Available online at www.sciencedirect.com

Cradle-loop barrels and the concept of metafolds in proteinclassification by natural descentVikram Alva1, Kristin K Koretke2, Murray Coles1 and Andrei N Lupas1

Current classification systems for protein structure show many

inconsistencies both within and between systems. The

metafold concept was introduced to identify fold similarities by

consensus and thus provide a more unified view of fold space.

Using cradle-loop barrels as an example, we propose to use

the metafold as the next hierarchical level above the fold,

encompassing a group of topologically related folds for which a

homologous relationship has been substantiated. We see this

as an important step on the way to a classification of proteins

by natural descent.

Addresses1 Department of Protein Evolution, Max-Planck-Institute for

Developmental Biology, Spemannstr. 35, D-72076 Tubingen, Germany2 Computational Chemistry Group, GlaxoSmithKline, Collegeville, PA

19426-0989, USA

Corresponding author: Lupas, Andrei N

(andrei.lupas@tuebingen.mpg.de)

Current Opinion in Structural Biology 2008, 18:358–365

This review comes from a themed issue on

Sequence and Topology

Edited by Nick Grishin and Sarah Teichmann

Available online 3rd May 2008

0959-440X/$ – see front matter

# 2008 Elsevier Ltd. All rights reserved.

DOI 10.1016/j.sbi.2008.02.006

IntroductionIn most proteins, folding is indispensable for activity. The

units of folding, that is the segments of a polypeptide

chain capable of folding independently, are called

domains. In most cases, folding occurs largely autonom-

ously (with some assistance from cellular folding factors),

but some domains will only fold if they receive an

appropriate signal from an adjacent domain or if they

encounter the correct scaffold, be this the membrane,

another protein, or a nucleic acid. Because of the central

importance of folding for the understanding the activity

of proteins, systems for classifying protein structures have

become an essential tool in molecular biology. These

systems differ in the way they are generated: while the

Dali Domain Dictionary is largely automated [1], relying

on the popular structure comparison program DALI,

CATH (Class-Architecture-Topology-Homology) com-

bines automated and manual processes [2] and SCOP

(Structural Classification of Proteins) is based mostly on

manual assignments [3]. SCOP particularly has become a

point of reference in the discussion of similarities and

Current Opinion in Structural Biology 2008, 18:358–365

dissimilarities between proteins. Despite showing a con-

siderable amount of agreement, each of these systems

offers its own view of fold space, with many differences

becoming apparent upon detailed, protein-by-protein

comparison [4,5]. In this review, we will discuss the

metafold concept as a method to achieve a more unified

view of fold space and propose modifications to this

concept, using cradle-loop barrels as an example.

The metafold conceptThe most important source of disagreement between

classification systems are differing domain definitions,

but in a fair number of cases, disagreements also arise

because the same domain is assigned to different folds. In

order to understand these problems, it is useful to con-

sider the definition of a fold as a conserved, topologically

distinct arrangement of secondary structures in a domain,

with extensions and insertions peripheral to the fold

treated as decorations. A fold change occurs when one

or more secondary structure elements within the fold alter

their nature and/or their topology. Clearly a wide latitude

in judgement is possible with respect to domain bound-

aries, to what constitutes a decoration, and to the degree

of topological change necessary to separate structures into

different folds. To alleviate the discrepancies arising from

subjective estimates of fold similarities and differences,

Daggett and coworkers introduced the metafold concept

as a consensus method designed to reveal fold similarities

relatively independently of the methods used to compare

protein domains [5]. In essence, domains are considered

part of the same metafold if their topological similarity is

recognized by multiple classification systems.

Although the metafold concept is clearly helpful in

obtaining a more unified view of fold space, we think

that its usefulness is limited by the fact that it does not

address a fundamental source of tension within and

between classification systems, namely the coexistence

of homologous and analogous classification criteria. Both

SCOP and CATH explicitly use homology for classifi-

cation, SCOP in the first two hierarchical levels (family

and superfamily) and CATH in the first level. Clearly,

what one percieves as homologous has a profound influ-

ence on drawing domain boundaries and on labelling

parts of the structure as decorations to a conserved core.

