Recoding of stop codons expands the metabolic potential of two novel Asgardarchaeota lineages

Asgardarchaeota have been proposed as the closest living relatives to eukaryotes, and a total of 72 metagenome-assembled genomes (MAGs) representing six primary lineages in this archaeal phylum have thus far been described. These organisms are predicted to be fermentative heterotrophs contributing to carbon cycling in sediment ecosystems. Here, we double the genomic catalogue of Asgardarchaeota by obtaining 71 MAGs from a range of habitats around the globe, including the deep subsurface, brackish shallow lakes, and geothermal spring sediments. Phylogenomic inferences followed by taxonomic rank normalisation confirmed previously established Asgardarchaeota classes and revealed four additional lineages, two of which were consistently recovered as monophyletic classes. We therefore propose the names Candidatus Sifarchaeia class nov. and Ca. Jordarchaeia class nov., derived from the gods Sif and Jord in Norse mythology. Metabolic inference suggests that both classes represent hetero-organotrophic acetogens, which also have the ability to utilise methyl groups such as methylated amines, with acetate as the probable end product in remnants of a methanogen-derived core metabolism. This inferred mode of energy conservation is predicted to be enhanced by genetic code expansions, i.e., stop codon recoding, allowing the incorporation of the rare 21st and 22nd amino acids selenocysteine (Sec) and pyrrolysine (Pyl). We found Sec recoding in Jordarchaeia and all other Asgardarchaeota classes, which likely benefit from increased catalytic activities of Sec-containing enzymes. Pyl recoding, on the other hand, is restricted to Sifarchaeia in the Asgardarchaeota, making it the first reported non-methanogenic archaeal lineage with an inferred complete Pyl machinery, likely providing members of this class with an efficient mechanism for methylamine utilisation. Furthermore, we identified enzymes for the biosynthesis of ester-type lipids, characteristic of bacteria and eukaryotes, in both newly described classes, supporting the hypothesis that mixed ether-ester lipids are a shared feature among Asgardarchaeota.


MAIN
The recently described Asgard archaea have been proposed as the closest living prokaryotic relatives to eukaryotes, supporting a two-domain tree of life 1,2 . Six Asgard lineages have been described (although see note added in proof), all of which are named after Norse gods; Lokiarchaeota, Thorarchaeota, Odinarchaeota, Heimdallarchaeota, Helarchaeota, and the recently proposed Gerdarchaeota 1, [3][4][5] . Asgard archaea were introduced as a superphylum 1 . However, a subsequent reclassification, based on taxonomic rank normalisation using relative evolutionary divergence (RED), indicates that this lineage is a phylum for which the name Asgardarchaeota was proposed together with the classes Lokiarchaeia, Thorarchaeia, and Heimdallarchaeia as Latin placeholder names until nomenclature types are designated 6,7 .
The inferred eukaryotic-like nature of the Asgardarchaeota, in particular the encoded plethora of eukaryotic signature proteins (ESPs), spurred initial speculations about possible eukaryotic contamination of the recovered metagenome-assembled genomes (MAGs) 8 . However, these arguments have since been refuted by analysing additional MAGs 9 , and long-read sequencing technologies yielding near-complete MAGs have confirmed that eukaryote-like features are integral to Asgardarchaeota genomes 10 . Furthermore, a recent, decade-long cultivation effort resulted in the first Asgardarchaeota co-culture, Candidatus Prometheoarchaeum syntrophicum strain MK-D1, a Lokiarchaeia representative from deep-sea sediments 11 . The authors obtained a closed genome encoding 80 ESPs and presented evidence for the transcription of these genes, supporting not only that Asgardarchaeota genomes are not chimeric assembly artefacts, but also that ESP genes are actively expressed by these archaea.
Insights into the metabolism of Asgardarchaeota based on functions inferred from MAGs, transcriptomics, and experimental data from the Lokiarchaeia culture indicate that members of this phylum are mostly anaerobic fermentative heterotrophs 5,11,12 , although at least one lineage has the potential for short-chain hydrocarbon oxidation 4 . In addition, some Heimdallarchaeia seem to have acquired oxygen-dependent pathways in their recent evolutionary history and were inferred to reduce oxygen or nitrate 13 . Heimdallarchaeia, Thorarchaeia and Lokiarchaeia encode the complete archaeal Wood-Ljungdahl pathway 14 , which could function as an electron sink or operate in reverse to oxidise organic substrates 12 . It was further hypothesised that cofactors reduced by Asgardarchaeota during organic carbon oxidation may be reoxidized by fermentative hydrogen production to fuel a syntrophic relationship with hydrogen-or formate-consuming organisms 12 . Study of Lokiarchaeia co-cultures containing Ca. P. syntrophicum MK-D1 confirmed several of these inferred functions. In particular, this archaeon uses small peptides and amino acids while growing syntrophically with a methanogen or a bacterial sulphate reducer through interspecies hydrogen and possibly also formate transfer 11 .
Despite this recent focus on Asgardarchaeota, we have likely only explored a small fraction of the diversity encompassed by this phylum. Microbial community profiling based on small subunit (SSU) rRNA gene sequences suggest that many novel Asgardarchaeota lineages are awaiting genomic discovery 5,14,15 . Here we describe 46 Asgardarchaeota MAGs obtained from coastal, hot spring and deep-sea sediments complemented by 25 MAGs extracted from public metagenomic datasets. This improved genomic sampling enabled us to resolve phylogenomic relationships, extend the rank normalisation analysis, and to propose two new classes, Ca. Sifarchaeia (see note added in proof) and Ca. Jordarchaeia, both named after Norse Gods. Based on metabolic reconstruction we infer both lineages to be heteroorganotrophic acetogens which make use of genetic recodings to enhance their metabolic capabilities, including the first reported complete archaeal pyrrolysine machinery outside of methanogens.

