A Tree of Life Based on Protein Domain Organizations

Kaoru Fukami-Kobayashi,* Yoshiaki Minezaki, Yoshio Tateno, and Ken Nishikawa

*RIKEN BioResource Center, RIKEN Tsukuba Institute, University of Tsukuba, Tsukuba, Ibaraki, Japan; School of Library and Information Science, University of Tsukuba, Tsukuba, Ibaraki, Japan; and Center for Information Biology and DNA Data Bank of Japan, National Institute of Genetics, Research Organization of Information and Systems, Mishima, Japan It is desirable to estimate a tree of life, a species tree including all available species in the 3 superkingdoms, Archaea, Bacteria, and Eukaryota, using not a limited number of genes but full-scale genome information. Here, we report a new method for constructing a tree of life based on protein domain organizations, that is, sequential order of domains in a protein, of all proteins detected in a genome of an organism. The new method is free from the identication of orthologous gene sets and therefore does not require the burdensome and error-prone computation. By pairwise comparisons of the repertoires of protein domain organizations of 17 archaeal, 136 bacterial, and 14 eukaryotic organisms, we computed evolutionary distances among them and constructed a tree of life. Our tree shows monophyly in Archaea, Bacteria, and Eukaryota and then monophyly in each of eukaryotic kingdoms and in most bacterial phyla. In addition, the branching pattern of the bacterial phyla in our tree is consistent with the widely accepted bacterial taxonomy and is very close to other genome-based trees. A couple of inconsistent aspects between the traditional trees and the genome-based trees including ours, however, would perhaps urge to revise the conventional view, particularly on the phylogenetic positions of hyperthermophiles.

Introduction As a number of genome projects produced the complete genome sequences for various species, genome-wide information has been employed to construct a genome tree for inferring evolutionary relationship of species. Accordingly, several methods for constructing a genome tree have been invented. Some authors compared the whole genomes with respect to the gene contents (Fitz-Gibbon and House 1999; Snel et al. 1999; Tekaia et al. 1999; Wolf, Rogozin, Grishin, et al. 2001; Korbel et al. 2002) to the domain contents (Gerstein 1998; Wolf et al. 1999; Lin and Gerstein 2000; Caetano-Anolles G and Caetano-Anolles D 2003; Winstanley et al. 2005; Yang et al. 2005), to the gene order (Wolf, Rogozin, Kondrashov, Koonin 2001; Korbel et al. 2002), or to the total evolutionary distances between a pair of orthologs (Wolf, Rogozin, Grishin, et al. 2001; Clarke et al. 2002), others dealt with multiple genes by concatenating the aligned sequences (Hansmann and Martin 2000; Brown et al. 2001; Katoh et al. 2001; Wolf, Rogozin, Grishin, et al. 2001), and still others synthesized a compound tree by uniting multiple trees constructed from individual genes (Daubin et al. 2001). In the pregenome era, we used to construct phylogenetic trees for individual genes or proteins, which have greatly contributed to elucidating unknown, unclear, or controversial phylogenetic positions of a number of species and populations. Unfortunately, however, trees constructed for different genes or proteins are often mutually incompatible (Wolf et al. 2002; Bapteste et al. 2004). The reasons for this could be due to imperfectness of a tree-construction method, stochastic nature of nucleotide substitution, polymorphism in a common ancestral population, horizontal gene transfer (HGT), and/or nondiscrimination of paralogous from orthologous genes. The primary reason, however, lies in the fact that most of the constructed trees are gene trees (Tateno et al. 1982) and not species trees. Because not all genes have evolved consistently with their
Key words: universal tree, protein domain, genome, phylogeny, hyperthermophile. E-mail: kfukami@brc.riken.jp.
Mol. Biol. Evol. 24(5):11811189. 2007 doi:10.1093/molbev/msm034 Advance Access publication March 1, 2007

host species owing to some of the biological factors mentioned above, a gene tree is sometimes inconsistent with the way the species in question have evolved. One way to resolve the inconsistency between the gene and species trees is to use as many genes in a genome as possible (Tateno et al. 1982). It has been reported that the more genes we use, the less inuence we suffer from the disturbances mentioned and the closer we can reach the correct tree (Rokas et al. 2003). It is thus desirable to use the entire and pertinent information on the whole genome, if we dare to construct a species tree as a rst approximation of the evolutionary relationship of species. Here, we report a new method for constructing a species tree based on protein domain organizations. We determine domain organization of each protein by the sequential order of the domains in the protein, and then dene a repertoire of protein domain organizations of an organism as a set of domain organizations of all proteins encoded in its genome. By using difference of the repertoires among species, a species tree is constructed. Proteins with the same domain organization are roughly regarded as homologous over their total length. Therefore, our method compares the whole genomes with respect to homologous gene contents by the medium of protein domains. Because our method is based not only on the contents of domains but also their order along the primary structure in each protein, it treats more detailed aspects of evolution than the methods based on the domain contents alone (Winstanley et al. 2005; Yang et al. 2005). Moreover, like the domain content methods, our method does not require an orthologous gene set of the species studied for which numerous complicated and burdensome procedures are unavoidable. Applying our method to 17 archaeal, 136 bacterial, and 14 eukaryotic organisms, we have constructed a tree of life that is quite consistent with the taxonomy widely accepted today. We then discuss consistent and inconsistent aspects between our tree and the traditional tree focusing particularly on the phylogenetic positions of hyperthermophiles. Materials and Methods Protein Domains We used the Protein families database of alignments and HMMs (Hidden Markov Models) (Pfam) (Bateman et al. 2004) version 11.0 to assign the Pfam domains to

