Spotted Lake (Figure 1A, Figure S1) is located in Osoyoos, British Columbia, Canada (49°4′40.86″ N, 119°34′3.01″ W) within Carboniferous to Permian green schist facies rocks, along with dolomites, quartzites, marbles and localized deposits of pyrite and pyrrhotite (Jenkins, 1918). Oxidation of these iron sulfides results in the generation of sulfuric acid and the subsequent weathering of the basin dolomites, yielding high levels of Mg²⁺ and SO42- that concentrate in the endorheic lake. As a result, Spotted Lake is rich in magnesium and sodium sulfate salts, and with a slightly alkaline pH. Due to low levels of precipitation in this region, summer evaporation leads to the formation of individual brine pools (Figure 1B), which are separated by mud mounds (Figure 1C) and surficial salt crusts (Jenkins, 1918; Cannon et al., 2012). Samples from Spotted Lake were collected in October 2010 (Table 1); in total, 4 individual ponds were surveyed and samples including water and sediment (top 5–10 cm) were aseptically collected in sterile containers. Water samples were collected first (without disturbing the sediment) in autoclaved plastic sterilization units with samples immediately sealed, and subsequently analyzed for pH and ion concentrations following Wilson et al. (2012). Water activity was measured in triplicate from two brine pools in the laboratory using an AquaLab Dew point activity meter 4TE, and on two sediment samples, at a temperature of 25°C. Sediments (45–100 g) from each pond were collected in 50 mL plastic conical tubes and immediately after collection were transferred to glass test tubes, which were sealed with rubber stoppers, purged with nitrogen, and sealed with a crimped metal band before being frozen at −20°C. All tubes were placed in a cooler with freezer packs for shipment, and stored at −80°C upon arrival. Soil samples from each pond were sent for inductively coupled plasma mass spectrometry (ICP MS) analysis, Bureau Veritas, Canada.

For in situ imaging of the environment, soil samples were fixed in glutaraldehyde, dehydrated in ethanol and critical-point dried to preserve cell structure using a Tousimis Auto Samdri 815 Series A Critical Point Dryer following Dykstra and Reuss (2003). Samples were then mounted, carbon coated and imaged using a Zeiss Merlin High-resolution Scanning Electron Microscope (SEM) at 1 kV. Soil samples were also imaged to assess viability: soil was stained using Live/Dead Baclight Bacterial Viability Kit (Life Technologies; now Thermo-Fisher Scientific, Waltham, MA) and then imaged on a Zeiss ApoTome 2.

DNA extraction was performed utilizing both a high-input process (MoBio) and a low-input process (Zymo) to explore differences in acquired sequencing data due to extraction protocols (Figure S3). (1) High input method: Genomic DNA (gDNA) was extracted using the MoBio Powersoil DNA isolation kit (Carlsbad, CA), and concentrated with Zymo Research Genomic DNA Clean and Concentrate. Gel electrophoresis and a NanoDrop Spectrophotometer (Thermo Scientific) were used to assess gDNA quality and concentration, respectively. (2) Low input method: gDNA extraction was performed (0.25 g from each of the four samples) using Zymo Research Soil Microbe DNA MicroPrep; eluted gDNA was further subjected to whole genome amplification using phi-29 (GE Healthcare Illustra Ready-To-Go GenomiPhi V3) to produce enough DNA for library construction.

The Ion Torrent PGM system (Rothberg et al., 2011), Ion Xpress Fragment Library and Ion Xpress Template kits were purchased from Ion Torrent Systems (Guilford, CT). Sequencing, library construction and template preparation were performed according to the 200 bp Ion Xpress Fragment Library and Template Preparation protocols, and libraries S1–S4 (high input), and Z1–Z4 (low input) for sediment samples 1–4 were constructed. A single sequencing run was then performed (Table S1) on a 316 chip using 500 flows (equivalent to 125 cycles).

Raw sequencing reads were analyzed with Phylosift v1.0.1 (Darling et al., 2014) using default parameters, which included quality trimming of FASTQ data. Microbial community structure was assessed directly from sequencing reads through comparison to reference sequences (37 near-universal single-copy genes, 16S and 18S ribosomal genes, mitochondrial genes, eukaryotic-specific genes, and hundreds of virus-specific genes). Raw reads were also submitted to MG-RAST to confirm Phylosift results, and also for further 16S rRNA phylogenetic and protein functional analyses (Meyer et al., 2008; Glass et al., 2010). For MG-RAST, the default quality control options for quality trimming, dereplication, and screening for common contaminants were used.

