Sulfur intermediates as new biogeochemical hubs in an aquatic model microbial ecosystem

Background The sulfur cycle encompasses a series of complex aerobic and anaerobic transformations of S-containing molecules, and plays a fundamental role in cellular and ecosystems level-processes, inuencing biological carbon transfers and other biogeochemical cycles. Despite their importance, the microbial communities and metabolic pathways involved in these transformations remain poorly understood, notably for inorganic sulfur compounds of intermediate oxidation states (thiosulfate, tetrathionate, sulte, polysuldes). Isolated and highly stratied, the extreme geochemical and environmental contexts of the meromictic ice-capped Lake A, in the Canadian High Arctic, provides an outstanding model ecosystem to resolve the distribution and metabolism of aquatic sulfur cycling microorganisms along redox and salinity gradients.

processes in the oxic freshwater layers, reductive reactions in the anoxic and sul dic bottom waters and genes for both transformations at the chemocline, and co-varied with bacterial abundance. Up to 154 different genomic bins with potential for sulfur transformation were recovered, revealing a panoply of taxonomically diverse microorganisms with complex metabolic pathways for biogeochemical sulfur reactions. Metabolism of sulfur cycle intermediates was widespread throughout the water column, cooccurring with sulfate reduction or sul de oxidation pathways. The genomic bin composition suggested that in addition to chemical oxidation, these intermediate sulfur compounds were likely produced by the predominant sulfur chemo-and photo-oxidizers at the chemocline and by diverse microbial organic sulfur molecule degraders.

