Diverse biosynthetic pathways in Arctic hydrothermal biofilms

Abstract

Biosynthetic gene clusters (BGCs) which encode diverse secondary metabolites are ubiquitous across microbiomes and support critical ecological functions. They also serve as attractive targets for new drug discovery. Here we combined genome-resolved metagenomics with long-read Nanopore RNA sequencing, yielding 1016 bacterial and 124 archaeal medium-to-high quality MAGs from previously unchartered microbial communities in Arctic hydrothermal vent biofilms. We identified 2965 BGCs from 870 metagenome-assembled genomes (MAGs) comprising a distinctive and rich diversity of BGCs, with ribosomally synthesized and post-translationally modified peptides (RiPPs) predominating across all samples. RiPPs and non-ribosomal peptide synthetases (NRPs), also known to encode metabolites with antimicrobial potential, are represented among the most expressed transcripts. Terpenes, though less expressed, contribute to microbial signaling and defense. Notably, we identify hydrogen cyanide (HCN) synthesis pathways in archaeal genomes, challenging the view that cyanogenesis is restricted to bacteria and eukaryotes. Our findings demonstrate that microbial adaptation to extreme environments favors RiPP-based biosynthesis and that HCN may play a role in archaeal ecological interactions. Moreover, microbial communities in Arctic hydrothermal vent biofilms provide a rich reservoir of unique bioactive compounds, with implications for drug discovery.

Biosynthetic genes produced from microbes have unique industrial properties, however, the diversity of such genes from extreme environments remains limited. Here, the authors use meta-omics of Arctic hydrothermal vent biofilms and uncover nearly 3000 genes, including a possibility for hydrogen cyanide production in archaea.

Introduction

Antibiotic resistance is a growing global crisis, responsible for over 1.27 million deaths annually as of 2019¹. This situation raises the need to find new antibiotics, particularly those derived from natural products, as many approved antibiotics have natural origins^1–3. These bioactive compounds are typically synthesized by BGCs, which are physically co-located sets of genes encoding the enzymes, transporters, and regulatory elements required for secondary metabolite biosynthesis⁴. BGCs play important ecological roles by mediating chemical interactions within microbial communities⁵. Their products, known as secondary metabolites, can function as antimicrobials, toxins, siderophores, or signaling molecules, facilitating competition, cooperation, or adaptation in diverse environments⁵. BGCs are recognized as an important source of clinically valuable natural products, such as antibiotics (e.g., streptomycin, erythromycin), anticancer agents (e.g., doxorubicin), and immunosuppressants (e.g., rapamycin)³. Due to their essential biological activities, BGC-derived metabolites remain central to drug discovery efforts^6,7. In response to rising antibiotic resistance, genome mining of uncultivated microorganisms, particularly those from underexplored environments, is increasingly employed to identify novel BGCs with therapeutic potential^6,8.

Deep-sea hydrothermal vent ecosystems represent one of the most extreme and dynamic habitats on Earth, where primary production is predominantly fueled by chemosynthesis rather than photosynthesis⁹. Since the first discovery in the late 1970s¹⁰, a vast microbial diversity has been discovered on a global scale^11–13. The use of single-cell and genome-resolved metagenomics has been a driver for microbial discoveries in hydrothermal and subseafloor environments^14–25. Coupled with improved analytical and computational tools, this has led to the discovery of novel metabolisms, novel microbial lineages, as well as known metabolisms in new lineages^26–29. Prokaryotic BGCs from various marine environments, including water columns, deep-sea sediments, cold-seep sediments, and bacterial mats, have been shown to encode a wide array of novel secondary metabolites, many with potential antimicrobial properties^21,30–34. In contrast, the microbial communities inhabiting deep-sea hydrothermal vent biofilms remain largely unexplored in this context, despite their biodiversity and potential to yield previously unrecognized biosynthetic pathways.

Biofilm-forming microorganisms play important ecological roles in vent ecosystems by structuring microbial assemblages and mediating nutrient flux³⁵. Given the spatial and resource limitations within biofilms, microbial interactions in marine biofilms occur through various mechanisms, including cooperation, communication, and competition³⁶. The production and secretion of secondary metabolites, such as toxins and antibiotics, are common strategies employed by microbes to compete within biofilm communities^36,37. Despite this, deep-sea hydrothermal vent biofilms have been largely overlooked as a source of secondary metabolites, and to date, BGCs in biofilm communities from Arctic Mid-Ocean Ridges (AMOR) hydrothermal systems have not been systematically described.

In this study, we explore the biodiversity and biosynthetic potential of biofilm microbial communities from the Loki’s Castle, Ægir Vent Field, and Seven Sisters vent field along the AMOR. By integrating Nanopore long-read metatranscriptomic data with BGC predictions from MAGs, we assess the transcriptional activity and site-specific expression profiles of BGCs across distinct vent ecosystems. Biofilm communities were taxonomically diverse, with bacteria dominating across all samples. Common deep-sea hydrothermal vent phyla such as Pseudomonadota, Bacteriota, Aquificota, and Campylobacteota were consistently present²¹. RiPPs represented the most prevalent BGC class, both at the genomic and transcriptomic levels, and were detected in both bacterial and archaeal lineages. Comparative analysis with the Minimum Information about a Biosynthetic Gene Cluster database (MIBiG)⁴ revealed that a large proportion of predicted BGCs do not cluster with any known reference, underscoring the uniqueness of hydrothermal vent biofilms as rich reservoirs of previously uncharacterized secondary metabolites. Notably, long-read transcriptomics data confirmed that many of these BGCs are actively expressed in situ, providing direct evidence that microbes within these extreme environments produce these secondary metabolites as part of their ecological strategy. This is the first comprehensive report on the microbial communities and biosynthetic capabilities of the Ægir vent field, and the first observation of the potential for HCN synthesis in the domain Archaea, specifically from an Asgardarchaeota-affiliated MAG. Together, our findings reveal the ecological importance of secondary metabolism in hydrothermal vent biofilms and highlight their untapped potential for natural product discovery.

