SSU rRNA gene methods

Environmental DNA was extracted from the sediments using the fastDNA Spin kit for sediments (Q-biogene, Carlsbad, CA). PCR reactions for amplification of community bacterial SSU rRNA genes contained environmental cave DNA, 1 X PCR buffer, 3 mM MgCl₂, 250 μM of each dNTP, 10 μM of each primer (27f and 1492r; Table ​Table1),1), and 0.5 U of Taq polymerase (DYnazyme II, Thermo Fisher Scientific, Lafayette, CO). Cycling conditions were a 94°C hot start for 2 min, followed by 25 cycles of 94°C for 30 s, 50°C for 30 s and 72°C for 2 min, and completed with a 5 min 72°C extension. For construction of SSU rRNA gene libraries for each cave sample, PCR products from multiple (4–8) clean PCR reactions (i.e., showing only a band of the expected size) were pooled, and purified using a PCR purification kit (QIAGEN, Valencia, CA). Libraries were generated using a TOPO TA cloning kit and chemically competent Escherichia coli TOP10F cells (Invitrogen, Carlsbad, CA). Cloned SSU rRNA genes were sequenced from both directions using primers T3 and T7, and the data used to assign OTUs using DOTUR (Schloss and Handelsman, 2005); a 97% sequence identity was used for OTU classification. One to three representative clones were chosen for each OTU and bidirectional sequencing of the complete gene performed (primers 704F and 926R, Table ​Table1).1). Taxonomic assignments were determined to the phylum level using the SILVA search and classify function (v1.2.11) with a minimum identity value of 0.80 using the SILVA taxonomy for classification (Pruesse et al., 2012). A maximum likelihood tree was then calculated using RAxML v7.2.6 (Stamatakis, 2006) using the general time reversible model of nucleotide substitution (Tavaré, 1986) for OTUs representing more than two clones. The tree was calculated 10 consecutive times using randomized starting trees to find the tree with the greatest likelihood to be the most representative. The tree files were then visualized within the software Mega and annotated using the vector-drawing program Inkscape.

Q-PCR methods

Bacterial biomass was estimated using a method modified from Einen et al. (2008) where ribosomal RNA genes are quantified using quantitative PCR (Q-PCR) and extrapolated to cells/g of material. Q-PCR reactions were performed using a StepOnePlus real-time PCR system (Life Technologies) with primers Univ533F and Bact684R (Table ​(Table1),1), DyNAmo Flash SYBR green PCR master mix (ThermoFisher Scientific, Waltham, MA) with absolute quantification against a standard curve. Each reaction contained 1X master mix, 0.5 μM of each primer, 1X ROX reference dye, and 1ng of environmental cave DNA or a serial dilution of plasmid DNA. Q-PCR reactions were 95° for 7 min as a hot start, followed by 35 cycles of 95°C for 5 s and 60°C for 30 s. PCR efficiency was 84%. Bacterial counts were then extrapolated to biomass using the conversion factor of 3.9 copies of rRNA genes/bacterial genome (Einen et al., 2008).

RubisCO and aclB

PCR amplification was attempted for the two main types of autotrophic CO₂ fixation mechanisms, RubisCO (Calvin-Benson-Bassham cycle) and ATP citrate lyase (reverse TCA cycle) to determine the possible presence of different carbon fixation pathways (Table ​(Table1).1). Each PCR reaction contained 1 U of Phire polymerase (Thermo Fisher Scientific, Waltham, MA), 1 X PCR Buffer, 200 μM dNTP, 2 μM each primer, 5 μg BSA, and 1 ng/μl environmental cave DNA. Cycling conditions began with a 2 min hot start at 98°C, followed by 30 cycles of 98°C for 10 s, 60°C for 20 s, and 72°C for 30 s and completed with a 2 min extension at 72°C. PCR product from positive reactions were excised from a 1.5% agarose gel and purified using the GeneJET extraction kit (Thermo Fisher Scientific). Purified PCR products were then ligated into the pJET1.2 vector (Thermo Fisher Scientific) at a 3:1 ratio and transformed into chemically competent cells (Active Motif, Carlsbad, CA). Both strands of inserts from 48 clones were sequenced from each clone library using primers Pjet1.2F and Pjet1.2R. The sequences were then translated and aligned with representative sequences from described isolates and uncultured clones. Phylogenetic trees were calculated using RaxML v7.0.3 (Stamatakis, 2006) using the Whelan and Goldman model of amino acid substitution (Whelan and Goldman, 2001).

