Do different surrogate methods detect lateral genetic transfer events of different relative ages?

Update

4

TRENDS in Microbiology

proteins [6]. The two TatC proteins have evolved two distinct functions: one is required for secretion of the phosphodiesterase PhoD and the other is required for secretion of YwbN, a protein of unknown function [7,8]. Remarkably, Gram-positive bacteria almost universally contain only two types of Tat subunit: TatC plus a single subunit with characteristics of both TatA and TatB (here designated TatAC). The only exception is found in streptomycetes, which contain TatA, TatB and TatC [9]. Our recent studies have shown that the presence of only one TatAC and one TatC subunit is sufficient to sustain a functional Tat system in B. subtilis [8]. Unexpectedly, this organism contains two substrate-specific minimal Tat machines, each composed of a specific TatAC–TatC subunit pair. On this basis, it seems that TatAC proteins of B. subtilis and other Gram-positive bacteria perform the functions of both TatA and TatB of E. coli (Figure 2b). Although E. coli TatA and TatB share some sequence similarity, it was believed that they perform distinct functions and cannot substitute for each other. The validity of this view was first challenged by the use of hybrid RR-signal peptide–colicin V reporter proteins, which were still exported in a TatC-dependent manner by strains lacking either TatA or TatB [10]. Recently, Freudl and co-workers have demonstrated unambiguously in E. coli that TatA and TatC can form an active minimal translocase [11]. For this purpose, the authors employed an artificial plasmid-encoded minimal E. coli TatA–TatC translocase that mediates low-level translocation of a TorA–MalE reporter protein. Suppressor mutations mapping in the extreme N-terminal domain of TatA (Figure 1) were found to compensate strongly for the absence of TatB. These exciting observations imply that TatA is a bifunctional component of the Tat machinery, not only in Gram-positive bacteria but also in E. coli. The question of whether E. coli TatB is also intrinsically bifunctional remains to be answered, although this might be the case because minimal TatB–TatC translocases display activity in Streptomyces lividans [12]. Taken together, the assignment of E. coli-like

Vol.14 No.1 January 2006

TatA or TatB functions to given TatA or TatB homologues is difficult on the basis of sequence information alone. It is, therefore, a major challenge for future research to elucidate the molecular basis of TatA bifunctionality. Acknowledgements Funding for the project, of which this work is a part, was provided by the CEU project LSHG-CT-2004-005257.

References 1 Palmer, T. et al. (2005) Export of complex cofactor-containing proteins by the bacterial Tat pathway. Trends Microbiol. 13, 175–180 2 Robinson, C. and Bolhuis, A. (2001) Protein targeting by the twinarginine translocation pathway. Nat. Rev. Mol. Cell Biol. 2, 350–356 3 Mori, J. and Cline, K. (2002) A twin arginine signal peptide and the pH gradient trigger reversible assembly of the thylakoid DpH/Tat translocase. J. Cell Biol. 157, 205–210 4 Alami, M. et al. (2003) Differential interactions between a twinarginine signal peptide and its translocase in Escherichia coli. Mol. Cell 12, 937–946 5 Dilks, K. et al. (2003) Prokaryotic utilization of the twin-arginine translocation pathway: a genomic survey. J. Bacteriol. 185, 1478–1483 6 Jongbloed, J.D.H. et al. (2000) TatC is a specificity determinant for protein secretion via the twin-arginine translocation pathway. J. Biol. Chem. 275, 41350–41357 7 Pop, O. et al. (2003) Sequence-specific binding of prePhoD to soluble TatAd indicates protein-mediated targeting of the Tat export in Bacillus subtilis. J. Biol. Chem. 278, 38428–38436 8 Jongbloed, J.D.H. et al. (2004) Two minimal Tat translocases in Bacillus. Mol. Microbiol. 54, 1319–1325 9 de Keersmaeker, S. et al. (2005) Structural organization of the twinarginine translocation system in Streptomyces lividans. FEBS Lett. 579, 797–802 10 Ize, B. et al. (2002) In vivo dissection of the Tat translocation pathway in Escherichia coli. J. Mol. Biol. 317, 327–335 11 Blaudeck, N. et al. (2005) Isolation and characterization of bifunctional Escherichia coli TatA mutant proteins that allow efficient Tat-dependent protein translocation in the absence of TatB. J. Biol. Chem. 280, 3426–3432 12 De Keersmaeker, S. et al. (2005) Functional analysis of TatA and TatB in Streptomyces lividans. Biochem. Biophys. Res. Commun. 335, 973–982 0966-842X/$ - see front matter Q 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.tim.2005.11.001

