The distributions of CATH domain structures in superkingdoms resemble those of SCOP structures and reveal a consistent evolutionary bias

Domain structures are unevenly distributed in the world of proteins and proteomes [1]. They distribute differently in superkingdoms Archaea (A), Bacteria (B) and Eukarya (E) and can be pooled into seven taxonomical groups depending on whether they are unique to a superkingdom (A, B and E) or are shared by two (AB, AE and BE) or three superkingdoms (ABE). The taxonomical groups can be visualized in a simple Venn diagram (Figure 2). Bias in the relative number of domains structures corresponding to each taxonomical group persists regardless of the classification used (CATH or SCOP) or the level of structural abstraction of the classification scheme (Figure 2). This bias cannot be attributed to non-vertical patterns of inheritance (e.g. the effect of horizontal transfer) since research groups have confirmed independently that convergent evolution is relatively rare (∼2–12%) at these high levels of structural conservation (e.g., [17], [18]).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Distribution of CATH domain structures among taxonomical groups of domain distribution in superkingdoms.

The percentage of domain structures shared by superkingdoms was considered as coarse estimate of evolutionary conservation of the hierarchical levels of classification. CATH domain censuses were derived from the present study. SCOP values were taken directly from published data [16], [18] and involve 1030 Fs, 1740 FSF and 2,397 FF defined by SCOP v. 1.73.

https://doi.org/10.1371/journal.pcbi.1003009.g002

Distribution biases among taxonomical groups show some striking features. First and as expected, higher taxonomical levels show higher levels of structural sharing between superkingdoms (especially ABE) than lower taxonomical levels, confirming the contention that they are evolutionarily more conserved. Second, the highest level of structural abstraction (CATH A) does not contain a single superkingdom-specific taxonomic group, suggesting that these groups represent sets of structures that are late evolutionary additions to the protein repertoire. Finally, ABE and BE domain structures are consistently the dominant taxonomic groups at all hierarchical levels, from FF to A. This final observation suggests they represent the most ancestral and common taxonomical groups. The most parsimonious corollary of these evolutionary patterns of domain distribution is that the ancient BE taxonomical group must arise by loss of archaeal-specific domain structures, suggesting Archaea is the most ancient superkingdom. As we will now show, this suggestion can be confirmed by phylogenomic reconstruction.

Structural chronologies of CATH domain structures uncover patterns of proteome diversification

We generated phylogenomic trees describing the phylogenetic relationship of 38 A, 1,152 T and 2,221 H domain structures (Figures 3 and 4). Tree distribution profiles and metrics of skewness suggest significant cladistic support (P<0.01). The trees were well resolved. However, internal branches for trees of Hs and Ts were poorly supported by bootstrap analysis, an expected outcome with trees of this size. Chronologies of evolutionary appearance [7] of CATH domain structures were derived directly from the phylogenomic reconstructions. The relative age of domains (nd) was measured on the trees as a relative distance in nodes from the hypothetical ancestor of domains at the base of the trees, and used to build the timelines. Since our method produces rooted trees that are highly unbalanced and reject the Yule and random speciation models [19] and since molecular speciation in our trees has clock-like behavior and does not depend on changes in domain abundance [20], nd was considered a good and most-parsimonious proxy for time. To study how domain structures distribute in proteomes, we calculated a distribution index (f), the number of species that use each structure given on a relative 0–1 scale. The f index was plotted along the timelines of domain structures, i.e. against nd (Figure 5). Three As (nd_A = 0–0.068), fifteen Ts (nd_T = 0–0.061) and fifteen Hs (nd_H = 0–0.049) were present in all proteomes examined (f = 1) and were the most ancient in the timeline. A list of the fifteen Hs is given in Table S1. The f of As decreased with increasing age. The f of Ts and Hs decreased with their increasing age until f approached zero at nd_T = 0.55 and at nd_H = 0.55, respectively. We term these ages “crystallization points” of the T and H structural chronologies, borrowing the idea of a phase transition from physics. At these time points, a steady decrease in f results in a large number of structures being specific to a small number of organisms. After crystallization, an opposite trend takes place, in which Ts and Hs increase their representation in genomes. In contrast, the architectural chronology that describes the appearance of As remained unaffected by the crystallization event since the losing trend of As started at nd_A = 0.56–0.60 but rarely reached zero (Figure 5).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Phylogenomic tree of CATH A domain structures.

