The genome of I. hospitalis consists of a single circular chromosome (Table 1). At 1,297,538 bp, the genome of I. hospitalis is the smallest among free-living organisms, which do not require a continuous association with another species and can replicate independently (Figure 1). Even the combined gene complement of I. hospitalis and N. equitans (1,434 and 556 protein-coding genes, respectively) is significantly smaller than that of average free-living bacteria (approximately 3,600 genes) or archaea (approximately 2,300 genes), based on the available completed genomes. The size distribution of 623 complete microbial genomes indicates that the 1-2 Mbp range includes both obligate symbionts/parasites as well as free living bacteria and archaea (Figure 1). The minimal genome for free-living organisms may therefore be on the order of 1 Mbp, several taxonomically and metabolically distant archaeal and bacterial lineages having independently reached near-minimal functional gene sets for their respective ecological niches.

The sizes of microbial genomes are the result of dynamic equilibria between contraction by deletions and expansion due to duplications, lateral gene transfer and insertion of mobile DNA. For free-living organisms with very large effective population sizes, genome streamlining is likely to be a selective consequence of reducing the metabolic burden to maintain DNA of little adaptive value, as illustrated by the genomes of such highly successful and widespread lineages as Prochlorococcus and Pelagibacter 2021. An alternative (but not necessarily exclusive) hypothesis links genome reduction to elevated mutation rates in large populations. Accumulation of mutations could lead to inactivation and loss of genes that make weak contribution to the fitness of the respective organisms 22. Ignicoccus, however, inhabits heterogeneous, geographically dispersed and relatively ephemeral hydrothermal marine environments. Such organisms generally have small effective populations and experience periodic bottlenecks and limited gene flow 23. Conceivably, in a case like this, genome contraction might have to do with the very active recombination and DNA repair that organisms inhabiting extreme environments employ for maintaining genomic integrity. Frequent recombination might not only efficiently remove deleterious mutations induced by the environmental conditions but also generate diversity and increase the fixation rate of adaptive alleles 2425. A high frequency of illegitimate, intra-chromosome recombination could also be effective in preventing genome expansion by increasing the frequency of deletions and counteracting gene duplication. This might explain the reduced genome size in many members of the Archaea and contribute to their proposed higher adaptability to chronic energy stress 26. While we expect these general principles to be valid in the Nanoarchaeum-Ignicoccus system as well, the co-evolution of these two organisms also left unique imprints on their physiology 23. The most striking effect of this co-evolution, however, is the massive gene loss in N. equitans, resembling that of obligate intracellular bacterial symbionts and, as an extreme case, that of eukaryotic organelles 12.

The recently published database of archaeal clusters of orthologous genes (arCOGs) provides a framework for comparing the I. hospitalis genomic data to genes from 41 previously sequenced archaeal genomes organized into sets of probable orthologs 27. Of the 1,434 annotated I. hospitalis protein-coding genes, 1,155 (80.5%) were assigned to arCOGs, a coverage that is the lowest among the Desulfurococcales (85% on average) and overall among thermophilic Crenarchaeota.

I. hospitalis lacks orthologs of 19 genes from the Crenarchaeota core (that is, genes that are represented in all 12 available genomes of thermophilic species of Crenarchaeota included in the arCOGs) 27 (Table S1 in Additional data file 1). None of these genes include components of information processing systems, indicating that these systems are largely intact in I. hospitalis despite the small genome. The missing genes encode, primarily, diverse metabolic enzymes, some of which - for example, thymidylate kinase - catalyze essential reactions. Conceivably, these enzymes are substituted for by distant homologs that so far remain undetected or by analogs.

Using the assignment of I. hospitalis genes to arCOGs, we applied weighted parsimony to perform a reconstruction of gene gain and loss events in archaea 2728, with an emphasis on the I. hospitalis lineage. The small genome size appears to be a result of gene loss that has vastly predominated the evolution of this lineage: it was inferred that approximately 484 arCOGs were lost, compared to the inferred gain of only 56. Approximately 946 arCOGs (1,094 genes, representing 76% of the I. hospitalis gene set) appear to have been inherited from the last common ancestor of the Desulforococcales, the order to which Ignicoccus belongs, together with Aeropyrum pernix, Hyperthermus butylicus and Staphylothermus marinus. The functional distribution of the lost genes is consistent with the fact that I. hospitalis is an obligate anaerobic autotroph. In contrast to A. pernix, numerous genes related to aerobic metabolism as well as catabolism and transport of amino acids, sugar and nucleotides were lost, along with many transcriptional regulators (Figure 2; Table S2 in Additional data file 1). An analysis of arCOGs that are present in N. equitans but absent in I. hospitalis does not suggest that the inferred gene loss in I. hospitalis was accompanied by transfer of potentially essential functions to the symbiont (Table S3 in Additional data file 1). Ignicoccus is far removed from the root of the tree of thermophilic Crenarchaeota (whether the tree is constructed for rRNA or various informational proteins), and the tree, including basal branches, is dominated by heterotrophs and mixotrophs (Figure S1 in Additional data file 2). Thus, the alternative scenario, namely, that Ignicoccus reflects the ancestral state for this entire group, is not supported by the phylogenetic analyses. However, this might reflect our incomplete sampling of the archaeal diversity and the bias towards isolation and characterization of heterotrophs. A better understanding of the direction of evolution in archaeal genome size and architecture will require a significant increase in the number and diversity of cultivated species and sequenced genomes, including close relatives of I. hospitalis and additional chemolithoautotrophs.