Genome Analysis

Do different surrogate methods detect lateral genetic transfer events of different relative ages? Mark A. Ragan, Timothy J. Harlow and Robert G. Beiko ARC Centre in Bioinformatics and Institute for Molecular Bioscience, University of Queensland, Brisbane, QLD 4072, Australia

Non-tree-based (‘surrogate’) methods have been used to identify instances of lateral genetic transfer in microbial genomes but agreement among predictions of different methods can be poor. It has been proposed that this Corresponding author: Ragan, M.A. ([email protected]). Available online 13 December 2005 www.sciencedirect.com

disagreement arises because different surrogate methods are biased towards the detection of certain types of transfer events. This conjecture is supported by a rigorous phylogenetic analysis of 3776 proteins in Escherichia coli K12 MG1655 to map the ages of transfer events relative to one another.

Update

TRENDS in Microbiology

Vol.14 No.1 January 2006

5

Box 1. Surrogate methods for detecting laterally transferred genes Diverse methods have been used to identify genomic regions that, by certain criteria, seem anomalous and might, therefore, be of exogenous origin. Some methods focus on statistical features within a sequence ‘window’ of arbitrary length that is incremented along the genome [19,20]. More interesting in the present context, however, are methods applied to individual genes: lists of Escherichia coli K12 MG1655 genes that have been identified as anomalous using each of four gene-centric methods [7] are compared with the corresponding list resulting from rigorous phylogenetic analysis [11]. It has long been recognized [21] that a gene can differ from others in the same genome in nucleotide composition [9,10], the frequencies of specific DNA ‘words’ of length k (k-mers) [22,23] and codon frequencies [24]. These mutually interdependent measures are simple to compute but, in the absence of a model, are problematic to interpret [22]. Moreover, it is not usually possible to identify a probable source lineage from compositional anomalies alone, particularly because so little natural biodiversity has been sampled. The GCC-based method developed by Lawrence and Ochman (the ‘GC method’) [9] is an example of a compositional method. Such compositional features of genes can be abstracted as a Markov model [15,25]. The second surrogate method employed is the Markov model developed by Hayes and Borodovsky (the ‘MM method’) [15].

Introduction Lateral genetic transfer (LGT) mediated by plasmids and phage has been recognized for many years [1] and can confer advantages to bacteria in strongly selective environments such as hospitals and chemical disposal sites. However, it is increasingly argued that LGT has had a much more central role in shaping the content and functional repertoires of many microbial genomes than first thought [2–6]. Evidence for LGT takes many forms and the underlying methods to analyze its occurrence can be grouped into two broad categories: phylogenetic, methods that are explicitly based on the inference and comparison of phylogenetic trees, and surrogate [7], methods that are not tree-based, although some do make use of phylogenetic information (Box 1). Genetic material that has introgressed into a lineage might bear a signature of its exogenous origin, and surrogate methods are designed to detect this signature, for example, as an atypical nucleotide composition or an unexpected phyletic distribution (Box 1). In practice, it is difficult to predict exactly what lateral events different surrogate methods should be expected to detect because, in each instance, much depends on the statistical and evolutionary distinctiveness of the introgressed DNA compared with its new host genome. Indeed, Lawrence and Hendrickson [8] indicate methodological robustness as one of the four criteria that are most crucial for understanding the extent and impact of LGT. It is known, however, that when applied to the genome of Escherichia coli K12 MG1655, four surrogate methods – one from each of the main types described in Box 1 – identify fewer genes in common than would be expected under a simple stochastic model [7]. To explain this surprising result, it was hypothesized that some, or all, of these surrogate methods preferentially detect LGT events of different relative antiquity [7]. Initial differences in nucleotide composition (e.g. after the introgression of foreign DNA into a new host genome) can be ‘ameliorated’ (eroded away) over a period of time www.sciencedirect.com

