Unveiling microbial interactions in stratified mat communities from a warm saline shallow pond

Summary

Modern phototrophic microbial mats are complex communities often used as analogs of major Precambrian ecosystems. Characterizing biotic, notably metabolic, interactions among different microbial mat members is essential to gain insights into the ecology and biogeochemistry of these systems. We applied 16S/18S rRNA metabarcoding approaches to characterize the structure of archaea, bacteria and protist communities from microbial mats collected along strong physicochemical (oxygen, salinity, temperature, depth) gradients in a shallow pond at the salar de Llamara (Chile). All mats were highly diverse, including members of virtually all known high-rank eukaryotic and prokaryotic taxa but also many novel lineages. Bacterial candidate divisions accounted for almost 50% of sequences in deeper mats, while Archaea represented up to 40% of sequences in some mat layers. Molecular phylogenetic analyses revealed six novel deeply divergent archaeal groups, along abundant and diverse Pacearchaeota and Woesearchaeota. Multivariate statistical analyses showed that local environmental conditions strongly influenced community composition. Co-occurrence network structure was markedly different between surface mats located in the oxygenated zone and mats located in transition and anoxic water layers. We identified potential biotic interactions between various high- and low-rank taxa. Notably, a strong positive correlation was observed between Lokiarchaeota and the poorly known candidate bacterial division TA06.

Introduction

Microbial mats are vertically laminated and phylogenetically diverse benthic microbial communities (Des Marais, 1990) that are embedded in an organic matrix made of extracellular polymeric substances (EPS; Dupraz et al., 2009). They are characterized by the presence of steep biogeochemical gradients generated and maintained by the metabolic activity of diverse functional groups (van Gemerden, 1993; Dupraz and Visscher, 2005). In the literature, “microbial mat” can refer to both lithifying (e.g. stromatolites) and, as in the present study, non-lithifying systems (Dupraz and Visscher, 2005). Stromatolites are abundant in the fossil record from the Proterozoic (2.5 Ga – 0.5 Ga) and represent major Precambrian ecosystems (Riding, 2000; Tice and Lowe, 2004; Allwood et al., 2006). Modern microbial mats are considered as relevant analogs of these ancient systems and used as models to understand their ecology (Dupraz and Visscher, 2005; Foster and Mobberley, 2010). Although less conspicuous than in the past, they are found worldwide in a variety of water bodies including lagoons, marine intertidal and subtidal zones, hypersaline settings, hot springs and freshwater rivers and lakes (Des Marais, 1990).

Several microbial diversity surveys based on the amplification and sequencing of prokaryotic small subunit ribosomal RNA (SSU rRNA) genes of saline (salinity > 3%) to hypersaline mats (salinity > 10%) have revealed the presence of complex communities distinct from the water column (Ma et al., 2010), containing members from most high-rank prokaryotic phyla. They include solar salterns in Israel (Sørensen et al., 2005), Mexico (Ley et al., 2006; Jahnke et al., 2008; Robertson et al., 2009; Harris et al., 2013) and Spain (López-López et al., 2010), coastal areas in the Persian Gulf (Abed et al., 2008), Australia (Allen et al., 2009; Wong et al., 2015; Suosaari et al., 2016) and the North Sea (Bolhuis and Stal, 2011; Bolhuis et al., 2013), the Great Sippewissett salt marsh (Armitage et al., 2012), lakes in Spain (Jonkers et al., 2003), The Bahamas (Baumgartner et al., 2009) and Christmas island (Schneider et al., 2013), salt flats (Demergasso et al., 2003; Rasuk et al., 2016) and wetlands (Farías et al., 2014) in Chile. The eukaryotic diversity has, however, largely been ignored and has been investigated in only a few studies (Feazel et al., 2008; Allen et al., 2009; Ma et al., 2010; Edgcomb and Bernhard, 2013). Although often heavily biased towards Metazoa, these studies revealed the presence of diverse phylogenetic lineages belonging to all major eukaryotic super-groups.

The Atacama desert, located in the North of Chile, is one of the driest places on Earth (Clarke, 2006) and is often used as a model to study both the dry and UV exposure limits of life (Warren-Rhodes et al., 2006). Northern Chile harbors numerous saline water bodies and salt crusts, locally called salares (Risacher et al., 2003). The exploration of the microbial diversity from these salares has also been mostly restricted to bacteria (Drees et al., 2006; Demergasso et al., 2008; Farías et al., 2014) and little is known about their archaeal and eukaryotic counterparts. The salar de Llamara, located at an altitude of 750 m, is one of them. The bacterial and, to a lesser extent, archaeal diversity of various lakes (Demergasso et al., 2004), evaporite domes (de los Ríos et al., 2010; Rasuk et al., 2014) and mats (Demergasso et al., 2003; Rasuk et al., 2016) has been documented from this area. Most high-rank bacterial phyla were represented, including numerous Candidate divisions. However, these studies remain too sparse and do not provide insight into the biotic and abiotic parameters that determine the composition of these complex layered communities. In particular, specific microbe-microbe interactions involving parasitism, predation and syntrophy, likely very important in these complex systems, remain unknown.

In this work, we applied high-throughput 16S/18S rRNA metabarcoding approaches to study the structure (composition) of archaeal, bacterial and protist communities of four stratified microbial mats located in a shallow pond at the salar de Llamara. Although the mats located a few centimetres apart, they were exposed to strong physicochemical (oxygen, salinity, temperature) gradients. Multi-variate analyses show that the composition of these rich and phylogenetically stratified microbial communities is partly determined by those physicochemical parameters. Co-occurrence networks indicate potential ecological associations between community members at different taxonomic resolution including, notably, potential positive and negative interactions among members of poorly known bacterial and archaeal candidate divisions.

