The complete sequence was achieved using the previously determined sequence of HBSQ001 as a scaffold to arrange 220 separate contigs derived from 454 sequence reads. Contig gaps were then spanned by PCRs to close the genome. The general features of the circular chromosome, and three plasmids carried by this strain (EMBL accessions FR746099- FR746102), are presented in Table 1. For comparison, the corresponding data of the Spanish strain HBSQ001 are also included. The chromosomes of both are similar in size (∼3.1 Mb) and %G+C content (47.8%). Both have two rRNA operons and 45 tRNA genes, and carry a similar number of predicted ORFs (C23^T: 2,894 and HBSQ001: 2,819). Curiously, the gene density is relatively low (79%) and there are over 300 pseudogenes (C23^T, 337; HBSQ001, 314), comprised largely of degraded transposases and conserved hypothetical proteins. The strains differ in their plasmids, with C23^T carrying one large plasmid (PL100, 100 kb) and two small plasmids of around 6 kb (PL6A, PL6B) while HBSQ001 contains a single, unrelated 47 kb plasmid (PL47). Analyses of tetra-nucleotide frequencies show that the sequences CTAG, GGCC and AGCT are strongly avoided on the main chromosome. GGCC is also avoided on the large plasmid and is absent in the small plasmids (Table S1).

Figure 1 presents the results of several global analyses of the main chromosome of C23^T. Major deviations from the average %G+C content (topmost plot) correlate closely with changes in tetramer frequency, as shown by the intense vertical bands in the TETRA plot below (third level). The second level graph shows the distribution of pseudogenes derived from non-transposase ORFs, and these are clearly associated with many of the variant regions identified in the adjacent TETRA and %G+C plots. Bacterial genomes often show large-scale organizational patterns, such as a systematic bias in the nucleotide composition of their leading and lagging strands, preferential placement of ORFs on the leading strand, and highly expressed genes close to the replication origin [22], [23]. In such cases, a plot of cumulative GC-skew versus genome position can show a simple, geometric pattern where the replication origin and the terminus occur near major inflections [24], [25]. In general, statistical deviations are much weaker in archaea so that cumulative GC-skew plots do not give a simple pattern (fifth level of Figure 1), nor does the GC-profile graph, a type of cumulative GC-skew that is more sensitive to local changes in %G+C content [26], shown just above it. However, in comparing the plots from the two strains (see later) one can distinguish between strong local deviations due to insertion of foreign DNA and weak positional deviations related to replication origins. Strong changes in the GC-profile correlate well with significant alterations in %G+C and tetramer composition. These atypical genome regions represent a mixture of unusual genomic features, described in detail below. Two prominent features are, (1) near the left end of the %G+C panel a distinct peak of higher %G+C, labeled hmuI (corresponding to a very long ORF encoding the halomucin gene with a highly biased codon usage) and, (2) a peak of lower %G+C at around 1.6 Mb that corresponds to a prophage, integrated into a tRNA gene.

Although the GC-skew plots were not able to directly identify likely replication origins, archaeal origin-of-replication sequences (ori) typically occur adjacent to Cdc6/Orc genes [27], [28], [29]. Haloarchaea can carry up to 17 copies of such genes [30] but the number of origins is usually much lower [29]. Strain C23^T possesses nine chromosomal Orc/Cdc6 genes, five of which appear to be complete (solid black arrowheads, CDC6 plot in Figure 1), and four are pseudogenes (smaller, unfilled arrowheads). In addition, two other Orc/Cdc6 genes are carried on the large plasmid, PL100 (see below).

The protein sequence of the first chromosomal Cdc6 ORF, Hqrw_1001, is most closely related to HVO_0001 of Hfx. volcanii and VNG2411G/OE4380F of Hbt. salinarum, both of which lie next to the main replication origin, oriC1 [28]. The ORFs surrounding Hqrw_1001 are also very similar to those found near oriC1 in other haloarchaea [21], [28], [29]. Upstream of Hqrw_1001 is the probable oriC1 of Hqr. walsbyi C23^T, characterized by sequence elements typical of replication origins of haloarchaea (Figure S1), including long inverted repeats surrounding an AT-rich sequence (potential DNA unwinding domain, DUE) [27], [28], [31]. Apart from Hqrw_1001, only two other complete homologues have sufficient, non-coding upstream sequences to contain potential further origins, Hqrw_3381 and Hqrw_3385. These lie very close to each other, are inward facing, and have ori-like sequence motifs nearby (Figure S1). These may represent one or two additional origins, and have been labeled oriC2 and oriC3 in Figure 1. As shown in the fifth and sixth panels of Figure 1, the predicted replication origins are near high points in the cumulative GC-skew graph, while the minimum (∼1.2 Mb) occurs almost mid-way between oriC1 and oriC2/oriC3, and is likely to include a site or region where replication terminates. Multiple replication origins are common in Archaea [28], [32].

All 45 tRNA genes are located on the main chromosome. Rather than an even distribution, they show some tendency to cluster near the predicted replication origins, and to avoid the region furthest away from the origins, where replication is likely to terminate (1.1–1.6 Mb). In contrast to the relatively relaxed organization of tRNA genes in Haloquadratum, thermophilic Archaea such as Sulfolobus display a strong clustering of tRNA genes [33]. There are two rRNA operons in Haloquadratum, the first (at ∼0.07 Mb) faces away and is very close to oriC1. The second (at ∼1.8 Mb) is almost diametrically opposite the first on the circular chromosome but is some distance from the potential oriC2/oriC3 at ∼2.4 Mb, again facing away. The two 16S rRNA genes differ by 3 bases while the corresponding 5S and 23S rRNA genes are identical. The next two lower panels in Figure 1 show the positions of protein-coding genes that are likely to be highly expressed; those for ribosomal proteins and RNA polymerase subunits. Their distribution is similar to the tRNA genes. The three CRISPR loci (crispr-1, -2 and -3) are indicated in the lowest panel of Figure 1. They occur within extended genomic regions of distinct tetra-nucleotide frequency, indicating foreign DNA. Overall, the genome appears to show some degree of global modeling, oriented around the predicted replication origins, but disturbances occur in regions likely to represent insertions of foreign DNA.

Haloquadratum is unusual in having a genome with a %G+C that is about 20 percentage points lower than all other genera within the family Halobacteriaceae. The most parsimonious explanation is that Haloquadratum has evolved from the higher level down to its current value. As expected from such a change, Hqr. walsbyi shows a strong bias towards codons with A or T in the 3^rd position, relative to the codon usage of other haloarchaea (ca. 60% A+T vs ca. 20%), while the bias is only slight for the 1^st and 2^nd position [7]. The strength of this bias indicates a rapid evolutionary drift towards a lower GC content. Due to this strong bias, ten different amino acids in this organism have NNT as the preferred codon in their synonymous codon sets (Table S2) [7]. However, as noted in the literature of codon-anticodon interactions, tRNA anticodons with A in the first position (which decode NNT codons), are generally uncommon, and in Haloquadratum there are no such tRNAs present in the genome. This means that despite a preference for NNT codons, all of these must be decoded by tRNA anticodons using G:U base-pairing, highlighting the view that “the anticodon-codon wobble base pair of a G₃₄ with a U3 … is almost isomorphic/isosteric to a Watson-Crick base pair.” [34]. tRNA base modifications also determine specificity, and have been examined best in Haloferax volcanii [35], but not yet in Haloquadratum.

Two tRNAs contain introns (tRNA-Trp, tRNA-Met2), while tRNA-Ile1 has a CAT anticodon that is modified [36], enabling it to switch its specificity and act as a TAT anticodon, and decode Ile (ATA) codons. The genome also contains two partial tRNA genes that are associated with prophages (see below).

Plasmid extracts of Hqr. walsbyi C23^T cells revealed a small, 6 kb multi-copy plasmid (∼30 copies/genome equivalent) designated PL6, and a much larger, low copy-number plasmid, PL100 (Figure 2A and B). Panel A shows a typical gel profile of un-restricted plasmid DNA, with the smaller plasmid band displaying super-coiled, open-circular and dimeric forms. Loading more plasmid extract (panel B) revealed the large plasmid, PL100, running above the fragmented chromosomal DNA band. Plasmid preparations from the previously sequenced Spanish isolate, HBSQ001, and from of a novel, Australian isolate of Haloquadratum, strain Bajool9 [16], were also included in Panel A. The Spanish strain lacks small plasmids and its large 47 kb plasmid (PL47) is not visible in this gel. The Bajool9 isolate was recovered from a saltern crystallizer in Bajool, Queensland, Australia [16], some 1700 km north of the crystallizer from which C23^T was isolated. It also harbours a ∼6 kb plasmid, as well as another, much smaller plasmid (faintly visible at the bottom of the lane, running between the 1 and 2 kb size markers).

