The inocula used during the present study was obtained by sampling formation waters in anaerobic conditions from four NG reservoirs which have been previously characterized and named as R1, R2, R3 and R4, as previously described (Bassani et al., 2023). Upon receipt, water samples were stored overnight at 4°C under N₂ atmosphere to ensure anaerobic conditions. Inoculation procedures were performed in a Whitley A85 anaerobic workstation (Don Whitley Scientific, West Yorkshire, UK) under a steady flow of H₂, CO₂ and N₂.

Batch coltures were performed in serum flasks of a total volume of 158 mL. Flasks were initially closed with rubber septa, flushed with N₂ to ensure anaerobic conditions and then sterilized in an autoclave (121°C x 80 min). Initially, batch assays were conducted without nutrient enrichment by adding 25 mL of sterilized formation waters to serum flasks seeded with 25 mL formation waters from each reservoir for a total working volume of 50 mL by providing either N₂, CH₄ 100% or a CH₄:H₂ blend at 50:50 ratio (%/%). Nonetheless, after 25 days no signs of microbial activity were detected (Supplementary Figure S1) and the provision of nutrients was performed in order to stimulate microbial growth. Similarly, in batch assays with nutrient enrichment 25 mL of nutrient media (DSMZ 141) were added to the serum flasks seeded with 25 mL of formation waters from the four reservoirs in order to reach a total working volume of 50 mL. For each reservoir, four conditions have been assessed: (i) one positive control (Glu), i.e., formation water plus DSMZ 141 media sparged with N₂ and supplemented with 10 mM glucose solution, providing easily accessible carbon source to the microbial consortium and monitoring microbial growth and activity in “ideal” heterotrophic conditions; (ii) one sample with formation water plus DSMZ 141 media supplemented with 100% CH₄ (CH₄), mimicking the presence of natural gas within the reservoir; (iii) one sample with formation water plus DSMZ 141 media supplemented with CH₄/H₂ mixture at a 90:10 ratio (%/%; H₂ 10%); (iv) one sample with formation water plus DSMZ 141 media supplemented with CH₄/H₂ mixture at a 50:50 ratio (%/%; H₂ 50%).

For each condition, 3 technical replicates were prepared and supplemented with 2 bar of gas (i.e., CH₄, CH₄/H₂ or N₂ for samples of the positive control) delivered in the headspace through a sterile gas-tight syringe and a three-way valve. Batches were incubated at 50°C and 200 RPM agitation during a 23-day experiment. Headspace pressure, liquid optical density (OD), and pH were measured using, respectively, a digital manometer, an OD_600nm spectrophotometer and a pH meter. Monitoring of the above-mentioned parameters was carried out at 3/4 day intervals.

To evaluate the concentration of short-chain VFA (scVFA) in the formation waters collected from the four reservoirs and their content after batch cultures, an HPLC methodology developed for the detection of acetic, propionic, n-butyric, iso-butyric, n-valeric, and iso-valeric acid adapted from Ricci et al. (2021) was applied in the present study. Quantification of organic acids was performed by HPLC using a Thermo-Fischer Dionex Ultimate 3,000 system (Thermo-Fischer, USA) coupled with a Thermo-Fischer Dionex Ultimate 3,000 Variable wavelength detector operating at 210 nm. For organic acids, the column (Metab AAC, ISERA GmbH, Düren, Germany) was eluted isocratically with 9 mM H₂SO₄ at a flow rate of 0.6 mL min-1 and an oven temperature of 40°C. Calibration was performed by generating individual stock solutions (100 mM) of the above-mentioned scVFA species by solubilizing their relative weight/volume in MilliQ water. Stock solutions were then used to prepare calibration standards containing all the acid species for concentrations between 10 and 1 mM. Data collected were screened against the VFA levels contained in the media and in the inactivated formation waters used as blanks allowing to differentiate between the degradation/generation of VFAs naturally occurring in the culture media from that caused by microbial activity.

Headspace gas concentrations were measured using gas chromatography (MicroGC Fusion, INFICON, Bad Ragaz, Switzerland) calibrated for the detection of H₂, O₂, N₂, CH₄, CO₂ and H₂S (%). 3 mL headspace gases were sampled at cultures inoculation (Day 0) and at the end of the cultivation cycle (Day 23) using a gas-tight luer lock syringe. Data displayed in this study are obtained from the average of single measurements performed on physiological triplicates for each condition tested and have a max RSD <0.5%.