These decisions, which have a substantial arbitrary com-

ponent, by necessity shape the higher levels of the

classifications. Furthermore, by placing homology at

the base and topological considerations into the upper

layers, both systems make the implicit assumption that

homologous proteins always have the same fold. As has

www.sciencedirect.com

The cradle-loop barrel metafold Alva et al. 359

become increasingly clear in recent years, this is not the

case and multiple events have been described that can

lead to fold change in evolution: point mutations, indels,

topological substitutions, circular permutations, strand

swaps, strand and hairpin invasions, 3D domain swaps,

and chimeric fusions [6,7,8�,9�,10]. In such cases,

homology typically trumps topological dissimilarity,

leading to the grouping of homologous proteins into

the same fold, even when they have undergone changes

that would cause analogous proteins to be classified into

separate folds.

In order to address the tensions arising in classification

systems from fold changes between homologous proteins,

we think that the metafold would more usefully be viewed

as the next hierarchical level above the fold, bringing

together groups of topologically similar folds for whom a

homologous relationship has been substantiated in at least

one case. This core group of member folds could be

usefully expanded by a periphery of candidate folds, which

are related to the core members by a topological change

known to occur between homologous proteins, albeit with-

out homology having been substantiated in that particular

case. Proteins known to be analogous to members of the

core group should be excluded from the metafold. Thus,

topologically similar proteins may end up in different

metafolds, based on evidence of their descent. In such

cases, these proteins will also need to be assigned to

different folds. Although this may seem to subvert the

concept of fold, we note that this concept has always had a

large arbitrary component and that some larger folds are

currently subdivided without easily recognizable reasons.

The goal of our proposal is to ultimately eliminate all

analogous criteria from protein classifications (more on this

subject at the end of this paper).

Although more complicated and also fuzzier than the

definition of Daggett and coworkers, our metafold defi-

nition offers several advantages: it addresses the

homology–analogy problem directly, it has considerably

greater explanatory power with regard to the fold space it

maps, it acknowledges explicitly that the classification is

more robust and better supported in some cases (the core

groups) than in others (the peripheries), and it provides a

first step toward a comprehensive classification of proteins

by descent.

In the following, we would like to explain our ideas on

metafolds using the example of cradle-loop barrels, which

we have studied in detail for almost a decade.

The cradle-loop barrel core groupOur interest in the cradle-loop barrels began with the

structure determination of VatN, the substrate recog-

nition domain of the AAA protein VAT [11]. The N-

terminal part of this domain (VatN-N) forms one of the

most topologically complex folds known, consisting of

www.sciencedirect.com

duplicated bbab units that are completely interdigi-

tated to form a six-stranded barrel. The b1–b2 loops of

each unit cross over the symmetry-related b20 strand,

giving the fold an unusual, knotted appearance

(Figure 1) and its name: the double-psi b-barrel [12].

In each bbab unit, the helix is connected to the last

strand by a conspicuous Gly-Asp motif, which we named

the GD box (Figure 2).

In order to understand how such a complex fold could

have evolved, we undertook an extensive bioinformatics

and structural study with the aim of identifying possible

precursors with simpler topologies. We focussed on sev-

eral groups of proteins that showed sequence similarity to

VatN-N. Of particular interest were dimeric bacterial

transcription factors, typified by Bacillus subtilis AbrB,

whose N-domains carried a single GD box and were

elaborated by an additional C-terminal b-strand. They

thus appeared to represent a homodimeric precursor to

the double-psi barrel, with a permuted fold that would

resolve the topological complexities of the latter. This led

us to propose this topology as the ancestral form [11]. Our

hypothesis was contradicted by the published structure of

AbrB-N (1EKT), which, instead of a barrel, showed a

side-by-side dimer of two three-stranded b-meanders

with two equatorial helices.

In order to understand this contradiction, we investigated

a group of sequences that were intermediate between

VatN-N and AbrB. Proteins of this group, typified by

PhS018 from Pyrococcus horikoshii, either have internal

sequence symmetry and carry two copies of the GD

box or are homodimers with one copy per subunit.