Conclusions
The Lake A microbial ecosystem provided an ideal opportunity to identify new features of the biogeochemical sulfur cycle. Our detailed metagenomic analyses across the broad physico-chemical gradients of this highly strati ed lake extend the known diversity of microorganisms and metabolic pathways involved in sulfur transformations over a wide range of environmental conditions. The results identify the importance of sulfur cycle intermediates and organic sulfur molecules as major sources of electron donors and acceptors for aquatic and sedimentary microbial communities in association with the classical sulfur cycle.
Background Page 3/26 The sulfur cycle encompasses complex energetic processes where sulfur (S) ions and molecules in different oxidation states, from the most reduced (-2: sul des, H 2 S/HS − ) to the most oxidized (+ 6: sulfate, SO 4 2− ), are transformed through oxidation, reduction, disproportionation [1] and eventually comporportionation [2] by taxonomically diverse microorganisms. Tightly interwoven with carbon, nitrogen and metal cycles, the sulfur cycle is tied to both cellular and ecosystem level-processes [3]. In marine sediments, sulfate is an ubiquitous electron acceptor and sulfate-reducing microorganisms have been estimated to contribute to up to 29% of the organic matter remineralisation [4]. Sulfate reducers can generate massive concentrations of sul des, which in turn serve as an electron donor for symbiotic or free-living sulfur-oxidizing microorganisms that recycle sul des into sulfate [5].
Although sulfate reduction and sul de oxidation have received most attention, our knowledge of the identity of microorganisms and metabolic pathways involved in these processes is limited and novel lineages of microorganisms mediating steps in the sulfur cycle remain to be discovered [6,7].
Furthermore, sulfur cycling is highly complex; sulfur transformations by microorganisms and the geochemical reactivity of reduced sulfur molecules with metal oxides generate several inorganic sulfur compounds of intermediate oxidation states (e.g., thiosulfate, S 2 O 3 2− ; tetrathionate, S 4 O 6 2− ; sul te, HSO 3 − ; polysul des: S 2 − n+1 ). These inorganic S compounds are all substrates for further microbial oxidation, reduction or disproportionation [3,8] and precipitation with reduced metals [9]. The rapid recycling of these sulfur species has biogeochemical signi cance, especially in low sulfate environments, where the rapid turn-over of these compounds provides an opportunity for shortcuts in the sulfur cycling and potentially sustains a large variety of microorganisms [10]. However, because e cient microbial scavenging drives their concentrations down to the picomolar range, the importance of these processes remains unrecognized. An alternative strategy to reveal these processes is to look for genomic evidences of the identity, ecology and functional properties of microbes metabolizing these compounds.
The genetic underpinning of sulfur transformations is still poorly resolved, and is complicated by the bidirectional activities of key enzymes, and by the diversity and complexity of many enzymatic pathways.
For example, the metabolic pathways involved in sulfur compounds disproportionation remains unknown and even the exact mechanisms involved in sul de oxidation remain little understood [1,11]. Another important, yet largely overlooked component of the sulfur cycle involves the utilisation and formation of organo-sulfur molecules (OSM). These labile metabolites including sulfonate (compounds with a R-SO 3 − functional group) and sulfonium such as dimethylsulphoniopropionate (DMSP) are produced by macroand micro-algae and may represent an important source of sulfur for certain microorganisms in freshwaters [12,13] and oceans, where genetic potential for transformation of OSM is widespread in marine bacteria [14].
Due to the abundance of sulfur compounds in sea water, much of our knowledge on sulfur cycle comes from marine sediments, which constitute a major biotope for sulfur cycling microorganisms [3]. However, the distribution of sulfur cycling microorganisms is not limited to marine environments, and sulfur cycling microorganisms also have important ecological roles in sulfate-poor environments such as wetlands and lakes [7,15].
Marine-derived lakes are natural laboratories for understanding the sulfur cycle. In the polar regions, melting ice sheets lead to an isostatic rebound of the continents, isolating fjords that become marinederived lakes. Melting snow and glaciers then discharge freshwater into these lakes that oats on the denser seawater. This density gradient is re-enforced by a mostly permanent ice cover, resulting in meromictic lakes that are layered physically and chemically with well-de ned environments, enabling the development of complex and stable microbial communities along light, nutrient and redox gradients [16]. The water column of these lakes is divided into a mixolimnion consisting of the oxic surface freshwater layer immediately beneath the ice, and an anoxic and sul dic monimolimnion, derived from the marine bottom layer. The chemocline at the interface between these layers is a zone of highest chemical reactivity in the lake, typically associated with elevated microbial activity and intense sulfur cycling [17,18]. Phototrophic or chemotrophic sulfur oxidizers that recycle sul des produced in the anoxic monimolimnion overlap or coexist in the chemocline, depending on oxygen and light penetration [19]. The activity of these microorganisms results in the production of sulfate and sulfur deposits outside the cells, leading to yellow or orange coloration of the water [20]. In addition, metal oxides (iron and manganese) frequently accumulate in meromictic lakes [21], leading to chemical oxidation of sulfur species.
The broad range of physico-chemical conditions in a single water column makes polar meromictic ecosystems ideal lake-size laboratories to investigate sulfur-cycling microorganisms and metabolic pathways. We hypothesized that sulfur cycle intermediate (SCI) molecules such as thiosulfate, tetrathionate, sul te and polysul des are a major source of energy for aquatic microbial communities, and that related biogeochemical processes would be evident in a polar meromictic lake because of the prolonged selection of microbial species and functions across stable biogeochemical gradients.
To test this hypothesis, we analysed the microbial community composition, abundance and metabolic potential in Lake A (83°00′N, 75°30′W), a deep marine-derived meromictic lake in the Canadian High Arctic (Fig. 1a), with a focus on sulfur cycling microorganisms across light and redox gradients. Applying genome-centric metagenomics, we recovered a large variety of sulfur cycling microorganisms with contrasted and complex metabolic pathways involved in sulfur transformations. We elucidated the importance of sulfur cycle intermediates as well as organic sulfur molecules in structuring the microbial community composition and function over the broad range of geochemical conditions that are represented in this ecosystem.

Results
Oxygen, salinity and temperature pro les of Lake A, measured at the sampling time (18 July 2017), showed that the water column under the ice was highly strati ed, with similar pro les to previous years [21]. The freshwater oxygenated mixolimnion extended to the chemocline, located from 12 to 24 m below the upper ice surface ice, and marine-derived saline, anoxic water occurred below 24 meters (Fig. 1b).

