The Structural Classification of Proteins (SCOP) database is a largely manual classification of protein structural domains based on similarities of their structures and amino acidsequences. A motivation for this classification is to determine the evolutionary relationship between proteins. Proteins with the same shapes but having little sequence or functional similarity are placed in different superfamilies, and are assumed to have only a very distant common ancestor. Proteins having the same shape and some similarity of sequence and/or function are placed in "families", and are assumed to have a closer common ancestor.
The SCOP database is freely accessible on the internet. SCOP was created in 1994 in the Centre for Protein Engineering and the Laboratory of Molecular Biology. It was maintained by Alexey G. Murzin and his colleagues in the Centre for Protein Engineering until its closure in 2010 and subsequently at the Laboratory of Molecular Biology in Cambridge, England. As of January 2014[update], the work on SCOP has been discontinued and the last official version of SCOP is 1.75 (released June 2009). The prototype of a new Structural Classification of Proteins 2 (SCOP2) database has been made publicly available. SCOP2 defines a new approach to the classification of proteins that is essentially different from SCOP, but retains its best features.
The source of protein structures is the Protein Data Bank. The unit of classification of structure in SCOP is the protein domain. What the SCOP authors mean by "domain" is suggested by their statement that small proteins and most medium-sized ones have just one domain, and by the observation that human hemoglobin, which has an α2β2 structure, is assigned two SCOP domains, one for the α and one for the β subunit.
The shapes of domains are called "folds" in SCOP. Domains belonging to the same fold have the same major secondary structures in the same arrangement with the same topological connections. 1195 folds are given in SCOP version 1.75. Short descriptions of each fold are given. For example, the "globin-like" fold is described as core: 6 helices; folded leaf, partly opened. The fold to which a domain belongs is determined by inspection, rather than by software.
The levels of SCOP are as follows.
- Class: Types of folds, e.g., beta sheets.
- Fold: The different shapes of domains within a class.
- Superfamily: The domains in a fold are grouped into superfamilies, which have at least a distant common ancestor.
- Family: The domains in a superfamily are grouped into families, which have a more recent common ancestor.
- Protein domain: The domains in families are grouped into protein domains, which are essentially the same protein.
- Species: The domains in "protein domains" are grouped according to species.
- Domain: part of a protein. For simple proteins, it can be the entire protein.
The broadest groups on SCOP are the protein fold classes. These classes group structures with similar secondary structure composition, but different overall tertiary structures and evolutionarily origins. This is the top level "root" of the SCOP hierarchical classification.
- All alpha proteins  (284)
- Domains consisting of α-helices
- All beta proteins  (174)
- Domains consisting of β-sheets
- Alpha and beta proteins (a/b)  (147)
- Mainly parallel beta sheets (beta-alpha-beta units)
- Alpha and beta proteins (a+b)  (376)
- Mainly antiparallel beta sheets (segregated alpha and beta regions)
- Multi-domain proteins (alpha and beta)  (66)
- Folds consisting of two or more domains belonging to different classes
- membrane and cell surface proteins and peptides  (58)
- Does not include proteins in the immune system
- Small proteins  (90)
- Usually dominated by metal ligand, heme, and/or disulfide bridges
- coiled-coil proteins  (7)
- Not a true class
- Low resolution protein structures  (26)
- Peptides and fragments. Not a true class
- Peptides  (121)
- peptides and fragments. Not a true class.
- Designed proteins  (44)
- Experimental structures of proteins with essentially non-natural sequences. Not a true class
- All alpha proteins  (284)
The number in brackets, called a "sunid", is a COP ique integer entifier for each node in the SCOP hierarchy. The number in parentheses indicates how many elements are in each category. For example, there are 284 folds in the "All alpha proteins" class. Each member of the hierarchy is a link to the next level of the hierarchy.
Each class contains a number of distinct folds. This classification level indicates similar tertiary structure, but not necessarily evolutionary relatedness. For example, the "All-α proteins" class contains >280 distinct folds, including: Globin-like (core: 6 helices; folded leaf, partly opened), long alpha-hairpin (2 helices; antiparallel hairpin, left-handed twist) and Type I dockerin domains (tandem repeat of two calcium-binding loop-helix motifs, distinct from the EF-hand).
