Introduction

Methane has been a key component in the atmosphere since the dawn of life on Earth (Kasting, 1993). Being one of the most potent greenhouse gases, methane has a crucial role in regulating the modern as well as ancient climate of Earth (Pavlov et al., 2000). Snowball events in the ancient Earth could be linked to declines in the concentration of methane, and modern Earth is now experiencing global warming at least in part owing to excess emissions of anthropogenic methane (Kasting, 2004; Bousquet et al., 2006). The biological production of methane in the atmosphere has been largely attributed to the activity of methanogens, a group of strictly anaerobic microbes which likely originated at some point in the mid to early-Archean (Battistuzzi et al., 2004; Ueno et al., 2006; Wordsworth and Pierrehumbert, 2013). It is suggested that methanogens could have dominated the biosphere in most of the Archean Eon, and methane emissions may have prevented the rapid cooling of the early Earth (Catling et al., 2001). However, populations of methanogens appeared to decline greatly at the end of the Archean Eon, likely due to the depletion of oceanic nickel, which is an essential cofactor for many key enzymes in the methanogenesis pathway and required for growth (Thauer et al., 2010), and the increasing environmental oxygenation during the Great Oxidation Event at ~2.5–2.3 Ga (Konhauser et al., 2009). Oxygenation could be detrimental directly to methanogens via the formation of deadly derivatives of O2 or reactive oxygen species (ROS), such as H2O2 and O2− radicals (Imlay, 2008).

Previous reports have demonstrated that methanogens neither abandon their overall dependence on nickel nor develop nickel-specific ligands to cope with low nickel availability when facing nickel famine (Hausrath et al., 2007; Thauer et al., 2010). Likewise, it remains elusive how ancestors of methanogens coped with the ROS damage. Nevertheless, they appeared to have survived those catastrophic events and proliferated into the modern era. Today, modern methanogens are virtually found in all types of anaerobic environments, and most of them are sensitive to oxygen (Fetzer et al., 1993; Fetzer and Conrad, 1993; Thauer et al., 2010; Yuan et al., 2011). However, it has been shown that some methanogens could survive oxygen stress for several hours to days (Kiener and Leisinger, 1983; Fetzer et al., 1993; Ueki et al., 1997; Ma and Lu, 2011), and they have also been found in typical oxidative environments such as upland soils (Angel et al., 2011; Angel et al., 2012). Likewise, physiological, ecological and genomic analyses have revealed that at least certain methanogen lineages have evolved strong antioxidant features (Fetzer and Conrad, 1993; Erkel et al., 2006; Angel et al., 2012; Lyu and Lu, 2015). However, how methanogens adapt to oxidative environments remain poorly understood. To gain more insight into this puzzle, here we sampled genomes from six well-established methanogen orders (Methanobacteriales, Methanocellales, Methanococcales, Methanomicrobiales, Methanosarcinales and Methanopyrales) and predicted their antioxidant features by analyzing functional genes relevant to ROS production, O2/ROS elimination and self-repairing systems. A survey of environmental 16 S rRNA genes of those methanogen lineages was also conducted to determine their distribution in oxidative environments. At last, a phylogenomic analysis was performed to estimate major diversification events of those methanogens and to put our results into an evolutionary perspective.

Subjects and methods

Comparative genomics

Functional genes associated to antioxidant potentials were extracted from the sampled methanogen genomes (Supplementary Table 1) and manually curated and re-annotated based on experimental evidence (Supplementary Methods). The presence of [4Fe-4S] clusters in methanogenesis redox enzymes was summarized from various experimental sources and verified via examining conserved [4Fe-4S] binding motifs for all strains by sequence alignments (Major et al., 2004). Hierarchical clustering of gene content was done by the Ward’s minimum variance method with JMP Pro, Version 12, SAS Institute Inc. Nonparametric tests were performed with Kruskal–Wallis analysis of variance followed by Mann–Whitney to investigate the statistical differences in gene abundance among methanogen classes.

Environmental gene survey

Environmental 16 S rRNA gene sequences were extracted from EnvDB and classified by RDP10 (Cole et al., 2009; Tamames et al., 2009). Habitat information was further extracted and verified from the literature as well as GenBank for each sequence. Complete methods and datasets were summarized in Supplementary Information.

