The quotation above neatly summarizes what is arguably one of the most important scientific discoveries of the twentieth century (rather remarkably, this quote is the entire abstract of Woese and Fox's groundbreaking article 1). So profound are its implications that the debate rages to this day: did Carl Woese and George Fox really discover a new domain of life, which is equal in status to bacteria and eukaryotes 23, or is it 'merely' an unusual branch of bacteria 4567? This debate is reflected even in the different names that, 25 years after their description as a distinct, third line of the evolution of life, are still applied to this group of organisms: on the one hand, archaea, in adherence with the three-domain interpretation, and on the other archaeabacteria, emphasizing the purported affinity with bacteria. Of course, Woese and Fox did not actually discover these unusual organisms; some of the would-be archaea have been known for decades and their unusual properties, such as extreme halophilic and extreme thermophilic phenotypes, have been described in considerable detail (see, for example, 8910). The revolutionary aspect of Woese and Fox's work was subtler and more profound: by comparing certain parts of the genomic sequences of various organisms, they came up with a three-domain classification of life, in which a group of prokaryotes they designated archaebacteria has been accorded the status of a distinct domain (subsequently renamed archaea, to emphasize the fundamental separation from other domains), on an equal footing with bacteria and eukaryotes. Numerous microbiologists had seen archaea before, but without Woese and Fox's foray into genome analysis no-one recognized these organisms for what they really were. Their way of comparing genome sequences was, by today's standards, extremely crude, as they analyzed not even sequences but oligonucleotide catalogues of rRNA genes. It is all the more astounding that the principal conclusion achieved with this 'primitive' approach stands to this day, 25 years and 16 complete (and several more nearly complete) archaeal genome sequences later (Table 1).

In the years following Woese and Fox's breakthrough 1, many unique features of archaea have become apparent. To begin with, many of these organisms thrive under conditions that, by the usual standards of biology, seem unimaginable, such as in the water in the vicinity of the hydrothermal vents called 'black smokers' heated to over-boiling temperatures and saturated with hydrogen sulfide, or in extreme salinity 111213. In the most extreme hyperthermophilic habitats, archaea are, in fact, the only detectable life forms. In more moderate environments, archaea coexist with bacteria and eukaryotes, and their ecological importance is being increasingly recognized 14. The first molecular biological studies showed that archaea are highly unusual and clearly distinct from bacteria at the molecular level. In particular, the structure of the membrane glycerolipids in archaea is different from that of bacterial and eukaryal cells, and archaea do not contain murein, the predominant component of bacterial cell walls 1516.

But the most striking differences between archaea and bacteria are seen in the organization of their information-processing systems. The structures of ribosomes and chromatin, the presence of histones, and sequence similarity between proteins involved in translation, transcription, replication and DNA repair all point to a closer relationship between archaea and eukaryotes than between either of these and bacteria 1718192021. Moreover, the key components of the DNA replication machinery - such as the polymerases involved in elongation and initiation and the replicative helicases - are not homologous, or at least not orthologous, in archaea and eukaryotes on the one hand, and bacteria on the other 1722. This observation led to the hypothesis that replication of double-stranded DNA as the principal form of replication of the genetic material was 'invented' twice, independently: once in bacteria and once in the ancestor of archaea and eukaryotes 2223. In contrast many - although not all - of the metabolic pathways of archaea more closely resemble their bacterial rather than eukaryotic counterparts 242526. These studies support the status of archaea as a distinct domain of life with specific connections to eukaryotes, and emphasize the unusual and unique nature of archaeal genomes.

The new age of archaea began in 1996 with the whole-genome shotgun sequencing of the first archaeal genome, that of Methanococcus (now Methanocaldococcus) jannaschii 27. The Methanococcus 'genomescape' at first looked largely mysterious, with clear functional assignments produced for only 38% of the genes 27. A more detailed computational analysis that pushed the methodology available at the time to its limits yielded general functional predictions for up to 70% of the genes, showing that a solid connection between the genomes of archaea and those of other, better known forms of life did exist 24. Nevertheless, the fact remained that, more than anything, the first sequenced archaeal genome revealed the depth of our ignorance of the biology of this remarkable group of organisms. Subsequent genome sequencing, while certainly less extensive than the devoted 'archaeologists' would wish, produced a rich sampling of genomes of taxonomically diverse archaea (Table 1). This set of completely sequenced genomes includes multiple representatives of the two major divisions of the archaea established by phylogenetic analysis of rRNA, namely the Euryarchaeota and the Crenarchaeota 3, as well as the principal ecological types of archaea, such as hyperthermophiles, moderate thermophiles, and mesophiles, as well as halophiles and methanogens; autotrophic and heterotrophic forms, and anaerobes and aerobes are also represented by multiple species (Table 1).

