The Gypsum Hill spring system (79° 25′ N; 90° 45′ W) was originally described by (Beschel, 1963; Pollard et al., 1999) and is discussed in detail by a series of subsequent studies (Pollard et al., 1999; Omelon et al., 2001, 2006; Andersen, 2002; Battler et al., 2013). Gypsum Hill is one of at least 6 identified perennial saline spring systems on Axel Heiberg Island. Gypsum Hill is located in Expedition Fiord approximately 3 km from the White and Thompson Glaciers on the NE side of Expedition river. The system is comprised of 30–40 small outlets concentrated over an area of approximately 3000 m² in a narrow band (∼300 m long × 30 m wide) that perennially discharge a combined average of 15–20 L s^-1 of anoxic [mean oxido-reduction potential (ORP) ∼325 mV] brines (7.5–8% salts) at the base of Expedition Diapiar (Gypsum Hill). Outlet morphology is highly variable and seasonally dynamic, ranging from well-defined pools, to patches of brine-saturated sediment. Individual outlets discharge microaerophilic (0.05–0.2 ppm dissolved O₂), hypersaline (8–9% salinity), and cold (-0.5–6.6°C) fluids at rates varying from <0.5–1.5 L s^-1. The spring waters are near neutral pH (6.9–7.2) characterized by high concentrations of dissolved ions (dominated by Na⁺, Cl^-, with lesser K⁺, Mg²⁺, Ca²⁺, SO₄^2-) and rich in both sulfate (2300–3724 mg^-1) and sulfide (25–100 p.p.m.; Andersen, 2002; Pollard, 2005; Perreault et al., 2007). While the temperature of individual outlets vary (–0.5–6.6°C) between one another, fluid temperature (3 or 4°C average) is relatively constant despite significant seasonal changes in air temperature (Pollard, 2005). The physiochemical parameters of the springs have been detailed by others (Pollard et al., 1999; Omelon et al., 2001), and remain largely unchanged from the original chemistry reported by Beschel (1963). It is interesting, as first noted by Beschel (1963), that if all ions are normalized to Na⁺, the spring fluids are a chemical match to seawater. Although the physical and chemical characteristics of the fluid discharge have been well characterized, the water source is poorly constrained. It has been hypothesized that the springs originate from subpermafrost salt aquifers and rise to the surface through the permafrost (Andersen, 2002); however, the source aquifer has not been defined and the subsurface hydrogeologic system is largely unknown. An isotopic study by Pollard et al. (1999) found that the Gypsum Hill spring water did not deviate from the established Local Meteoric Water Line suggesting aquifer recharge under climatic conditions similar to modern. While Pollard et al. (1999) had originally presented five possibilities the most favorable including recharge from surface water bodies through tailks, ancient ground water circulation, and deep circulation of glacially pressurized water, Andersen (2002) developed a combined flow and thermal model to demonstrate that the brine discharge may originate in Phantom Lake, a glacially dammed lake several kilometers away.

Field work was conducted over two years in July 2013 and July 2014. During the summer of 2013 mats forming in a local flood plain characterized by local patchy color variation over a centimeter scale were identified proximal to outflow channels originating from Gypsum Hill outlet 4 (Figure 2). The microbial mats observed were 1–2 mm thick and underlain by dark-black silty water saturated sediments. While surface microbial mat communities are observed on a yearly basis during the summer research season (July–August, personal observations), the extent, appearance (color), range, and location of the mats can change seasonally. Over the course of three field seasons (2013–2015), we observed some mats in different locations, of different sizes, color, and appearance throughout the spring flood plain while other mat areas surrounding the predominant springs (i.e., high continuous flow with little changes over a 15 year observation period) were relatively consistent. During winter the springs become snow covered and it is likely that the flood plains, while not directly visible, are significantly disturbed if not absent. These observations suggest a seasonality and transient nature to the mat communities and a very limited duration of activity during the high Arctic summer months.

Areas dominated with green filamentous streamers (green mat; GM) were observed to be spatially associated with areas of red-brown sediment mats (red mat; RM). Samples of both GM and RM were collected July 10, 2013 by using a sterile 15 ml tube to scoop mat material beneath the air–water interface. Care was taken not to disturb or include the underlying sediment. Samples were frozen in the field and stored at -20°C until DNA extraction. During July 2014, two replicate sediment samples were collected from Gypsum Hill outlet 4 (GH-4, sediment A and sediment B). Sediment samples were frozen in the field and stored at -20°C until DNA extraction. Thus, a total of four samples were selected for downstream analysis, GH4-sediment A and GH4-sediment B from Gypsum Hill outlet 4 (GH-4) and green mats and red mats from the surficial flood plain.

