Updated clusters of orthologous genes for Archaea: a complex ancestor of the Archaea and the ...

15 Pages·2012·0.61 MB·English

Wolfetal.BiologyDirect2012,7:46
http://www.biology-direct.com/content/7/1/46
RESEARCH Open Access
Updated clusters of orthologous genes for
Archaea: a complex ancestor of the Archaea and
the byways of horizontal gene transfer
Yuri I Wolf*, Kira S Makarova, Natalya Yutin and Eugene V Koonin
Abstract
Background: Collections of Clusters of Orthologous Genes (COGs) provide indispensable tools for comparative
genomic analysis, evolutionary reconstruction and functional annotation ofnew genomes. Initially, COGs were
made for allcomplete genomes ofcellularlife forms that were available atthe time. However,with the
accumulation of thousands ofcomplete genomes, construction of a comprehensive COG set has become
extremely computationally demanding and prone to error propagation, necessitating the switch to taxon-specific
COG collections.Previously, wereported thecollectionof COGs for 41 genomes of Archaea (arCOGs).Here we
present a majorupdate of thearCOGsand describeevolutionary reconstructionsto reveal general trends in the
evolution of Archaea.
Results: The updated version ofthe arCOG database incorporates 91% of thepangenomeof 120 archaea (251,032
protein-coding genes altogether) into10,335 arCOGs. Using this new set of arCOGs, we performed maximum
likelihood reconstruction of thegenome content ofarchaeal ancestral forms and gene gain and loss events in
archaeal evolution. This reconstruction shows that the last Common Ancestor of the extant Archaea was an
organism of greater complexity than most oftheextant archaea, probably withover 2,500 protein-coding genes.
The subsequent evolution of almost all archaeal lineages was apparently dominatedby gene loss resulting in
genome streamlining. Overall, inthe evolution ofArchaea as well as a representative set of bacteria that was
similarly analyzedfor comparison, gene losses are estimatedto outnumbergene gains at least 4to 1. Analysis of
specific patterns ofgene gain in Archaea shows that, although some groups, in particular Halobacteria, acquire
substantially more genes than others,onthewhole, gene exchange between major groups ofArchaea appearsto
be largely random,with nomajor ‘highways’of horizontal gene transfer.
Conclusions: The updated collection of arCOGs is expectedto become a key resource for comparative genomics,
evolutionary reconstruction and functional annotation ofnew archaeal genomes. Given that, inspite of themajor
increase inthe number ofgenomes, theconserved core of archaeal genes appears to be stabilizing, the major
evolutionary trendsrevealed here have a chance to stand the testof time.
Reviewers: This article was reviewed by(for complete reviews seethe Reviewers’Reports section): Dr.PLG, Prof. PF,
Dr. PL (nominated by Prof. JPG).
Keywords: Archaea, Orthologs,Horizontal gene transfer
*Correspondence:[email protected]
NationalCenterforBiotechnologyInformation,NLM,NationalInstitutesof
Health,Bethesda,MD20894,USA
© Wolfetal.;licenseeBioMedCentralLtd.ThisisanOpenAccessarticledistributedunderthetermsoftheCreative
CommonsAttributionLicense(http://creativecommons.org/licenses/by/2.0),whichpermitsunrestricteduse,distribution,and
reproductioninanymedium,providedtheoriginalworkisproperlycited.
Wolfetal.BiologyDirect2012,7:46 Page2of15
http://www.biology-direct.com/content/7/1/46
Background been extensively used for functional annotation of new
Agenome-wide evolutionaryclassification ofgenesises- genomes (e.g., [19,20], comparative analysis of gene
sential for the entire enterprise of genomics including neighborhoods [21-23] and other connections between
both functional annotation and evolutionary reconstruc- genes, as implemented in the popular STRING tool [24];
tion. The construction of such a classification for a large target selection in structural genomics (e.g., [25]); and
set of diverse genomes is never an easy task due to the various genome-wide evolutionary analyses [6,8]. Subse-
complexity of evolutionary relationships between genes quently, the COGs have been employed as the seed for
to which gene duplication, gene loss and horizontal gene the EggNOG database that was constructed using
transfer (HGT) all make major contributions. The inter- improved algorithms for graph-based automatic con-
play of all these evolutionary processes makes accurate struction oforthologousgene clusters[26,27].
delineation of orthologous and paralogous relationships The methods for the construction of COGs and other,
between genes extremely complicated [1-3]. Accurate similar clusters of putative orthologous genes cannot
identification of orthologs and paralogs is central to guarantee correct identification of the orthologous and
functional characterization of genomes because ortho- paralogous relationships between genes due to the afore-
logs typically occupy the same functional niche in differ- mentioned complexity of the evolutionary processes.
ent organisms whereas paralogs undergo functional The original COG analysis of small numbers of genomes
diversification duplication via the processes of neofunc- involved the final step of manual curation that was im-
tionalization and subfunctionalization [3-5]. Clear differ- portant for detecting and resolving problems that were
entiation between orthologs and paralogs is equally not adequately addressed by the automatic procedure.
