Lithogenic hydrogen supports microbial primary production in subglacial and proglacial environments.

Life in environments devoid of photosynthesis, such as on early Earth or in contemporary dark subsurface ecosystems, is supported by chemical energy. How, when, and where chemical nutrients released from the geosphere fuel chemosynthetic biospheres is fundamental to understanding the distribution and diversity of life, both today and in the geologic past. Hydrogen (H2) is a potent reductant that can be generated when water interacts with reactive components of mineral surfaces such as silicate radicals and ferrous iron. Such reactive mineral surfaces are continually generated by physical comminution of bedrock by glaciers. Here, we show that dissolved H2 concentrations in meltwaters from an iron and silicate mineral-rich basaltic glacial catchment were an order of magnitude higher than those from a carbonate-dominated catchment. Consistent with higher H2 abundance, sediment microbial communities from the basaltic catchment exhibited significantly shorter lag times and faster rates of net H2 oxidation and dark carbon dioxide (CO2) fixation than those from the carbonate catchment, indicating adaptation to use H2 as a reductant in basaltic catchments. An enrichment culture of basaltic sediments provided with H2, CO2, and ferric iron produced a chemolithoautotrophic population related to Rhodoferax ferrireducens with a metabolism previously thought to be restricted to (hyper)thermophiles and acidophiles. These findings point to the importance of physical and chemical weathering processes in generating nutrients that support chemosynthetic primary production. Furthermore, they show that differences in bedrock mineral composition can influence the supplies of nutrients like H2 and, in turn, the diversity, abundance, and activity of microbial inhabitants.

Subglacial and proglacial environments host abundant, diverse, and active microbiomes supported by combinations of nutrients derived from relict organic matter, minerals, and the interaction of water and minerals (1–5). In catchments containing abundant sedimentary carbonates and shales, the biological oxidation of pyrite (FeS2) has been suggested to be a primary control on the composition of subglacial sediment microbiomes (5–7), likely due to its potential role in supplying reductant to drive primary production (8). Far less is known of the nutrients that support communities inhabiting subglacial and proglacial systems overlaying silicate (e.g., granite, gneiss, basalt) bedrock. This is a major knowledge gap considering that silicate-based glaciers are prevalent habitat types on Earth both today and in the geological past (2, 9) and potentially on other icy planetary bodies such as Saturn’s moon Enceladus (10).

Hydrogen (H2) is a common metabolic substrate in many rock-hosted ecosystems (11–14), including subglacial environments (15, 16). H2 can be generated abiotically through the reduction of water by ferrous iron minerals (17, 18), the radiolysis of water (13), or the activity of one of several classes of microbial enzymes (19, 20). In addition, H2 can be produced through silica radical-based mechanisms (21), which have recently been suggested to be of importance in generating H2 in cold (∼0 °C), dark subglacial ecosystems (15, 16, 22). In this process, mechanical shearing of silicates is proposed to generate surface radicals that, when exposed to water, produce H2 through reactions 1 and 2:Moreover, this mechanism may help to explain why abundant members of microbial communities in various subglacial ecosystems appear to be dependent on H2 as an electron donor in the absence of evidence for other known biotic or abiotic pathways for H2 production (3, 15, 23–26).

H2 may be expected to be an even more important electron donor supporting primary production in mafic (basaltic) subglacial and proglacial systems when compared to those underlain by felsic (granitic) or carbonate bedrock. In addition to containing abundant silicates, basalts are typically enriched in iron [8–13% by weight, (27)]. The iron in basalts has been linked to production of H2 from water in subsurface ecosystems in the Columbia River (CR) basin, Washington, United States (18). For example, in laboratory experiments containing anoxic water buffered at pH 6.0, H2 production from crushed CR basalt was 4.1 and 5.5 nmol⋅g−1⋅d−1 at 30 and 60 °C, respectively. When water–rock reactions were buffered at pH 7.0, rates decreased to 2.4 and 2.9 nmol H2⋅g−1⋅d−1 at 30 and 60 °C, respectively. Assuming Arrhenius kinetics, an extrapolation of these rates to temperatures relevant to glacial meltwaters (∼0 °C) yields anticipated H2 production rates of 2.8 and 2.0 nmol⋅g−1⋅d−1 at pH 6.0 and 7.0, respectively. These rates are similar to those (∼5 nmol H2⋅g−1⋅d−1) obtained in crushing experiments using a variety of silicate minerals incubated at 0 °C (15, 16). Thus, glaciers overriding basaltic bedrock might be expected to generate H2 via either the silica radical- or iron-based mechanisms—or both—which may in turn support more abundant and active hydrogenotrophic communities than those in carbonate-hosted subglacial systems.

Here, we examine the role of H2 in supporting microbial communities in proglacial and subglacial sediments collected from contrasting basaltic and carbonate/shale catchments: Kötlujökull, Iceland (KJ), and Robertson Glacier, Alberta, Canada (RG), respectively. These sediments, collected beneath or near the terminus of each glacier, represent recently exposed subglacial material generated by glacial comminution. Furthermore, the microbial communities they host are likely seeded by and highly reflective of those existing in the subglacial environments of their host catchments, as has been reported previously (4, 6). We hypothesized that H2 would be elevated in meltwaters and/or sediment porewaters from the former when compared to the latter. In turn, we hypothesized that the abundance and activity of hydrogenotrophic autotrophs would be higher in the basaltic system due to selection for such organisms during the assembly of those communities. Our results are discussed in consideration of the role of bedrock type in supporting H2-dependent primary production in diverse subglacial and proglacial sediment habitats in contemporary and past environments on Earth and in subglacial habitats on extraterrestrial icy planetary bodies such as Enceladus.

Materials and Methods

Field Site Descriptions.

KJ (63°36′N, 18°52′W) drains the southeast flank of Mýrdalsjökull, a temperate ice cap in southern Iceland atop Katla volcano, and is the largest outlet on its east side. As of 1997, the glacier ranged in elevation from 1,200 to 220 m over a length of 15 km, with its terminal ice front stretching nearly 12 km on a flat outwash plain known as Mýrdalssandur (28). The bedrock beneath KJ consists primarily of Upper Pleistocene-aged hyaloclastites, pillow lava, and associated basaltic sediments (29). The iron and silica content of the basaltic and rhyolitic rock in the Katla volcanic system range from ∼4 to 17% and 46 to 74% by weight, respectively (30).

RG (50°44′N, 115°20′W) drains the northern flank of the Haig Icefield in Peter Lougheed Provincial Park, Kananaskis Country, Alberta, Canada. The bedrock in this catchment is of Upper Devonian age and consists of impure limestones, dolostones, and dolomitic limestones with interbeds of shale, siltstone, and sandstone (31, 32). Sediments from RG consist of 25 to 56% carbonate with the balance potassium feldspar (20 to 49%), quartz (18 to 25%), muscovite (0.5 to 3%), and pyrite (0 to 1%) (33). As of 2011, the glacier spanned an elevational gradient of 2,900 to 2,370 m with a length of ∼2 km, terminating on a flat till plain.

Sample Collection and Field-Based Analyses.

