The two silicate minerals, quartz (Madagascar), alkali feldspar (Norway), were purchased from the Geology Superstore (http://www.geologysuperstore.com). Mineral samples were washed with 18.2 MΩ cm⁻¹ water and dried at 75°C for over 18 h. Minerals were then wrapped in several layers of paper roll (to avoid metal–rock interaction) and broken down using a sledgehammer on a metal plate. Crushed feldspar samples were washed again using 18.2 MΩ cm⁻¹ water, the 63 μm to 2 mm size fraction was collected and dried at 100°C prior to furnacing at 450°C for 4 hours to remove any trace organics. Quartz samples were washed using 18.2 MΩ cm⁻¹ water, the 63 μm to 2 mm size fraction was collected and dried at 100°C prior to furnacing at 1,000°C for 2 hours, to remove any organics and in an attempt to remove fluid inclusions (Bray et al., 1991; Kuznetsov et al., 2012). Crushed minerals where then milled under an oxygen-free nitrogen atmosphere (zero grade, BOC), using a Fritsch Planetary Mono Mill Pulverisette six at 500 r.p.m. for 30 min. Roughly 15 g of sample were crushed in the ball mill at any one time. Gas samples of the ball mill headspace were taken both before and after milling the samples. This process was repeated 10 times to generate sufficient crushed sample for sacrificial incubations of the minerals. After milling, the minerals were transferred to a Coy Vinyl Anaerobic Chamber (CoyLabs, MI, United States), previously flushed repeatedly with zero grade N₂ until Oxoid™ Resazurin Anaerobic Indicator (BR0055B, Thermo Fisher Scientific Inc.) indicated anaerobic conditions. Within this chamber, crushed samples were transferred into a large serum vial, which was sealed with grey butyl rubber stoppers (previously boiled in 1 M NaOH for 1 h and rinsed 6 × with 18.2 MΩ cm⁻¹ water), and crimp sealed. The headspace of the vial was then flushed with oxygen-free-N₂ for 2 min and stored in the dark until incubation.

Six × 6 g triplicate samples, one for each time point, were weighed into 50 mL borosilicate serum bottles (previously acid washed, rinsed 6 × with 18.2 MΩ cm⁻¹ water, and furnaced at 450°C for 4 h). They were placed into an anaerobic chamber which had been repeatedly flushed with N₂ until an Oxoid™ Resazurin Anaerobic Indicator showed that the headspace was anoxic. The serum bottles were then sealed with thick blue butyl rubber stoppers, previously boiled in 1 M NaOH for 1 h, soaked in de-ionised (DI) water overnight and rinsed 6 times with 18.2 MΩ cm⁻¹ water, which in turn were crimp sealed. The sealed vials were flushed for 2 min with oxygen-free-N₂ and stored at 0°C until the addition of water and the start of the incubation experiments. Triplicate blank vials were prepared for each time point following the same methods. Once all vials were crimped and flushed, approximately 8 mL of 18.2 MΩ cm⁻¹ water (flushed with N₂ for over an hour) at 0°C were added to each vial.

Serum bottles were analysed prior to the addition of water by over pressurising the bottles with 10 mL of oxygen-free N₂ (zero grade, BOC). A 10 mL headspace gas aliquot was transferred into a 5.9 mL double-wadded Exetainer® (Labco, Lampeter, United Kingdom) using a gas-tight syringe. Approximately 8 mL of 18.2 MΩ cm⁻¹ water (at 4°C and sparged with oxygen-free N₂ for over an hour) was then added to the serum bottles. The headspace in the vials was subsampled after 24 h following the same method. Headspace samples were then taken at 48, 120, 240, 360 and 720 h after the addition of water to the vials, following this same procedure. A 5 mL gas subsample was taken from the exetainers and analysed using an Agilent 8860 Gas Chromatographer (Agilent Technologies, Santa Clara, CA, United States). Concentrations of CH₄ and CO₂ were determined by a Flame Ionization Detector (FID). The sample loop was 0.5 mL, Helium (He) was used as the carrier gas, and a Porapak Q 80–100 mesh, 2 m × 1/8 inch × 2 mm SS column and a methaniser were used to distinguish the compounds. The concentrations of H₂ and O₂ were determined using a Thermal Conductivity Detector (TCD), using a 1 mL sample loop, Argon (Ar) as the carrier gas, and a Hayesep D 80–100 mesh, 2 m × 1/8 inch SS column, in series with a molecular sieve 5a, 60–80 mesh, 8 ft × 1/8 inch column. The oven temperature was set at 30°C for the initial 4 min, and then the temperature ramped up at rate of 50°C min^-1 until the oven reached a temperature of 200°C. This temperature was maintained for 2.5 min, when the run was concluded. The concentrations of headspace gases were calculated based on a standard-curve generated from the dilution of a mixed gas standard (173738-AH-C, BOC). Standards were run daily and gave a CV of 1.0% (n = 15) for H₂, with a limit of detection (LOD) of 9.6 ppm, equivalent to 4.4 nmol g⁻¹, a CV of 0.8% (n = 15) for CH₄, with a LOD of 0.1 ppm, equivalent to 0.04 nmol g⁻¹ and a CV of 0.9% (n = 15) for CO₂, with a LOD of 0.1 ppm, equivalent to 0.05 nmol g⁻¹. The ideal gas law was used to convert to molar concentrations, and concentrations were corrected for dilution during sampling and for gases dissolved in the water using Henry’s Law (where relevant). The results were also blank corrected for each time point and normalised to dry sediment mass.

