Introduction

Deep-sea microbiology is still poorly understood, despite the fact that the ocean is one of the largest biotopes on Earth and critically important for global nutrient cycling. Molecular diversity approaches based on the amplification, cloning and sequencing of 16S rRNA genes have revealed a variety of deep-sea archaeal, bacterial and picoeukaryotic lineages that are often very divergent from cultivated species. Analyses of metagenomic libraries from marine, including deep-sea, plankton provide a powerful strategy for obtaining additional information about the structure, ecology and evolution of microbial communities inhabiting the marine ecosystem by allowing access to genes of as yet uncultured microorganisms (Stein et al., 1996; Beja et al., 2000a, 2000b, 2002a, 2002b; López-García et al., 2004; Venter et al., 2004; DeLong et al., 2006; Martin-Cuadrado et al., 2007; Rusch et al., 2007).

Archaea have been detected throughout the oceanic water column and are quantitatively important members of picoplankton in the deep ocean, which suggests that organisms from this domain might have a fundamental role on global biogeochemical cycles (Karner et al., 2001; Francis et al., 2005; Herndl et al., 2005). Pelagic microbial communities include psychrotolerant, free-living, as yet uncultivated marine archaea, and might also include anaerobic, particle-attached, methanogenic archaea. So far marine archaea identified in planktonic 16S rRNA gene libraries or metagenomic libraries fall into one of the following four lineages: Group I within the Crenarchaeota, and Groups II–IV within the Euryarchaeota (DeLong, 1992; Murray et al., 1998; López-García et al., 2001; Bano et al., 2004). Recently, a new subgroup of crenarchaeota distantly related to the extensive Group I was identified, the so-called Group 1A, which clustered some fosmid sequences from the Pacific ALOHA water column (DeLong et al., 2006) with the environmental clone pSL12 from a Yellowstone hotspring (Barns et al., 1996).

Being elusive to isolation and growth in pure culture, little is known about marine archaeal physiology, and only two complete genomes from non-thermophilic crenarchaeota are available to date, that of Nitrosopumilus maritimus SCM1, and that of the sponge symbiont Cenarchaeum symbiosum A (Hallam et al., 2006a). N. maritimus, which was isolated from a laboratory fish tank to become the first cultivated representative of Group I Crenarchaeota, grows by aerobically oxidizing ammonium to nitrite, a metabolic ability previously unknown for the archaea (Konneke et al., 2005). The abundance of Group I archaeal cells in the deep ocean paralleled with significant expression levels of archaeal genes putatively encoding the ammonia monooxygenase subunit A (amoA) strongly supports the interpretation that at least a large fraction of the marine crenarchaeota are nitrifiers (Wuchter et al., 2006; Lam et al., 2007). Given the Earth's volume occupied by deep oceans, this would render archaea critical players in the N cycle at a global scale. Furthermore, deep-sea ammonium-oxidizing archaea may also be ecologically important for carbon cycling, as they appear to be autotrophic (Ingalls et al., 2006) or at least mixotrophic. Indeed, C. symbiosum possesses the enzyme genes for the newly described 3-hydroxyproprionate/4-hydroxybutyrate autotrophic CO2 fixation pathway (Hallam et al., 2006a, 2006b; Berg et al., 2007) and the Global Ocean Sampling database contains high proportions of genes for one key enzyme in this route, suggesting that this archaea-specific pathway is widely distributed in oceans (Berg et al., 2007; Rusch et al., 2007). However, although shot-gun sequencing approaches (for example, Global Ocean Sampling analyses) allow the association of functional genes to given environments, linking metabolic genes to specific phylogenetic lineages directly is only possible for very low-diversity biotopes (Tringe et al., 2005). By contrast, sequencing of large genomic clones from metagenomic libraries allows assigning functional genes to particular lineages provided that appropriate phylogenetic markers colocalize in the same genome fragment. Through this approach proteorhodopsin-like genes, possibly acquired by horizontal gene transfer from phototrophic bacteria (Beja et al., 2000a), were found in bona fide Group II Euryarchaeota inhabiting photic layers, though not in euryarchaeota from deeper waters (Frigaard et al., 2006).

The structure and gene content of a limited number of sequenced large metagenomic clones carrying ribosomal RNA genes affiliated to marine Group I Crenarchaeota and Group II Euryarchaeota have been published to date (Stein et al., 1996; Schleper et al., 1998; Beja et al., 2000b, 2002a; López-García et al., 2004; Moreira et al., 2004; Frigaard et al., 2006). These studies were fundamentally descriptive; a few of them compared the sequences of two genomic fragments containing identical or nearly identical 16S rRNA sequences. The first of them compared two strains of C. symbiosum, which revealed an unexpected degree of genomic variation that might imply considerable functional diversity within single populations of coexisting microbial strains (Schleper et al., 1998, 2005). Likewise, a pioneer comparison of two large genomic fragments of planktonic crenarchaeota from Antarctic surface water (fosmid 74A4) and North Pacific 200-m deep plankton (fosmid 4B7) reported significant genomic divergence between fragments with relatively similar 16S rRNA gene sequences (Beja et al., 2002a). This suggested that this phylogenetic marker might not reflect the actual genomic and physiological diversity underlying the species or operational taxonomic unit (OTU) level, which is usually considered to be at ⩾97–98% nucleotide identity at this locus.

Comparing genomic sequences of the same OTUs from different geographic locations or from samples characterized by different environmental parameters can reveal interesting information about the microevolution and ecological adaptations of marine archaea. Yet genomic data for deep-sea Group I Crenarchaeota and Group II Euryarchaeota are rather poor, with most of the genomic information available (including the two complete genome sequences) corresponding to DNA samples or isolates from the photic zone. Furthermore, so far there is no genomic information about members of Euryarchaeota Groups III and IV and Crenarchaeota Group 1A. In an effort to generate additional information about marine deep-sea planktonic archaea, we have sequenced and analysed 22 archaeal genome fragments (∼770 Mb) identified in metagenomic fosmid libraries of deep plankton from geographically distant and very different oceanic water masses: Ionian Sea (3000 m), Adriatic Sea (1000 m), South Atlantic (1000 m) and Antarctic Polar Front (500 m). We present the first genomic data about Group 1A Crenarchaeota and Group III Euryarchaeota. Comparative genomic analyses of Groups I and II genomic fragments belonging to the same OTUs (based on 16S rDNA similarity) from different geographical locations suggest markedly different genome dynamics between members of these two archaeal groups.

Materials and methods

Sample collection and metagenomic library construction

The characteristics of the metagenomic planktonic libraries from which a selection of archaeal genomic clones are analysed in this study are summarized in Table 1. The 500 m-deep Polar Front cosmid library (DeepAnt) was constructed in two sequential steps yielding, initially, 6107 clones (López-García et al., 2004) and later extended by 3200 additional clones (Moreira et al., 2006). The 3000-m deep Ionian Sea fosmid library (KM3) was constructed as described previously (Martin-Cuadrado et al., 2007). For the 1000-m deep South Atlantic library (SAT1000), 100 l sea water was collected in December 1998 during the oceanographic cruise DHARMA98 and for the 1000-m deep Adriatic library (AD1000), 200 l was collected during the 2003 May Urania cruise in the Gulf of Manfredonia. In both cases, sea water was collected with a CTD rosette and sequentially filtered on board through a 5 μm pore size polycarbonate filter and through 0.22-μm pore size Sterivex filters (Durapore; Millipore, Billerica, MA, USA) using a peristaltic pumping system. Sterivex filters retaining the 0.2–5 μm diameter planktonic cells were conserved in lysis buffer (40 mM EDTA, 50 mM Tris/HCl, 0.75 M sucrose) at −20 °C until DNA extraction. Filters were thawed on ice and then treated with 1 mg ml−1 lysozyme and 0.2 mg ml−1 proteinase K (final concentrations). Nucleic acids were extracted with phenol–chloroform-isoamyl alcohol and chloroform–isoamyl alcohol, and then concentrated using a microconcentrator (Centricon 100; Amicon, Millipore, Billerica, MA, USA). DNA integrity was checked by agarose gel electrophoresis. Fosmid genomic libraries were constructed from approximately 0.2 μg (SAT1000) and 10 μg (AD1000) of DNA from the 0.2- to 5-μm plankton fraction using the CopyControl Fosmid Library Production Kit (Epicentre) as described by the manufacturer's instructions yielding 5413 and 38 704 clones, respectively. All fosmid libraries were constructed and stored at the University of Paris-Sud, Orsay.

