Main

Of all of the biogeochemical cycles on Earth, the methane cycle may be most tightly linked to climate. Methane (CH4) is a greenhouse gas roughly 30 times more potent than carbon dioxide (CO2), and approximately 1 gigatonne is produced annually by methanogenic (methane-producing) archaea that inhabit anoxic environments2. The efflux of methane into the atmosphere is mitigated by methane-oxidizing microorganisms (methanotrophs). In oxic environments, CH4 is consumed by aerobic bacteria that use methane monooxygenase (MMO) and O2 as a terminal electron acceptor3, whereas in anoxic environments, anaerobic methanotrophic archaea (ANME) use a reverse methanogenesis pathway to oxidize CH4, the key enzyme of which is methyl-CoM reductase (MCR)4,5. Some ANMEs rely on a syntrophic partner to couple CH4 oxidation to the reduction of terminal electron acceptors, yet Methanoperedens (ANME-2d, phylum Euryarchaeota) can directly couple CH4 oxidation to the reduction of iron, nitrate or manganese6,7. Some phenomena have been suggested to modulate rates of methane oxidation. For example, some phages can decrease rates of methane oxidation by infection and lysis of methane-oxidizing bacteria8, and others with the critical subunit of MMO9 probably increase the ability of their host bacteria to conserve energy during phage replication. Here we report the discovery of novel extrachromosomal elements (ECEs) that are inferred to replicate within Methanoperedens spp. Their numerous and diverse metabolism-relevant genes, huge size and distinctive genomic architecture distinguish these archaeal ECEs from all previously reported elements associated with archaea10,11,12 and from bacteriophages, which typically have one or a few biogeochemically relevant genes13,14. We hypothesize that these novel ECEs may substantially impact the capacity of Methanoperedens spp. to oxidize methane.

Genome structure and features

By analysis of whole-community metagenomic data from wetland soils in California, USA (Extended Data Fig. 1), we discovered enigmatic genetic elements, the genomes for three of which were carefully manually curated to completion (Methods). From sediment samples from the Rifle, Colorado aquifer15, we recovered partial genomes from a single population related to those from the wetland soils; the sequences were combined and manually curated to ultimately yield a fourth complete genome (Methods). All four curated genomes are linear and terminated by more than 1-kb inverted repeats. The genome sizes range from 661,708 to 918,293 kb (Fig. 1a, Extended Data Table 1 and Supplementary Table 1). Prominent features of all genomes are 25–54 regions composed of perfect tandem direct repeats (Fig. 1b and Supplementary Table 2) that are novel (Extended Data Fig. 2) and occur both in intergenic regions and in genes where they usually introduce perfect amino acid repeats (Supplementary Table 2). All genomes have two replichores of unequal lengths and initiate replication at the chromosome ends (Extended Data Fig. 3). Each replichore carries essentially all genes on one strand (Fig. 1a). Although the majority of genes are novel, approximately 21% of the predicted proteins have best matches to proteins of Archaea (Extended Data Fig. 4a), and the largest group of these have best matches to proteins of Methanoperedens spp. (Extended Data Fig. 4b). Of note, the GC contents of the four genomes are approximately 10% lower than those of previously reported and coexisting Methanoperedens species (Fig. 2a). We rule out the possibility that these sequences represent genomes of novel Archaea, as they lack almost all of the single-copy genes found in archaeal genomes and sets of ribosomal proteins that are present even in obligate symbionts (Extended Data Figs. 5 and 6a and Supplementary Tables 3–6). There are no additional sequences in the datasets that could comprise additional portions of these genomes. Thus, they are clearly neither part of Methanoperedens spp. genomes nor parts of the genomes of other archaea.

Fig. 1: Borgs share overall genomic features.
figure 1

a, Genome replichores (arrows) and coding strands (black bars) for aligned pairs of the four complete (Black, Purple, Sky and Lilac) and one near-complete (Orange) Borg. Blocks of sequence with identifiable nucleotide similarity are shown in between each pair (coloured graphs linked by lines; y axes show similarity). b, Genome overviews showing the distribution of three or more perfect tandem direct repeats (gold rods) along the complete genomes. Insets provide examples of local elevated GC content associated with certain gene clusters and within gene and intergenic tandem direct repeats (gold arrows).

