As the least-understood domain of the tree of life, the Archaea is a major frontier of microbiology and biogeochemistry research1,2. Known Archaea shoulder key roles in biogeochemical cycles3 by mediating unique or unusual metabolisms such as methanogenesis, anaerobic methane oxidation, anaerobic alkanotrophy, and chemolithotrophic ammonia oxidation4,5. Although some groups of Archaea can be readily cultivated in the lab, less than 0.1% of the known species diversity of Archaea has been described as pure cultures6. The explosion of archaeal metagenome-assembled genomes (MAGs) in recent decades has greatly expanded our scope of view of the true diversity of Archaea, but still more than half of archaeal taxonomic ‘blind spots’ have not yet been described carefully in the literature and await ‘discovery’7.

Due to their physicochemical extremes—especially temperature and pH - geothermal ecosystems are rich in extremophilic Archaea, while more moderate habitats within geothermal areas are conducive to less extremophilic Archaea. Prokaryotes and their viruses dominate in high-temperature geothermal ecosystems because no known eukaryotes can grow above ~65 °C; and within these systems, Archaea are known to dominate at the high-temperature and low pH extremes, since no known Bacteria can grow above 95 °C8 or below pH 29,10. Physicochemical extremes are very common in terrestrial geothermal ecosystems because subsurface phase separation and varying degrees of mixing with shallow water can lead to wide variations in spring temperature and chemistry11. In these systems, acidic pools represent vapor-dominated end-members along with sulfur-depositing fumaroles; the low pH of these systems derives from the enrichment of hydrogen sulfide in the vapor phase, followed by oxidation of the hydrogen sulfide to sulfuric acid12. Alkaline springs represent water-dominated end-members and are characterized by high content of total dissolved solids, while high-temperature, high-pH spring sources are significantly understudied11,13.

One magmatic hydrothermal region that harbors springs with such physicochemical extremes is the Tengchong geothermal spring system in Southwest China, which belongs to the Mediterranean-Himalayan Geothermal Belt that resulted from collision of the Indian Plate and the Eurasian Plate. Pronounced variations regarding the physicochemical parameters have been observed among different geothermal springs, with reported temperatures up to 96 °C and pH values from ≤1.8 to ≥9.311,14,15. Despite decades of microbiology research at Tengchong geothermal springs, very few archaeal isolates have been described, and all belong to the genera Sulfolobus, Acidianus, and Metallosphaera16,17,18,19, within the order Sulfolobales. Recently, metagenomics has uncovered the potential metabolisms of some yet-uncultivated archaeal lineages in Tengchong geothermal springs, including chemolithotrophic Caldarchaeales20,21, symbiotic Aenigmatarchaeota22 and Panguiarchaeales23, carbohydrate-metabolizing Bathyarchaeia24, ammonia-oxidizing Nitrosocaldaceae25, anaerobic methylotrophic class EX4484-20526, and several lineages potentially involved in methane/alkane metabolism27.

However, while these studies provided important insights into specific lineages of Archaea, they did not provide a high-level view into the full diversity and potential functionality of Archaea as a whole, or the forces controlling their diversity and distribution. Here, we applied deep metagenomic and metatranscriptomic sequencing to diverse sediment samples from a variety of geothermal springs in Tengchong. A total of 152 metagenomes were acquired from 48 springs over a period of 6 years, and nearly 3000 archaeal MAGs were successfully reconstructed. By analyzing these MAGs and focusing on highly transcribed genes from metatranscriptomes, we describe the diversity, distribution, and potential roles of Archaea in the biogeochemical cycling of carbon, sulfur, nitrogen, and hydrogen. Our study expands the global cache of archaeal genomes and illustrates their potential ecological roles and interactions in terrestrial geothermal ecosystems.


Nearly 3000 archaeal genomes adequately represent the Archaea diversity in Tengchong geothermal springs

A total of 152 sediment samples covering 48 geothermal pools/streams located in Tengchong County (Yunnan Province, China) and spanning six years from 2016 to 2021 were collected and metagenomically sequenced (Supplementary Data 1). A comprehensive investigation of archaeal diversity was conducted by initially scanning for genes encoding ribosomal protein S3 (rpS3). Using this procedure, we successfully identified 27,436 archaeal and bacterial rpS3 sequences, which clustered into 8757 representative rpS3 sequences with 99% as the sequence identity cutoff (see Methods). Taxonomic assignments based on the Genome Taxonomy Database (GTDB) revealed that archaeal rpS3 genes could be classified into 13 phyla, 29 classes, and 64 orders (Supplementary Fig. 1 and Supplementary Data 2). Further bioinformatic analysis resulted in the construction of 2949 archaeal MAGs meeting currently accepted criteria as medium to high quality28, with a median of 89.6% completeness and 0.8% contamination (Supplementary Fig. 2, Supplementary Data 3 and 4). Moreover, these MAGs had a median of 70 scaffolds, and over 90% (2,678) of the MAGs had ≤200 scaffolds. The quantity of our reconstructed MAGs is equivalent to about half of the existing archaeal genomes in the GTDB (total 6062 Archaea genomes in GTDB v.207 until Jan. 2023)29. Although approximately 10% of the archaeal MAGs did not contain rpS3 (Fig. 1c), the other 90% of the MAGs represent 40% of the total archaeal representative rpS3 by number, and importantly represent 82% of the archaeal representative rpS3 when they were weighted by sequence coverage in the metagenomes (Fig. 1d). Also, according to the GTDB nomenclature, these MAGs encompass 12 of the 13 phyla and 25 of the 29 classes detected, and the taxonomy of the MAGs is highly consistent with that derived by the representative rpS3 (Fig. 1b, Supplementary Fig. 1, Supplementary Data 5 and 6). Overall, these results show that the 2,949 archaeal MAGs recovered in our study adequately represent the archaeal community in the Tengchong geothermal spring system.

Fig. 1: Archaeal diversity in 152 geothermal spring metagenomes.
figure 1

a Composition and relative abundance of archaeal classes in microbial community based on rpS3 gene sequences. Detailed information including the relative abundance of each rpS3 gene in each microbial community is reported in Supplementary Data 2. b Composition and relative abundance of archaeal classes in archaeal community based on rMAGs. Colors indicate archaeal classes: green series, DPANN superphylum; blue series, Thermoproteota; red series, Asgardarchaeota. c The percent of archaeal MAGs harboring rpS3 genes among all 2949 archaeal MAGs. d The percent of rpS3 genes associated with archaeal MAGs normalized to the total archaeal community based on count and coverage of rpS3 genes.