RESULTS AND DISCUSSION
Sampling sites and community profiling An in silico SSU rRNA gene survey, based on SILVA (r132) 16 revealed 99 sites around the globe, predominantly from anoxic marine and freshwater sediments, as potential Asgardarchaeota habitats for metagenomic recovery (Fig S1). Subsequently, we collected samples for shotgun sequencing from sites in Queensland, Australia, with similar characteristics and based on SSU rRNA gene screening discovered Asgardarchaeota at relative abundances of up to 2.7% in anoxic sediments from two brackish lakes at the Sunshine Coast (Fig. S2a-c). We extended our search to deep sea sediments and detected Asgardarchaeota in anoxic cores from the Hikurangi Subduction Margin of the Pacific Ocean with relative abundances reaching 11.6% in core segments 1.5-634.7 m below the seafloor (mbsf), with the highest abundances reported for depths >100 mbsf (Fig. S2d, e). Additionally, we identified two hot spring sediments, from Mammoth Phylogeny and rank normalised taxonomy of Asgardarchaeota. A maximum-likelihood inference was performed using IQ-TREE under the LG + C10 + F + G + PMSF model, based on a multiple-sequence alignment of up to 122 protein markers (subsampled to 42 amino acids per marker) for 143 Asgardarchaeota MAGs, and 1637 archaeal representatives of non-Asgard lineages in GTDB release r95 (1780 taxa, 5124 sites). The tree was rooted on the Undinarchaeota. Branches with bootstrap support >0.9 are indicated by purple dots. Habitat information for Asgardarchaeota MAGs recovered in this study is shown with black and white symbols at the branch tips corresponding to the symbols in the figure legend. The two outer layers indicate the presence of inferred selenocysteine (Sec, orange) and pyrrolysine (Pyl, blue) encoding systems in each MAG: colour-filled circles represent a complete Sec/Pyl-encoding system, i.e. genes required for the Sec/Pyl biosynthesis and insertion, and the corresponding tRNA; semicircles represent a partial set of detected genes; empty circles indicate the presence of only a tRNAsec or tRNApyl. Asgardarchaeota classes are highlighted in different colours: Bright cyan -Sifarchaeia; dark yellow -Jordarchaeia; Light pink -Thorarchaeia; orange -Lokiarchaeia; Sky blue -Heimdallarchaeia; Light yellow -Odinarchaeia; Asgard hot vent group (AHVG) and Asgard Lake Cootharaba group (ALCG) -black with bold nodes. Note, that the previously proposed lineages 'Gerdarchaeota' and 'Helarchaeota' are represented by GTDB the order-level lineages Helarchaeales and JABLTI01, respectively. Each of the arrows represents functions assigned to predicted proteins encoded in the respective genomes (Sifarchaeia, 4 MAGs, max. completeness 94.1%; Jordarchaeia, 5 MAGs, max. completeness 95.0%). Pie charts next to the arrows/enzymes indicate the proportion of MAGs that encode a certain enzyme (see Tables S4-S10 for more information). A black arrow indicates that this enzymatic step is encoded by at least one MAG of a given class (Sifarchaeia or Jordarchaeia), and a solid grey arrow indicates an enzymatic step that is missing from a given class. The reactions indicated by black dashed lines are partially encoded and/ or are tentative. The dashed red lines represent potential interspecies electron transfer. Note that Sifarchaeia and Jordarchaeia genomes encode an acetyl-CoA synthetase (acs) which preferentially acts in the direction of acetate and ATP production. Abbreviations: APS-adenylyl sulphate, PAPS -3′-Phosphoadenylyl sulphate, PEP-phosphoenolpyruvate, H4F-tetrahydrofolate, H4MPT-tetrahydromethanopterin, DMAdimethylamine, MMA-trimethylamine, LPS-lipopolysaccharide, CoM -coenzyme M, Oxa-oxaloacetate, Cit -isocitrate, 2-Oxo -2-oxo-glutarate, Suc-CoA -succinyl-coenzyme A, Suc -succinate, Fum -fumarate, Mal -malate, ETF -electron transfer flavoprotein, ETFQO -ETF-ubiquinone oxidoreductase, MQ -menaquinone, Fd -ferredoxin, Pi -inorganic phosphate, PPi -inorganic pyrophosphate, red -redox, rTCA -reverse tricarboxylic acid cycle, WLP -Wood-Ljungdahl pathway.
Genome recovery, phylogenomics and taxonomic rank normalisation Metagenomic analysis of the selected lake, deep-sea and hot spring sediments yielded a set of 46 Asgardarchaeota MAGs, which were supplemented with 25 MAGs recovered from the NCBI Sequence Read Archive (SRA) (Table S1). Overall, the 71 MAGs have an average estimated completeness of 78.7 ± 15.3% with an estimated contamination of 3.8 ± 2.3% (Table S1). The GC content ranged from 28.8 to 48.4%, and the average genome size was estimated to be~4 Mbp (Table S1 and Fig. S3).
We inferred evolutionary relationships via maximum-likelihood and Bayesian trees (Table S2) from trimmed multiple-sequence alignments of 122 and 53 archaeal single-copy marker proteins, respectively 17,18 . Our phylogeny was further evaluated by inferring trees from (1) alignments post removal of compositionally biased sites to increase tree accuracy for distantly related sequences, and (2) alignments of alternative concatenated marker sets including 16 ribosomal proteins (rp1) 19 and 23 ribosomal proteins (rp2) 20 . All phylogenomic inferences of our extended dataset confirmed the monophyly of previously proposed Asgardarchaeota lineages and recovered four novel lineages within this phylum ( Fig. 1 and S4-11). Next, we applied the taxonomic rank normalisation approach implemented in the Genome Taxonomy Database (GTDB) 6,7 to assign ranks to Asgardarchaeota lineages. Our results support the rank of class for Thorarchaeia, Odinarchaeia, Heimdallarchaeia and Lokiarchaeia ( Fig. 1 and Table S3). The previously proposed phyla "Helarchaeota" and "Gerdarchaeota" were robustly recovered within the classes Lokiarchaeia and Heimdallarchaeia, and represent the GTDB order-level lineages Helarchaeales and JABLTI01, respectively ( Fig. 1 and Table S3). Two of the novel lineages found in the present study comprising 4 and 5 MAGs, were robustly recovered in all phylogenies ( Fig. 1 and S4-S11) and were assigned the rank of class based on their RED values and independence from other classes within Asgardarchaeota. A pangenomic analysis based on protein clusters further supported considerable differences between the novel and existing classes (Fig. S12). We propose the names Ca. Sifarchaeia class nov. (see note added in proof) and Ca. Jordarchaeia class nov., derived from the gods Sif and Jord in Norse mythology. For simplicity, these candidate classes will be referred to as Sifarchaeia and Jordarchaeia in the remainder of the manuscript. We designated type genomes 21 in both lineages (see proposal of type material) and provide a detailed metabolic reconstruction for both classes below. The phylogenetic placement of the two remaining novel lineages, comprising only two MAGs each, from lake and subsurface sediments, respectively (Table S1 and Fig. S2), was not consistent among trees inferred from different models and marker sets (Fig. S4-11), and we therefore assign them the placeholder names, Asgard hot vent group (AHVG) and Asgard Lake Cootharaba group (ALCG). We foresee that the phylogeny of both lineages will be resolved as more Asgardarchaeota genomes become available.
To evaluate the placement of Sifarchaeia and Jordarchaeia with regard to eukaryotes, we inferred a tree based on 15 markers conserved in the Archaea and eukaryotes 22 . This inference confirmed previous results by placing Heimdallarchaeia as a sister group to Eukarya within Asgardarchaeota, whereas Sifarchaeia and Jordarchaeia clustered with the remaining lineages in this phylum (Fig. S13). The detection of numerous eukaryotic signature proteins (ESPs) in Sifarchaeia and Jordarchaeia ( Fig. S14 and Table S4) further supports a close relationship between Asgardarchaeota and eukaryotic organisms. However, the patchy distribution of ESPs in these and other Asgardarchaeota lineages (Fig. S14), and the observed lack of organelle-like structures in the Lokiarchaeia culture 11 , suggests that the ESPs encoded in extant Asgardarchaeota are reminiscent of genes present in the last Asgard archaeal common ancestor (LAsCA) and are likely to perform different functions than their eukaryotic homologues.
Core metabolism and electron transport Based on metabolic inference, we propose that Sifarchaeia and Jordarchaeia are hetero-organotrophs ( Fig. 2 and Table S5-S10). This lifestyle is similar to the predicted metabolism of the cultured Lokiarchaeum Ca. P. syntrophicum MK-D1, where short-chain fatty acids including acetate are produced via central metabolic pathways 11 . However, unlike MK-D1 which produces these short-chain fatty acids via the fermentation of amino acids, Sifarchaeia and Jordarchaeia appear to be mostly restricted to oxidising fatty acids or lactate to acetate (Fig. 2).
Fatty acids are likely to be utilised via the canonical β-oxidation pathway predicted in both lineages with acetate and ATP generated through acetyl-CoA synthases (Acd) (Fig. 2 and Suppl. Text). The ability to oxidise fatty acids is common in Archaea such as Archaeoglobus 23 , has been suggested for Asgardarchaeota lineages previously 5,12 , and was recently predicted in an alkane oxidising lineage 4 . Electrons derived from oxidising these fatty acids could establish a membrane potential since Sifarchaeia and Jordarchaeia encode genes for complex I (dehydrogenase) and complex V (ATP synthase) of the electron transport chain (Fig. 2). Notably, complex I lacks the reduced cofactor oxidising subunits NuoEFG, which form the NADH dehydrogenase module. Therefore, we hypothesise that energy conservation in Sifarchaeia and Jordarchaeia depends on electron transfer by reduced ferredoxin (Fig. 2), similar to the membrane-bound fpo-like complex of the acetoclastic methanogens 24 . Electrons from the Nuo complex could be transferred to menaquinone, since Sifarchaeia and Jordarchaeia encode a near-complete biosynthesis pathway for this quinone (Table S9 and Suppl. Text), and subsequently to an unidentified terminal electron acceptor, or alternatively to a membrane-bound hydrogenase for H 2 generation. The latter has been proposed for syntrophic microorganisms 25 including the cultured Lokiarchaeia strain MK-D1 11 . However, Sifarchaeia and Jordarchaeia likely use a divergent mechanism since they do not encode the H 2 producing electron transfer complex FixABCD-HdrABC (Suppl. Text) identified in MK-D1 and other Asgardarchaeota 11 .
Acetate may also be generated from D-lactate by the encoded putative D-lactate dehydrogenase (Dld) via pyruvate oxidoreductase (PorABCD) and acetyl-CoA synthases (Acd) (Figs. 2 and S15). The presence of up to 9 and 13 copies of Dld genes in Sifarchaeia and Jordarchaeia MAGs, respectively, suggest that Dlactate oxidation is important in their metabolism (Table S9). Furthermore, most of these Dld genes are collocated with a heterodisulfide reductase (Hdr) subunit D complex (Fig S15a) that would allow electrons, generated from the lactate oxidation, to reduce coenzyme M (CoM) -coenzyme B (CoB) (Fig. 2 and Suppl. Text). Then a hydrogen evolving NiFe hydrogenase Hdr-Mvh would facilitate the reoxidation of the predicted CoM-CoB heterodisulfide by generating H 2 and oxidised ferredoxin 11 encoded in Sifarchaeia and Jordarchaeia MAGs. Alternatively, in Sifarchaeia, both coenzymes might be re-oxidised by the encoded thiol:fumarate reductase which catalyses the reduction of fumarate, with CoB and CoM as electron donors, to succinate and heterodisulfide CoM 26 . The hydrogen produced by this electron-confurcating NiFe hydrogenase Hdr-Mvh complex (Fig. 2), as proposed for strain MK-D1 11 , is likely utilised by Sifarchaeia and Jordarchaeia for indirect interspecies electron transfer. Similarly, both lineages might be able to transfer electrons via formate, catalysed by the encoded formate dehydrogenase (Fig. 2), to syntrophic partners as predicted to occur in the MK-D1 enrichment culture 11,26 . Such a symbiotic relationship would also complement the amino acid and vitamin needs of Sifarchaeia and Jordarchaeia, which lack genes encoding the biosynthesis of the amino acids proline, tyrosine and phenylalanine, and additionally alanine biosynthesis genes are missing in Jordarchaeia (Table S9). Vitamin biosynthesis genes not detected in both novel lineages include biotin, and peridoxin (Table S9).
While these organic acids appear to be key in the metabolism of these novel lineages, members of both classes encode enzymes catalysing the transfer of methyl groups, such as methylated amines, to CoM, similar to a pathway previously reported for methylotrophic methanogens 27 . However, Sifarchaeia and Jordarchaeia are missing genes for methyl-CoM reductase (Mcr), the enzyme catalysing the final step in methane formation. Instead, both novel lineages encode two catalytic subunits of a putative tetrahydromethanopterin (H 4 MPT) coenzyme-M methyltransferase (MtrAH). This predicted two-subunit enzyme differs from the eight-subunit membrane-associated complex in methanogens (Suppl. Text) and has also been reported in Methanomassiliicoccales 28 . The authors of this study proposed that mtrAH encodes a H 4 F/H 4 MPT-CoM methyltransferase in these hydrogen dependent methylotrophic methanogens.
Similarly, Sifarchaeia and Jordarchaeia could use this enzyme to catalyse the reverse reaction to facilitate the transfer of methyl groups from methyl-CoM to methyl-H 4 MPT, and subsequently to acetyl-coenzyme A (CoA) to be reduced to acetate for energy conservation (Fig. 2), although the bioenergetics of this potential reaction remain unclear. The putative H 4 MPT-CoM methyltransferase may also oxidise the methyl groups via the reverse archaeal Wood-Ljungdahl pathway (WLP; H 4 MPT-dependent). Alternatively, the WLP could function in the opposite direction to autotrophically fix carbon dioxide using hydrogen as an electron donor, however we did not detect genes of uptake hydrogenases, i.e., Sifarchaeia and Jordarchaeia lack genes for group 1 NiFehydrogenases (Table S10).
Besides the possibility for utilising methylamines, genomes of both novel classes encode enzymes to break down complex carbohydrates Fig. 3 Selenocysteine recoding in Asgardarchaeota. a Proposed mechanism of selenocysteine biosynthesis in Asgardarchaeota. b Proposed selenocysteine insertion in Asgardarchaeota. Presence of genes in each class are indicated with coloured circles: filled circlehalf or more of the MAGs encode the given gene; half-filled circleless than half of the MAGs encode the given gene. Note that the eukaryotic SECIS-binding protein 2 (SBP2) is missing from all Archaea (indicated by a question mark). c tRNAsec in Sifarchaeia MAG "lw60_2018_gm2_56" and d tRNAsec in Jordarchaeia MAG "LHC_bin_1308". Highlighted are the acceptor arm (red), D arm (yellow), anticodon arm (green), variable arm (blue) and the T arm (purple). e Phylogenetic tree of selB, inferred with IQ-TREE (PMSF C10 model) from a TrimAl-trimmed alignment of SelB genes from archaeal, bacterial and eukaryotic genomes. Tree was rooted between the bacterial and archaeal-eukaryotic clade. Asgardarchaeota sequences are highlighted with different colour labels: bright cyan -Sifarchaeia; dark yellow -Jordarchaeia; light pink -Thorarchaeia and orange -Lokiarchaeia.
via glycoside hydrolases including β-galactosidase and α-amylase, and carbohydrate esterases (Table S8). The resulting glucose could be utilised via the Embden-Meyerhof-Parnas (EMP) pathway to generate pyruvate, for subsequent oxidation to acetate, or to be metabolised by the encoded partial reverse TCA cycle (Fig. 2).