Domains within a fold are further classified into superfamilies. This is a largest grouping of proteins for which structural similarity is sufficient to indicate evolutionary relatedness and therefore share a common ancestor. However, this ancestor is presumed to be distant, because the different members of a superfamily have low sequence identities. For example, the two superfamilies of the "Globin-like" fold are: the Globin superfamily and alpha-helical ferredoxin superfamily (contains two Fe4-S4 clusters).
Protein families are more closely related than superfamilies. Domains are placed in the same family if that have either:
- >30% sequence identity
- some sequence identity (e.g., 15%) and perform the same function
The similarity in sequence and structure is evidence that these proteins have a closer evolutionary relationship than do proteins in the same superfamily. Sequence tools, such as BLAST, are used to assist in placing domains into superfamilies and families. For example, the four families in the "Globin-like" superfamily of the "Globin-like" fold are Truncated hemoglobin (lack the first helix), Nerve tissue mini-hemoglobin (lack the first helix but otherwise is more similar to conventional globins than the truncated ones), Globins (Heme-binding protein), and Phycocyanin-like phycobilisome proteins (oligomers of two different types of globin-like subunits containing two extra helices at the N-terminus binds a bilin chromophore). Families in SCOP are each assigned a concise classification string, sccs, where the letter identifies the class to which the domain belongs; the following integers identify the fold, superfamily, and family, respectively (e.g., a.1.1.2 for the "Globin" family).
Within a family are protein domains. Proteins are placed in the same protein domain if they are isoforms of each other, or if they are essentially the same protein, but from different species. This is done manually. The "protein domains" are further subdivided into species. ("Protein domains" are not on separate pages in the current release of SCOP; in pre-SCOP, they are on separate pages). For example, the "Globins" family contains >80 structurally characterised members, including Leghemoglobin from Yellow lupin and Soybean, and Hemoglobin, alpha-chain from human, horse and deer.
PDB entry domains
A "TaxId" is the taxonomy ID number and links to the NCBI taxonomy browser, which provides more information about the species to which the protein belongs. Clicking on a species or isoform brings up a list of domains. For example, the "Hemoglobin, alpha-chain from Human (Homo sapiens)" protein has >190 solved protein structures, such as 2dn3 (complexed with cmo), and 2dn1 (complexed with hem, mbn, oxy). Clicking on the PDB numbers is supposed to display the structure of the molecule, but the links are currently broken (links work in pre-SCOP).
Most pages in SCOP contain a search box. Entering "trypsin +human" retrieves several proteins, including the protein trypsinogen from humans. Selecting that entry displays a page that includes the "lineage", which is at the top of most SCOP pages. The page includes the following information.
- 1. Root: scop
- 2. Class: All beta proteins 
- 3. Fold: Trypsin-like serine proteases 
- barrel, closed; n=6, S=8; greek-key
- duplication: consists of two domains of the same fold
- 4. Superfamily: Trypsin-like serine proteases 
- 5. Family: Eukaryotic proteases 
- 6. Protein: Trypsin(ogen) 
- 7. Species: Human (Homo sapiens) [TaxId: 9606] 
Searching for "Subtilisin" brings up the protein, "Subtilisin from Bacillus subtilis, carlsberg", with the following lineage.
- 1. Root: scop
- 2. Class: Alpha and beta proteins (a/b) 
- Mainly parallel beta sheets (beta-alpha-beta units)
- 3. Fold: Subtilisin-like 
- 3 layers: a/b/a, parallel beta-sheet of 7 strands, order 2314567; left-handed crossover connection between strands 2 & 3
- 4. Superfamily: Subtilisin-like 
- 5. Family: Subtilases 
- 6. Protein: Subtilisin 
- 7. Species: Bacillus subtilis, carlsberg [TaxId: 1423] 
Although both of these proteins are proteases, they do not even belong to the same fold, which is consistent with them being an example of convergent evolution.
Comparison to other classification systems
SCOP classification is more dependent on manual decisions than the semi-automatic classification by CATH, its chief rival. Human expertise is used to decide whether certain proteins are evolutionary related and therefore should be assigned to the same superfamily, or their similarity is a result of structural constraints and therefore they belong to the same fold. Another database, FSSP, is purely automatically generated (including regular automatic updates) but offers no classification, allowing the user to draw their own conclusion as to the significance of structural relationships based on the pairwise comparisons of individual protein structures.
