Photosynthesis is the primary driver of energy dynamics in terrestrial and marine ecosystems, where light energy is harnessed for the conversion of atmospheric CO2 to reduced carbon. Avenues for understanding alternate non-photosynthetic primary production strategies are limited to subterranean or deep-sea ecosystems that function in the absence of sunlight. A significant amount of research has been devoted to characterizing primary production in ecosystems such as hydrothermal vents (reviewed by Nakagawa and Takai, 2008) and sulfidic caves (Sarbu et al., 1996; Chen et al., 2009; Engel et al., 2010; Jones et al., 2012), where supplies of reduced sulfur, hydrogen or methane support rich chemolithoautotophic activity; however, the energy dynamics of carbonate caves are less well defined. Carbonate cave communities are presumed to be sustained by allocthonous carbon sourced from photic surface ecosystems and entering the cave with vadose-zone drip water, surface water flow or the behavior of macrofauna (Laiz et al., 1999; Simon et al., 2003; Barton et al., 2004). In contrast, limited-access carbonate caves in semiarid and arid regions are highly oligotrophic owing to low carbon levels in surface soils, low mean annual rainfall and small openings that prevent large scale macrofauna exchange with surface ecosystems. As such, these caves provide a window for analyzing the metabolic flexibililty of microbial communities in an aphotic oligotrophic habitat with potential similarity to diverse globally dominant terrestrial and marine environments, including subsoil to bedrock layers, oligotrophic aquifers and the deep ocean. Energy dynamics elucidated from these aphotic ecosystems may also be applicable to comparably oligotrophic photic systems such as arid deserts, glacial and polar ice, and even extraterrestrial planetary subsurface environments.

In addition, analysis of the energy dynamics of carbonate caves provides information concerning the potential influences of microbial activity on carbon sequestration in speleothems, the secondary carbonate deposits commonly found in caves. The latter application is particularly relevant to the widespread study of the isotopic composition of speleothems for reconstruction of recent (Quaternary) climate change (Wang et al., 2005). Microbial contributions to speleothem isotopic signatures are not well understood, however, research indicates that microbial activity enhances calcium carbonate precipitation (Contos et al., 2001; reviewed by Barton and Northup, 2007; Banks et al., 2010), and that carbon isotope fractionation rates vary with different microbial CO2-fixation pathways (reviewed by Berg et al., 2010). Research characterizing the functional profiles of speleothem microbial communities will enhance our understanding of the metabolic potential and energy dynamics of oligotrophic karst environments and provide critical information for applications such as the analysis of speleothem isotopic signatures.

Kartchner Caverns is a limited-access cave that developed within a Lower Carboniferous age Escabrosa limestone formation in the semiarid Whetstone Mountains of southeastern Arizona, USA (Jagnow, 1999). The only natural entrance is a small blowhole, largely limiting macrofauna exchange to bats, insects and small rodents, amphibians and reptiles. Human access has been tightly controlled since the beginning of cave development (Tufts and Tenen, 1999), and environmental conditions within the cave have been continuously monitored since 1989 (Toomey and Nolan, 2005). Current information on carbonate cave metabolic potential has been largely inferred from 16S rRNA gene molecular surveys rather than functional analyses (for example, Barton et al., 2004, 2007; Ortiz et al., 2013). A broad pyrotag survey in Kartchner Caverns revealed an unexpectedly high bacterial diversity on speleothem surfaces with an average of 1994 operational taxonomic units (OTUs) per speleothem that were classified in 21 phyla and 12 candidate phyla, and appeared to be dominated by organisms associated with heterotrophic growth (Ortiz et al., 2013). Studies in other carbonate caves have proposed that cave microbes are translocated soil heterotrophs (Laiz et al., 1999; Simon et al., 2003) supported primarily by allochthonous carbon compounds entering the cave with air currents or water percolating from the surface (Barton et al., 2004). In contrast, our pyrotag survey revealed an overlap of just 16% between cave OTUs and those of a soil community from above the cave. Phylogenetic associations to cave OTUs suggested the presence of chemolithoautotrophy within the cave.

The objective of this study was to examine the primary production metabolic capabilities of Kartchner Caverns speleothem communities using (1) an analysis of the functional and taxonomic composition of a cave speleothem metagenome and (2) comparisons of the genomic profiles of multiple cave metagenomes to other terrestrial and marine environments to identify potentially important strategies developed by carbonate cave microbes to survive the unique challenges of this oligotrophic, subterranean environment. These analyses were driven by the hypothesis that although speleothem communities are dominated by heterotrophic bacteria, the source of energy for heterotrophs is at least partially derived from key chemolithoautotrophic activities.

Materials and methods

Sampling, DNA extraction and sequencing

This study represents a metagenomic analysis of surface microbial communities sampled from a complex cave formation located in a remote room of Kartchner Caverns accessed through the Echo Passage. The formation was sampled in October 2009 and consists of multiple stalactites descending from a common drapery (Figure 1).

Figure 1
figure 1

Map of Kartchner Caverns. The map indicates the location of tour trails (Rotunda and Big Room trails), the Echo Passage and Big Wall sites (BW2) used for metagenomic and qPCR analysis in the current study, and the sites sampled in previous studies (BW1 and SA). The insert shows the complex cave formation sampled for the Echo Passage speleothem metagenome analysis in Kartchner Caverns. The surfaces of stalactites a, b and c were swabbed for the Echo Passage metagenome analysis.