Full size image
Fig. 2: Borg and Methanoperedens spp. genomic features and abundance patterns.
figure 2

a, The average genome GC contents of Borgs and Methanoperedens spp. are distinct. The black line denotes the median, and the dashed lines show the interquartile range. b, Groups of related Methanoperedens spp. (rows) correlate with groups of Borgs (columns) across a set of 50 samples. The asterisks indicate two-sided Pearson correlations above 0.92 with FDR-corrected P values below 2.0 × 10–20 that suggest that Brown, Green, Orange, Beige and Ochre Borgs associate with one group of Methanoperedens spp., Olive, Cyan, Gold, Apricot and Rose associate with a second group, and Black associate with a third group.Black asterisks indicate best association with a Methanoperedens genome (correlation ≥ 0.92, P ≤ 1 × 10–20); grey asterisks indicate association with a scaffold containing the Methanoperedens L11 marker gene (correlation ≥ 0.92, P ≤ 1 × 10–20). c, Frequency of genes in different functional groups in the four complete Borg genomes. d, Comparison of the protein family composition of Borgs and Methanoperedens spp. Clustering on the basis of shared protein family content highlights groups of Borg-specific protein families (blue shading) and protein families shared with their hosts (orange shading). The full clustering, including diverse archaeal mobile elements, is shown in Extended Data Fig. 5. PEGA, surface layer protein; PHA, polyhydroxyalkanoate.

Full size image

Abundances of Methanoperedens spp. and some ECEs are tightly correlated over a set of 46 different wetland soil samples (43 genomes were included in the analysis; Extended Data Fig. 6b). This observation supports other indications that these ECEs associate with Methanoperedens and suggests that specific ECEs have distinct Methanoperedens spp. hosts (Fig. 2b). This is true for one ECE whose abundances correlate reasonably well with a specific host group, in which ECE to Methanoperedens spp. abundance ratios range from 2:1 to 8:1. Given their up to approximately 1-Mb length, there may be more ECE DNA in some host cells than host DNA. The Borg sequences are much more abundant in deep, anoxic soil samples (Extended Data Fig. 7a,b).

A few percent of the genes in the genomes have locally elevated GC contents that approach, and in some cases match, those of coexisting Methanoperedens spp. (Fig. 1b). This, and the very high similarity of some protein sequences to those of Methanoperedens spp., indicates that these genes were acquired by lateral gene transfer from Methanoperedens spp. Other genes with best matches to Methanoperedens spp. genes have lower GC contents (closer to those of these ECEs at approximately 33%), suggesting that their DNA composition has partly or completely ameliorated since acquisition16.

Archaeal ECEs include viruses17, plasmids18 and minichromosomes, sometimes also referred to as megaplasmids10,11,12. The genomes reported here are much larger than those of all known archaeal viruses, some of which have small, linear genomes12, and at least three are larger than any known bacteriophage19. These linear elements are larger than all of the reported circular plasmids that affiliate with halophiles, methanogens and archaeal thermophiles. We did not detect genes for plasmid partitioning or conjugative systems, rRNA loci or encoded viral proteins (Supplementary Table 3), and the genomes were markedly different from recently reported Methanoperedens spp. plasmids20. The distinctly lower GC content and variable copy number argue against their classification as archaeal minichromosomes12,21. Thus, we cannot confidently classify the ECEs as viruses, plasmids or minichromosomes. Moreover, the protein family profiles are quite distinct from those of archaeal and bacterial ECEs (Fig. 2d and Extended Data Fig. 5). Some bacterial megaplasmids have been reported to be very large and linear, but they typically encode few or no essential genes22, and if they contain repeats, they are interspaced (that is, not tandem)23. Each distinctive feature of the ECEs has been reported in microbial genomes, plasmids or viruses, but the combination of these features in these huge ECEs is unique. Thus, we conclude that the genomes represent novel archaeal ECEs that occur in association with, but not as part of, Methanoperedens spp. genomes. We refer to these as Borgs, a name that reflects their propensity to assimilate genes from organisms, most notably Methanoperedens spp.