Phylogenetic analysis

Genome sequences were sampled from IMG/ER (https://img.jgi.doe.gov/cgi-bin/mer/main.cgi). Genes encoding the 22 ribosomal proteins were extracted from the sampled genomes (Supplementary Table 1). Ribosomal proteins were aligned by MAFFT (v7.107) (Katoh and Standley, 2013). Concatenation of sequence data was done by Geneious R8.1 (Kearse et al., 2012). The E-INS-I algorithm was used for MAFFT. The alignments were filtered by eliminating any position that was missing data from 90% or more out of the total aligned taxa unless otherwise mentioned. Phylogenetic trees were constructed by either RAxML (v8.2.9) constrained by the bootstrapping autoMRE stopping criteria or MrBayes (v3.2.6) through Bayesian inference (BI) (Ronquist and Huelsenbeck, 2003; Stamatakis et al., 2008). Both algorithms were hosted by the CIPRES Science Gateway (Miller et al., 2010), where a WAG protein matrix model with corrections for a gamma distribution with four discrete rate categories and a proportion of invariant sites (WAG+G+I) were implemented for the phylogenetic analyses. Whenever necessary, another protein model (LG+G+I) was also implemented in both RAxML and MrBayes, testing if a tree topology remained robust after changing the substitution model. This model was chosen out of the 56 different substitution models examined by Mega 7 based on the best fit Bayesian Information Criterion (Kumar et al., 2016). For BI, the following parameters were also implemented, nrun=3, ngen=1 000 000, nchains=4, samplefreq=10, nst=6, Nucmodel=protein, rates=invgamma, sump burnin=50 000, sumt burnin=50 000 and burin frac=0.25. A few pre-runs were performed to optimize the parameters to ensure the log probability plateaued after the burnin setting and the Markov Chain Monte Carlo samplings converged before the ngen setting.

Both RAxML and MrBayes trees were subjected to molecular dating with the RelTime module in Mega 7 (Tamura et al., 2012; Kumar et al., 2016). Relative node ages were first obtained by using the ribosomal protein alignment data with the same aforementioned WAG+G+I or LG+G+I model in a maximum-likelihood framework. Reliable calibration anchors reported previously were then used to scale the relative node ages into absolute dates (Battistuzii and Hedges, 2009; Marin et al., 2017). RelTime belongs to the latest version of molecular dating algorithms, which appears to outperform the sophisticated and time-consuming Bayesian methods while retaining comparable accuracy (Tamura et al., 2012; Kumar and Hedges, 2016).

Results and discussion

Antioxidant features across methanogens

Three strategies could be envisaged to respond to the oxygenation challenge: (i) limiting ROS production, (ii) reducing accumulation of ROS within the cell and (iii) self-repairing for ROS damage. Studies have revealed that all strategies are essential to ensure survival of both aerobes and anaerobes under oxidative stress (Gort and Imlay, 1998; Imlay, 2002; Hillmann et al., 2008; Imlay, 2008). Accordingly, we retrieved the functional genes involved in these strategies from methanogen genomes, and the results shown below grouped those methanogens into two clusters based on their distinct antioxidant features.

ROS production

The amount of ROS production is mainly related to the presence of quinone-derived electron carriers and [4Fe-4S] clusters in the central metabolism (Imlay, 2002). This is owing to the formation of superoxide (O2−) and hydrogen peroxide (H2O2) when oxygen reacts with the quinone-like electron carriers (for example, flavin and ubiquinone) (Müller, 1987; Massey, 1994). The reactivity of oxygen with those electron carriers is owing to the formation of semiquinone derived radicals (for example, flavosemiquinone and ubisemiquinone) (Müller, 1987; Massey, 1994). The H2O2 further oxidizes [4Fe-4S] clusters and generates hydroxyl radicals (HO·) by the Fenton reaction (Walling, 1975).

Being the hub of energy production and biosynthesis, the methanogenesis pathway is probably the main source for ROS production. Hierarchical Clustering analyses based on functional gene content revealed two different clusters (Figure 1). The two clusters correspond to the Class I and II methanogens proposed by the Brochier-Armanet group (Brochier-Armanet et al., 2011; Petitjean et al., 2015). The Class I methanogens include Methanobacteriales, Methanococcales and Methanopyrales, while the Class II comprises Methanocellales, Methanomicrobiales and Methanosarcinales. Therefore, a systematic shift in gene content patterns for methanogenesis would likely be associated with the evolution of methanogens at the class level.