Thirty-two swabs moistened with sterile deionized H2O were used to sample the surface area (4 cm2 per swab) of three stalactites from the Echo Passage speleothem using previously described protocols (Ortiz et al., 2013). Previous research revealed variability in community composition along the surface of a single speleothem, but indicated that this variability is less than that observed between distinct speleothems. (Legatzki et al., 2011). The composite sample created using 32 swabs to sample the majority of the surface area of three closely associated speleothems descending from a single drapery was designed to incorporate potential variability in community composition along the speleothem surface. The swabs were sonicated (20 s), vortexed (1 min) and sonicated again in sterile deionized H2O, and then removed. The remaining supernatant was centrifuged at 14 000 g for 10 min. The resulting pellet was resuspended in 978 ml sodium phosphate buffer and total genomic DNA was extracted using the FastDNA spin kit for soils (MP Biomedicals, Solon, OH, USA) following the manufacturer’s protocol optimized to enhance DNA recovery from low template samples (Solis-Dominguez et al., 2011). Finally, 509 ng of DNA extract at a concentration of 9.8 ng μl−1 was provided to the Arizona Genomic Institute for sequencing in one direction using the Rapid Library Preparation method for GS-FLX-Titanium pyrosequencing (454 Life Sciences, Roche Diagnostic Corporation, Branford, CT, USA).

Drip-water collection

Drip-water samples were collected monthly from January to December 2011 below a stalactite in the Big Wall room (Figure 1, BW2). Water was collected for a period of 5–23 days each month until an average volume of 220 ml had been recovered. Dissolved organic carbon was measured following acidification to remove inorganic carbon, using a Shimadzu VCSH total organic carbon analyzer (Columbia, MD, USA) with a solid state module (SSM-5000A). NO2-N and NO3-N was analyzed using a Dionex ICS-1000 (ThermoScientific, Sunnyvale, CA, USA). NO2-N was below the detection limit in all samples.

Data processing, assembly and annotation

Total sequence data obtained from the Echo Passage sample were 291 Mbp (930 939 reads of an average read length of 313 bp). Low quality reads were filtered from the data set based on the following criteria: a quality score two s.d. smaller than the average, a length two s.d. smaller or larger than the average and ambiguous bases (Ns) (Huse et al., 2007; Hurwitz et al., 2012). The resulting 769 420 reads were then dereplicated (; Gomez-Alvarez et al., 2009), leaving 506 618 reads that were assembled using the Newbler software (454 Life Sciences, Roche Diagnostics Corporation) using a minimum length of 80 bp and a minimum sequence identity of 96%. The assembled reads were submitted to the Integrated Microbial Genomes with Microbiomes Samples Expert Review (IMG/M ER) pipeline (Markowitz et al., 2009) for gene prediction and annotation. In addition, the unassembled reads were analyzed by BLASTX ( against the NCBI non-redundant nucleotide (NR) database, then evaluated for taxonomic assignments using the MEtaGenome ANalyzer (MEGAN v4.69.4) software with its default settings. MEGAN uses a lowest common denominator algorithm to assign reads to a taxa (Huson et al., 2007). Details on in-house python scripts for analysis of large data sets and BLAST adaptive strategies are documented at: The data for the Echo Passage metagenome is available through the IMG/M database, IMG-ID 2189573024.

Sequence data from two stalactites and one calcite-coated wet rock surface located below the Big Wall in the Rotuda Room area of the cave (Figure 1, BW2) were generated by a concurrent study and were included for all comparative analyses. Sampling procedures, data processing, assembly and annotation for the BW2 metagenomes followed the same protocols described above (Julia Neilson, personal communication). The three BW2 surfaces were comprehensively sampled with a total of 60, 96 and 84 swabs, respectively, and produced DNA yields of 2900, 446 and 229 ng of DNA per surface. The average sequence read length of the three BW2 samples was 382–392 bp. The BW2 data is archived in the IMG/M database and will be released upon publication of the related manuscript.

Comparative analysis

Relative gene abundance was profiled by comparing the Echo Passage cave metagenome to 15 other metagenomes (Supplementary Table 1) that included the three Kartchner Caverns BW2 cave metagenomes and 12 publicly available metagenomes obtained from the IMG/M database ( that had been pre-processed (quality filter, assembly and so on) before submission to the IMG/M database. The twelve IMG/M metagenomes included four from each of the following environments; deep ocean, bulk soil and rhizosphere soil. Gene prediction and annotation for all samples was done on assembled sequence reads using the IMG/M ER pipeline. Hierarchical cluster analysis was performed using the IMG/M ER/Compare Genomes/Genome Clustering tool by function using the Clusters of Orthologous Groups of proteins (COGs) classification system. The relative gene abundance for each metagenome sample was calculated for a given COG category/function by dividing the number of genes in that category by the total number of genes with COG function assignment in order to reduce annotation bias (Delmont et al., 2011), and averages were generated for each environment. The COG classification system was selected because it assigned functions to the largest number of gene sequences for each metagenome. Statistical significance was determined by one-way analysis of variance (ANOVA) and the Tukey–Kramer honestly ignificant difference (HSD) test (P<0.05) as implemented in JMP9 (SAS Institute Inc., Cary, NC, USA).

Analysis of specific genes

The metabolic analysis of the Echo Passage metagenome was performed using both the Kyoto Encyclopedia of Genes and Genomes (KEGG) and the COGs classification systems. KEGG maps of carbon fixation and nitrogen metabolism for the Echo Passage metagenome were obtained from the IMG/M ER pipeline. The Echo Passage protein sequences for specific enzymes involved in these two processes were downloaded from the site and analyzed by BLASTP against the NCBI-NR database. BLAST results were then uploaded to MEGAN for taxonomic analysis and visualization using default settings.