Using criteria based on the features of the four complete Borgs, we searched for additional Borgs in our metagenomic datasets from a wide diversity of environment types. From the wetland soil, we constructed bins for 11 additional Borgs, some of which exceed 1 Mb in length (Extended Data Table 1 and Supplementary Table 1). Other Borgs were sampled from the Rifle, Colorado aquifer, discharge from an abandoned Corona mercury mine in Napa County, California, and from shallow riverbed pore fluids in the East River, Colorado. In total, we recovered genome bins for 19 different Borgs, each of which was assigned a colour-based name. We found no Borgs in some samples, despite the presence of Methanoperedens spp. at very high abundance levels (Extended Data Fig. 7). Thus, it appears that these ECEs do not associate with all Methanoperedens spp.

Pairs of the four complete Borg genomes (Purple, Black, Sky and Lilac) and three fragments of the Orange Borg are alignable over much of their lengths (Fig. 1a). The Rose and Sky Borg genomes are also largely syntenous (Extended Data Fig. 8a) and were reconstructed from different samples that contain these Borgs at very different levels of abundance. Despite only sharing a less than 50% average nucleotide identity across most of their genomes, the genomes have multiple regions that share 100% nucleotide identity, one of which is approximately 11 kb in length (Extended Data Fig. 8b,c). This suggests that these two Borgs recombined, indicating that they recently coexisted within the same host cell.

Borg gene inventories

Many Borg genomes encode mobile element defence systems, including RNA-targeting type III-A CRISPR–Cas systems that lack spacer acquisition machinery, a feature previously noted in huge bacterial viruses19. An Orange Borg CRISPR spacer targets a gene in a mobile region in a coexisting Methanoperedens spp. (Extended Data Fig. 8d), further supporting the conclusion that Methanoperedens spp. are the Borg hosts.

The four complete genomes and almost all of the near-complete and partial genomes encode ribosomal protein L11 (rpL11), and some have one or two other ribosomal proteins (Extended Data Fig. 6a). The rpL11 protein sequences form a group that places phylogenetically sibling to those of Methanoperedens spp. (Extended Data Fig. 9), further reinforcing the link between Borgs and Methanoperedens spp. Four additional rpL11 sequences were identified on short contigs from the wetland group with the Borg sequences and probably represent additional Borgs (Supplementary Table 1). The topology of the rpL11 tree, and similar topologies observed for phylogenetic trees constructed using other ribosomal proteins, MCR proteins, electron transfer flavoproteins and aconitase, may indicate the presence of translation-related genes in the Borg ancestor (Extended Data Fig. 9 and Supplementary Information).

The most highly represented Borg genes encode glycosyltransferases, which are proteins involved in DNA and RNA manipulation, transport, energy and the cell surface (PEGA and S-layer proteins). Also prevalent are many genes encoding membrane-associated proteins of unknown function that may impact the membrane profile of their host (Fig. 2c). At least seven Borgs carry a nifHDK operon for nitrogen fixation, also predicted in Methanoperedens spp. genomes, and may augment the influence of the host on nitrogen cycling (Fig. 1b, Supplementary Information and Supplementary Table 6). Potentially related to survival under resource limitation are genes in at least ten Borg genomes for synthesis of the carbon storage compound polyhydroxyalkanoate (PHA), a capacity also predicted for Methanoperedens spp.24. Other stress-related genes encode tellurium resistance proteins that do not occur in Methanoperedens spp. genomes (Supplementary Table 5). All Borgs carry large FtsZ-tubulin homologues that may be involved in cell division, and proteins with the TEP1-like TROVE domain protein that also do not occur in Methanoperedens spp. genomes (Supplementary Table 5). These may form a complex similar to Telomerase, Ro or Vault ribonucleoproteins, although their function remains unclear25. Several Borgs encode two genes of the tricarboxylic acid cycle (citrate synthase and aconitase; Supplementary Information).