The vials were opened immediately after gas sampling, and the liquid overlying the sediment was filtered using 0.22 µm Sartorius™ Minisart™ High Flow PES in-line filters. A 5 mL aliquot was used for pH measurements and H₂O₂ analysis, using DMP [method detailed in Baga et al. (1988); Gill-Olivas et al. (2021)].

The remainder of the filtrate was frozen immediately after filtering and stored at −20°C until further analysis for nutrients and organic acids. Anions and organic acids, including acetate (LOD: 1.1 ppb, equivalent to 0.02 nmol g⁻¹; CV: 1.4%), formate (LOD: 0.7 ppb, equivalent to 0.02 nmol g⁻¹; CV: 2.3%), F⁻ (LOD: 1.9 ppb, equivalent to 0.13 nmol g⁻¹, CV: 8%), Cl⁻ (LOD: 1.1 ppb, equivalent to 0.04 nmol g⁻¹, CV: 2.6%), NO₂ ⁻ (LOD: 2.2 ppb, equivalent to 0.06 nmol g⁻¹, CV: 2.2%), NO₃ ⁻ (LOD: 1.8 ppb, equivalent to 0.04 nmol g⁻¹, CV: 1.9%), and SO₄ ^2- (LOD: 3 ppb, equivalent to 0. 4 nmol g⁻¹; CV: 7%) were analysed using a Dionex ICS 6000 (ThermoScientific), fitted with a Dionex IonPac™ AS11-HC 4 µm column. Cations, including: Na⁺ (LOD: 0.3 ppb, equivalent to 0.50 nmol g⁻¹; CV: 0.3%), K⁺ (LOD: 0.4 ppb, equivalent to 0.36 µmol g⁻¹; CV: 0.6%), Mg²⁺ (LOD: 1.6 ppb, equivalent to 0.26 µmol g-1; CV: 1.0%) and Ca²⁺ (LOD: 1.1 ppb, equivalent to 0.35 µmol g-1; CV: 1.1%), were determined using a Dionex IC 5000 Analytical (Dionex, Sunnyvale, CA, United States), fitted with an IonPac™ CS12A 2 mm column.

Total dissolved iron (dFe) was determined by automating and implementing the ferrozine method described in Viollier et al. (2000) on a Gallery Automated Photometric Analyzer (Thermofisher), with an LOD of 13 ppb, equivalent to 1.58 nmol/g.

The raw concentrations of ions in solution, the concentration of headspace gases and pH values were used to determine saturation indices (SI) of quartz, calcite and aragonite. The thermodynamic equilibria software package PHREEQ-C (version 2.18.5570), using the standard (phreeqc.dat) database, and using an extended Debye-Hückel model to calculate ion activities from concentrations (Parkhurst and Appelo, 1999), was used to calculate the SI with respect to these minerals.

The specific surface area (SSA) of materials was measured using a NOVA 1200e BET Analysis System. Approximately 1 g of dried sample was loaded into a pre-calibrated sample cell. The whole system was then evacuated and dried at 200°C for a minimum of 2 h (Clausen and Fabricius, 2000). The surface area of materials was measured using nitrogen gas as an adsorbent at 77 K. The system was not calibrated using a certified standard, therefore the values obtained were indicative, rather than absolute.

