The samples for this study were acquired over the course of three expeditions to SURF in Lead, South Dakota (USA) in Sept. 2013, Oct. 2013, and Feb. 2014 (Yates Shaft, latitude 44.352157°, longitude -103.750503°). Detailed geochemical data were acquired only on the last two trips. Oxidation-reduction potential (ORP), conductivity, pH, temperature, and total dissolved solids (TDS) were measured in situ with an Ultrameter II 6PFC^E (Myron L Company). Redox sensitive species (DO, ∑S²⁻, Fe⁺⁺, Mn²⁺, NO⁻₃, NO⁻₂, NH⁺₄, SiO₂, ∑PO⁻³₄) were measured using Hach DR/2400 portable field spectrophotometers and associated reaction kits (Hack Company, Loveland, CO).

Samples for major anions, cations, and dissolved gasses were preserved for laboratory analysis. Major anions were measured using a Metrohm 850 Professional Ion Chromatograph equipped with are Metrosep A Supp 5, 250 × 4 mm column and conductivity detector. The mobile phase was 3.2 mM Na₂CO₃, 1.0 mM NaHCO₃, and 2.5% acetonitrile following at 0.7 ml/min. Cations were measured using an Agilent 4100 Microwave Plasma Atomic Emission Spectrometer. Samples were diluted 1:10 or 1:100 in 5% HNO₃, and concentrations were calculated relative to synthetic standards prepared in the same manner. Dissolved gas samples were collected by either the bubble stripping method (Alter and Steiof, 2005) where in-line filtering was possible (sites B, D, 6, 9), or via bulk fluid collection into evacuated serum vials (2, 3A 5, 8). The bubble stripping method has significantly higher sensitivity and is preferable when possible. Headspace concentrations of He, H₂, O₂, N₂, CO₂, CH₄, CO, ethane, and propane were measured with a Shimadzu GC-2014ATF headspace GC equipped with Haysep 80/100 (5 m) and MS-5A 60/80 (2.5 m) molecular sieve columns and TCD and FID detectors. Dissolved gas concentrations were calculated from headspace gas concentrations based on solubility constants of each gas at the analytical temperature and pressure and comparison to standard gas mixtures.

Geochemical data from each site were used to calculate the Gibbs energy yields of 140 potential catabolic reactions. These reactions coupled the electron acceptors O₂, NO⁻₃, MnO₂, Fe₃O₄, SO⁻²₄, and S⁰ with the electron donors Fe⁺⁺, CH₄, H₂, S⁰, NH₄⁺, HS⁻, CO, and Mn⁺⁺ (Table 1). Values of Gibbs energy yields were calculated using

where ΔG_r represents the Gibbs energy of reaction r under in situ conditions, K_r and Q_r stand for the equilibrium constant and activity product of the rth reaction, R refers to the gas constant, and T denotes temperature in kelvin. Values of K_r were calculated using the revised-HKF equations of state (Helgeson et al., 1981; Tanger and Helgeson, 1988; Shock et al., 1992), the SUPCRT92 software package (Johnson et al., 1992), and thermodynamic data taken from Bricker (1965), Hem et al. (1982), Shock and Helgeson (1988, 1990); Shock et al. (1989), Sverjensky et al. (1997), Schulte et al. (2001). Values of Q_r for each reaction were calculated from

where a_i designates the activity of the ith species and v_i,r indicates the stoichiometric coefficient of the ith species in the rth reaction. Activities were determined using geochemical data from each field site and the program SPEC8 (Geochemist's Workbench 8, Aqueous Solutions LLC). ΔG_r is presented in units of kilojoules per mole of electron transferred, kJ (mol e⁻)⁻¹. In order to scale energy availability to the limiting reactant, the Gibbs energy calculations are also presented in terms of energy densities, E_r, which are calculated by

where [i] refers to the concentration of the limiting electron donor or acceptor (LaRowe and Amend, 2014). The metric represented by Equation (3) has been shown to correlate with biomass abundance better than the typical reporting of ΔG_r in molar units (LaRowe and Amend, 2014).