As expected, the composition of Archaea varied greatly across samples (Fig. 1a, b and Supplementary Fig. 1). The average and maximum abundance of Archaea based on reads mapping to MAGs was 34.7% and 95.2%, respectively, and 54 samples were dominated by Archaea (>50% relative abundance; Supplementary Data 6). This is further supported by qPCR experiments conducted on several recent samples collected from geothermal spring sediments (Supplementary Table 1). In agreement with many previous thermal spring surveys30,31, we found that Thermoplasmata and Thermoproteia were the most abundant lineages across our sites (Fig. 1a, b). Thermoplasmata dominated (maximum abundance, 70.2%) the DRTY pools, which are acidic (pH <5.5), followed by Thermoproteia, Nitrososphaeria, and Bathyarchaeia, as well as Micrarchaeia and Nanoarchaeia, which has recently been validated as Nanobdellia32 (Fig. 1b, Supplementary Data 1 and 6). However, Thermoproteia dominated in most other samples, including both acidic springs (some DRTY, ZZQ pools) and alkaline springs (pH ≥ 8.5; including ZMQR/L, SRBZ, GMQS/P pools), with a maximum abundance of 93.7%. The composition of other Archaea in Thermoproteia-dominant communities in acidic springs was similar to those in Thermoplasmata-dominated acidic springs, whereas in alkaline springs other Archaea included Nitrososphaeria_A, Aenigmatarchaeia, Methanomethylicia, Nanoarchaeia (Nanobdellia), and Archaeoglobi. The remaining springs (such as JZ, GXS, QQ, HNT pools) contained various Archaea that together contributed up to 45% of the read abundance. Given the high abundance of Archaea in the Tengchong spring system, we further investigated their diversity and metabolic potential.

MAGs are diverse and represent hundreds of unknown species

To better explore the diversity of these Archaea, the 2949 archaeal MAGs were dereplicated at the strain level into 603 representative MAGs (rMAGs) based on 99% average nucleotide identity (Supplementary Data 3). The rMAGs represent the majority of archaeal phyla (12 of 18, 66.7%), as defined in the GTDB, and 27 classes, 67 orders, 147 families, 265 genera, and 475 species (Fig. 2). They comprise the first genomic representatives of four orders (within Micrarchaeia, Korarchaeia, and Bathyarchaeia), 26 families, and 131 genera. Additionally, they include one phylum (B1Sed10-29), three classes (belonging to Halobacteriota, Thermoplasmatota, Thermoproteota), 30 orders, 90 families, and 107 genera that defined in GTDB without nomenclature and effective descriptions. Of the 475 species represented by the rMAGs, 82.5% do not have a previous species representative in GTDB, and another 13.9% do not have a nomenclature. Thus, the Archaea in Tengchong geothermal springs are poorly explored and they are of great research value.

Fig. 2: Newly identified archaeal species are broadly distributed across the Archaea.
figure 2

a Phylogenetic tree of 603 archaeal rMAGs recovered from Tengchong geothermal springs. Taxonomic labels are based on the GTDB. Triangles behind taxonomic labels indicate the order/class existed in GTDB but without nomenclature; stars behind taxonomic labels indicate the order is newly discovered in this study. Red pie charts indicate the prevalence of each order across 152 samples. Orange and yellow pie charts indicate the maximum relative abundance of each order in microbial and archaeal communities, respectively. Dark blue and light blue columns indicate the numbers of archaeal MAGs and rMAGs contained in the corresponding order, respectively. Purple stacked histograms indicate the number of families, genera, and species that are newly identified (light part) and existing but without nomenclature (dark part). Purple pie charts indicate the percentage of newly identified species among all species in each order. The relative abundance of each order in microbial communities is available in Supplementary Data 6. The generated tree in newick format is provided in FigShare repository at b Number and percentage of taxa with newly identified and existing but not formally named at each taxonomic level for 603 archaeal rMAGs.

The identified rMAGs are taxonomically broadly distributed (Fig. 2). In particular, they span a broad diversity within the Thermoproteota (7/8 classes) and several phyla known for their small cell size, small genome size, and episymbiotic lifestyles (Altiarchaeota, Iainarchaeota, B1Sed10-29, Micrarchaeota, Aenigmatarchaeota and Nanoarchaeota (Nanobdellota)), with hotspots of diversity and novelty in the orders B26-1 (Bathyarchaeia), Sulfolobales, Caldarchaeales, and CG10238-14 (within Aenigmatarchaeia). Further, many Thermoproteota orders were both prevalent and abundant across the 152 samples, especially Sulfolobales, Thermofilales, Thermoproteales, Caldarchaeales, Nitrososphaerales, and B26-1, suggesting their ecological importance. Small Archaea known or interpreted to be episymbionts were widely distributed, but at low abundance. Only Micrarchaeales, CABMEP01 (within Micrarchaeia), CG10238-14, Nanoarchaeales (Nanobdelalles), and Parvarchaeales had relative abundances greater than 1% in at least one metagenome (Supplementary Data 6). By contrast, fewer than 10 rMAGs were assigned to orders within Thermoplasmatota, Hadarchaeota, Halobacteriota, Methanobacteriota, and Asgardarchaeota, except for the more diverse orders Archaeoglobales, Thermoplasmatales, and Hadarchaeales. It is noteworthy that ARK-15 (Thermoplasmata order), with only two rMAGs, was abundant (up to 66%) in some of the metagenomes. In addition to abundant species, these rMAGs also represent diverse low-abundance species with several representatives, further confirming this collection of genomes represents the diversity of the Archaea community well. In summary, the identified MAGs in geothermal springs not only improve genomic representatives within several archaeal lineages, but also substantially increase the known archaeal phylogenetic diversity across the tree of life.

pH and temperature are key drivers of archaeal composition and potential function

By applying a generalized linear model (GLM), we found that microbial communities correlated with a variety of environmental factors (Supplementary Table 2). Among them, pH and temperature emerged as the two strongest correlates and they most likely affect the archaeal community at both species (χ2 = 40.8 and 29.2, respectively, both p < 0.001) and functional (χ2 = 46 and 34.2, respectively, both p < 0.001) levels. Sulfate and nitrite were weaker correlates at both species (χ2 = 8.42 and 3.84, respectively, both p < 0.05) and functional (χ2 = 8.18 and 4.00, respectively, both p < 0.05) composition of Archaea. Our analyses of both taxonomic composition and potential metabolic functions showed remarkable variation according to pH (Supplementary Fig. 3a-d). Statistical differences among MAGs present in acidic (pH <5.5), neutral (5.5 ≤ pH <8.5) and alkaline (pH ≥ 8.5)33 springs were evident (ANOSIM test, R = 0.807 and p = 0.001 for rMAGs; R = 0.638 and p = 0.001 for KOs). Additionally, grouping samples by temperature revealed significant differences among mesothermal (<60 °C), thermal (60–80 °C), and hyperthermal (≥80 °C)34 communities (ANOSIM test, R = 0.275 and p = 0.001 for rMAGs; R = 0.255 and p = 0.001 for KOs; Supplementary Fig. 3f, h), suggesting temperature is a secondary factor that also influences the taxonomic and functional diversity of geothermal spring archaeal communities. The composition and potential function of Archaea from thermal communities was similar to those from mesothermal communities but vastly different from those hyperthermal springs based on both the ordination and correlation coefficient (Supplementary Fig. 3e–h). There was no significant difference in archaeal communities when grouped by sampling season, despite previous reports of seasonal effects on chemistry and community composition in Tengchong springs35 (Supplementary Fig. 3i–l).