Nucleotide sequence accession numbers

The SSU rRNA and the Rubisco sequences representing the OTUs analyzed in this paper have been submitted to GenBank and assigned accession numbers KJ623626—KJ623652 and KJ623653—KJ623698, respectively.

Bacterial abundance and microbial diversity based on SSU rRNA genes

We used molecular phylogenetic analysis to examine the microbial communities in the three caves. The bacterial small subunit (SSU) ribosomal RNA gene was PCR amplified using domain specific primers (Table ​(Table1).1). PCR amplifications using Archaea-specific primers were negative for all three caves. Bacterial numbers for the cave samples determined by quantitative PCR (Q-PCR) of small subunit ribosomal genes (Einen et al., 2008) were 1.6–40 × 10⁶ cells/g sediment (Table ​(Table3),3), or up to about 10 fold higher than found in the glassy rind of seafloor basalts (Einen et al., 2008). The PCR products were cloned and sequenced (results for the Eukarya are reported elsewhere, Connell and Staudigel, 2013). For Harry's Dream, Warren Cave, and Hubert's Nightmare we sequenced 102, 82, and 78 clones which grouped into, respectively, 34, 11, and 18 Operational Taxonomic Units (OTUs; 97% sequence identity cutoff) (Tables ​(Tables44–6) and Chao1 indices of 70.1, 35.5, and 58.5. Distinct Bacterial communities were identified in each cave (Figure ​(Figure2).2). The phylogeny of the organisms found in the clone libraries was frequently similar to cultured organisms or 16S rRNA gene sequences from other volcanic, cave, sediment or cold environments (Figure ​(Figure33).

Open in a separate window

Figure 2

Pie chart showing the percentage of the major phyla identified in each cave by 16S rDNA analysis. The identification of the other, minor phyla are given in Tables ​Tables44–6.

Open in a separate window

Figure 3

Phylogenetic tree of the small subunit SSU ribosomal RNA genes of Bacteria from the caves.

The highest diversity of clones was found in Harry's Dream (Figure ​(Figure2,2, Table ​Table4).4). This finding is consistent with Harry's Dream being a more energy (light)-rich environment and, in fact, it was the only cave where representatives of well-recognized phototrophs (Cyanobacteria) were abundant. Chloroflexi were also abundant in this cave, but while there are known phototrophic members in this phylum they are anoxygenic phototrophs which would not be expected to be present. Rather, the vast phylogenetic diversity in the phylum Chloroflexi suggests broad metabolic capabilities within this group. The Harry's Dream Chloroflexi were dominated by the Ktedonobacteria class of Bacteria, representing about 58% of the Chloroflexi sequences (Table ​(Table4).4). Harry's Dream also had the only group of unclassified Bacteria for any of the three caves. The other abundant phyla include the Proteobacteria (primarily Betaproteobacteria) and the Acidobacteria.

In Warren Cave, the majority of the sequences fell into two phyla, the Chloroflexi (almost entirely Ktedonobacteria) and the Acidobacteria, representing about 66% and 31% of all the sequences, respectively. Only two minor OTUs were from other phyla: OTU4 was closely related to various Hyphomicrobium spp. (Alphaproteobacteria), a taxon known to oxidize manganese (Ghiorse and Ehrlich, 1992), while OTU10 (Planctomycetes) displayed no cultured relatives. Chloroflexi 16S rRNA gene sequences are often found in environmental DNA surveys and metagenomes (Huber et al., 2006), even in Antarctica (Cary et al., 2010; Pearce et al., 2012), but are usually only a minor part of the microbial community, so finding them in such relative abundance here was surprising. Acidobacteria are often the most abundant bacterial phylum in soils (Janssen, 2006) and are generally thought to be versatile aerobic heterotrophs adapted to low-nutrient conditions (Ward et al., 2009). They were the only group present in all three caves. Both the Chloroflexi and Acidobacteria phyla contain a large diversity of uncultured representatives, so the scope of the metabolic diversity within these phyla remains unknown, but their abundance implies that the Warren Cave environment provides conditions for their success.