Genomic sequencing revealed that the PL6 plasmid band observed in C23^T consists of two closely related plasmids, designated PL6A and PL6B. These were in approximately equal proportions in the population, as judged by restriction digests and sequenceing reads (data not shown). Their %G+C is significantly higher than that of the chromosome (51–52% versus 47.8%). They show high sequence identity, particularly over the first two ORFs, but this drops sharply at around 3.2 kb, (arrowed in Figure 3), following immediately after a motif consisting of two direct repeats of the sequence ACAGATTA bordered by an inverted repeat. Both plasmids have six predicted ORFs that show a similar organizational pattern. The corresponding ORFs between plasmids are predicted to code for proteins of similar size and amino acid sequence (Figure 3).

The six annotated ORFs of PL6A or PL6B show few protein motifs that could give clues to function. None show significant homology to any of the known Rep, Cdc6, MCM helicases or DNA polymerases, nor are there any detectable oriC1-like origin-of-replication-box (ORB) motifs, so the actual mode of plasmid replication remains uncertain. However, archaeal plasmids are known to display considerable variation in their replication systems [37], [38]. Helix-turn-helix motifs, indicative of DNA binding, are predicted for Hqrw_6001, 6002 and 6005 (and the corresponding PL6B ORFs), and the latter protein also contains a P-loop ATPase-like motif (see also below). Intriguingly, Hqrw_6007 and Hqrw_7007 were detected by mass spectrometry in purified cell membrane preparations of C23^T (Table S3), a cellular location consistent with their predicted C-terminal membrane anchors. The N-termini of these proteins are similar to members of the CopG/MetJ family of repressors, indicating probable binding to target sites on PL6. If so, the bifunctional properties of these proteins suggest they could tether the PL6 plasmids to the cell membrane. Repeats and inverted repeat sequences occur within the two long intergenic regions found between Hqrw_6007 and Hqrw_6001, and between Hqrw_6002 and Hqrw_6004. A long, inverted repeat occurs in both plasmids at a similar position (nt 107–123, PL6A), and is flanked by ORFs with predicted DNA binding motifs. Cumulative nucleotide skew plots that include thymidine show a strong inflection in the intergenic sequence at about 3.1 kb (a highly conserved region), with a maximum in the intergenic sequence between the sixth and first ORFs (Figure 3). The arrangement of these conserved repeats at almost diametrically opposite sites (0.1 and ∼3.2 kb), and their correlation with inflections in nucleotide skew, suggest they may represent replication origins and termini (assuming a theta mode of replication).

The nucleotide sequences of the PL6 plasmids were not similar to the main chromosome, or to PL100 of C23^T, or to the genome of HBSQ001, or other reported haloarchaeal plasmids. Partial sequencing of the 6 kb plasmid in the Bajool9 strain revealed it to be similar in sequence and gene organization to PL6A or PL6B (85% nt identity, data not shown). BLASTN searches of the CAMERA metagenomic database gave numerous matches to DNA sequences recovered from a saltern crystallizer pond in the USA (Table 2). Hundreds of matches were found in the larger dataset of environmental sequences recovered from Lake Tyrrell, a hypersaline lake in Australia (Table 2). Sequence reads from this dataset could be retrieved that showed high nucleotide identity to PL6, and these could be assembled into contigs of up to 6 kb that displayed co-linear sequence similarity to these plasmids (data not shown). While no matches to metagenomic data from the Santa Pola saltern in Spain were observed, recently available sequences from a viral metagenome of the same saltern showed several TBLASTX matches to PL6 ORFs Hqrw_6005 and Hqrw_7005 (see later). The presence of Hqr. walsbyi 16S rRNA gene sequences in the metagenomic data from the three sites confirmed the presence of this species in all three datasets. The data indicate that PL6-like plasmids are widely distributed.

Comparing PL6 to the genome sequences of other microorganisms, the second ORF (Hqrw_6002/7002) gave significant matches to numerous archaeal homologues. In their genomic contexts, these homologues are frequently adjacent to a previously described group of well-conserved halovirus genes, initially identified from the sequence of halovirus His2 [39]. A block of four genes including that for the major virus capsid protein of His2 is found in many haloarchaeal genomes as well as pHK2, a small haloarchaeal plasmid. Although these genomic homologues do not appear to be His2-like prophages [39] their distribution and conservation somehow relates haloviruses and haloarchaeal plasmid genes at specific sites of genomic integration [39], [40], [41]. Two recently described pleomorphic haloviruses, HRPV-1 and HHPV-1, also contain a similar block of homologues related to His2, plasmid pHK2 and the same genomic loci [42], [43], further consolidating this link. A summary of the relationships between PL6, the three haloviruses, and the related loci from six different genera of haloarchaea is presented in Figure 4. In most cases, homologues of PL6 ORF Hqrw_6002 are found either immediately adjacent to the His2-related gene cluster or separated by a small number of intervening genes. In the case of Hfx. volcanii, the intervening genes are three consecutive insertion sequences. Archaeoglobus, which is not a member of the family Halobacteriaceae but belongs to the same archaeal phylum (Euryarchaeota), also contains a His2 VP1 homologue, which is near a virus integrase. The proximity, consistent orientation, and wide phylogenetic distribution of this gene arrangement speaks of an important and complex network of DNA cross-talk; one that spans many of the known genera of haloarchaea and involves a diverse group of plasmids and haloviruses. In some respects, these appear similar to Mavericks, a class of mobile elements found in Eukaryotes [44].

The PL6 homologues Hqrw_6005 and Hqrw_7005 were specifically related to halovirus His1 protein His1V_gp16, and to an ORF of Nmn. pharaonis (NP3284A). The latter ORF occurs within a 13 kb region that also includes a CRISPR system, and which has been inserted into a chromosome-integrated copy of the Nmn. pharaonis plasmid PL23. All these homologues possess a conserved P-loop ATPase motif. Halovirus His1 is the type species of the Salterprovirus group, to which halovirus His2 also belongs [39], so the relationship observed in this study between PL6, His1, and genomic loci carrying His2-related genes provides further support for a virus link between these elements. Indeed, there is a strong possibility that PL6 is a provirus.

We propose that this family of archaeal genomic loci that include halovirus, and now PL6 homologues, be referred to as virus and plasmid related elements (ViPREs). Understanding the mechanisms, significance and impacts of ViPREs on the evolution of haloarchaea (and perhaps other Euryarchaeota) and their viruses is clearly a high priority.

The large plasmid of C23^T is 100,258 bp in length, 4 percentage points lower in G+C than the main chromosome (43.9%) and predicted to code for 83 ORFs. It is more than twice the size of the 47 kb plasmid in HBSQ001 (PL47), and although the two plasmids appear to be largely unrelated they both carry a preponderance of hypothetical or conserved hypothetical proteins without discernable function. ORFs with assigned function on PL100 include those involved in plasmid maintenance or replication, restriction endonucleases, methylases and helicases. Two cdc6 gene homologues, Hqrw_5030 and Hqrw_5083, are situated approximately opposite each other on the plasmid, and their upstream regions possess ORB-like repeat motifs consistent with archaeal replication origins (Figure S1, OriP1 and OriP2). Nucleotide skew analyses indicated only small inflections near both, but as there are no RepH homologues, replication is most likely initiated near these cdc6 genes.

Plasmid PL100 was unstable in laboratory cultures of C23^T, and was lost from the culture used to prepare DNA for genome sequencing. Some sequences of PL100 had been recovered from an earlier, preliminary genome sequencing effort, and this allowed PCR primers to be designed to detect the presence of the plasmid. Screening cultures for successful PCR amplification identified one (out of 6) that had retained the plasmid, and DNA from this source was used to PCR amplify across contig gaps and complete the sequence (see Materials and Methods). From these observations, PL100 is readily dispensable, at least under laboratory conditions.