By 2009, the original SCOP database manually classified 38,000 PDB entries into a strictly hierarchical structure. With the accelerating pace of protein structure publications, the limited automation of classification could not keep up, leading to a non-comprehensive dataset. The Structural Classification of Proteins extended (SCOPe) database was released in 2012 with far greater automation of the same hierarchical system and is full backwards compatible with SCOP. In 2014, manual curation was reintroduced into SCOPe to maintain accurate structure assignment. As of February 2015, SCOPe 2.05 classified 71,000 of the 110,000 total PDB entries.
SCOP2 is a prototype classification system that aims to more the evolutionary complexity inherent in protein structure evolution. It is therefore not a simple hierarchy, but a network connecting protein superfamilies representing structural and evolutionary relationships such as circular permutations, domain fusion and domain decay. Consequently, domains are not separated by strict fixed boundaries, but rather are defined by their relationships to the most similar other structures. As of February 2015, the SCOP2 prototype classifies 995 PDB entries.
- ^Andreeva, A.; Howorth, D.; Chandonia, J. -M.; Brenner, S. E.; Hubbard, T. J. P.; Chothia, C.; Murzin, A. G. (2007). "Data growth and its impact on the SCOP database: New developments". Nucleic Acids Research. 36 (Database issue): D419–D425. doi:10.1093/nar/gkm993. PMC 2238974. PMID 18000004.
- ^Hubbard, T. J.; Ailey, B.; Brenner, S. E.; Murzin, A. G.; Chothia, C. (1999). "SCOP: A Structural Classification of Proteins database". Nucleic Acids Research. 27 (1): 254–256. doi:10.1093/nar/27.1.254. PMC 148149. PMID 9847194.
- ^Lo Conte, L.; Ailey, B.; Hubbard, T. J.; Brenner, S. E.; Murzin, A. G.; Chothia, C. (2000). "SCOP: A Structural Classification of Proteins database". Nucleic Acids Research. 28 (1): 257–259. doi:10.1093/nar/28.1.257. PMC 102479. PMID 10592240.
- ^Andreeva, A.; Howorth, D.; Brenner, S. E.; Hubbard, T. J.; Chothia, C.; Murzin, A. G. (2004). "SCOP database in 2004: Refinements integrate structure and sequence family data". Nucleic Acids Research. 32 (90001): D226–D229. doi:10.1093/nar/gkh039. PMC 308773. PMID 14681400.
- ^Murzin, A. G.; Brenner, S.; Hubbard, T.; Chothia, C. (1995). "SCOP: A structural classification of proteins database for the investigation of sequences and structures"(PDF). Journal of Molecular Biology. 247 (4): 536–540. doi:10.1016/S0022-2836(05)80134-2. PMID 7723011.
- ^PDB: 2DN1; Park SY, Yokoyama T, Shibayama N, Shiro Y, Tame JR (July 2006). "1.25 Å resolution crystal structures of human haemoglobin in the oxy, deoxy and carbonmonoxy forms". J. Mol. Biol. 360 (3): 690–701. doi:10.1016/j.jmb.2006.05.036. PMID 16765986.
- ^Lo Conte, L.; Brenner, S. E.; Hubbard, T. J.; Chothia, C.; Murzin, A. G. (2002). "SCOP database in 2002: Refinements accommodate structural genomics". Nucleic Acids Research. 30 (1): 264–267. doi:10.1093/nar/30.1.264. PMC 99154. PMID 11752311.
- ^ ab"What is the relationship between SCOP, SCOPe, and SCOP2". scop.berkeley.edu. Retrieved 2015-08-22.
The Structural Classification of Proteins (SCOP) database has facilitated the development of many tools and algorithms and it has been successfully used in protein structure prediction and large-scale genome annotations. During the development of SCOP, numerous exceptions were found to topological rules, along with complex evolutionary scenarios and peculiarities in proteins including the ability to fold into alternative structures. This article reviews cases of structural variations observed for individual proteins and among groups of homologues, knowledge of which is essential for protein structure modelling.
- homology modelling
- metamorphic proteins
- protein structure evolution
- Structural Classification of Proteins (SCOP)
Over the past two decades, the Structural Classification of Proteins (SCOP) database has become an essential resource in many areas of protein research . Initially designed to assist structural biologists in the analysis of structural similarities between proteins, SCOP facilitated the development of tools and algorithms and it has been successfully used in protein structure prediction and large-scale genome annotations [2,3]. SCOP also contributed to our understanding of protein repertoire, including how proteins relate to each other and how their structures and functions evolved . Each grouping in the classification was the product of a careful, systematic analysis of protein structures and a detailed knowledge of protein function and evolution. Many distant evolutionary relationships between proteins were first discovered during their analysis for classification in SCOP [5–7]. Some of these have never been described in the literature and thus the SCOP database has become a repository for many interesting research findings.
