Natural cobalt–manganese oxide nanoparticles: speciation, detection and implications for cobalt cycling

Sample collection

We collected samples from six localities comprising four separate deposit types in a July 2022 field campaign (see Fig. 1 and Table 1 for sample locations and an image of each site). During sampling, the weather was dry and sunny (daytime maxima 20–27°C) at the northern sampling sites (NT and Qld), and dry but cooler and partly cloudy (daytime maxima 15–20°C) at the southern sampling sites (SA and NSW). The localities are:

Fig. 1.

Map of sampling locations including images of the arid to semi-arid landscapes.

No sites had any nearby water (neither lakes nor permanent streams) at the time of sampling. Samples were collected using sterile equipment under aseptic conditions, using nitrile gloves at all times. Sieves and trowels were rinsed with methylated spirits between each sampling site, and soils were collected from a depth of 10–20 cm, sieved through a 2-mm plastic mesh and collected into pre-sterilised vials. Full descriptions of the development of these sampling methods are provided in Fairbrother et al. (2013) and Reith et al. (2010, 2012).

Extraction of Co- and Mn-based nanoparticles from soil

For SP ICP-MS analysis, NPs were extracted from soils and suspended in an aqueous phase. A protocol for NP extractions was adopted from Gao et al. (2020). Briefly, 0.5 g of soil sample were accurately weighed, placed in a 50-mL polypropylene centrifuge tube and mixed with 50 mL of 2.5-mM tetrasodium pyrophosphate (TSPP, prepared in ultra-pure water with a final pH of 9.6). After shaking in a constant temperature water bath for 30 min (25°C) and following 30 min of sonication, samples were settled at room temperature for 16–18 h in the dark, allowing larger soil particles to precipitate. Supernatants containing extracted NPs were subsequently filtered through 0.45-µm syringe filters (Nylon Filter, Watman, GE Healthcare Life Sciences, UK) and diluted 1:100 with ultrapure water prior to SP ICP-MS analysis.

Single particle inductively coupled plasma–mass spectrometry (SP ICP-MS)

SP ICP-MS was used to quantitatively detect single Co-based and Mn-based NPs in the soil sample extracts and to study their characteristics on a single particle basis. An 8900 series ICP-MS system (Agilent Technologies, Santa Clara, CA, USA) was equipped with platinum cones and x-lenses, and operated with MassHunter software (Agilent Technologies) in ‘single particle’ mode (quadrupole dwell time 0.1 ms). A Scott-type double-pass spray chamber (Glass Expansion, Melbourne, Vic., Australia) was cooled to 2°C and a MicroMist concentric nebuliser (Elemental Scientific, Omaha, NE, USA) was used for sample nebulisation. For the analysis and calibration of SP data sets, a previously developed open-source Python-based data processing platform ‘SPCal’ (ver. 1.1.2, see https://github.com/djdt/spcal/releases/tag/v1.1.2) was used (Lockwood et al. 2021). Size values were calculated assuming particles are spherical with an average spinel-like formula (CoMn₂O₄, density 5.09 g cm^–3; Habjanič et al. 2014). Ionic response factors were calculated based on a 10 ng g^–1 ICP-MS standard. Transport efficiencies were calculated by the ‘size method’ (Lockwood et al. 2021) analysing 100-nm-sized Au NPs (nanoComposix, San Diego, CA, USA, in 0.2-mM citrate). SP ICP-MS results are reported by site in Table 2, with full results in Supplementary Table S1. Representative size distributions are shown in Fig. 2.

Fig. 2.

Size distributions of Co-based NPs for samples (a) Queen Sally sample QS1 and (b) Copperado sample Copp5. Sizes were estimated using mass fractions and densities of CoMn₂O₄.

