The core was dried in vacuum for 24 h within a MBraun antechamber and embedded using EPO-TEK 301-2 resin (JP Kummer Semiconductor Technology GmbH, Augsburg, Germany). 10 cycles of soft vacuum and ambient pressure were applied. After a low temperature curing for 48 h, a longitudinal cut was performed, approximately 1.5 cm below the plane used for DNA sampling, using a diamond wire saw (MURG 394, WELL Diamond Wire Saws SA, Le Locle, Switzerland). XRF maps of sodium, silicon, phosphorus, potassium, sulfur, calcium, titanium, vanadium, chromium, manganese, iron and nickel (all K-edge) were obtained at the Crystal Growth Facility (EPFL), with an EDAX Orbis PCMicro EDXRF analyzer system (AMETEK Inc., Berwyn, Pennsylvania, United States). Boxplots of counts of iron, sulfur, and calcium fluorescence signals (representing atom-percent values for each element, at scanning spots where the sulfur signal was above 10%) were generated using OriginPro, version 2022b (OriginLab Corporation, Northampton, Massachusetts, United States). t-tests were achieved (Excel Analysis ToolPak, previously testing equality of variance with F-test) to assess statistical significance of the minerals shifts between biotic reaction and abiotic control (Supplementary Figure S5). More details are available in Supplementary Text S2.3.

Sand-bentonite samples of about 2.5 cm in length and 1.5 cm in width were obtained from the right half of each reactor in anoxic conditions. The samples were chosen based on visually observable features such as dark color and/or red/brownish halos (Supplementary Figure S2). These samples were placed into custom-built sample holders made of poly-(methyl methacrylate), dried in a high vacuum for 12 h, and stabilized in EPO-TEK 301-2 resin. A minimum of 6 soft-vacuum/ambient pressure cycles were applied to remove trapped gas bubbles. After 48 h of a low-temperature curing, the embedded sample surfaces were gradually exposed by rough removal of excess resin using a Dremel^® 3,000 tool. Fine wet grinding using silicon carbide grinding paper of three grit grades (Struers Inc., Cleveland, Ohio, United States) and ethanol, analytical reagent grade (Thermo Fisher Scientific Inc., Waltham, Massachusetts, United States) was performed manually until an even and shiny sample plane was achieved. The prepared samples were sealed within a nitrogen atmosphere by packing in three layers of Mylar^®. Synchrotron radiation μXRF maps and Fe K-edge XANES spectra were collected at beamline 2–3 of the Stanford Synchrotron Radiation Lightsource at the Stanford Linear Accelerator Laboratory (Menlo Park, California, United States). Fe-XANES data were collected in the energy range of 6,927 eV to 7,520 eV. μXRF imaging and S K-edge XANES was performed at beamline I-18 at the Diamond Light Source (Didcot, Oxfordshire, United Kingdom) using the Data Analysis WorkbeNch (DAWN) (Basham et al., 2015). The S K-edge XANES were collected in the energy range of 2,400 eV to 2,600 eV. Both Fe and S XANES spectra were analyzed using Linear Combination Fitting (Athena software) (Ravel and Newville, 2005).

The experimental set-up aimed to establish a link between the evolving microbiome and the availability of H₂, the main electron donor, and sulfate, the major electron acceptor. Close to the H₂ inlet, the availability of H₂ is high and bioavailable sulfate is expected to be rapidly depleted due to the activity of SRB. As H₂ was applied daily, the total quantity provided to each reactor is proportional to the duration of the experiment (Table 1). Water flow was controlled by the sand-bentonite back-pressure, resulting in variable flow across time and reactors (Supplementary Figure S3). Reactors 1 and 2 received between 0 and 4 μL/min of porewater (Supplementary Figure S3C); in total, reactor 1 received 10.1 mmoles of sulfate from the porewater while reactor 2 received 8.1 mmoles (Table 1). Reactor 3 received a varying flow of water, on average 4 μL/min until day 57, which sharply increased after day 57 with an average of 20 μL/min until the end of the experiment (Supplementary Figure S3A); in total, reactor 3 received 16.1 mmoles of sulfate from the porewater (Table 1). Reactor 4 received on average 12 μL/min, with a sharp decrease towards the end of the experiment down to 0 μL/min on day 70 (Supplementary Figure S3B); in total, reactor 4 received 22.9 mmoles of sulfate from the porewater (Table 1).