Optimal (P<0.01) most parsimonious A (26,323 steps; CI = 0.3738, RI = 0.7655; g1 = −0.427) tree was reconstructed from a protein domain census in 492 completely sequenced genomes. The phylogeny was plotted into circular tree diagram and cartoon representations of the core structures labeled with each CATH id were mapped onto the leaves of the tree. The Venn diagram shows the diversity of A in the three superkingdoms, Archaea, Bacteria and Eukarya.

https://doi.org/10.1371/journal.pcbi.1003009.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. Phylogenomic trees of CATH T (A) and H (B) domain structures.

Optimal (P<0.01) most-parsimonious T (392,769 steps; CI = 0.0251, RI = 0.7488; g1 = −0.169) and H (658,425 steps; CI = 0.0149, RI = 0.7444; g1 = −0.144) trees were reconstructed from a protein domain census in 492 completely sequenced genomes. The phylogenies reconstructed from a genomic census of 1,152 Ts and 2,221 Hs in 492 proteomes, where all 492 characters were parsimoniously informative. Terminal leaves are not labeled because they would not be legible. The Venn diagram shows the diversity of Ts and Hs in the three superkingdoms, Archaea, Bacteria and Eukarya.

https://doi.org/10.1371/journal.pcbi.1003009.g004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 5. Architectural chronologies of CATH A, T and H domain structures.

Three phases or epochs (I, II and III) in the timeline delimit the appearance, crystallization and diversification of As (A), Ts (B) and Hs (C) in all three superkingdoms (top panels) and in Archaea, Bacteria, and Eukarya (bottom panels). Individual plots show the relationship of f (distribution Index) and age of domain structures defined at A (nd_A), T (nd_T) and H (nd_H) levels of structural abstraction.

https://doi.org/10.1371/journal.pcbi.1003009.g005

To uncover hidden patterns of organism diversification in our dataset, we divided structures according to their distribution in superkingdoms and constructed three separate structural chronologies for the genomes of each superkingdom at A, T and H levels of structural abstraction (Figure 5). Taxonomical groups of domain structures were identified in the time plots with different colors. We previously observed that a superkingdom must ‘lose’ a significant number of SCOP structures before the evolutionary appearance of the first superkingdom-specific ‘signature’ structure [7]. In our study, this loser trend of domain structures was also observed for the CATH annotated genomes in each superkingdom. This observation strengthens our claim of reductive evolution in protein domains of the lineages that emerge from the cellular urancestor (the last universal common ancestor; LUCA) that we find is functionally complex [11]. The loser trend of SCOP and CATH structures reveals the primordial birth of Archaea followed by the birth of Bacteria and Eukarya. For example, the complete loss of Hs first starts in Archaea (nd_H = 0.176) with the membrane-bound lytic murein transglycosylase D (chain A) H domain (3.10.350.10). Its appearance is congruent with the loss of the first SCOP FSF in Archaea (nd_FSF = 0.174), the LysM domain (d.7.1), observed in previous studies [7]. Both domain definitions are very much similar in how they describe functions in the cell. Analysis of domain distribution in Archaea shows that the vast majority of ancient Ts and Hs that were lost in proteomes were present in all superkingdoms (ABE; colored grey). These were followed by AB (orange), A (wine) and few AE (red) structures, most of which started to appear after the crystallization point and during the superkingdom specification and organismal diversification epochs [7]. Clear decreases in structural representation (f-value) also occurred in Bacteria and Eukarya, but involved fewer and younger structures. Analysis of domain distribution in Bacteria shows that AB and B structures (dark yellow) started to increase representation after the crystallization point, leading towards their diversification and specification. Similarly, the eukaryotic chronology showed that comparatively younger architectures [e.g. BE (blue) and E (green)] increased their popularity among the eukaryal lineages. The appearance and distribution of the seven taxonomical groups of T and H structures was unfolded in the timelines using boxplots describing the range of nd_T and nd_H values and measures of central tendency for each group (Figure 6). Only domains shared by the three superkingdoms (ABE) span the entire chronology, from the origin of proteins (nd = 0) to the present (nd = 1). These structures represent instantiations of the domain content of the urancestor but their late appearance may also indicate events of horizontal transfer between lineages. Boxplots for BE, AE and AB explain relationships among superkingdoms over time. The BE boxplot is the most ancient of the three, suggesting Archaea diversified early by reductive evolution. The A, B and E boxplots reflect the history of ‘signature’ structures that are unique to individual superkingdoms. These signatures appear first in Bacteria and then concurrently in Archaea and Eukarya, an observation that is congruent with timelines derived from SCOP domains [7]. Despite its early specification, Archaea tends to acquire Archaea-specific structures very late in evolution and their number is limited when compared to Bacteria and Eukarya. This may stem from very strong adaptive pressures that were historically imposed by lifestyle. Archaea are very simple organisms that usually live in harsh and extreme environments [21]. We believe their extremophilic lifestyles impose constraints on their molecular make up that: (i) limit the possibility of acquiring new structures, and (ii) induce a constant selective pressure to maintain a minimal structural set necessary for survival. We therefore propose that Archaea maintained a minimal set of structures while losing structures by strong reductive evolution. We note that signature As exhibit very low f values, suggesting these molecular designs were acquired as adaptations to new environments and lifestyles. The appearance of structures shared by only two superkingdoms was also revealing. For example, the AE boxplot's upper whisker approached nd_H = 1, implying a recent relationship between Archaea and Eukarya. Comparatively, the nd values for SCOP FSFs for the AE taxonomical group was nd_FSF = 0.85, supporting the late appearance of the interaction [7]. Note that a sister relationship between Archaea and Eukarya is usually used to claim the canonical bacterial rooting of the tree of life [22], but that in our studies this relationship is only supported by domain structures that are quite derived. It is also noteworthy that the early loser trend in the BE taxonomic group, made explicit by smooth decreases in f-values in the timeline, occurs in the absence of signature domain structures specific to superkingdoms (Figure 5). This weakens other evolutionary scenarios of superkingdom origin, including chimerism mediated by massive horizontal gene transfer (endosymbiosis or fusion) processes, and the possibility that phylogenetic signal of these events (e.g. those between Bacteria and Eukarya) would make Archaea appear artificially ancient in phylogenomic reconstructions (see below).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 6. Cumulative frequency plots of CATH H and T domain structures.