Total DNA was extracted from each sample using the RNA PowerSoil Total RNA Kit (MoBio, Carlsbad, CA, United States) and the RNA PowerSoil DNA Elution Accessory Kit (MoBio). Amplicons of partial 16S rRNA (bacteria) were produced for each sample using barcoded primers: forward 27Fmod (5′-AGRGTTTGATCMTGGCTCAG-3′) and reverse 519 (5′-TTACCGCGGCTGCTGGCAC-3′) targeting the V1–V3 hypervariable regions as in (McFrederick et al., 2013). High-throughput targeted-amplicon sequencing was performed by Mr. DNA (Molecular Research LP) (Shallowater, TX, United States) using the GS FLX+ Titanium platform (Roche, Branford, CT, United States) for the outflow channel samples (GM and RM). High-throughput targeted-amplicon sequencing was performed by Genome Quebec using the GS FLX+ Titanium platform for the sediment samples (GH-4).

Sequences were analyzed and classified using the suggested Mothur 454 protocol (Schloss et al., 2009). Briefly,.sff files were separated into.fasta and.qual files using ‘sff.info’ where flow = T. Sequence error was reduced as per Schloss et al. (2009) RM and GM sequences were trimmed using flowgrams with the ‘trim.flows’ command with the following parameters: order = A, pdiffs = 5, bdiffs = 2, maxhomop = 8, minflows = 450, maxflows = 450, while spring sediment samples were trimmed with the parameters: order = B, pdiffs = 5, bdiffs = 2, maxhomo = 8, minflows = 650, maxflows = 650. Sequence error was further reduced by ‘shhh.flows’. Sequences were reduced to only unique sequences using ‘unique.seqs’ and unique sequences that passed quality control for the sediment and mat samples were combined. Sequences were aligned to the Mothur-interpreted Silva database (version Silva.nr_v123) with the ‘align.seqs’ command. Aligned sequences were reduced to only the overlapping region using ‘screen.seqs’ and ‘filter.seqs’ (vertical = T, trump = .). Sequences were again reduced to only unique sequences using ‘unique.seqs’ then preclustered into OTUs consisting of two or fewer different base pairs using the ‘pre.cluster’ command (diffs = 2). Chimeras were removed using the uchime program within Mothur. A distance matrix was generated with ‘dist.seqs’ (cutoff = 0.15) and OTUs were formed based on this distance matrix using ‘cluster’. All sequences were classified using the Silva database mentioned above using ‘classify.seqs’.

A distance matrix generated of uncorrected pairwise distances between aligned sequences was constructed to compare the molecular diversity between samples. Distance matrices were generated for raw data and rarefied sample sets. The ‘count.groups’ command was used to determine that GH 4 spring GH4-sediment sample B had the lowest number of sequences, the ‘sub.sample’ was therefore used to rarefy all samples to only 4670 sequences. OTUs were classified using the ‘classify.otu’ followed by ‘phylotype’ and ‘make.shared’ so that a taxonomic comparison analysis could be completed. A representative sequence with the smallest maximum distance to the other sequences for each OTU was obtained and megaBLAST was used to find the closest uncultured and sequenced organism in the literature. Alpha diversity statistics were calculated on the OTU tables after rarefying all data sets to 4670 sequences.

Taxonomic stacked bar chart were rendered in R using the plot_bar function in phyloseq (McMurdie and Holmes, 2013). A heat map was constructed using the plot_heatmap function in phyloseq (McMurdie and Holmes, 2013) with method = NMDS and distance = bray.

Initial imaging of samples was conducted at the BioTron advanced imaging facility, Western University, London, Canada. Samples were lightly crushed using a sterile agate mortar and pestle and fragments transferred to carbon tape affixed to titanium TEM stub mounts and imaged at 40 Pa on a Hitachi 3400-N Variable Pressure Scanning Electron Microscope with an accelerating voltage of 10–15 kV at a working distance of 5–10 mm. High resolution imaging was conducted at the Western University Nanofabrication Facility. Samples were mechanically crushed using a mortar and pestle into ∼1 mm fragments under sterile conditions and fixed in 2% glutearalderhyde overnight at room temperature. Following fixation, samples were dehydrated in ethanol by subsequent 15-min immersions in increasing concentrations of ethanol (50, 75, and 100%) followed by a final submersion in 100% ethanol. The dehydrated samples were critical point dried with a Samdri-PVT-3B CP dryer (Tousimis Research Group) in the Biotron imaging facility to preserve cell structure. Samples were then manually transferred to carbon tape (M. E. Taylor Engineering part # DSCC-12) affixed to titanium TEM imaging stubs using a binocular microscope. Each sample was plasma coated with ∼10 nm of amorphous osmium using a Filgen OPC80T Osmium Plasma Coater with an OsO₄ source. High-resolution imaging was carried out under high vacuum on a LEO Zeiss 1540XB FIB/SEM under an accelerating voltage of 1 kV and a working distance of 3.8 mm.