2007 The Authors. This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/ by-nc/2.0/uk/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

1182 Fukami-Kobayashi et al.

the open reading frame (ORF) sequences in the Genomes TO Protein structures and functions (GTOP) database (Kawabata et al. 2002) as of May 2004 (ftp://spock.genes. nig.ac.jp/pub/gtop/Old_04Jul/). GTOP contains all ORFs of the organisms for which the whole-genome sequence has been reported or the complete set of cDNA clones is sequenced. The Pfam domains were assigned for the ORFs by HMM (Durbin et al. 1998) using GA (gathering cutoffs) scores recorded for individual protein families (Eddy 2003). Assignment of the Structural Classication of Proteins (SCOP) superfamily domains was also carried out by applying reverse PSI-Blast (position specic iterative Blast [Basic Local Alignment Search Tool]) (Altschul et al. 1997) to the SCOP database (Murzin et al. 1995) version 1.65. The cutoff e value was 103. We primarily adopted the Pfam domains for construction of a repertoire of protein domain organizations. Although the SCOP domains have been commonly used in the domain contents methods (Gerstein 1998; Wolf et al. 1999; Lin and Gerstein 2000; Caetano-Anolles G and Caetano-Anolles D 2003; Winstanley et al. 2005; Yang et al. 2005), we supposed that the Pfam domains were more suitable to the present study for the following 2 reasons. First, the Pfam domains are dened by sequence conservation and have clear evolutionary relationship, which is important to phylogenetic analysis. Second, the Pfam domains are more extensively assigned to the ORF regions and are expected to bring out more information from genome sequences than the SCOP domains: the former covered 50.5% of the ORF regions, whereas the SCOP domains covered only 40.2%, when measured in the residue basis in GTOP. In order to check our expectation and robustness of our method, we applied our method also to the SCOP superfamily domains. Domain Organizations and Repertoires The domain organization of a protein is dened as the sequential order of the domains in the protein. For example, if a protein has domains A, B, and C along its primary structure in that order, its domain organization is dened as ABC (g. 1). It is distinguished from the domain organization BAC, which consists of the same domains but their order is different, and is also distinguished from the domain organization ABCC, which has an extra C domain. The domain organization could also be composed of one domain. Regions to which the same domain is assigned are guaranteed to reect their common ancestry, and therefore, proteins with the same domain organization are roughly regarded as homologous over their total length. The repertoire of protein domain organizations of an organism is dened as the set of domain organizations of all proteins encoded in the genome of the organism. Only the presence or absence of a domain organization was considered for the construction of the repertoire of each organism. Evolutionary Distances and a Phylogenetic Tree Let us consider 2 organisms A and B that diverged from a common ancestor O. Let us also denote that the numbers of domain organizations of A and B are nA and nB, respectively, and the number of domain organizations shared by A and B is nAB. A and B initially share the same

FIG. 1.Protein and its domain organization. The domain organization of a protein is dened as the sequential order of the domains in the protein. The domain organization could also be composed of one domain.

repertoires (i.e., nA 5 nAB 5 nB). As A and B diverge, nAB decreases because they independently accumulate new domain organizations and lose extant ones. Assuming that the gain and loss respectively follow the Poisson process, the evolutionary distance between O and A, dOA, is given by dOA 5 lnnAB =nA : 1

Similarly, the evolutionary distance between O and B, dOB, is obtained. The values of dOA and dOB are more or less the same in most cases. However, if one of the organisms, B for instance, has a substantially smaller number of the domain organizations than A, dOA is much larger than dOB. The excess of dOA is mostly due to the shrinkage of the genome size of B from the original one, which often occurs in parasites or symbionts. To reduce this adverse effect, we dened the geometric mean of dOA and dOB as the evolutionary distance between A and B given by dAB 5 dOA 3dOB :
1=2

Actually, we have explored other ways to dene dAB, by employing arithmetic mean and harmonic mean and by taking a smaller value between dOA and dOB and conrmed that geometric mean showed the best performance among them. In this sense the employment of geometric mean is rather empirical, yet it also has some basis: because the geometric mean weighs a smaller value between the 2, the effect of the genome shrinkage on the evolutionary distance is effectively reduced. To see this clearly, let us consider an extreme case in which the genome shrinkage occurred in B and no other changes occurred in A and B, then dOB is zero by denition, and thus dAB is expectedly zero. The evolutionary distance between every pair of the organisms studied was computed, and a tree of life was accordingly constructed by the Neighbor-Joining (NJ) method (Saitou and Nei 1987). Bootstrap Probability We evaluated the reliability of each internal branch of the constructed tree by the bootstrap method (Felsenstein

A Tree of Life Based on Domain Organizations

1183

FIG. 2.Correlation between the number of ORFs and the number of domain organizations. Each spot represents an organism of Archaea (triangle), Bacteria (square), or Eukaryote (diamond). The 2 numbers are roughly proportional to one another, except for plant organisms (marked with arrows). Both numbers for eukaryotic organisms are distinctively larger than those of archaeal or bacterial organisms.

FIG. 3.Correlation between the number of domain organizations and the proportion of shared domain organizations. The correlation was computed only for the bacterial organisms. Each spot represents a bacterial organism, and the proportion of a given organism was obtained by averaging all the proportions computed between the organism and the other bacterial organisms.