Results

Microbial composition across biofilm communities

To resolve the microbial diversity and biosynthetic capacity of Arctic hydrothermal vent biofilms, we performed an integrated metagenome- and metatranscriptome-based study of nine unique hydrothermal biofilms (Fig.1, Table 1). The biofilms were collected from three distinct hydrothermal vent fields along AMOR, the shallow-water Seven Sisters vent field, the deep-sea Ægir vent field, and the deep-sea Loki’s Castle vent field. From the biofilms, we recovered 1147 MAGs. In addition, long-read nanopore metatranscriptomic data were obtained from seven of the biofilms. Sequencing information is given in Supplementary Data 1 and Supplementary Data 2.

Fig. 1. Sampling locations and biofilm collection across Arctic mid-ocean ridge hydrothermal vent fields.

Sampling locations shown in the map. A–C and E–H intra-vent field sampling of biofilms from the Ægir Vent Field (Table 1; Metagenome ID M24, M25, M26, M36, M37, M38 and M39), D the Barite Field within the Loki’s Castle Vent Field (Table 1; Metagenome ID M33), and I the Seven Sisters Vent Field (Table 1; Metagenome ID M42). Sampled biofilms using a hydraulic biosyringe attached to the ROV arm are indicated by a red arrow. Samples were collected during four sampling campaigns by AMOR on board the research vessel G.O. SARS (Summer of 2017, 2020, 2021, and 2022). The map image is the intellectual property of Esri and is used herein under license. Copyright © 2026 Esri and its licensors. All rights reserved.

Table 1.

Samples and collection sites used in the study

Sample ID	Vent field	Metagenome ID	Metatranscriptome ID	Depth (m)	BioProject
GS21-ROV-106-ROV432	Ægir Vent Field	M24	c24	2288	PRJNA1301805
GS20-ROV9-BS1	Ægir Vent Field	M25	c25	2288	PRJNA881934
GS20-ROV9-BS2	Ægir Vent Field	M26	NA	2288	PRJNA1301797
GS17-BB	Loki’s Castle Vent Field (Barite field)	M33	c33	2390	PRJNA1301837
GS22-107-ROV569-BS1	Ægir Vent Field	M36	c36	2306	PRJNA1301809
GS22-107-ROV569-BS2	Ægir Vent Field	M37	c37	2308	PRJNA1301812
GS22-107-ROV579-BS1	Ægir Vent Field	M38	c38	2304	PRJNA1301820
GS22-107-ROV579-BS2	Ægir Vent Field	M39	c39	2317	PRJNA1301824
GS21-106-ROV435A-BS	Seven Sisters Vent Field	M42	NA	165	PRJNA1301830

After de-replication, 1140 medium-to-high quality bacterial and archaeal MAGs (minimum contig size >2000 bp) were retained (Supplementary Data 3). Among them, 124 were classified using GTDB-tk with database version R226 into 10 archaeal phyla, while 1016 MAGs belonged to 51 bacterial phyla. Within bacteria, Pseudomonadota was the most prevalent phylum, detected in all samples (327 MAGs), followed by Bacteroidota (156) (Fig. 2A). Thermoproteota was the dominant archaeal phylum, present in six of the nine metagenomes (Fig. 2B). Detailed taxonomic information is provided in the supplementary materials (Supplementary Data 3). Sample M25 exhibited the highest biodiversity, containing 254 MAGs from 27 bacterial and 10 archaeal phyla (Fig. 2A, B). Bacterial MAGs dominated all samples, but archaeal abundance was noticeably high in M25 and M39, with 60 and 24 archaeal MAGs, respectively. The archaeal taxa present suggest an elevated temperature in these samples, supported by the measured sediment temperature using a live temperature probe connected to the remotely operated vehicle (ROV). Just below the biofilm in M39, the temperature was measured at 33 °C, and at 10 cm sediment depth, the temperature was elevated to 96 °C. The recently described phylum JALSQH01 from deep-sea hydrothermal deposits²¹ was classified into three MAGs from sample M25. This suggests that this new taxon is more widely distributed and specific to the hydrothermal vent ecosystem.

Fig. 2. Taxonomic composition of bacterial and archaeal MAGs in AMOR biofilms.

The number of MAGs identified in each of the nine biofilms, grouped by phylum within: A Bacteria; B Archaea. Taxonomic classifications are based on the Genome Taxonomy Database (GTDB v. R226).

Relative abundance analysis based on mapped reads to de-replicated MAGs using CoverM revealed that Pseudomonadota dominated seven of the nine biofilm communities, while Campylobacterota prevailed in samples M26 and M36. Notably, Aquificota demonstrates significant relative abundance in samples M25 (3.53%), M26 (11.85%), and M42 (10.56%). Bacteroidota is also an important bacterial phylum within the biofilm communities, contributing >1% to the total relative abundance in all samples except M26. Overall, between 20 and 75.3% of the sequenced microbial community was represented by the MAGs in the study. Detailed information on the relative abundance of each phylum is provided in the supplementary materials (Supplementary Data 3). The presence of well-known deep-sea groups such as Pseudomonadota, Campylobacterota, Bacteroidota, Aquificota, Chloroflexota, and Planctomycetota aligns with other studies on deep-sea microbial communities^21,33,38,39. Desulfobacterota was detected at >1% relative abundance in samples M37, M38, and M39, suggesting adaptation to sulfur-rich environments. Sample M39 exhibited the highest diversity of this phylum, with 22 MAGs classified as Desulfobacteria. M25 showed the greatest diversity within the archaeal phylum Thermoproteota, with 31 MAGs classified to this phylum, accounting for 6.47% of its metagenome. 29 of these MAGs belonged to Sulfolobales, a group specialized in sulfur metabolism and known for thriving in high-temperature, acidic, and alkaline environments^40,41. Halobacteriota was another archaeal phylum frequently observed in our biofilms, exhibiting a relative abundance of 1.2% in sample M38, where 4 MAGs were identified, and 2.32% in M39, with 8 MAGs detected.