important for the reconstruction of evolutionary scenar- This step ceased to be feasible with the rapid increase in
ios[6-9]. the number of sequenced genomes whereas the compu-
In principle, orthologous and paralogous relationships tational cost of the analysis has steeply increased. There-
between genes have to be disentangled by means of fore, along with the development of improved, lower
comprehensive phylogenetic analysis of entire families of complexity algorithms for identification of orthologous
homologous genes in the compared genomes [2,10-13]. gene clusters [1,15,16], several smaller scale projects
However, for the case of numerous, diverse genomes, have been conducted in which COGs were constructed,
such comprehensive phylogenomic analysis remains annotated and analyzed in detail for compact groups of
both an extremely labor-intensive and an error-prone bacteria such as the Thermus-Deinococcus group [28],
process. Accordingly, several methods have been devel- Cyanobacteria [29], and Lactobacillales [19]. Along
oped that aimat the identification of setsof likely ortho- these lines, we have delineated the set of COGs for 41
logs without performing comprehensive phylogenetic genomes of archaea [30]; this data set that we denoted
analysis; benchmark comparisons indicate that some of arCOGs has become an important tool for archaeal gen-
these methodsperform aswell ifnot,insome cases,bet- omeanalysis [31-34].
ter than phylogenomic approaches [1,14-16]. Generally, Here we present a major update of the arCOGs that
these non-phylogenomic approaches in orthology infer- includes 120 archaeal genomes and use it for evolution-
ence are based on partitioning graphs of genome- ary reconstructions that seem to provide insights into
specific best hits for all genes (typically, compared in the majortrends ofarchaealevolution.
form ofproteinsequences)from the analyzed set ofgen-
omes. The key underlying assumption of this approach Results and discussion
is that the sequences of orthologous genes are more UpdateofarchaealCOGdatabase
similar to each other than to the sequences of any other The updated arCOG database includes protein sequences
genesfrom the comparedgenomes. from 120 completely sequenced genomes. Altogether,
The best hit graph approach, supplemented by add- 251,032protein-codinggenes(91%ofthetotalgenecom-
itional proceduresfordetectingco-orthologousgenesets plement) were assigned to 10,335 clusters. The coverage
and for treating genes encoding multidomain proteins, of individual genomes by arCOGs ranged from 99%
was first implemented in the Clusters of Orthologous (strains of Sulfolobus islandicus and Methanococcus
Groups (COGs) of proteins [17]; the acronym COG has maripaludis with abundant close relatives in the set) to
been subsequently reinterpreted to simply denote Clus- 73% (Nanoarchaeum equitans, the sole sequenced repre-
ters of Orthologous Genes [3]. The original COG set of sentative of the phylum Nanoarchaeota). In the current
1997 included only 7 complete genomes, all that were set of archaeal genomes, 129 arCOGs are strictly ubiqui-
available at the time [17]. The latest comprehensive tousand32morearCOGsareubiquitoustotheexclusion
COG collection released in 2003 incorporated ~70% of of N. equitans; in the original version of the arCOGs, the
the protein-coding genes from 69 genomes of prokar- corresponding numbers were 166 and 50 [30]. With the
yotes and unicellular eukaryotes [18]. The COGs have addition of new genomes, the size of the strictly universal
Wolfetal.BiologyDirect2012,7:46 Page3of15
http://www.biology-direct.com/content/7/1/46
gene set inevitably decreases due to lineage-specific gene genes [41-46]. The current set of 10,335 arCOGs
losses and possibly also annotation errors [35,36]. In real- includes 6,736 phyletic patterns of which 5,998 (89%)
ity, however, the sets of conserved archaeal genes could are unique. Similar to our 2007 observations, we found
be stabilizing. Indeed, analysis of the commonality distri- that the most common patterns represent genes that
bution [37,38] for the new arCOG collection gives esti- are conserved in well-defined archaeal clades, such as
mates for the size of the “core” (highly conserved) and all 120 Archaea, 4 Thaumarchaeota or 3 species of
the “shell” (moderately conserved) components of the Methanosarcinagenus (Table 1 and Figure 3).
archaeal pangenome that are almost unchanged since Analysis of the current arCOG set reveals the funda-
2007 (current estimates of ~220 and ~2,200 vs. ~230 and mental limitation of phyletic patterns as the basis for
~2,200,respectively,forthe2007arCOGset).Bycontrast, phylogenomic analysis. The early, small genomic data
addition of the new genomes substantially increased sets were readily amenable to direct pattern counting (e.
(from ~5,200 to ~7,400) the repertoire of rare archaeal g. in the original 1997 COGs that included 7 species,
genesthatbelongtothevariable“cloud”(Figure1). about1/3ofthepossible patternswith3ormorespecies
One of the immediate applications of clusters of were found in the actual data), the current set of 120
orthologous genes is phylogenomic reconstruction (i.e. archaeal species allows for ~1036 possible patterns of
identification of patterns of lineage-specific gain and loss which only a tiny fraction (~1/1032) are actually
of genes) for the respective group of organisms, in the observed. An overwhelming majority of the observed
case of arCOGs, the Archaea. For this reconstruction, a patterns are unique and even the most frequent patterns
‘species tree’ is required as a template. We employed the represent at most 1.6% of the arCOGs. Even under un-
arCOGs of ribosomal proteins to construct a maximum realistically restrictive models of gene content evolution
likelihood tree from a concatenated alignment of riboso- that prohibit horizontal gene transfer, random loss of
mal proteins (Figure 2, Additional file 1). Generally, the non-essential genes alone results in an exponential de-
tree agrees well with the archaeal taxonomy and with crease in the number of non-unique phyletic patterns
the recently published results of phylogenetic analysis with thenumber ofgenomesinthedata set.
