealed superfamily-specific folds in which the core β-­sheet structures are tailored into unique shapes. PSI also greatly increased the number of protein topologies with high contact order, which is known to limit the success of current ab initio structure-prediction methods (Bonneau et al., 2002), thus providing invaluable high-resolution templates for modelling. Without previous pre­conceptions, comparisons of some SG structures revealed dramatic structural variations in related proteins that go beyond the expectations based on their sequence similarity. These provide convincing examples of how protein folds can evolve without compromising the integrity of the structure of the functional site.

The plethora of structural data delivered over the past decade by SG and independent groups revealed numerous examples of atypical structural features and structural variations that have challenged many longstanding tenets in protein science (Andreeva et al., 2007; Andreeva & Murzin, 2006). Amongst these, for instance, is the discovery of the deep trefoil knot (Nureki et al., 2002). SG has determined the structures of several knotted proteins, which in turn helped to dispel one of the oldest dogmas in molecular biology prohibiting knots in protein chains. In addition, the old folding paradigm `one sequence – one structure' is increasingly being challenged as more and more structural variations are observed in protein families and their individual members. It has become evident that the protein fold is neither physically nor evolutionarily invariant (Andreeva & Murzin, 2006).

These structural variations found amongst homologous as well as individual proteins create nontrivial structural relationships at any `evolutionary' level of SCOP and increase the structural diversity within families and superfamilies. In essence, the classification of a new protein in SCOP depends on its relationship to protein(s) of known structure. If there is strong evidence that a protein is homologous to other protein(s) in SCOP then it is classified into an existing superfamily. The fold of these evolutionary-related proteins is an attribute that describes the given evolutionary lineage. If a protein is not homologous to any protein in SCOP and has a fold that differs from any known fold in composition, architecture or topology, then it is classified as a novel fold. Paradoxically, a protein with a novel fold may well be structurally similar to (but still distinct from) previously classified proteins, whereas a protein classified as a new member of an existing superfamily may display novel features in architecture or topology.

In this article, we survey protein structures resulting from various SG efforts, in particular proteins structurally characterized by the Joint Center for Structural Genomics (JCSG) and published in this special issue of Acta Crystallographica Section F (Fig. 1 and other images below). Here, we focus on some interesting examples of novel protein architectures and unusual topologies that have given new insights into protein folds and evolution.

Figure 1 Gallery of selected protein structures determined by the JCSG (see also Figs. 5, 6 and 7). (a) Acetoacetate decarboxylase (ADC) subunit (PBD entry 3c8w ). β-Strands in the double-barrel β-sheet are shown as coloured arrows; other secondary-structure elements and loops are shown as silver coils. (b) DUF1089 protein PA1994 (PDB entry 2h1t ) coloured by rainbow. An `unswapped' monomer is shown, a large β-sheet of which is folded into a spiral roll. (c) DUF1831 protein lp2179 (PDB entry 2iay ) coloured by secondary structure: red, α-helix; yellow, β-strand; green, loop. (d) DUF1470 protein Jann2411 (PDB entry 3h0n ) coloured by secondary structure, with the N-terminal subdomain coloured as in (c) and the C-terminal subdomain coloured using an alternative palette: cyan, α-helix; purple, β-strand; pink, loop. The sphere represents the zinc ion. (e) DUF1488 protein Shew3726 (PDB entry 2gpi ) coloured by secondary structure, with the rare left-handed β-X-β unit coloured using an alternative palette. (f) DUF2006 protein NE1406 (PDB entry 2ich ) viewed along the pseudo-twofold axis that relates its similar barrel domains. The topologically equivalent β-strands in both domains and in ADC (a) are shown in the same colour.

2. Novel architectures

The main building blocks of protein folds are α-helices and β-sheets, which pack together enclosing clusters of nonpolar residues or hydrophobic cores. In theory, the number of different possible globular arrangements of these blocks around a single hydrophobic core is limited. All of the simple `two-layer' architectures have already been observed (Chothia et al., 1997). Nevertheless, new architectural types are still being uncovered. Typically, these new architectures are multilayer and/or made of customized building blocks such as nonpolar helices that can be buried in the protein interior, combinations of long and short helices, β-sheets that are tailored into particular shapes etc. Such customized secondary structures are usually coded by particular or atypical sequence patterns, which tend to be superfamily-specific.