DNA was obtained from filters and biofilms using a modified phenol-chloroform extraction within 30 days of each sample collection. This included three rounds of physical (freeze, thaw, vortex) and chemical (lysozyme) disruption of the cell wall prior to phenol-chloroform extraction. Controls during DNA extraction and polymerase chain reaction (PCR) confirmed that there were no contaminating nucleic acids during the extraction process (data not shown). Whole genomic DNA was sent to Molecular Research DNA (Shallowater, TX, USA). Primers 515 forward and 806 reverse (Caporaso et al., 2011) were used to amplify the hypervariable 4 region of the 16S rDNA gene according to previously described methods (Dowd et al., 2008). Following amplification, all PCR products from different samples were mixed in equal concentrations and purified using Agencourt Ampure beads (Agencourt Bioscience Corporation, MA, USA). Samples were sequenced on Roche 454 FLX titanium instruments using recommended reagents and following manufacturer's guidelines.

Sequences were analyzed using the programs QIIME (Caporaso, 2010) and Mothur (Schloss et al., 2009), and compared against the SILVA 111 database (Quast et al., 2012). Sequences were quality filtered by removing from the dataset any sequence that did not have an exact match to the proximal primer, contained fewer than 200 or greater than 350 bases, had more than one ambiguous nucleic acid base (Ns), or had a quality score less than 25. Sequences were clustered using the USEARCH algorithm (Edgar, 2010) at a 97% identity cutoff, and taxonomy was assigned to the lowest possible level using the SILVA 111 database (www.arb-silva.de). It is important to note that these data are not strictly quantitative, and thus all percentages reported below should be taken as relative abundance only. All statistical analyses (PCA etc.) were performed using the Matlab® Statistics Toolbox.

The eight samples presented in this work were taken from the 800, 4700, and 4850 ft levels from open borehole fluids, capped manifolds, and pools within the mine tunnels at SURF. Replicate analyses from the October 2013 and February 2014 sampling trips are presented for three samples (B, D, and 8). Sampling sites on the 800 ft level, shown in plan view in Figure 1A, feature fluids cascading from diamond drill holes into the mine tunnel, often forming thick accumulations of apparent iron oxides and biofilms extending from the source point (Figure 1C). Microterrace structures of microbial biofilms are present (Figure 1D) as are fine white filaments (not shown). Fluid samples from the deeper levels (see map in Figure 1B) are considerably more diverse. At this depth, boreholes that produce significant flow must be capped with manifolds or plugged with cement to prevent flooding. A subset of these capped sites were opened and sampled (Samples B, D, and 3A, Figure 1G). Site 8 is a diamond borehole that was plugged, but is now producing a large amount of fluid (~2 L/min) through a fracture in the seal (Figure 1E). The last two sampling sites are from pools that are presumably fed with reducing fluids through fissures in the mine walls (Sites 6 and 9, Figure 1F).

The solute chemistry of the sampled fluids reflects the diversity of environments within SURF and the history of water-rock-microorganism interactions. Field measurements (Table 2) reveal a narrow range of temperature (10–32.8°C) and pH (6.55–8.46), but sharp differences in redox chemistry, documented with ORP measurements (79 to -292 mV) and Σ S²⁻ (0.00–2778 μg/L). Significant ranges are also observed in some nutrient concentrations, with nitrate and ammonium varying from below detection limit (bdl) to very high levels (~5 and 0.5 mg/L, respectively). Phosphate concentrations were consistently low but detectable (up to 0.11 mg/L), and nitrite was bdl at all sites. Major anion and cation concentrations also show significant variability (Table 3). Sulfate (96–4400 mg/L) is the dominant anion in all samples, followed by chloride (13–219 mg/L); bicarbonate levels (2.6–15 mM) are elevated at most locations. In most samples, Na⁺ is the dominant cation (27–2125 mg/L), but concentrations of Ca²⁺ (6–456 mg/L) and Mg²⁺ (5–90 mg/L) are as high or higher than Na⁺ at the 800′ level sites. Dissolved gas concentrations are reported in Table 4. Hydrogen is universally low but detectable in some samples (0.01–1.2 nM). In contrast, He was detected in all samples from the 4850′ level except Pool 9, ranging from 0.26 to 8.6 nM. Dissolved CO₂ concentrations range from 131 to 1212 nM and correlate well with HCO⁻₃ concentrations. Hydrocarbon gasses co-occur, although methane is more abundant than ethane by a factor 8–202.

