Samples from GB were collected on three cruises abroad the R/V New Horizon in 2004 and 2005 as described previously (Dick and Tebo, 2010). Cells for DNA and RNA extraction were filtered directly from the bottle onto a 142 mm 0.2 mM polycarbonate filter membrane by N₂ gas pressure filtration at <5 psi. Filters were preserved immediately in RNA later (Ambion, Austin, TX, USA) as recommended by the manufacturer, incubated at 4°C overnight, then stored at -20°C for the duration of the cruise and at -80°C upon return to shore. To minimize the degradation of RNA during collection, samples were kept under in situ conditions (cold and dark) during CTD retrieval and then processed immediately on deck. Although, we cannot exclude the possibility that some changes in the RNA pool occurred during sample collection (it took approximately one hour to recover CTDs from 2000 m) and processing, these artifacts should affect all samples evenly and are circumvented to some extent by (i) the vast majority of recovery time is under water at temperatures that do not deviate substantially from those of plume waters, and (ii) the comparative nature of this study (plume versus background). Metadata and chemical/physical characteristics of samples are present in Supplementary Table S1 and more detail as previous reports (Lesniewski et al., 2012; Anantharaman et al., 2013).

DNA was extracted in June of 2012 as described in detail previously (Dick and Tebo, 2010; Lesniewski et al., 2012; Anantharaman et al., 2013; Li et al., 2015). Briefly, ¼ filters were cut into small pieces and cells were lysed with bead beating and a chemical solution (300 mM EDTA, 300 mM NaCl, 300 mM Tris, pH 7.5, 1.5% SDS, 50 mM DTT, 0.5 N acetate and lysozyme). DNA was then purified with Montage PCR purification filter units and eluted in TE buffer. RNA was extracted in June 2012 from 1/4 filters with the mirVana mRNA Isolation kit (Ambion), treated with DNase I to remove DNA, and concentrated and re-purified using the RNeasy MinElute Kit (Qiagen, Valencia, CA, USA). Total RNA was quantified by RiboGreen (Invitrogen, Carlsbad, CA, USA). RNA amplification by random priming and complementary DNA (cDNA) synthesis was performed as described previously (Frias-Lopez et al., 2008; Shi et al., 2009; Lesniewski et al., 2012). Shotgun sequencing of DNA and cDNA were performed with Illumina HiSeq2000 PE 100 with HiSeq Cluster Kit v4 (San Diego, CA, USA) at the University of Michigan Sequencing Core.

Metagenomic de novo assembly was performed as described previously (Li et al., 2015). In brief, short DNA reads were dereplicated (100% identity over 100% lengths) and trimmed using Sickle¹. The trimmed and dereplicated pair-end short reads were used for de novo assembly by IDBA-UD with the following parameters: --mink 50, --maxk 92, --step 4 or 6, --min_contig 500. Gene calling and annotations were done using the DOE Joint Genome Institutes (JGI) Integrated Microbial Genomes pipeline².

Assembled metagenomic sequences were assigned into putative taxonomic groups by binning with emergent self-organizing maps (ESOM) based on the analysis of sequence tetranucleotide frequencies (Dick et al., 2009a; cutoffs: minimum scaffold size = 5 Kb, maximum scaffold size = 10 Kb). To identify appropriate taxonomic levels of each genomic bins, we used following order of workflow: (i) identification of SSU rRNA genes on scaffolds by BLASTN (Altschul et al., 1990) against the Silva SSU Database version 115 (Quast et al., 2013); (ii) ESOM binning with reference genomes; (iii) taxonomic clustering of annotated genes by protein UBLAST with the IMG database (>50%) of identified open reading frames (Markowitz et al., 2008). Some genomic bins with multiple closely related genomes based on the presence of multiple copies of single-copy conserved genes were further separated by plotting differential coverage of scaffolds and their GC contents (Supplementary Table S2). The completeness and contamination of genomes within bins were estimated by counting single-copy conserved genes using CheckM with default setting, which is an automated method for assessing the quality of microbial genomes recovered from isolates, single cells, and metagenome (Parks et al., 2015; Supplementary Table S3).