Statistical analysis of the Na⁺, K⁺, Mg²⁺ and Ca²⁺ concentrations was conducted using the Dunn test in RStudio. There was no clear temporal trend for the concentrations of these cations, thus the values were pooled into a single population for each treatment (see Table 1). The results of this analysis were used to determine if there was any significant difference between the waters incubated with quartz, alkali feldspar or in empty serum vials.

Analysis of the gas in the headspace of the ball mill after crushing showed differences between the two minerals. No H₂ was detected during crushing of quartz or feldspar. Quartz released quantifiable concentrations of CO₂ (average of 8.1 nmol mL⁻¹, equivalent to 12 nmol g⁻¹) during dry crushing, but CH₄ concentrations were below the LOD (Table 2). By contrast, there was some CH₄ in the alkali feldspar samples (average of 0.6 nmol mL⁻¹, equivalent to 0.14 nmol g⁻¹), but CO₂ was below the LOD once corrected for the initial concentration in the headspace. The O₂ concentrations before and after crushing were similar, although quartz had a slightly higher average concentration (422 nmol mL⁻¹; Table 2), the concentrations were more variable between crushes (118 nmol mL⁻¹; Table 2).

Figure 1 shows the release of CH_4, H₂ and CO₂ from both minerals during the incubations. Both quartz and alkali feldspar produced quantifiable concentrations of CH₄. Concentrations of CH₄ increased linearly during the course of the incubation. Alkali feldspar produced slightly more CH₄ than the quartz samples, reaching a maximum of 0.4 nmol mL⁻¹ (equivalent to 0.16 nmol g⁻¹) and 0.1 nmol mL⁻¹ (equivalent to 0.05 nmol g⁻¹), respectively (Table 3; Figure 1A). The incubations produced significant concentrations of H₂. Alkali feldspar produced considerably more H₂ than quartz, with a median value of 8.7 nmol/g (average of 46 nmol mL⁻¹) after 24 h and increasing to 16 nmol/g (average of 96 nmol mL⁻¹) after 720 h of incubation. The concentration of H₂ produced by quartz was below the limit of detection until the 48 h time point (median value 3.8 nmol/g, average of 16 nmol mL⁻¹), and had a maximum median value at 240 h (4.7 nmol/g, or an average of 30 nmol mL⁻¹) and then decreased again (Figure 1B; Table 3). There was some CO₂ in the atmosphere of the incubations, after the blank correction (see Table 3). CO₂ concentrations in the headspace of the alkali feldspar incubations were equal to or below 0 nmol mL⁻¹ (Table 3). In contrast, the concentration of CO₂ in the headspace of quartz incubations increased over time (from 3.3 to 43 nmol g⁻¹ median value, Figure 1B).

There was silicate dissolution during the incubation of both quartz and alkali feldspar. The concentration of truly dissolved silica (dSi) increased quickly for crushed quartz, despite being oversaturated with respect to Quartz (Figure 2B). The median concentration of dSi after 24 h was 0.32 μmol/g, increasing to 2.0 μmol/g (median value) over the course of the incubation (Figure 2A). The concentration of dSi in alkali feldspar started slightly higher (medium value of 1.1 μmol/g) and increased at a slower rate, reaching a median value of 1.5 μmol/g at the end of the incubation (Figure 2A). The concentrations of dSi appeared to have stabilised after 240 h, when concentrations in these incubations become oversaturated with respect to Quartz (Figure 2B).

The concentrations of dissolved cations in solution showed some significant different between quartz and alkali feldspar incubations (Figure 3). The concentrations of monovalent ions (Na⁺ and K⁺) in the quartz incubations where not significantly different from the blanks (Figures 3A,B). Concentrations of divalent cations (Mg²⁺ and Ca²⁺) where low, but significantly higher than those in the blanks and the alkali feldspar incubations (Figures 3C,D). The solutions incubated with crushed alkali feldspar showed concentrations of monovalent ions significantly higher than the blanks (Figures 3A,B), whereas concentrations of Ca²⁺ were lower than those found in Quartz incubations, but still significantly different to those of the blanks (Figure 3C). The concentration of Mg²⁺ was not significantly different in alkali feldspar incubations than in blanks (Figure 3D).