At KJ, dangerously high water levels on 11 October 2016 dictated that water and sediments be collected from the Múlakvisl meltwater stream at a proglacial location (63°32.099′N, 18°50.421′W) ∼1 km downstream from the glacier terminus. Saturated sediments were collected from the stream bank using a flame-sterilized stainless-steel spoon and were placed into sterile 500-mL jars. Sediments were frozen on site with dry ice and maintained at −20 °C during transport to and storage at Montana State University. Saturated sediments from RG were collected from an ice cave at the terminus of the glacier on 14 September 2009, as reported previously (8), and are subglacial in nature given the sampling location beneath the glacier. At both sample locations, the surface layer (1 cm) of sediments was removed before samples were collected and preserved for analyses.

Electrical conductivity (EC) and pH were measured using a SevenGo Duo pro pH/Ion/Conductivity probe (Mettler Toledo). Meltwaters were collected with a syringe and were immediately transferred to clean glass serum vials, stoppered and crimp sealed, and then subjected to dissolved oxygen and alkalinity titrations within 5 h of sampling. Samples were allowed to equilibrate to ambient temperature in sealed vials prior to taking measurements. Dissolved oxygen was measured using a Mettler Toledo OptiOx probe. Total alkalinity was determined using a Hach digital titrator to titrate samples with 1.6 N sulfuric acid to an endpoint pH of 4.5. The pH was monitored using a SevenGo Duo pro pH/Ion/Cond probe.

Total sulfide was determined using the methylene blue reduction method (34). KJ meltwater was filtered through a 0.2-μm nylon membrane button filter directly into assay reagents in the field. Assay tubes were then wrapped in aluminum foil and subjected to absorbance analyses at the laboratory within 1 wk of collection using a Genesys 10S Vis spectrophotometer (Thermo Scientific) and a wavelength of 670 nm. Samples for determining concentrations of dissolved gases other than sulfide (H2, CH4, and CO2) were collected using a modified bubble-strip method, as described previously (35). Briefly, a 60-mL Luer-lock syringe was filled with 20 mL of N2 gas. Twenty milliliters of KJ meltwater were then introduced into the syringe followed by vigorous shaking for 1 min to allow for gas equilibration. The gas bubbles were then transferred to glass vials containing a saturated sodium chloride solution for storage until analysis by gas chromatography, as previously described (36).

Total Organic Carbon and Total Nitrogen Determinations.

KJ and RG sediment samples were dried overnight at 80 °C, then hand milled in an acid-washed (overnight in 10% HNO3) porcelain mortar and pestle and sieved to 63 μm. Triplicate subsamples were acidified with 2 N HCl in silver foil cups until acid addition did not induce bubbling during a 2-h incubation. Acidified subsamples were then washed twice with diH2O and dried for 24 h at 80 °C after each wash. Triplicate acidified (Ag foil-wrapped) and triplicate unacidified samples were then wrapped in tin foil and analyzed on a Costech ECS 4010 (Costech Analytical Technologies) equipped for carbon and nitrogen analyses using a combustion temperature of 980 °C, a reducing oven temperature of 650 °C, a gas chromatography (GC) oven temperature of 65 °C, and a 4-m length GC column. External acetanilide standards were used to create C and N calibration curves. Total organic carbon (TOC) was determined as the difference between total carbon in unacidified and acidified samples.

Most Probable Number Assay Preparation.

Most probable number (MPN) assays were conducted to determine the abundance of cultivatable cells associated with RG and KJ glacial sediments capable of coupling H2 oxidation to autotrophic growth. Clean, acid-washed (overnight with 5% HNO3) 24-mL serum vials were filled with 9 mL of M9 minimal medium (37) with the following modifications: The concentration of MgSO4 was reduced to 200 µM and carbon sources were omitted. Medium was deoxygenated by purging with 0.2-μm nylon filter-sterilized nitrogen (N2) gas passed over heated (210 °C) and H2-reduced copper shavings. The headspace of vials was flushed with H2 for 5 min and then CO2 was added to achieve a final headspace mixing ratio of 80% H2 and 20% CO2. Oxidants, including Na2SO4, the ferric iron-containing minerals hematite and ferrihydrite, and NaNO3 were prepared in sterile anoxic M9 salts medium and added to a final concentration of 1 mM. Hematite particles (<5 µm) were purchased from Sigma-Aldrich. Ferrihydrite was synthesized as previously described (38). For aerobic MPNs, 16 mL (at ambient pressure) filter-sterilized air was added to each vial via syringe. Anoxic and filter-sterilized Wolfe’s vitamins and SL-10 trace elements solutions (37) were added to each bottle to final dilutions of 1 mL⋅L−1.

Glacial sediments were thawed overnight at 4 °C and 5 g of thawed sediment (wet weight) were added to 27 mL of autoclaved and cooled M9 salts to generate a ∼1:10 (vol/vol) sediment slurry. Vials were sealed and deoxygenated as described above and were kept on ice during all manipulations. Serial 1:10 dilutions of this slurry (initial dilution factor, 101) were made in sterile, anoxic M9 medium out to a sediment dilution factor of 107. Finally, MPN series were inoculated in triplicate by transfer of 1 mL of the appropriate sediment dilution to each experimental vial. Three abiotic control vials were prepared for each condition. These were autoclaved for 30 min at 121 °C, incubated for 24 h at 4 °C to allow for germination of spores, and reautoclaved. All vials were incubated at 4 °C.

MPN Assessment Using DNA Evidence.

Following 6 mo of incubation, MPN vials were qualitatively judged to be positive or negative for growth by measuring the amount of extractable DNA in each microcosm. Two milliliters of culture were removed from each MPN assay microcosm and were concentrated by centrifugation (14,000 × g, 15 min, 4 °C). Pellets were overlain with 400 μL of CTAB buffer and 200 μL of a 500 mg⋅L−1 molecular-grade skim milk solution (Hardy Diagnostics) and DNA was then extracted following the CTAB S protocol (39). DNA was resuspended in 50 µL of molecular-grade H2O and the concentration was determined with the Qubit dsDNA HS Assay kit and fluorometer (Molecular Probes).

Microcosms were determined to be positive for growth if >2 ng of DNA was recovered from a 2-mL culture sample. This amount of DNA was empirically determined to be associated with the growth of cells in MPN vials based on recovery of ≤1 ng⋅mL−1 in autoclave killed controls. Microcosms were assayed until all three microcosms in two sequential serial dilutions had been judged negative. Then the pattern of positive and negative growth across three serial dilutions, along with the sediment dilution factors associated with those microcosms, was used to estimate the MPN of cells capable of growth under the provided enrichment conditions present in each gram of proglacial or subglacial sediment inoculum. Selection of three appropriate serial dilutions, proper accounting for sediment dilution factor, and the tables used in estimating MPN are described in the US Food and Drug Administration’s Bacteriological Analytical Manual (40).

MPN Assessment Using Metabolic Evidence.

Metabolic activity assays were used in parallel with DNA extractions to generate MPN estimates correlated with the use of experimentally supplied oxidants. In microcosms with CO2 only, acetate and methane were measured. Sulfide was assayed in microcosms amended with sulfate. Ferrous iron was measured in hematite- and ferrihydrite-amended microcosms and nitrite was assayed in microcosms amended with nitrate. Microcosms were judged to be positive if the concentration of the assayed metabolite was 3 standard deviations (SDs) higher than its average concentration in three killed controls. MPN was then estimated from the pattern of positive and negative microcosms as described above.