PhS018 turned out to have yet a third fold, forming a

singly interdigitated, six-stranded barrel (Figure 1) [13�].We named this topology the RIFT barrel for its wide-

spread occurrence in ancient proteins, such as the ribo-

somal protein L3, the N-domain of the F1 ATPase, and

translation factors of the Ef-Tu family. This topology was

clearly related to that of double-psi barrels by a strand

swap of the symmetry-related b2/b20 strands, but was not

visibly related to the published AbrB fold, 1EKT. In

particular, the conserved GD boxes resembled closely

those of VatN-N (Figure 2), but had an entirely different

conformation from 1EKT.

Given this discrepancy, we decided to redetermine the

structure of AbrB-N, which turned out to neither

resemble the published structure (which was sub-

sequently retracted), nor a permuted form of the

double-psi barrel. Rather, additional C-terminal strands

were inserted into the RIFT barrel to form an eight-

stranded architecture with two pairs of interdigitated b-

hairpins (Figure 1), leading us to name this fold the

swapped–hairpin barrel [14]. Significantly, the GD-box

region now resembled the equivalent regions closely in

double-psi and RIFT barrels (Figure 2).

Current Opinion in Structural Biology 2008, 18:358–365

360 Sequence and Topology

Figure 1

Topological map of the cradle-loop barrel metafold. In all structures, b-strands are colored in green, cradle loops in black, helices in yellow,

b-strands inserted into the barrel in blue, and elements not part of the barrel in gray. The b2 and b20 strands in the RIFT and double-psi barrels

are shown in red and pink, respectively. In homodimeric structures, monomers are distinguished by light and dark colors. Homologous connections,

as substantiated by HMM–HMM comparisons [17], are indicated by red arrows. The structures shown are: (I) swapped-hairpin barrels: AbrB

Current Opinion in Structural Biology 2008, 18:358–365 www.sciencedirect.com

The cradle-loop barrel metafold Alva et al. 361

Figure 2

The GD box (reproduced with permission from [13�]). (a) Superposition of the conserved bab element from PhS018 (2GLW, both repeats), VatN-N

(1CZ4, both repeats), AbrB-N (1YFB), and the addiction antidote MazE (1MVF). A difference in the paths of the b2/b20 strands (red) distinguishes the

double-psi barrel of VatN-N. (b) Detailed view of the GD box. The left view shows a superimposition of the GD boxes of the proteins in panel (a),

showing the neighboring secondary structure elements and the side chains of the conserved aspartic acid residues. The right view shows the first GD

box of PhS018 and its hydrogen-bonding network.

Our results on VatN-N, PhS018 and AbrB-N confronted

us with a problem of nomenclature, as the three

barrels were clearly homologous, but equally clearly

had different folds. In search of a term that would

describe their relationship, we chose to define them

as a metafold and denote them as cradle-loop barrels for

the distinctive profile conferred by their characteristic

b1–b2 loops.

Thus, in our evolutionary scenario, an ancestral, homo-

dimeric RIFT barrel gave rise to swapped-hairpin bar-

rels by strand invasion and to double-psi barrels by

fusion and strand swapping. We also found sequence

similarity indicative of homology between RIFT bar-

rels and yet a fourth fold, the C-terminal domain of

bacterial fluorinating enzyme: this is related to the

RIFT topology by a strand invasion from the second

cradle loop into the space between b10 and b20, yielding

a seven-stranded barrel. Jointly, these proteins map out

a network of homologous but topologically distinct

folds (Figure 1).

The cradle-loop barrel peripheryWe placed the RIFT barrel at the center of the cradle-

loop network, because of its simpler topology and its

occurrence in a wide range of ancient proteins. We

included all proteins with a RIFT barrel fold as candidate

(Figure 1 Legend Continued ) (homodimeric barrel, 1YFB) and MraZ (mono

2GLW) and the homology model of MTpME2200 Orf5 based on PhS018 (ho

b6 to N-terminus, 1QY6), and B3 barrel (1WID); and (III) double-psi barrels:

(1NH0), the C-terminal domain of bacterial fluorinating enzyme (1RQP), and

www.sciencedirect.com

groups, even though we initially had no evidence for their

homologous relationship to the proteins we had used to

define the metafold. We did this in order to map out the

possible relationships that would be explored next most

usefully. Recently, we substantiated a homologous

relationship between the core group and one candidate

group, the riboflavin kinases (SCOP b.43.5, CATH

2.40.30, see Table 1), by identifying a family of archaeal

proteins that bridge the evolutionary space between the

two. These proteins are similar in sequence to both core

and candidate group, and have CTP-dependent riboflavin

kinase activity [15��].

