Soil was collected in the surroundings of the backstop berm of a military shooting range located in Brocheten, Switzerland (47° 19′ 5218″ N and 007° 34′ 6546″ E; canton Solothurn). The soil originated from a longer-term (> 30 years) contaminated shooting range soil characterized by nutrient-poor, moderately dry pasture (BSB + Partner, Ingenieure und Planer 2007).

Hotspots of Sb and Pb concentration were determined using a handheld XRF device upon collection. There, soil was taken from the surface layers (0–30 cm) and, after roots and sods were removed, was homogenized, air-dried (< 2 wt% water content), passed through a sieve (≤ 2 mm), and stored in buckets in the dark. The soil texture qualified as coarse sand (7.60% humus, 27.10% pebbles, 27.30% silt) according to the international system for particle size distribution analysis and contained 16.48 ± 0.31% CaCO3 and 4.41% organic carbon. Soil pH was 7.40 ± 0.01 (measured in 0.01M CaCl2). It displayed a cation exchange capacity of 0.34 ± 6.72 meq/100 g soil. The total metalloid concentrations (Mn, Fe, Sb, and Pb) in the soil samples collected were quantified by energy-dispersive X-ray fluorescence spectroscopy (SPECTRO XEPOS, AMETEK, Germany). The mean concentrations in the soil used in this study were 1255 ± 49 mg kg−1 Mn, 42,018 ± 535 mg kg−1 Fe, 27 ± 4 mg kg−1 Sb, and 766 ± 84 mg kg−1 Pb. The certified reference material “BCR-176R fly ash” (European Community Bureau of Reference, BCR, Sigma-Aldrich, Switzerland) was used for calibrating XRF analyses, yielding 92% of the certified expected elemental recovery (Table S1).

In order to determine metal partitioning/distribution in the soil before and after treatment in the reactors (i.e., to constrain how metals are bound or associated with different soil phases), a four-step sequential extraction/digestion on certified reference material “BCR-701” was used according to BCR recommendations (European Comission 2012), followed by ICP-MS analysis of the single extracted fractions. More precisely, metals were extracted by four consecutive steps into the following fractions (F1-F4): “exchangeable” F1 (step 1 with 0.11 M acetic acid at pH 2), “reducible” F2 (step 2 with 0.5 M hydroxylammonium chloride at pH 2), “oxidizable” F3 (step 3 with 8.8 M hydrogen peroxide at pH 2), and “residual” F4 (step 4 with aqua regia). The non-extractable fraction of Sb was calculated by difference between the total Sb content (determined by XRF) and Sb extracted in F1 to F4.

The soil–water system of the bioreactors was simulated using thermodynamic equilibrium modeling with Visual MINTEQ database (Gustafsson 2014) implemented in Geochemist Workbench version 12.0 (Champaign, Illinois). The full model containing all species in the database, as well as a reduced model (suppressing different Sb oxides), is presented in the Supplementary information (Figure S1a, Figure S1b). The computed Pourbaix diagrams indicate that, at neutral pH, Sb(V) reduction to Sb(OH)3 would occur at a redox potential lower than Eh =  ~ 125 mV.

Regarding Mn (Figure S1c), the diagram suggested that at Eh < 450 mV, the conversion of Mn(III) (as BixByte) to dissolved Mn2 + would be expected.

To safely suppress both Sb and Fe reduction (at <  − 250 mV, Figure S1d), a minimal redox potential of Eh =  ~ 400 mV was chosen for the redox-stat. The range of 450 mV < Eh < 350 mV is hereafter referred to as “manganese-reducing conditions.”

The experiments were performed in two 1-L continuously stirred-tank bioreactors (CSTR, Multifors, INFORS HT, Bottmingen, Switzerland) in mesophilic (21 ± 5 °C) conditions at a pH of 7.0 ± 0.3 (Figure S2). Soils were added at a solid–liquid ratio of 10% (w/v), and the slurry was stirred at 70 rpm. The hydraulic retention time was set to 48 h, where the bioreactor was fed with an anoxic artificial rainwater solution with the following composition (in g L−1): 2.13 NaCl, 3.63 MgCl2·7H2O, 3.06 CaCl2·6H2O, 0.37 KCl, 1.53 NaNO3, and 2.98 Na2SO4. The headspace of the bioreactors was constantly flushed with N2 during operation. The bioreactors were covered with aluminum foil to protect them from light and thus to prevent photooxidation, as well as algae growth. The principal aim of this study was to acquire a comprehensive understanding of the distinct contribution of Mn reduction in Sb mobilization. To achieve this, in one bioreactor (RMnR), the redox potential was set to “manganese-reducing conditions” (see the “Thermodynamic modeling” section), as previously described (Rajpert et al. 2018). In RMnR, the redox potential was stabilized by sporadic air injections into the reactor headspace. As an adaption to the setup used by (Rajpert et al. 2018), instead of a standard digital redox—processor, a NI CompactRIO and LabVIEW (National Instruments, Ennetbaden, Switzerland) was used to control the gas valves (compressed air input), which allowed us also to record Eh data in 30-min intervals instead of manual readouts. In the other bioreactor, which served as a control (RCTRL), we let the redox conditions develop naturally, allowing for the establishment of much more reducing conditions in the incubated soil.