SP ICP-ToF-MS

To investigate the Co:Mn composition of individual particles, SP ICP-ToF-MS was used. A Vitesse ICP-ToF-MS system from Nu instruments (Wrexham, UK, see https://www.nu‐ins.com/products/hr‐mc‐icp‐ms/vitesse) was employed in single particle mode recording mass spectra from 42 to 250 amu at 12.85 kHz (corresponding to a spectra saving interval of ~80 µs). Data were recorded using NU CODAQ VITESSE software (ver. 1.5.8267.1) and processed by a modified version of SPCal (ver. 1.1.2) from Lockwood et al. (2021). The ICP-ToF-MS instrument was equipped with a concentric nebuliser and its transport efficiency was determined to be 0.8%, obtained by analysing a diluted 100-nm Au NP standard from nanoComposix (San Diego, CA, USA, in 2-mM citrate buffer) and a 10 ng g^–1 ionic standard while determining the uptake flow rate. To distinguish background and noise from SP signals, Compound Poisson statistics (α = 10⁻⁶) were applied while using a log-normal approximation method as discussed in a recent study (Gonzalez de Vega et al. 2023). Elements with at least 200 data points per million data points above the particle decision limit were selected for a correlative analysis of coinciding signals and for further calibrations. Here, contiguous signal regions above the signal mean with at least one point above the decision limit were accumulated and calibrated using respective ionic standards diluted to 10 ng g^–1. To isolate particle events associated with Co and Mn, a compositional filter was used to limit the analysis solely to Co- and Mn-containing particles. Following calibration, molar ratios of single particles were evaluated in a scatter plot. A linear fit was used and reported as an average Co:Mn molar ratio. Data not containing values for Co:Mn ratios did not contain both Co and Mn with masses over the mass detection limits, which ranged between 20 and 80 ag depending on the ionic background values of sample extracts. Particle compositions were determined using hierarchical agglomerative clustering of normalised elemental mass fraction, with a Euclidean distance cutoff of 20% (Tharaud et al. 2022). This rather large distance threshold was chosen to determine the overall mean Co:Mn molar ratios in single particles rather than cluster numbers and their individual composition.

Synchrotron X-ray analyses

Regolith samples were prepared for analysis by double impregnation in epoxy resin, following the method described in Missen et al. (2022a) and briefly re-described here. These Co-rich regolith samples are deposited in the Museums Victoria Petrology collection (sample number E20470). First, ~0.5 g of resin was added to a 0.7-mL vial (top diameter ~7 mm) of the sample and allowed to cure. Subsequently, groups of six vials were placed upside down in a 2.5-cm mould and ~3.0 g of resin was again added to encase these six vials into a 2.5-cm diameter puck. A small rotary saw was used to cut the pucks to a height of ~0.6 cm, then polishing and grinding of the puck to a surface fineness of 1 μm was performed using oil-based lubricants to ensure that no clays or water-soluble phases dissolved during the polishing. The synchrotron analyses were undertaken at the ID21 beamline of the European Synchrotron Radiation Facility (ESRF), France with data preserved online (Missen et al. 2022b). Data were collected at the Co K-edge (7709 eV).

XFM (X-ray fluorescence microscopy) and XANES (X-ray absorption near-edge spectroscopy) measurements were performed using the scanning X-ray microscope at beamline ID21, ESRF. XFM mapping was performed at 7900 eV with a dwell time of 0.1 s using a 0.3-μm vertical × 0.7-μm horizontal Kirkpatrick–Baez focused beam with an average flux of 1.2 × 10¹⁰ photons s^–1, the energy selection was carried out by means of a Si (111) double crystal monochromator. XRF (X-ray fluorescence) emission spectra were obtained with a silicon drift detector (SDD) of area 80 mm² of active area (SGX Sensortech, RaySpec Ltd, UK). The XRF maps (Fig. 3) were processed using an open source code (see https://gitlab.com/EduardoV/xas_tools) combined with PyMCA software (ver. 5.9.2, see https://www.silx.org/doc/PyMca/dev/index.html; Solé et al. 2007) to fit the element emission lines, correct detector dead time and normalisation by incoming beam intensity. From the Co distribution in the sample, points of interest were selected for Co K-edge XANES scans using a 7709-eV reference energy in fluorescence yield (FY), beginning 10 eV before and finishing 140 eV after the reference energy. A step size of 0.5 eV and a dwell time of 0.1 s was used for collecting XANES scans, and scans were generally completed in triplicate. XRF and X-ray absorption spectroscopy (XAS) measurements were performed at the ESRF, France (Proposal number EV-506; Missen et al. 2022b).

Fig. 3.

XFM maps presented in RGB (red–green–blue) format showing distribution of Co (red) relative to iron (green) and Mn (blue). The characteristic association of Co and Mn is shown by the purple overlap colour of red (Co) and blue (Mn). Pixel size for the maps is 200 × 200 nm². (a) Queen Sally (subsample QS1), one of the few samples with cobalt oxide NPs separate from Mn. (b) Mount Cobalt (subsample MC4). (c) Pyrite Hill (subsample PHill2). (d) Copperado (subsample Copp4).

