H2 is probably the most important substrate for terrestrial subsurface lithoautotrophic microbial communities. Abiotic H2 generation is an essential component of subsurface ecosystems truly independent of surface photosynthesis. Here we report that H2 concentrations in fracture water collected from deep siliclastic and volcanic rock units in the Witwatersrand Basin, South Africa, ranged up to two molar, a value far greater than observed in shallow aquifers or marine sediments. The high H2 concentrations are consistent with that predicted by radiolytic dissociation of H2O during radioactive decay of U, Th, and K in the host rock and the observed He concentrations. None of the other known H2-generating mechanisms can account for such high H2 abundance either because of the positive free energy imposed by the high H2 concentration or pH or because of the absence of required mineral phases. The radiolytic H2 is consumed by methanogens and abiotic hydrocarbon synthesis. Our calculations indicate that radiolytic H2 production is a ubiquitous and virtually limitless source of energy for deep crustal chemolithoautotrophic ecosystems.
 One constraint on the extent of subsurface microbial life is whether sufficient energy exists to sustain a minimal metabolism [McKay, 2001]. Pedersen  and Stevens and McKinley  have proposed that some subsurface microbial communities rely on substrates derived from geochemical processes rather than from photosynthetically derived organic carbon. These lithoautotrophic ecosystems necessitate abiogenically produced H2. H2 generates substantial energy for various electron acceptors [Amend and Shock, 2001] and its high diffusivity makes it readily available to microorganisms in confined pore spaces. H2 is also crucial for Fischer-Tropsch (F-T) synthesis of organic compounds [Cody et al., 2000; Rushdi and Simoneit, 2001; Sherwood Lollar et al., 2002] where H2, catalyzed by metallic phases, reacts with CO/CO2. The C and H isotopic signatures of these hydrocarbons are distinct from those of thermogenic hydrocarbons and microbial CH4 [Sherwood Lollar et al., 2002]. Microorganisms utilizing F-T synthesized organics as substrates would be independent of photosynthetically derived organic matter.
 The few dissolved H2 analyses for deep subsurface environments vary dramatically from <1 nM to mM [Haveman and Pedersen, 1999; Marine, 1979; Stevens and McKinley, 1995; Vovk, 1987]. These data when combined with the low abundance of dissolved organic compounds and evidence from microcosm experiments suggest that autotrophic assimilation dominates over heterotrophic metabolisms in these environments [Pedersen, 2001; Stevens and McKinley, 1995] and that fermentation is not the source of the H2. Abiotic H2 production by water-rock interaction and/or tectonic activity have been proposed [Chapelle et al., 2002; Stevens and McKinley, 1995], but whether these processes can sustain the observed microbial ecosystems remains uncertain. Vovk  proposed that radiolysis of water explains high subsurface H2 concentrations. His theory is consistent with observation of H2-rich fluid inclusions from a natural fission reactor at Oklo, Gabon [Savary and Pagel, 1997]. Other abiotic H2 generation processes have been published, but a systematic comparison of them for one subsurface environment has not been undertaken until this study.
 To investigate the origins of high H2 concentrations in the deep subsurface and to determine whether water radiolysis plays a role in H2 generation, we analyzed the radiolytic and radiogenic products, dissolved H2 and He, in 24 groundwater samples collected from 6 Au and coal mines of the Witwatersrand Basin, South Africa. We evaluated the thermodynamic potential of each candidate H2 generation reaction under the in situ conditions to account for the observed H2 concentrations.
2. Geological Background and Analytical Methods
 Witwatersrand Basin stratigraphy consists of the 2.9 Ga Witwatersrand Supergroup (mainly quartzite), overlain by the 2.7 Ga Ventersdorp Supergroup volcanic sequence and either the 2.5 Ga Transvaal Supergroup dolomite or the Permo-Carboniferous Karoo sandstone and shale [Robb and Meyer, 1995]. Subvertical dykes and fractures compartmentalize the hydrological system [Omar et al., 2003]. Au mining operations depressurize this system and release dissolved gases associated with either brines or relatively fresh fracture water recharged from the overlying aquifers. High H2 concentrations (up to 33% volume) have been detected in fracture water [Cook, 1998] and within mineral fluid inclusions (2–5% volume) [Drennan et al., 1998].
