The two environments sampled within the Soudan Mine are shown in Figure 1. Groundwater is not very abundant in the banded iron formations of the mine at our sampling depth of 714 m below the surface. However, small amounts of water emerge steadily from exploration boreholes that extend downward from the deepest level of the mine. The two sets of samples were collected from water trickling out of two such boreholes that are separated by about 100 meters (Figure 1). This water is a calcium, sodium, chloride solution about twice as salty as seawater. It is anoxic, with up to 150 ppm of dissolved ferrous iron and variable enrichments of several trace elements. At both locations, the water emerging from the boreholes produces cm-scale "Black" environments that appear to extend down into the borehole. The water flowing away from each borehole, on the floor of the mine tunnel, is exposed to the oxygenated mine atmosphere, and transitions to a sequence of "Red" environments within a few cm of the orifices. The oxidized environments are continuously fed by anoxic water flowing from the boreholes. The water in the borehole, which yielded the Black sample, as well as a number of similar sites found throughout the mine, has a pH of 6.70 and redox potential of -142 mV. Some of the Black areas are associated with bubbling of gas. The Black sediment contained 5.8 × 10⁵ microbes per ml. X-ray diffraction analyses of the minerals in this area show that chlorite-serpentine [(Mg,Al)₆(Si,Al)₄O₁₀(OH)₈], clinochlore, ferroan [(Mg,Fe)₆(Si,Al)₄O₁₀(OH)₈], quartz, and silinaite [LiNaSiO₅·HCl] are present in the Black sediments. Water slowly flows from the borehole into the stream running down the main mine tunnel. As the water comes in contact with oxygen in the passageway, the pH rapidly decreases to 4.37 and redox potential increases to -8 mV. The Red sample contained 1.2 × 10⁶ microbes per ml, and these sediments include goethite [FeO(OH)], followed by szaibelyite [(Mg,Mn)BO₂(OH)], and sussexite [(Mn,Mg)BO₂(OH)].

DNA was purified from the two samples, amplified using the GenomiPhi procedure (GE Healthcare, Piscataway, NJ), and then sequenced by 454 Life Sciences. A summary of the sequence characteristics determined using the pyrosequencing technique is shown in Table 1. The raw sequence reads and quality scores [see Additional files 6 and 7] are provided in compressed format.

The two samples produced more than 70 Mbp of sequence data from over 700,000 sequences, and there was no significant skew in the sequence data (as measured by dinucleotide frequency) when the data generated by pyrosequencing was compared to complete genome sequences.

The two sequence libraries were compared to the 16S rDNA database from the Ribosomal Database Project16. As shown in Figure 2, the Black sample was dominated by Actinomycetales such as Brevibacterium and Corynebacterium that volatilize sulfur via an organic intermediate and can also break down complex heterocyclic and polycyclic ring structures1718. In contrast, members of the Chromatiales, including the genera Chromatiaceae, Thiobacillus, and Halothiobacillus, dominate the Red sample. These chemoautotrophic Bacteria often use the Calvin-Benson-Bassham cycle to fix CO₂ through the oxidation of iron or sulfur, and consequently they would be expected to be present in samples from an iron-rich deposit. These two communities are fundamentally different both from each other and from the community identified in the Iron Mountain metagenome7. The community in the Red sample has a much higher species richness than the Black sample, and the differences between the Soudan and Iron mines reflect the iron composition (hematite versus pyrite), temperature, and pH of the various environments7.

A16S clone library was created from the Red sample to validate the 454 sequencing approach. Ninety-six clones were sequenced using traditional techniques, and compared to the 16S rDNA database from the Ribosomal Database Project 16. The congruity between the 16S genes sequenced in the 454 library and the 16S sequences from the clone library, as shown in Fig. 2, is quite remarkable.

We also used the 16S sequences to evaluate the randomness of the library. An analysis of 160 bacterial genome sequences in the SEED database 1519 with annotated 16S genes showed that about 1 in 10⁵ bases is from a 16S gene. Based on this estimate, as a rule of thumb the Soudan samples are expected to contain approximately 3,000 bases of 16S sequence in total, or approximately 30 sequences. Twenty four sequences were found to have significant similarity (with an E value less than 1 × 10^-5 and a match of 50 bp or more) to 16S rDNA from the Black sample and seventy six sequences were found to have significant similarity to 16S rDNA from the Red sample.

Sequences from both libraries were compared to the SEED database, a curated database of microbial genomes 1519. The annotations using the SEED interface primarily occur through the development of subsystems, a technique pioneered by the Fellowship for Interpretation of Genomes1520. Subsystems are groups of genes that function together, such as the genes whose products are involved in a metabolic pathway, or the group of genes whose products make a cellular structure. A summary of the subsystem hits are shown in Figure 3, and all matches to subsystems are provided as supplemental data [see Additional files 1, 2 and 3]. These subsystems show that the pyrosequencing generates sequences that represent a large swathe of central metabolism in each of the environments. Common metabolic potential that is expected to be present in sulfur-utilizing chemoautotrophs is represented in the mine libraries, including the Calvin-Benson cycle, inorganic sulfur assimilation, amino acid biosynthetic genes, and so on. The comparison of the subsystem similarities suggested the simple hypothesis that groups of genes (or subsystems) important to a particular environment will be enriched in that environment. To distinguish between ecologically important differences and differences caused by sampling error, a method was devised to identify those subsystems that are statistically significantly overrepresented in one sample when compared to another 11.

