Introduction

Investigations have described microorganisms in the seafloor (Sogin et al., 2006; Mason et al., 2010), in continental crust (Davidson et al., 2011) and beneath ice sheets (Lanoil et al., 2009). A deep microbial biosphere may be a vast reservoir of diversity (Whitman et al., 1998), a refuge for life (Sleep and Zahnle, 1998) and the setting of the origin of life itself (Martin et al., 2008). Subglacial lakes, such as those under East Antarctica, are analogs to possible extraterrestrial habitats (Siegert, 2005). In the absence of sunlight or organic matter from the surface, subglacial life must rely on ambient geochemical energy (Hoehler, 2004). In subglacial volcanic settings, mixing of volcanic gases, fluids and minerals, as well as oxygenated glacial meltwater, creates chemical disequilibrium (Giggenbach, 1980; Arnórsson et al., 1983; Oelkers and Gíslason, 2001; Symonds et al., 2001), and the available energy can be harvested by microorganisms in coupled reduction–oxidation reactions (Gaidos et al., 1999). Populations of such microbes may descend from the original inhabitants of the unglaciated landscape, but may also be released into a lake during basal melting of the glacier in which they were originally entombed (Christner et al., 2000; Miteva et al., 2004).

In Iceland, the Vatnajökull ice cap covers several active volcanoes. Meltwater drains to minima in the glaciostatic potential at its bed and fills three subglacial lakes: one in the Grímsvötn caldera and two beneath depressions or ‘cauldrons’ (Icelandic katlar, singular ketill) in the glacier surface to the northwest of Grímsvötn (Björnsson, 2002). The latter two lakes periodically discharge in subglacial floods (jökulhlaups) to the Skaftá river 40 km away and are called the Skaftárkatlar lakes. Forty km to the northeast, glacial melt from a hydrothermal field feeds Kverkfjallalón lake, which is hydraulically disconnected from the Skaftá system. This lake was formerly subglacial, but is now exposed to the atmosphere (Figure 1).

Figure 1
figure 1

Map of the Vatnajökull ice cap in Iceland showing locations of the Grímsvötn, Skaftárkatlar and Kverkfjöll lakes, and the subglacial conduit through which the Skaftárkatlar lakes drain into the Skaftá river.

Full size image

Previous investigations of the Grimsvotn and west Skaftárketill lakes found their waters to be dilute (total dissolved solids <200 p.p.m.), cold (<5 °C) and low in organic matter (<<1 mg l−1 C) owing to melting of the overlying, nutrient-poor ice (Ágústsdóttir and Brantley, 1994; Gaidos et al., 2004; Jóhannesson et al., 2007; Gaidos et al., 2009). The Grímsvötn lake was sampled in June 2002, 3 months after a jökulhlaup, and consisted of oxygenated glacial melt, but the west Skaftárketill lake, explored in June 2006, was anoxic and sulfidic, with hydrothermal input (Jóhannesson et al., 2007). The pelagic microbial communities are also distinct: Grímsvötn contains a low abundance (2 × 104 cells ml−1) of diverse heterotrophic bacteria (Gaidos et al., 2004), whereas west Skaftárketill has a higher biomass (5 × 105 cells ml−1) dominated by a few bacterial taxa related to anaerobes and microaerobes, with no evidence for archaea (Gaidos et al., 2009). Differences in chemistry can be explained by varying inputs of meltwater, volcanic gases and hydrothermal fluid. Differences in the microbial communities can be ascribed to environmental selection, provided cells affiliated with these groups are carried from other anoxic sites to the glacier surface and remain viable during passage to the lake. Vatnajökull ice contains ∼4 × 103 cells cm−3, but these appear neither to be viable nor related to the Skaftárketill taxa (Gaidos et al., 2004).

Here we present results from additional sites under Vatnajökull, including the east Skaftárketill lake, a jökulhlaup in the outlet to the Skaftá river and the Kverkfjallalón lake. These molecular data include the first paired, long-read pyrosequence libraries of the 16S rRNA gene based on different primer sets. Our findings indicate that the dominant bacterial taxa are widely distributed beneath the ice of Vatnajökull, suggesting a subterranean source for this subglacial microbiome.