 Twenty-four fracture water samples were collected from exploration boreholes in five Au mines and one coal mine, Beatrix, Kloof, Driefontein, Mponeng, Evander and Sasol, at depths of 700 to 3300 m. The boreholes were isolated from the mining environment by a stainless steel telescoping packer (some internally lined with nylon tubing) attached with a valve-control manifold. Dissolved gases were stripped into an overturned funnel and captured in preevacuated sealed serum vials, which were subsequently over-pressurized by several mL of fracture water to prevent air contamination. Noble gas samples were collected in Cu tubes [Lippmann et al., 2003]. Samples for aqueous geochemistry were collected by filtering (0.2 μm) and diverting fracture water to serum vials with or without preservatives added. The samples were stored in a −20°C freezer or 4°C refrigerator in the field lab, and later transported back to USA on dry ice for analyses.
 H2 and CH4 were analyzed by a Kappa-5 Reduced Gas Analyzer (RGA 5, Trace Analytical, Sparks, MD) equipped with a Hg vapor detector (reduced gas detector; RGD) for H2 and a flame ionization detector (FID) for CH4. Gases were extracted from sample bottle using a gas tight syringe and diluted to a concentration less than 10 ppm. A total volume of 5 ml diluted sample was injected through an injection port and divided evenly into two streamlines for FID and RGD where HgO was converted to Hg vapor. Compositional analyses of gas samples were also performed at the Stable Isotope Laboratory of the University of Toronto. A Varian 3400 gas chromatograph (GC) equipped with a FID was used to determine the concentrations of CH4, C2H6, C3H8 and C4H10. The hydrocarbons were separated on a J&W Scientific (Folsom, CA) GS-Q column (30 m × 0.32 mm ID) with a He gas flow and the following temperature program: initial 60°C hold 2.5 minutes, increase to 120°C at 5°C min−1. A Varian 3800 GC, equipped with a micro-thermal conductivity detector (μTCD) and a Varian Molecular Sieve 5A PLOT fused silica column (25 m × 0.53 mm ID), was used to determine the concentrations of the inorganic gas components (H2, He, Ar, O2, CO2 and N2). To determine the concentrations of Ar, O2 and N2, the He carrier gas flow rate was 3 mL min−1 and the temperature program was: initial 30°C hold 6 minutes, increase to 80°C at 15°C min−1, hold 4 minutes. To determine the CO2 concentration, the He carrier gas flow rate was 50 mL min−1 and the temperature program was: initial 60°C, increase to 250°C at 20°C min−1, hold 6 minutes. To determine the concentrations of H2 and He, the Ar carrier gas flow rate was 2 mL min−1 and temperature program was: initial 10°C hold 10 minutes, increase to 80°C at 25°C min−1, hold 7 minutes. For water samples with >1% H2 the RGA and Varian 3400 GC analyses agreed to within a factor of 2 despite the 1000 times dilution for the RGA analyses. All analyses were run in triplicate and mean values are reported. Reproducibility for triplicate analyses was ±5% relative error. Noble gas analyses, including He, were performed on Cu tube samples using the procedures of Lippmann et al. .
 The measured H2 and He concentrations were corrected for loss through diffusion to the fracture headspace that is created behind the tunnel face as the fracture is dewatered [Lippmann et al., 2003]. The correction was based on the observation that the concentration of the dissolved atmospheric noble gases, 20Ne, 36Ar, 84Kr, 132Xe, progressively increased with increasing atomic mass in a manner consistent with diffusive loss at in situ temperatures from initial concentrations that were in equilibrium with the atmosphere during recharge [Lippmann et al., 2003]. The deficiency of other dissolved gases can be adjusted according to the differences between their diffusion coefficients and that of 20Ne and the following relationship [Lippmann et al., 2003]:
where Ci is the measured concentration of specific gas i, Cio is the concentration of specific gas i in equilibrium with atmosphere composition, and Di is the diffusion coefficient [Jaehne et al., 1987] of specific gas i at the in situ temperature Ts.