Mixed membrane lipids and the great lipid divide
Both Sifarchaeia and Jordarchaeia encode all genes for the synthesis of archaeal ether-type lipids, but in addition, Sifarchaeia encode enzymes for the biosynthesis of ester-type lipids, characteristic of Bacteria and Eukarya (Fig. S16). This finding aligns with previous reports of ester lipid biosynthetic pathways in Asgard lineages 29 , supporting the hypothesis that mixed ether-ester lipids are a shared feature among Asgardarchaeota. Subsequently, this trait could have been lost in some subordinate lineages, including Jordarchaeia (Fig.  S16). Phylogenetic inference of a key ester-type lipid gene supports the finding that archaeal homologues are distinct from their bacterial counterparts 30 and showed some Lokiarchaeia genes clustering with eukaryotic homologues, albeit with low support values (Fig. S17). The great lipid divide between bacteria and archaea has been further eroded by the discovery of ester-type lipid genes in members of the Poseidoniales (Marine Group II archaea) 31 , and functional validation of ether-type lipid genes in the Fibrobacteres-Chlorobi-Bacteroidetes (FCB) superphylum 32 . This suggests, together with the reported extensive interdomain horizontal gene transfer of several membrane lipid biosynthesis genes 30 , that the lipid divide thought to distinguish the domains of life is more permeable than previously thought.