1985) modied by us for the domain organization repertoire data. The modication is as follows. Let us consider a table of m rows and n columns, in which the rows and the columns stand for m organisms and n domain organizations respectively, and presence or absence of a domain organization of an organism is recorded in each cell. In bootstrap resampling, n columns were randomly sampled with replacement from the original table to make a resampled table. Using the resampled table, evolutionary distances and then a tree were obtained by the same methods as those mentioned in the above section. These processes were repeated 100 times. We then compared the topology of each generated tree with that of the original tree and counted the number of occurrences of each branch of the original tree for the 100 generated trees. The percentile number of the occurrences for each branch is the bootstrap probability for the branch in this case. Construction of a Network Using the evolutionary distances used for a tree construction, we also constructed a network with NeighborNet (Bryant and Moulton 2004) to reveal some of the conicting signals that might harbor in the distance data. Results Properties of Domain Organization Repertoires We selected 17 archaeal, 136 bacterial, and 14 eukaryotic organisms, for which the whole-genome sequence or the complete set of cDNA clones is sequenced. Each archaeal and eukaryotic organism represents its species, and the bacterial organisms are derived from 105 species, some of which contain different strains. The names, codes, numbers of ORFs, genome sizes, and National Center for Biotechnology Information (NCBI) taxonomy of the organisms analyzed are listed in supplementary table S1 (Supplementary Material online).

The number of the domain organizations extracted from each organism (nX) ranged between 266 and 4,762 among the 167 organisms, and 18,302 domain organizations were extracted from the organisms as a whole. As expected, nX was distinctively larger for eukaryotic organisms than for archaeal or bacterial organisms. In prokaryotes, nX is larger for a bacterial organism than for an archaeal organism, except for a parasitic or symbiotic bacterial organism such as Mycoplasma, Rickettsia, Chlamydia, or Buchnera. In general, nX is roughly proportional to the number of ORFs in an organism (g. 2), although plant organisms have small nX for their number of ORFs, as compared with other organisms. This is not due to ineffective detection of the Pfam domains in the plant proteins because the domains are no less extensively assigned to the ORF regions in the plant genomes (e.g., the Pfam cover rate for A. thaliana is 40.2%) than in the other eukaryotic genomes (e.g., 38.2% for human). The most probable explanation for the relatively small nX in the plants is the recent genome duplication in them (The Arabidopsis Genome Initiative 2000; Goff et al. 2002; Yu et al. 2005), by which the number of ORFs was doubled, whereas that of domain organizations was not. We then counted nAB, the number of domain organizations shared by organism A and organism B, for every pair of the 167 organisms. The proportion of the shared nAB to the total nA or nB ranged between 0.03 and 1.00. The proportions for a pair of organisms, nAB/nA and nAB/nB, are expectedly larger as the organisms A and B relate more closely. The proportion also tends to increase, as the total nA deceases (g. 3). This is consistent with the fact that most genes in a small genome are essential and ubiquitous. Construction of a Domain Organization Tree Using the nAB/nA and nAB/nB values, we computed an evolutionary distance for every pair of the 167 organisms and constructed a phylogenetic tree of them as shown in gure 4 (see Supplementary g. S1, Supplementary Material online for an enlarged version of the tree). The organisms in

1184 Fukami-Kobayashi et al.

100

ecol0 ecol1 ecol2 ecol3 sfle0 sfle1 100 styp0 100 styp2 styp1 100 baph0 100 buch0 baph1 83 ypes0 100 ypes1 ypes2 bflo0 plum0 wbre0 hduc0 100 99 hinf0 pmul0 vcho0 100 vvul0 100 vvul1 vpar0 sone0 paer0 100 pput0 96 psyr0 neur0 100 bbro0 100 bpar0 99 bper0 rsol0 cvio0 100 nmen0 nmen1 cbur0 100 xaxo0 100 xcam0 100 xfas0 xfas1 100 atum0 atum1 smel0 100 mlot0 100 bmel0 92 bsui0 100 bjap0 rpal0 100 100 rcon0 rpro0 ccre0 cjej0 86 hhep0 95 100 100 hpyl0 hpyl1 wsuc0 aaeo0 gsul0 bbac0 100 lint0 lint1
100 100 100 97 100

Bacteria
-proteobacteria

-proteobacteria

-proteobacteria -proteobacteria
-proteobacteria
bbur0

100 100 99 87

97 97

ctep0 bsub0 bhal0 oihe0 bant0 100 bcer1 bcer0 99 saur0 100 saur1 100 saur2 sepi0 100 linn0 lmon0
100 93 97 96 100

ccav0 cmur0 ctra0 cpne0 100 cpne1 cpne3 cpne2 rbal0 bthe0 pgin0
92 100 95

tpal0

Spirochaetes

Chlamydiae
Bacteroidetes

Proteobacteria

Bacillales

100

efae0 ljoh0 lpla0 llac0 100 saga0 saga1 spyo0 100 spyo1 spyo2 100 100 spyo3 100 spne0 spne1 smut0
100 100

Lactobacillales

81 97 100 87 98 87 100 100

86

100 100

ctet0 Clostridia tten0 tmar0 fnuc0 blon0 cdip0 100 ceff0 cglu0 100 100 98 100 mbov0 mtub0 100 mtub1 100 94 mlep0 100 100 save0 scoe0 100 twhi0 twhi1 anab0 100 telo0 syne0 96 98 pmar0 100 pmar2 100 pmar1 syne1 gvio0 100 drad0 tthe0 Deinococcus-Thermus halo0 100 mace0 mmaz0 mjan0 100 mkan0 84 100 mthe0 mmar0 100 aful0 100 96 paby0 100 phor0 pfur0 100 taci0 tvol0 90 aper0 100 paer1 100 ssol0 stok0 hsap0 100 mmus0 100 rnor0 96 100 drer0 trub0 93 100 cint0 100 cbri0 cele0 100 dmel0 100 scer0 100 spom0 Fungi ncra0 atha0 osat1