Microbial Community Composition
Microbial community composition and abundance was investigated by 16S rRNA (rRNA) and 16S rRNA gene (rDNA) sequencing and quanti cation (qPCR) down the water column with discreet samples from 2 m, 6 m, 14 m, 22 m, 34 m, 40 m, 55 m, 65 m. In addition, 16S rDNA sequences were extracted from the metagenome dataset and analyzed along with the amplicon data (Fig. 1d). Bacteria represented nearly 100% of the 16S rRNA genes in both qPCR and metagenomic data in the upper freshwater samples and 90% in the saline water samples (Supplementary Figure S1), therefore only the bacterial community is detailed in this study. The bacterial community compositions of replicate samples were very similar ( Fig. 1c and Supplementary Figure S1). Although the number of rRNA genes may uctuate depending on species and physiological state, taxonomic pro les from 16S rRNA and 16S rDNA were congruent, suggesting that these communities were likely active (Supplementary Figure S2). The results were also congruent with the microbial community composition recovered from metagenomic data with the exception of SAR11 and Patescibacteria lineages, which appeared to be underestimated by amplicon sequencing ( Fig. 1c and Supplementary Figure S2).

Depth Distribution Of Sulfur Cycling Genes
Shotgun metagenomes from the eight depths were sequenced to evaluate the metabolic potential of Lake A microbial communities. A total of 10378 different genes (KEGG Orthologues) were identi ed in the metagenomes. Hierarchical clustering of the overall metabolic potentials of the samples was congruent with the clustering based on taxonomic pro les and geochemical data, with two major clusters: the freshwater samples (2, 6 and 14 m) and the saline anoxic samples (34, 40, 55 and 65 m); and the oxycline sample (22 m) branching separately on the dendrogram (Fig. 2). A large number of genes involved in sulfur cycling were identi ed in all samples, and were distributed vertically according to the physicochemical gradients (Fig. 2).

Sulfur-oxidizing populations
The potential for thiosulfate oxidation was widespread in the community (Figs. 3 and 4). Numerous Parcubacteria, Actinobacteria and Chloro exi bins included genes for the thiosulfate:quinone dehydrogenase (doxD), while the gene for the thiosulfate dehydrogenase (tsdA) was mainly detected in Alphaproteobacteria and Bacteroidetes. Genes of persul de dioxygenase fused with a rhodanese domain (pdo_rho), potentially involved in thiosulfate oxidation [22] were widespread in the Bacteroidetes and Planctomycetes bins (Fig. 3). The SoxABCDXYZ complex for thiosulfate oxidation was also identi ed in Alphaand Beta-proteobacteria (Fig. 3). This complex was coupled with oxidative DsrAB genes in Alphaproteobacteria, indicating potential for H 2 S oxidation to sulfate. Oxidative DsrAB genes and genes for sul de (fccAB) and sulfur oxidation through the sul de:quinone oxidoreductase (SQR) were identi ed in the Chlorobiaceae as well as in few Alphaproteobacteria bins. Betaproteobacteria bins harbored genes for sul de oxidation (soeAB and sorB). SQR gene was also detected in Cyanobacteria, Actinobacteria, Bacteroidetes, Gammaand Deltaproteobacteria. Sulfur and thiosulfate oxidation through Hdr-like complex [23] was identi ed in one alphaproteobacterial bin.