Sulfate is expected to serve as an electron acceptor. In the reactors, sulfate comes from the borehole porewater (approx. 15 mM, Figure 3 and Supplementary Table S3) and dissolution of gypsum from bentonite (Maanoja et al., 2020). Porewater was sampled and analyzed weekly during the H₂ injection phase, and the evolution of sulfate, sulfide, and ferrous iron is presented in Figure 3. Sulfate concentration followed a decreasing trend from the start of the experiment until the sacrifice of the reactors. Sulfate is reduced to sulfide, which abiotically precipitates with dissolved ferrous iron to form iron sulfide. The initial ferrous iron concentration in reactor 4 (80 μM) was lower than in the three other reactors (120 μM). Ferrous iron was depleted in the four reactors starting on day 10. During the entire experiment, any additional iron coming from the fresh porewater inflow was removed before reaching the outlet. Sulfide was not detected in the borehole porewater. In the reactors, the sulfide concentration remained one order of magnitude lower than the sulfate concentration. In reactors 3 and 4, the concentration of sulfide increased slightly to reach, respectively, 40 μM and 140 μM when sacrificed. In reactors 1 and 2, the sulfide concentration increase was greater, reaching, respectively, 360 μM and 240 μM. At the end of the experiment, the sulfide concentration has increased relative to the initial time point in all four reactors, but it showed temporal variations: a decrease from day 80 to 100 in reactor 1, and from day 80 until the end of the experiment in reactor 2.

Considering sulfate reduction as stoichiometrically linked to H₂ oxidation, we aimed at calculating the overall sulfate reduction rates in the four reactors during the whole experiment. To this end, we compared two calculations methods: a linear regression and a global mass balance. The sulfate rates calculated are summarized in Table 2. The linear regression model was applied to sulfate concentration curves (linear trend indicated in Figure 3A, detail in Supplementary Table S4). R-squared values obtained for reactors 1, 2, and 4 are above 0.95, which indicates that a zero-order model adequately captures the microbial activity in the reactors.

The above sulfate reduction rate was calculated assuming a batch system with the initial sulfate concentration corresponding to that after saturation with borehole porewater during 5 to 30 days (time 0 in Figure 3). This is clearly an assumption that results in an underestimated rate as it does not account for additional sulfate sources: (i) further dissolution of gypsum (dissolution during the initial equilibration released 1.71 to 2.96 mmol, while total gypsum available represents approximatively 8.2 mmol, Supplementary Table S1 and Supplementary Text S3); and (ii) inflow of fresh borehole water with a sulfate concentration of 15 mM. For reactors 1 and 2, the sulfate concentration increases during the 1^st week (Figure 3) evidencing the limitations of the batch reasoning.

V in tot : total water volume injected in the reactor (L) SO 4 2 − in : sulfate concentration in the inflow (mM)n_{gypsum dissolution}: quantity of sulfate present in gypsum (10 mmol, Supplementary Text S3)V_w and   SO 4 2 − out w total outflow (L) and average concentration (mM) in the outflow during week wV_voids: volume of voids in the reactor (L) SO 4 2 − out end : sulfate concentration in the outflow at the end of the experiment (mM)

A global sulfate mass balance is considered as a second approximation of the sulfate reduction rate (Eq. 2) and the results summarized in Table 2. The rates are 2 to 5 times lower than the linear reduction rate, and there is a higher variability amongst the reactors.

Due to microbial sulfate consumption, the reactor water is expected to become undersaturated with respect to gypsum, resulting in its dissolution. Furthermore, sulfide produced precipitates with ferrous iron to form iron–sulfur minerals (e.g., mackinawite FeS, greigite Fe₃S₄, or pyrite FeS₂) (Duverger et al., 2020). Patches of black color of the sand-bentonite matrix were observed after incubation with H₂ (Supplementary Figure S2).

