Shewanella sp. ANA-3 strain WT and the non-As(V)-respiring arrA deletion mutant, S. ANA-3 strain ARRA3, were used in these experiments. The bacterial strains were kindly donated by Professor Dianne Newman (California Institute of Technology). Both strains were grown on aerobic lysogeny broth (LB) agar plates at 30°C for 24 h. Isolated colonies were used to inoculate aerobic LB in Erlenmeyer flasks and incubated at 30°C/150 rpm for 24 h until late-logarithmic phase. Five ml of this aerobic culture was transferred to inoculate serum bottles containing 100 ml of anoxic minimal growth medium [in g/L: K₂HPO₄ 0.225, KH₂PO₄ 0.225, NaCl 0.46, (NH₄)₂SO₄ 0.225, MgSO₄⋅7H₂O 0.117, NaHCO₃ 4.2, fumaric acid 4.64; in ml/L: sodium lactate 2.59, trace elements, and vitamin solutions 10 ml each] (Saltikov et al., 2003). These bottles were incubated statically in the dark at 30°C. After 48 h (late-logarithmic phase), the cells were washed twice with anoxic 30 mM bicarbonate buffer by centrifuging at 2,509g/30 min, and the resulting pellet was re-suspended in a small volume of the same buffer, keeping the anoxic conditions throughout this washing process. The concentration of cells was measured as OD₆₀₀ in a Jenway 6715 UV/Vis spectrophotometer and then diluted for inoculation into the arsenic-Fe(III)-(oxyhydr)oxide thin film experiments.

Amorphous Fe(III)-(oxyhydr)oxide (ferrihydrite) was prepared by dissolving 16.2 g of FeCl₃ in 500 ml of deionized water (_dH₂O), the pH was raised to 7.0 by adding a 10 M NaOH solution under constant stirring and the resulting solution was washed with _dH₂O and centrifuged (2,688g/20 min), with the process repeated six times (Schwertmann and Cornell, 2000). The final concentration of biologically available Fe in the ferrihydrite suspension was 140 mmol per liter slurry. An As(V) solution as sodium arsenate (Na₂HAsO₄⋅7H₂O, Sigma-Aldrich) was prepared and mixed to the ferrihydrite to give a final concentration of 12% mol/mol As/Fe. This mixed solution was stirred on a roller shaker for 24 h to promote the sorption of As(V) on the Fe(III) mineral surface sites. Thin films of arsenical Fe(III)-(oxyhydr)oxide were prepared on boron-doped silicon wafers (7.2 mm × 7.2 mm × 0.5 mm) by pipetting 40 μl of the 12% mol/mol As/Fe solution and allowing to dry overnight.

Microbial incubation samples for NanoSIMS analysis were prepared in 15 ml anoxic serum bottles. The thin film-coated Si wafers were placed vertically in a plastic holder that was fixed to the bottom of the bottle with silicon grease. Freshwater medium (in g/L: NaHCO₃ 2.5, NH₄Cl 0.25, NaH₂PO₄⋅H₂O 0.6, KCl 0.1, vitamin mix, and mineral mix 10 ml each) (Wilkins et al., 2007) was added to the serum bottles (7 ml) and amended with 20 mM ¹³C-labeled sodium lactate [¹³CH₃CH(OH)CO₂Na, Sigma-Aldrich^®]. In these experiments, the arsenical Fe(III)-(oxyhydr)oxide was the sole terminal electron acceptor, and ¹³C-labeled lactate was used as the electron donor, where ¹³C was expected to be assimilated by the metabolically active cells and so allow their identification during the experiment. Cells of both S. ANA-3 strains were inoculated in three replicates under anoxic and sterile conditions to give a final OD₆₀₀ of 0.5, following a procedure reported previously (Newsome et al., 2018). The bottles were incubated at 30°C in the dark for 11 days. Two control experiments were prepared, S. ANA-3 WT with no electron donor added (“no donor”), to assess the effect of cells with no metabolic activity, and uninoculated freshwater medium (“no cells”), to assess abiotic reactions.