Sulfur-reducing populations
Potential for thiosulfate reduction through thiosulfate reductase (PhsC gene) was identi ed in Bacteroidetes bins recovered from anoxic samples (Figs. 3 and 4). Tetrathionate reductase genes (ttrBC) were mainly found in Rhodospirillales bins while genes coding the octoheme tetrathionate reductase (otr) were detected in few Desulfobacteraceae, Anaerolineae, Chlorobiaceae and Bacteroidetes bins. The sulfate reduction pathway (DsrAB, AprAB, Sat and Qmo genes) was identi ed in Deltaproteobacteria bins as well as in one Chloro exi, two Planctomycetes and three Parcubacteria (Candidatus Nealsonbacteria, Ca. Zixibacteria and Ca. Abyssubacteria) bins recovered from the saline anoxic waters (Figs. 3 and 4). Anaerobic sul te reductase genes (asrAB) were detected in all Omnitrophica bins as well as in two Planctomycetes and two Parcubacteria bins. Polysul de reductase genes (psrA) were found in Deltaproteobacteria as well as in some Marinimicrobia and Bacteroidetes bins. Sulfhydrogenase genes (hydABCD), involved in S 0 and polysul de reduction were also identi ed in genomic bins. However, while HydACD genes were detected in 8 Parcubacteria bins, including Ca. Kuenenbacteria, the gene coding the beta subunit (hydB) was detected only in half of these bins. By contrast hydB alone was also detected in Omnitrophica, Marinimicrobia and WOR1 (Ca. Saganbacteria) bins (Fig. 3). NADH-dependent persul de reductase genes (npsr), involved in persul de, polysul de or S 0 reduction were identi ed in Deltaproteobacteria, Parcubacteria (Ca. Abyssubacteria and Candidate Division KSB1) as well as in two Planctomycetes bins. A fusion of DsrE, Npsr, and TusA genes and including two rhodanese regions was also identi ed in the metagenomic sequences and in two bins related to Ca. Nealsonbacteria and Phycisphaeraceae (Fig. 3), suggesting a novel gene. Rhodanese regions and TusA protein are involved in sulfur binding and intracellular transport while both DsrE and Npsr are involved in sulfur reduction, suggesting a role in intracellular elemental sulfur reduction for this new gene.

Microbial Network For Sulfur Metabolism
Based on the identi ed genes, metabolic capabilities of the Lake A populations were inferred to determine which S-containing molecules were potentially produced or consumed throughout the water column (Fig. 5). Up to 100 different bins (65% of the bins with S -cycling genes) were predicted to be associated with hydrogen sul de production or consumption. Metabolic potentials involving the production or consumption of polysul des (89 bins, 58%), thiosulfate (72 bins, 47%) and tetrathionate (61 bins, 40%) were also widespread in the Lake A community (Fig. 5). Metabolic potential associated with sul te and sulfate was less represented in the community with only 46 bins (30%) associated with sul te and 30 bins (19%) with sulfate (Fig. 5).

Discussion
Ice-capped, the meromictic Lake A is an extreme microbial ecosystem, where strong and persistent environmental gradients provide a natural model for broader understanding of aquatic biogeochemical cycles. The water physico-chemical strati cation pro le measured in 2017 in the Lake A has been observed since 1974 [18][19][20], indicating a highly stable system and allowing extrapolation of geochemical pro les from historical data. Based on a previous complete geochemical characterization of the lake waters [21] high concentrations of sulfate occur in the chemocline and in the anoxic waters with increasing sul de concentrations with depth ( Fig. 1). High manganese concentrations were also detected at the chemocline while a peak of iron was observed few meters below ( Fig. 1 [21]). Finally, at similar snow and ice cover, photosynthetically available radiation (PAR) was detected down to 20 meters (Fig. 1  [24]). These extrapolations were supported by the depth distribution of green sulfur bacteria (Chlorobium), that con rmed light and sul des transition zones around 24 meters. Together these observations indicate a stable and intense redox gradient throughout the water column for microbial selection and growth.
RNA and DNA-based 16S amplicon and metagenomic sequencing from the High Arctic Lake A waters revealed multiple contiguous layers of complex yet stable and potentially active microbial communities, with putative metabolism aligning with the geochemical gradients of the lake (Fig. 1). The microbial community of the oxic mixolimnion beneath the ice was consistent with cold freshwater communities, with lineages of the Verrucomicrobia, Bacteroidetes, Actinobacteria, Cyanobacteria and Betaproteobacteria, which are frequently observed in lakes and rivers [25][26][27]. At the chemocline, alphaproteobacterial chemotrophic sulfur oxidizers and phototrophic sulfur oxidizers (Chlorobiaceae), both previously observed in microbial surveys of Antarctica [17,28] and temperate meromictic lakes [19,29] co-occurred since the lower depth limits of the Lake A photic and aerobic zones coincided (Fig. 1).
The chemocline microbial community also shared major similarities with marine communities with, for example, strong proportions of Marinimicrobia (SAR406), Pelagibacter (SAR11) and Deltaproteobacteria SAR324, which are frequently detected in seawater, hadal waters in the deep ocean and oxygen minimum zones [30][31][32]. By contrast, the microbial community of the saline anoxic monimolimnion showed homologies with deep-sea hypersaline anoxic basin communities, notably with sequences related to Chloro exi MSBL5, Desulfobacteraceae MSBL7, Planctomycetes MSBL9 and Cloacimonadales MSBL8 [33]. At these depths, the microbial community also shared similarities with anoxic and sul dic marine sediments, where Deltaproteobacteria SEEP SRB1 and Desulfarculaceae, Atribacteria, Omnitrophica and Chloro exi members also ourish [34]. Taken together these results reveal that, cascading along its geochemical gradients, the Lake A water column hosts a panoply of microorganisms, relevant to a broad range of environments and environmental conditions from oxic freshwaters to anoxic marine sediments.