cace0 cper0

mgal0 mgen0 mpne0 uure0 mpul0 mpen0 past0

Mollicutes

Firmicutes

Actinobacteria

Cyanobacteria

Euryarchaeota

Crenarchaeota

Archaea

Metazoa

Viridiplantae

Eukaryota

FIG. 4.A phylogenetic tree based on the domain organization repertoires. The tree includes 17 archaeal, 136 bacterial, and 14 eukaryotic organisms. The organisms in the tree are color coded according to the NCBI taxonomy (Benson et al. 2000; Wheeler et al. 2000). The Pfam (Bateman et al. 2004) domain organizations were used for constructing the domain organization repertoire of each organism. Evolutionary distance between every pair of the 167 organisms was computed based on their domain organization repertoires, and the tree was constructed by the NJ method (Saitou and Nei 1987). The number on an internal branch shows a percentage of bootstrap probability (Felsenstein 1985). Only bootstrap values equal to or higher than 80% are shown.

A Tree of Life Based on Domain Organizations

1185

the phylogenetic tree are color coded following the NCBI taxonomy (Benson et al. 2000; Wheeler et al. 2000). The topology of the tree is quite consistent with the NCBI taxonomy and other traditional ones (Garrity et al. 2004) in the following 5 aspects. First, the organisms are clearly divided into the 3 superkingdoms, Archaea, Bacteria, and Eukaryota. Secondly, all the 3 eukaryote kingdoms and most prokaryote phyla show monophyly at a high bootstrap probability. Monophyly is also observed in many bacterial sections, a lower rank to phylum. Although Beta-/ Gammaproteobacteria sections are not clearly separated, this is rather due to historical misclassication of Betaproteobacteria as a separate phylum. Thirdly, most of the branching patterns of the bacterial phyla are in agreement with their evolutionary relationships that are widely accepted today, although bootstrap support for the branching patterns is not signicant. For example, it is considered now that Deinococcus-Thermus phylum diverged very early in the evolution of Bacteria (Gupta 1998; Brown 2003), which is also shown in our tree. Our tree also shows early divergence of Cyanobacteria phylum, clustering of gram-positive phyla Actinobacteria and Firmicutes together, and close relationships of Chlamydiae, Bacteroides, and Spirochaetes to Proteobacteria. Fourthly, the branching patterns of the eukaryotic organisms are in agreement with their evolutionary relationships in the traditional taxonomy, although the branches in the animal (Metazoa) lineage and those in the plant (Viridiplantae) lineage are respectively longer and shorter than those in the other lineages. The longer branches in the animal lineage probably reect an increase in the rate of change of domain organization repertoires in this lineage. Finally, our tree places parasitic and symbiotic organisms in the appropriate taxonomic positions. Many genome trees do not show the correct positions of those organisms and cluster them together regardless of their taxonomic positions, whereas some genome trees resolve the problem and show the correct positions (Korbel et al. 2002; Kunin et al. 2005; Yang et al. 2005). Our method has also overcome the problem. These 5 aspects together conrm that the difference of the domain organization repertoires among organisms is a reliable measure for inferring their evolutionary lineages. There are, however, also inconsistent aspects between the traditional tree and ours, which are also found in genome trees constructed by other authors. First, our tree shows that Euryarchaeota is not monophyly in the Archaea cluster. Whereas one of the Euryarchaeota organisms, Halobacterium sp. NRC-1 (halo0), is placed on a separate basal branch of the cluster, 2 Euryarchaeota organisms, Thermoplasma acidophilum DSM1728 (taci0) and Thermoplasma volcanium GSS1 (tvol0), cluster together with Crenarchaeota rather than the other Euryarchaeota organisms. The same relationships in the Archaea cluster have also been obtained in other genome trees (Korbel et al. 2002; Wolf et al. 2002; Yang et al. 2005). Secondly, the Bacillales section in our tree does not show monophyly (Wolf, Rogozin, Grishin, et al. 2001; Korbel et al. 2002; Wolf et al. 2002; Yang et al. 2005). Thirdly, the Spirochaetes organisms in our tree do not make a cluster (Yang et al. 2005). Fourthly, the Proteobacteria cluster in our tree includes thermophilic bacterium, Aquifex aeolicus (aaeo0) (Wolf et al. 2002; Yang et al. 2005).