In order to compare Spotted Lake metagenomes (Table S2) with previously-analyzed metagenomes, keyword searches for salt-associated metagenomes in addition to a few other sets such as air (in order to include data representing exogenous environmental seeding) were conducted. The resulting list was filtered to exclude virus-specific datasets, contigs/assemblies, and datasets associated with a specific organism, giving a final list of MG-RAST metagenomes for comparison (Table S3). For each of these metagenomes a lowest common ancestor (LCA) analysis was performed at the species level using MG-RAST with the default settings (max e-value 10⁻⁵, min identity cutoff 60%, min alignment length cutoff 15 bp). An abundance matrix was constructed with one row per metagenome and one column for each unique taxonomic key across all metagenomes. The unique set of taxonomic keys was generated at each taxonomic level and a separate PCA analysis was done for each taxonomic level from domain to species (see Supplementary Material, pg. 2).

Water pH ranged from 8.0 to 8.3, and a_w was 0.98 for the water column, and ranged from 0.96 to 0.99 within the sediment. Water chemistry measurements revealed brine compositions consisting of SO₄ (2.8 M), Mg (2.1 M), and Na (1.9 M), with minor contributions from K and Cl (Table 1), nearly identical to the 1933 historical measurements (McKay, 1935). Total salinity was measured at 37.1% with an approximate molar ratio of MgSO₄:Na₂SO₄ of 20:9 consistent with previous identification (Cannon et al., 2012) of minerals including epsomite (MgSO₄•7H₂O) and mixed Mg-Na salts in various hydration states (Figure 2A), e.g., blöedite, konyaite. ICP-MS (Table 2) revealed very low silica amounts, <8%, and CaO levels ranging from 17 to 25%. The percentage total for the major oxides ranged from 62 to 66%, which was representative of the fact that a large component of the sediment was comprised of salt and not detectable using the methodology employed. When subtracted from the ideal total of 100%, the remaining ~33−37% was inferred to be sulfate salt, corresponding with the sulfate concentrations of the water column. Total organic carbon was low, ranging from 1 to 3%.

Spotted Lake was found to be host to a diverse set of organisms, from macroscopic brine shrimp (Figure 2B) to a wide range of bacteria (Figure 2C). Imaging of the community through Live/Dead staining revealed a viable population (Figure 2D). The abundance assessment of the S1–S4 and Z1–Z4 libraries (representing “high input” vs. “low input” modalities) pooled, revealed respectively 90 and 88% Bacteria, 10 and 4% Eukarya, and 3 and 3% Archaea (Figure 3). The S1–S4 and Z1–Z4 libraries varied from each other (Figures 4A–D), especially at the sub-domain level. Within the high input libraries, dominant bacterial phyla included Bacteroidetes (33.2%), Proteobacteria (21.4%), and Firmicutes (14.1%). Alternatively, the low-input Z1–Z4 library had a dominant contribution from the Firmicutes (21.3%) only. The highest abundance proteobacterial classes in S1–S4 were δ/ε-proteobacteria (11.2%), γ-proteobacteria (6.1%), and α-proteobacteria (3.2%). While highly diverse at the family to species level, halophiles were well represented, including the genus Halomonas, previously found in hypersaline (NaCl) environments, representing 3% of γ-proteobacterial sequences, and the family Rhodobacteraceae (9%), members of which are common in seawater, where they play a key role in marine carbon cycling (Pujalte et al., 2014). Also present were sulfate reducers of the family Desulfobacteraceae (46%), largely comprised of organisms most closely related to the genus Desulfotignum, an obligate anaerobe capable of both chemoorganotrophy and chemolithotrophy (Kuever et al., 2001). Sequences belonging to the classes of bacteria lacking a cell wall were also represented: Mollicutes (8%) and Haloplasmatales (1%), the former of which is typically a parasite of eukaryotes (Skennerton et al., 2016), and the latter which is found only in hypersaline environments (Antunes et al., 2008).

Our results underscore the potential for bias in community characterization between “low-input” and “high-input” methodologies, with implications for future Martian expeditions. Not all low-input samples yielded evidence of Archaeal sequences (Figure 4A), but consistently demonstrated the presence of β-proteobacteria, which were largely absent in the “high-input” libraries (Figure 4D). Utilizing small sample sizes can result in an over- or underrepresentation of species due to the location of the subsample within the larger context, especially where physicochemical boundaries are present, thus care must be taken to extract multiple sub-samples that adequately represent the larger sample of interest. Beyond the issues noted with small sample sizes, whole genome amplification with phi-29 polymerase does result in a preference toward A+T rich genomes (Yilmaz et al., 2010), which may account for the dominance of the Firmicutes in these samples.

