Description of the sampling sites, sample processing and storage, and physico-chemical characteristics of PHF have been reported previously (Monnin et al., 2014; Quéméneur et al., 2014; Postec et al., 2015). Briefly, four active carbonate chimneys were sampled at four distinct sites of the Bay of Prony, including the “Bain des Japonais” (BdJ), the sites (ST) ST07 (also called “Aiguille de Prony”), ST09 and ST12. The BdJ site is located on the foreshore of the Carenage Bay and is uncovered at low tide. It consists of many fluid outlets building a carbonate plateau. There, small needle-like brittle structures of 10–20 cm high and 3–6 cm in diameter were collected. The submarine sites ST07, ST12, and ST09, located in the Prony Bay, were sampled at 16 meters below sea level (mbsl), 38 mbsl, and 48 mbsl, respectively. They form large carbonate edifices, carrying several chimneys of up to several meters high. The chimneys are porous structures, mainly composed of brucite [Mg(OH)₂], aragonite and calcite (CaCO₃) and magnesium carbonates (MgCO₃) (Pisapia et al., 2017). The external walls of the mature submarine chimneys are colonized by various invertebrates (e.g., ascidia, sponges, corals). The top of such chimneys was collected by scuba-diving in each submarine site. They were kept at 4–8°C during their transportation (∼3 h) to the laboratory, where they were immediately processed. Cross sections (4 cm thick) of these chimneys top were realized aseptically; peripheral parts of the sections (2 cm thick) were cautiously removed using a chisel and sub-samples of the inside parts of the chimneys, bathed with the alkaline fluid, were collected and stored at –80°C until DNA extraction, as detailed previously (Quéméneur et al., 2014; Postec et al., 2015).

DNA extraction and quantification procedures have been described previously (Postec et al., 2015). Briefly, about 0.4 g of chimney slurry was transferred into a sterile 2 mL tube containing glass beads (lysing matrix E from MP BioMedicals) and 0.8 mL of lysis buffer (Tris.HCl 100 mM, NaCl 100 mM, EDTA 50 mM, pH 8). Cell lysis was achieved with chemical reactions (adding Sodium Dodecyl Sulfate and Lauryl Sarcosine) and mechanic disruption using a Fast Prep homogenizer. DNA was extracted using phenol:chloroform:isoamyl alcohol and chloroform:isoamyl alcohol. DNA from each sample or dilution was used as template for 16S rRNA amplification. The hypervariable region V1–V3 of the 16S rRNA gene was amplified using the pair of primers 27F (5′-AGAGTTTGATCCTGGCTCAG-3′)/533R (5′-TTACCGCGGCTGCTGGCAC-3′) and 21F (5′-TCYGKTTGATCCYGSCRGA-3′)/529R (5′-TCGCGCCTGCTGCRCCCCGT-3′) for bacteria and archaea, respectively. The forward and reverse primers were barcoded (Supplementary Table S1). PCR reactions were performed in a 50 μl mixture containing Platinium^TM Taq High Fidelity (Life Technologies) 1 unit, High Fidelity PCR Buffer 1X, dNTP mixture 0.2 mM each, MgSO4 2 mM, primer mix 0.3 μM each, and 2 μl template DNA. The protocol mentioned below was used for amplification: initial denaturation at 95°C for 2 min, followed by 25 cycles of denaturation 95°C for 45 s, primers annealing at 54°C for 45 s, elongation at 72°C for 1 min 30 s, and a final elongation step at 72°C for 7 min. At least three PCR products from each combination of sample and primer pairs were pooled and then purified with QIAquick Gel Extraction Kit (Qiagen). An equimolar amount of each PCR pool was finally used for library construction (addition of the adaptors by ligation) and pyrosequencing using 454 GS FLX Titanium technology (Beckman Coulter Genomics).

