A highly reactive precursor in the iron sulfide system

Iron sulfur (Fe–S) phases have been implicated in the emergence of life on early Earth due to their catalytic role in the synthesis of prebiotic molecules. Similarly, Fe–S phases are currently of high interest in the development of green catalysts and energy storage. Here we report the synthesis and structure of a nanoparticulate phase (FeSnano) that is a necessary solid-phase precursor to the conventionally assumed initial precipitate in the iron sulfide system, mackinawite. The structure of FeSnano contains tetrahedral iron, which is compensated by monosulfide and polysulfide sulfur species. These together dramatically affect the stability and enhance the reactivity of FeSnano. Mackinawite is commonly assumed to be the first solid phase in the iron sulfide system. Here, the authors report the existence of a highly reactive nanocrystalline solid phase that is a necessary precursor to the formation of mackinawite.

T he existence of solid precursors or intermediates prior to the formation of stable crystals is increasingly established, although many experimental and analytical challenges to characterize such entities remain 1 . Evidence for such highly reactive intermediate precursors, frequently structurally different from their bulk counterparts, is available for oxic systems; for example, metastable amorphous phases precede the formation of crystalline CaCO 3 polymorphs 2,3 , while nanocrystalline phases are necessary precursors for gypsum formation in the CaSO 4 system 4,5 , and Feoxo Keggin precede the formation of ferrihydrite in the Fe-OH system 6,7 . However, so far, in anoxic systems, and in the lowtemperature Fe-S system in particular, only aqueous Fe-S clusters 8,9 , and a conventionally assumed solid phase with a disordered and nanoparticulate mackinawite-like structure are known [8][9][10][11][12][13] .
Despite increasing evidence of the existence of multiple solid precursors in the Fe-S system 1 , the nature, structure and stability of early formed solid Fe-S precursors with structures different to that of mackinawite have so far not been documented. Such Fe-S phases have been hypothesized as potential membrane catalysts for the formation of prebiotic molecules and life's emergence on early Earth [14][15][16] . Furthermore, such reactive Fe-S phases are of prime interest for the potential green catalytic conversion of atmospheric CO 2 16,17 , and for the development of sustainable, clean, and low-cost energy storage technologies 18,19 .
With this work, we document the existence, identity and structure of a highly reactive nanocrystalline solid Fe-S precursor phase that is structurally different to mackinawite and that is a required precursor to the formation of mackinawite. This phase is, therefore, an important component of all further transformation reactions to more stable phases in the Fe-S system. For example, in anoxic low-temperature environments the formation of pyrite (FeS 2 ) proceeds via this phase through the mackinawite pathway (e.g., mackinawite → greigite (Fe 3 S 4 ) [ ± → marcasite (FeS 2 )] and pyrite (FeS 2 ) [9][10][11][20][21][22] .

Results
Existence of FeS nano precursor. Using two distinct and highly controlled chemical and fully anaerobic approaches (slow titration and a novel diffusion method; see Supplementary Methods) allowed us to quantify all stages in the nucleation, growth, stabilization, and transformations of solid phases in the Fe-S system from aqueous ions to crystalline mackinawite. We show that, contrary to previous studies [8][9][10][11][12] , an additional solid-phase FeS precursor does exist. We have named this phase FeS nano and characterized it through multiple complementary high-resolution microscopic and spectroscopic techniques (Table 1).
Slowly titrating an aqueous Fe 2+ solution with NaHS, or diffusing H 2 S gas into an aqueous Fe 2+ solution and allowing the pH to increase to just below 4.5, led to the formation of faint gray precipitates (Supplementary Figs. 1-3; Supplementary Tables 1, 2) that formed when less than 2% of the total initial ferrous iron was consumed ( Supplementary Fig. 3a). We anaerobically separated, dried and analyzed these precipitates by X-ray diffraction (XRD) and the patterns revealed the presence of a phase with d-spacings of 12.1, 9.3, and 7.6 Å (Table 1 and Fig. 1a). Such d-spacings are entirely different from the Bragg peaks characterizing mackinawite (main d-spacing 5.0 Å; Fig. 1a arrow). Although lattice plane distances of up to 6.7 Å were reported for disordered mackinawite 12,23 , the presence of three distinct dspacings larger than 7 Å in our precipitate-from both the titration and diffusion experiments (Fig. 1a)-clearly documents the existence of a different, new solid Fe-S phase.