including the monophyly of the ‘TACK superphylum’, a The rapidly increasing proportion of unique phyletic
large assemblage of archaeal phyla that includes Thau- patterns calls for a more coarse-grained comparison
marchaeota, Aigarchaeota (with the single current repre- whereby non-identical but similar patterns are treated as
sentative, Candidatus Caldiarchaeum subterraneum), members of the same group. However, standard cluster-
CrenarchaeotaandKorarchaeota[39,40]. ing and ordination techniques perform poorly on phy-
letic pattern data because of the difficulty of objectively
PhyleticpatternsofarCOGs assessing the similarity between the observed phyletic
The original 1997 study of the COGs [17] included the patterns of (ar)COGs. A proper similarity measure must
first analysis of phyletic patterns, i.e. patterns of take into account relationships that go beyond simple
presence-absenceofgenesfromagivenCOGinthegen- distances between 120-dimensional binary vectors pri-
omes of the analyzed organisms. Subsequently, phyletic marily because different dimensions are not independent
patterns proved useful in describing the evolutionary or correlated ina simple manner, but are connected bya
history of lineages and functional relationships between complex network of phylogenetic and environmental
relationships between the corresponding organisms. As
a step toward a more biologically meaningful analysis of
10000
phyletic patterns, we compared evolutionary scenarios
thatareimpliedbythesepatterns.
1000
s
G
O
of arC 100 GWaeinsr,elcoosnsesstrauncdteadnctehsetraplossttaetreisorinparroCbOaGbsilities of gene
o.
n presence in ancestral nodes and the probabilities of the
10
associated gene gain and loss events using the Count
method of Csűrös and Miklós[9]. The reconstruction
1
wasbasedonbinary patterns (i.e.ignoringparalogs),and
0 20 40 60 80 100 120
no. of genomes the topology ofthe ribosomalproteintree (Figure 2)was
Figure1AcommonalityplotforArchaealprotein-coding used as the guide. The results of the present reconstruc-
genes.DiamondsshowthenumberofarCOGsthatincludethe tion (Figure 4) that was based on 10,335 arCOGs repre-
givennumberofdistinctgenomes.Dashedredlines,decomposition sented in 120 species of archaea generally agree with the
ofthedataintothreeexponents(“cloud”,“shell”and“core”[37,38]);
earlier observations made with the original arCOG col-
solidredline:thesumofthethreecomponents.
lection using maximum parsimony [30] as well as the
Wolfetal.BiologyDirect2012,7:46 Page4of15
http://www.biology-direct.com/content/7/1/46
Methanosarcinales
Methanocella
Methanomicrobiales
Halobacteriales
Archaeoglobales
Thermoplasmatales Euryarchaeota
AciduliprofundumbooneiT469
Methanococcales
Methanobacteriales
MethanopyruskandleriAV19
Thermococcales
NanoarchaeumequitansKin4 M Nanoarchaeota
Sulfolobales
Desulfurococcales
Crenarchaeota
Thermoproteales
ThermofilumpendensHrk5
Thaumarchaeota Thaumarchaeota
Caldiarchaeumsubterraneum Aigarchaeota
CandidatusKorarchaeumcryptofilumOPF8 Korarchaeota
0.2
Figure2PhylogenyofuniversalArchaealribosomalproteins.TheapproximateMaximumLikelihoodtreewasreconstructedusingFastTree
[51,52].
outcome of the ML reconstruction reported by Csűrös (genome streamlining) during the diversification of re-
and Miklós [9]. The genome of the last archaeal com- centfamily-levelgroups (Figure 4).
mon ancestor (LACA) is inferred to have contained Gene gains that are associated with the ancestor of a
genes from at least ~1725 arCOGs of which 970 could clade conceivably reflect the innovations that led to the
be identified with high confidence (posterior probability diversification and define the biological characteristics
>0.9). Taking into account the characteristic level of par- that differentiate the given clade from other groups.