Many Borg genes are predicted to have roles in redox and respiratory reactions. The Black Borg encodes cfbB and cfbC, genes involved in the biosynthesis of F430, which is the cofactor for MCR, the central enzyme involved in methane oxidation by Methanoperedens spp. The similarity in GC content of Borg cfbB and cfbC and protein sequences of coexisting Methanoperedens spp. suggests that these genes were acquired from Methanoperedens spp. recently. The Blue and Olive Borgs encode cofE (encoding coenzyme F420:L-glutamate ligase), which is involved in the biosynthesis of a precursor for F420. The Blue and Pink Borgs have an electron bifurcating complex (Supplementary Information) that includes d-lactate dehydrogenase. Eight Borgs encode genes for biosynthesis of tetrahydromethanopterin, a coenzyme used in methanogenesis, and ferredoxin proteins, which may serve as electron carriers. The Green and Sky Borgs also encode 5,6,7,8-tetrahydromethanopterin hydro-lyase (Fae), an enzyme responsible for formaldehyde detoxification and involved in pentose-phosphate synthesis. Also identified were genes encoding carbon monoxide dehydrogenase (CODH), plastocyanin, cupredoxins and many multihaem cytochromes (MHCs). These results indicate substantial Borg potential to augment the energy conservation by Methanoperedens spp. This is especially apparent for the Lilac Borg.

Host-relevant gene inventory of Lilac Borg

We analysed the genes of the complete Lilac Borg genome in detail as, unlike the other Borgs, the Lilac Borg co-occurs with a single group of Methanoperedens spp. that probably represent the host (Fig. 3 and Supplementary Table 7). Remarkably, this Borg genome encodes an MCR complex, which is central to methanogenesis and reverse methanogenesis. The mcrBDGA cluster shares high (75–88%) amino acid sequence identity with that of the coexisting Methanoperedens spp. genome. This complex is also encoded by a fragment of the Steel Borg. For both the Lilac and the Steel Borgs, the GC content of the region encoding this operon is elevated relative to the average Borg values. Methanoperedens spp. pass electrons from methane oxidation to terminal electron acceptors (Fe3+, NO3− or Mn4+) via MHCs26,27,28. The Lilac Borg genome encodes 16 MHCs with up to 32 haem-binding motifs within one protein. By analogy with experiments showing that cyanophages with a photosystem gene increase host fitness, we suggest that MHC genes may increase the capacity of Methanoperedens spp. to oxidize methane9,29. However, this needs to be tested experimentally. Membrane-bound and extracellular MHC may diversify the range of Methanoperedens spp. extracellular electron acceptors.

Fig. 3: Cell cartoon illustrating capacities inferred to be provided to Methanoperedens spp. by the coexisting Lilac Borg.
figure 3

Like all Borgs, this Borg lacks the capacity for independent existence, and we infer that it replicates within host Methanoperedens spp. cells. Borg-specific proteins are those that were not identified in the genome of coexisting Methanoperedens spp. Borg-encoded capacities are grouped into the major categories of energy metabolism (including the MCR complex involved in methane oxidation), extracellular electron transfer (including MHCs) involved in electron transport to external electron acceptors, central carbon metabolism (including genes that enable production of polyhydroxybutyrate (PHB)) and stress response/defence (including production of compatible solutes). Locus codes are listed in Supplementary Table 7.

Full size image

The Lilac Borg encodes a functional NiFe CODH, but this is fragmented in some genomes. Other genes for the acetyl-CoA decarbonylase–synthase complex are present only in Methanoperedens spp. The CODH is located in proximity to a cytochrome b and cytochrome c, so electrons from CO oxidation could be passed to an extracellular terminal acceptor such as Fe3+ in an energetically downhill reaction. This would allow the removal of toxic CO and may contribute to the formation of a proton gradient that can be harnessed for energy conservation.

The Lilac Borg has a gene resembling the γ-subunit of ethylbenzene dehydrogenase (EBDH), which is involved in transferring electrons liberated from the hydroxylation of ethylbenzene and propylbenzene30. This EBDH-like protein is located extracellularly, and given haem-binding and cohesin domains, it may be involved in electron transfer and attachment.