The notion of protein evolution, incorporated in SCOP, allowed grouping of proteins based not only on their structural features but also on their common evolutionary origin. Depending on the degree of evolutionary divergence and structural similarity, discrete units (domains) are hierarchically organized into families and superfamilies. These are further grouped into structural folds, defined by the domains' topology and architecture, and classes reflecting their secondary structure composition. The classification of proteins in SCOP depends on their relationships to proteins with known 3D structure and their identification typically includes a sequence similarity search against a database of structurally characterized proteins. Close evolutionary relationships between proteins, e.g. family relationships, are usually detectable with sequence search methods such as BLASTP or FASTA. At the superfamily level, most of the distant relationships are detectable using iterative PSI-BLAST, hidden Markov models or profile–profile searches . These preliminary classification steps are very similar to the initial steps for the identification of templates in template-based protein structure modelling (also known as homology or comparative modelling) [9–12], which usually begin by searching a sequence database of proteins with known 3D structures using the target sequence as a query. Once a suitable template is selected, all current methods create an alignment of the target and template sequences and this alignment is further used as input to build a 3D model for the target protein. Template-based methods rely on two important assumptions: that proteins fold into one stable folded structure and that homologous proteins fold into similar structures. Current methods can produce reliable and accurate protein structure models when suitable templates are selected and the degree of structural conservation between the full length target and template protein is substantial.
Since the SCOP database was established in 1995, the amount of structural data has grown nearly 40-fold. The classification protocol has changed over this time, allowing better evaluation of sequence–structure relationships for classified proteins and the quality of alignments produced by different sequence comparison algorithms . Numerous exceptions observed to topological rules, along with complex evolutionary scenarios and unusual protein features prompted the development of SCOP2, a successor of the SCOP database . Here, I review selected cases of structural variations and peculiarities in individual proteins and among group of homologues. Knowledge of these cases may be of use in essential steps of protein structure modelling such as the selection of structurally and biologically relevant templates or for improving the target-template sequence alignments by considering evolutionary information about the structural variations of both the target and template proteins.
Conformational transitions in proteins
Conformational changes in proteins have been known for a long time and are crucial to many biological processes . These range from a subtle side-chain displacement or a loop flexibility to a large domain motion involving hinge regions that are not constrained by packing forces. In some proteins, short ‘chameleon sequences’ can undergo more dramatic changes and adopt alternative secondary structures. Chameleon sequences are more common to intrinsically disordered proteins , but they can also be found in globular domains. For example, in hypoxia-inducible factor prolyl hydroxylase 2 (PHD2), a region located in the active site vicinity (β2β3 loop) undergoes transition from an extended β to an irregular conformation upon binding to HIFα peptide (PDB 3HQR) . Similarly, upon oligomer formation, an α to β conformational transition is observed in the α-apical domain of the thermosome (Figure 1A) . These conformational changes, although quite dramatic, usually involve relatively short stretches of amino acid residues.
Side by side comparison of alternative conformers of: (A) α-apical domain of the thermosome: I) isolated domain (PDB 1ASS), II) domain from the closed thermosome (PDB 1A6E); the region that undergoes a secondary structural transition from α to β is indicated with a black arrow and coloured in orange in the secondary structure plot; (B) Mad2: I) O-Mad2 (PDB 1DUJ), II) I-Mad2 (PDB 3GMH, chain B), III) C-Mad2 (PDB 3GMH, chain E); the regions that undergo a structural change and a β-to-α transition are coloured in light blue and in orange respectively; (C) apolipoprotein A: I) lipid-free form (PDB 2A01), II) lipid-bound form (PDB 2MSD); (D) RfaH: I) closed form (PDB 2OUG), II) open form (PDB 2LCL). The secondary structure plots refer only to portions of each structure that are shown in particular colours (e.g. green in (A), grey in (B), blue in (C), red in (D)). All figures were prepared using Pymol (http://www.pymol.org).