Results

Composition of microbial mat communities

We characterized the composition of eukaryotic and prokaryotic communities from quality-filtered sequences obtained by pyrosequencing of 16S and 18S rRNA gene amplicons. Biological replicates were carried out for all the samples. In addition, we amplified and sequenced in parallel 16S rRNA gene fragments using universal prokaryotic (bacteria + archaea) and bacteria-specific primers for comparison. These constitute extra technical replicates for bacteria. Reads from the 16 different samples (14 in the case of eukaryotes, see Material & Methods) were pooled in order to define OTUs that could be fully compared among samples. We obtained a total of 437 archaeal OTUs, 2768 bacterial OTUs and 586 eukaryotic OTUs from both universal prokaryotic and eukaryotic primers. A total of 4105 quality-filtered OTUs was produced from the primer set specific to Bacteria (Supporting information Fig. S1A and Supporting information Table S3). The eukaryotic component of LLA9-B-2b was largely dominated by annelid sequences (6691 out of the 6699 eukaryotic quality-filtered sequences in this sample). This implies that the (very different) prokaryotic diversity observed in this sample (not shown) was likely influenced by annelid microbiomes. Accordingly, LLA9-B-2b was removed from any subsequent analyses.

We calculated various ecological indexes to assess the diversity of bacteria, archaea and eukaryotes in these microbial communities (Supporting information Table S3). Overall, the prokaryotic communities were highly diverse and even, potentially reflecting a high level of biotic interactions, as increased species interactions often correlate with higher evenness (Seymour and Altermatt, 2014). The eukaryotic communities were characterized, to the exception of the top layer mat samples (LLA9-A-1, LLA9-B-1 and LLA9-C-1), by a lower diversity dominated by a few abundant OTUs. NMDS analyses revealed a high similarity of community compositions inferred from replicate samples of each layer. In addition, the bacterial community structure inferred from both bacterial and universal prokaryotic primer sets was similar (Fig. 2A). In the following we thus considered only the results obtained with the prokaryotic primer set, as it gives access to the relative proportions of both Bacteria and Archaea. NMDS analyses also revealed highly congruent OTU composition across all replicates for both eukaryotic and universal prokaryotic primer sets (Fig. 2 B and C).

Fig. 2.

Non-Metric multi-Dimensional Scaling (NMDS) plots showing differences in (A) bacterial, (B) prokaryotic (archaea+bacteria) and (C) eukaryotic community composition between all samples. They were based on (A) the relative proportions of high-ranked bacterial phyla inferred from sequences obtained with both bacterial and prokaryotic primer sets and (B-C) OTU frequencies.

Relative proportions of high-rank prokaryotic and eukaryotic taxa detected in the different mat layers are shown in Fig. 3 (see Supporting information Table S4 for details on low-frequency prokaryotic taxa). In general, bacteria were highly diverse, with most high-rank bacterial phyla represented, including a wide variety of Candidate Divisions which, in deep layers, accounted for up to 20-45% (Fig. 3 A and B). Archaea were also abundant and diverse (see below), accounting for up to ca. 40% of prokaryotic sequences and 20% of OTUs in some layers (Supporting information Fig. S2). Mats located in the oxygenated area of the pond were largely dominated by bacteria. The surface layers of these mats (LLA9-A-1 and LLA9-B-1) were similar at high-rank taxa (Fig. 3 A and B) and OTU levels (Fig. 2B), containing many photosynthetic microorganisms. They notably harboured the highest proportion of Cyanobacteria (20-25%, mainly Chroococcales; Supporting information Fig. S3), along numerous Bacteroidetes, Verrucomicrobia, Alphaproteobacteria (mainly Rhodobacterales and Rhodospirillales; Supporting information Fig. S4A) and Deltaproteobacteria (mainly Desulfobacterales and Myxococcales; Supporting information Fig. S4B). LLA9-B-2, just below LLA9-B-1, contained abundant Actinobacteria, Chloroflexi (exclusively Anaerolineae, a non-phototrophic lineage; Supporting information Fig. S5A) and Fibrobacteres. Potential photosynthetic members were represented by Cyanobacteria (5-10% of the sequences) and Alphaproteobacteria (Rhodobacterales and Rhodospirillales in LLA9-B-2b). In these mats from the oxic zone, the presence of Archaea was minor (0-8%). However, the structure of mat LLA9-C, situated in the transition zone, was completely different (Fig. 3A). Indeed, Archaea were highly abundant and were dominated by diverse OTUs affiliated to Pacearchaeota and Woesearchaeota (both formerly Euryarchaeota DHVEG-6 cluster; Castelle et al., 2015) and to Thermoplasmata (Fig. 3A). Sequences belonging to Lokiarchaeota, a recently characterized group (Spang et al., 2015), were also detected. These Archaea thrived along numerous Bacteroidetes, belonging to the Sphingobacteriia (particularly in LLA9-C-1; Supporting information Fig. S5B), and Planctomycetes, while Cyanobacteria were less abundant (<10%, mainly Oscillatoriales; Supporting information Fig. S3). Potential anoxygenic phototrophs were detected in significant proportions in LLA9-C-1 (mainly Rhodobacterales and Rhodospirillales; Supporting information Fig. S4A). Candidate divisions were particularly abundant in the two deeper mats; where they corresponded mostly to Aminicenantes (former OP8), Parcubacteria (former OD1) and TA06 (Fig. 3B). A few OTUs from Candidate divisions dominated the two deepest mats; they were affiliated to Parcubacteria (OTU-59015 in LLA9-C-2 and LLA9-C-2b), Aminicenantes (OTU-61397 in LLA9-D-2/2b), Latescibacteria (former WS3; OTU-59035 in LLA9-D-2b and LLA9-C-3b) and TA06 (OTU-59384 in LLA9-D-1/1b and LLA9-C-3/3b, OTU-59763 in LLA9-C-1/1b and LLA9-C-2/2b; Supporting information Fig. S6). The prokaryotic community of the deepest mat, LLA9-D also comprised Archaea (up to 25%), along with other moderately abundant groups that included Bacteroidetes, Chloroflexi, Deltaproteobacteria (mostly sulfate-reducers; Supporting information Fig. S4B), Planctomycetes and Spirochaetes (Fig. 3A). Between 3-5% of the sequences of LLA9-D-1/1b corresponded to Cyanobacteria in an area where the conditions are permanently anoxic (Fig. 1). These cyanobacteria were related to various Oscillatoriales, although most of the sequences (92% and 80%, respectively) fell into the same OTU-59050 (Supporting information Fig. S3).

