Metagenomic assembly and genome reconstruction
We used two different strategies for metagenomic assembly and genomic binning of the eighteen metagenomic datasets from Deception Island volcano. First, reads were assembled using IDBA-ud (Peng et al., 2012) (-mink 50, -maxk 92, -tep 4, -min_contig 1000) and then genomic binning was performed through MaxBin 2.0 (Wu et al., 2016). Contigs were annotated using the Integrated Microbial Genomes & Microbiomes (IMG/M) system (Markowitz et al., 2009).
Furthermore, reads were co-assembled using MEGAHIT v. 1.0.2. (Li et al., 2015), discarding contigs smaller than 1000 bp. Then contigs were binned using anvi’o v. 5 following the workflow described by Eren et al. (2015). Reads for each metagenome were mapped to the co-assembly using bowtie2 with default parameters (Langmead & Salzberg, 2012). A contig database was generated using the ‘anvi-gen-contigs-database’. Prodigal (Hyatt et al., 2010) was used to predict open reading frames (ORFs). Single-copy bacterial and archaeal genes were identified using HMMER v. 3.1b2 (Finn et al., 2011). The program ‘anvi-run-ncbi-cogs’ was used to annotate genes with functions by searching for them against the December 2014 release of the Clusters of Orthologous Groups (COGs) database (Galperin et al., 2015) using blastp v2.10.0+ (Altschul et al., 1990). Predicted protein sequences were functionally and taxonomically annotated against KEGG with GhostKOALA (genus_prokaryotes) (Kanehisa et al., 2016). Individual BAM files were profiled using the program ‘anvi-profile’ with a minimum contig length of 4 kbp. Genome binning was performed using CONCOCT (Alneberg et al., 2013) through the ‘anvi-merge’ program with default parameters. We used ‘anvi-interactive’ to visualize the merged data and identify genome bins. Bins were then manually refined using ‘anvi-refine’, and completeness and contamination were estimated using ‘anvi-summarize’.
Bins generated by the assembly and co-assembly approaches were quality checked through CheckM v. 1.0.7 (Parks et al., 2015), which is based on the representation of lineage-specific marker gene sets. Bins were taxonomically classified based on genome phylogeny using GTDB-Tk (Chaumeil et al., 2020).

Taxonomic and functional annotation of metagenome-assembled genomes (MAGs)

Bins were defined as a high-quality draft (>90% complete, <5% contamination), medium-quality draft (>50% complete, <10% contamination) or low-quality draft (<50% complete, <10% contamination) metagenome assembled-genome (MAG), according to genome quality standards suggested by (Bowers et al., 2017). We selected 11 MAGs based on their medium or high-quality and taxonomy, preferably selecting groups related to extremophiles or associated to sulfur and nitrogen metabolisms. Annotation of all predicted ORFs in MAGs was performed using prokka v.14.5 (Seemann, 2014). Further, proteins were compared to sequences in the KEGG Database through GhostKOALA (genus_prokaryotes) (Kanehisa et al., 2016) and in the SEED Subsystem through RASTtk (Brettin et al., 2015). Phenotypes were predicted using the PICA framework (Feldbauer et al., 2015) and PhenDB (https://phendb.csb.univie.ac.at/).