Supernatants were withdrawn on days 0, 4, 8, and 11 of incubation to measure aqueous Fe(II), As species and flavins. Fe_total and As_total were only measured on the last day of incubation. The 0.5 M HCl-extractable Fe(II) and total biologically available Fe(III) were quantified by reacting the aqueous samples with a ferrozine solution and measuring the absorbance at 562 nm, following a standard methodology (Lovley and Phillips, 1986). As(III) and As(V) species in solution were quantified by diluting the samples in _dH₂O and analyzed by inductively coupled plasma mass spectrometry (ICP-MS, 7500CX; Agilent Technologies, United States). Fe_total and As_total were measured by acidifying the aqueous samples in 2% v/v HNO₃ and analyzed by inductively coupled plasma atomic emission spectroscopy (ICP-AES, Perkin–Elmer Optima 5,300 DV). FMN and riboflavin were measured in the supernatants using HPLC with a fluorescence detector, as detailed elsewhere (Newsome et al., 2018). Riboflavin and FMN solutions (0.1–1.0 μM) were used as references. Plots and statistical analyses were conducted using OriginPro 2019b^®.

On day 11, all the Si wafers (samples and controls) were removed from the serum bottles, preserved by chemical fixation with glutaraldehyde and dehydrated sequentially with ethanol in an anaerobic cabinet, as detailed previously by Newsome et al. (2018). The wafers were removed from the last ethanol solution, air dried in an anaerobic cabinet overnight and once dry, coated with 10 nm of Pt using a sputter coater.

Scanning electron microscopy (SEM) imaging was used to localize areas of interest with cells and mineral, using a FEI Quanta 650 FEG SEM with a 15 kV beam. The SEM image acquisition was kept to large areas to reduce beam damage to the cells as much as possible. The wafers were stored under anoxic conditions between analyses.

Two wafers of each sample were analyzed in a NanoSIMS 50 L ion microprobe (CAMECA, France) using a 16 keV Cs⁺ primary ion beam. The primary ion beam was scanned over the surface of the samples with a current of 1.08–0.78 pA to obtain a spatial resolution of ≈400–100 nm (aperture D1 = 3–5). The CAMECA mass resolving power values of 5,500 and 7,000 were used (ES = 3 and AS = 2) for detectors 2 (mass 13) and 7 (mass 75), respectively. Iron metal, gallium arsenide and a clean silicon wafer were used as reference materials to improve mass resolution and avoid peak overlaps from molecular interferences, such as ¹²C¹H^– at mass 13, ²⁸Si₂¹⁶O^– at mass 72, and ⁵⁶Fe¹⁹F^– at mass 75 (as detailed in the next section). The following negative secondary ions were collected simultaneously: ¹²C, ¹³C, ¹²C¹⁴N, ²⁸Si, ⁵⁶Fe¹²C, ⁵⁶Fe¹⁶O, and ⁷⁵As were analyzed using a double focusing mass spectrometer. Images were collected at a dwell time of 5,000 μs px^–1 with a pixel resolution of 256 × 256. Additionally, an ion-induced secondary electron (SE) image was obtained. The ¹²C, ¹³C, and ¹²C¹⁴N signals were used to identify the biomass, and the ¹³C/¹²C ratio was used to identify active cells by their ¹³C accumulation (above the 1.11% natural ¹³C/¹²C abundance); ²⁸Si was used to image the wafer surface; ⁵⁶Fe¹²C and ⁵⁶Fe¹⁶O were used as ⁵⁶Fe proxies because ⁵⁶Fe has a low ionization under Cs⁺ bombardment (Wilson, 1995) [⁵⁶Fe¹⁶O is easily detected on the Fe(III) oxyhydroxide, whereas ⁵⁶Fe¹²C is easily detected on the cell surface], and along with ⁷⁵As, these secondary ions were used to map the arsenical Fe(III) oxyhydroxide. Implantation of Cs⁺ ions at a dose of 1 × 10¹⁷ ions cm^–2 was used to remove the Pt coating and reach steady state on the sample surface before collecting chemical images (McPhail and Dowsett, 2009).