We used XRF for elemental mapping of iron, sulfur, and calcium on one half of each reactor. Sulfur hotspots (i.e., with atomic percent above 10) expected to correspond to pre-existing gypsum (calcium sulfate) or to iron sulfide precipitation were selected to study the correlation between atomic percent of sulfur and calcium, or sulfur and iron. The mean iron to calcium ratio observed at sulfur spots for each reactor significantly shifted from approximatively 1.1 (abiotic control) to 1.5 (reactor 2, p < 0.05), up to 4.6 (reactor 1, p < 0.05) (Supplementary Figure S5C). Moreover, the decrease in calcium atomic percent at sulfur spots is statistically significant for the four biotic reactors compared to the abiotic control (p < 0.00001 for reactor 1, 2, and 4, p < 0.005 for reactor 3, Supplementary Figure S5A). In contrast, XRF data do not provide conclusive evidence of the change in iron concentration at sulfur hotspots (Supplementary Figure S5B). In reactors 1, 3, 4 and in the abiotic control (reactor not exposed to H₂), the average iron to sulfur ratio is 0.45, close to 0.5, corresponding to pyrite, which is initially present in Wyoming bentonite MX-80 (0.60% by weight, Supplementary Table S1). In reactor 2, the iron to sulfur ratio is lower than in the abiotic control (0.17 compared to 0.45, p < 0.05). The number of sulfur hotspots in this reactor was greater than in any other reactors, including the abiotic control (225 compared to values ranging from 18 (reactor 4) to 40 (reactor 3)). t-test results are detailed in Supplementary Table S5.

We used μXANES to analyze sulfur speciation. For the abiotic control (no H₂), the best fit for the sulfur data was obtained with gypsum (Supplementary Table S7). For 14 out of the 18 samples exposed to H₂, the best fits were obtained with pyrite as the main sulfur phase and with mackinawite as the second most abundant phase. In one sample, mackinawite was identified as the main phase by fitting. Poor quality of the standard data impeded obtaining very good values for statistical indicators of the goodness-of-fit, but the values confirm qualitatively adequate fits. Iron XANES spectra analysis confirmed the presence of pyrite and mackinawite in the samples, with higher fit quality than the sulfur XANES spectra (Supplementary Table S8). XANES mapping of iron evidences the presence of a pyrite precipitate of 1 mm² in reactor 2 (Supplementary Figure S6). Unfortunately, the standards did not allow the identification of all the iron phases, with for instance the presence of an unknown iron phase in the middle of a ferrihydrite spot in reactor 3 (Supplementary Figure S7, sample 20 in Supplementary Table S8).

The 35 sand-bentonite samples collected from the middle plane of each reactor and swabs from equipment parts of the reactors were analyzed to quantify biomass (16S rRNA gene copies number per gram of sand-bentonite) and phylogeny (16S rRNA gene amplicon sequencing). Here, we first discuss results obtained by averaging the relative abundance and biomass quantification results of the samples for each reactor (Figure 2 and Supplementary Tables S9–S11).

When comparing 16S rRNA gene copies number from the initial dry sand and dry bentonite, to the sand-bentonite samples post-incubation, it is clear that growth has taken place within all four reactors: the average 16S rRNA gene copies number per gram of material in reactors post incubation, dry bentonite, and dry sand are respectively: 8.2 × 10⁸, 8.2 × 10⁵, and 3.7 × 10⁵ 16S rRNA gene copies/gram substrate (Supplementary Figures S8, S9). The reactors were run for different durations and the flow rate was variable, resulting in different amounts of total sulfate available (Section 3.1). We expected reactor 4, that received the most water and sulfate (Table 1) to host the highest biomass, but on average, reactor 3 has significantly higher biomass growth than the other reactors (p < 0.01 for reactor 1, p < 0.05 for reactor 2, and p < 0.001 for reactor 4, Supplementary Figure S9 and Supplementary Table S12). The flow in reactor 4 decreased sharply at the end of the experiment [from 20 μL/min at day 60 to 0 μL/min at day 70, with on average 3 μL/min during the last 10 days (Supplementary Figure S3)], meanwhile in reactor 3 it increased up to 32 μL/min, with on average 20 μL/min during the last 10 days (Supplementary Figure S3). Altogether, this suggests that biomass growth measured at the end of the experiment depends on the flow rate (and thus sulfate availability) in the reactors at the end of the experiment rather than during the whole experimental run.