alogy and the number of genes in the transient “cloud” Among the Archaea, the maximum number of gene
(Figure 1),the genomesize ofLACAcan beestimated at gains was detected at LACA (Table 2 and Figure 3);
around 2600 genes, which puts this ancestor form on thesegainsnecessarilyincludebothgenesinheritedfrom
the high end of genomic complexity by the standards of the last universal common ancestor of cellular life and
the extant archaeal genomes (Additional file 2). The his- genes acquired during the stem phases of archaeal evo-
tory of most archaeal lineages appears to have involved lution. Among the internal branches of the archaeal
gradual, moderate genome growth or genomic stasis in tree, the branch leading to the common ancestor of
the deep branches followed by extensive gene loss Halobacteriales is associated with >1600 gene gains,
bringing the inferred size of the ancestoral halobacter-
Table1PhyleticpatternsinarCOGs ial genome to >3000 genes. Other branches character-
Patternfrequency Patterndescription No.ofSpecies ized by high rates of gene gain include Sulfolobales,
163 allThaumarchaeota 4 Thermococcales, Methanomicrobia and others. Of spe-
cial interest is the gain of 431 genes assigned to the
159 allMethanosarcina 3
common ancestor of the proposed TACK-superphy-
129 allArchaea 120
lum. Although the inference of gene gain depends on
116 allPyrobaculum+Thermoproteus 7
tree topology and therefore cannot be construed as
114 allHalobacterium 2
direct evidence of the monophyly of any group, such a
100 allHalobacteriales 16
high number of gains indicates a strong signal of
Wolfetal.BiologyDirect2012,7:46 Page5of15
http://www.biology-direct.com/content/7/1/46
420 Methanosarcinales
439 Methanocella
Methanomicrobiales
1,642 Halobacteriales
Archaeoglobales
416 Thermoplasmatales Euryarchaeota
Aciduliprofundum boonei T469
704 Methanococcales
Methanobacteriales
Methanopyrus kandleri AV19
705 Thermococcales
1,725 Nanoarchaeum equitans Kin4 M Nanoarchaeota
737 Sulfolobales
Desulfurococcales Crenarchaeota
Thermoproteales
Thermofilum pendens Hrk5
494 Thaumarchaeota Thaumarchaeota
431
Caldiarchaeum subterraneum Aigarchaeota
Candidatus Korarchaeum cryptofilum OPF8 Korarchaeota
0.2
1 7 1 3 9 1 3 9 9 6 4 0
0 4 4 3 8 2 6 5 2 1 1 0
9 1 3 4 1 3 1 1 1 1 1 1
1, 1,
Figure3PhyleticpatternsandinferredgenegainpatternsinArchaea.Themostfrequentphyleticpatternsareshowntotherightofthe
treeasblocksofgenomeswherethegeneispresent.Theinferrednumberofgenegainsisindicatedforthetreebranches.Themostfrequent
gainpatternsareshownasgreendotsassociatedwithtreebranches.ThepatternwithintheMethanosarcinalescladereferstothe
Methanosarcinagenus.Thefuzzypatterntotheleftofthetreerootreferstothe“zero-gain”inferencewithoutaconfidentassignmenttoany
particularclade.
shared gene content among the archaeal phyla that COG (0.87 in bacteria vs. 0.71 in archaea) and a higher
constitute theTACK superphylum. fraction of COGs with multiple (estimated total number
>1.5) gene gains (47% in bacteria vs. 39% in archaea).
ComparisonoftheratesofgenegainandlossinArchaea Thus, although the overall modes of genome-scale evo-
andBacteria lution are similar in both prokaryotic domains, intra-
For comparison with Archaea, we analyzed the phyletic domain gene exchange seems to have played a greater
patterns of 50 bacterial species in the 2003 COG data role in the evolution of bacteria. This finding does not
set [18]. The evolutionary scenarios were reconstructed appear surprising given the typical greater complexity of
using the Count software with the same parameters as bacterialcomparedtoarchaealcommunities.
used for Archaea and the ribosomal protein phylogeny
[39] as the guide tree topology (Table 3). Despite the dif-
ference in sampling density and breadth (the COG set of PatternsofgenegaininArchaea
2003 covers a relatively small fraction of the bacterial di- Phyletic patternsforminthe courseofevolutionbygene
versity), the estimated numbers of gene losses per COG gain (largely via HGT), vertical transmission through
and the gain-loss ratio are very similar for the two pro- speciation, and gene loss. Of these processes, gene gain
karyotic domains of life. Strikingly, Archaea and Bacteria seems to be of greatest interest. As shown above, gene
display essentially the same, four-fold excess of losses loss is much more frequent, appears to be largely ran-
over gains. For Bacteria, the reconstruction yielded dom in terms of which clades are affected and generally
20-25% more secondary gene gains (in addition to one appears to occur at approximately constant rate over
default gain in the respective last common ancestor) per long spans of evolution (a form of molecular clock) [6].
Wolfetal.BiologyDirect2012,7:46 Page6of15
http://www.biology-direct.com/content/7/1/46
Methanosarcinales
Methanocella
Methanomicrobiales
Halobacteriales
Archaeoglobales
Thermoplasmatales Euryarchaeota
<1000
AciduliprofundumbooneiT469
1000-1400
Methanococcales
1400-1800
Methanobacteriales
MethanopyruskandleriAV19 1800-2200
Thermococcales
2200-2600
NanoarchaeumequitansKin4 M Nanoarchaeota
260-30000
Sulfolobales
>3000
Desulfurococcales Crenarchaeota
Thermoproteales
ThermofilumpendensHrk5
Thaumarchaeota Thaumarchaeota
Caldiarchaeumsubterraneum Aigarchaeota
CandidatusKorarchaeumcryptofilumOPF8 Korarchaeota
0.2
Figure4InferredancestralgenomesinArchaea.Thesquareboxesatthebasesofcladesindicatethenumberoffamiliesintheinferred
ancestralgenomes;therectanglesatthetipsofcladesindicatethenumberoffamiliesintheextantgenomeswithintheclade.Squareboxesat
thetipsindicatesinglegenomes;rectangularboxesindicatemultiplegenomes.