The β-sheet is a very versatile building block and comes in many different shapes and sizes. In addition to the two major architectural types of β-sheet proteins, β-sandwich and β-barrel, a number of minor types have been discovered, including β-helices, β-propellers, β-prisms, β-clips, barrel–sandwich hybrids etc. (Chothia & Murzin, 1993). Below, we describe two new β-sheet architectures found in structures determined by the JCSG (Fig. 2).

Figure 2 New β-sheet architectures. (a) A bifurcated X-shaped β-sheet can fold upon itself on both sides, forming a double barrel. (b) A very large β-sheet can be folded into a β-spiral roll with overlapping edges. This architecture combines features of both β-­barrel and β-sandwich. In both parts, for simplicity, the arrows denote β-strands but do not define the strand directionality. The actual β-sheets of these architectures may comprise parallel and antiparallel strands.

3. Unusual topologies

Several empirical rules regarding protein topology have been established by previous analyses of protein structures. One of them postulates that secondary-structure elements that are adjacent in sequence are adjacent in the structure. That is, protein structures generally have a low contact order (Bonneau et al., 2002). Other rules describe topological preferences, such as the absence of knots, the right-handedness of connections between two parallel β-strands etc. (Chothia & Finkelstein, 1990; Sternberg & Thornton, 1976, 1977). These rules have been the subject of many studies looking for their physical and/or biological bases. The statistical stringency of these rules is now in need of revision owing to the recent influx of new protein structures. Moreover, many more known exceptions to these rules are found in recent protein structures with unusual topologies. Some of these are discussed in this section.

3.2. Knotted protein structures

Until recently, deep knots had not been observed in protein chains and therefore their formation was considered to be impossible. In recent years, several structural genomics centres have determined structures of various knotted proteins, in particular proteins from a fast-growing superfamily of putative methyltransferases. All the members of this superfamily contain an obligatory deep trefoil knot that forms the binding site for an S-adenosylmethionine cofactor (Lim et al., 2003). A different trefoil knot, the smallest of its kind, has been discovered in the structure of the uncharacterized protein MJ0366 (PDB entry 2efv ; T. S. Kumarevel, P. Karthe, S. Kuramitsu & S. Yokoyama, unpublished work; Fig. 3). This protein is classified into the ribbon–helix–helix (RHH) superfamily of DNA-binding proteins in SCOP and it is very likely that it may exhibit a similar DNA-binding function. This knotted fold, comprising two RHH motifs arranged as in the typical RHH dimeric fold and connected by a linker, probably resulted from a gene-duplication/fusion event. Interestingly, a fully functional protein from the two linked RHH subunits of the Arc repressor was artificially created long before the discovery of the first deep knot (Robinson & Sauer, 1996), but it was not recognized at the time that the single-chain protein could potentially form a knotted fold.

Figure 3 A trefoil knot in the structure of the uncharacterized protein MJ0366 (a duplicated RHH motif).

3.3. Globular oligomers of noncompact subunits and high-contact-order monomeric structures

It was assumed that in the beginning all proteins were monomeric and some then evolved into oligomeric structures. A very small number of known globular oligomers composed of interlocking noncompact subunits were considered to be an exception confirming this rule. In recent years, the PSI has contributed to the discovery of many new families, the individual members of which form this type of oligomer (Fig. 4a). These include the following SCOP families: YejL-like (Pfam 07208; DUF1414), HP0242-like (Pfam 09442; DUF2018), YopT-like (Pfam 09467), AF2212-like (Pfam 01954; DUF104), YonK-­like (Pfam 09642), AF2331-like, SMc04008-like (Pfam 06844; DUF1244) and CsrA-like (Pfam 02599) (Murzin et al., 1995). Interlaced obligatory oligomers have also been discovered in some new members of the previously defined superfamilies, for example in the structure of a putative all α-helical NTP pyrophosphohydrolase (PDB entry 2rfp ) determined by the JCSG (Han, Elsliger et al., 2010).