Materials and methods

Drilling, sampling and temperature data

The east Skaftáketill lake was sampled during a 3-day interval in June 2007 by two boreholes drilled through the 280-m-thick ice sheet by a sterilizing hot water drill (Thorsteinsson et al., 2008). Boreholes A and B, separated by 170 m, were at 64.48702°N 17.50583°W (1593 m above sea level (a.s.l.)) and 64.48757°N 17.50915°W (1588 m a.s.l.), respectively. The depth of the lake was 103 m at both locations. Multiple temperature profiles were obtained in each borehole using a Starmon mini-temperature recorder with an estimated accuracy of ±0.02 °C after zero-point adjustment. Depth and bottom temperatures were obtained with a Geokon 4500SHSR-7.5 MPa vibrating wire pressure sensor and thermometer connected to a Campbell CR-10X data logger on the glacier surface. Samples were collected in a 1 l gas-tight sampler (Mount Sopris 2FSA-1000). Contamination of the lake and reverse contamination of samples were avoided using the techniques described in Gaidos et al. (2009). Samples A1–A3 were retrieved from 284, 331 and 379 m below the ice surface and B1–B4 from 284, 336, 377 and 388 m. Samples were kept under anaerobic conditions at 4 °C during transport to the laboratory and processed within 3–4 days. In June 2009, 750 ml of the Kverkfjallalón lake (64.669°N, 16.685°W, 1675 m a.s.l.) was collected from the surface and depths of 4, 6 and 9 m using the same sampler. This was reduced by Na2S solution (0.05% (w v−1) final concentration) and kept at 4 °C during transport to the laboratory. A jökulhlaup sample was collected from the Skaftá river near its outlet point at the margin of the Skaftárjökull outlet glacier of Vatnajökull (64.256°N, 18.136°W) on 11 October 2008, close to the time of maximum discharge (1290 m3 s−1).

Chemical analysis

Samples for determination of major and trace elements were filtered on-site through a 0.2-μm cellulose acetate membrane with a Teflon filter or, using polypropylene syringes, into polypropylene bottles and acid-washed amber glass bottles. The water pH was analyzed on site using a pre-calibrated combination glass electrode (Metrohm, Riverview, FL, USA). The measurements were carried out at ∼25 °C to prevent errors related from pH dependence on temperature. Samples for major anions (CO2, Cl, F, SO4, S2O3) were analyzed on-site within ∼20 min of sampling using ion chromatography (Kaasalainen and Stefánsson, 2011). Dissolved H2S was measured using precipitation titration on-site with Hg-acetate solution (Kaasalainen and Stefánsson, 2011). Samples for major elemental analysis (Si, B, Fe, Al, Na, K, Ca, Mg) were collected into polypropylene bottles and acidified with Suprapur HNO3 (0.5 ml acid to 100 ml sample) and analyzed using inductively coupled plasma emission spectrometry (Spectro, Kleve, Germany) (Skoog and Leary, 1992). Dissolved inorganic nutrients NO2−, NO3− and NH4+ were collected into high-density polyethylene bottles and preserved by freezing (−18 °C) until analyzed, to prevent any organic growth or decay. Dissolved nutrients were analyzed colorimetrically. Before analysis of NH4, the samples were acidified using H2SO4 (0.5 ml of 1% H2SO4 to 100 ml sample) (APHA, 2005).

Enrichment cultures

Media for enrichments of chemolithotrophic and chemo-organotrophic organisms were prepared as described in Gaidos et al. (2009) using 0.5 ml sample and 4.5 ml 0.2-μm filtered water from the subglacial lake (the latter alone as a negative control). Cultures were incubated under aerobic (ambient headspace) and anaerobic conditions at 3 °C, 60 °C and 80 °C. Each enrichment was prepared in a Hungate culture tube. WN2 enrichments contained 0.01% yeast extract, vitamin solution, Balch element solution (Balch et al., 1979), S0 and resazurin, and were incubated under pure N2 and 0.025% final wt v−1 Na2S•9H2O. WO2 enrichments were the same as WN2, except incubated aerobically with ambient headspace. C–J enrichments had addition of 50 μl of 100 × yeast-acetate medium and were incubated with 80%/20% H2/CO2 and 0.025% final wt v−1 Na2S•9H2O. A–J enrichments were the same as C–J, but incubated aerobically with ambient headspace. Additional enrichments used R2A medium and 162 Thermus medium (Degryse et al., 1978), both aerobically with ambient headspace, and Thermotoga (‘Toga’) medium (Marteinsson et al., 1997) and YPS medium (Marteinsson et al., 2001) under pure N2 headspace. Growth in enrichments was confirmed with phase-contrast microscopy (Olympus BX51). Growth was observed every 2 days for 5 weeks and then every 2–6 days for several months. Where growth occurred, serial dilutions were performed in complement medium to isolate strains.