The close relationship of C23^T and HBSQ001 is clearly seen in a nucleotide alignment of the two chromosomes (Figure 5), where only minor shifts from a straight line due to indels occur; up-shifts indicating insertions in HBSQ001 and downshifts for insertions in C23^T. There are no large-scale rearrangements – the two nucleotide sequences are co-linear, sharing 98.6% nucleotide similarity. Panel C is a different representation of the aligned chromosomes using the Artemis Comparison Tool (ACT) that shows pattern of variation between the chromosomes. Together with panel A, the two graphs present a typical pattern of intra-species variation seen in many prokaryotes: a backbone core sequence with high similarity that is interspersed with strain-specific indels.

Twelve prominent regions of strain-specific variation (divergent regions DV1 - DV12) are indicated above the ACT alignment. Below the alignment, the positions of genomic islands (GI 1-4) of HBSQ001, as described previously by Cuadros-Orellana et al. [13], are shown. Genomic islands are regions of high variability that were identified when the HBSQ001 genome was compared to autochthonous, metagenomic DNA sequences. The DV and GI sequences are discussed in more detail below, but it is clear that the GIs also correspond to regions of major difference between HBSQ001 and C23^T seen in the current study, i.e. GI I  =  DV2, GI 2  =  DV6, GI 3  =  DV11, GI 4  =  DV12. Panels B and D plot the variation in tetramer frequency of the two strains, and show that some regions of divergent sequence composition are common to both (such as the halomucin genes at around 0.6 Mb, DV1) while others are strain-specific, often coinciding with large indels and indicative of foreign DNA insertions. Panel E compares the cumulative GC-skew plots of both chromosomes, which show major differences at positions corresponding to large indels. The dramatic drop in the HBSQ001 plot at around 2.6 Mb is due to a strain-specific integration of a prophage (see below). The prophage provides HBSQ001 with two additional cdc6 genes compared to C23^T, but these are closely spaced and outward facing, and are not associated with nearby ori-like sequence motifs. The replication origins used by HBSQ001 are probably the same as the corresponding sequences in C23^T, as these are conserved, but the cdc6 gene (HQ2952A) corresponding to that near oriC2 of C23^T has been inactivated by an insertion. The adjacent ori could still function, as replication origins in haloarchaea can use Cdc6 proteins synthesized from genes other than the adjacent copy [29].

The close similarity of the two strains enabled a systematic and precise analysis of all the differences between their chromosomes. A custom PERL-based script (as outlined in Materials and Methods) was applied that could perform a base-by-base comparison of very similar sequences while being tolerant to insertions and deletions of any size as well as large sequence duplications. Trivial differences between the strains were excluded (e.g. point mutations), based on criteria defined in Materials and Methods. Although the constraints were relatively relaxed, the resulting set contained only 512 non-trivial differences, emphasizing the very close relationship of the two strains. These differences were manually inspected to eliminate simple cases of regions with enhanced sequence divergence, and this left a total of 360 differences, representing the final set of strain-specific sequences (Tables S4 and S5).

The remainder of the two chromosomes defines the common shared sequences of the two strains, comprising about 84% of each genome. These sequences are 98.6% identical at the DNA level and are absolutely synteneic. They do not show even a single genome rearrangement (inversion or transposition). Sequence identity remains high even in intergenic regions (which represent a significant proportion of the genome in Haloquadratum) and in genes that are not well conserved among haloarchaea. Therefore, we prefer the term “shared sequence” instead of “core sequence” as the latter implies restriction of the analysis to only the well-conserved sequences of both genomes.

In other studies comparing strains of the same prokaryotic species, core sequence similarities of 98–99% are commonly observed but with intergenic regions excluded. Also, these genomes usually show inversions and rearrangements. For example, in E.coli, a recent study of 61 genomes found that they can vary in size by more than 1 Mb (4.6–5.7 Mb), that gene order is not strongly conserved (i.e. extensive inversions/rearrangements), and that any particular E. coli genome contains only about 1/5th of the core genes of the species, while the remaining 4/5ths sample a pan-genome estimated to contain about 15,000 gene families [45]. In the current study, the complete lack of genome rearrangements together with a 98.6% sequence identity, even when intergenic regions are included (which amount to more than 20% of the sequence), indicate a remarkable level of similarity between the two strains of Haloquadratum walsbyi, particularly given that they originate from samples taken at near-maximal distance on Earth.

Strain-specific sequences contribute 16% of the genetic material in each strain, and include numerous integration and excision events. It is an apparent paradox that so many of these events have occurred while the shared sequences do not even show a single large-scale genome rearrangement. The availability of such highly conserved genome sequences, with synteneic shared sequences, provided an outstanding opportunity to examine, in fine detail, the differences between these strains. This revealed two unsuspected biological processes that have moulded these genomes: (a) repeat-mediated deletions and (b) deletion-coupled insertions. These processes, which are described in detail below, were able to be discerned with such clarity because (i), the positions of insertions and deletions could be determined unambiguously to single-base resolution, and (ii), the number of individual cases is large enough to exclude the possibility that they were random events.

Each of the 360 strain-specific sequences represents some type of insertion or deletion (indel), which could be classified into one of several categories (Figure 6, Table S4). About half of the indels represent mobile genetic elements (see below) while the other half are unrelated to known mobile genetic elements. The smallest indels are just longer than 20 bp (the lower cut-off size for manual inspection) while the largest is more than 100 kb (in DV12). Some of the short indels were sub-classified as “polyrepeats”, and consist of a short (e.g. 6-mer) repeated sequence (allowing for some variability), where the number of copies differs between the two strains. For polyrepeats, an exact point of insertion/deletion cannot be defined. Several polyrepeat indels are located within coding regions, where the proteins from the two strains have e.g. dipeptide repeats of different length but with the reading frame conserved.

An unexpected observation was that the majority of the insertions are coupled to deletions such that the two strains possess unrelated DNA sequences integrated at exactly the same position within the chromosome. Deletion-coupled insertions differ from “hot-spots of integration” where insertions and deletions usually occur within a short distance but commonly at distinct positions. Of several possible explanations for this phenomenon, we favour a process where insertion and deletion are mechanistically coupled, so resulting in unrelated sequences at exactly the same position.

A random process can be excluded for statistical reasons, as 64 independent deletion-coupled insertions were identified (Figure 6, Table S4), and these showed no evidence of sequence conservation that could be interpreted as a targeting. The deletion-coupled insertions are also unlike mobile, site-specific pathogenicity islands (PI) found in bacteria [46], as those occurring at the same site are usually closely related and characteristically produce direct repeats upon integration [46], [47]. The results of the current study indicate that deletion-coupled insertions originate by an as yet undescribed mechanism that couples the insertion of foreign DNA to the simultaneous loss of chromosomal genes, an apparently risky strategy for individual cells as it removes part of their own genome. Our data imply that the majority of the insertions of foreign genetic material are coupled to a deletion.

Insertions and associated deletions varied widely in size, with the smallest described being just over 20 bp, the applied minimal cut-off (although even shorter replacements were observed). Both, the deleted sequence and the replacing inserted sequence may be long (16–43 kb), while the longest example was 104 kb (in HBSQ001), corresponding to a 429 bp unrelated sequence in C23^T. Some are likely to have a major biological impact, for example, the exchange of the S-layer glycoprotein in HBSQ001 occurs on a 43.7 kb vs 8.2 kb deletion-coupled insertion region (DV2) that has been reported as genetic island GI1 [14]. Ín this case, the deleted/inserted genes are homologous and thus deletion-coupled insertion results in what has been described as divergent genes in a conserved synteneic context [13]. The exchange of the cell surface glycoprotein may be responsible for the observed morphological difference between the cell walls of the two strains [8]. The Cas/CRISPR systems for virus/plasmid defence, which are exclusively found in C23^T, are also located on such deletion-coupled insertions (DV3 and DV6).

Another unexpected finding are differences that we interpret as resulting from repeat-mediated deletion events, where a sequence of usually <20 bp in one strain is found directly repeated at both ends of a strain-specific sequence while only a single copy of the repeat exists in the corresponding position in the other strain (Table S6). In the analyzed set, there are 16 independent cases of such indels, with direct repeats ranging from 10–19 bp, and an additional 11 cases involving repeats ranging from 7–9 bp. There are also examples where the repeats are 6-mers (3), 5-mers (5) and 4-mers (6). Although short direct repeats such as 4-mers may occur by chance, less than one case (0.4) would be expected in a set of 99 indels.