Thus, we focus here on the patterns of gene gain, ignor- dimensional (given 120 analyzed genomes) binary
ingpossible subsequentgeneloss. presence-absence patterns, we obtain 238-dimensional
Unlike parsimony-based methods that reconstructbin- (the tree has a total of 238 branches, including the
ary evolutionary scenarios, Count produces profiles of branchleadingtotheroot)binarygainpatterns.
posterior probabilities of events. To convert these into Remarkably, despite the near doubling of the dimen-
patterns, we use the simple definition of a “likely event”: sionality, the likely gain patterns provide considerably
if a gene gain probability is >0.5 for a particular branch, more coarse-grained material for comparisons. The
we mark it as a likely gain. Thus, instead of 120- 10,335 arCOGs form 1,878 gain patterns, of which 77%
Table2InferredgenegainsinArchaea Table3Comparativeanalysisofgenegainsandlossesin
No.ofgains Clade ArchaeaandBacteria
1,725 LACA arCOGs bac.COGs(2003)
1,642 Halobacteriales families 10,335 4,149
737 Sulfolobales species 120 50
705 Thermococcales gains 17,680 7,769
704 deepEuryarchaeota(Eury-withoutThermococcales) gains/family 1.71 1.87
494 Thaumarchaeota acquisitions/family 0.71 0.87
439 Methanomicrobia losses 74,690 32,355
431 TACKsuperphylum losses/family 7.23 7.80
420 Methanosarcina loss/gainratio 4.22 4.16
416 Thermoplasmatales single-gain(<1.5) 61% 53%
Wolfetal.BiologyDirect2012,7:46 Page7of15
http://www.biology-direct.com/content/7/1/46
are unique (compared to 6,736 presence-absence pat- p<0.0001 for the internal branches; 36% of the internal
terns with 89% of them being unique). The most com- branch gains involve multiple branches) (Additional file 3).
mon (apart from the obvious overall winner, the single This observation, again, is compatible with a largely ran-
gaininLACA)gainpattern(Figure3andTable4),asin- dom gene exchange. Nevertheless, some branches appear
gle gain in the Halobacteriales ancestor, is inferred for to significantly deviate from the expected behavior: for ex-
1,189 arCOGs. With one exception, all frequent (>300) ample, 94 of the 109 arCOGs likely acquired by the com-
gain patterns involve a single gain. The only exception mon ancestor of Thaumarchaeota and Aigarchaeota were
is the “zero-gain” pattern that was assigned to 901 also acquired elsewhere, which is over than twice more
arCOGs. These genes are scattered among archaeal thanexpected;conversely,onthePyrobaculum-Thermopro-
clades in such a way that COUNT could not confi- teus clade, the number of multiple-event gains was almost
dently assign a gain to any particular branch (e.g. twicelessthanexpected(54outof265).
arCOG11374 is found in Methanobacterium sp. AL-21
and in Methanosphaera stadtmanae DSM 3091 that RoutesofgeneexchangeinArchaea
are neither sister nor very distant species; informally, it To explore the preferred routes of intra-archaeal gene
appears likely that this gene has been transferred from exchange, we focus on the 1,267 arCOGs with 2 pre-
bacteria to some Methanobacteriaceaeancestor). dicted gains not involving LACA. This pattern likely
The frequency distribution of the number of gains per indicatesasinglegeneexchangewithinarchaea(oralter-
arCOG, plotted either in the discrete form for likely natively, two independent acquisitions by different
gains (Figure 5a) or in the continuous form for the sums archaeal clades from outside of the domain; however,
of posterior probabilities (Figure 5b), indicates that the this is a distinctly less parsimonious solution). The most
single-gain patterns represent 61-67% of the data and frequently occurring 2-gain pattern (Table 6, Figure 6,
the number of arCOGs with multiple acquisitions Additional file 4) is the Thermoplasmatales-Sulfolobales
declines exponentially. The excellent fit of the number pair, which was found in 16 arCOGs; an unexpectedly
of gene gains to the exponential decay function implies large number of shared genes between Sulfolobus and
that horizontal transfer of genes of each arCOG occurs Thermoplasma, suggestive of preferential HGT, has been
inarandomfashion. reported previously [47]. Not surprisingly, many pairs of
clades frequently exchanging genes involve Halobacter-
Multiplegenegainsinarchaea iales, the archaeal group that seems to be generally most
The inferred multiple gene gains in the same arCOG pronetoHGT (see above).
could provide insights into the history of intra-domain If the paths of gene exchange are random, the number
gene exchange. Altogether there are 2,495 arCOGs with of exchanges between a pair of clades is expected to be
≥2 likely (posterior probability >0.5) gains (excluding proportional to the product of the numbers of gains on
those with a gene gain in LACA). These families are these clades. We found that these variables were indeed
implicatedinatotal of6,190gene acquisitions. correlated, relatively weakly but significantly (rS=0.39,
Over 20% of the multiple gains involve the ancestral p<0.0001 for 196 pairs occurring more than once).
branch of the Halobacteriales (Table 5) which is the ab- However, the most frequently observed exchanges be-
solute leader among Archaea with respect to involve- tween clade pairs are an order of magnitude more fre-
ment in gene exchange as either the donor or the quent than expected by chance, an overwhelmingly
acceptor. Other prominent gene exchange participants unlikely fluctuation for the majority of these patterns
are Thermococcales,Thermoplasmatales and Sulfolobales. (Additional file 4). Taken together, these findings indi-
Generally, the number of multi-gain events on archaeal cate that, although HGT between the lineages was not
tree branches strongly correlates with the overall number completely random, no major “highways” [48] of intra-
of gains (rS=0.83, p<0.0001 for the entire tree; rS=0.93, domain gene exchange existed in the history of archaea.
