Introduction

Cold seeps are characterized by the emission of subsurface fluids into the seafloor and occur widely at active and passive continental margins1,2. The upward fluids are often rich in methane and other hydrocarbons which sustain seabed oasis composed of various microorganisms and faunal assemblages3,4. The primary process that fuel complex cold seep ecosystems is the anaerobic oxidation of methane (AOM), conjointly operated by a consortium of anaerobic methane-oxidizing archaea (ANME) and sulfate-reducing bacteria (SRB)5,6. AOM removes approximately 80% of upward venting methane, acting as an efficient methane filter7. Additionally, deep-sea cold seep sediments also contain diverse and abundant diazotrophs that might contribute substantially to the global nitrogen balance8. Cold seeps are therefore biologically and geochemically significant on a global scale.

The venting fluids can significantly influence the sedimentary environment of seep sites, resulting in changes of chemical characteristics of sediments9. In particular, arsenic (As), one of the most abundant elements in the Earth’s crust, are anomalously enriched in seep sediments10,11,12,13,14. The anomalous As enrichment could be attributed to the ascending fluids that could capture As and other metals when passing through thick shaly formations10,14; or the so-called particulate iron shuttle effect9,11,13. As is also toxic metalloid in nature that, upon exposure, can cause negative effects for all living things15. Depending on the physicochemical conditions, As can be found in different oxidation and methylation states, showing various levels of toxicity and bioavailability16. In marine environments, arsenate (As(V)), and arsenite (As(III)) are the dominant forms of inorganic As17. It is assumed that microbes have evolved a genetic repertoire related to As cycling, dated back to at least 2.72 billion years ago18,19. As biotransformation processes include As detoxification to mitigate toxicity and As respiration to conserve energy. The As detoxification is mainly achieved by two steps: reduction of As(V) to As(III) by cytoplasmic As(V) reductases (arsC gene) with homology to either glutaredoxin (arsC1 gene) or thioredoxin (arsC2 gene) family and subsequent extrusion of As(III) via As(III) efflux permeases (arsB and acr3 genes)20,21 (Fig. 1). Another As detoxification mechanism involves the methylation of As(III) to methylarsenite (MAs(III)) by the As(III) S-adenosylmethionine (SAM) methyltransferase (arsM gene)22 (Fig. 1). Although MAs(III) intermediates are more toxic than As(III), they do not accumulate in cells and can be detoxified through several different pathways. MAs(III) can be further methylated by ArsM and volatilized, extruded from cells via the MAs(III) efflux permease (arsP gene)23, oxidated to less toxic MAs(V) by the MAs(III)-specific oxidase (arsH gene)24, or demethylated to less toxic As(III) by the C-As lyase (arsI gene)25. As respiration consists of the chemolithotrophic oxidation of As(III) by As(III) oxidase (aioAB/arxAB genes) and dissimilatory As(V) reduction by respiratory As(V) reductase (arrAB genes)15,26 (Fig. 1). Taken together, microbes have a huge potential effect on the biogeochemical cycling and toxicity of As.

Fig. 1: Diagram of the microbial transformations of As.
figure 1

As(III), arsenite; As(V), arsenate; MAs(III), trivalent methylarsenite; MAs(V), pentavalent methylarsenate. As(III) efflux permease: ArsB/Acr3; cytoplasmic As(V) reductase: ArsC; respiratory As(V) reductase: ArrA; As(III) oxidase: AioA/ArxA; As(III) S-adenosylmethionine (SAM) methyltransferase: ArsM; C-As lyase: ArsI; MAs(III) efflux permease: ArsP; MAs(III)-specific oxidase: ArsH.

Full size image

So far, As-transforming microbes and As-related genes have been widely investigated in various natural environments, including polluted and pristine soils27,28, terrestrial geothermal springs29,30,31, wetlands32,33, pelagic oxygen-deficient zones34, groundwater35, etc. For example, metagenomic and metatranscriptomic analyses revealed that Aquificae were the key players for the arsC-based detoxification in Tengchong geothermal springs29. A global survey also described the phylogenetic diversity, genomic location, and biogeography of As-related genes in soil metagenomes36. Only recently, the behavior of As biotransformation has been reported in the deep-sea realms, i.e. hadal trench of the Challenger Deep37. Deep-sea ecosystems cover 67% of Earth surface and have extremely high densities of microbes (up to 1000× greater than surface waters) which play a critical role for long-term controls on global biogeochemical cycles38,39. The environmental conditions in deep seafloor cold seeps differ greatly from those in the aforementioned ecosystems, such as low temperatures, high pressure, darkness and the presence of seepage activities1. Thus, the investigation of As-related genes and microbes at seeps will expand our current knowledge on As metabolisms and allow us to discover new lineages containing As-related genes.