Fig. 3.

Bar charts showing the relative proportion of 16/18S rRNA gene sequences in Llamara mats affiliated to high-rank (A) prokaryotic taxa, (B) bacterial candidate divisions (the percentage of these sequences among total bacteria is indicated in brackets) and (C) eukaryotic taxa.

Fig. 1.

Sampling site and microbial mat fragments collected in a pond from the salar de Llamara. Mats were sampled along physicochemical gradients (depth, salinity, oxygen, temperature), as indicated on the right panel.

Eukaryotes were also present in Llamara mats, being more abundant and diverse in the oxic zone mats (Fig. 3C). Choanoflagellida, Bacillariophyta, Ciliophora and Palmophyllales represented collectively more than 80% of eukaryotic sequences in the LLA9-A-1 and LLA-B-1, where a few abundant OTUs dominated (Supporting information Fig. S7). LLA9-B-2 was much less diverse; nearly 95% of the sequences belonged to a unique ciliate OTU (OTU-2538). Below the chemocline, the patterns of diversity were distinct between the different mat layers (Fig. 3C), possibly because of stochastic effects due to a lower total number of sequences. The mat from the transition zone, LLA9-C, was comparatively more diverse. Its surface layers were dominated by heterotrophic Stramenopiles, Cryptophytes and ciliates, represented by few abundant OTUs (bicosoecid OTU-578, cryptophyte OTU-1269, ciliate OTU-2106; Supporting information Fig. S7). Its deeper layer, LLA9-C-3 was dominated by fungal sequences, essentially the OTU-81 related to Chytridiomycota (Supporting information Fig. S7-S8), although Amoebozoa accounted for ca. 10% of sequences (Fig. 3C). Fungi, essentially dominated by Ascomycota (OTU-319) were also the most abundant in the deepest and hottest mat, where they accounted for at least 60% of eukaryotic sequences. LLA9-D-2b represents an exception, since most of the detected eukaryotic OTUs belonged to Bacillariophyta. The presence of diatoms in this dark layer may seem surprising. Although contamination during sampling cannot be fully excluded, this explanation is unlikely. Not only did we subsample with care using sterilized tools, but both at the OTU (Fig 2B) and high-taxonomic level (Fig.3A and B), the prokaryotic diversity from this subsample is highly consistent between the two replicates and do not show the presence of any photosynthetic taxa. The most likely explanation for this observation is the presence of diatom traces that got buried as the mat grew. Because most eukaryotes do not likely thrive in deep layers, the stochastic amplification of genes from decaying diatoms may result in apparent dominance. In reality, this needs to be interpreted as scarcity of autochthonous eukaryotes in that layer. Sequences related to Metamonada (Excavata), anaerobic flagellates, were exclusively found in LLA9-D-1.

Novel archaeal groups

Compared to most ecosystems, Llamara microbial mats (especially in transition and anoxic zones) contained high relative proportions of archaea (Fig. 3A and Supporting information Fig. S2). BLAST analyses suggested that a significant fraction of archaea did not accommodate within known groups. Therefore, we built a phylogenetic tree using representative sequences of all archaeal OTUs plus 11 longer, Sanger sequences from the same mats to refine taxonomic assignments (Fig. 4 and Supporting information Fig. S9). We thus confirmed the affiliation of many sequences to known archaeal taxa. This was the case of Thermoplasmata, which were diverse and abundant in all Llamara mats but LLA9-A (44 OTUs, 6043 reads). However, this diversity was largely dominated by a single OTU (OTU-59849, 3854 reads; Supporting information Fig. S10) affiliated to the Marine Benthic Group D. In addition, 12 OTUs (252 reads) branched close to Methanomassiliicoccus luminyensis (Supporting information Fig. S9), an archaeon able to reduce methanol with hydrogen and that represents a recently discovered order of methanogens (Borrel et al., 2013). Two large groups of OTUs were affiliated to Pacearchaeota and Woesearchaeota, encompassing most of the archaeal diversity identified (309 OTUs, 6110 reads), with many low-frequency OTUs. Lokiarchaeota, the closest archaeal relatives of eukaryotes discovered so far (Spang et al., 2015), were mostly represented by a single OTU (OTU-61378, 80% of Lokiarchaeota sequences). Only two OTUs affiliated to Bathyarchaeota (former Miscellaneous Crenarchaeotic Group; Fig. 4), a rather abundant and widespread group in sediments (Meng et al., 2014; Fillol et al., 2016).

Fig. 4.

Approximate maximum likelihood phylogenetic tree of representative 16S rRNA gene archaeal OTU sequences detected in Llamara mats. Lineages containing more than one OTU were collapsed. The percentage represented by archaeal sequences among the prokaryotic community of each mat is shown after the mat name. Llamara Mat Group (LMG) 1 to 6 correspond to groups with no close relatives in the databases. The full tree is available in Supporting information Fig. S9.

Several deep-branching monophyletic clades of OTUs had no close relatives in databases. We thus defined six novel groups of archaea; termed LMG-1 to LMG-6 (for Llamara Mat Group) (Fig. 4). LMG-2 was sister to the Halobacteria, a class fairly common in saline waters and sediments (Ma et al., 2010). If LMG-4 might have some affinity with methanogens, LMG-1 and LMG-3 are too deeply branching to hypothesize any potential metabolism from neighbouring groups. Finally, LMG-5 and LMG-6 branched among the TACK superphylum. They were not very diverse in terms of OTUs but still represented a substantial number of sequences in all mats but LLA9-A. LMG-5 was dominated by one OTU particularly abundant in LLA9-D-2 and LLA9-C-3 that represented 65% of sequences (OTU-61816). LMG-6 also contained one dominant OTU mostly present in the deepest mat layers (OTU-59982, 78% of sequences; Supporting information Fig. S9).