FeS nano characteristics and transformation to mackinawite. The relatively weak nature of these large d-spacing XRD peaks shows that the new Fe-S phase is nanoparticulate, a characteristic also confirmed through high-resolution transmission electron microscopy (HR-TEM; Fig. 1b-e). However, a Pseudo-Voigt fitting procedure used on the XRD patterns allowed us to derive crystallite sizes, which suggested the coexistence of crystallites of~23 and~44 nm in size (Supplementary Fig. 4a and Supplementary Discussion). In contrast, the TEM images most often revealed highly monodispersed nanocrystals~2 nm in diameter (Table 1 and Fig. 1b). The larger crystallite sizes derived from XRD peak fitting likely reflect the self-assembly or fusion of individual crystals to bigger agglomerates, as occasionally also observed by TEM ( Supplementary Fig. 4b, c); furthermore, the hump at 10.8 2θ (8.2 Å) in the diffraction pattern from the diffusion reaction ( Fig. 1a) results from short-range order in these larger structures. Although nanoparticle aggregates were also observed in TEM, during this sample preparation the samples were deposited onto the TEM grids from a dilute suspension. In contrast, upon drying the bulk samples for XRD analyses such aggregation was unavoidable.
Fast Fourier transform (FFT) of images of several individual nanoparticles from different replicate experiments corroborated the large d-spacings (~12,~9, and 7 Å; Fig. 1d, e; Supplementary  Table 3) from the XRD, indicating that the thus formed nanoparticles remained the same without further transformation. At the same time, energy dispersive analyses indicated a nonstoichiometric phase only containing iron and sulfur (Fig. 1f).
This new FeS nano phase could be stabilized by keeping the reacting solutions strictly anaerobic and at pH below 4.5. Although our data are from ambient temperatures, such conditions have been inferred as possible locus for Fe-S phase formation in microniches in acidic deep-sea hydrothermal vents 15 . When the pH was increased above 4.5 ( Supplementary  Fig. 3a), our new FeS nano transformed extremely rapidly ( Supplementary Fig. 3a) to mackinawite, as evidenced by the replacement of the large d-spacings characteristic of FeS nano by Bragg peaks smaller than 6 Å, characteristic of mackinawite (e.g., Supplementary Fig. 3c; Supplementary Table 3).
The Raman spectra of FeS nano ( Fig. 2a and Supplementary  Fig. 5) showed two Fe-S asymmetric stretching vibrations at 204 and 215 cm −1 and a symmetric Fe-S stretching vibration at 274 cm −1 (Fig. 2a; Supplementary Table 4). The broadening of the peak at 204 cm −1 indicates a high level of disorder at the local bond scale and a weakening in the interionic bonding caused by an expansion in the lattices 24 .
Combining Raman with X-ray photoelectron spectroscopic (XPS) analyses of FeS nano also revealed the presence of variable proportions of long and short chain polysulfides (Fig. 2a, b, e; Supplementary Table 4) in its structure. The proportions of di-(S 2 2− ) and polysulfides (S n 2− ) in the FeS nano phase contributed 14% and 17%, respectively to the integrated intensity signal, while the largest contribution (up to 62%) was from monosulfide species (S 2− ) ( Fig. 2b; Table 1 and Supplementary Table 6). Polysulfides and Fe III species were previously documented in XPS analyses of crystalline Fe-S minerals (pyrite, pyrrhotite, and troillite), but in most cases their presence was attributed to surface oxidation during sample handling [25][26][27][28][29] . Similarly, polysulfides, Fe II -S and Fe III -S species found in mackinawite have been interpreted as a consequence of weathering 30 .