Archaea are more abundant under extreme pH conditions and show a strongly bimodal distribution

Archaeal community composition varied significantly between acidic, neutral, and alkaline springs. Additionally, correlation analyses reveled strong relationships between the relative abundance of Archaea and pH (Supplementary Fig. 4a, b). When the springs were divided into acidic or alkaline groups, both extremes of pH favored archaeal abundance, with stronger relationships in acidic springs (Pearson test, R = −0.76, p = 5.6 × 10−14) than alkaline springs (Pearson test, R = 0.57, p = 2.7 × 10−7). Quantitative data obtained through qPCR also indicated that in acidic springs, the absolute abundances of Archaea was greater than those of Bacteria (Supplementary Table 1). This trend is consistent with the known competitive advantage of Archaea over Bacteria at several physicochemical extremes36. However, in alkaline springs, both pH (Pearson test, R = 0.55, p = 5.5 × 10−7) and archaeal abundance (Pearson test, R = 0.67, p = 1.1 × 10−10) correlated with temperature (Supplementary Fig. 4c–f). Further, analysis of the relationship between pH and abundance of the 20 most abundant archaeal orders showed Thermoproteia and Thermoplasmata dominating, but with distinct patterning by pH (Supplementary Fig. 5). Sulfolobales dominated in many acidic springs. For example, Sulfolobaceae dominated in DRTY-4, −5, −10 and ZZQ pools (Supplementary Fig. 56 and Supplementary Data 6), consistent with its abundance in other acidic thermal springs37,38. Sulfolobales were also abundant in some alkaline springs. For example, Desulfurococcaceae were abundant in ZMQL/R and GMQP pools, with a maximum relative abundance of 87%. Two Thermoproteia orders, Thermoproteales and Thermofilales, were more abundant in alkaline springs, as well as Nezhaarchaeales. In contrast, two orders of Thermoplasmata, ARK-15 and Thermoplasmatales, were enriched in acidic springs, with maximum abundance of 66% and 34%, as well as other acidophilic lineages like Micrarchaeia, Parvarchaeales, JACIWG01, and Conexivisphaerales. Several lineages with lower abundance were more abundant in neutral springs, such as Caldarchaeales, B26-1, Nitrososphaerales, and EX4484-205, whose maximum abundances are 15%, 14%, 8% and 9%. We also observed that some widespread small Archaea, including Micrarchaeales, Parvarchaeales, and CABMEP01, existed exclusively in acidic springs, most likely because of their symbiotic/parasitic relationships with acidophilic hosts39,40,41 (Supplementary Fig. 7). Overall, the archaeal communities showed a selective enrichment in response to pH.

Thermoproteota are likely key players in the carbon cycle

An analysis of potential carbon metabolisms encoded by rMAGs within 12 representative springs that were sampled more than four times was conducted to investigate metabolic potential (Fig. 3). Given the strong relationships between pH and potential archaeal functions (Supplementary Fig. 3c, d), the pattern of potential autotrophic pathways according to pH was examined. In neutral springs, the Wood-Ljungdahl pathway dominated, and B26-1 was the largest archaeal contributor (Figs. 3a, d, 4). Thermoproteales endoding a complete set of genes involved in the dicarboxylate-hydroxybutyrate (DC/4-HB) cycle including the key enzymes acetyl-CoA/propionyl-CoA carboxylase (K01964, K15036, K15037) were abundant in both acidic and alkaline springs (Supplementary Data 7). Additionally, in acidic springs, members of Sulfolobales were abundant and encoded either the 3-hydroxypropionate/4-hydroxybutyrate (3HP/4HB; with pyruvate synthase and PEP carboxylase being the key enzymes (K01007, K01595)) or DC/4-HB cycles, as supported by the detection of a complete set of genes for both pathways. Surprisingly, some members of the Sulfolobales were highly abundant in alkaline springs but only possessed the 3HP/4HB cycle, suggesting that 3HP/4HB might also be an important pathway for carbon fixation in such environments. Genes encoding potential for methane metabolism, notably methyl-coenzyme M reductase (mcrABG), were restricted to low-abundance orders, such as putative methanogens within the Archaeoglobales, Methanotrichales, Methanomassiliicoccales, Nezhaarchaeales, Korarchaeales, Methanomethylicales, Nitrososphaerales and putative methanotrophs/alkanotrophs in the Hadarchaeales (Fig. 4), which were described in detail previously27. Even though some members of the Archaeoglobales and Nezhaarchaeales in Tengchong and other terrestrial spring systems have recently been shown to be methanogenic42,43, the number of transcripts mapping to mcrABG in our datasets were low.

Fig. 3: Metabolic profile of archaeal communities under different pH conditions.
figure 3

Schematic diagrams of the carbon and hydrogen cycle (a), sulfur cycle (b), and nitrogen cycle (c) within the archaeal communities in geothermal springs. Legends for each cycle are shown on the left; gray lines indicate pathways that were not detected. The panels on the right indicate the reactions encoded by the archaeal communities in the different springs, and the thickness of the line indicates the relative abundance of Archaea putatively involved in the reaction. The geothermal springs are grouped by pH: the three on the upper left are acidic, the three on the upper right are alkaline, and the five on the lower are neutral. Only the geothermal springs that were sampled ≥ 4 times are shown, and the values of relative abundance represent the average. d Metabolic weight score (MW-score; see Methods) of archaeal communities in different geothermal springs.

Fig. 4: Metabolic potential of archaeal orders in the geothermal spring metagenomes.
figure 4

Black circles indicate the existence of a complete pathway, that is, all genes in the pathway are present. Gray circles indicate the existence of an incomplete pathway. White circles indicate the absence of any genes in the pathway. For each archaeal order, pathways containing all genes in at least one rMAG are considered representative of the order. The counts of functional genes for each MAG are summarized in Supplementary Data 7.

Acetate is a prevalent electron donor and carbon source in both oxic and anoxic environments, and species-specific uptake of acetate was recently studied in a Tengchong spring44. Furthermore, acetate oxidation with low potential electron acceptors, such as partially oxidized sulfur compounds, can be an important intersection of the carbon cycle and other element cycles45. Genes for acetate oxidation were observed in most springs (average 8–60% coverage except DRTY-3 and DRTY-9; Fig. 3a). Particularly, the acetyl-CoA synthetase (acs) gene was detected among many archaeal orders, likely functioning in acetate oxidation to acetyl-CoA (Fig. 4 and Supplementary Data 7). The importance of acetate oxidation was further supported by metatranscriptomic sequencing, with reads mapping to acs in many springs (Fig. 5a). In addition to acetate oxidation, archaeal genes related to fermentation were abundant across all representative springs (over 53% coverage) further confirming the presence of a baseline population of heterotrophic Archaea (Fig. 3a). The highly expressed pyruvate:ferredoxin oxidoreductase (por) and related poorly characterized tungsten-dependent aldehyde ferredoxin oxidoreductases (aor) are important for both sugar and amino acid fermentations and were recently shown to have proliferated in the Caldarchaeales family Wolframiiraptoraceae in Tengchong and other geothermal ecosystems21 (Fig. 5a). Ferredoxins Fer primarily facilitating the transfer of electrons from glucose and/or amino acids to oxidoreductases were commonly detected across multiple orders within Thermoproteota46 (Supplementary Data 7). Additionally, the widespread existence and expression of acetate-CoA ligases (acdAB; Fig. 4 and Supplementary Data 8) indicates that many Archaea can oxidize acetyl-CoA for energy conservation. However, when carbon sources become limited or acetate becomes the sole carbon source, these Archaea can utilize the reverse ACD and POR enzymes to assimilate acetate as a carbon substrate45,47,48. Interestingly, another acetogenesis pathway, the phosphate acetyltransferase (pta)-acetate kinase (ack) pathway, was also detected in small Archaea (e.g., Pacearchaeales and Woesearchaeales), Asgardarchaeota, and Thermoproteota. The pta-ack pathway is thought to be a bacterial-type acetogenesis pathway, although it has been found in a few Archaea49,50. Our discovery of widespread pta-ack pathway enzymes suggests that this so-called Bacteria-specific pathway is more widely distributed in Archaea than previously thought.