Additional filters can be added to reduce the rate of false positives; for example, Nakamura et al. [25] eliminated genes with nucleotide contents that approximate a Markov model trained on the highly expressed region of genes that encodes ribosomal proteins in each genome. A gene of sufficiently distant exotic origin would be expected to exhibit a different pattern of pairwise similarity scores against its putative orthologs (reciprocal best matches) in other genomes than is typical of endogenous genes in the same genome. The phylogenetic discordance method of Clarke et al. (the ‘PD method’) [17] was used to identify anomalous patterns of pairwise matches. By analogy with construction of a tree from a distance matrix, an anomalous pattern of similarity relationships must specify a topologically conflicting tree [17]. It would be anomalous for a gene not to have orthologs among genomes of closely related taxa because its occurrence would be difficult to explain using a small number of non-lateral processes such as vertical descent, mutational sequence change and gene loss. Anomalous phyletic distributions are captured by the fourth surrogate method, the distributional profile method of Ragan and Charlebois (the ‘DP method’) [16].

ranging from decades to a few hundred million years [9,10] in response to the new regime of, for example, polymerases, repair machinery and tRNA abundances. Distinctive nucleotide compositions might be lost quicker than other features, making surrogate methods based on nucleotide composition preferentially efficacious in identifying recent LGT events or in identifying older events in which the introgressed DNA had an unusually pronounced compositional bias. Other surrogate approaches (e.g. those based on distributional profiles) might be less sensitive to fine-scale statistical features, as long as gene homology remains recognizable. A phylogenetic approach to LGT in the genome of Escherichia coli K12 The lack of consensus among surrogate methods led Beiko et al. [11] to assess LGT among prokaryotes using a computational pipeline (a series of high-throughput computational steps) to reconstruct the evolutionary history of putatively orthologous proteins. This approach was applied to 220 240 proteins in 144 genomes, yielding 22 432 individual phylogenetic trees, of which 19 672 were fully or partially resolved at 95% Bayesian posterior probability. The Bayesian posterior probability of a tree is equal to the frequency of that feature within a set of serially sampled trees, and is interpreted as the probability that the corresponding phylogenetic hypothesis is true. Among the 19 672 trees with some degree of resolution, 5823 were topologically incongruent with the reference supertree. A novel approach for comparing topologies [11] enabled identification of the nodal depth (i.e. relative age) of most individual LGT events inferred for each genome, including that of E. coli K12 MG1655. Intersecting these results with predictions from each of the four surrogate methods (see Ref. [7] and Box 1) leads to a test of the hypothesis that the different methods preferentially detect LGT events of different ages. The genome of E. coli K12 MG1655 encodes 4243 proteins, of which 3776 are clustered into ortholog families

Update

6

TRENDS in Microbiology

Vol.14 No.1 January 2006

Mapping the depth of acquisition events Each edit (instance of topological incongruence) that involves a direct ancestor of E. coli K12 MG1655 is classified as either ‘obligate’, that is, an ancestor occurs in all resolutions, or ‘possible’, whereby an ancestor occurs in at least one alternative resolution. In a minority of these instances, a lineage ancestral to E. coli K12 MG1655 could be identified as either the donor or the recipient of the laterally transferred gene, given the data. All instances of topological incongruence have been assigned to one of six classes. The first three classes contain those instances in which an ancestor is: necessarily the recipient (and not the donor) of a laterally transferred gene; necessarily the donor (and not the recipient); and necessarily involved but in an indeterminate way. The other three classes identify those cases for which an ancestral lineage was implicated among some,

Table 1. Counts of topological incongruence in Escherichia coli K12 MG1655 Class Recipient Donor Indeterminate a

Possiblea 73 76 587

Obligate 60 17 390

Excludes obligate.

[12] with four or more members each [13]. When compared to the reference supertree, 1617 of the resulting trees contain topologically incongruent bipartitions with R95% Bayesian posterior support. A minimal edit path (resolving the topological difference by inferring single or serial LGT events on the reference tree) can be inferred for 1550 of these trees. Within this set, an ancestor of the extant E. coli K12 MG1655 genome is potentially implicated in 1203 resolutions.