Some potentially important branches of archaea are still missing from sequence databases, however, such as the mysterious Korachaeota, which might have branched off the trunk of the phylogenetic tree prior to the divergence of the remainder of the archaea 28, and the equally intriguing Nanoarchaea that so far seem to have the smallest genomes of all known cellular life forms 2930. These lacunae notwithstanding, the available sampling of archaeal genomes is substantial and is complemented by an even greater diversity of bacterial and eukaryotic genomes that are available for comparative analysis. This article critically assesses the contribution of comparative genomics to our understanding of the functional systems of archaeal cells and their evolution. We pose the following question: what have we learned from comparisons of archaeal genomes that could not easily have been learned by other, more traditional approaches? We suggest some tentative answers, as we see them. What follows is a viewpoint from behind a computer terminal; we realize that, from the experimenter's bench, the perspective might be somewhat different.

From the beginning of comparative genomics, it has been obvious that genome comparisons will yield valuable functional and evolutionary information only within a framework of the rational classification of genes and proteins. In our view, perhaps the most natural form of such a classification is a system of orthologous gene sets, which allows a researcher to analyze the evolutionary fate of each individual gene 31. Orthologs are homologous genes that evolved from a single ancestral gene in the last common ancestor of the compared genomes, whereas paralogs are genes related via duplication within a genome 323334. When duplication(s) succeeds speciation, a family of paralogs in one species should be considered orthologous to the corresponding family in the other species 34. Insomuch as orthologous relationships are correctly defined, phyletic (or phylogenetic) patterns of orthologous gene sets help in the prediction of gene functions and provide clues to the prevailing trends in genome evolution (a phyletic pattern is defined, simply, as the pattern of representation of genomes in each orthologous set) 26313536. These phyletic patterns are captured in the database of Clusters of Orthologous Groups of proteins (COGs) 37, and here we use COGs for a systematic survey of archaeal genomes (most of the phyletic pattern analyses can be done directly on the COG website by using the phyletic pattern search tool 38).

The most common phyletic patterns found in archaea are shown in Table 2. Not unpredictably, the top pattern consists of the 313 COGs that are represented in all archaeal genomes sequenced so far. What is more remarkable is that this apparent conserved core of archaeal genomes has undergone only limited shrinkage since the time it was first defined by comparative analysis of four archaeal genomes 39 (Figure 1). Extrapolating from the effect (or rather the near lack thereof) of the latest additions to the collection of archaeal genomes on the size of the conserved core of archaeal genes, we are compelled to conclude that around 300 genes are shared by all archaea, encode essential functions and have not been subject to non-orthologous gene displacement during archaeal evolution (non-orthologous gene displacement is a widespread phenomenon whereby a gene responsible for an essential function is displaced by an unrelated or distantly related gene responsible for the same function 40).

Of the COGs represented in all archaea, 16 so far have no members from other domains of life and comprise a unique archaeal genomic signature, whereas 61 are exclusively archaeo-eukaryotic. The majority of the pan-archaeal genes are known to be involved in, or are implicated in, information processing, particularly translation and RNA modification (Figure 2). Strikingly, among the 61 COGs that are uniquely shared by archaea and eukaryotes, only two do not, technically, belong to the information-processing machinery (COG1936, a nucleotide kinase, and COG3642, a protein kinase typically fused to a metalloprotease domain); the 10 uncharacterized COGs in this category consist of proteins whose predicted biochemical activity (GTPase, methyltransferase or RNA-binding protein) suggests a role in translation or RNA modification.

Thus, phyletic pattern analysis strongly supports the identity of archaea as a distinct group of organisms with a stable, conserved core of genes that primarily encodes proteins involved in the replication and expression of the genome. Furthermore, there is clearly a subset of genes, again primarily associated with information processing, that is shared by archaea and eukaryotes, to the exclusion of bacteria; this is compatible with the archaeo-eukaryotic affinity suggested by phylogenetic analyses of rRNA and proteins involved in translation, transcription and replication. The fact that this archaeo-eukaryotic component is quantitatively small, however, shows that the process of evolution has been more complex than simple vertical inheritance and has involved extensive horizontal gene transfer (HGT) between archaea and bacteria, at least outside the core gene set 242541. An intensely mixing pool of genes coding for metabolic enzymes, structural components of the cell and other proteins outside the central information-processing machinery might have existed after the divergence of bacteria and archaea but prior to the separation of the major archaeal and bacterial lineages.