Quantitative PCR analysis

The relative abundance of bacteria, archaea and fungi present in the cave was compared with soil from above the cave by quantitative PCR (qPCR) using a CFX96 Real-Time PCR System (Bio-Rad Laboratories, Hercules, CA, USA). The analysis was conducted using DNA extracts from four cave surfaces sampled near BW2 in Kartchner Caverns (Figure 1). The surfaces included a dry rock wall (DRW), a wet rock coated with calcite veneer (WRW) and two speleothems (A and B). Soil samples were collected at a depth of 15–25 cm from three sites above the cave including near the cave sink hole next to the natural cave entrance (SH), above the Rotunda Room portion of the cave (OR) and next to the rain gauge (RG) in the saddle between the two hills over the cave. The soils were collected and processed as described by Drees et al. (2006). Soil cover above the cave is minimal being classified as primarily exposed bedrock surfaces with intermittent pockets of soil development typically <1 m in depth. The sample depth was selected to avoid the influence of eolian deposition of surface materials while preserving the rhizosphere influence of the photic surface ecosystem. Total genomic DNA from cave and soil samples was extracted using the FastDNA Spin Kit for soil and amplified in a 10-μl qPCR reaction with 1 × SsoFast EvaGreen Supermix (Bio-Rad Laboratories), 400 μg ml−1 unacetylated bovine serum albumin solution, 400 nM of each primer and 400 pg DNA extract. Amplifications used 16S rDNA primers 338F/518R and 931F/1100R for bacteria and archaea, respectively, (Einen et al., 2008) and 18S rDNA primers nu-SSU-1196F/nu-SSU-1536R for fungi (Borneman and Hartin, 2000; Castro et al., 2010). Standards for each domain were made from linearized plasmids (pGEM- T Easy; Promega, Madison, WI, USA) containing the SSU rRNA gene fragments from Escherichia coli JM109, Halogeometricum borinquense ATCC 700274 and Alternaria alternata for bacteria, archaea and fungi, respectively. The qPCR amplification conditions were: 98 °C for 3 min, followed by 45 cycles of 98 °C for 5 s and 60 °C for 5 s (6 s for fungal qPCR). Sample traces were considered quantifiable if they fell within the range of reproducible standard traces for the respective standard curves. Otherwise, the SSU rRNA gene was labeled undetectable.

The following controls were included in all reactions. The positive control used DNA extracted from the surface of an above-ground rock using the same swab and extraction protocols used on cave surfaces. Good amplification of bacteria, archaea and fungi was consistently observed from the positive control. No template-negative controls were run for each domain, and values obtained were consistently below the range of the standard curves confirming the absence of significant influence from reagent contaminants. Technical triplicates were averaged for each sample. Potential inhibition was evaluated using plasmid puc18 added to all DNA extracts, and was amplified with the primers M13F and M13R. No inhibition was detected. Statistical significance was determined with a two-tailed t-test using the JMP9 software (SAS Institute Inc.).


Analysis of drip-water chemistry

Drip-water flow rates fluctuated during the year with the highest rates tracking the winter rains and the summer monsoons (Figure 2). Nitrate levels consistently exceeded total dissolved organic carbon concentrations, and peak concentrations for both carbon and nitrogen predictably followed the Sonoran Desert spring bloom, which typically occurs during March and April. A nitrate peak was also observed in October that was not accompanied by a parallel increase in total dissolved organic carbon.

Figure 2
figure 2

Analysis of drip-water samples collected monthly from January to December 2011 below a stalactite in the Big Wall room (Figure 1, BW2). Drip-water flow rates and drip-water dissolved organic carbon and nitrate–nitrogen concentrations are represented.

Echo Passage metagenome overview and taxonomic composition

The analysis of assembled reads from the Echo Passage speleothem metagenome (IMG/M ER pipeline) resulted in 365 407 predicted genes of which 50% had predicted known functions (Table 1). Based on the COGs classification system, the most abundant metabolic categories represented were amino acid transport and metabolism (10%), energy production and conversion (8%), and replication, recombination and repair (7%) (Supplementary Figure 1). Analysis using the KEGG database also showed amino acid and energy metabolisms among the most abundant categories (Supplementary Figure 2).

Table 1 Echo Passage metagenome sequence statistics

Taxonomic analysis of the unassembled reads using a combination of MEGAN software and the NCBI-NR database produced classifications for 69% of the reads (Figure 3). Of those classified, Bacteria were dominant (85%) followed by Archaea (10%), Eukaryota (5%) and viruses (0.13%). Within the bacterial domain, 54% of the sequences could be further assigned to a phylum and these were dominated by Proteobacteria (52%), Actinobacteria (13%) and Planctomycetes (7.5%). This bacterial taxonomic distribution supports recent pyrotag and clone analyses performed in two other regions of Kartchner Caverns (Figure 1, BW1 and SA) that also found Protoebacteria and Actinobacteria to be the dominant phyla (Legatzki et al., 2011; Ortiz et al., 2013). However, the specific Echo Passage speleothem taxonomic profile was distinct from the nine other previously characterized speleothems, supporting our previous observation that bacterial community composition varies among speleothems.