Quantifiable concentrations of acetate and formate were found in both incubations (Figures 4A,B, respectively). Concentrations of acetate were significantly higher in alkali feldspar than in quartz incubations (Dunn’s test, p<0.001), and in both cases concentrations were significantly higher than those in blanks (Dunn’s test, p<0.01 and <0.001 for quartz and alkali feldspar, respectively). The formate concentrations were not significantly different between samples, and were significantly higher than in blanks (Dunn’s test, p<0.001). Finally, dFe was also measured in concentrations above the blank values (Dunn’s test, p<0.01), but not significantly different from each other (Figure 5).

Incubations of sample with water under anoxic conditions did not produce quantifiable H₂O₂ concentrations, given an LOD of 0.3 µM.

The SSA of crushed quartz was measured to be 21 m²/g and the SSA of crushed alkali feldspar was 8.1 m²/g.

The quartz had been furnaced at 1,000°C in an attempt to remove or at least to minimise the potential influence of fluid inclusions (Bray et al., 1991), and this is consistent with there being no quantifiable release of either CH₄ or H₂ during dry crushing. The release of CO₂ during crushing is most likely due to carbonate mineral impurities present in the sample, some of which dissociated into gas during crushing [Eqs 9,10; Martinelli and Plescia (2004); Martinelli and Plescia (2005)]. The trace amounts of carbonates, which may have been converted to CaO during furnacing, would have then been exposed to water during the incubation, producing CO₂, Ca²⁺ and, often, Mg²⁺ as they dissolve (Figures 1C, 3C). This suggests that even trace concentrations of carbonate minerals (<0.01%; Supplementary Figure S1) in silicate minerals can noticeably influence biogeochemical reactions, as has been noted in field studies of waters draining granitic terrain (White et al., 2005).

Crushing silicates produces both Si and SiO surface free radicals (Narayanasamy and Kubicki, 2005), which can react with water to produce H₂ and H₂O₂, respectively (Eq. 2–5). Further, it has been suggested when H₂O is split to produce H₂ it must produce stochiometric equivalent concentrations of oxidants (Sherwood Lollar et al., 2021). This is not immediately apparent during our incubations, where H₂ generation is apparent, but H₂O₂ was below the LOD in all samples. Organic matter in the mineral samples was minimised by furnacing, so it is unlikely that oxidation of organic matter is a sink for H₂O₂.

The presence of carbonate could account for the lack of H₂O₂ in the quartz incubations, since carbonate is a known HO· scavenger (Buxton and Elliot, 1986). Further, interactions between carbonate and OH·, generated by radiolytic decomposition of H₂O, has been suggested as an abiotic source of acetate and formate in fracture waters [Eqs 14–16 (Costagliola et al., 2017; Sherwood Lollar et al., 2021)]. Both acetate and formate were measured in the quartz and alkali feldspar incubations, and it is possible that the low concentrations found in the quartz incubations could be generated by carbonate and HO· interactions. By contrast, the higher concentrations of these ions in the feldspar incubations may arise from dissolution of fluid inclusions in the crushed mineral, which was only furnaced to 450°C, rather than to 1,000°C for quartz, the latter to minimise the potential influence of fluid inclusions on the incubations.

We note that the production of HO· can have both positive and negative effects on subglacial ecosystems, potentially producing VFAs, through the reaction of HO· with organic matter (Anastasio et al., 2007) or with trace carbonates (Costagliola et al., 2017). These VFA’s are readily available carbon sources to these subglacial environments which have otherwise very limited long term inputs of organic matter. However, H₂O₂ and HO· can also be very damaging to living organisms (Ikai et al., 2010; Sampson et al., 2012). Potentially, our results suggest that even trace amounts of carbonate can rapidly react with HO·, thus potentially protecting microbes in these environments and producing potential energy substrate sources simultaneously.

Quartz is often considered a relatively unreactive mineral, due to its low solubility and slow dissolution kinetics (Crundwell, 2017). However, abrasion increases the solubility of both quartz and feldspar. Figure 2 shows that silicate dissolution is taking place (Figure 2A), even when PHREEQC modelling suggests that this solution is oversaturated with respect to quartz (Figure 2B). This enhanced silicate solubility is a consequence of the excess surface free energy imparted to the minerals during crushing (Petrović et al., 1976; Petrovich, 1981). The large surface area created during crushing also promotes more rapid rates of dissolution. Further, quartz comminution and wetting produces low concentrations of H₂ (Figure 1B). Hence, even relatively “unreactive” minerals such as quartz can produce compounds of biological utility when abraded.