Finally, the Firmicutes cluster includes another thermophilic bacterium, Thermotoga maritima (tmar0), and fusobacteria, Fusobacterium nucleatum (fnuc0) (Yang et al. 2005). Those inconsistent aspects are not conclusive to state that our tree is incompatible with the traditional trees for the following 2 reasons. First, some branches involved in the inconsistent clusters have too low bootstrap probabilities to be valid. Our tree cannot resolve deep phylogeny, which is a general feature of trees of life. The branch clustering Thermoplasmata and Crenarchaeota, for example, shows less than 50% bootstrap probability, and so do the branches that prevent Spirochaetes from being monophyly. Secondly, a taxonomic unit sometimes corresponds to a paraphyletic group such as reptiles in the traditional classication (Graur and Li 2000). Because the Bacillales organisms make a paraphyletic group in the Firmicutes cluster in our tree, and so do the Euryarchaeota organisms in the Archaea cluster, they can be considered as taxonomic units. A SCOP Tree and a Domain Contents Tree We also applied our method to the SCOP superfamily domains in place of the Pfam domains (supplementary g. S2, Supplementary Material online). As can be seen in gures 4 (or supplementary g. S1, Supplementary Material online) and supplementary gure S2 (Supplementary Material online), the 2 trees have similar topology. The SCOP tree, however, is a little more different from the traditional taxonomy than the Pfam tree. The Cyanobacteria phylum clusters with the Proteobacteria phylum in the SCOP tree. The phyla Chlamydiae, Bacteroides, and a part of Spirochaetes are more closely related to the Firmicutes phylum in the SCOP tree than in the traditional taxonomy. In addition, the overall bootstrap probabilities are lower in the SCOP tree than in the Pfam tree. We considered, therefore, the Pfam tree better than the SCOP tree. The difference in the 2 trees could be attributed to the difference in the amount of information between the Pfam and SCOP domains as mentioned in Materials and Methods. We also constructed a tree using the Pfam domains instead of the domain organizations (supplementary g. S3, Supplementary Material online) to see if the domain organizations bring a better result than the domain contents. The domain contents tree has more similar topology to the domain organization tree than the SCOP tree: it exhibits an early diversication of the Cyanobacteria phylum as well as the domain organization tree. However, like the SCOP tree, the domain contents tree shows closer phylogenetic relationship between the Firmicutes phylum and Chlamydiae, Bacteroides, and a part of Spirochaetes than the traditional taxonomy indicates. In addition, it does not show the Mollicutes section as monophyly, and the overall bootstrap probabilities of the domain content tree are lower than those of the domain organization tree. These results demonstrate that the tree constructed from the domain organizations is superior to the tree constructed from the domain contents alone. This is mainly because the change of domain organization repertoires is caused not only by the gain or loss of a domain but also by the reorganization of existing domains, whereas the change of domain contents is related only with the former.

1186 Fukami-Kobayashi et al.

Bacteria
-proteobacteria
0.1
rcon0 rpro0 xfas0 atum0 xfas1 atum1 xcam0 -proteobacteria mlot0 bsui0 xaxo0 Spirochaetes bbac0 smel0 bmel0 cbur0 bjap0 rpal0 lint0 rbal0 lint1 ccre0 pgin0 bthe0 ctep0 cdip0 blon0 ceff0 mbov0 cglu0 mtub0 mtub1 mlep0 scoe0 save0 twhi0 twhi1 drad0 tthe0 ecol3 sfle1 ecol1 ecol2 buch0 styp0 baph1 sfle0 baph0 bflo0 styp1 plum0 hduc0 wbre0 styp2 ypes1 hinf0 ypes2 pmul0 ecol0 ypes0 vcho0 vvul0 vvul1 vpar0 sone0 psyr0 pput0 paer0 nmen0 cvio0 nmen1 neur0 bper0 bbro0 bpar0 rsol0 hpyl1 hpyl0 hhep0 cjej0 wsuc0 aaeo0 gsul0

-proteobacteria -proteobacteria -proteobacteria -proteobacteria

Bacteroidetes

Actinobacteria

Euryarchaeota
mace0 mmaz0 mmar0 mkan0 mjan0 mthe0

-proteobacteria
ccav0 ctra0 cmur0 cpne2 cpne1 cpne3 cpne0

DeinococcusThermus

Chlamydiae
tmar0 tten0 cace0 cper0 ctet0 fnuc0 gvio0 pmar2 anab0 pmar1 telo0 syne1 syne0 pmar0 tpal0 bbur0

Archaea

aful0 paby0 phor0 pfur0 aper0 paer1 ssol0 taci0 halo0 stok0 tvol0

Spirochaetes

Crenarchaeota

Cyanobacteria

bsub0 oihe0 bhal0 sepi0 bant0 saur1 bcer1 saur0 bcer0 saur2 lmon0 linn0

Clostridia

Bacillales
ncra0 spom0 scer0 atha0 osat1

mpen0 mgen0 spne1 mpne0 spne0 past0 mgal0 smut0 uure0 mpul0 saga1 saga0 lpla0 llac0 efae0 spyo2 ljoh0 spyo0 spyo1 spyo3

Mollicutes

Lactobacillales

Fungi

Viridiplantae

dmel0 cele0 cbri0 cint0 trub0 rnor0 hsap0 mmus0 drer0

Eukaryota

Metazoa

FIG. 5.A network based on the domain organization repertoires. The network was constructed with NeighborNet (Bryant and Moulton 2004) using the same evolutionary distances as used for the tree in gure 4. The network exhibits not only similar evolutionary relationship to that of the gure 4 tree but also some conicting signals harbored in the distance data, which are represented by many parallelograms in the network.

Construction of a Domain Organization Network Using the evolutionary distances computed from the Pfam domain organizations, we constructed a network as shown in gure 5 (see Supplementary g. S4, Supplementary Material online, for an enlarged version of the network). The network exhibits similar evolutionary relationships among the organisms to that of the Pfam tree. In addition, it reveals some conicting signals, which are represented by many parallelograms in the network. Discussion One of the new and strong points of our method is to use domain organization for construction of a phylogenetic tree. A tree of life necessarily incorporates all the available species from the 3 superkingdoms, Archaea, Bacteria, and Eukaryota. In addition, it is desirable to construct it from whole genome not from a limited number of genes. Therefore, a method for constructing a tree of life should be applicable to a huge amount of genome data. In our method, a protein domain is considered as an evolutionary unit,

instead of an amino acid residue. Such a coarse grained way greatly compresses the genome data and enables us to incorporate all the available species into a tree. Although protein domains do not represent the whole genome of an organism, our method shows one of the most feasible ways to cover a skyrocketing amount of genome data. Our method scales the change of a domain organization repertoire, that is, appearance and disappearance of domain organization in an organism, which is a rare event in the course of evolution (Fukami-Kobayashi et al. 1993; Apic et al. 2003; Fukami-Kobayashi et al. 2003). This feature is advantageous for elucidating phylogenetic relationships of remotely related species. On the other hand, our method treats more detailed aspects of evolution than the methods based on the domain contents because the domain organization considers not only domain contents of a protein but also their sequential order. The domain organization tree demonstrated a better result than the domain contents tree, which may suggest that domain contents does not give sufcient information and domain organization gives us moderately reduced information for construction of a tree of life.