The reduced frequency of duplicated genes (paralogs) in I. hospitalis compared to all other archaea except N. equitans (Figure 3) and the absence of transposable elements support the hypothesis of genome streamlining. Furthermore, approximately 180 chromosomal gene clusters that are typically conserved in archaea are disrupted in the genome, including some of the ribosomal operons as well as those encoding the proteasome components, ATP synthase and DNA topoisomerase VI. As it is unlikely that so many gene clusters and operons have independently assembled in archaeal lineages not directly related, the architecture of the I. hospitalis genome suggests that recombination events have resulted in gene cluster fragmentation, deletions, and may have restricted gene family expansion. On the other hand, it is notable that several families of paralogous genes are uniquely expanded in I. hospitalis (Table S4 in Additional data file 1). The most intriguing is the presence of 10 genes that encode WD40-repeat-containing proteins. Proteins containing WD40 repeats are among the most abundant and highly conserved in eukaryotes, where they are key structural components of a variety of macromolecular complexes 29. Proteins containing these repeats are also widely scattered among archaea and bacteria but are mostly encoded in (relatively) large genomes 30. In particular, among archaea, we have detected comparable expansions of WD40-containing proteins only in Methanosarcinales, a group of Euryarchaeota that displays significant gene gain 27. Conceivably, the WD40-proteins of I. hospitalis are involved in the organization of specific protein complexes and/or cellular compartments, and potentially might contribute to the interaction with N. equitans. Similarly, I. hospitalis encodes 9 proteins containing the V4R domain and 12 proteins containing the CBS domain, both small-molecule-binding domains that are likely to be involved in metabolic regulation and signaling 3132 (Table S4 in Additional data file 1). Considering the homology identified between the V4R domain and a component of the eukaryotic Golgi vesicle transport machinery 33, some of the expanded V4R gene family members also might be implicated in the unique vesicle formation process that has been observed in Ignicoccus 9.

In addition to streamlining, selection for reducing metabolic cost in I. hospitalis may have impacted its proteome composition. In hyperthermophiles, certain biases in amino acid usage have been associated with side chain physical and chemical properties that contribute to increased protein stability 3435. For example, a preference for lysine over arginine has been attributed to a greater flexibility of the lysine side chain, which entropically stabilizes the folded state of proteins 36. While the overall amino acid usage in the N. equitans-I. hospitalis proteomes follows the distribution observed for other hyperthemophiles, there is a significant increase in lysine over arginine usage in I. hospitalis relative to the values that could be predicted from the GC content (Figure 4; note that the two positively charged amino acids, lysine and arginine, are often interchangeable in proteins but are encoded by contrasting codons, namely AAA/G for lysine, and CGX and AGA/G for arginine, hence the strong correlation of the abundance of these amino acids with the GC content). This discrepancy could be explained by selection at the genomic level against using the metabolically more expensive arginine. Arginine biosynthesis in Ignicoccus is predicted to proceed via carbamoyl-phosphate and would require five ATP equivalents, whereas lysine, synthesized from 2-oxoglutarate via the aminoadipate pathway, would use two ATP equivalents (Figure 4). Metabolic cost and nutrient availability have been proposed to play a selective role in the evolution of genome size, GC content and amino acid use in organisms that inhabit oligotrophic or energetically poor environments 202637. Since sulfur-hydrogen respiration is energetically weak 38, such genomic and proteomic adaptations allow I. hospitalis not only to be a competitive vent colonizer but also to support N. equitans. At present, in the absence of sequence data from other species of Ignicoccus, we cannot distinguish if the relationship with N. equitans has directly influenced these genomic features of I. hospitalis.

The cell-cell contact between I. hospitalis and N. equitans seems to present an opportunity for extensive lateral gene transfer (LGT). LGT is considered to play a major role in microbial genome evolution and is well-documented in symbiotic systems and in environmental microbial communities 39404142. Recent LGT events are readily detected with various methods based on nucleotide composition or codon usage, but methods that rely on protein sequence similarity and phylogenetic trees are more informative for ancient LGT events 43. To analyze the I. hospitalis genome for potential LGT events, we therefore combined automatic genome-wide phylogenetic reconstruction using PyPhy 44 with similarity searches and COG distribution analysis. The LGT candidates were further analyzed using hand-curated alignments and maximum likelihood phylogenetic analyses. Identifying the LGT direction requires analysis of conflicts between the topologies of the corresponding gene trees and the adopted species tree. The position of N. equitans within the Archaea is controversial and ranges from representing a distinct and basal phylum 11216 to being a derived member of order Thermococcales from the Euryarchaeota 17. Many gene trees identify the Thermococcales as an early diverging lineage, which further complicates this distinction. Ignicoccus on the other hand has been confidently assigned to order Desulfurococcales from the Crenarchaeota based on phylogenetic and arCOG analysis. Therefore, when attempting to infer direction of LGT, we relied on the phylogenetic placing of N. equitans and I. hospitalis genes relative to other crenarchaeal homologues, especially those from the Desulfurococcales (Aeropyrum, Hyperthermus and Staphylothermus).