Production of acetate was determined using a 1-mL culture sample via ion chromatography using a Metrohm 930 Compact IC Flex ion chromatograph (Metrohm) equipped with a 25-mm Metrosep A Supp 5 anion column and 930 Compact IC Flex 1 conductivity detector. Column oven temperature was set to 40 °C. A 3.2 mM Na2CO3 plus 1 mM NaHCO3 solution was used as eluent at a flow rate of 0.7 mL⋅min−1. The lower limit of acetate detection, as determined by standard curves, was 2.4 μM. Production of methane was determined from a 1-mL headspace sample via gas chromatograph as described previously (36). Production of sulfide was determined from a 0.75-mL culture sample using the methylene blue reduction method (34). Production of soluble or particulate ferrous iron was determined in a 0.5-mL culture sample using the ferrozine method (41). To prevent oxidation, samples were immediately acidified by addition of HCl to a final concentration of 0.5 M. Production of nitrite was determined in a 1-mL culture sample using a Griess assay modified as described previously (42).

MPN (Meta)genome Sequencing.

Total genomic DNA from select MPN microcosms was subjected to Illumina library preparation and paired‐end sequencing (2 × 150 bp) with the NovaSeq platform at the Genomics Core Facility at the University of Wisconsin–Madison. Sequences were assembled, binned, and annotated as previously described (43) (detailed in ). Manual BLASTp searches were conducted to search for specific proteins involved in CO2 fixation pathways, dissimilatory sulfate reduction, dissimilatory nitrate reduction, putative ferric iron reduction pathways, and reversible H2 oxidation (i.e., [NiFe]- and [FeFe]-hydrogenases). BLAST searches were conducted using bait sequences specific to the subunits harboring the active site as well as requisite accessory subunits of each protein complex of interest. Matches with an E value <1 × 10−6, >30% amino acid homology, and covering >60% of the length of the BLASTp bait sequence were considered positive hits. Hits were further screened based on the presence of amino acid residues unique to select functional genes, e.g., the conserved cysteinyl residues demarcating [NiFe]-hydrogenases from membrane bound oxidoreductases (Mbx, Nuo) (44). Putative [NiFe]- and [FeFe]-hydrogenases were confirmed and classified using the HydDB tool (45). The assembled metagenomic data are available under National Center for Biotechnology Information (NCBI) BioProject ID PRJNA622799.

H2 Oxidation Assays.

H2 oxidation assays were conducted in acid-washed 160-mL serum vials prepared as described above for MPN assays. Vials were filled with artificial Skálm Pore Water (SPW) medium (for KJ sediment samples) or artificial RG Pore Water (RGPW) medium (for RG sediment samples) made in milliQ H2O. The composition of SPW was derived from a previous geochemical analysis of the Skálm River, an outlet of KJ to the north of Múlakvisl (46). The composition of this medium is as follows: SiO2 (500 µM), NaCl (500 µM), KCl (25 µM), CaCl2 (250 µM), and MgCl2 (150 µM). The pH of the medium was adjusted to 6.8 with 1 M NaOH. RGPW medium composition was derived from geochemical analysis of RG outflow waters and has been described previously (47). Serum bottles were filled with 60 mL of SPW or RGPW medium for uninoculated controls or 48 mL for heat-killed sediment controls and experimental vials. Vials containing medium were autoclaved and then cooled to 4 °C. Twenty grams of sediment (wet weight) were added to each vial except for uninoculated controls, followed immediately by stoppering and deoxygenation with N2 on ice as described above. Assays to measure rates of H2 transformation under aerobic conditions were not subjected to deoxygenation. Heat-killed sediment controls were autoclaved twice after inoculation as described above.

Twenty-five milliliters of headspace gas was removed from each microcosm and replaced with filter-sterilized CO2 to achieve a final headspace mixing ratio of 20%. Oxidant solutions (prepared in sterile anoxic SPW or RGPW medium using salts described above) were added to microcosms to achieve a final concentration of 1 mM. The uninoculated and heat-killed controls had the same headspace and medium composition as the CO2-amended rate assays, with extra medium added to uninoculated microcosms to account for the lack of sediment. Finally, H2 was added to a final headspace mixing ratio of 500 ppmv (∼1.7 μmol H2 per microcosm). Vials were incubated at 4 °C upside-down to minimize potential H2 diffusion through stoppers.

The headspace mixing ratio of H2 was monitored over time via gas chromatography. For subsampling, 1 mL of ultrapure N2 gas was added to microcosm headspace and mixed well. One milliliter of the mixed headspace was then removed and diluted 1:4 with ultrapure N2. Gases were allowed to equilibrate and the concentration of H2 was determined as previously described (36). Gas phase concentrations were converted to dissolved aqueous phase concentrations according to Henry’s law, as previously described (11). Total microcosm H2 content was calculated as the sum of headspace and aqueous phase H2. The maximum rate of net H2 oxidation was calculated by applying linear regressions to subsets of each data series, as indicated in . The inverse of the slope of each regression, as presented in , was taken to be the rate of net H2 oxidation. Following termination of the incubation, microcosms were assayed for concentrations of total ferrous iron (soluble and particulate), acetate, and methane, as described above.

CO2 Fixation Assays.

Rates of CO2 fixation were determined using previously described methods (8). Briefly, acid-washed 160-mL serum vials were filled with 55 mL of SPW or RGPW medium and autoclaved before inoculation with 10 g wet weight of KJ or RG sediments, respectively. After inoculation, microcosms were amended with oxidant solutions/suspensions to achieve a final concentration of 1 mM, as described above. Inoculated microcosms were purged with N2 gas as described above. The headspace of microcosms was then flushed with filter-sterilized H2 gas. For oxic microcosms, 33 mL each of filter-sterilized air and CO2 were added to the headspace and mixed before 66 mL of mixed gas were removed, leaving 100 mL of headspace comprising 60% H2, 20% air, and 20% CO2. In all other microcosms, 25 mL of filter-sterilized CO2 were added and mixed before 25 mL of mixed gas were removed, for final headspace composition of 80% H2 and 20% CO2. Heat-killed controls were prepared by double-autoclaving as described above. Assays were initiated by the addition of 5 μCi of NaH14CO3.

The amount of 14C incorporated into biomass was determined by uniformly resuspending sediments and removing a 1-mL subsample aseptically using an N2-purged and sterile syringe and needle. Subsamples were acidified to pH <2 by addition of 12 N HCl, vortexed, and allowed to vent in a fume hood for 1 h to volatilize and off-gas unreacted NaH14CO3. Samples were then spun down for 10 min at 17,000 × g and supernatant was removed. Pellets were washed three times by adding 1 mL of 0.1 M HCl, vortexing, and spinning down as described above. Finally, washed pellets were resuspended in 1 mL of milliQ water, transferred to scintillation vials, and mixed with 10 mL of CytoScint Liquid Scintillation Cocktail (MP Biomedicals). The radioactivity associated with each sample was then quantified with a Tri-Carb 2900TR liquid scintillation counter (PerkinElmer).

A standard curve was constructed to account for quenching of photons by sediments suspended in the scintillation cocktail. Microcosms not spiked with 14C were prepared and subsamples were processed as described above. NaH14CO3 was then added in the amount of 40, 70, and 200 pCi per scintillation vial. Triplicate subsamples were prepared at each 14C activity for vials that contained sediments and those that did not, and activity was counted by liquid scintillation as described above.