Atbest, there seem to exist weakly preferred “byways” of
HGT.
Table4InferredgenegainpatternsinArchaea
Patternfrequency Patterndescription No.ofSpecies Conclusions
1,189 Halobacteriales 16 The updated version of the arCOG collection incorpo-
rates 91% of the pangenome of 120 archaea into 10,335
1,147 LACA 120
arCOGs. This new set of arCOGs is expected to become
901 scattered(noconfidentgains) N/A
a key resource for comparative genomics, evolutionary
433 Sulfolobales 13
reconstruction and functional annotation of archaeal
341 Thermococcales 12
genomes that undoubtedly will be appearing at an in-
321 Methanosarcina 3
creasing pace. Notably, between this new arCOG
Wolfetal.BiologyDirect2012,7:46 Page8of15
http://www.biology-direct.com/content/7/1/46
a
1
0 2 4 6 8 10
0.1
cy 0.01
n
e
u
q
e
fr 0.001
0.0001
0.00001
number of gains
b
0 2 4 6 8 10
d.f.
p.
g
o
l
number of gains
Figure5Distributionofthegainpatternsbythenumberofgains.a. Numberofconfidentlypredictedgainsinapattern(p>0.5).b.The
sumofposteriorgainprobabilitiesinapattern(kernel-smoothedprobabilitydensity).Dottedline,thebest-fittingexponent.
collection and the original, 2007 version, the conserved subsequent evolution of almost all archaeal lineages was
genecoreofArchaeahasnotsubstantiallyshrunk,suggest- gene loss leading to genome streamlining. Overall, in the
ingthatthepresentcompositionofthiscorecouldbeclose evolution ofArchaeaaswell asa representativeset of bac-
todefinitive.Wedescribeheresomeresultsoftheongoing teria that we analyzed for comparison, gene losses are esti-
work on reconstruction of the genome content of archaeal mated to outnumber gene gains at least four to one. We
ancestral forms and gene gain and loss events. This recon- furtherinvestigatedthespecificpatternsofgenegaininAr-
structionclearlyindicatesthatthelastcommonancestorof chaea and found that, although some archaeal groups, in
the extant Archaea was a complex organism, most likely
with over 2,500 genes, and that the principal trend in Table6Inferredgenegainpatternsinvolving2Archaeal
clades
Table5InferredgenegainsinArchaea Patternfrequency Patterndescription
No.ofgains Clade 16 Thermoplasmatales–Sulfolobales
565 Halobacteriales 14 deepEuryarchaeota–Thaumarchaeota
296 Thermococcales 13 Halobacteriales–Thaumarchaeota
247 Thermoplasmatales 13 deepEuryarchaeota–Methanomicrobiales
245 Sulfolobales 12 Thermoplasmata–Thermococcales
204 Thaumarchaeota 12 Halobacteriales–Methanocella
199 deepEuryarchaeota 12 Halobacteriales–Thermoplasmatales
166 Archaeoglobales 12 Halobacteriales–Methanohalobiumevestigatum
123 TACKsuperphilum 11 Halobacteriales–Archaeoglobusfulgidus
104 Methanobrevibacterruminantium 11 Halobacteriales–Archaeoglobales
98 Methanosarcina 10 Halobacteriales–Thaum-+Aigarchaeota
Wolfetal.BiologyDirect2012,7:46 Page9of15
http://www.biology-direct.com/content/7/1/46
MethanohalobiumevestigatumZ 7303
Methanosarcinales
Methanosarcina
Methanocella
Methanomicrobiales
HalorhabdusutahensisDSM 12940
Halobacteriales
Halobacterium
Archaeoglobales ArchaeoglobusfulgidusDSM 4304
Thermoplasmatales Euryarchaeota
AciduliprofundumbooneiT469
Methanococcales
MethanobacteriumAL 21
Methanobacteriales
MethanobrevibacterruminantiumM1
MethanopyruskandleriAV19
Thermococcales
NanoarchaeumequitansKin4 M Nanoarchaeota
Sulfolobales
Desulfurococcales
Crenarchaeota
Thermoproteales
ThermofilumpendensHrk5
Thaumarchaeota Thaumarchaeota
Caldiarchaeumsubterraneum Aigarchaeota
CandidatusKorarchaeumcryptofilumOPF8 Korarchaeota
0.2
Figure6ThebywaysofhorizontalgenetransferamongArchaea.Linesconnectthecladesthatformthemostfrequentphyleticpatterns
withtwoinferredgains(differentcolorsareusedforvisualdifferentiationonly).OneofthetwocladesisthelikelyoriginoftherespectivearCOG
andtheotheristhelikelyacceptoroftheHGT.
particularHalobacteria,aremorepronetogeneacquisition algorithm [1]ontheresultsofall-against-allBLAST
than others, on the whole, gene exchange within Archaea [49]search.
appears to be largely random, so that there are no major (cid:1) Multiple alignments of the initial cluster
‘highways’ofgenetransfer. members constructed using the MUSCLE
The arCOG data is available at <ftp://ftp.ncbi.nih.gov/ program [50] were used as PSSMs for a PSI-
pub/wolf/COGs/arCOG/>. BLAST search [49] against the database of
Archaea proteins; significantly similar proteins
Methods (domains) were added to the corresponding
ConstructionofarchaealCOGs original clusters.