The purpose of this study was to decipher the microbial transformation of As in cold seep sediments at a global scale. Here, we applied a relatively comprehensive data set of 87 sediment metagenomes and 33 metatranscriptomes derived from 13 geographically diverse cold seeps across global oceans (Supplementary Fig. 1; Supplementary Data 1), to investigate As-associated genes and their host microbes. This study aims to address the following questions: (i) biogeography of As cycling genes across global cold seeps; (ii) phylogenetic diversity and distribution of As cycling genes across global cold seeps; (iii) interactions between As metabolisms and biogeochemical cycles of carbon and nitrogen.

Results

As cycling genes are widespread and active across global cold seeps

To gain a broad view on biogeography of As cycling genes, we determined their abundances from 87 sediment metagenomes collected from 13 globally distributed cold seeps. Considering that sulfate respiration is one of the most important microbial redox processes in cold seep sediments40, the dsrA was used as the target gene to compare with As cycling genes. We found that genes related to As detoxification were prevalent in these cold seep samples (Fig. 2) and their abundances were higher than those of dsrA genes (Supplementary Figs. 2, 3; Supplementary Data 2). The arsM and arsP genes that respectively produce volatilized methylated organoarsenicals and mediate its subsequent expulsion outside cell, were the most abundant ones. The arsC1/arsC2 genes for cytoplasmic As(V) reduction and the acr3 gene for As(III) extrusion also dominated most cold seep samples. Moreover, As detoxification genes, i.e. arsM, arsP, arsC1/arsC2, acr3, were actively expressed in the sediment metatranscriptomes from Haima and Jiaolong seeps along with gas hydrate deposit zones of Qiongdongnan and Shenhu, revealing in situ microbial activities on As detoxification (Supplementary Fig. 4). Seep microbes might utilize both methylation and cytoplasmic As(V) reduction strategies to overcome potential toxic effects of exceptional As accumulation at cold seeps. Alternatively, methylation is not strictly a detoxification pathway but also an antibiotic-producing process with MAs(III) being a primitive antibiotic41, which could provide additional competitive advantages. However, the function of arsM in anoxic environments and its contribution to As cycling have yet to be verified. Our results contradict previous findings demonstrating that arsM are less common in soils36 and hot springs29 than arsC, but in line with those found in hadal sediments37. The discrepancy in As detoxification mechanisms between terrestrial and deep-sea ecosystems could be attributed to their huge variations in the habitats and geographical locations. When comparing the abundances of As(III) efflux pumps, we observed that arsB was in much lower abundance than acr3 (Fig. 2a). Previous studies also reported an abundance of acr3 over arsB in forest soils and wetlands32,36,42. This is likely because Acr3 proteins are more ancient and have greater phylogenetic distribution as compared with ArsB19. Conversely, genes related to energetic As respiratory oxidation (aioA) and reduction (arrA/arxA) were less abundant in all cold seep samples as compared with As detoxification genes. Despite of this, respiratory genes were transcriptionally active, as evidenced by the detection of arrA transcripts in the Jiaolong seep (up to 15.9 TPM, Supplementary Fig. 4).

Fig. 2: The global distribution of potential genes involved in As cycling at cold seeps.
figure 2

a The abundances of As cycling genes across the 87 cold seep metagenomes. The abundance of each gene was normalized by the gene length and sequencing depth and represented as GPM (genes per million) values. b The partial least squares discrimination analysis (PLS-DA) plots based on the abundances of As-cycling genes (n = 87). Similarity values among the samples of different sediment depths and types of cold seep were examined using a 999-permutation PERMANOVA test. Source data is available in Supplementary Data 2.