 The corrected values were converted to the dissolved concentrations following the procedure of Andrews and Wilson . Flow rate measured during sampling varied by less than 10%. The uncertainty for the correction for diffusion loss is ±10%. Error propagation yielded an uncertainty less than ±20% for the reported values (after the correction for diffusive loss).
 The cations were measured by an Optima 4300 DV inductively coupled plasma-atomic emission spectrometer (Perkin-Elmer, Wellesley, MA). Anions were measured with a DX-320 ion chromatograph (Dionex, Sunnyvale, CA). Dissolved inorganic carbon was determined by an infrared spectrometer (Licor 6352, Lincoln, NE). The total organic carbon (TOC) was measured from acidified samples as CO2 generated by catalytic combustion using an infrared detector (Tekmar Dohrman DC-190, Perkin-Elmer, Wellesley, MA) and represented nonpurgeable organic fraction. The pH and Eh of the fracture water were measured in the field. The uncertainty for aqueous chemistry is ±5%.
 The microcosm experiment for H2-utilzing metabolisms was carried out by supplying the headspace with 80% of H2 and 20% of CO2, and 0.05% (final concentration) of Na2S as a reducing agent. The control sample was prepared in the same way but autoclaved right after the purging of headspace gases. The samples were incubated at the in situ temperature and the abundances of headspace gases were measured by a RGA 5 and compared to those of the control.
3. Dissolved H2 and He Abundances
 H2 abundances ranged up to 27% of the combustible gases, and the measured concentrations spanned over five orders of magnitude with a maximum of 7.41 mM (Table 1). The sample with the second highest measured H2 concentration (4.98 mM for sample 12) was collected from a 3-day-old, tens of meters long, diamond-drilled borehole into metavolcanic rocks at 3.0 kilometers below land surface (kmbls). The ∼10 L min−1 flow rates for the high-salinity (0.23 M Cl−), nonmeteoric fracture water and gas (mostly CH4, H2 and He) indicate that the borehole had been flushed by >2000 borehole volumes of water (Table 1) prior to sampling. By contrast the lowest measured H2 concentration (16 nM for sample 4) was collected from a one-week-old borehole intersecting the contact between quartzite and a diabase dyke at 1.29 kmbls where the flow rates of low-salinity (0.04 M of Cl−) meteoric fracture water and gas (mostly CH4, N2 and He) [Ward et al., 2004] were 37.5 and 0.04 L min−1, respectively. The H2 concentrations were not correlated with depth, salinity, pH, rock type, borehole age, fracture water age or any other measurement (correlation coefficients were all < 0.5), but the highest concentrations were typically found in the deeper, highly saline, nonmeteoric, older fracture water [Lippmann et al., 2003].
Table 1. Characteristics of Groundwater Samples From the Witwatersrand Basin
Borehole volumes flushing borehole prior to sampling are minimum estimates based on time since intersection, borehole volume, and flow rate at the time of sampling, which is typically much less than initial flow rate.
Data from Ward et al. ; δ13C and δ2H of CH4 for methane origin: B, biogenic origin; A, abiogenic origin.
All analyses were run in triplicate, and mean values are reported. Reproducibility for triplicate analyses was ±5% relative error. The measured values were converted to the dissolved concentrations following the procedure of Andrews and Wilson . Flow rates measured during sampling varied by less than 10%. The uncertainty for the correction for diffusive loss is ±10%. Error propagation yielded an uncertainty less than ±20% for the reported values. Corrected H2 concentrations were referred to the original H2 abundances corrected for consumption by abiogenic CH4 formation and/or diffusive loss. The calculations were only restricted to the samples with abiogenically isotopic signatures due to the uncertainties whether H2-consuming methanogens were solely responsible for microbial H2 utilization.
Data for samples 1–8a, 9–11 and 16 from Ward et al. . MB, Methanobacterium; MSae, Methanosaeta; ML, Methanolobus; MSar, Methanosarcina; MSpir, Methanospirrillum; ND, Archaea were not detected. MB for sample 14 was just above detection limit.
NA, not available.