Figure 3
figure 3

Taxonomic affiliations for Echo Passage metagenomic sequences. Taxonomic associations were determined by a combination of BLASTX analysis of all the Echo Passage sequences against the NCBI-NR database and the MEGAN software. (a) Domain distribution; (b) bacterial distribution; (c) archaeal distribution; and (d) eukaryotic distribution. Percentage values represent the genes assigned to a particular taxon relative to the total number of genes assigned to that specific category.

MEGAN assigned 82% of the archaeal sequences to phyla (Figure 3), and these were dominated by Thaumarchaeota (76%) followed by Euryarchaeota (17%) and Crenarchaeota (7.6%). The majority of the Thaumarchaeota sequences (53%) were associated with marine archaea. Further analysis revealed that 13% of the Thaumarchaeota reads were classified as the Nitrosopumilus maritimus, an ammonia-oxidizing marine archaeon (member of group I.1a) (Konneke et al., 2005), and 4.7% were associated with Nitrososphaera gargensis (member of group I.1b), an ammonia-oxidizer frequently found in terrestrial environments (Hatzenpichler et al., 2008). Within the Eukaryota, 79% of the sequences could be classified of which 23% were fungi, accounting for just 0.9% of the total classified community. Other significant members of the classified eukaryotic community include Metazoa (37%), Viridiplantae (16%) and Alveolata (5%).

Domain distributions were further evaluated by SSU rDNA qPCR, for comparison with metagenome results and soils above the cave (Figure 4). As with the metagenome analysis, the qPCR results showed the cave communities to be dominated by Bacteria followed by Archaea. Fungal genes were below the detection level. Fungal genes amplified from the rock used as a positive control (see Materials and methods) were more abundant than those present in the soils, confirming that the absence of fungal genes on cave surfaces was not an artifact of the sampling procedure or the primers used. Bacterial abundance in cave communities was comparable to the soil samples, however, archaeal abundance was significantly higher in the cave than in the soil communities. In contrast, fungal abundance was significantly greater in the soil communities than on cave surfaces.

Figure 4
figure 4

qPCR analysis of the SSU rRNA genes of bacteria, archaea and fungi in cave and soil samples. Cave samples included dry rock (DRW) and wet calcite-coated rock (WRW) walls and two speleothems (A and B) that were all located near the Big Wall area (BW2) of Kartchner Caverns. Soil samples were collected near the sink hole next to the natural cave entrance (SH), above the Rotunda Room portion of the cave (OR) and next to the rain gauge (RG) in the saddle between the two hills over the cave. Bars for archaea labeled with different letters represent samples with significantly different 16S rRNA copy numbers (two-tailed t-test; P<0.01). No significant differences were observed for bacteria. Fungi were below the detection level for cave samples.

Comparative metagenomic analysis

A comparative metagenomic analysis of the assembled reads was performed using 4 Kartchner cave metagenomes and 12 publically available metagenomes from deep-ocean, bulk soil and rhizosphere soil samples as explained in Materials and methods (Supplementary Table 1). These metagenome groups will be referred to as cave, ocean, soil and rhizosphere, respectively, for the remainder of this paper. Hierarchical cluster analysis grouped the four cave samples in a separate clade more closely associated with soil and rhizosphere samples than with ocean samples (Figure 5). The cluster analysis suggests that the cave communities represent distinct and potentially specialized terrestrial microbial communities.

Figure 5
figure 5

Hierarchical cluster analysis of the 16 selected metagenomes. The tree was generated using an agglomerative algorithm based on pair-wise comparisons of the 16 metagenomes. This comparative metagenomic tool is provided by the IMG/M ER pipeline that used the predicted COG functions for each metagenome.

Over- and under-representation of cave COG categories were evaluated by comparison with soil, rhizosphere and ocean environments. COG categories were classified as similar if no significant difference was found between the cave and any of the ocean, soil or rhizosphere habitats (Figure 6a), or as variable if the cave was significantly different from at least one of the other three habitats (P<0.05; Figure 6b).

Figure 6
figure 6

Average relative abundance of COG categories for cave, deep ocean, bulk soil and rhizosphere soil metagenomes. (a) COG categories for which cave metagenomes showed no significant difference to any of the other three habitats. Letters representing COG categories: energy production and conversion (C); cell cycle, cell division and chromosome partitioning (D); amino acid transport and metabolism (E); lipid transport and metabolism (I); cell wall, membrane and envelope biogenesis (M); cell motility (N); inorganic ion transport and metabolism (P); secondary metabolites biosynthesis, transport and catabolism (Q); function unknown (S); and intracellular trafficking, secretion and vesicular transport (U). (b) COG categories for which cave metagenomes differed significantly from at least one of the other three habitats. Letters representing COG categories: nucleotide transport and metabolism (F); carbohydrate transport and metabolism (G); coenzyme transport and metabolism (H); translation, ribosomal structure and biogenesis (J); transcription (K); replication, recombination and repair (L); post-translational modification, protein turnover and chaperones (O); general function prediction only (R); signal transduction mechanisms (T); and defense mechanisms (V). Bars represent 1 s.d. from the mean. Environments labeled with different letters within a COG category differ significantly in gene abundance (ANOVA, n=4, P<0.05).