Figure 1
figure 1

Hierarchical clustering topology for the methanogenesis gene content (Supplementary Table 2) among the methanogens, which are color coded as blue and green, corresponding to the previously proposed Class I and Class II methanogens (Brochier-Armanet et al., 2011; Petitjean et al., 2015), respectively. The clustering was done by the Ward’s minimum variance method with JMP Pro.

Full size image

This shift could potentially reduce the ROS production for the Class II methanogens through two different mechanisms in the methanogenesis pathway. One, the Class I methanogens use a flavin/ferrodoxin mediated electron bifurcation mechanism (Thauer et al., 2008; Kaster et al., 2011), whereas the Class II Methanosarcinales mainly use a cytochrome/methanophenazine-mediated electron transfer mechanism for methanogenesis (Table 1 and Supplementary Table 2) (Welander and Metcalf, 2005; Thauer et al., 2008; Kulkarni et al., 2009). The electron bifurcation mechanism requires the formation of flavosemiquinone, which could react with oxygen to form O2− and H2O2 (Buckel and Thauer, 2012). In hydrogenotrophic Methanosarcinales spp., flavin proteins are not involved in the cytochrome/methanophenazine mechanism (Welander and Metcalf, 2005). In aceticlastic and methylotrophic Methanosarcinales spp., a Fpo or Rnf flavin protein complex participates in the electron transfer (Kulkarni et al., 2009; Spring et al., 2010; Schlegel et al., 2012). However, current evidence favors a direct electron transfer instead of an electron bifurcation mechanism for the Fpo or Rnf complex, which would not require a formation of flavosemiquinone (Catlett et al., 2015). A bifurcation mechanism has recently been proposed for the cytoplasmic Hdr complex, but it may only play an auxiliary role when needed for maximal thermodynamic efficiency (Yan et al., 2017). Two, although the Class II Methanomicrobiales and Methanocellales have been predicted to use the bifurcation mechanism, they reduce their [4Fe-4S] motifs by about 70% in their methanogenesis pathway compared to the Class I methanogens (Table 1). This could reduce HO· production through the Fenton reaction. This view was supported by the Hierarchical Clustering analysis that showed Methanomicrobiales and Methanocellales merged into the Class I once enzymes with reduced numbers of [4Fe-4S] motifs were removed from the bifurcation-based methanogenesis pathway (Supplementary Figure 1). Likewise, Hierarchical Clustering solely based on those [4Fe-4S] enzymes separated Methanomicrobiales and Methanocellales from the Class I (Supplementary Figure 2). Interestingly, the same analysis now moved Methanosarcinales into the Class I, suggesting an unexpected link between Methanosarcinales and the Class I methanogens in the context of [4Fe-4S] enzymes (Supplementary Figure 2).

Table 1 Redox enzymes potentially contributing to ROS production from methanogenesis pathways
Full size table

O2/ROS elimination

O2/ROS elimination is achieved by a variety of antioxidant enzymes in model microbes (Imlay, 2008; Imlay, 2013). By working together, these enzymes can reduce O2 to H2O, and transform H2O2 and O2− into less-toxic products such as O2 and H2O. HO· can only be indirectly eliminated by reducing HO· production via iron storage to avoid the Fenton reaction (Neilands, 1993; Touati et al., 1995). Many of such eliminations are redox-dependent reactions and require reducing power to proceed, which usually comes directly from small redox proteins (Lu and Holmgren, 2014). These redox proteins also serve as a buffering system to keep cellular redox system from becoming overly oxidized (Susanti et al., 2014).

Overall, common O2/ROS elimination enzymes were present in all methanogens regardless of class affiliation except for the complete absence of catalase in the Class I methanogens (Table 2 and Supplementary Table 3). However, Class II methanogens possessed statistically more genes encoding antioxidant enzymes and small redox proteins than their Class I counterparts. These observations suggest that a core pool of antioxidant genes is shared by both classes of methanogens, but this pool expands numerically in the Class II methanogens. However, the statistical difference for this expansion disappeared after correction for genome size. The genomes of the Class II methanogens were ~1.7 times larger than those of the Class I methanogens (Table 2). In other words, the genome size and the pool of antioxidant genes expand by the same extent in the Class II methanogens. On one hand, the expansion of antioxidant gene set could simply be a byproduct of genome expansion. If this is true, an across-the-board expansion of antioxidant genes would be expected. Apparently, this did not happen, and the F420H2 oxidase even exhibited a reduction (Table 2). On the other hand, a numerical expansion of certain but not all antioxidant genes in the Class II methanogens could reflect a specific adaptation to oxidative environments.