Protein sets for 120 completely sequenced genomes of (cid:1) The clusters with approximatelycomplementary
Archaea were downloaded from the NCBI FTP site. phyleticpatterns and highinter-cluster sequence
New members of the previously established arCOGs similarity weremerged.
were identified using the PSI-BLAST search with core
arCOG alignments as the PSSMs. New arCOGs were
constructed largely as previously described [30]. Briefly, Phylogeneticanalysis
theprocedureinvolved thefollowing steps: Themaximumlikelihoodphylogenetictreeofthe conca-
tenated alignment of ribosomal proteins universal for ar-
(cid:1) Initial clusters basedontrianglesofsymmetrical chaea was constructed using the FastTree program
besthitswere constructed usingamodified COG [51,52]aspreviouslydescribed [39].
Wolfetal.BiologyDirect2012,7:46 Page10of15
http://www.biology-direct.com/content/7/1/46
Reconstructionofgenegainandlosseventsduringthe sensulato?Inotherwords,couldwedefinealessarbitrarybarrierforthe
evolutionofArchaea phylumlevelwithinthearchaeabasedonphylogenomicinformation?A
commentalongthoselineswouldbewelcome.
Reconstruction of gene gain and loss during the evolu-
tion of Archaea was performed using the program Authors’response:Wecannotandshouldnotaddressissuesofformal
COUNT [53]. Maximum likelihood inference for gene taxonomyinthisarticle.Whatreallymattersisthecladisticpattern;which
cladesaregiventhestatusof“phylum”andwhicharesubsumedwithinother
presence, gain and loss at all branches of the phylogen-
phylamattersonlyoperationally.
etictreewasobtainedfor2-categoryevolutionarymodel.
-Theauthorsdetectahigherintra-domaintransferinbacteriathanin
archaea,anddonotfindit“surprisinggiventhetypicalgreatercomplexityof
Additional files
bacterialcomparedtoarchaealcommunities”.However,itisnotself-evident
thatgreatercommunitycomplexitynecessarilycorrelateswithmore
Additionalfile1:Thearchaealribosomephylogeny.The horizontalgenetransfer.Mostlikely,thereisanimportantphylogenetic
phylogenetictreereconstructedforconcatenatedalignmentsof effect,withsomelineagesbeingmorepronethanothersevenwithin
ribosomalproteinsusingFastTreeprogram,Newickformat. bacteria.Theobservationofahigherintra-domaintransferinbacteriacould
bebiasedbythetaxonomicsamplingandalsothemorelimitedsampling
Additionalfile2:TheinferredLACAgenomecontent.Posterior
sizeforbacteria.Canthesepossibilitiesbediscarded?
probabilitiesforthepresenceofanarCOGinLACAgenome.
Additionalfile3:Gainsandmultiplegainsonthearchaealtree Authors’response:Thepossibilitythattheresultsofanycomparativegenomic
branches.ThesumsofposteriorgainprobabilitiesforallarCOGsand analysisareaffectedbytaxonomicsamplingbiaseffectivelycanneverbe
arCOGsinvolvedinmultiplegaineventsforthearchaealtreebranches discardedgiventhattheexistingsamplesrepresentatinysliceoftheentire
(nodenumberscorrespondtolabelsintheAdditionalfile1). existingbiodiversity.Nevertheless,someofthesamplingbiaseffectscanbe
Additionalfile4:High-frequencytwo-gainpatterns.Theobserved mitigatedbyusingthetree-basedprobabilisticinferenceofancestralstatesand
andexpectednumbersofthetwo-gainpatterns(nodenumbers evolutionaryevents(e.g.thepotentialeffectofthesamplingbiasintroducedby
correspondtolabelsintheAdditionalfile1;p-valuecomputedusing inclusionof7strainsofSulfolobusislandicusinourdatasetisreducedbythe
expectationunderthePoissondistribution). factthattheirgenecomplementsarenotcountedindependentlybutarelargely
sharedbythecommonbranchofS.islandicusandS.solfataricus).Regardless,
wetrustthatthiscommentandresponsewillremindthereaderofinevitable
Competinginterests incompletenessofcomparativegenomicanalyses.
Theauthorsdeclarethattheyhavenocompetinginterests.