The average relative abundance of the 35 sand-bentonite samples varies across reactors but all include sulfate reducers, methanogens, both known hydrogenotrophs, as well as fermenters (Figure 4 and Supplementary Tables S9, S10).

Among sulfate reducers, Desulfocurvibacter genus has a similar average relative abundance in the four reactors, ranging from 3.8% (reactor 3) to 7.3% (reactor 2). Sulfate reducers from the Proteobacteria phylum are well represented with four ASVs identified to be part of the Desulfobacteraceae or Desulforomonadaceae families (Figure 4). Desulfuromonadaceae (ASV7) has the highest overall relative abundance (6.8%, Supplementary Tables S9, S10). Other sulfate reducers are Firmicutes, with Desulfotomaculum spp. identified in reactors 2 and 3 and Desulfallas spp. identified in all reactors. Three ASVs from the Deltaproteobacteria class could not be assigned down to the genus level, however, many families of this class are sulfate reducers. Data analysis using the RDPnaive algorithm (which is less restrictive) predicted the attribution of ASV6 to the order Desulfobacterales and of ASV10 from the family Peptococcaceae to genus Desulfallas (data not shown). If ASVs 6 and 10 are indeed sulfate reducers, the relative abundance of sulfate reducers is high in all reactors: 38.7% (reactor 2), 42.2% (reactor 4), 46.9% (reactor 1), and 50.2% (reactor 3) (Supplementary Tables S9, S10). Taking ASVs individually, we observe that only in reactor 3 does an ASV corresponding to a sulfate reducer have the highest relative abundance on average (ASV3 from Desulfobacteraceae family with 16.5%). In the other reactors, the most abundant ASV is from the genus Pseudomonas or Sunxiuqinia.

Methanosarcina genus is the only representative of methanogens. Averaging the 35 sand-bentonite samples in each reactor, the genus was detected as the 4^th, the 7^th, and the 13^th most abundant ASV in reactors 1, 2, and 3, respectively. This genus was also detected in reactor 4, but the average number of reads was lower than the number of reads in the Kitome, and thus the presence of this genus in the sand-bentonite matrix could not be confirmed. Activity of this methanogen was confirmed via the detection of methane in all four reactors during the experiment (Supplementary Figure S4). Indeed, detection of methane in the gas phase of reactor 4 supports the presence of Methanosarcina spp. in this reactor, despite its low abundance based on reads count.

Fermenters do not participate in H₂ oxidation but foster availability of low molecular weight organic carbon substrates for the growth of H₂ oxidizers. Attribution of fermentative metabolism based on genus is an approximate endeavor as many bacteria capable of fermentation are also able to catalyze other reactions. For instance, members of the genus Pseudomonas can ferment many substrates but also carry out denitrification (Carlson and Ingraham, 1983) or iron reduction (Vasil, 2007). Therefore, while we classify Pseudomonas and Sunxiuqinia genera as fermenters here, they could also be involved in other metabolisms. These two genera are part of the top five most abundant ASVs, with relative abundance ranging from 10.3 to 17.6%, and 6.4 to 22.4%, respectively. Other putative fermenters are from Symbiobacterium genus., with a homogeneous relative abundance across the reactors, and from Gracilibacter genus.

Differences between reactor microbiomes were evidenced using PCA (Figure 5). Reactors 1, 2, and 3 present similarities with the overlapping of the clouds formed by the samples plotted on the two first principal component axes, while reactor 4 plots separately.