Cumulative frequency distribution plots plotted against the respective for T (A) and H (B) domain structures. Bottom plots show boxplots describing nd ranges for the seven taxonomic groups of T (C) and H (D) structures that are unique to individual superkingdom (A, B, E) or shared by two (AB, BE, AE) or all (ABE) superkingdoms. Numbers of T and H structures belonging to each taxonomic group are also indicated.

https://doi.org/10.1371/journal.pcbi.1003009.g006

Trees of proteomes derived from the CATH genomic census confirm the early emergence of Archaea

We previously reconstructed trees of proteomes from a genomic census of SCOP domains and made inferences about the rooting of the tree of life [7], [11], [16]. We found trees of proteomes reconstructed from ancient domain structures were rooted paraphyletically in Archaea while trees reconstructed using derived structures exhibited the canonical rooting with Bacteria emerging at their base. We also revealed how parasitic and symbiotic lifestyles can complicate phylogenetic interpretation [7], [16]. The proteomes of organisms that are parasitic or that establish symbiotic relationships with other organisms have frequently experienced reductive evolution, discarding enzymatic and cellular machineries in exchange for resources from their hosts. Since their inclusion can lead to incorrect phylogenetic trees, we excluded proteomes from all but 295 free-living (FL) organisms and reconstructed rooted trees that most parsimoniously describe their evolution. The FL set included 41 archaeal, 189 bacterial, and 65 eukaryotic organisms. The tree of FL proteomes reconstructed from a census of H domain structures supported the trichotomy of the superkingdoms (Figure 7). The number of bacterial proteomes was however overrepresented in the FL-tree and could cause long-branch attraction during phylogenetic reconstruction possibly leading to incorrect deep phylogenetic relationships. Since taxon sampling can also affect phylogenomic inference [23], we randomly sampled equal numbers of proteomes per superkingdom (a maximum of 41) and generated replicated trees of proteomes. Reconstruction of equally sampled FL proteomes improved tree resolution and bootstrap support values of deep branches. More importantly, the trees consistently showed a paraphyletic rooting in Archaea and the derived placement of monophyletic Bacteria and Eukarya (Figure 7). We also reconstructed trees of FL proteomes from three subsets of phylogenetic characters: ancient H structures common to all superkingdoms corresponding to the architectural diversification epoch (nd_H<0.176), H structures of intermediate ancestry corresponding to the superkingdom specification epoch (0.176<nd_H<0.55) and H structures that are derived and reflect the organismal diversification epoch (0.55<nd_H). The proteome tree reconstructed from the most ancient H structures was rooted paraphyletically in Archaea, reflecting their early segregation through the minimalist strategy. Reconstructions from H structures of intermediate ancestry produced trees with three clades corresponding to the three superkingdoms that were rooted in Archaea. Finally, reconstructions from H structures that were derived yielded the canonical tree of life rooted in Bacteria. It is noteworthy that the rooting of these trees reflects the early appearance of Bacteria-specific domain structures (Figure 7, see trees reconstructed using most ancient, ancient and younger characters sets). We note the split of Archaea in three groups in the tree reconstructed from ancient H structures. We believe this anomaly stems from using subsets of characters in phylogenomic reconstructions and from the existence of a ‘modern effect’ [11] imposed by relatively recent changes in abundance of domain structures belonging to the ABE taxonomic group. Both factors impoverished phylogenetic signal and obscured deep phylogenetic relationships. The modern effect is an embodiment of recent evolutionary processes affecting ancient repertoires, the effects of which must be identified and removed when reconstructing the set of domain structures present in the urancestor [11].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 7. A phylogenomic tree of proteomes generated from the equally sampled dataset of FL proteomes.