Although the Lilac Borg lacks genes for a nitrate reductase, it encodes a probable hydroxylamine reductase (Hcp) that may scavenge toxic NO and hydroxylamine byproducts of Methanoperedens spp. nitrate metabolism. As the hcp gene was not identified in coexisting Methanoperedens spp., the Borg gene may protect Methanoperedens spp. from nitrosative stress. Proteins such as H2O2-forming NADH oxidase (Nox) and superoxide dismutase (SOD) may protect against reactive oxygen species. An alkylhydroperoxidase, two probable disulfide reductases and a bacterioferritin all may detoxify the H2O2 byproduct of Nox and SOD. The Lilac Borg also encodes genes that probably augment osmotic stress tolerance. This Borg, but not Methanoperedens spp., provides genes to make Nε-acetyl-β-lysine as an osmolyte. An aspartate aminotransferase links the tricarboxylic acid cycle and amino acid synthesis, producing glutamate that can be used for the production of the osmolyte β-glutamate. More importantly, perhaps, it has recently been established that a bacterial homologue of this single enzyme can produce methane from methylamine31, raising the possibility of methane cycling within the Borg–Methanoperedens spp. system.

The Lilac Borg has three large clusters of genes. The first may be involved in cell wall modification, as it encodes large membrane-integral proteins with up to 17 transmembrane domains, proteins for polysaccharide synthesis, glycosyltransferases and probably carbohydrate-active proteins. The second contains key metabolic valves that connect gluconeogenesis with mannose metabolism for the production of glycans. One gene, encoding fructose 1,6-bisphosphatase (FBP), was not identified in the Methanoperedens spp. genomes and may regulate carbon flow from gluconeogenesis to mannose metabolism. In between these clusters are 12 genes with PEGA domains with similarity to S-layer proteins. Cell-surface proteins, along with these PEGA proteins, account for approximately 13% of all Lilac Borg genes. We conclude that functionalities related to cell wall architecture and modification are key to the effect of these ECEs on their host, perhaps triggering cell wall modification for adaptation to changing environmental conditions (Fig. 3).

Conclusions

Borgs are enigmatic ECEs that can approach (and probably exceed) 1 Mb in length (Extended Data Table 1). We can neither prove that they are archaeal viruses or plasmids or minichromosomes, nor prove that they are not. Although they may ultimately be classified as megaplasmids, they are clearly different from anything that has been previously reported. It is fascinating to ponder their possible evolutionary origins. Borg homologous recombination may indicate movement among hosts, thus their possible roles as gene transfer agents. It has been noted that Methanoperedens spp. have been particularly open to gene acquisition from diverse bacteria and archaea6, and Borgs may have contributed to this. The existence of Borgs encoding MCR demonstrates for the first time (to our knowledge) that MCR and MCR-like proteins for metabolism of methane and short-chain hydrocarbons can exist on ECEs and thus could potentially be dispersed across lineages, as is inferred to have occurred several times over the course of archaeal evolution17,32. Borgs carry numerous metabolic genes, some of which produce variants of Methanoperedens spp. proteins that could have distinct biophysical and biochemical properties. Assuming that these genes either augment Methanoperedens spp. energy metabolism or extend the conditions under which they can function, Borgs may have far-reaching biogeochemical consequences, with important and unanticipated climate implications. Confirmation that Borgs impact the rate of oxidation of methane by Methanoperedens and extend the conditions under which these archaea can function will require experimental evidence. This could be pursued by establishing cultures that include Methanoperedens with and without Borgs and comparison of the methane oxidation rates, with testing performed under a range of geochemical conditions.