More recent HGT, which has emerged as a major aspect of prokaryotic evolution in general 26424344, was apparently prominent in all archaea, although gene exchange with bacteria seems to have been much less extensive in hyperthermophiles than in mesophiles such as Methanosarcina or even Halobacterium 4445. Apparent preferential HGT has been noticed between archaea and hyperthermophilic bacteria, such as Aquifex and Thermotoga; when compared to bacterial mesophiles these bacteria have many more proteins with greater similarity to archaeal than to bacterial homologs 4647. With HGT, or more precisely the pivotal role of HGT in evolution, remaining a controversial subject 48, this conclusion has been disputed on the grounds that Aquifex and Thermotoga might be early-branching bacteria retaining ancestral features in many protein sequences 49. But this argument seems untenable simply because of the obvious split of the gene complements of these bacteria into 'garden variety' bacterial genes and 'archaeal' genes 50. The reality of horizontal gene flow from archaea to thermophilic bacteria becomes even more tangible upon examination of the proteins encoded in the genome of Thermoanaerobacter tengcongensis 5152, which contains many more 'archaeal' genes than appear in other bacteria of the Bacillus-Clostridium group and to which the early-branching argument would not apply.

Although archaeal hyperthemophiles do not appear to have many genes acquired via HGT from bacteria, at least after the divergence of the archaeal lineages, horizontal gene exchange between archaea themselves might have been extensive. Strikingly, even within the conserved core of archaeal genes, major diversity of phylogenetic tree topologies has been observed (53 and Y.I. Wolf and E.V.K., unpublished observations). As noted by Nesbo and coworkers 53, "the notion that there is a core of nontransferable genes...has not been proven and may be unprovable". These findings do not invalidate the notion of a core of indispensable genes that are conserved across archaea but suggest a wide spread of xenologous gene displacement, whereby an essential gene is displaced by an ortholog from a distant lineage, typically via an intermediate stage of redundancy 44.

Other phyletic patterns that are common among archaea seem primarily to reflect HGT or gene loss prevalent in individual archaeal lineages (Table 2). Thus, Methanosarcina, a mesophile with by far the largest genome among the sequenced archaeal genomes, is represented in numerous COGs that have no other archaeal members but are present in various groups of bacteria. This organism, which coexists with a diverse bacterial biota, appears to be a veritable sink for horizontally acquired bacterial genes 5455. Similar, if less dramatic, evidence of apparent horizontal gene transfer was seen in Halobacterium, Sulfolobus, and A. fulgidus (Table 2; 44). Of further note are the patterns of genes that are ubiquitous in one of the major branches of archaea, namely Euryarchaeota or Crenarchaeota, but are missing from the other branch. While quantitatively small, the set of euryarchaea-specific genes includes those for several crucial cellular functions, such as the two subunits of DNA polymerase II and the FtsZ GTPase that is required for cell division in Euryarchaeota and bacteria but missing from Crenarchaeota and eukaryotes.

Phyletic patterns can be used for interesting and potentially useful forays into functional genomics - more specifically for the identification of the genomic cognates of particular phenotypes. The most dramatic phenotypic characteristic of archaea is hyperthermophily, and attempts have been made to use the phyletic pattern approach to identify a gene set typical of hyperthermophiles. Strikingly, there is only one COG that is represented in all hyperthermophiles (both bacteria and archaea) but not in any other sequenced genomes, the reverse gyrase (56; COG1110). Reverse gyrase consists of a topoisomerase and a helicase domain and functions to introduce negative supercoiling into DNA; this activity is apparently required for DNA replication and gene expression at extreme high temperatures 57. But 'clean' phyletic patterns that have an unequivocal association with a given phenotype are an exception rather than the rule, so flexible pattern selection approaches have been employed. Our recent analysis of phyletic patterns enriched in archaeal and bacterial hyperthermophiles yielded around 60 COGs potentially related to this phenotype 58. About one quarter of these COGs encode parts of a predicted DNA repair system that is largely characteristic of thermophiles (59 and see below). The remaining COGs in this set suggest the existence of a transcriptional regulator that might be involved in adaptation to hyperthermal environments, and a distinct class of enzymes, the S-adenosyl methionine (SAM)-radical enzymes, whose chemistry is likely to be particularly efficient under these conditions 58. Finally, a substantial number of COGs are specific for methanogens or shared by the methanogens and A. fulgidus (Table 2 and 45). Many of these include known or predicted enzymes involved in methanogenesis and associated metabolic pathways 4560; others remain to be characterized and are likely to encode additional components of these pathways.