Interestingly, in Hubert's Nightmare, a dark, but much colder cave with significantly lower airflow, the microbial community was quite distinct from Warren Cave even though they are adjacent to (within about 50 m of) each other. There were several phyla that were present or in much greater abundance than the other caves: Verrucomicrobia, Actinobacteria, and Bacteroidetes. There was also an abundance of Proteobacteria clones in this sample that included roughly equal percentages of Betaproteobacteria and Alphaproteobacteria; Hubert's Nightmare was the only cave with more than two clones from the Alphaproteobacteria. Whether the difference in the microbial community in Hubert's Nightmare as compared to Warren Cave was due to its shallower sediment depth, its lower degree of weathering (the sediment was coarser), the colder temperature or some other environmental factor can only be speculated upon.

Screening for potential chemolithoautotrophy based on carbon fixation genes

We screened for the presence of potential chemolithoautotrophic microorganisms in the three caves using PCR amplification of key genes for the dominant autotrophic CO₂ fixing pathways: Those encoding ribulose-1,5-bisphosphate carboxylase/oxygenase (RubisCO) form I and form II (cbbL and cbbM, respectively), the primary CO₂ fixing enzyme found in aerobic chemoautotrophs, and for ATP citrate lyase (aclB) diagnostic of the reverse tricarboxylic acid (rTCA) cycle, another key pathway important in CO₂ fixation in both aerobes and anaerobes. RubisCO cbbL genes were detected, while PCR amplification of aclB was negative in all samples. Among the known types of RubisCO there are two major forms that catalyze net CO₂ fixation (Tabita et al., 2008). Form I, comprised of 8 large and 8 small subunits, is typically found in aerobic bacteria and chloroplasts and is believed to be the more recently evolved aerotolerant form with a higher specificity for carboxylase activity than for oxygenase activity. Among the Form I RubisCOs are the “red” and “green” types. The green type Form I RubisCO is the common RubisCO in plants, green algae, cyanobacteria and some purple bacteria while the red type Form I RubisCO is found in non-green algae and includes some autotrophic Alpha- and Beta-proteobacteria. Form II, comprised of large subunit dimers, is most frequently found in organisms living in environments with relatively high CO₂ concentrations, including anoxic and low oxygen environments and is generally believed to be the more ancient form. Three additional forms of RubisCO, forms III, IV and a hybrid form II/III, whose role in CO₂ fixation remains unclear have been described (Tabita et al., 2008; Wu et al., 2009). Most of the RubisCO sequences found were related to the form I red type RubisCO (Figure ​(Figure4).4). The majority of the Form I green type RubisCO were from Harry's Dream, which is not surprising given that it is seasonally influenced by sunlight. The most unexpected results from our study was the discovery of highly diverse form I (cbbL) RubisCO gene sequences most deeply rooted in the red-like clade that either cluster with sequences from uncommon (thermo)acidophilic species [e.g., Warren Cave (I) OTU1] or possibly from completely new clades [e.g., Warren Cave (I) OTU2] (Figure ​(Figure4).4). No form II (cbbM) RubisCO gene sequences were detected. Overall, these results are consistent with the caves being a fully oxic environment. The large number of diverse RubisCO sequences in the deeply rooted red clades having no or few closely related cultured representatives combined with the relatively low phylogenetic (16S SSU rRNA) diversity found in the caves (Figures ​(Figures2,2, ​,3)3) is paradoxical and is discussed further below.