The circular cladogram of the most parsimonious rooted tree describes the evolution of 123 equally sampled proteomes and was generated from genomic abundances of 2221 Hs. Terminal nodes of Archaea (A: 41 proteomes), Bacteria (B: 41), and Eukarya (E: 41) were labeled in red, blue, and green, respectively. Also the total character set was divided into three independent character sets e.g. Most Ancient (nd_H 0∼0.176), Ancient (nd_H 0.176∼0.318) and Younger (nd_H 0.318∼1) characters set. These character sets resulted in three trees of proteomes that reflected the behavior of the tree over different character sets. Root branches are indicated with arrows.

https://doi.org/10.1371/journal.pcbi.1003009.g007

Modern Archaeo-Eukaryotic architectural sharing questions the canonical tree of life

The structural chronology, especially at H level, unveils a relatively recent (perhaps ongoing) sharing of protein architectures between archaeal and eukaryal genomes. The timeline reveals that while AE domain structures appeared for the first time when Archaea and Eukarya acquired their superkingdom-specific signature structures, the vast majority of them appeared quite late in evolution (e.g., Figure 6D). This was unanticipated. This finding inspired us to resolve the phylogenetic contribution of each structural character set in the tree of proteomes. Interestingly, characters that are shared by archaeal and eukaryal genomes exhibited high retention index (RI) values (Figure 8), indicating that the sharing pattern did not result from annotation artifacts. The RI measures the amount of synapomorphy (features that are shared and derived) expected from a data set that is retained as synapomorphy on a cladogram. Boxplots of structural character sets shared by the seven taxonomical groups were also plotted (Figure 8). Since low RI values indicate high levels of homoplasy (i.e. non-vertical phylogenetic signal), the low values of bacterial signature structures confirm the high incidence of horizontal gene transfer that exists in the bacterial superkingdom. In turn, the relatively high RI levels of the common ABE group is surprising. Most members of the group include very ancient structures (Figure 8), many of which were part of the urancestor. High RI levels in this taxonomical group challenge the common assumption that horizontal transfer was rampant during early life [22].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 8. The extent of synapomorphy exhibited by phylogenomic characters (H) in the trees of proteomes.

(A) Boxplots for retention index (RI) values of characters specific to seven taxonomical groups. (B) Mean RI for each taxonomical group was plotted with its standard error. (C) RI is plotted against the age (nd_H) of each character, colored according to its specific taxonomical group. (D) RI is plotted against the f distribution index of each, same coloring scheme were used as of (C).

https://doi.org/10.1371/journal.pcbi.1003009.g008

These RI boxplots are powerful enough to explain the relationships of superkingdoms in our tree of proteomes. The AE boxplot is the only one exhibiting very high RI values. In turn, bacteria-specific characters had the most dispersed RI boxplot. Hence, archaeal and eukaryotic lineages share good signal characters that are very recent and are widely present; their high f values indicate for example their presence in most of archaeal and eukaryotic proteomes (Figure 5C). More than 30 years ago, Woese and Fox [24] defined the existence of three ‘aboriginal’ lines of descent – superkingdoms Archaea, Bacteria and Eukarya. The microbial Archaea and Bacteria lines were conceptualized as ‘urkingdoms’ of deep origin that were qualitatively different from the eukaryotic kingdoms. This prompted reconstructions of a tripartite tree of life and later proposals of the early rise of Bacteria with rooting determined using paralogous gene couples (e.g., EF-Tu/EFG). This classical (canonical) tree topology induces sister lineages corresponding to Archaea and Eukarya and an exclusive common ancestor of both. Many archaeal components involved in informational systems (e.g. translation, replication and transcription) and transmission of genetic information show a higher sequence similarity with their eukaryotic homologue than their bacterial homologue [25], [26]. For instance, more than 30 ribosomal proteins are shared between the Archaea and Eukarya that are not present in Bacteria [27]. Moreover, Archaea and Eukarya also share a similar base excision repair system that is different than the system in bacteria [28]. If the phylogenetic signal in the sequence of these RNA and protein molecules adequately depicts history, these findings would explain the evolutionary link between Archaea and Eukarya and the topology of the canonical tree of life that emerges in some phylogenetic studies from their close relationship. However, many genes do not share the archaeal and eukaryal link and the canonical root must be considered contentious. Remarkably, the tree of proteomes reconstructed using the modern structural character set in our experiments (Figure 7, epoch III or younger character sets) is the only tree with the canonical topology that places the root branch in Bacteria. This topology mostly results from protein domain structures of very recent origin that are shared between Archaea and Eukarya. We contend that these very recent domains retain good phylogenetic signal, especially in their sequences, and will be less affected by processes of mutation saturation. Consequently, the close evolutionary relationship of Archaea and Eukarya in trees of life derived from analyses of these sequences [22], [24] can be considered an artifact of the focus on sequence.