Figure 4 (a) The structure of UPF0352 protein CPS2611 (PDB entry 2ota ; S. M. Vorobiev, W. Zhou, M. Su, J. Seetharaman, H. Wang, H. Janjua, K. Cunningham, L.-C. Ma, C. Liu, T. B. Acton, R. Xiao, G. T. Montelione, L. Tong & J. F. Hunt, unpublished work) is an example of an obligatory oligomer. It is composed of two interlocking noncompact subunits, which are coloured orange and blue. (b) A representative structure of the DinB-like family member (PDB entry 2f22 ; JCSG, unpublished work). It contains two interlocking structural repeats, which are shown in green and red.

These new structural insights change our understanding of the formation and evolution of oligomeric structures. An evolutionary scenario in which monomers could evolve from oligomeric structures with a single hydrophobic core seems to be equally plausible. Indeed, many monomeric globular folds composed of structural repeats may have evolved from globular assemblies of single-repeat subunits by gene-duplication/fusion mechanisms. Ancestral non­globular interlocking subunits could result in protein structures of higher contact order, such as the representatives of the DinB-like family, as first characterized by the JCSG (Fig. 4b).

4. Fold evolution

Nowadays, it is widely accepted that the protein fold can change and that these changes may affect not only the peripheral elements of the structure but also the core elements (Andreeva & Murzin, 2006; Grishin, 2001a; Murzin, 1998).

An example is the recently discovered nontrivial structural relationship between the global regulatory protein CsrA (PDB entry 1vpz ; Rife et al., 2005) and the C-terminal domain of YqeH GTPase (PDB entry 3ec1 ; Sudhamsu et al., 2008; Fig. 5). The CsrA structure, which was first determined by the JCSG, revealed a dimer of interlocked subunits forming a single domain with an atypical β-sandwich architecture. The YqeH-domain structure (3ec1 ) consists of two compact structural repeats organized into a β-sandwich fold of very similar architecture. The topologies of these two sandwiches are related by segment swapping. Interestingly, CsrA has recently been shown to bind RNA and it seems plausible that YqeH could also have this function. A hypothetical evolutionary link between the two domains would be a CsrA-like dimeric protein with compact subunits packed side-by-side like the YqeH repeats.

Figure 5 Structures of the CsrA dimer (a) and the C-terminal domain of YqeH (b). The individual subunits of CsrA are coloured cyan and blue, whereas the YqeH structural repeats are shown in yellow and red. (c) Stereoview of the superimposition of the CsrA dimer (blue) and the C-terminal domain of YqeH (orange).

4.2. Fold changes following a duplication/fusion event

The JCSG has determined the structures of two different members of the DUF1285 family (PDB entries 2re3 and 2ra9 ) with divergent sequences (Han, Bakolitsa et al., 2010). Their comparison allows the identification of structurally conserved and variable regions and improves the initial classification. The 2re3 /2ra9 fold can be divided into two structurally similar domains, the mutual arrangement of which is conserved owing to a tight association. The N-terminal domain is more conserved than the C-­terminal domain in both sequence and structure. The C-terminal domain of 2re3 contains a barrel-like β-sheet, resulting in some structural similarities to the PH-­like fold. The N-terminal domains of both 2re3 and 2ra9 show no global similarities to known protein structures, except for the structure of the C-terminal domain of 2re3 . The consensus fold of these domains comprises an α-helix flanked at each end by a three-stranded meander β-sheet (Fig. 6). The presence of two copies of this unique fold in the same protein structure (2re3 ) suggests a gene-duplication/fusion event, analogous to 2ich and many other examples in SCOP. The ancestral structure probably contained a few additional structural elements at the N-terminus. These elements could have evolved into the N-terminal tail of the N-terminal domain or the barrel-like extension to the N-terminal β-sheet of the C-terminal domain, as observed in the 2re3 structure. During evolution, the C-­terminal β-­meander of the consensus fold, which is absent in the C-­terminal domain of 2ra9 , was probably lost in this and some other structures. Recent analysis has suggested that DUF1285 belongs to a new superfamily that also includes the UPF0598 and PfamB PB002487 families (L. Aravind, personal communication; N. V. Grishin, personal communication; https://iole.swmed.edu/~grishin/2re3/2re3.htm ). All three families are clearly taxa-specific: DUF1285 is found in α-, γ- and δ-­proteobacteria, PB002487 in β-proteobacteria and UPF0598 in Metazoa. The distinct phylogenetic distributions suggest that these families may have evolved different functional and structural features. Interestingly, a member of PB002487 has already been targeted for structure determination by the JCSG (Target ID 392148, gi:91785099).