A sequence that occurs in only one strain represents an “indel” as there is no a priori information to indicate whether it is due to an insertion in one strain or a deletion in the other unless a distinction can be made based on additional information, such as gene truncation, preservation of conserved gene clusters, or tetra-nucleotide analysis. In nearly all repeat-associated indels (10 out of 11) where a distinction between deletion and insertion was possible, a deletion was identified as the cause.

Repeat-mediated deletion events appear to be common among copies of the small mobile element HqIRS46, which contains a direct repeat of 22 bp length (Table S6). Repeat-mediated deletion shortens the element of 390 bp to concatenated terminal sequences of 52 bp.

In a comparison of the genomes of Hbt. salinarum strains NRC-1 and R1, three indels were noted [21] that probably also represent repeat-mediated deletions. One was a precise deletion of a copper-binding domain of the HcpB protein in NRC-1 (via a 32 bp direct repeat). The second was a 133 bp deletion in the promoter of the rRNA operon in R1 (via a 27 bp direct repeat). The third case, a 10007 bp strain-specific sequence (flanked by an 8 bp repeat) present only in NRC-1, was originally believed to be an insertion in the latter strain. Subsequent low pass sequencing of other Halobacterium strains has shown that this is a deletion in R1 (F. Pfeiffer, D. Oesterhelt, unpublished results). These examples of repeat-mediated deletions extend this biological principle to a different genus within the Halobacteriaceae, suggesting that they are widespread and possibly much more common than previously realised.

These types of events are reminiscent of the repair of double-strand breaks (DSB) by a pathway described as micro-homology-mediated end-joining (MMEJ), a process that has been well studied in eukaryotes, and more recently in Archaea [48], [49]. However, we think that the observed cases of repeat-mediated deletion in the present study (and in Halobacterium) have occurred independently of double-strand breaks, as these would be expected to occur at random positions within a genome if generated by radiation or chemical radicals. It is extremely unlikely that the 4 observed core deletions in the small mobile element HqIRS46 resulted from 4 independent double-strand breaks that, by chance, have all occurred within this repeat. Also, it is quite unlikely that double-strand breaks in Halobacterium have – by chance – occurred between the prominent, closely-spaced repeats which occur in the copper-binding domain of halocyanin and in the rRNA promoter region.

The core sequences removed after repeat-mediated deletion events show major size variations, with documented examples ranging from 10 bp to 34 kb (Table S6). Even shorter core deletions have been observed but the resulting events are below our length cut-off of 20 bp (for the sum of core and repeat length). In several cases, the strain not affected by the deletion shows slight sequence variation in the “direct repeat”. It can be assumed that the two copies have been mutated to identity in the other strain prior to the deletion event. Besides the deletion in HqIRS46, there are only four repeat-mediated deletions where the direct repeats are longer than 20 bp (22, 43, 139, 734 bp, respectively). In two cases (43 bp and 22 bp), this is a partial tRNA duplication related to prophage integration (see below).

We propose that repeat-mediated deletions represent an, as yet, undescribed process that contributes to genetic variability. The significance of the present study is that these events have been so clearly documented in natural isolates rather than in experimentally mutagenised systems. The large number of events must reflect a process that is more common than anticipated. The observed similarity between repeat-mediated deletions described here and the MMEJ mechanism of double-strand break repair may indicate that MMEJ may make use of components that are involved in the repeat-mediated deletion process.

It should be emphasized that one prominent source for direct repeats are target duplications associated with transposon insertion. The described repeat-mediated deletion would be an efficient mechanism for trace-free removal of integrated transposons and thus may represent a defence system against such selfish DNA elements. In this context, the high reversion rate of transposon-triggered mutations in Halobacterium may be significant [50]. Upon exit from a donor site, a transposon may leave a double-strand break (which unless repaired would lead to loss of the chromosome) or leave the target duplication as a footprint. In the latter case, most Halobacterium transposons would introduce a frame-shift, as target duplications of 5, 8, and 10 bp are most common. Thus, regain of function would be unlikely. The high reversion rate of transposon-triggered mutations can be nicely explained by the proposed mechanism of repeat-mediated deletions.

In prokaryotes, Cas/CRISPR systems provide a sequence-specific defense barrier against incoming foreign DNA such as viruses or plasmids, and the genes involved often display a high rate of change [19]. The two Haloquadratum strains differ radically in their Cas/CRISPR systems, as summarized in Figure 7. Strain C23^T carries two complete sets of Cas genes, one preceding the crispr-1 locus, and a second, distinct set located between the flanking crispr-2 and crispr-3 loci. The DR sequence found in crispr-1 differs from that present in the other two CRISPR loci, consistent with their distinct Cas genes. There are 85 spacers spread across the three loci but crispr-1 has duplicates of two spacers, and crisprs-2 and -3 share an identical spacer, making a total of 82 unique spacers. Strain HBSQ001 has no Cas genes and only 5 complete spacers. A MITE (HqIRS37) has inserted within the spacer between the third and fourth copy of the DR (counting from the leader sequence end). The residual HBSQ001 crispr locus is closely related to crispr-2 of C23^T as they share the same leader and DR sequences (labeled DR2 in Figure 7), but the lack of Cas genes in HBSQ001 means that the CRISPR system in that strain is non-functional.

The first CRISPR locus in C23^T has no homologue in HBSQ001, as it is located on a long deletion-coupled insertion between the two strains (25.4 kb vs 6.6 kb), representing one of the variable regions (DV3, Table 3; #88 of Table S5). The segment carrying crispr-2 and -3 is also part of a long deletion-coupled insertion between the strains (15.5 kb vs. 5.2 kb, #294), one of 21 strain-specific sequences that constitute hypervariable region DV6 (genomic island GI 2).

A summary of the significant sequence matches between CRISPR spacer sequences and haloarchaeal or halovirus sequences is given in Table S7. None of the CRISPR spacers exactly matched known haloviruses or haloarchaeal plasmids, but a less stringent search against the PL6 plasmids revealed one C23^T spacer that gave an 80% match (29/36 nt) to a sequence within ORF Hqrw_6004 of plasmid PL6A (crispr-2, number 17, Table S7), and the predicted amino acid sequence of this spacer and the corresponding section of the PL6A ORF are nearly identical (11/12 aa). This is not surprising given the widespread carriage of PL6-like plasmids in Haloquadratum. Twelve C23^T spacers were found to closely match metavirome (virus metagenome) sequences from the Santa Pola saltern, Spain [51]. Four of these spacers perfectly matched metavirome target sequences. On the other hand, 4 of the metavirome contigs were targeted by more than one spacer, suggesting they belong to viruses that frequently interact with Haloquadratum. The close sequence relationships found between C23^T CRISPR spacer sequences and the Spanish saltern metavirome provide persuasive evidence that the dominant virus types detected in the saltern from which HBSQ001 was isolated are also present in Australia.

Although no spacer sequences were shared by the two strains, four spacers matched sequences adjacent to the CRISPRs of the other strain (allowing for 1–2 mutations; Figure 7, Table S7). These spacer sequences target two Cas genes and a predicted endonuclease. Three spacers of HBSQ001 closely match sites within the block of Cas genes found between the crispr-2 and crispr-3 loci of C23^T. One of these matching sequences was found after reconstructing the ancestral HBSQ001 spacer by removing the inserted HqIRS37 MITE (Table S7). On the other hand, one spacer of C23^T (crispr3/spacer20) exactly matches a 35 nt sequence in HBSQ001 that is found within an ORF (HQ2151) located just 1.2 kb away from the CRISPR remnant of that strain. The HQ2151 protein is predicted to be a HNH restriction endonuclease (COG3440), genes for which are often carried by bacteriophages, haloviruses [52] and plasmids. Spacer sequences matching integrated viruses or plasmids in other strains of the same species have been previously reported, eg. in Sulfolobus islandicus [20], consistent with the immunity function of CRISPRs.