Fig. 5: Metatranscriptomic read mapping to archaeal MAGs in geothermal springs.
figure 5

Top 10 highest expressed genes associated with carbon metabolism (a), sulfur metabolism (b), and nitrogen metabolism (c) in the four reference springs. Cells in gray indicate either the archaeal genomes contained no ortholog or no expression of the ortholog is detected. d Percent relative abundance of metatranscriptomic reads mapping to the top five archaeal orders in each reference spring. The top 100 transcriptional rank of functional genes of rMAGs from the most abundant orders in the reference springs, including DRTY-16_202007_bins_21 (ARK-15; e) and DRTY-16_202007_bins_2 (Thermoplasmatales; f) in DRTY-16 pool, JZ-2_201912_bins_104 (Nitrososphaerales; g) in JZ-2 pool, JZ-1_201705_bins_112 (Nitrososphaerales; h) in QQ pool, DRC_201705_bins_79 (Sulfolobales; i) and DRC_201705_bins_77 (Thermoproteales; j) in JZ-3 pool. FPKM: Fragments Per Kilobase of transcript per Million.

The dominance of rMAGs coding for organic carbon metabolism (over 92% coverage) across all sites suggests the presence of a “baseline” of heterotrophic Archaea (Fig. 3a). However, genes related to heterotrophic processes were distinct in the three pH groups, likely indicating preferences for distinct organic carbon substrates (Fig. 3d), mirroring differences in dissolved organic carbon pools in acidic and alkaline springs in Tengchong and elsewhere51. For example, Archaea in neutral springs tended to encode more families of carbohydrate-active enzymes (CAZymes) than those in acidic springs, even though they contained fewer CAZymes per rMAG (Supplementary Fig. 8). Those in alkaline springs harbored a similar number of CAZymes families as those in neutral springs, but they had more CAZymes per rMAG. Archaea in acidic springs encoded fewer CAZymes per rMAG and fewer CAZyme families, and polysaccharide lyases were nearly absent. Although many acidic springs are known to accumulate terrestrial organic carbon51, these patterns suggest that polysaccharides may be acid-hydrolyzed naturally in acidic springs, rendering CAZymes less important. We note that order EX4484-205, also known as Brockarchaeota, is a major contributor to CAZymes in the circumneutral springs, consistent with previous reports26 (Supplementary Fig. 9), suggesting that EX4484-205 may play an important role in degrading diverse complex organics in neutral springs. Similarly, several orders of Thermoproteia contained high numbers of CAZymes in both acidic and alkaline springs. Sulfolobales CAZymes associated with the metabolism of amino sugars and Thermoproteales/Thermofilales CAZymes associated with starch/hemicellulose degradation were both highly expressed in the alkaline JZ-3 spring (Supplementary Data 9). Additionally, in acidic DRTY-16 spring, starch-degrading CAZymes encoded by Thermoplasmatales and ARK-15 were expressed. The potential carbon metabolisms encoded by archaeal communities in acidic springs showed two distinct patterns (Fig. 3d). In some pools, notably DRTY-3 and DRTY-9 (DRTY-3/-9), the heterotrophic “baseline” dominated, while in others, notably DRTY-5 and DRTY-7 (DRTY-5/-7), Archaea encoded those functions plus carbon fixation and acetate oxidation. This difference correlated with the high abundance of Thermoproteia in DRTY-5/-7 (Fig. 1), suggesting these Thermoproteia may be important for those functions.

Potential for dissimilatory sulfate reduction is enriched in polyextreme geothermal springs mainly by Thermoplasmata in acidic springs and Thermoproteales in alkaline springs

For Archaea to thrive in extreme habitats, it would be opportune to diversify to exploit diverse energy sources. Genes supporting sulfite oxidation and sulfate reduction were ubiquitous (Fig. 3b, d), and the key gene sulfate adenylyltransferase (sat) was expressed in all four metatranscriptomic datasets (Fig. 5b). Both metagenomic and metatranscriptomic analyses showed the prevalence of the dissimilatory sulfite reductase (dsrAB) and adenylylsulfate reductase (aprAB) gene clusters across both acidic and alkaline springs, suggesting the importance of dissimilatory sulfate reduction (DSR) (Figs. 3b,  5b), although very few measurements of DSR in high-temperature terrestrial geothermal ecosystems have been reported52. In acidic geothermal springs, DSR was likely mediated by the most abundant Archaea, ARK-15, whereas the high expression of both dsrAB and aprAB in some Thermoproteales suggested distinct sulfate-reducing Archaea in neutral and alkaline springs (Fig. 5j). Notably, the most abundant ARK-15 rMAG lacks aprAB genes, indicating a potential dependency on partners to provide sulfite. The sulfide derived from DSR or from the spring sources themselves12 could be oxidized using sulfide:quinone oxidoreductase (sqr) to form zero-valent sulfur, a crucial intermediate in the biogeochemical sulfur cycle that is widespread in terrestrial and marine geothermal ecosystems53. In acidic springs, many rMAGs within the orders ARK-15 and Thermoplasmatales possessed these genes and their transcripts were also abundant; likewise, many rMAGs encoding these genes belonged to the Thermoproteales in alkaline springs (Fig. 5 and Supplementary Data 8). In contrast to these dissimilatory pathways in the polyextreme springs, Archaea in neutral springs expressed the suite of genes necessary for assimilatory sulfate reduction (Fig. 5b). Additionally, the broad occurrence of sulfur dioxygenases (sdo) suggested chemolithotrophic sulfur oxidation in neutral springs, especially by B26-1, Caldarchaeales, Nitrososphaerales, and EX4484-205 (Figs. 3d, 4). These sdo genes were also highly expressed in neutral JZ-2 pool and predominantly map to Nitrososphaerales (Supplementary Data 8). We also noted a high number of transcriptional reads mapping to thiosulfate sulfurtransferase (sseA) for thiosulfate oxidation to form sulfite in neutral and alkaline pools, and tetrathionate reductase (ttrAB), for reduction of tetrathionate to thiosulfate, in alkaline JZ-3 pool. In summary, both metagenomic and metatranscriptomic data suggested an active dissimilatory sulfur cycle mediated by Archaea in acidic pools, which may be driven by the concentration of sulfur in the vapor phase initially as hydrogen sulfide. The data suggest that hydrogen sulfide can be oxidized by Archaea, and the resulting sulfate is used by DSR. In alkaline springs, the data suggest that Archaea can oxidize reduced sulfur and then utilize various sulfur compounds including tetrathionate, thiosulfate, sulfate, and sulfite to conduct DSR or form zero-valent sulfur.