Some proteins, however, undergo much larger structural rearrangements, leading to a conformational transition from one stable folded state to another. These so-called metamorphic proteins  exist in multiple conformations and undergo conformational transitions that involve a major rearrangement of both their secondary structural elements and their entire hydrogen bonding network, repacking of their interior and, in most known cases, exposure of a new binding interface. This new binding interface is usually associated with a new function that is exhibited by one of the conformers but not by the others, and hence the structural transitions observed in metamorphic proteins play an important role in their molecular function. One of the first known examples are the serpins, which upon proteolytic cleavage undergo irreversible structural changes associated with their inhibitory mechanism . Their close homologue ovalbumin (30% identity), for example, is not subject to similar conformational changes . More recently, several proteins have been shown to exist in an equilibrium of multiple conformational states and can reversibly change their structures. Mitotic arrest deficient 2 (Mad2) was first described to exist in three conformations, latent (open) O-Mad2, (intermediate) I-Mad2 and activated (closed) C-Mad2 [22,23]. In the latter, the C-terminal region refolds into an irregular structure, the so-called ‘safety belt’, and a β-hairpin that replaces the N-terminal strand in the O-Mad2 structure. The N-terminal strand shifts and undergoes a transition from β to α conformation in the activated conformer. The formation of the ‘safety belt’ is used to topologically entrap Mad2 binding partners containing the so-called MIM motif. In complex with its binding partners, C-Mad2 can recruit additional copies of O-Mad2 and convert them into an intermediate I-Mad2 that is a structural hybrid of the two conformers (Figure 1B). In contrast with Mad2, lymphotactin undergoes complete rearrangement of all stabilizing interactions in order to convert from a monomeric chemokine fold to a dimeric β-sandwich fold . The chemokine-like conformer binds to XCR1 GPCRs whereas the dimeric conformer lacks this ability, but instead it interacts with cell-surface glycosaminoglycans.
In many aspects chloride intracellular channel protein 1 (CLIC1) has the most complex scenario for structure and function transitions. CLIC1 is a chloride ion channel that exists as both a globular soluble and a transmembrane form. Soluble CLIC1 exists in equilibrium between monomeric and dimeric states. The monomeric form has a typical GST fold with N-terminal thioredoxin-like domain that undergoes a structural transition to an all α-helical conformation upon dimerization . This conformational switch results in the exposure of a large hydrophobic surface that contributes to the dimeric interface. Only the dimeric form can interact with membrane lipids. Upon binding to the lipid surface, the same N-terminal region becomes a transmembrane helix that penetrates the lipid bilayer and via self-association forms the channel pore .
Conformational transitions induced by a change of the environment are intrinsic features of some α-helical proteins. For example, upon contact with lipids apolipoprotein A undergoes a change from a four helical up-and-down bundle to a ring-like structure that wraps around the lipids (Figure 1C) . The lipid-free form of apolipoprotein is involved in various interactions with cellular receptors whereas the lipid-bound form is involved in a lipid transport. Similarly, saposins undergo conformational changes from closed monomeric to open dimeric form in the presence of lipids [28,29]. The death domain of protein kinase Pelle (Pelle-DD) adopts a six helical bundle in solution, characteristic of the death domain family, but in the presence of MPD (2-methyl-2,4-pentanediol), the structure of Pelle-DD refolds into a single helix .
A striking structural and functional transition is observed for the RfaH transcription factor, the C-terminal domain of which undergoes a transition from an α-helical hairpin to an SH3 β-barrel, converting it from a transcription into a translation factor (Figure 1D) . RfaH is a member of a conserved ubiquitous multigene family of transcription factors. The α-helical conformer masks the RNA polymerase binding interface in the N-terminal domain and this autoinhibition is essential to avoid functional interference with its paralogue, NusG. Both RfaH conformers are functionally active: the α-hairpin binds to the ribosome and activates translation whereas the β-barrel form has a function similar to NusG.
Many more examples of structural transitions are known, such as for fibronectin , T-cell receptor α , KaiB , etc. Little is known about the exact mechanisms that drive these conformational changes. The functional requirement for some proteins to form and maintain an accurate and specific active or binding site probably exerts a strong selective pressure to adopt only one stable folded structure. For other proteins, however, conformational transitions provide an elegant way of switching between different molecular functions. Our current state of knowledge about the large structural rearrangements of certain proteins does not have any predictive power but it has some important implications for protein structure modelling. Particularly, it is essential for the selection of relevant templates and in finding the structural conformer that is more suitable for modelling. Given that many methods use non-redundant sequence databases derived by using sequence similarity clustering, it is currently up to the user to identify the most appropriate template and its relevant conformer for a particular modelling problem.