To evaluate abundances of HM, SRB and AB during batch cultivation of formation waters collected from R1, R2, R3 and R4 the qPCR protocol described by Nyyssönen et al. (2012) targeting genomic sequences of both mcrA and dsrB and relying on the use of primer couples ME1/ME3R and DSRp2060F/DSR4R was adopted. Similarly to HM and SRB, AB abundance was assessed by qPCR targeted the fhs gene encoding for formyl-tetrahydrofolate ligase, using the FHS 2 and FTHFSr primers previously described by Xu et al. (2009). The protocol consists of a Sybr-green qPCR assay using the Quantitect Sybr Green Kit (Qiagen, Hilden, Germany), single reactions volume was 25 μL, with all primer couples added at a final concentration of 0.3 μM and with 2.5 μL genomic DNA template. The assays were carried out with a Qiagen Rotor-Q thermal cycler (Qiagen, Hilden, Germany), fluorescence was detected within the range of green (510 nm) in a final 25 μL reaction volume. Amplification conditions were as follows:

Initial denaturation: 15 min 95°C;

Denaturation: 40 s 95°C;

Annealing: 40 s 55°C;

Elongation: 40 s 72°C;

Final elongation: 6 min 72°C.

qPCR specificity was tested through melting curve analysis consisting of an initial denaturation for 10 s at 95°C followed by re-annealing at 55 degrees with a gradual T increase up to 95°C while fluorescence was continuously detected. The lowest n°copies/ml detectable by the used methodology were 1.33 × 10², 1.85 × 10² and 1.76 × 10² copies/ml for mcrA, dsrB and fhs, respectively. Samples with values below this threshold were not reported.

Results obtained were statistically analyzed by means of a one-way ANOVA test. The results reported in the present work are those with p < 0.03. Results of statistical analysis on significant samples are described in the Supplementary Figures S2–S4.

A 0-D model was developed using the COMSOL® Multiphysics platform, a finite element modeling program designed to solve a vast array of partial differential equations (PDEs), which has emerged as an innovative and succesfull approach to groundwater modeling (Li et al., 2009; Jin et al., 2011; Jin and Kirk, 2016; Liu and Liu, 2017; Sainz-Garcia et al., 2017).

The developed model allows determine the growth kinetics and gas/VFA production and consumption, accounting for the concentration in moles of target chemicals and their presence in both gas and liquid phases.

Governing equations for the four main microbial processes, represented by reaction equations outlined in Supplementary Table S1, are implemented in the model simulations (Hoehler et al., 1998; Conrad, 1999; Jin and Bethke, 2005; Feldmann et al., 2016; Jin and Kirk, 2016; Berta et al., 2018; Delattre et al., 2020; Strobel et al., 2023; Tremosa et al., 2023).

The model takes into account/includes equilibrium reactions for gas-water interactions, focusing on potential increases in dissolved carbonate and sulfate in reservoir water for bacteria and archaea reactions. Equilibrium calculations are based on the mass action law, considering all species used in this study and their corresponding equilibrium constants. Supplementary Table S1 details the equilibrium phases, mass action equations and constants (Hemme and van Berk, 2018; Louca et al., 2019).

Monod model is a standard in modeling microbial growth under substrate limited conditions. Due to the dependence of microbial growth and decay on substrate and electron acceptor availability the “double Monod model” has been proposed to address this (Megee et al., 1972; Strobel et al., 2023), as outlined in the Supplementary Table S1.

The model accounts for changes in key components across both gas and liquid phases. The concentration of components within the liquid phase, or solubility, is generally represented using Henry’s Law. A crucial assumption here is the equilibrium between concentrations in both phases.