For genes encoding peptidases and carbohydrate metabolizing enzymes, all annotated genes in bacterial genomes were searched against public databases of peptidases (MEROPS; Rawlings et al., 2014) and Carbohydrate-Active enZYmes (CAZYs; Lombard et al., 2014) with E-value < 10^-10 by BLASTP. To further improve the accuracy after this screening, all hits were also compared to the non-redundant NCBI protein database (version 092012). Only those that had top hits to peptidases and carbohydrate metabolizing enzymes [glycoside hydrolases (GHs); polysaccharide lyases (PLs); carbohydrate esterases (CEs); auxiliary activities (AAs)] were considered. The extracellular peptidases were further confirmed based on the identification of extracellular transport signals using SignaIP (Petersen et al., 2011), POSRTb (Yu et al., 2010) or PRED-SIGNAL (Bagos et al., 2009), and these tools are widely used to predict the presence and location of signal peptides and their cleavage sites in amino acid sequences from different organisms. Genes related to carbohydrate metabolizing enzymes were further classified into different groups according the prediction of CAZYs (Lombard et al., 2014).

Abundance of transcripts for each genomic bins or functional genes was determined by mapping all non-rRNA transcripts to the assembled fragments or functional genes using Burrows–Wheeler Aligner (BWA) with default settings and normalizing to their sequence length and total number of non-rRNA transcripts in each metatranscriptomic library (Li and Durbin, 2009).

Alignment of ribosomal protein S3, 16S rRNA genes were performed by MUSCLE using default parameters followed by manual refinement (Edgar, 2004). Phylogenetic analysis of representative ribosomal protein S3 were inferred by Maximum Likelihood implemented in Mega 6.0 (Tamura et al., 2013) using the Tamura-Nei and passion model after testing by ProTest (Darriba et al., 2011) and bootstrapped 1000 times. The phylogeny of 16S rRNA genes were inferred by maximum likelihood implemented in RaxML using the GAMMAGTR algorithm and bootstrap 1000 times in ARB software (Ludwig et al., 2004).

Scaffolds from metagenomic assembly are available from DOE JGI-IMG/MEG with Taxon Objects ID 3300001683 and NCBI with accession number LWDU00000000. The original short reads of metagenome and metatranscriptomes are available from NCBI with accession number SRP075597, SRX134768, and SRX134769, respectively.

The number of DNA and cDNA short reads from plume and background are listed in Supplementary Table S1. Metagenomic assembly and binning by tetranucleotide frequencies coupled with coverage and GC content resulted in a total of 74 bacterial genomes (Figure 1; Supplementary Table S2). Among them, eight genomes predicted by the CheckM package to have relative high contamination were removed from further analysis. Of the remaining 66 genomes, 53 and 42 of these bacterial genomes were estimated to be >60 and >80% complete, respectively (Supplementary Table S3). To determine the taxonomic identities of these bacterial genomes, 16S rRNA genes (>500 bp) identified within the genomes (38 genomes) were used for phylogenetic analysis (Figure 2). For remaining genomes (28 genomes), we used phylogeny of ribosomal protein S3 (Supplementary Figure S1) and the binning of reference genomes to confirm their taxonomic identities (Castelle et al., 2013; Hug et al., 2013). These results indicated that the 66 genomes belong to 10 different bacterial phyla, including Actinobacteria (six genomes), Bacteroidetes (seven genomes), Chloroflexi (three genomes), Deferribacteres (three genomes), Firmicutes (one genome), Gemmatimonadetes (one genome), Nitrospirae (two genomes), Planctomycetes (seven genomes), Proteobacteria (Alphaproteobacteria, six genomes; Betaproteobacteria, one genome; Deltaproteobacteria, five genomes; Gammaproteobacteria, 20 genomes), and Verrucomicrobia (four genomes; Supplementary Table S2).