Conservation of protein structure during evolution
Proteins are the evolutionary products of various molecular events operating at gene level such as point mutations, nonhomologous recombination, transposition, juxtaposition, exon rearrangement, gene or exon duplications, etc. Mutations of many amino acids in proteins do not affect or have only marginal effect on structure and stability. Therefore, unless there is a selective pressure for a conformational change, the structures of homologous proteins should be similar. Generally, proteins performing the same molecular function diverge with speciation of organisms and hence their structures tend to be more conserved than their sequences. An example is the structural conservation observed in the SCOP family of Sm-like proteins. These proteins fold into a partly open β-barrel and associate in hetero- or homoheptameric ring structures  that serve as platforms for versatile protein–protein and protein–RNA interactions. The requirement to maintain the oligomer symmetry that is essential for the protein function exerts a strong evolutionary pressure to maintain the 3D shape and, despite the low sequence similarity (10–30% sequence identity over 65 residues), all members have very similar structures (Figure 2A). The most conserved sequence features of this family are two Gly residues that play a role in maintaining the barrel curvature typical for all Sm-like proteins. At the level of ∼50% sequence identity, it is likely that proteins have very similar 3D structures. There are, however, exceptions to this rule and there are homologous proteins having very similar sequences but globally different structures. In the Cro family of repressors, for instance, Pfl6 and Xfaso1, share 45% sequence identity over 55 residues. Their structures retain the local structural similarity of the DNA binding motif at their N-termini but, despite a high sequence similarity, they adopt very different structures at their C-termini . In Xfaso1 this region is α-helical whereas in Pfl6 it folds into β-sheet stabilized by dimerization (Figure 2B).
(A) Superposition of Sm-proteins. Structures are shown in ribbon and coloured as follows: in yellow–Sm D1 (PDB 1B34, chain A), in green–Sm D2 (PDB 4PJO, chain D), in blue–Sm D3 (PDB 1D3B, chain A), in red–Sm B (PDB 1D3B, chain B), in black–Sm F (PDB 1N9R, chain A). A sequence logo showing the degree of amino acid conservation derived from the structure-based sequence alignment is shown below. (B) Side by side comparison of the structures of two Cro-proteins. I) Pfl6 (PDB 2PIJ) and II) Xfaso1 (PDB 3BD1); BLASTP pair-wise sequence alignment with 45% identity over 55 residues and one 5 residue gap; (C) fold decay event in the glutamate synthase central domain; the FMN-binding domain is shown in purple (PDB 1OFD, chain A, residues 840–1210) and the central domain in blue (PDB 1OFD, chain A, residues 490–735); structurally equivalent regions are shown in cartoon and the rest in ribbon. (D) Large insertion in an α-helix in the structures of two nonspecific endonucleases. I) Nuclease A from Anabaena sp. (PDB 1ZM8); II) Nuclease A from Streptococcus agalactiae (PDB 4QH0).
Events such as transposition, nonhomologous recombination, alternative splicing etc., can result in insertions or deletions and sometimes can significantly alter the structure of protein gene products. For example, the proteins belonging to the SCOP α/β hydrolase superfamily exhibit large deletions or insertions of secondary structural elements and even entire domains in order to accommodate different substrates. The common structural core of these homologous proteins, however, remains conserved, particularly near the active site and the nucleophile elbow motif (PDB 5AJH, 4J7A, 1THG, 3I2K). The evolutionary scenario with the glutamate synthase family is quite different: the FMN-binding domain was duplicated and fused and then the duplicated domain underwent a large deletion of three β/α units, resulting in an incomplete barrel (Figure 2C). Deletion events of this kind that affect the structural cores of homologous proteins are not uncommon. A similar event occurred in the structure of a nonfluorescent flavoprotein in which the remaining structural parts retain significant sequence similarity (36% identity) to its homologue, luciferase (PDB 1NFP, 1LUC). Insertions and deletions can also occur within secondary structural elements. Some members of the nonspecific endonucleases superfamily, for instance, contain a loop bisecting a long α-helix that borders the enzyme active site (Figure 2D). The length of this loop varies between 9 and 13 residues in different homologues but interestingly the conformation of the α-helix before and after the insertion does not deviate.