Influence of abiotic parameters on community structure

Although continuous, and located a few centimeters away, each mat sample developed under a set of specific abiotic factors, with oxygen values dropping to zero while salinity and temperature values increased by two fold between the surface and the bottom of the pond (Fig. 1). All these differences in environmental parameters correlated with the different structure of the prokaryotic communities as revealed by a Mantel test linking environmental and community pairwise distances between mat layers (r = 0.4094; p = 0.022). Indeed, only 3.4% of the total OTUs (126 out of the 3676 identified) were shared by the four mat samples (Supporting information Fig. S1B). They mostly affiliated to Bacteroidetes (Sphingobacteria), Hydrogenedentes (former BRC1), Planctomycetes (Phycisphaerae), Alphaproteobacteria, Deltaproteobacteria, Spirochaetes and Verrucomicrobia (Supporting information Table S5). A Kruskal-Wallis test (H = 27.99, p = 1.22e-07) showed that these cosmopolitan OTUs were not evenly distributed between top and bottom mat layers, which was consistent with differences in local environmental conditions.

In addition, a Canonical Correspondence Analysis (CCA) revealed distinct relationships between mat layers and specific parameters (Fig. 5). Microbial communities from the two shallowest mats were most influenced by the presence of oxygen, especially the superficial layers (LLA9-A-1, LLA9-B-1). Comparatively, intermediate and bottom mats and mat layers were, at various degrees, positively correlated with lower oxygen levels and higher salinity and temperature. Our results indicate that an important part of prokaryotic community composition variation is explained by environmental parameters (depth below water level and below mat surface, oxygen concentration, position in the mat, salinity and temperature).

Fig. 5.

Canonical correspondence analysis (CCA) plot of 16S rRNA gene sequence frequencies and a set of physicochemical parameters. b.w.l: below water level; b.m.s: below mat surface.

Biotic interactions within microbial mat communities

To identify potential biotic interactions within the dominant, prokaryotic communities of Llamara mats, we constructed several co-occurrence networks based on the most abundant OTUs and using stringent criteria. First, we constructed four independent networks to identify potential key interactions between dominant high-rank taxa within each of the mats. The network corresponding to the shallowest mat was characterized by many highly connected taxa. A mix of positive and negative correlations was detected, with the presence of Cyanobacteria excluding potential anoxygenic phototrophs (Alphaproteobacteria, Gammaproteobacteria and Chloroflexi; Fig. 6A), reflecting their stratified location within the mat. Bacteroidetes and Cyanobacteria were also positively associated in LLA9-B, where they were highly abundant (Fig. 6B). In these two mats, Alphaproteobacteria appeared to play a central role. The networks of the two deeper mats (Fig. 6C and D), less interconnected, harbored many poorly known phyla. The bacterial candidate division TA06 and Lokiarchaeota, even though they were not among the most abundant taxa, exhibited the only strong positive correlation detected in mat C (Fig. 6C). Finally, the network topology of mat D suggests numerous excluding relationships, i.e. negative correlations inferred from the observation that those taxa do not co-occur, and a central role for Latescibacteria in a cluster also involving LMG-5, Hydrogenedentes and Parcubacteria. Deltaproteobacteria and Chloroflexi were involved in interactions across all mats.

Fig. 6.

Co-occurrence networks showing potential interactions at high-rank taxa level in Llamara microbial mats. (A) LLA9-A, (B) LLA9-B, (C) LLA9-C and (D) LLA9-D. The thickness of the edges represents the strength of the correlation between taxa; thin edges being negative correlations and thick edges being positive correlations. The size of the pie charts illustrates the frequency represented by each taxon with the mat. The distribution of each taxon in the various mat layers is indicated within the pie charts.

To try to identify interactions at a finer level among widely distributed OTUs, we built a network based on OTUs present in at least ten samples (n=69), which included a large variety of bacterial and archaeal taxa (Fig. 7). The most abundant OTUs (n=3) were found to be positively correlated to one or several other OTUs. Indeed, the cyanobacterial OTU-61815 appeared to be strongly connected to Bacteroidetes (OTU-61527), Gammaproteobacteria (OTU-61583) and Verrucomicrobia (OTU-61396). Aminicenantes, through OTU-61397, formed a module potentially important in the deep mat layers with several representatives of Planctomycetes (OTU-60620), Deferribacteres (OTU-61663) and LMG-5 (OTU-61816 and OTU-59962). Finally, the OTU-61333, affiliated to Planctomycetes, was only correlated to Spirochaetes (OTU-59111). Some moderately frequent OTUs also seemed to be important, as they were connected with a variety of potential partners. They notably included Bacteroidetes (OTU-61410), Alphaproteobacteria (OTU-62466), Gammaproteobacteria (OTU-61583), as well as Lokiarchaeota (OTU-61378).

Fig. 7.

Co-occurrence network built using the 69 OTUs present in at least 10 mat samples. The thickness of the edges represents the strength of the correlation between two OTUs; thin edges being negative correlations and thick edges being positive correlations. The size of the pie charts illustrates the frequency represented by each OTU across the four mats. The distribution of each OTU in the various mat layers is indicated within the pie charts.

Discussion

Phototrophic microbial mats: stratified complex systems strongly influenced by environmental factors

Diversity surveys of archaeal, bacterial and eukaryotic components in microbial phototrophic mat communities are still scarce in the literature, even though the study of these systems is relevant in many respects. First, their persistence over geological times makes them natural models to improve our understanding of the evolution of life throughout the Precambrian. Also, because they are little studied, contemporary microbial mats represent potential reservoirs of phylogenetic novelty. Finally, these phylogenetically and functionally diverse communities exhibiting close cell-cell interactions can serve to study the interplay of various metabolisms at various scales and infer how this influences community structure.