Table 2 Comparison of gene abundance for O2/ROS elimination enzymes between the Class I and Class II methanogens
Full size table

Four specific observations support this latter view (Table 2). (i) Both F420H2 oxidase and NO/O2 reductase oxidize O2 into H2O, but the latter has a much higher Km for O2 (Gomes et al., 2002; Silaghi-Dumitrescu et al., 2005), whereas the former is deactivated when cells are exposed to air (Seedorf et al., 2004, 2007). In addition, NO/O2 reductase detoxifies NO, an inhibitor of methanogenesis which could be abundant at the oxic-anoxic interface (Klüber and Conrad, 1998; Rodrigues et al., 2006). The almost absence of F420H2 oxidase in the Class II methanogens is striking given that these methanogens still use F420H2 as main electron carriers for methanogenesis (Welander and Metcalf, 2005). Therefore, a shift from F420H2 oxidase dominance in the Class I methanogens to NO/O2 reductase dominance in the these methanogens could suggest an adaptation of the Class II members to oxidative conditions (Table 2 and Supplementary Figure 3). (ii) Rubredoxin and thioredoxin are small redox proteins that have been shown to have important roles in a variety of antioxidant processes for methanogens (Susanti et al., 2014). Thioredoxin is a stronger reducing agent than rubredoxin, transferring electrons at redox potentials ~−300 to −120 mV and 0±100 mV, respectively (Åslund et al., 1997; Balmer et al., 2004; Lin et al., 2005). Compared with the Class I methanogens, the rubredoxin system is numerically comparable while the thioredoxin system is substantially expanded in the Class II methanogens (Table 2), indicating a potential enhancement of the redox-buffering system in the Class II members. (iii) Transmembrane thioredoxin proteins were also commonly predicted in the Class II methanogens, but rarely in the Class I methanogens (Table 2). In addition to the thioredoxin domain present either in the cytoplasmic or periplasmic side, those transmembrane thioredoxin proteins also have two or three cysteine residues in the transmembrane region, located within the vicinities of both sides of the membrane (Figure 2). It has been previously shown that electron transfer could happen between those cysteine residues and the thioredoxin domain (Krupp et al., 2001; Deshmukh et al., 2003). Therefore, the transmembrane thioredoxin proteins may enable electron shuffle between the cytoplasmic and periplasmic spaces, which may help with redox recovery around the cellular membranes under oxidative stress (Figure 2). And (iv), the Class I methanogens appear to use F420H2 to regenerate the reduced thioredoxin, whereas the Class II methanogens probably use NADPH or NADH instead (McCarver and Lessner, 2014; Susanti et al., 2016). Although F420H2 may be an efficient electron donor when H2 levels were high in the atmosphere of an ancient and anoxic Earth, it would not be as good as NADPH or NADH in the face of an oxygenated Earth where H2 levels were low (Nisbet and Sleep, 2001; Lollar et al., 2014). Not only is it less likely to be reduced at lower H2 concentrations, but it reacts chemically with O2 and, thus is less stable than NADPH and NADH under oxidative conditions (Seedorf et al., 2004; Susanti et al., 2016).

Figure 2
figure 2

Predicted transmembrane structures for (a), Ccd-like protein using Mtc_2346 of Methanocella conradii as a reference; (b), Ccd-like protein with a transmembrane CxxC motif, using Mtc_1625 of Methanocella conradii as a reference; (c), DsbD-like protein with a periplasmic CxxC motif, using Mtc_0222 of Methanocella conradii as a reference; and (d), DsbD-like protein with a cytoplasmic CxxC motif, using Mbar_A1175 of Methanosarcina barkeri as a reference. Presumably, electrons could be relayed between the cytoplasmic (in) and periplasmic space (out), through those transmembrane proteins’ cysteine residues located in the cellular membrane and the thioredoxin domain located either in the cytoplasmic or the periplasmic space (Krupp et al., 2001; Deshmukh et al., 2003). Trx, thioredoxin domain. TM, transmembrane helix. C, cysteine residues. S-, sulfide moieties of the cysteine residues. Cartoons were drawn using DOG 1.0 (Ren et al., 2009). See more details in Supplementary Methods.