Table 1 Characteristics of sampling sites and planktonic metagenomic libraries used in this study
Full size table

Screening of libraries and sequencing of genomic clones

The DeepAnt and KM3 libraries were screened for the presence of archaeal 16S rRNA genes as previously described (López-García et al., 2004; Martin-Cuadrado et al., 2007). Clones from the AD1000 and SAT1000 libraries were pooled in groups of 96, and DNA was extracted from pooled cultures using the QIAprep Spin Miniprep Kit (Qiagen, Valencia, CA, USA). The libraries were PCR-screened for the presence of archaeal 16S rRNA genes using combinations of the archaea-specific forward primers 21F (5′-TTCCGGTTGATCCTGCCGGA), Ar109 (5′-AC(G/T)GCTGCTCAGTAACACGT), ANMEF (5′-GGCTCAGTAACACGTGGA) and the prokaryotic-specific reverse primer 1492R (5′-GGTTACCTTGTTACGACTT). PCR reactions were carried out under the following standard conditions: 35 cycles (denaturation at 94 °C for 15 s, annealing at 50 °C for 30 s, extension at 72 °C for 2 min) preceded by 2 min denaturation at 94 °C and followed by 7 min extension at 72 °C. PCR products were partially sequenced. A total of 28, 30 and 9 archaeal 16S rDNA sequences were identified in the KM3, AD1000 and Sat1000 libraries, respectively. After partial sequencing and preliminary phylogenetic analyses, 12 fosmid clones affiliating to crenarchaeota and 9 to euryarchaeota were selected for complete sequencing. Sequencing was carried out at the Broad Institute at the Massachusetts Institute of Technology thanks to the Betty and Gordon Moore Foundation ‘Marine Microbiology Initiative’ (http://www.moore.org/marine-micro.aspx). An additional euryarchaeotal clone, DeepAnt-15E7, which we had identified and sequenced (Genome Express, Meylan, France) from a cosmid library of 500-m deep picoplankton at the Antarctic Polar Front (López-García et al., 2004) was included in this analysis, making a total of 22 archaeal clones analysed (Table 2). All sequences were obtained in a single contig except KM3-136-D10 and SAT1000-49-D2, which were assembled in two different ones.

Table 2 General information about the archaeal genome fragments analysed in this study from various metagenomic libraries
Full size table

Annotation and analysis of genome fragments

Protein coding genes were predicted using the annotation package GLIMMER (Delcher et al., 1999), SEEDs (Overbeek et al., 2005) and were further manually curated. Spacers were subsequently searched against the non-redundant database (http://www.ncbi.nlm.nih.gov/) using BLAST (Altschul et al., 1997) to ensure that no open reading frame (ORF) had been missed. Identified ORFs were compared to known proteins in the non-redundant database using BLASTX. All hits with an E-value greater than 10−5 were considered nonsignificant. COGNITOR was used for COG (Clusters of Orthologous Groups) assignments as well as COG functional categories (Tatusov et al., 2001). tRNAs were identified using tRNAscan-SE (Lowe and Eddy, 1997). Putative protein transmembrane domains were predicted using TMHMM 2.0 (Krogh et al., 2001). The GEECEE and CUSP programs from the EMBOSS package (Rice et al., 2000) were used to calculate GC content and codon usage, respectively. For comparative analyses, reciprocal BLASTN and TBLASTXs searches between the different fosmids were carried out, leading to the identification of regions of similarity, insertions and rearrangements. To allow the interactive visualization of genomic fragment comparisons, we used Artemis Comparison Tool ACTv.6 (Carver et al., 2005) and MAUVE (Darling et al., 2004). To identify possible recombination regions between fosmids, we applied different programs available in the RDP version program package (Martin et al., 2005), including GENECONV (Sawyer, 1999), RDPv.beta14 and MaxChi (Posada and Crandall, 2001). The SimPlot graphic (Ray, 1998) shown in Figure 6 was done with the following parameters: window 1000 pb, step 20 pb, GapStrip: On, Kimura ( two parameter) and T/t 2,0.

Figure 6
figure 6

Detection of recombination at the translation elongation factor 2 (EF-2) locus within the crenarchaeotal OTU A (SAT1000-23-F7, AD1000-202-A2 and KM3-34-D9). The curves show the similarity at nucleotide level of AD1000-202-A2 and KM3-34-D9 to the region comprised between open reading frames (ORFs) 22 and 36 in fosmid SAT1000-23-F7. The inlets on the right show the phylogenetic relationships (neighbour-joining trees) retrieved within the OTU A from EF-2 compared to the concatenate of the different ORFs in this region.

Full size image

Phylogenetic analysis

Archaeal 16S rRNA gene sequences detected in genomic clones were aligned using MUSCLE (Edgar, 2004) with their closest relatives in databases as identified by BLAST (Altschul et al., 1997), those from the ALOHA water column (DeLong et al., 2006) and those from available Group I crenarchaeotal genomes and selected Groups I and II archaeal genome fragments. Alignments were manually edited using the ED program of the MUST package (Philippe, 1993). Gaps and ambiguously aligned positions were excluded from our analyses, yielding 1184 and 1118 positions available for constructing the trees of crenarchaeota and euryarchaeota, respectively. Maximum likelihood trees were reconstructed using TREEFINDER (Jobb et al., 2004) applying a general time reversible model of sequence evolution, and taking among-site rate variation into account by using a four-category discrete approximation of a Γ distribution plus invariant sites. ML bootstrap proportions were inferred using 1000 replicates. Phylogenetic trees were viewed using the program TREEVIEW (Page, 1996). For the intergenic transcribed spacer (ITS) analysis (Figure 2b), a neighbour-joining tree was constructed using nearly all positions (159), including gaps to maximize the signal, using the programs available in the MUST package. Protein sequence alignments were generated with MUSCLE and ClustalW (Chenna et al., 2003) and edited manually as necessary. Gblocks was used to eliminate ambiguously aligned positions (Castresana, 2000). Phylogenetic analyses were then performed using the MEGA4 phylogenetic tool software package (Zmasek and Eddy, 2001; Kumar et al., 2004). To construct the concatenate of the euryarchaeotal Group II ribosomal proteins only L3, L4, L23, L2/LB, S19, L22, S3, L29, S5, L30 and L15 were used to include KM3-136-D10 in the analysis.

Figure 2
figure 2

(a) Maximum likelihood phylogenetic tree based on the rRNA genes contained in the fosmids of crenarchaeota analysed in this study. Fosmid clone names indicate the metagenomic library of provenance: KM3, Ionian Sea, 3000 m depth; AD1000, Adriatic Sea, 1000 m depth; SAT1000, South Atlantic, 1000 m depth; DeepAnt, Antarctic Polar Front, 500 m depth; HF, North Pacific followed by a number indicating their depth (DeLong et al., 2006). Previously sequenced environmental fosmid clones are indicated as such, the rest of clone names correspond to environmental sequences, [accession number]. (b) Neighbour-joining tree of the intergenic transcribed spacer (ITS) sequences adjacent to the 16S rRNA gene in various crenarchaeotal fosmids.

Full size image

Accession numbers

The sequences have been deposited in GenBank under the accession numbers AC215088–AC215101, AC215103–AC215109 and EU556724.

Results

Archaeal fosmids in deep-sea metagenomic libraries

To gain insights into the genomic diversity of deep-sea marine archaea, we screened four metagenomic libraries (Table 1) of meso- and bathypelagic plankton (size fraction 0.2–5 μm) from various oceanic locations looking for the presence of genomic clones containing archaeal 16S rRNA genes, a selection of which were completely sequenced and analysed. Two of the metagenomic libraries had been constructed previously: the cosmid library DeepAnt (Antarctic Polar Front, 500 m depth) (López-García et al., 2004; Moreira et al., 2006) and the fosmid library KM3 (Ionian Sea, 3000 m depth) (Martin-Cuadrado et al., 2007). For this study, we constructed and screened two additional fosmid libraries: SAT1000 (South Atlantic, 1000 m depth) and AD1000 (Adriatic Sea, 1000 m depth). With 9307, 20 757, 5413 and 38 704 genomic clones, respectively, that is, a total of ∼74 000 genomic clones with an average insert size of 35 kb; this represents approximately 2.6 Gb of stored genetic information (Table 1).