Further functional and evolutionary information can be extracted from complementary phyletic patterns, which are the signature of non-orthologous gene displacement 2661. Although the complementarity is, most often, only partially due to redundancy in some species, several cases of near-perfect complementarity among archaea are notable, such as the two classes of unrelated lysyl-tRNA synthetases 6263, and two forms of thymidylate synthase that are also unrelated to each other 6164. Below, when discussing functional genomics of the archaea, we return to the use of conserved and complementary phyletic patterns for functional prediction.

Comparative genomics nowadays includes a new variety of phylogenetic analysis, which for short has been dubbed genome-tree construction. Under this approach, phylogenetic trees are built not from the sequences of a single gene (such as an rRNA) but from concatenated sequences of multiple genes (proteins), from other, integral measures of the evolutionary distance between genomes (for example, the median of the distribution of evolutionary rates between orthologs), or from non-sequence-based measures such as the similarity of gene repertoire and gene orders 65. Generally, it appears that trees produced from concatenated alignments of gene products that are not particularly prone to HGT yield the best resolution 666768. All genome-tree analyses unequivocally supported the monophyly of archaea and the monophyly of Crenarchaeota. Beyond that, however, the genome-tree topology is not necessarily compatible with that of rRNA-based trees. Thus, genome-tree analysis cast doubt on the bifurcation of Euryarchaeota and Crenarchaeota being the first split in archaeal evolution; in some of these analyses, Halobacterium and Thermoplasma branch off first, suggesting that Crenarchaeota are a highly derived lineage that evolved from within Euryarchaeota 66. The same versions of genome-trees strongly suggest monophyly of methanogens, which is compatible with their distinct gene repertoire and life style 45; but alternative trees constructed from concatenated multiple alignments of a different assortment of translation machinery components support the original divergence of Crenarchaeota and Euryarchaeota but reject the monophyly of methanogens 2169. It appears that a robust phylogeny of archaea will require many additional genome sequences and perhaps also further refinement of phylogenetic methods dealing with long branches and with large amounts of data. The reconstruction of the best approximation of archaeal phylogeny is of interest not so much in and of itself, but more in terms of clarifying the tempo and mode of evolution of this remarkable group of organisms. A definitive tree topology will help answer fundamental questions, such as whether methanogenesis evolved only once or several times, whether the role of histones in chromatin formation is ancestral or derived in the archaeo-eukaryotic lineage, and even the exact evolutionary relationship between archaea and eukaryotes.

Phylogenetic trees can also be employed for reconstruction of the gene sets of ancestral life forms. Given a species tree topology and phyletic patterns of the maximum possible number of orthologous gene sets (or COGs), the most parsimonious evolutionary scenario, which includes the minimum possible number of elementary events, can be reconstructed using various parsimony algorithms 7071. The elementary events included in this type of analysis are gene gain and gene loss. Gene gain in a given lineage may occur either as emergence of new genes (COGs), primarily via duplication with subsequent radical divergence, or as HGT from other lineages. The relative likelihood of gene loss and gene gain (the gain penalty) substantially affects the reconstructed evolutionary scenario and the gene composition of the reconstructed ancestral genomes - but this parameter is a major unknown. Nevertheless, examination of the gene sets for the last universal common ancestor (LUCA) derived with different gain penalties showed, perhaps rather unexpectedly, that the assumption of equal probabilities of gains and losses (a gain penalty of 1) yields a reasonable reconstruction of the main functional systems of the cell 71.

We therefore applied our version of the weighted parsimony algorithm 70, with that assumption, to the updated set of bacterial, archaeal and eukaryotic genomes (also assuming the dichotomy of Euryarchaeota and Crenarchaeota suggested by rRNA trees and some of the genome-trees) and the results are schematically shown in Figure 3 (see also additional data file). This reconstruction suggests that the common ancestor of archaea could have had around 900 genes, with substantial gene gain but only minimal gene loss compared to the more ancient common ancestor of the archaeo-eukaryotic lineage. Obviously, the conserved core of the pan-archaeal genes is a subset of the reconstructed ancestral gene set, but it seems striking that approximately two thirds of the ancestral genes have been lost from at least one of the sequenced archaeal genomes (Figure 3).