Pyrosequencing of the V1–V3 region of the microbial 16S rRNA generated 188,058 sequences. Raw nucleotide sequence data were submitted to the European Nucleotide Archive under project accession number PRJEB21795. These sequences were filtered according the following criteria: read length between 150 and 500 bp, no ambiguous bases, average quality score ≥25, and no error allowed in primer and barcode sequences. A total of 74,457 and 95,916 high quality sequences were obtained for the bacteria and archaea, respectively. Alignments and clustering to Operational Taxonomic Units (OTUs) were carried out using the QIIME 1.9.1 workflow (http://www.qiime.org/) (Caporaso et al., 2010b). Chimera sequences were identified and removed using UCHIME (Edgar et al., 2011). Multiple sequence alignments were performed using PyNAST (Caporaso et al., 2010a). OTUs were defined by clustering sequences using UCLUST (Edgar, 2010) with a pairwise distance threshold value of 3%, a cutoff currently used for species demarcation while reducing the potential inflation of the number of OTUs due to pyrosequencing errors. To limit the impact of spurious OTUs, low abundance OTUs accounting for less than 0.005% of total sequences were removed as recommended by Bokulich et al. (2013). Taxonomic assignment of the filtered sequences was performed using the RDP classifier algorithm (Wang et al., 2007) with a minimum bootstrap confidence of 80% against the SILVA database (Quast et al., 2013) release for QIIME (v.123). The OTU table was rarefied (i.e., downsampled to the sample with the smallest set of sequences) to reduce sequencing depth heterogeneity between samples. There is currently no standardized threshold for the rare biosphere. Here, rare OTUs were defined as OTUs comprising 0.005 to 0.2% of sequences per sample, an empirical threshold previously adopted by Hugoni et al. (2013). OTUs comprising 0.2 to 1% of sequences per sample were considered as intermediate and OTUs representing more than 1% of sequences per sample were considered as abundant. The alpha diversity within the four samples was estimated using the Shannon (Shannon and Weaver, 1949) and Simpson (Simpson, 1949) indices, and the beta diversity among the samples was measured using weighted UniFrac (Lozupone and Knight, 2005) metric. Taxonomic composition, and diversity results were visualized using the R package phyloseq v.1.14.0 (McMurdie and Holmes, 2013). BLASTn (Basic Local Alignment Search Tool) searches against the NCBI (National Center for Biotechnology Information) database of nucleotides (nt) and against the RefSeq database were performed for some OTU sequences. To infer the phylogeny of Methanosarcinales, cultivated members of this order and environmental clones were added to the sequences of abundant or ubiquitous PHF OTUs. The V1–V3 regions of these sequences were aligned with MAFFT v7.123b (Katoh et al., 2002) and a maximum likelihood tree was inferred using the software SeaView (Gouy et al., 2010) with 1000 bootstrapped trials.

The correctly formatted amplicons and clone libraries of five serpentinizing ecosystems were downloaded from public databases (SRA and Genbank). These data constituted a subset of known serpentinizing systems, including four continental environments: Cabeço de Vide Aquifer (Tiago and Veríssimo, 2013), The Cedars (Suzuki et al., 2013), Voltri ophiolite (Quéméneur et al., 2015), Leka ophiolite (Daae et al., 2013), and one submarine system: the famous Lost-City Hydrothermal Field (Schrenk et al., 2004; Brazelton et al., 2006). A closed-reference OTU picking process was carried out to compare the sequences of amplicons and clone libraries covering different regions of the 16S rRNA gene. The sequences were clustered against a reference sequence database (SILVA database, v123). The OTUs shared among samples were visualized with Anvi’o v2.3.2 (Eren et al., 2015).

The bacterial species richness of the sampling sites comprised 1,979 distinct OTUs (Supplementary Figure S1A). However, the species richness distribution varied significantly from one site to another, with at least a twofold decrease in OTU abundance for BdJ, ST07, and ST09 compared to the ST12 site (1203 OTUs, Supplementary Table S2). The presence of more than 780 OTUs, specific to the ST12 site (Supplementary Figure S1A) and comprising mainly of rare taxa, might result from a lower in situ pH condition (around 9.2), in comparison with the other studied PHF sites (pH > 9.8) (Monnin et al., 2014). This trend has previously been observed in the Coast Range Ophiolite serpentinizing site, where circumneutral pH wells contained many rare taxa that were absent in the high pH ones (Twing et al., 2017), most likely because these latter drastic conditions selected the few most alkaliphilic members within the prokaryotic community. The overall species richness was considerably lower in the archaea (310 OTUs) than in the bacteria, and also showed an imbalance in OTU distribution across the four sites (Supplementary Figure S1B and Supplementary Table S2). The Shannon and Simpson indexes indicated lower archaeal diversity, as compared to the bacterial one (Supplementary Figure S2 and Supplementary Table S2). This lower richness and diversity of archaea is not specific to PHF as it has been generally observed in various environments globally (Aller and Kemp, 2008), including deep-sea hydrothermal vents (Anderson et al., 2015) and Lost-City (Brazelton et al., 2010). The rarefaction curves had a near-asymptotic shape (Supplementary Figure S3) suggesting that the sequencing effort was suitable to capture the near-full diversity of this serpentinizing environment.

