You are on page 1of 7

Downloaded from genome.cshlp.

org on November 3, 2012 - Published by Cold Spring Harbor Laboratory Press

The net of life: Reconstructing the microbial phylogenetic network
Victor Kunin, Leon Goldovsky, Nikos Darzentas, et al. Genome Res. 2005 15: 954-959 Access the most recent version at doi:10.1101/gr.3666505

Supplemental Material References

http://genome.cshlp.org/content/suppl/2005/06/22/15.7.954.DC1.html This article cites 37 articles, 21 of which can be accessed free at: http://genome.cshlp.org/content/15/7/954.full.html#ref-list-1 Article cited in: http://genome.cshlp.org/content/15/7/954.full.html#related-urls

Creative Commons License Email alerting service

This article is distributed exclusively by Cold Spring Harbor Laboratory Press for the first six months after the full-issue publication date (see http://genome.cshlp.org/site/misc/terms.xhtml). After six months, it is available under a Creative Commons License (Attribution-NonCommercial 3.0 Unported License), as described at http://creativecommons.org/licenses/by-nc/3.0/. Receive free email alerts when new articles cite this article - sign up in the box at the top right corner of the article or click here

To subscribe to Genome Research go to: http://genome.cshlp.org/subscriptions

© 2005, Published by Cold Spring Harbor Laboratory Press

1999. and genome conservation—a novel genome-based method combining gene content and sequence similarity (Kunin et al. where all edges are of a similar strength. sequence alignments (Doolittle 1981). Here we present a first attempt to reconstruct the history of the microbial world. Cambridge CB10 1SD.org on November 3. We propose that genes might propagate extremely rapidly across microbial species through the HGT network.3666505. Yet. fax 44-1223-494471. which is the most prevailing mode of inheritance (Kunin and Ouzounis 2003a). and use an ancestral state inference algorithm to map HGT events on the tree. Rokas et al.org . 2002. and a correct representation should reflect HGT events.” We use available tree reconstruction methods to infer vertical inheritance. We term the bulk of horizontal gene flow between tree nodes as “vines. 2 Corresponding author. recording both horizontal and vertical gene transfer. Tekaia et al.] Following the legacy of Darwin’s Origin of Species (Darwin 1859).genome. 2003). the bulk of the genes are still transferred by vertical gene transfer. Phylogenetic trees have been derived from compositional signatures (Fox et al. Ouzounis2 Computational Genomics Group. Makarenkov and Legendre 2004). Article and publication are at http://www. average ortholog similarity. 2001.org/cgi/doi/10. and genome conservation (see Methods) for OFAM data and gene content for STRING data. to avoid possible bias from a single tree reconstruction method. Similarly. we derived genomic trees with three independent methods: gene content. with a significant amount of horizontal gene transfer (HGT) present (Boucher et al. Gusfield et al. most current methods for phylogenetic reconstruction depict evolutionary history of organisms as a tree. We discuss the major properties of this complex phylogenetic network based on a multitude of genome comparisons.” and demonstrate that multiple but mostly tiny vines interconnect the tree.1 Leon Goldovsky. small-world nature. We also describe a weighting scheme used to estimate the number of genes exchanged between pairs of organisms. Martin 1999).genome. ISSN 1088-9051/05. Lin and Gerstein 2000. The European Bioinformatics Institute. 954 Genome Research www. including trees based on gene content (Fitz-Gibbon and House 1999. a finding with important implications for genome evolution. 2002.Downloaded from genome. California 94598. EMBL Cambridge Outstation. The summary of the evolutionary events reconstructed with each method is presented in Table 1. USA.1101/ gr. Kunin and Ouzounis 2003a). All these tree-like representations of evolution have an inherent drawback. 2012 . 1980). E-mail ouzounis@ebi. United Kingdom It has previously been suggested that the phylogeny of microbial species might be better described as a network containing vertical and horizontal gene transfer (HGT) events.cshlp. we present a first attempt to reconstruct such an evolutionary network. which we term the “net of life. 2004). but more like a tree.Published by Cold Spring Harbor Laboratory Press Letter The net of life: Reconstructing the microbial phylogenetic network Victor Kunin. [Supplemental material is available online at www. 2005). gene order (Korbel et al. its existence is not questioned. Article published online before print in June 2005. The strong influence of HGT led to a proposal that presentation of microbial phylogeny as a tree is inaccurate as instances of HGT are not recorded in this presentation (Doolittle 1999. mostly addressing recombination (but not HGT) as a form of nonvertical inheritance (Wang et al. 2004). using certain organisms as hubs. We demonstrate that vertical inheritance constitutes the bulk of gene transfer on the tree of life. 2003). Snel et al. For a scaffold depicting vertical gene transfer we use established tree reconstruction methods. or alignments of artificially concatenated conserved orthologs (Brown et al. With genome sequencing technology. a well-established consensus between evolutionary biologists is that the genomic history of most microbial species is mosaic. average ortholog similarity (Clarke et al.genome. we used two data sets: OFAM (see Methods) and groups of orthologs defined by STRING (von Mering et al. Here. and discuss the patterns of gene propagation through the network. 1999. Korbel et al. Thus. Walnut Creek.uk. 2002). methods based on complete genome sequences appeared. and Christos A. 2002). 1 Present address: DOE Joint Genome Institute. all phylogenetic reconstructions so far have presented microbial trees rather than networks. In analogy. It is evident that although HGT is readily detectable. 2003. demonstrate its scale-free.genome. dealing solely with vertical inheritance (Bapteste et al. and topological analyses of tree structure (Piel et al.). the net of life is not a grid. www. Nikos Darzentas. 2001. Our results strongly suggest that the HGT network is a scale-free graph. 2003). with robust branching stems connected by thin climbing vines. Results Data To ensure that our results are not affected solely by the orthology data (see Methods).org 15:954–959 ©2005 by Cold Spring Harbor Laboratory Press.org. Yet. the widely accepted view that the phylogenetic history of genomes should be represented as a network rather than a tree has not been realized yet. Although the quantification of the evolutionary effect of the HGT is still a subject of an ongoing debate (Snel et al. on which we document the instances of horizontal transfer that intertwine the tree.ac. Attempts to deal with this issue include algorithmic solutions for network-like tree reconstruction.