As characterizing the phylogenetic composition represents the first step to understand these complex microbial systems, we carried out microbial diversity surveys on four mats distributed along physicochemical gradients (increasing depth, salinity and temperature; decreasing oxygen) of a shallow (30-35 cm depth) pond (Fig. 1 and Supporting information Table S1). In this system, oxygen levels decreased from surface to bottom, the latter being completely anoxic. Salinity increased from marine-like levels in the top 10 cm to moderately halophilic conditions (4.3 to 6.6% salinity) at the bottom of the pond. While the pH remained relatively constant (7.16-7.66), temperature increased from 28°C on the pond surface to 52°C at the surface of the deepest mat (LLA9-D). Interestingly, the temperature measured on several points below LLA9-D reached only 30°C on average (data not shown), suggesting that the temperature increase might be due to the microbial activity of moderately thermophilic lineages.

Our study represents the first attempt to characterize the structure of complex communities at such small spatial scale (0-30 cm). In particular, this approach allows following change according to environmental parameters without the influence of geography. We found a strong similarity of community structures in both biological (Fig. 2 B and C) and technical (Fig. 2A) replicates, highlighting that our experimental design accurately accounted for local heterogeneity effects, a fundamental parameter rarely taken into account in the literature. Llamara microbial mats were highly diverse, with a number of OTUs comparable to what is regularly observed in soil or sediments, the most diverse environment types (Lozupone and Knight, 2007). These stratified communities are exposed to marked physicochemical gradients within mats as well as in the water column. These abiotic factors explain in large part differences in community composition. The availability of various electron donors and acceptors combined with varying local environmental conditions likely favours the occurrence of different energy metabolisms at as many microniches in these complex ecosystems.

A reservoir of novel prokaryotic clades

Llamara complex mat communities harboured a phylogenetically diverse collection of bacterial and archaeal OTUs (Fig. 4-5), many of which were very distantly related to their closest cultured representative. Consequently, we could only hypothesize potential metabolisms for most of them. The two mats situated in the oxic zone (LLA9-A and LLA9-B) were mostly composed of relatively well characterized bacterial phyla, including abundant members of Bacteroidetes, Cyanobacteria, Proteobacteria and Verrucomicrobia. Below the chemocline defining an oxic/anoxic and salinity transition (mats LLA9-C and LLA9-D), however, bacterial candidate divisions and archaea accounted for up to 75% of the sequences. The bacterial candidate divisions included a collection of widely distributed poorly known phyla (Solden et al., 2016) that could represent more than 15% of the bacterial domain diversity (Brown et al., 2015). Four of them (Aminicenantes, Latescibacteria, Parcubacteria and TA06) were particularly abundant in our mats. Aminicenantes are metabolically diverse anaerobes found in a wide range of environmental conditions (e.g. oxygen, salinity, temperature) in both marine and terrestrial habitats (Farag et al., 2014). Latescibacteria were particularly abundant in bottom layers of transition and anoxic zones (LLA9-C-3 and LLA9-D-2). They belong to the superphylum Fibrobacteres-Chlorobi-Bacteroidetes and are composed of fermentative anaerobes that could possibly be involved in EPS degradation (Youssef et al., 2015). Parcubacteria and TA06 are commonly found in anoxic environments and include fermentative organisms (Peura et al., 2012; Wrighton et al., 2012, 2014; Baker et al., 2015). One parcubacterial lineage has also been shown to be an ectosymbiont of algal endosymbionts in a ciliate (Gong et al., 2014). Consistently with studies showing the presence of Oscillatoriales able to perform anoxygenic photosynthesis in anoxic, saline, environments (Oren, 2015), we detected a substantial number of cyanobacterial sequences related to Oscillatoriales in mat D.

Despite recent progress (e.g. García-Maldonado et al., 2014; Petitjean et al., 2014; Spang et al., 2015; Solden et al., 2016), most archaeal phyla still remain phylogenetically and ecologically poorly described. In Llamara mats, Thermoplasmata were particularly abundant, notably through one OTU affiliated to the Marine Benthic Group D (Supporting information Fig. S9). This group includes thermophilic organisms potentially involved in protein remineralization (Lloyd et al., 2013). They are very common in the anoxic marine sub-sea floor, which seems consistent with the presence of most Thermoplasmata representatives below the chemocline, where the conditions were slightly thermophilic (Fig. 3A). Members of the former DHVEG-6 clade, recently renamed Pacearchaeota and Woesearchaeota (Castelle et al., 2015), were abundant in mat LLA9-C. Organisms from these groups were originally found in deep-sea hydrothermal vent systems (Takai and Horikoshi, 1999) but later detected in marine sediment, terrestrial soils and freshwater plankton as well (Teske and Sørensen, 2008; Ortiz-Alvarez and Casamayor, 2016). Although there is no cultured representative of these two lineages, recent in-depth genomic analyses showed that they have small genomes notably harboring genes involved in carbon and hydrogen metabolisms (Castelle et al., 2015). The OTUs affiliated to these groups were extremely diverse and found in low frequency, suggesting that these organisms could be highly specific parasites or symbionts. Finally, we detected below the chemocline the presence of abundant and unknown archaeal lineages (LMG-1 to 6) belonging to both Euryarchaeota and the TACK syperphylum (Fig. 4). Overall, though the community structure of the two deepest mats (LLA9-C and LLA9-D) was similar to previously described archaea-rich mat communities (Sørensen et al., 2005; Schneider et al., 2013), this study extends our current knowledge of the diversity of contemporary microbial mat communities to novel archaeal lineages and highlights the abundance and diversity of bacterial candidate divisions in these systems. Our extensive 16S rDNA-based survey therefore confirms that Llamara microbial mats are important reservoirs of prokaryotic diversity.