Results showed a significant difference between the cave communities and at least one of the other three habitats (ocean, soil or rhizosphere) for 50% of the COG categories (Figure 6b). Among these variable categories, significant over-representation in the cave communities was observed for replication, recombination and repair (L). The over-represented genes in this category were primarily associated with uncharacterized proteins involved in DNA repair (COGs 1336, 1337, 1343, 1367 and 1518). Under-representation in cave communities was found for carbohydrate transport and metabolism (G), though a significant difference was only found when comparing the cave with soil and rhizosphere metagenomes. Specifically, differences were observed for genes associated with a monosaccharide ABC-transporter; COGs 1129, 1879 and 1869 were significantly lower in the cave and ocean compared with the soil and rhizosphere metagenomes. These three genes together with COG1172, which was significantly under-represented in the cave relative to all three environments, represent the four components of the ribose ABC-transporter, a two-component system involved in importing nutrients into the cell. Under-representation of this transporter likely reflects the low-nutrient availability of this ecosystem (Lauro et al., 2009). Within carbohydrate metabolism, all genes for major pathways such as glycolysis, pentose phosphate pathway and the Entner–Doudoroff pathway were detected in the Echo Passage metagenome.

Specific patterns that differentiated the oligotrophic ecosystems (cave and oceans) from the typically richer ecosystems (soil and rhizosphere) were also analyzed. The signal transduction (T) and defense mechanism (V) COGs were found to be significantly lower in the oligotrophic environments than in the richer ones. Over-represented categories included coenzyme transport and metabolism (H), translation, ribosomal structure and biogenesis (J), and post-translational modification, protein turnover and chaperones (O). Transcription (K) represented the only cave COG with abundance more similar to the soil and rhizosphere samples than to the ocean (Figure 6b). The covariance of the oligotrophic ecosystems was of particular interest given that the cluster analysis indicated that the cave communities were generally more similar to both soil and rhizosphere than to the ocean metagenomes.

Cave energy and nutrient dynamics inferred from the Echo Passage metagenome

CO2 fixation

Potential primary production strategies in Kartchner Caverns were of primary interest because of the low total dissolved organic carbon levels in cave drip water (Figure 2). KEGG analysis of the Echo Passage metagenome identified putative genes from all six known CO2-fixation pathways in the metagenome (Supplementary Figure 3A and 3B), however, only the Calvin–Benson–Bassham (CBB) and the Arnon–Buchanan reverse tricarboxylic acid (rTCA) cycles were fully represented. Putative RuBisCO genes (representing the CBB cycle, n=22) identified in the Echo Passage metagenome were further analyzed for taxonomic identification using MEGAN. Genes belonging to both Bacteria and Archaea were identified and the relative abundance of those with domain assignments (50%) was 10:1 (Bacteria to Archaea). Just 14% could be classified below the domain level using MEGAN (Supplementary Figure 4A); therefore, BLASTP searches against the NCBI-NR database were conducted with the unidentified genes to obtain further taxonomic associations. Top hits indicated that 45% of the RuBisCO genes in the Echo Passage metagenome were associated with Proteobacteria. Genus level associations were only obtained for the betaproteobacterial genus Nitrosospira (82% amino acid sequence identity) and the actinobacterium Acidithiomicrobium (84% identity), both of which are known to fix CO2 using the CBB cycle (Utaker et al., 2002; Norris et al., 2011).

Seventy-nine putative genes corresponding to the key rTCA cycle indicator gene, ATP–citrate lyase, were also classified with MEGAN (Supplementary Figure 4B) and 89% could be assigned to a domain with a 6:1 ratio of Bacteria to Archaea. Of the bacterial genes, 44% could be assigned to Proteobacteria (n=27), 11.5% to Nitrospirae (n=7), 11.5% to Actinobacteria (n=7) and 8% to Chloroflexi (n=5). Of these phyla, previous research has documented autotrophic growth by both Proteobacteria and Nitrospirae using the rTCA cycle (Berg, 2011). The assignment of nine ATP–citrate lyase genes to the archaeal domain is of particular interest for future study because there is conflicting evidence in the literature concerning the ability of archaea to assimilate CO2 using the rTCA cycle (Strauss et al., 1992; Ramos-Vera et al., 2009; Berg et al., 2010).

Additional insights into CO2-fixation mechanisms present in the Echo Passage community were obtained using the COG classification system. This analysis identified 130 putative genes for the aromatic ring hydroxylase (COG2368), also known as 4-hydroxybutyryl-CoA dehydratase. For Archaea, 4-hydroxybutyryl-CoA dehydratase is considered as the key indicator enzyme in two archaeal CO2-fixation mechanisms (Berg et al., 2010): the 3-hydroxypropionate/4-hydroxybutyrate (HP/HB) cycle (Berg et al., 2007) and the dicarboxylate-4-HB (DC/HB) cycle (Huber et al., 2008). This gene is considered as a marker for CO2 fixation by Thaumarchaeota (Zhang et al., 2010). In contrast, in bacteria, it is associated with fermentation rather than CO2 fixation (Gerhardt et al., 2000). Taxonomic analysis using MEGAN could confirm 24% of these genes belonging to Archaea, 8.5% of which were Thaumarchaeota (Supplementary Figure 4C). Not all the genes for the HP/HB or the DC/HB cycles were identified in the Echo Passage metagenome, which is not surprising given the paucity of oligotrophic environmental isolates in current databases used for gene annotation. The abundance of putative archaeal genes for 4-hydroxybutyryl-CoA dehydratase and ATP–citrate lyase supports the hypothesis that autotrophic archaea contribute to cave ecosystem carbon assimilation.