Community dissimilarity analysis based on Bray-Curtis and Sørensen indices revealed distinct clustering among the samples. Bray-Curtis dissimilarities, which account for relative abundance, showed that sample M26 was significantly different from the others (0.70−0.89), with high bootstrap confidence intervals (upper CI > 0.95 in most pairwise comparisons) (Supplementary Fig. 1). In contrast, samples such as M24, M36, and M39 exhibited lower dissimilarity values (e.g., M24 and M39, 0.49), indicating more similar community structures. This was supported by Sørensen dissimilarities, which are based on the presence or absence of taxa, with M26 showing high dissimilarity (e.g., M26 and M37, 0.81; M26 and M42, 0.86), suggesting a distinct taxonomic composition (Supplementary Fig. 2). Permutation tests showed that most pairwise similarities did not differ significantly (p > 0.1). These results indicate that M26 harbors a microbial community that is compositionally distinct from the other samples. Meanwhile, samples M24, M36, M38, and M39 appear to form a more cohesive group in terms of both shared taxa and relative abundances.

Linking diversity and transcription of BGCs across hydrothermal biofilms

Metagenomic analysis of nine hydrothermal biofilm samples revealed a total of 2965 BGCs across 870 MAGs (Fig. 3A). The dominant phylum, Pseudomonadota, encoded the largest number of BGCs (n = 1258), followed by Bacteroidota (n = 299) and Myxococcota (n = 277). BGCs were distributed across 46 of 51 bacterial and 7 of 10 archaeal phyla, with Nitrospirota exhibiting the highest BGC density. Only 11.97% (n = 355) of BGCs were complete (i.e., not located at contig edges, and thus less likely to be truncated), while the majority (n = 2610) were fragmented. RiPPs were the most abundant BGC class (n = 1151), followed by other (n = 638), Terpene (n = 394), NRP (n = 383), Polyketide (n = 236), Hybrid (n = 162), and Saccharide (n = 1) (Fig. 3B, Supplementary Fig. 6A, Supplementary Data 4). RiPP-like clusters, reclassified from bacteriocins in antiSMASH since the release of version 6.0⁴², were the most frequently predicted subtype (n = 633) (Fig. 3B; Supplementary Fig. 3A). RiPPs were consistently dominant across all biofilm samples (Fig. 3C), suggesting a conserved ecological role. In comparison, archaeal MAGs encoded 50 BGCs across seven phyla, with lower diversity and representation limited to RiPP, Other, and Hybrid classes (Fig. 4).

Fig. 3. Diversity of BGCs and predicted products in AMOR biofilms.

A Abundance of individual BGC classes, B Predicted product types, with circle sizes indicating relative abundance. C Composition of BGC classes across nine metagenomes, excluding saccharides. BGCs are categorized based on the biosynthetic classes of their core enzymes, as assigned by antiSMASH v7.0 and BiG-SLiCE v2.0.0.

Fig. 4. Distribution of BGCs and predicted metabolites in archaeal MAGs.

The number and types of BGCs identified across archaeal phyla from nine biofilms, categorized by BGC class and predicted metabolite. Taxonomic classifications are based on GTDB release R226. BGC classes and predicted products were assigned using antiSMASH v7.0 and BiG-SLiCE v2.0.

To assess functional expression, metatranscriptomic profiling using long-read nanopore sequencing was successful on seven of the nine biofilm samples. A total of 1450 BGCs were actively transcribed across 519 MAGs, spanning 41 microbial phyla: 35 bacterial (Fig. 5) and 6 archaeal (Fig. 6). Expression patterns closely mirrored metagenomic abundance, whereby Pseudomonadota, which encoded the highest number of BGCs, also exhibited the highest transcriptional activity (n = 616 expressed BGCs), followed by Bacteroidota (n = 164), Desulfobacterota (n = 136), and Myxococcota (n = 135). Sample M39 showed the highest overall BGC expression (n = 411 clusters) (Fig. 5B). The MAG Barite_M33_B89 (Pseudomonadota, candidate genus JABSOG01) exemplified the link between genomic potential and transcriptional output. antiSMASH identified 107 BGC regions containing core biosynthetic domains in this MAG, spanning NRPS (n = 74), polyketides (n = 15), RiPPs (n = 8), Hybrids (n = 6), and Other (n = 4), and exhibited the highest number of expressed clusters (n = 75), suggesting that this MAG may represent a potentially important biosynthetic contributor within its biofilm community.

Fig. 5. BGC richness and expression across Bacterial phyla in the AMOR biofilms.

A normalized BGC counts across bacterial phyla in all samples, with phyla represented by a single MAG indicated in bold. Taxonomic classifications are based on GTDB release R226, and B expression profiles of bacterial BGC classes in AMOR biofilms. The heatmap shows the expression levels of predicted BGC classes across the seven AMOR biofilms. Expression levels were estimated using BiG-MAP v1.0.0 and normalized as average log₁₀-transformed TPM values. BGC detection and classification were performed using antiSMASH v7.0 and BiG-SLiCEv2.0.0.

Fig. 6. BGC richness and expression across Archaeal phyla in AMOR biofilms.

A Normalized BGC counts across archaeal phyla in all samples, and B expression profiles of archaeal BGC classes in the AMOR biofilms. The heatmap shows the expression levels of predicted BGC classes across the seven AMOR biofilms. Expression levels were estimated using BiG-MAP v1.0.0 and normalized as the average log₁₀-transformed TPM values. Taxonomic classifications are based on the GTDB release R226; BGC detection and classification were performed using antiSMASH v7.0 and BiG-SLiCEv2.0.0.

RiPPs were the most transcriptionally active BGC class (n = 532), consistent with their genomic prevalence (Figs. 5B and 6B). Expression levels, quantified as transcripts per million (TPM), are detailed in Supplementary Data 5. Statistical analysis (Kruskal–Wallis with Dunn’s post-hoc test) revealed significant differences in expression among BGC classes (Supplementary Fig. 6B; Supplementary Data 7). RiPPs showed significantly higher expression than Hybrid (p = 0.0238), Polyketide (p = 0.0008), and NRP (p = 0.0002) clusters after Bonferroni correction (Supplementary Data 7). Similarly, the Other class showed significantly higher expression compared to NRP (p = 0.0127) and Polyketide (p = 0.0179). No significant differences were observed among Hybrid, NRP, Polyketide, and Terpene groups (p > 0.05).