The PCA of the microbial community associated with equipment evidences similarity amongst the four reactors (Supplementary Figure S10). The top ASVs are from the genera Pseudomonas, Desulfocurvibacter, Desulfobacteraceae (ASV3), Methanosarcina, Gracillibacter, Desulfallas, Sunxiuqinia (Supplementary Table S13). Methanosarcina genus was preferentially located at the water inlet of reactors 1 and 2 (data not shown). Desulfocurvibacter genus was observed at the water inlet, outlet (where it has the higher relative abundance), and on the H₂ tube, suggesting the dependence of this genus on H₂ (data not shown).

The spatial distribution of biomass and ASVs in each reactor also contains useful information that can be mined as a result of the spatially-resolved sampling. Biomass concentration was compared across vertical columns for each reactor and evidenced that growth had taken place mainly in the middle vertical column in each reactor, where the highest availability of H₂ is expected (Supplementary Figure S8). Indeed, for all reactors, at least one of the off-center columns showed significantly (p < 0.05) lower 16S rRNA gene copies number per gram of substrate than the central column (Supplementary Table S13). Notably, in reactor 3, the number of 16S rRNA copies is lower on either side of the central column (p < 0.05). For reactor 2, and reactor 4, three columns out of four have significantly lower number of 16S rRNA copies than the central one (p < 0.05). For reactor 1, only the outer left column presents significantly lower number of copies (p < 0.05). In this reactor, growth appears to have taken place homogeneously. t-test results for all reactors are detailed in Supplementary Table S13.

Maps of individual taxa within the microbiome (Figure 6) show that sulfate reducers, particularly member of Desulfocurvibacter genus, grow around the H₂ outlet. In reactor 3, Desulfocurvibacter spp. were outcompeted by other SRBs, and the middle column of the reactor is occupied by a member of Desulfobacteraceae family (which is the most abundant ASV in this reactor). In reactors 1, and 2, the most abundant sulfate reducer (from the Desulfuromonadaceae family) occupies mainly the outer left and bottom of the reactor (reactor 1) or is homogeneously distributed (reactor 2).

In reactor 2, Methanosarcina spp. grew near the H₂ outlet, with a similar growth pattern as that encountered for Desulfocurvibacter spp. However, in this reactor the highest relative abundance of Methanosarcina spp. is observed at the top right corner and corresponds to the highest number of 16S rRNA gene copies in this reactor. In reactor 1, members of Methanosarcina genus grew where there is no SRB growth; in reactor 3, they grew at the top left corner, overlapping with members of Desulfocurvibacter genus.

The sulfate reduction rate was calculated using two approaches: a linear regression, and a global mass balance. The rates obtained by linear regression were two to five time higher (on average, 1.752 ± 0.191 μ mol day   cm 3 ) than the global mass balance calculation (on average, 0.44 ± 0.14 μ mol day cm 3 ). The rates obtained by linear regression are similar for the four reactors (RSD of 11%, highest rate in reactors 2 and 4, and lowest in reactors 1 and 3, but no statistically significant differences), despite different running conditions (duration, water flow, parallel/serial setup). The global mass balance captures differences (RSD of 24%), especially related to water flow (highest rate in reactors in the parallel setup, lowest rate in reactors in the serial setup). The linear regression considered a batch system with no additional input of sulfate and gave an average value for the entire duration of the experiment. On the contrary, the global mass balance takes into consideration sulfate inputs (dissolution, fresh borehole water) and loss of sulfate in the outflow (Eq. 2). The mass balance is thus more complete and is considered to be a better model. The difference between the two rates highlights the limited quantitative information provided by the evolution of the concentration at the outflow, given the existence of preferential flow paths, and the heterogeneous sulfate dissolution and consumption in the porous matrix.