It can be assumed that the crispr3/spacer20 sequence in C23^T was collected precisely to defend against incoming mobile elements carrying the HQ2151A endonuclease, so the presence of this gene in the Spanish isolate probably represents a successful invasion by such an element despite the CRISPR protection. If the ancestor of the Spanish strain also carried a spacer against this endonuclease, then its integration into the chromosome would have triggered a deletion of at least the cognate spacer, and possibly more, as has been observed in other Archaea [53], [54] It seems likely that nearly the complete Cas/CRISPR locus has been deleted in HBSQ001 by such an event, leaving only the observed remnant. Deletions within CRISPRs [55] are well documented, as is the loss of entire sets of Cas genes [56]. It is not clear why the inter-species spacer targets are close to CRISPRs but it may relate to the common occurrence of Cas/CRISPR genes on plasmids, which would be sensed as foreign DNA when transferred to new strains. This is consistent with phylogenetic studies in Bacteria showing that Cas/CRISPR systems appear to spread by lateral gene transfer [57], and indeed the evidence from tetra-nucleotide composition suggests that both CRISPR loci are contained on regions of foreign DNA. It is likely then that plasmid transfers between Hqr. walsbyi strains occur at an appreciable rate, although the mechanism(s) involved have yet to be experimentally determined. Overall, the spacer/target sequence relationships observed in the CRISPRs of the two strains, and between the Spanish saltern virome and spacers of C23^T, show that the two strains are exposed to incursions by closely related organisms, including viruses and plasmids. This evidence points to a rapid, global dispersal system that links the salterns in Australia and Spain, as if they were a global pond (discussed later).

Strain C23^T carries an orthologue of halomucin, HmuI, an extremely long protein first described for HBSQ001 [7]. The HBSQ001 protein is predicted to be 9,159 aa (27,477 bp) while that of C23^T is 7,837 aa. HmuI is secreted, and surrounds the cells as a cloud of protein as shown by a specific antibody stain (results not shown). It is a highly glycosylated (sialylated) protein, and could play a role in protecting the cell against conditions of desiccation or extremely low water activity [7]. Although the HmuI of C23^T is considerably shorter, the two halomucins retain blocks of high sequence conservation with more than 90% sequence identity, especially in the N-terminal and C-terminal regions. They possess the same major features, an N-terminal Sec-pathway signal sequence, stretches of multiple direct repeats that span much of the protein, and a serine/glycine rich region near the C-terminus. However, the C-type lectin-like (CTLD) domains carried by the HBSQ001 protein are absent in the C23^T HmuI. Comparison at the DNA sequence level indicates that the hmuI genes must have diverged in situ from a common ancestral gene rather than by lateral gene transfer events as both are embedded within a common genomic region that extends into each gene for 119 codons at the N-terminus, and for 559 codons at the C-terminus. Between the flanking core sequences the genes vary in sequence, and this region is designated as divergent region (DV1, #24). The repetitive domain structure of halomucin is likely to have promoted length changes by standard homologous recombination or repeat-mediated deletion, and two independent deletions can be discerned; one of 2.4 kb (HBSQ001 codons 283-1075) which removed the pair of CTLD domains, and one of 1.3 kb (HBSQ001 codons 5078-5538). There are also several other, smaller insertions and deletions. When the HmuI of HBSQ001 was first described in 2006 there were no prokaryotic homologues in the sequence databases but current BLASTP searches of the Genbank database show matches to large, secreted proteins from a number of recently sequenced Archaea (including haloarchaea) and Bacteria. These proteins, often annotated as large exoproteins, GLUG-domain proteins, or filamentous haemagglutinins, are considerably shorter in length than even the C23^T HmuI, with most being between 1,000–2000 aa. The closest sequence was that of Ig-like domain-containing protein Swol_1078 of Syntrophomonas wolfei (E = 10⁻⁴⁴). Two groups of matching sequences were observed; those similar to the HBSQ001-specific CTLD domain containing region (aa 500 – 1,060), and the great majority that matched to the repeats (approx. aa 1,500–5,500 of C23^T HmuI). The similar repeat domains of the latter group of proteins probably reflect a common structural characteristic, such a filamentous shape, and it is possible that HmuI provides an attachment function.

Two additional proteins with characteristics similar to halomucin (very large and probably secreted glycoproteins) have been reported for HBSQ001: Hmu2 (2885 residues) and Hmu3 (2079 residues). As these are encoded on strain-specific sequences (hmu2 on #50, hmu3 on #692), similar proteins are not present in C23^T.

The cell walls of most haloarchaea, excluding halococci, typically comprise a single protein layer, the surface or S-layer that lies directly outside the cell membrane. Experimental evidence from a few haloarchaeal species indicates that the S-layer is composed of only one protein that is usually heavily glycosylated, and forms an interlocking two-dimensional protein sheet attached to the cell membrane [58], [59]. Haloarchaeal genome sequences often have multiple genes annotated as cell surface (glyco)proteins, based largely on the presence of secretion and glycosylation signals. Strain C23^T has twelve such ORFs but only one, Hqrw_1237 (822 aa), is closely related to known S-layer glycoproteins. The highest similarity is to the S-layer protein of Har. japonica (Q9C4B4, 64% aa identity). A second, much smaller ORF, Hqrw_1408 (222 aa), present in both strains, shows about 40% aa identity to the C-terminal 140 residues of known S-layer proteins. The remaining annotated surface proteins share no significant similarity to S-layer proteins. Mass spectrometry analysis of membrane preparations of C23^T revealed the presence of Hqrw_1237, as expected, while Hqrw_1408 was not detected. In addition, three of the other predicted surface proteins were unambiguously identified; Hqrw_1240, Hqrw_1641 and Hqrw_2184. The functions of the latter proteins are unknown.

The probable S-layer glycoprotein genes (csg) of C23^T (Hqrw_1237) and of HBSQ001 (HQ1207A) are located on a long deletion-coupled insertion (43.7 kb vs 8.2 kb, #50). In the latter strain, this had previously been reported to be a hypervariable region, and denoted as genomic island GI 1 [13]. The S-layer protein of C23^T aligns to homologues from other halophiles beginning from its N-terminus, while the orthologue of HBSQ001 appears to have an extended N-terminal region. Only from aa 193 onwards does it show significant sequence identity (55%) to the C23^T homologue Hqrw_1237 (starting at pos. 55), and to the S-layer glycoproteins from other halophiles. However, the N-terminal 192 aa show only very distant homology (if any) to proteins from C23^T or other halophiles. Nevertheless, the protein has a signal sequence and is one of the major membrane proteins in HBSQ001 (Fusetti & Pohlmann, unpublished results). Thus, it is likely to be the S-layer glycoprotein of the Spanish isolate.

The similarities in the C-terminal region between the probable S-layer glycoproteins from the two strains could have been caused by a genome rearrangement within the coding region, but this is not the case, as the level of nucleotide sequence conservation (55%) is drastically lower than the average similarity of other orthologue pairs (97%). The genomic breakpoint has been unambiguously positioned and is located outside of the csg gene. Cuadros-Orellana et. al [13] suggested that some of the variability of this region (GI1) could be due to recombination with related sequences elsewhere on the chromosome. However, this seems less plausible in C23^T as there is little intragenomic similarity. Thus, in contrast to the situation described for the halomucin gene, the csg gene has most likely been acquired by lateral gene transfer in the Spanish isolate. A third form of the cell surface glycoprotein was found on fosmid eHwalsbyi 559 also within the same synteneic gene context [13].

The HBSQ001 specific 43.7 kb DV2 sequence codes not only for a distinct csg and the additional hmu2 gene, but also for 5 additional proteins that are probably secreted and glycosylated. These differences are likely to be responsible for the three-layered cell wall structure observed in HBSQ001 compared to the two-layered surface of C23^T [8]. The combination of a long ca. 150 aa N-terminal extension on the S-layer protein of HBSQ001, and the additional cell surface glycoproteins in this strain, are probably important in altering the outer cell surface to evade viral infection, and may have helped HBSQ001 survive such attacks despite the absence of a functional CRISPR system. There is also a potentially significant difference in the glycosylation pattern of surface and secreted proteins between the two isolates. Sialic acid synthesis genes are present in HBSQ001 but absent in C23^T, as they occur on a deletion-coupled insertion (#690, within DV12).