Values of Gibbs energies for 140 redox reactions (see Table 1) were calculated with Equations (1–3) using geochemical data given in Tables 2–4. Of these reactions, 98 were exergonic in at least one sampling locality, and in Figures 2A,B, the Gibbs energies for each site are represented in kJ per mole of electron transferred and in Joules per kg of water (energy density), respectively. The order from top to bottom in Figure 2 is from most exergonic (~100 kJ/mole e⁻) to least exergonic (near 0 kJ/mole e⁻) determined for the Manifold B site (dark gray circles). Colored horizontal bars indicate the electron acceptor and show O₂, NO⁻₃, and MnO₂ (in gray, green, and purple) are the most exergonic whereas Fe₃O₄, SO²⁻₄, S⁰, and CO (in brown, orange, yellow, and light blue) provide the least amount of energy, ranging from little to none. Sample locations plot near one another, suggesting similar energetic potential between sites. In Figure 2B, Gibbs energies are recast as energy densities using Equation (3). Here the colored horizontal bars indicate the electron donor of each reaction. In this view, energy yield is high using S⁰, HS⁻, NH⁺₄, Fe⁺⁺, and MnO₂, but low using CH₄, CO, H₂. The distribution between sites is also very large, with individual reactions ranging from highly exergonic to not at all from one site to the other (e.g., Rxn. #27 at site B vs. 3A). The differences between presentations of Gibbs energies and sources of variability are discussed below.

A total of 27 established bacterial and archaeal phyla along with 16 candidate phyla were detected in the 10 water and biofilm samples (Figure 3). All sites are dominated by Bacteria; Archaea are limited to 5% Euryarchaeota in Borehole 5, and 10, 12, and 2.5% Thaumarchaeota in Borehole 5, Pool 9, and Manifold D, respectively. No Crenarchaeota were observed in this study. While a slight primer bias against Archaea is contributing to the observed abundance patterns, this bias should not more specifically exclude the Crenarchaeota. In addition, the general dearth of Archaea was confirmed with sequence independent methods, such as intact polar lipid analysis (data not shown).

Within the Bacteria, the Proteobacteria (18–81%) tend to dominate at all sites. However, the class-level breakdown varies from almost exclusively Betaproteobacteria (55% in Borehole 2) to including Deltaproteobacteria (17–19% in Boreholes 5 and 6, and Manifolds B and D) and/or Gammaproteobacteria (26% in Borehole 8). Also common are the Nitrospirae (8 of 10 samples, with up to 19% at Borehole 5 internal biofilm), the Chloroflexi (all 10 sites, with 20% in Borehole 8), and the Planctomycetes (all samples, with 4–12% in Boreholes 5 and 8 and Pool 9). The manifold samples (B and D) contain far more Firmicutes (Clostridia) and Actinobacteria than other samples, and Pool 6 is anomalously high (25%) in Bacteroidetes. Phyla for which there are no cultured representatives (candidate phyla) comprise a notable percentage of the sequence diversity at most sites (0.3–19%). Unassigned sequences (despite multiple attempts at identification and implementation of multiple pipelines) comprise 2–29% of our libraries, and vary significantly by site. The variability suggests that these taxa are not merely one group for which our pipeline is insensitive, but rather a diversity of unknown organisms. All sequences are accessible on the Short Read Archive (NCBI) database under BioProject PRJNA262938.

The observed distribution of aqueous species suggests varying levels of water rock interaction, equilibration with the atmosphere, and active cycling of redox sensitive elements between samples. In order to more critically evaluate the relationship between different sites and chemical species, we performed a principle component analysis of all measured geochemical parameters (Figure 4). This analysis reveals strong segregation between the chemistry of fluids taken from manifold sites compared to boreholes. The first principle component (PCA 1) appears to be driven by the dissolved oxygen (DO) concentration and ORP with more oxidizing fluids appearing on the right and reducing fluids on the left. Of the sites that were sampled twice, samples from October plot rightward of those from February, reflecting improved sampling techniques that exclude atmospheric oxygen contamination. PCA 2 groups samples with high total sulfide and ammonium concentration together in the lower half, while samples with high conductivity and total dissolved solids cluster in the upper left hand quadrant (e.g., Manifolds B and 3A). The specific geochemical parameters that also plot in this quadrant (e.g., Sr²⁺, SO²⁻₄, Ca²⁺, K⁺, Na⁺, Fe²⁺, He) suggest increased water rock interaction. The helium is presumed to derive from alpha decay, indicating prolonged isolation of these fluids from the atmosphere.