0.7

Proportion (count per number of proteins)

0.6

0.5

0.4

0.3

0.2

0.1

0 19

Aquifex

18 Firmicutes

17 Spirochaetes

16 Cyanobacteria

15 Actinobacteria

14 ε-proteobacteria

13 α-proteobacteria

12 β-proteobacteria

11

Xanthomonas

Shewanella

Vibrio

10

Coxiella

9

8

Pseudomonas

7

Haemophilus

6

Buchnera

Salmonella

5

Yersinia

4

3

E. coli CFT073

Shigella

E. coli K12 MG1655

2

E. coli O157:H7

1

0

Phyletic depth of inferred LGT event TRENDS in Microbiology

Figure 1. Phyletic depths of transfer events inferred using surrogate methods. The bars pertain to the set of genes identified as anomalous by each of four surrogate methods (GC content, red; Markov model, yellow; phylogenetically discordant sequences, green; distributional profiles, blue) and that were also identified as putative transfer events using a phylogenetic method [11] and could thus be assigned a phyletic (nodal) depth. The x-axis is a flattened representation of the supertree inferred by Beiko et al. [11], with Escherichia coli K12 MG1655 at the far left, and the root to the right of Aquifex. Branches from the ancestral lineage of E. coli K12 are shown with one of the genus, division or phylum names associated with that branch. The height of each bar represents the proportion of the intersection set for each surrogate method that might have been acquired at the phylogenetic depth indicated on the x-axis (numbers 1–19). Note that each individual gene could have more than one associated depth: for instance, if a protein tree contained no sequences from genus Yersinia, then it would be impossible to assign the corresponding transfer event uniquely to either edge four or edge five (xaxis). Although the entire range of possible transfer depths of such events is shown here, the statistical tests described in the main text assign a mean depth to each event (e.g. 4.5 in the hypothetical case described). www.sciencedirect.com

Update

TRENDS in Microbiology

but not all, alternative edit paths, either as recipient, donor or indeterminately. The counts in each class are shown in Table 1. Only events in which the E. coli K12 MG1655 genome is potentially the recipient are relevant to the hypothesis being tested here, so only the ‘recipient’ and ‘indeterminate’ classes were examined for overlap with surrogatemethod predictions. These four classes of LGT events were pooled and used to examine the intersections with the sets of genes identified by each of the four surrogate methods. It is appropriate to include the events of indeterminate direction because only transfers in which a direct ancestor of the E. coli K12 MG1655 genome was the recipient should feature in these intersections (because the surrogate methods were applied only to E. coli K12 MG1655). Figure 1 shows that the GCC-based (GC) and Markov model (MM) surrogate methods preferentially identified transfer events that have occurred since the divergence of the enteric bacteria, whereas the phylogenetic discordance (PD) and distributional profile (DP) methods found many events that precede the divergence of major proteobacterial and g-proteobacterial lineages. The mean depths of events identified by GC and MM were 4.3 and 5.5 nodes, respectively; PD and DP detected transfers with average depths of 6.8 and 7.0 nodes. However, the standard deviation of these mean values was w4.0 in each case, calling into question whether the observed differences are statistically significant. Two statistical tests were used to assess the significance of the differences among these mean depths. First, analysis of variance (ANOVA) on the pooled set of mean depths identified a significant difference between the groups (pw3.0!10K5, three degrees of freedom). Tests of assumptions showed no violation of equality of variance, and only slight violation of multivariate normality (for the GC and MM methods). As ANOVA is robust to even moderate violations, these results do not suggest the need for a non-parametric approach. Second, Tukey’s test showed that the mean depth of transfers found using the GC method is significantly less than the mean depths found using DP or PD (p!0.001 in both cases). The mean depth found using MM is significantly less than those found using DP or PD at p%0.07, although not at p%0.05. The difference in mean depths of events found using PD versus DP, or GC versus MM, is not significantly different (pO0.19 in both cases). The sensitivity of surrogate versus phylogenetic methods Lawrence and Ochman [14] argue that recognition of laterally transferred regions requires a suite of approaches. Methods based on nucleotide composition (such as GC [10] and MM [15]), for example, lose discriminative power rapidly as introgressed regions are ameliorated [9,10], whereas methods based on distributional profiles (DP [16]) or unusual patterns of sequence similarity (PD [17]) might fail to detect transfers among closely related lineages. Analysis of phylogenetically validated events in the genome of E. coli K12 MG1655 supports this interpretation, and is consistent with the proposal that laterally transferred regions of different www.sciencedirect.com