New Microbial Agents And Metabolic Pathways For Sulfur Transformation
From surface layers to the bottom, most of the genomic bins (61.4%) recovered from Lake A included genes for sulfur cycling (Figs. 3 and 4). Furthermore, bacterial abundance in Lake A was correlated with the average number of metabolic pathways for sulfur transformation per bin (R 2 = 0.69, p = 0.04), supporting the notion that sulfur cycle represents a major process in Lake A waters and involves a large diversity of microorganisms. Reconstruction of genomic bins highlighted that in addition to the conventional taxa associated with the classical sulfur cycle in meromictic saline lakes such as sulfatereducing Desulfobacteraceae, sulfur-oxidizing Alphaproteobacteria and phototrophic sulfur oxidizing Chlorobiaceae [17], various lineages with poorly known ecological functions are also involved in sulfur transformations. Among these lineages, key genes of sulfur metabolism were identi ed in Ca. WOR1, SAR86, Lentisphaerae, Aminicentantes, Marinimicrobia, Calditrichaeota, Omnitrophica and Parcubacteria, thereby expanding the known diversity of sulfur cycling bacteria ( Fig. 3 and Supplementary Table S1).
A strong functional redundancy in sulfur transformation pathways was detected throughout the water column, with taxonomically diverse microorganisms having similar metabolic pathways (Figs. 3 and 4). For example, sul de oxidation potential through SQR and the SoxABCXYZ complex was identi ed in Alphaproteobacteria (Rhodospirillales) and Gammaproteobacteria SAR86. Oxidative DsrAB genes were also identi ed in half of the Rhodospirillales bins and in the Deltaproteobacteria SAR324, while the Hdrlike complex, also involved in sul de oxidation was discovered in another Alphaproteobacteria bin, congruent with experimental evidence in the Alphaproteobacterium Hyphomicrobium denitri cans [23]. In addition, the SoxABCXYZ complex coupled with SoeAB genes were detected in Betaproteobacteria bins whereas SQR, FccAB and the oxidative DsrAB genes were ascertained in the Chlorobiaceae bin [28] (Fig. 3). These multiple pathways for sul de and sul te oxidations accumulated in the freshwater and chemocline layers (Fig. 4), suggesting that sul des sustain multiple ecological niches in aquatic environments over space and/or time. If the occurrence of these various sulfur-oxidizing pathways and lineages at the chemocline is supported by the chemical pro les, their identi cation in the upper freshwater samples, coupled with the co-detection of sulfonate degradation genes (Fig. 2) suggests that organic sulfur molecules may also support sulfur-oxidizing populations in non-sul dic waters, multiplying the availability of ecological niches and allowing functional redundancy.
In the anoxic saline layer, the dissimilatory sulfate reduction pathway (Sat, Qmo, AprAB and DsrAB genes) occurred in the Deltaproteobacteria bins as expected [7], but was also found in genomic bins a liated with Chlorolexi, Planctomycetes, Calditrichaeota and Parcubacteria (Ca. Nealsonbacteria, Ca. Abyssubacteria and Ca. Zixibacteria) (Fig. 3). These results provide an ecological context for these new lineages of sulfate reducers, that were previously proposed after mining of combined metagenomic datasets [6]. Our metagenomic survey also predicted a sul te reduction potential (AsrAB genes) for Omnitrophica members (Figs. 3 and 4), supporting data mining [6], as well as for few Planctomycetes and Patescibacteria populations in the sul dic waters of the monimolimnion.
Patescibacteria, Planctomycetes and Chloro exi phyla showed the strongest variability of genomic potential within their lineages (Figs. 3 and 4). Each of these phyla included populations predicted to gain energy from thiosulfate oxidation, sulfate and sul te reduction as well as polysul de/elemental sulfur reduction or oxidation. A new fusion gene probably involved in elemental sulfur/polysul de reduction was also identi ed in two genomic bins a liated with Patescibacteria and Planctomycetes phyla. Sequence comparison with public databases indicated that this gene is also present in a single-cell genome related to the Planctomycetes-derived phylum of the Kiritimatiellaeota, isolated from a deep continental microoxic subsurface aquifer [35], suggesting that this gene might be relevant in microoxic conditions. Interestingly, Planctomycetes, Chloro exi and Ca. Nealsonbacteria (Patescibacteria) genomic bins also included numerous genes (> 10 per bin) coding for sulfatases. These hydrolytic enzymes potentially release sulfate from sulfated organic matter [36], providing additional electron acceptors throughout the water column. Together these results extend the diversity of sulfur cycling microorganisms and metabolic pathways. They suggest new fundamental roles in sulfur cycling for members of the Patescibacteria, Planctomycetes and Chloro exi in aquatic environments, with strong ecological niche differentiation within member of these lineages.