The energetic calculations carried out in this study suggest that there is a considerable amount of energy to be had from catalyzing chemolithotrophic reactions. However, Gibbs energy can be normalized in multiple ways (e.g., moles of e⁻, moles of reactant, moles of product, per kg water, etc.), each of which highlights different aspects—as has been shown by others (McCollom, 2000; Boettger et al., 2013; LaRowe and Amend, 2014). In Figure 2, Gibbs energy yields are presented in terms of Joules per mole of electrons transferred and also in energy density units of Joules per kg of water. Each of these presentations yields unique structure and identifies different key variables.

Figure 2A presents the standard view of Gibbs energy for each reaction in terms of Joules per mole of electrons as well as the distribution of electron acceptors. We see that the most strongly exergonic reactions utilize O₂, NO⁻₃, or MnO₂ as electron acceptors, similar to the canonical view of electron acceptor potency. Also, there is a relatively tight clustering between different sites, suggesting a broad similarity between energy available at each site, despite large geochemical differences. Outliers are reactions involving dissolved gasses whose activities vary dramatically. From this figure we can see which reactions are most exergonic if sufficient reactants are available. In contrast, Figure 2B presents Gibbs energy in terms of energy density, weighting the theoretical energy yield (shown in 2A) by the availability of limiting reactants. Here, the difference in energy availability for many of the reactions spans several orders of magnitude between sites and some electron donors (S⁰, HS⁻, NH⁺₄, and Fe²⁺) yield consistently more energy (0.1–100 J/kg water) than others. This presentation of Gibbs energy indicates the importance of local geochemistry to energy supply. On this plot, energy yield does not decrease systematically from the top to the bottom of the figure, suggesting different controls than in Figure 2A.

To explore this point further, Figure 5 condenses the data points from Figure 2 into histograms that more specifically illustrate the energy available from various electron acceptors (A, C) and electron donors (B, D). Panels A and B (Gibbs energy per mole e⁻ transferred) show a bimodal distribution, with a larger peak centering on 0 kJ (mol e⁻)⁻¹ and a smaller, broader peak extended from about −60 to −90 kJ (mol e⁻)⁻¹. Note in Panel A (binned by electron acceptor) that reactions with O₂, NO⁻₃, and MnO₂ are more exergonic and are clearly separated from the less exergonic reactions with Fe₃O₄, SO⁻²₄, S⁰, HCO⁻₃, and CO. In Panel B (binned by electron donor), no trend is observed (Figure 5B). For example, Fe²⁺ oxidation (red bars) spans nearly the entire range from +30 to −100 kJ (mol e⁻)⁻¹. Does this mean that aerobes and nitrate reducers should dominate the subsurface at SURF?

Unfortunately for the aerobic microbes, the calculations summarized in panel A and B fail to account for limiting reactants in the systems. Oxygen is a high-energy electron acceptor where available, but where it is limiting this potential cannot be realized. Figures 5C,D recast the Gibbs energy calculations as Gibbs energy densities. Figures 5C,D show a single peak with a maximum at 10⁻² J (kg H₂O)⁻¹ with tails to very high and low energy yields. Binning this data in terms of electron acceptors shows little relationship to energy (Figure 5C). For example, reactions using nitrate (green) as the electron acceptor range from 10⁻¹⁰ to 10 J (kg H₂O)⁻¹. Conversely, structure emerges when the calculated energies are binned by electron donor (Figure 5D), with clearly defined profitable (e.g., S⁰, NH⁺₄) and poor (H₂, CO) electron donors. Comparison of the left and right sides of Figure 5 reveals the distribution of molar vs. volumetric Gibbs energy. This analysis illustrates the value of scaling ΔG_r calculations by the limiting reactant for a more realistic view of potential metabolisms. In previous studies, energy density appears to show a stronger relationship to in situ microbial communities than unweighted energetic calculations (McCollom, 2000; LaRowe and Amend, 2014). The relationship between energy yield and the corresponding identities of microbial populations is discussed below.