Functional roles of expressed BGCs in biofilm adaptation

Several BGCs linked to microbial adaptation were both genomically encoded and transcriptionally active. Aryl polyene (APE) BGCs (n = 253), known to enhance oxidative stress tolerance and biofilm formation⁴³, were consistently expressed across all metatranscriptomes, particularly in Pseudomonadota, Bacteroidota, Campylobacterota, Desulfobacterota, Myxococcota, Acidobacteriota, Bdellovibrionota, Planctomycetota, Verrucomicrobiota, Omnitrophota and Zhuqueibacterota (Fig. 7, Supplementary Fig. 3B, Supplementary Data 5). Ectoine BGCs (n = 14), primarily from Pseudomonadota, were detected in eight metagenomes and actively transcribed in M24, M36, M37, and M38, suggesting roles in osmotic and thermal stress mitigation⁴⁴. BGCs associated with nutrient acquisition and regulation included phosphonates (n = 17), thioamitides (n = 8), and siderophores (n = 2)^30,45. Quorum sensing and biofilm regulation were supported by homoserine lactone (hserlactone) BGCs (n = 98) and ranthipeptides (n = 51)^46–48. Hserlactone BGCs were present in all metagenomes, and transcriptional activity was detected in all seven screened metatranscriptomes, indicating the importance of this BGC in biofilm development (Supplementary Data 5).

Fig. 7. Expression profiles of BGC products in AMOR biofilms.

Heatmaps show the expression levels of predicted BGC products across seven AMOR biofilms, separated by domain in A Bacteria and B Archaea. Expression levels were estimated using BiG-MAP v1.0.0 and normalized as average log₁₀-transformed TPM values.

Hydrothermal biofilms harbored a diverse array of potential antimicrobial BGCs, many of which were actively transcribed (Fig. 7). Among RiPP subclasses, we detected thiopeptides (n = 23)⁴⁹, lanthipeptides (n = 20)^39,50, lasso peptides (n = 8)⁵¹, sactipeptides (n = 6)⁵², and linear azol(in)e-containing peptides (LAPs; n = 4)⁵³. Several of these RiPP families include members with known antimicrobial activities^50,54,55. Additional antimicrobial classes included β-lactones (n = 73)⁵⁶, N-acyl amino acids (n = 65)⁵⁷, resorcinols (n = 34)⁵⁸, NRPS (n = 203)⁵⁹, and polyketide synthases (T1PKS: n = 78; T3PKS: n = 96)^60,61. Although many of the detected BGCs are linked to antimicrobial activity in other organisms, their functional roles within these hydrothermal biofilms are not yet clear. Evidence indicates that such metabolites can act as signaling compounds at low concentrations, and some BGCs may remain silent unless triggered by specific environmental conditions^6,54. The MAG Barite_M33_B89 (phylum Pseudomonadota, candidate genus JABSOG01) was a major contributor to antimicrobial potential, expressing BGCs across multiple classes (Supplementary Data 5). These BGCs spanned five classes: NRPS (74), polyketides (15), RiPPs (8), Hybrids (6), and Other (4). Given that NRPs, polyketides, and RiPPs are renowned for producing diverse antimicrobial compounds^8,30,62, this understudied genus, JABSOG01, represents a compelling candidate for future studies on novel natural bioactive compounds. The MAG Aegir_M24_B72, classified within the phylum Bacteroidota, contained the longest BGC (105.77 kb; Supplementary Data 4), predicted to encode an NRPS–T1PKS hybrid.

Network analysis of biosynthetic gene clusters

To evaluate the uniqueness of BGCs recovered from biofilm communities, we performed a comprehensive network analysis comparing BGCs identified in nine biofilm samples with those curated in the MIBiG database (version 3.1). In total, 5450 BGCs were analyzed, comprising reference BGCs from MIBiG and 2953 BGCs from our metagenomic dataset. These BGCs were clustered into 4953 gene cluster families (GCFs), of which 4643 were singletons, and 310 contained two or more BGCs (Fig. 8A; Supplementary Fig. 7, Supplementary Data 6). Strikingly, only a single GCF included BGCs from both the biofilm-derived dataset and the MIBiG reference set, underscoring the pronounced divergence between these two sources. This minimal overlap highlights the distinct biosynthetic repertoire of biofilm-associated microorganisms and suggests the presence of novel metabolic pathways not represented in existing databases. The high proportion of singleton GCFs further supports the hypothesis that biofilm communities harbor a largely untapped reservoir of unique biosynthetic potential. Pairwise cluster sharing analysis revealed substantial variability in BGC composition among biofilm samples. For instance, samples M24 and M39 shared 75 GCFs, as did M25 and M39, while M24 and M25 shared 50 GCFs (Fig. 8B). Sample M36 exhibited moderate overlap with M39 (46 shared GCFs), M25 (25), and M24 (22), indicating partial biosynthetic similarity. In contrast, M26 was markedly distinct, sharing only 9 GCFs with M39 and between 2–4 GCFs with other samples. Notably, only a single GCF was shared between MIBiG and the biofilm samples M36, M37, and M38, further reinforcing the novelty of the biofilm-derived BGCs.

Fig. 8. Cluster analysis of BGCs from AMOR biofilms.

A Sequence similarity network of GCFs identified using BiG-SCAPE v1.1.5 (similarity cutoff = 0.3). Each node represents a BGC, colored by class as assigned by BiG-SCAPE. Circular nodes indicate BGCs from AMOR biofilms, and triangular nodes represent reference BGCs from the MIBiG v3.1 database. Singletons are excluded for clarity. B UpSet plot showing the number of shared GCFs across biofilms, based on the sequence similarity network.