In the era of comparative genomics, experimental studies on a genomic scale lag woefully behind computational studies. The great majority of the genes in most species will never be studied experimentally, and our understanding of the biochemistry and physiology of the respective organisms therefore depends on the transfer of information from functionally characterized orthologs 2672. For both bacteria and eukaryotes, such transfer is facilitated by the availability of a vast body of experimental data on model organisms, such as Escherichia coli, Bacillus subtilis, the yeast Saccharomyces cerevisiae or the fruit fly Drosophila melanogaster. The situation is quite different for archaea because, some genetic studies of mesophilic archaeal species notwithstanding 73, there is, so far, no satisfactory model system; this results primarily from the fact that most of these organisms grow slowly and are hard to cultivate. The functions of most of the archaeal genes have therefore been predicted by sequence analysis. Moreover, on many occasions the similarity between an archaeal protein and its functionally characterized homolog is so low that computational methods for sequence analysis have to be extended to the limit of their power.

A substantial fraction of the functional predictions for archaeal proteins appear 'trivial' in the sense that the respective proteins are highly conserved orthologs of well-characterized proteins from model organisms and, for all practical purposes, the validity of the prediction is beyond reasonable doubt (which is not to say that there are no important details of the functions of these proteins that can be uncovered only by experiment). For many other proteins, however, the prediction remains only a pointer to the probable biochemical function while the biology remains a mystery. A rough breakdown of the state of functional characterization of several archaea with sequenced genomes is given in Figure 4. The substantial fraction of genes for which only general, typically biochemical, prediction is available, is testimony to the current limited understanding of archaeal biology. Moreover, even some of the more definitive predictions only serve to emphasize the biological differences between archaea and the bacterial or eukaryotic models from which the predictions are inferred (Table 3). A good example is the archaeal ortholog of the bacterial DNA primase (DnaG), which is a highly conserved protein present in all archaea 24. The discovery of a predicted bacterial-type primase in archaea was unexpected, given that the archaeal replication system is orthologous to that of eukaryotes and, in particular, archaea encode the two subunits of the eukaryotic-type primase (COG1467 and COG2219; it should be noted parenthetically that detection of the large primase subunit itself required extremely careful sequence analysis due to the low similarity to the eukaryotic ortholog 22). Given that the niche of the replicative primase seems to be occupied by the eukaryotic-type enzyme 7475, the DnaG ortholog is likely to have a critical role in repair, but beyond this general idea its function has yet to be determined by direct experimentation; such experiments have the potential to reveal completely new repair systems and pathways. Other proteins implicated in repair as a result of exhaustive sequence analysis, such as the putative nucleases encoded by COG1833 and COG1628 (Table 3), illustrate the same point: the biochemical activities are predicted but the biology remains to be investigated experimentally.

Some of the other functional predictions inferred from sequence analysis directly help filling glaring gaps in otherwise well-characterized pathways of archaeal metabolism. A good example of such focused prediction is the identification of an archaeal fructose-1,6-bisphophate aldolase, an indispensable glycolytic enzyme, which was first predicted computationally to be a member of the DhnA family of aldolases by our group 76 and subsequently identified experimentally 77. In the same vein, during work for this article, we predicted the missing archaeal aconitase, an essential enzyme of the tricarboxylic acid cycle (Table 3; K.S.M. and E.V.K., unpublished observations).

The identities of a considerable number of proteins responsible for essential functions in archaea remain a mystery. Perhaps the most notable case is the missing cysteinyl-tRNA synthetase of thermophilic methanogens. Cysteine is incorporated into the proteins of these organisms as readily as in any others, but they lack an ortholog of cysteinyl-tRNA synthetase. Two different solutions for this paradox have been proposed, one involving an uncharacterized protein that has been proposed to be a 'third class' of aminoacyl-tRNA synthetases 78, and the other based on the apparent ability of the archaeal prolyl-tRNA synthetase to couple tRNA^Cys with cysteine 79. The first hypothesis has been refuted by our group upon more detailed sequence analysis 80, however, and the second did not seem to be compatible with subsequent structural studies 81. The real cysteinyl-RNA synthetase of methanogens seems still to be hiding among uncharacterized proteins. Gaping holes also remain in archaeal pathways of isoleucine biosynthesis 82, heme biosynthesis 83, biotin biosynthesis 26, and several others.