The bacterial community composition of the PHF sites encompassed 22 different phyla (including six candidate phyla) reflecting a high biodiversity. The phylum Proteobacteria was preponderant in the intertidal BdJ (78.4% of sequences, 244 OTUs), while the three submarine sites were mainly composed of Chloroflexi (50.7–54.8%, 150–161 OTUs) and, to a lesser extent, of Proteobacteria members (21.1–27.4%, 250–603 OTUs) (Figure 1). Remarkably, all other bacterial phyla shared by all the PHF sites were essentially recovered at much lower abundance. These bacterial phyla comprised Cyanobacteria (0.2–16.0%), Actinobacteria (0.2–4.8%), Bacteroidetes (0.2–3.1%), Firmicutes (0.1–5.3%), Planctomycetes (0.2–1.6%), and the candidate phylum Parcubacteria (formerly candidate division OD1) (0.01–0.6%). By contrast, the Acetothermia, Acidobacteria, Gemmatimonadetes, and Nitrospirae phyla were only retrieved from submarine sites, representing 7.4% of the total sequences. Relative abundances of some bacterial phyla detected in our study were not in agreement with those reported by sequencing clone libraries (Quéméneur et al., 2014; Postec et al., 2015). This was the case for the Firmicutes and the Parcubacteria, which were found in much lower proportion in this study. Nevertheless, significant shifts in phyla proportions have already been observed in the same chimney edifice over a 6 year survey (Postec et al., 2015). Our pyrosequencing approach also allowed the identification of 6 novel phyla that had not been previously detected. Members of all these phyla, i.e., Fibrobacteres, Lentisphaerae, Marinimicrobia, PAUC34f, Saccharibacteria, and Verrucomicrobia, were found among the large diversity of the ST12 site. On the contrary, several phyla previously found in PHF: Atribacteria, Microgeomates, Omnitrophica, and candidate division NPL-UPA2 (Quéméneur et al., 2014; Postec et al., 2015; Mei et al., 2016; Pisapia et al., 2017) were not represented in this study. This discrepancy may be explained by the use of different PCR primers targeting different regions of the 16S rRNA gene of bacteria and archaea. It may also be the consequence of the ongoing evolution of the new generation sequencing technology (i.e., 454 Roche vs. Illumina). That raises a major issue regarding the lack of standards for 16S rRNA gene surveys, which reduces the reproducibility and complicates comparisons between studies (Escobar-Zepeda et al., 2015). Additionally, the sampling area of the chimneys (either central or more peripheral) could also explain these taxonomic discrepancies. Indeed, the highly porous structure of the chimney interiors is exposed to steep physicochemical gradients (in terms of salinity, pH, redox potential, nutrients) along their transverse axis, defining microhabitats colonized by contrasted microbial communities.