Full size image

Self-repairing system

ROS damages DNA bases by oxidizing purines and pyrimidines and is mutagenic (Dalhus et al., 2009). It also oxidizes membrane lipids into phospholipid hydroperoxides and reacts with sulfur amino acids of proteins to form disulfide bonds or methionine sulfoxide, leading to dysfunction of membrane and protein structures (Manevich et al., 2002; Weissbach et al., 2005). In addition, ROS causes iron release from iron–sulfur (FeS) clusters (Djaman et al., 2004). Cells can repair these damages to DNA, lipids, protein and FeS clusters via DNA base repair, phospholipid hydroperoxides reduction, disulfide bonds (S–S) and sulfoxide moiety (S=O) reduction, and FeS assembly, respectively.

Genes encoding DNA base repair, S–S and S=O reduction enzymes were either moderately or strongly enriched in the Class II methanogens compared to the Class I members (Table 3 and Supplementary Table 4). A similar trend was evident for phospholipid hydroperoxides reduction mediated by the peroxiredoxins already shown in Table 3. The high enrichment of cytoplasmic S–S reduction enzymes in the Class II methanogens was consistent with the higher abundance of thioredoxins (Table 3), which were largely regenerated by thioredoxin reductases (Supplementary Table 4).

Table 3 Comparison of gene abundance for enzymes repairing oxidative damage between class I and class II methanogens
Full size table

In terms of FeS assembly, most Class I methanogens could only use sulfide but not cysteine as a sulfur source, owing to the absence of genes encoding cysteine desulfurase (Liu et al., 2010). In contrast, a cysteine desulfurase gene namely iscU was ubiquitous in the Class II methanogens, and some members even possessed an additional cysteine desulfurase gene sufS. This ability of using cysteine as a sulfur source for FeS assembly may ensure adaptation to oxidative environments for the Class II methanogens, as sulfide is mostly depleted under oxidative conditions (Lyu and Lu, 2015). In addition, two types of FeS carriers (that is, A-type and ApbC) were encoded by many Class II methanogens except the Methanomicrobiales, while only one type (that is, ApbC) was encoded by the Class I methanogens, suggesting a further adaptation of many Class II members to oxidative environments. In bacteria, there is a shift between expression of the two types of FeS carriers during the adaptation to aerobic respiration from anaerobic fermentation (Outten et al., 2004). In summary, like the O2/ROS elimination system, self-repairing genes also appear to be more enriched in the Class II methanogens than the Class I members.

Class II methanogens more frequently detected in oxidative environments

Our comparative genomic analyses suggest a potential metabolic shift from the Class I to the Class II methanogens, resulting in a richer set of antioxidant features in the latter. However, encoding more antioxidant genes does not necessarily relate to different abilities to cope with oxidative stress. To gain more insight on this matter, an environmental 16 S rRNA gene sequences survey was performed to link gene sequence identity to its corresponding habitat. If the Class II methanogens are more adapted to oxidative environments than the Class I, they should be more frequently detected in a range of oxygenated habitats.

Extracted from EnvDB, a database that links prokaryotic taxa to their habit information, a total of 212 and 485 out of ~5000 sequences could be characterized as possessing origins from ‘oxic’ and ‘oxic-anoxic interface’ habitats, respectively (Table 4 and Supplementary Table 5) (Tamames et al., 2009). There was a substantial difference in the distribution of sequences affiliating with the two classes of methanogens. In particular, after normalization of sample size, sequences belonging to the Class II methanogens were nine times more frequent in oxygenated habitats than sequences of the Class I methanogens (Table 4). Importantly, this difference remained valid across a wide range of habitats including ocean, rice soil, subsurface, upland and wetland (Table 4), indicating a universal adaptation to oxic environments for the Class II methanogens. For the ‘interface’ sequences, the initial analysis showed no difference between the two classes of methanogens, after the sample size normalization (Table 4). However, a closer examination of the data suggested that the ‘interface’ sequences affiliated to the Class I methanogens were retrieved mostly in association with insects. Upon removal of all ‘insect’ associated sequences, sequences affiliating with the Class II methanogens were six times more frequent in the remaining ‘interface’ samples than sequences associated with the Class I methanogens (Table 4). Likewise, the already scarce ‘oxic’ sequences affiliated to the Class I methanogens were exclusively found in association with plants in rice and upland soils, suggesting that Class I methanogens found in oxygenated environments may not be exposed and adapted to oxygenic conditions but instead inhabit anoxic niches inside or near a host organism (Supplementary Table 5). In summary, the environmental data was in good agreement with the comparative genomics results, suggesting the Class II methanogens are more adapted to oxidative environments than the Class I methanogens.