Beyond straightforward (even if highly sensitive) sequence analysis, a powerful approach to the prediction of functions involves analysis of various forms of genomic context, or establishing 'guilt by association' 2684858687. The associations employed to infer gene functions may be manifest at different levels, including the phyletic patterns discussed above, juxtaposition of domains in multidomain proteins, clustering of genes in (predicted) operons, co-expression, and protein-protein interaction. The last two of these types of data, obtained through transcriptomic and proteomic efforts, are becoming increasingly important in the functional genomics of eukaryotes and, to a somewhat lesser extent, bacteria, but are so far unavailable for archaea. The main type of context information in archaea has therefore been obtained by analyzing conserved elements of gene order and multidomain proteins. Only a relatively small fraction (10-15%) of each archaeal genome is covered by evolutionarily conserved gene strings that can be predicted to form operons 87. Nevertheless, by comparing gene orders in multiple genomes, partially conserved gene neighborhoods can be reconstructed and examination of some of these leads to predictions of functional systems whose existence has not previously been suspected (Table 3).

The most notable illustrations of this approach (both from our own group) are the prediction of the archaeal exosome 88 and a potential new repair system typical of archaeal and bacterial thermophiles 59. The eukaryotic exosome is a multisubunit complex that consists of RNAses, helicases and RNA-binding proteins and is involved in the exonucleolytic degradation of various classes of RNA 899091. During comparative analysis of gene order in prokaryotic genomes, it was observed that a distinct set of genes, some of which encode orthologs of eukaryotic exosome components, form a partially conserved predicted superoperon, which includes in total over 15 genes (although none of the archaeal genomes contains every one of these within the predicted superoperon). In addition to RNAses and RNA-binding proteins (with an RNA helicase apparently encoded in a separate operon), the exosomal superoperon also encodes a proteasome subunit and a subunit of prefoldin, a co-translational molecular chaperone (88 and Figure 5a). Thus, these observations point to the existence of a multifunctional macromolecular complex that could couple post-translational protein folding with regulated, ATP-dependent degradation of RNA and proteins. This complex remains to be discovered experimentally, and the potential implications for new functional and physical interactions in eukaryotes are also open to experimental study.

A more sophisticated comparison of gene orders, which required special algorithms for delineation of partially conserved genomic neighborhoods 92, led us to predict a distinct DNA repair system that is most prevalent in thermophiles and includes genes for a predicted novel DNA polymerase, a helicase, two nucleases and several uncharacterized genes, at least one of which could encode a novel nuclease (59 and Figure 5b). Furthermore, this neighborhood contains multiple, diverged versions of a gene coding for a protein with a probable structural role dubbed RAMP (repair-associated mysterious protein). The proliferation of RAMP genes (Figure 5b) is an example of a potentially adaptive lineage-specific expansion of a gene family; such expansions are discussed below in greater detail.

Additional, simpler cases of functional prediction via 'guilt by association' are illustrated in Figure 5c,d,e. The gene for the uncharacterized protein represented by COG1711 (Figure 5c) forms an evolutionarily highly conserved gene pair with the gene for the clamp subunit of DNA polymerase (ortholog of the eukaryotic PCNA). The orthologs of COG1711 proteins are conserved in all eukaryotes, and this protein might be an essential but still uncharacterized component of the archaeo-eukaryotic DNA replication machinery (K.S.M. and E.V.K., unpublished observations). The gene represented by uncharacterized COG1909 is squeezed between genes for RNA polymerase subunits and that for a ribosomal protein (Figure 5d). Examination of the multiple alignments that lead to this COG shows conservation of polar residues compatible with an enzymatic function (K.S.M. and E.V.K., unpublished observations). There are no readily detectable eukaryotic orthologs for this protein, which is therefore likely to be an archaea-specific enzyme with a house-keeping function.