2003).630 635. due to lower coverage of genomes (Table 1). 2002). 2).e. Distribution of HGT vine widths. data Tree reconstruction method Organisms HGT events Gene loss transfers Also.cshlp. we had to select a meaningful threshold to depict the inWe have also examined HGT vines that are reported to be ferred events. particularly when two related organisms are positioned distantly on a tree. as well in that genes often travel between organisms as groups rather than other cases of HGT detected between species with similar life singletons (Boucher et al. We thus conclude that the HGT styles.005 88. Phylothe diameter of the network fluctuating between five and six. caused either by specific genetic mechanisms the underlying (vertical inheritance) tree from the net of life. Unlike the global properties of the network.” proConnectivity of the network viding a medium to acquire and redistribute genes in the microTo investigate the properties of the HGT network.943 288. We found Pirellula sp. The distribution of HGT vine widths. top of the list of (terminal) nodes with the largest number of HGT is shown in Figure 1.056 thus we limited the analysis to the STRING Gene content 98 9968 32. which are virtually identical and independent on the data set. All data sets and trees produce virtually partners. and possess complex lowest thresholds.Published by Cold Spring Harbor Laboratory Press Phylogenetic network reconstruction chitecture bias from our analysis and examined results consistent beOrthology Vertical tween different tree architectures. 2012 . the ing plant roots and forming—symbiotic in case of Bradyrhizobium network also demonstrates power law distribution of connectiv(Kiers et al. Furthermore.328 OFAM Gene content 165 36. 2004). since the tree architectures are different.225 leaves (terminal nodes) of the tree.. HGT vine width distribution When examining 165 microbial genomes for the network hubs.. and 10. japonicum and E. we experimented with several thresholds. and OFAM Genome conservation 165 39. Salanoubat at low thresholds. one HGT vines is observed between the Bradyrhizobium genus (or (a single HGT). possibly containing more falseet al. Table 1. we removed bial communities. ancestral OFAM Average ortholog similarity 165 39. Genome Research www. or by virtue of their close proximity to and interaction with other Since our inference of HGT vine widths is probabilistic (see Methspecies in their environmental niches. or Bradyrhizobium japonicum. 1). network is likely to have a power-law distribution of connectivity. Both bacteria are reported to have acquired large result of noise inevitable when a probability model is examined number of genes horizontally (Kaneko et al. following a power law provides certain hints for HGT events in this species (Glockner et (Table 2A). 2002. i. these hubs can serve as bacterial “gene banks. to investigate the connectivity of the HGT wide and consistent across data sets and trees. Careful analysis of the genes that are transferred positive instances. 2003). penetratWhen higher thresholds are chosen for the analysis. there is evidence for HGT between B.org 955 . the sequenced genomes from contemporary species. In conclusion. 2003) and parasitic in case of Ralstonia (Alfano and ity of nodes (Table 2B). and Erwinia carotovora always at the number of genes transferred between any two nodes on the tree. This power-law signal is obscured at the plants. 2003). identical frequency distribution (Fig. and thus be scale-free. Summary of settings and results from various experimental designs HGT champions We aimed at investigating the HGT network in search of hubs and the widest HGT vines.org on November 3. Interestingly. Irrespectively of the tree used and sometimes the broader Rhizobiales group) of Alpha Proteobactedata set. namely. We suggest that this deviation from the power law is a et al. ods). tude. genetically distant. carotovora in the literature (Streit et al. subsequently fixdifferent trees and data sets (Table 3).589 84. Genin and Boucher 2004)—relationships with the tree used (Fig. Both cause tumor-like structures. Incorrect tree architecture can cause the mistaken inference of high amounts of HGT. Thus. namely. five. One of the widest network. ated within the genome.834 640.genome. certain species came out on the top of the connectivity list We define the HGT vine width as a summary of all horizontal with a remarkable consistency between the results obtained from transfer events between two nodes on the tree. where many nodes appear to have high conmolecular mechanisms to interact with the host plants (Sawada nectivity.385 89.e. with the STRING data shifted by an order of magnial. inner nodes (i. We thus aimed to exclude tree ar- Figure 1.951 646. Our usage of thresholds higher than one for between the two bacteria can help to understand the mechaevidence of HGT is indeed reinforced by biological observations nisms of pathogen–host interactions in these species. both these species are soil bacteria. the exact number of predicted gene transfers between two nodes is highly dependent on the tree structure. once again irrespectively of the data set or Collmer 2004.Downloaded from genome. with ria and the Beta Proteobacterium Ralstonia solanacearum. the HGT network displays small-world behavior. the original genome report for Pirellula sp.791 states) are often incomparable.