A small fraction of I. hospitalis genes (approximately 6%) appear to have been transferred from lineages within Euryarchaeota, while approximately 4% seem to be of bacterial origin (Figure 5). Many of those genes encode subunits of protein complexes involved in energy metabolism or transporters and might have been acquired by I. hospitalis as small clusters or operons. Examples of putative 'bacterial' gene clusters include those encoding bacterial type polysulfide reductase (Igni528-530), the multisubunit putative Ech hydrogenase (Igni542-546, Igni1144-148) and a nitrate reductase-like complex (Igni1377-1379). Among the clusters of apparent origin from Euryarchaeota are genes encoding ABC-type transporters for antibiotics and molybdate (Igni146-147, Igni1340-1343) as well as a 2-oxoacid:ferredoxin oxidoreductase complex (Igni1075-1078). Other genes encoding characteristic proteins of Euryarchaeota are scattered in the genome (for example, the CrcB-like integral membrane protein Igni921, a 6Fe-6S prismane cluster-containing protein Igni960, micrococcal thermonuclease Igni1343, thermophilic glucose-6-phosphate isomerase Igni415). If N. equitans is a derived member of Thermococcales, as some gene trees and genomic analyses suggest 1727, then some of the putative euryarchaeal LGTs in the I. hospitalis genome might actually represent transfers from N. equitans. Such transfers could have occurred during extensive genome degradation suffered by N. equitans associated with elimination of metabolic functions, similar to cases of nuclear transfer of symbiont genes during eukaryotic organelle formation. Additional LGTs from bacteria and/or archaea, including N. equitans, might be hidden in the large number of genes (>600 or approximately 40% of the open reading frames) that either lack detectable homologs or are placed unresolved within the Archaea due to insufficient phylogenetic signal.

One of the possible outcomes of LGT in symbiotic associations involves orthologous gene displacement in the recipient genome and maintenance of the gene in the donor genome as well. In the N. equitans-I. hospitalis system, we identified 13 such cases, in which the orthologs in both genomes are each other's closest homologues (Figure 5). Several of the genes appear to have been transferred from N. equitans to I. hospitalis, including ones encoding valyl-tRNA synthetase (Igni220-Neq252), tyrosyl-tRNA synthetase (Igni347-Neq389) and a type IV endonuclease (Igni1092-Neq77a) (Figure 6a; Figure S2 in Additional data file 2). Two genes involved in recombination and repair that form a predicted operon in N. equitans (an AP endonuclease 2 family and a DEAD/DEAH box helicase, NEQ368-369) have also been transferred to I. hospitalis, either as independent events or becoming separated later by genomic rearrangement (Igni0112 and 0145). Genes encoding aminoacyl tRNA synthetases and recombination and repair proteins are frequently exchanged in microbial communities and might increase the fitness of recipient organisms, for example, by conferring antibiotic resistance in the case of aminoacyl-tRNA synthetases 4546.

A similar case of lateral transfer likely involved the gene encoding leucyl aminopeptidase (LAP), Igni738-Neq412 (Figure 6b). LAPs are ubiquitous in bacteria and eukaryotes but their presence in archaea is so far strictly limited to the Desulfurococcales and the Cenarchaeales. While no specific function has been described so far for archaeal LAPs, in bacteria they are multifunctional proteins, with roles in protein turnover as well as in transcription control and recombination 47. The absence of LAP in Euryarchaeota, in Korarchaeum cryptofilum (a potentially basal archaeal lineage with affinities with the Crenarchaeota 48) as well as in two of the four Crenarchaeota orders for which genomic data are available may suggest that the Desulfurococcales and Cenarchaeales acquired the gene via LGT from bacteria. The phylogenetic analysis places the I. hospitalis gene close to that of N. equitans but not part of the Desulfurococcales clade. The high level of sequence similarity between the N. equitans and I. hospitalis LAP genes (40%) surpasses that between any of the other Desulfurococcales (approximately 30%). However, the direction of the transfer is uncertain. The exclusion of the I. hospitalis LAP from the clade formed by the other Crenarchaeota homologs suggests that the Ignicoccus gene may have been acquired from N. equitans followed by orthologous gene displacement. Based on this scenario, the original presence of LAP in N. equitans would be at odds with its purported affiliation with the Euryarchaeaota and specifically the Thermococcales, which are lacking leucyl aminopeptidases. The alternative hypothesis, transfer of the LAP gene from I. hospitalis to N. equitans, is challenged by the separation of the Ignicoccus-Nanoarchaeum clade from the other Desulfurococcales. Complete genome sequences of other Ignicoccus or Nanoarchaeota species may help distinguish between these competing hypotheses.

Genetic information processing in I. hospitalis, as inferred from the genome sequence, is typical of the Crenarchaeota. Orthologs of two family B DNA polymerases are present in the genome (Igni62, 690); one corresponds to the aphidicolin-resistant DNA polymerase I (polA), and the other to the aphidicolin-sensitive DNA polymerase II (polB) of Aeropyrum pernix 49. No orthologs of the third family B DNA polymerase or Euryarchaeota-type heterodimeric DNA polymerase were found. Unlike other archaeal genomes, the genes coding for replication initiation/origin recognition factor (Orc1/Cdc6) are not co-localized with the predicted origin of replication 5051, a characteristic potentially related to general operon fragmentation in I. hospitalis. Unlike other archaea, including I. hospitalis, that possess DNA primases consisting of a small (catalytic) and large (structural) subunits, N. equitans seems to encode a single-subunit primase (NEQ395) in which the small subunit is fused to the carboxy-terminal domain of the large subunit 52 (EVK, unpublished observations). This may be the result of extreme genome contraction in this organism, possibly linked to its symbiotic lifestyle. Similarly, an important molecular machine absent in N. equitans but present in I. hospitalis is the RNase P complex (RNA and four separate proteins subunits, rpp14, 21, 29 and 30). It has been recently shown that tRNA processing in N. equitans is RNase P-independent, most likely because genome shrinkage led to the evolution of leaderless tRNAs that was followed by the loss of all five RNAse P complex genes 53.