The range of the rates calculated is five to ten-fold higher than the sulfate reduction rate of 0.14 to 0.20 μ mol day   cm 3   reported by Bagnoud et al. (2016c) for a similar Opalinus Clay rock derived microbial community but within a porewater-filled borehole. This suggests that the porous material (representing backfill in the future repository) provides more favorable growth and/or metabolic conditions. Considering a 1:4 stochiometric ratio, the H₂ oxidation rate was estimated to be 2–7 μ mol day   cm 3 using the average of the results from the mass balance and the linear regression, respectively. Literature-reported rates of H₂ oxidation by sulfate reducers cover a wide range: 1.2 to 312 μ mol day   cm 3 as reported by Thaysen et al. (2020). Dohrmann and Krüger (2023) measured H₂ oxidation in the formation fluid in a natural gas field and found the pressure-dependent rate to vary between 0.123–0.325 μ mol day   cm 3 at ambient pressure and to increase up to 4 μ mol day   cm 3 at 100 bar.

It is most useful for a DGR to compute the H₂ consumption rate as a function of the volume of backfill material. The rate is estimated to be 0.9–3.2 mol day   m backfill 3 (calculation detailed in Supplementary material, Supplementary Text S4). Considering a production of 25 million m³ (900 × 10⁶ mol) of H₂ during a safety period of 1,000 years for the storage of radioactive waste, and constant H₂ oxidation rate, less than 1% of the backfill would be required for the microbial consumption of all the H₂ produced. There are clearly many approximations associated with these calculations: for instance, partial saturation of the backfill and limited water availability, sulfate depletion, or the dependence of microbial activity on temperature and pressure (Dohrmann and Krüger, 2023). In reality, one can expect variable H₂ production and consumption rates as a function of time.

In the work by Bagnoud et al. (2016c), the stoichiometry of sulfate reduction was observed to be close to 8.5 moles of H₂ per mole of sulfate reduced, which was attributed to other reduction reactions coupled with H₂ oxidation (i.e., reduction of intermediate valent state sulfur species, clay iron reduction) and H₂ losses. In our study, sequencing of the 16S rRNA gene did not provide conclusive information on the presence of homoacetogens and iron reducers. Nonetheless, given that ASVs from the class Clostridia, and the family Desulfuromonadaceae were detected, and are taxa that harbor homoacetogens and iron reducers, respectively, we cannot exclude a role for those metabolisms in H₂ consumption. Furthermore, the presence of hydrogenotrophic methanogens highlights the limitations of using sulfate reduction as a proxy for H₂ oxidation rate.

Mapping sulfate-reducing taxa indicates that, while some genera (e.g., Desulfocurvibacter) have grown close to the H₂ tube outlet, and, thus, their growth may be limited by H₂ availability, other ASVs capable of sulfate reduction were distributed homogeneously within all reactors. The homogeneous distribution suggests the use of other electron donors (e.g., low-molecular weight organic acids). Bentonite can release organic carbon (Maanoja et al., 2020) which could serve as an electron donor. So can the degradation of autotrophic biomass. If the contribution of organic carbon to sulfate reduction is substantial, the sulfate reduction rate could be an overestimation of the H₂ consumption rate. However, lactate and acetate were never detected in the outflow and total dissolved organic carbon was not measured, preventing meaningful assessment of the contribution of organic carbon to this metabolism. Rates measured by Bagnoud et al. (2016c) presented evidence that sulfate reduction did not account for all H₂ consumption as the amount of H₂ consumed was 8.5-fold greater than that calculated by the stoichiometric ratio with sulfate. However, in the same experiment, consumption of organic carbon supports the occurrence of heterotrophic growth, possibly of the sulfate reducers (Bagnoud, 2015). For the present study, this suggests that not all sulfate reduction is attributable to H₂ oxidation, but nonetheless we can expect that more H₂ is consumed than sulfate reduced, due to other metabolisms.