Current trees of life built for example from sequence concatenation, such as those in refs. [29], [30], include genes encoding for multidomain proteins (e.g. aminoacyl-tRNA synthetases). Some of these domains are of recent origin and may fall within the derived domain set we have analyzed. We claim that strong phylogenetic signal in the sequence of these domains likely drives the reconstructed topologies. Instead, weak phylogenetic signal embedded in the sequences of older and universal domains is swamped by the recent archaeo-eukaryotic signal that is in part responsible for the canonical tree. Our focus on CATH domain structure (not gene sequence) can dissect the differential contribution of old and recent protein domains that belong to the proteome-encoding gene repertoire. A similar focus on deep phylogenetic signal in RNA structure has also shown the basal placement of Archaea in phylogenetic reconstructions from tRNA, RNase P RNA and 5S rRNA [31]–[35], including analysis of paralogy in tRNA [35]. For example, a timeline of accretion of helical RNA substructures of RNase P complexes showed the most ancient substructures were universal and harbored the core catalytic activities of the endonuclease [34]. However, the first substructures that were lost were specific to Archaea and this episode occurred before molecules were accessorized with superkingdom-specific substructures. The early origin of Archaea was also shown in trees that describe the structural evolution of RNase P RNA, which placed archaeal molecules at its base. These results obtained by studying the evolution of RNA structure clearly parallel the evolutionary patterns of CATH domain accumulation of this study. Clearly, deep phylogenetic signal in protein and RNA structure is free from the limitations of gene sequence and associated non-vertical patterns arising from horizontal gene transfer but more importantly from domain rearrangement and can therefore reveal historical patterns without bias [36]. Here we show the importance of considering the age heterogeneity of a biological repertoire, in this case the proteome, when making phylogenetic statements.

Chronologies of CATH architectures reveal evolutionary patterns of structural diversification

The architectural chronology of As is evolutionarily more conserved than chronologies of Hs and Ts (Figure 5A). The timeline shows that As are widely shared and are refractory to loss in genomic lineages. In fact, very few As are lost in superkingdoms (4 in Archaea, and one each in Bacteria and Eukarya) and are thus very old and popular in the world of organisms. The 3-layer (αβα) sandwich (3.40) is the most abundant and ancient of all proteins. The orthogonal bundle (1.10) and the α/β-complex (3.90) are equally abundant and are the second and third most ancient architectures. Remarkably, the phylogenomic tree of As shows that comparatively simpler shape structural designs are more favored than complex designs and in general are more ancient, appearing at the base of the tree. Architectural complexity was here evaluated on empirical grounds by focusing on the topology and regularity of spatial arrangements of secondary structures in a structural design. For example, the most ancient 3.40 and 1.10 architectures involve simple arrangements of secondary structure that can be very diverse in different structural variants while more recent shape designs are spatially more convoluted and regular (Figures 3). As time progresses the complexity in architectural make up of structural designs also increases (Figure 3). The few As that are lost in superkingdoms are quite complex and as expected their appearance is quite derived. The first loss occurred in Eukarya (nd_A = 0.76) with the very complex Clam architecture, and then in Archaea and Bacteria. We note that Archaea loses four As quite late and in a row, showing that the pervasive reductive trends of Archaea described above extend almost to the present. This also reflects the conservative nature of extremophilic Archaea, which are not in need of modern structural designs. Bacteria loses the most recent A structural design, Box (2.80), at nd_A = 1, which is shared by both archaeal and eukaryal genomes. Box is involved in nucleotide excision repair, a molecular function that has a unique place in cellular defense because of its wide substrate range and its ability to virtually remove all base lesions from a genome. Ögrünç et al. [28] reported a similar base excision repair system used in Archaea and Eukarya and argued that a different set of proteins are employed by the bacterial nucleotide repair system. Interestingly, the f index for Box in Archaea (f = 1) and Eukarya (f = 0.96) again indicates a recent sharing of structural designs between archaeal and eukaryal organisms. Architectures constitute the second highest level of structural abstraction in CATH, and because of their high conservation it is difficult to clearly delimit the three epochs of the protein world. In contrast, our results indicate CATH H and SCOP FSF are the most suitable levels to uncover the evolution of domain structures in genomes. These levels of abstraction are structurally and evolutionarily conserved. They preserve deep phylogenetic signatures and are variable enough to dissect evolutionary history of proteomes and molecular functions.