Secondary electron scanning electron microscopy (SEM) was used to evaluate cell-substrate associations in the mat samples. Imaging of GM samples revealed the presence of completely intact and undamaged diatoms as well as a massive biofilm (Figures 6A,B) while specific cell-substrate associations were not observed. As a result of limited samples and the inability to fix immediately in the field, the cells comprising the biofilm are misshapen and desiccated resulting in a puckered, non-canonical morphology despite attempts to preserve cellular morphology using standing fixation and critical point dying protocols prior to electron microscopy. Even though cellular morphology is not adequately preserved for morphological identification, electron microscopy did reveal differences between the red and green mat samples. The dominant bacterial taxon, Thiomicrospira, identified through 16S rRNA targeted-amplicon sequencing in the green mat sample is not known to form biofilms (Niederberger et al., 2009) and does not have a known coccoid morphology. The observed biofilm could be formed by algal cells. Algae were not targeted in the by the bacteria-specific primer sets used in the targeted-amplicon sequencing, and would also account for the green color characterizing the GM samples.

Scanning electron microscopy of RM samples revealed intimate associations between cocci and rod-shaped bacteria and the mineral substrate through filament-like extensions of extracellular substance (Figure 6C). The dominant rod-shaped cells are consistent with the morphology of Marinobacter and Psychroflexus. The rough surface texture apparent on the cells imaging in RM is suggestive of mineral precipitation (Ronholm et al., 2014). Cell-coating secondary mineral precipitation can occur through either passive or active biomineralization processes. A cultivation study of a supraglacial spring system Borup Fiord pass on Ellesmere Island, Canada (Gleeson et al., 2011) isolated a psychrophilic, sulfide-oxidizng consortia dominated by Marinobacter sp. that precipitated S (0). Subsequent dilution-to-extinction experiments indicated that Marinobacter sp. was the dominant sulfide oxidizer. While the majority of Marinobactor species are known for their ability to degrade hydrocarbons, the Marinobacter sp isolated by Gleeson et al. (2011) is the first known species to oxidize sulfur. This is consistent with a study conducted by Perreault et al. (2008) that identified soxB, a thiosulphate oxidation gene, in Marinobacter sp. NP40. Microscopy showed variably mineralized cells. The mineral coatings of the red mat (RM) cells may have preserved cellular morphology without the need for fixation techniques. The striking morphological difference between the RM and GM communities supports the differences in the genetic diversity as revealed by 16S rRNA targeted-amplicon sequencing. The observation of specific cell-substrate associations in the RM samples suggest that substrate-cell interactions play an important role in the RM communities, and the observation of putative mineral coatings in the RM cells suggests a role for biomineralization processes and perhaps lithoautotrophy involved sulfide oxidation.

There are substantial differences between the mat and GH spring sediment communities for classified OTUs at >1% abundance level within the Gammaproteobacteria clade. The Gypsum Hill spring sediment samples are dominated by Thiotrichales (>90%), similar to previous studies, while the samples from the mats have a significant Alteromonadales component (Figure 5). The community composition varies between the visually distinct mats suggesting spatially integrated, yet, biologically distinct bacterial communities. This division may reflect patchy geochemical variation, niche partitioning, or dynamic colonization as discussed in the following sections.

The core bacterial community shared between all samples is consistent with those found in studies comparing the bacterial composition of streamers forming in snow-covered run-off channels during the winter months to the source sediment of Gypsum Hill (Niederberger et al., 2009). The similarity in community structure between the streamers and Gypsum Hill source spring suggests a common source and it can be concluded that the bacterial complement in the runoff channels is, at least in part, derived from the spring outlet. In contrast, the core bacterial community between the mat samples and the source spring reported here comprises less than 1% of the total richness and there was a significant difference in the diversity between the two mat samples and between the mat samples and the source GH4 sediment. Given the significant difference in environment between the GH spring and flood plain, this difference in community structure is not surprising, however, supports the hypothesis that the community within the GH spring is not sourced from surficial microbial communities and is the surficial expression of deeper, subsurface community. We suggest that the seasonal outflow forming the flood plain provides an ephemeral habitat for transient surface microorganisms to colonize.