Cell counts and fluorescent in situ hybridization

For flow cytometry, samples were fixed with 1% (v v−1) glutaraldehyde for 20 min at room temperature before staining with 1:1000 solution of Sybrgreen I (Molecular Probes, Eugene, OR, USA). Cells were enumerated by a FACS (Aria II) flow cytometer and samples were delivered at a calibrated rate for 2 min. Forward- and right-angle light scattering and DNA fluorescence data were collected. Fluorescent in situ hybridization (FISH) hybridizations and enumeration by microscopy were as described in Gaidos et al. (2009). We used three universal probes for bacteria (EUB I/II/III), two universal probes for archaea (ARC344 and ARC 944) and four probes previously designed to target the dominant lineages in the west Skaftá lake: Acetobacterium, Sulfuricurvum, Sulfurospirillum and Paludibacter (Supplementary Table S1). Enumeration of each taxon was normalized by total bacterial counts with EUB I/II/III. We determined the efficiency of our probes by comparing them with our pyrosequence libraries (described below). The Sulfuricurvum, Sulfurospirillum and Paludibacter probes matched 95%, 96% and 96%, respectively, of v3–v5 sequences with the same taxonomic assignment. For the v4–v6 libraries, the respective fractions are 97%, 76% and 93%, respectively. The Acetobacterium probe does not target a part of the ribosomal gene that was pyrosequenced.

Clone library construction and sequencing

DNA was extracted as described previously (Gaidos et al., 2009). Polymerase chain reaction amplification was performed according to the protocol in Skírnisdóttir et al. (2001), with primers targeting either the archaeal or bacterial domains (Supplementary Table S1). Only reactions with bacterial primers yielded detectable product; all amplification attempts with archaea-specific primers using gradients, different dilutions and alternative cycling protocols were unsuccessful. All positive controls using the archaeon Thermococcus barophilus produced detectable product. Polymerase chain reaction product was cloned by the TA method using a TOPO TA cloning kit (Invitrogen, Carlsbad, CA, USA). Plasmid DNA from single colonies was isolated and sequenced using a BigDye Terminator Cycle Sequencing Ready Reaction Kit on an ABI sequencer (PE Applied Biosystems, Foster City, CA, USA). All clones were sequenced using the reverse primer 805R (Supplementary Table S1). Chimera detection was performed using Greengenes (DeSantis et al., 2006). Sequences were aligned and taxonomically classified using neighbor-joining and parsimony trees constructed by the ARB package (Ludwig et al., 2004). These sequence data have been submitted to the GenBank database under accession numbers JQ697981–698334.

Pyrosequencing

Pyrosequencing of 70–100 nucleotides (nt) of the v6 variable region in the 16S ribosomal gene was performed with a 454 GS-FLX (Roche, Basel, Switzerland) on samples B1, B3 and B4 from the east Skaftárkatlar lake, a Kverkfjallalón sample from 4 m and the Skaftá river sample. Template was first verified by amplification using primers 967F and 1046R, the same used for our study of the west Skaftá lake (Supplementary Table S1). The primers were subsequently used for 5′→3′ pyrosequencing, with the addition of the linker sequence plus a unique 4-base ‘barcode’ to the forward primer for parallel sample sequencing. Tags shorter than 60 nt were discarded when trimmed by the GS FLX software. The resulting v6 libraries comprise 59 772 reads. Two sets of reactions targeting the v3–v5 and v4–v6 regions of the bacterial 16S rRNA gene were performed, and pyrosequencing carried out on a 454 GS-FLX with Titanium chemistry as part of the Census of Deep Life (Sogin et al., 2010). Details of library construction are given in the Supplementary Information. New primers (Supplementary Table S1) were designed to capture >95% of known eubacterial diversity based on the SILVA version 106 database (Pruesse et al., 2007). They match 50–100% of GenBank 16S sequences from members of all known phyla, with the exception of a few small phyla such as OP11 and SR1. Sequencing was from the v3 and v6 ends of the respective libraries. Raw sequences were filtered, trimmed and matched to a reference sequence database constructed from SILVA by the VAMPS pipeline (Huse et al., 2010). The distribution of v3–v5 read lengths has peaks at 415 and 440 nt, and that of v4–v6 reads peaks at 475 and 487 nt. After quality control and removal of likely non-ribosomal interlopers, pyrosequencing runs on east lake extracts yielded 150 586 v3–v5 reads and 147 979 v4–v6 reads. Filtering of chimeras reduced these to 150 586 and 147 959 reads, respectively. Statistics of individual libraries are provided in Supplementary Table S2. These data have been submitted to the NCBI Short Read Archive as SRA 050216, 050305, 050636 and 050638.