Utilization Of Sulfur Cycle Intermediates
Sulfur cycle intermediates (SCIs: thiosulfate, tetrathionate, sul te, polysul des, elemental sulfur) have a large biogeochemical signi cance in anoxic and marine environments, creating shortcuts around the classic sulfur cycle [1,8]. The potential for oxidation and reduction of these sulfur molecules was widespread in the Lake A microbial community, with taxonomically diverse lineages potentially using SCIs as electron donors or acceptors (Fig. 3). The number of genes for SCI metabolism and sulfate reduction was similar, suggesting that SCI utilisation might represent a quantitatively important process in Lake A sulfur cycling (Fig. 2). Furthermore, the number of genomic bins with SCI utilization genes exceeded the number of bins with sulfate reduction and hydrogen sul de oxidization pathways and SCIs were found as major hubs in the sulfur metabolic network (Fig. 5), indicating a wide diversity of microorganisms able to process SCIs.
The potential to use SCIs was shared between specialists that use only a limited range of these molecules, and generalists that could potentially metabolise a broad range of sulfur compounds including sulfate or hydrogen sul de. The specialists included some members of the Parcubacteria with the potential limited to thiosulfate oxidation, Omnitrophica with only genes for sul te reduction and Bacteroidetes populations with the metabolic potentials for thiosulfate and polysul de oxidation. By contrast, generalists were mainly represented by members of the Deltaproteobacteria or Alphaproteobacteria lineages with a large suite of sulfur transformation genes, suggesting high variability in substrate utilization (Fig. 4).
The taxonomically diverse microorganisms observed here are likely fuelled by microbial phototrophic and chemotrophic hydrogen sul de oxidations that generate SCIs of various oxidation states [8], as well as by abiotic oxidation of hydrogen sul de with iron and manganese oxides present in elevated concentrations in the Lake A ( Fig. 1) [1]. Although SCIs have not been measured in Lake A, a sul dic smell and a yelloworange color of the water below 22 m was detected during sampling supporting the presence of polysul des and aqueous elemental sulfur in the water and the metabolic potentials detected in metagenomic dataset. Together these results indicated a strong ecological role for SCIs by providing an energy source for a diverse and abundant microbial community in both fresh, brackish and saline waters.