In our case, the presence of polysulfides on the surface of the FeS nano phase were compensated by a 58:42 ratio in the Fe II -S to Fe III -S species (Fig. 2c, f; Table 1 and Supplementary Table 6) and they did not originate by surface oxidation (Fig. 2d, g). The presence of both Fe III and polysulfides in the FeS nano phase rather than only ferrous iron and monosulfides indicate defects in the initial atomic arrangements, which explain the highly reactive nature of FeS nano , with the polysulfide pool providing a source of S for further redox reactions. Polysulfides are likely formed by electron transfer from monosulfides to the Fe III bonded in the FeS nano structure. Polysulfide species can be accommodated between the Fe and S arrangements of the FeS nano phase, as shown by density functional theory calculations (Supplementary Fig. 8 and text in SM), leading to the large d-spacings documented in the diffraction patterns and TEM images (Fig. 1). Such an expansion caused by the incorporation of similar molecular species has recently been evidenced in iron sulfide green rusts formed by reacting cysteamine with iron oxides 31 . That reaction reduced Fe III from iron oxide nanoparticles to Fe II , yielding a Fe-S layered type material with large d-spacings that were the result of the intercalation of cysteamine molecules in the green rust interlayers. Furthermore, it has even been suggested that precursors of mackinawite involve Fe II Fe III hydroxide species related to the green rust group; however, no experimental work has demonstrated their existence and it is generally assumed that Fe II species lead the formation of the first condensed Fe-S phase 10 .
Local chemistry of Fe in FeS nano . The oxidation states and coordination of Fe in our synthetic FeS nano and in the transformation end-product mackinawite were derived from Fe K-edge X-ray absorption near-edge spectroscopy spectra (Supplementary Figs. 9, 10) and revealed clear differences in the preedge, edge jump and main peak (marked as regions I-III in Fig. 3a, b). The preedge in the FeS nano spectrum was 65% less intense than in mackinawite (i.e., total integrated area in FeS nano = 0.09 and in mackinawite = 0.26; Supplementary Table 9) indicating a different site symmetry. The shift in the centroid to higher energy indicates a slightly higher Fe III content in the FeS nano . The lower intensity and higher energy of the edge jump in the FeS nano spectrum, compared to the mackinawite spectrum, are also consistent with the presence of Fe III in the FeS nano structure.
The local environment of Fe in our FeS nano phase, as analyzed by extended X-Ray absorption fine structure (EXAFS), revealed an average coordination of a first shell containing four S atoms in a tetrahedral coordination at a distance of 2.23 Å and a second shell containing at least two Fe atoms at 4.10 Å (Supplementary Figs. 11-13; Table 1 and Supplementary Table 11). There is a clear difference with mackinawite, in particular in the Fe-Fe bond   11 ).

Discussion
All combined, the above information allows us to derive the formation mechanism for the new FeS nano phase. In solution, the interaction between Fe II atoms and aqueous H 2 S at pH ≤ 4 leads to an atomic arrangement composed of one Fe atom and four S atoms in a tetrahedral coordination. This first atomic arrangement is the earliest stage in the nucleation of a solid and it follows on from the previously proposed aqueous clusters (FeS aq or Fe 2 S 2 ) 13 . The acidic environment promotes the stabilization of both the aqueous Fe-S species 9,32,33 and their transition to the first structure solid Fe-S phase. We clearly demonstrate that this first solid in the Fe-S system is not mackinawite but a new phase, FeS nano . With time, some of the Fe II from the initial Fe-S tetrahedra will lose electrons (oxidation to Fe III ). Such a partial "oxidation" leads to the formation of a layered solid with an unsaturated structure, with any excess of Fe 2+ being released back into the solution. The Fe III in turn will partially oxidize the aqueous H 2 S to form a series of polysulfide species to be accommodated between the Fe-S layers, forming the highly disordered FeS nano phase with large d-spacing and low Fe:S ratios (Table 1). This mechanism can be compared to the sulfidation of mackinawite, troillite and greigite in an O 2 -free environment [34][35][36] , yet in our case it describes the formation of a first solid phase from aqueous ions in the Fe-S system. These results clearly document the pathways of formation and existence of a new solid FeS nano phase, which is the first solid     20,22 . All preparations were performed inside a glovebox (Coy-Laboratory Products Inc.), which was filled with a 5% H 2 and 95% N 2 gas mixture.