The oxidation of sulfur, sulfide, ferrous iron, ammonium, and the reduction of manganese oxides are strong in situ energy sources, but how does energy yield vary by site or with temperature? Oxidation of reduced sulfur species yields large amounts of energy per kg H₂O at all sites [10⁻³–10³ J (kg H₂O)⁻¹], regardless of electron acceptor or temperature (Figure S1). Sulfide oxidation coupled to methanogenesis, the reverse of the consortia mediated anaerobic oxidation of methane (AOM) (Orphan, 2001), is highly exergonic at all sites (#102). A similar consortium of methanogens and sulfide oxidizing bacteria could potentially take advantage of this large energy yield in this system. In addition, iron oxidation is a visibly important process within the SURF environment, resulting in iron-rich biofilms at Boreholes 2 and 5 as well as Pool 9 (see Figure 1 for images), and our calculations confirm significant energy yield for iron oxidation to ferrihydrite and goethite (Figure S2). Mineralogy influences these calculations with higher Gibbs energies for the more crystalline products (Goethite > Ferrihydrite). The relative energy yield between sites is determined by the concentration of ferrous iron rather than the oxidant supply (Manifold D >> Pool 9) and increases with temperature. Ammonium oxidation is another significant source of energy, particularly coupled to O₂, S⁰, or MnO₂ reduction (Rxns. # 8, 74, 133; Figure S3). Exergonicity is controlled by the concentration of ammonium with high ammonium sites (e.g., Borehole 8 and Pool 6) yielding energy densities consistently one to two orders of magnitude higher than those with low ammonium concentrations (e.g., Pool 9 and Manifold D). In this case, a negative correlation between energy yield and temperature is seen with higher energy density present at lower temperatures. Finally, manganese oxidation is exergonic at all sites (Rxn. #7, Figure S4), with Gibbs energy yields on par with that of ammonium oxidation and little relationship with temperature.

Clearly, there are many possible exergonic catabolic strategies in the SURF system; however, a number of commonly discussed subsurface metabolisms are absent from this list. Perhaps the most commonly cited subsurface energy source is hydrogen (Nealson et al., 2005), which is an attractive reductant because it has abiogenic sources that are decoupled from surface processes such as the radiolysis of water (Lin et al., 2005; Blair et al., 2007). Hydrogen oxidation with oxygen is a highly exergonic reaction when viewed per mole of electrons (Rxn. # 1; Figure 2 or Figure S4A); however, hydrogen concentrations are simply too low in this system to produce significant energy density (Figure S4B). It is possible that hydrogen oxidizers are consuming hydrogen rapidly as it is generated, creating a cryptic cycle. While it is difficult to evaluate cryptic cycling for this or really any metabolisms using the methodologies described above, molecular data can shed some light on these potential cycles, (discussed below). In addition to hydrogen oxidation, methanogenesis, sulfate reduction, and iron reduction were also considered as potential sources of energy in SURF sites (Rxn. # 92, 42, 117; Figure S4), but were found to yield little to no energy and only at the lowest temperatures. These results contrast with observations of sulfate reducing phyla and probable biogenic methane, but it should be noted that these calculations were performed for only chemolithotrophic reactions.

Now that there is compelling evidence for metabolisms that could yield significant amounts of energy in this environment, it is incumbent to ask if microbes capable of mediating these metabolisms are in fact present. Microbes closely related to known chemolithotrophs (based on 16S rRNA gene sequences) are found in abundance; however, the scope of this analysis is incomplete, because many sequences are from uncharacterized strains. To identify likely physiologies, sequencing data were analyzed at the family level for each site, and the families were classified as physiotypes based on the predominant metabolism of cultured representatives as identified in Bergey's Manual of Systematic Bacteriology and in recent publications of newly defined or cultured groups by Iino et al. (2010), Yamada (2006), Bollmann et al. (2014), Chivian et al. (2008). Where metabolisms were unknown (such as for the unclassified groups, Candidate Phyla, or OTUs with only coarse phylogenetic affiliation), the family was placed into the physiotype marked “?”. At some sites this bin constitutes >50% of the OTUs sampled (Borehole 2), whereas in others it is <10% (Pool 6 biofilm). When the family level proved too metabolically diverse to categorize, the family was put into the physiotype labeled “mixed.” The group labeled “P/S?” consists of organisms related closely to oxygenic or anoxygenic photosynthesizers that are unlikely to be mediating these metabolisms in the dark mine environment.