Table 4 Abundance of 16 S rRNA sequences in different habitatsa
Full size table

Divergence time estimated for the two methanogen classes

To gain more insight in the evolutionary history of the Class I and Class II methanogens, phylogenomic and dating analyses were conducted to infer the divergence time for the methanogens. In contrast to previous studies that considered almost all ribosomal markers, the analyses done here were based on a restricted number of ribosomal markers to minimize potential phylogenetic bias caused by horizontal gene transfer. The 22 ribosomal markers included in our analyses were considered to be inherited vertically across archaeal species (Matte-Tailliez et al., 2002).

In general, both the Maximum-likelihood (ML) and BI trees exhibited identical topologies at the order level (Figure 3a). Consistent with previous studies, the Class I methanogens formed a monophyletic group (Bapteste et al., 2005; Anderson et al., 2009; Forterre, 2015; Petitjean et al., 2015). This cluster was strongly supported by the BI trees, but with weak bootstrap supports by the ML trees. As already documented in previous studies (Forterre, 2015; Petitjean et al., 2015), the uncertainties for the ML tree may stem from the exact placement for each of the three orders within the Class I, owing to the unstable position of Methanopyrales. It could either form a sister lineage with the Methanobacteriales–Methanococcales cluster (Bapteste et al., 2005; Anderson et al., 2009; Petitjean et al., 2015), or with Methanobacteriales to the exclusion of Methanococcales (Forterre, 2015; Petitjean et al., 2015). The Class II methanogens, however, did not form a monophyletic group. Instead, Methanocellales and Methanosarcinales formed a cluster with Halobacteria to the exclusion of Methanomicrobiales (Figure 3a). Although this topology was strongly supported by the BI trees it had only moderate support by the ML trees (Figure 3a). About 30% of the remaining ML trees either grouped the Class II methanogens into a monophyletic group (Figure 3b), to the exclusion of Halobacteria (Andam and Gogarten, 2011; Yutin et al., 2012; Sorokin et al., 2017) or recovered two monophyletic groups (Figure 3c), containing Methanosarcinales–Methanocellales and Methanomicrobiales–Halobacteria, respectively (Brochier-Armanet et al., 2011; Petitjean et al., 2015; Raymann et al., 2015).

Figure 3
figure 3

(a) Maximum-likelihood (ML) tree modeled with either ‘WAG+G (4)+I’ or ‘LG+G (4)+I’ matrix based on a concatenation of 22 ribosomal proteins sampled from methanogens Methanopyrales, Methanobacteriales, Methanococcales, Methanomassiliicoccales, Methanomicrobiales, Halobacteria, Methanocellales and Methanosarcinales, with other euryarchaeal sequences from Archaeoglobi, Thermoplasmatales, Thermococci and Thermoprotei as references (Supplementary Table 1). The tree was rooted by Nanoarchaeum equitans. The ribosomal proteins include Rpl2p, Rpl15p, Rpl18p, Rpl22p, Rpl23p, Rpl30p, Rpl3p, Rpl44e, Rpl4p, Rps10p, Rps13, Rps15p, Rps17e, Rps19e, Rps19p, Rps2p, Rps3p, Rps4p, Rps5p, Rps6e and Rps7p. Node values indicate bootstrap supports (first/second value for WAG/LG) for the ML tree and posterior probabilities (third/fourth value for WAG/LG) for the Bayesian tree, respectively. The bar represents the number of changes per sequence position out of a total of 3067 positions. The same tree topology on the order level was reproduced by Bayesian inference with MrBayes. Nevertheless, alternative ML tree topologies shown in (b) and (c) were also found, and the three tree topologies at the class level differed only at the phylogenetic position of Halobacteria. Shown in Table 5, time estimates for the indicated nodes (open square) were inferred for all the three tree topologies with internal calibration points (open diamond) derived from previous studies (Battistuzii and Hedges, 2009, Marin et al., 2017).