93 2. the consistency between the results of medium. according to the four methods used y = a * xk.e. Scalefree networks display identical properties when any random subset of the complete network is sampled. the connectivity distribution appears as a decreasing straight line on a log-log scale (Fig.95 b K R2 Goodness-of-fit is expressed as the coefficient of determination (R2) defined as R2 = 1 SSE/SSM.84 2. In this case. In fact. 2004).org . we required a data set of orthologs across all currently sequenced species.8 6. or the preferential involvement of specific functional classes (Nakamura et al. This has a profound ecological meaning and strong implications for genome evolution. HGT vines require a third dimension. 2).8 11. However. the reconstruction and understanding of the net of life will improve with better phylogenetic representation of sequenced organisms. or geographical distribution.and high-confidence HGT vine width thresholds. This presentation has the limitation of an inherent inability to depict HGT events. 1997) to find best bidirectional hits across 165 microbial genomes in COGENT database release 184 (Janssen et al. Discussion The strongest limitation of the types of the network reconstruction presented here is the inability of the ancestral state inference methods to precisely establish the donor organism for a HGT event. as Supplemental material.25 1.54 2. Methods In order to reconstruct the phylogenetic network of microbial species. We used BLASTP (Altschul et al.7 9.org on November 3. 2012 . we predict that the initial donor and final acceptor organisms might have nothing in common in terms of phylogenetic origin. k) for (A) HGT vine widths (Fig. HGT more often occurs between related organisms. this type of HGT causes orthologous gene replacement.8 11. Connectivity of the HGT network. Often a HGT event is inferred across nodes of the tree that existed at different time periods. and SSM is the sum of squares around the mean. ecological niche.77 0. a small-world structure means that a substantially beneficial gene appearing in any organism can swing across species barriers and reach any other organism via a very small number of HGT events.94 0. Although the probabilistic schema described in the Methods section was designed to reduce the impact of this phenomenon. In this study. as well any input data used in this study.” Although this effect might influence the character of the inferred network. We are currently working on incorporation of detecting homologous HGT events in the phylogenetic network.96 0.genome. When the coverage is low. Our results suggest that the connectivity of microbial HGT network has a power-law behavior. including all species identifiers (Janssen et al. An example of such drawing is shown in Figure 3. based on the scale-free model. Thus.55 0. the average shortest path between any two of its nodes (termed “network diameter”) involves traversing only relatively few nodes. where beyond a tree drawn in the conventional two-dimensional space. Figure 2. 1997). The currently acceptable representation of phylogenetic data is in the form of a tree-like structure in a two-dimensional space.95 0. In a small-world network. and the prediction should be read as “the donor is a progeny of the node. We propose to represent the phylogenetic data in the form of a three-dimensional tree.93 2.. 2). indicates that the properties of the phylogenetic network reported here are genuine and realistic.Downloaded from genome. and communicate indirectly through the “hubs” in the network of life. we did not address this mechanism. In the context of the HGT network. participating nodes can be drawn closer in the third dimension. Another limitation is our inability to determine the correct path across organisms when multiple HGT events happened. this prediction of our hypothesis has an independent verification from the “experiment” of antibiotics-resistance genes that are known to spread extremely rapidly across species (Jacoby 1996). Although most of the reported instances of drug resistance involve pathogenic bacteria. 956 Genome Research www. When convergence of gene content is particularly high.Published by Cold Spring Harbor Laboratory Press Kunin et al. i. with real data from this study. 2003). followed by homologous recombination (Vulic et al.83 11. Rather than introducing new protein families into a genome. A method to correctly infer HGT donors should greatly improve reconstruction of the network.7 7.68 0.4 5. 1) and (B) the connectivity of the network (Fig. 2003). To eliminate paralogy. often referred to as a “dendrogram” (meaning tree-graph in Greek).0 1. The hubs of the HGT network presented here might partially result from the phylogenetic coverage of the sequenced species. we focus instead on events that introduce novel protein families into genomes. Table 2.9 2. where SSE is the sum of squared errors. GeneTrace determines the donor group of organisms rather than a particular donor species. The full tree is available in VRML format. multiple HGT events accumulate on long branches.68 0. suggesting that our conclusions should not be strongly affected by an ever-increasing number of genomes. where a = exp(b) = eb Method A Average ortholog similarity Gene content Genome conservation STRING B Average ortholog similarity Gene content Genome conservation STRING 6. A network in which connectivity of nodes distributes as a power-law has also scale-free and small-world properties. and an artificial “hub” might appear. The GeneTrace method applied here uses phylogenetic distribution as a marker of HGT events. identification of the exact order and direction of HGT events would drastically improve reconstruction of the network.cshlp.72 0. Parameters for the power-law distribution (b.88 2.