Other scenarios of protein structure evolution and structural changes in homologous protein families have been described elsewhere [37–42]. The knowledge of protein families, their conserved features and structural variations is a prerequisite for better quality model building. Human expertise is also essential to distant homology recognition and the modelling of homologous but structurally divergent proteins. Looking back in retrospect, two approaches in protein structure prediction, distant homology recognition in CASP2 (Critical Assessment of protein Structure Prediction) and hybrid template assemblies in CASP4, were pioneered by the main author of SCOP, Alexey Murzin. His detailed knowledge of protein structures allowed his successful prediction in CASP4 of a novel topology for target T0104, which still remains unique among the known P-loop containing proteins .
Using evolutionary information about the target and the template can be helpful to improve the quality of the target/template alignments or to define specific alignment constraints in template-based modelling. Evolutionary information, however, can sometimes introduce a bias and affect the performance of some secondary structure prediction methods. This can happen in multigene families where a particular structural feature has been lost in some lineages. For instance, secondary structure prediction methods that exploit evolutionary information fail to predict the second helix in the p53 tetramerization domain in bony fishes that is otherwise absent from other vertebrate p53 proteins .
Proteins with unusual topologies
Folding pathways of proteins tend to follow an energetically favourable route leading to a stable, low energy conformation. Several empirical rules were established during early analyses of protein structures, underlining basic topological principles and preferences [45–47]. Some of these postulated that secondary structural elements that are adjacent in sequence make a contact in three dimensions, that is, protein structures tend to have a low contact order . In order to fold into a stable globular structure, it was reasoned that α-helical and β-sheet secondary structure elements should associate tightly and pack closely to form a hydrophobic core of a protein. Topological features such as crossing loops and left-handed β–α–β connections were considered energetically unfavourable and very rare. Similarly, knots in the polypeptide chain were postulated as highly improbable due to a large entropic barrier to folding and the intrinsically difficult process of formation of knotted topology. Nowadays, exceptions to each of these rules have been observed (Figure 3). Some of these previously considered rare and improbable features appear to be characteristic of highly represented protein families. The superfamily of RNA methyltransferases containing a deep trefoil knot, for example, consists of numerous families, many members of which have been structurally characterized recently . Another example is the vast expanding superfamily of DinB/YfiT-like putative metalloenzymes that fold into high contact order structures and probably originated from an interlocked dimeric homologue (Figure 3E). Prediction of long range interactions in proteins still remains a difficult problem. Topological restraints in structure modelling are now increasingly being used in order to improve the prediction accuracy. Their stringency should be carefully considered or complex folds and knotted topologies may never be predicted.
(A) Loop crossing in the structure of DinI (PDB 1GHH); (B) trefoil knot in the MJ0366 structure (PDB 2EFV); (C) left-handed β–α–β connection in the structure of a protein with unknown function shew_3726 (PDB 2GPI); a black arrow indicates the location of the unusual topological feature; a schematic drawing of each feature is shown next to each structure for clarity, (D) structure of the hexadeca-haem cytochrome Hmc that does not possess a compact hydrophobic core (PDB 1GWS); (E) high contact order structure of DinB protein (PDB 2F22).
This review was an attempt to provide a brief, and very selective, overview of our current understanding of how proteins evolved and function, and give a hint of possible implications to structure modelling. It is noteworthy that although exceptions have been found for nearly every rule defined in the past, these do not disprove the rule. Many homologous proteins fold into similar structures and their structures are more conserved than their sequences. Importantly, every group of related proteins has its own evolutionary history and perhaps underwent events that may not be observed in other protein families. Evolutionary changes are not restricted to the peripheral elements of a protein domain but can also affect the structural core. Many proteins adopt a single, unique, well defined three-dimensional structure under native conditions. By contrast, others exist in multiple conformational states and hence provide new insights into how protein structures and functions can evolve through the process of conformational transitions.
I thank Cyrus Chothia and Alexey Murzin for valuable discussions, Gareth Bloomfield, Chris Johnson and Dave Howorth for reading of the manuscript and suggestions.
New Developments in Protein Structure Modelling for Biological and Clinical Research: Held at Charles Darwin House, London, U.K., 8 December 2015
Abbreviations: CASP, critical assessment of protein structure prediction; CLIC1, chloride intracellular channel protein 1; Mad2, mitotic arrest deficient 2; SCOP, Structural Classification of Proteins
This is an open access article published by Portland Press Limited on behalf of the Biochemical Society and distributed under the Creative Commons Attribution Licence 4.0 (CC BY).