The mathematical model is based on the two-film theory and the “film” between the phase is the resistance against the mass transfer. As no reaction takes place in the gas phase, the change of the substrates hydrogen and carbon dioxide over time is dominated by the mass transfer (Li, 2017). With a mass-transfer model emulating molecular flow between water and gas phases, by combining mass transfer with microbial reactions, we can derive equations for the production and consumption of H₂, H₂S, CO₂, and CH₄ within the liquid phase. Mass transport is described using a generalized approach (Hagemann et al., 2016) on the COMSOL^® multi-physics platform, as presented in the Supplementary Table S1. To effectively address mass balances, fluid dynamics, and kinetic equations the following boundary conditions and model specific assumptions were made:

Boundary Conditions:

Model Assumptions:

The validation of the model involves verifying its operational efficacy through comparison between simulated and experimental data, under identical operating conditions, utilizing the COMSOL® platform. A model-fit procedure is adopted, wherein the values of unknown parameters, namely the biochemical kinetics constants, are adjusted to minimize the discrepancy between model predictions and available experimental data. A specific comparative analysis between the simulated and experimental data is conducted via statistical error analysis. This method implements a constrained search procedure, thus necessitating the specification of lower and upper parameter limits. These limits were selected based on the range of values most commonly observed in relevant literature. As delineated in the preceding section, comprehensive knowledge on mass transfer and bacterial growth kinetics has been wholly integrated into formulating the modeling equations, with the resultant model being implemented within the COMSOL® computing platform. To derive suitable estimates for the parameters vital for mathematical representation, reliance is placed on either experimentally determined values or an inference procedure fitting model simulations to observed data. In particular, comparisons between simulated and experimental data are typically executed using a statistical error analysis based on the Levenberg–Marquardt method, complemented with the second least-squares regression technique to minimize the residual (Vasile et al., 2021). The exploration of parameter space ceases when the model simulation optimally aligns with the experimental data.

In the initial phase of the modeling work, the model will be utilized to verify its operational efficacy by scrutinizing the concordance between the simulation estimates and the experimental measurements for growth rate, VFA, methane, carbon dioxide, hydrogen sulfide and hydrogen evolution derived from lab tests. The conditions listed in Table 2 were simulated along all the modeling campaign. The model operation were scrutinized by comparing them with the experimental data reported in paragraph 3.4 and in the Supplementary Information. The parameters pertaining to biochemical kinetics, gas–liquid equilibria, and bacterial growth were adjusted using the experimental data. Furthermore, a comparison was made between the results obtained from the simulations and the experimental data in terms of the number of HM-AB-SRB cells, headspace gas pressure and composition, and VFA produced.

During the experiments, the cultures O.D. and pressure variations were monitored to evaluate the microbial growth and activation. When cultured in the absence of nutrients, the formation waters of the four reservoirs did not display any sign of microbial growth. O.D. and pressure measurements did not reveal significant changes when compared to negative controls thus underlying limiting conditions for microbial growth (Supplementary Figure S1).

Data related to O.D. and pressure values for the batch tests performed in the presence of nutrients are reported in Figure 2. The batches prepared from the R1 formation water supplemented with gas mixtures reached O.D. 0.5 within the first 9 days, whilst in glucose-fed batches further increase was detected until day 20. Batches of formation water from R3 glucose-supplemented samples did not exhibit increases in the O.D. values which were reported below 0.2. Nonetheless, R3 supplemented with gas showed remarkable O.D. increase rates within the first 13 days with values ranging from 0.6 to 1 in batch cultures with headspace filled with 50% H₂ and CH₄, respectively, (Figure 2). Conversely, from what we observed in cultures seeded with R1 and R3 formation waters, the inoculant collected from reservoirs R2 and R4 did not show significant variations of the O.D. during the experiment, with values remaining frequently <0.3 (Figure 2).

Along with the O.D. variations, the headspace partial pressure was monitored during the whole experiment and the data are illustrated in Figure 2. A common trend was observed in most of the tested conditions, with the average headspace pressure reported to decrease ≈ by 0.1–0.3 bars by the end of the experiment. Nonetheless, with pressure losses observed in positive control samples (ΔP ≤0.3 bar) falling within the same range of those reported for most of the batch cultures, the observed variations are attributed to the numerous samplings performed during the batch experiment. Only the batch cultures from R1 and R3 with 50% H₂ in the headspace displayed ΔP higher than the positive control ones, with ΔP_R1H250% = 0.13 bar and ΔP_R3H250% = 0.1 bar, respectively.

Given the importance that VFAs have as metabolic intermediates in microbial metabolism, their quantification and monitoring could elucidate interactions occurring within the mixed microbial communities of the reservoirs. The presence and quantification of five species of VFAs, namely acetic, propionic, n-butyric/iso-butyric and n-valeric/iso-valeric acid, were monitored using HPLC during batch experiment, results of the analysis are resumed in Figure 3.