see Methods). 2012 . All methods are implemented as it appears definition. Three-dimensional representation of the net of life. The list of species representing the major hubs in the HGT network and their connectivity ranking in the three trees considered Figure 3. STRING adopts the similarity content conservation STRING Organism definition of orthologs as groups of Pirellula sp. Audit. Enright.org 957 . please refer to Supplemental material. another organism are represented (L. 1997). To reconstruct the microbial (HGT vine width threshold is set to 10. This data set is accessible at 2005). I.. we clustered these hits by using Markov clustering algorithm (MCL) used three methods of genome-based phylogenetic reconstruc(Enright et al. There are many methods for the reconstruction of phylogenetic trees (see Introduction). and the third combines the two meathe resulting protein families used for the analysis described sures to achieve maximum precision and contrast (Kunin et al. for full names. none guarantees we used only bidirectional best hits across genomes. all these methhttp://cgg.. herein as the “OFAM” data set. resulting in Average 98 prokaryotic species.M. To avoid biases generated by any single tree.ebi. Peregrin-Alvarez. N. N. as nodes on green branches. also known as clusters of orClostridium acetobutylicum 4 4 10 5 thologous genes (COGs) (Tatusov et Chromobacterium violaceum 6 10 9 Absent al. Darzentas. the widths of both the vertical inheritance branches and the horizontal inheritance vines correspond to the numbers of gene families transferred by either mechanism. gene content (Korbel et al. we used orthology information for 110 species from Table 3. Red lines correspond to the vines representing HGT. after excludortholog Gene Genome ing Eukaryotes. 2002). Goldovsky. Lópezservation of gene content. 2005). Jenssen. While being based on complete genomes.uk/services/ortho-fam/. P. For visualization purposes. A. Also.Published by Cold Spring Harbor Laboratory Press Phylogenetic network reconstruction the STRING database (von Mering et al.ac. Absent signifies absence of the organism in the input data.). phylogenetic tree. only values for HGT vine width >30 are shown. in prep. we required a ranking. et al. however. i. Inner nodes of the tree are ignored during the phylogenetic network. The tree backbone was generated by using the average gene similarity approach (see Methods). 2002). The first method derives phylogenetic distances from conD. 2002).e. and genome conservation (Kunin et al. J.genome.cshlp. We call larity between genomes. 2003). Certain key species and taxa are labeled. B. Archaea.Downloaded from genome. Bacteria are shown as nodes on cyan branches. Ahrén. Genome Research www.J. average gene similarsures that all genes that had at least one bidirectional best hit in ity (Clarke et al. from which we cross-linked 106 species to COGENT. The root is represented as a yellow sphere.org on November 3. the second uses only sequence simiBigas. We then 100% accuracy. ods produce phylogenies that are remarkably similar to the clasTo ensure that the results are not an artifact of the orthology sical 16S rRNA trees. The radius of the nodes is proportional to the estimated gene content size (in terms of number of gene families). 2 1 1 Absent homologs built from at least one tripBradyrhizobium japonicum 3 3 2 4 let of best-matching pairs of seErwinia carotovora 5 2 4 Absent quences. M. Smith. Cases. The exhaustive nature of this schema ention.