Figure 6A illustrates the distribution of physiotypes within the sampling locations of SURF. At the largest scale it is clear that there is enormous variability in the potential metabolisms present, which is not surprising given the variety and concentrations of electron donors and acceptors at SURF, and therefore the amount of energy available from a diverse set of catabolic strategies. There are distinct differences between the metabolic potential in filtered fluids vs. biofilm samples, as well as between samples from the 800′ and 4850′ levels. In general, the fluids contain a greater proportion of unclassified metabolisms (and phyla, Figure 3) where biofilms contain more heterotrophic constituents. This observation is consistent with trophic structuring and heterotrophic turnover that would be expected in a microbial biofilm or mat. Heterotrophy in samples that are open to the atmosphere is mostly performed by aerobic heterotrophs whereas borehole sites contain more mixed and anaerobic heterotrophic representatives. A similar switch can be observed in the sulfur system where sulfide oxidation yields to sulfate reduction as the prevailing oxidation state progresses from more oxidizing to more reducing sites.

In most cases, particular families dominate the physiotypes at a given site. The most extreme example of this is the Fe_ox physiotypes that are almost exclusively composed of the betaproteobacterial family Gallionellaceae, which constitutes 54.4% of sequences in borehole 2 fluid. The energy yield of this reaction at site 2 is correspondingly high, up to 0.3–0.4 J per kg H₂O (Rxn. # 2, 3). Generally, the abundance of the Fe_ox physiotype was correlated with increasing energy density from iron oxidizing reactions (r = 0.44, see Figure S5). Only a single known iron reducing bacterium was identified, Geobacteriaceae which was found exclusively in pool 6. This is one of the few sites where this reaction is energy yielding (Rxn. #119). The CH_{4, ox} physiotype is dominated by the Methylococcaceae and most common in Borehole 8. The energy density for aerobic methanotrophy here is correspondingly high (Rxn. 14, 10^−1.5 J/kg H₂O). Other dominant families from potentially chemolithoautophic physiotypes include: H_{2, ox}—Hydrogenophilaceae, N_ox—Nitrospiraceae, Thaumarchaeota, S_ox—Thiotrichaceae, Ectothiorhodospiraceae, S_red—Desulfobacteraceaea, Desulfobulbaceae, Peptococcaceae/Desulforudis.

Comparison of these physiotype abundances to the energy yields shown in Figure 5 illustrates the correspondence between energy density and resident microbial populations. For instance, organisms capable of oxidizing S⁰ and HS⁻ are found in abundance in locations where there are significant amounts of energy available—per kg H₂O—for these reactions (Figure S5). The same correlation can be made for iron oxidation and nitrogen oxidation. The presence of sulfate reducing and hydrogen oxidizing bacteria do not agree well with energetic predictions. In the case of hydrogen oxidation, the presence of H_{2, ox} communities where measured hydrogen concentrations are too low to yield significant energy, requires that it is produced and consumed by proximal consortia, and either cannot be accurately measured and/or never accumulates in the system. The occurrence of this process is supported in our dataset by the predominance of the H_2,ox physiotype in biofilm as opposed to fluid samples.

To further elucidate the relationship between geochemistry and microbial physiotypes, a principle component analysis of both sets of measurements was performed (Figure 6B). Here, vectors for each physiotype are plotted in black and geochemical species in gray, in a principle coordinate system derived from both systems at each site. Species and physiotypes that co-vary in abundance should point in similar directions, whereas physiotypes that are dependent on two chemical species will appear between the two chemical vectors. SURF sample sites are plotted in symbols and their positions reflect proximity to their constituent physiotypes and chemical species.