At the time of the experiment, it was observed that in samples from both R1 and R3, variations in the concentrations of VFAs occurred in all conditions tested. Acetate and propionate were the most represented VFAs species although within different concentration ranges (Figure 3). In samples from R1 cultures acetate was reported between 2.13 and 11.81 mM by day 20 with positive controls fed with glucose displaying the highest concentrations (11.81 mM). In samples from R3 acetate was reported between 1.72 and 5.21 mM. For what concerns propionate, concentrations between 0.68–1.21 mM and 0.01–1.21 mM were observed in R1 and R3 samples by day 20, respectively. Along with acetate and propionate, other scVFAs species were detected although in lower and variable amounts. In particular, during the cultivation of formation waters from the four reservoirs traces of i-butyric, butyric and i-valeric acid were detected (Figure 3).

While variations in scVFAs levels occurred in R1 and R3 cultures, data collected from cultures inoculated with R2 and R4 formation liquids evidenced consistent differences with the other reservoirs. Samples collected from cultures supplied with glucose and CH₄ displayed no sensible variations in levels of considered scVFAs species. R2 samples in which the serum headspace was filled with mixtures of 50% H₂ and 10% H₂ traces of propionate, butyrate and i-valerate were reported whilst in R4 samples provided with 50% H₂ a peak 3.06 mM peak of acetate was detected on day 23 (Figure 3).

Regarding pH, batch cultures obtained from R1 and R3 stabilized their pH between 7.5 and 8 by day 23, whilst cultures obtained from R2 and R4 reached pH between 6.5 and 7. pH values measured during batch cultivation were slightly higher or within the ranges previously reported (Bassani et al., 2023). Nonetheless increase in average pH is likely to be attributed to interactions with the provided media, which could have partially altered the pH. This is also sustained by the fact that in batch culture performed without external nutrient provision pH values remained stable along the experiment (Data not shown) and close to those previously reported by chemical characterization Bassani et al. (2023).

To better define the growth of the microbial clusters of interest (i.e., HM, SRB and AB) during batch cultivation of formation waters of R1, 2, 3 and 4, qPCR analysis was performed on samples collected at the beginning of the test on day 0 (Table 3), in the mid-phase at day 13 and at the end at day 23.

qPCR results obtained from the different conditions tested are reported in Figure 4, along with the concentrations of headspace gases measured at the beginning (day 0) and at the end of the experiment (day 23). qPCR analysis on day 0 samples revealed the initial composition of the inoculant obtained from R1, 2, 3 and 4. Fhs gene was the most abundant among those used as targets with values of 1.09 × 10⁴, 1.79 × 10² and 1.04 × 10⁴ copies/ml in R1, R2 and R3, respectively. mcrA gene was detected in R1 and R3 at 2.42 × 10³ and 5.72 × 10² copies/ml respectively, whilst in R2 samples mcrA copies/ml were reported below the threshold of detection. dsrB copies/ml were the least abundant in all samples and were frequently reported below the threshold of detection of the devised technology. Among the tested reservoir samples from R4 were the only ones that gave no significant response to the assessed gene markers with copies/ml of mcrA, dsrB and fhs being consistently below their relative threshold of detection.