In order to obtain a more complete view of the potential

homologous protein space, we think that candidate

groups would also usefully include proteins that have

at most one topological change relative to the core group,

provided that this change is known to occur between

homologous proteins. Our current list of cradle-loop bar-

rels includes a series of candidate topologies, related in

this way to the three core topologies: RIFT, swapped-

hairpin and double-psi (Table 1).

For the RIFT barrel, two variants are obtained by circular

permutation and three by the insertion of an additional

strand into the barrel. The elongation factor/aminomethyl-

transferase common domain (SCOP: b.44) is formed by the

meric barrel, 1N0E); (II) RIFT barrels: PhS018 (monomeric barrel,

modimeric barrel), EF-Tu b1 to C-terminus, 1D2E), V8 protease

VatN-N (1CZ4), the C-terminal domain of AstA (1YLE), HIV-1 protease

the PK barrel (1A49). Details of the structures are given in Table 1.

Current Opinion in Structural Biology 2008, 18:358–365

362 Sequence and Topology

Table 1

The cradle-loop barrel metafold

SCOP code Superfamilya CATH code Protein, PDB accession, remarks

RIFT barrel proteins

Homodimeric barrel

– Af2212 from Archaeoglobus fulgidus – Af2212, 2NWT chains A and B, single GD, missing a-helix

Monomeric barrel

– PhS018 from Pyrococcus horikoshii * – PhS018, 2GLW, both GD

b.43.2 FucI/AraA C-terminal domain-like 3.20.14 L-Fucose isomerase, 1FUI chain A res 356–591, insert in both

cradle loops

L-Arabinose isomerase, 2AJT chain A res 329–498

b.43.3 Translation proteins+ 2.40.30 EF-Tu, 1EFC, chain A res 205–296, C-term GD, a1 replaced

with b-strand

Ribosomal protein L3, 1JJ2, chain B res 40–78, 189–206,

260–314, both GD, insert in cradle-loop 1, a1 replaced with

b-strand

Ribosomal protein L35ae, 1SQR chain A, missing a2

RimM, 2F1L chain A res 7–95

PF1791, 2EY4 chain C res 1–73, both GD, missing a1 and a2

b.43.4 Riboflavin synthase domain-like* 2.40.30 Riboflavin synthase, 1I8D chain A res 6–92, both GD

Ferredoxin reductase, 1GAW chain A res 36–156, both GD,

missing a1

NADPH-cytchr reduct, 1JA1 chain A res 281–324, 452–518,

N-term GD, missing a1

b.43.5 Riboflavin kinase-like* 2.40.30 Riboflavin kinase, 1N08, missing a2

b.49.1 N-domain of F1 ATP synthase, a and b subunits 2.40.30 F1-ATPase b subunit, 1W0J chain A res 28–91, both GD,

missing a1

b.49.2 Alanine racemase C-terminal domain-like* 2.40.37 Alanine racemase, 1BD0 chain A res 243–336, both GD

ODC, 7ODC chain A res 2–43, 284–418, C-term GD

b.49.3 Aminopeptidase/glucanase lid domain 2.40.30 YsdC, 1VHE chain A res 73–163, C-term GD

b.129.2 PG0164-like* – PG0164, 2D9R chain A res 20–104, C-term GD

e.56.1 YaeB-like – HI0510, 1XQB chain A res 1–63, 99–136, C-term GD

f.46.1 HlyD-like secretion proteins* – MexA, 1VF7 chain A res 26–37, 173–226, C-term GD

b1 circularly permuted to C-terminus

b.44.1 EF-Tu/eEF-1a/eIF2-g C-terminal domain 2.40.30 EF-Tu, 1D2E chain A res 349–451, both GD, missing a1 and a2

b.44.2 Aminomethyltransferase b-barrel domain+ – GcvT, 1WOS chain A res 279–361, both GD, missing a1 and a2

b6 circularly permuted to N-terminus

b.45.1 FMN-binding split barrel 2.30.110 FMN bdg protein, 1FLM chain A

FMN-bdg protein MTH152, 1EJE chain A

UbiD, 2IDB chain A res 6–325

b.45.2 PilZ domain-like+ – Hypothetical protein VCA0042, 1YLN chain A res 138–248,