Eukaryotes in microbial mats

Halophilic and halotolerant microbial eukaryotes are found among a variety of taxa, including especially Stramenopiles (e.g. Bacillariophyta, Bicosoecida, Chrysophyceae), Alveolata (e.g. Ciliophora) and Fungi (Edgcomb and Bernhard, 2013). Unlike prokaryotes, eukaryotes are not generally suited to habitats characterized by combined anoxic, salty and relatively hot physicochemical environments. Accordingly, mining of eukaryotic rRNA sequences in metagenomic data produced from the same samples showed that they accounted for at most 3% of the total rRNA sequences in the top layers (LLA9-A-1, LLA9-B-1 and LLA9-C-1) and less than 1% in the lower layers, where the environmental conditions are harsher (data not shown). Yet, Llamara eukaryotes included taxa relatively rare in other environments that are potentially important for understanding eukaryotic evolution, as the relationships between many eukaryotic groups still remain uncertain (Katz, 2012). For instance, we detected sequences having high sequence similarities to the Microhelida order of Palpitomonas (mostly in LLA9-A-1 and LLA9-B-1 layers; Yabuki et al., 2012). This lineage corresponds to deeply-branching heterotrophic organisms that could possibly hold key information on the early evolution of photosynthetic eukaryotes (Yabuki et al., 2010). Palmophyllales is another interesting group usually found in low-light intensity marine habitats (Leliaert et al., 2011). They are likely located a few millimetres below the surface in Llamara mats, where the light intensity is limited compared to the surface. To our knowledge the present study represents the first report of Palmophyllales in saline mats. Finally, Fungi are diverse and common in hypersaline water columns and mats (Edgcomb and Bernhard, 2013) where they are potentially involved in EPS degradation (Cantrell and Duval-Pérez, 2013). They encompassed most of the eukaryotic community diversity below the chemocline in Llamara mats and are likely thermophilic, as the temperature reaches up to 52°C at the surface of LLA9-D.

Potential biotic interactions and future directions

Microbe-microbe interactions occur at small scale and are therefore much more difficult to identify than abiotic interactions, which can be assessed through the measure of a set of environmental variables and the use of multivariate statistical analyses. Correlation networks based on OTU presence/absence and abundance are emerging as a powerful tool to generate hypotheses on which interactions could be biologically relevant and should be further investigated (Weiss et al., 2016). This approach was all the more suited in the present case that most prokaryotic lineages that we were able to detect belonged to poorly characterized groups. While negative correlations (co-occurrence never observed) can be a way to identify metabolic competition, separated niches or any other relationship precluding co-occurrence, positive correlations (co-occurrence frequently or always observed) encompass a variety of relationships that include parasitism, predation and, particularly relevant to complex mat communities associated to redox gradients, metabolic syntrophy. At the level of high-rank taxa, network topology strikingly differed between the most superficial mat from the oxic zone (Fig. 6A), composed of two strongly connected modules, and the rest of mats (Fig. 6 B to D). Although the cause of these contrasting patterns remains unclear, it is tempting to speculate that this reflects different global trophic strategies. Cascading metabolic activities starting from oxygenic primary production and carried out mostly by members of relatively well-known, dominant taxa might occur in the upper mat. By contrast, more balanced and intricate bi- or multi-partite biotic, potentially syntrophic, associations might dominate in deeper mats, making it difficult to detect interactions between high-rank taxa at the abundance threshold that we applied in this work.

Interestingly, most correlations identified at the co-occurrence OTU-level network reconstruction were positive. In particular, we identified in the deeper layers a module of potentially interacting OTUs related to an annamox Planctomycetes and a thermophilic fermentative Deferribacteres (Caldithrix; Miroshnichenko et al., 2010), along OTUs belonging to LMG-5 archaea and Aminicenantes, a widely distributed group of largely unknown metabolic capacities (Farag et al., 2014) (Fig. 7). Aminicenantes is one of the bacterial candidate divisions harboring members with small genomes, suggesting that they are involved in symbiotic relationships (e.g. parasitism, syntrophy) with others organisms (Kantor et al., 2013; Rinke et al., 2013; Wrighton et al., 2014; Brown et al., 2015). In mat LLA9-C, another candidate division, TA06, was found to be positively correlated to Lokiarchaeota, suspected to be involved in the hydrogen cycle in anaerobic environments (Sousa et al., 2016) (Fig. 6C). This particular potential association deserves to be further explored given the importance of this, and possibly related, lineages in understanding early eukaryotic evolution (Spang et al., 2015). Finally, the network constructed for mat LLA9-D showed strong correlations between LMG-5 archaea and potential fermenters, including Hydrogenedentes (Nobu et al., 2015), Latescibacteria (Youssef et al., 2015) and Parcubacteria (Wrighton et al., 2014). Altogether, these findings highlight the ecological complexity of Llamara mats and suggest a central role for hydrogen-related metabolisms in these systems. Unravelling the type of biotic interactions (i.e. parasitism, predation and metabolic syntrophy), especially between poorly known but abundant lineages, identified here as positive correlations will be crucial to understand the functioning of these ecosystems.

Experimental Procedures

Sampling and measurement of physicochemical parameters

Mat samples were collected in March 2012 in a small pond of the salar de Llamara, at the North of the Atacama desert (21°16'7.37"S, 69°37'4.01"W). The pond was shallow (30-35 cm deep) yet stratified, with an oxic upper water layer of 15-20 cm above the anoxic, S-rich (data not shown) layer. Small invertebrates were observed swimming in the oxic part of the water column. Some confined pockets in the anoxic part of the pond were whitish, possibly due to the presence of planktonic anaerobic sulfide reducers (Grant and Bathmann, 1987). Mats were collected at four sites across one of the arms (ca. 3 m wide, 9 m long) of an irregular shallow pond, along both depth and physicochemical gradients (Fig. 1 and Supporting information Table S1). Physicochemical parameters (conductivity, oxygen and temperature) were measured on several points of the water column using a multi-parameter probe Hanna HI9828. Salinity values were calculated using the measured temperature and conductivity values (Aminot and Kérouel, 2004). Mats LLA9-A and LLA9-B were located in the oxic zone, whereas LLA9-C was in the oxic-anoxic transition zone and LLA9-D in the most anoxic, hot and saline area of the pond. Mat fragments of ca. 10 x 15 cm of surface, and up to 10 cm depth, were collected and their layers further subsampled as much as their consistency allowed it. We were able to separate LLA9-B and LLA9-C in, respectively, two (LLA9-B-1 and LLA9-B-2) and three (LLA9-C-1, LLA9-C-2, LLA9-C-3) distinct layers. The surface, colored layers of the highly gelatinous LLA9-D mat were not disposed horizontally, but formed a series of intricate peaks with hardened internal white areas that made it impossible to separate all observable layers, even though this was the thickest mat. We therefore sampled two main layers, LLA9-D-1, encompassing the coloured laminated region of the mat occupying the uppermost 4-6 cm, and LLA9-D-2, corresponding to the underlying, spongy biomass. Finally, mat LLA9-A was relatively thin and lacked consistency, such that it could not be subsampled (LLA9-A-1). For each layer, we collected, as replicates, two mat pieces located, horizontally, a few cm away to correct for potential local heterogeneity effects. In the following, replicate samples are indicated by a “b” following the sample name. Subsamples were immediately placed in 5 ml cryotubes and fixed with ethanol (≥ 70%) and RNA later (Ambion™, Fisher Thermo Scientific Inc.). They were stored frozen at -20°C until DNA extraction.