The archaeal community of the PHF sites mainly comprised potential methanogens and autotrophic ammonia-oxidizer representatives among Euryarchaeota and Thaumarchaeota phyla, respectively. The BdJ and ST12 sites were dominated by members of Methanosarcinales (99.8% of sequences, 38 OTUs and 94.6%, 44 OTUs, respectively) within the class Methanomicrobia (Figure 2). All abundant and ubiquitous OTUs of Methanosarcinales were similar to the sequences from other serpentinite-hosted ecosystems, as shown by OTU-based phylogeny (Figure 3). The first set of OTUs constituted a sister group of Lost-City Methanosarcinales (LCMS) (Schrenk et al., 2004) whereas the remaining OTUs were closely associated with the sequence of The Cedars Methanosarcinales (TCMS) (phylotype Ced_A01) (Suzuki et al., 2013). Although the present study reported a more accurate diversity than the previous studies based on clone libraries, no new phylotypes were identified within the Methanosarcinales group. Interestingly, although first detected in a terrestrial serpentinizing system (The Cedars), TCMS-like archaea were detected at lower abundance in the intertidal BdJ site than in the submarine sites of PHF. Moreover, here the overall abundance of the LCMS-like was two-fold higher than the abundance of the TCMS-like (Figure 3). This result is in opposition with the ratio previously observed in clone libraries and thus contradicts the hypothesis that the predominance of TCMS-like archaea in PHF was due to the low salinity of fluids (of meteoric origin such as in The Cedars) (Postec et al., 2015). Finally, these two specific phylotypes of Methanosarcinales were ubiquitous in all PHF sites. In contrast, the ST07 and ST09 sites were dominated by the Thaumarchaeota Marine Group I (69.4%, 179 OTUs and 62.4%, 194 OTUs, respectively), which was detected in all submarine sites despite its low abundance in the ST12 site (Figure 2). The Marine Group I is frequently observed in the global ocean (Auguet et al., 2010; Fernàndez-Guerra and Casamayor, 2012), which may explain why they are found only in submarine sites of PHF. In addition, the Thaumarchaeota Soil Crenarchaeotic Group was also detected in the submarine sites at very low-abundance level (0.1 to 1.9%, 3 to 6 OTUs) and may originate from the surrounding hills having been carried away by run-off waters. Other less represented archaeal classes belonged to the Thermoplasmata (mainly Marine Group II) (0.1–0.7%, 7–18 OTUs), Methanobacteria (0.1%, 2 OTUs at BdJ), Terrestrial Group of Thaumarchaeota (0.1%, 2 OTUs in ST07) and presumed thermophiles of the Thermococci class (0.1%, 1 OTU at ST12) (Figure 2). Finally, the remaining OTUs were affiliated to the Deep Sea Euryarchaeotic Group within the Aenigmarchaeota phylum and to the Miscellaneous Crenarchaeotic Group that were not previously detected in PHF. The archaea of the Miscellaneous Crenarchaeotic Group are widespread in marine sediment and are essentially retrieved from anoxic environments such as deep oceanic subsurface sediments or mud volcanoes (Kubo et al., 2012). Since the temperature of PHF fluids measured at the chimney outlets did not exceed 43°C, the presence of archaea closely related to hyperthermophilic species (e.g., Thermococcus spp.) originating from deep-sea environments was surprising. Despite their low abundance, such representatives of the Thermococcales were, however, previously detected in PHF, even in clone libraries, and the high GC content of their sequence strongly suggested they were bona fide hyperthermophiles (Postec et al., 2015). It is thus likely that they originated from subsurface hyperthermophilic microbial communities fed by serpentinization reactions and carried up to the surface by the fluid as previously suggested (Postec et al., 2015).