No diffusion correction required from noble gas analyses.
 The measured He concentrations (mostly 4He) (Table 1) ranged over three orders of magnitude and were not correlated with the H2 data, but were correlated with Cl− (correlation coefficient = 0.98). Depending on the assumptions regarding crustal fluxes of 4He, 40Ar and 134,136Xe, the He concentrations correspond to subsurface residence times of 1.5 to ∼20 Myr [Lippmann et al., 2003] during which time the Witwatersrand Basin was at present pressure and temperature conditions [Omar et al., 2003].
 Nine samples were corrected for degassing into the fracture headspace as the fracture was dewatered using noble gas analyses [Lippmann et al., 2003]. The corrections increased H2 and He concentrations by ∼10× (Table 1) but did not impact the H2 to He ratios due to their comparable diffusion coefficients [Jaehne et al., 1987]. Our data overlap published dissolved H2 and He concentrations for groundwater from Precambrian Russian [Vovk, 1987] and Fennoscandian Shields [Haveman and Pedersen, 1999] and the Triassic Dunbarton Basin [Marine, 1979] (Figure 1).
 Although anomalously high H2 concentrations that have been reported for shallow sedimentary boreholes are artifacts of the drilling process [Bjerg et al., 1997; Bjornstad et al., 1994], the H2 concentrations reported herein are not due to drilling for the following reasons:
 1. Bjornstad et al.  and Bjerg et al.  used percussion drilling into shallow unconsolidated sands with no drilling fluid circulation, whereas mine boreholes are drilled with high-rpm diamond bits cooled by mine water circulating at a rate of L min−1.
 2. Bjornstad et al.  reported that tiny fragments of metal casing generated by pulverization of the rock during percussion drilling created a slurry that reacted with static groundwater to produce H2 at a rate of ∼1–10 nM s−1, but that flushing the borehole with groundwater removed this effect within ∼12 hours. In mine drilling the circulating water removes cuttings and bit fragments from the borehole before they have time to react. Furthermore, mine boreholes are drilled under pressure so that upon intersection of a fluid-filled fracture (at least several MPa) residual mine water and cuttings are removed by the flushing of thousands of borehole volumes of fracture water (Table 1).
 3. Bjerg et al.  reported H2 concentrations of ∼50 μM during drilling and a slow decline over days, whereas H2 concentrations in mine boreholes are observed to slowly increase over days after drilling (Table 1). Four samples of sample 14 over 54 days (∼2700 to 5700 borehole volumes) documented increasing H2 concentrations (0.3 to 3.7 mM) with increasing gas/water flow rate ratios and constant gas composition (H2 ranged from 10–12 mole%). The H2 concentration of sample 8c immediately after fracture intersection (∼10 borehole volumes) was ∼9 times less than that 2 weeks later (sample 8d; ∼5000 borehole volumes). The increasing gas/water flow rates probably represent degassing of water within the fracture as the fracture is drained as inferred from atmospheric noble gas analyses [Lippmann et al., 2003].
 4. Bjerg et al.  observed <5 nM H2 in adjacent boreholes at <4 m distance, indicating that H2 migrated through permeable aquifer to adjacent wells, whereas the boreholes reported herein are separated from other boreholes by hundreds of meters of low-permeability (<μDarcy) rock and because all drilling is below formation pressure any drilling-produced H2 migrates out of the borehole with the drilling fluid, not into and through the rock formation.
 5. Bjerg et al.  observed a longer term, anomalously high H2 concentration (∼100 nM) created by slowly pumping groundwater (100 mL min−1) through 10 m of 2-cm diameter Black Iron casing (∼2000 seconds of reaction time) and no anomalously high H2 for stainless steel casing. The boreholes used in this study had either a 1 m long steel casing at the outlet or no casing at all and for the observed flow rates water would have only 1 to 30 seconds to interact with the steel to produce H2. H2 concentrations and gas composition exhibited no correlation with water flow rates or the presence or absence of steel casing.