The following materials were utilised for calibration of the beamline and for comparison with analytical samples: Co metal foil (Co⁰), cobalt(II) oxide (Co^IIO), cobalt(II,III) oxide (Co₃O₄ i.e. Co^IICo^III₂O₄), a sample labelled heterogenite (Museums Victoria Mineralogy collection, sample number M42912), which further analysis showed contained some Co^II as well as Co^III, and Co^III-bearing Mn oxide from Copperado. Principal component analysis (PCA) was applied on all μXANES data, performed with Orange software (ver. 3.36.2, see https://orangedatamining.com/; Demšar et al. 2013), including the add-on ‘Spectroscopy’ (Toplak et al. 2017). PCA was implemented on the second derivative of the spectra (using the Savitzky–Golay method, second polynomial order and 15 points window) and a vector normalisation. The spectra from clusters observed in the PCA plots were grouped using the k-means method and subsequently normalised following the XAS normalisation widget in Orange software (Spectroscopy add-on).

Bulk regolith analysis

Determination of bulk Co and Mn contents in the regolith was performed by ALS Geochemistry. Samples (previously sieved through a 2-mm plastic mesh as described above) were pulverised until 85% of the sample passed through a 75-μm filter, then analysed by a method employing an aqua regia digestion followed by ICP-MS analysis (ALS code ME-MS41L). Co and Mn results for each site are reported in Table 2 and were used to determine the ratio of nanoparticulate Co and Mn to Co and Mn in the <2-mm regolith fraction, with full results by sampling location in Supplementary Table S1.

Cobalt- and manganese-based nanoparticles

Both Co- and Mn-containing NPs were extracted and detected in all 28 soil samples using SP ICP-MS, with a summary of results shown in Table 2. Two different SP ICP-MS methods were used. First, a quadrupole-based SP ICP-MS set-up was employed, which is characterised by its high detection power and therefore its ability to detect even small Co or Mn particulate masses. However, a disadvantage of quadrupole-based SP ICP-MS is that Co and Mn could not be analysed simultaneously. Therefore, SP ICP-ToF-MS was used in a subsequent step enabling the simultaneous acquisition of almost all element signals in a single particle. Despite lower detection power, this enabled the determination of compositions on a single particle level. As a focus of this study, molar ratios between and Co and Mn were determined, and values are listed in Table 2 (ratios for each sampling site are in Supplementary Table S1). However, in a non-target approach, other elements associated with Co were investigated qualitatively. Samples without reported mass composition either had no particles containing both Mn and Co, or particles were too small to be detected in SP ICP-ToF-MS, which had approximately one order of magnitude higher mass detection limits compared to quadrupole-based SP ICP-MS (size detection limit, sDL, for Mn and Co were 31.6 and 27.7 nm respectively). Owing to the characteristics of quadrupole and ToF-based SP ICP-MS, the former was used to determine mass and size distributions, whereas the latter was employed for compositional analyses. In the present study, we found evidence that a majority of Co and Mn is present in a CoMn₂O₄ phase, which was therefore selected as proxy to estimate size distributions. However, particulate mass distributions of elements can be determined without knowledge of phases allowing a less biased view (Table 2). The three highest concentrations of extractable Co-based NPs were found in three of the four weathering sulfarsenide Queen Sally samples (1262 ± 21, 1390 ± 41 and 1236 ± 30 ng NP Co g^–1 regolith, subsequently written as nanograms of Co per gram for simplicity), with an average number of 2.19 ± 0.04 × 10⁸ NPs g^–1. Only sample QS4 had a lower concentration (71 ± 1 ng Co g^–1), which was likely due to sampling further from the deposit rather than from the actively weathering zone. The two samples with the lowest average number of Co NPs were found in sampling sites at siderite-hosted Young’s Cobalt prospect (0.69 ± 0.08 and 0.94 ± 0.18 ng Co g^–1, average number of NPs (1.10 ± 0.09) × 10⁷ NPs g^–1). Other samples with high average concentrations of Co-containing NPs included those from the other weathering sulfarsenide deposit Mount Cobalt (average 229 ± 7 ng Co g^–1 from (2.40 ± 0.04) × 10⁸ NPs g^–1, n = 7, varying between 30 ± 1 and 867 ± 25 ng Co g^–1) and lateritic Copperado (average 36 ± 1 ng Co g^–1 from (1.28 ± 0.03) × 10⁸ NPs g^–1 with n = 5, dominated by 129 ± 3 ng Co g^–1 in sample Copp4, with an average of 12.9 ± 0.4 ng Co g^–1 in the other four samples). Size distributions of the Co-based NPs typically showed an average size of 55 ± 38 nm, with significant variation in mean sizes with 110 nm for QS1 and 35 nm for Copp5 (Fig. 2).