Open in a separate window

Figure 4

Phylogenetic tree of RubisCO genes showing the sequences from the three caves in relation to previously known sequences. Inset: a conceptualized version of the phylogenetic tree of the RubisCO genes showing the main branches of RubisCO gene sequences from cultured organisms (red lines), previously identified uncultured environmental sequences (green lines) and novel sequences from the Mt. Erebus caves (blue lines).

Warren cave

Warren Cave is dark with highly oligotrophic sediments, making it an excellent example of a DOVE and a model system for studying microbial ecosystems deriving their essential energy sources from inorganic substances including those in rocks [e.g., Fe(II), Mn(II)] (Kelly et al., 2008) and volcanic gasses. That there is a low diversity of microbes in this cave makes it easier to evaluate the potential energy yielding pathways in this ecosystem from what is known about the metabolisms found in the Chloroflexi and Acidobacteria and their genomes. Mt. Erebus magmatic gas emissions are rich in CO₂, contain CO and H₂, but are essentially devoid of CH₄, and H₂S (Oppenheimer and Kyle, 2008; Moussalam et al., 2012). Thus gaseous CO as seen in other volcanic systems (King and Weber, 2007), H₂ and reduced metals in the rocks appear to be the primary energy sources available to the microbial communities. To our knowledge CO has not been detected in the caves, although its absence could be attributed to efficient microbial utilization. Trace amounts of H₂ and CH₄ above air background have been detected in some caves (Fischer et al., 2013).

The phylogenetic affiliations of the major Warren Cave bacterial OTUs are consistent with the oligotrophic nature of the ecosystem. The phylogenetic diversity of Bacteria is low and there is an unusual abundance of two major phyla, the Chloroflexi and the Acidobacteria. There is no evidence of human contamination such as sequences of clones from microbes associated with the human body. Although the SSU rRNA clone library results cannot be taken to quantitatively indicate the abundance of a given taxon because of varying SSU rRNA gene copy number in different organisms, the dominance of these two phyla suggests a relatively low diversity environment and that these organisms are highly adapted to local conditions. In contrast, the large diversity of RubisCO cave sequences indicates there are a variety of organisms with autotrophic carbon fixation potential. Reconciling the SSU rRNA phylogeny with the abundance of RubisCO genes represents an enigma. As far as is known, Acidobacteria do not grow chemoautotrophically although they are recognized as being oligotrophic and possess predicted cox genes encoding CO dehydrogenase (CODH) (King and Weber, 2007; Ward et al., 2009), suggesting they may utilize CO to supplement their energy needs. Within the Chloroflexi at least one member (Oscillochloris trichoides DG6) has RubisCO (Ivanovsky et al., 1999) and this gene sequence groups with our newly-discovered, deeply rooted cave RubisCO sequences (Figure ​(Figure4).4). Major groups within the Chloroflexi (like Ktedonobacter racemifer, to which two of the major Warren Cave OTUs, OTUs 1 and 3, are phylogenetically related; Figure 5) also possess the CODH large subunit gene coxL (DOE Joint Genome Institute, http://www.jgi.doe.gov/) (Weber and King, 2010; King and King, 2014). Thus, CO oxidation appears to be a common, but not universal trait of organisms in the phylum Chloroflexi (King and King, 2014) and may also be important to the energetics of Mt. Erebus DOVEs. Many aerobic CO oxidizers with coxL also possess RubisCO for CO₂ fixation (King and Weber, 2007) although many are not able to grow with CO as the only energy source. Interestingly, Ktedonobacter and related species are the most dominant groups in the high altitude volcanoes in the Atacama desert (Lynch et al., 2012) and on cinder deposits of Kilauea volcano in Hawaii (Weber and King, 2010; King and King, 2014) clearly indicating the versatility of this group of organisms to be adapted to extremely oligotrophic conditions and withstand extreme environments. Although Chloroflexi and Acidobacteria are found relatively nearby in soils from the Antarctic Dry Valleys or the hydrothermally active Tramway Ridge (Mt. Erebus; Figure ​Figure1),1), their abundance there is much lower and the similarity of the sequences with the Warren Cave SSU rRNA sequences is low (<83%). We postulate that the Chloroflexi CO oxidizers and some of the other minor bacterial taxa are responsible for autotrophic CO₂ fixation in this system.