Vol.14 No.1 January 2006

7

antiquities are best discovered by coordinated use of complementary methods [14]. The events identified phylogenetically in this analysis do not constitute the complete set of genes that have been subjected to transfer because phylogenetic methods cannot reliably detect transfers between sister taxa in a tree and because the strict posterior probability threshold collapses many phylogenetic relationships for which support is weak. Although they are computationally much more intensive than surrogate methods such as these, phylogenetic methods are less dependent on subtle nucleotide-level signatures that could be unevenly distributed [18] and subject to amelioration. More importantly, the best methods of sequence alignment and phylogenetic inference are grounded in a model-based statistical framework. Thus, instances of LGT inferred from phylogenetic trees can be examined in a proper statistical context in terms of compatibility with key evolutionary assumptions and with respect to statistical support for the answer obtained. Acknowledgements We thank the Australian Research Council for support (grant number CE0348221).

References 1 de la Cruz, F. and Davies, J. (2000) Horizontal gene transfer and the origin of species: lessons from bacteria. Trends Microbiol. 8, 128–133 2 Ochman, H. et al. (2000) Lateral gene transfer and the nature of bacterial innovation. Nature 405, 299–304 3 Koonin, E.V. et al. (2001) Horizontal gene transfer in prokaryotes: quantification and classification. Annu. Rev. Microbiol. 55, 709–742 4 Boucher, Y. et al. (2003) Lateral gene transfer and the origins of prokaryotic groups. Annu. Rev. Genet. 37, 283–328 5 Jain, R. et al. (2003) Horizontal gene transfer accelerates genome innovation and evolution. Mol. Biol. Evol. 20, 1598–1602 6 Lerat, E. et al. (2005) Evolutionary origins of genomic repertoires in bacteria. PLoS Biol. 3, e130 7 Ragan, M.A. (2001) On surrogate methods for detecting lateral gene transfer. FEMS Microbiol. Lett. 201, 187–191 8 Lawrence, J.G. and Hendrickson, H. (2003) Lateral gene transfer: when will adolescence end? Mol. Microbiol. 50, 739–749 9 Lawrence, J.G. and Ochman, H. (1997) Amelioration of bacterial genomes: rates of change and exchange. J. Mol. Evol. 44, 383–397 10 Lawrence, J.G. and Ochman, H. (1998) Molecular archaeology of the Escherichia coli genome. Proc. Natl. Acad. Sci. U. S. A. 95, 9413–9417 11 Beiko, R.G. et al. (2005) Highways of gene sharing in prokaryotes. Proc. Natl. Acad. Sci. U. S. A. 102, 14332–14337 12 Harlow, T.J. et al. (2004) A hybrid clustering approach to recognition of protein families in 114 microbial genomes. BMC Bioinformatics 5, 45 13 Beiko, R.G. et al. (2005) A word-oriented approach to alignment validation. Bioinformatics 21, 2230–2239 14 Lawrence, J.G. and Ochman, H. (2002) Reconciling the many faces of lateral gene transfer. Trends Microbiol. 10, 1–4 15 Hayes, W.S. and Borodovsky, M. (1998) How to interpret an anonymous bacterial genome: machine learning approach to gene identification. Genome Res. 8, 1154–1171 16 Ragan, M.A. and Charlebois, R.L. (2002) Distributional profiles of homologous open reading frames among bacterial phyla: implications for vertical and lateral transmission. Int. J. Syst. Evol. Microbiol. 52, 777–787 17 Clarke, G.D. et al. (2002) Inferring genome trees by using a filter to eliminate phylogenetically discordant sequences and a distance matrix based on mean normalized BLASTP scores. J. Bacteriol. 184, 2072–2080 18 Bernaola-Galvan, P. et al. (2004) Quantifying intrachromosomal GC heterogeneity in prokaryotic genomes. Gene 333, 121–133

8

Update

TRENDS in Microbiology

19 Karlin, S. et al. (1997) Compositional biases of bacterial genomes and evolutionary implications. J. Bacteriol. 179, 3899–3913 20 Worning, P. et al. (2000) Structural analysis of DNA sequence: evidence for lateral gene transfer in Thermotoga maritima. Nucleic Acids Res. 28, 706–709 21 Me´digue, C. et al. (1991) Evidence for horizontal gene transfer in Escherichia coli speciation. J. Mol. Biol. 222, 851–856 22 Wang, B. (2001) Limitations of compositional approach to identifying horizontally transferred genes. J. Mol. Evol. 53, 244–250