To determine the amount of CO2 fixed in each microcosm, the following calculations were used. Radioactivity measured in disintegrations per minute was converted to microcuries and activity in microcuries was corrected for sediment quenching using the curve described above. Carbon uptake (Cuptake) was calculated using Eq. below, where DIC is the sum of inorganic carbon added to each microcosm and present in growth medium, APOC is the measured activity of sample particulate organic carbon, and ADIC is the activity of the H14CO3− spike added to each microcosm:

Results were normalized to grams dry weight sediment (gdws) and the means and SDs of measurements on triplicate microcosms are reported. Rates of microbial carbon fixation were determined by applying linear regressions to subsets of the Cuptake datasets presented in . The slopes of these regressions are presented in and were taken to be the maximum rates of carbon fixation.

Inoculation and Monitoring of Transfer Cultures.

MPN cultures inoculated with KJ sediments and amended with hematite were transferred at a 1:10 (vol/vol) ratio into fresh microcosms (70-mL serum bottles, 30 mL of M9 medium prepared and amended as described above for MPN assays). Production of cells was monitored by filtering cells onto 0.22-μm black polycarbonate filters (MilliporeSigma), staining cells with SYBR Gold nucleic acid gel stain (Life Technologies) at a final dilution factor of 10,000, and enumeration via epifluorescence microscopy using an Evos FL microscope (Life Technologies).

Results and Discussion

Sources of H2 in Glacial Meltwaters.

The concentration of dissolved H2 in the outflow stream of KJ, a basalt hosted glacier in southern Iceland, was 426.8 ± 248.2 nM (n = 3) in October 2016 (∼1 km from glacier terminus). The variability in the H2 measurement may be due to the highly turbulent nature of the stream flow at the time of sampling. At RG, a carbonate-hosted system in southern Alberta, Canada, the concentration of H2 in meltwater sampled from the terminus of the glacier was 41.6 nM in October 2014 (48). Unfortunately, the variability associated with this measurement cannot be assessed due to the lack of replicate samples. Other measured geochemical parameters are reported in Table 1.

Table 1.

Geochemical measurements for glacial meltwaters and sediments from Kötlujökull, Iceland, and Robertson Glacier, Alberta, Canada

Sample site	pH	EC, μS/cm	DO, ppm	Fe2+, mg/L	S2−, µM	H2, nM	CH4, nM	CO2, µM	DIC, µM	TOC, mg⋅gdws−1	TN, μg⋅gdws−1
Kötlujökull	6.8	82.1	10.3	ND	3.00	426.8 (248.2)	1.1 (0.1)	77.3 (50.2)	ND	0.38 (0.14)	BLD
Robertson	8.8*	32.5*	11.3*	0.05†	0.31†	41.6†	10.8†	ND	603‡	24.8 (14.0)	212.7 (54.2)

SDs of triplicate analyses are denoted in parentheses, where available. The limit of detection for nitrogen, given the instrumentation and techniques used, was 27 μg⋅gdws−1. Abbreviations: BLD, below limit of detection; DIC, dissolved inorganic carbon; DO, dissolved oxygen; EC, electrical conductivity; gdws, gram dry weight sediment; ND, not determined; TN, total nitrogen; TOC, total organic carbon.

Data from Boyd et al. (47).

Data from Canovas (48).

Data from Boyd et al. (8).

The concentration of H2 measured in KJ meltwaters was similar to or in excess of concentrations measured in waters typically thought of as being enriched in H2, including high-temperature volcanic influenced hot spring systems where dissolved H2 concentrations rarely exceed 1 µM (11, 12). Given that KJ drains Katla Volcano, it cannot be ruled out that the H2 detected in the meltwaters may be volcanic in origin. However, unlike KJ waters, whose low conductivities (82 µS/cm) are consistent with a meteoric origin, waters influenced by volcanic input tend to have much higher conductivities [average >1,700 µS/cm (49)]. This observation, combined with the ∼15-km distance between the KJ sampling location and the primary vent of Katla Volcano, suggests that other sources are likely contributing H2 to meltwaters at KJ leading to the high concentrations measured. However, without knowing the precise conduits emanating from the volcano and their proximity to KJ, a hydrothermal source for H2 in meltwaters cannot be excluded.

Previous studies have shown that pulverization of silicate minerals followed by exposure to water can generate H2 at temperatures ranging from ∼0 to 35 °C (15). Based on the detection of silica radicals following mineral pulverization, a cataclastic mechanism of H2 generation was proposed [i.e., reactions 1 and 2 (15)]. Furthermore, it was suggested that this cataclastic mechanism mimics the process of glacial comminution of silicate bedrock (15) and thus could explain the unusually high H2 concentrations detected in meltwaters from several subglacial habitats (23). However, this study only accounted for gas that was generated upon exposure of pulverized minerals to water and thus excluded the potential for gas to be released from rock inclusions during mechanical crushing. More recent studies have shown that fluid inclusions in muddy carbonates and shales (including those sampled from RG) contain H2 that is released during rock crushing (16) and could be responsible for the H2 detected in RG meltwaters.

The minerology of basalts and other mafic rocks makes possible another H2-generating mechanism: the reduction of water by ferrous silicates, as proposed for the Columbia River Basalt group of central Washington (18). This mechanism might be enhanced in subglacial systems that are actively exposing fresh iron-bearing minerals (e.g., fayalite) capable of reacting with water. Importantly, like muddy carbonates and shales, fluid inclusions in basalts have also been shown to contain H2 that is suggested to derive from iron-based reduction of water when the basalts were crystalizing at high temperature (>400 °C) (50). Thus, glacial comminution or fracturing of the host basalt bedrock could expose fresh iron and silicate minerals capable of reacting with water and producing H2 via mechanisms described above or could release H2 from fluid inclusions.

Regardless of the relative contributions of each of these mechanisms to the availability of H2 in subglacial and proglacial systems, Icelandic volcanic rocks, being enriched in both silica and iron (27, 30, 51) relative to carbonates/shales such as those dominating the RG catchment (33), may thus be predicted to generate more H2 during glacial comminution. Our field data support this prediction, with an order of magnitude more dissolved H2 measured in the meltwaters of KJ than in those of RG. Importantly, the dissolved H2 measured in meltwaters from both systems was higher than would be expected based on equilibration with atmospheric H2 alone [<0.5 nM, given an atmospheric concentration of ∼500 parts per billion (52)], consistent with a source for H2 in the glacial bedrock/sediments as described above. Furthermore, H2 at concentrations even lower than those measured in KJ or RG meltwaters has been shown to support biological H2 oxidation in other systems (53), and H2 oxidizing microbial populations have been recently reported in subglacial ecosystems in Antarctica (26). Together, these observations suggest the potential for H2 oxidation to support microbial life in both KJ and RG sediment communities.

An additional question concerns the source of oxidants capable of supporting H2-dependent growth in subglacial habitats. The generation of H2 due to reduction of water by either cataclastic silica radical- or ferrous iron-based reactions must achieve charge/redox balance, which should occur via an accompanying oxidation reaction. In the case of silica-radical–based reduction of water, it has been suggested that hydroxyl radicals are formed and that these can oxidize ferrous iron or sulfide in bedrock minerals, leading to formation of ferric iron or sulfate (16). Likewise, reduction of water by ferrous iron in minerals is accompanied by iron oxidation to yield ferric iron (18, 50). In subglacial systems, constant flow of ice and meltwater along with comminution of underlying bedrock could transport both H2 (however produced) and oxidized minerals downstream (54) to environments with geochemical conditions that are favorable for the reverse of these reactions to occur.