The O 2 -free conditions inside the glovebox were ensured through a Pd catalyst that reacted with the 5% H 2 to reduce any oxygen. The dry environment in the glovebox was maintained with silica gel. Full details of the preparation of O 2 -free water is provided in the Supplementary Methods. Once the starting solutions were prepared (100 mM Fe 2+ (aq) and 150 mM NaHS (aq) ), they were used for the titration experiments that were carried out in a Labfors reactor (Infors HT, Switzerland- Supplementary Fig. 1). The reactor system consists of a double-jacketed glass reaction vessel and an overhead unit with port connections for pH-Eh electrodes, inlets and outlets connections for gas and liquids and a built-in stirrer. The addition of reagent solutions was controlled through an automated precision metering pump unit, and the controller recorded pH and Eh. Before the start of an experiment, the Labfors reactor was flushed for 1 h with N 2 (99,9995%) and the headspace of the reactant solutions was maintained anaerobic by a continuous N 2 flow warranting a slight positive pressure of the gas in the reactor. The N 2 gas outlet from the N 2 gas bottle was connected to the reactor inlet through an oxygen trap (LIOT-4, Agilent Technologies) to remove any remaining oxygen in the bottled N 2 . Furthermore, a gas wash bottle containing a 6 M NaOH solution was connected in the gas outlet of the reactor to trap any excess H 2 S gas as an additional safety measure. To sample the solutions, a sampling system was attached to one of the reactor outlets via a T-connector. The system was assembled from flow-through, twoway valves linked to two syringes, one for flushing the system with O 2 -free nitrogen and a second (also N 2 gas preflushed) syringe for removal of a sample ( Supplementary  Fig. 1). All experiments started by anaerobically transferring the light blue Fe 2+ solution into the reactor vessel; this initial Fe 2+ solution had a pH of 4.1. Once transfer was achieved and the overhead N 2 gas flow was set to be constant, the initial Fe 2+ solution was slowly titrated to pH 6-7 through the slow addition (0.47 mL/min) of the also anaerobic NaHS solution. The increase in pH from 4.1 to just below pH 7.0, required almost 9 h of continuous NaHS addition. The pH, Eh, volume of NaHS added, and time were recorded automatically every minute.
In the initial stages of the experimental work and to find optimal conditions in terms of time and pH ramp conditions, we carried out a series of test titration experiments with much higher concentrations of NaHS (1000, 500, and 150 mM; Supplementary Table 1). From these tests we determined that the optimal experimental parameters were slow titration with 150 mM NaHS. Thereafter, titration experiments with this concentration of NaHS were repeated three times to demonstrate that the data are highly reproducible in terms of time, pH ramp and added volume of NaHS. The continuously recorded pH profile allowed us to quantify exactly the volume of NaHS needed to reach certain pH conditions and thus to choose the most adequate sampling times. Samples were removed from the reactor at different pH values (defined by NaHS addition and time) through the double syringe sampling system. Once in the sample syringe, the samples were transferred immediately into an inflatable glove bag filled with oxygen-filtered N 2 gas for transfer back into the CoyLaboratory glovebox. Once in the glovebox, samples were filtered using a vacuum filtration kit and 0.02 µm pore size polycarbonate membranes to retain the solids.