Owing to the prevalent association of Co with Mn (see Synchrotron: NP agglomerates, Co oxidation state and elemental associations below), we also analysed the concentration of Mn-based NPs in the regolith samples using SP ICP-MS. The concentration of Mn-based NPs was highest at the Calvert lateritic Mn–Co oxide prospect, averaging 358 ± 14 ng NP Mn g^–1 regolith, subsequently written as nanograms of Mn per gram (highest value 486 ± 24 ng Mn g^–1 at sample Calv1) from an average of (1.62 ± 0.05) × 10⁸ NPs g^–1. The other Mn–Co prospect Copperado had the next highest nanoparticulate Mn contents averaging 49 ± 2 ng Mn g^–1 from (8.1 ± 0.4) × 10⁷ NPs g^–1. Sample Copp4 had the highest NP Mn content (111 ± 6 ng Mn g^–1) of the 5 Copperado sites as well as having the highest NP Co content. Across the other four sites, the average concentrations of Mn-based NPs were consistently between 9 and 15 ng Mn g^–1, with 9.5 ± 0.5 ng Mn g^–1 at Queen Sally, 10.8 ± 0.6 ng Mn g^–1 at Mount Cobalt, 14.8 ± 0.7 ng Mn g^–1 at Young’s Cobalt prospect and 10.9 ± 0.7 ng Mn g^–1 at Pyrite Hill. The minimum value across these sites is 4.3 ± 0.3 ng Mn g^–1 at sample MC3 and maximum value was 30 ± 1 ng Mn g^–1 at nearby site MC5.

Using SP ICP-ToF-MS enabled the study of particle compositions. Besides the calibration of the Co:Mn molar ratio, other elements associated with Co on a single particle basis were investigated. Fig. 4a, b shows the transient raw data of SP ICP-ToF-MS and demonstrates the coincidental detection of Co (yellow) with different elements including Ti (blue), Mn (red), Fe (green) and As (cyan) on a single particle level. Fig. 4c shows the composition of Co-containing NPs found in sample PHill4 as a representative sample (weathering pyrite environment). It is evident that three different groups were identified. The largest fraction (61.2%) contained Mn and Co at an approximate ratio of 3:1 as plotted in Fig. 4d. It was determined that 22.4% of the NPs contained Ti as well as Co and Mn (approximate Ti:Co:Mn ratio 5:2:1) and 16.4% contained mainly Ti and Fe (with minor Co and Mn) in individual particles. The compositional analysis of Co-NPs found that most particles were associated with Mn. However, the relative fraction of Mn to Co varied across different sampling locations. Fig. 4e shows representative sample MC6 (weathering sulfarsenide), for which the highest molar fraction of Co relative to Mn was contained in individual particles (84.7%). Furthermore, respectively 6.3 and 5.5% additionally contained Fe or Ti, and 3.5% of particles consisted of As and Co.

Fig. 4.

(a, b) Transient raw data of SP ICP-ToF-MS. (c) Molar composition of three groups of Co-containing NPs found in sample PHill4 as representative sample. (d) Distribution of Mn and Co in the 61.2% of Pyrite Hill sample PHill4 Mn–Co oxide NPs at an approximate ratio of 3:1 Mn:Co. (e) Representative Mount Cobalt sample MC6, for which significantly more Co relative to Mn was contained in individual NPs.