Pyrosequence analysis

Further sequence library analysis used MOTHUR version 1.22 (Schloss et al., 2009, 2011), custom Perl scripts and the R statistical package (R Development Core Team, 2008). Raw v6 libraries were filtered of reads with lengths shorter than 50 nt, lengths longer than 110 nt (to remove dimers), ambiguous base calls or average call quality factor <30. The v3–v5 or v4–v6 sequences with GAST distances >0.25 are probably non-ribosomal and were removed. Unique v3–v5 and v4–v6 sequences were aligned to the Greengenes core alignment (DeSantis et al., 2006). The v6 sequences were aligned to the SILVA bacterial alignment (Pruesse et al., 2007) and those that started after position 31 189 or ended before position 3255 were removed. Missing or incompletely covered ends were filtered and non-unique sequences were again removed from the resulting alignment. Sequences were ‘pre-clustered’ to account for amplification and sequencing errors (Huse et al., 2010). On the basis of an average 454 GS-FLX error rate of 1% (Gilles et al., 2011) and the need to reduce sequences to a practical number, pre-clustering criteria of 1 nt difference for v6, 2 nt for v3–v5 and 4 nt for v4–v6 were adopted. Chimeras were identified and removed using Chimera Slayer (Haas et al., 2011) and UCHIME (Edgar et al., 2011), with the data sets themselves as the references. Calculation of distances between sequences penalized gaps by −1 and excluded end gaps. The v3–v5 and v4–v6 sequences were clustered into operational taxonomic units (OTUs) using a 3% average distance cutoff; we adopted a 10% cutoff for v6-based OTUs (Gaidos et al., 2009). To compare α-diversities without regard to sampling effort, we constructed 1000 bootstrap libraries of each real library with the size of the smallest library, and the average and standard deviation of the abundance-based coverage estimator of species richness (Lee and Chao, 1994) were computed. The distinctiveness of libraries was determined using the integral form of the Cramer–von Mises test. Dissimilarity distances between libraries were calculated using UPGMA trees based on either Czekanowski–Bray–Curtis distances (Bray and Curtis, 1957; Czekanowski, 1909) or Horn–Morisita distances (Horn, 1966); bootstrap support was based on 1000 Monte-Carlo simulations of the data and a majority consensus tree constructed using PHYLIP (Felsenstein, 1989). Taxonomic assignments used the Ribosomal Database Project Bayesian classifier (Wang et al., 2007) and bacterial and archaeal subsets of the reference database of Huse et al. (2010). We required 80% consensus of members for OTU classification.

Results

Physiochemical conditions

Temperature profiles of the east Skaftárketill lake are near 4 °C and about 0.2 °C above the maximum density temperature of pure water (Supplementary Figure S1). There is no indication of a colder bottom water mass as in the west lake (Jóhannesson et al., 2007), but a single profile beneath borehole A contains some suggestion of this phenomenon. Like the west lake, all east lake samples were anoxic, with average sulfide (H2S) of 16 p.p.m., and average dissolved CO2 of 105 p.p.m. Dissolved solids totaled 250–300 p.p.m., with SiO2 and Na the dominant constituents (mean concentrations of 100 and 30 p.p.m., respectively) (Supplementary Figure S2 and Supplementary Table S3). Iron and sulfate (SO4) concentrations were lower than in the west lake, with mean values of 0.2 and 1.7 p.p.m., respectively. NO2 was marginally detected (sample average of 46 p.p.b. above blank levels), but neither NO3 nor PO4 were detected above blank levels of ∼10 p.p.b. The east lake appears uniform for most species (variation of a few percent), except for SO4, H2S and CO2, which vary by 95%, 32%, and 14%, respectively. H2S and CO2 de-gas from samples retrieved from depth, but SO4 does not. Sulfate concentration increases with depth under both boreholes and is higher at borehole A (Figure 2).

Figure 2
figure 2

Profiles of sulfate concentration normalized to mean values (black lines) in the east Skaftárketill water column under boreholes A (circles) and B (triangles). Gray lines are profiles of normalized SiO2, Na+ and Cl− concentrations, indicating that these biologically inert species are well mixed in the water column, and that sulfate is being rapidly produced and/or consumed. The mean concentrations of sulfate are 2.58 p.p.m. (a) and 0.96 p.p.m. (b).