Methods

Sampling and creation of metagenomic datasets

We analysed sequences from sediments of an aquifer in Rifle, Colorado, USA, that were retrieved from cores from depths of 5 m and 6 m below the surface15 in July 2011, and cell concentrates from pumped groundwater from the same aquifer collected at a time of elevated O2 concentration in May 2013. Discharge from the Corona Mine, Napa County, California, USA, was sampled in December 2019. Shallow pore water was collected from the riverbed at the East River, Crested Butte, Colorado sampled in August 2016. Soil was sampled from depth intervals between 1 cm and 1 m from a permanently moist wetland located in Lake County, California. Wetland soils were sampled in late October and early November 2017, 2018 and 2019. DNA was extracted from each sample (DNeasy PowerSoil Pro) and submitted for Illumina sequencing (150-bp or 250-bp reads) at the QB3 facility, University of California, Berkeley. Reads were adapter and quality trimmed using BBduk33 and sickle34. Filtered reads were assembled using IDBA-UD35 and MEGAHIT, gene predictions were established using Prodigal36 and USEARCH37 was used for initial annotations34,35,37,38. Functional predictions and predictions of tRNAs followed previously reported methods19.

Genome identification, binning and curation

Hundreds of kilobytes of de novo-assembled sequences were identified to be of interest as potential novel ECEs first based on their taxonomic profile. The taxonomic profiles were determined through a voting scheme in which the taxonomy is assigned at the species to domain level (Bacteria, Archaea, Eukaryotes and no domain) by comparison with a sequence database (protein annotations in the UniProt and ggKbase: https://ggkbase.berkeley.edu/) when the same taxonomic assignment received >50% votes. Assembled sequences selected for further analysis had no taxonomic profile, even at the domain level. The majority of contigs of interest had more genes with similarity to those of archaea of the genus Methanoperedens spp. than to any other genus (see Extended Data Fig. 4). The second feature of interest was dominance by hypothetical proteins yet absence of genes that would indicate identification as phage or viruses or plasmids.

These initially identified large fragments were manually curated to remove scaffolding gaps and local assembly errors, to extend and join contigs with the same profile, GC and coverage, and then to extend the near-complete sequences fully into their long terminal repeats. The last step required reassignment of reads mapped at one end and at double depth to both ends. The fully extended sequences had no unplaced reads extending outwards, despite genome-wide deep coverage. Given this, and the absence of any fragments that could potentially be part of a larger genome, it was concluded that sequences represented linear genomes.

In more detail, our curation method involved mapping of reads to the de novo fragments and extension within gaps and at termini using previously unplaced reads that we added based on overlap or by the relocation of misplaced reads (these could often be identified based on improper paired read distances and/or wrong orientation). Local assembly errors were sought by visualization of the reads mapped throughout the assembly and identified based on imperfect read support, or where a subset of reads was partly discrepant and discrepancies involved sequences that were shared by tandem direct repeats of the same region (that is, the tandem direct repeat regions were collapsed during assembly). De novo-assembled sequences often ended in tandem direct repeat regions because repeats fragment assemblies. To resolve local assembly errors, gaps were inserted and reads relocated to generate the sequence required to fill the gaps. This ensured comprehensive essentially perfect agreement between reads and the final consensus sequence. In some cases, the tandem direct repeat regions had greater than the expected depth of mapped reads and no reads spanned the flanking unique sequences. In these cases, the repeat number was approximated to achieve the expected read depth, but some arrays may be larger than shown. GC skew and cumulative GC skew were calculated using iRep39 for the fully manually curated complete genomes, and the patterns were used to identify the origins and terminus of replication. The pattern of use of coding strands for genes (predicted in Bacterial Code 11) was compared with these origin and terminus predictions to resolve genome organization. The curated sequences were searched for perfect repeats of lengths of 50 or more nucleotides using Repeat Finder in Geneious. When repeat sequences overlapped, the unit of direct repeat was identified and the length of that repeat, number of repeats, location (within gene versus intergenic) and genome position were tabulated. Once the features characteristic of the ECEs of interest had been determined, we sought related elements. Sequences of interest were identified based on (1) credible partial alignment with the complete sequences, (2) no domain-level profile, (3) GC content of 30–35%, (4) regions with three or more direct tandem repeats scattered throughout the genome fragment, and (5) more best hits to Methanoperedens spp. proteins than to proteins from any other organisms. If scaffolds met criterion (1) they were immediately classified as targets. If they met most or all of the other criteria and had similar coverage values, they were binned together with other scaffolds from the same sample with these features. Often, ends of some of the contigs in the same bin overlapped perfectly and could be joined, increasing confidence in the bin quality. Genome sequences were aligned to each other using Mauve40. Where anomalously high (perfect) sequence identity suggestive of recent recombination was detected between Borgs, reads mapped to the region were visualized to verify that the assembly was correct (that is, not chimeric; also see information in the Extended Data).