 The potential for H2 and acetate production from decomposition of complex organic compounds by fermentative microorganisms (mechanism 1) [Boone et al., 1989] was evaluated by calculating the Gibbs free energy of fermentation reaction for propionate under the in situ conditions:
The concentrations of reactants and products were derived from field and laboratory measurements (Table 2). The calculation yielded positive free energies for all the samples except samples 7, 8c and 9 (Table 3) and indicated that fermentation is prohibited in most samples because of the high concentrations of dissolved H2. In addition to propionate, the uncharacterized carbon in the TOC pool can also be the potential substrate for fermentation. One extreme scenario is to assume that the residual TOC after subtraction of acetate, formate, propionate, and cell biomass is equivalent to butyrate. This is the maximum amount of organic matter available for fermentation. The fermentation of the hypothetical butyrate was assumed to produce acetate, H2, and bicarbonate (reaction 2 in Table 3). The free energies were positive for most samples with the exception of samples 4, 6, 7 and 8d (Table 3). Of the other investigated potential substrates, including long-chain carboxylic acid (C5 to C12), benzoate, malonate, benzene and toluene, the fermentation reaction yielded positive free energies for most samples. The failure of microbial fermentation to explain the high H2 concentrations in the Witwatersrand Basin therefore is not dependent on the assumed organic substrate.
Table 2. Geochemical Compositions of Groundwater Samples Used in the Free Energy Estimationa
 H2 production can also occur by oxidation of Fe2+-bearing silicates (e.g., olivine and augite) with reduction of water (mechanisms 2 and 3) [Coveney et al., 1987; Stevens and McKinley, 2000]. These minerals are absent in the strata of the Witwatersrand Basin because of the low-grade greenschist facies metamorphism during the early Proterozoic. The Witwatersrand Supergroup contains chlorite- and chloritoid-bearing shales, whereas the Ventersdorp Supergroup volcanic units are composed of hornblende replacing clinopyroxene phenocrysts with chlorite, epidote and minor clinozoisite, tremolite and actinolite present in the matrix [Schweitzer and Kroener, 1985]. Minor ultramafic intrusives are composed of hornblende pseudomorphs of clinopyroxene with chloritic groundmass. The greatest potential for H2 production by these two mechanisms therefore arises from the 1.4 Ga Pilanesberg dolerite/syenite composite dykes and the 200 Ma Karoo dolerite dykes, respectively. The Pilanesberg dykes contain mafic phenocrysts of augite partially altered to hornblende and encompassed by biotite and magnetite, and minor chlorite and actinolite are present [Ferguson, 1973; Van Niekerk, 1962]. The Karoo dolerites are composed of similar mafic mineral assemblage with fewer hydrous phases. In order to assess the potential of these mechanisms, equilibrium H2 concentrations were calculated for oxidation of these Fe-bearing assemblages. The calculations were based on the assumption that the Fe-bearing products for the reaction were either magnetite or goethite and quartz. Our calculations indicated that equilibrium H2 concentrations for magnetite and hydrous Fe-bearing phases, such as hornblende and greenalite, can only yield sub-nM H2 at the in situ pH and pe. Fayalite and ferrosilite can generate H2 concentrations ranging from μM to tens of mM, but they are absent in these geological formations. The equilibrium H2 concentration derived from oxidation of the Fe-bearing clinopyroxene end-member, hedenbergite, is inversely related to the pH and Ca2+ activity (Figure 2). In the deep, highly saline fracture water where mM H2 was found, this equilibrium H2 concentration falls well below the observed concentrations (line c in Figure 2). In the low-salinity, circum neutral fracture water that typifies shallower fracture zones, the equilibrium H2 concentration falls well above the μM H2 observed (line d in Figure 2). From a strictly thermodynamic perspective, the oxidation of Fe-bearing clinopyroxene associated with the sparse mafic intrusions can account for some of the observed H2 concentrations in the shallower fracture zone, but not the extremely high H2 concentrations found in the deepest, Ca-rich, saline fracture water.