To further explore the relative importance of autotrophy in the cave, a COG-based comparative analysis of the CBB, rTCA and HP/HB-DC/HB indicator enzymes was performed using the set of 16 metagenomes (soil, rhizosphere, cave and ocean) described previously (Figure 7). Statistical analysis showed significant over-representation of RuBisCO genes (COG1850, CBB cycle) in the cave relative to the other three metagenomes. The ATP–citrate lyase (COG2301, rTCA) and the 4-hydroxybutyryl-CoA dehydratase (COG2368, HP/HB and DC/HB) were also more abundant in both oligotrophic ecosystems (cave and ocean) than in the soil and in the rhizosphere, however, the differences were not significant due to the high variability between samples. A KEGG-based analysis supported the COG results for RuBisCO and ATP–citrate lyase (not shown), but the 4-hydroxybutyryl-CoA dehydratase was not identified using the KEGG database.

Figure 7
figure 7

Heat map showing average relative abundance of COGs representing key carbon fixation enzymes identified in the cave, deep ocean, bulk soil and rhizosphere soil metagenomes. COG IDs: COG1850, RuBisCO, key enzyme for the CBB; COG2301, ATP–citrate lyase, enzyme representing the Arnon–Buchanan cycle (rTCA) and COG2368, 4-hydroxybutyryl dehydratase, key enzyme for the HP/HB and DC/HB cycles. Values labeled with different letters within a single COG differ significantly in gene abundance (ANOVA, n=4, Tukey–Kramer HSD, P<0.05). HSD, honestly significant difference; rTCA, reverse tricarboxylic acid.

Nitrogen metabolism

The availability of reduced compounds as energy sources for primary production in carbonate caves is presumed to be limited to the host rock and drip water because of the absence of specific energy sources (for example, reduced sulfur compounds present in sulfidic caves). Drip-water inorganic nitrogen appeared particularly relevant to the Kartchner Caverns ecosystem because NO3-N concentrations in cave drip water consistently exceeded the dissolved organic carbon levels (Figure 2). An extensive analysis of 42 nitrogen cycling genes was performed revealing a diversity of nitrification, nitrate reduction and ammonia assimilation genes in the Echo Passage speleothem metagenome (Supplementary Figure 5). Based on drip-water chemistry, we were particularly interested in nitrification as a potential energy source for this oligotrophic cave ecosystem. Twenty-three ammonia monooxygenase (amoA) genes involved in the first step of nitrification were detected in the Echo passage metagenome and were found using MEGAN classification to be equally distributed between the archaeal and bacterial domains (Supplementary Figure 6). Genes for hydroxylamine oxidase, a key enzyme required for ammonia oxidation by bacteria (AOB), but not for ammonia-oxidizing archaea (AOA) (Hallam et al., 2006; Kim et al., 2011) were not detected, suggesting that AOA are the predominant ammonia oxidizers in the cave ecosystem.

Genes for nitrite oxidoreductase, responsible for the oxidation of nitrite to nitrate, were not identified by gene annotation based on KEGG and COG databases. Nevertheless, BLASTP searches against the Echo Passage metagenome using the IMG/M ER BLAST tool revealed 62 Echo genes with 30–97% identity (E-value<1e-5) to a putative nitrite oxidoreductase enzyme (YP_003798871) from the metagenome of Candidatus Nitrospira defluvii, a known nitrite oxidizer (Lucker et al., 2010).

Ammonia assimilation, denitrification and dissimilatory nitrate reduction pathways were all well represented in the Echo Passage metagenome (Supplementary Figure 5). In addition, the marker gene (nrfA) (Kraft et al., 2011) for dissimilatory nitrate reduction to ammonium (DNRA) (Supplementary Figure 5) was detected. Comparative metagenomic analysis revealed strong representation of COG0004 (ammonia permease), COG1251 (NAD(P) H-nitrite reductase large subunit) and COG2146 (nitrite reductase–ferredoxin) in the speleothem communities (Figure 8). COG2146, involved in assimilatory reduction of nitrate and DNRA, was significantly over-represented in the cave metagenomes relative to the other environments. The NAD(P)H-nitrite reductase large subunit is also associated with pathways for both assimilatory nitrate reduction and DNRA (KEGG).

Figure 8
figure 8

Heat map showing average relative abundance of COGs for putative nitrogen metabolism genes identified in the cave, deep ocean, bulk soil and rhizosphere soil metagenomes. COG IDs: COG0004, ammonia permease; COG1140, nitrate reductase beta subunit; COG1251, NAD(P)H-nitrite reductase; COG1348, nitrogenase reductase subunit NifH (ATPase); COG2146, nitrite reductase–ferredoxin; COG2223, nitrate/nitrite transporter; COG3180, putative ammonia monooxygenase; COG4263, nitrous oxide reductase; and COG5013, nitrate reductase alpha subunit. Values labeled with different letters within a single COG differ significantly in gene abundance (ANOVA, n=4, Tukey–Kramer HSD, P<0.05). HSD, honestly significant difference.

Finally, key genes for nitrogen fixation, including the nifD and nifK genes, which encode the α- and β-subunits of the dinitrogenase enzyme, were not identified by the KEGG classification system. COGs analysis identified only the nifH and nifF genes that encode for dinitrogenase reductase and flavodoxins, respectively.


The functional and taxonomic analyses presented in this study provide new insights into the ecosystem dynamics and survival strategies of microbial communities colonizing speleothem and rock surfaces in the oligotrophic Kartchner Caverns habitat. In this study, we focused specifically on surface communities because we believe them to be dynamic and most responsive to the on-going flux in drip-water nutrient inputs. Information from this study can inform future work focused on microbial communities embedded in calcite formations that might provide a more long-term temporal archive of past environmental conditions and microbial influences on calcite precipitation.