Rates of Net H2 Oxidation.

To begin to assess the role of H2 in supporting microbial life in subglacial and proglacial habitats, potential rates of net H2 oxidation were determined in microcosm assays amended with H2 and CO2 with or without one of several common oxidants, including SO42−, hematite, ferrihydrite, NO3−, and O2 (Fig. 1 ). In microcosms prepared without sediments (uninoculated) or with autoclaved sediments (heat killed), net H2 loss was equivalent to loss due to sample removal. This indicates that rates of net H2 oxidation exceeding those measured in abiotic controls can be ascribed to biological activity.

Fig. 1.

H2 content (A and C) and carbon fixation (B and D) in microcosms inoculated with proglacial or subglacial sediments from KJ (A and B) or RG (C and D), respectively, and incubated in the dark at 4 °C. The means and SDs of measurements from triplicate microcosms are presented. Abbreviations: Fh., ferrihydrite; gdws, gram dry weight sediment; Hem., hematite; HK, heat-killed control; Uninoc., uninoculated control.

All microcosms exhibited an initial lag phase, defined as the time between introduction of the substrate (H2) and its detectable consumption by microbial activity (Fig. 1 ). In KJ microcosms, the lag phase lasted no more than 12 d. In contrast, the lag phase for RG sediments lasted 18 d under most conditions. While RG sediments were frozen in storage for several years between their collection in 2009 and their use as inocula here, the lag times observed in this experiment match closely with those observed in other growth experiments that used RG sediments stored for between 1 and 5 y (8, 23, 47). Indeed, similar lag times were associated with the onset of CO2 fixation using the same sediments collected from RG in 2009 but frozen for 2 y (8) versus 9 y in the present study. This suggests that any deleterious effects of long-term storage on RG sediment microbiota were minimal.

It has been suggested that the duration of the lag phase can be used as a proxy for the degree of adaptation by a given microorganism (or community of microorganisms) to use a given substrate (55, 56), with shorter lag times indicative of a greater degree of adaptation. In this context, the results of the current study suggest that hydrogenotrophic members of the microbial communities inhabiting KJ sediments may be better adapted to take advantage of H2 as an electron donor than those of RG sediments. Alternatively, hydrogenotrophs may be numerically more abundant at KJ than RG.

The most straightforward approach to addressing the likelihood of these possibilities would be to use molecular methods to determine the abundance of organisms with the genomic capacity to oxidize H2. However, numerous attempts to extract genomic DNA from KJ sediments, including using methods specifically developed for basaltic sediments (57), failed to yield detectable (>0.5 ng DNA per gdws) DNA. Moreover, extracts did not yield PCR amplicons when probed using universal 16S rRNA gene primers. This prevented application of molecular approaches to characterize endogenous sediment communities in KJ and to estimate their abundance.

To further examine the role of H2 oxidation in supporting subglacial and proglacial sediment communities, the maximum rate of net H2 oxidation was determined. This rate ranged from 2.9 to 9.3 (median, 7.6) nmol⋅d−1⋅gdws−1 for KJ microcosms and from 0.8 to 2.3 (median, 1.1) nmol⋅d−1⋅gdws−1 for RG microcosms (Fig. 2 and ). Maximum rates of net H2 oxidation in KJ microcosms were up to 10-fold higher than those in RG microcosms, adding further support to the hypothesis that the microbial communities associated with KJ sediments are better adapted to use H2 as a reductant. For perspective, maximum net rates of H2 oxidation in KJ were within an order of magnitude of those (∼84 nmol H2⋅d−1⋅gdws−1) measured in Antarctic desert surface soils incubated at a slightly higher temperature [10 °C (58)].

Fig. 2.

Maximum H2 oxidation rates in microcosms containing proglacial or subglacial sediments from KJ or RG, respectively, when incubated in the dark at 4 °C. Linear regressions were applied to selected data from the sets represented in Fig. 1 , and the inverse of the slope of each regression is reported. Error bars represent the SD associated with the slope of each regression. Abbreviations: Fh., ferrihydrite; gdws, gram dry weight sediment; Hem., hematite; HK, heat-killed control; Uninoc., uninoculated control.

Rates of Net CO2 Fixation.

A second set of microcosm assays was conducted to measure rates of net CO2 fixation among members of the KJ and RG sediment communities and to gauge whether autotrophy is coupled to hydrogenotrophy in either community, in addition to the extent to which this is similarly (or not) influenced by oxidant amendment. Each microcosm contained H2 and CO2 (a fraction labeled as 14CO2) and was amended with O2, NO3−, hematite, SO42−, or was unamended (CO2 only) (Fig. 1 ). No microcosms were amended with ferrihydrite because, in H2 oxidation assays, there was no discernable difference in activity between microcosms amended with ferrihydrite and those amended with hematite.

In heat-killed control microcosms containing sediments from KJ and RG, net uptake of 14CO2 did not occur (Fig. 1 ). In contrast, net uptake of 14CO2 began to increase after an initial lag phase in all test microcosms. In KJ microcosms, the lag phase lasted 7 to 14 d, depending on the supplied oxidant, whereas in RG microcosms it lasted ∼28 d regardless of oxidant. The lengths of these lag phases match closely with those measured for net H2 oxidation in KJ microcosms, while they are roughly 1.5 times longer than those measured for net H2 oxidation in RG microcosms. This suggests that H2 may be the primary reductant supporting CO2 fixation for at least part of the KJ community, while its primary role in RG may be to supplement facultatively autotrophic or mixotrophic (with respect to electron donor) microbial populations. Support for this possibility comes from the relatively high levels of TOC in RG subglacial sediments (24.8 ± 14.0 mg⋅gdws−1; Table 1) and their porewaters [∼60 µM (47)], a characteristic that would otherwise be expected to promote chemoheterotrophy or chemolithoheterotrophy over chemolithoautotrophy, as has been shown for other natural systems (59) and cultures (60, 61). In contrast, the amount of TOC in KJ sediments is below the practical limit of quantitation for the instrumentation and techniques used in this study (<1.06 mg⋅gdws−1). Extrapolation beyond the lower quantitation limit based on the slope of the calibration curve indicates that TOC in KJ sediments is roughly two orders of magnitude lower than in RG sediments (0.38 ± 0.14 mg gdws−1; Table 1).

The maximum rate of net CO2 fixation ranged from 0.37 to 2.39 (median, 0.44) nmol C⋅d−1⋅gdws−1 for KJ and from 0.04 to 1.98 (median, 0.08) nmol C⋅d−1⋅gdws−1 for RG microcosms (Fig. 3 and ). Rates of net CO2 fixation were 4.8- to 12.2-fold higher in KJ microcosms than in RG microcosms for oxidants other than NO3−. The dramatic increase in both maximum CO2 fixation rate and total accumulated fixed CO2 in microcosms amended with NO3− is a notable feature of each dataset, and is consistent with previous evidence indicating the potential for NO3− reduction in subglacial environments from a variety of catchments (62, 63), including those from RG (47). For comparison, a previous measurement of the rate of CO2 fixation (1.2 ± 0.7 nmol C⋅d−1⋅gdws−1) in unamended (no added H2 or exogenous oxidant) microcosm assays containing RG sediments incubated over a period of 176 d (8) falls between the low rates described above and that observed in RG microcosms amended with NO3−, which was 1.98 ± 0.39 nmol C⋅d−1⋅gdws−1.