Synchrotron: NP agglomerates, Co oxidation state and elemental associations

Agglomerates (in the physical, not statistical sense) of Co-based NPs were at the most 20 μm in cross-sectional breadth, but were generally <1 μm in size (Fig. 3). Using XANES, the most prevalent Co oxidation state across all samples was a mixture of +II and +III (Fig. 5 and 6), although the +III oxidation state was generally more dominant. Several samples showed distinct clusters of spectra in PCA figures (Fig. 5; clusters determined mathematically by the k-means method) when plotting the variation of sample spectra to the analytical standards. Detailed analysis of our synthetic and natural compounds led to the following choices for the standards used as references for the k-means method: cobalt(II) oxide for Co^II, the heterogenite sample for mixed Co^II/Co^III and a rock crust of Co^III-bearing Mn oxide from Copperado for Co^III. Heterogenite and lithium cobalt(III) oxide (LiCo^IIIO₂) were intended as the initial standards for Co^III but were shown to contain minor traces of Co^II based on their XANES spectra, thus the most oxidised sample we analysed (Co^III-bearing Mn oxide from Copperado) was chosen as the Co^III standard. The case of heterogenite may be that the heterogenite sample contains minor Co^II with a potential formula of HCo^II_xCo^III_1−xO_2−x÷2 (x likely <0.2) or more likely that a Co^II bearing mineral such as guite (Co^IICo^III₃O₄; Lei et al. 2022) was intermingled with an especially heterogeneous heterogenite sample.

Fig. 5.

PCA comparisons for sampling sites using the k-means method to select three groups (Co^II, Co^II–Co^III and Co^III character) showing (a) Queen Sally sample QS1, (b) Pyrite Hill sample PHill4, (c) Young’s Cobalt prospect sample YCP1 and (d) Copperado sample Copp4 as representative samples. Standards included to generate the PCA figures are Co^IIO (right hand side of PCA figures, circles), natural heterogenite (central area, cross symbols), Co^III-bearing Mn oxide (left hand side, plus symbols) and finally the sample itself (triangles) in the PCA maps for comparison.

Fig. 6.

First derivative of XANES spectra (grouped in PCA using Orange software then averaged) for the samples (a) Queen Sally sample QS1, (b) Pyrite Hill sample PHill4, (c) Young’s Cobalt prospect sample YCP1 and (d) Copperado sample Copp4.

Clustering showed that samples may contain groups of Co NPs with 1, 2 or 3 clusters of oxidation states represented (Fig. 6). Localities such as PHill4 (all mixed oxidation state, Fig. 5b, 6b) and Copp4 (all Co^III, Fig. 5d, 6d) contained Co-based NPs with just a single oxidation state group represented. Young’s Cobalt prospect sample YCP1 contained distinct clusters of Co^II and Co^III-based NPs but no mixed oxidation state NPs (Fig. 5c, 6c). Other sampling localities (e.g. QS1 and MC1 from weathering sulfarsenide localities; QS1 shown in Fig. 5a, 6a) showed the presence of all of Co^II, Co^III and mixtures of the two oxidation states in the regolith. These observations suggest that Co-based NPs form with different mineral identities (e.g. heterogenite NPs with mainly Co^III, spinel-based NPs with both Co^II and Co^III, Co^II arsenates or oxides with Co^II).

XFM results showed that the major association of Co in the regolith samples was not with Fe, despite its prevalence in all samples. Instead, in most analysed localities, Co and Mn were observed in close association, with the Co and Mn signal in the XRF maps typically overlapping in all samples except some parts of the Co-rich Queensland sulfarsenide localities. No association with S was observed in these supergene samples. This observation led to the hypothesis, subsequently verified by SP ICP-ToF-MS, that the NPs detected by SP ICP-MS are not formed of individual Co- (or Mn-) oxide NPs but comprise both Co and Mn in Co–Mn oxide NPs.

Cobalt cycling in terrestrial environments

In Fig. 7, we present a diagram of cobalt cycling in surface environments with a focus on Co(–Mn) nanoparticulate phases. Co cycling in surface environments begins with dissolution of solid Co-bearing phases to form soluble Co^II (Fig. 7, Process 1). Sulfides and sulfarsenides are typical primary hosts of cobalt (including examples with both cobalt as a trace element and as a mineral-defining species) along with Co hosted by primary oxides and carbonates (Co within lateritic deposits may be considered as forming later in this cycle). After dissolution, several processes can take place. Formation of secondary oxysalt minerals (Fig. 7, Process 2) may be followed by redissolution back to Co^II (Fig. 7, Process 6), however formation of Co^III oxide minerals is less likely to be followed by redissolution due to the higher stability of these phases – for instance, heterogenite does not dissolve unless the Co^III is reduced back to Co^II (Dehaine et al. 2021).

Fig. 7.

Schematic of cobalt-based nanoparticles in the environment.