Table 2 shows subsystems that were determined to be statistically more common, with 95% confidence, in either the Red or Black samples from the Soudan mine. The subsystems that are overrepresented in a metagenome can yield significant insights into the microbial ecology of the environment. A few specific examples are detailed below.

A previous study used Sanger sequencing to determine the metagenome of the Iron Mountain community7. The environmental differences (such as the difference in temperature) account for the predominant differences between the microbial communities. The organismal differences are reflected in the individual biochemistries of the samples [see Additional files 4 and 5]. For example, the AMD metagenome contains significantly more occurrences of Archaea-specific subsystems such as those involved in protein biosynthesis than the Soudan samples. The AMD sample has a preference for CO₂ fixation and simple carbohydrate metabolism when compared to either of the Soudan samples. There are also many currently unexplained differences between subsystems found in these environments that must relate the biology of the organisms to the chemistry of the environment.

The SEED database used for these studies contained 351 subsystems. The vast majority (83%) of subsystems were present in one or more of the sequenced metagenomes, and over half (52%) of the subsystems are present in every metagenome. A comparison of the subsystem classification reveals trends between the metagenomes (Figure 4). For example, oxygenic photosynthesis is prevalent in samples that are naturally illuminated such as the Sargasso Sea10. This analysis also suggested that phosphorous metabolism is more prevalent in oceanic surfaces rather than terrestrial environments. Comparisons of the Minnesota Farm metagenome6 with the Soudan Mine metagenomes, also from Minnesota, showed important differences in the production and consumption of secondary metabolites, membrane transport, and fatty acid metabolism. The complete lists of statistically significantly different subsystems between both Red or Black samples and each of the previously published metagenomes are supplied as supplemental material [see Additional files 4 and 5].

Samples were collected from several sites in the Soudan Mine. This analysis concerns the sample collection at two sites on Level 27 (714 m below the surface; Figure 1). Water and sediments were sampled from the two locations shown in Figure 1 giving the "Black" (reduced) sample and "Red" (oxidized) sample. Microbes were concentrated from these samples by filtration with 0.22 μm Sterivex units. Microbial counts were enumerated by staining the samples with SYBR-Gold (Invitrogen, Carlsbad, CA) and visualization with an epifluorescent microscope 21. DNA was extracted from the microbial sample using either the Ultra Clean Soil DNA Kit or Power Soil Kit (MolBio, Boulder, CO). The DNA was amplified with GenomiPhi (GE Healthcare, Piscataway, NJ) in an Eppendorf thermal cycler (Eppendorf, Westbury, NY) using multiple reactions containing 50–100 ng of the isolated DNA as template and the manufacturer's recommended protocols. After amplification, the resulting DNA was purified with silica columns (Qiagen, Valencia, CA) and concentrated by ethanol precipitation. The DNA was resuspended in water to a final concentration of 0.3 mg/ml. Approximately 10 μg of each sample was sequenced using the pyrosequencing technology (454 Life Sciences, Branford, CT).

Bacterial-specific primers 27F (5'-AGAGTTTGATCMTGGCTCAG-3') and the universal 1492R primer (5'-TACGGYTACCTTGTTACGACTT-3') 22 were used to amplify the 16S rDNA genes. PCR products were cloned into the pCR^®4-TOPO^® vector as recommended by the manufacturer (Invitrogen, Carlsbad, CA).

Water samples were collected by filtering the water through 0.2 μm filters into clean bottles. Field measurements of pH, E_h, temperature and conductivity were conducted in situ. The sediment samples were collected as slurries with a pipette resting on and in the sediments. Those slurries were transferred to clean centrifuge tubes, allowed to settle by gravity and then the fluid was decanted.

Major anions in the water were determined by GC (Dionex IGS-2000, Sunnyvale, CA) and major and trace elements by ICP/MS (Thermo Electron PQ ExCell, Franklin, MA). The mineral identifications are based on XRD (Bruker-AX D500 X-ray Diffractometer, Germany) measurements. The X-ray peaks were relatively small. Much of the sediment was apparently not well crystallized.

The unassembled sequences provided by 454 were compared to the SEED database using the BLASTX algorithm on the Teragrid cluster at Argonne National Laboratories1523. All BLAST searches were performed using an expect value cutoff of 1 × 10^-5. At this cutoff approximately 3 of the observed hits would be expected to occur at random23.

The BLASTN algorithm was used to identify 16S genes from release 9 of the RDP database 1624. These BLAST searches were also performed using an expect value cutoff of 1 × 10^-5 and a minimum sequence match length of 50 nt.

The statistical analysis of subsystems present in each sample was performed essentially as described elsewhere 11. The presence or absence of subsystems between two data sets was determined using 20,000 replicates of samples of 5,000 subsystems each. The 95% confidence interval for the median was constructed using the 0.025 and 0.975 percentiles.