Full size image

Biomass, cell and cultivable diversity

Cell counts and FISH results are summarized in the upper left panel of Figure 3. The bacterial cell concentration in the east lake samples average 4.4±2.2 × 105 ml−1 by microscopy (4.5 × 105 ml−1 by flow cytometry). Exclusion of sample B4 (with only 1.5 × 104 ml−1) yields an average of 5.1±1.3 × 105 ml−1. Taken together, Acetobacterium-, Sulfurospirillum- and Sulfuricurvum-specific probes hybridized with 34–100% of all cells in each sample. Hybridizations with the Paludibacter probe were detected only in sample B4 and at low significance (average 1 cell per field). Growth was observed in all enrichments incubated at 3 °C but not in the negative control (Supplementary Table S4). Ribosomal sequences in clone libraries derived from enrichments of two samples (A3 and B2 in WN2 and Toga media, respectively) had 80–99% homology to sequences in clone libraries from the east lake samples. None of 17 isolates from the 3 °C lake enrichments (Supplementary Table S4) was affiliated with clone sequences from lake samples (Supplementary Table S4 and Supplementary Figure S3). Growth at higher temperature was observed only for enrichments from sample B4 (in C–J, A–J and 162-1 media at 60 °C, and in Toga, 162-1, C–J, A–J, WN2 and YPS media at 80 °C). Sequences in a clone library from the hot enrichment are related to cultivated or uncultivated thermophiles (Supplementary Table S5); 74% show 99% homology to the uncultivated bacterium SRI-280 originally detected in an Icelandic hot spring (Skírnisdóttir et al., 2001) and affiliated with Caldiserica (formerly OP5) (Mori et al., 2009).

Figure 3
figure 3

Taxonomic affiliations in seven east Skaftárketill lake samples (A1–B4) and a jökulhlaup sample. Clockwise from upper left: cells hybridizing with taxon-specific FISH probes relative to universal bacterial (EUB I-II-III) probes, OTUs in clone libraries, OTUs in v4–v6 pyrosequencing libraries and OTUs in v3–v5 libraries (the latter two defined by a 3% average sequence distance cutoff). Bars are labeled by the sample and the number below each label is its depth below the glacier surface (in meters). The number to the right of the FISH bars is the total cell counts in 105 ml−1, and the height of each bar is proportional to the count. The number to the left of the clone bars is the size of the library. FISH was not performed on the jökulhlaup sample and data from the west Skaftárketill lake (Gaidos et al., 2009) are shown instead.

Full size image

Molecular (DNA-based) diversity

No polymerase chain reactions with archaeal primers yielded detectable product. Bacterial clone libraries from the east Skaftárkatlar lake are dominated by Acetobacterium- and Sulfuricurvum-related sequences, with a smaller number of clones related to Geobacter, Sulfurospirillum and Desulfosporosinus (upper right panel of Figure 3). The vast majority of pyrosequencing reads are also taxonomically affiliated with these five taxa (Figures 3 and 4 and Table 1). Community richness and evenness statistics based on our pyrosequence libraries are reported in Supplementary Table S6. Clustering with a cutoff of 3% average sequence difference assigns reads from our east lake v3–v5 and v4–v6 libraries to 321 and 1500 OTUs, respectively. Sequences from the aggregated v6 libraries from three east lake samples cluster into 154 OTUs (cutoff of 10%). More than 90% of reads in all of these libraries are assigned to the five dominant taxa (Figure 4 and Table 1). This is analogous to the statistics of the v6 library from the west Skaftárketill lake (Gaidos et al., 2009), which was re-analyzed alongside the other v6 libraries. However, not all major taxa are the same (Figure 4). Thermus- and Paludibacter-related sequences constitute 9.7% and 8.3%, respectively, of the west lake library, but «1% of the east lake libraries, while Geobacter-related sequences are comparatively rare (∼0.6%) in the west lake library.

Figure 4
figure 4

Composition of v6 pyrosequencing libraries for samples from the west and east Skaftárkatlar subglacial lakes, a jökulhlaup in the Skaftá river and the Kverkfjallalón subaerial lake. OTUs used was defined by a 10% average difference clustering criterion and their Bayesian classification used the VAMPS v.6 references database (Huse et al., 2008) based on the Silva taxonomy (Pruesse et al., 2007).