N-term GD

Hypothetical protein VCA0042,1YLN chain A res 23–137,

N-term GD

b.47.1 Trypsin-like serine proteases 2.40.10 V8 protease, 1QY6 chain A, two tandem barrels

Trypsin, 1HJ9 chain A, two tandem barrels

NSP4 proteinase, 1MBM chain A, two tandem barrels

3C cys protease, 1CQQ chain A, two tandem barrels

b.106.1 Phage tail proteins+ 2.40.10 Tail attachment protein gpF3, 1K0H chain A, missing a1 and a2

Baseplate structural protein GP27, 1K28 chain D res 4–120

b.140.1 Replicase NSP9 – Replicase NSP9, 1QZ8 chain A, missing a1

e.53.1 QueA-like – QueA, 1VKY chain A res 64–142, N-term GD, missing a1

b-strand inserted from cradle-loop 1

b.58.1 PK b-barrel domain-like+ 2.40.33 Pyruvate kinase, 1A49 chain A res 116–217, both GD,

missing a1 and a2

YuaD, 1ORU chain A, both GD, insert in both cradle

loops, missing a2

b-strand inserted from cradle-loop 2

b.141.1 Bacterial fluorinating enzyme, C-domain* – 50-Fluoro-50-deoxyadenosine synthase, 1RQP chain A

res 193–298, both GD

Additional C-terminal b-strand

b.142.1 DNA-binding pseudobarrel domain 2.40.330 DNA-bdg protein RAV1, 1WID chain A, C-term GD

EcoRII, 1NA6 chain A res 4–178, C-term GD

Current Opinion in Structural Biology 2008, 18:358–365 www.sciencedirect.com

The cradle-loop barrel metafold Alva et al. 363

Table 1 (Continued )

SCOP code Superfamilya CATH code Protein, PDB accession, remarks

Swapped-hairpin barrel proteins

Homodimeric barrel

b.129.1 AbrB/MazE/MraZ-like* 2.10.260 MazE, 1MVF chains D and E, GD

AbrB, 1YFB chains A and B, GD

Monomeric barrel

b.129.1 AbrB/MazE/MraZ-like* – MraZ, 1N0E chain A, both GD

Double-psi barrel proteins

b.52.1 Barwin-like endoglucanases 2.40.40 Eng V, 2ENG, insert in cradle-loop 1

Pollen allergen PHL P 1, 1N10 chain A res 1003–1145

Barwin, 1BW3, insert in both cradle loops

b.52.2 ADC-like* 2.40.40 Pyruvoyl dep. ADC, 1PPY chain A, both GD

DMSO reductase, 1EU1 chain A res 626–780, N-term GD,

insert in cradle-loop 1

VAT-N, 1CZ4 chain A res 1–91, both GD

e.29.1 b and b0 subunits of DNA dept RNA-pol 2.40.270 DD RNAP b, 1SMY chain C res 668–698, 832–1004

DD RNAP b0, 1SMY chain D res 625–740, C-term GD

b6 circularly permuted to N-terminus

b.50.1 Acid proteases 2.40.70 HIV-1 protease, 1NH0 chain A, a1 replaced by two extended

b strands

Endothiapepsin, 1OEW chain A res 10–140

b2 and a1 deleted

d.108.1 Acyl-CoA N-acyltransferases* – AstA, 1YLE chain A res 273–340, C-term GD

a Homology between superfamilies was evaluated with HHsearch [17]. Superfamilies showing sequence similarity indicative of homology fall into two

separate networks, marked with (*) and (+), as described in the text. We currently have no evidence for homology between these two networks.