 The equilibrium H2 concentration from oxidation of FeS to FeS2 (mechanism 4) is inversely related to the dissolved HS− and depends on the pH with a constant H2 concentration for pH < 7 and diminishing H2 concentration with increasing pH above 7 (line a versus b in Figure 2). For sulfidic, saline, deep fracture water, the H2 concentration in equilibrium with this reaction falls well below the observed mM H2 concentrations (line a in Figure 2). For shallow, low-salinity fracture water, the H2 concentration in equilibrium with this reaction lies above the observed μM H2 concentrations despite the low sulfide concentrations (line b in Figure 2). The pyrite precipitation reaction can account for some of the low H2 concentrations, but it cannot produce tens of mM H2 concentrations observed in the deeper, alkaline fracture water.
 H2 can be generated by the thermal decomposition of long-chain hydrocarbons and small carboxylic acids (acetate and propionate) in the presence of metallic sulfide or metals as catalysts under the hydrothermal conditions (above 200°C and 350 bars) (mechanism 5) [Seewald, 2001]. The free energy for organic dehydrogenation reactions were positive for samples with H2 concentrations >1 μM (Table 3), precluding a thermogenic origin of the H2 in these samples. Although thermogenic H2 is compatible with the free energy calculations for the samples with <1 μM H2 (samples 4, 5, 7, 8c and 9 in Table 3), it is not consistent with the isotopic analyses of the C1–4 hydrocarbon compounds [Ward et al., 2004]. This suggests that thermodecomposition of hydrocarbons and carboxylic acids can account for nM H2 concentrations, if the reaction rates are not too sluggish at temperatures <60°C. H2 production by this mechanism, however, cannot account for the observed μM to mM H2 concentrations. Nor is the thermodecomposition of the alkanes consistent with the observed C and H isotopic compositions [Ward et al., 2004].
 H2 generation from reduction of H2O by broken Si-O bonds has been proposed as a mechanism to explain H2 anomalies associated with active fault zones (mechanism 6) [Kita et al., 1982]. On the basis of the experimental data of Kita et al. , quartzite or granite with 1% porosity that are completely crushed to less than 0.5 mm in a fault zone will produce ∼1 to 10 μM H2. This does not explain the mM H2 concentrations observed in our fracture zones even if they were tectonically active fault zones, which they are not.
4.2. Radiolytic Model
 Radiolytic H2 is produced through dissociation of water by α, β and γ particles released during radiogenic decay of U, Th and K [Spinks and Woods, 1990]. These particles dissociate H2O into e−, H+, H• and OH• radicals, H2 and H2O2, which then react within μs to yield H2, O2, and H2O2. The 4He concentration is the sum of the α dose and the 4He that is diffused from the solid matrix into the water. The H2 and He generation rates depend primarily on the U, Th and K concentrations of the strata and not on environmental conditions (e.g., pH or pe), and they increase only slightly with decreasing porosity because of the complementary increase in bulk U, Th and K concentrations [Hoffmann, 1992].
where i represents an α, β, or γ irradiation, Enet (J kg−1 sec−1) is the net absorbed dose rate of pore water, W is the weight ratio of pore water to rock (3.7 × 10−4 for a porosity of 0.1% and rock density of 2.7 g cm−3), Ei (J kg−1 sec−1) is the apparent dose rate from decay of U, Th, and K, Si is the stopping power of silicate matrix (Sα: 1.5, Sβ: 1.25, Sγ: 1.14) [Hoffmann, 1992], Gi (mole J−1) is H2 yield per unit of absorbed energy (0.96, 0.6 and 0.4 molecules (100 eV)−1 for α, β, and γ radiation, respectively) [Harris and Pimblott, 2002; Spinks and Woods, 1990], and Y (mole kg−1 sec−1) is the H2 production rate. We calculated the apparent dosage rate for a specific particle and corrected the effect of stopping power for silicates using equation (3) to obtain the net dosage rate (Table 4). H2 production rates for a specific particle were derived from the product of the net dosage rate and H2 yield and were summed together to obtain the bulk H2 production rate using equation (4). Diminishing radiogenic U, Th and K concentrations by decay were not considered because the noble gas analyses indicated that the water residence time was less than 100 Ma [Lippmann et al., 2003]. 4He production rate was calculated on the basis of the same rock chemistry used for H2 production.