Finally, uncharacterized COG1545 consists of genes encoding putative zinc-ribbon-containing proteins that form a stable gene pair with the gene for acetyl-CoA acetyltransferase, a central enzyme of fatty acid biosynthesis (Figure 5e). Both these genes show remarkable paralogous expansion in several archaea, probably as a result of a series of duplications of the gene doublet. It appears likely that proteins from COG1545 form a complex with acetyl-CoA acetyltransferase, with the zinc-ribbon protein regulating and/or stabilizing the enzyme. The predictions depicted in Figure 5c,d,e and other similar ones (87; and K.S.M. and E.V.K., unpublished observations) are not particularly precise, even in terms of the biochemical activity of the respective proteins. Nevertheless, guilt by association implicates each of these proteins in specific biological functions, and the evolutionary conservation of both the proteins themselves and the gene order all but proves that their functions are essential. Thus, these proteins appear to be excellent targets for experimental studies, which have the potential to reveal new facets of central cellular processes in archaea.

Comparative-genomic analysis of prokaryotes and eukaryotes points to lineage-specific expansion (proliferation) of paralogous gene families as a major means by which organisms adapt to their specific environment and lifestyle 939495. A number of such expansions are seen in archaea but in most cases we have, at best, only a vague understanding of the associated biology; several examples are given in Figure 6. The expansion of two groups of permeases in Thermoplasma and Sulfolobus (Figure 6a) clearly reflects the heterotrophic metabolism of the former 96 and the chemo-organotrophic lifestyle of the latter 97. The specific proliferation of ferredoxin in methanogens (Figure 6b) is also easily explained by the role of these proteins in the oxidoreduction reactions of methanogenesis 98. The remaining two cases in Figure 6c,d are much more enigmatic. The congruent proliferation of the transcription-initiation factors TFIIB and TFIID in Halobacterium (Figure 6c) might point to unusual aspects of transcription regulation in this archaeon but the details remain obscure. The proliferation of two subunits of a predicted nucleotidyltransferase in several archaea 99100 (Figure 6d) is of special interest and might have something to do with thermal adaptation, but the actual functions and even the substrates of these enzymes remain a mystery. Other lineage-specific expansions, such as that of distinct families of predicted ATPases in Methanocaldococcus and Pyrococcus, or a specific family of RadA(RecA)-like ATPases and the UspA-family of NTP-binding proteins in several archaeal species 39, suggest the existence of unusual pathways, perhaps involved in stress response and signal transduction, but the actual biology associated with these expansions can only be uncovered experimentally.

Archaeal comparative genomics is a young field and so far, as we have seen, largely predictive. But a few experimental studies have already been instigated as a result of comparative-genomic predictions. The discovery of the archaeal fructose-1,6-bisphosphate aldolase mentioned above 7677 is a case in point, and several other examples of experimental validation of predictions are given in Table 3. It does not seem to be chance that these examples all involve metabolic enzymes for which the specific reaction could be predicted precisely. Validation is likely to be much more difficult for proteins of other functional groups, such as putative repair enzymes, for which the actual substrates are harder to predict.

For some conserved archaeal proteins, functions cannot be predicted computationally despite considerable effort. Several important discoveries have been made by experimental characterization of such mysterious proteins. The most notable cases include the archaeal Holliday-junction resolvase, which is not related to its functional analog in bacteria 101102103, and DNA polymerase II, a highly conserved euryarchaeal protein that is not found outside this lineage and shows no detectable sequence similarity to any other proteins 104105. Additional examples of direct experimental determination of the functions of archaeal proteins that could not be predicted by computational techniques (at least not before the experiment had been reported) are given in Table 3.

Especially notable is the story of the Alba protein, a DNA-binding component of chromatin in Crenarchaeota 106107. As noted above, crenarchaea lack histones and in these organisms Alba appears to be the main chromatin protein, in a striking case of non-orthologous gene displacement. But orthologs of Alba are also present in thermophilic Euryarchaeota and in some eukaryotic lineages, where its functions remain to be elucidated. The most remarkable discovery regarding Alba is the regulation of its interaction with DNA and with the chromatin-associated protein deacetylase Sir2 via lysine acetylation and deacetylation 106108. In eukaryotes, regulation of chromatin dynamics via acetylation and deacetylation occurs through histone tails 109. Thus, a special case of non-orthologous gene displacement seems to have taken place whereby the regulation mechanism is conserved but the actual substrates are different in archaea and eukaryotes. To add an extra twist to the story, Thermoplasma lacks both histones and Alba but has the bacterial DNA-binding protein HU, pointing to three distinct solutions to the problem of chromatin organization in archaea 107.