circular permutation of b6 to the N-terminus and the

FMN-binding split barrel (b.45), trypsin-like serine pro-

teases (b.47), replicase NSP9 (b.104), phage tail proteins

(b.106), and the QueA-like barrel (e.53) by the circular

permutation of b1 to the C-terminus. In the PK b-barrel

domain (b.58), the barrel is elaborated by the insertion of a

b-strand originating from the first cradle loop, and in the

aforementioned C-terminal domain of bacterial fluorinat-

ing enzyme (b.141) by the insertion of a b-strand from the

second cradle loop. In the DNA-binding pseudobarrel

(b.142), the barrel is extended by the insertion of an

additional C-terminal strand. This barrel might have been

formed either by strand insertion, or by the fusion of two

different half barrels, the RIFT monomer and the

swapped-hairpin monomer (Figure 1). In the RIFT fold

space, we are observing the emergence of a second hom-

ologous network, based on translation proteins (b.43.3) and

including members of three variant RIFT folds (b.44.2,

b.45.2, b.106.1, b.58.1), as listed in Table 1. We have as yet

no evidence for the homology of this network to our core

cradle-loop network.

For the double-psi barrel, one variant arises by circular

permutation of b6 to the N-terminus, as seen in acid

proteases (b.50), and one by the deletion of b2 and a2, as

seen in acyl-CoA N-acyltransferases (d.108). Although the

acyltransferases have clear sequence similarity to double-

psi barrels, aspartic proteases do not. Nevertheless, their

inherent chaperone-like activity may point to a distant

evolutionary connection with the double-psi barrel

domains of AAA-ATPases [16].

www.sciencedirect.com

We do not currently have fold variants for the swapped-

hairpin barrels, but we note that some monomeric

proteins of this group have lost strand b10 [14], pointing

to further topological variants in the network.

OutlookThe combination of sequence and fold similarity in

generating current protein classifications, that is of hom-

ologous and analogous criteria, introduces contradictions

between and within systems. We see the metafold as a

useful concept for addressing one source of contradic-

tions, namely homologous fold change. Although prim-

arily sequence-driven and thus based on homology, the

metafold concept still uses fold similarity in order to

identify candidate folds. Ultimately, however, contradic-

tions can only be addressed comprehensively by elim-

inating all analogous classification criteria. At that point,

we would obtain a classification of proteins by natural

descent, conceptually related to the Linnaean system for

organisms, albeit with multiple roots, as proteins are not

monophyletic. We think that such a system is highly

desirable, as homology offers a rich source of structural,

functional and mechanistic information, while analogy is

comparatively uninformative and often misleading.

Indeed the central role of model systems in modern

biology can only be understood in terms of extrapolation

by homology, since few researchers would be interested

in Danio, Drosophila or Caenorhabditis for their own sake.

If a classification by natural descent is so desirable, why

has analogy played such a great role in all classification

Current Opinion in Structural Biology 2008, 18:358–365

364 Sequence and Topology

efforts so far? We would argue that this was by default, as

homologous criteria were not available: sequence search

methods were not sufficiently developed to reveal remote

homology, sequence databases were too sparse to allow

for efficient connections in sequence space, and there

were too few structures to validate distant relationships.

Also, proteins are not monophyletic, so that — unlike in

the Linnaean system — analogy was the default assump-

tion for observed similarities.

This situation is changing. With the emergence of profile

search methods and, more recently, with methods based

on the comparison of profile Hidden Markov Models [17],

bioinformatic tools have reached considerable sensitivity.

Sequence databases have been growing fairly steadily by

about one order of magnitude every 5 years [18], and

currently contain about 5 � 106 proteins. Given a global

proteome of about one trillion proteins (�108 species with

�104 protein-coding genes each), at current rates we

might know the sequence of most proteins on Earth in

about a quarter century. Of possibly greater immediate

impact, there are now some 700 complete genomes from

all over the tree of life, also growing by about one order of

magnitude every 5 years [19]. In parallel, the number of

structures known to atomic resolution, currently at

5 � 104, has been growing steadily, if more slowly, by

about one order of magnitude every 9 years [20] (the

number of ‘non-redundant’ structures with at most 30%

pairwise sequence identity is about one-fifth this size).