Transporters
Sifarchaeia and Jordarchaeia encode several ABC transporters for the uptake of essential trace compounds, including tungstate 33 , which has been shown to enhance the growth of methanogens 34 and could provide a similar benefit to both classes (Fig. 2, S18 and Table S9). In addition, Sifarchaeia possess a low-affinity inorganic phosphate transporter that also functions as a major uptake system for arsenate 35 . To mitigate the toxicity of arsenate, both classes may be able to actively expel arsenate from their cells by reducing it to the less toxic arsenite 36 , which can then be pumped out of the cell by the ATP-consuming arsenite exporter (Fig. 2).

Expanding metabolic capabilities by recoding of stop codons
Based on inferred proteins and codon usage we predict that Sifarchaeia and Jordarchaeia increase their amino acid synthesis repertoire and consequently their metabolic potential through localised recoding strategies. These include the recoding of the stop codons opal (UGA) and amber (UAG) to incorporate the rare 21st and 22nd amino acids, selenocysteine (Sec) and pyrrolysine (Pyl), respectively, through distinct recoding processes. Both novel classes encode the archaeal/eukaryotic-type Sec biosynthesis machinery (Fig. 3). We also detected a single selenocysteine t-RNA (tRNAsec) in Sifarchaeia and Jordarchaeia MAGs (Fig. 3c, d) and confirmed previous reports of this tRNA in Lokiarchaeia and Thorarchaeia 37,38 , but did not identify a tRNAsec in Heimdallarchaeia or Odinarchaeia (Table S11). Remarkably, the tRNAsec in all Sifarchaeia and some Lokiarchaeia had unusual insertions and deletions, negating previously proposed domainspecific characteristics. For example, the Sifarchaeia tRNAsec has a short 6 bp D-stem ( Fig. 3 and S19a), a feature that has been attributed to eukaryotes and bacteria, whereas archaeal tRNAsec were thought to generally possess a 7 bp D-stem 39 . Our tRNAsec phylogeny recovered most recoded Asgardarchaeota lineages as monophyletic groups clustering with methanogens and eukaryotes, albeit with low bootstrap support values likely due to the short alignment length (Fig. S19b). The recovery of monophyletic tRNAsec groups that match the species tree suggest that horizontal gene transfers (HGTs) may not be common in the evolutionary history of tRNAsec, despite the reported frequent and extensive gene duplication of tRNAs in general 40 .
Furthermore, Sifarchaeia and Jordarchaeia, as well as Lokiarchaeia and Thorarchaeia, encode enzymes to correctly charge this tRNA in order to synthesise a functional selenocysteine tRNAsec (Sec-tRNAsec) using the archaeal/eukaryotic-type Sec biosynthesis pathway. This process involves an initial mischarging of tRNAsec with serine by seryl-tRNA synthetase, then phosphorylation by phosphoseryl-tRNA kinase and conversion into a functional tRNAsec by Sec synthase using selenophosphate formed by selenophosphate synthetase (SPS) from selenium (Fig. 3a) 41 . We found no evidence for the presence of a bacterial-type Sec biosynthesis pathway in Asgardarchaeota, despite previous reports of a bacterial Sec synthase (SelA) (Fig. 3a) in Thorarchaeia MAG SMTZ1-83 38 . Instead, we suggest that the contig harbouring SelA in this MAG is likely bacterial contamination (Table S12), leading us to posit that Asgardarchaeota rely solely on selenophosphate-dependent synthesis of Sec-tRNASec (Fig. 3a).
Sec insertion in Sifarchaeia and Jordarchaeia and all other Sec recoded Asgardarchaeota lineages could be mediated by the Secspecific elongation factor (SelB), which connects the selenocysteine insertion sequences (SECIS), an RNA element that forms a stem-loop structure during Sec insertion ( Fig. 3b and S20), to the ribosome with the help of the SECIS-binding protein 2 (SBP2). Phylogenetic analysis of SelB and SPS supports a predominantly vertical inheritance of both genes and a separation of bacterial and archaeal/ eukaryotic orthologs ( Fig. 3e and S21-S22). Within the archaeal/ eukaryotic branch, the genus Methanopyrus was identified as the deepest branching lineage in both trees, and Asgardarchaeota formed a monophyletic sister group to eukaryotes, although with low bootstrap support. We did not detect SBP2 homologues in Asgardarchaeota, consistent with previous reports that Archaea do not encode this elongation factor, and implying that this key enzyme evolved after eukaryogenesis 37,41 . We found 12-25 and 19-25 predicted SECIS elements (the site where Sec insertion occurs) in Sifarchaeia and Jordarchaeia MAGs to facilitate synthesis of two and three selenoproteins in Sifarchaeia and Jordarchaeia repestively (Table S13). The detected selenoproteins were located 30-500 bases upstream of the corresponding SECIS element (Fig. S20a), a distance range previously observed in Archaea and Eukarya 37,42 . All three selenoproteins detected in Sifarchaeia and Jordarchaeia, a heterodisulfide reductase (Hdr) subunit A, a peroxiredoxin (Prx), and a F420-non-reducing hydrogenase iron-sulfur (Mvh) subunit D, are also present in Lokiarchaeia 37 . This suggests that selenoproteins are common to all Asgardarchaeota, which likely depend on the increased catalytic activity of Sec-containing proteins, such as HdrA, as part of their energy conservation strategies (Fig. S2). Indeed, it has been experimentally verified that selenoproteins can provide up to a hundred times increased catalytic activity over cysteine, its sulphur-containing analogue 43 . Furthermore, the selective advantage of selenoenzymes is not restricted to increased efficiency but may also include the ability to function on a broader range of substrates 44 , and under challenging conditions such as oxidative stress 45 . In the case of Sifarchaeia, Jordarchaeia, and Lokiarchaeia, the Sec-containing protein in the encoded HdrABC-MvhADG-NiFe-hydrogenase complex (Fig. 2) may increase the efficiency of this H 2 evolving electronconfurcating enzyme complex. Further support for a Secenhanced metabolism among Asgardarchaeota are sulphate permeases (SulP), encoded in three Sifarchaeia and several Lokiarchaeia genomes, and predicted to import sulphate and related oxyanions such as selenate, the oxidised form of selenium [46][47][48] . Subsequently, selenate can be reduced and incorporated into proteins during translation as selenocysteine 49 .
The first non-methanogenic archaeal Pyl recoding We detected a second recoding solely present in Sifarchaeia which affects the amber (UAG) stop codon and could allow this class to use the rare 22nd amino acid pyrrolysine (Pyl). The presence of a Pyl tRNA, all required Pyl biosynthesis genes, and specific Pylencoded proteins suggests that this recoding provides Sifarchaeia with an efficient mechanism for methylamine utilisation, despite an unusually high UAG stop codon usage.
Sifarchaeia encode a complete Pyl encoding system including all three Pyl biosynthesis proteins (PylB, PylC, PylD) and a pyrrolysyl-tRNA synthetase (PylS) to charge the pyrrolysine tRNA (tRNApyl, pylT) (Fig. 4a) 50 . Unlike selenocysteine (Fig. 3b), no specific proteins or insertion sequences are required for the tRNApyl insertion, which has been proposed to directly compete with the translation termination release factor for UAG codons (Fig. 4b) 51 . While Pyl genes in Archaea usually form an uninterrupted pylTSBCD cluster, Sifarchaeia show a pattern similar to Methanohalobium evestigatum 52 , in which the pylS gene is~6 Kb distant from pylBCD, separated by a NAD kinase and several hypothetical proteins (Fig. 4c). The tRNApyl of Sifarchaeia, encoded by pylT (Fig. S23), is located upstream of pylS and displays a classic cloverleaf secondary structure with an unusual acceptor stem tail that discriminates the Sifarchaeia tRNApyl from the CCA tails of previously reported archaeal and bacterial homologues (Fig. 4d) 53 . Remarkably, the reported low usage (<6%) of the UAG stop codon in Pyl-containing Archaea 51,54 does not apply to Sifarchaeia. Instead 21% of their CDSs are terminating with UAG ( Fig. 4e and Table S14), a percentage corresponding to UAG frequencies of Pyl-encoding bacteria 55 . How a misspecification of Pyl-tRNApyl to the frequent UAG stop sense codons is avoided remains unknown, although possible mechanisms exist (see below).
Pyl recoding has only been reported previously in archaeal methanogens belonging to the phyla Thermoplasmatota and Halobacteriota, the class Methanomethylicia (Verstraetearchaeota, sensu NCBI taxonomy), and from the candidate lineage Persephonarchaea MSBL1 54,56,57 . Thereby, experimental validations of Pyl synthesis have been focused on the genus Methanosarcina (Halobacteriota) [58][59][60] . Several bacterial phyla, including Firmicutes and Desulfobacterota, also possess Pyl recoding thought to be acquired from Archaea via multiple HGTs 61 , but this recoding is absent in eukaryotes 53 . In addition to Sifarchaeia, we identified PylSBCD genes for the first time in Hydrothermarchaeota and Bathyarchaeia representatives by screening GTDB genomes 7 , and most gene phylogenies support a novel cluster containing both representatives, together with Methanomethylicia, and Sifarchaeia ( Fig. 4f and S24-S28).
The major role of Pyl recoding in Archaea, methanogenic and non-methanogenic alike, is methylamine utilisation, since Pyl is foremost incorporated in the active sites of methyltransferases 54,62 . Indeed, Sifarchaeia encode several methyltransferases, including monomethylamine methyltransferase (MtmB) and dimethylamine methyltransferase (MtbB), with the latter possessing a Pyl recoding, making it the only in-frame UAG stop codon in Sifarchaeia (Table S15 and Fig. 4c). Thereby, MtbB, together with Methylcobamide:CoM methyltransferase (MtbA), could methylate the cognate corrinoid protein (MtbC), which in turn methylates coenzyme M (CoM) 52 . This cascade of encoded methyl transfers could allow Sifarchaeia to convert methyl groups directly to acetate for energy conservation (see above). Hence, maintaining the Pyl-recoding seems essential, since MtbB requires Pyl, which was hypothesised to activate and orient methylamines as substrates for the corrinoid protein MtbC 52 . How Sifarchaeia, with their high percentage of UAG stop codons, control the specificity of Pyl insertions versus protein termination remains to be determined, however, it has been suggested that environmental conditions such as the presence of methylamines could selectively activate Pyl biosynthesis 51 . Indeed, the Firmicute Acetohalobium arabaticuma was recently found to expand its genetic code to include Pyl only in the presence of trimethylamines (TMA), but to down-regulate the transcription of the entire Pyl operon when TMA was absent 55 .