Both strains carry two different bacterio-opsin-like genes, annotated as bop1 and bop2, along with a single halo-opsin gene (hop). To check for the expression of these and other membrane proteins, cell membranes of the C23^T strain were purified using the protocol for isolating purple-membranes of Halobacterium salinarum [60], and a red-membrane band was observed at the same position on sucrose gradients as purple membranes from Halobacterium salinarum. SDS-PAGE revealed a single, prominent Coomassie stained band of identical apparent MW to Bop of Hbt. salinarum (data not shown), and MS analysis of the excised band confirmed it to be BopI (Table S3). In the previous study of HBSQ001, BopI was also stated to be present in the cell membrane [7]. Minor protein components of the purified membranes of the C23^T strain were also detected, among which were BopII, Hop and Brp-like protein (carotenase) (Table S3). Recently, BopII of Hqr. walsbyi has been found not to pump protons, and its actual function in the cell is yet to be determined [61].

The genomes of both strains were examined for prophages using a number of methods including searches for insertions into tRNA genes, similarity to known halovirus (or prophage) genomes, and to highly conserved virus genes (e.g. terminase).

As mentioned above, genomic island 3 (GI 3, DV11, #658) may represent a strain specific prophage in HBSQ001. Another likely prophage of C23^T occurs at nt 1615793–1635538, where a 19.7 kb virus-like region has integrated into the tRNA-Ala gene. This is a strain specific insertion (#370, DV7) with an average %G+C content of 44.5, about 3 percentage points lower than that of the entire chromosome, and an altered tetra-nucleotide composition. A phage integrase is adjacent to the right end, which is marked by a 22 nt duplication of the tRNA. The ORFs between are typical of bacteriophages, including DNA polymerase, restriction/modification enzymes and a cluster of several ORFs that overlap at start/stop codons and are likely to be transcribed as a single unit. The DNA polymerase (Hqrw_2609), which is adjacent to the left end of the tRNA gene, is most similar to that of halovirus HF1, but is interrupted by a frame-shift, so this prophage is either defective or relies on the host enzyme for replication.

Common to both strains is a partial duplication of the tRNA-Thr1 gene that occurs 6.7 kb further downstream (nt 103,049–103,105 in C23^T). The region between contains genes for hypothetical and conserved hypothetical proteins, along with many pseudogenes, some being cdc6 homologues. The partial tRNA-Thr1 copy is imperfect, with 5 mismatches over 57 nt, and is missing 15 nt at the 5′ end. There are no integrase genes nearby. A similarly truncated tRNA-Thr gene is also carried by haloviruses HF1 and HF2, but these occur nearby a predicted integrase gene, presumably to allow host integration [52]. It is likely that this 6.7 kb region of Haloquadratum is a remnant of either a prophage or an integrative plasmid.

As discussed above, plasmid PL6 may represent a prophage in C23^T. Whether the larger plasmid, PL100, is a prophage or a plasmid is difficult to predict as it shares no significant similarity to known haloviruses and has a high proportion of hypothetical or conserved hypothetical ORFs. While ORF Hqrw_5003 shows almost perfect similarity to a metavirome sequence derived from the Santa Pola saltern in Spain (ADE29299), and several other ORFs show significant similarity to halovirus or bacteriophage ORFs, many of these are similar in function to genes also typically carried by plasmids, such as endonuclease, helicase, integrase, and Cdc6 homologues. The presence of a Cdc6 homologue cannot be taken as evidence that PL100 is simply a plasmid, as halovirus BJ-1 carries such a gene [62].

Four hypervariable regions (genomic islands, GI 1–4) of HBSQ001 were described by Cuadros-Orellana et al. [13] (Fig 5). These show high variability between the HBSQ001 genome and environmental sequences recovered from the saltern from which this strain was isolated (i.e. autochthonous DNA). Since GIs 1–4 largely correspond to divergent regions (DV) observed in the current study (described above), only relevant information that has not already described will be given here. Genomic Island 1 corresponds to DV2 (see above). Genomic Island 2 is 185 kbp in length (nt 1269154 – 1454284, #264 – #314 of Table S5) and corresponds to a similarly sized region, DV6, in C23^T (203 kb; nt 1268551 – 1472326). The latter region harbors many IS elements, phage-associated, and metabolic genes. Unlike most other GIs and DVs, which display a single large indel between the two strains, GI 2 shows 21 independent indels or deletion-coupled insertions, several of which exceed 10 kb. These strain-specific sequences are interspersed among core sequences ranging from as little as 43 bp to 12 kb. Commonly, the indels are separated by not more than 5 kb (Table S5). The whole region has an increased G+C content (50.0%), an altered tetra-nucleotide frequency, is devoid of tRNAs and ribosomal protein genes, and is located mid-way between ori1 and ori2/ori3. GI 2 (DV6) is particularly abundant in non-transposase related pseudogenes (Figure 1). The clustering of numerous distinct indels/replacements in a probable replication termination region is curious, but appears similar to observations in strains of Sulfolobus islandicus recovered from widely spaced sites (see Figure 3 of [33]).

GI 2 (DV6) begins abruptly in HBSQ001 just after nt 1269163 (strain-specific region #264) where a large deletion has removed most of ORF HQ2048A. Comparison with C23^T shows this is a deletion because the corresponding C23^T ORF, Hqrw_2212, remains intact. The ends of the undeleted region in C23^T have a direct repeat, ACATCATTCT, while in HBSQ001 there is only one copy directly at the deletion border. This is one of the cases of a repeat-mediated deletion, which has eliminated just over 16 kb from the HBSQ001 genome. Fortuitously, among the autochthonous DNA sequences recovered from the Santa Pola saltern in Spain by [13], was a 15 kb fosmid clone sequence (EF583981) that shared a similar %G+C as Hqr. walsbyi but did not match any sequence in HBSQ001. In the current study this sequence was found to be 99% identical to C23^T across its entire length, and mapped within this 16 kb deletion in GI 2 (nt 1268551–1284680). This example further supports the tight relationship between Hqr. walsbyi populations present at the two distinct global sites.

A significant strain-specific difference found within GI 2 (DV6) is the CRISPR array and associated Cas genes of C23^T (see above). In this case C23^T has 15.5 kb relative to 5.2 kb in HBSQ001 (#294). The comparative sequence data is consistent with HBSQ001 suffering a deletion relative to C23^T, as the leader and first DR are contiguous with the shared sequence between the two strains (Figure 7), but after this HBSQ001 has only a few remnant spacers left while C23^T retains a full set of spacers as well as Cas genes.

Mobile, integrative genetic elements, because of their repetitive and propagative nature, are significant drivers of cellular evolution, inactivating genes and generating multiple copies that can lead to deletions and other genomic rearrangements [67]. Knowledge of these elements in a genome can help explain its structure, evolution and propensity for change, but identifying them can be difficult, particularly if they are novel, short, or show some variability in sequence. The unexpectedly close similarity of the two Haloquadratum strains provided an extraordinary advantage in detecting such elements. A surprising finding from these comprehensive searches is that despite both strains carrying numerous mobile genetic elements, their shared sequences have remained synteneic. Rather, there is a strong tendency for MGEs to target other MGEs.

Both strains carry large numbers of MGEs: a total of 528 for C23^T and 536 for HBSQ001. These are of various types, including canonical transposons (13 families), IS605-type transposons (29 families) and three categories of transposase-free mobile repeats: MITEs (miniature inverted-repeat transposable elements, 5 families), PATEs (palindrome-associated transposable elements, 6 families), and SMRs (small mobile repeats, 5 families). Since MITEs lack a transposase, their mobilisation relies on an appropriate transposase provided in trans, which also determines the type of target duplication (if any). A MITE cannot be considered a degenerate version of the transposon supplying the transposase, as their sequences are unrelated except for the few terminal bases of the inverted terminal repeat. SMRs also contain ITRs but do not cause target duplications and, unlike MITEs, their ITRs show no sequence similarity to any of the transposons. Currently, their mode of mobilization is enigmatic. We observed a strong targetting preference for several of the SMR families (see below). PATEs differ from MITEs and SMRs in being devoid of ITRs, but they contain near-terminal palindromic sequences. During the analyses described below, these were found to be related to those of IS605-type transposons. Indeed, some PATEs appear to have originated from IS605-type transposon by repeat-mediated deletion of the core sequence, and so represent degenerate versions with only the fused terminal sequences.