-ThenumberofinferredancestralgenesforLACA,butalsoforthelast
Authors’contributions commonbacterialancestorisrelativelyhighandthenumberofgenelosses
KSMandYIWcollectedthedata;KSM,NY,EVKandYIWanalyzedthedata;YIW outnumbersgenegainsbyatleastafactorof4.Giventheimportanceof
andEVKwrotethemanuscriptthatwasreadandapprovedbyallauthors. horizontalgenetransferinprokaryoticevolutionandthefactthatmany
genesevolvebyduplicationandneofunctionalization,theideaofagene-
Reviewers’reports richancestralgenomeevolvingbyimportantgenelossseemsabitodd.Is
thisrealordoesthisresultfromourincapacitytoidentifygenetransfer
Reviewer1:Dr.PurificacionLopez-Garcia,CentreNationaldela eventsand/orfailuretoidentifysometypeofgenecoalescentprocessesin
RechercheScientifique,France thepast?Whatarethepotentialbiasesofthisinferredvalue?
Thismanuscriptpresentsanupdateofthearchaealclusteroforthologous Authors’response:Theideaofadeepancestorthatwasmorecomplexthan
genes(arCOGs)basedon120archaealgenomesequenceswhich,although anaveragemodernmemberofthedescendantcladeprobablyis“abitodd”(or
stilllimitedfacetothewidearchaealdiversity,isalargeimprovement evenmorethanabit)tomanybiologists.Wesubmit,however,thatthisoddity
comparedtotheinitialsetofgenomesusedtodefinethefirstarCOGs.The isprimarilyavestigeofthetraditionalbeliefintheincreaseofcomplexitywith
arCOGsbecomethereforemuchmorecomprehensiveand,onthissole evolutionarytime,anotionthatmightbeintuitivebutisnotactuallyborneby
basis,theworkisalreadyimportantanduseful.Inaddition,theauthorstake anysubstantialevidence[54].Giventheimportanceofgenomestreamlining,
advantageofthisanalysistomakesomeinferencesaboutarchaealgenome especiallyintheevolutionofprokaryotes,thereisnoreasontoaprioridismiss
evolution.Theauthorsestimateinthiswaythenumberofgenesatancestral thepossibilitythatgenelossonaverageoutweighsgenegaincausedbyHGTas
nodes(includingthelastcommonarchaealancestor,LACA)andthe wellasduplicationfollowedbyneo/subfunctionalization(thelatterprocess
estimatednumberofgenegainsatdifferentnodesonaguidephylogenetic beingrelativelylessimportantinprokaryotes).Theinferenceofthecomplex
treebasedonhighlyconservedribosomalproteins.Thenumberofgenes ancestoris“real”inthesensethatthisisthemaximumlikelihoodestimate
sharedbydistantarchaealgroupsisusedtoevaluatethelevelofintra- basedonthebestpossibleestimationoftheparametersoftheevolutionary
domaingenetransferandcomparedwiththatobservedina(smaller)setof process(genegainandlossrates)fromtheavailabledataonphyleticpatterns.
bacterialgenomes.Overall,Ifindthearticleinterestinganduseful. Thus,HGTistakenintoaccountimplicitlyhere,underthegenegainrate.What
Ihaveafewcommentsthattheauthorsmightwanttoaddress: ismeantby“sometypeofgenecoalescentprocessesinthepast”isnotquite
cleartous.Admittedly,thecurrentinferencemethodologyissomewhatcrude,
-WolfetalreconstructanarchaealphylogenybasedonthearCOGsof beingbasedonphyleticpatternsandnottakingintoaccountthephylogenies
ribosomalproteinsasguideforsubsequentinferencesofgenomeevolution. ofindividualgenes.Suchacomprehensiveanalysisisbeyondthecapabilitiesof
Accordingtotheauthorsthetreeagreeswithknowntaxonomicschemes thecurrentlyaccessiblemaximumlikelihoodmodels.However,giventhe
andpointstothemonophylyofthe“TACK”supergroup(Thaumarchaeota– apparent(approximate)randomnessoftheHGTprocesses,itappearshighly
Aigarchaeota–Crenarchaeota-Korarchaeota).Thisisdifficulttosayfroman unlikelythatifandwhensuchanalysisbecomesfeasible,itchangesthe
unrootedtree,becausetherootmightliesomewhereelseinthetree,e.g. conclusionsqualitatively.Withregardtothepossiblebiases,thisagaingoes
betweentheKorarchaeotaandtheremaining“TAC”orelsewhere,inthe backtothetaxonomicsamplinganditspotentialincompleteness.Iftomorrow
presenceofanappropriateoutgroup.Atanyrate,theTACKmonophyly adeepbranchofArchaeawithverysmallgenomesisdiscovered,thiswillresult
appearstobesupportedbyalargenumberofsharedgenes(431).Atthis inadownwardreassessmentofthecomplexityofLACA.
point,itmaybetimelytoconsideramoresystematics-orientedcomparison
intermsofthescaleofgenelossandgainandgenomeevolutionbetween Theinferenceofcomplexdeepancestorsisneitherentirelynewnorlimitedto
theEuryarchaeotaandthe“TACK”super-phylum.Areweinfrontof theevolutionofArchaea.Manyevolutionaryreconstructionsandtheoretical
comparableevolutionaryscalesintermsofphylogenomicbreadth?Ifthatis modelsofgenomeevolutionpointtoextensivegenelossasoneofthekey
thecase,couldtheTACKsuper-phylumbeassimilatedwithCrenarchaeota evolutionaryprocesses[55-57].Perhapsparticularlyimpressiveare