These libraries were scanned by PCR using various 16S rDNA archaeal-specific primer combinations revealing a relatively high proportion of archaeal clones (López-García et al., 2004; Martin-Cuadrado et al., 2007 and this study). We obtained approximately 75 archaeal 16S rDNA sequences from all the libraries, which were compared by BLAST to sequences in databases and classified in different archaeal groups after a preliminary phylogenetic analysis (data not shown). Using the 16S rDNA-containing clones as a proxy, the relative abundance of different archaeal groups appeared to vary depending on the libraries (Figure 1). Although Group I Crenarchaeota are thought to constitute the majority of the archaeal fraction in deep waters (Karner et al., 2001; DeLong et al., 2006), our study revealed important differences among libraries, with Groups II and III Euryarchaeota dominating some of them. This might partly reflect biases in euryarchaeota detection due to differential primer/probe specificity, a problem that is limited in the present study by the use of a variety of primer combinations. In any case, it certainly reflects real variation in archaeal community composition of different sea water masses and geographic locations. Since the Mediterranean basin is rather warm and with very particular physico-chemical characteristics (Table 1), different archaeal composition would be expected for the Adriatic and Ionian libraries compared to open ocean libraries. However, while crenarchaeota appeared dominant, though not overwhelmingly, in the Ionian KM3 library, they accounted for less than one-third of the archaea in the Adriatic one (Figure 1). Similarly the South Atlantic and Antarctic Polar Front libraries, from geographically close sampling stations (Table 1) albeit from quite different water masses (Supplementary Figure S1), exhibited an opposite dominance of archaeal groups, with crenarchaeota clearly dominating the SAT1000 library and Group II Euryarchaeota the DeepAnt one. Notably, the usually scarce Group III Euryarchaeota appeared to be the most represented euryarchaeotal clade in the South Atlantic library.

Figure 1
figure 1

Relative proportion of archaeal genomic clones containing 16S rRNA genes in the metagenomic libraries from meso- and bathypelagic plankton used in this study.

Full size image

From the preliminary 16S rDNA archaeal phylogenetic tree, we selected 22 archaeal genomic clones for complete sequencing and comparative analysis based on two major criteria: (i) they represented lineages about which there was no genomic information available as yet (for example, Group 1A Crenarchaeota, Group III Euryarchaeota) or (ii) they had nearly identical 16S rDNA sequences and could therefore be ascribed to a single OTU (here defined as lines having 16S rDNAs with ⩾98% identity) but came from distant oceanic locations (Table 2; Figures 2 and 3). Of the 22 selected fosmids, 12 belonged to the Crenarchaeota and, of these, 3 were more closely related to the recently identified Group 1A, initially defined by DeLong et al. (2006) on the basis of the environmental sequence pSL12 and the Pacific ALOHA fosmids HF770-041I11 and HF4000-039O16. However, the monophyly of the Group 1A with the addition of these environmental sequences remains only weakly supported, the marine planktonic sequences forming a clade distinct from the Yellowstone hot spring clones pSL12, OPPD032 and YNP-ObP-A62 (Figure 2a). The remaining nine crenarchaeotal fosmids belonged to the widespread Group I and comprised three distinct cosmopolitan OTUs (A, B and C) with, for each, one fosmid from the Adriatic, one from the Ionian and a third one from the South Atlantic (Figure 2a). In support of their pan-oceanic distribution, these three OTUs included environmental sequences from yet another oceanic basin, the Pacific Ocean, at the ALOHA water column (HF clones) and the Suiyo Seamount (SSM264-EA01, MaSc-NEA01 and MaSc-NEA03) (Figure 2a). From the 10 euryarchaeotal genomic clones sequenced, 3 were ascribed to the Group III and the rest to deep-sea Group II members, as none of the 16S rDNA sequences from our libraries branched with the phototrophic euryarchaeota (Frigaard et al., 2006). The recently isolated Aciduliprofundum boonei, an obligate thermoacidophilic sulphur- or iron-reducing heterotrophic deep-sea vent archaeon representative of the DHVE2 group (Reysenbach et al., 2006), branched closer to the Thermoplasmatales than to the Groups II and III (Figure 3 and data not shown). As in the case of crenarchaeota, we sequenced fosmids from the same pan-oceanic OTUs coming from at least one Mediterranean and one South Atlantic or Antarctic location. Again, that these OTUs are cosmopolitan was further supported by the formation of tight clades with Pacific sequences from the ALOHA water column (HF clones), Suiyo Seamount hydrothermal sea water (SSM263-NA03) or ridge flank crustal fluids in the North East Pacific (pIVWA104, CTD005-33A) (Huber et al., 2006) (Figure 3).

Figure 3
figure 3

Maximum likelihood phylogenetic tree based on the rRNA genes contained in the fosmids of euryarchaeota analysed in this study. Fosmid clone names indicate the metagenomic library of provenance: KM3, Ionian Sea, 3000 m depth; AD1000, Adriatic Sea, 1000 m depth; SAT1000, South Atlantic, 1000 m depth; DeepAnt, Antarctic Polar Front, 500 m depth; HF, North Pacific followed by a number indicating their depth (DeLong et al., 2006). Previously sequenced environmental fosmid clones are indicated as such, the rest of clone names correspond to environmental sequences, [accession number].

Full size image

The archaeal genomic fragments sequenced ranged from 22.6 to 41.8 kb, with an average size of 35 kb (Table 2). All fosmids were assembled in a single contig with the exception of KM3-136-D10 and SAT1000-49-D2, which were assembled in two contigs. Through alignments between SAT1000-49-D2 and the other fosmids, we concluded that the most probable orientation of the two contigs was that shown in Figure 4, and we joined them artificially by a stretch of 650 ‘N’ that most likely represented the region missed during sequencing. KM3-136-D10 appeared to lack a larger region, so that we kept the two contigs separated in our analyses. On average, gene density was relatively high, with >90–91% coding regions in all fosmids except those from Group I Crenarchaeota, which had slightly less dense coding regions (86.9% on average). These genome fragments were annotated and the gene content and organization compared within the different archaeal groups as follows.

Figure 4
figure 4

Comparative genomic organization of fosmids from crenarchaeota. A neighbour-joining tree relating the different 16S rRNA markers found in the respective fosmids is shown on the left as reference. Conserved genomic regions between fosmids are indicated by grey shaded areas, grey intensity being a function of sequence similarity by BLASTN. ORFs are numbered from left (ORF 1 in each case) to right as shown in Supplementary Table S1. Particular ORFs mentioned in the text are highlighted by a graphic code. Note that SAT1000-49-D2 is a composite of two contigs joined by a stretch of 650 nucleotides (labelled as ‘N’).

Full size image

Genomic organization and detection of high intra- and inter-genomic recombination levels in Group I Crenarchaeota

All Group I Crenarchaeota genome fragments had a very low GC content (Table 2), with an average of 35.9%, a value closer to that of the free-living N. maritimus SCM1 genome (33.9%) than to that of the symbiont C. symbiosum (57.4%). Genome organization was remarkably variable (Figure 4), despite the fact that the nine Group I crenarchaeotal fosmids sequenced formed three well-defined, though rather close (especially OTUs B and C), OTUs in terms of 16S rRNA phylogeny (Figure 2a). This variability was particularly evident at the level of gene and genome fragment rearrangements within and among OTUs (Figure 4), in spite of similar overall gene content, gene similarity (Supplementary Table S1) and small portions of co-linear regions being conserved. This is suggestive of a high level of intra-genomic recombination leading to gene shuffling.

Remarkably, ITS between the 16S and 23S rRNA genes, which are generally very variable, were extremely conserved. Thus, although they were rather short (much shorter than, and very different from, those of soil crenarchaeota), having 130 bp on average, their phylogenetic analysis allowed retrieving trees that were overall congruent with 16S rRNA trees even using simple p-distance-based trees considering all positions (including gaps) (Figure 2b and Supplementary Figure S1). This degree of conservation had been observed previously in environmental sequences (Garcia-Martinez and Rodriguez-Valera, 2000) and the Group I crenarchaeote clone DeepAnt-EC39 (López-García et al., 2004). The most obvious difference with 16S rRNA trees was the split of OTU C, since the ITS sequence of the OTU C clone SAT1000-49-D2 was identical to, and branched with, those of OTU B (Figure 2 and Supplementary Figure S2). This behaviour was suggestive of a recombination event between members of these two OTUs. To detect other possible recombination events between the different OTU clones at the 16S-23S rDNA transition region, we made p-distance-based phylogenetic trees using different contiguous windows along an alignment of the 16S rDNA+ITS+23S rDNA including all positions (Supplementary Figure S2A). The tree obtained with the last 300 positions of the 16S rRNA gene was congruent with the maximum likelihood phylogenetic tree obtained for this whole gene (Figure 2a). OTUs A, B and C were clearly distinct and the clone DeepAnt-EC39 branched with moderate support at the base of OTU C. The tree obtained using the adjacent 189 positions corresponding to the ITS region showed, as already seen, the clustering of OTU C clone SAT1000-49-D2 with OTU B (100% identity). DeepAnt-EC39 displayed a longer branch within the OTU B, which might be explained by recombination with environmental lines (compare Figure 2b and Supplementary Figure S2A). The tree obtained using the contiguous first 300 positions of the 23S rRNA gene was surprising as it showed an extreme high conservation (100% identity) of this region within the three OTUs, A, B and C, DeepAnt-EC39 being now identical to members of the OTU C. This level of conservation is possibly not related to a particular essential function of this region of the 23S rRNA in Group I Crenarchaeota, as the phylogenetic distance between N. maritimus and the fosmid 74A4, which can be used as internal control in the analyses, remains approximately the same in all the trees made. Finally, the tree obtained with the following 300 positions in the 23S rRNA (positions 300–600) retrieved the same phylogenetic relationships as did the 16S rRNA partial and global trees and the rest of the 23S rRNA gene.