But how do we reconcile the low 16S rRNA phylogenetic diversity of Warren Cave with the high diversity of RubisCO sequences? There are several possible reasons. First, the simplest explanation is that we undersampled the phylogenetic diversity of organisms by only analyzing 82 16S rDNA clones but have a better representation of the RubisCO from the 48 clones sequenced from each cave. This would be especially true if the two dominant groups, Acidobacteria and Chloroflexi, have a large number of copies of the small subunit (16S) ribosomal genes biasing our phylogenies toward those groups. In particular, the abundance of Ktedonobacteria-like sequences within the Chloroflexi may be overrepresented because they have a very large bacterial genome with eight copies of the 16S rRNA gene (Chang et al., 2011). In contrast, the Acidobacteria have only one or two copies (Lee et al., 2009). A second explanation is that the most abundant organisms by 16S rRNA gene analysis are not the organisms with the RubisCO genes (nor do they have the rTCA cycle for CO₂ fixation which also would have been detected) and our PCR methods are selectively amplifying the RubisCO genes from the remaining low abundance autotrophic microbial populations present in the cave. Thirdly, although unlikely, it is possible that Warren Cave was successfully colonized by a relatively low diversity of microbes (at the phylum level), but, with time, their functional genes have further evolved and diverged relative to their ribosomal genes. Something similar to this has been proposed to explain the high level of sequence heterogeneity in soils from Mars Oasis, Alexander Island, Antarctica (Pearce et al., 2012). Finally, it could be that this discrepancy is simply due to the resolution at which we are analyzing sequences, grouping ribosomal sequences at higher taxa rankings than which the RubisCO sequences are grouped. A combination of these explanations is likely.

A concurrent study of the fungi based on ITS sequences in Warren Cave revealed moderate fungal diversity with all taxa belonging to the phyla Basidiomycota and Ascomycota (Connell and Staudigel, 2013). Many of those identified are known to function as saprophytes but fungi are also known to oxidize CO (Conrad and Seiler, 1980), abilities compatible with an oligotrophic lifestyle. One of the most common taxa found was a black yeast, Aureobasidium pullans, commonly found on rocks in polar habitats and other oligotrophic habitats (e.g., Atacama desert, Himalayas) (Onofri et al., 2007). Although diverse fungi were undoubtedly present, from the results so far we are not able to evaluate the relative abundance of Bacteria and fungi in any of the caves examined.

Hubert's Nightmare

Hubert's Nightmare is a small cave system that was visited in 2008 and 2010. In 2012 it was inactive and frozen over. Because of the close proximity of Hubert's Nightmare to Warren Cave (less than 50 m, in the same slope area of Mt. Erebus) and the uniformity of sediment compositions of most other Mt. Erebus ice caves (Table ​(Table2),2), we expect Hubert's Nightmare to have the same sediment composition as Warren Cave. Thus, the main differences between the two caves are the temperature of the fumarolic vents where the sediments were collected (also creating differences in cave humidity) and the degree of weathering of the sediments sampled. Hubert's Nightmare sediments had both a lower temperature and amount of weathering (coarser grain size) than Warren Cave sediments, as well as a slightly higher pH. The lower temperature indicates a lower heat (and hence gas) flux from the vents. Therefore, when compared to Warren Cave, the overall gas chemistry in Hubert's Nightmare is likely to be closer to air because vent gasses will be subjected to more dilution by the atmosphere. This lower flux of potential gaseous energy sources combined with the higher pH may account for the differences in microbial community composition and cell numbers as compared to Warren Cave. Additionally, the cell abundance for Warren Cave (Table ​(Table3)3) may be inflated if the Ktedonobacteria have more copies of the SSU 16S ribosomal RNA gene (Chang et al., 2011). Although the major microbial groups represented in the clone libraries are not recognized for having chemolithoautotrophic metabolism, like in Warren Cave, there was a large diversity of type I RubisCO sequences present indicating the potential exists in the Hubert's Nightmare microbial community. We do not know whether some of these RubisCO sequences could belong to members of the Verrucomicrobia, at least one member of which is an autotrophic methanotroph that uses the CBB cycle for CO₂ fixation (Khadem et al., 2011).