Full size image

To determine the distribution characteristics of As cycling genes, each metagenome was categorized in terms of its sediment depth (i.e. surface: <1 mbsf; shallow: 1–10 mbsf; deep: >10 mbsf). Metagenomes were also grouped based on the type of cold seep, including gas hydrate, seep (i.e. oil and gas/methane seep), and volcano (mud/asphalt volcano)1, respectively. The partial least squares discrimination analysis (PLS-DA) revealed dissimilarity in As cycling genes among different sediment layers (Fig. 2b; F = 4.3504, p = 0.001, R2 = 0.10267, 999-permutations PERMANOVA test). The distribution traits of As cycling genes in surface sediments were separated from deep sediments and more similar to those in shallow ones (Fig. 2b). The abundance of prevalent As cycling genes such as acr3, arsC2 and arsM in deep sediments were significantly higher as compared with those in shallow and surface sediments (Supplementary Fig. 2). As cycling genes in different types of cold seep were also different from each other (Fig. 2b; F = 3.5246, p = 0.004, R2 = 0.07742, 999-permutations PERMANOVA test). Dominant As cycling genes in gas hydrates displayed higher abundances relative to those in seeps and volcanos (Supplementary Fig. 3). Hence, the distributions of As-associated genes were influenced by a combination of sediment depths and types of cold seep. The higher As cycling gene abundances observed in our deep or gas hydrate-associated samples could be correlated with a high level of environmental As, as what was described in As-rich altiplanic wetlands32. In the Nankai Trough, As with unknown sources was demonstrated to actively release into sediment layers where methane hydrates occur (As concentration of 14 ppm in gas hydrate-bearing sediments vs av. 6.4 ppm for the whole sediment core)17.

Microbes involved in As cycling varied across seep habitats

To profile taxonomic diversity of As-related microbes, a total of 1741 species-level metagenome-assembled genomes (MAGs, 95% average nucleotide identity) were reconstructed from these 87 cold seep metagenomes (Supplementary Data 3). Of these, 1083 MAGs spanning 9 archaeal and 63 bacterial phyla as well as one unclassified bacterial phylum were potentially involved in As cycling at cold seeps (Supplementary Data 4). Metagenomic read recruitments revealed that the recovered 1083 As-related MAGs accounted for 1.8–62.8% cold seep communities (Fig. 3 and Supplementary Data 4). Taxonomic compositions of As-related microbiome across different types of cold seep displayed pronounced variations (Fig. 3). In the sediments derived from oil and gas/methane seep, As-related microbes contained mostly Methanogasteraceae (i.e. ANME-2c) and Methanocomedenaceae (i.e. ANME-2a) within Halobacteriota phylum, ETH-SRB1 within Desulfobacterota phylum, JS1 within Atribacterota phylum as well as Anaerolineae and Dehalococcoidia within Chloroflexota phylum. The As-related microbes in gas hydrate sediments were dominated by bacterial lineages, highlighted by Atribacterota (JS1) and Chloroflexota (Anaerolineae and Dehalococcoidia). Nevertheless, in asphalt/mud volcano sediments, the compositions of As-related microbes were diverse in different samples. The clear distinctions in As-related microbiomes across different seep habitats suggested an important role driven by environment selection. Multiple parameters, including sediment temperature, sediment depth, water depth, methane concentration, and geographic distance have been demonstrated to cause these variations43,44. Additionally, our results show that Chloroflexota outnumbered Atribacterota in sediment samples with lower Fe(II) concentration (av. 16.51 µmol/L), while Atribacterota dominated over Chloroflexota in sediment samples with higher Fe(II) concentration (av. 81.54 µmol/L). It is possible that iron oxyhydroxides control the mobilization of As45 and thus affect As-related microbial communities (Fig. 3).

Fig. 3: The community structures of microbiome involved in As cycling at cold seeps.
figure 3

The relative abundance of each MAG was estimated using CoverM. The compositions of microbiome involved in As cycling across different types of cold seep were clustered based the Bray–Curtis distance. Orange and red asterisks denote samples with lower and higher concentrations of Fe(II), respectively. Detailed statistics for As-related microbiome are provided in Supplementary Data 4.