The membrane composition of hyperthermophiles is specifically adapted to reduce proton and ion permeability, which increase with temperature 54. Cyclic tetraether-type lipids (caldarchaeol) that are present in the cytoplasmic membrane of I. hospitalis and in the cell membrane of N. equitans are especially associated with low permeability 13. In contrast, the absence of caldarchaeol in the outer membrane of Ignicoccus and the presence of protein pores 11 indicate potentially less restrictive exchanges with the environment through the outer membrane. With only eight types of transporters, almost all predicted to be specific for inorganic ions or export of intracellular solutes (Figure 7), N. equitans is unlikely to import by itself all of the required metabolic precursors from its host. Consistent with its streamlined genome and autotrophic lifestyle, I. hospitalis also encodes very few transporters (<3% of its proteome), the lowest number among the sequenced species of Crenarchaeota. The types of transporters and their inferred specificities are described in Figure 7. A number of inferred subunits of ABC transporters were found in membrane preparations of I. hospitalis cells, showing that these proteins are expressed in significant amounts 55. An unexpected finding for an obligate autotroph was the presence of genes encoding two ABC transporters for oligopeptides and branched amino acids. Under laboratory conditions, it was indeed found that addition of peptides improved growth of I. hospitalis 2, suggesting that, in its natural environment, this organism might be opportunistic in utilizing such resources. The different lipid and protein compositions between the cytoplasmic membrane and the outer membrane of I. hospitalis 1013 suggest the existence of specific partitioning mechanisms. The genome encodes a predicted gene (Igni479) from the LolE permease family, an ATP-dependent transport system involved in lipoprotein release that has been shown in Buchnera to transport lipids targeted to the outer membrane across the inner membrane 56 and that might play a role in I. hospitalis membrane synthesis.

While some proteins may spontaneously insert in the membrane, most transport into and across the membrane requires the function of specialized cellular systems 57. All the components of the Sec pathway were identified in the I. hospitalis genome, including the 7S RNA gene component of the signal recognition particle (Figure 7). Even though potentially functional protein secretion complexes, including the euryarchaeal-specific SecDF are encoded in its genome, N. equitans lacks an identifiable 7S RNA gene. Since that component is critical for the assembly of a functional signal recognition particle, it might be synthesized as two separate transcripts, such as some of the tRNAs 58, or might be imported from the host. The Tat system, which transports folded protein across the membrane, is present in I. hospitalis but absent in N. equitans. For all proteins that are targeted for translocation, the signal peptide has to be removed either during or after translocation. The protease that removes some of the signal peptides in archaea, signal peptidase I, was identified in both genomes (Igni153 and Neq432). I. hospitalis also encodes a type IV prepillin peptidase (PibD, MEROPS family A24A, Igni1405), which processes membrane and secreted proteins with a class III signal peptide, including proteins involved in motility (flagellin) and pili formation 59. Neither I. hospitalis nor N. equitans appear to have flagellins, although several genes potentially associated with archaeal flagellar or pili assemblies were identified in both genomes (flaI, flaJ). While the cells do not appear to be motile, certain appendages and pili-like structures have been observed in electron micrographs 606162 and might play a role in the interaction between the two organisms.

I. hospitalis is the first archaeon with sulfur-based autotrophy for which a complete genome sequence is hereby reported. Metabolic reconstruction (Figure 7) points to simple and efficient strategies that fit a streamlined genome. Nitrogen assimilation is predicted to rely on readily available ammonia, the most economical strategy in reduced environments 63. Ammonia could be acquired through an AmtB transporter (Igni1293), which is apparently co-transcribed with the gene for the nitrogen regulatory protein PII (glnK, Igni1294). These genes are widely present in bacteria and most members of the Euryarchaeota but are nearly absent from Crenarchaeota, and probably have been laterally transferred to I. hospitalis from a euryarchaeon (Figure 5). GlnK controls the transport of ammonium ions by interacting with AmtB and also activates a type of glutamine synthase (GS) that fixes the ammonia into glutamine. GS is present in all sequenced archaea with the exception of N. equitans and, by assimilating ammonia into the amide group of L-glutamine, makes it available to downstream glutamine-dependent amidotransferases. One such enzyme is glutamate synthase (GltS, Igni408), which is predicted to catalyze the reductive transfer of the amide group from glutamine to 2-oxoglutarate, resulting in glutamate and an amino group donor for transamination reactions. The domain architecture of GltS in I. hospitalis is unique and contains a GXGXG structural domain 64 followed by a ferredoxin (4Fe-4S) and a glutamine synthetase FMN-binding domain. The glutamate synthase domain I (GlxB domain) is expressed as a separate polypeptide (Igni407) and it is unclear if these two proteins actually assemble to form glutamine synthase or if GlxB functions independently as a type II glutamine aminotransferase. The two genes are likely cotranscribed with an aspartate/aromatic aminotransferase (Igni406), suggesting a tight coupling of the transamination processes. The GS/GltS operates very efficiently at low concentrations of ammonia and substitutes for the alternative ammonia incorporation mechanisms that use glutamate dehydrogenase (GDH) 63. GDH catalyzes the reversible conversion of glutamate to 2-oxoglutarate and represents an alternative route for both deamination and ammonium incorporation. It has been suggested that GDH provides the major nitrogen assimilation mechanism in most hyperthermophiles 65, but this is clearly not the case with Ignicoccus. This is likely due to the absence of a steady source of exogenous amino acids; thus, the cells must rely on free ammonia, present at concentrations too low for GDH to operate at. However, in N. equitans, which lacks the ammonium transporter or a GS/GtlS enzyme pair, limited nitrogen metabolism could rely on a GDH (NEQ077), likely using glutamate imported from the host. The transfer of glutamate from I. hospitalis to N. equitans has been detected experimentally 3. It is not clear why N. equitans has retained a GDH gene among its very few encoding metabolic enzymes. One possibility could be that GDH would contribute to the cell redox potential by oxidative deamination of glutamate.