A Tree of Life Based on Domain Organizations

1187

Our method has the following 6 features that would perhaps make it superior to the methods using gene or protein sequences in the construction of a tree of life. The rst 3 features are also inherent in other genome-based methods using gene contents, gene order, or total evolutionary distances. 1) It can incorporate the information of a protein even if the protein is not shared among the species in question. On the other hand, a sequence method can use only genes or proteins that are shared among all the species studied. When constructing a tree of superkingdoms, one has to use limited genes such as ribosomal RNA (rRNA), ribosomal protein, elongation factor Tu, and aminoacyl-tRNA synthetase (AARS) genes, which are almost exclusively involved in information storage and processing (Ciccarelli et al. 2006). 2) It does not require sequence alignment for the proteins in question. When comparing among superkingdoms, it is difcult to obtain reliable alignments even for the most conservative amino acid or nucleotide sequences. That is true of rRNAs (Hasegawa and Hashimoto 1993; Gupta 1998). In addition, the results of tree construction depend substantially upon which parts of the alignment are used as conservative regions (Hansmann and Martin 2000). 3) It is resistant to mutational saturation, base (or amino acid) composition biases, and heterogeneity of evolutionary rates among different evolutionary lineages. Those pitfalls prevent sequence methods from constructing the correct tree for remotely related species (Gribaldo and Philippe 2002). 4) The computational time required is roughly proportional to the number of genomes, because most of the time is spared for mapping domains in each genome. If we conduct a round robin blasting of all genes of a genome against the other, the computational time will be roughly proportional to the square number of genomes. As the number of genomes is increased, this problem will become more and more serious. 5) It does not require orthologous gene sets. Computation of orthologous gene sets is accompanied with numerous complicated and burdensome procedures, and certainly error prone. 6) It judges proteins homologous or not by the medium of protein domain organizations. It is practically difcult to dene homologous proteins using sequence similarity because the optimum cutoff criteria, Blast score/e value and coverage of homologous regions, for example, vary from protein to protein. Our method takes simpler and more feasible strategy. In addition, our method discriminates distantly related species in appearance due to genome shrinkage from that in reality and consequently places parasitic and symbiotic organisms in the appropriate taxonomic positions in a tree. Although some genome-based methods (Korbel et al. 2002; Kunin et al. 2005; Yang et al. 2005) achieve the correct positioning by normalizing genome sizes, our method achieve it by taking a geometric mean over the evolutionary distances (see Materials and Methods for details). However, no method is free from drawbacks, and our method is no exception. The factor that may most seriously inuence not only our method but also other genome-based methods in general is a massive amount of HGTs (Kunin and Ouzounis 2003) supposed to have occurred at the early stage of the cellular-organism evolution (Aravind et al. 1998; Brown 2003). There is now pretty strong evidence that there is a considerable amount of HGTs in genomes.

It has also been shown that many genes in eukaryotes originated from bacteria and mitochondria (Rivera and Lake 2004; Embley and Martin 2006). Recently, it has been reported that even vitally important genes such as the recA/RAD51 family genes were acquired via endosymbiotic gene transfer (Lin et al. 2006). Dutilh et al. (2004) tried to correct the HGT effect in their gene-content method, and showed that positioning of Halobacterium at the base of Archaea was attributed to extensive levels of directed HGT in this organism. As shown in the network of gure 5, our distance data computed from domain organization repertoire harbor some conicting signals, which mainly came from HGT. We have to keep in mind, therefore, that even a genome-based tree is just an approximation of evolutionary relationship of species and might not accurately describe their actual evolutionary process. This is a major issue in molecular phylogeny and is too often overlooked. Whereas the overall topology of our tree is in good agreement with the conventional one, there are a number of inconsistencies between them as already described in Results. Among them, the phylogenetic positions of hyperthermophilic bacteria are particularly noteworthy. Contrary to the 16S rRNA tree (Woese 1987; Brown 2003) or the trees constructed from protein sequences (Brown et al. 2001; Katoh et al. 2001), hyperthermophilic bacteria, A. aeolicus (aaeo0) and T. maritima (tmar0), are not the earliest branching lineages in our tree. Especially, tmar0 has traditionally been located at the deepest branch of eubacteria (Woese 1987; Brown 2003), whereas our tree shows that it is closely related to Clostridia in the Firmicutes cluster with 98% bootstrap value. On the other hand, aaeo0 is located next to Deltaproteobacteria, although its bootstrap value is low. We can reject the attribution of HGTs in this case: if HGTs had occurred in these hyperthermophilic bacteria at the early stage of the cellular-organism evolution, they would have moved the phylogenetic positions of these species to earlier ones. Only a hyperthermophilic bacterium, Thermus thermophilus HB27 (tthe0), is shown to have diverged very early in the bacterial lineage. Similar results on the phylogenetic positioning of hyperthermophilic bacteria are reported also in other genome trees (Korbel et al. 2002; Wolf et al. 2002; Yang et al. 2005). Regarding this issue, it is interesting that some conicting signals are seen between T. thermophilus HB27 (tthe0) and the Archaea cluster in gure 5. Our tree together with other genome trees, therefore, may urge to reconsider the conventional view that thermophilic organisms are the earliest cellular life-forms. The conventional view has been derived from gene trees constructed from nucleotide or amino acid sequences and could be suffered from nucleotide or amino acid composition biases characteristic to hyperthermophilic DNAs/proteins (Hasegawa and Hashimoto 1993; Cambillau and Claverie 2000; Fukuchi and Nishikawa 2001; Nakashima et al. 2003). The biases would make hyperthermophilic DNAs/ proteins of Bacteria and Archaea resemble in sequences and bring them together toward the root of their gene tree. Our method, as already discussed, is completely free from such biases. Other genome trees (Korbel et al. 2002; Wolf et al. 2002; Yang et al. 2005) are also free from such biases and show similar results to ours. Therefore, the