Measurement of aqueous Fe(II) by ferrozine showed that by day 11, insoluble Fe(III) had been mobilized to soluble Fe(II) by both S. ANA-3 strains incubated with ¹³C-lactate as the electron donor (26 μM by the WT and 21 μM by the ARRA3) (Supplementary Figure 2A). No Fe(II) was detected in the “no donor” or in the “no cells” controls (Supplementary Figure 2A). The Fe_total solubilized at day 11 (29 μM by the WT and 22 μM by the ARRA3) was measured by ICP-AES (Figure 1A), and the values were comparable to the total Fe(II) levels measured by ferrozine. However, ICP-AES also allowed the detection of small quantities of Fe_total in the “no cells” controls (1 μM) by day 11 (Figure 1A), suggesting that the ferrihydrite thin film was partially soluble under the experimental conditions. Slightly higher Fe_total levels were measured in the “no donor” control (9 μM), suggesting that the washed cells retained some reducing power. In earlier exploratory work (Supplementary Figure 2B), Fe_total was only mobilized in incubations with cells and electron donor, and its concentration increased with incubation time.

The As_total measured by ICP-AES showed that S. ANA-3 WT solubilized higher amounts of As (71 μM), although this value was not significantly higher (p > 0.05, Tukey–Kramer test) than the ARRA3 mutant (45 μM), S. ANA-3 WT “no donor” (41 μM) and the “no cells” control (38 μM) (Figure 1A). The As mobilization observed in the “no cells” control implies that As was partially desorbed under these experimental conditions, as observed previously (Newsome et al., 2018), and consequently, approximately 50% of the As mobilized in the experiments with cells is likely to be due to abiotic desorption. S. ANA-3 WT released 83% of the total As in the form of As(III), whereas the ARRA3 cells released negligible (0.2%) As(III), as expected given its lack of an As(V) reductase (Figure 1B and Supplementary Figure 3). The “no donor” control released 33% of As_total in the form of As(III), which was somewhat surprising, although this again may be explained by the presence of residual endogenous reduced cofactors within the cells at the start of the experiment (Figure 1B and Supplementary Figure 3). It is noteworthy that As(III) levels remained constant from day 4 of incubation in this control experiment, whereas the ¹³C-incubated WT showed an increasing As(III) release throughout the period of incubation. Similarly, solubilized As(V) increased through time in all experiments and controls. The abiotically desorbed As from the “no cells” control was 99.9% As(V) (Figure 1B and Supplementary Figure 3). There were slight discrepancies between the As_total quantification (Figure 1A) and the total As estimation obtained by adding the two As species (Figure 1B), potentially due to the sample preparation, where the As_total by ICP-AES samples was extracted in HNO₃, possibly solubilizing more As, in contrast to the ICP-MS samples that were prepared in _dH₂O.

The concentration of flavins (FMN and riboflavin), which can act as extracellular electron shuttles to Fe(III) oxides, was monitored in the aqueous phase. Flavins were detected at similar levels by day 11 in all cultures except in the “no cells” control (Figure 2), where the “no donor” control released 205 nM total flavins, followed by ¹³C-incubated S. ANA-3 WT (186 nM) and ARRA3 (185 nM). From the start of the experiment, FMN was detected in the supernatants of all experiments with added cells, and its concentration decreased over the incubation time (Figure 2), whereas riboflavin concentrations increased (Figure 2). Although the original freshwater medium contained 61 nM riboflavin, none was detected in the “no cells” control, suggesting that all the riboflavin detected in the experiments inoculated with Shewanella spp. was produced by the cells.

Scanning electron microscopy showed that the majority of cells (>90%) of both S. ANA-3 strains incubated with ¹³C-labeled lactate were observed at distances of more than a micron away (1–20 μm) from the Fe(III) mineral on the Si wafer surface (Figure 3). A small proportion of the cells of both S. ANA-3 strains were observed on the Si wafer but adjoining the Fe(III) mineral (Figure 3C). Very few cells were observed to be present directly on the mineral surface, in contrast to previous experiments with G. sulfurreducens (Newsome et al., 2018), which is known to require direct contact with insoluble Fe(III) minerals to respire them. Scarce cells were observed in the “no donor” control, suggesting a much lower biomass production than the ¹³C-incubated WT and ARRA3 samples, as expected, due to the lack of an organic substrate to use as electron donor and carbon source (Figure 3E).