Pathways for the synthesis of almost all amino acids can be recognized in the I. hospitalis genome, with the exception of proline and homocysteine. Some of the enzymatic activities involved in I. hospitalis amino acid biosyntheses have been detected experimentally and labeling experiments have been used to reconstruct most the pathways 66. The genome also encodes the predicted enzymes of purine, pyrimidine, NAD, riboflavin/FAD, pyridoxal and CoA biosynthesis. The mevalonate pathway for the synthesis of the characteristic archaeal membrane archaeol- and caldarchaeol-type lipids appears to be complete (Figure 7), although enzymes involved in some of the steps have not yet been characterized in archaea 6768.

I. hospitalis utilizes a novel and so far unique autotrophic CO₂fixation pathway, termed the dicarboxylate/4-hydroxybutyrate cycle 69. The individual steps of the pathway have been investigated experimentally in detail and most have been confirmed biochemically 666970 (Figure 7). Acetate in the form of acetyl-CoA is carboxylated by a pyruvate ferredoxin oxidoreductase enzyme complex (Igni1256-1259) to form pyruvate, which is then converted to phosphoenolpyruvate by pyruvate:water dikinase (Igni1113). The source of acetyl-CoA may be linked to two adjacent genes potentially encoding an acetyl-CoA synthase (Igni256, 257). Normally, that enzyme is encoded as a single polypeptide. The two genes in I. hospitalis may encode the enzyme as two subunits requiring post translational assembly or, alternatively, the two open reading frames could indicate a pseudogene. The level of acetate in the environment where I. hospitalis has been isolated from is not known, but the genome encodes a putative sodium-acetate symporter (Igni454) and uptake of acetate has been confirmed experimentally 66. Since Ignicoccus can grow in the laboratory in the absence of acetate, the genome might also encode an acetogenesis mechanism using CO₂. One potential route is direct reduction of CO₂ to formate using hydrogen, catalyzed by a putative membrane formate dehydrogenase complex. Genes encoding a membrane complex with similarity to both nitrate reductase and formate dehydrogenase were identified as a likely operon acquired from a bacterium (Igni 1377-1380). However, since neither nitrate respiration in I. hospitalis cultures 2 nor the biochemical activity of formate dehydrogenase in cell extracts 66 were detected, the cellular function of that complex remains unclear.

The archaeal-type PEP carboxylase 71 catalyzes the second CO₂ incorporation reaction, which results in the formation of oxaloacetate, an important precursor for amino acid biosynthetic pathways (Figure 7). Reactions catalyzed by malate dehydrogenase, fumarase, succinate dehydrogenase and succinyl CoA-ligase lead to the synthesis of succinyl-CoA. Until recently, the fate of succinyl-CoA was unclear and, as the reactions that would close the cycle were not apparent based on experimental data or genomic information, the mechanism of acetyl-CoA regeneration remained unknown. Huber et al. 69 recently discovered that I. hospitalis uses a novel strategy to connect, through succinyl-CoA, the partial reductive citric acid cycle with the 4-hydroxybutyrate route of acetyl-CoA regeneration (Figure 7). Based on this finding, the proposed dicarboxylate/4-hydroxybutyrate cycle appears to be energetically less costly than other carbon fixation cycles operating in archaea 69, further supporting the notion that I. hospitalis combines a streamlined genome with efficient metabolic strategies.

Phylogenetic analysis of the two I. hospitalis gene clusters encoding oxoacid:ferredoxin oxidoreductase complexes indicates that one of them (Igni1256-1259) belongs to the pyruvate:ferredoxin oxidoreductase family and, therefore, is the likely catalyst for acetyl-CoA carboxylation. The other complex (Igni1075-1078) has a close affinity to a family with oxoglutarate specificity with no close homologs in Crenarchaeota (Figure S3 in Additional data file 2), suggesting acquisition by lateral transfer. The functional inference is based on phylogenetic partitioning of archaeal oxoacid:ferredoxin oxidoreductase genes into distinct clades that correspond to enzymes specific for pyruvate, valerate/isovalerate, or 2-oxoglutarate or that have mixed specificity 727374. In addition, alignments of the I. hospitalis alpha and beta subunit sequences revealed the presence of motifs conserved in archaeal and bacterial enzymes specific for pyruvate (Igni 1258-1259) or 2-oxoglutarate (Igni 1077-1078) 757677 (Figure S4 in Additional data file 2).