To compare the genomic and transcriptomic abundance of these genomic bins, DNA and cDNA reads were mapped to the assembled sequences and results were normalized to the sequence length and total number of DNA and cDNA short reads in metagenomic and metatranscriptomic libraries, respectively. From the genomic evidence, the most abundant bacteria in the GB plume included members of SUP05 (Bin82-2, Bin82-1, and Bin94), Actinomycetales (Bin70 and Bin63), Gammaproteobacteria methanotrophs (Bin75, Bin74-2, Bin74-1, Bin66, and Bin65), Planctomycetes (Bin81, Bin102, Bin53, Bin39, Bin44, Bin62, and Bin64), Verrucomicrobia (Bin48, Bin84, Bin49, and Bin40), Acidimicrobium (Bin50), and unknown Gammaproteobacteria (Bin110-2; Figure 3). However, transcriptomic evidence indicated that members of SUP05, methanotrophs, unknown Gammaproteobacteria, Planctomycetes, SAR202 (Bin59), and SAR324 (Bin58) are the most transcriptionally active bacteria groups (>4 × 10⁶), while members of Verrucomicrobia and Acidimicrobium and have relative low transcripts (1.6~2.7 × 10⁶) in the GB plume and background (Figure 3). Interestingly, most of these bacteria had higher transcript abundances in the plume than in the background with few exceptions from two unknown Gammaproteobacteria (Bin78 and Bin79) and Alteromonadales (Bin95-1 and Bin95-2; Figure 3).

Of the 66 identified bacterial genomes, 59 genomes from 10 bacterial phyla contained multiple genes encoding extracellular peptidases, ranging from 0.3/Mb to 9.2/Mb. A member of Gemmatimonadetes (Bin68) had the most abundant extracellular peptidase genes (9.2/Mb), following by members of Erythobacter from Alphaproteobacteria (Bin42-2 and Bin42-1), Rubricoccus and Flavobacteriaceae from Bacteroidetes (Bin46, Bin67-1, and Bin67-2), and Nocardioides from Actinobacteria (Bin88; Figure 4; Supplementary Table S4). Transcripts for extracellular peptidases were also detected from 52 bacterial groups in the GB plume. Among them, members of Methylococcaceae (Bin74-2, Bin66, Bin74-1, and Bin65), Planctomycetes (Bin81, Bin102, Bin44, Bin64, and Bin39), Gemmatimonadetes (Bin68) and unknown Gammproteobacteria (Bin110-2) had the most abundant transcripts for extracellular peptidases in the GB plume, while members of Methylococcaceae (Bin74-2 and Bin74-1) and Gemmatimonadetes (Bin68) also have high extracellular peptidase transcript abundance in the background (Figure 4; Supplementary Table S4).

The detected bacterial extracellular peptidases belong to 58 different families (Figure 5A). Among them, families S01C (DegP serine endopeptidases), S41A (C-terminal processing peptidase), S08A (subtilases), M16B (metalloendopeptidases), and M23B (lysostaphin) were the most abundant in the metatranscriptomes, representing 58.8 and 50.2% of all bacterial extracellular peptidase transcripts in the GB plume and background, respectively (Figure 5A). These five families of extracellular peptidases are related to serine endopeptidases (S01C, S41A, and S08A) and metalloendopeptidases (M16B and M23B), which cleave peptides at different sites of proteins, suggesting comprehensive protein degradation by deep-sea bacteria and potential use of protein as a growth substrate. Most extracellular peptidases had higher transcript abundance in the plume than in background, with only a few exceptions such as the families M16C (eupitrilysin), S49X (periplasmic serine protease), T02 (isoaspartyl aminopeptidases) and M03A (peptidyl-dipeptidase; Figure 5A).

In addition to extracellular peptidases, all bacterial genomes present in here were also searched for carbohydrate-metabolizing enzymes, including GHs, PLs, CEs, and AAs. A total of 65 bacterial genomes contain at least one copy of genes related to carbohydrate-metabolizing enzymes, suggesting the carbohydrate metabolisms are widespread in deep-sea bacteria identified in the GB hydrothermal plume (Figure 4; Supplementary Table S4). Members from Planctomycetes (Bin62, Bin102, Bin39, Bin64, and Bin44), Verrucomicrobia (Bin84, Bin49, Bin40, and Bin48), Bacteroidetes (Bin67-2, Bin89, Bin46, Bin108-2, and Bin67-1), Deferribacteres (Bin57) and Gemmatimonadetes (Bin68) have the most abundant genes for GHs (>6/Mb), while members of Alphaproteobacteria (Bin37), Actinobacteria (Bin51), Planctomycetes (Bin39, Bin62, Bin64), and Chloroflexi (Bin59) have high CEs coding gene abundance (>4/Mb). For PLs and AAs, all bacterial genomes identified here had relatively low gene abundance (Figure 4). Transcripts for GHs and CEs were detected from 56 and 44 bacteria groups (Figures 5B,C), while only 5 and 31 bacteria groups have transcripts related to PLs and AAs in the GB hydrothermal plume (Supplementary Figure S2). Among these bacteria, members of Methylococcaceae (Bin65), Planctomycetes (Bin102, Bin81), SUP05 (Bin82-2), Deferribacteres (Bin57) and Verrucomicrobia (Bin48) have the most abundant transcripts for GHs, while members of Methylococcaceae (Bin74-2), SAR202 (Bin59), and Planctomycetes (Bin102 and Bin44) have high CE transcripts. Furthermore, GHs and CEs have much higher transcript abundance than those related to PLs and AAs (Figure 4; Supplementary Table S4).