Not all the cells observed by NanoSIMS accumulated ¹³C, and these were considered to be metabolically inactive. The majority of the active cells of both S. ANA-3 strains were observed at a distance from the Fe(III)-(oxyhydr)oxide mineral on the uncoated Si wafer surface of all the areas analyzed for each strain (N = 94 for the WT and N = 71 for the ARRA3) (Figure 4 and Supplementary Figures 4, 5). A few active cells were observed to be in direct contact with the Fe(III)-(oxyhydr)oxide surface, either on the uncoated Si wafer surface but adjoining the Fe(III)-(oxyhydr)oxide mineral (N = 17 for the WT and N = 22 for the ARRA3) or growing directly on the Fe(III)-(oxyhydr)oxide surface (N = 5, only in the WT) (Figure 4C). The mean ¹³C/¹²C accumulation of S. ANA-3 WT was 7.4% ± 5.7 for cells on the uncoated Si wafer, 4.7% ± 1.7 for cells adjoining the mineral and 3.9% ± 0.5 for cells growing on the Fe(III) mineral surface; this implies that overall, the WT cells accumulated 3–6 times more ¹³C than the natural ¹³C/¹²C ratio (Supplementary Figure 6A). The ARRA3 mutant cells accumulated a mean ¹³C/¹²C of 7.1% ± 4.5 for cells on the uncoated wafer and 5.2% ± 0.6 for cells adjoining the Fe(III) mineral, implying an accumulation of up to 4–6 times the natural ¹³C/¹²C ratio (Supplementary Figure 6B). The mean ¹³C/¹²C of cells on the uncoated Si wafer compared to other locations on the sample [adjoining and/or growing directly on the Fe(III) mineral] was not significantly higher (p > 0.05, Tukey–Kramer test) in cells of both S. ANA-3 strains. Chemical fixation can dilute the ¹³C fraction in cells (Musat et al., 2014), leading to an underestimation of the incorporation rates of isotopically-labeled lactate, although the estimated ¹³C/¹²C ratio was appropriate to locate metabolically active cells and the calculation of absolute ratios was not the aim of this work. The areas analyzed in NanoSIMS can be seen in Supplementary Figures 4, 5 for the WT and ARRA3 strains, respectively.

As and Fe on the Fe(III) mineral surface were mapped using NanoSIMS imaging, where ROIs were selected in all samples to compare normalized ⁵⁶Fe¹⁶O and ⁷⁵As counts (Supplementary Figure 7). ⁷⁵As and ⁵⁶Fe¹⁶O showed a positive correlation in all samples, as shown by linear regression fitting, where R² was above 0.6 in all plots (Supplementary Figure 7). The ⁷⁵As counts as a % of ⁵⁶Fe¹⁶O counts were compared (Figure 5); the mean ⁷⁵As/⁵⁶Fe¹⁶O ratio was 6.0% ± 6.7 on the Fe mineral of the S. ANA-3 WT, 1.1% ± 0.4 in S. ANA-3 ARRA3, 0.9% ± 0.8 in the S. ANA-3 WT “no donor” and 0.2% ± 0.1 in the “no cells” control. The ⁷⁵As/⁵⁶Fe¹⁶O ratio on the mineral in the WT sample was significantly higher than that on the mineral areas of the ARRA3 and control samples at the 99% confidence level (p < 0.01, Tukey–Kramer test).