The function of the predicted 2-oxoglutarate:ferredoxin oxidoreductase (OGOR) complex in I. hospitalis remains unclear. 2-Oxoglutarate serves as an entry point in glutamate and lysine biosynthesis and is also linked to the biosynthesis of several other amino acids as shown by carbon tracing and inferred from genomic data 66 (Figure 7). In heterotrophic archaea and bacteria, oxoacid:ferredoxin oxidoreductases are involved in amino acid and sugar fermentation reactions that generate reduced ferredoxin and ATP, although OGOR has been ascribed a biosynthetic function, namely, generation of succinyl-CoA from 2-oxoglutarate 7879. By contrast, in anaerobic autotrophs, OGOR is a key enzyme in the reverse citrate cycle, where it catalyzes the fixation of CO₂ on succinyl-CoA with the formation of 2-oxoglutarate 70. However, this reaction has not been detected in I. hospitalis cell extracts and carbon isotope tracing does not support its occurrence in laboratory cultures 66. In fact, succinyl-CoA produced by the first half of the carbon fixation cycle is reduced by succinyl-CoA reductase to succinic semialdehyde and channeled into the hydroxybutyrate pathway 69. The same reaction has been shown to connect the 3-hydroxypropionate with the 4-hydroxybutyrate pathways in another recently discovered novel carbon fixation cycle in the crenarchaeaote Metallosphaera sedula 80. However, unlike the succinyl-CoA reductase from M. sedula, which uses NADPH as electron donor, the enzyme in I. hospitalis requires reduced ferredoxin 69. A speculative role for OGOR could be to provide a ferredoxin-based electron shuttle between oxoglutarate and succinyl-CoA, at the expense of a fixed carbon (Figure 7). Such coupling has been shown to be important in the anaerobic metabolism based on aromatic compounds in Thauera aromatica, OGOR providing benzoyl-CoA reductase with reduced ferredoxin 75. In Ignicoccus, under active growing conditions, such a reaction based on de novo synthesized 2-oxoglutarate does not seem advantageous as it would increase the succinyl-CoA pool at the loss of an acetyl-CoA while sufficient reduced ferredoxin may be supplied by a hydrogenase. However, under limited CO₂ and H₂ conditions, 2-oxoglutarate from the internal pool or derived from exogenous amino acids and peptides could keep the 4-hydroxybutyrate part of the cycle active and generate acetyl-CoA for maintenance functions. Experimental studies will be needed to test this hypothesis and identify the specificity of the predicted OGOR complex.

Under laboratory conditions, the only energy yielding reaction that sustains the metabolism of I. hospitalis is the oxidation of molecular hydrogen coupled to the reduction of elemental sulfur. While energetically weak (-6.7 kcal/mol) 38, there are indications that this type of respiration might have been used by ancient microbes of the early Archaean 5. Details of bioenergetic reactions and the mechanisms for generating the membrane chemiosmotic potential in anaerobic hyperthermophilic archaea are still not well understood. Minimal enzymatic components that are required include a membrane hydrogenase complex, a sulfur reductase and an electron transport chain between them. In I. hospitalis, there appear to be two clusters of genes encoding subunits of the sulfur/polysulfide reductase complex. The first such cluster (Igni801-803) contains the catalytic reductase (SreA), a 4Fe-4S ferredoxin (SreB) and the membrane anchoring component NrfD (SreC) with eight transmembrane domains. NrfD is thought to participate in the transfer of electrons from the quinone pool into the terminal components of the Nrf pathway. Elsewhere in the genome, a gene cluster (Igni528-530) that appears to be of bacterial origin contains a different NrfD, a periplasmic FeS ferredoxin, as well as a membrane protein with four putative heme binding sites that may serve in the electron transfer chain through the membrane, possibly binding menaquinone. This gene cluster is also present in the related archaeon Hyperthermus butylicus 81, suggesting the possibility that it was transferred between the two archaeal lineages after one of them likely acquired it from a delta proteobacterium. Two types of reductase complexes might therefore assemble in I. hospitalis, archaeal and bacterial. In other sulfur reducers a periplasmic polysulfide-sulfur transferase (a member of the rhodanese family) facilitates the transfer of low concentrations of polysulfide to the reductase. I. hospitalis is the only crenarchaeote that is missing a rhodanese family gene. This could be a result of growing under relatively neutral pH, where polysulfide concentrations may be high enough. Therefore, access of polysulfide to the cytoplasmic membrane, where the reductase complex is likely located, could occur by diffusion across the large periplasmic space after passage though the outer membrane pores.