Full size image

All three topologies were subjected to molecular dating, calibrated by three divergence time points reported previously (Battistuzii and Hedges, 2009) and re-used in a recent study (Battistuzii and Hedges, 2009; Marin et al., 2017). Those divergence time points were initially derived using a Bayesian timing method with two constraints (Battistuzii and Hedges, 2009), a minimum of 3.46 Ga for the origin of methanogenesis evidenced by isotopically light carbon (Ueno et al., 2006), and a maximum of 4.20 Ga divergence within Archaea inferred by the midpoint of the time range estimated for the last ocean-vaporizing event (Sleep et al., 1989). It was found that ancestors of the Class I methanogens first branched off the euryarchaeal tree at 3.4–3.3 Ga (nodes c1–c3) and then started to diverge into the three orders at 3.2–3.1 Ga (nodes b1–b3) (Table 5). Despite the uncertain topologies regarding the Class II methanogens, their divergence time estimates remained relatively stable, that is, they branched off the euryarchaeal tree at 2.3–2.2 Ga (nodes f1–f3). Likewise, their further divergence into Methanocellales and Methanosarcinales remained almost unchanged at ~2.0 Ga (nodes h1–h3). Depending on its phylogenetic placements, Halobacteria diverged from the Class II methanogens from 2.4–2.1 Ga (nodes g1–g3). Although using a restricted ribosomal proteins’ data set did not resolve the uncertainties regarding the exact phylogenetic positions of the Class II methanogens, the results were able to provide time estimates indicating that the divergence of the Class I and Class II methanogens might have occurred before and around the time of the Great Oxygenation Event, respectively.

Table 5 Divergence times (Ma) and confidence intervals (CI) among the euryarchaea
Full size table

Conclusions

In this study, comparative genomic analyses and environmental sequences surveys identified two distinct methanogen clusters, corresponding to the Class I and Class II methanogens, respectively, as proposed previously based on phylogenomic studies (Brochier-Armanet et al., 2011; Petitjean et al., 2015). The results in the present study suggested that the Class I methanogens had limited abilities for oxidative adaptation. In contrast, the Class II methanogens possessed expanded genomic features potentially enabling better adaptation to oxidative environments. Three general evolutionary mechanisms towards a stronger tolerance to oxidative stress for methanogens at the class level could be envisaged from our analyses. The first is the usage of enzymes producing less ROS or enhancing ROS removal in largely unchanged pathways, which has been observed in the evolution of electron bifurcation-based hydrogenotrophic methanogenesis and O2/ROS elimination pathways. The second is to employ a completely new pathway resulting in less ROS production, which has been found in the shift from an electron bifurcation to a cytochrome/methanophenazine based methanogenesis pathway. The third is the expansion and diversification upon a core antioxidant system, which has been seen in the O2/ROS elimination and self-repairing pathways.

Our phylogenomic and dating analyses further suggested that the Class I methanogens might have evolved before the Great Oxygenation Event. The Class II methanogens, on the other hand, might have evolved later, possibly expanding their antioxidant features as a response to the abruptly elevated oxidative stress. This would not only protect them from an increasingly oxidized Earth but enabled access to nickel available in the oxidized layer during the ‘nickel famine’ that began at ~2.5 Ga (Konhauser et al., 2009). Alternatively, it is possible that the metabolic changes between the two methanogen classes observed here were due to recent niche specialization rather than an ancient adaptation to oxygenation. However, this would seem counterintuitive given that the lately evolved Class II methanogens mainly modified or expanded upon the many antioxidant features already present in the early evolved Class I methanogens, which comprise many hyperthermophilic organisms that thrive in habitats resembling those environments where microbial ancestors first arose (Boussau et al., 2008; Liu and Whitman, 2008). Nevertheless, studies on ancient gene reconstruction may shed more light on the evolutionary history of both methanogens and their antioxidant systems. On the other hand, discoveries of new methanogen lineages will greatly contribute to this effort, especially those that possess ancient metabolisms. Indeed, new methanogens are continued to be found, including but perhaps not limited to the seventh methanogen order Methanomassiliicoccales (Dridi et al., 2012; Borrel et al., 2013), the Candidatus ‘Methanofastidiosum’ belonging to the WSA2 euryarchaeal class (Riviere et al., 2009; Nobu et al., 2016), the Candidatus ‘Methanonatronarchaeia' representing the halophilic SA1 euryarchaeal group (Eder et al., 2002; Sorokin et al., 2017), and even some putative methanogens outside the Euryarchaeota (Evans et al., 2015; Vanwonterghem et al., 2016). Considering that most of those newly discovered methanogens appear to be methylotrophic, it may be speculated that they evolved later since the main substrates available in ancient environments would have favored methanogenic growth on H2 and CO2 (Martin, 2011).