Fe and As were observed to be co-located with cells in the NanoSIMS analysis using large fields of view (30–50 μm). Therefore, we selected single active cells to collect NanoSIMS depth profiles at higher spatial resolution and prolonged acquisition to image trace Fe and As in active cells of both S. ANA-3 strains. On the surface of the cells, low signal of Fe (⁵⁶Fe¹²C and ⁵⁶Fe¹⁶O) and As (only in the WT cells) was imaged (Figures 6A,B,G,H). Although ⁵⁶Fe¹²C was expected to produce a more intense signal than ⁵⁶Fe¹⁶O due to the naturally high ¹²C content on the cell surface, the ion counts were similar. For the ARRA3 strain, Fe was observed on the cell surface but not As (Figures 6G,H); in contrast, the WT exhibited Fe and As (Figures 6A,B). The ⁵⁶Fe¹⁶O and ⁷⁵As counts on the cells were much lower than those on the Fe(III) mineral surfaces (Supplementary Figures 4, 5), where ⁵⁶Fe¹⁶O and ⁷⁵As counts on the Fe(III) minerals were at least one order of magnitude higher than those on the cells. Additional 3D models of single-cell depth profiles of duplicate cells of both S. ANA-3 strains are provided as Supplementary Figure 8.

NanoSIMS imaging showed that cells of both strains accumulated ¹³C at variable levels above natural abundance, which could reflect, for example, different growth stages, stochastic gene expression, etc. (Peteranderl and Lechene, 2004; Zimmermann et al., 2015). A small number of active WT cells (N = 4) were observed on the Fe(III) surface. We hypothesize that these cells directly transferred electrons from c-type cytochromes on the cell surface to the Fe(III)-(oxyhydr)oxide mineral, obviating the use of endogenously produced flavins. The majority of active cells (90% of the total) were imaged at a distance from the Fe(III) mineral, suggesting that the dominant Fe(III)-(oxyhydr)oxide reduction mechanism in S. ANA-3 is not through direct cell–mineral contact but through electron shuttling under these experimental conditions. This is in agreement with previous observations of Shewanella species not requiring direct contact with the Fe(III) mineral that they were respiring, instead using flavins (endogenous or exogenous) to mediate the extracellular transfer of electrons in non-contact scenarios (von Canstein et al., 2008; Edwards et al., 2015; Xu et al., 2016). It has been suggested that the energy available from the reduction of Fe(III) would make it worthwhile for S. ANA-3 cells to invest energy in producing and releasing flavins (Covington et al., 2010), which in themselves have high reducing capacities with reduction potentials of −208 and −219 mV for riboflavin and FMN, respectively (van der Zee, 2002). The amount of flavins secreted in all samples in this experiment was in the sub-micromolar level, in accordance to previous reports for other Shewanella species (Marsili et al., 2008; von Canstein et al., 2008). Flavins have proven to be effective mediators for reductive Fe(III) dissolution at nanomolar concentrations, and it has been observed that above 1 μM, the rate of Fe(III) dissolution is not enhanced (Wang et al., 2015). Moreover, the conversion of FMN to riboflavin has been reported in cultures of different Shewanella strains (von Canstein et al., 2008). In addition to flavins, other molecules acting as electron shuttles could also be at play, such as the elusive and only recently identified ACNQ (2-amino-3-carboxy-1,4-naphthoquinone) (Mevers et al., 2019). The inactive cells imaged by NanoSIMS were most likely present in the starting inoculum and did not actively grow via the oxidation of the organic matter analog (¹³C-lactate).

Flavins can catalyze As(III) oxidation in oxic and anoxic conditions (Pi et al., 2019), and it was hypothesized that these redox-active molecules may catalyze As(V) reduction in anoxic systems. However, the demonstrated secretion of flavins yet lack of As(III) production in the ARRA3 experiment suggests that bacterially produced flavins were not involved in As(V) reduction under the conditions tested, and instead only played a role in Fe(III) respiration.

The high concentration of flavins measured in the supernatant of S. ANA-3 WT “no donor” control could indicate cell death and lysis. In this control experiment, Fe(II) was undetectable at the end of incubation, confirming that both the electron donor and shuttle are required to complete the transfer of electrons. In natural environments, organic matter typically plays this role, and its availability limits Fe(III) respiration (Shi et al., 2016). Moreover, a percentage of soluble As(V) was reduced to As(III), possibly utilizing the residual intracellular reduced cofactors (for example, NADH, accumulated during biomass production) and indicating that this intracellular respiration mechanism was favored over the extracellular respiration of the insoluble mineral. These results stress the potential role of As(V)-respiring bacteria in As(III) mobilization.