Marinobacter and Psychroflexus, the two most abundant OTUs in the red mat, and second and third most abundant in the green mat, respectively (Figure 4 and Table 3), are not found in the GH4 spring sediment samples in this study, but have been identified previously from the Gypum Hill outlet (Perreault et al., 2008). In this study, the absence of Marinobacter and Psychroflexus in the GH sediments account for a large degree of the variation between the GH spring sediment samples and the mat samples. The absence of these phyla in the GH sediments in the present study may reflect the large degree of community variation. The top six OTUs, based on relative abundance, are summarized in Table 3. The dominant OTU in the green mat is 99% similar to Marinobacter psychrophilus at the 16S rRNA level, a nitrite reducer isolated from sea-ice in the Canadian basin (Zhang et al., 2008). The second most dominant, 95% similar to Psychroflexus sediminis, isolated from a hypersaline lake in China (Chen et al., 2009), is an obligate aerobic nitrate reducer. It is interesting to note that Psychroflexus sediminis produces non-diffusible orange carotenoid pigments (Chen et al., 2009) and this could explain the observable difference in coloration between the red and green mats. Thiomicrospira, the dominant taxon in the GH4 sediment samples, is an obligate chemolithoautotrophic sulfur oxidizer (Knittel et al., 2005). The dominance of Marinobacter as described here from the RM sample, has not been previously reported from the cold saline springs on AHI and suggests that in contrast to the outflow channels, the mat communities are not sourced directly from the spring. It would be pertinent to assess the microbial diversity in the active layer beneath the surficial mats to assess spatial variation in community structure with respect to changing physical and chemical conditions.

Here we show the presence of a microbial niche in summer surface microbial mats associated with flood plains likely fed by Gypsum Hill runoff channels. The microbial composition of these mats is unique with respect to other bacterial communities reported from other Arctic springs on AHI (e.g., Perreault et al., 2007, 2008; Niederberger et al., 2010; Lay et al., 2013; Lamarche-Gagnon et al., 2015). We suggest that seasonal shift produces ephemeral surface habitats that are not populated by through spring runoff channels. The fact that the surface is only habitable in the summer months suggests seasonality to the activity of these communities.

While results from this study are hint at a surficial community active only during the summer months, future metagenomics and metatranscriptomic analyses would lead to a better understanding of the metabolic potential and active members inhabiting these mat communities. A deeper assessment of the mat cyanobacterial diversity would allow for a more thorough comparison to other mat communities and replicate analyses are required to better understand seasonal variability and stability.

At present the surface conditions on Mars are inconsistent with the formation of standing bodies of water. The subsurface represents the most temporally extensive habitable and potentially inhabited environment on Mars. Water ice is present in the near Martian subsurface (10 s of cm; Bandfield, 2007) with polar layer deposits upwards of 1 km thick (Picardi et al., 2005). Recent radar observations have demonstrated the existence of large volumes of ground ice in the near surface environment (Stuurman et al., 2016). Modern subsurface water may be present in pockets due to radiogenic heating and lithostatic pressure and the presence of brines depressing the freezing point (Clifford et al., 2010); droplets of salts on the Phoenix lander legs demonstrate the thermodynamic stability of brines on Mars (Rennó et al., 2009). Both recurring slope lineae and (Ojha et al., 2015) Martian slope streaks (Bhardwaj et al., 2017) have been hypothesized as indicators of trainset water activity on the modern Martian surface. If these features represent surface expressions of subsurface briny aquifers (e.g., Stillman et al., 2016), transient liquid water on the surface could potentially transport subsurface microbial communities, or evidence of these communist to the surface. Alternatively, RSL features may be formed in the absence of flowing liquid water through phenomena such as deliquescence (Heinz et al., 2016). The presence of salts depresses the freezing point of water raising the depth at which liquid water could theoretically be stable. However, the presence of salt can also lower the water activity below that required for life (Tosca et al., 2008). The absence of sulfates and chlorates associated with clay-rich terrains on Mars suggests that liquid water was not always characterized by extreme salinity indicating potentially more habitable conditions in the past.

On Axel Heiberg Island, the perennial cold spring such as Gypsum Hill represent the surface expression of subsurface briny aquifers flowing through thick permafrost. Our results here suggest that the communities within these springs are distinct from the transient surficial communities that form in flood plains during the summer months. Detailed sampling over the course of several consecutive seasons is required to better understand the variability in the surface communities and their persistence through the winter. Understanding the geomicrobiology and microbial activity in hypersaline cold springs like Gypsum Hill shed insight into the potential habitability of putative subsurface Martian brines. The development of the mat communities demonstrates the potential for ephemeral environments linked to seasonal melt events, and/or flooding to host diverse and unique microbial communities.