DNA extraction, amplification and sequencing

DNA was extracted using the Power Biofilm™ DNA Isolation Kit (MoBio, Carlsbad, CA, USA) according to manufacturer’s instructions. 18S rRNA gene fragments of approximately 550 bp, encompassing the V4 hypervariable region, were PCR-amplified using as primer set EK-565F and 18s-EUK-1134-R-UNonMet, biased against Metazoa (Bower et al., 2004). Prokaryotic 16S rRNA gene fragments of approximately 290 bp, encompassing also the V4 hypervariable region, were amplified using the primer set U515F/U806R. Bacterial 16S rRNA gene fragments were also amplified, in order to assess the accuracy of the universal prokaryotic primer set in catching the bacterial diversity. Bacterial rRNA gene fragments of approximately 470 bp, encompassing the V3 to V5 hypervariable regions, were thus amplified using the primers 341F and 805R. Forward and reverse primers were tagged with 20 different 10-bp molecular identifiers (MIDs) to allow pooling and later sorting of PCR amplification products from 20 distinct samples. Both primer and MID sequences are available in Supporting information Table S2. The 25-µl amplification reaction mixtures contained 0.5-3 µl of eluted DNA, 1.5 mM MgCl₂, 0.2 mM of deoxynucleotide (dNTP) mix, 0.3 µM of each primer and 0.5 U Taq DNA polymerase (Platinum Taq DNA Polymerase, Invitrogen, Carlsbad, CA, USA). The amplification conditions consisted of 35 cycles (94°C for 30 s, 55-58°C for 30-45 s, 72°C for 90 s) preceded by 2 min denaturation at 94°C, and followed by 5 min extension at 72°C. Amplicons from at least 5 independent PCR products for each sample were pooled together and then purified using the QIAquick PCR purification kit (Qiagen, Hilden, Germany), according to the manufacturer’s instructions. The same amounts of purified amplicons from 20 samples were pooled. Amplicons were pyrosequenced using the 454 GS FLX Titanium technology from Roche (Beckman Coulter Genomics, Danvers, MA, USA). In parallel, near-complete archaeal 16S rRNA gene sequences (~1,500 bp) were amplified from LLA9-D-1 and LLA9-D-2 using 21FQ/1492R as primer set (Supporting information Table S2). PCR reactions were performed under the following conditions: 35 cycles (94°C for 15 s, 55°C for 30 s, 72°C for 2 min) preceded by 2 min denaturation at 94°C, and followed by 7 min extension at 72°C. Clone libraries were constructed using the TopoTA cloning kit (Invitrogen, Carlsbad, CA, USA) according to the manufacturer’s instructions. Clone inserts were sequenced from both forward and reverse ends in order to retrieve the full sequence of the insert (Beckman Coulter Genomics, Takeley, United Kingdom). Pyrosequences have been deposited at NCBI under the BioProject number PRJNA357675.

Sequence analysis

We obtained 158064 , 173270 and 251063 raw pyrosequences for the prokaryotic, bacterial and eukaryotic primer sets, respectively. A series of filters were applied to each dataset, in order to retain only high-quality sequences, as previously described (Bachy et al., 2013; Simon et al., 2015). First, only pyrosequencing reads having exact forward and reverse primer sequences were kept. After the elimination of the primer sequences, the remaining sequences were analysed with AMPLICONNOISE (Quince et al., 2011) to further eliminate errors introduced during PCR reactions or 454 sequencing. Filtered reads were then clustered into operational taxonomic units (OTUs) with an identity cut-off of 97% (prokaryotes) or 98% (eukaryotes) over their entire length using CD-HIT (http://weizhongli-lab.org/cd-hit/; Li and Godzik, 2006). Unfortunately, the sequencing of the eukaryotic amplicons produced sequence ends of lower quality for LLA9-D-1 and LLA9-D-1b. The elimination of the low quality portion of these reads resulted in shorter sequences compared to the other samples. To avoid the risk of misplacing sequences from LLA9-D-1/1b in the wrong OTUs, we decided to build OTUs independently for these two samples. Singletons, i.e. sequences that did not cluster with any other sequence across all samples, were eliminated as a precaution. The most abundant sequence in each OTU was used as reference. OTU reference sequences were assigned to high-level taxonomic groups based on sequence similarity using BLAST and a reference database made of both 16S rDNAs (Silva SSU Ref NR 115; Quast et al., 2013) and 18S rDNAs (Silva SSU Ref NR115 + PR2 database; Guillou et al., 2013). A maximum e-value cut-off of 10e-5 was applied to affiliate sequences to particular taxa. The sequences present in all OTUs were then attributed to the different samples according to their MIDs. Finally, potential chimerical OTUs were eliminated by a stringent procedure in which OTUs including sequences from at least two different samples were treated separately from OTUs composed of sequences from only one sample. Prokaryotic OTUs (i.e. obtained from both prokaryotic and bacterial primer sets) present in several samples were retained if the coverage of their representative sequence with their best hit was ≥0.9, but their taxonomic affiliation was designated as “uncertain” when their identity was <0.85. Prokaryotic OTUs including sequences from only one sample were kept if they combined a coverage ≥0.9 and an identity ≥0.95 with their best hit. The rest of these OTUs was checked using UCHIME (with the RDP classifier training database v9 as reference, Edgar et al., 2011). They were validated if they were not detected as chimeras, with minimum coverage and identity values of 0.9. Eukaryotic OTUs present in several samples were kept if they combined coverage and identity values ≥0.7 and ≥0.8 with their best hit, respectively. However, their taxonomic affiliation was changed to “uncertain” in cases where the coverage with their best hit ranged from 0.7 to 0.8. OTUs present in only one sample were retained if they presented both coverage and identity values ≥0.9 with their closest BLAST hit. The other eukaryotic OTUs were then checked by UCHIME (using PR2 database as reference) and KEYDNATOOL (http://www.keydnatools.com/) and by comparing BLAST hits recovered from independent sequence fragments (sequences were split in two and three fragments). We retained only OTUs that were validated by the three different approaches and having, in addition, a coverage >0.7 and an identity ≥0.9 with their best hit. Finally, OTUs associated to chloroplasts and Metazoa were removed from our dataset for subsequent analyses. After filtering, we retained 125885 (prokaryotic primers), 114578 (bacterial primers) and 207951 (eukaryotic primers, including LLA9-D-1/1b) sequences (Supporting information Table S3).