The metagenomic analysis revealed that the cave microbial communities clustered apart from soil, rhizosphere and ocean environments, but were more similar to the soil and rhizosphere metagenomes than to the ocean. Despite the closer association observed between the three terrestrial metagenome groups, a greater covariance of under- and over-represented COG categories was observed for the two low-nutrient ecosystems (ocean and cave) than for the three terrestrial metagenomes. The two COG categories that were significantly under-represented in both the cave and ocean metagenomes (T and V) were previously identified among genomic markers for confirming the trophic strategy of uncultured marine oligotrophs (Lauro et al., 2009). In addition, our previous characterizations of Kartchner speleothem community diversity identified clones and cultured isolates with 99% and 98% identity, respectively, to classic marine oligotrophs such as Sphingopyxis alaskensis and Polaromonas aquatica (Ikner et al., 2007; Ortiz et al., 2013), indicating that community members from the two habitats are phylogenetically related, suggesting potential ecological similarities between these oligotrophic communities. Taken together, these results allow speculation that carbonate cave communities originated from soil ecosystems, but that the specific low-nutrient environmental conditions of Kartchner Caverns have selected for oligotropic functional profiles that parallel trophic strategies found in oligotrophic marine environments.

Energy dynamics of the Echo Passage microbial community

Metagenomic analysis of the carbon and nitrogen metabolic pathways along with the associated functional taxonomic composition provides strong evidence for a chemolithoautotrophic component to the Echo Passage speleothem microbial community. Full analysis required creative use of multiple databases and analysis strategies to compensate for the limited representation of cave microbes and oligotrophic metagenomes in current databases. First, an abundance of genes representing all six known CO2-fixation pathways was detected revealing the existence of a microbial community with diverse autotrophic potential. Putative RuBisCo genes representing the CBB cycle were significantly over-represented relative to soil, rhizosphere and oceans, and the abundances of both ATP–citrate lyase (rTCA cycle) and 4-hydroxybutyryl-CoA dehydratase (HP/HB-DC/HB cycles) genes were comparable to the nutrient-limited ocean environments and greater than both soil and rhizosphere. The abundance of HP/HB genes was of particular interest because enzymes in this cycle use bicarbonate as the active inorganic carbon species, whereas bicarbonate is not a RuBisCo substrate. The HP/HB pathway is hypothesized to be advantageous for chemolithoautotrophic marine archaea because bicarbonate availability under slightly alkaliphilic conditions (for example, ocean water) is significantly higher than dissolved CO2 (Berg, 2011), conditions that may also apply to this carbonate cave ecosystem where drip water pH averages 8.0.

The diversity of key genes representing the CBB, rTCA and HP/HB pathways suggests the potential for community CO2 fixation under diverse environmental conditions. Each pathway has different energy demands as reviewed by Berg (2011) and demonstrated by the example of a gammaproteobacterial marine endosymbiont that uses the relatively energy-expensive CBB cycle under high-energy conditions and the energetically more favorable rTCA cycle under low-energy conditions (Markert et al., 2007). The majority of Echo Passage RuBisCo genes (CBB) were associated with Proteobacteria, the dominant phylum in the community. In contrast, the ATP–citrate lyase genes for the rTCA cycle were assigned to a greater diversity of phyla including Proteobacteria, Nitrospirae, Actinobacteria, Chloroflexi and the domain, Archaea. The chemolithoautotrophic Nitrospirae exemplify chemoautotrophic K-strategists capable of growing on nitrite substrate concentrations 10-fold lower than that required by the well-characterized Nitrobacter species (Bartosch et al., 2002). Putative oxidoreductase genes with similarity to C. Nitrospira defluvii were identified in the Echo metagenome, allowing us to speculate that chemolithoautotrophic nitrite-oxidizing Nitrospirae are key primary producers in this oligotrophic community. This hypothesis is supported by previous 454-pyrotag surveys that found Nitrospirae on all speleothem surfaces sampled within the cave (Ortiz et al., 2013). Nitrospirae have also been identified globally in cave communities, including Altamira Cave, Spain (Portillo et al., 2008; Porca et al., 2012), Niu Cave, China (Zhou et al., 2007), Pajsarjeva jama, Slovenia (Porca et al., 2012), Nullarbor caves, Australia (Holmes et al., 2001; Tetu et al., 2013) and Spider and Lechuguilla caves, New Mexico (Northup et al., 2003).

A second key insight offered by this work is the potential contribution of Archaea to the energy dynamics of this oligotrophic karst ecosystem. qPCR analysis revealed that archaeal abundance in Kartchner rock and speleothem communities is significantly higher than in soil communities from above the cave. A recent global 16S rRNA pyrotag analysis found an average archaeal abundance in soils of 2% with the abundance inversely correlated with C:N ratio, suggesting that Archaea can tolerate or even exploit low-nutrient conditions (Bates et al., 2011). The taxonomic analysis of the Echo Passage metagenome found that unassembled archaeal sequences comprised 10% of the metagenome.