184: 2072–2080. Borzym.E. M. ———.. Itoh. D.org on November 3. Kawashima. Mosaic bacterial chromosomes: A challenge en route to a tree of genomes.A. Eddhu. Korbel. Rev. Type III secretion system effector proteins: Double agents in bacterial disease and plant defense. M. Genet. Nesbo. et al.. References Alfano. C. 2001. T. 42: 107–134. Genin. CAO acknowledges support from the UK Medical Research Council and IBM Research.. 25: 3389–3402. Goldovsky. Douady. J. Comput.. Idesawa.. Med...R. E. Science 284: 2124–2129. Leigh. T. Z.E. Sci. Clarke.. E. 2002). DNA Res. and Langley. and Collmer.F. and House. Case.Published by Cold Spring Harbor Laboratory Press Kunin et al. A. V... Miller. while the minimal number of edges required to connect all nodes (n) by HGT is n 1. Marshall. and Ouzounis. on the Genome Phylogeny Server (http://cgg.J. S. Douady. Enright.. Ludwig. J. M. T. Host sanctions and the legume-rhizobium mutualism.uk/cgi-bin/ gps/GPS. M... Janssen. Only results consistent across different trees and with consistently high jacknife scores (Kunin and Ouzounis 2003b) are considered robust. C... A. or 2/n. G. and Gojobori. An efficient algorithm for large-scale detection of protein families. Dyer. strain 1.G. G.. Bioinformatics 19: 1412–1416..genome. Bioinformatics 19: 1162–1168..B.. Hespell. R. P. Gusfield. B..J. GeneTRACE: Reconstruction of gene content of ancestral species. 2002. Murray.F.J. 2004. W.A. C. J.S. S. Optimal. 2002. Genome Res. as these can only identify recently acquired genes and are not designed to reconstruct early events. 2003. 2000. Annu. R. D. to avoid multiple counts of a single HGT event. Balch. A. Rev. and Doolittle.. Sasamoto. and Ouzounis. R.. and Charlebois. Glockner. D. Boudreau. 1859.. Tanner. Universal trees based on large combined protein sequence data sets. M.F. K. W. ———. M. H.. Minamisawa. Wolfe. Heitmann..O. Stackebrandt.O. Iriguchi. Genome Res. we then assign the value of 2/3 as a probability for each possible event to be depicted correctly (and 1/3 for an incorrect detection).. Lessons learned from the genome analysis of Ralstonia solanacearum. Walsh. et al... Thus. Consider a situation when a protein family appears twice in distant sections of a tree. Boucher.A. J.A. R. Rousseau. Audit. and Ouzounis. Brown.. Y.F. Bauer. W. Enright. M. Kunin. Annu. Sato. S... L. 1999... Measuring genome conservation across taxa: Divided strains and united kingdoms. as below.pl) and described elsewhere (Kunin et al... and Boucher. Ahren. 2002... Matsuda. We thus used GeneTrace—a method that identifies HGT from the phylogenetic distribution of protein families on the tree of life (Kunin and Ouzounis 2003b). we sum up all probabilities of transfer for each gene family transferred between the two nodes and term the resulting edge as “vine” and the weight of the edge as “vine width. Jacoby. Piel. Bioinformatics 19: 1451–1452.J. Boucher.. SHOT: A web server for the construction of genome phylogenies.. Trends Genet.. We thus adopted a schema for normalization of the number of transferred genes..T. S. previously labeled by GeneTrace as arising from HGT. 13: 1589–1594. R. Teeling. V. Huynen. The small-world dynamics of tree networks and data mining in phyloinformatics. K. Bioinform.R.. Sanderson.E.. C. Rev.T. K. K.J. Biased biological functions of horizontally transferred genes in prokaryotic genomes.. 2004. On the origin of species by means of natural selection.J. 2003. Kunin.Downloaded from genome. West. Kunin. A limitation of the GeneTrace approach to reconstructing HGT events is its inability to distinguish between the donor and the acceptor genomes (Kunin and Ouzounis 2003b). C. The balance of driving forces during genome evolution in prokaryotes. Acknowledgments We thank members of the Computational Genomics Group for useful discussions. Phylogenetic reconstruction and lateral gene transfer. L. Science 209: 457–463. V. Zhang. Just as there are many methods to reconstruct phylogenetic trees. 2003. 