In these experimental conditions, As was released abiotically, suggesting that these experimental conditions supported As(V) desorption, and if new mineral phases were formed, these did not promote complete As(V) sequestration. In the present work, phosphate was added in higher amounts (3 mM) than typically found in natural environments, to promote bacterial growth, and the high P/As ratio also potentially contributed to abiotic As(V) desorption (Biswas et al., 2014). Given the characteristics of the samples, it was not possible to recover and analyze (unaltered) the low amounts of secondary minerals produced (< 30 mg).

The ⁷⁵As/⁵⁶Fe¹⁶O% was a useful estimation to investigate the fate of these ions on the solid mineral surface; however, quantification is not straightforward using NanoSIMS, where different artifacts affect the secondary ion yield (Hoppe et al., 2013). Under these experimental conditions, sorption and desorption phenomena took place, along with potential secondary mineralization, which challenge the quantitative interpretation of these results. For these reasons, we are cautious with this data interpretation and only attempt to explain these observations in relation to the complementary aqueous phase observations. The significantly higher ⁷⁵As/⁵⁶Fe¹⁶O% on mineral areas of ¹³C-incubated S. ANA-3 WT (Figure 5), compared to the results in strain ARRA3 and the control experiments, suggests an active mobilization of both As and Fe from the mineral surface by the WT cells, whereas lower ⁷⁵As/⁵⁶Fe¹⁶O% could suggest preferential As mobilization, driven by Fe(III) reduction and abiotic desorption. This agrees with the higher As_Total and Fe_Total mobilized to the aqueous phase in the ¹³C-incubated WT sample (Figure 1). The lower ⁷⁵As/⁵⁶Fe¹⁶O% observed in samples where only or mostly As(V) was solubilized was previously observed in the local environment of the non-As(V) reducer G. sulfurreducens (Newsome et al., 2018). These results could also indicate an As(III) sequestration–precipitation effect by interaction with Fe(II) and/or organic matter (Kocar et al., 2006; Liu et al., 2020), which were only present in the WT sample, a finding that was similarly observed in incubations with other Shewanella strains (Jiang et al., 2013; Lee and Hur, 2014). This is contrary to the expectation that solubilized As(V) would re-sorb or precipitate more readily on the Fe(III) mineral, where additionally, competitive sorption could take place under these relatively high phosphate conditions (Zhu et al., 2011; Biswas et al., 2014).

Moreover, these NanoSIMS results contrast with the aqueous phase geochemistry results, where both S. ANA-3 strains solubilized As(III) and/or As(V), which remained in solution through the incubation period analyzed. However, the NanoSIMS assessment is limited to the analysis of small regions (1–3 μm²) on the solid mineral, which only describes this phenomenon at the microenvironment of cells and is complementary to what is being observed in the bulk surrounding aqueous environment. Our findings suggest that, overall, the Fe(III)-(oxyhydr)oxide [and any post-reduction secondary Fe(II)-containing minerals] had insufficient sites to support As(III) and/or As(V) re-sorption, which may be explained by the low Fe(III) mineral surface compared to the large volume of aqueous medium, where these experimental conditions were designed to mimic the natural groundwater environment.

The advantages of being able to respire both As(V) and Fe(III) were demonstrated by enhanced As and Fe mobilization in both the aqueous and solid phases in the S. ANA-3 WT compared to ARRA3, implying that As(V)-respirers, which are frequently Fe(III)-respirers as well (Reyes et al., 2008), may play a key role in As mobilization in sedimentary/anoxic environments.