An interesting and unique case is found in the DRTY-6 pool. The pH of DRTY-6 pool gradually decreased through time from neutral to acidic, accompanied by an increased abundance of Archaea, among which the class Thermoplasmata was dominant (Fig. 1a and Supplementary Fig. 10). Meanwhile, we observed a gradual increase in DSR pathway genes. This transformation reinforces the link between Thermoplasmata and the DSR pathway as a response to the enrichment of sulfate in acidic springs. Surprisingly, the abundance of archaeal genes for sulfur oxidation decreased as DRTY-6 acidified through time. We also noticed a gradual change in potential carbon metabolisms among the archaeal community; for example, the abundance of genes for inorganic carbon fixation, acetate oxidation, and hydrogen metabolism decreased. This is consistent with a decline in the relative abundance of Thermoproteia. Combining these trends with decreases in carbohydrate and hydrogen metabolism, we surmise that the acidification of DRTY-6 may have led to a shift in the predominant archaeal pathways for energy conservation, with more genes related to DSR rather than sulfur oxidation.

Multiple strategies for utilization of nitrogen resources in N-limited geothermal springs

According to the GLM results mentioned above, nitrite significantly correlated with the archaeal species and functional composition in geothermal springs (Supplementary Table 2). However, the abundance of genes for dissimilatory nitrogen metabolisms in Archaea in Tengchong springs were limited relative to carbon and sulfur, which is consistent with previous studies24,54 (Fig. 3c, d). The concentrations of dissolved inorganic nitrogen are low in most neutral and alkaline springs35. Nitrogen is supplied to many terrestrial geothermal springs as ammonia55, which can be oxidized to nitrite by thermophilic ammonia-oxidizing Archaea (AOA)56,57. We detected a high number of transcriptional reads mapping to ammonia monooxygenase (amoCAB), suggesting active ammonia oxidation by AOA (Fig. 5 and Supplementary Data 8). Both amoCAB and nitrite reductase (nirK) were highly expressed by MAGs assigned to Nitrososphaerales inhabiting neutral and alkaline springs. The high expression of nirK by Nitrososphaerales suggests that the greenhouse gas N2O may be released by nitrifier denitrification (Fig. 5c). Ammonia oxidation to nitrite or nitrate could provide substrates for assimilatory or dissimilatory metabolisms. The archaeal communities in these springs encode more genes associated with dissimilatory nitrate reduction (DNR) than assimilatory nitrate reduction (Fig. 4), suggesting AOA provide sufficient oxidized nitrogen to drive denitrification, although in other geothermal ecosystems, the rate of denitrification is limited by the nitrate/nitrite supply56. Despite the expression of genes involved in the DNR pathway in DRTY-16, the complete lack of expression of amoCAB suggests that the dominance of ammonium over ammonia may inhibit ammonia oxidation at low pH (Fig. 5c, Supplementary Data 8). Community-level metabolic reconstructions revealed that incomplete denitrification may be an important archaeal pathway particularly in neutral springs (Fig. 3c, d).

In addition, Archaea living in geothermal springs may obtain nitrogen from organic molecules. ARK-15 rMAGs expressed formamidase (E3.5.1.49) in acidic springs, which suggests transformation of methanamide to ammonia (Fig. 5c and Supplementary Data 8). Some Sulfolobales (Desulfurococcaceae members) in neutral and alkaline springs expressed nitronate monooxygenase (ncd2), which converts nitroalkane into nitrite. Urease (ureAB) transcripts mapping to Nitrososphaerales suggests they may produce ammonia from urea in JZ-2 pool. Surprisingly, ureAB were also highly expressed in alkaline JZ-3 pool by members of the Jordarchaeales, suggesting broad use of urea in geothermal ecosystems. Moreover, the expression of glutamate dehydrogenase (gudB), which interconverts glutamate and ammonia, was much higher than glutamine synthetases (glnA, gltBD) that couple ammonium assimilation to glutamate synthesis (Fig. 5c). gudB was encoded by ARK-15/Thermoplasmatales and Thermoproteales from acidic DRTY-16 and alkaline JZ-3 pools, respectively (Fig. 5e, f, j).

Archaeal hydrogen metabolism is widely distributed in geothermal springs

Molecular hydrogen occurs ubiquitously in geothermal springs from both geological and biological sources, and it is a widely utilized electron donor and intercellular electron shuttle for microbial growth and survival58. The potential metabolism of hydrogen by the archaeal community in geothermal springs was different under different pH conditions (Fig. 3a, d). Hydrogenases were identified in nearly half of the archaeal rMAGs (270 out of 603, 45%; Supplementary Data 10), including [NiFe]-, [FeFe]- and [Fe]-hydrogenases (Supplementary Fig. 11). The H2-oxidizing [Fe]-hydrogenases, which are specific to Methanobacteriales4, although mcrABG were not detected in the Methanobacteriales rMAGs from this study. Several Group A [FeFe]-hydrogenases, known to be involved in coupling ferredoxin oxidation to fermentative H2 evolution, were detected in Micrarchaeia and Nanoarchaeia (Nanobdellia), as described from single-amplified genomes of Iainarchaeota previously59. However, [FeFe]- and [Fe]-hydrogenases were far less abundant compared with [NiFe]-hydrogenases.

[NiFe]-hydrogenases were the most widespread hydrogenases, occurring in 11 phyla, 22 classes, and 38 orders of Archaea (Supplementary Data 10 and Supplementary Fig. 11). This pattern mirrors the overall diversity of [NiFe]-hydrogenases, which are divided into four clades: H2-uptake Group 1 (membrane-bound) and 2 (cytosolic), bidirectional Group 3, and H2-evolving Group 459,60. Group 1a,g,j [NiFe]-hydrogenases were distributed in a few orders of Thermoproteota. These enzymes are O2-sensitive and mediate anaerobic H2-dependent sulfate, nitrate, metal, heterodisulfide, and fumarate respiration60. Group 2e [NiFe]-hydrogenases are a unique lineage of aerobic uptake hydrogenases and were only found in aerobic hydrogenotrophic Thermoproteia61. As expected, a Group 2e [NiFe]-hydrogenase was encoded by a Sulfolobales rMAG (DRTY-4_201912_bins_2) that was highly abundant in many acidic springs. Group 3 [NiFe]-hydrogenases are cytosolic bidirectional enzymes coupling with different cofactors: F420 (3a), NADP (3b), heterodisulfide reductase (3c), and NAD (3d). Group 3b and 3d [NiFe]-hydrogenases were encoded by the neutral-abundant B26-1, EX4484-205, Nitrososphaerales, and Caldarchaeales; these are O2-tolerant and are considered as redox valves that interconvert electrons between H2 and NAD(P)H depending on the availability of exogenous electron acceptors62. These Group 3b and 3d [NiFe]-hydrogenases may help these Archaea to cope with oxygen fluctuations. Group 3c [NiFe]-hydrogenases form functional complexes with heterodisulphide reductase (MvhADG-HdrABC) and perform the electron bifurcation process that couple the endergonic reduction of ferredoxin with H2 by coupling it with the exergonic reduction of heterodisulphide with H263. This hydrogenase was prevalent in Halobacteriota, Thermoplasmatota, and Thermoproteota. Moreover, the Archaea rMAG dataset encoded diverse membrane-bound H2-evolving Group 4 [NiFe]-hydrogenases, with several subgroups associated with different electron donors, namely formate (4b) and ferredoxin (4d, 4e, 4 g, 4 h, 4i). These enzymes have a respiratory function that releases electrons from low-potential donors to reduce protons, preserving the free energy liberated during electron transfer as a proton- or sodium-motive force63. The majority of Group 4 [NiFe]-hydrogenases were assigned to subgroup 4b and 4 g and were prevalent in members of Sulfolobales and Thermofilales enriched in alkaline springs, along with B26-1 in neutral springs. Additionally, ferredoxins involved in hydrogen production by facilitating electron transfer to Group 4 [NiFe]-hydrogenases were also commonly observed among Archaea46 (Supplementary Data 7). Altogether, the diverse hydrogenases likely sustain H2-based respirations, fermentations, and various electron-bifurcation and energy-conversion mechanisms in these archaeal communities.