Analyses of alpha diversity were abundance weighted and calculated using the Shannon Diversity Index (H'). H' ranged from 599 to 704, with an average of 666 for high-input samples, and 680 to 1,122, with an average of 802, for low-input samples. To compare species abundance across all metagenomes (Figure 5A), a lowest common ancestor (LCA) analysis was performed, using classification down to the species level. This yielded hits in 17,433 unique taxonomic categories, though half of the abundance was captured by only 16 taxonomic categories (Figure 5B). Assessment of the metagenomic datasets using these highly abundant taxonomic categories revealed some similarities between Spotted Lake and both hypersaline and Antarctic environments (Figure 5C).

Principal components analysis (PCA) on the LCA abundance data (Figure 6A) revealed a similar pattern of explanatory power at each level of taxonomic depth: the first three principal components (PCs) explained 58–62% of the taxonomic variation from class to species. Thus, we focused on the species level analysis (Figures 6C–E), which provides the most specific taxonomic classification at a cost of slightly reduced explained variance for higher-order PCs.

Abundance analysis revealed that the hypersaline environments studied could be distinguished by three main taxonomic signatures: (archaeal) Halobacteria (associated with the first principal component, or PC1), unclassified bacterial sequences (PC2), and Cyanobacteria (PC3). These three main directions of variation represent the observed combinations of taxonomic abundances: For example, metagenomes from Chula Vista water samples have moderate to high PC1 scores and are dominated by Halobacteriaceae (Figure 5C). Occupying another niche are microbial mat samples from Guerrero Negro, which have low PC1 and range from low to high PC3 scores that are inversely related to PC2, so that high levels of Microcoleus are associated with low levels of uncharacterized bacteria, and vice versa (Figure 5D); see Table S3 for metagenomic datasets.

To describe the strength between these taxonomic signatures and each PC, a figure of merit (FoM) was defined, which is a score ranging from 0 to 1, where 0 indicates no association with a PC score, and 1 indicates perfect correlation (a PC is uniquely associated with variation of the specified taxonomic signature). Higher-order PCs are similarly associated with specific taxonomic signatures (Figure 6E) with high figures of merit. These associations corroborate the patterns of abundance visible in Figure 5C. A similar analysis was also carried out using SEED Subsystems database (Overbeek et al., 2005) for functional analysis of protein sequences (Figure 6B). Functional analysis (Figures 6F,G) revealed a tighter grouping of hypersaline environments than found based on abundance (Figures 6C,D). Functional variation was associated with phages (PC1), protein biosynthesis (PC2), and clustering-based subsystems (PC3). Metagenomes derived from Ace Lake, Antarctica, clustered in a similar fashion to Spotted Lake along PC2 and PC3 (Figure 6G), in part due to high abundance of clustering-based subsystems, a label given to genes with unknown function but presumed functional coupling (appearance together within multiple genomes, such as within an operon). The function-based PCs were less associated with specific functions than in the abundance analysis (Figure 6H; figure of merit 0.44 ± 0.18 vs. 0.69 ± 0.12 for abundance analysis, mean ± s.d.).

Hypersaline environments have been shown to preserve biological material on timescales exceeding that of non-saline systems, thus, the presence of paleo-sulfate lakes on Mars is of particular interest for exobiology, as these deposits may retain evidence of previous microbial habitation on the planet. The biological characterization of Spotted Lake is important in order to further delineate the limits to microbial growth in hypersaline environments: do such conditions on present-day or ancient Mars represent potentially habitable, versus merely organic-preserving, environments? On Mars today, climate is mainly regulated by its obliquity (Mischna et al., 2013), which in recent epochs has varied with a 124,000-year cycle. This cycle alternatively stabilizes and destabilizes ground ice in non-polar regions of Mars, which could alter the probability or frequency of near-surface liquid water brines. While MgSO₄ has minimal (−3.6°C) freezing point depression, it can undergo supersaturation allowing for an additional (4–6°C) cooling below this eutectic without precipitation, and at some concentrations can be cooled up to 15°C below the eutectic before becoming solid (Toner et al., 2014). Freezing and salt precipitation events as well as their associated latent heat may help buffer these deposits against low Martian temperatures and diurnal fluctuations. It is plausible that salt precipitation and drying-out events could also result in the entrapment of high molarity fluids that still may maintain transient water activity levels high enough to support microbial activity. Any organisms in such brines would have to cope with osmotic stress, desiccation, and freeze-thaw stresses, as well as cosmic irradiation if within 1–2 m of the surface.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.