Overall, the reported rate of sulfate reduction can be viewed as a conservative range due to three important restrictions. First, the sulfate-reducing biofilm within the reactor was not fully developed as evidenced by the remaining sulfate within the outflow water after more than 100 days of daily H₂ injections (Figure 3). Second, the H₂ oxidation rate calculated via sulfate reduction observed at the outflow does not account for sulfate consumption in the unconnected porosity. Third, the H₂-oxidizing community within the sand-bentonite matrix is not exclusively composed of SRB, but also of methanogens. Thus, the actual H₂ oxidation rate is likely to be higher.

To decipher whether the most abundant ASVs originated from the borehole water or the sand or bentonite solids, DNA from the dry sand and clay, and from borehole porewater, was extracted and sequenced (Supplementary Tables S2, S14). The microbiomes from the borehole porewater and the dry materials are clearly different from that of the reactors, except for Pseudomonas genus, that is present everywhere. This ASV is one of the most abundant in the borehole water and was detected in the coarse sand and in two out of three dry bentonite samples as well. None of the other 25 most abundant ASVs from reactors, dry sand, dry bentonite, or borehole water, overlap. The microbiome obtained from dry clay materials may not represent the full diversity of the community due to the inability to detect low-abundance species. This is evidenced when comparing the microbial diversity of dry clay and that of the same clay placed in water with an electron donor/acceptor (Vachon et al., 2021). Therefore, it is not a trivial exercise to pinpoint the origin of the microbial community identified in the 4 reactors. However, similarity in the microbial community amongst three of the four reactors (Figure 5), even though their preparation occurred up to a month apart, suggests that the origin of the microbiome is likely to be the dry materials or the borehole porewater, and not the equipment or preparation and handling conditions. Members of Gracillibacter, and Peptococacceae genera could originate from the borehole water (number of reads > 500). ASVs from Paenibacillaceae, which was among the most abundant families in the borehole water, did not grow in the reactors. ASVs from the order Clostridiales represent some of the most abundant genera in the borehole porewater, but do not correspond to the ASVs from the class Clostridia detected in the reactors. Finally, surprisingly, few reads (88 reads on average, with 45 reads in the Kitome considered to be contamination) were detected in the reactor for the most abundant borehole porewater sulfate reducer, Desulfosporosinus genus, suggesting that it did not thrive in the sand-bentonite porous medium amended with H₂. Altogether, comparing the microbiomes of the dry materials, the borehole porewater, and the reactors did not allow to conclusively disentangle the origin of the most abundant genera in the reactors.

Sampling the internal surfaces of the equipment (water inlet/outlet, gas trap, gas inlet, steel and Plexiglas surfaces) provided the opportunity (i) to decipher the potential contribution of the microbiome associated with the equipment to sulfate consumption and methane production, (ii) to identify whether conditions in the sand-bentonite support growth of a matrix-specific community, distinct from the one on the equipment surfaces. PCA highlighted a marked difference of the community in the sand-bentonite in reactor 4 compared to that in the three other reactors, which was not reflected in the sulfate reduction rate. Contrary to the sand-bentonite matrix, samples obtained from the equipment were similar in the four reactors (overlapping of clouds on equipment PCA, Supplementary Figure S11) and Desulfocurvibacter, Pseudomonas, and Methanosarcina genera were detected with the highest relative abundance (Supplementary Table S13). This observation suggests that the equipment community might contribute non-negligibly to sulfate reduction rates, and would explain why the rates from the linear-regression are similar across reactors despite differences in the sand-bentonite matrix microbial community. Additionally, it suggests the development of distinct communities within the sand-bentonite porous medium in each reactor that differ from that on the equipment. Thus, the sand-bentonite microbial community in each reactor may have been shaped by the specific flow rate, that impacts the availability of solutes as well as the ratio of H₂ to sulfate (Table 1).