Ignicoccus depends on molecular hydrogen as the sole electron donor. A single predicted operon contains the genes encoding the large and small subunits of a hydrogen uptake NiFe hydrogenase, including the large and small subunits (Igni1366-1369). The heterodimer is exported to the periplasm through the twin-arginine translocation (TAT) system and is assembled with a 4Fe-4S ferredoxin and a membrane protein anchor containing histidine residues that might bind a b-type heme 82. The formation of the metal-containing active site and the assembly of the hydrogenase is a complex process requiring multiple accessory proteins 83, all of which appear to be encoded in the I. hospitalis genome (Figure 7). Hydrogen oxidation is coupled with electron transfer through FeS centers and a putative membrane cytochrome to the quinone pool of the respiratory chain, which contributes to the generation of a membrane potential that drives ATP synthesis. The quinone appears to be associated with the membrane component of the hydrogenase and that of polysulfide reductase, with the exchange of the electrons likely involving formation of respiratory 'supracomplexes' 84. A separate 'energy-converting' Ni-Fe hydrogenase family complex (Ech), which is evolutionarily related to the energy-conserving NADH:quinone oxidoreductase (complex 1), appears to be encoded by genes in two clusters (Igni542-546 and Igni1144-1146). This hydrogenase is the likely catalyst in maintaining the pool of reduced ferredoxin.

The I. hospitalis genome also contains a four gene putative operon with close homologues among the bacterial respiratory periplasmic nitrate reductases (Igni1377-1380). Similarity to formate dehydrogenases was also detected, so the function of the complex is not clear, as nitrate cannot serve as a sole electron acceptor in Ignicoccus 260. In bacteria, depending on the composition of the complex, periplasmic nitrate reduction can either contribute to the generation of the proton gradient or serve as an electron sink, eliminating excess reducing equivalents from the cytoplasm 85.

A complete membrane A-type ATPase is predicted to be encoded in the genome of I. hospitalis, in contrast with only a subset of subunits in N. equitans 12. While N. equitans might be unable to synthesize ATP, the presence of a predicted nucleoside diphosphate kinase (Neq307) suggests that regeneration of the NDP pool is feasible, which might reduce its host dependency by recycling (Figure 7). Since it has few ion transporters and no genes encoding membrane hydrogenases or oxidoreductases, it is unknown if N. equitans can independently maintain a membrane potential or whether it needs to acquire such capabilities from its host.

As an obligate anaerobe, I. hospitalis requires a mechanism to deal with the toxicity of reactive oxygen species. A superoxide reductase is present (Igni1348) and could detoxify superoxide resulting from oxygen reduction by transition metals. According to a recently proposed mechanism 86, a ferrocyanide complex bound within the superoxide reductase active site may scavenge the superoxide by one-electron redox chemistry while the superoxide reductase iron site remains reduced. The resulting peroxide could be transferred to soluble organic compounds, resulting in the formation of alkyl peroxides that can be reduced by peroxiredoxin. A gene encoding a member of this family is encoded in the genome (Igni459) and a recent proteomic analysis of I. hospitalis in laboratory cultures has shown that its product is an abundant cytosolic protein 55.

Although the recognition and exchange mechanisms between I. hospitalis and N. equitans remain elusive, the available genomic and ultra-structural data suggest some possible ways of interaction between the two organisms. Since the transporters in both species are few and provide limited specificities, they are unlikely to comprise the main route of metabolite acquisition by N. equitans. Similarly, transfer of protein complexes to N. equitans from the host by secretion, especially for membrane components, would violate topological and signal sequence constraints of the translocation machinery. Potential vehicles for the transport of metabolites and proteins from I. hospitalis to N. equitans appear to be the large and variably shaped vesicles and tubes that emerge from the host's cytoplasm 910. Such structures could provide transient or even constant contact between the two cytoplasms once the physical contact between the cells has been established, possibly fulfilling the metabolic and energetic requirements of N. equitans. This would also allow it to carry out limited respiration, transport and ATP synthesis and may explain how detached N. equitans cells or cells not in direct contact with the host can survive for some time. Electron microscopy studies have indicated that some of the I. hospitalis periplasmic vesicles fuse with the outer membrane, which likely results in their contents being released into the environment 910. This release of small molecules and, perhaps, peptides might provide chemical cues to N. equitans for host recognition and attachment. Since neither of the two organisms appears to be motile, the actual mechanism by which they find each other and become attached in the turbulent hydrothermal vent environment remains enigmatic.

Recent ultra-structural and physiological studies have shown that a physical connection can form between the two organisms 362. Three-dimensional reconstructions point to a dynamic type of interaction, some N. equitans cells contacting the outer membrane of I. hospitalis in places where the host periplasmic space is wide and contains cytoplasmic vesicles while others are attached to regions with a very narrow periplasm and displaying fibrilar structures 62. The steps and molecular determinants of the cell-cell recognition and interaction and the membrane and periplasm dynamics remain uncharacterized. The cytoplasmic membrane of Ignicoccus itself is highly 'corrugated', as shown in sections and three-dimensional reconstructions, thereby increasing its surface significantly; in addition, it spontaneously evaginates in the absence of N. equitans 291062. Therefore, the physiological role of the conglomerate of tubes and vesicles and the significance of the wide periplasmic space probably extends beyond their possible connection to N. equitans. As energy generation resides at the level of the cytoplasmic membrane, these structures could provide a substantially increased respiratory surface confined in the space surrounded by the outer membrane, analogous to the eukaryotic mitochondrial cristae. Vesicles might concurrently transport specific lipids and proteins to the outer porous membrane, which in this case would serve not only as a protective barrier but also for controlling gas and solute exchange. This could represent a mechanism enabling Ignicoccus species to rely exclusively on the low energetic yield of the sulfur-hydrogen respiration to sustain an elevated turnover of cellular components at high temperature. Combined with the obligate CO₂ autotrophy and efficient metabolism, such adaptations might allow Ignicoccus to outcompete heterotrophs in colonizing emerging hydrothermal vent niches that are still poor in dissolved organic compounds.