Previously unrecognized biosynthetic capacities in archaea

A total of 60 BGCs associated with HCN synthesis were identified, including one in the archaeal MAG Aegir_M39_B184, taxonomically assigned to the class Heimdallarchaeia within the phylum Asgardarchaeota (Supplementary Fig. 4A and 4B). HCN is a well-characterized virulence factor predominantly found in bacterial taxa such as Pseudomonas spp⁶³. Its biosynthesis is mediated by HCN synthase, encoded by the canonical hcnABC operon, which catalyzes the oxidative decarboxylation of glycine to yield carbon dioxide and HCN⁶⁴, a process termed cyanogenesis. This pathway is tightly regulated by environmental conditions, including iron and phosphate availability, as well as microaerobic conditions⁶⁴.

In contrast to the tripartite hcnABC operon of Pseudomonas fluorescens (BGC0002345, MIBiG 3.1), the putative HCN synthase in Aegir_M39_B184 is encoded by two genes, hcnAB and hcnC (Protein file: https://doi.org/10.6084/m9.figshare.30788630). Structural modeling using the AlphaFold3⁶⁵ webserver (https://deepmind.google/science/alphafold/) and comparative superimposition with the structural model of hcnA and hcnB from P. fluorescens (Supplementary Fig. 5A), support the presence of a complete and potentially functional HCN synthase. A high structural similarity was also observed for the hcnC subunit (https://doi.org/10.6084/m9.figshare.30788579). Furthermore, transcriptomic data confirmed expression of both genes (Supplementary Data 5). Although the identified proposed hcnAB and hcnC genes were located on a contig edge, high-scoring blastp hits to other Archaea, in particular Heimdallarchaeia from Guayamas Basin⁶⁶ containing additional flanking genes, support the presence of a complete HCN cluster in Archaea (Supplementary Data 8). Comparative genomic analysis using antiSMASH v8.0.4, revealed a homologous BGC architecture in another Asgardarchaeota genome, Candidatus Prometheoarchaeum syntrophicum strain MK-D1 (NZ_CP042905.1), in addition to the Guayamas MAGs (Supplementary Fig. 4C). A superimposed model of hcnAB from Aegir_M39_B184 and a Candidatus Heimdallarchaeaceae archaeon MAG from Guayamas Basin (MHA1303415.1) not located at a contig edge, show high structural similarity (Supplementary Fig. 5B). Overall, this supports the potential for archaeal cyanogenesis.

Additionally, the MAG Aegir_M39_B193, classified within the phylum Thermoproteota, harbors a BGC predicted to encode the biosynthesis of 2-deoxystreptamine, a central scaffold of aminoglycoside antibiotics such as streptomycin, kanamycin, and gentamicin⁶⁷. This compound is predominantly associated with bacterial lineages, particularly within the phyla Actinomycetota and Bacillota^68–70. The presence of aminoglycoside biosynthetic pathways in archaeal genomes is exceedingly rare and, to our knowledge, has not previously been reported in hyperthermophilic archaea such as Thermoproteota⁷, particularly with supporting transcriptomic evidence. To limit the probability of these features being part of contaminating contigs, an additional quality assessment was made using MAGpurify2. The genes of the HCN synthase and the aminoglycoside pathway were retained on contigs with low contaminant probability scores (both 0.05), classified to the same taxa as the overall MAGs, class Heimdallarchaeia within the phylum Asgardarchaeota and class Thermoproteia within the phylum Thermoproteota, respectively.

Discussion

Microbial biofilms from extreme environments, shaped by extreme environmental pressures and adapted through cooperative and competitive interactions^71,72, represent promising reservoirs of novel bioactive compounds. Despite their potential, the biosynthetic landscape of Arctic deep-sea hydrothermal biofilms remains largely unexplored. In this study, we investigated nine biofilms from three Arctic hydrothermal vent sites, analyzing BGC diversity across 1140 archaeal and bacterial MAGs. Our findings underscore the potential biosynthetic richness of these biofilms and highlight their promise as untapped sources of antimicrobial agents. The microbial consortia inhabiting Arctic hydrothermal biofilms exhibit remarkable taxonomic and metabolic diversity, encompassing a broad-spectrum of bacterial and archaeal phyla. BGCs serve as adaptive molecular tools, conferring selective advantages that promote microbial persistence and competitive success in these extreme environments. Notably, the M26 biofilm displayed a distinct community composition despite originating from the same vent field (Ægir vent field) as six other samples, suggesting that biofilm assembly is influenced not only by environmental parameters but also by intra-community interactions. Diversity metrics based on abundance and presence/absence data support niche specialization, as shown by the unique BGC repertoire identified in M26. These patterns point to a metabolically distinct microbial consortium structured within this biofilm. The average number of BGCs per MAG (2.60) in the AMOR biofilms exceeds the global marine average, emphasizing the influence of habitat on secondary metabolism³⁴. For instance, Antarctic soils yielded 1,417 bacterial BGCs with no archaeal BGCs, whereas cold seep communities contained 238 archaeal BGCs among 487 MAGs, predominantly encoding RiPPs^30,73. These observations suggest that selective pressures in deep-sea ecosystems favor the retention and expression of RiPP BGCs as a survival strategy. RiPPs emerged as the most abundant BGC class across all nine biofilms, contrasting with other extreme environments such as Antarctic soils and Whalers Bay sediments, where terpenes predominate^73,74. This divergence suggests that adaptation to hydrothermal conditions may favor RiPP biosynthesis, potentially due to their roles in microbial antagonism and communication. The consistent dominance of RiPPs across samples, rather than being driven by outliers, indicates community-wide selection for these compounds. This pattern aligns with global surveys of marine microbial genomes, which also report RiPPs and terpenes as the most prevalent BGC classes^7,34. Moreover, RiPP BGCs in our dataset encode a diverse array of potential antimicrobial peptides, some of which have previously demonstrated in vitro antimicrobial activity³⁴. For example, thiopeptides, detected in both the metagenomic and metatranscriptomic datasets, are well-known for their potent antibiotic activity against Gram-positive bacteria^49,75,76. Sactipeptides, also detected in both the metagenomic and metatranscriptomic datasets, exhibit broad-spectrum activity against Gram-positive and Gram-negative bacteria^50,77–79. While some members of these RiPP classes are known to exhibit antimicrobial activity, such functional properties cannot be inferred without detailed characterization of individual clusters. The expression of these antimicrobial peptides in AMOR biofilms suggests that they play an important role in microbial competition within the biofilm.