Transposase-free MGEs (MITEs, PATEs, and SMRs) are far more numerous than transposons (Table 4) and are responsible for the majority of the strain differences caused by MGEs (Figure 6). The distribution of these elements between the two strains of Haloquadratum is similar except for two families of SMRs that are only present in C23^T (see below). MGEs appear to be relatively evenly distributed around the C23^T chromosome (data not shown) but the large plasmid, PL100, is completely free of SMRs, possibly reflecting a recent acquisition.

In comparing MGEs between the two strains, the following categories were recognized: (a) common elements, which occur at corresponding genome positions, (b) strain-specific element insertions (SSEIs), which represent one type of strain-specific elements, and where the adjacent sequence is contiguous in one strain and interrupted by the mobile element in the other, (c) elements which occur within larger, strain-specific sequences (indels), and so are likely to have either been carried into the cell as part of a segment of foreign DNA, or have inserted into such a segment after its integration into the chromosome. In several cases, indels start and/or terminate with an MGE. This may imply that MGEs can trigger deletion events that may also remove neighbouring sequences.

The mobility of MGEs was inferred from the presence of strain-specific insertions (SSEIs, case b), and although these could possibly also result from the removal of a hypothetical corresponding element in the other strain, e.g. by repeat-mediated deletions via the flanking target site duplications, we believe most SSEIs represent insertions. On this basis, the most active MGEs in C23^T are the SMR HqIRS55 (10 SSEIs), the C23^T-specific transposase ISHwa4 (8 SSEIs), and the PATE HqIRS56 (7 SSEIs).

The key features of all Haloquadratum MGEs are summarized in Table S8, in which complete transposons are referred to by their ISFinder database numbers [68], prefixed with ISHwa, while the other elements are assigned numbers prefixed with HqIRS (for Haloquadratum IS-Related Sequences).

Twelve of the 13 distinct transposon families detected in the two strains have been previously described in Halobacterium [21] or other haloarchaea, but the remaining one, ISHwa2, is related to Tn5, a transposon widely distributed in Bacteria. Although ISHwa2 occurs multiple times in both strains it is not found in any of the other 78 completely sequenced Archaea, except for a small remnant in Natronomonas pharaonis. This distribution is indicative of a lateral gene transfer event from a bacterial representative, and the transposase of ISHwa2 most closely matches homologues in Microcoleus chthonoplastes, a cyanobacterial species known to inhabit hypersaline environments.

Transposon mobility is likely in at least 10 of the families as examples of SSEIs were observed but they were not equally active: seven of the 10 transposon families show SSEIs exclusively in strain C23^T, while the remaining 3 transposon families show SSEIs in both strains.

In some transposon families, the pattern of SSEIs and their distribution within and between strains indicates that the transposon was carried into the cell via an integration event of foreign DNA providing a source of infection for the rest of the genome. As detailed in Text S1, this is exemplified by a 43 kb C23^T-specific sequence (DV10, #578), which carries a total of 10 transposons (or fragments thereof), several glycosyltransferase genes and one phage integrase homolog. It is likely that DV10 is responsible for the ‘outbreak’ of four transposons (ISHwa4, ISHwa8, ISHwa12, and ISHwa13). Although the DV10-specific copies of ISHwa8 and ISHwa12 are broken, this could have occurred after their spread to other genome locations (see Text S1).

MITEs depend on a suitable transposase supplied in trans by a regular transposon, with which they share a similar inverted terminal repeat sequence as well as the length of an eventual target duplication [69]. Four of the five distinct families of MITEs detected in Hqr. walsbyi are non-coding and shorter than 500 bp, and the most frequent of these have been previously described [69]. The ISH2-type element HqIRS32 is the longest MITE (ca. 520 bp) and is the only example containing an ORF likely to code for a protein (but not a transposase). There is evidence from Halobacterium that this ORF in ISH2 is expressed [70]. Homologous short ORFs were detected in (HqIRS33 on DV10, HqIRS34).

The mobility of some MITEs, as reflected by SSEIs, was curious, as no functional copy of the appropriate transposase was found elsewhere in the genome. For example, MITE HqIRS39 has an ITR similar to that of transposon HqIRS12 but all copies of the HqIRS12 transposase are disrupted by frame-shifts or in-frame stop codons (see also Text S1).

Three families of SMRs (HqIRS42, HqIRS43, and HqIRS55, grouped as SMR-A) share a number of common features despite a lack of sequence similarity: they are about 110 bp long, have prominent inverted terminal repeats, do not generate target duplications, are relatively frequent (up to 54 copies in one genome), and seem to have a strong and identical target site preference, namely to target PATEs and IS605-type transposons, inserting within their near-terminal palindromes (see also below). As an example, the relatively frequent PATE HqIRS56 is regularly targeted by SMR-A elements (9 of 19 in C23^T and 12 of 23 in HBSQ001).

The other two families of SMRs (HqIRS40 and HqIRS41, grouped as SMR-B) occur only in C23^T. SMR-Bs are similar to SMR-As in that they target other MGEs but show little targeting preference. HqIRS41 is exceedingly short, being only 80 bp long.

HqIRS46 is a PATE with a direct repeat of 22 bp, representing the two near-terminal palindromes (Table S9). In four cases, the element is complete in one strain while the core and one copy of the 22-mer have been deleted in the other strain. Core deletions, which lead to a terminal-only version of the element with only 52 bp, have all the characteristics of repeat-mediated deletions (Table S6). However, there are not only examples where the core has been deleted in one of the strains, there are also cases where terminal-only copies of HqIRS46 occur at the same genome position in both strains. In these cases it cannot be decided if the core has been deleted independently in the two strains or if the terminal-only version of the element has been mobilized.

An equivalent core deletion occurs for IS605-type transposon ISHwa23, which codes for an IS1341-type transposase (TnpB). The deleted region is flanked by a 19 bp direct repeat which, again, resembles the near-terminal palindrome (Table S6). In this case, the element is shortened from a 1500 bp transposon to a 52 bp PATE. As described in the Text S1, this core deletion removes a transposon that must have been mobilized before. There are 6 SSEIs of ISHwa23 in HBSQ001, but in all six cases the terminal-only 52 bp version is found. This may indicate that this extremely short transposon-derived PATE can be mobilized.

Several other of the IS605-type elements were found to exist as terminal-only PATEs. In addition, several elements contain other types of internal deletions that remove most of the transposase genes. In several transposon remnants that carry IS200-type and IS1341-type transposase genes in divergent orientation, large internal regions are deleted so that only short C-terminal remnants of the two transposons persist. One PATE, HqIRS71, shows high sequence similarity to an extended terminal segment of certain IS605-type transposons. This includes part of the C-terminal coding region, but all ORFs on HqIRS71 are disrupted and so HqIRS71 is considered to be non-coding.

In summary, IS605-type transposons frequently suffer deletions in Haloquadratum. This is one of the reasons why most of the IS1341-type transposons are pseudogenes in this organism. On several occasions, direct repeats are found adjacent to the deleted region and thus we propose that the mechanism of repeat-mediated deletion provides a defense mechanism against IS605-type transposons and related PATEs in Haloquadratum.

The extraordinary genome conservation and shared CRISPR spacer relationships observed between geographically distant isolates of Hqr. walsbyi speak of a rapid global dispersal, possibly by airborne salt particles or migratory birds. The strong, global coherence of this species also argues for a strong selection against change and/or a high competitive fitness in hypersaline waters, perhaps related to protection from desiccation by halomucin. It seems that hypersaline waters all over the world represent a type of global pond for this organism. The genomes differ mainly by indels, deletion-coupled insertions, and movements of mobile genetic elements, with deletions occurring frequently via a potentially novel mechanism involving small direct repeats. The high number of pseudogenes, particularly of transposases, in both strains probably results from this, and may reflect a host protective mechanism against mobile genetic elements. Deletion-coupled insertions indicate the uptake and integration of foreign DNA but the mechanisms underlying the generation of these identically placed but unrelated sequences remain to be understood. Strain C23^T was probably infected by four distinct transposons upon integration of a foreign sequence of 42.9 kb. At least two distinct mechanisms are used to attack transposons. Repeat-mediated deletions permit trace-free removal of canonical transposons, taking advantage of the associated target duplications. The same mechanism can also inactivate IS605-type transposons (and associated PATEs) by causing element core deletions. In addition, the SMR-A category of small mobile repeats preferentially targets IS605-type transposons and PATEs. The link between PL6-like plasmids and conserved genomic loci related to haloviruses (ViPREs), points to a close relationship between plasmids and viruses in haloarchaea, the details of which remain to be understood.