qPCR analysis also revealed that copies/ml of fhs gene targeting AB were the most abundant between analyzed samples, with their number increasing significantly (p < 0.0001; Supplementary Figures S2, S3) along the incubation reaching by day 23 values between 6.01 × 10⁶–8.7 × 10⁸ copies/ml and 1.53 × 10⁵–1.85 × 10⁸ copies/ml in samples from R1 and R3, respectively. Similarly, fhs gene marker was the most abundant in R2 samples cultivated in H2-supplemented cultures. Nonetheless, copies/ml of the target gene were reported at lower amounts (8.4 × 10³–3.4 × 10⁶ copies/ml) than both R1 and R3. mcrA copies/ml, used for targeting the presence of HM, were measured at several orders of magnitude below (1.33 × 10²–1.37 × 10⁷) those of fhs in several samples from R1, 2 and 3. Although most of the variations observed for mcrA copies/ml were not statistically relevant, significant (p < 0.0001) increases in target gene copies abundance was reported in samples from R1 and R3 cultivated under 50% H₂ headspace with copies/ml on day 23 counting at 1.03 × 10⁶ and 1.37 × 10⁶, respectively (Figure 4). For what concerns SRB copies/ml of the target gene used in the present study, dsrB, were often reported below or slightly above the threshold level of the devised methodology (1.85 × 10² copies/ml) in cultures from R2, R3 and R4 (Figure 4). An increase in dsrB copies/ml value was observed in R1 samples with control cultures supplemented with glucose stabilizing by day 23 at 9.03 × 10⁵ copies/ml, whilst other R1 cultures reported values between 1.24–4.3 × 10⁴ copies/ml. Further relevant increases in target gene copy numbers were only observed in R2 and R3 samples cultured under H₂ 10 and 50% headspace with dsrB copies/ml stabilizing between 3.75 × 10² and 1.4 × 10⁴ copies/ml. Data regarding headspace gas concentrations in the different conditions tested for the four reservoir waters are plotted along with qPCR data in Figure 4. By the end of the experiment on day 23, all cultures from both R1 and R3 displayed an increase in CO₂ headspace concentrations with values ranging between 1.5–5.5% and 2.5–4.5%, respectively. Furthermore, both R1 and R3 positive control cultures and those gassed with 100%CH₄ displayed production of H₂ in low amounts (0.5–1.5%). In most of the tested conditions, along with the reported increase in CO₂, a proportional decrease in the concentrations of either N₂, CH₄ or CH₄/H₂ mixture occurred. No traces of CH₄ were detected in positive control samples, whilst its measurement in other samples was covered by the CH₄ already present in the headspace atmosphere. Finally, traces of H₂S were detected only within R1 positive controls and cultures supplemented with H₂ 10% at values in the range between 0.045 and 0.05%, whilst no traces of the gas were reported in other cultures where SRB growth was detected.

With respect to the cultures from R2 and R4, the monitoring of the headspace gas concentrations did not reveal significant changes in any of the tested conditions.

A modeling procedure was used to simulate microbial activity in the four reservoirs of interest for the case studies analyzed. Due to the effects that indigenous microbial communities could have on the stored gases, variations of headspace pressure and gas concentrations/partial pressures (i.e., CH₄, H₂ CO₂ and H₂S) were considered along with possible pressure drops due to the liquid and gas sampling. The results of the simulations on headspace gas compositions, pressures and VFA production are shown in Figure 5 (R1) and Supplementary Figures S6, S7 (R2, R3 and R4) in comparison to data obtained from experimental trials with external nutrient provisions. In these figures, each color represented in the legend corresponds to a specific case study: green for glucose tests, magenta for 100% CH4, blue for 10% H2, and red for 50% H2. Experimental gas phase data on Day 0 and 23 are depicted with solid symbols, while modeled values are represented by filled symbols (all the simulated and experimental data illustrated in Figure 5 and Supplementary Figures S6, S7 are reported in Supplementary Table S9). The figures also illustrate experimental trends for pressure and VFAs, which are shown with filled symbols, whereas the simulated trends are reported with dashed lines.

Results of the simulations indicates that the trends of pressure decreases are similar in almost all the analyzed cases and are mainly due to gas and liquid sampling and the dissolution of gases in the liquid culture, confirming the accuracy of the generated model. The model also provides information on the variation of moles present in the gas phase for each component. In particular, small H₂ productions are recorded in cases with glucose or 100%CH₄ only for R1 and R3. Slight consumption of hydrogen was recorded for all cultures, when H₂ was initially injected in the headspace. For all four reservoirs hydrogen consumption was almost negligible (< 0.1%) for H₂ 10%. Instead, in cases with H₂ 50%, the simulations report hydrogen consumptions around 1.2–1.6% compared to the initial moles for R1 and R3, and less than 0.2% for R2 and R4. As for CH₄, the simulations do not show any production in R2 and R4 in all four case studies. For R1 and R3, a small production of methane is recorded, less than 0.003% relative to the initial moles present, only in cases where H₂ is present from the beginning, particularly in the case with 50% of H₂. Such results obtained from the modeling seemingly highlight higher microbial activity in R1 and R3 when compared to R2 and R4, in accordance with the experimental results. Similar trends are observed when analyzing the growth kinetics of HM, AB, and SRB modeled with the growth model (Supplementary Table S1). The results shown in Figure 6 for R1 and Supplementary Figure S8 for R2, R3, R4 in the Supplementary material display the simulated trends of cell number growth for the three considered microbial classes.