From an ecological perspective, APE BGCs, encoding widespread bacterial pigments⁴³, were consistently detected and transcriptionally active across all samples, indicating their potential role in microbial fitness and biofilm stability. Phosphonate BGCs, which contribute to marine phosphorus cycling⁴⁵, were identified in 17 MAGs across five biofilms and seven bacterial phyla. Due to their frequent occurrence in marine and cold seep prokaryotic genomes^30,31, transcriptomic activity was detected from three samples (M25, M33, and M39) in five bacterial phyla (Pseudomonadota, Bacteroidota, Myxococcota, Campylobacterota, and Bdellovibrionota). In contrast, siderophore BGCs were rare, with only two detected (M25 and M26, phylum Pseudomonadota) and no expression detected, suggesting limited iron acquisition strategies in these biofilms.

Thioamitide BGCs, previously reported in the archaeal phylum Halobacteriota from cold seeps³⁰, were also identified in our dataset, spanning bacterial phyla Actinomycetota and Pseudomonadota, as well as archaeal phyla Halobacteriota (order Alkanophagales). These clusters mediate thioamidation, a biochemical process replacing an amide oxygen with sulfur⁸⁰. Prior studies proposed that Halobacteriota members utilize thioamide pathways to modify methyl-coenzyme M reductase, facilitating anaerobic oxidation of methane (AOM)³⁰. Consistent with this, MAGs Barite_M33_B83 and Aegir_M39_B54 (order Alkanophagales) exhibited transcriptomic expression of a thioamitide BGC, indicating active AOM metabolism. Another Alkanophagales MAG (Aegir_M38_B4) expressed ranthipeptides, metabolites previously considered bacterial and only recently reported in archaea^48,81, expanding their known distribution and suggesting a role in Halobacteriota adaptability. Collectively, these findings demonstrate that BGC-encoded metabolites contribute to microbial resilience in extreme environments and influence biofilm dynamics through selective pressures, interspecies signaling, and physiological protection. Furthermore, NRPS BGCs, another prominent class of antimicrobial producers, were detected in eight of nine biofilms. M33 was notably enriched, harboring 91 of the 203 NRP generic clusters. Transcriptomic data revealed active expression of 128 NRP generic BGCs, including 65 from M33, affiliated with diverse hydrothermal vent phyla such as Pseudomonadota, Planctomycetota, Campylobacterota, Chloroflexota, Calditrichota, Bacteroidota, Desulfobacterota, Myxococcota, Verrucomicrobiota, and Zhuqueibacterota. These data suggest that NRPS-derived metabolites play a significant role in microbial competition and niche establishment. Terpenes, though less abundant than RiPPs and NRPS, were detected in all metagenomes (394 BGCs), exclusively in bacterial taxa. Their known roles in microbial signaling and defense⁸², coupled with transcriptomic evidence from 162 BGCs from all screened metatranscriptomes, highlight their functional relevance³⁴.

The discovery of HCN biosynthetic pathways in archaeal genomes expands the known metabolic repertoire of this domain and challenges the long-standing assumption that cyanogenesis is restricted to bacteria and eukaryotes. This suggests that archaea may utilize HCN as a chemical mediator in ecological interactions, particularly within the competitive and chemically dynamic environment of deep-sea biofilms. Given HCN’s established role in microbial antagonism and defense⁸³, its presence in archaea raises intriguing questions about convergent evolution or horizontal gene transfer events. The broader taxonomic distribution of cyanogenic BGCs observed here, extending beyond traditional producers such as Pseudomonas and Cyanobacteria⁸⁴, indicates that HCN synthesis may be a more widespread and ecologically relevant trait in marine microbial communities. The expression of these BGCs in multiple biofilm samples further supports their functional role in situ, suggesting that cyanogenesis contributes to niche differentiation, microbial competition, and defense strategies in hydrothermal ecosystems. These findings prompt a reevaluation of archaeal roles in chemically mediated interactions and highlight the need for further investigation into the evolutionary origins and regulatory mechanisms of archaeal cyanogenesis. However, additional biochemical and/or cultivation efforts, integrating comparative genomics, transcriptomics, and metabolomics, are needed to verify the HCN pathway in archaea and to elucidate the functional significance of HCN production and its impact on biofilm community dynamics.

In summary, the diversity and expression of BGCs across AMOR biofilms reflect finely tuned adaptive strategies to the physicochemical pressures of hydrothermal vent ecosystems. The broad-spectrum of predicted bioactive compounds, many with potential antimicrobial properties, positions these biofilm-associated microbiomes as a valuable and underexplored resource for biotechnological and pharmaceutical innovation. As our analyses are based on MAGs of varying completeness, some predicted BGCs (e.g., those located at contig edges) may represent partial clusters or isolated biosynthetic genes rather than complete operons. This limitation is inherent to rule-based prediction tools such as antiSMASH and may also affect downstream clustering and novelty assessment using BiG-SCAPE⁸⁵. Therefore, while the overall diversity of predicted BGCs in AMOR biofilms remains robust, the absolute number of clusters should be interpreted with caution, and future work is required to validate their completeness and presence in situ.

Methods

Sampling sites and collection

Biofilm samples were collected at three hydrothermal locations, both from shallow water and deep-sea, along the AMOR. In the current study 9, biofilms were collected: one from the shallow-water even Sisters vent field (125 m depth)⁸⁶, one from the diffuse venting area at the Loki’s Castle vent field (Barite field, 2340 m depth)⁸⁷, and seven samples from the Ægir vent field (2310 m depth)⁸⁸ (Table 1). Samples were collected using the Ægir6000 ROV on board the R/V G.O. SARS, equipped with a biosyringe (a hydraulic sampling cylinder) connected to the ROV manipulator arm. Sampling was conducted during sampling campaigns in 2017, 2020, 2021 and 2022 to AMOR, led by the Centre for Deep Sea Research at the University of Bergen. Samples retrieved onboard R/V G.O. SARS were centrifuged at 6000 x g for 5 minutes and the supernatant wewasemoved. Aliquots were frozen in liquid nitrogen and stored at −80 °C until processing.