Further analysis demonstrates that families of GH0, GH43, GH77, GH3, GH23, GH117, and GH103 for GH are the most abundant in the metatranscriptomes, representing 71.8 and 70.2% of all bacterial GH transcripts in the GB plume and background (Figure 5B). These dominant families are related to aldose 1-epimerases, arylsulfatases, sulfatases, beta-lactamases, beta-N-acetylglucosaminidases, lytic transglycosylases, N-acetyl-beta-hexosaminidases, and many uncharacterized GHs. While for CEs, families of CE11 and CE0 are the most abundant in the metatranscriptomes, responding to 63.0 and 60.5% of all bacterial CE transcripts in the GB plume and in the background (Figure 5C). Finally, similar with the extracellular peptidases, all detected families of GHs and CEs have much higher transcripts in the plume than in the background (Figure 5).

To better understand the most active heterotrophic bacteria, the extracellular peptidases and carbohydrate-metabolizing enzymes in Planctomycetes, Gemmatimonadetes, and Verrucomicrobia were characterized further. Members of Planctomycetes are the most abundant and transcriptionally active heterotrophic bacteria in the GB hydrothermal plume (Supplementary Figure S3). Seven near-complete genomes of Planctomycetes with distinct genomic and transcriptomic abundance were identified (Figure 6). These seven genomes contained a total of 122 extracellular peptidase genes from 33 different families. Among them, transcripts for 27 families were detected, representing 27.9% of all bacterial extracellular peptidase transcripts in the GB plume (Figure 6A). For carbohydrate metabolizing enzymes, a total of 369 genes for 29 different families of GHs and 134 of CE genes belonging to 10 families were also identified in the seven Planctomycete genomes, representing 31.9 and 26.6% of all bacterial transcripts for GHs and CEs in the GB plume, respectively (Figure 6B). Similarly, transcripts involved in PLs and AAs identified in members of Planctomycetes also accounted for 35.4 and 69.3% of all related bacterial transcripts in the GB plume. These results indicate that members of Planctomycetes are involved in scavenging various organic matter in the GB plume. Interestingly, these Planctomycetes bacteria also have many genes and transcripts involved in the hydrolysis of sulfate esters, such as arylsulfatase, choline sulfatases and sulfatases, indicating Planctomycetes also respond for the utilization of organic sulfur compounds in the deep-sea (Supplementary Figure S4).

Only one genome related to Gemmatimonadetes was identified in the GB hydrothermal plume. This Gemmatimonadetes genome was estimated to be 85.8% complete and contained high abundance of genes for extracellular peptidases (the most of all bacterial genomes recovered) and carbohydrate metabolizing enzymes (Figure 4). Furthermore, this deep-sea Gemmatimonadetes bacterium also had high transcript abundance for carbohydrate metabolizing enzymes and extracellular peptidases, accounting for 1.5 and 5.5% of all bacterial carbohydrate metabolizing enzyme and extracellular peptidase transcripts, respectively (Figure 6A). For members of Verrucomicrobia, four genomes were identified in the GB hydrothermal plume, ranging in completeness from 44.4 to 99.8%. In these Verrucomicrobia, only eight extracellular peptidase genes were identified, but 59 genes encoding carbohydrate metabolizing enzymes were found, the most of all bacteria identified in this study. In addition, Verrucomicrobia bacteria had high abundance of transcripts for carbohydrate metabolizing enzymes, but relative lower transcripts for extracellular peptidases (Figures 4 and 6).