CATH architectures become more complex in evolution

To obtain a detailed view of architectural discovery and usage over time, we grouped As into 10 major structural designs: sandwiches, bundles, barrels, prisms, horseshoes, rolls, solenoids, propellers, complexes and other (a category with structural designs that could not be clearly grouped into the main categories) (Table 1 and Figure 9). We found that most sandwiches, bundles, barrels, complexes and rolls have high f values (f∼1) and rather simple structural designs (Figure 9). In turn, structural designs such as propellers, horseshoes, solenoids (2 Solenoid, 2.150), prisms, trefoil and box, have low f values (f = 0.85–0.10) and are very complex. Under the assumption that widespread and abundant designs are old, complex folds appear to have evolved later than simpler folds. We also mapped the appearance of T and H structures harboring individual A designs, plotting nd_H and nd_T values for Hs and Ts belonging to each of the 38 known As (Figure 10).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 9. Architectural chronologies of CATH A domain structures colored according to structural design.

As shown in Table 1 we grouped the 38 As into 10 larger sets of general structural designs. As were plotted against their age (nd_A) and f distribution indices, whereas each A was colored according to their general structural design group.

https://doi.org/10.1371/journal.pcbi.1003009.g009

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 10. Cumulative frequency distributions of Ts and Hs belonging to a particular A along timeline of domain structures.

Plots A and B describe the evolutionary appearance of T and H domain structures, respectively. These two plots uncover patterns of diversification of structural designs in architectures over time. For example, the evolutionary accumulation of Ts and Hs belonging to the oldest architecture, the 3-layer (αβα) sandwich (3.40), occurs early but at different rates than Ts and Hs belonging to the orthogonal bundle (1.10) and 2-layer sandwich (3.30). The same pattern can be seen in (B), where the accumulation of the 4-layer sandwich (1.20) surpasses that of the α-β complex (3.90), even if 3.90 is more older than 1.20.

https://doi.org/10.1371/journal.pcbi.1003009.g010

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Grouping of CATH A level structures into 10 general categories.

https://doi.org/10.1371/journal.pcbi.1003009.t001

The structural makeup of the most ancient 3-layer (αβα) sandwich (3.40) architecture (Figure 3) represents the central theme of the most ancient SCOP FFs [37]. These structures consist of repeating α-β-α supersecondary units, such that the outer layer of the structure is composed of helices packing against a central core of parallel β-sheets. Many enzymes, including most of those involved in glycolysis, are α/β layered proteins and are cytosolic [38]. These α/β structures harbor repeats of the α-β-α arrangement (e.g., the α-β-α-β-α sequence). The β-strands are parallel and hydrogen bonded to each other, while the α-helices are all parallel to each other but are antiparallel to the strands. Thus the helices pack against the sheet forming a sandwich-like structure. We note that the β-α-β-α-β (αβα) subunit, often present in nucleotide-binding proteins, represents the Rossmann structural motif found in proteins that bind nucleotides, especially the cofactor NAD(H) [39].