Recoding evolutionary history in Asgardarchaeota
While the evolutionary history of Pyl encoding is still debated, a structure-based phylogeny suggested that PylS was present in the last universal common ancestor (LUCA) 61,63 . Similarly, it has been argued that Sec recoding is an ancient archaeal trait considering the highly conserved nature of the Sec incorporation machinery 37 , and the fact that the genes involved are not always physically linked in an operon, which impedes its propagation between lineages via horizontal transfer 64 . Our Pyl and Sec trees indicate primarily vertical evolution of these genes (Figs. S20-28), suggesting that HGT is an infrequent event in the evolution of both traits in archaea. Therefore, we suggest that the last Asgardarchaeota common ancestor possessed both the Pyl and Sec recoding. Subsequently, Pyl was lost in the branch leading to Heimdallarchaeia and eukaryotes, and also in Jordarchaeia and Odinarchaeia (Fig. 5). Lokiarchaeia and Thorarchaeia also lack the Pyl gene cluster (Figs. 5 and Fig. 1 and Table S15), but we detected tRNApyl sequences in genomes from both lineages which could be remnants of an ancient Pyl trait that has since been lost. The roles of these tRNAs are unknown, however, they could function as sources of various small noncoding RNA species 65 . Sec recoding, on the other hand, remained present in most Asgardarchaeota lineages and was only lost in Heimdallarchaeia and Odinarchaeia (Fig. 5). Maintaining these presumably ancient recodings could be driven by selective metabolic advantages, i.e. the catalytic advantages of Sec-containing enzymes and the importance of Pyl for active sites of methyltransferases (see above).

Inferred ecology of novel Asgardarchaeota lineages
The low relative abundances of Asgardarchaeota in our available samples (0.1-2.67%, Fig S2) impeded visualisation and multiomics approaches, and limited the interpretations of ecological roles of Sifarchaeia and Jordarchaeia to the analysis of physicochemical metadata, taxonomic and functional community profiles, and features inferred from genomic reconstructions.
Based on our chemical analysis, we found that the sulphate levels in lake Weyba sediments, from which Sifarchaeia MAGs were recovered, were higher than in a neighbouring lake (Table S16) and comparable to levels in anoxic deep sea sediments 4,66 , which suggested that this site is a suitable habitat for sulphate reducers. Indeed, we detected dissimilatory sulphite reductase (dsrAB) genes, encoding a key enzyme in sulphate reduction 67 , in Lake Weyba metagenome assemblies (Table S17). Furthermore, SSU rRNA gene-based community profiles of Sifarchaeia-containing samples revealed the presence of taxa  (Fig. S13), showing inferred gain and loss of selenocysteine (Sec) and pyrrolysine (Pyl) recoding from the last Asgard archaeal common ancestor (LAsCA) to extant taxa in the Asgardarchaeota including eukaryotes. Partial loss is defined as the loss of Pyl biosynthesis and insertion genes while retaining the Pyl-tRNA.
assigned to sulphate-reducing bacteria (SRB), including Desulfobacterota (formerly Deltaproteobacteria), with a combined relative abundance of up to 31.5% (Table S18). Given that the only cultured Asgardarchaeum Ca. Prometheoarchaeum syntrophicum strain MK-D1 has been maintained in co-cultures with a methanogen or with the SRB Halodesulfovibrio 11 , it is tempting to speculate that Sifarchaeia form a similar syntrophic relationship with SRB by providing formate and hydrogen equivalents while receiving certain amino acids and vitamins, which this novel lineage cannot synthesise (Table S9). While methanogens were absent from Lake Weyba metagenomes (Table S18), this result is not surprising, since SRB are believed to outcompete methanogens under non-limiting sulphate concentrations due to the increased energetic efficiency in acquiring common substrates over methanogenesis 68 . Nevertheless, we detected a low number of methyl-CoM reductase (mcr)-like genes in Lake Weyba metagenomes, which however, were assigned exclusively to Helarchaeales and other Lokiarchaeia lineages. Helarchaeales have been inferred to possess the potential to anaerobically oxidise short-chain hydrocarbons 4 and are therefore unlikely to represent a hydrogen-consuming, syntrophic partner for Sifarchaeia. In addition, it has been concluded that the absence of a classical membrane-bound hydrogenase in Helarchaeales eliminates the possibility that hydrogen is a major syntrophic electron carrier 4 .
Community profiles of Jordarchaeia-containing samples vary considerably, but all include <0.9% methanogens (Table S17 and S18) compared to a slightly higher percentage (2.5%) of SRB. These results were consistent with a lower number of Mcr genes compared to dsr genes in these samples. Nevertheless, members of both groups, methanogens and SRB, could function as a syntrophic partner for Jordarchaeia. Additionally, hydrogenotrophy is pervasive in geothermal systems, particularly among members of the Aquificales and diverse archaea 69 , providing additional potential metabolic partners for thermophilic Jordarchaeia.