Organic Sulfur Molecules As Sci Sources
Oxidized organic sulfur molecules, such as sulfonate and sulfonium are produced by the phytoplankton as osmoprotectants, antioxidants [37] or predator deterrents in aquatic environments [38], and are an important source of sulfur and carbon for pelagic bacteria in the ocean [14]. These compounds are also frequently detected in metabolomes of diatoms [39]. Eukaryotic microalgae were detected in the oxic mixolimnion in the metagenomic dataset (e.g., Chrysophyceae, Ochrophyta, Chlorophyceae, data not shown) and in a previous amplicon survey [24], therefore the presence of eukaryotic sulfur metabolites in the Lake A would not be surprising. The genetic capacity for OSM degradation was widely distributed, occurring in 56% of the genomic bins (Figs. 3, 4 and 5). Proteobacterial lineages were detected as major sulfonate degraders in the mixolimnion and chemocline. Since these degradation processes release sul te in the water, our results suggest that organic sulfur molecules might have an important ecological role, providing sulfur compounds of intermediate oxidation states in aerobic and microaerobic aquatic systems regardless of the salinity and sulfate concentration.
Genes for DMSP utilisation, identi ed in the freshwater and brackish waters of the lake were also detected in Alphaproteobacteria (Rhodospirillales and SAR11/Pelagibacter bins) and Actinobacteria (Acidimicrobiia), as reported in surface oceans [14]. In addition, numerous genes for respiration of dimethylsulfoxide (DMSO) were identi ed in the anoxic monimolimnion and in Desullfobacteraceae, Chloro exi and Bacteroidetes bins, suggesting that algal metabolites could sink within senescent phytoplankton from the upper water column and be used as an alternative energy source by anaerobic microbial populations in the lower water column of Lake A.

Conclusions
Isolated and highly strati ed, the High-Arctic meromictic saline Lake A offered an outstanding opportunity to investigate microbial metabolism along oxygen, sulfate, sul de and salinity gradients. Although the chemocline harbored the taxonomically and functionally most diverse and abundant microbial community, de ning this layer as an hotspot for microbial activity and sulfur transformations as in other meromictic lakes [17,19], genes for sulfur transformations were identi ed throughout the water column, and provided new insights into sulfur cycling microorganisms over a large range of environmental conditions. The pathways and taxa involved in sul de oxidation and sulfate reduction were complex and diverse. However, the metagenomic dataset revealed that sulfur transformations were not limited to these classic processes and that multiple sulfur cycling pathways could be simultaneously operating throughout the water column, with taxonomically diverse populations using sulfur cycle intermediates as electron donors or acceptors. Genes for organic sulfur molecule degradation and respiration were also abundant and widely distributed in the microbial community, suggesting that phytoplankton metabolites might also be a major source of energy for freshwater and marine bacteria. Our data extend the diversity of sulfur cycle lineages and metabolic pathways in aquatic ecosystems and emphasize the ecological importance of sulfur cycle intermediates as key hubs for electron ow and energy production over a wide range of environmental conditions.

Sample Collection And Nucleic Acid Extraction
In summer 2017 (18 July 2017), three independent 24 cm-diameter holes were drilled through the ice Nucleic acids (DNA and RNA) of two of the replicates samples per depth were extracted from the same Sterivex lters using Qiagen Allprep DNA/RNA Mini Kit. The Sterivex cartridges were opened and the membrane lters were cut into small pieces before the lysis steps, as previously described [40]. All steps of the nucleic acid extractions, from the opening of the lters to the nucleic acid resuspension in autoclaved, ltered and UV-treated water, were carried out in a sterile laminar ow cabinet. Negative control (no template) of nucleic acids extraction was simultaneously carried out. The DNA extracts were stored at -20 °C until library preparation. For RNA extracts, two additional DNase steps (DNase I, Ambion, Foster City, CA, USA) were carried out to remove any trace of carried over DNA. The absence of DNA contamination was con rmed by ampli cation of 16S rRNA genes with bacterial primers using the RNA extracts (undiluted and diluted ten times) as template, with no product detected after 35 PCR cycles. The RNA was then immediately converted to cDNA using a High-Capacity cDNA Reverse Transcription kit (Applied Biosystems, Foster City, CA, USA) and stored as cDNA at -20 °C until library preparation.