Taken together, these observations strongly suggest that this region of the rrn operon is a hotspot for inter-genomic homologous recombination. This would explain why the ITS, in principle not subjected to strong purifying selection, is so astonishingly conserved within and between different OTUs. Intra-chromosomal recombination at the rrn operon is unlikely since these organisms possibly bear a single rrn operon. Two elements support this idea. First, C. symbiosum and N. maritimus genomes harbour one single rrn operon and, second, though synteny was not perfectly conserved, there were more or less long syntenic regions involving all the Group I Crenarchaeota fosmids examined here, which would be extremely improbable if these crenarchaeota contained more than one rrn operon (Figure 4). Therefore, recombination likely takes place with exogenous genomic DNA from more or less closely related archaea.

Regarding protein gene content and organization, some protein genes were very well conserved in our fosmids. These included, immediately upstream the 16S rDNA, the genes encoding glutamate semialdehyde aminotransferase, a tetra-tricopeptide-repeat-rich protein (TPR) and an unknown conserved transmembrane protein (Figure 4 and Supplementary Table S1). Though not equally conserved, genes for a NADP(H)-dependent nitroreductase or the transcriptional regulator AsnC, followed by a thioredoxin reductase were often found immediately downstream of the 23S rRNA gene. BLASTP comparisons of the proteins found in these Group I crenarchaeotal fosmids with the non-redundant database (E-value cutoff of 1e−5) showed that most of them had best hits with proteins of N. maritimus SCM1 (56.8% of blasted proteins), C. symbiosum A (11.3%) and the previously described marine crenarchaeotal clones DeepAnt-EC39 (21.0%), 4B7 (5.7%), 74A4 (0.7%) as well as other uncultured crenarchaeota. Hits with thermophilic crenarchaeota (Sulfolobus acidocaldarius DSM 639) were observed only occasionally (0.6%). In 1.4% of the cases, the best hits were bacterial proteins (0.7% of them were indeed cyanobacteria) followed by a slightly less similar protein in the N. maritimus SCM1 or C. symbiosum A genomes. These observations might be consistent with previous analysis of DeepAnt-EC39 showing that some genes had probably been acquired by horizontal gene transfer, likely by an ancestor of Group I Crenarchaeota, from euryarchaeota or bacteria (including well-supported cases of transfers from cyanobacteria) (López-García et al., 2004).

Although all the genes found in our bathypelagic fosmids had orthologues in both N. maritimus SCM1 and C. symbiosum A, overall synteny was not maintained, except for the three genes upstream the 16S rDNA mentioned above. However, co-linearity was preserved in clusters of genes that were positioned in varying locations along the fosmids suggesting the occurrence of frequent rearrangements and recombination (Figure 5). To explore further the occurrence of recombination events at protein coding genes in our Group I crenarchaeotal fosmids, we looked for potential recombination spots in conserved syntenic regions within the different OTUs (Supplementary Tables S2 and S3). In the case of OTU A comprising fosmids AD1000-202-A2, KM3-34-D9 and SAT1000-23-F7, we aligned an overall syntenic region of 19 652 bp corresponding to ORFs 22–36 in SAT1000-23-F7, which included the glutamate semialdehyde aminotransferase, TPR and adjacent hypothetical protein conserved genes as well as the translation elongation factor 2 (EF-2), a glucose dehydrogenase, a nitroreductase, two transcriptional regulators and seven additional conserved hypothetical protein genes (Supplementary Table S1). The Mediterranean KM3-34-D9 and the South Atlantic SAT1000-23-F7 had 93.85% nucleotide identity on average, whereas that between the two Mediterranean clones (Adriatic and Ionian) KM3-34-D9 and AD1000-202-A2 was slightly lower, 88.11%. These values supported the inclusion of these fosmids within a single OTU, in accordance with values expected for a single prokaryotic species (Konstantinidis and Tiedje, 2005). These data argue, again, in favour of true oceanic ubiquity for this OTU. We scanned this alignment with recombination breakpoint-detection methods and found up to eight potential sites (Supplementary Table S2). Among them, one of the most evident inter-genomic recombination events leading to genetic exchange within populations involved the EF-2 gene, as it exhibited a much higher similarity with its homologue in AD1000-202-A2 (96.2% nucleotide identity) than to KM3-34-D9 (75.3%) in contrast to the rest of the aligned region (Figure 6).

Figure 5
figure 5

Regions of co-linearity shared by some of the deep-sea crenarchaeota fosmids studied here and the genome of Nitrosopumilus maritimus and crenarchaeotal genomic clones 4B7 and 74A4.

Full size image

Considering together the six genomic clones that defined the closely related OTUs B and C, they exhibited lower resemblance of gene organization among them than OTU A fosmids. This might imply a higher intra-genomic recombination rate, at least in the area surrounding the rrn operon, resulting in extensive genome reshuffling. For instance, in fosmid KM3-47-D6, synteny was restricted nearly exclusively to the genes upstream the rrn operon. Furthermore, 96% of the ORFs downstream of the rrn operon were not at all syntenic between OTU C clones KM3-47-D6 and AD1000-207-H3, and OTU B clones AD1000-56-E4 and SAT1000-21-C11. Nevertheless, we found, once again, clusters of genes that kept local synteny while being scrambled within and between members of OTUs B and C (Figure 4). We were able to align a fragment of 9663 bp corresponding to a syntenic region covering the ORFs 26–29 in SAT1000-49-D2, including also the genomic clone DeepAnt-EC39. Scanning this alignment with recombination breakpoint-detection programs, up to 9 potential recombination sites were identified (Supplementary Table S3). Inter-genomic recombination events might even involve members of more distant OTUs. Thus, one of the potential recombination events detected involved the NADPH nitroreductase of the OTU C fosmids KM3-47-D6 and SAT1000-49-D2, which showed an atypical clustering, being much more similar to that of the OTU A fosmid KM3-34-D9 (96.2 and 95.8% nucleotide identity, respectively) than to any of their homologous counterparts in their respective OTUs. In addition, intra-genomic recombination can lead to the translocation of relatively big genome fragments. For example, a set of genes keeping strict or nearly strict synteny is observed downstream of the rrn operon in several, but not all, members of OTUs B and C but upstream the rrn operon in the sequences belonging to OTU A (boxed regions in Figure 4). This translocation is about 16 000 bp long and includes genes for two oligopeptide permeases, a helicase, the chaperone DnaJ, a membrane protein probably implicated in the uptake of heavy metals (DedA), two TPR repeat-containing proteins, a SAM domain protein, several hypothetical proteins and EF-2.

Taken together, all these observations provide strong support for the idea that these Group I archaeal genomes have high intra- and inter-genomic recombination rates, at least in the vicinity of the rrn operon. This conclusion was further strengthened by the observed ratio of protein genes having the same transcriptional sense as the rrn operon versus those with an opposite orientation. As can be seen in Figure 7, only 50–60% of protein genes in our Group I Crenarchaeota fosmids had the same orientation as the rrn operon. These values were strikingly different from those observed for the Group II Euryarchaeota genomic clones (see below) revealing different genome dynamics for these two archaeal groups.

Figure 7
figure 7

Relative proportions of coding sequences with opposed transcriptional orientations in the archaeal genomic clones annotated in this study including all open reading frames (ORFs) (higher panel) or excluding the spc operon encoding the ribosomal proteins (lower panel). *Data derived from two contigs.