Vol.14 No.1 January 2006

23 Hooper, S.D. and Berg, O.G. (2002) Gene import or deletion: a study of the different genes in Escherichia coli strains K12 and O157:H7. J. Mol. Evol. 55, 734–744 24 Daubin, V. et al. (2003) The source of laterally transferred genes in bacterial genomes. Genome Biol. 4, R57 25 Nakamura, Y. et al. (2004) Biased biological functions of horizontally transferred genes in prokaryotic genomes. Nat. Genet. 36, 760–766 0966-842X/$ - see front matter Q 2005 Published by Elsevier Ltd. doi:10.1016/j.tim.2005.11.004

Research Focus

Trafficking arms: oomycete effectors enter host plant cells Paul R.J. Birch1, Anne P. Rehmany2, Leighton Pritchard1, Sophien Kamoun3 and Jim L. Beynon2 1

Scottish Crop Research Institute, Invergowrie, Dundee, UK, DD2 5DA Warwick HRI, University of Warwick, Wellesbourne, Warwick, UK, CV35 9EF 3 Department of Plant Pathology, Ohio Agricultural Research and Development Centre, Ohio State University, Wooster, OH 44691, USA 2

Oomycetes cause devastating plant diseases of global importance, yet little is known about the molecular basis of their pathogenicity. Recently, the first oomycete effector genes with cultivar-specific avirulence (AVR) functions were identified. Evidence of diversifying selection in these genes and their cognate plant host resistance genes suggests a molecular ‘arms race’ as plants and oomycetes attempt to achieve and evade detection, respectively. AVR proteins from Hyaloperonospora parasitica and Phytophthora infestans are detected in the plant host cytoplasm, consistent with the hypothesis that oomycetes, as is the case with bacteria and fungi, actively deliver effectors inside host cells. The RXLR amino acid motif, which is present in these AVR proteins and other secreted oomycete proteins, is similar to a host-cell-targeting signal in virulence proteins of malaria parasites (Plasmodium species), suggesting a conserved role in pathogenicity.

Oomycete plant pathogens Oomycetes, despite apparently sharing morphological features with some fungal plant pathogens (including hyphae, appressoria, haustoria and spores), belong to the kingdom Stramenopiles and are, therefore, more closely related to brown algae and diatoms. Plant-pathogenic oomycetes are responsible for economically and environmentally devastating epidemics such as the 1846 Irish potato famine (caused by Phytophthora infestans) and the current sudden oak death epidemic (caused by Phytophthora ramorum) in California, USA [1]. Oomycetes Corresponding authors: Birch, P.R.J. ([email protected]), Beynon, J.L. (jim. [email protected]). Available online 13 December 2005 www.sciencedirect.com

can be host specific or can exhibit a wide host range. A biotrophic mode of nutrition that requires access to living host plant cells at an early stage in the establishment of infection is characteristic of most Phytophthora species and all Peronosporaceae (downy mildews) and Albuginaceae (white rusts). Peronosporaceae and Albuginaceae are obligately biotrophic and cannot be cultured easily. Oomycetes share with many bacterial, fungal and nematode plant pathogens the requirement for living host tissue for at least part of the infection cycle. To establish infection, these pathogens must evade, suppress or manipulate host defenses. This biotrophic requirement presents a point of vulnerability when all such pathogens are detected because invaded plant cells can induce resistance responses, including a localized programmed cell death called the hypersensitive response (HR) [2]. In this article, we highlight recent discoveries that improve the understanding of how oomycetes manipulate plant hosts and that reveal a dynamic evolutionary battle between plant and pathogen to achieve and evade detection, respectively. Plant–pathogen battle sites: the host cytoplasm and apoplast After 15 years of forward genetic studies, many resistance (R) proteins have been identified that function as a surveillance system to detect pathogen effectors [3]. Once detected, these effectors are termed avirulence (AVR) proteins. Each AVR protein is believed to be detected by a specific R protein in what is known as the gene-for-gene interaction, which often triggers the HR [2]. Many bacterial plant pathogens synthesize effector proteins and deliver them into host cells where they manipulate host defenses, including the HR [4]. Delivery