Table 4. Parameters and Results for Calculations of Model H2 and He Productionsa
Enet-α, ev s−1 g−1
Enet-β, ev s−1 g−1
Enet-γ, ev s−1 g−1
H2 Rate, nM yr−1
He Rate, nM yr−1
Enet, net dosage for α, β, or γ irradiation.
Up. Wits Supergroup
Lw. Wits Supergroup
Up. Vredefort Crust
Lw. Vredefort Crust
 The calculation yields H2 and 4He generation rates ranging from 10−1 to ∼1 nM yr−1 and 10−4 to 10−2 nM yr−1, respectively. For the U concentrations estimated from Xe isotope analyses of the fracture water [Lippmann et al., 2003], the H2 and 4He production rates are 102 nM yr−1 and 1 nM yr−1, respectively. Although the high U concentrations of the ore zones elevate the production rates of H2 and 4He, their minor volumetric abundance limits their contribution to <10% of the total H2 and 4He production rates for the Witwatersrand strata. The predicted radiolytic H2 concentration for each stratigraphic unit (cross symbols in Figure 1) was calculated by multiplying the measured He concentration by the theoretical H2/He production ratio.
4.3. Comparison of Modeling Results With Analytical Data
 The predicted H2-He relationship bounds the five high H2 concentrations (samples 7, 12, 14c, 15a and 15b in Figure 1), while nine other samples (8a–d, 9, 11, 14a–b and 14d) with similar H2 concentrations but more He lie slightly below the predicted line. Given that the predicted radiolytic H2 concentrations are well constrained by the measured He, U, Th, and K concentrations and radiolysis theory and that the diffusivities of H2 and He are similar, the most logical explanation for H2 data that are less than that predicted by radiolysis is H2 consumption by abiotic reactions and/or microbial H2 oxidation.
 The C and H isotopic compositions of C1–4 hydrocarbons were used to infer the presence or absence of H2-consuming processes. Samples 9–15a (Figure 1 and Table 1) yielded C1–4 isotopic compositions indicative of abiogenic processes [Ward et al., 2004]. Six samples (samples 7, 8a–d and 15b in Table 1) yielded CH4 isotopic values consistent with a mixture of 80–85% abiogenic CH4 and 15–20% of microbial CH4 [Ward et al., 2004]. In the 15 abiogenic CH4-bearing samples, the amount of H2 consumed by the abiotic CH4 formation (4H2 + CO2 ↔ CH4 + 2H2O) was calculated using a H2:CH4 stoichiometric ratio of 4:1. This H2 concentration was added to the measured H2 concentration (w/o correction for diffusive loss) to obtain initial H2 concentrations ranging from 1.9 mM to 1.6 M (Figure 1 and Table 1). Correction of the H2 concentration for the H2 consumed by the measured abiogenic CH4 shifts samples 9–15a closer to the predicted radiolytic H2 concentration (Table 1). These slightly higher H2 concentrations are still consistent with the predictions of radiolysis. A similar correction to the H2 concentrations from methanogen-bearing fracture water cannot be performed as some or all of the CH4 could have originated from acetoclastic methanogenesis.
 Samples 1–8, 15b and 16 (Figure 1 and Table 1) possessed CH4 isotopic compositions indicative of methanogenesis [Ward et al., 2004]. This inference was consistent with the results derived from cultivation, microcosm experiments and 16S rDNA sequence analyses (Table 1). One H2-utilizing methanogen has been isolated from a 2 kmbls fracture zone near to the location of samples 1–5 [Bonin and Boone, 2004]. Microcosm experiments showed that sample 3 produced CH4 during a two-month incubation at the in situ temperature with H2 in the headspace, whereas sample 12 did not (Figure 1). 16S rDNA sequences affiliated with methanogens were detected for samples 1, 6, 8a, 8b, 14a, 14c, 14d, 15a, 15b, and 16 (Table 1). The abiogenic H2 consumption process appears to dominate in fracture zones >2 kmbls, whereas biogenic H2 consumption is more prevalent in fracture zones <2 kmbls.