The most important insight obtained by SP ICP-MS and synchrotron XFM was the formation of Co(–Mn) oxide NPs (Fig. 7, Process 5). These Co-containing NPs may form from physical weathering of Co-bearing Fe- or Mn-oxides (Fig. 7, Process 7) or may be mediated by a microbial process. Co(–Mn) oxide NPs may re-dissolve in the presence of rain, surface and ground waters, but the main two processes expected to be key to NP mobility are sorption (Fig. 7, Process 8) to either organoclays or to Fe-, Mn- or Co-oxide minerals. Redox drivers for the dynamic Fe (Kappler et al. 2021) and Mn (Li et al. 2021) cycles are likely to govern Co mobility due to their general predominance over Co in most surface environments, although the availability and behaviour of Co oxidation states is different to that of Mn and Fe. Desorption (Fig. 7, Process 9) may release the Co(–Mn) oxide NPs back to being free NPs, whereas plant-based uptake of Co (Fig. 7, Process 3) may occur directly from soluble Co^II in pore water and from Co(–Mn) oxide NPs (either free or sorbed on organoclays). From plants, Co can be incorporated into other biota where a range of interactions may take place. Although these interactions are beyond the scope of this paper, they are likely important for the cycling of Co through living organisms and back into the environment (e.g. decomposition of plants enriched in Co would result in Co being returned to the environment: Fig. 7, Process 10).

The extent to which microbes are involved in mediating cobalt cycling processes is not yet known. Iravani and Varma (2020) showed that differing conditions for Co-(hydr)oxide NP formation leads to different crystallinity and morphology of the products, processes that are complicated further in the natural environment by the presence of other elements such as Fe and Mn. Microbacterium sp. MRS-1 has been shown to produce Co₃O₄ from soluble Co^II cations (i.e. laboratory evidence for Fig. 7, Process 5) (Sundararaju et al. 2020). Although Co-oxide based NPs are likely less toxic to most microbes than soluble Co^II, several studies have also determined their high toxicity when present as Co₃O₄. These results indicate the potential toxicity of produced NPs and their capacity to alter a local biome by their toxicity towards non-Co-resistant species including microbes (Khan et al. 2015; Moradpoor et al. 2019) and plants (Ghodake et al. 2011) (see Fig. 7, Processes 3 and 11).

In the warm to hot Australian arid and semi-arid environments where the sampling campaign was conducted, yearly rainfall is low, and soils typically have low moisture content. The degree to which Co is cycled in Co-rich environments varies, with dissolution rates (especially of ephemeral phases, compared to releasing Co from Co-rich Fe–Mn oxides) at a maximum following rainfall. In this regard, the Fe–Mn oxide-hosted cobalt prospects of Copperado and Calvert are different to their southern counterparts, with a clear distinction between dry and wet seasons despite the overall semi-arid conditions.

Association of Co with Mn and other elements

In most of our analysed samples, Co was found associated with Mn through both SP ICP-ToF-MS and XRF. The latter revealed groups of NPs varying in size from <1 µm to up to 20 µm (Fig. 3). The only larger sizes (up to 100 µm) of Co-enriched regolith zones were identified in highly Co-enriched samples from Mount Cobalt and Queen Sally (880- and >6600-ppm Co average concentrations at these two sites respectively). The chemical form of the Co-containing NPs is likely to be a Co–Mn oxide. The pure end-members of the Co and Mn oxyspinel family are respectively guite, Co₃O₄ (Lei et al. 2022), and hausmannite, Mn₃O₄ (Baron et al. 1998). Potential intermediates CoMn₂O₄ (Habjanič et al. 2014) and MnCo₂O₄ (Das Sharma et al. 2011), possible species of the Mn–Co oxide phases observed by XRF in this study, have not been described as minerals. The SP ICP-ToF-MS data show that Mn is generally the dominant element in the Co–Mn oxide NPs, with an example average 1:3 Co:Mn molar ratio at sampling site Phill4 shown in Fig. 4d. Particles with Co > Mn were observed at two sampling sites in weathering sulfarsenide localities (Co:Mn molar ratio of 1.79 at QS2 and MC1; Supplementary Table S1). However, for the samples in which ToF-MS was able to determine a Co:Mn molar ratio (13 samples), the ratio is <1 in 11 out of 13 samples being between 0.1 and 0.5 in nine of the samples, including all five non-sulfarsenide samples for which a ratio was calculable. Thus, the formula for most of the non-sulfarsenide Co–Mn NPs may be approximated as (Co_xMn_1−x)Mn₂O₄ if a spinel model is used. The association of Co with Mn has been observed in lateritic environments, as Mn^III and Mn^IV are able to oxidise Co^II into less mobile Co^III (either by inorganic or biologically mediated processes), allowing for structural incorporation into Mn oxides (Tanaka et al. 2013; Simanova and Peña 2015; Newsome et al. 2020).