12: 406–411. P. 47: 169–179. We could not use methods that are based on identification of biased GC content or codon usage. R. R. 11: 195–212.. S. Comput.L. Magrum. W. V.. 1996...cshlp. In this case. I. 2003.A. R. 10: 808–818.J... Beck. Uchiumi. C.A. and patchy presence of the gene family in distantly related clades implies HGT.T. 2003. This gives us the probability that each of the edges describes a valid HGT event as (n 1)/(n(n 1)/2). 2003a. Nakamura. and Bork. T. Whole-genome trees based on the occurrence of folds and orthologs: Implications for comparing genomes on different levels. Gapped BLAST and PSI-BLAST: A new generation of protein database search programs. Biol. Natl.. Watanabe. Cases. Trends Microbiol.A. Annu.. 37: 283–328. Genet. Nucleic Acids Res. 9: 189–197. Doolittle... Martin. J. M. and Denison. COmplete GENome Tracking (COGENT): A flexible data environment for computational genomics. This method was shown to have at least 90% accuracy on simulated data (Kunin and Ouzounis 2003b) and at least 81% accuracy on biological data (Kunin and Ouzounis 2003a). 2002. Lombardot. P. efficient reconstruction of phylogenetic networks with constrained recombination. Nucleic Acids Res. being capable of reconstructing HGT events on most levels on the tree of life.. W. absence of a gene in some members of a clade indicates gene loss.. Fox.org . S. To describe the inferred sum of all HGT events between two nodes within an evolutionary net. and Donoghue. Y. 958 Genome Research www.F. L. C. Assuming equal probability for all possible scenarios.. 30: 1575–1584.. Genet. Gibson. P.. T. at least one HGT event may be necessary to explain the phylogenetic distribution of the family. J.J. N. Schaffer. Consider now a protein family that has three dispersed roots within a tree.. D. Annu. A. Beiko.. 2012 . Rev.. Makarenkov. M.. W. 100: 8298–8303. Whole genome-based phylogenetic analysis of free-living microorganisms. Nat. and Stanhope.. Y. G. Darwin. For STRING data. 1999. 33: 616–621. 18: 158–162. D. and Ouzounis.J. Thus.A. Italia. J. However. J. B. we used a gene content tree constructed according to (Korbel et al.” Altschul.L. The phylogeny of prokaryotes. C. Harte. 28: 281–285. we assign a probability of 2/n. Kiers. T.ac. Kaneko.D. Maniloff. at least two horizontal transfer events are necessary to explain the distribution.J.. C. Nat..A. K.. Bapteste. W. there are several available methods to identify HGT events. 2005. F. Thus. and Legendre. Papke.S. 2004. Nucleic Acids Res. Complete genomic sequence of nitrogen-fixing symbiotic bacterium Bradyrhizobium japonicum USDA110. Acad.. Bioessays 21: 99–104. Proc. R. S. 2005). M.. J. the number of all possible connections is n(n 1)/2. a gene that was extensively transferred horizontally creates links between all lineages that possess the gene. Fitz-Gibbon. Y. regardless whether they were involved in the particular transfer or not. Goldovsky. Phytopathol. 36: 760–766. From a phylogenetic tree to a reticulated network. Phylogenetic classification and the universal tree.J.L. W. Nucleic Acids Res. 42: 385–414. J..A. 2003b.. London.. Antimicrobial-resistant pathogens in the 1990s... 1980. E.ebi. Then.A. Ragan.. Similar amino acid sequences: Chance or common ancestry? Science 214: 149–159. A.H. 2004. 1997. Zhang.H. C. R. S. Lin.. and Gerstein. Phytopathol. simple linking of all nodes creates three possible edges for horizontal transfer. GeneTrace assumes that presence of a gene family in multiple members of a clade reveals its ancestral nature. Lateral gene transfer and the origins of prokaryotic groups.A. Bacteriol. Inferring genome trees by using a filter to eliminate phylogenetically discordant sequences and a distance matrix based on mean normalized BLASTP scores. A. M. Gade. J. 27: 4218–4222.. J. Snel. and Lipman. 2: 173–213. Biol. T. Nakamura. to each node that connects independent origins of a protein family. 1981. Van Dongen. Nature 425: 78– 81. Madden. H. 2004.E. Janssen. Kube. et al. and Doolittle. Complete genome sequence of the marine planctomycete Pirellula sp. C.. 2004. 1999..