Given that we currently recognize �105 protein families

[21] and this number is unlikely to rise by more than at

most one order of magnitude (if indeed it will rise at all), it

seems likely that most protein families will have at least

one member of known structure within the next 10–20

years [22�]. By targeting proteins from a wide range of

families that have remained unexplored, without regard

to the availability of functional information, structural

genomics initiatives are playing a key role in this effort

[23]. The wide availability of sequence and structure data

for most families is essential for substantiating remote

homology, since sequence similarity as a function of

structure similarity is a powerful discriminator between

homology and analogy (M Remmert et al., unpublished).

We therefore wish to argue that it has become possible to

start removing analogous criteria across protein classifi-

cations and move toward a system based on natural

descent.

In recent months, structural genomics initiatives have

been discussed critically in a string of papers appearing in

the main structural biology journals. One criticism has

been that their goal of covering fold space is not useful, as

fold space is structurally continuous in the sense that most

folds consist of recurring subdomain-sized fragments

[24,25] and attempts at structure classification are there-

fore ultimately futile [26]. We disagree with this argu-

ment because, while structure space may be

Current Opinion in Structural Biology 2008, 18:358–365

geometrically continuous, it is evolutionarily discontinu-

ous [10]. Natural descent therefore provides a robust basis

for protein classification. The primary goal of structural

genomics is not to chart hitherto unknown and in most

cases peripheral areas of fold space, it is to map structure

space onto sequence space, allowing us to merge the

currently separate sequence-based and structure-based

approaches into one comprehensive protein classification.

AcknowledgementThis work was supported by institutional funds from the Max PlanckSociety.

References and recommended readingPapers of particular interest, published within the annual period ofreview, have been highlighted as:

� of special interest�� of outstanding interest

1. Dietmann S, Park J, Notredame C, Heger A, Lappe M, Holm L: Afully automatic evolutionary classification of protein folds: DaliDomain Dictionary version 3. Nucleic Acids Res 2001, 29:55-57.

2. Greene LH, Lewis TE, Addou S, Cuff A, Dallman T, Dibley M,Redfern O, Pearl F, Nambudiry R, Reid A et al.: The CATH domainstructure database: new protocols and classification levelsgive a more comprehensive resource for exploring evolution.Nucleic Acids Res 2007, 35:D291-D297.

3. Andreeva A, Howorth D, Chandonia JM, Brenner SE, Hubbard TJ,Chothia C, Murzin AG: Data growth and its impact on the SCOPdatabase: new developments. Nucleic Acids Res 2008,36:D419-D425.

4. Hadley C, Jones DT: A systematic comparison of proteinstructure classifications: SCOP, CATH and FSSP. Structure1999, 7:1099-1112.

5. Day R, Beck DA, Armen RS, Daggett V: A consensus view of foldspace: combining SCOP, CATH, and the Dali DomainDictionary. Protein Sci 2003, 12:2150-2160.

6. Grishin NV: Fold change in evolution of protein structures.J Struct Biol 2001, 134:167-185.

7. Kinch LN, Grishin NV: Evolution of protein structures andfunctions. Curr Opin Struct Biol 2002, 12:400-408.

8.�

Andreeva A, Murzin AG: Evolution of protein fold in thepresence of functional constraints. Curr Opin Struct Biol 2006,16:399-408.

This article along with Ref. [9�] describe examples of homologous foldchange by a range of different mechanisms. The level of structural insightdisplayed in both papers is impressive.

9.�

Andreeva A, Prlic A, Hubbard TJ, Murzin AG: SISYPHUS —structural alignments for proteins with non-trivialrelationships. Nucleic Acids Res 2007, 35:D253-D259.

Like Ref. [8�], this article presents examples of homologous fold change.In addition, it introduces a database of manually curated structuralalignments, which allows users to explore the fold changes interactively.

10. Lupas AN, Koretke KK: Evolution of protein folds. InComputational Structural Biology. Edited by Peitsch M, SchwedeT. World Scientific Publishing Co.; 2008.