Arsenic is often a concern when discussing the environmental impacts of cobalt sulfarsenide mining and ore processing. In the presence of iron oxides and under basic conditions, arsenic is usually stabilised as sorbed complexes on the (oxyhydr)oxides or even as a variety of iron arsenate in minerals such as scorodite (Filippou and Demopoulos 1997; Majzlan et al. 2012; Drahota et al. 2016). In the weathering environment around the Mount Cobalt and Queen Sally mines, the cobalt and arsenic concentrations are sufficiently high that cobalt arsenate minerals such as erythrite form on the surface of weathering ore (Kloprogge et al. 2006; Nagashima et al. 2021). In general, arsenic deportment and mobility in tailings has been better studied than that of cobalt, such as at abandoned As–Sb mines in Slovakia (Majzlan et al. 2011). However, with the operation of cobalt arsenide mines such as Bou Azzer, Morocco (Ahmed et al. 2009; Subías et al. 2022), and potential for sites such as Mount Cobalt to reopen to bolster cobalt supplies, understanding the combined behaviour of cobalt and arsenic is important to ensure neither of these elements become environmental contaminants.

Importance of nanoparticles for cycling of Co in surface environments

We detected Co- and Mn-containing NPs in all samples across all analysed sites using SP ICP-MS, encompassing different deposit and climate types as well as widely varying bulk cobalt concentrations, and found that they were associated on a single particle basis. Different mineralogical hosts for cobalt in the outcropping cobalt-rich rocks (Fe–Mn oxides, cobalt sulfarsenides, cobalt-rich pyrite and cobalt-rich carbonates) all contain cobalt-based NPs in surrounding soils. The host rocks provide a source for Co to be released into the environment, and regardless of original lithology (excepting the sulfarsenide sites) XRF maps showed the association of Co and Mn in nanoparticulate regolith phases. High As and comparatively low Mn contents at Queen Sally and Mount Cobalt result in the Co–Mn association not being observed as strongly at these two localities, but it is notable that even at low-Co and carbonate-hosted Young’s Cobalt prospect and the pyrite-hosted (i.e. high Fe) Pyrite Hill, the Co–Mn association in weathered regolith samples is strong.

The chemistry and reactivity of the Co-containing NPs is key to the behaviour of the Co stored within them. Unlike reactive elemental or sulfide NPs formed by some elements in the environment, oxide NPs are typically relatively stable (Nair et al. 2022). As minerals, the Co^III oxyhydroxide mineral heterogenite may persist for almost 1 × 10⁹ years (Decree et al. 2014) and oxyspinels such as hausmannite may also be stable for billions of years, for example the hausmannite deposits of Wessels-type ore in the Kalahari manganese field (Vafeas et al. 2019). Only more acidic conditions are likely to dissolve Co–Mn oxide phases (e.g. an acid rain event, or the acid-generating dissolution of nearby sulfarsenides or sulfides at relevant sites).

Links to anthropogenic cobalt

The concentrations of Co in natural environments are increasing due to anthropogenic factors, namely the use of metals like Co for new energy storage and generation technologies. Environments that are naturally enriched in Co provide the ideal sites to study Co (bio)geochemical processes, and conclusions drawn on Co cycling in naturally enriched environments are applicable to terrestrial landscapes in which Co is artificially enriched. Our study incorporates both largely undisturbed sites (Copperado, Calvert, Broken Hill) and previously disturbed sites at various stages of remediation (Queen Sally, Mount Cobalt, Young’s Cobalt prospect).

Our results suggest that the promotion of more basic conditions to encourage the formation of Mn–Co and Co-dominant (oxyhydr)oxides when processing Co-rich waste materials will result in the best chance of Co being immobilised and not being released to the environment. Acidic conditions should be avoided when possible to prevent Co, As and other critical or deleterious elements from being dissolved into surface waters. Although Co sulfide and sulfarsenide weathering may release acid, iron sulfides such as Co-bearing pyrite are more likely to be the larger contributors to acid mine drainage environments due to their volume compared to primary Co minerals in most environments. Desulfurisation of sulfides prior to mine waste disposal is likely to prevent mobilisation of Co in tailings dams.