CONCLUSION
In the present study, we applied taxonomic rank normalisation to genome phylogenies including 71 novel Asgardarchaeota genomes and propose two novel Candidatus classes, Sifarchaeia and Jordarchaeia, which have the potential to convert C1 compounds into organic products as methylotrophic acetogens. Thereby, both classes utilise a methanogen-like pathway but do not encode homologues of the key enzyme methyl-CoM reductase (Mcr). This absence, together with the inferred Mcr-like enzymes in Helarchaeales 4 , and our detection of an McrA-like gene in Lokiarchaeia outside the order Helarchaeales (Fig. S29), suggests that pathways for the utilisation of methane and other hydrocarbon gases, or remnants thereof, played an important role in the evolution of Asgardarchaeota. We further reveal recoding as an ancient trait in this phylum, which allows the incorporation of the rare amino acids selenocysteine (Sec) and pyrrolysine (Pyl) into selected proteins, possibly yielding benefits from enhanced catalytic properties of Sec-and Pyl-containing enzymes. Thereby, Pyl, which is restricted to Sifarchaeia (see note added in proof), with remnant tRNAs in Thorarchaeia and Lokiarchaeia, likely supports efficient methylamine utilisation, and possibly represents another relic from a methylotrophic methanogen or methanotrophic ancestor. Next to Desulfobacterota, living as endosymbionts in a gutless marine oligochaete 70 , Sifarchaeia are only the second lineage inferred to encode both, Sec and Pyl containing proteins. Considering that Sifarchaeia and the symbiotic Desulfobacterota were recovered from anaerobic marine sediments, this type of environment may be a hotspot for stop codon recodings.
Our results support previous reports of a close relationship between Asgardarchaeota and eukaryotes, based on phylogenetic inferences, the detection of various encoded eukaryotic signature proteins and of enzymes for the biosynthesis of bacterial/ eukaryotic-type ester lipids in Sifarchaeia and Jordarchaeia and other lineages in this phylum. We envision that future recoveries of additional Asgardarchaeota MAGs, in concert with culturebased approaches, will further fuel phylogenomic and metabolic reconstructions and lead to the experimental verification of encoded functions, thereby ultimately shedding more light on the origin of eukaryotes.

Note added in proof
During the final stages of review of this manuscript, three papers were published that collectively describe seven new Asgard phyla (and a number of subordinate lineages) based on 39 novel MAGs: Hermodarchaeota 71 , Sifarchaeota 72 , Kariarchaeota, Hodarchaeota, Borrarchaeota, Baldrarchaeota and Wukongarchaeota 73 . These genomes have not been included in the analyses presented in our study due to their recent publication, however, an additional phylogenetic inference indicates that one of our new classes is synonymous with Sifarchaeota and Borrarchaeota (Fig. S30). Due to its publication priority, we have used Sifarchaeota as the base name, noting that this lineage represents a class (Sifarchaeia; see proposal of higher ranks) according to rank normalisation, which we use throughout this manuscript. We also propose the intermediate ranks of family and order, and a corrected spelling of the genus Ca. Sifarchaeotum 72 . Furthermore, the species represented by MAG "lw60_2018_gm2_56" in our study belongs to the genus Ca. Borrarchaeum proposed by Liu et al. 73 , which in turn belongs to the class Sifarchaeia (Fig. S30). We propose the name Ca. Borrarchaeum weybense for this species, see proposal of type material. Additionally, we confirmed pyrrolysine recoding in other members of Ca. family Borrarchaeaceae, but not in the two other MAGs representing the class Sifarchaeia (Fig. S30) suggesting that this type of recoding has been lost in some members of this class.

Proposal of type material
Candidatus Borrarchaeum weybense. Candidatus Borrarchaeum weybense (wey.ben'se. N.L. neut. adj. weybense of or pertaining to Lake Weyba, a saltwater lake in Queensland, Australia). Inferred to be a hetero-organotroph with genetic code expansions (recodings) allowing the incorporation of the rare 21st and 22nd amino acids selenocysteine and pyrrolysine. This uncultured lineage is represented by the genome "lw60_2018_gm2_56", NCBI BioSample SAMN19461863, recovered from Lake Weyba sediments, and defined as high-quality draft MAG 74 with an estimated completeness of 94.08% and 3.74% contamination, the presence of a 23S, 16S and 5S rRNA gene and 16 tRNAs.
Candidatus Jordarchaeum madagascariense. Candidatus Jordarchaeum madagascariense (ma.da.ga.scar.i.en'se. N.L. neut. adj. madagascariense of or pertaining to Madagascar, an island country in the Indian Ocean). This uncultured lineage is represented by the genome "EB_bin_7", NCBI BioSample SAMN19461862, recovered from elephant bird fossils in Madagascar, with an estimated completeness of 95.02% and a contamination of 2.41%, the presence of a 23S, 16S and 5S rRNA gene and 6 tRNAs.

Sample collection and DNA extraction
Sunshine Coast Lakes sediment. Lake sediment samples from Lake Cootharaba (LC) (−26.28°, 152.99°) and Lake Weyba (LW) (−26.44°, 153.06°) were sampled using sterilised one-metre PVC pipes. LC sediments at depths from 5 cm to 25 cm and LW sediments at depths from 5 cm to 60 cm were sampled in 5 cm intervals in December 2018 and November 2019. Salinity of lake water was recorded using a Seawater Digital Refractometer (Milwaukee, US). Collected sediments were flash frozen in alcohol and dry ice, and delivered to ALS Environmental testing, Brisbane, Australia for chemical analysis. DNA was extracted within four hours of sampling using the PowerSoil DNA Isolation kit (MoBio, USA) following the manufacturer's protocol.
Hikurangi subduction margin sediment. Deep-sea sediment samples of Hikurangi Subduction Margin were sampled by the International Ocean Discovery Program (IODP) Expedition 375 scientists onboard 75 . Sampling holes were drilled at four sites: U1518 (an active fault near the deformation front; sampling depths range from 0 mbsf to 494.90 mbsf), U1519 (the upper plate above the high-slip slow slip event source region; sampling depths range from 0 mbsf to 640.00 mbsf), U1520 (the incoming sedimentary succession in the Hikurangi Trough; sampling depths range from 0 mbsf to 1045.75 mbsf) and U1526 (atop the Turanganui Knoll Seamount; sampling depths range from 0 m to 83.60 mbsf) 76 . Sediment cores were sub-sampled shipboard using 5 ml syringes, which were stored and shipped on dry ice until they reached the laboratory and were then stored at −80°C until DNA extraction. To minimise possible contamination, we trimmed off the outer centimetre of each sample and used the inner sediment core for DNA extraction. To optimise DNA extraction for these low biomass samples, 300 mg sediments were first mixed with G2 DNA/ RNA Enhancer beads (Ampliqon, Denmark). The subsequent DNA extraction steps were conducted using the PowerSoil DNA Isolation kit (MoBio, USA) following the manufacturer's protocol.