In this ordination, sites are broadly distributed across the diagram with clustering between 800′ level samples, manifold sites, and sulfidic sites from 4850′ level. Physiotypes are similarly distributed with clusters of CH_{4, ox}, S_ox, and H_{2, ox} in the upper left, N_ox and Fe_ox on the right, and S_red and Het_anaerobic in the lower left. An example of a physiotype that depends on multiple chemical species is Fe_ox, which appears directly in between the DO and Fe⁺⁺ vectors, illustrating its dependence on both variables. The same is true for CH4_ox and N_ox. This also suggests that O₂ is the terminal electron acceptor for these metabolisms, rather than NO⁻₃, SO²⁻₄—or other oxidants. Aerobic heterotrophy is very closely associated with DO, suggesting that oxygen, not organic matter is limiting in these systems. In contrast, there is a close correspondence between the chemical species HS⁻ and H₂ and their oxidizing phylotypes S_ox, and H_2,ox, respectively, suggesting that reductant supply is the determining factor for these metabolisms. Mixed heterotrophy appears antithetical to anaerobic and aerobic heterotrophy, suggesting that it is a combination of the two rather than a poorly characterized phylum.

The diagram shown in Figure 6B can be used to infer metabolisms for the unknown physiotypes. For instance, in the case of sulfate reduction, it can be seen that this physiotype (S_red) plots in the lower left, very near to Het_anaerobic and “?” physiotypes. Sulfate reduction with hydrogen was shown to be a poor source of energy in this system (see Chemolithotrophy), which seemingly contradicts the abundance of SRBs in a number of sites. Notably, the S_red vector does not fall in between H₂ and SO₄, and instead appears near that for anaerobic heterotrophy. Perhaps these sulfate reducers are using organic matter rather than H₂ as an electron donor. Many of the SRBs identified in SURF are capable (and often require) heterotrophic modes of growth, lending support to this idea. The big unknown in these, and in most ecosystems, is the metabolic potential of the unclassified organisms. While this analysis is not detailed enough to suggest a specific metabolism, the location of the “?” physiotype near S_red and in the neighborhood of CH_{4 genesis} and Het_anaerobic appears to suggest at the very least that these organisms are likely anaerobes. Further efforts to isolate these unknown organisms and characterize their metabolic potential through genomic studies are required to further resolve these questions.

This is not the first study to quantify environmental energy availability; previous studies have targeted diverse environments including terrestrial hot springs, submarine hydrothermal systems, igneous marine basement, and marine sediments (McCollom, 2000; Amend et al., 2003, 2011; Shock et al., 2005, 2010; Spear et al., 2005; LaRowe et al., 2008; Vick et al., 2010; LaRowe and Amend, 2014; Teske et al., 2014). While the sites are unrelated, there appears to be an upper limit on ΔG_r of approximately −120 kJ (mole e⁻)⁻¹. This value reoccurs in a diversity of environments, including terrestrial hydrothermal systems, marine sediments, marine basement, and shallow marine hydrothermal systems. In different environments, different reactions are the most exergonic: for example, aerobic iron oxidation to magnetite in Yellowstone hot springs evaluated in Shock et al. (2010), the knallgas reaction in Juan de Fuca Ridge basement fluids described in Boettger et al. (2013), and nitrite reduction with H₂ in an acidic thermal fluid on Vulcano Island (Italy) noted in Amend et al. (2003); several other reactions, however, yield similarly high energy. In the SURF environments described here, the most exergonic reaction is aerobic carbon monoxide oxidation, with a ΔG_r of −118 kJ/mole e⁻. These most exergonic reactions use O₂, NO⁻₃, or NO⁻₂ as the terminal electron acceptor, which is consistent with the canonical electron tower model of microbial metabolism. If this were the whole story, it would have significant and grave implications for subsurface life that is truly isolated from surface processes, not to mention life in the Archean oceans.

In this light, the reversal of fortunes that occurs when ΔG_r is instead presented as energy density is most revealing. Energy yields switch from being very dependent on the electron acceptors O₂, NO⁻₃, and MnO₂ to instead depending on the electron donors S⁰, HS⁻, NH⁺₄, Fe⁺⁺, and Mn⁺⁺. Notably, this switch is present almost wherever both formulations of ΔG_r are presented. For instance, Boettger et al. (2013) saw energy yield switch from being primarily dependent on O₂ and NO⁻₃ to depending on CH₄ and H₂ with a range of electron acceptors. Similar trends are observed in LaRowe and Amend (2014) with the energy density emphasis on sulfur species. Similar to this study, they found that energy density calculations correspond more closely to resident populations than kJ (mole e⁻)⁻¹ calculations. Viewing energetic potential in terms of energy density may both more accurately represent environmental systems and introduce a number of potentially profitable, surface independent, metabolisms to fuel the DSB.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.