Full size image

Expanded diversity of microbial lineages containing As detoxification genes

Among these As detoxification genes, acr3, arsC1/arsC2, arsM and arsP were widely distributed in bacteria and archaea, while other As detoxification genes (arsB, arsI and arsH) were sparsely distributed (Fig. 4). The acr3 gene is typically affiliated with Proteobacterial, Firmicutes, Actinobacterial and other bacterial sequences36,42,46. Our study observed an unexpectedly wider phylogenetical diversity of acr3 than previously reported. Notably, Asgardarchaeota including Lokiarchaeia, Thorarchaeia, Sifarchaeia, LC30, along with Heimdallarchaeia and Wukongarchaeia described as the most likely sister group of eukaryotes, are firstly documented to have genetically capability for As(III) extrusion. The greater diversity of As detoxification genes found in Asgardarchaeota phylum further point to their ancient origin19. Furthermore, a considerable number of candidate bacterial phyla without cultured representatives (e.g. 4484–113, AABM5-125-24 and RBG-13-66-14) were also equipped with such an ability. Though their functional redundancy as As(III) efflux pumps, arsB was more phylogenetically conserved as compared with acr3 and simply restricted to Alphaproteobacteria, Gammaproteobacteria and Campylobacterota (Fig. 4). This observation is in agreement with previous reports comparing the diversity of arsB to acr336,42,46. The arsM gene was relatively uncommon in terrestrial soil microorganisms29,36. In contrast, this study showed that the arsM genes in seep microbes have a great taxonomic diversity similar to acr3 genes, including Chloroflexota, Proteobacteria, Atribacterota, Asgardarchaeota, Hydrothermarchaeota, Thermoplasmatota, Thermoproteota as well as other currently unidentified bacterial phyla (e.g. 4484-113, AABM5-125-24 and RBG-13-66-14) (Fig. 4). Among these, Atribacterota, Asgardarchaeota and the candidate bacterial phyla stated above have not previously been implicated in As methylation19,36. For cytoplasmic As(V) reduction, Asgard archaeal lineages all lacked corresponding genes (arsC1 and arsC2). The underlying causes to their absence in Asgardarchaeota are unclear. It’s likely that Asgardarchaeota lost cytoplasmic As(V) reduction genes during evolution or possess different enzyme systems. In general, these data advance our understanding on the phylogenetical diversity of As detoxification genes and highlight the potentially important role played by archaea in As cycling, Asgardarchaeota particularly.

Fig. 4: Phylogenetic distribution of As-cycling genes.
figure 4

Left bar plot showing the total number of genomes encoded in each phylogenetic cluster assigned by GTDB-Tk based on GTDB r207 release. Right bubble plot showing the number of As-cycling genes encoded within each phylogenetic cluster. Detailed information on phylogenetic diversity of As-cycling genes is provided in Supplementary Data 5.

Full size image

Diverse seep lineages are identified to perform As respiration

In addition to mitigating toxicity, some microorganisms can respire the redox-sensitive element of As to reap energetic gains (i.e. arsenotrophy), either via chemoautotrophic As(III) oxidation (aioAB/arxAB) or anaerobic As(V) respiration (arrAB)47,48. The alpha subunits of these arsenotrophic enzymes form distinct clades with the dimethylsulfoxide (DMSO) reductase superfamily34. This superfamily also includes other enzymes critical in respiratory redox transformations, e.g. Nap and Nar. Here, we identified two AioA, three ArxA and 17 ArrA protein sequences, respectively. A phylogenetic analysis of recovered arsenotrophic protein sequences showed that they all clustered together with known AioA/ArxA and ArrA proteins (Fig. 5a). Functional As bioenergetic aioA/arxA and arrA genes are generally found together with other necessary accessory genes. The aioA of As(III) oxidizing microorganisms always forms an operon with aioB and other genes involved in As detoxification and metabolisms (e.g. aioD, aioXSR, arsR)15,26. Arx is demonstrated to be a variant of Arr and these two enzymes have a similar genetic arrangement. The arrA/arxA gene is always found together with the arrB/arxB and often with the arrC/arxC and arrD/arxD15,26. The genomic organization analysis showed that identified two aioA, three arxA and 11 of 17 arrA genes all had corresponding accessory genes (Fig. 5b), further confirming their potential identities as arsenotrophic enzymes.