Full size image

Metabolic potential of bathypelagic Group I Crenarchaeota

Most of the genes identified encoded proteins involved in informational processes and many corresponded to conserved hypothetical proteins also present in other marine crenarchaeota (Supplementary Table S1). However, we also identified some genes that might provide clues as to the metabolic potential of these microbes. Thus, we found several genes encoding 3-hydroxipropionate cycle proteins involved in the putative carbon fixation pathway in the chemolithotrophic metabolism of marine crenarchaeota (Hallam et al., 2006a, 2006b). KM3-47-D6 contained a gene for a methylmalonyl-CoA mutase (mcmA1) and a methylmalonyl-CoA epimerase (mce), and KM3-86-C1 and SAT1000-49-D2 harboured an enoyl-CoA hydratase gene (fadB). In AD1000-202-A2 and KM3-34-D9, a region downstream of the rrn operon contained a gene for biotin ligase (birA), a protein required for assembly and activation of the carboxylase complex (Rodionov et al., 2002). Next to birA genes was a succinyl-CoA synthetase (alpha subunit), an enzyme that mediates the formation of succinyl-CoA from succinate in the reductive tricarboxylic acid pathway. Though these genes are not the key enzymes of the reductive tricarboxylic acid pathway cycle, the 3-hydroxipropionate pathway, or its variant, the 3-hydroxypropionate/4-hydroxybutyrate autotrophic CO2 assimilation pathway recently discovered in archaea, including C. symbiosum and N. maritimus (Berg et al., 2007), their presence in these archaeal fosmids might indeed indicate that CO2 fixation is possible in these organisms.

Multiple membrane transporters were tentatively identified on the basis of transmembrane segment predictions (Supplementary Table S1), but due to low similarity to any other transporters in the database, it is difficult to predict the kind of molecules that they might incorporate into the archaeal cells. A urea transporter was identified using BLASTX at the 5′ end of KM3-86-C1, which had the highest similarity to a homologous sequence in C. symbiosum A (80% amino-acid similarity) followed by several eukaryotic hits, the closest of which came from the picoplanktonic alga Ostreococcus tauri (66% amino-acid similarity). However, neither N. maritimus SCM1 nor other sequenced archaea or archaeal fragments possessed a homologue of this gene. From the rest of transporter genes, only two, found in fosmids SAT1000-49-D2, SAT1000-23-F7 and AD1000-207-H3, could be clearly assigned to oligopeptide or dipeptide permease proteins (OppC and DppB types) (Supplementary Table S1). The potential for incorporating oligopeptides in the cell, together with the possession of carbon fixation genes, suggests that, similarly to C. symbiosum A (Hallam et al., 2006a, 2006b), these meso- and bathypelagic crenarchaeota may behave as mixotrophs.

Genome fragments from Group 1A Crenarchaeota

The rest of our crenarchaeotal clones, namely, KM3-153-F8, AD1000-325-A12 and AD1000-23-H12, fell apart in the recently defined Crenarchaeota Group 1A. The clones studied here exhibited 16S rDNA identity levels below 97%, representing quite distinct OTUs, particularly AD1000-23-H12, with less than 86% 16S rDNA identity with the two others (Figure 2). Group 1A appears scarce in open oceanic libraries (DeLong et al., 2006; Figure 1). However, in a previous study of the Ionian sampling site (Zaballos et al., 2006) several 16S rRNA sequences (clusters Arch 1, 2, 3 and 4) that can be ascribed to the 1A group were detected, although none was found in samples of similar depth from the Greenland Sea, indicating that this group might be more abundant in the bathypelagic Mediterranean. Along these lines, we identified and sequenced three clones in our Mediterranean libraries, but failed to detect any in the South Atlantic or Polar Front samples. This is the first report of genomic sequences from members of this group. Their average GC content was very low, 33.8, 31.5% if we exclude 16S rDNA (having 52.37–53.98% GC) (Table 2 and Supplementary Table S1). This value approached the average GC content in the rest of free-living planktonic Group I Crenarchaeota (35.9%), including the sequence of N. maritimus SCM1 (33.9%).

In accordance with their distant phylogenetic positions, there was no synteny between AD1000-23-H12 and the two other clones beyond the rrn operon (Figure 4 and Supplementary Table S1). Alternatively, since the number of ribosomal operons of members of the Group 1A is not known, we cannot rule out that the region sequenced belongs to a different ribosomal RNA operon. Also, as expected, the genes identified in all these three genomic fragments were totally different from those found in Group I Crenarchaeota. The lack of other Group 1A crenarchaeotal sequences in databases may partially explain why only 52.5% of the best BLAST hits were crenarchaeotal proteins. The remaining best BLAST hits were bacterial (31.2%) and euryarchaeotal (16.2%) proteins, a situation already observed during the analysis of the Group I clone DeepAnt-39 that was suggestive of high horizontal gene transfer rates (López-García et al., 2004) (Supplementary Table S1).

The clones that were phylogenetically closer, AD1000-325-A12 and KM3-153-F8 (93.9% 16S rRNA identity) exhibited good synteny in their relatively small overlapping region (Figure 4). However, mean gene similarity was relatively low (43.7% nucleotide identity and 65.45% amino-acid similarity over the 8356 bp region that could be aligned excluding the rrn operon). This shared region upstream of the rrn operon contained a cluster of genes involved in the synthesis of aromatic amino acids. Downstream of it, both fosmids bore a luciferase-like protein, the next two genes being different in the two fosmids. Upstream of its overlapping region with AD1000-325-A12, KM3-153-F8 also contained a cluster of genes implicated in the biosynthesis of purine nucleotides. The more distant clone AD1000-23-H12 carried a set of genes that could be involved in nitrogen fixation. Immediately downstream of the rrn operon were genes for a putative molybdate ABC transporter (NifC-like ABC-type porter) followed by a homologous ORF similar to a dinitrogenase iron–molybdenum cofactor biosynthesis gene in Clostridium phytofermentans ISDg (E-value 10e−64), which is required for nitrogen fixation in this organism (Mehta et al., 2005; Leigh and Dodsworth, 2007). The remaining genes of this clone are mostly housekeeping and gave no clue as to possible metabolic capabilities that could be hypothesized for these archaea.

Organization of genomic fragments from Group II Euryarchaeota

Group II Euryarchaeota are reliably found in 16S rDNA and metagenomic surveys, and were present to a varying extent in our metagenomic libraries from different oceanic locations (Figure 1). To date, only one mesopelagic genomic sequence was available for this group, the 500-m deep planktonic clone DeepAnt-JYKC7 from the Antarctic Polar Front, which included one 16S rRNA gene and a ribosomal protein cluster (spectinomycin or spc operon) (Moreira et al., 2004) (Table 1). In addition, a few other genomic fragments for this group were available at the time of writing this work, all of them from the photic region. They were the 23S rDNA-containing BAC clone 37F11 (Beja et al., 2000b), the fosmid EF100-57A08 (Rich et al., 2007) and three fosmids containing proteorhodopsin-like genes adjacent to the 16S rDNA, of which H70-59C08, contained also the spc operon (Frigaard et al., 2006). We sequenced seven genomic clones from different deep-sea marine libraries, five of which were distributed in two different pan-oceanic OTUs (OTU E including AD1000-18-D2, KM3-130-D10 and SAT1000-15-B12 and OTU F including KM3-72-G3 and DeepAnt-15E7) according to the previously applied criterion of a 16S rRNA identity ⩾98% (Figure 3). From the remaining clones, KM3-85-F5 was quite close to these two OTUs, all the sequences forming with them and other environmental sequences a cluster with ⩾96% identity at the 16S rRNA level. This larger cluster was also relatively close to a group of sequences that contained the previously sequenced DeepAnt-JYKC7 (Figure 3). KM3-136-D10 was, in contrast, very distantly related to that cluster, having a significantly lower GC content, 45.31% versus an average of 58.7% for the remaining Group II euryarchaeotal clones (Table 2). Accordingly, the analysis of the codon usage and amino-acid composition for DeepAnt-15E7 and KM3-130-D10, both having high GC content, showed that triplets possessing G or C at second and third position were favoured for several amino acids, especially for abundant ones such as Gly (GCC), Arg (CGC), Ala (GCC), Val (GTC), Ser (AGC), Leu (CTG/CTG), Glu (GAG) and Lys (AAG) (Supplementary Figure S3). All fosmids from Group II (as well as those from Group III, see below) euryarchaeota had the 16S rRNA and 23S rRNA genes separated in the genome (Figure 8 and Supplementary Table S1), a general feature observed also for the Thermoplasmatales (Ruepp et al., 2000).