Public data acquisition
Potential Asgardarchaeota containing metagenomes were identified in the NCBI Sequence Read Archive (SRA) using SingleM (https://github.com/ wwood/singlem). This software uses single-copy marker genes to search for public metagenomes containing reads that match a bacterial or archaeal lineage of interest. The search for Asgardarchaeota reads yielded matches for seven corresponding study IDs (SRP029382, SRP061771, ERP013176, SRP077065, SRP049601, DRP003377 and SRP098167) in the SRA database (Table S1). Information from all NCBI sequencing runs from each study was collected, but only shotgun metagenomic sequence runs were downloaded for our analysis.

Small subunit rRNA gene community profiles
To obtain microbial community profiles, we aligned the reads of all shotgun sequenced samples to the SILVA 132_99 database 16 and classified the reads into operational taxonomic units (OTUs) using CommunityM (https://github.com/dparks1134/CommunityM) under default settings.

Metagenome assembly, binning and bin dereplication
The raw reads generated from the Sunshine Coast Lake and Hikurangi Subduction Margin sediment DNA were first processed using SeqPrep (https://github.com/jstjohn/SeqPrep) under default settings to merge overlapping paired-end reads and trim adaptors. Pre-processed pairedend reads were then assembled using metaSPAdes genome assembler v3.13.0 79 with default settings. The raw reads from the Geothermal spring sediments were assembled using ALLPATHS 80 . Reads obtained from SRA were assembly using metaSPAdes with default settings. BamM (http:// ecogenomics.github.io/BamM/) was then used to map sequences back to the assemblies. Next, binning was performed with uniteM (https://github. com/dparks1134/unitem) using selected methods (metabat_sensitive, metabat2, maxbin_107, maxbin_40 and groopM) under the default settings. CheckM 81 was then applied to calculate estimated completeness, contamination as well as strain heterogeneity. For metagenomeassembled genomes (MAGs) binned via multiple binning methods, the average nucleotide identity (ANI) was calculated, and MAG pairs with ANI >99% were de-replicated by keeping the MAG with the highest quality, defined as completeness − 4 * contamination).

Phylogenomics, rank normalisation and pangenomics
A total of 143 Asgardarchaeota genomes, including MAGs recovered from samples in this study, extracted from public SRA datasets, and downloaded from Genbank 82 with an estimated quality (completeness − 4 × contamination) over 40% were included in the downstream analysis. The multiplesequence alignment of selected MAGs was generated using gtdb-tk 83 based on 122 archaeal-specific marker proteins (Table S2). Maximum likelihood (ML) phylogenies for archaeal genomes were inferred using IQ-Tree 1.6.9 84 under the LG + C10 + F + G + PMSF model. Statistical support was estimated on a set of 1480 archaeal genomes (including 1377 non-Asgard archaea GTDB species representatives from GTDB release 05-RS95) using 100 bootstraps replicated under the same model ( Figs. 1 and S4-5). In addition, ML trees of trimmed alignments, from which we removed compositionally biased sites to increase tree accuracy for distantly related sequences prior to concatenation, using BMGE 85 or Divvier 86 , were evaluated with the same method (Figs. S6-7).
To further confirm the phylogenetic placement of Asgardarchaeota lineages, three additional ribosomal protein marker sets were used to create alignments: 16 ribosomal proteins defined in Hug et al. 19 , a subset of 23 proteins used by Rinke et al. 20 and a subset of 53 from the 56 top ranked archaeal marker proteins assessed in Dombrowski et al. 18 . Proteins were aligned to Pfam and TIGRfam HMMs using HMMER 3.1b2 (http:// hmmer.org) with default parameters. The alignments were subjected to phylogenomic analysis using IQ-Tree 1.6.9 84 under the LG + C10 + F + G + PMSF model (Fig. S8-10). Bayesian trees were inferred with Phylobayes 87 for a subset of 44 genomes (incl. 34 Asgardarchaeota) under the CAT + GTR + G4 model (Fig. S11). Four independent Markov chains were run for 43,000 generations. After a burn-in of 10%, convergence was achieved for all chains (maxdiff < 0.1). All phylogenetic trees inferred in this study are summarised in Table S2. Trees were viewed and annotated by iTOL 88 .
The ranks of Asgardarchaeota lineages were normalised with the tool PhyloRank (https://github.com/dparks1134/PhyloRank) based on the relative evolutionary divergence (RED) values, as implemented in the Genome Taxonomy Database (GTDB) 6,7 ; https://gtdb.ecogenomic.org/). In brief, PhyloRank linearly interpolates the RED values of internal nodes according to lineage-specific rates of evolution under the constraints of the root being defined as zero and the RED of all present taxa being defined as one. To account for the influence of the root placement on RED values PhyloRank roots a tree multiple times, at the midpoint of each phylum with two or more classes. The RED of a taxon is then calculated as the median RED over all these tree rootings, excluding the tree in which the taxon was the outgroup. The RED intervals for each rank were defined as the median RED value ±0.1 to serve as a guide for the normalisation of taxonomic ranks from genus to phylum in GTDB. Taxonomic assignments follow the naming formation and hierarchy of standard taxonomic categories based on their nomenclature types defined by the International Code of Nomenclature of Prokaryotes and recent proposals to amend the Code [89][90][91] . We also consider recommendations on quality standards for genomes considered as types see Chuvochina et al. 21 and references therein.
For example, the recent proposal to formalise the rank of phylum under the Code provision 90 with the addendum by Whitman et al. 91 defines that phylum names are to be formed by the addition of the suffix -ota, such as Asgardarchaeota. A detailed description of the archaeal GTDB taxonomy including nomenclature curation workflows is provided in Rinke et al. 6 .
To review the evolutionary relationship between Asgardarchaeota and eukaryotes, we used GraftM 93 for the identification of orthologues of 15 ribosomal proteins used in a previous studies 1, 22 . Eukaryotic hits were confirmed according to their NCBI annotation. The collected sequences for each marker gene were aligned with MAFFT v7.455 94 and concatenated. The concatenated alignment was then trimmed by TrimAl v1.4 95 with '-gappyout' selection. Maximum-likelihood tree was calculated by IQ-TREE 84 under 'LG + C60 + F + G + PMSF' model. Statistical branch support was calculated using 100 bootstraps under the same model.