The orthogonal bundle (1.10) and α-β-complex (3.90) appear immediately after the 3-layer (αβα) sandwich (3.40) design. The orthogonal bundle consists of a 3–4 α-helix bundle and is found in a number of different proteins, most of which associate with membranes. Due to physical constraints imposed by the lipid bilayer of membranes the list of possible membrane protein structures is limited to either bundles [40], [41] or barrels [42], [43]. In many cases the α-helices are part of a single polypeptide chain and are connected to each other by three loops. In the 4-helix bundle proteins the interfaces between the helices consist mostly of hydrophobic residues while polar side chains on the exposed surfaces interact with the aqueous environment. A number of cytokines consist of 4-helix bundles, such as interleukin-2, interleukin-4, human growth hormones, and the granulocyte-macrophage colony-stimulating factor (GM-CSF) [38] and DNA binding proteins (e.g., transcription factors, repressors proteins) [44]. CATH has grouped the complex shaped structures into the ‘complex’ bin, until alternative assignment methods are developed. The α/β-complex architecture groups together all those designs that include significant α and β secondary structural elements in a mixed fashion. Examples of α/β-complex proteins include bacterial and mammalian pancreatic ribonucleases [45], Zn metallo-proteases and DNA topoisomerases [46]. Two kinds of barrel structures are the most ancient and abundant in the protein world, the α/β-barrel (3.20) and the β-barrel (2.40) [6], and both appeared at about the same time (nd_A = 0.13). The α/β-barrel is composed of eight α-helices and parallel β-strands that alternate along the peptide backbone. The α/β-TIM barrel is the most prominent example of α/β-barrel and is widely present in enzymes of central metabolism [47]. A β-barrel is a large β-sheet that twists and coils to form a closed structure in which the first strand is hydrogen bonded to the last. β-strands in β-barrels are typically arranged in an antiparallel fashion. Barrel structures are commonly found in porins and other proteins that span cell membranes and in proteins that bind hydrophobic ligands in the barrel center, such as lipocalins [48]. The roll is a complex nonlocal structure in which 3–4 pairs of antiparallel β-sheets, only one of which is adjacent in sequence, are ‘wrapped’ in 3D space to form a barrel shape [49]. Rolls appear for the first time at nd_A = 0.3.

A number of distinct and more complex architectures appear later on in the chronology, including solenoids, horseshoes, prisms, propellers and trefoils. Solenoid proteins, with their arrays of repeating motifs, tend to have elongated structures that contrast with the majority of globular proteins whose polypeptide chains follow more complex trajectories [50]. These are constructed from tandem structural repeats arranged in superhelical fashion, a feature that is important for many cellular processes [51]. Solenoid proteins constructed from HEAT repeats [52] and armadillo repeats [53], [54] constitute the principal transport receptors. A key structural property that differentiates solenoid proteins from other structured proteins is the lack of contacts between distal regions of protein sequence (sequence-distal contacts). For this reason, solenoid proteins are often more flexible than other structured proteins and this flexibility is an important feature of their specific functions [50]. The solenoid structure appears for the first time at nd_A = 0.46. The α-horseshoe protein appears at nd_A = 0.4, is a super helical structure made up of a number of three α-helical orthogonal bundle repeats. The α-β horseshoe appeared at nd_A = 0.56, consists of several α/β-repeating units [55]. The structure of the ribonuclease inhibitor, a cytosolic protein that binds strongly to any ribonuclease that may leak into the cytosol, takes the concept of the repeating α/β unit to the extreme [55]. The structure is made of a 17-stranded parallel β-sheet curved into an open horseshoe shape, with 16 α-helices packed against the outer surface. Prisms are similar to solenoids in geometry but completely different in connectivity. A more self-contained β-sheet forms each face of a triangular prism. They appear late at nd_A = 0.86. The trefoils consist of an unusual β-sheet formed by six β hairpins arranged with three fold symmetry into ‘Y’ like structures [56] and are also quite derived (nd_A = 1).

Models of evolution of CATH and SCOP domain structures are congruent

The most ancient and popular architecture, the 3-layer (αβα) sandwich (3.40), harbors the most ancient and abundant topology, the Rossmann fold (3.40.50) and the most ancient and abundant superfamily, the P-loop containing nucleotide triphosphate hydrolases (3.40.50.300). Despite differences of topology and ranking within databases [57], this H structure of CATH is analogous to the “P-loop containing nucleotide triphosphate hydrolase” FSF (c.37.1) of SCOP [4], since both have Rossmann fold topology and also agree on their keyword definitions. A careful analysis of CATH and SCOP structures phylogenies show that the ancient domains structures at T (3.40.50) and H (3.40.50.300) levels are in global agreement with timelines of F (c.37) and FSF (c.37.1) structures [7]. Despite differences in domain definitions of tertiary structure in CATH and SCOP, the remarkable conservation of evolutionary signal indicates both classification systems effectively preserve evolutionary information in protein structure and uncover global patterns of origin and diversification that are for the most part congruent.