Fig. 5: The metabolic potential of the arsenotrophic gene-carrying MAGs.
figure 5

a A maximum-likelihood tree of the DMSO reductase family, with protein sequences identified as associated with arsenotrophic enzymes in this study. Bootstrap values are generated from 1000 replicates. Bootstrap values ≥70 are shown. Scale bar indicates amino acid substitutions per site. b The genomic context of the aioA, arxA and arrA genes in MAGs containing arsenotrophic genes. c Heatmap showing the predicted metabolism in potential As-respiring microbes. Detailed annotation is presented in Supplementary Data 6. The completeness of each pathway was calculated using the DRAM Distill function.

Full size image

The aioA/arxA genes are uncommon in soil microbiomes and mostly found in Proteobacteria15,36,49. By assigning the taxonomy, aioA/arxA genes recovered here belonged to Gammaproteobacteria (n = 3) and Alphaproteobacteria (n = 2), consistent with previous findings (Figs. 3 and 5c). Nevertheless, 17 arrA genes were phylogenetically affiliated with seven distinct bacterial lineages: Bacteroidota (n = 1), Chloroflexota (n = 4), Deferribacterota (n = 1), Desulfobacterota (n = 8), Desulfobacterota_I (n = 1), Gammaproteobacteria (n = 1), and Nitrospirota (n = 1) (Figs. 3 and 5c). Despite that several other bacterial lineages (i.e. Deferribacterota, Firmicutes and Chrysiogenetes) are reported to contain arrA genes, most known As(V)-respiring microorganisms are assigned to proteobacterial clades15,36,49. Our findings of the arrA-containing Bacteroidota, Chloroflexota and Nitrospirota, expand the database of putative dissimilatory As(V) reducers.

As respiration are potentially critical to central metabolisms in cold seeps

Microbially mediated As respiration has been verified to influence biogeochemical cycles of carbon and nitrogen, e.g. chemoautotrophic As(III) oxidation coupled with denitrification50,51. Here, functional annotations identified near-complete calvin and reductive TCA carbon fixation pathways in aioA/arxA-carrying Alphaproteobacteria (n = 2) and Gammaproteobacteria MAGs (n = 3) (Fig. 5c). Terminal reductase systems were also recognized in aioA/arxA-carrying MAGs, i.e. nitrate reductase (narGHI). The cooccurrence of these genes suggests that the As(III) oxidation may help support autotrophic carbon fixation and nitrate reduction.

In addition, five arrA-carrying MAGs possessed genes for AssA (Fig. 5c), which mediate the first step of anaerobic activation of alkanes via fumarate addition52. Phylogenetic analysis revealed that identified AssA sequences were phylogenetically close to archaea-type and Group V AssA53 (Supplementary Fig. 5). These potential hydrocarbon degraders were classified as Chloroflexota (n = 2), Deferribacterota (n = 1), Desulfobacterota (n = 1), and Bacteroidota (n = 1). Methane, the simplest hydrocarbon, has been demonstrated to stimulate As(V) respiration during the process of anaerobic oxidation of methane54. Similarly, the occurrence of both AssA and ArrA indicated that heterotrophic MAGs stated above may also employ As(V) as electron acceptor for anaerobic degradation of multi-carbon alkanes. Genes encoding carbohydrate-active enzymes (CAZymes) targeting various complex carbohydrates were also present in these arsenotrophic MAGs, including chitin, pectin, starch and plyphenolics (Fig. 5c).

Notably, arsenotrophic MAGs may function as potential nitrogen fixers introducing new nitrogen to local environment. Genes encoding for the catalytic component of nitrogenase (i.e. nifHDK) were detected in one arxA-carrying (Gammaproteobacteria, n = 1) and six arrA-carrying (Gammaproteobacteria, n = 1; Desulfobacterota, n = 4; Desulfobacterota_I, n = 1) MAGs (Fig. 5c). It has been previously reported that As(III) oxidation can fuel biological nitrogen fixation in tailing and metal(loid)-contaminated soils55,56. The data present here further complement that diazotrophs could also fix N2 using energy obtained from dissimilatory As(V) reduction.