Harry's Dream

Evidence for photosynthetic microbes was found in Harry's Dream, the only cave we studied that receives periodic light. Regardless, the TOC content and cell numbers were lower than found in Warren Cave, even though the TOC is comparably low for both sites (Table ​(Table5).5). In both caves we would expect a similar demand for the organic carbon from heterotrophic bacteria, so we do not believe this could explain the difference in TOC content or cell abundance. Rather, it could be due to the location of the fumaroles relative to the surrounding rock topology. In Harry's Dream the fumarole was located on a small ledge that is relatively elevated when compared with the surrounding topography. In Warren Cave the fumarole was in a depression located under a ledge where melt water and condensing steam running down lava pillows could help transport organic carbon and cells from surrounding areas and collect in the depression. Alternatively, the higher cell numbers for Warren Cave relative to Harry's Dream could simply reflect the greater abundance of organisms with high SSU rRNA gene copy number (e.g., the Chloroflexi).

A more curious aspect of Harry's Dream cave is the high abundance of 16S SSU gene sequences (OTUs 1 and 2) closely related to deeply branching cyanobacteria from a clade (WD272) that has never been characterized (Figure ​(Figure3)3) coupled with the complete lack of correlated cyanobacteria-related RubisCO sequences. Green type RubisCO sequences were found in Harry's Dream but they were more closely related to RubisCOs from known chemolithoautotrophs (Figure ​(Figure4).4). Thus WD272 (1) may possess RubisCO genes more closely related to chemolithoautotrophs, (2) they do not use the CBB cycle, or (3) they are not actually autotrophs.

The abundance of sequences within the phylum Chloroflexi in Harry's Dream led us to consider whether some of these sequences could be representative of phototrophic bacteria. However, all the sequences belong to classes other than the photosynthetic class (also called Chloroflexi). Nevertheless, because of the broad phylogenetic diversity represented within the phylum Chloroflexi and the known metabolic versatility within the class Chloroflexi, the possibility for photoautotrophy or photoheterotrophy within other taxa within the phylum in Harry's Dream is possible. In addition to the cyanobacteria, some Acidobacteria are known to be phototrophic (Bryant et al., 2007).

Two of the OTUs (6 and 13) from Harry's Dream phylogenetically clustered in the Hydrogenophilales, an order known for being thermophilic and capable of oxidizing hydrogen (Hyashi et al., 1999). The cultivated organisms from this order have an optimum temperature of ~50°C and are facultative chemolithoautotrophs. Although the temperature of the soils we collected was low compared to the optimum temperatures for the cultured isolates, it is not uncommon to find thermophilic bacteria or bacteria phylogenetically related to thermophilic bacteria in cold environments. Additionally, that these organisms are facultative chemolithoautotrophs would seem to be a good lifestyle for organisms living in ice caves where changing conditions such as gas flux and temperature occur throughout the year. This observation, plus the abundance of RubisCO genes in Harry's Dream that are related to the red or deeply rooted red type RubisCOs found in the other caves, suggests that similar chemolithoautotrophs, even if in minor abundance, exist in all three caves.

We thank the mountaineers, Brian Hasebe and Alasdair Turner, who supported our field research on Mt. Erebus and Alasdair Turner for the photograph in Figure ​Figure1.1. This work was supported by the NSF Office of Polar Programs awards ANT-0739731 to BT, ANT-0739696 to LC and ANT-0739712 to HS. BT gratefully acknowledges fellowship support from Durham University's (U.K.) interdisciplinary Institute of Advanced Studies on their focused theme of “light,” in relation to the work here, life in the absence of light.