Statistical analysis

All statistical analyses were conducted using the R SOFTWARE (http://cran.r-project.org; R Development Core Team, 2013). To assess overall differences between microbial community compositions, we constructed Non-Metric multi-Dimensional Scaling (NMDS) plots based on pair-wise Bray-Curtis distances (Bray and Curtis, 1957) between all samples on the basis of OTU frequencies. Because we used different primer sets, the NMDS analysis conducted on the bacterial community structure and based on amplicons from both bacterial and prokaryotic primer sets was done using the relative abundances of high-rank bacterial taxa. These analyses were done with the package “vegan” (version 2.0-10; Oksanen et al., 2014). A Wisconsin standardization was applied in the case of the eukaryotic OTUs to balance the influence of the most and least abundant OTUs. Diversity (Simpson, 1949), evenness (Pielou, 1966) and richness indexes were also determined using the R package “vegan”. Chao1 richness indices (Chao, 1984) were computed separately for each sample using raw OTU counts and the “estimateR” function. To test whether the prokaryotic communities correlated with environmental parameters in the different mats, we conducted Mantel tests on a matrix of Bray–Curtis dissimilarities based on prokaryotic OTU frequencies and a matrix of Euclidean distances based on physicochemical parameters (depth below water level, depth below mat surface, temperature, oxygen concentration and salinity), using the “vegan” package. Canonical Correspondence Analyses (CCA) including prokaryotic OTU frequencies and the physicochemical parameters mentioned above were performed using the “ade4” package (version 1.7-4). Venn diagrams showing the number of OTUs shared by, or exclusive to, the different samples (excluding the eukaryotic OTUs for LLA9-D-1/1b), were built using the “gplots” package (version 2.16.0; Warnes et al., 2015). Heatmaps were generated using the “pheatmap” package (version 1.0.7). Finally, the H statistic of the Kruskal-Wallis test (R function kruskal.test) was computed using the average frequencies of the 126 OTUs shared across all four mats, that were calculated for both oxic (LLA9-A-1, LLA9-B-1 and LLA9-C-1) and anoxic mat layers (LLA9-B-2, LLA9-C-2, LLA9-C-3, LLA9-D-1 and LLA9-D-2), to test whether their distribution in the top and bottom layers was statistically similar.

Phylogenetic reconstruction

A phylogenetic tree was built with the archaeal OTUs obtained from massive 454 sequencing, the Sanger sequences obtained from gene libraries and a broad taxonomic selection of reference sequences from the NCBI. Sequences were aligned using MAFFT with the E-INS-i algorithm, 200PAM/k=2 as scoring matrix and a gap penalty of 1.53 (Katoh and Standley, 2013). Conserved positions in the alignments were selected using BMGE with default parameters (Criscuolo and Gribaldo, 2010) and used to reconstruct a phylogenetic tree using FASTTREE 2.1.8 and the GTR + CAT model of nucleotide evolution (Price et al., 2010). Finally, the tree was visualized with the GraPhLAn software (Asnicar et al., 2015); for each OTU, its frequency among Archaea in Llamara mats was plotted in their corresponding outer circle.

Network reconstruction

Five co-occurrence networks were reconstructed using prokaryotic read abundances as a starting point and considering all sample replicates independently. One network was reconstructed for pervasive OTUs (present in at least 10 samples out of 16) in the four microbial mats. The sparsity of this matrix was low, 26.38%. For the remaining four networks, one per mat (all layers included), only OTUs with more than 5% of sequences were retained. These OTUs were then affiliated to high-rank taxonomic groups. Matrix sparsity was very low in all cases (0.0%, 11.9%, 0.0% and 5.8% for mats A, B, C and D, respectively). The five matrices were treated identically for network reconstruction, as follows. Co-occurrences of the reads were calculated with SparCC (Friedman and Alm, 2012) for each mat layer. Ten iterations were used to estimate the median correlation of each pair and the statistical significance of the correlations was calculated by bootstrapping with 500 iterations. Correlations were then sorted for statistical significance (p < 0.001) and R > 0.7 or R < -0.7. Networks were built and visualized using ad hoc scripts in R with the aid of the igraph package (http://igraph.org/).

Originality-Significance Statement.

Despite their interest as potential models of major Precambrian ecosystems, modern phototrophic microbial mats remain poorly studied and the biotic, notably metabolic, interactions among different members of these complex communities are fundamentally unknown. To learn more about how biotic and abiotic parameters structure this kind of communities, we applied metabarcoding approaches and co-occurrence networks to several layers of microbial mats located along strong redox gradients in a warm saline pond at the Atacama desert. Our study reveals a great variety of archaea and bacteria from novel, divergent groups, and potential interactions between some of these poorly known taxa.