Diffusion experiments. All diffusion experiments were performed inside of the CoyLaboratory glovebox. Although the Fe 2+ solutions were prepared the same way as for the titration experiments (Mohr's salt [(NH 4 ) 2 Fe(SO 4 ) 2 ·6H 2 O]), the source of sulfide for the diffusion experiments was H 2 S gas that was generated by reacting HCl with solid Na 2 S·9H 2 O crystals (Sigma Aldrich 99.999%). For each diffusion experiment, 25 mL of a freshly prepared Fe 2+ solution were transferred inside the glovebox into a 100 mL Schott Bottle (Duran ® ). In another Schott bottle,~1 g of Na 2 S·9H 2 O was mixed with 10 mL of 6 M HCl to generate H 2 S gas. This gas was allowed to diffuse and be transferred into the bottle containing the Fe 2+ solution through gas-tight tubing and thus increase the pH of the iron solution to~4.3. The diffusion was allowed to proceed for 1 h. A 6 M NaOH solution was used as a trap to capture any excess of H 2 S in the reactor (Supplementary Fig. 2). Twelve diffusion experiments (Supplementary Table 2) were performed and the end pH of the mixed solutions was measured with a calibrated pH meter after 1 h of reaction. At the end of each diffusion experiment the solid precipitates were separated from the solution using a vacuum filtration kit and 0.02 µm pore size polycarbonate membranes. The filtrated solution was kept for total iron analysis by inductively coupled plasma optical emission spectrometry (ICP-OES; see below). The diffusion experiments were always performed fresh and the immediately separated solids were re-suspended in O 2 -free ethanol, deposited on an analysis holder or substrate for analysis (see below) and left to dry inside the glovebox. These solid samples were analyzed with a variety of solid analyses methods, which each had special sample handling requirements (see details below and Supplementary Table 2). For each of these analyses, samples were removed from the glovebox on their respective substrates, but in sealed, double-jacketed containers that were placed inside of three layers of oxygen-free N 2 gas-filled and sealed bags.
Solution characterization. Total iron concentrations were analyzed in the supernatant solutions after separation of the solid phases after immediately diluting and acidifying with 1% HNO 3 . Iron measurements were carried out by ICP-OES (Thermo Fisher iCAP 7400 Radial) with a calibration between 1 and 100 ppm. The quality of the measurements was controlled through duplicate analyses of the samples and analysis of QC check-standards before and after the analysis.
Solid characterization methods. For all solid analyses all sample handling and mounting steps were done inside the anaerobic chamber. Once mounted, all samples were sealed in special holders or containers for analyses or transport to analytical instruments.
Powder XRD. For XRD analysis, the as-synthesized anaerobic solid samples were filtered and redispersed in degassed ethanol and then mounted onto a flat silicon crystal inside an airtight Bruker XRD holder. This procedure was performed at all times in the anaerobic chamber under a strict and controlled oxygen-free environment. The airtight holder was placed inside an airtight container and was than inserted into another airtight container for transport to, and mounting into, the sample holder of the XRD instrument. The thus mounted samples were analyzed with a Bruker D8 diffractometer using a scan range between 2°and 70°2θ and at a scan rate of 0.05°2θ/min. The XRD patterns were collected over 18-24 h in order to resolve even small and low angle peaks; during measurements changes in the background sometimes occurred (i.e., between 10°and 13°2θ), yet the data is presented without further processing.
High-resolution transmission electron microscopy. Aliquots (either neat, diluted or redispersed in ethanol) of the solid samples were deposited onto holey carbon TEM grids (Agar Scientific), which were transferred into a special anaerobic transfer holder (Gatan 648 Double tilt). This allowed the safe anaerobic transfer of the highly reactive samples to the TEM instrumentation. All samples were freshly synthesized just immediately prior to TEM analyses. The effect of the dilution with ethanol was tested and compared to a nonethanol diluted sample and no changes were observed in the resulting images or spectral and diffraction information.
HR-TEM imaging, selected area electron diffraction and energy dispersive X-ray (EDX) spot analysis were acquired using a Field-Emission-Gun-Transmission-Electron microscope (FEI Tecnai TF20) fitted with a CCD Camera (Gatan Orius SC600A) and an EDX spectrometer (Oxford Instruments 80 mm 2 X-max). The microscope was operated at 200 kV. To avoid beam damage, the exposure for image collection was limited to maximum 0.25 s. Images were analyzed using the ImageJ software 37 . Atomic plane distances were calculated from the fast FFT analyses of several particles from the each acquired high-resolution image and from different areas in each image. The lattice planar distances were then calculated from the reciprocal values of the distance from the spots to the center (Supplementary Table 3).