Samples (15 mL) were collected in duplicates from the reactors through a sterile needle and syringe and ultra-centrifuged (Amicon Ultra-4 Centrifugal Filter Unit, 30 kDa MWCO; 4500 rcf, 21 °C, 15 min). The concentrations of total dissolved Sb, Pb, Mn, and Fe were quantified using an Agilent 8800 qqq-ICP-MS (Agilent Technologies AG, Basel). Prior to analysis, the samples were diluted (1:200) in 3% HNO3 (Merck, Switzerland). All elements were measured in helium-collision mode, monitoring masses 55Mn, 56Fe, and 121Sb. The limits of quantification (LOQ) were determined as ten times the standard deviation (σ) of a set of blanks (n = 10), while the limit of detection (LOD) was determined as three times σ. The determined LOQ values for Sb, Mn, and Fe are 0.69 µg L−1, 0.35 µg L−1, and 1.19 µg L−1, respectively. Correspondingly, the LOD values for Sb, Mn, and Fe are 0.21 µg L−1, 0.10 µg L−1, and 0.36 µg L−1, respectively.

Antimony speciation analyses were carried out by LC-ICP-MS (Lintschinger et al. 1998) (Agilent Technologies AG, Basel, 1260 pump series). The samples for Sb speciation were handled in the glove box, filtered through 0.45-mm PVDF filters, stabilized in the degassed mobile phase (10 mM Na-EDTA, 1 mM phthalic acid, and 2% methanol at pH 4.5), stored at 4 °C, and measured within 72 h after sampling. An anionic exchange column (Hamilton PRP-X100, 250 × 4 mm, 10 µm) with an isocratic elution was used for the separation. The injection volume was 100 µL, and the flow rate was 0.8 mL min−1. To improve nebulization, the flow rate was split in half before entering the nebulizer of the ICP-MS. The calibration standards of Sb(III) and Sb(V) were freshly prepared from Sb2O3 dissolved in 2M HCl (Merck, Switzerland) and from KSb(OH)6 (Merck, Switzerland) dissolved in water, respectively. The dissolved reduced iron (Fe2+) concentration was measured spectrophotometrically using 1,10-phenanthroline (Fadrus and Malý 1975). From the measured trace metal concentration changes in the reactor, trace metal mobilization rates were calculated as described in Equation S3 and Equation S4. Dissolved organic carbon (DOC) was quantified using a total organic carbon analyzer (TOC-L Shimadzu).

Microbial genomic DNA was extracted from the soil samples using the DNeasy PowerSoil Pro Kit (Qiagen, Netherlands). DNA was quantified using a Qubit™ dsDNA HS Assay Kit (Thermo Fischer Scientific, Waltham, MA, USA). Gene-targeted sequencing of the 16S ribosomal RNA was performed using the Quick-16S™ NGS Library Preparation Kit (Zymo Research, Irvine, CA). The 515F-806R primers were used as they were designed to amplify prokaryotes (Bacteria and Archaea) and target the V3-V4 region of the 16S rRNA gene. The sequencing library was prepared in real-time PCR machines to control the cycles and prevent PCR chimera formation. The final PCR products were quantified with qPCR fluorescence and pooled together based on equal molarity. The final pooled library was cleaned with Select-a-Size DNA Clean & Concentrator™ (Zymo Research, Irvine, CA) and quantified in Qubit. The final library was sequenced on Illumina® MiSeq™ with a V3-V4 reagent kit. Raw sequencing reads were processed in R using dada2 (Callahan et al. 2016), including the removal of primer sequences, quality control, estimation of error rates, and detection and removal of chimeras. The resulting sequence table was aligned to version 138 of the SILVA ribosomal RNA database (Quast et al. 2013), specifically the non-redundant dataset 99. Subsequently, a phyloseq object was constructed using the phyloseq package (McMurdie and Holmes 2013), which encompassed an amplicon sequence variant (ASV) table, a taxonomy table, and sample data. Further analysis was conducted using the R packages phyloseq (McMurdie and Holmes 2013) and vegan (Oksanen et al. 2020). Raw sequencing data are deposited on NCBI SRA archive under bioprocess accession number PRJNA997570.

A multiple linear regression to a significance level of p < 0.05 was employed to determine possible correlations between Sb total in the effluents and Fe, Mn, Pb, and DOC concentrations, as well as the redox potential and pH for the entire experiment (see Table S4).