The metatranscriptomic reads were mapped against arsenotrophic MAGs to depict the gene expression profile at the genome level. Genes for dissimilatory As(V) reduction were transcriptionally active in the Jiaolong seep, as evidenced by the detection of arrA transcripts (21.7−340.1 TPM, Supplementary Data 7). In addition, transcripts of assA (9.2 − 6306.5 TPM) and nifH (155.1 − 28016.9 TPM) were identified in the Jiaolong seep, Shenhu area and Qiongdongnan Basin, implying anaerobic degradation of hydrocarbons and nitrogen fixation were actively expressed for these arsenotrophic microbes. No transcriptomic sequences related to arrA, assA and nifH genes were detected in the Haima seep. Nevertheless, it does not mean that genes of interest are not transcribed in situ because it is difficult to recover enough RNA from deep-sea samples and RNA can get lost during the process of deep-sea sampling57.

Our findings point towards a previously unrecognized arsenotrophs at seeps, impacting both carbon and nitrogen cycling. However, we acknowledge that cultivation experiments with As-respiring isolates are ultimately needed both to elucidate their lifestyle and confirm functionality for As-dependent carbon fixation, hydrocarbon and carbohydrate degradation as well as nitrogen fixation.

Discussion

Microbial transformation of As has been well documented and characterized in environments such as ocean water, groundwater and geothermal springs, but the knowledge on gene- and genome-level As cycling in deep sea (e.g. cold seep) is limited. Our study demonstrated that As methylation and cytoplasmic As(V) reduction were the predominant detoxification mechanisms employed by cold seep microbiomes. These results substantially expanded the diversity of As detoxification genes to a broader microbial community including Asgardarchaeota and a great number of candidate bacterial phyla. In addition, diverse arsenotrophic lineages are also identified, including Bacteroidota, Chloroflexota, Nitrospirota, etc, which also potentially participate in carbon and nitrogen biogeochemical cycling. This study provides a detailed understanding of As biotransformation in a complex microbiome in deep-sea realms, which could have significant implications for addressing environmental issues. Our results will also provide insights for microbial evolution in the early ocean with harmful metal(loids), e.g. As, as a driving force58.

Methods

Metagenomic and metatranscriptomic data sets

The 87 metagenomes and 33 metatranscriptomes analyzed in this study are derived from 13 globally distributed cold seep sites (Supplementary Fig. 1). Among them, 65 metagenomes and 10 metatranscriptomes were compiled from our previous publications8,59, and other 22 metagenomes were downloaded from NCBI Sequencing Read Archive (SRA). A detailed description of sampling locations and sequencing information for metagenomic and metatranscriptomic data is given in Supplementary Data 1.

Bioinformatic analyses

DNA reads pre-processing, metagenomic assembly and binning were performed with the function modules of metaWRAP (v1.3.2)60. First, the metaWRAP Read_qc module was used to trim raw sequencing DNA reads. Then the filtered DNA reads were individually assembled with the metaWRAP Assembly module using Megahit61 or metaSPAdes62 with default settings (detailed assembly statistics are summarized in Supplementary Data 1). In addition, metagenomic reads from the same sampling station (n = 10) were also co-assembled using Megahit with the default settings. Thereafter, MAGs were recovered from contigs with the length longer than 1 kb using the metaWRAP Binning module (parameters: -maxbin2 -concoct -metabat2) or the VAMB tool63 (v3.0.1; default parameters; detailed binning statistics are summarized in Supplementary Data 1). Further refinement of MAGs was performed by the Bin_refinement module of metaWRAP (parameters: -c 50 -x 10), and CheckM (v1.0.12)64 was used to estimate the completeness and contamination of these MAGs. All MAGs were dereplicated at 95% average nucleotide identity (ANI) using dRep (v3.4.0; parameters: -comp 50 -con 10)65 to obtain representative species MAGs. This analysis provided a non-redundant genome set consisting of 1741 species-level MAGs.

Raw metatranscriptomes were quality filtered with the Read_qc module of metaWRAP (v1.3.2)60 as described above. The removal of ribosomal RNAs was conducted with sortmeRNA (v2.1)66 in the quality-controlled metatranscriptomic reads.

Non-redundant gene catalog construction

Genes were predicted on contigs (≥1 kb) from the assemblies using the METABOLIC pipeline (v4.0)67, which resulted in 33,799,667 protein-coding genes. Clustering of the predicted proteins was performed with MMseqs2 (v13.45111)68 using the cascaded clustering algorithm at 95% sequence similarity and 90% sequence coverage (parameters: -c 0.95 -min-seq-id 0.95 -cov-mode 1 -cluster-mode 2) following the ref. 69. This process yielded a total of 17,217,131 non-redundant gene clusters.