Raman spectroscopy. The dried samples were transferred into a specially designed Raman analysis sample holder composed of two thin microscope cover slips, in between which the sample was sealed with silicone grease. The samples were freshly prepared and once in the sealed, airtight holder they were transferred from the NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-05493-x ARTICLE NATURE COMMUNICATIONS | (2018) 9:3125 | DOI: 10.1038/s41467-018-05493-x | www.nature.com/naturecommunications anaerobic chamber to the Raman instrument and stored until analysis in sealed airtight jars. Raman analyses were performed using a Horiba LabRam HR 800 microscope operated with lasers at excitations of 633 and 473 nm and using a 50× magnification objective, with gratings of 600 or 1800 ln/mm to give a resolution better than 1 cm −1 . The instrument was calibrated using a silicon standard (main Raman shift of 520.7 cm −1 ) and Raman signals were acquired via a 1024 CCD detector. The instrument was operated in confocal mode using a 150-300 s detector exposure time and 20-30 spectral accumulations.
Several trials to determine the best combination of analysis parameters (e.g., laser wavelength, laser power, exposure time, number of accumulations, and background) were carried out prior to collection of the actual spectra; during these tests spectra were collected on different spots of various specially synthesized iron sulfide test samples. In addition, on a few samples a sequence of spectra was collected over time to evaluate any effects of the laser beam on the samples. From these tests the optimal conditions for analyses (473 nm laser using between 0.1% and 1% intensity~1-10 mW) were determined. Using these parameters, we collected high-resolution spectra on all samples at wavelengths between 100 and 800 cm −1 . For each sample, several spots were analyzed by collecting 30 spectra with 150 s of exposure each. Once the spectral stack was collected, the data were evaluated to assure that no changes were induced during repeat analyses, after which the spectra were averaged.
All spectra were analyzed using the LabSpec 6 software and peaks were fitted using a combination of Lorentzian and Gaussian models. The goodness-of-fit was evaluated using the reduced χ 2 . The positions of the Raman vibrations were derived from the fits. Band positions and identities were assigned and compare with literature data (Supplementary Tables 4, 5).
X-ray photoelectron spectroscopy. Samples diluted in ethanol were deposited on gold substrates and then transferred to the XPS instrument inside double-walled plastic bags that were inside airtight jars. To mount the samples into the XPS instrument, a disposable glove bag was installed over the XPS sample interlock system. This glove bag was purged with O 2 -free argon several times. The samples inside the airtight jar were transferred into the instrument, which was immediately evacuated to ultra-high vacuum (~10 −7 Pa) and samples were kept under vacuum overnight prior to analyses. During analyses the vacuum was maintained between 10 −7 and 10 −9 Pa. XPS spectra were collected using a Thermo Escalab 250 XPS instrument equipped with a monochromatic AlK α X-ray source (75-150 W). Calibration of the binding energies was performed using the carbon 1 s peak at 285 eV. Survey scans were collected between 200 and 800 eV with a pass energy of 160 eV. Depth of interaction was between 5 and 7 nm. The spot size was 500 μm and the analyses were done with a power of 150 W. High-resolution spectra were collected at the binding energies of Fe (700-740 eV), S (160-170 eV), O (525-545 eV) and C (280-295 eV), with a pass energy of 20 eV and a step size of 0.1 eV. The process was repeated after argon etching, 5 times each for 1 min, removing about 3-5 nm from the surface of the already analyzed samples.