E. 2004. 12: 17–25. http://cgg.org on November 3. J. 1997. Genome sequence of the plant pathogen Ralstonia solanacearum. Jaeggi.M. M.L. King.R.uk/services/ortho-fam/. 94: 9763–9767. accepted in revised form May 2. Genet. Proc. K. Appl. Genin. F. Lazcano. Genomes in flux: The evolution of archaeal and proteobacterial gene content. B. Tekaia.ebi. Dionisio. Natl.cshlp. Cattolico. ———. Liesegang. 1997. A. Raasch. and Broughton. Genome Res. Tatusov. Bacteriol. 2003. Taddei. C. M.. A. 49: 155–179. Brottier. J. and Snel. 2003.. Bork.uk/cgi-bin/gps/GPS. STRING: A database of predicted functional associations between proteins.genome. Perret. X. J. 9: 550–557. and Radman. 2003. Comput. D. 1999. Biol.. Vulic... Snel. M. A genomic perspective on protein families.Downloaded from genome. and Lipman. L. R. The genomic tree as revealed from whole proteome comparisons.. F... and Zhang.ac. F. M.. Molecular keys to speciation: DNA polymorphism and the control of genetic exchange in enterobacteria. M. Genome Research www. W. Nature 415: 497–502..J. 8: 69–78. M. J. Genome phylogeny based on gene content. An evolutionary hot spot: The pNGR234b replicon of Rhizobium sp. 186: 535–542. Gen.ebi.. Streit. R.. Nat.B. L.. P. Koonin. and Huynen. Nature 425: 798–804. Perfect phylogenetic networks with recombination. Genome Res.ac. 2001. Web site references http://cgg. 2012 . Wang.V. C. Billault. H..A. Genome Phylogeny Server. Schmitz. L. C.. 2002. Mangenot. et al.. Camus.... Science 278: 631–637. 2002. S. and Dujon.. W.. P. and Carroll.. Nucleic Acids Res. Bork..pl. Artiguenave. Zhang. S. 21: 108–110. Changing concepts in the systematics of bacterial nitrogen-fixing legume symbionts. N.. J.Published by Cold Spring Harbor Laboratory Press Phylogenetic network reconstruction Rokas. Sawada. L. Staehelin. Gouzy.L. B. 2005.. S. Acad..J. P.J. D.D. 31: 258–261.org 959 . Arlat. Schmidt. A. Microbiol. F. S.... Kuykendall. 1999. Williams. H.. Salanoubat. and Young.. Deakin. B.. von Mering. J.. OFAM data set.. Received January 6.C. B. Sci. Genome-scale approaches to resolving incongruence in molecular phylogenies... strain NGR234. W... 2005.A. Huynen.