Some Archaea may be competitive or cooperative

We focus not only on the metabolic potential of individual archaeal orders, but also attempt to explore the larger biogeochemical processes and metabolic interactions, including competitive or cooperative dynamics within the archaeal community. As mentioned above, Archaea were not abundant in the relatively temperate springs (low-temperature neutral pH springs), where most ecosystem functions are likely performed by Bacteria (Fig. 1a). Therefore, we built co-occurrence networks based on communities dominated by Archaea (>50% in relative abundance) in polyextreme (thermoacidic and thermoalkaline) springs. The networks constructed based on the investigation of rpS3 genes demonstrated that there are only few links between Archaea and Bacteria, and most of these links exhibited significant negative correlations. This suggested a potential competitive rather than cooperative relationship between Archaea and Bacteria (see microbial networks at To enhance the identification of potential beneficial functional interactions among Archaea with greater resolution and precision, we extended our analysis to include archaeal rMAGs in constructing co-occurrence networks (Fig. 6a). In thermoacidic springs, the network exhibits high phylogenetic diversity, including 14 archaeal orders. The dominant lineage, Thermoplasmata, may compete with another dominant lineage, Sulfolobales, as illustrated by negative correlations between the two cohorts (Fig. 6b). This competition may be attributed to overlapping functional niches (Fig. 6d), and the outcome of the competition is determined by temperature. When temperature exceeds 62 °C, Sulfolobales prevails over Thermoplasmata (Supplementary Fig. 12). Conversely, as the temperature decreases, Thermoplasmata gains a competitive advantage over Sulfolobales. Another factor, the high concentration of sulfate in DRTY pools, may further favor Thermoplasmata, especially the ARK-15 subgroup, members of which are capable of respiring sulfate or sulfite, providing an extra niche in the competition with other Archaea. Alternatively, some positive co-occurrence patterns suggest cooperativity (Fig. 6b, c). In thermoacidic springs, Thermoplasmatales and Conexivisphaerales may cooperate on sequential redox transformations of sulfur (Fig. 6b, d). Thermoplasmatales oxidize sulfide to elemental sulfur, which can be further oxidized into sulfite or reduced to sulfide by Conexivisphaerales. We conjecture that the partial DSR pathway of dominant ARK-15 species may require the reduction of sulfate by Thermoplasmatales and the low-abundance Conexivisphaerales for sulfite supply, though no significant relationship was observed between them. In contrast, fewer Archaea encode pathways for sulfate reduction in thermoalkaline springs, mainly members of the Thermoproteales and Caldarchaeales (Fig. 6c, e). Additionally, one module in the network in thermoalkaline springs consists of abundant Sulfolobales and Thermoproteales, while the other comprises less abundant Archaea, including small Archaea from CG10238-14 and Anstonellales (Fig. 6c). Although distinct, thermoacidic geothermal springs also host numerous low abundance small Archaea in the network, such as Micrarchaeales, CABMEP01, and Parvarchaeales, suggesting the importance of these poorly studied Archaea. Given the expression of CAZymes and rbcL gene in Micrarchaeales and CABMEP01, these less abundant small Archaea may participate in the degradation of complex organic carbon and nucleotides. This could potentially serve as a source of monosaccharides and phosphorus sources or as part of a parasitic or predatory relationship with their hosts such as ARK-15 and Thermoplasmatales (Fig. 6d, e; Supplementary Data 11). In general, the analysis of rMAGs supports the general notion that competition and cooperation between Archaea may be common, especially in springs where Archaea dominate over Bacteria.

Fig. 6: Potential metabolic interactions within archaeal communities in Archaea-dominant geothermal springs.
figure 6

a Maximum relative abundance of archaeal orders involved in co-occurrence networks in acidic (pH <5.5) and alkaline (pH ≥8.5) geothermal springs. The co-occurrence networks between archaeal orders in thermoacidic (b) and thermoalkaline (c) springs. Schematic diagram of distinct archaeal orders involved in co-occurrence networks and their potential roles in the biogeochemical cycles of carbon, nitrogen, and sulfur metabolism in acidic (d) and alkaline (e) geothermal spring habitats. The color of circle indicates class, which is consistent with Fig. 1. The number in circle indicates the order assigned to same class.


We provide a large resource of archaeal genomes from a unique magmatic-geothermal spring area of mainland China that hosts physicochemically diverse springs. These 2,949 reconstructed genomes and over four million protein sequences considerably expand the known phylogenetic diversity of Archaea in terrestrial geothermal systems and expand the total number of archaeal genomes in the GTDB by over 48%. The integration of metagenomic and metatranscriptomic sequencing allowed us to examine the metabolic potential and gene expression and enabled us to develop conceptual models of potential metabolic interactions between Archaea at the ecosystem scale. Additionally, we provide evidence that microbially generated low pH not only controls community composition but by doing so it alters the abundance of functions for important biogeochemical cycling reactions. Within the dataset, temperature was also strongly correlated with archaeal diversity.

A fundamental goal of ecology is to understand how environmental parameters impact the composition and function of biological communities. For a long time, temperature and pH have been reported as the two most influential factors affecting the distribution of microbial communities in thermal springs, where temperature is usually considered to be the main driving factor. The general belief is that both microbial richness and diversity show a unimodal relationship with temperature, with diversity peaking around 20 to 40 °C64,65,66,67. However, many samples in these studies have a narrow pH range of 6–8, so pH effects may be underestimated. In contrast, several other studies with a wider range of pH suggest microbially generated low pH is the dominant correlate with the distribution and diversity of microbial communities in thermal springs68,69,70. Springs in those studies are divided into two distinct clusters, acidic (2–5) and neutral (6–8) pH. There are very few high-temperature alkaline springs with pH above 8 in these studies. The largest data set of 925 hot springs studied by Power et al. also showed that temperature only has a significant effect above 70 °C70. Those results are similar to our study, which showed that archaeal composition and function are primarily correlated with pH at temperature below 80 °C (Supplementary Fig. 3). Unlike previous studies based on 16 S rRNA gene sequencing, the current study avoids primer bias and assumptions about function based on single gene amplicons. Additionally, our large data set spans a very wide range of pH (2–10) and temperature (23–100 °C) and is a good representation of the physicochemical diversity within terrestrial geothermal ecosystems worldwide. Our study shows that neither pH nor temperature alone is sufficient to explain the community composition, but emphasizes the importance of pH, especially in acidic springs buffered by sulfuric acid, as the strongest control.