Full size image
Table 1 Major taxa in summed analyses of east Skaftárketill lake samples (%)
Full size table

The lake community is extremely uneven, as demonstrated by a rank abundance plot of the long-read OTUs (Supplementary Figure S4). This is also indicated by low values of the Shannon’s H index (1–1.8) and high values of Shannon’s D index (0.28–0.49) (Supplementary Table S6). Both ACE and Chao1 estimators predict that the actual OTU richness is several thousand, that is, 3–4 times the observed values, but rarefaction analysis shows that, despite the depth of sequencing, in none of the libraries has sampling reached saturation (Supplementary Figure S5). This is a result of the extreme unevenness of the taxonomic distribution: continued sequencing only yields more reads of the same dominant taxa.

The fractional abundances of the major OTUs in the pyrosequence libraries are in quantitative agreement (Supplementary Figure S6). However, the observed and inferred numbers of OTUs in all v4–v6 libraries (and their aggregate) are higher than in the respective v3–v5 libraries (Supplementary Figure S5 and Supplementary Table S6), despite similar library sizes and a uniform clustering criterion. The dashed line in the inset of Supplementary Figure S4 shows that the v4–v6 excess is in low-abundance OTUs. However, richness estimates are sensitive to clustering criterion (Kim et al., 2011), and increasing the cutoff to 4% in the aggregate v4–v6 library decreases the number of OTUs to 615, while decreasing it to 2% in the v3–v5 library increases the number of OTUs to 1179. Evidence for differences between lake samples is marginal: variation in effort-adjusted ACE* between v3 and v5 libraries is not significant (χ2=3.9, N=7), but that between v4 and v6 libraries is significant (χ2=95.2, N=7) (Supplementary Table S6). Pyrosequencing libraries for a given locus are indistinguishable from one another based on Czekanowski–Bray–Curtis distances (95% confidence). UPGMA trees constructed from Morisita–Horn distances between libraries finds relationships that are congruent between v3–v5 and v4–v6, and with strong bootstrap support (Supplementary Figure S7), there is no apparent relation with depth or water chemistry, and this may be an artifact of extraction efficiency.

Clone and long-read pyrosequencing libraries from the jökulhlaup sample are dominated by a single Sulfuricurvum-related taxon with ⩾90% of clones/reads and the unevenness of the abundance distribution is similar to the east lake. The observed diversity is higher than in the east lake, although the predicted richness is roughly the same. Sulfuricurvum is not nearly as dominant in the v6 library and the observed and predicted richness are intermediate of the two long-read libraries. The Kverkfjallalón v6 library consists mostly of a single taxon affiliated with genus Xenophilus in the Comamonadaceae (Betaproteobacteria). A heat map and dendogram based on Bray–Curtis distances between all the v6 libraries clearly demonstrates the similarity of the west and east lake communities and the distinctiveness of the jökulhlaup and Kverkfjallalón samples (Figure 5).

Figure 5
figure 5

Heat map and relational dendogram based on Bray–Curtis distances between v6 pyrosequence libraries for three samples from the east Skaftárketill lake, the west lake, a jökulhlaup sample and the Kverkfjallalón subaerial lake. Lighter-shaded squares indicate more similar pairs of libraries.

Full size image

As expected, cultivable diversity is only a small fraction of total diversity in these environments. Only 13.7% (3%) of v3–v5 (v4–v6) OTUs have relatives (⩾97% identity) in clone libraries from our 3 °C anaerobic enrichments, and these OTUs represent only 1.0% (1.6%) of all reads in the pyrosequence libraries. In all, 82% (97%) of these v3–v5 (v4–v6) reads belong to a single Desulfosporosinus-related taxon. Conversely, all but 2 of the 62 clones from our enrichments have relatives in the pyrosequence libraries; these are affiliated with Acetobacterium, Desulfosporosinus, Geobacter, Sulfurospirillum and Sulfuricurvum. Nine (82) OTUs in our v3–v5 (v4–v6) libraries have relatives in the clone library from our 60/80 °C B4 enrichment, but these comprise only 1.2% (3.5%) of all reads. Nearly all (99.5% (97%)) such v3–v5 (v4–v6) sequences belong to a single OTU classified by the Ribosomal Database Project as Caldiserica. Isolates from cold enrichments (data not shown) represent only 0.36% (0.02%) of all reads in our v3–v5 (v4–v6) libraries.