Genome fragments were phylogenetically profiled to establish relatedness to sequences in public databases. Sequences were classified as having no detectable hit if the protein had no similar database sequence with an E < 0.0001.

Correlation analyses

Reads from each sample were aligned to each Methanoperedens and Borg genome. Alignments were performed using bbmap41 using the following parameters: editfilter = 5, minid = 0.96, idfilter = 0.97, ambiguous = random. The number of reads aligning to each genome was then parsed into a matrix and the correlation between abundance patterns for Methanoperedens and Borg genomes was then calculated using Pearson correlation metric as implemented in scipy42. Correlation between a Methanoperedens genome and a Borg genome was deemed significant if the Pearson correlation between the two genomes was higher than 0.92. The code used for this analysis is available through Zenodo (https://doi.org/10.5281/zenodo.6887003).

CRISPR–Cas analysis

Borg and Methanoperedens-encoded CRISPR repeats and spacers were identified using CRISPRDetect43. The coding sequences from this study were searched against Cas gene sequences reported from previous studies44 using hmmsearch with E < 1 × 10−5 to identify the full locus. Matches were checked using a combination of hmmscan and BLAST searches against the NCBI nr database and manually verified by identifying colocated CRISPR arrays and Cas genes. Spacers extracted from between repeats of the CRISPR locus were compared with sequence assemblies from the sites where Borgs were identified using BLASTN-short 45. Matches with alignment length of more than 24 bp and 1 or less mismatch were retained and targets were classified as bacteria, phage or other. CRISPR arrays that had 1 or less mismatch, were further searched for more spacer matches in the target sequence by finding more hits with three or less mismatches.

Protein and gene content analysis

After the identification and curation of Borg genomes and accumulation of usearch annotations for coding sequences, functional annotations were further assigned by searching against PFAM r32, KEGG, pVOG. Transmembrane regions in proteins were predicted with TMHMM. All Methanoperedens genomes and genome assemblies, as well as 1,153 archaeal viruses and ECEs were downloaded from the NCBI RefSeq database. Open reading frames were predicted using Prodigal, and all proteins from Borg genomes and the reconstructed ECE database were clustered into protein families and compared across genomes as previously described19. In brief, the coding sequences were clustered into families using a two-step procedure; first an all-versus-all sequence search was performed using an E value cut-off of 1 × 10−3, sensitivity of 7.5 and coverage of 0.5, and a sequence similarity network was built on the basis of the pairwise similarities and the greedy set cover algorithm to define protein subclusters. The resulting subclusters were grouped into protein families using a comparison of hidden Markov models. For subfamilies with probability scores of at least 95% and coverage at least 0.50, a similarity score (probability × coverage) was used as weight of the input network in the final clustering using the Markov clustering algorithm, with 2.0 as the inflation parameter. These clusters were defined as the protein families.

Functional annotation

Genes of interest were further verified and compared using the conserved domain search in NCBI and InterproScan46 to identify conserved motifs within the amino acid sequence. MHCs were identified based on three or more CxxCH motifs within one gene. The cellular localization of proteins was predicted with Psort (v3.0.3) using archaea as the organism type. Proteins were compared using blastp and aligned using MAFFT47 v.7.407 to visualize homologous regions and check conserved amino acid residues that constitute the active site or are required for cofactor and ligand binding.

Phylogenetic trees

For each gene, references were compiled by BLASTing the corresponding gene against the NCBI nr database, and their top 50 hits clustered by CD-HIT using a 90% similarity threshold48. The final set of genes was aligned using MAFFT v.7.407, and a phylogenetic tree was inferred using IQTREE v.1.6.6 using automatic model selection49 and visualized using iTOL50. Synteny plots were generated using Mauve51 and gene clusters through Adobe Illustrator and gggenes.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.