Most -omics studies of Archaea focus on the dominant lineages, while low-abundance lineages are often overlooked, especially groups of small Archaea such as Nanoarchaeota (Nanobdellota), Aenigmatarchaeota, and Micrarchaeota, as well as Asgardarchaeota. Our metatranscriptomic analysis showed that some small Archaea, such as CABMEP01 (within Micrarchaeia) and CG10238-14 (within Aenigmatarchaeia), expressed multiple carbohydrate-metabolizing enzymes, peptidases, and carbohydrate and amino acid transporters (Supplementary Data 11). Furthermore, they encoded a number of transposases with high expression, suggesting frequent gene exchange, which may be relevant to their symbiotic lifestyle71. In addition, Micrarchaeia was the only member in acidic DRTY-6 spring expressing ribulose-bisphosphate carboxylase (rbcL), which is involved in the AMP pathway catabolizing nucleotides as reported previously72 (Supplementary Data 8). Interestingly, members of the Jordarchaeales, an Asgardarchaeota order, were the only archaeal lineages that expressed urease and a urea transport system in JZ-3 pool (Supplementary Data 11). These enzymes may help Jordarchaeales utilize exogenous urea and convert it into CO2 and ammonia, which has never been reported in Jordarchaeales, which generally have poor nitrogen-metabolizing capacity73. In short, low-abundance small Archaea and Asgardarchaeota occupy unique ecological niches in these geothermal ecosystems. Although methanogens were not abundant in these springs, microcosm experiments demonstrated that they were active but temperature-sensitive41. That study noted a shift in the predominant methanogenic pathway in a member of the Archaeoglobales from hydrogenotrophic to methylotrophic when incubation temperatures increased from 65 to 75 °C. And both genomic inference and mcr expression under methanogenic conditions strongly suggested hydrogenotrophic methanogenesis by a member of the Nezhaarchaeales in situ.

We note that some key biochemical functions are encoded by coexisting but taxonomically distinct orders of Archaea (Fig. 6). This functional redundancy is widespread in microbial communities in various environments74,75,76,77. In our geothermal springs, a high functional redundancy was observed with respect to energy-transducing metabolic pathways78, including polysaccharide degradation, acetogenesis, acetate oxidation, thiosulfate disproportionation, and sulfide oxidation, which are conducted by abundant but phylogenetically diverse Archaea (Fig. 6). These functions are likely to be fundamental and important archaeal community processes in geothermal ecosystems79, and support polysaccharide, acetate and sulfur compound as important energy source. In addition, in our dataset, functional redundancy seemed to be more prevalent in polyextreme conditions. An example is the distribution of CAZymes, which were more abundant and widespread in the archaeal community in acidic/alkaline springs relative to neutral springs within the same functional cluster (Supplementary Fig. 7). The identity of Archaea possessing the same function can vary substantially across space or time. For example, some Thermoproteales mediate DSR in thermoalkaline springs, while the same function is likely mediated by Thermoplasmata in thermoacidic springs (Fig. 6b,d). This taxonomic variability may result from the unique ecophysiology of these ecologically equivalent organisms that leads them to be selected by different physicochemical characteristics76,79. In general, functional redundancy is a common aspect of microbial communities, which is linked to the capacity to withstand environmental pressures and fluctuations.


Sample collection, DNA extraction, sequencing, and metagenomic assembly

This study was conducted in compliance with all relevant ethical standards and regulations. We obtained permits for all sample collection, supported by Yunnan Tengchong Volcano and Spa Tourist Attraction Development Corporation. Geothermal spring sediment samples were collected and transferred with informed consent, under protocols approved by the Ethical Review Committee at the University of Science and Technology of China. A total of 152 sediment samples were collected from 48 geothermal springs spanning six years (2016–2021) located at Tengchong county in Yunnan province, China. These geothermal springs span a wide range of physiochemical parameters with temperature ranging from 23 to 100 °C and pH ranging from 2.0 to 9.7 (Supplementary Data 1). Detailed method for sample collection, DNA extraction, and metagenomic sequencing have been described in Hua et al.20. Raw reads generated on Illumina Hiseq 4000 sequencer were preprocessed using custom Perl scripts80 ( Quality reads for each sample were de novo assembled separately using SPAdes81 (v.3.9.0) with parameters: -meta -k 21,33,55,77,99,127.

Genome binning, curation and dereplication

For each sample, quality reads were mapped on assembled scaffolds with length >2500 bp using BBMap (v.38.92; with following parameters: k = 15 minid = 0.97 build = 1. The generated.bam file was used to calculate sequence depth and further to conduct genome binning using three tools including: MaxBin282 (v.2.2.7), CONCOCT83 (v.1.1.0) and MetaBAT84 (v.2.12.1). The best bins were determined using DASTool85 (v.1.1.3). Clean reads for each selected bin were recruited using BBMap (the same parameters as mentioned above) and were reassembled using SPAdes (v.3.15.2) with the following parameters: -careful -k 21,33,55,77,99,127. The genome quality including completeness, contamination, heterogeneity of all bins except small Archaea with tiny sizes and episymbiotic lifestyle (previously known as DPANN Archaea) was evaluated using CheckM86 (v.1.1.3). For small Archaea bins, their completeness was estimated to calculate the occurrence frequency of genes among a set of 48 single-copy genes2. Contamination and heterogeneity of these bins were determined using CheckM. The genome quality for all bins was recorded in Supplementary Data 3. To obtain genomes with high quality, scaffolds with abnormal sequence depth and multiple marker genes were manually removed as previously described24. Scaffolds with depth <[quartile1st – 1.5 × (quartile3rd - quartile1st)] and > [quartile3rd + 1.5 × (quartile3rd - quartile1st)] were treated as outliers. Finally, a total of 2,949 Archaea bins were obtained with completeness > 50% and contamination <10%. The program dRep87 (v.3.2.2) was applied to dereplicate genomes at 99% ANI (strain level). A total of 603 representative MAGs (rMAGs) were picked out for further metabolic analysis.

Taxonomic assignment and phylogenetic inference of assembled rMAGs

The taxonomy of all 2949 archaeal genomes was initially determined using GTDB-Tk88 (v.2.1.0). Phylogenetic analysis was constructed and used to confirm the taxonomic assignments. A total of 603 representative MAGs from present study and reference genomes from GTDB were picked out to infer the phylogeny. Briefly, multiple sequence alignments of 53 concatenated conserved archaeal marker genes were retrieved from GTDB-Tk. The poorly aligned regions were eliminated using trimAl89 (v.1.4.rev22) with the parameter: -automated1. Phylogeny was inferred using IQ-TREE90 (v.1.6.12) with 1,000 ultrafast bootstrapping iterations. The best model of LG + F + R10 was determined using ModelFinder91, which is well supported by Bayesian information criterion (BIC). The phylogenetic tree was visualized using iTOL92 (v.6).

Taxonomic assignment of rpS3 genes

Gene calling was conducted using Prodigal93 (v.2.6.3) (parameters: -p meta) for scaffolds >500 bp in each metagenome. The candidate rpS3 protein sequences were retrieved by searching all proteins against curated rpS3 database (see rpS3 database at which sequences ≥ 200 amino acids using AMPHORA294 (v.2.0). To identify rpS3 protein sequences with accuracy, three steps were performed: (i) Hits with score <40 from AMPHORA-based searches were filtered; (ii) Protein sequences with length <60 amino acids were eliminated; and (iii) All remaining sequences were searched against NCBI-nr database using BLASTp program and only hits identified as rpS3 genes were kept. Across all metagenomes, a total of 27,436 rpS3 protein sequences were identified. The confirmed rpS3 protein sequences were subsequently clustered at 99% sequence identity using USEARCH95 (v.11.0.667) with following options: -sort length -id 0.99 -maxrejects 0 -maxaccepts 0 -centroids. This resulted in the generation of 8,757 representative archaeal and bacterial rpS3 proteins at species level.