Genome-resolved metagenomics

DNA was extracted using FastDNA Spin Kit for Soil (MP Biomedicals) from frozen samples and sent to the Norwegian Sequencing Centre (University of Oslo, Norway) for shotgun metagenomic sequencing. 150-bp paired-end sequencing was performed using an Illumina NovaSeq S4 flow cell. Raw reads were scanned and filtered for quality, duplication rate, and adapter contamination using fastp (v0.23.2)⁸⁹.

Filtered reads from individual biofilms were assembled into contigs by MEGAHIT (v1.2.9)⁹⁰ with parameters --min-contig-len 2000 --presets meta-large. Contigs with lengths longer than 2000 bp were subjected to automatic MAG reconstruction using metaWRAP (v1.3.2)⁹¹ which implements the combinatorial use of MaxBin2 (v2.2.6)⁹², CONCOCT (v1.0.0)⁹³, and MetaBat2 (v2.12.1)⁹⁴ for MAG refinement. The quality of each MAG was estimated with CheckM1 (v1.0.12) and CheckM2 (1.1.0)⁹⁵ in addition to the presence/absence of rRNA and tRNA determined by Barrnap v0.9⁹⁶ and tRNAscan v2.0.12⁹⁷, respectively. Low-quality genomes following the MIMAG standard⁹⁸ (completeness <50% and contamination >10%) were removed from downstream analysis, with the exception of 25 MAGs first estimated to be medium-quality by checkM1. High-quality MAGs are defined with completeness ≥90% and contamination ≤5%, and include at least one copy of 16S, 23S, and 5S rRNA genes and ≥18 unique tRNAs. Medium-quality MAGs are defined with completeness ≥90% and contamination ≤10%. In addition to the quality assessment described above, key MAGs were further purified using MAGpurify2⁹⁹ to reduce the possibility of contaminating contigs using --probability_threshold 0.7 for removal of contaminants.

De-replication of MAGs from all 9 biofilms was performed at the species level using dRep (v3.4.0)¹⁰⁰ with parameter settings -l 40000 -comp 50 -con 10 --S_ani 98. Although stringent, an ANI value of 98% was selected to balance genome diversity while minimizing cross-mapping issues and avoiding redundancy >98%¹⁰¹. Relative abundance of de-replicated MAGs in each biofilm was estimated by mapping filtered paired-end reads from each biofilm sample to MAGs using CoverM v0.6.1 (https://github.com/wwood/CoverM). Taxonomic classification was performed using the Genome Taxonomy Database^102,103 via the Genome Taxonomy Database Toolkit¹⁰⁴ (GTDB-Tk, v2.5.2 – GTDB R226 taxonomy) with the classify_wf using default parameters.

RNA extraction, complementary DNA synthesis, and nanopore sequencing

RNA was extracted using the Quick-DNA/RNA Miniprep Kit (Zymo Research, USA), and subsequent RNA purification steps utilized HighPrep RNA Elite beads (MagBio Genomics Inc., USA). Ribosomal RNA (rRNA) was removed from total RNA samples for all samples, except for c25, with the Pan-Prokaryote riboPOOL (Bacteria and Archaea) (siTOOLs Biotech GmbH, Germany). Polyadenylation was performed using E. coli Poly(A) Polymerase (New England Biolabs, USA). Complementary DNA (cDNA) synthesis was carried out using the Ligation Sequencing Kit V14, and sequencing was performed on the MinION Mk1C device (Oxford Nanopore Technologies plc, UK). The full description of the adapted protocol following the Nanopore sequencing guidelines for microbial metatranscriptomics is available at protocols.io (https://doi.org/10.17504/protocols.io.dm6gpzz9plzp/v1). Sequencing utilized the Oxford Nanopore Flow Cell R10.4.1 (FLO-MIN114, Oxford Nanopore Technologies, UK).

Biosynthetic gene clusters; screening and expression profiling

Gene annotation was conducted on MAG sequences using the Prokka annotation pipeline (version 1.14.6)¹⁰⁵. The dataset was divided into two separate runs for archaeal and bacterial MAGs. The following parameters were applied: --addgenes, --force, --increment 10, --compliant, --centre NorSeq, --norrna, and --notrna. Secondary metabolite BGCs were identified and analyzed using antiSMASH (version 7.0)¹⁰⁶ with the following options: --cc-mibig, --cb-general, --cb-knownclusters, --cb-subclusters, --asf, --pfam2go, --smcog-trees, --rre, and --genefinding-tool prodigal. Predicted BGCs were subjected to downstream analysis with BiG-SLiCE (version 2.0.0)¹⁰⁷ to extract features, construct GCF models, and assign GCFs.

A network analysis of recovered BGCs was performed using BiG-SCAPE v1.1.5¹⁰⁸. The program was executed in ‘--auto’ mode. Networks were computed with the ‘mix’ and ‘mibig’ options, which analyze all classes together and incorporate MIBiG v3.1¹⁰⁹ BGCs into the network. The ‘--include singletons’ option was enabled to visualize singleton BGCs. Multiple raw distance cutoff values (ranging from 0.1 to 1.0) were tested, and networks generated with a 0.3 cutoff were selected for further analysis. Lower cutoff values are more suitable for grouping BGCs that produce identical compounds, whereas higher cutoffs provide a broader view of related product families. The resulting sequence similarity matrices were visualized using Cytoscape v3.10.3¹¹⁰. Additionally, the network annotation table generated by BiG-SCAPE was included to differentiate between the two sets of BGCs and to categorize different BGC classes.