Modeled results align with those obtained experimentally, emphasizing that AB are the most active microbial cluster in formation waters from R1, R2, R3 and R4. Regarding methanogenic metabolisms, a slight growth in cell numbers is recorded only for R1 and R3 in cases with H₂ 50% and H₂ 10%, but it is much lower than the values obtained for AB. Low microbial activity and cell number growth was also found for SRB, which report even lower levels than HM. The discrepancy in the growth levels of microbial classes seemingly points out minimal HM and SRB activity in the investigated reservoirs. Throughout the simulations, the monitoring of microbial activity confirmed that R2 and R4 have shown lower or decreased microbial growth compared to R1 and R3. Dominant presence of AB was observed in all four reservoirs, resulting particularly remarkable in R1 and R3. This dominance relates to the increased production of various acidic species, such as acetate. This predominance of acetogenic metabolisms is made more evident when analyzing the quantities of VFA produced, along with their mass balances, carbon and hydrogen uptake, as well as the selectivity of carbon and hydrogen distribution metabolized by microbes in various products generated by the three selected microbial classes, mainly CH₄, VFA, CO₂, and H₂S. Carbon uptake in moles of CH₄, CO₂, acetate, propionate and butyrate and hydrogen uptake in VFA, H₂S, CH₄, as calculated by the model are presented in Supplementary Tables S10, S11 of the Supplementary material. From the simulation results of considered cases, the majority of carbon is utilized for VFA production, whilst the remaining part is used for CO₂ production. Only in certain cases a minimal part is utilized for methane production, specifically for R1 and R3 in cases with H₂ 50% and H₂ 10%. The model reproduces a similar type of partitioning for hydrogen as depicted in Supplementary Table S3. As with carbon, most of the uptaken hydrogen is utilized for the production of VFA, with only a minimal part (<1%) being used for the production of methane and H₂S. Particularly for H₂S, in cases where hydrogen is not provided in the headspace gas mixture, simulations show minimal usage of acetate for its production. Higher uptake of carbon and hydrogen is observed in R1 and R3 in cases were H₂ was initially present in headspace gas mixture (i.e., H₂ 10% and H2 50%). A direct consequence of hydrogen and carbon uptake data is the selectivity data reported in Figure 7, for all four reservoirs at the end of fermentation.

For R1 amended with glucose, carbon is divided into 10% in CO₂, and the remaining 90% in VFA; in the case with 100%CH₄, about 15% goes to CO₂ and 85% to VFA. We have similar trends also for cases with 10%H₂ and 50%H₂, with carbon percentages destined for VFA greater than 90% and less than 1% for CH₄, which is present only in these two case studies. Also, the variability of generated products in terms of VFA increases for case studies with hydrogen present from the beginning. The results obtained for R3 are very similar to those obtained for R1 in terms of carbon division between CO₂, CH₄, and VFA. Unlike in R1, there is greater selectivity toward CO₂, always greater than 12%, and lower selectivity toward acetate (about 10–20% less than R1), mainly due to the higher production of other VFAs such as propionate, derived from acetate itself (Supplementary Table S1). For instance, for R3, in the case with H₂ 50%, there is about 17% in CO₂, 33.5% in acetate, 44.3% in propionate, 6.6% in butyrate, and less than 1% in CH₄.