Since GTDB provided a phylogenetically consistent and rank normalized genome-based taxonomy for prokaryotes, rpS3 gene sequences from GTDB with well-documented taxonomy information were extracted to build a custom rpS3 gene database, which could be further used to assign taxonomy at different levels for rpS3 genes retrieved from metagenomes. Briefly, gene sequences with length >300 bp were kept and were further clustered using USEARCH at 100% identity. To eliminate rpS3 genes that misassigned to the corresponding genomes due to the biases caused by metagenome assembly and genome binning, taxonomic assignments were conducted for all rpS3 genes identified from GTDB with themselves as database using (v.1.9.1) program in QIIME96: -m rdp -c 0.60. Items were discarded if the assigned taxonomy is inconsistent with the taxonomy retrieved from GTDB. The retained rpS3 gene sequences were further clustered at 99% identity using USEARCH to establish a final version of rpS3 database at species-level. 8,757 representative rpS3 gene sequences were searched against the curated database with the same procedure as mentioned above. Among them, 2500 rpS3 genes (28.5%) can’t be assigned at phylum-level. These gene sequences were further searched against the protein sequences of curated rpS3 database using BLASTx program: -evalue 1e-5 -num_descriptions 5 -num_alignments 5. The lowest concordant level of the five top hits was used to assign taxonomy for each undetermined sequence.

Relative abundance of rMAGs and rpS3 genes

Clean reads were mapped on 603 rMAGs and scaffolds with length >500 bp using BBMap respectively. The relative abundance of each rMAG in one sample was calculated as the percentage of reads mapped to the rMAG to the total reads mapped to the metagenome. Only rMAG with a relative abundance more than 0.01% is counted. Similarly, clean reads were mapped on 8757 representative rpS3 gene sequences for each sample using BBMap and the coverage information was calculated by dividing recruited reads to the length of gene length. The relative abundance of each rpS3 in one sample was calculated by dividing the coverage of one rpS3 to the total coverage of all rpS3. A table was generated to record the abundance information of each archaeal phylum, class and order based on rMAGs and representative rpS3 genes in each sample (Supplementary Data 6).

Genome and community-level metabolic annotation

Functional annotation was conducted using METABOLIC97 (v.4.0) with all representative archaeal MAGs and clean reads of each sample as inputs. Briefly, gene calling was performed for each rMAG using Prodigal with “-p single” option to identify open reading frames (ORFs). ORFs were subsequently searched against the KEGG, TIGRfam, Pfam and collected HMM profiles associated with key biogeochemical cycling using HMMER98 (v.3.3.2). The MW-scores (metabolic weight score) were calculated by METABOLIC which describes the contribution of one function/pathway relative to all functions in all Archaea since only archaeal genomes were taken as inputs.

The metabolic pathway associated with elemental biogeochemical cycling were further assessed by searching all ORFs in rMAGs against the KOfam database (KEGG release 103) using the KofamScan99 (v.1.3.0; E value < 10−5). The archaeal order was determined to contain the pathway if all genes involved in this pathway can be detected in at least one rMAG in a given order. To ensure the accuracy, only rMAGs with scaffold numbers <200 were taken into consideration. The gene repertoire against Carbohydrate-Active enZYmes (CAZy) Database100 for searching carbohydrate degradation capacities was explored for each rMAG as well.

Phylogenetic tree reconstruction of hydrogenase

Hydrogenase proteins were initially identified using METABOLIC across 603 archaeal rMAGs in this study, and were further classified into different functional groups using a web tool of Hydrogenase database (HydDB)60. Reference sequences of [Fe], [FeFe] and [NiFe] hydrogenases in Archaea were downloaded from the HydDB. The obtained protein sequences were aligned using MUSCLE101 (v.3.8.31) and trimmed using TrimAL (-gt 0.05 -cons 50). Phylogeny was generated using IQ-tree with 1,000 ultrafast bootstrapping iterations and VT + F + R9 as the best model.

Metatranscriptomic analysis

Geothermal spring sediment samples collected in Jan 2021 and Jul 2022 were conducted for metatranscriptomic sequencing. Sufficient RNA was only obtained from four samples, including JZ-2 (pH 7.47, 36.0 °C) in Jan 2021 and DRTY-16 (pH 2.11, 38.1 °C), QQ (pH 6.80, 70.1 °C), JZ-3 (pH 9.98, 64.7 °C) in Jul 2022. Total RNA was extracted using the RNeasy PowerSoil Total RNA kit (QIAGEN). Total RNA was subsequently sent out for metatranscriptomic sequencing, following with rRNA removal and mRNA enrichment. The generated paired-end reads ( ~114 Gbp) were quality-controlled using Sickle (v.1.33; with the following parameters: -t sanger --quiet -l 50. Trimmed transcripts were mapped to the identified ORFs of 603 archaeal rMAGs using BBMap with the same parameters as above. The fragments per kilobase of transcript per million (FPKM) values were extracted from the output. The heatmap figure was built based on gene expression data using the pheatmap package (v.1.0.12) in RStudio (v.1.2.5033).

Statistical analyses

A total of 132 geothermal spring samples with both pH and temperature measured were picked out and applied for ordination analysis. To assess the resemblance of samples and to identify main factors affecting the clustering pattern, principial coordinates analysis (PCoA) was conducted based on Bray-Curtis distance of archaeal community composition and functional annotation (KOs) data. Three factors were considered including sampling season (summer and winter), pH (acidic: pH <5.5; neutral: 5.5 ≤ pH <8.5; alkaline: pH ≥8.5)33, and temperature (mesothermal: temperature <60 °C; thermal: 60 ≤ temperature <80 °C; hyperthermal: temperature ≥80 °C)34. Analysis of Similarity (ANOSIM) was used to examine the significant differences among different sample groups. Pearson correlation was applied to test the correlation between pH and temperature with the relative abundance of specific archaeal lineages. The generalized linear model (GLM) was applied to establish connections between archaeal community at strain/function level and environmental parameters using the glm function in the “stats” package (v. 4.2.2)102 in R platform ( Statistical significance (p < 0.05) was assessed by conducting likelihood ratio tests (LRT) on the linear models, utilizing Type-II ANOVA in the “car” package (v. 3.0-10)103 in R.

Network-based co-occurrence analysis

A comprehensive dataset comprising 48 acidic and 31 alkaline geothermal springs, spanning both temporal and spatial dimensions, was employed for unveiling the co-occurrence patterns of archaeal orders (Supplementary Data 1). To enhance clarity and streamline the analysis, archaeal rMAGs occurring in fewer than half of samples were excluded, resulting in the retention of 83 rMAGs for subsequent network construction. The co-occurrence networks were constructed using SparCC104 with default parameters, and the robustness of the networks was ensured by employing 100 bootstrap samples to infer pseudo-P values. Only correlations deemed statistically significant (p  <  0.05, two-sided) and robust (absolute value of correlation coefficient greater than 0.6) between pairwise rMAGs were considered, contributing to the inference of a reliable network. Visualization of the network was performed using Gephi (v.0.10.1)105.

Reporting summary

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