I. hospitalis KIN4I cells (DSMZ strain 18386) were grown as described in 2. DNA was isolated from frozen cells using an alkaline lysis followed by proteinase K digestion method 87. Sequencing and assembly were performed at the DOE Joint Genome Institute, Walnut Creek, CA, USA using the standard microbial genome sequencing pipeline 88 based on a combination of 3-, 6- and 40-kb (fosmid) DNA libraries. The Phred/Phrap/Consed software package was used to assemble and assess quality 89. Possible miss-assemblies were corrected and gaps between contigs were closed by editing in Consed, custom primer walks or PCR amplification and sequencing. The estimated error rate in the completed genome sequence of I. hospitalis is less than 1 in 50 kbp.

Automated gene prediction was performed by using the output of Critica complemented with the output of Glimmer as part of the genome annotation pipeline at Oak Ridge National Laboratory (ORNL), Oak Ridge, TN, USA. The predicted coding sequences were translated and used to search the National Center for Biotechnology Information (NCBI) nonredundant database, UniProt, TIGRFam, Pfam, PRIAM, KEGG, COG, and InterPro databases. The tRNAScanSE tool 90 was used to find tRNA genes, whereas ribosomal RNAs were found by using BLASTn against the ribosomal RNA databases. The RNA components of the protein secretion complex and the RNaseP were identified by searching the genome for the corresponding Rfam profiles using INFERNAL 91. Transporter proteins were initially identified based on similarity to transporter categories in GOG and Pfam and were further analyzed using the Transporter Classification Database 92. Additional gene prediction analysis and manual functional annotation was performed within the Integrated Microbial Genomes (IMG) platform developed by the Joint Genome Institute, Walnut Creek, CA, USA 93. The complete genome sequence has been deposited in GenBank [GenBank:CP000816.1].

Analysis of the I. hospitalis and N. equitans genomes were carried out using the IMG system 93. The genes referred to throughout the text and figures correspond to the assigned open reading frame numbers in the two genomes. Putative operons were identified using the method of Overbeek et al. 94. Structure fold prediction of membrane proteins with no detectable similarity to other database sequences was performed using Phyre 95. To calculate the frequency of paralogs in the different archaeal genomes, blastclust analyses were performed using the translated coding sequences and varying the similarity threshold for sequence inclusion into clusters. To derive the amino acid usage statistics for archaeal genomes, the percentages of amino acids encoded within each protein were first calculated and used to determine the overall percentage for the whole proteome. The frequency for each amino acid use was then analyzed graphically relative to the GC content of the genome considering that the GC content can influence codon usage. Archaeal COG analyses for I. hospitalis and N. equitans were performed as described 27. Analysis of genome sizes of bacteria and archaea was based on genomic data available in IMG (March 2008 version). A table containing the accession numbers for all the genomes as well as all the numerical parameters and classification used in the analysis is provided as Additional data file 3.

To identify the potential presence of laterally transferred genes in I. hospitalis, we first used the Pyphy system 44 to automatically calculate individual phylogenetic trees for every gene in the genome. Briefly, each protein sequence was blasted against a local version of a non-redundant protein database (SWISS and TREMBL) and sequences with significant hits (<10e-6) were retrieved and aligned with the query sequence using CLUSTALW. Phylogenetic trees were then constructed using PAUP* with the neighbor joining and parsimony methods with 100 bootstrap replicates. Because the automatic 'phylogenetic connection' calculated by Pyphy and displayed as the phylome map of the genome was at times affected by poor bootstrap support values or unresolved trees, we visually inspected each tree and, when sufficient confidence was present, a broad phylogenetic connection to the Crenarchaeaota, Euryarchaeota or Bacteria was assigned to the I. hospitalis gene. For numerous genes, although the Ignicoccus gene was clearly of archaeal type, either the phylogenetic signal was insufficient or the evolutionary history of that gene across Archaea involved numerous potential LGTs. Such genes have been generically designated as 'archaeal'. When no close homologues for I. hospitalis genes were found or the phylogenetic trees included archaeal and bacterial genes but were not sufficiently resolved, such genes were designated as 'unknown' phylogenetic type. Finally, when the closest hit and the resulting phylogenetic trees indicated a N. equitans gene as the closest homologue, those genes were designated as potential LGTs within the N. equitans-I. hospitalis system. We also used the arCOG analysis to improve the phylotyping information for some of the functional gene categories that were not resolved by phylogenetic analysis.

Genes representing potential LGTs within the N. equitans-I. hospitalis system were subjected to a more extensive phylogenetic analysis. Sequence alignments were obtained using a combination of alternative methods as implemented on the M-Coffee web server 96. Following manual alignment curation and masking of regions with high variability that could not be confidently aligned, the amino acid substitution model best fit for each gene was chosen using the software Modelgenerator v84 97. Maximum likelihood phylogenetic trees were constructed using PhyML v2.4.4 98 using the parameters identified by Modelgenerator. Alternative tree topologies were also explored using a combination of the software Tree Puzzle and PROML/PHYLIP, as previously described 99. The protein sequence alignments used to generate the trees for several inferred laterally transferred genes are provided as Additional data file 4.