Figure 8
figure 8

Comparative genomic organization of fosmids from euryarchaeota. A neighbour-joining tree relating the different 16S rRNA markers found in the respective fosmids is shown on the left as reference. Conserved genomic regions between fosmids are indicated by grey shaded areas, grey intensity being a function of sequence similarity by BLASTN. Open reading frames (ORFs) are numbered from left (ORF 1 in each case) to right as shown in Supplementary Table S1. Particular ORFs mentioned in the text are highlighted by a graphic code.

Full size image

Comparing to the situation within Group I Crenarchaeota, the most striking observation was the high degree of synteny shown by the different euryarchaeotal clones for similar (within OTUs) but also disparate (among distant OTUs) 16S rRNA genetic distances (compare Figures 4 and 8). Part of this synteny was due to the presence of the spc operon encoding ribosomal proteins and other ribosome-related proteins (translation initiation factor SUI1 and protein translocase SecY), though not only. Thus, synteny was particularly well maintained further upstream of the spc operon whenever this region was sequenced, including DeepAnt-JYKC7 (Moreira et al., 2004) and the clone KM3-136-D10, which was very divergent in terms of 16S rRNA sequence (88.7% identical to that of KM3-130-D10) (Figure 3) but also included genes encoding an ATP-dependent RNA helicase, the B subunit of a heterodisulphide reductase, a characteristic protein with three fused domains, Hsp60 and a conserved transmembrane protein (Figure 8 and Supplementary Table S1). As done for the crenarchaeota, and since co-linearity was generally well conserved, we applied current programs for the detection of recombination breakpoints, but we failed to detect any with confidence (data not shown). The genomic stability and apparent absence of intra-genomic recombination events leading to rearrangement observed in the surroundings of the 16S rRNA gene contrasts markedly with the active intra- and inter-genomic recombination and gene shuffling occurring in Group I Crenarchaeota around the rrn operon. This difference is further reinforced by the transcriptional orientation of genes in this area relative to that of the 16S rRNA gene, as approximately 90% of the genes had the same transcriptional sense as the 16S rRNA gene. This trend for genes to be oriented in the same direction was also clearly visible even when we removed the spc operon cluster from our analysis (Figure 7).

Though overall synteny was conserved, the genetic distance at protein markers was noticeable even between closely related OTUs. For instance, sequence divergence at the spc cluster from two clones picked up as examples from the closely related OTUs E (KM3-130-D10) and F (DeepAnt-15E7) was somewhat higher than expected for such highly conserved informational proteins. The average nucleotide and amino-acid identities were only 78.29 and 85.65%, respectively. As a reference, we compared the sequence divergence for spc operon genes in three different Methanosarcina species, M. mazei Go1, M. barkeri strain Fusaro and M. acetivorans C2A (with a 16S rRNA identity ca. 97%). Nucleotide identity values were 84.68% (Go1 versus Fusaro) and 91.5% (Fusaro versus C2A), in parallel with 89.85 and 93.66% amino-acid similarity, respectively. The relatively low similarity values observed for Group II were not due to potential gene exchange by horizontal gene transfer of ribosomal protein genes from foreign donors, since phylogenetic trees for the concatenated sequences as well as all individual trees were congruent with the phylogenetic relationships shown by the 16S rRNA (data not shown). They appear to be simply explained as the consequence of high mutational activity. Even if most ribosomal protein genes had a very small ratio of non-synonymous versus synonymous substitutions (dN/dS) showing strong purifying selection, in some cases we obtained evidence suggesting that positive selection may be occurring. For example, the ribosomal protein S3 in DeepAnt-15E7 (OTU F) and KM3-130-D10/SAT1000-15-B12 (OTU E) gave dN/dS values of 1.357 and 1.355, respectively. Whether this is the consequence of actual positive selection in response to particular selective pressures, for instance antibiotic resistance, or other causes remains speculative at this point. At any rate, our observations would be consistent with mutation playing a more prominent role than intra-genomic, and perhaps also inter-genomic, recombination in the genome evolution of Group II Euryarchaeota. In fact, although a high level of homologous recombination between members of the same or related OTUs could explain the high level of synteny conservation, this would be at odds with the apparent high accumulation of mutations observed that should be neutralized under the effect of intense homologous recombination.

Metabolic clues from Group II Euryarchaeota genome fragments

In agreement with their meso- and bathypelagic provenance, none of our Group II Euryarchaeota genomic clones possessed the proteorhodopsin-like gene that was found immediately downstream of the 16S rRNA gene in some euryarchaeota from the photic region (Frigaard et al., 2006). Apart from the possible phototrophy of those surface euryarchaeota after the horizontal acquisition of a proteorhodopsin-like light-sensitive proton pump, the only hints about possible metabolic capabilities for marine Group II Euryarchaeota were those pointed out by the previous analysis of clone DeepAnt-JYKC7. It harboured a cluster of two protein genes that might be co-transcribed encoding the putative components of an anaerobic respiratory chain (heterodisulphide reductase subunit B and a multidomain protein with a ferredoxin, a flavodoxin and a succinate dehydrogenase/fumarate reductase Fe–S subunit domain) (Moreira et al., 2004). As mentioned above, these genes were also found in our five fosmids overlapping this region (Figure 8 and Supplementary Table S1).

In addition, some genes indicating potential metabolic capabilities were detected downstream of the 16S rRNA genes in the clones that covered that region, that is, DeepAnt-15E7 and KM3-72-G3 (OTU F), as well as KM3-130-D10 and AD1000-18-D2 (OTU E). They contained a glutamine ABC transport system with the periplasmic subunit, two permeases, and an ATP-binding protein. At the further downstream end of fosmid AD1000-18-D2, two genes coding for the carbamoyl-phosphate synthase complex, involved in both the urea cycle and the biosynthesis of arginine and pyrimidines, were identified. They were found also in the surface clones EF100-57A08 (Rich et al., 2007), HF70-39H11 (Frigaard et al., 2006) and the river sediment clone GZfos14B8 belonging to anaerobic methanotrophic archaea (Hallam et al., 2004). Although the region downstream of the 16S rRNA gene was syntenic in our clones, only a few of these genes were conserved in the surface clone H70-59C08 (Rich et al., 2007), an ATP-NAD kinase protein, a membrane protein flotillin-like and a 3-ketoacyl-CoA thiolase. Perhaps the most interesting observation in this region downstream of the 16S rRNA gene came from the identification of a molybdopterin oxidoreductase, a 4Fe–4S ferredoxin (iron-binding protein) and a TraB/PrgY-like protein. Interestingly, the dsm operon described for Halobacterium sp. NRC1 involved in the anaerobic respiration of dimethylsulphoxide (DMSO) (Muller and DasSarma, 2005) has a similar structure. Although not all the genes were the same, equivalent domains were always found. This might indicate that these euryarchaeota use DMSO as an electron acceptor.

In summary, these Group II euryarchaeotal fragments contained not only ribosomal and other housekeeping genes but seemed enriched in putative anaerobic respiration components, which might suggest that they gain energy by anaerobically respiring various substrates.

Genome fragments from Group III Euryarchaeota

Euryarchaeota Group III was first identified from environmental 16S rRNA gene libraries by Fuhrman and Davis (1997) in samples retrieved from the Northeast Pacific at 500 and 300 m depth, and was later detected at 941 m depth in the Alboran Sea (Mediterranean) (Massana et al., 2000) and at 3000 m depth in the Antarctic Polar Front (López-García et al., 2001), suggesting that they were specific inhabitants of the deep ocean. Group III members were also detected in metagenomic libraries throughout the ALOHA water column in the Pacific, although at lower frequencies in surface waters (DeLong et al., 2006). Though they are generally little represented in environmental studies, they appeared more relatively abundant in our SAT1000 library (Figure 1). We sequenced three fosmid clones from three different libraries corresponding to three geographic locations. Two of them, AD1000-40-D7 and SAT1000-53-B3 belonged to a same pan-oceanic OTU (99.1% 16S rDNA identity, OTU D) (Figure 3), had similar low GC content (Table 2) and were largely syntenic (Figure 8 and Supplementary Table S1). Despite their relatedness at the 16S rDNA locus, the similarity of the overlapping ORFs at amino-acid level was only 88.88%. The more distant clone KM3-28-E8 (92.7% identity to AD1000-40-D7) had higher GC content (54.8%) and had only one gene in common with the two other fosmids, the first truncated ORF, coding for a hypothetical protein bearing only 31% similarity to its homologue in the other clones. The information about genome organization and dynamics in this group of archaea is very limited, and very little can be said from comparing only these three fosmids. Nevertheless, since only approximately 50% of the ORFs had the same transcriptional orientation as the 16S rRNA gene (Figure 7), a higher rearrangement frequency could be hypothesized for these archaea than that observed for the Group II in the same genomic area.