Alkali feldspar dissolution kinetics are more rapid than quartz (Lerman, 1988). The water chemistry in the feldspar incubations produced a silicate weathering signature, with monovalent ions present in higher concentrations than divalent ions (Figure 6), consistent with there being no trace carbonate in the feldspar, as well as high dSi concentrations (Figure 2). The concentration of dSi, increased rapidly after the addition of water, whereas concentrations of Na⁺ and K⁺ where immediately high and stayed relatively stable, as did the pH solution (pH 11.2; Figure 7), consistent with the results from PHREEQC models and hydrolysis being the initial reaction in solution (Keller and Reesman, 1963; Garrels and Christ, 1965). Hydrolysis produces a leached surface layer on the feldspar surface, depleted in base cations. Si dissolves much more slowly from this leached layer. Further dissolution of base cations is via diffusion through the leached layer, which is also slow (Lerman, 1988). The concentration of dSi increased more rapidly in quartz incubations than in alkali feldspar incubations, potentially because the surface area of the quartz (21 m²/g) was much higher than that of the alkali feldspar (8.1 m²/g). Both solutions are supersaturated with respect to crystalline quartz because comminution produces areas with high excess surface energy on the crushed minerals, such as amorphous zones, dislocations and adhering microparticles (Petrović et al., 1976; Petrovich, 1981). This makes the surface super-soluble with respect to crystalline quartz, and hence the solution becomes supersaturated as Si dissolves from these sites.

The SSA of crushed silicate samples has been linked to the generation of Si free radicals and thus H₂ production (Telling et al., 2015). However, quartz produced considerably less H₂ than alkali feldspar (Figure 1B) despite the higher measured SSA. Increased H₂ generation has also been linked to the presence of Al-OH, which are often found in sheet silicates, which can also react with Si· and generate H· (Kameda et al., 2004). These Al-OH bonds could also be present in alkali feldspar samples and be a potential source of additional H₂. A potential control on the products of abrasion is the different shearing that these minerals undergo. Potentially, the structure of these minerals favours either homolytic or heterolytic cleavage, resulting in either H· or H⁺, respectively (Delogu, 2011). Delogu (2011) found crushing quartz increased both H· and H⁺ simultaneously, as the generation of active sites produces H·, which recombine to form H₂ molecules, and silanol groups (SiOH), which can generate H⁺. However, the data presented here did not follow this trend. The pH of water incubated with crushed quartz did decrease slightly relative to the ultra-pure water incubations, potentially through the mechanisms outlined above. Yet, the high pH generated by surface hydrolysis of the feldspar confounds easy comparison of the experiments in terms of distinguishing between homolytic and heterolytic effects.

Even simple silicate minerals, such quartz and feldspar, contain trace amounts of other minerals which make it difficult to unequivocally determine the mechanochemical reactions initiated by abrasion. These experiments, using naturally occurring minerals, rather than ultrapure minerals with a minimal fluid inclusion content, produces results which are more representative of natural subglacial systems where abrasion is taking place, and highlights some of the potential interactions that occur between bulk and trace minerals. However, these interactions make it challenging to identify the individual reactions taking place in order to quantify the effects of these interactions. This is even the case for the quartz incubations, which were furnaced at 1,000°C in an attempt to remove fluid inclusions and organic matter, and was considered pure (with less than 0.01% of impurities), yet produced CO₂, CH₄ and acetate and formate upon crushing. These can be produced by the reaction of carbonates with free radicals, as suggested above (Section 4.1), however, this is difficult to confirm without quantifying fluid inclusions, if present, and their composition.

Feldspar samples were furnaced at 450°C to remove any OM contamination, however this temperature does not remove fluid inclusions. Fluid inclusions can be trapped within the mineral structure during the crystallization of magma. The composition of these fluid inclusions is highly variable, containing traces of traces of gases and fluid from the environment where they were formed. Fluid inclusions in crystalline rocks, such as quartz and granitoids (which usually include feldspars), are often rich in hydrocarbon gases (Potter and Konnerup-Madsen, 2003). Further, it is challenging to quantify the fluid inclusions present within minerals. Thus, it is difficult to differentiate between the reactions taking place when freshly abraded surfaces react with water, and the compounds released from these fluid inclusions. There was no evidence of carbonate weathering in the solute of the feldspar samples, therefore it is likely that at least some of the acetate, formate and CH₄ measured during the incubation of feldspar samples originates from fluid inclusions.