The last subject we have to briefly touch upon is structural genomics of the archaea. The ultimate goal of the structural genomics enterprise is determining the three-dimensional structure for all proteins, or at least for all sufficiently different proteins encoded in the genomes of diverse life forms 110. This goal is far from being reached, and targets for structural determination have been prioritized by different researchers on the basis of different principles, from nearly random choice to relatively elaborate strategies, including the use of the COG database 111112113114115. The development of structural genomics so far has been a mixture of success, when informative and interesting structures have been solved, and mild disappointment in cases when the structure determination did not seem to shed any light on a protein's function. Structural genomics could be particularly important in the case of archaea, for which a miniscule number of structures had been solved prior to the launch of structural genomic initiatives, and in which proteins often show low similarity to bacterial or eukaryotic homologs, making homology modeling difficult.

Notable developments that illustrate both the benefits and the pitfalls of structural genomics, are the concerted effort on 'structural proteomics' of Methanothermobacter thermoautotrophicus 116 and a similar project on M. jannaschii 117. The elucidation of the structure of the M. jannaschii protein MJ0577 117 is an excellent case for the power of structural genomics. Analysis of this structure and accompanying biochemical experiments revealed a distinct nucleotide-binding domain that is distantly related to the catalytic domains of class I aminoacyl-tRNA synthetases and belongs to the so-called HUP fold of nucleotide-binding domains 118. Together with comprehensive sequence analysis, the determination of this structure provided the structural, functional and evolutionary context for the UspA protein family, which is specifically expanded in archaea 39. The exact function(s) of these proteins remains unknown but, in this case, structural genomics ensured a substantial functional insight. On several other occasions, however, determination of the structures of archaeal proteins has failed to provide clear functional clues; these remain structures in search of a function.

The first sequenced archaeal genome was a veritable terra incognita. Six years after that sequence appeared, the archaeal genomescape looks quite different. The principal landmarks have been mapped and now, when a new archaeal genome is released, we largely know what to expect from it. Computational approaches to comparative genomics, combining in-depth sequence and structure comparison with genome context analysis, have led to the reconstruction of the central functional systems of archaeal cells. But these approaches have also produced numerous isolated predictions of biochemical activities of archaeal proteins that remain to be fitted into a general picture, and this can be done only through 'wet' experiments, although new genome sequences will substantially help by enriching the genomic context. A shrinking but still notable set of archaeal genes includes those that encode highly conserved proteins without any clue to function; solving these mysteries has the potential to bring out truly new biology. Furthermore, in this article we have not even touched upon important aspects of archaeal genomics, such as the in-depth studies of the translation system, which have revealed several highly unusual, remarkable mechanisms and enzymatic systems 63119 or the identification of regulatory sites in DNA and patterns of transcription regulation 120121. The latter avenue of research is still in its infancy but will certainly grow in scale once more archaeal genomes, and in particular closely related ones, are sequenced.

Because of the lack of established model systems for archaeal experimental biology and the resulting difficulty with large-scale experimentation, clues from genome comparison are even more crucial for archaeal functional genomics than they are in the case of bacteria or eukaryotes. So far, the input of comparative genomics into actual experiments has been less prominent than we would hope. Simply put, it is not often that experimenters rush to test predictions produced by in silico genome comparison and, furthermore, it is even rarer that targets for functional characterization are carefully prioritized on the basis of how unusual and fundamental the predictions are. As discussed above, however, the few cases when such tests have been performed are encouraging. It is our hope that the future belongs to a much tighter integration of comparative, structural and functional genomics.

Beyond functional studies, archaeal genomics is fundamental to our understanding of two critical transitions in the evolution of life. The first is the primary split between the bacterial and archaeo-eukaryotic lineages, which might have involved the origin of the DNA-replication machinery and of the large, double-stranded DNA genomes themselves 2223, and the second is the origin of eukaryotes 122. With regard to the latter problem, archaea are a particularly valuable source of information because, on many occasions, they seem to have retained primitive traits while eukaryotes have undergone major changes. A characteristic example is the small DNA polymerase subunit, which has all the hallmarks of an active phosphatase in archaea, but not in eukaryotes, in which the phosphatase activity is predicted to be inactivated 123. Indubitably, archaea resemble the common ancestor of the archaeo-eukaryotic line of descent more closely than eukaryotes do, so archaeal genomics is our best chance to reconstruct this critical intermediate in the evolution of life. We are confident that comparative archaeogenomics has a bright future, with major progress in both the functional and the evolutionary avenues of research expected within the next few years.

The list of genes in the reconstructed gene set of the last common ancestor of archaea is available (additional data file 1).