11. Coles M, Diercks T, Liermann J, Groger A, Rockel B,Baumeister W, Koretke KK, Lupas A, Peters J, Kessler H: Thesolution structure of VAT-N reveals a ’missing link’ in theevolution of complex enzymes from a simplebetaalphabetabeta element. Curr Biol 1999, 9:1158-1168.

12. Castillo RM, Mizuguchi K, Dhanaraj V, Albert A, Blundell TL,Murzin AG: A six-stranded double-psi beta barrel is shared byseveral protein superfamilies. Structure 1999, 7:227-236.

13.�

Coles M, Hulko M, Djuranovic S, Truffault V, Koretke K,Martin J, Lupas AN: Common evolutionary origin of

www.sciencedirect.com

The cradle-loop barrel metafold Alva et al. 365

swapped-hairpin and double-psi beta barrels. Structure 2006,14:1489-1498.

This paper establishes the homologous relationship between DNA-bind-ing swapped-hairpin barrels and double-psi barrels with chaperoneactivity via proteins with an intermediate topology, which are most likelyalso DNA binding.

14. Coles M, Djuranovic S, Soding J, Frickey T, Koretke K, Truffault V,Martin J, Lupas AN: AbrB-like transcription factors assume aswapped hairpin fold that is evolutionarily related to double-psi beta barrels. Structure 2005, 13:919-928.

15.��

Ammelburg M, Hartmann MD, Djuranovic S, Alva V, Koretke KK,Martin J, Sauer G, Truffault V, Zeth K, Lupas AN et al.: A CTP-dependent archaeal riboflavin kinase forms a bridge in theevolution of cradle-loop barrels. Structure 2007, 15:1577-1590.

This paper makes two important contributions. It provides the firstcharacterization of an archaeal riboflavin kinase and it shows thatarchaeal riboflavin kinases are intermediate in sequence betweenbasal, DNA-binding cradle-loop barrels and bacterial and eukaryoticriboflavin kinases, providing an evolutionary bridge. The paper makespredictions about the sequence of mutations necessary to transform aDNA-binding protein into an enzyme, first bringing about CTP bindingand subsequently conferring the ability to transfer the g-phosphate toriboflavin.

16. Hulko M, Lupas AN, Martin J: Inherent chaperone-like activity ofaspartic proteases reveals a distant evolutionary relation todouble-psi barrel domains of AAA-ATPases. Protein Sci 2007,16:644-653.

17. Soding J: Protein homology detection by HMM-HMMcomparison. Bioinformatics 2005, 21:951-960.

www.sciencedirect.com

18. The GenBank entry in Wikipedia. http://en.wikipedia.org/wiki/GenBank.

19. The Genomes OnLine Database v 2.0. http://www.genomesonline.org/gold_statistics.htm.

20. The PDB Statistics hyperlink on the Protein Data Bank home page.http://www.rcsb.org/pdb/statistics/contentGrowthChart.do?content=total&seqid=100.

21. The InterPro database. http://www.ebi.ac.uk/interpro/.

22.�

Marsden RL, Lewis TA, Orengo CA: Towards a comprehensivestructural coverage of completed genomes: a structuralgenomics viewpoint. BMC Bioinformatics 2007, 8:86.

Structural genomics approaches are essential for a comprehensivestructural characterization of proteins. This paper presents an excellentdescription of the current structural coverage of protein families anddiscusses further developments needed to bring this effort to a success-ful conclusion.

23. Burley SK, Bonanno JB: Structuring the universe of proteins.Annu Rev Genomics Hum Genet 2002, 3:243-262.

24. Taylor WR: Evolutionary transitions in protein fold space. CurrOpin Struct Biol 2007, 17:354-361.

25. Kolodny R, Petrey D, Honig B: Protein structure comparison:implications for the nature of ‘fold space’, and structure andfunction prediction. Curr Opin Struct Biol 2006, 16:393-398.

26. Honig B: Protein structure space is much more than the sum ofits folds. Nat Struct Mol Biol 2007, 14:458.

Current Opinion in Structural Biology 2008, 18:358–365