Discussion

Paired libraries of v3–v5 and v4–v6 pyrosequences, v6 pyrosequence libraries, clone libraries and FISH give a concordant description of the bacterial community in the east Skaftárketill lake (Figure 3 and Supplementary Figure S6, as well as Table 1). Our data also demonstrate that the east and west Skaftárkatlar lakes are dominated by a few bacterial taxa affiliated with the genera Acetobacterium, Geobacter, Sulfurospirillum, Sulfuricurvum and Desulfosporosinus (Figures 3 and 4). Thermus- and Paludibacter-affiliated sequences are major components of the west Skaftárketill library but not the east Skaftáketill libraries, although they are present. This difference may reflect actual variation in community composition in time or space, but it is based on a single west Skaftá sample.

There are also differences between v3–v5 and v4–v6 libraries generated from the same sample: The greater taxonomic richness of the v4–v6 libraries compared with the v3–v6 libraries (Supplementary Table S6) cannot be ascribed to library size (Kemp and Aller, 2004), and Schloss (2010) found that the observed number of OTUs increased along the 16S gene (v1–v9) at the same genetic distance (but see Kim et al., 2011). Moreover, our v4–v6 primers may span greater taxonomic richness than the v3–v5 primers. The uncultivated TM7 division appears in our v4–v6 but not v3–v5 libraries, although ProbeCheck (Loy et al., 2008) indicates this is not an artifact of primer coverage. These results caution against interpreting richness estimates based on a single locus.

Our failure to detect a significant archaeal population in these lakes stands in contrast to other subglacial environments in which archaea have been detected (Boyd et al., 2010; Stibal et al., 2012), and archaeal-dominated subsurface chemolithotrophic communities (Chapelle et al., 2002). In all, 9 reads out of ∼382 000 in our bacterial pyrosequence libraries assigned to the Archaea may be a result of nonspecific amplification, and it is possible that archaea are present in a benthic community we have not yet sampled. Growth in 3 °C enrichments demonstrates that some lake taxa are cold-adapted; we speculate that differential response to low temperature (Cavicchioli, 2006) and/or level of ambient energy (Valentine, 2007) might be responsible for the bacterial dominance in these lakes.

The Skaftárkatlar community differs markedly from those in the Kverkfjallalón lake and the jökulhlaup in the Skaftá river (Figure 4). The Kverkfjallalón water column is dominated by a single affiliate to Xenophilus (Comamonadaceae), a group of aerobic soil betaproteobacteria that are widespread in subglacial environments (Yde et al., 2010). With one exception, the Kverkfjallalón pyrosequence library is devoid of the major Skaftárkatlar taxa, while the jökulhlaup assemblage is a mix of the two communities (Figures 3 and 4). The Sulfuricurvum-affiliated taxon is ubiquitous in our Vatnajökull samples (Table 1). The cultivated representative of this genus is a microaerobic sulfur-oxidizing lithoautotroph (Kodama and Watanabe, 2004), traits that may promote its widespread distribution. The genome sequence should provide clues about this organism’s metabolic versatility (Lucas et al., 2010).

Phylogenetic analysis of sequences in respective clone libraries from our samples reveals that the major Vatnajökull OTUs are each other’s closest known relatives (Figure 6). This points to efficient contact between the Skaftárkatlar and Kverkfjallalón lakes, despite their separation. Contours of the water driving potential at the glacier bed show the lakes to be separated by barriers exceeding 50 m and there is no direct contact between them at the glacier bed (Figure 7). We speculate that exchange occurs by movement of groundwater through the permeable tephra and basalt that underlies western Vatnajökull (Flowers et al., 2003). Unattached cells are readily transported through basalt (Entry and Farmer, 2001; Marteinsson et al., 2001). The high similarity of Sulfuricurvum sequences from all of our samples and enrichments suggests that it derives from a single population beneath Vatnajökull. We propose that these subglacial lakes were colonized from a deeper aquifer, rather than from the overlying ice, and that the Skaftárkatlar community was selected from this deeper biota by factors such as anoxia, low temperature, high hydrogen and the slow introduction of oxygen. Successful enrichment of thermophiles from samples and identification of thermophile relatives supports this hypothesis, but a direct investigation that samples the aquifer beneath Vatnajökull is needed.