Haloquadratum walsbyi C23^T ( = JCM 12705^T = DSM 16854^T) was cultivated in DBCM2 medium as described previously [8]. Total DNA was extracted from cell pellets using the Qiagen Genomic-tip DNA extraction kit (Qiagen, GmbH). Initial sequencing was performed using the Roche GS FLX sequencing platform with LR70 reagents. A total of 69.6 Mb data was collected and assembled using the Newbler software (provided with the sequencing instrument). A combined total of 3.07 Mb resulted from the initial round of assembly with 99.78% of the bases being Q40 or higher, at an average coverage of 22.5 fold. The resulting contigs were ordered according to the genome of HBS001, which gave an early indication of the complete synteny between these two strains. PCR on genomic DNA provided templates to primer walk across the contig gaps. Assembly of the 454 contigs and PCR sequences utilized the Phred–Phrap–Consed package [74].

The reliability of the 454 sequencing results was analyzed. Bases with a low quality values (< 25 from the Newbler software) were checked. Of these 186 could be compared to data from a preliminary sequencing effort. Only 2 differences were observed. Low quality bases within polynucleotide runs, which potentially cause frame-shifts were compared by BLASTX to proteins from haloarchaea with or without the low quality base. This also indicated that most of the low quality bases were correct. Additional potential errors in the 454 sequenced contigs were identified by comparison with the sequence of the closely related HBSQ001 strain, particularly at positions where the C23^T base was of low quality and differed from the corresponding base in the other strain. PCR amplification and re-sequencing of these regions in the C23^T genome detected a total of 42 errors. Seven additional errors were detected when proteomic data resulted in the identification of pseudogenes. The total number of errors thus computes to 15/Mb. The sequences of primers used for PCR and primer walking are available on request from the corresponding author. The sequences of the main chromosome, PL100, PL6A and PL6B have the following EMBL accession numbers: FR746099, FR746100, FR746101 and FR746102.

The TaKaRa LA Taq kit (Takara Bio Inc., Japan) was used for all PCRs. Template was genomic DNA from Haloquadratum walsbyi C23^T (described above). Reaction cycles and conditions were varied depending upon the expected product length and template difficulty (ISH elements), but a typical scheme was as follows: 50 µl reaction, ∼100 ng DNA: initial denaturation at 94°C, 1 min; 25 cycles of 94°C/30 sec, 58°C/30 sec, 72°C/4 min; final extension 72°C/4 min. Product size and quality was assessed by agarose gel electrophoresis of a 2 µl sample. Where reactions contained subsidiary, unspecific products, the major band was purified by preparative agarose gel electrophoresis. All products were purified to remove primers and salts using the Wizard SV gel and PCR cleanup kit (Promega, GmbH). Sequencing of PCR products was performed using the ABI PRISM BigDye terminator method at the MPI of Biochemistry core sequencing facility.

Several sequencing reads from an earlier, preliminary sequencing effort could not be matched to the genome sequence based on 454 sequencing. Some of these could be assembled into contigs, indicating that they were not simple contaminants. PCR primers designed from these contigs were then used to detect the presence of a plasmid in some of the subcultures of C23^T. As the original sequencing reads were mate pairs, this enabled contig adjacencies to be determined, which could be used to close most gaps by a PCR approach using (as a template) plasmid DNA purified from a positive subculture detected in the previous PCR. Due to the low overall sequence coverage, non-assembled singletons in the original dataset were also utilized in phred-phrap assembly, and these enabled further extension of remaining contigs, which could be incorporated by further PCR-based gap closure. Final closure of PL100 was achieved by an all-against-all PCR approach on the few remaining contigs.

The genome annotation and core analyses were performed using the HaloLex system (http://www.halolex.mpg.de, [75]). Further analyses were done using the MIGenAS server (http://www.migenas.org, [76]). The common core of the two strains shows 98.6% sequence identity. Accordingly, there must be a complete consistency with respect to ORF annotations in this shared sequence, including annotation of pseudogenes. An attempt has been made to keep the ORF annotations in the two strains as consistent as possible.

A number of specific bioinformatics tools were used, including the following for: CRISPRs (http://crispr.u-psud.fr/), genome alignment using MUMmer (http://mummer.sourceforge.net/), 7S RNA detection (http://bio.lundberg.gu.se/srpscan/), rRNA detection (http://www.cbs.dtu.dk/services/RNAmmer/), tRNA detection with tRNA-scan (http://lowelab.ucsc.edu/tRNAscan-SE/) and Aragorn (http://130.235.46.10/ARAGORN/), tetramer analysis (http://www.megx.net/tetra) [77], genome comparison/alignment (http://asap.ahabs.wisc.edu/mauve/), cumulative skew plots with GenSkewApp (http://www.helmholtz-muenchen.de/en/mips/services/analysis-tools/genskew/index.html), GC-Profile (http://tubic.tju.edu.cn/GC-Profile/), Z-curve plots (http://tubic.tju.edu.cn/zcurve/), GC% with a 2.5 SD cutoff [78], and elements of the EMBOSS suite of bioinformatics programs found within the eBioX program (http://www.ebioinformatics.org). Pilfind was computed at http://signalfind.org/.

A PERL-based script was applied to compare the two chromosome sequences. This script allows a base-by-base comparison but is robust against sequence duplications and indels of any size. The script interprets the pair of sequences as an alternate set of “runs”, defined as subsequences that are completely identical, and “connectors”, which are the divergent sequences that occur between runs. Runs are identified by “word matches”, using exclusively words that are unique in each of the sequences. These words are expanded on both sides as long as the sequences remain identical, which results in a run of a given length. In an initial scaffolding phase, long words (15-mers) are used and only the longest runs (summing up to 30% of the total chromosomal sequences) are considered. The serial order of the runs is determined, which is preserved within synteneic regions of the chromosomes. The result of this phase is an ordered set of runs that build a scaffold of the chromosome sequences. In a second phase, each of the intermediate sequences between runs is subjected to the same procedure except that a short word length (3 bases) is applied and the longest identified run is immediately intercalated into the scaffold. This intercalation creates up to two new but shorter intermediate sequences, one on each side of the newly intercalated run. Intermediate sequences for which no additional run can be identified are “connectors”. Because the algorithm uses trinucleotides as the shortest word size, and thus the minimal length of a run, it performs a base-by-base comparison of the chromosomes. All sequence differences are then represented in the connectors. The procedure allows any length of the connector, from zero (in case of an indel) to more than 100 kb (a huge strain-specific region). As an example, point mutations are connectors of length one in both sequences.

The algorithm can handle sequences that are full of duplications, a common problem in the genomes of halophilic archaea. Only words that are unique in each of the sequences are applied for matching, which excludes words originating from duplications. After this initial matching, sequences are extended on both sides as long as they remain identical. No restrictions apply in this step, so that co-localized copies of duplicated sequences will become part of the corresponding runs. Sequence deviations between the co-localized copies do not cause any problems as they are handled in the second phase, being intermediate sequences between runs. At this stage, additional copies located elsewhere in the genome are no longer relevant.

Most connectors represent trivial sequence differences (e.g. point mutations). For the current project, connectors were selected as non-trivial if they fulfilled one of the following criteria: (a) a length difference of at least 20 bp, (b) the number of single-base differences exceeds 10 in a single connector (which is defined not to have even a single unique trinucleotide match), or (c) the sequence identity is below 50% (which is used only for connectors of at least 7 bp length). These selection criteria are highly relaxed, allowing a detailed comparison of the two chromosomes to be performed. The selected set of 512 non-trivial connectors was inspected manually. The region of difference, and 100–200 bases on each side, were compared to the genome sequences using BLASTN. Cases representing just an enhanced sequence variability (as obvious from a contiguous blast alignment which is free of long gaps) were skipped. The remaining set contained 360 strain-specific sequences. The remainder of the chromosomes, excluding the strain-specific sequences, represents the shared sequence.