A higher signal of surface Fe and As was detected through NanoSIMS single-cell depth profiles, in contrast to intracellular Fe and As (Figure 6 and Supplementary Figure 8). This suggests preferential Fe and As accumulation at the cell surface, which is likely in a medium with soluble Fe(II) and As(III) (Shiraishi et al., 2018), although NanoSIMS does not discriminate the valence of the elements analyzed. In one study, Shewanella putrefaciens strain CN32 accumulated iron in the form of intracellular granules (mixed Fe valence) when growing with Fe(III) oxyhydroxides as terminal electron acceptors, but this is the only Shewanella strain (and indeed only study) to show these granules (Glasauer et al., 2007). In this work, S. ANA-3 did not contain any structures that resembled intracellular Fe granules, rather, the Fe counts imaged in NanoSIMS were scattered. Thus, the ⁵⁶Fe¹²C and ⁵⁶Fe¹⁶O signals detected on the cell surface of both Fe(III)-respiring strains could have originated from the membrane-bound c-type cytochromes that are active during this electron transfer process. Only the WT strain showed ⁷⁵As on the cells, where As(III) was detected as the predominant aqueous As species. We hypothesize that the ⁷⁵As imaged in the WT could be As(III/V) located at the periplasm, where As(V) respiration takes place. The exact accumulation mechanism of As at the cell surface in the WT is unknown; however, As(III) has a high affinity for thiol groups in proteins and can bind to lipids, carbohydrates, amines, amides and aromatic groups on the cell wall (Shen et al., 2013). This suggests that in reducing conditions, solubilized As(III) will likely bind to more sites than anticipated, including organic/biological surfaces, which may deter its mobilization.

We identified two possible mechanisms for insoluble Fe(III) reduction by S. ANA-3 in our experimental system; most cells were found at a distance from the Fe(III)-(oxyhydr)oxide, with Fe(III) reduction most likely mediated by flavins acting as electron shuttles, whereas a small proportion of cells grew in direct contact with the Fe(III) mineral. Our aqueous geochemistry results showed that As(III) was solubilized throughout the course of the experiment by the WT strain (Supplementary Figure 3A), and complementary early development work showed a similar trend for Fe solubilization measured by ICP-AES (Supplementary Figure 2B) in the ¹³C-incubated samples. This indicates that As(V) respiration did not precede dissimilatory Fe(III) reduction. Moreover, S. ANA-3, like other dissimilatory As(V)-respiring bacteria, has a respiratory As(V) reductase located in the periplasm, and this would make the direct reduction of As(V) sorbed to solid Fe(III)-(oxyhydr)oxides unlikely. Our findings indicate that the majority of S. ANA-3 WT cells did not directly contact the solid phase As(V), that was sorbed to the Fe(III)-(oxyhydr)oxide, and therefore the reduction of As(V) is most likely to have occurred once it was mobilized to the aqueous phase. This was in accordance with data from cultures of a Clostridium strain observed to reduce only soluble As(V) (possibly as a detoxification mechanism) (Langner and Inskeep, 2000) and another study where Alkaliphilus oremlandii OhILAs only respired soluble As(V) (Tian et al., 2015).

Even though the characterization of the spatial and temporal fluctuations of As in impacted ecosystems is challenging, we propose a multistep Fe(III) and As(V) reduction mechanism by S. ANA-3 in the conditions studied; (i) the cells first reduce the insoluble Fe(III)-(oxyhydr)oxide, mainly through flavins used as electron shuttles, (ii) this reduction leads to the solubilization of Fe(II) and desorption of As(V), (iii) once in solution, As(V) traverses the outer membrane (for example, through porins) and is reduced to As(III) in the periplasm and finally, (iv) some As(III) binds to cells (observed as ⁷⁵As in 3D reconstructions of the WT, but not present in ARRA3), although most of it diffuses from the cells and accumulates in the aqueous phase. As(V) respiration led to the release of As(III); however, Fe(III) reduction was a prerequisite for As(V) respiration; thus, Fe(III) reduction was the controlling mechanism of As mobilization under the conditions tested in this work. We propose that this model of Fe(III)-(oxyhydr)oxide reduction followed by As mobilization could explain the mechanism of As release in reducing aquifers, supported, for example, by recent field data (Gnanaprakasam et al., 2017; Mihajlov et al., 2020). The geochemical conditions in As-impacted sites dictate the fate of arsenic. As(III) will likely persist at lower aquifer depths and under reducing conditions, as there are fewer As sorption sites available, for example, due to lower loadings of Fe(III) minerals, unless As(III)-oxidizing bacteria or minerals with high sorption properties are present (Gorny et al., 2015).