Proposed type material
MAGs proposal as type material were selected considering MIMAG standards 74 and following the recommended practice for proposing nomenclature type material 21 .
Hydrogenase. We collected [NiFe]-, [FeFe]-and [Fe]-hydrogenase sequences from the study of Greening et al. 98 to create a Blast database, which was used to query the 143 Asgardarchaeota genomes to search for potential hydrogenase genes. The sequence hits with e-values < 1e-20, scores >100, and sequence identities >30% were then submitted to HydDB 99 for further identification of hydrogenase subgroups.
Lipid membrane biosynthetic genes. KEGG orthologs of ester/ether lipid biosynthesis genes were used to investigate the potential of membrane lipid synthesis in Asgardarchaeota. To calculate the phylogenetic tree of glycerophosphoryl diester phosphodiesterase (UgpQ), we included genes used from a previous study of UgpQ phylogeny 29 . Eukaryotic UgpQ sequences were obtained from UniprotKB (http://www.uniprot.org) based on assignments to PF03009, including only sequences categorised as "Protein Existence [PE]" with the UniprotKB levels "Evidence at protein level" and/or "Evidence at transcript level". Asgardarchaeota UgpQ homologues were identified with blastp 100 against KO K01126 by only retaining sequences with a maximum e-value of 1e-30. Collected UgpQ sequences were aligned using HMMER 3.1b2 (http://hmmer.org) against Pfam PF03009.
In addition, as the lipopolysaccharide ABC transporter genes were exclusively detected in Sifarchaeia MAGs, we inferred phylogenetic trees to rule out the possibility of mis-annotation. Sequences of lipopolysaccharide transport system ATP-binding protein (TagH, COG1134) and lipopolysaccharide transport system permease protein (TagG, COG1682) were collected from the NCBI conserved domain database. Collected sequences together with Sifarchaeia hits for each COG were aligned using MAFFT v7.455 94 , respectively. Maximum-likelihood trees of UgpQ, TagH, and TagG were initially inferred by FastTreeMP 101 with Wag+Gamma model and subsequently with IQtree 84 under 'LG + C60 + F + G + PMSF' model with 100 bootstraps.
ESP identification. High-quality Asgardarchaeota genomes (complete-ness>90%; <10% contamination; n = 38) were selected to search for eukaryotic signature proteins (ESPs) listed in the annotation table in Zaremba-Niedzwiedzka et al. 1 The analysis was limited to high-quality MAGs in order to minimise false negative hits. The resulting information was used to complete the ESP presence/absence table (Table S4). We used Prodigal 102 for gene prediction and hypothetical genes were annotated by InterProScan 5.31 97 to screen for ESP homologues with certain IPR domains. As for ESPs denoted by the COG database, we downloaded sequences for each COG entry from the NCBI conserved domain database 103 . The COG sequences were passed to GraftM 0.13.1 93 to create GraftM packages, which were then used to query Asgardarchaeota genes, with 'graftM create' and 'graftM graft' functions under default settings, respectively. Hits were further confirmed by blastp 100 against the NCBI non-redundant protein database (https://blast.ncbi. nlm.nih.gov/Blast.cgi).
Selenocysteine encoding system. We used Secmarker 0.4 39 with the Infernal score threshold of 40 to detect the presence of tRNAsec in the Asgardarchaeota genomes and all archaeal and bacterial GTDB release 04-RS89 genus-dereplicated genomes. The detected tRNAsec sequences were aligned with MAFFT v7.455 94 and trimmed by a minimum consensus of 40% 104 . Maximum-likelihood tree of tRNAsec was inferred using IQtree with 100 bootstraps under the VM + F + I + G4 model, which was selected by IQ-TREE's ModelFinder module 84 . Seblastian 105 with default settings was applied to search for both selenocysteine insertion sequences and selenoproteins in Asgardarchaeota MAGs. The detected selenoproteins were verified by comparing the annotations to the corresponding Prokkaannotated genes with similar positions.
Genes encoding enzymes responsible for selenocysteine biosynthesis and insertion were decided by annotation methods described above. Additionally, as the Thorarchaeota MAG "SMTZ1-83" is the only Asgardarchaeota genome proposed to encode SelA 38 , we blasted the genes present on the contig (LRSK01000263.1) containing selA, using blastp 100 under NCBI non-redundant protein sequences database. The results are shown in Table S12, and reveal that this contig is most likely a contamination.
Since homologues of genes encoding SelB and SPS have been reported in archaeal, bacterial and eukaryotic genomes 37 (Mariotti et al.), we hypothesised that these genes might be valuable to better understand the evolution of selenocysteine recoding. Bacterial and eukaryotic SelB and SPS sequences were selected and downloaded from UniprotKB (http:// www.uniprot.org) to cover diverse taxonomic groups. Archaeal SelB and SPS sequences were collected from the order Methanococcales, two Methanopyrus genomes, and Asgardarchaeota, whose genomes were reported to be tRNAsec-positive. The collected gene sequences were aligned with MAFFT v7.455 94 and trimmed by TrimAl v1.4 95 with '-automated1' selection. Maximum-likelihood trees were calculated by IQ-TREE 84 under 'LG + C10 + F + G + PMSF' model with 100 bootstraps.
Pyrrolysine encoding system. The presence of tRNApyl in Asgardarchaeota MAGs was determined by Prokka 1.14.6 96 . All genes of tRNApyl containing contigs of Thorarchaeia and Lokiarchaeota MAGs were compared against NCBI nr with blastp 100 to screen out possible contamination (Table S19).
Genes encoding enzymes responsible for pyrrolysine (Pyl) biosynthesis (PylS, PylB, PylC, PylD) and insertion (RF1) were detected by annotation methods described above. To explore the evolution of the Pyl system, we collected protein sequences of PylSBCD cluster genes (PylSc, PylSn, PylB, PylC, PylD) from the GTDB release 03-RS86 genus-dereplicated genomes. This was achieved by hmmsearch (Sean R. Eddy, http://hmmer.org) against HMM models of TIGR03912 (pyrrolysine--tRNA ligase, N-terminal region), TIGR02367 (pyrrolysine--tRNA ligase, C-terminal region), TIGR03910 (pyrrolysine biosynthesis radical SAM protein), TIGR03909 (pyrrolysine biosynthesis protein PylC), and TIGR03911 (pyrrolysine biosynthesis protein PylD). Homologues of PylB, PylC and PylD that were not located on the same contigs were excluded, since these genes encode enzymes for pyrrolysine biosynthesis, and were only reported to be in close proximity. All genomes with at least two Pyl genes, which equals 50% of the required genes, were included in the downstream analysis. The collected sequences for each gene were aligned with MAFFT v7.455 94 and trimmed by TrimAl v1.4 95 with '-automated1' selection. The PylS alignment was created by concatenating sequences of PylSn and PylSc. Maximum-likelihood trees were calculated by IQ-TREE 84 under 'LG + C10 + F + G + PMSF′ model with 100 bootstraps. Then we concatenated the above alignments in the order of pylSBCD, with the absence of certain genes represented by gaps. The contaminated alignment was trimmed by TrimAl v1.4 95 with '-gt 0.4' selection and further trimmed to exclude columns with less than 40% of consensus. Sequences with <80% remaining amino acids were removed, resulting in a final alignment of 62 protein sequences with 1103 columns. A maximum-likelihood tree was calculated with IQ-TREE 84 under 'LG + C10 + F + G + PMSF' model with 100 bootstraps.
To search for Pyl-containing genes, we applied a strategy described previously 106 . In brief, we compared the annotation of each UAGterminating CDS in all Sifarchaeia MAGs with the annotation of its downstream neighbouring CDS. In cases of matching annotations, both CDS were fused in silico as a unique CDS and predicted as potentially Pyl incorporating.
D-lactate dehydrogenases. Gene annotations of the encoded putative Dlactate dehydrogenase (Dld) KEGG orthologs in Sifarchaeia and Jordarchaeia MAGs were verified using Pfam and TIGRfam HMMs (Table S20).

DATA AVAILABILITY
The raw reads and genome sequences from the metagenomes described in this study are available at NCBI under multiple BioProjects: PRJNA678545 (Sunshine Coast lakes) and PRJNA678552 (Hikurangi Subduction Margin). Genome sequences assembled and binned from public metagenomes and described in this study are available at NCBI under the BioProject PRJNA678817. All datasets generated and/ or analysed during this study, including genome sequences, are available in our data repository at Zenodo.