 Although the H2O2 produced during radiolysis has not been detected in the fracture water samples, 5 to 500 mM of SO42− has been measured in the rock pore water (T. C. Onstott et al., unpublished data, 2004). These concentrations are higher than observed in the fracture water and likely represents the oxidation product of sulfide in the rock by radiolytically produced O2, OH• and H2O2. The S2− and Fe2+ derived from anaerobic microbial metabolisms (sulfate reduction and iron reduction) could also be converted by these oxidants to their oxidized equivalents, providing a cycling mechanism that would remove the deleterious oxidants while sustaining anaerobic metabolism.
5. Subsurface H2 Flux
 Steady state diffusive crustal H2 fluxes produced by water radiolysis in the Witwatersrand Basin were estimated from equations (5) and (6):
where R (μM yr−1) is the net production rate of H2, J (μmoles m−2 yr−1) is the H2 flux, D (m2 yr−1) is the diffusion coefficient [Jaehne et al., 1987], dC/dz (μM m−1) is the concentration gradient, and ϕ is the effective porosity of rocks (∼1%). Calculated H2 fluxes for the top 20 km of the Witwatersrand Basin are ∼8 μmoles m−2 yr−1 assuming no H2 consumption. This is less than an H2 flux of ∼50 μmoles m−2 yr−1 calculated for a shallow aquifer at Lake City, South Carolina based on acetate and formate fluxes reported for a shale-sandstone transition [McMahon and Chapelle, 1991] and a hypothetical fermentation reaction:
 Both fluxes are substantially less than the 104 to 106 μmoles m−2 yr−1 H2 fluxes reported for intertidal and subtidal microbial mats [Hoehler et al., 2002]. H2 fluxes and turn over rates increase dramatically from the less productive lithoautotrophic to highly productive, solar powered, photosynthetic ecosystems. In surface microbial mats, the flux of fixed organic carbon from photosynthetic microorganisms fuels fermentative bacteria, thereby enhancing H2 production and H2 concentrations are controlled either by diffusion to the atmosphere or H2-utilizing microorganisms. In shallow aquifers, fermentative microorganisms capable of utilizing dead, photosynthetically fixed organic carbon generate a lower biogenic H2 flux. The H2 concentrations controlled by H2-utilizing anaerobes remain at levels sufficient to keep the fermentation reaction exergonic and capable of sustaining the fermentative microorganisms. The estimated radiolytic H2 flux from the shale aquitard to sandstone aquifer would contribute only ∼1% (∼0.5 μmoles m−2 yr−1) to this total H2 flux [Onstott et al., 1998], indicating that biogenic H2 still dominates over abiotic H2 production in shallow aquifer ecosystems.
6. Conclusions and Implications
 In deep subsurface environments like the Witwatersrand Basin where the fracture water flow is very slow and organic carbon is scarce, the influence of fermentative production would be much less than in shallow environments and H2 concentrations are well above the threshold for fermentative reactions, indicating that biogenic contributions are negligible. Although other abiotic mechanisms could be contributing to the total H2 production in the Witwatersrand Basin, they are not required as radiolytic H2 production more than suffices to explain the observed H2 concentrations (up to 2 molar) and to support the lithoautotrophic communities detected in the fracture water. Unlike all the other mechanisms which depend on a combination of geochemical and mineralogical species, the radiolysis of water only requires water and K (or other radioactive elements) and will diminish with the slow decay of the radiogenic elements. A radiolytically driven power source therefore seems a plausible alternative for subsurface life on other planets and satellites where surface conditions prohibit photosynthesis, such as Mars and Europa [McKay, 2001].
 This work is supported by NSF grants (EAR-9978267) for the LExEn (Life in Extreme Environments) Program. We thank the team members of the Witwatersrand Microbiology project and geologists of Driefontein, Evander, Mponeng, Kloof, Beatrix, and Sasol mines for their assistance in field sampling and coordination of logistical supply. We also acknowledge E. Van Heerden and D. Litthauer (Univ. of Free State) for providing laboratory space and assisting with sample collection and R. Wilson (SRK-Turgis Tech.) for setup of the field laboratory and logistical assistance. We thank L. D. Labeyrie and two anonymous reviewers for their critical reviews and helpful comments.