For R2 and R4, the simulation results tell us that in glucose cases, more than half of the carbon uptake ends in CO₂, the remaining part in acetate. In H₂ 10% cultures, about 15% in CO₂ and the remaining part in acetate and propionate, and in cases with H₂ 50% about 5% in CO₂ and the remaining part in VFA. Similarly, selectivity of hydrogen uptake, points out that almost all hydrogen is used for VFA production, whilst less than 1% is used for production of H₂S and CH₄ in traces. Particularly for H₂S, a more accentuated presence is noticed when H₂ is initially present in tested gas mixtures. Extending the simulation time of the models from 23 days corresponding to the experimental time up to 90 days of simulated time provides information on what could happen to the bioprocesses until various limitations such as lack of nutrients or carbon or sulfate sources come into play. The results on the selectivity for carbon and hydrogen at 90 days are reported in Supplementary Figure S12 in the Supplementary material. As can be seen from the simulation results, the selectivity concerning hydrogen remains almost unchanged. The selectivity concerning carbon instead shows slightly different values compared to those obtained on 23 days. In particular, the selectivity concerning VFA generally increases (about 10%), and among the VFAs themselves, the selectivity toward acetate decreases in favor of other fatty acids such as propionate. For what concerns the trials that do not consider the addition of nutrients to the liquid phase, the simulation data are in line with the experimental results, underlining absence of activity and growth for the considered microbial clusters (i.e., AB; HM; SRB).

Given the importance that H₂ has in the program for full decarbonization by 2050, the development of valuable storage strategies is a paramount objective. In this context, UHS strategies gained interest due to the availability of large storage volumes in underground formations (Reitenbach et al., 2015; Verga, 2018; Pan et al., 2021; Muhammed et al., 2022). With several studies underlining the presence of specific microbial communities that might affect underground gas storage, a better understanding of their activity is required (Gniese et al., 2014; Valk et al., 2020; Molíková et al., 2022).

In the present study, we showed the presence of microbial activity in formation waters collected from four depleted gas reservoirs of potential interest from the perspective of UHS. However, the testing conditions were different from the actual reservoir conditions (e.g., pressure and nutrient supply), yet the activity was very limited.

Modeling of microbial risks associated with UHS encompasses a multitude of physical, chemical, and biological processes. In particular, hydrogenotrophic microbial processes play a significant role in UHS (Hagemann et al., 2016; Jin and Kirk, 2016) with the integration of flow and transport of nutrients, microbial population dynamics, and biochemical reactions being essential for the model’s final derivation.

Analyzing the production trends of acetogenic, methanogenic, and sulfate-reducing metabolisms offers crucial insights into microbial interactions and their overall impact on the efficacy and sustainability of hydrogen storage processes. A thorough model-based evaluation of these metabolic pathways allows for a nuanced understanding of microbial contributions, potential challenges, and enhance the performance of UHS strategies. In the present study a modeling approach was applied to microbial cultures seeded with formations water from reservoir R1, R2, R3 and R4 with the modeled results being scrutinized and compared with the experimental data reported in this work. Overall, data derived from the modelling are in accordance with the experimental ones, highlighting for the considered microbial populations a predominant activation of AB when compared to HM and SRB when external nutrients are provided. With particular regard to H₂, modeled data defined that highest H₂ consumption was reported in H₂ 50% amended cultures with only 1.2–1.6% of the initial H₂ molar content consumed in R1 and R3, whilst <0.2% was consumed in R2 and R4. Along with its consumption, the model allowed to define H₂ yields in the different metabolic products (i.e., VFAs, CH₄ and H₂S) of the considered microbial clusters. According to modeling data, most of the consumed H₂ would be redirected to VFA generation with <1% being consumed in methanogenesis and sulfate reduction.

Both experimental and modeling data reported in the present study seemingly underline low microbial effects on stored H₂ mixtures with the contribution of hydrogenotrophic clusters like HM and SRB being almost negligible in all tested reservoir fluids. According to the data reported in this work, low microbial densities and competition for H₂ with more active AB could be addressed as the main reason behind low activity displayed by the two hydrogenotrophic clusters.

This hypothesis is further sustained by the data regarding carbon yields for different products of considered microbial metabolisms. In fact, modeled data evidence how AB related metabolisms present the highest carbon yields when compared to HM and SRB, with most of the available carbon being incorporated in AB metabolic products (i.e., VFAs, CO₂). When conditions without external nutrient provision were simulated by the developed model, results aligned with the experimental data indicating absence of activation of the microbial clusters of interest in UHS. Along with the experimental ones, data obtained from the model seemingly indicate that microbial activation is strongly affected by nutrient availability, with hydro-chemical conditions similar to those found in the reservoir likely limiting the activity of the indigenous microbial communities. Model simulation estimates on the four reservoirs are generally in accordance with the experimental data presented in this work, underscoring how the developed model is grounded in robust foundations and could potentially be extended to field application for a better understanding of the microbial risk associated with UHS.