Fig. 3.

Maximum CO2 fixation rates within microcosms containing proglacial or subglacial sediments from KJ or RG, respectively, when incubated in the dark at 4 °C. Linear regressions were applied to selected data from the sets represented in Fig. 1 , and the slope of each regression is reported. Error bars represent the SD associated with the slope of each regression. Abbreviations: gdws, gram dry weight sediment; Hem., hematite; HK, heat-killed control. Microcosms were not amended with ferrihydrite for this experiment given the similar rates of H2 oxidation in ferrihydrite- and hematite-amended microcosms.

Given the extremely low nitrogen content of sediments from both catchments (KJ, <27 µg⋅gdws−1; RG, 212.7 ± 54.2 µg⋅gdws−1; Table 1) and their correspondingly high organic C:N ratios, it is possible that the addition of NO3− simply relieved fixed N limitation, thereby increasing the metabolic activities of all members of these communities. However, given that in RG microcosm assays H2 oxidation commenced prior to CO2 fixation, it is unlikely the two processes are tightly linked in the metabolism of a single microbial population. In any case, the higher rates of both H2 oxidation and CO2 fixation in KJ microcosms, and their matching lag times, together point to the linking of these processes in KJ sediments much more strongly than in RG sediments.

KJ and RG microcosms amended with ferrihydrite and hematite, as well as RG microcosms amended with SO42−, exhibited maximum rates of net H2 oxidation and CO2 fixation that were similar to unamended (CO2 only) controls (Figs. 2 and 3). This suggests that dominant primary producers are either 1) using CO2 as an oxidant (i.e., they are acetogens or methanogens) or 2) utilizing an oxidant or reductant endogenous to their source sediments in lieu of those supplied experimentally. Microcosms were therefore probed for evidence of these two possibilities. Neither acetate nor CH4 were detected in the medium or headspace of any of the microcosm assays. Similarly, no microcosm medium contained dissolved sulfide, even when SO42− had been added. Ferrous iron was detected in microcosms amended with ferrihydrite (KJ, mean ± SD, 22.7 ± 9.0 μM; RG, 74.8 ± 62.3 μM), hematite (KJ, 40.7 ± 7.6 μM; RG, 20.3 ± 5.1 μM), SO42− (KJ, 33.2 ± 5.4 μM; RG, 39.5 ± 27.9 μM), and CO2 only (KJ, 20.9 ± 2.9 μM; RG, 8.7 ± 4.4 μM), but was also present in KJ heat-killed control microcosms (13.2 ± 0.9 μM), which might be expected given the Fe(II) content of the iron rich basaltic sediments at KJ. NO2−, an intermediate in the reduction of NO3−, was present in NO3−-amended microcosms from both catchments (KJ, 131.4 ± 8.5 μM; RG, 5.8 ± 2.7 μM), suggesting that NO3− may stimulate both hydrogen oxidation and CO2 fixation in both systems.

Abundance of Hydrogenotrophs and Oxidant Coupling.

To further investigate the abundance of hydrogenotrophs and the oxidants that couple with H2 oxidation in KJ and RG sediment populations, an MPN cultivation approach was undertaken (Fig. 4). Microcosm assays containing M9 minimal medium, with H2 as the sole added reductant and CO2 as the sole added carbon source, were amended with O2, NO3−, ferrihydrite, hematite, SO42−, or were not amended (CO2 only). Surprisingly, there were between 2- and 90-fold more cultivatable hydrogenotrophs in RG sediments than in those from KJ. This observation is consistent with the higher amounts of DNA recovered from RG sediments (7.4 ng⋅gdws−1) when compared to KJ sediments (<0.5 ng⋅gdws−1). To the extent that cultivation and DNA extraction/recovery bias was minimized in these experiments, these data suggest that the faster onset and higher rates of microbial H2 oxidation and CO2 fixation activity observed in KJ sediment microcosms described above are unlikely to result from a greater in situ abundance of hydrogenotrophic cells. Rather, they suggest that the community inhabiting KJ sediments is assembled to include populations better adapted to efficiently utilize available nutrients such as H2, which is at a concentration in KJ meltwaters that is nearly an order of magnitude greater than observed in RG meltwaters.

Fig. 4.

Abundance of hydrogenotrophic cells capable of autotrophic growth as determined by most probable number (MPN) assays containing dilutions of proglacial or subglacial sediments from KJ or RG, respectively. The MPN was assessed first by quantifying total extractable DNA (A) and further refined by measurement of the metabolic products of CO2, SO42−, Fe(III), or NO3− reduction (methane and/or acetate, sulfide, ferrous iron, and nitrite, respectively) (B). Microbial growth was detected in CO2- and SO42−-amended microcosms without associated reduction of CO2 or SO42−. This observation led to the hypothesis that growth under these conditions is supported by heterotrophy. Error bars represent 95% confidence intervals. Abbreviations: Ace., acetate; Fh., ferrihydrite; gdws, gram dry weight sediment; Hem., hematite.

A related possibility is that the apparent abundance of hydrogenotrophs at RG reflects populations that are mixotrophic and/or can facultatively oxidize H2. Evidence in support of this possibility comes from physiological and genomic studies of the dominant primary producer in RG sediments, Thiobacillus sp. RG5 (8). This strain is supported primarily by products (e.g., thiosulfate) of abiotic FeS2 oxidation, but its genome also encodes a group 1 [NiFe]-hydrogenase (7) that is predicted to function in H2 oxidation (20). While growth experiments aimed at evaluating H2 transformation were not conducted with Thiobacillus sp. RG5, they have been conducted with a close relative, Thiobacillus denitrifications. T. denitrificans also encodes a group 1 [NiFe]-hydrogenase but was shown to be unable to use H2 as sole electron donor when growing autotrophically via NO3− reduction (64). This suggests the possibility for mixotrophic energy metabolism as it relates to H2, as has been shown for other organisms (60, 61).

MPNs were evaluated by quantifying DNA and reduced products from supplied oxidants. When MPNs were estimated using total extractable DNA, the abundance of hydrogenotrophs in both KJ and RG sediments differed by less than an order of magnitude among experimental conditions (supplied oxidants) (Fig. 4). KJ sediment MPNs estimated by total extractable DNA were not substantially different from those estimated by the detection of Fe2+ for ferrihydrite- and hematite-amended microcosms, nor were DNA-based MPN estimates different from those based on detection of NO2− for NO3−-amended KJ and RG microcosms (Fig. 4). However, MPNs estimated based on the detection of products of the reduction of SO42− and CO2 did substantially differ from those based on total extractable DNA (Fig. 4 ). Given that sediments from both RG and KJ are known to contain iron oxides () (65, 66), it is possible that sulfide generated from SO42− reduction was oxidized by Fe(III) in a cryptic cycle similar to what has been described for other freshwater sediment ecosystems (67). However, the similar numbers of hydrogenotrophs in the SO42− (also containing CO2) and CO2-only conditions, combined with the absence of detectable methane or acetate in MPN and microcosm assays, indicate instead that the microbial populations enriched under the CO2-only and SO42−-amended conditions (for both KJ and RG sediments) likely possess mixotrophic metabolisms in which H2 oxidation is not necessarily coupled to autotrophy. As such, these cells may have been supported by remnant organic carbon in sediments used to inoculate the MPN enrichment series.