Searching for As cycling genes

In this study, 11 well-characterized marker genes70,71 were selected to assess their potential influence to the As biogeochemical cycle. These genes include eight As detoxification genes (acr3, arsB, arsC1, arsC2, arsP, arsH, arsI, and arsM) and three As respiratory genes (aioA, arrA, and arxA). A hidden Markov model (HMM)-based search was performed to identify As-related genes in non-redundant gene catalogue by using hmmsearch function in HMMER package (v3.1b2)72. The HMM profile searches and score cutoffs for 11 As-related genes were taken from Lavy et al. (2020)71.

Taxonomic and functional profiling of MAGs

As-related MAGs were taxonomically annotated using the classify_wf function of the GTDB-Tk toolkit (v2.1.1)73 with default parameters against the GTDB r207 release. For all MAGs, gene calling and metabolic pathway prediction were conducted with the METABOLIC pipeline (v4.0)67. Functional annotation of genomes was also carried out by searching against KEGG, Pfam, MEROPS and dbCAN databases using DRAM (v1.3.5)74. The identification of As-related genes in MAGs was performed by searching against As-related HMM profiles from Lavy et al. (2020)71 as reported above. Genes involved in anaerobic hydrocarbon degradation were screened using BLASTp (identity >30%, coverage >90%, e < 1 × 10–20) against local protein databases53.

Abundance calculations

At the contig level, the relative abundances of genes related to As cycling across 87 metagenomes were calculated from non-redundant gene catalog using the program Salmon (v1.9.0)75 in the mapping-based mode (parameters: -validateMappings -meta). GPM (genes per million) values were used as a proxy for gene abundance as describe in ref. 74. At the genome level, the relative abundance of each MAG was profiled by mapping quality-trimmed reads from the 87 metagenomes against the MAGs using CoverM in genome mode (https://github.com/wwood/CoverM) (v0.6.1; parameters: -min-read-percent-identity 0.95 -min-read-aligned-percent 0.75 -trim-min 0.10 -trim-max 0.90 -m relative_abundance).

To calculate the transcript abundances of As-related genes, we also mapped clean reads from the 33 metatranscriptomes to non-redundant gene catalog or arsenotrophic MAGs. The transcript abundance of each gene was calculated as the metric-TPM (transcripts per million). GPM or TPM values were normalized based on the gene length and sequencing depth.

Phylogenetic analyses of functional genes

For phylogeny inference, protein sequences of functional genes were aligned with MAFFT (v7.490, -auto option)76, and gap sequences were trimmed using trimAl (v. 1.2.59, -automated1 option)77. Maximum likelihood phylogenetic trees were constructed for each genes using IQ-TREE (v2.12)78 with the following options: -m TEST -bb 1000 -alrt 1000. Branch support was estimated using 1000 replicates of both ultrafast bootstrap approximation (UFBoot)) and Shimodaira-Hasegawa (SH)-like approximation likelihood ratio (aLRT). Reference protein sequences for As-based respiratory cycle were obtained from Saunders et al. (2019)34. Reference protein sequences for fumarate addition were derived from Zhang et al. (2021)53. All the tree files were uploaded to Interactive tree of life (iTOL; v6)79 for visualization and annotation.

Statistical analyses

Statistical analyses were done in R (v4.0.4-v4.1.0) with the following descriptions. Normality and homoscedasticity of data were evaluated using Shapiro-Wilk test and Levene’s test, respectively. One-way analysis of variance (ANOVA) and least significant difference (LSD) test were conducted to evaluate the variations of each gene across different sediment depths and types of cold seeps. The partial least squares discrimination analysis (PLS-DA) was performed based on the GPM values of As cycling genes with R package ‘mixOmics’. The permutational multivariate analysis of variance (PERMANOVA) was employed to test whether As cycling genes shifted among different sediment depths and types of cold seeps using ‘adnois’ function in vegan package. All PERMANOVA tests were performed with 9999 permutations based on Bray–Curtis dissimilarity.