Nanopore sequencing reads were basecalled using Dorado (version 0.5.1) (https://github.com/nanoporetech/dorado) with low-quality reads removed (--min-qscore 10) and adapter sequences trimmed (--trim). Ribosomal RNA sequences were filtered out using SortMeRNA v.4.3.6¹¹¹. The resulting cDNA sequences were mapped against their corresponding metagenomic datasets using BiG-MAP (v.1.0.0)¹¹² to profile the expression of BGCs. The BiG-MAP.map pipeline was modified to replace the default aligner with Minimap2 (v.2.26-r1175)^113,114 to ensure more accurate mapping of Nanopore long reads. Modifications to the BiG-MAP.map module were permitted under the tool’s MIT license, and the modified script is provided in the Data Availability section of this paper. The relative abundance of biosynthetic genes across metagenomes was also estimated using the BiG-MAP.map module. For Nanopore metatranscriptomic reads, the -U option (unpaired mode) was applied, while Illumina metagenomic reads were processed using the -I1 and -I2 paired-end options. Abundance values were reported as TPM for metatranscriptomic datasets and genes per million (GPM) for metagenomic datasets.

Statistical analyses and data visualizations

All statistical analyses were conducted using Python (version 3.13.2). Principal component analysis of microbial communities across biofilm samples was performed using the scikit-learn package (v.1.6.1). Bray–Curtis dissimilarities based on relative abundances of MAGs were calculated using the pdist function from the scipy package (v.1.15.2). Pairwise community similarity based on microbial order presence-absence data was quantified using the Dice–Sørensen coefficient. Statistical significance of the observed similarities was assessed using a permutation test (1000 permutations per pair). Kruskal–Wallis tests comparing the expression of different BGC classes were performed using the kruskal function from the scipy package. Post-hoc Dunn’s tests to identify significant pairwise differences among groups were conducted with the scikit-posthocs package (v.0.11.4), using TPM for each BGC class across samples. Data visualizations were created using RAWGraphs (v.2.0.0)¹¹⁵ and Python’s Matplotlib library (v.3.10.1), with all source data provided in the supplementary materials. No statistical method was used to predetermine sample size. No data were excluded from the analyses. The experiments were not randomized, and blinding was not relevant to the current study. To ensure the reproducibility of the metatranscriptomic pipeline used in the study, the full protocol is made available on protocols.io: https://doi.org/10.17504/protocols.io.dm6gpzz9plzp/v1.

Reporting summary

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

Author contributions

T.T.N. and R.S. designed the study. T.T.N., R.S. and I.H.S. collected samples at the AMORs and contributed to formal analysis. T.T.N. and R.S. conducted bioinformatic analysis and visualization. T.T.N. and R.S. wrote the manuscript, and all authors reviewed and revised the manuscript.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Funding

T.T.N. and R.S. are supported by the Norwegian Research Council, project DeepSeaQuence; Grant No. 315427. IHS is supported by the University of Bergen, and sampling campaigns to the AMORs are supported by the University of Bergen through the Trond Mohn Foundation-funded centre for deep-sea research; Grant No. TMS2020TMT13. Open access funding provided by University of Bergen.

Data availability

Metadata of all samples and metagenomes in the study is deposited at NCBI under BioProjects listed in Table 1. Sequence data of individual MAGs has been deposited under the respective BioSample accession numbers listed in Supplementary Data 3, and all raw sequencing read data has been deposited to the Sequence Read Archive (SRA) with the following accession numbers: SRR36181650 (M24) (https://www.ncbi.nlm.nih.gov/sra/?term=SRR36181650), SRR36183670 (M25) (https://www.ncbi.nlm.nih.gov/sra/?term=SRR36183670), SRR36184633 (M26) (https://www.ncbi.nlm.nih.gov/sra/?term=SRR36184633), SRR36184876 (M33) (https://www.ncbi.nlm.nih.gov/sra/?term=SRR36184876), SRR36184925 (M36) (https://www.ncbi.nlm.nih.gov/sra/?term=SRR36184925), SRR36185683 (M37) (https://www.ncbi.nlm.nih.gov/sra/?term=SRR36185683), SRR36199626 (M38) (https://www.ncbi.nlm.nih.gov/sra/?term=SRR36199626), SRR36199942 (M39) (https://www.ncbi.nlm.nih.gov/sra/?term=SRR36199942) and SRR36203984 (M42) (https://www.ncbi.nlm.nih.gov/sra/?term=SRR36203984). Nanopore cDNA raw reads have been deposited to SRA under accession numbers: SRR36202719 (c24) (https://www.ncbi.nlm.nih.gov/sra/?term=SRR36202719), SRR36203181 (c25) (https://www.ncbi.nlm.nih.gov/sra/?term=SRR36203181), SRR36217904 (c33) (https://www.ncbi.nlm.nih.gov/sra/?term=SRR36217904), SRR36218254 (c36) (https://www.ncbi.nlm.nih.gov/sra/?term=SRR36218254), SRR36218304 (c37) (https://www.ncbi.nlm.nih.gov/sra/?term=SRR36218304), SRR36218718 (c38) (https://www.ncbi.nlm.nih.gov/sra/?term=SRR36218718), SRR36218944 (c42) (https://www.ncbi.nlm.nih.gov/sra/?term=SRR36218944). A FigShare project (https://doi.org/10.6084/m9.figshare.c.8159183) has been established for the study which includes GenBank and index files of all identified BGCs, in addition to the network file generated from BiG-SCAPE (https://doi.org/10.6084/m9.figshare.30762029), a protein sequence file of the hcnAB and hcnC from the Heimdallarchaeion Aegir_M39_B184 (https://doi.org/10.6084/m9.figshare.30788630), and a figure of the superimposed AlphaFold3 models of the hcnC from the Heimdallarchaeion Aegir_M39_B184 and Pseudomonas fluorescence PS619 (https://doi.org/10.6084/m9.figshare.30788579). Source data are provided with this paper.

Code availability

The source code and data for all statistical analyses are available through Zenodo at https://doi.org/10.5281/zenodo.20057600¹¹⁶. The Zenodo repository represents the archived version of the GitHub repository associated with this paper and ensures long-term accessibility and reproducibility.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-77384-5.