The data were processed with the CasaXPS software and the background subtraction was done via the Shirley method, while fitting of the spectra was performed according to standard methods 38,39 . The high-resolution Fe2p spectra were analyzed for the 2p 3/2 envelope and fitted using a single peak for low spin Fe II species as in previous works 26,40 . (Supplementary Table 6). For Fe III , multiplet splitting peaks were considered as predicted by the crystal field theory 41 . To build a model, several constrains were imposed (i.e., area of the peak, full width at half maximum (FWHM) and peak position) and the fitting did not rely on the combination of individual spectra as a fingerprint, but on the contribution of peaks with certain constrains that are characteristic for each specific species. The area under the curve (with a suitable baseline applied) was fitted with the minimum number of components to ensure best possible fit. The components were constrained so that the FWHM was maintained, whilst allowing free fitting of the peak area and position to ensure a good but scientifically viable fit. Where components were used to account for different bonding regimes, these were taken from the literature. Although this approach only provided initial values for peak positions, it still allowed for optimization of the peak areas through the fitting algorithm.
The FWHM was set at 1.8 eV for FeS nano and 1.0 eV for mackinawite. A satellite peak was included to fit the mackinawite spectrum. The S2p spectra were fitted using the 2p 1/2 and 2p 3/2 doublets, the first one was set to be half the area of the second (Supplementary Table 6). The fits were compared with data from the literature for different iron sulfides (Supplementary Tables 7 and 8). All spectra were corrected to allow for slight variations, using the C-C peak at 285 eV.
X-ray absorption spectroscopy. Fe K-edge X-ray absorption spectra (XAS) were collected at station I18 at the UK Diamond Light Source using a Si(111) monochromator with an energy resolution (ΔE/E) of 1.4 × 10 -4 . The monochromator was calibrated using a Fe-foil at 7111.99 eV. Samples of the synthesized FeS nano and mackinawite phases were sealed inside the anaerobic chamber in between double sticky Kapton/mylar layers (with a total thickness of <200 µm) and transported anaerobically to the station in jars filled with O 2 -free nitrogen. The particle size of the synthetic mackinawite used as a reference was~5 nm as it was formed from the transformation of nanoparticulate FeS nano and it was aged only for short periods of time. The mackinawite spectra were collected between 6911 and 7580 eV, and the FeS nano spectra between 6911 and 7400 eV. Both measurements were performed using a step size of~0.5 eV. We also collected XAS transmission spectra from wustite, hematite and magnetite for reference.
The raw data were aligned, averaged, normalized and background-subtracted using the Athena software package 42 . The X-ray absorption near-edge structure (XANES) region of the Fe K-edge XAS spectra exhibit a characteristic preedge feature between 7109 and 7116 eV, composed of two overlapping peaks whose centroid depends on the contributions from Fe II and Fe III present 43,44 . Comparison with literature values revealed that our calibration was~0.9 eV higher than Wilke et al. 2001 43 .
Data from the EXAFS region were analyzed with the Artemis software package 42 . To test for Fe in tetrahedral and octahedral coordination in the FeS nano phase, theoretical models based on the structures of greigite and mackinawite were generated using the FEFF6 and ATOMS software packages 45 . Single scattering paths were considered in the fits (Fe-S, first shell and Fe-Fe, second shell) based on the greigite structure. However, only the Fe-S path could be considered for the mackinawite data as there was no contribution of the Fe-Fe path to the data. In the fits, the amplitude parameter (S 0 2 ) was fixed to 0.80 for both paths, which was derived from the synthetic mackinawite standard. The energy shift (ΔE 0 ) was constrained to be the same for the two paths and it was set as a fitting parameter, as well as the disorder parameter in the distribution of interatomic distances (σ 2 ). Due to the small size of the particles, the coordination numbers (N) are not expected to be round figures and thus the coordination numbers could also be set as fitting parameters. To construct the fits, independent variables (N var ) were at least half of the independent points (N ind ). The data were then fitted through a least squares approach in the k-space between 2.8 and 6.5 Å −1 using multiple k-weights. The quality of the fits was assessed with the reduced χ 2 and the R-factor 46 . Coordination numbers (N), bond distances (R), Debye-Waller factors (σ 2 ) and inner potential corrections (ΔE) were extracted from the fits.
Data availability. The authors declare that data supporting the findings of this study are available within the paper and in the supplementary information file or are available from the corresponding author upon reasonable request.