As in the case of Group II archaea, a significant number of genes in SAT1000-53-B3 and AD1000-40-D7 were housekeeping. Many of them coded for components of the translation apparatus, such as a valine tRNA, an alanyl-tRNA synthetase, an aspartyl-tRNA synthetase, a putative metal-dependent hydrolase related to alanyl-tRNA synthetase, the signal recognition particle protein Srp54 and the translation termination factor aRF1 (Supplementary Table S1). Others were related to DNA repair (for example, an endonuclease IV homologue) or transcription (a putative σ54-related protein). A peptidyl-prolyl cis-trans isomerase (ppiD), not previously described in archaea, was also identified. PPIase is an enzyme that accelerates protein folding by catalysing the cis-trans isomerization of proline imidic peptide bonds in oligopeptides (Dartigalongue and Raina, 1998). Two genes (ORF 11 and 17 in SAT1000-53-B3, and ORF 18 and 24 in AD1000-40-D7) could be involved in the recently described and unusual wyeosine post-transcriptional modification of tRNA. The wyeosine family of tricyclic ribonucleosides is one of the more complex and structurally interesting post-transcriptional modifications in tRNAs, and has so far only been found in the archaeal and eukaryal domains (Zhou et al., 2004). The first ORF was homologous to a TYW3 domain protein, which is usually involved in wybutosine (yW) biosynthesis in yeast, and the second ORF was homologous to a SAM-dependent methyltransferase related to tRNA (uracil-5-)-methyltransferase modifications (de Crecy-Lagard, 2007).

Regarding potential metabolic clues, one of the most significant proteins found in AD1000-40-D7 and SAT1000-53-B3 was a homologue of an ammonium permease, AmtB, with an associated histidine kinase domain. Proteins with similar structure have been identified so far only in the bacterial species Parvibaculum lavamentivorans DS-1 (65% similarity), Planctomyces maris DSM 8797 (62%) and in the anaerobic ammonium oxidizer Candidatus Kuenenia stuttgartiensis (60%). Adjacent to this gene in AD1000-40-D7, there was a duplication of the histidine kinase domain present in this putative ammonium transporter. Next to it was a gene homologous to CheY, a chemotaxis response regulator. This two-component system combination is evocative of an ammonium sensor and transporter resembling those found in ammonia-oxidizing bacteria (Arp et al., 2007). Fosmid SAT1000-53-B3 also contained genes for the carbamoyl-phosphate synthase complex, as in the Euryarchaeota Group II (see above). The more distant clone KM3-28-E8 harboured a gene potentially involved in anaerobic respiration (Acyl-CoA dehydrogenase short-chain-specific gene), as well as three putative genes that might be involved in C fixation by the 3-hydroxipropionate pathway or one variant of it. These were 3-hydroxyacyl-CoA dehydrogenase (fadB), malonyl-CoA synthetase (fadD) and enoyl-CoA hydratase (fadB) (Supplementary Table S1). Whether these enzymes are specifically involved in autotrophic C fixation or in other pathways of central metabolism, for example, fatty acid synthesis, remains uncertain.

Discussion

Marine pelagic archaea are among the most widespread and abundant marine microbes, having been found in many marine environmental 16S rRNA gene libraries since they were first discovered (DeLong, 1992; Fuhrman et al., 1992). Despite their obvious interest, relatively little is known about their ecology and evolution. Fortunately, the accumulation of molecular, genomic and metagenomic data combined with environmental biochemistry and the first cultivation success has led to a fundamental breakthrough concerning the ecological role of the most easily encountered marine archaea, the Group I Crenarchaeota, which appear to be aerobic nitrifiers (Konneke et al., 2005; Hallam et al., 2006a, 2006b; Wuchter et al., 2006; Lam et al., 2007). However, this type of information is missing for most of the other planktonic archaea that also appear to be widespread in oceans, and nothing is known in general about archaeal population dynamics and evolution in various oceanic regions. To gain additional information about marine pelagic archaea, and since the vast majority of them remain uncultivable, we detected, sequenced and comparatively analysed genomic clones that possessed archaeal 16S rRNA genes from four meso- and bathypelagic planktonic metagenomic libraries representing different oceanic regions and water masses (Table 1). We selected and analysed both clones belonging to the divergent and less abundant Group 1A Crenarchaeota and Group III Euryarchaeota, for which nearly nothing is known, and a number of very closely related clones forming defined OTUs within the Group I Crenarchaeota and Group II Euryarchaeota. Our study targeted three objectives. First, metagenomic library screening would generate information about the relative distribution of different archaeal groups in various deep-sea regions. In addition, sequencing genomic clones carrying a reference phylogenetic marker would eventually lead to (i) the identification of metabolic genes and hypothesizing functions for given archaeal lineages and (ii) the characterization of genome structure in comparable genomic surroundings, which might elucidate particular lineage-specific genome dynamics and lead to hypotheses about evolutionary trends.

It is commonly accepted that Group I Crenarchaeota are more abundant in oligotrophic, deep waters, whereas Group II Euryarchaeota are either more abundant in surface waters or equally rare throughout the water column (Karner et al., 2001; Herndl et al., 2005), something that seemed to be confirmed by some metagenomic studies (DeLong et al., 2006; Martin-Cuadrado et al., 2007). However, extensive comparative deep-sea studies to fully appreciate the distribution and potential phylogeography of marine pelagic archaea are lacking and, therefore, drawing general conclusions from few locations and collection dates may be unreliable. In this sense, our screening of four metagenomic libraries from the Antarctic Polar Front, the South Atlantic, and the Mediterranean Ionian and Adriatic seas, containing from ca. 5500 to 39 000 clones, revealed important different relative proportions of Groups I and 1A Crenarchaeota and Groups II and III Euryarchaeota in the different libraries (Figure 1). Since all potential cloning biases were similar, and since at least Groups I and II archaea carry likely one single 16S rRNA gene (which can be deduced from their comparative genome organization), these differences could reflect actual relative frequencies in the environment. Many more studies on spatial and temporal relative abundances of marine pelagic archaea will be needed to understand their population dynamics. Despite differences in relative abundances, we did not observe any clear phylogeographical pattern. In contrast, for the more abundant archaeal Groups I (Crenarchaeota) and II (Euryarchaeota), we identified clones belonging to the same OTUs (that is, having ⩾98% identical 16S rRNA genes) from geographically very distant metagenomic libraries (Mediterranean and South Atlantic/Antarctic Polar Front). Several of these OTUs also include environmental sequences from Pacific 16S rRNA gene and/or metagenomic libraries (Figures 2 and 3). This suggests that at least some of what might actually be archaeal species have a true pan-oceanic distribution, though their relative abundances may vary spatially and temporally. Stratification in the water column possibly determines the strongest spatial pattern. Indeed, most of our clones clustered with deep-sea sequences, and we were unable to detect putatively phototrophic euryarchaeota (Frigaard et al., 2006) in our metagenomic libraries.

The protein genes identified in the 22 archaeal genomic fragments sequenced corresponded in large part to housekeeping machinery (translation, transcription, DNA repair, etc.) and to some conserved hypothetical proteins. However, a number of genes pointed out to some clues indicating potential metabolic capabilities in different archaeal lineages. We did not detect any amoA gene in Group I Crenarchaeota that might confirm the suspicion that they are ammonia oxidizers, as is the case of N. maritimus (Konneke et al., 2005), C. symbiosum (Hallam et al., 2006a) and some Group I soil fosmids (Treusch et al., 2005). Yet, we detected various genes, having homologues in C. symbiosum and N. maritimus, involved in 3-hydroxyproprionate-based pathways of CO2 fixation, suggesting that these organisms can also fix carbon autotrophically. Interestingly, genes encoding oligopeptide transporters were found in clones from two Group I archaeal OTUs, suggesting that at least some Group I archaeal species could be mixotrophic, as is the case of C. symbiosum (Hallam et al., 2006a, 2006b). Mixotrophy is being increasingly detected in marine microbes. It includes not only phototrophic bacteria (Beja et al., 2000a) and euryarchaeota (Frigaard et al., 2006) but also lithoheterotrophic microbes able to gain extra energy from CO oxidation in surface (Moran and Miller, 2007) and deep-sea water (Martin-Cuadrado et al., 2007). In a recent work, amplified and sequenced amoA genes at different depths in the North Pacific Subtropical Gyre showed a strong phylogenetic partitioning with depth. This was correlated with quantitative distributions of Group 1A sequences and was interpreted as an indication of Group 1A being putative ammonia oxidizers as well (Mincer et al., 2007). We did not identify any amoA gene in the three clones that we sequenced from this archaeal group. Nonetheless, we found genes involved in nitrogen fixation in one of them. Genes for nitrogen fixation have also been found in some hyperthermophilic methanogens (Mehta and Baross, 2006) and other deep sea so far uncultivated archaea (Mehta et al., 2005). At any rate, this observation brings to light the possibility that marine pelagic crenarchaeota are not only key but also multifaceted players in the nitrogen cycle.