Figure 6
figure 6

Neighbor-joining trees of sequences from our Iceland 16S rRNA clone libraries, databases and cultivates showing phylogenetic relationships within OTUs affiliated with (a) Acetobacterium, (b) Sulfurospirillum, (c) Sulfuricurvum and (d) Geobacter. The scale bar represents the expected % of substitutions per nucleotide position and a marine crenarchaeon was used as an outgroup. The cluster in Acetobacterium represents 256 clones with >99% sequence similarity (137 and 119 from the west and east Skaftárkatlar lakes, respectively). The cluster in Sulfurospirillum contains 25 west and 8 east lake clones with >99% similarity. The cluster in Sulfuricurvum contains 84 west lake (17 from enrichments at 3 °C), 196 east lake, 22 jökulhlaup and 30 Kverkfjallalón lake clones with >99% similarity. The cluster in Geobacter contains 1 west and 8 east lake clones with >99% similarity.

Full size image
Figure 7
figure 7

Contours of the water driving potential (in meters) (Cuffey and Paterson, 2010) derived from a 2010 LiDAR map of Vatnajökull in the vicinity of the Skaftárkatlar lakes superimposed on a hillshade of the ice surface shortly after both lakes were drained in jökulhlaups (data from the Icelandic Meteorological Office). North is up, the map is about 10 km across, and the whiter area is missing data. The calculations use the bedrock geometry of Björnsson and the water levels in boreholes to the lakes in 2006 and 2007 as reference values. The lakes are separated by a potential barrier exceeding 50 m in height, and independently drain through subglacial channels to the southwest. Another barrier exceeding 100 m separates these from the Kverkfjallalón lake 40 km to the northeast. The same features appear in a map constructed from a 1998 aerial SAR survey (data from the Institute of Earth Sciences) when the Skaftárkatlar lakes were nearly full (data not shown).

Full size image

We speculate that the dominant bacterial taxa in the oligotrophic Skaftárkatlar lakes fill niches in a chemolithotrophic ecosystem that includes sulfur cycling and is fueled by H2, CO2, Fe(III) and O2 (Figure 8). On the basis of the physiologies of cultivated relatives, we propose that the Acetobacterium-related taxa are homoacetogens, combining CO2 and H2 into acetate (CH3COOH). Homoacetogens can outcompete methanogens at low temperature and high H2 concentration (Kotsyurbenko et al., 2001; Drake et al., 2002). We also propose that the Geobacter taxon carries out ferric iron reduction using H2 as its energy source (Nevin et al., 2005). Further, that the relative of Sulfuricurvum kujiense oxidizes sulfide, elemental sulfur (S0) and thiosulfate (S2O3) at low oxygen concentrations (Kodama and Watanabe, 2004). We propose that the Desulfosporosinus member uses H2 to reduce sulfate, thiosulfate or sulfite (SO3−), to elemental sulfur (Ramamoorthy et al., 2006; Vatsurina et al., 2008; Lee et al., 2009; Alazard et al., 2010; Abicht et al., 2011), whereas the Sulfurospirillum member completes the sulfur cycle, reducing elemental sulfur, thiosulfate and sulfite, to sulfide (Finster et al., 1997; Stolz et al., 1999; Jensen and Finster, 2005; Kodama et al., 2007).

Figure 8
figure 8

Schematic of possible chemolithotrophic pathways in the Skaftárkatlar lakes and proposed assignments of the major taxa based on established metabolisms of cultivated relatives. White pathways consume O2; gray pathways consume H2. Fe(II) will rapidly react with sulfide to form insoluble pyrite (Stumm and Morgan, 1996). No taxon has been associated with the knallgas reaction (of H2 with O2).

Full size image

Future investigations of the Vatnajökull lakes can test these hypotheses by measuring concentrations of relevant substrates, determining rates of activity and establishing the physiologies of the major taxa. Sulfate exhibits a concentration gradient in the east lake (Figure 2), indicating that it is produced and/or consumed more rapidly than the lake mixes. However, in addition to biological sulfur cycling, sulfate can be formed by abiotic sulfide oxidation and SO2 disproportionation, and removed by dilution with glacial melt. The Skaftárketill lakes contain volcanic CO2 and presumably Fe(III) (in mafic minerals), and we are also reasonably certain that O2 enters the lakes as the overlying ice shelf melts but have not detected it, probably because it reacts with sulfide to sub-micromolar levels on the ∼1-year residence time of the lakes (Jóhannesson et al., 2007). Although, we expect H2 from serpentinization and we detected it in the west lake, we have been unable to quantify this sparingly soluble gas because of loss to bubbles during depressurization of retrieved samples. In situ profiling of O2 and H2, the former at nanomolar sensitivity (Revsbech et al., 2009), should be a high priority of future investigations.