Quantitative Pcr
The abundance of bacterial and archaeal 16S rRNA genes was estimated on the two replicate samples per depth using quantitative PCR (qPCR) with primers Bact1369f/Bact1492r [41] and Arc787f/Arc1059r [42], respectively. Quanti cation was performed in triplicate with a range of template concentrations (0.1, 0.5, 1 ng of DNA) to compensate for any PCR inhibition. Genomic DNA extracted from Methylomonas methanica (DSM25384) and Methanosarcina acetivorans (DSM2834) were serially diluted to construct standard curves (concentration ranged from 10 2 to 10 6 16S rRNA genes per reaction). The R 2 of standard curves obtained by qPCR were above 0.99, PCR e ciencies were above 88.7%, and melting curves showed no trace of non-speci c ampli cations. Threshold cycle (Ct) of the samples ranged from 15 to 25 cycles whereas Ct of negative controls (water) were all after 37 cycles. The qPCR results were expressed in terms of 16S rRNA gene numbers per millilitre of water sample (Fig. 1d).
Samples and negative controls of nucleic acids extraction, transcription and PCRs were sequenced using an Illumina MiSeq v3 kit at the IBIS/Laval University, Plate-forme d'Analyses Génomiques (Québec, QC). Reads were assembled into single paired-end sequences, curated and clustered into OTUs (97% sequence similarity) as detailed in a GitHub repository (https://github.com/CruaudPe/MiSeq_Multigenique). OTUs detected in negative controls were removed from the analysis as described [27]. Taxonomic a liations of the reads were determined with Mothur [46] using BLAST against Silva database release 132 as reference [47].

Metagenomic Library Preparation, Sequencing And Analysis
One metagenome per sample depth (8 metagenomes) was constructed using a Nextera XT Library Kit (Illumina, San Diego, CA, USA). The 8 metagenomes were pooled equimolarly then sequenced in two Illumina MiSeq (2*300 bp) runs and one Illumina NexSeq run (2*150 bp) at the Institut de Biologie Integrative et des Systèmes (IBIS) sequencing platform (Univeristé Laval, Canada) and at the CGEB -Integrated Microbiome Resource (Dalhousie University, Canada) respectively. Datasets were quality ltered using the Trimmomatic tool [48], with default settings. Paired-end joining was done using FLASH2 [49]. The 16S rRNA reads longer than 110 bp were isolated from metagenomic reads using REAGO 1.1 [50], and taxonomic assignments were performed as for the 16S rRNA gene amplicons.
Each metagenome was assembled separately from paired-end reads passing quality ltering using SPAdes [51]. Assembled contigs and mapping les (BAM les generated using BBmap) were uploaded to the Department of Energy Joint Genome Institute (DOE-JGI) IMG/MER analysis pipeline [52] for gene calling and functional annotation. To account for differences in sequencing depth between samples, metagenomes were normalized to the size of the smallest dataset.

Binning And Functional Characterization
For metagenome assembled genome reconstruction, all quality ltered sequences were pooled and coassembled using MEGAHIT [53]. Read coverage of the contigs was carried out using bwa-mem (http://bio-bwa.sourceforge.net), followed by contig binning using MetaBAT-2 [54] with contigs longer than 2000 bp. The completeness and contamination level of the combined genomic bins were then evaluated using CheckM [55]. Only bins with a contamination level under 5% and completeness above 50% were analysed. Genetic composition of genomic bins was then explored using KEGG [56] and MetaCyc [57] pathway mappers with genes identi ed by IMG/MER in the co-assembly. The results were manually checked and the presence of speci c pathways was determined by detection of key genes as detailed in the Supplementary Material.

Statistical analysis
Statistical analyses of the data set (Student t-test, Non-Parametric Multivariate Analysis Of VAriance (NPMANOVA), Bray-Curtis-based dissimilarity index calculations and correlation-based clustering, Non Metric Multidimensional Scale (NMDS)) were carried out according to recommendations of the Guide to Statistical Analysis in Microbial Ecology [58], using PAST software [59].       Depth distribution of metabolic potentials identi ed in the genomic bins. Each point represents a genomic bin with the color corresponding to its taxonomy. Figure 5 Metabolic network of the Lake A. . a) Genomic bins are represented as circles with color corresponding to their taxonomic a liation and size corresponding to the average number of reads mapping on the genomic bin. Genomic bins are connected to sulfur compounds (white circle with pie chart) predicted to be utilized (blue link) or produced (purple link). For each sulfur molecule, pie chart represents the proportion of consumption (blue) and production (purple) and the size of the pie chart is proportional to the number of connections. b) Metabolic network overlaid with the potential for organic sulfur molecule utilisation in green.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.