1188 Fukami-Kobayashi et al.

phylogenetic positions of hyperthermophiles in the present study are expected to be more realistic than the traditional ones. Of course, in order to draw the conclusive lineages, we would need more genome data of hyperthermophiles and other evidence from various biological viewpoints. Supplementary Materials Supplementary table S1 and gures S1S4 are available at Molecular Biology and Evolution online (http://www.mbe.oxfordjournals.org/). Acknowledgments We thank Mr S. Sakamoto for his help and advice on the use of GTOP. This work was supported in part by Grant-in-Aid for Scientic Research on Priority Areas (C) Genome Biology from the Ministry of Education, Culture, Sports, Science and Technology of Japan. Funding to pay the Open Access publication charges for this article was provided by the Japan Society for the Promotion of Science Grant #16255006. Literature Cited
Altschul SF, Madden TL, Schaffer AA, Zhang J, Zhang Z, Miller W, Lipman DJ. 1997. Gapped BLAST and PSI-BLAST: a new generation of protein database search programs. Nucleic Acids Res. 25:33893402. Apic G, Huber W, Teichmann SA. 2003. Multi-domain protein families and domain pairs: comparison with known structures and a random model of domain recombination. J Struct Funct Genomics. 4:6778. Aravind L, Tatusov RL, Wolf YI, Walker DR, Koonin EV. 1998. Evidence for massive gene exchange between archaeal and bacterial hyperthermophiles. Trends Genet. 14:442444. Bapteste E, Boucher Y, Leigh J, Doolittle WF. 2004. Phylogenetic reconstruction and lateral gene transfer. Trends Microbiol. 12:406411. Bateman A, Coin L, Durbin R, et al. (13 co-authors) 2004. The Pfam protein families database. Nucleic Acids Res. 32:D138D141. Benson DA, Karsch-Mizrachi I, Lipman DJ, Ostell J, Rapp BA, Wheeler DL. 2000. GenBank. Nucleic Acids Res. 28:1518. Brown JR. 2003. Ancient horizontal gene transfer. Nat Rev Genet. 4:121132. Brown JR, Douady CJ, Italia MJ, Marshall WE, Stanhope MJ. 2001. Universal trees based on large combined protein sequence data sets. Nat Genet. 28:281285. Bryant D, Moulton V. 2004. Neighbor-net: an agglomerative method for the construction of phylogenetic networks. Mol Biol Evol. 21:255265. Caetano-Anolles G, Caetano-Anolles D. 2003. An evolutionarily structured universe of protein architecture. Genome Res. 13:15631571. Cambillau C, Claverie J-M. 2000. Structural and genomic correlates of hyperthermostability. J Biol Chem. 275:3238332386. Ciccarelli FD, Doerks T, von Mering C, Creevey CJ, Snel B, Bork P. 2006. Toward automatic reconstruction of a highly resolved tree of life. Science. 311:12831287. Clarke GD, Beiko RG, Ragan MA, Charlebois RL. 2002. Inferring genome trees by using a lter to eliminate phylogenetically discordant sequences and a distance matrix based on mean normalized BLASTP scores. J Bacteriol. 184:20722080. Daubin V, Gouy M, Perriere G. 2001. Bacterial molecular phylogeny using supertree approach. Genome Inform. 12:155164.