Monitoring the outflow sulfate concentration associated with the linear-regression analysis has the advantage of giving information on temporal variations in sulfate consumption, and thus, the activity of sulfate reducers. The linear-regression fitted the sulfate consumption data for reactors 1, 2, and 4 adequately (R-squared superior to 0.95) indicating a steady-state activity during the experiment, but less so for reactor 3 (R-squared of 0.79). The low R-squared value for the reactor 3 linear model is explained by an increase in sulfate concentration after 57 days, which is likely due to a steep increase in the water inflow rate (Supplementary Figure S3). Sulfate may not have been consumed by the microbial community rapidly enough to overcome this abrupt increase. Another possibility is the creation of preferential water flow paths not yet colonized by sulfate-reducing communities. If only the low water flow regime is considered (first 50 days) for reactor 3, a linear sulfate reduction rate of 2.569 μ mol day   cm 3   is observed, with a R-squared of 0.96. This is the highest rate measured in the four reactors, and matches sequencing results which indicated that, on average, SRB were present with a higher relative abundance in this reactor (Results 3.4). However, the underlying reasons for greater growth and activity of SRB in this specific reactor are not understood.

Methane was detected in the gas phase of all reactors. Methanosarcina spp. are versatile methanogens capable of all three methanogenic pathways (Madigan et al., 2018), thus no direct conclusion of its role in H₂ consumption can be extracted from this system. However, its growth and the detection of methane strongly support methanogenesis as an important metabolism stimulated by H₂ (directly or indirectly). Methanogenesis consumes 5 moles of gas, for 1 mole of CH₄ produced, thus participating to pressure decrease despite gas production (4H₂ + CO₂ ➔ CH₄ + 2H₂O). The average relative abundance of Methanosarcina spp. was higher in reactors 1 and 2, which received the least water, and thus the least sulfate. Detection of this genus in reactor 3, which received variable water flow rate (very low for the first 2/3 of the experiment and high in the last third), suggests growth during the first period when the reactor received little sulfate. The number of reads associated to Methanosarcina spp. was below that of the kitome in all samples from reactor 4, which received by far the highest total porewater volume (Table 1). Not taking into consideration gypsum dissolution as a source of sulfate, total H₂ to sulfate ratio in reactors 1 and 2 is above the expected stoichiometry (respectively 6.2:1 and 4.9:1 vs. 4:1, Table 1), meaning that the sulfate provided could not have oxidized all H₂, pointing to CO₂ as a potential additional electron acceptor. Indeed, there is circumstantial evidence for hydrogenotrophic metabolism by members of Methanosarcina genus by the fact that its growth is localized at the point at which H₂ is delivered into reactor 2. In addition, the relative abundance of Desulfocurvibacter genus is lower than in reactors 1 and 4 at the same spot. Therefore, Methanosarcina genus is hypothesized to be outcompeted by SRB when sulfate is not limited but grow actively in case of sulfate depletion.

Representatives of Archaea have been detected in Opalinus Clay rock at very low abundances (Stroes-Gascoyne et al., 2011; Mitzscherling et al., 2023), and there is no direct evidence of their activity to date (Poulain, 2006; Stroes-Gascoyne et al., 2007; Vinsot et al., 2017). They are also found at extremely low abundance in borehole water (only the genus Methanolobus was detected) (Vinsot et al., 2014; Bagnoud et al., 2016a). This is likely due to the absence of an abundant electron donor and the abundance of sulfate, both of which favor SRB over methanogens. The higher substrate affinity of SRB to H₂ as compared to that of methanogens and homoacetogens accounts for SRB’s competitive advantage (Kristjansson et al., 1982; Muyzer and Stams, 2008). When methanogens and SRB are present, the latter can maintain H₂ at a low steady-state concentration, at which methanogenesis is endergonic (Hoehler et al., 2001). In the subsurface, methanogen-, or methanogen/homoacetogen-dominated communities were detected in water with sulfate concentration around 1 mM (i.e., 15-fold lower than in Opalinus Clay porewater) (Kotelnikova and Pedersen, 1997, 1998; Chapelle et al., 2002). In the sand/bentonite backfill, methanogens could be indigenous to bentonite, as evidenced by the detection of methane during bentonite incubation (Maanoja et al., 2020). Unfortunately, phylogeny was not investigated in that work.

This result highlights that metabolisms other than sulfate reduction should be considered as microbial H₂ sinks in a DGR backfill.