The bacteria taxonomic profiles observed for the three submarine sites were very similar, while the one observed for the intertidal site (BdJ) differed significantly (Figure 4A). This pattern of distribution, previously observed in a study on PHF (Quéméneur et al., 2014), may be explained by the influence of the marine environment regime (intertidal vs. submarine) variation in the composition of the end-member fluids (Monnin et al., 2014) or the resulting differences observed in the mineralogy of the carbonate chimneys (Pisapia et al., 2017). The relationship between species and sampling sites was mainly driven by a few dominant OTUs in the ordination plot (Figure 4A). Among them, five abundant OTUs in submarine sites were assigned to the class Dehalococcoidia (DEH, phylum Chloroflexi) and were related to clones retrieved from the serpentinizing springs of The Cedars (Suzuki et al., 2013). Two of these dominant OTUs (#915 and #1117) were assigned to the candidate order MSBL5 (Mediterranean Sea Brine Lake group 5) and three (#1432, #1706, and #1947) to the division GIF9 (Groundwater InFlow clone 9) (Supplementary Table S3). The candidate orders MSBL5 and GIF9 were originally detected in a deep-sea anoxic hypersaline lake (Daffonchio et al., 2006) and in a contaminated groundwater-fed bio reactor (Alfreider et al., 2002), respectively. These five dominant OTUs are quite divergent from any cultivated species (the closest similarity is found for OTU1947 with Dehalogenimonas alkenigignens, at 82% identity). The metabolic capabilities of these candidate groups are still largely unknown. Even for the entire class Dehalococcoidia, our knowledge of metabolism is essentially limited to that of the few cultivated representatives, which have the ability to oxidize hydrogen and reduce organohalogenic compounds (Kaster et al., 2014; Biderre-Petit et al., 2016). However, a single cell sequencing of a Dehalococcoidia phylotype, abundant in marine sediments, suggested that this bacterium may replace organohalides by dimethyl sulfoxide as a terminal electron acceptor (Wasmund et al., 2013). In the BdJ site, 4 OTUs (#1298, #153, #1806, #279) representing more than a third of all sequences (Figure 4A) were assigned to the haloalkaliphilic genus Roseibaca (Alphaproteobacteria), a sister genus of the phototrophic anaerobic sulfur-oxidizing Rhodobaca, previously detected as dominant in BdJ (Mei et al., 2016). Representatives of the Roseibaca genus are known to be rose-colored (Labrenz et al., 2009) and could in part explain the dark rose color of the biofilm covering the exteriors of BdJ chimneys. The BdJ site also contained two dominant OTUs (#607 and #763) affiliated to the Comamonadaceae family (Betaproteobacteria) showing 98% identity with Hydrogenophaga defluvii (Supplementary Table S3). Members of the genus Hydrogenophaga are chemoorganotrophic or chemolithoautotrophic, thriving by oxidation of hydrogen under microaerophilic conditions (Willems, 2014). Despite the potential involvement of Firmicutes in the organomineralization of carbonates chimneys in PHF (Pisapia et al., 2017), only one abundant OTU of this phylum was identified in submarine sites and assigned to uncultured anaerobic Syntrophomonadaceae. This OTU, however, differed from the one (KM207235 HPst091-1-1) reported to be dominant in juvenile chimneys of PHF (Pisapia et al., 2017), most likely because of the differences of age of the chimneys studied (i.e. mature vs. juvenile). Members of this family are anaerobes that usually grow in syntrophic association with hydrogenotrophic and formate-utilizing methanogens by thriving energy from oxidation of carboxylic acids (Sobieraj and Boone, 2006). These anaerobic oxidations are thermodynamically feasible only when the H₂ or formate produced is concomitantly consumed by methanogens or other syntrophic partner. Recently, syntrophic associations were shown to permit the conversion of organic acids and alcohols to methane under extremely haloalkaline conditions (Sorokin et al., 2016). In PHF, such syntrophic associations between uncultivated members of the Syntrophomonadaceae and the Methanosarcinales could be a relevant strategy to overcome energetic limitations, due to the high pH conditions and the lack of electron acceptors.

For archaea, no differential segregation of OTUs between intertidal and submarine sites could be observed, in contrast to the trend previously reported (Quéméneur et al., 2014). Nevertheless, the beta-diversity analysis linked BdJ to ST12 (Figure 4B), with both sites harboring abundant OTUs related to the LCMS (their annotation was refined through phylogenetic analysis, Figure 3), whereas ST09 clustered with ST07 because they shared abundant OTUs related to Candidatus Nitrosopumilus (Thaumarchaeota) and numerous OTUs of Marine Group I (Figure 4B and Supplementary Table S4). In light of these beta-diversity results, the archaeal community seemed less site-dependent than the bacterial one, as exemplified by the two groups of Methanosarcinales OTUs (LCMS and TCMS) found in both intertidal and submarine serpentinizing systems of PHF sites.

A comparison of microbial communities across six geographically distant serpentinizing systems was carried out at the OTU level (at 97% identity threshold) used as a proxy of the prokaryotic species that is the most commonly found unit in microbial biogeographic studies. For practical reasons, this comparison was limited to the fraction of known microorganisms (i.e., those which could be assigned to a taxon in the Silva database), thus the results cannot be extrapolated to the whole prokaryotic community. Despite this limitation, some biogeographic patterns could be observed. The general OTU composition appeared to be quite different from one serpentinizing system to another and no OTU was common to all sites, but several related phylotypes were recovered from more than two systems; among them, a few were detected exclusively in serpentinizing systems. The distribution pattern of PHF phylotypes in other geographic sites and its implications is discussed below.