The observation of similar MPN estimates derived from DNA extraction and Fe2+ detection in hematite- and ferrihydrite-amended KJ sediments (Fig. 4), combined with evidence for both net H2 oxidation and CO2 fixation in hematite- and ferrihydrite-amended KJ microcosms (Figs. 1–3), point to potential coupling of H2 oxidation, iron reduction, and CO2 fixation in KJ sediments. To further investigate this possibility, microcosms containing M9 minimal growth medium amended with H2, CO2, and hematite were inoculated with MPN cultures exhibiting both microbial growth and iron reduction. Due to the dilution of glacial sediment during preparation of MPN cultures and their subsequent transfer into M9 medium, the transfer of endogenous oxidants and organic carbon to these cultures was presumed to be negligible [transfer cultures estimated to contain 0.6 µg TOC⋅mL−1 after ∼600-fold (vol/vol) dilution of sediments]. Continued growth of a single morphotype was observed, and this growth was associated with the production of Fe2+ (Fig. 5). Given that CO2 was the only provided carbon source and H2 the only provided reductant, these observations point to this population coupling H2 oxidation with the reduction of hematite to fuel autotrophic growth. While H2 oxidizing, chemolithoautotrophic Fe(III) reducers are known to occur in high-temperature environments (68, 69) and have recently been identified in low-pH, moderate-temperature environments (70), to our knowledge such organisms have not previously been identified in low-temperature environments of circumneutral pH.

Fig. 5.

Activity and growth of a KJ most probable number (MPN) culture amended with H2, CO2, and hematite when transferred into fresh medium containing these components. (Meta)genomic sequencing of DNA extracted from this culture reveals a single metagenome assembled genome affiliated (94% RpoB ID) with Rhodoferax ferrireducens. Culture Fe2+ content was calculated as the concentration of Fe2+ measured in the experimental microcosm minus that measured in the abiotic control. Error bars represent SEM. Abbreviations: AC, abiotic control; LOD, limit of detection.

MPN (Meta)genomic Sequence Analysis.

Nutrient amendments that promoted both H2 oxidation and CO2 fixation were targeted to identify the population(s) responsible for the observed activities and to probe mechanisms of H2 oxidation in these taxa. These amendments included NO3−, hematite, ferrihydrite, SO42−, and CO2 alone for KJ sediments and NO3− for RG sediments. MPN cultures grown with each of these amendments were subjected to DNA extraction and metagenomic sequencing. The composition of each sequenced MPN community is shown in Table 2, while the completeness and contamination of each recovered metagenome assembled genome (MAG) is presented in .

Table 2.

Composition of communities from the most dilute MPN assays amended with H2, CO2, and (where indicated) other oxidants and containing proglacial or subglacial sediments from Kötlujökull or Robertson Glacier, respectively

Glacier	Supplied oxidant	% Binned populations	Taxon	RpoB identity	Hydrogenase (subgroup)	CO2 fixation marker	Terminal reductase
Kötlujökull	CO2	92.9	Glaciimonas sp. PCH181	91.9%	HupUV (2b), HoxYH (3d)	CbbSL
		3.9	Cellulomonas sp. WB94	99.2%	HydA (A1)	None
		3.3	Phycicoccus sp. Soil748	93.5%	HyhBGSL (3b)	None
	SO42−	88.5	Glaciimonas sp. PCH181	92.0%	HupUV (2b), HyaABC (1d)	CbbSL
		3.5	Phycicoccus sp. Soil748	93.8%	HyhBGSL (3b)	None
		4.3	Cellulomonas sp. WB94	99.2%	HydA (A1)	None
	Hem.*	97.3	Rhodoferax ferrireducens	93.7%	HupUV (2b), HyaABC (1d)	CbbM	NarGHJI
		2.7	Polaromonas sp. CF318	95.6%	HupUV (2b)	None	NarGHJI
	NO3−	100.0	Rhodoferax ferrireducens	93.7%	HupUV (2b), HyaABC (1d)	CbbM	NarGHJI
Robertson	NO3−	96.7	Polaromonas sp. A23	96.9%	None	CbbSL	NarGHJI
		3.3	Polaromonas glacialis	98.1%	None	4-Hydroxybutyryl-CoA dehydratase	NarGHJI

Community DNA was extracted and subjected to metagenomic sequencing to reveal the relative abundance of metagenome assembled genomes (MAGs) in the community, the taxonomic composition of MAGs via RNA polymerase β-subunit (RpoB) homology, and the presence of protein homologs that would allow for H2 oxidation, CO2 fixation, and the coupling of H2 oxidation with supplied oxidants. Abbreviations: CbbM, the large subunit of the proteobacterial form II (L2) ribulose-1,5-bisphosphate carboxylase/oxygenase; CbbSL, the small and large subunits of ribulose-1,5-bisphosphate carboxylase/oxygenase involved in the Calvin cycle, respectively; Hem., hematite; HoxYH, the small and large subunits of the group 3d bidirectional [NiFe]-hydrogenase (20); HupUV, the small and large subunits of the group 2b sensory [NiFe]-hydrogenase; HyaABC, the small, large, and cytochrome subunits of the group 1d uptake [NiFe]-hydrogenase, respectively; HydA, the catalytic subunit of [FeFe]-hydrogenases; HyhBGSL, the iron–sulfur, diaphorase, small, and large subunits of the group 3b bidirectional [NiFe]-hydrogenase, respectively; NarGHJI, the α, β, molybdenum cofactor assembly chaperone, and γ subunits of the dissimilatory nitrate reductase complex, respectively.

The MAGs described for Kötlujökull hematite-amended MPN assays were recovered from a culture transferred into fresh medium, the growth and Fe(III) reduction activity of which are depicted in Fig. 5.

Sequencing of genomic DNA from KJ MPN assays amended with H2/CO2 only and with H2/CO2/SO42− yielded the same three MAGs. The most abundant MAG in the H2/CO2-only and the H2/CO2/SO42−-amended MPNs (93% and 89% of total reads, respectively) exhibited close affiliation (92% identity at the amino acid level of the RNA polymerase β-subunit [RpoB] for both bins) with Glaciimonas, a metabolically flexible genus common to glacial ecosystems (e.g., refs. 71–73). Both Glaciimonas MAGs encode homologs of [NiFe]-hydrogenases and ribulose-1,5-bisphosphate carboxylase/oxygenase (RuBisCO), consistent with an ability to grow autotrophically using H2 as an electron donor. However, despite being 99% similar at the level of RpoB sequence identity, the two MAGs differed in the complements of encoded [NiFe]-hydrogenases. Both Glaciimonas MAGs encode homologs of group 2b H2 sensing [NiFe]-hydrogenases that regulate transcription and maturation of oxidative [NiFe]-hydrogenases (20, 74). In addition to the group 2b homolog, the Glaciimonas MAG from the H2/CO2/SO42−-amended microcosm encodes a group 1d oxidative [NiFe]-hydrogenase and the MAG from the H2/CO2-amended microcosm encodes a bidirectional group 3d [NiFe]-hydrogenase. The presence of both an H2 sensing and an oxidative [NiFe]-hydrogenase points to adaptation to efficiently utilize H2 when it is available while potentially minimizing costs associated with de novo protein and cofactor biosynthesis. However, the presence of a bidirectional hydrogenase in the H2/CO2 Glaciimonas MAG, rather than an oxidative hydrogenase as found in the H2/CO2/SO42− MAG, casts doubt on the reliance of this population on H2 as its sole source of electrons during growth and could be further evidence of mixotrophic or facultatively autotrophic metabolism. Homologs of the dissimilatory bisulfite reductase, a gene required for reduction of SO42−, were not detected in either Glaciimonas MAG, consistent with the absence of detectable sulfide in KJ MPN and activity assay microcosms amended with SO42−.