Figure 6 A consensus fold of the N- and C-terminal domains of the DUF1285 family. (a) Side-by-side comparison of the N-terminal domains of 2ra9 (green) and 2re3 (red) and the C-­terminal domain of 2re3 (blue). Nonconserved additional regions are shown in grey. (b) Stereoview of the superimposition of the common parts of the three domains.

5. Metamorphic proteins

A small but growing number of `metamorphic' proteins adopt different folded conformations for the same amino-acid sequence under native conditions (Murzin, 2008). Unlike prions, they undergo reversible conformational changes. The recent discoveries of metamorphic proteins that are capable of independent interconversion and of an abrupt fold change in a protein lineage suggest a general nature for this phenomenon.

The Midwest Centre for Structural Genomics (MCSG) has determined two representative homopentameric structures of the YbjQ-like family (DUF74). In one of them (PDB entry 1y2i ; J. S. Brunzelle, G. Minasov, X. Yang, L. Shuvalova, F. R. Collart & W. F. Anderson, unpublished work) the constituent subunits have identical conformations and are related by fivefold symmetry. In the other pentamer (PDB entry 1vr4 ; J. S. Brunzelle, L. K. McNamara, X. Yang, G. Minasov, L. Shuvalova, F. R. Collart & W. F. Anderson, unpublished work) the subunits are similarly arranged around the fivefold axis but adopt different conformations. One of the subunits has a conformation that is identical to the conformation of the 1y2i subunits, whereas a subunit on the opposite side of this pentamer has an alternative folded conformation. The central part of the latter forms a β-sheet instead of α-­helices and makes different intersubunit contacts (Figs. 7a and 7b). The equivalent regions in the remaining subunits of the 1vr4 pentamer are partly invisible.

Figure 7 Metamorphic proteins. (a, b)Side-by-side comparison of alternatively folded subunits of the DUF74 pentamer (PDB entry 1vr4 ). Chain A and chain D (b) are coloured according to their secondary structure. The adjacent subunits are coloured as follows: chain B, light blue; chain C, magenta; chain E, pink. (c, d) Side-by-side comparison of the Sfri0576-like family structures 2q3l (c) and 2ook (d). The equivalent nonpolar residues that are exposed on the 2q3l surface and buried in the 2ook core are shown in stick representation.

Such metamorphic proteins may have evolved in other families and can account for the abrupt fold changes (Murzin, 2008; Roessler et al., 2008). One recent example of this type of change was revealed by the structural comparison of two closely related proteins that are members of the Sfri0567-like family characterized by the JCSG (Kumar et al., 2010). These proteins share 54% sequence identity but adopt notably different conformations referred to as `open' and `closed'. In the `open' structure (PDB entry 2q3l ) two long helices define the walls of a deep groove, whereas in the `closed' (PDB entry 2ook ) structure these helices are refolded so that the groove is occluded (Figs. 7c and 7d). A substantial number of solvent-exposed nonpolar residues in the `open' structure become buried in the hydrophobic core of the `closed' structure. The different conform­ational states observed in 2q3l and 2ook are stabilized by the formation of different dimeric (and crystal) contacts.

6. Concluding remarks

One of the primary advantages of the PSI SG initiative is that it promotes discovery-driven rather than hypothesis-driven research. This approach allows the capture of many unexpected protein relationships that provide important new insights into protein-structure evolution. Structural genomics has made major contributions to the discovery of new protein topologies and architectures and thus ultimately has accelerated our understanding of protein folding.

The Protein Structure Initiative has produced a great amount of data which we are just beginning to understand and appreciate. Some of these data will also await the release of complementary information on homologous proteins and/or any experimental biological insights in order to reveal their real value. Protein structures that are single representatives of folds bear the potential for new discoveries in the near future. Their prize will be given when the structure of another representative of this fold is determined.