Thaumarchaeota were dominant among archaea and represented the third most abundant phylum overall. Among the Thaumarchaeota, MEGAN found 18% of the reads to be associated with ammonia-oxidizing archaea. Importantly, taxonomic associations to this phylum were also identified for both the ATP–citrate lyase (rTCA cycle) and 4-hydroxybutyryl-CoA dehydratase (HP/HB and DC/HB cycles) CO2-fixation genes. In addition, half of the amoA genes identified in the Echo Passage metagenome were classified as Archaea. Previous studies have identified Thaumarchaeota as chemoautotrophic ammonia oxidizers (Zhang et al., 2010; Pester et al., 2011) adapted to low substrate conditions (Martens-Habbena et al., 2009). Further, AOA in soils have been shown to fix CO2 using the HP/HB cycle (Pratscher et al., 2011). Archaeal amoA genes have also been amplified from a stalactite sample taken from a mine adit in Colorado (Spear et al., 2007). Finally, a previous analysis of archaeal community structure on the SA speleothems in Kartchner Caverns (Figure 1) identified OTUs that were phylogenetically associated with the AOA, N. gargensis (Legatzki et al., 2011). Taken together, these results suggest that Thaumarcheota represent a second key group of chemolithoautotrophic primary producers in the Echo Passage speleothem community.

A recent metagenomic analysis of microbial slime in the subterranean aquatic Weebubbie cave below Australia’s Nullarbor Plain found a similar pattern of primary production driven by inorganic nitrogen metabolism (Tetu et al., 2013). Similar to Kartchner, Weebubbie microbial communities included an abundance of ammonia-oxidizing Thaumarchaeota, although the Weebubbie community had a much higher relative abundance of Thaumarchaeota, and the AOA:AOB ratio was more similar to that found in marine environments. Thus, the Kartchner Caverns ecosystem again appears to represent an oligotrophic terrestrial counterpart to the energy dynamics observed in oligotrophic marine habitats such as Weebubbie cave.

The presence of archaea has been previously documented in numerous terrestrial carbonate caves (Northup et al., 2003; Chelius and Moore, 2004; Gonzalez et al., 2006; Macalady et al., 2007; Legatzki et al., 2011), but their physiological role has remained elusive. This metagenomic analysis strongly suggests that Thaumarchaeota represent a second group of nitrification-based primary producers in Kartchner Caverns. The data also indicate the potential involvement of Archaea in alternate primary production strategies due to the abundance of archaeal CO2-fixation genes not assigned specific taxonomic classifications.

Molecular diversity and functional gene characterizations of Movile Cave in Romania provide an intriguing contrast to these carbonate caves (Chen et al., 2009). Similar to Kartchner, Movile Cave is aphotic and sustained by chemolithoautotrophy, however, the atmosphere in Movile is rich in hydrogen sulfide and methane, and supports high microbial productivity and rich fauna. Microbial mats sustained by sulfur- and ammonia oxidizers contain a diversity of autotrophic bacterial phylotypes, but no archaea. The absence of archaea is in stark contrast to their abundance in Katchner, Weebubbie and the other carbonate caves listed above. The contrast between these cave ecosystems suggests that low-nutrient carbonate caves provide a template for evaluating the role of archaea in oligotrophic terrestrial ecosystems with potential application to subsurface soils, low-nutrient aquifers and even subsurface extraterrestrial ecosystems.

A final cave ecosystem survival strategy highlighted by this study was the over-representation of DNA repair enzyme genes. These enzymes belong to the RAMP (Repair Associated Mysterious Proteins) superfamily, a group of proteins for which no specific function is known, but that clearly associate with DNA repair mechanisms. The overabundance of DNA repair genes is surprising given the absence of typical DNA-damaging agents such as ultraviolet light. We hypothesize that the exceedingly high calcium concentrations in cave drip water (Legatzki et al., 2012) are the source of stress. Research in carbonate caves indicates that cave bacteria precipitate calcium carbonate as a mechanism to overcome calcium toxicity (Banks et al., 2010). Similarly, previous work has linked high levels of calcium ions in eukaryotic cells with DNA strand breakage (Cantoni et al., 1989). Specifically, it has been shown that when cells are under oxidative stress, calcium homeostasis is disrupted, leading to increases in intracellular calcium concentrations that result in the activation of nucleases that damage DNA. Thus, we hypothesize that the abundance of DNA repair enzymes is an adaptation of cave microbes to the stress caused by high calcium concentrations in the cave ecosystem.


This Echo Passage speleothem metagenome analysis combined with previously published pyrotag surveys provides strong evidence for carbonate cave microbial communities specifically adapted to low nutrient and high calcium conditions, and most probably sustained at least in part by an inorganic nitrogen-based primary production strategy with contributions from both bacteria and archaea. The diversity of CO2-fixation pathways represented in the Echo Passage metagenome suggests that the Kartchner speleothem communities are primed to exploit the observed seasonal fluctuations in drip-water nutrient content, a hypothesis that can be tested in future temporal studies, using qPCR to target the CO2-fixation genes identified in this work. In addition, unique nutrient conservation strategies, for example, DNRA, may be present such as suggested by the presence of the nrfA gene and the over-representation of COG2146 (nitrite reductase-ferrodoxin). Research has shown that DNRA capability is phylogenetically widespread (Kraft et al., 2011), is the dominant or sole nitrate consumption pathway in a diversity of soils (reviewed in Rütting et al., 2011) and may represent a unique strategy of oligotrophic microbial communities to minimize nitrogen loss. Despite numerous studies in terrestrial and aquatic systems, there is little consensus concerning the relative importance of this pathway to nitrogen dynamics or the specific environmental factors that control DNRA activity in soils (Rütting et al., 2011). Finally, the metagenomic analysis of novel microbial communities is typically constrained by the limited number of relevant organisms represented in the KEGG, COG and NCBI-NR databases. However, this analysis demonstrates that even with these limitations, key indicators of trophic dynamics in Kartchner Caverns were identified that provide new insights into the primary production potential and survival strategies of this carbonate cave microbial community with potential relevance to a diversity of terrestrial oligotrophic habitats.