Sequencing of DNA extracted from hematite- and ferrihydrite-amended KJ MPNs failed. However, a single transfer of the H2/CO2/hematite-amended MPN into fresh medium grew (Fig. 5), and this culture was subjected to DNA extraction and (meta)genomic sequencing. Both the hematite KJ MPN transfer culture and NO3−-amended KJ MPN culture comprised a dominant MAG that was closely related (94% RpoB sequence identity) to Rhodoferax ferrireducens, a genus that is commonly identified in subglacial habitats (62, 75) including RG (3, 6). Interestingly, like one of the Glaciimonas MAGs, the Rhodoferax MAG encoded homologs of a group 2b H2 sensing [NiFe]-hydrogenase and a group 1d uptake [NiFe] hydrogenase, again pointing to adaptation to effectively respond to H2 availability and use it as a component of energy metabolism. The MAG also encoded RuBisCO and dissimilatory nitrate reductase (NarGHJI), consistent with the stimulation of H2 oxidation and CO2 fixation by amendment with NO3−. Importantly, Rhodoferax ferrireducens strains have been shown to be metabolically flexible, displaying growth phenotypes as diverse as photoheterotrophy, aerobic heterotrophy, and fermentation (76). To our knowledge, however, H2-dependent chemolithoautotrophic growth has not so far been observed in this genus.

Despite evidence for iron reduction in KJ MPN assays, activity assays, and transfer cultures containing Rhodoferax, we were unable to identify genes encoding the porin cytochrome complexes that have been implicated in Fe(III) reduction in other bacteria. Notably, both the Pcc system used by Geobacter and several other metal-reducing organisms (77, 78) and the Mtr system that plays a similar role in Shewanella (79, 80) were absent from the Rhodoferax MAGs, despite each being estimated to be 99% complete. Shi et al. (79) previously reported the discovery of homologs of MtrABC and accessory proteins encoded in the genome of R. ferrireducens using a BLAST-based approach. However, a BLASTp analysis of amino acid sequences of MtrABC in Shewanella oneidensis against the genome of R. ferrireducens only revealed a homolog of the MtrA subunit (e value <1E-70). No homologs of MtrABC were detected in the KJ-derived Rhodoferax MAGs. Importantly, dissimilatory iron reduction has been demonstrated in R. ferrireducens (81) despite this lack of evidence for the presence of known Fe(III) reduction pathways in its genome or in the KJ-derived Rhodoferax MAGs.

The only nutrient that stimulated both net H2 oxidation and CO2 fixation at RG was NO3− (Fig. 1 ). For this reason, only the NO3−-amended RG MPN was subjected to metagenomic sequencing and analysis. Two MAGs were identified in this MPN and were both affiliated with Polaromonas strains. While both MAGs encoded homologs of RuBisCO and dissimilatory nitrate reductase (NarGHJI), neither encoded homologs of [FeFe]- or [NiFe]-hydrogenases or proteins involved in their maturation. Thus, it is not clear how these organisms were growing in MPN assays unless they were using an endogenous reductant, such as organic carbon, FeS2, or products derived from FeS2 oxidation. Importantly, a transfer of this MPN into the same medium did not yield growth, consistent with the use of an endogenous reductant in the MPN assays.

Conclusions

Both silicate- and iron-bearing minerals can generate H2 through cataclastic radical and reductive mechanisms in the presence of water. These minerals are chemically altered in the process, however, thereby preventing sustained reactivity unless they are resurfaced. Glacial comminution is one mechanism of generating fresh mineral surfaces capable of reacting with water on a continual basis. Thus, subglacial environments have the potential to provide continuous sources of chemical energy, in the form of disequilibrium between H2 and oxidized minerals, that can support chemolithotrophic microbial primary production in subglacial and downstream proglacial environments.

Meltwaters from KJ, which overlays a basaltic catchment, had an order of magnitude more dissolved H2 than those from RG, which sits in a carbonate catchment. The sediment community at KJ exhibited shorter lag times and significantly faster rates of H2 oxidation and CO2 fixation than those measured for the RG sediment community. Metabolic activity assays and MPN experiments show that at least two populations of autotrophic hydrogenotrophs inhabit KJ sediments: one metabolically flexible mixotroph (Glaciimonas sp.) and one that can, at a minimum, utilize H2, CO2, and Fe(III) or NO3− (Rhodoferax sp.). In RG sediments, however, differing lag times between H2 oxidation and CO2 fixation experiments suggest that the two processes are not tightly coupled. A potential explanation for this comes from a prior study that showed the dominant primary producer in RG sediments depended primarily on FeS2 or derivatives of its oxidation (i.e., thiosulfate) as reductants (7, 8). Thus, H2 oxidation in RG sediment populations may be attributable to a mixotrophic or chemolithoheterotrophic metabolism, as has been recently demonstrated in other taxa (60, 61).

(Meta)genomic sequencing of the most dilute MPN assays from KJ identified a relative of the NO3−- and Fe(III)-reducing Rhodoferax and a population of Glaciimonas as the dominant H2 oxidizing autotrophs and/or mixotrophs in this system. Transfers of the Rhodoferax MPN culture into fresh medium further demonstrated the ability of this population to grow autotrophically using the H2/Fe(III) redox couple. This represents the only report to date of chemolithoautotrophic, H2-dependent iron reduction in a nonhyperthermophilic or acidophilic organism, and experiments to further characterize this bacterium are underway.

Collectively, these results indicate that differences in bedrock mineral composition likely influence supplies of lithogenic H2 and oxidants and that these, in turn, influence the diversity, abundance, and activity of H2-dependent autotrophs in subglacial and proglacial habitats. More broadly, they further underscore the importance of physical and chemical weathering processes in releasing nutrients capable of sustaining microbial primary production in these habitats, corroborating recent evidence that such communities could be supported by chemolithotrophic primary production, independent of photosynthetically fixed carbon (8). This finding has important implications for the survival of terrestrial life through Snowball Earth periods of globally extensive glaciation (82, 83) and for the possibility of discovering life on other icy planetary bodies (84). Indeed, the recent discovery of H2 gas in plumes of material erupting from the icy crust of Enceladus (85) points to active rock–water interactions occurring beneath the ice surface and implies a subsurface geochemical system with potentially important parallels to that of KJ and other glaciers in basaltic terrains.