Furthermore, members of the Group III Euryarchaeota might also utilize the oxidation of ammonia as a means to gain energy, as we detected genes for a histidine kinase two-component system that resembled that found in ammonia oxidizing bacteria, including marine planktonic planctomycetes and the anammox planctomycete K. stuttgartiensis. Whether Group III Euryarchaeota can actually oxidize ammonia and whether they do it aerobically or anaerobically should deserve future investigations. Finally, regarding the potential energy metabolism of marine Group II Euryarchaeota, we systematically observed the presence of genes possibly involved in an anaerobic respiration chain that had already been described in the Antarctic Polar Front clone DeepAnt-JYKC7 (Moreira et al., 2004). In addition to this, the presence of a putative DMSO respiration cluster suggested that these organisms could use DMSO as electron acceptor. Although the significance of DMSO respiration in the global turnover of marine organo-sulphur compounds derived from algal remains needs to be established, the fact that several marine bacteria including members of the genera Shewanella and Rhodobacter carry a dms operon (McCrindle et al., 2005), as well as halophilic and perhaps Group II Euryarchaeota, might point to a significant role of this form of anaerobic respiration for the marine sulphur cycle.

Metagenomic analyses are beginning to make possible the study of population genomics in the environment. One of the most interesting observations of our work comes from the realization that differential genomic dynamics appear to be operating in different archaeal phylogenetic groups. This was particularly evident between Group I Crenarchaeota and Group II Euryarchaeota, for which we had the largest sampling. Thus, whereas gene shuffling by intra-genomic recombination was very important around the rrn operon within and among OTUs in Group I Crenarchaeota (Figure 4), co-linearity was strikingly maintained even across relatively distant OTUs within Group II Euryarchaeota (Figure 8). This opposite trend for equivalent genomic regions around the likely single-copy 16S rRNA gene in the capacity to rearrange was further supported by the relative orientation of genes in these genome fragments. Whereas approximately 90% of the genes in Group II genome fragments had the same transcriptional orientation as the 16S rRNA gene, only half of the genes in Group I clones did (Figure 7). Likewise, inter-genomic recombination leading to both genetic exchange within populations of the same or related OTUs (via homologous recombination, for example, at the rrn operon and the EF-2 gene; Supplementary Figures S2A and Figure 6) and with far more distant organisms resulting in horizontal gene transfer (López-García et al., 2004), appeared important in Group I Crenarchaeota. By contrast, we failed to detect with confidence potential recombination breakpoints in Group II Euryarchaeota fosmids. This and the fact that sequence divergence by mutation is relatively high in these organisms tend to support lower inter-genomic homologous recombination rates in Group II Euryarchaeota, which would otherwise counterbalance mutational activity, as compared to Group I Crenarchaeota. This indicates that intra- and inter-genomic recombination is an active process shaping genome evolution in Group I members. By contrast, sequence divergence in protein genes relative to 16S rRNA genetic distances appears to be much higher in Group II Euryarchaeota than in Group I Crenarchaeota. Moreover, evidence of positive selection was observed in one gene coding for the ribosomal protein S3. Therefore, mutation fixation might be the predominant mechanism in the evolution of Group II euryarchaeotal genomes. Of course, these observations are restricted so far to a small genomic region around the apparently single 16S rRNA gene in the genome, which we are using here as a proxy to hypothesize similar trends at the whole genome scale. This hypothesis may be supported by the general observation that the rrn operon surroundings are normally regions harbouring important housekeeping genes that are highly expressed and, therefore, subjected to higher selection than other genomic regions such as the terminus of replication. Nonetheless, falsifying this hypothesis will require full genome comparison of these archaea in the future.

Interestingly, excluding the truncated genomic clone KM3-136-D10, all Group II Euryarchaeota genome fragments had high GC content (average 58.7%), whereas all Group I Crenarchaeota had low GC content (35.9%) (Table 2). Differences in GC content in different phylogenetic clades might be simply the consequence of phylogenetic history of particular lineages. Yet they might be as well indicators of particular lifestyle strategies and genomic constraints. For instance, intracellular parasites tend to possess AT-rich genomes, since there is a mutational bias towards AT derived from the relaxation of selective pressures acting on some gene categories, such as DNA repair genes (Rocha and Danchin, 2002; Moran, 2003), while organisms living in environments where competition is strong and selective pressures high would tend to maintain higher GC genomes. According to this view, there could be two alternative hypotheses to explain the differences in GC content observed. First, it could be speculated that marine Group I Crenarchaeota are actually free-living planktoners (C. symbiosum, whose genome is GC rich, is a sponge symbiont and would have to compete with the sponge microbiota), whereas Group II Euryarchaeota might rather live associated with sinking particles and be therefore subjected to strong competition and selection. Though we pre-filtered samples by 5 μm-pore size filters, it might be possible that some particle-associated microbiota was recovered. Indeed, the GC values of the total sequence determined for different size ranges in the Sargasso shotgun metagenomic database decreases along with the pore size of the filter used, although there might be some potential bias introduced by the capture of eukaryotic cells using larger pore sizes (Supplementary Figure S4). This hypothesis could be eventually supported by the presence in the Group II Euryarchaeota of genes involved in anaerobic respiration (more likely to occur within particulate matter) or resembling those of planctomycetes, which are thought to be physically associated with particles (Woebken et al., 2007).

However, this first hypothesis would be at odds with fluorescent in situ hybridization studies showing that both marine pelagic crenarchaeota and euryarchaeota are free-living and not associated with particles (Woebken et al., 2007). If this is indeed the case and both archaeal groups were actually planktonic and free-living, a second, alternative hypothesis to explain the observed differences in genomic GC content could be that Group I Crenarchaeota had shorter generation times, behaving more like r-strategists or ‘bloomers’, as compared to Group II Euryarchaeota, which would have longer generation times relative to Group I Crenarchaeota, sustaining growth when resources are low and behaving as K-strategists or ‘survivalists’. These different lifestyle strategies are indeed related to the N/P ratios in the cell in marine pelagic ecosystems (Klausmeier et al., 2004; Arrigo, 2005). Having low GC genomes is a way to save N, a limiting element in oligotrophic oceanic waters (AT having seven N atoms, one N atom less than GC pairs, making a significant difference at the whole genome scale). Therefore, pelagic microbes that grow faster and reach larger population sizes would tend to accumulate low GC genomes not as a consequence of selective pressure relaxation (as occurs in parasites) but, in contrast, because there would be a strong selective pressure for r-strategists to save N under environmental conditions where this resource is limiting. In fact, this trend has been observed in free-living marine Prochlorococcus species genomes. These organisms undergo a strong purifying selection, so that low GC genome content has been interpreted as a strategy to save energy and nutrients (Dufresne et al., 2005). This tendency to have lower GC genomes as a consequence of strong purifying selection under nutrient limitation would be particularly favoured in Group I Crenarchaeota as, being ammonia oxidizers, they would tend to deviate most of the available reduced nitrogen (ammonia) to dissimilatory energy metabolism rather than to assimilatory pathways in a globally N-limiting oligotrophic ocean.

In conclusion, our comparative genomic analyses reveal remarkably different genomic dynamics in Groups I and II marine archaea that could be linked to opposite life strategies, both being subject to strong selection, in these two groups of organisms. In Group I Crenarchaeota, the possession of low GC genomes in potentially faster growing microbes (as compared to Group II Euryarchaeota) might be correlated with higher recombination levels. In fact, recombination appears to be higher in AT-rich regions in chromosomes (terminus of replication, insertion elements). In planktonic Group II Euryarchaeota, strong selection pressures endured by survivalist strategists would favour GC-rich genomes and low recombination frequencies. These hypotheses remain open to be tested in the future.