We note that levels of structural abstraction above H and FSF unify domains that may not be necessarily homologous. In other words, T and A in CATH and F in SCOP may show episodes of structural convergence. This could complicate evolutionary interpretations. The fact that the same evolutionary patterns observed using H domain structures in this study (and FSF structures in previous studies; reviewed in [1]) could be recovered at higher levels of the structural hierarchy is encouraging and suggests that the influence of convergent processes at these higher levels is limited and that the classifications do in general a good job in capturing true evolutionary information.

Major conclusions

In this study we follow the history of protein fold structures and proteomes in the tripartite world of organisms. Instead of generating trees of life from protein sequence with standard methods, we use a genomic structural census and robust cladistics methods to build trees of domain structures and proteomes. Structural phylogenies describing the evolution of CATH domains at A, T and H levels of structural abstraction revealed patterns of reductive evolution and the three epochs in the evolution of the protein world that were previously proposed [7]. Structural diversification patterns match those observed in the analysis of SCOP domain structures [7], [16], [58]. Reconstruction of phylogenomic trees of proteomes describing the evolution of lineages confirms Archaea is the most ancient superkingdom.

Provided assumptions of our phylogenomic method are considered valid, six major findings summarize novel results and take advantage of the ability of CATH to better describe topological features of protein structure:

1. Structural designs that are architecturally simpler are ancient and highly abundant in the extant world of proteins and organisms. We find the 3-layer (αβα) sandwich cytosolic architecture and the orthogonal bundle that often associates with membranes are the most ancient and are preferentially involved in metabolic activities. The origin of proteins thus lies at the interface of primordial membranes and cytoplasm. Bundles and barrels that populate membranes soon follow. Metabolic and membrane proteins thus appear crucial for the early biochemistry of primordial cells [37].
2. Structural designs that are architecturally complex, such as prisms, propellers, 2-solenoid, super-roll, clam, trefoil and box are derived and less favored in the world of organisms. These designs are generally specific to groups of organisms and have been probably adopted for specialized functions.
3. Although CATH and SCOP differ significantly in their protein domain definitions and in the hierarchical partitioning of fold space, we find that both protein structural classifications classify a protein on very similar theoretical grounds by taking into account their structural, functional and evolutionary roles. Remarkably, CATH and SCOP structures harbor similar phylogenetic signatures and reveal patterns of origin and diversification that are congruent.
4. Structural chronologies provide evidence that Archaea established the first organismal divide by losing a substantial number of domain structures early in evolution. We speculate this reductive evolutionary process reflects the environmental pressure of an ancient extremophilic lifestyle that forced maintenance of a minimal domain repertoire. Our analysis provides additional support to the archaeal rooting of the tree of life obtained from evolutionary analysis of SCOP domains (e.g. [7], [16]), RNA structure [31]–[34], tRNA paralogs [35], and a phylogenomic analysis of abundance of Gene Ontology terms in functiomes (Kim, Nasir and Caetano-Anollés, ms. submitted).
5. Structural chronologies uncover a recent trend of sharing of domain structures between Archaea and Eukarya that continues to the present and involves complex architectures such as the Box (2.80) design that is involved in nucleic acid repair.
6. Finally, we also speculate that this modern archaeo-eukaryotic architectural sharing pattern is the most probable reason for the bacterial rooting of the canonical tree of life reconstructed from changes in sequence. In contrast to structure, sequence evolution is more dynamic and prone to phylogenetic signal loss [36]. It is therefore likely that most useful phylogenetic signal in these sequence studies is drawn from structures that have been developed quite recently in evolution.

We note that these conclusions entrust CATH with the ability to properly apportion domain structures in fold space and are only valid if assumptions of character argumentation are valid.

Our trees of domain structures define timelines that trace back the history of innovation, diversification and distribution of protein structural designs. Our finding that protein architectures tend to become more complex in evolution is very significant. In a previous study, analysis of β-barrel structures revealed that the curl and stagger and complexity of the connectivity of supersecondary structures increases in evolution [12]. The very early appearance of multilayered sandwich structures is also compatible with the finding that the most ancestral folds share a common architecture of interleaved β-sheets and α-helices [12]. An even more recent study shows that 36 out of the 54 most ancient FFs harbor α/β/α-layered sandwich structures [37]. The very early appearance of the P-loop hydrolase motif in the first FF, the ABC transporters, was associated with a built-in lateral bundle, which resembles the trans-membrane domains of transporter proteins. This suggests that first proteins contained sandwich and bundle structures and were associated with the membranes of primordial cells. Remarkably, P-loop hydrolase folds and bundles make up important membrane complexes, such as ion channels and transporters. Their very early origin highlights a crucial links between the origin of proteins and the origin of cells.