Durbin R, Eddy S, Krogh A, Mitchison G. 1998. Biological sequence analysis: probabilistic models of proteins and nucleic acids. Cambridge: Cambridge University Press. Dutilh BE, Huynen MA, Bruno WJ, Snel B. 2004. The consistent phylogenetic signal in genome trees revealed by reducing the impact of noise. J Mol Evol. 58:527539. Eddy S. 2003. HMMER users guide version 2.3.2 [Internet] [cited 2007 March 12]. Available from: ftp://ftp.genetics.wustl.edu/ pub/eddy/hmmer/CURRENT/Userguide.pdf. Embley TM, Martin W. 2006. Eukaryotic evolution, changes and challenges. Nature. 440:623630. Felsenstein J. 1985. Condence limits on phylogenies: an approach using the bootstrap. Evolution. 39:783791. Fitz-Gibbon ST, House CH. 1999. Whole genome-based phylogenetic analysis of free-living microorganisms. Nucleic Acids Res. 27:42184222. Fukami-Kobayashi K, Tateno Y, Nishikawa K. 2003. Parallel evolution of ligand specicity between LacI/GalR family repressors and periplasmic sugar-binding proteins. Mol Biol Evol. 20:267277. Fukami-Kobayashi K, Tomoda S, Go M. 1993. Evolutionary clustering and functional similarity of RNA-binding proteins. FEBS Lett. 335:289293. Fukuchi S, Nishikawa K. 2001. Protein surface amino acid compositions distinctively differ between thermophilic and mesophilic bacteria. J Mol Biol. 309:835843. Garrity GM, Bell JA, Lilburn TG. 2004. Taxonomic outline of the prokaryotes release 5.0. Bergeys manual of systematic bacteriology. 2nd ed. New York: Springer. Gerstein M. 1998. Patterns of protein-fold usage in eight microbial genomes: a comprehensive structural census. Proteins. 33: 518534. Goff SA, Ricke D, Lan T-H, et al. (55 co-authors). 2002. A draft sequence of the rice genome (Oryza sativa L. ssp. japonica). Science. 296:92100. Graur D, Li W-H. 2000. Fundamentals of molecular evolution. Sunderland (MA): Sinauer Associates, Inc. Gribaldo S, Philippe H. 2002. Ancient phylogenetic relationships. Theor Popul Biol. 61:391408. Gupta RS. 1998. Protein phylogenetics and signature sequences: a reappraisal of evolutionary relationship among archaebacteria, eubacteria, and eukaryotes. Microbiol Mol Biol Rev. 62:14351491. Hansmann S, Martin W. 2000. Phylogeny of 33 ribosomal and six other proteins encoded in an ancient gene cluster that is conserved across prokaryotic genomes: inuence of excluding poorly alignable sites from analysis. Int J Syst Evol Microbiol. 50(Pt 4):16551663. Hasegawa M, Hashimoto T. 1993. Ribosomal RNA trees misleading? Nature. 361:23. Katoh K, Kuma K, Miyata T. 2001. Genetic algorithm-based maximum-likelihood analysis for molecular phylogeny. J Mol Evol. 53:477484. Kawabata T, Fukuchi S, Homma K, Ota M, Araki J, Ito T, Ichiyoshi N, Nishikawa K. 2002. GTOP: a database of protein structures predicted from genome sequences. Nucleic Acids Res. 30:294298. Korbel JO, Snel B, Huynen MA, Bork P. 2002. SHOT: a web server for the construction of genome phylogenies. Trends Genet. 18:158162. Kunin V, Ahren D, Goldovsky L, Janssen P, Ouzounis CA. 2005. Measuring genome conservation across taxa: divided strains and united kingdoms. Nucleic Acids Res. 33: 616621. Kunin V, Ouzounis CA. 2003. GeneTRACE-reconstruction of gene content of ancestral species. Bioinformatics. 19:1412 1416.

A Tree of Life Based on Domain Organizations

1189

Lin J, Gerstein M. 2000. Whole-genome trees based on the occurrence of folds and orthologs: implications for comparing genomes on different levels. Genome Res. 10:808818. Lin Z, Kong H, Nei M, Ma H. 2006. Origins and evolution of the recA/RAD51 gene family: evidence for ancient gene duplication and endosymbiotic gene transfer. Proc Natl Acad Sci USA. 103:1032810333. Murzin AG, Brenner SE, Hubbard T, Chothia C. 1995. SCOP: a structural classication of proteins database for the investigation of sequences and structures. J Mol Biol. 247: 536540. Nakashima H, Fukuchi S, Nishikawa K. 2003. Compositional changes in RNA, DNA and proteins for bacterial adaptation to higher and lower temperatures. J Biochem. (Tokyo). 133:507513. Rivera MC, Lake JA. 2004. The ring of life provides evidence for a genome fusion origin of eukaryotes. Nature. 431:152155. Rokas A, Williams BL, King N, Carroll SB. 2003. Genome-scale approaches to resolving incongruence in molecular phylogenies. Nature. 425:798804. Saitou N, Nei M. 1987. The neighbor-joining method: a new method for reconstructing phylogenetic trees. Mol Biol Evol. 4:406425. Snel B, Bork P, Huynen MA. 1999. Genome phylogeny based on gene content. Nat Genet. 21:108110. Tateno Y, Nei M, Tajima F. 1982. Accuracy of estimated phylogenetic trees from molecular data. I. Distantly related species. J Mol Evol. 18:387404. Tekaia F, Lazcano A, Dujon B. 1999. The genomic tree as revealed from whole proteome comparisons. Genome Res. 9:550557.

The Arabidopsis Genome Initiative. 2000. Analysis of the genome sequence of the owering plant Arabidopsis thaliana. 408: 796815. Wheeler DL, Chappey C, Lash AE, Leipe DD, Madden TL, Schuler GD, Tatusova TA, Rapp BA. 2000. Database resources of the National Center for Biotechnology Information. Nucleic Acids Res. 28:1014. Winstanley HF, Abeln S, Deane CM. 2005. How old is your fold? Bioinformatics. 21:i449i458. Woese CR. 1987. Bacterial evolution. Microbiol Rev. 51:221271. Wolf YI, Brenner SE, Bash PA, Koonin EV. 1999. Distribution of protein folds in the three superkingdoms of life. Genome Res. 9:1726. Wolf YI, Rogozin IB, Grishin NV, Koonin EV. 2002. Genome trees and the tree of life. Trends Genet. 18:472479. Wolf YI, Rogozin IB, Grishin NV, Tatusov RL, Koonin EV. 2001. Genome trees constructed using ve different approaches suggest new major bacterial clades. BMC Evol Biol. 1:8. Wolf YI, Rogozin IB, Kondrashov AS, Koonin EV. 2001. Genome alignment, evolution of prokaryotic genome organization, and prediction of gene function using genomic context. Genome Res. 11:356372. Yang S, Doolittle RF, Bourne PE. 2005. Phylogeny determined by protein domain content. Proc Natl Acad Sci USA. 102:373378. Yu J, Wang J, Lin W, et al. (117 co-authors). 2005. The genomes of Oryza sativa: a history of duplications. PLoS Biol. 3:e38.

William Martin, Associate Editor Accepted February 19, 2007